project tests refactoring

tests/opencl/BlackScholes/BlackScholes.cl

@@ -0,0 +1,101 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+
+#if(0)
+    #define EXP(a) native_exp(a)
+    #define LOG(a) native_log(a)
+    #define SQRT(a) native_sqrt(a)
+#else
+    #define EXP(a) exp(a)
+    #define LOG(a) log(a)
+    #define SQRT(a) sqrt(a)
+#endif
+
+
+///////////////////////////////////////////////////////////////////////////////
+// Predefine functions to avoid bug in OpenCL compiler on Mac OSX 10.7 systems
+///////////////////////////////////////////////////////////////////////////////
+float CND(float d);
+void BlackScholesBody(__global float *call, __global float *put,  float S,
+					  float X, float T, float R, float V);
+
+///////////////////////////////////////////////////////////////////////////////
+// Rational approximation of cumulative normal distribution function
+///////////////////////////////////////////////////////////////////////////////
+float CND(float d){
+    const float       A1 = 0.31938153f;
+    const float       A2 = -0.356563782f;
+    const float       A3 = 1.781477937f;
+    const float       A4 = -1.821255978f;
+    const float       A5 = 1.330274429f;
+    const float RSQRT2PI = 0.39894228040143267793994605993438f;
+
+    float
+        K = 1.0f / (1.0f + 0.2316419f * fabs(d));
+
+    float
+        cnd = RSQRT2PI * EXP(- 0.5f * d * d) * 
+        (K * (A1 + K * (A2 + K * (A3 + K * (A4 + K * A5)))));
+
+    if(d > 0)
+        cnd = 1.0f - cnd;
+
+    return cnd;
+}
+
+
+///////////////////////////////////////////////////////////////////////////////
+// Black-Scholes formula for both call and put
+///////////////////////////////////////////////////////////////////////////////
+void BlackScholesBody(
+    __global float *call, //Call option price
+    __global float *put,  //Put option price
+    float S,              //Current stock price
+    float X,              //Option strike price
+    float T,              //Option years
+    float R,              //Riskless rate of return
+    float V               //Stock volatility
+){
+    float sqrtT = SQRT(T);
+    float    d1 = (LOG(S / X) + (R + 0.5f * V * V) * T) / (V * sqrtT);
+    float    d2 = d1 - V * sqrtT;
+    float CNDD1 = CND(d1);
+    float CNDD2 = CND(d2);
+
+    //Calculate Call and Put simultaneously
+    float expRT = EXP(- R * T);
+    *call = (S * CNDD1 - X * expRT * CNDD2);
+    *put  = (X * expRT * (1.0f - CNDD2) - S * (1.0f - CNDD1));
+}
+
+
+
+__kernel void BlackScholes(
+    __global float *d_Call, //Call option price
+    __global float *d_Put,  //Put option price
+    __global float *d_S,    //Current stock price
+    __global float *d_X,    //Option strike price
+    __global float *d_T,    //Option years
+    float R,                //Riskless rate of return
+    float V,                //Stock volatility
+    unsigned int optN
+){
+    for(unsigned int opt = get_global_id(0); opt < optN; opt += get_global_size(0))
+        BlackScholesBody(
+            &d_Call[opt],
+            &d_Put[opt],
+            d_S[opt],
+            d_X[opt],
+            d_T[opt],
+            R,
+            V
+        );
+}

tests/opencl/BlackScholes/Makefile

@@ -0,0 +1,65 @@
+RISCV_TOOLCHAIN_PATH ?= $(wildcard ../../../../riscv-gnu-toolchain/drops)
+POCL_CC_PATH ?= $(wildcard ../../../../pocl/drops_riscv_cc)
+POCL_INC_PATH ?= $(wildcard ../include)
+POCL_LIB_PATH ?= $(wildcard ../lib)
+VORTEX_RT_PATH ?= $(wildcard ../../../runtime)
+VX_SIMX_PATH ?= $(wildcard ../../../simX/obj_dir)
+
+CC  = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gcc
+CXX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-g++
+DMP = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objdump
+HEX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objcopy
+GDB = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gdb
+
+VX_SRCS =  $(VORTEX_RT_PATH)/newlib/newlib.c
+VX_SRCS += $(VORTEX_RT_PATH)/startup/vx_start.S
+VX_SRCS += $(VORTEX_RT_PATH)/intrinsics/vx_intrinsics.S
+VX_SRCS += $(VORTEX_RT_PATH)/io/vx_io.S $(VORTEX_RT_PATH)/io/vx_io.c
+VX_SRCS += $(VORTEX_RT_PATH)/fileio/fileio.S
+VX_SRCS += $(VORTEX_RT_PATH)/tests/tests.c
+VX_SRCS += $(VORTEX_RT_PATH)/vx_api/vx_api.c
+
+VX_CFLAGS = -nostartfiles -Wl,-Bstatic,-T,$(VORTEX_RT_PATH)/startup/vx_link.ld
+
+CXXFLAGS =  -g -O0 -march=rv32im -mabi=ilp32
+CXXFLAGS += -ffreestanding # program may not begin at main()
+CXXFLAGS += -Wl,--gc-sections # enable garbage collection of unused input sections
+CXXFLAGS += -fno-rtti -fno-non-call-exceptions # disable RTTI and exceptions
+CXXFLAGS += -I$(POCL_INC_PATH) -I.
+
+VX_LIBS = -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
+QEMU_LIBS = $(VORTEX_RT_PATH)/qemu/vx_api.c -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
+
+PROJECT=BlackScholes
+
+all: $(PROJECT).dump $(PROJECT).hex
+
+lib$(PROJECT).a: BlackScholes.cl
+	POCL_DEBUG=all POCL_DEBUG_LLVM_PASSES=1 LD_LIBRARY_PATH=$(RISCV_TOOLCHAIN_PATH)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -o lib$(PROJECT).a kernel.cl
+
+$(PROJECT).elf: main.cc lib$(PROJECT).a
+	$(CXX) $(CXXFLAGS) $(VX_CFLAGS) $(VX_SRCS) main.cc oclBlackScholes_common.h oclBlackScholes_launcher.cpp oclBlackScholes_gold.cpp $(VX_LIBS) -o $(PROJECT).elf
+
+$(PROJECT).qemu: main.cc lib$(PROJECT).a
+	$(CXX) $(CXXFLAGS) main.cc $(QEMU_LIBS) -o $(PROJECT).qemu
+
+$(PROJECT).hex: $(PROJECT).elf
+	$(HEX) -O ihex $(PROJECT).elf $(PROJECT).hex
+
+$(PROJECT).dump: $(PROJECT).elf
+	$(DMP) -D $(PROJECT).elf > $(PROJECT).dump
+
+run: $(PROJECT).hex
+	POCL_DEBUG=all $(VX_SIMX_PATH)/Vcache_simX -E -a rv32i --core $(PROJECT).hex -s -b 1> emulator.debug
+
+qemu: $(PROJECT).qemu
+	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -d in_asm -D debug.log $(PROJECT).qemu
+
+gdb-s: $(PROJECT).qemu
+	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -g 1234 -d in_asm -D debug.log $(PROJECT).qemu
+
+gdb-c: $(PROJECT).qemu
+	$(GDB) $(PROJECT).qemu
+
+clean:
+	rm -rf *.elf *.dump *.hex

tests/opencl/BlackScholes/main.cpp

@@ -0,0 +1,248 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+
+// standard utilities and systems includes
+#include <oclUtils.h>
+#include <shrQATest.h>
+#include "oclBlackScholes_common.h"
+
+////////////////////////////////////////////////////////////////////////////////
+// Helper functions
+////////////////////////////////////////////////////////////////////////////////
+double executionTime(cl_event &event){
+    cl_ulong start, end;
+
+    clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &end, NULL);
+    clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_START, sizeof(cl_ulong), &start, NULL);
+
+    return (double)1.0e-9 * (end - start); // convert nanoseconds to seconds on return
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// Random float helper
+////////////////////////////////////////////////////////////////////////////////
+float randFloat(float low, float high){
+    float t = (float)rand() / (float)RAND_MAX;
+    return (1.0f - t) * low + t * high;
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// Main program
+////////////////////////////////////////////////////////////////////////////////
+int main(int argc, char **argv)
+{
+    cl_platform_id   cpPlatform;       //OpenCL platform
+    cl_device_id*    cdDevices = NULL; //OpenCL devices list (array)
+    cl_context       cxGPUContext;     //OpenCL context
+    cl_command_queue cqCommandQueue;   //OpenCL command que
+    cl_mem                             //OpenCL memory buffer objects
+        d_Call,
+        d_Put,
+        d_S,
+        d_X,
+        d_T;
+
+    cl_int ciErrNum;
+
+    float
+        *h_CallCPU,
+        *h_PutCPU,
+        *h_CallGPU,
+        *h_PutGPU,
+        *h_S,
+        *h_X,
+        *h_T;
+
+    const unsigned int   optionCount = 4000000;
+    const float                    R = 0.02f;
+    const float                    V = 0.30f;
+
+    shrQAStart(argc, argv);
+
+    // Get the NVIDIA platform
+    ciErrNum = oclGetPlatformID(&cpPlatform);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
+    shrLog("clGetPlatformID...\n"); 
+
+    //Get all the devices
+    cl_uint uiNumDevices = 0;           // Number of devices available
+    cl_uint uiTargetDevice = 0;	        // Default Device to compute on
+    cl_uint uiNumComputeUnits;          // Number of compute units (SM's on NV GPU)
+    shrLog("Get the Device info and select Device...\n");
+    ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_DEFAULT, 0, NULL, &uiNumDevices);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
+    cdDevices = (cl_device_id *)malloc(uiNumDevices * sizeof(cl_device_id) );
+    ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_DEFAULT, uiNumDevices, cdDevices, NULL);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
+
+    // Get command line device options and config accordingly
+    shrLog("  # of Devices Available = %u\n", uiNumDevices); 
+    if(shrGetCmdLineArgumentu(argc, (const char**)argv, "device", &uiTargetDevice)== shrTRUE) 
+    {
+        uiTargetDevice = CLAMP(uiTargetDevice, 0, (uiNumDevices - 1));
+    }
+    shrLog("  Using Device %u: ", uiTargetDevice); 
+    oclPrintDevName(LOGBOTH, cdDevices[uiTargetDevice]);
+    ciErrNum = clGetDeviceInfo(cdDevices[uiTargetDevice], CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(uiNumComputeUnits), &uiNumComputeUnits, NULL);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
+    shrLog("\n  # of Compute Units = %u\n", uiNumComputeUnits); 
+
+    // set logfile name and start logs
+    shrSetLogFileName ("oclBlackScholes.txt");
+    shrLog("%s Starting...\n\n", argv[0]); 
+
+    shrLog("Allocating and initializing host memory...\n");
+        h_CallCPU = (float *)malloc(optionCount * sizeof(float));
+        h_PutCPU  = (float *)malloc(optionCount * sizeof(float));
+        h_CallGPU = (float *)malloc(optionCount * sizeof(float));
+        h_PutGPU  = (float *)malloc(optionCount * sizeof(float));
+        h_S       = (float *)malloc(optionCount * sizeof(float));
+        h_X       = (float *)malloc(optionCount * sizeof(float));
+        h_T       = (float *)malloc(optionCount * sizeof(float));
+
+        srand(2009);
+        for(unsigned int i = 0; i < optionCount; i++){
+            h_CallCPU[i] = -1.0f;
+            h_PutCPU[i]  = -1.0f;
+            h_S[i]       = randFloat(5.0f, 30.0f);
+            h_X[i]       = randFloat(1.0f, 100.0f);
+            h_T[i]       = randFloat(0.25f, 10.0f);
+        }
+
+    shrLog("Initializing OpenCL...\n");
+        // Get the NVIDIA platform
+        ciErrNum = oclGetPlatformID(&cpPlatform);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+
+        // Get a GPU device
+        ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_DEFAULT, 1, &cdDevices[uiTargetDevice], NULL);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+
+        // Create the context
+        cxGPUContext = clCreateContext(0, 1, &cdDevices[uiTargetDevice], NULL, NULL, &ciErrNum);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+
+        //Create a command-queue
+        cqCommandQueue = clCreateCommandQueue(cxGPUContext, cdDevices[uiTargetDevice], CL_QUEUE_PROFILING_ENABLE, &ciErrNum);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+
+    shrLog("Creating OpenCL memory objects...\n");
+        d_Call = clCreateBuffer(cxGPUContext, CL_MEM_READ_WRITE, optionCount * sizeof(float), NULL, &ciErrNum);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+        d_Put  = clCreateBuffer(cxGPUContext, CL_MEM_READ_WRITE, optionCount * sizeof(float), NULL, &ciErrNum);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+        d_S    = clCreateBuffer(cxGPUContext, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, optionCount * sizeof(float), h_S, &ciErrNum);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+        d_X    = clCreateBuffer(cxGPUContext, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, optionCount * sizeof(float), h_X, &ciErrNum);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+        d_T    = clCreateBuffer(cxGPUContext, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, optionCount * sizeof(float), h_T, &ciErrNum);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+
+    shrLog("Starting up BlackScholes...\n");
+        initBlackScholes(cxGPUContext, cqCommandQueue, (const char **)argv);
+
+    shrLog("Running OpenCL BlackScholes...\n\n");
+        //Just a single run or a warmup iteration
+        BlackScholes(
+            NULL,
+            d_Call,
+            d_Put,
+            d_S,
+            d_X,
+            d_T,
+            R,
+            V,
+            optionCount
+        );
+
+#ifdef GPU_PROFILING
+    const int numIterations = 16;
+    cl_event startMark, endMark;
+    ciErrNum = clEnqueueMarker(cqCommandQueue, &startMark);
+    ciErrNum |= clFinish(cqCommandQueue);
+    shrCheckError(ciErrNum, CL_SUCCESS);
+    shrDeltaT(0);
+
+    for(int i = 0; i < numIterations; i++){
+        BlackScholes(
+            cqCommandQueue,
+            d_Call,
+            d_Put,
+            d_S,
+            d_X,
+            d_T,
+            R,
+            V,
+            optionCount
+        );
+    }
+
+    ciErrNum  = clEnqueueMarker(cqCommandQueue, &endMark);
+    ciErrNum |= clFinish(cqCommandQueue);
+    shrCheckError(ciErrNum, CL_SUCCESS);
+
+    //Calculate performance metrics by wallclock time
+    double gpuTime = shrDeltaT(0) / numIterations;
+    shrLogEx(LOGBOTH | MASTER, 0, "oclBlackScholes, Throughput = %.4f GOptions/s, Time = %.5f s, Size = %u options, NumDevsUsed = %i, Workgroup = %u\n", 
+        (double)(2.0 * optionCount * 1.0e-9)/gpuTime, gpuTime, (2 * optionCount), 1, 0);
+
+    //Get profiling info
+    cl_ulong startTime = 0, endTime = 0;
+    ciErrNum  = clGetEventProfilingInfo(startMark, CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &startTime, NULL);
+    ciErrNum |= clGetEventProfilingInfo(endMark, CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &endTime, NULL);
+    shrCheckError(ciErrNum, CL_SUCCESS);
+    shrLog("\nOpenCL time: %.5f s\n\n", 1.0e-9 * ((double)endTime - (double)startTime) / (double)numIterations);
+#endif
+
+    shrLog("\nReading back OpenCL BlackScholes results...\n");
+        ciErrNum = clEnqueueReadBuffer(cqCommandQueue, d_Call, CL_TRUE, 0, optionCount * sizeof(float), h_CallGPU, 0, NULL, NULL);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+        ciErrNum = clEnqueueReadBuffer(cqCommandQueue, d_Put, CL_TRUE, 0, optionCount * sizeof(float), h_PutGPU, 0, NULL, NULL);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+
+    shrLog("Comparing against Host/C++ computation...\n"); 
+        BlackScholesCPU(h_CallCPU, h_PutCPU, h_S, h_X, h_T, R, V, optionCount);
+        double deltaCall = 0, deltaPut = 0, sumCall = 0, sumPut = 0;
+        double L1call, L1put;
+        for(unsigned int i = 0; i < optionCount; i++)
+        {
+            sumCall += fabs(h_CallCPU[i]);
+            sumPut  += fabs(h_PutCPU[i]);
+            deltaCall += fabs(h_CallCPU[i] - h_CallGPU[i]);
+            deltaPut  += fabs(h_PutCPU[i] - h_PutGPU[i]);
+        }
+        L1call = deltaCall / sumCall; 
+        L1put = deltaPut / sumPut;
+        shrLog("Relative L1 (call, put) = (%.3e, %.3e)\n\n", L1call, L1put);
+
+    shrLog("Shutting down...\n");
+        closeBlackScholes();
+        ciErrNum  = clReleaseMemObject(d_T);
+        ciErrNum |= clReleaseMemObject(d_X);
+        ciErrNum |= clReleaseMemObject(d_S);
+        ciErrNum |= clReleaseMemObject(d_Put);
+        ciErrNum |= clReleaseMemObject(d_Call);
+        ciErrNum |= clReleaseCommandQueue(cqCommandQueue);
+        ciErrNum |= clReleaseContext(cxGPUContext);
+        //oclCheckError(ciErrNum, CL_SUCCESS);
+
+        free(h_T);
+        free(h_X);
+        free(h_S);
+        free(h_PutGPU);
+        free(h_CallGPU);
+        free(h_PutCPU);
+        free(h_CallCPU);
+
+       if(cdDevices)free(cdDevices);
+
+        shrQAFinishExit(argc, (const char **)argv, ((L1call < 1E-6) && (L1put < 1E-6)) ? QA_PASSED : QA_FAILED );
+}

tests/opencl/BlackScholes/oclBlackScholes_common.h

@@ -0,0 +1,50 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+
+
+
+#include <oclUtils.h>
+
+
+
+////////////////////////////////////////////////////////////////////////////////
+// Process an array of optN options on CPU
+////////////////////////////////////////////////////////////////////////////////
+extern "C" void BlackScholesCPU(
+    float *h_Call, //Call option price
+    float *h_Put,  //Put option price
+    float *h_S,    //Current stock price
+    float *h_X,    //Option strike price
+    float *h_T,    //Option years
+    float R,       //Riskless rate of return
+    float V,       //Stock volatility
+    unsigned int optionCount
+);
+
+
+////////////////////////////////////////////////////////////////////////////////
+// OpenCL Black-Scholes kernel launcher
+////////////////////////////////////////////////////////////////////////////////
+extern "C" void initBlackScholes(cl_context cxGPUContext, cl_command_queue cqParamCommandQue, const char **argv);
+
+extern "C" void closeBlackScholes(void);
+
+extern "C" void BlackScholes(
+    cl_command_queue cqCommandQueue,
+    cl_mem d_Call, //Call option price
+    cl_mem d_Put,  //Put option price
+    cl_mem d_S,    //Current stock price
+    cl_mem d_X,    //Option strike price
+    cl_mem d_T,    //Option years
+    cl_float R,    //Riskless rate of return
+    cl_float V,    //Stock volatility
+    cl_uint optionCount
+);

tests/opencl/BlackScholes/oclBlackScholes_gold.cpp

@@ -0,0 +1,92 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+
+
+
+#include <math.h>
+#include "oclBlackScholes_common.h"
+
+
+
+///////////////////////////////////////////////////////////////////////////////
+// Rational approximation of cumulative normal distribution function
+///////////////////////////////////////////////////////////////////////////////
+static double CND(double d){
+    const double       A1 = 0.31938153;
+    const double       A2 = -0.356563782;
+    const double       A3 = 1.781477937;
+    const double       A4 = -1.821255978;
+    const double       A5 = 1.330274429;
+    const double RSQRT2PI = 0.39894228040143267793994605993438;
+
+    double
+        K = 1.0 / (1.0 + 0.2316419 * fabs(d));
+
+    double
+        cnd = RSQRT2PI * exp(- 0.5 * d * d) * 
+        (K * (A1 + K * (A2 + K * (A3 + K * (A4 + K * A5)))));
+
+    if(d > 0)
+        cnd = 1.0 - cnd;
+
+    return cnd;
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// Black-Scholes formula for both call and put
+///////////////////////////////////////////////////////////////////////////////
+static void BlackScholesBodyCPU(
+    float& call, //Call option price
+    float& put,  //Put option price
+    float Sf,    //Current stock price
+    float Xf,    //Option strike price
+    float Tf,    //Option years
+    float Rf,    //Riskless rate of return
+    float Vf     //Stock volatility
+){
+    double S = Sf, X = Xf, T = Tf, R = Rf, V = Vf;
+
+    double sqrtT = sqrt(T);
+    double    d1 = (log(S / X) + (R + 0.5 * V * V) * T) / (V * sqrtT);
+    double    d2 = d1 - V * sqrtT;
+    double CNDD1 = CND(d1);
+    double CNDD2 = CND(d2);
+
+    //Calculate Call and Put simultaneously
+    double expRT = exp(- R * T);
+    call = (float)(S * CNDD1 - X * expRT * CNDD2);
+    put  = (float)(X * expRT * (1.0 - CNDD2) - S * (1.0 - CNDD1));
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// Process an array of optN options
+////////////////////////////////////////////////////////////////////////////////
+extern "C" void BlackScholesCPU(
+    float *h_Call, //Call option price
+    float *h_Put,  //Put option price
+    float *h_S,    //Current stock price
+    float *h_X,    //Option strike price
+    float *h_T,    //Option years
+    float R,       //Riskless rate of return
+    float V,       //Stock volatility
+    unsigned int optionCount
+){
+    for(unsigned int i = 0; i < optionCount; i++)
+        BlackScholesBodyCPU(
+            h_Call[i],
+            h_Put[i],
+            h_S[i],
+            h_X[i],
+            h_T[i],
+            R,
+            V
+        );
+}

tests/opencl/BlackScholes/oclBlackScholes_launcher.cpp

@@ -0,0 +1,125 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+
+
+
+#include <oclUtils.h>
+#include "oclBlackScholes_common.h"
+
+static cl_program cpBlackScholes;   //OpenCL program
+static cl_kernel  ckBlackScholes;   //OpenCL kernel
+static cl_command_queue cqDefaultCommandQueue;
+
+extern "C" void initBlackScholes(cl_context cxGPUContext, cl_command_queue cqParamCommandQueue, const char **argv){
+    cl_int ciErrNum;
+    size_t kernelLength;
+
+    shrLog("...loading BlackScholes.cl\n");
+        char *cPathAndName = shrFindFilePath("BlackScholes.cl", argv[0]);
+        shrCheckError(cPathAndName != NULL, shrTRUE);
+        char *cBlackScholes = oclLoadProgSource(cPathAndName, "// My comment\n", &kernelLength);
+        shrCheckError(cBlackScholes != NULL, shrTRUE);
+
+    shrLog("...creating BlackScholes program\n");
+        //cpBlackScholes = clCreateProgramWithSource(cxGPUContext, 1, (const char **)&cBlackScholes, &kernelLength, &ciErrNum);
+        cpBlackScholes = clCreateProgramWithBuiltInKernels(context, 1, &device_id, "BlackScholes", NULL);
+        shrCheckError(ciErrNum, CL_SUCCESS);
+
+    shrLog("...building BlackScholes program\n");
+        ciErrNum = clBuildProgram(cpBlackScholes, 0, NULL, "-cl-fast-relaxed-math -Werror", NULL, NULL);
+
+        if(ciErrNum != CL_BUILD_SUCCESS){
+            shrLog("*** Compilation failure ***\n");
+
+            size_t deviceNum;
+            cl_device_id *cdDevices;
+            ciErrNum = clGetContextInfo(cxGPUContext, CL_CONTEXT_DEVICES, 0, NULL, &deviceNum);
+            shrCheckError(ciErrNum, CL_SUCCESS);
+
+            cdDevices = (cl_device_id *)malloc(deviceNum * sizeof(cl_device_id));
+            shrCheckError(cdDevices != NULL, shrTRUE);
+
+            ciErrNum = clGetContextInfo(cxGPUContext, CL_CONTEXT_DEVICES, deviceNum * sizeof(cl_device_id), cdDevices, NULL);
+            shrCheckError(ciErrNum, CL_SUCCESS);
+
+            size_t logSize;
+            char *logTxt;
+
+            ciErrNum = clGetProgramBuildInfo(cpBlackScholes, cdDevices[0], CL_PROGRAM_BUILD_LOG, 0, NULL, &logSize);
+            shrCheckError(ciErrNum, CL_SUCCESS);
+
+            logTxt = (char *)malloc(logSize);
+            shrCheckError(logTxt != NULL, shrTRUE);
+
+            ciErrNum = clGetProgramBuildInfo(cpBlackScholes, cdDevices[0], CL_PROGRAM_BUILD_LOG, logSize, logTxt, NULL);
+            shrCheckError(ciErrNum, CL_SUCCESS);
+
+            shrLog("%s\n", logTxt);
+            shrLog("*** Exiting ***\n");
+            free(logTxt);
+            free(cdDevices);
+            exit(666);
+        }
+
+    //Save ptx code to separate file
+    oclLogPtx(cpBlackScholes, oclGetFirstDev(cxGPUContext), "BlackScholes.ptx");
+
+    shrLog("...creating BlackScholes kernels\n");
+        ckBlackScholes = clCreateKernel(cpBlackScholes, "BlackScholes", &ciErrNum);
+        shrCheckError(ciErrNum, CL_SUCCESS);
+
+    cqDefaultCommandQueue = cqParamCommandQueue;
+    free(cBlackScholes);
+    free(cPathAndName);
+}
+
+extern "C" void closeBlackScholes(void){
+    cl_int ciErrNum;
+    ciErrNum  = clReleaseKernel(ckBlackScholes);
+    ciErrNum |= clReleaseProgram(cpBlackScholes);
+    shrCheckError(ciErrNum, CL_SUCCESS);
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// OpenCL Black-Scholes kernel launcher
+////////////////////////////////////////////////////////////////////////////////
+extern "C" void BlackScholes(
+    cl_command_queue cqCommandQueue,
+    cl_mem d_Call, //Call option price
+    cl_mem d_Put,  //Put option price
+    cl_mem d_S,    //Current stock price
+    cl_mem d_X,    //Option strike price
+    cl_mem d_T,    //Option years
+    cl_float R,    //Riskless rate of return
+    cl_float V,    //Stock volatility
+    cl_uint optionCount
+){
+    cl_int ciErrNum;
+
+    if(!cqCommandQueue)
+        cqCommandQueue = cqDefaultCommandQueue;
+
+    ciErrNum  = clSetKernelArg(ckBlackScholes, 0, sizeof(cl_mem),   (void *)&d_Call);
+    ciErrNum |= clSetKernelArg(ckBlackScholes, 1, sizeof(cl_mem),   (void *)&d_Put);
+    ciErrNum |= clSetKernelArg(ckBlackScholes, 2, sizeof(cl_mem),   (void *)&d_S);
+    ciErrNum |= clSetKernelArg(ckBlackScholes, 3, sizeof(cl_mem),   (void *)&d_X);
+    ciErrNum |= clSetKernelArg(ckBlackScholes, 4, sizeof(cl_mem),   (void *)&d_T);
+    ciErrNum |= clSetKernelArg(ckBlackScholes, 5, sizeof(cl_float), (void *)&R);
+    ciErrNum |= clSetKernelArg(ckBlackScholes, 6, sizeof(cl_float), (void *)&V);
+    ciErrNum |= clSetKernelArg(ckBlackScholes, 7, sizeof(cl_uint),  (void *)&optionCount);
+    shrCheckError(ciErrNum, CL_SUCCESS);
+
+    //Run the kernel
+    size_t globalWorkSize = 60 * 1024;
+	size_t localWorkSize = 128;
+    ciErrNum = clEnqueueNDRangeKernel(cqCommandQueue, ckBlackScholes, 1, NULL, &globalWorkSize, &localWorkSize, 0, NULL, NULL);
+    shrCheckError(ciErrNum, CL_SUCCESS);
+}

tests/opencl/BlackScholes/oclUtils.h

@@ -0,0 +1,198 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+ 
+#ifndef OCL_UTILS_H
+#define OCL_UTILS_H
+
+// *********************************************************************
+// Utilities specific to OpenCL samples in NVIDIA GPU Computing SDK 
+// *********************************************************************
+
+// Common headers:  Cross-API utililties and OpenCL header
+#include <shrUtils.h>
+
+// All OpenCL headers
+#if defined (__APPLE__) || defined(MACOSX)
+    #include <OpenCL/opencl.h>
+#else
+    #include <CL/opencl.h>
+#endif 
+
+// Includes
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+
+// For systems with CL_EXT that are not updated with these extensions, we copied these
+// extensions from <CL/cl_ext.h>
+#ifndef CL_DEVICE_COMPUTE_CAPABILITY_MAJOR_NV
+  /* cl_nv_device_attribute_query extension - no extension #define since it has no functions */
+  #define CL_DEVICE_COMPUTE_CAPABILITY_MAJOR_NV       0x4000
+  #define CL_DEVICE_COMPUTE_CAPABILITY_MINOR_NV       0x4001
+  #define CL_DEVICE_REGISTERS_PER_BLOCK_NV            0x4002
+  #define CL_DEVICE_WARP_SIZE_NV                      0x4003
+  #define CL_DEVICE_GPU_OVERLAP_NV                    0x4004
+  #define CL_DEVICE_KERNEL_EXEC_TIMEOUT_NV            0x4005
+  #define CL_DEVICE_INTEGRATED_MEMORY_NV              0x4006
+#endif
+
+// reminders for build output window and log
+#ifdef _WIN32
+    #pragma message ("Note: including shrUtils.h")
+    #pragma message ("Note: including opencl.h")
+#endif
+
+// SDK Revision #
+#define OCL_SDKREVISION "7027912"
+
+// Error and Exit Handling Macros... 
+// *********************************************************************
+// Full error handling macro with Cleanup() callback (if supplied)... 
+// (Companion Inline Function lower on page)
+#define oclCheckErrorEX(a, b, c) __oclCheckErrorEX(a, b, c, __FILE__ , __LINE__) 
+
+// Short version without Cleanup() callback pointer
+// Both Input (a) and Reference (b) are specified as args
+#define oclCheckError(a, b) oclCheckErrorEX(a, b, 0) 
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the platform ID for NVIDIA if available, otherwise default to platform 0
+//!
+//! @return the id 
+//! @param clSelectedPlatformID         OpenCL platform ID
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_int oclGetPlatformID(cl_platform_id* clSelectedPlatformID);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Print info about the device
+//!
+//! @param iLogMode       enum LOGBOTH, LOGCONSOLE, LOGFILE
+//! @param device         OpenCL id of the device
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclPrintDevInfo(int iLogMode, cl_device_id device);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get and return device capability
+//!
+//! @return the 2 digit integer representation of device Cap (major minor). return -1 if NA 
+//! @param device         OpenCL id of the device
+//////////////////////////////////////////////////////////////////////////////
+extern "C" int oclGetDevCap(cl_device_id device);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Print the device name
+//!
+//! @param iLogMode       enum LOGBOTH, LOGCONSOLE, LOGFILE
+//! @param device         OpenCL id of the device
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclPrintDevName(int iLogMode, cl_device_id device);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the id of the first device from the context
+//!
+//! @return the id 
+//! @param cxGPUContext         OpenCL context
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_device_id oclGetFirstDev(cl_context cxGPUContext);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the id of the nth device from the context
+//!
+//! @return the id or -1 when out of range
+//! @param cxGPUContext         OpenCL context
+//! @param device_idx            index of the device of interest
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_device_id oclGetDev(cl_context cxGPUContext, unsigned int device_idx);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the id of device with maximal FLOPS from the context
+//!
+//! @return the id 
+//! @param cxGPUContext         OpenCL context
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_device_id oclGetMaxFlopsDev(cl_context cxGPUContext);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Loads a Program file and prepends the cPreamble to the code.
+//!
+//! @return the source string if succeeded, 0 otherwise
+//! @param cFilename        program filename
+//! @param cPreamble        code that is prepended to the loaded file, typically a set of #defines or a header
+//! @param szFinalLength    returned length of the code string
+//////////////////////////////////////////////////////////////////////////////
+extern "C" char* oclLoadProgSource(const char* cFilename, const char* cPreamble, size_t* szFinalLength);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get the binary (PTX) of the program associated with the device
+//!
+//! @param cpProgram    OpenCL program
+//! @param cdDevice     device of interest
+//! @param binary       returned code
+//! @param length       length of returned code
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclGetProgBinary( cl_program cpProgram, cl_device_id cdDevice, char** binary, size_t* length);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get and log the binary (PTX) from the OpenCL compiler for the requested program & device
+//!
+//! @param cpProgram                   OpenCL program
+//! @param cdDevice                    device of interest
+//! @param const char*  cPtxFileName   optional PTX file name
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclLogPtx(cl_program cpProgram, cl_device_id cdDevice, const char* cPtxFileName);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get and log the Build Log from the OpenCL compiler for the requested program & device
+//!
+//! @param cpProgram    OpenCL program
+//! @param cdDevice     device of interest
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclLogBuildInfo(cl_program cpProgram, cl_device_id cdDevice);
+
+// Helper function for De-allocating cl objects
+// *********************************************************************
+extern "C" void oclDeleteMemObjs(cl_mem* cmMemObjs, int iNumObjs);
+
+// Helper function to get OpenCL error string from constant
+// *********************************************************************
+extern "C" const char* oclErrorString(cl_int error);
+
+// Helper function to get OpenCL image format string (channel order and type) from constant
+// *********************************************************************
+extern "C" const char* oclImageFormatString(cl_uint uiImageFormat);
+
+// companion inline function for error checking and exit on error WITH Cleanup Callback (if supplied)
+// *********************************************************************
+inline void __oclCheckErrorEX(cl_int iSample, cl_int iReference, void (*pCleanup)(int), const char* cFile, const int iLine)
+{
+    // An error condition is defined by the sample/test value not equal to the reference
+    if (iReference != iSample)
+    {
+        // If the sample/test value isn't equal to the ref, it's an error by defnition, so override 0 sample/test value
+        iSample = (iSample == 0) ? -9999 : iSample; 
+
+        // Log the error info
+        shrLog("\n !!! Error # %i (%s) at line %i , in file %s !!!\n\n", iSample, oclErrorString(iSample), iLine, cFile);
+
+        // Cleanup and exit, or just exit if no cleanup function pointer provided.  Use iSample (error code in this case) as process exit code.
+        if (pCleanup != NULL)
+        {
+            pCleanup(iSample);
+        }
+        else 
+        {
+            shrLogEx(LOGBOTH | CLOSELOG, 0, "Exiting...\n");
+            exit(iSample);
+        }
+    }
+}
+
+#endif

tests/opencl/BlackScholes/shrQATest.h

@@ -0,0 +1,238 @@
+/*
+* Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+*
+* Please refer to the NVIDIA end user license agreement (EULA) associated
+* with this source code for terms and conditions that govern your use of
+* this software. Any use, reproduction, disclosure, or distribution of
+* this software and related documentation outside the terms of the EULA
+* is strictly prohibited.
+*
+*/
+
+#ifndef SHR_QATEST_H
+#define SHR_QATEST_H
+
+// *********************************************************************
+// Generic utilities for NVIDIA GPU Computing SDK 
+// *********************************************************************
+
+// OS dependent includes
+#ifdef _WIN32
+    #pragma message ("Note: including windows.h")
+    #pragma message ("Note: including math.h")
+    #pragma message ("Note: including assert.h")
+    #pragma message ("Note: including time.h")
+
+// Headers needed for Windows
+    #include <windows.h>
+	#include <time.h>
+#else
+    // Headers needed for Linux
+    #include <sys/stat.h>
+    #include <sys/types.h>
+    #include <sys/time.h>
+    #include <stdio.h>
+    #include <stdlib.h>
+    #include <string.h>
+    #include <stdarg.h>
+    #include <unistd.h>
+    #include <time.h>
+#endif
+
+#ifndef STRCASECMP
+#ifdef _WIN32
+#define STRCASECMP _stricmp
+#else
+#define STRCASECMP strcasecmp
+#endif
+#endif
+
+#ifndef STRNCASECMP
+#ifdef _WIN32
+#define STRNCASECMP _strnicmp
+#else
+#define STRNCASECMP strncasecmp
+#endif
+#endif
+
+
+// Standardized QA Start/Finish for CUDA SDK tests
+#define shrQAStart(a, b)      __shrQAStart(a, b)
+#define shrQAFinish(a, b, c)  __shrQAFinish(a, b, c)
+#define shrQAFinish2(a, b, c, d) __shrQAFinish2(a, b, c, d)
+
+inline int findExeNameStart(const char *exec_name)
+{
+    int exename_start = (int)strlen(exec_name);
+
+    while( (exename_start > 0) && 
+            (exec_name[exename_start] != '\\') && 
+            (exec_name[exename_start] != '/') )
+    {
+        exename_start--;
+    }
+    if (exec_name[exename_start] == '\\' || 
+        exec_name[exename_start] == '/')
+    {
+        return exename_start+1;
+    } else {
+        return exename_start;
+    }
+}
+
+inline int __shrQAStart(int argc, char **argv)
+{
+    bool bQATest = false;
+    // First clear the output buffer
+    fflush(stdout);
+    fflush(stdout);
+
+    for (int i=1; i < argc; i++) {
+        int string_start = 0;
+        while (argv[i][string_start] == '-')
+           string_start++;
+        char *string_argv = &argv[i][string_start];
+
+        if (!STRCASECMP(string_argv, "qatest")) {
+           bQATest = true;
+        }
+    }
+    
+    // We don't want to print the entire path, so we search for the first 
+    int exename_start = findExeNameStart(argv[0]);
+    if (bQATest) {
+        fprintf(stdout, "&&&& RUNNING %s", &(argv[0][exename_start]));
+        for (int i=1; i < argc; i++) fprintf(stdout, " %s", argv[i]);
+        fprintf(stdout, "\n");
+    } else {
+        fprintf(stdout, "[%s] starting...\n", &(argv[0][exename_start]));
+    }
+    fflush(stdout);
+    printf("\n"); fflush(stdout);
+    return exename_start;
+}
+
+enum eQAstatus {
+    QA_FAILED = 0,
+    QA_PASSED = 1,
+    QA_WAIVED = 2
+};
+
+inline void __ExitInTime(int seconds)
+{
+    fprintf(stdout, "> exiting in %d seconds: ", seconds);
+    fflush(stdout);
+    time_t t;
+    int count;
+    for (t=time(0)+seconds, count=seconds; time(0) < t; count--) {
+        fprintf(stdout, "%d...", count);
+#ifdef WIN32
+        Sleep(1000);
+#else
+        sleep(1);
+#endif
+    }
+    fprintf(stdout,"done!\n\n"); 
+	fflush(stdout);
+}
+
+
+inline void __shrQAFinish(int argc, const char **argv, int iStatus)
+{
+    // By default QATest is disabled and NoPrompt is Enabled (times out at seconds passed into __ExitInTime() )
+    bool bQATest = false, bNoPrompt = true, bQuitInTime = true;
+    const char *sStatus[] = { "FAILED", "PASSED", "WAIVED", NULL };
+	
+    for (int i=1; i < argc; i++) {
+        int string_start = 0;
+        while (argv[i][string_start] == '-')
+           string_start++;
+
+        const char *string_argv = &argv[i][string_start];
+        if (!STRCASECMP(string_argv, "qatest")) {
+           bQATest = true;
+        }	
+        // For SDK individual samples that don't specify -noprompt or -prompt, 
+        // a 3 second delay will happen before exiting, giving a user time to view results
+        if (!STRCASECMP(string_argv, "noprompt") || !STRCASECMP(string_argv, "help")) {
+            bNoPrompt = true;
+            bQuitInTime = false;
+        }
+        if (!STRCASECMP(string_argv, "prompt")) {
+            bNoPrompt = false;
+            bQuitInTime = false;
+        }
+    }
+
+    int exename_start = findExeNameStart(argv[0]);
+    if (bQATest) {
+        fprintf(stdout, "&&&& %s %s", sStatus[iStatus], &(argv[0][exename_start]));
+        for (int i=1; i < argc; i++) fprintf(stdout, " %s", argv[i]);
+        fprintf(stdout, "\n");
+    } else {
+        fprintf(stdout, "[%s] test results...\n%s\n", &(argv[0][exename_start]), sStatus[iStatus]);
+    }
+    fflush(stdout);
+    printf("\n"); fflush(stdout);
+    if (bQuitInTime) {
+        __ExitInTime(3);
+    } else {
+        if (!bNoPrompt) {
+            fprintf(stdout, "\nPress <Enter> to exit...\n");
+            fflush(stdout);
+            getchar();
+        }
+    }
+}
+
+inline void __shrQAFinish2(bool bQATest, int argc, const char **argv, int iStatus)
+{
+    bool bQuitInTime = true;
+    const char *sStatus[] = { "FAILED", "PASSED", "WAIVED", NULL };
+	
+    for (int i=1; i < argc; i++) {
+        int string_start = 0;
+        while (argv[i][string_start] == '-')
+           string_start++;
+
+        const char *string_argv = &argv[i][string_start];
+        // For SDK individual samples that don't specify -noprompt or -prompt, 
+        // a 3 second delay will happen before exiting, giving a user time to view results
+        if (!STRCASECMP(string_argv, "noprompt") || !STRCASECMP(string_argv, "help")) {
+            bQuitInTime = false;
+        }
+        if (!STRCASECMP(string_argv, "prompt")) {
+            bQuitInTime = false;
+        }
+    }
+
+    int exename_start = findExeNameStart(argv[0]);
+    if (bQATest) {
+        fprintf(stdout, "&&&& %s %s", sStatus[iStatus], &(argv[0][exename_start]));
+        for (int i=1; i < argc; i++) fprintf(stdout, " %s", argv[i]);
+        fprintf(stdout, "\n");
+    } else {
+        fprintf(stdout, "[%s] test results...\n%s\n", &(argv[0][exename_start]), sStatus[iStatus]);
+    }
+    fflush(stdout);
+    
+    if (bQuitInTime) {
+        __ExitInTime(3);
+    }
+}
+
+inline void shrQAFinishExit(int argc, const char **argv, int iStatus)
+{
+    __shrQAFinish(argc, argv, iStatus);
+
+    exit(iStatus ? EXIT_SUCCESS : EXIT_FAILURE); 
+}
+
+inline void shrQAFinishExit2(bool bQAtest, int argc, const char **argv, int iStatus)
+{
+    __shrQAFinish2(bQAtest, argc, argv, iStatus);
+
+    exit(iStatus ? EXIT_SUCCESS : EXIT_FAILURE);
+}
+
+#endif

tests/opencl/BlackScholes/shrUtils.h

@@ -0,0 +1,642 @@
+/*
+* Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+*
+* Please refer to the NVIDIA end user license agreement (EULA) associated
+* with this source code for terms and conditions that govern your use of
+* this software. Any use, reproduction, disclosure, or distribution of
+* this software and related documentation outside the terms of the EULA
+* is strictly prohibited.
+*
+*/
+
+#ifndef SHR_UTILS_H
+#define SHR_UTILS_H
+
+// *********************************************************************
+// Generic utilities for NVIDIA GPU Computing SDK 
+// *********************************************************************
+
+// reminders for output window and build log
+#ifdef _WIN32
+    #pragma message ("Note: including windows.h")
+    #pragma message ("Note: including math.h")
+    #pragma message ("Note: including assert.h")
+#endif
+
+// OS dependent includes
+#ifdef _WIN32
+    // Headers needed for Windows
+    #include <windows.h>
+#else
+    // Headers needed for Linux
+    #include <sys/stat.h>
+    #include <sys/types.h>
+    #include <sys/time.h>
+    #include <stdio.h>
+    #include <stdlib.h>
+    #include <string.h>
+    #include <stdarg.h>
+#endif
+
+// Other headers needed for both Windows and Linux
+#include <math.h>
+#include <assert.h>
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+
+// Un-comment the following #define to enable profiling code in SDK apps
+//#define GPU_PROFILING
+
+// Beginning of GPU Architecture definitions
+inline int ConvertSMVer2Cores(int major, int minor)
+{
+	// Defines for GPU Architecture types (using the SM version to determine the # of cores per SM
+	typedef struct {
+		int SM; // 0xMm (hexidecimal notation), M = SM Major version, and m = SM minor version
+		int Cores;
+	} sSMtoCores;
+
+	sSMtoCores nGpuArchCoresPerSM[] = 
+	{ { 0x10,  8 }, // Tesla Generation (SM 1.0) G80 class
+	  { 0x11,  8 }, // Tesla Generation (SM 1.1) G8x class
+	  { 0x12,  8 }, // Tesla Generation (SM 1.2) G9x class
+	  { 0x13,  8 }, // Tesla Generation (SM 1.3) GT200 class
+	  { 0x20, 32 }, // Fermi Generation (SM 2.0) GF100 class
+	  { 0x21, 48 }, // Fermi Generation (SM 2.1) GF10x class
+	  { 0x30, 192}, // Fermi Generation (SM 3.0) GK10x class
+	  {   -1, -1 }
+	};
+
+	int index = 0;
+	while (nGpuArchCoresPerSM[index].SM != -1) {
+		if (nGpuArchCoresPerSM[index].SM == ((major << 4) + minor) ) {
+			return nGpuArchCoresPerSM[index].Cores;
+		}
+		index++;
+	}
+	printf("MapSMtoCores SM %d.%d is undefined (please update to the latest SDK)!\n", major, minor);
+	return -1;
+}
+// end of GPU Architecture definitions
+
+
+// Defines and enum for use with logging functions
+// *********************************************************************
+#define DEFAULTLOGFILE "SdkConsoleLog.txt"
+#define MASTERLOGFILE "SdkMasterLog.csv"
+enum LOGMODES 
+{
+    LOGCONSOLE = 1, // bit to signal "log to console" 
+    LOGFILE    = 2, // bit to signal "log to file" 
+    LOGBOTH    = 3, // convenience union of first 2 bits to signal "log to both"
+    APPENDMODE = 4, // bit to set "file append" mode instead of "replace mode" on open
+    MASTER     = 8, // bit to signal master .csv log output
+    ERRORMSG   = 16, // bit to signal "pre-pend Error" 
+    CLOSELOG   = 32  // bit to close log file, if open, after any requested file write
+};
+#define HDASHLINE "-----------------------------------------------------------\n"
+
+// Standardized boolean
+enum shrBOOL
+{
+    shrFALSE = 0,
+    shrTRUE = 1
+};
+
+// Standardized MAX, MIN and CLAMP
+#define MAX(a, b) ((a > b) ? a : b)
+#define MIN(a, b) ((a < b) ? a : b)
+#define CLAMP(a, b, c) MIN(MAX(a, b), c)    // double sided clip of input a
+#define TOPCLAMP(a, b) (a < b ? a:b)	    // single top side clip of input a
+
+// Error and Exit Handling Macros... 
+// *********************************************************************
+// Full error handling macro with Cleanup() callback (if supplied)... 
+// (Companion Inline Function lower on page)
+#define shrCheckErrorEX(a, b, c) __shrCheckErrorEX(a, b, c, __FILE__ , __LINE__) 
+
+// Short version without Cleanup() callback pointer
+// Both Input (a) and Reference (b) are specified as args
+#define shrCheckError(a, b) shrCheckErrorEX(a, b, 0) 
+
+// Standardized Exit Macro for leaving main()... extended version
+// (Companion Inline Function lower on page)
+#define shrExitEX(a, b, c) __shrExitEX(a, b, c)
+
+// Standardized Exit Macro for leaving main()... short version
+// (Companion Inline Function lower on page)
+#define shrEXIT(a, b)        __shrExitEX(a, b, EXIT_SUCCESS)
+
+// Simple argument checker macro
+#define ARGCHECK(a) if((a) != shrTRUE)return shrFALSE 
+
+// Define for user-customized error handling
+#define STDERROR "file %s, line %i\n\n" , __FILE__ , __LINE__
+
+// Function to deallocate memory allocated within shrUtils
+// *********************************************************************
+extern "C" void shrFree(void* ptr);
+
+// *********************************************************************
+// Helper function to log standardized information to Console, to File or to both
+//! Examples: shrLogEx(LOGBOTH, 0, "Function A\n"); 
+//!         : shrLogEx(LOGBOTH | ERRORMSG, ciErrNum, STDERROR);
+//! 
+//! Automatically opens file and stores handle if needed and not done yet
+//! Closes file and nulls handle on request
+//! 
+//! @param 0 iLogMode: LOGCONSOLE, LOGFILE, LOGBOTH, APPENDMODE, MASTER, ERRORMSG, CLOSELOG.  
+//!          LOGFILE and LOGBOTH may be | 'd  with APPENDMODE to select file append mode instead of overwrite mode 
+//!          LOGFILE and LOGBOTH may be | 'd  with CLOSELOG to "write and close" 
+//!          First 3 options may be | 'd  with MASTER to enable independent write to master data log file
+//!          First 3 options may be | 'd  with ERRORMSG to start line with standard error message
+//! @param 2 dValue:    
+//!          Positive val = double value for time in secs to be formatted to 6 decimals. 
+//!          Negative val is an error code and this give error preformatting.
+//! @param 3 cFormatString: String with formatting specifiers like printf or fprintf.  
+//!          ALL printf flags, width, precision and type specifiers are supported with this exception: 
+//!              Wide char type specifiers intended for wprintf (%S and %C) are NOT supported
+//!              Single byte char type specifiers (%s and %c) ARE supported 
+//! @param 4... variable args: like printf or fprintf.  Must match format specifer type above.  
+//! @return 0 if OK, negative value on error or if error occurs or was passed in. 
+// *********************************************************************
+extern "C" int shrLogEx(int iLogMode, int iErrNum, const char* cFormatString, ...);
+
+// Short version of shrLogEx defaulting to shrLogEx(LOGBOTH, 0, 
+// *********************************************************************
+extern "C" int shrLog(const char* cFormatString, ...);
+
+// *********************************************************************
+// Delta timer function for up to 3 independent timers using host high performance counters 
+// Maintains state for 3 independent counters
+//! Example: double dElapsedTime = shrDeltaTime(0);
+//! 
+//! @param 0 iCounterID: Which timer to check/reset. (0, 1, 2)
+//! @return delta time of specified counter since last call in seconds.  Otherwise -9999.0 if error
+// *********************************************************************
+extern "C" double shrDeltaT(int iCounterID);
+
+// Optional LogFileNameOverride function
+// *********************************************************************
+extern "C" void shrSetLogFileName (const char* cOverRideName);
+
+// Helper function to init data arrays 
+// *********************************************************************
+extern "C" void shrFillArray(float* pfData, int iSize);
+
+// Helper function to print data arrays 
+// *********************************************************************
+extern "C" void shrPrintArray(float* pfData, int iSize);
+
+////////////////////////////////////////////////////////////////////////////
+//! Find the path for a filename
+//! @return the path if succeeded, otherwise 0
+//! @param filename        name of the file
+//! @param executablePath  optional absolute path of the executable
+////////////////////////////////////////////////////////////////////////////
+extern "C" char* shrFindFilePath(const char* filename, const char* executablePath);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing single precision floating point data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFilef( const char* filename, float** data, unsigned int* len, 
+              bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing double precision floating point data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFiled( const char* filename, double** data, unsigned int* len, 
+              bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing integer data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFilei( const char* filename, int** data, unsigned int* len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing unsigned integer data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is 
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFileui( const char* filename, unsigned int** data, 
+               unsigned int* len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing char / byte data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is 
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFileb( const char* filename, char** data, unsigned int* len, 
+              bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing unsigned char / byte data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFileub( const char* filename, unsigned char** data, 
+               unsigned int* len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing single precision floating point 
+//! data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+//! @param epsilon  epsilon for comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFilef( const char* filename, const float* data, unsigned int len,
+               const float epsilon, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing double precision floating point 
+//! data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+//! @param epsilon  epsilon for comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFiled( const char* filename, const float* data, unsigned int len,
+               const double epsilon, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing integer data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFilei( const char* filename, const int* data, unsigned int len,
+               bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing unsigned integer data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFileui( const char* filename, const unsigned int* data, 
+                unsigned int len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing char / byte data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFileb( const char* filename, const char* data, unsigned int len, 
+               bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing unsigned char / byte data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFileub( const char* filename, const unsigned char* data,
+                unsigned int len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Load PPM image file (with unsigned char as data element type), padding 
+//! 4th component
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param OutData  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+//! 
+//! Note: If *OutData is NULL this function allocates buffer that must be freed by caller
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrLoadPPM4ub(const char* file, unsigned char** OutData, 
+                             unsigned int *w, unsigned int *h);
+
+////////////////////////////////////////////////////////////////////////////
+//! Save PPM image file (with unsigned char as data element type, padded to 
+//! 4 bytes)
+//! @return shrTRUE if saving the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param data  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrSavePPM4ub( const char* file, unsigned char *data, 
+               unsigned int w, unsigned int h);
+
+////////////////////////////////////////////////////////////////////////////////
+//! Save PGM image file (with unsigned char as data element type)
+//! @return shrTRUE if saving the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param data  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrSavePGMub( const char* file, unsigned char *data, 
+              unsigned int w, unsigned int h); 
+
+////////////////////////////////////////////////////////////////////////////
+//! Load PGM image file (with unsigned char as data element type)
+//! @return shrTRUE if saving the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param data  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrLoadPGMub( const char* file, unsigned char** data,
+                  unsigned int *w,unsigned int *h);
+
+////////////////////////////////////////////////////////////////////////////
+// Command line arguments: General notes
+// * All command line arguments begin with '--' followed by the token; 
+//   token and value are seperated by '='; example --samples=50
+// * Arrays have the form --model=[one.obj,two.obj,three.obj] 
+//   (without whitespaces)
+////////////////////////////////////////////////////////////////////////////
+
+////////////////////////////////////////////////////////////////////////////
+//! Check if command line argument \a flag-name is given
+//! @return shrTRUE if command line argument \a flag_name has been given, 
+//!         otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param flag_name  name of command line flag
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCheckCmdLineFlag( const int argc, const char** argv, 
+                     const char* flag_name);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type int
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumenti( const int argc, const char** argv, 
+                        const char* arg_name, int* val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type unsigned int
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentu( const int argc, const char** argv, 
+                        const char* arg_name, unsigned int* val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type float
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentf( const int argc, const char** argv, 
+                        const char* arg_name, float* val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type string
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentstr( const int argc, const char** argv, 
+                          const char* arg_name, char** val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument list those element are strings
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  command line argument list
+//! @param len  length of the list / number of elements
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentListstr( const int argc, const char** argv, 
+                              const char* arg_name, char** val, 
+                              unsigned int* len);
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparef( const float* reference, const float* data,
+             const unsigned int len);
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two integer arrays
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparei( const int* reference, const int* data, 
+             const unsigned int len ); 
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two unsigned integer arrays, with epsilon and threshold
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param threshold  tolerance % # of comparison errors (0.15f = 15%)
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareuit( const unsigned int* reference, const unsigned int* data,
+            const unsigned int len, const float epsilon, const float threshold );
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two unsigned char arrays
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareub( const unsigned char* reference, const unsigned char* data,
+              const unsigned int len ); 
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two integers with a tolernance for # of byte errors
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+//! @param threshold  tolerance % # of comparison errors (0.15f = 15%)
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareubt( const unsigned char* reference, const unsigned char* data,
+             const unsigned int len, const float epsilon, const float threshold );
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two integer arrays witha n epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareube( const unsigned char* reference, const unsigned char* data,
+             const unsigned int len, const float epsilon );
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays with an epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparefe( const float* reference, const float* data,
+              const unsigned int len, const float epsilon );
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays with an epsilon tolerance for equality and a 
+//!     threshold for # pixel errors
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparefet( const float* reference, const float* data,
+             const unsigned int len, const float epsilon, const float threshold );
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays using L2-norm with an epsilon tolerance for 
+//! equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareL2fe( const float* reference, const float* data,
+                const unsigned int len, const float epsilon );
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two PPM image files with an epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param src_file   filename for the image to be compared
+//! @param data       filename for the reference data / gold image
+//! @param epsilon    epsilon to use for the comparison
+//! @param threshold  threshold of pixels that can still mismatch to pass (i.e. 0.15f = 15% must pass)
+//! $param verboseErrors output details of image mismatch to std::err
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparePPM( const char *src_file, const char *ref_file, const float epsilon, const float threshold);
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two PGM image files with an epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param src_file   filename for the image to be compared
+//! @param data       filename for the reference data / gold image
+//! @param epsilon    epsilon to use for the comparison
+//! @param threshold  threshold of pixels that can still mismatch to pass (i.e. 0.15f = 15% must pass)
+//! $param verboseErrors output details of image mismatch to std::err
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparePGM( const char *src_file, const char *ref_file, const float epsilon, const float threshold);
+
+extern "C" unsigned char* shrLoadRawFile(const char* filename, size_t size);
+
+extern "C" size_t shrRoundUp(int group_size, int global_size);
+
+// companion inline function for error checking and exit on error WITH Cleanup Callback (if supplied)
+// *********************************************************************
+inline void __shrCheckErrorEX(int iSample, int iReference, void (*pCleanup)(int), const char* cFile, const int iLine)
+{
+    if (iReference != iSample)
+    {
+        shrLogEx(LOGBOTH | ERRORMSG, iSample, "line %i , in file %s !!!\n\n" , iLine, cFile); 
+        if (pCleanup != NULL)
+        {
+            pCleanup(EXIT_FAILURE);
+        }
+        else 
+        {
+            shrLogEx(LOGBOTH | CLOSELOG, 0, "Exiting...\n");
+            exit(EXIT_FAILURE);
+        }
+    }
+}
+
+// Standardized Exit
+// *********************************************************************
+inline void __shrExitEX(int argc, const char** argv, int iExitCode)
+{
+#ifdef WIN32
+    if (!shrCheckCmdLineFlag(argc, argv, "noprompt") && !shrCheckCmdLineFlag(argc, argv, "qatest")) 
+#else 
+    if (shrCheckCmdLineFlag(argc, argv, "prompt") && !shrCheckCmdLineFlag(argc, argv, "qatest")) 
+#endif
+    {
+        shrLogEx(LOGBOTH | CLOSELOG, 0, "\nPress <Enter> to Quit...\n");                  
+        getchar();                                                           
+    }       
+    else 
+    {
+        shrLogEx(LOGBOTH | CLOSELOG, 0, "%s Exiting...\n", argv[0]); 
+    }
+    fflush(stderr);                                                         
+    exit(iExitCode);
+}
+
+#endif

tests/opencl/DotProduct/DotProduct.cl

@@ -0,0 +1,29 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+ 
+ __kernel void DotProduct (__global float* a, __global float* b, __global float* c, int iNumElements)
+{
+    // find position in global arrays
+    int iGID = get_global_id(0);
+
+    // bound check (equivalent to the limit on a 'for' loop for standard/serial C code
+    if (iGID >= iNumElements)
+    {   
+        return; 
+    }
+
+    // process 
+    int iInOffset = iGID << 2;
+    c[iGID] = a[iInOffset] * b[iInOffset] 
+               + a[iInOffset + 1] * b[iInOffset + 1]
+               + a[iInOffset + 2] * b[iInOffset + 2]
+               + a[iInOffset + 3] * b[iInOffset + 3];
+}

tests/opencl/DotProduct/Makefile

@@ -0,0 +1,65 @@
+RISCV_TOOLCHAIN_PATH ?= $(wildcard ../../../../riscv-gnu-toolchain/drops)
+POCL_CC_PATH ?= $(wildcard ../../../../pocl/drops_riscv_cc)
+POCL_INC_PATH ?= $(wildcard ../include)
+POCL_LIB_PATH ?= $(wildcard ../lib)
+VORTEX_RT_PATH ?= $(wildcard ../../../runtime)
+VX_SIMX_PATH ?= $(wildcard ../../../simX/obj_dir)
+
+CC  = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gcc
+CXX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-g++
+DMP = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objdump
+HEX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objcopy
+GDB = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gdb
+
+VX_SRCS =  $(VORTEX_RT_PATH)/newlib/newlib.c
+VX_SRCS += $(VORTEX_RT_PATH)/startup/vx_start.S
+VX_SRCS += $(VORTEX_RT_PATH)/intrinsics/vx_intrinsics.S
+VX_SRCS += $(VORTEX_RT_PATH)/io/vx_io.S $(VORTEX_RT_PATH)/io/vx_io.c
+VX_SRCS += $(VORTEX_RT_PATH)/fileio/fileio.S
+VX_SRCS += $(VORTEX_RT_PATH)/tests/tests.c
+VX_SRCS += $(VORTEX_RT_PATH)/vx_api/vx_api.c
+
+VX_CFLAGS = -nostartfiles -Wl,-Bstatic,-T,$(VORTEX_RT_PATH)/startup/vx_link.ld
+
+CXXFLAGS =  -g -O0 -march=rv32im -mabi=ilp32 
+CXXFLAGS += -ffreestanding # program may not begin at main()
+CXXFLAGS += -Wl,--gc-sections # enable garbage collection of unused input sections
+CXXFLAGS += -fno-rtti -fno-non-call-exceptions # disable RTTI and exceptions
+CXXFLAGS += -I$(POCL_INC_PATH) -I.
+
+VX_LIBS = -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
+QEMU_LIBS = $(VORTEX_RT_PATH)/qemu/vx_api.c -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
+
+PROJECT=DotProduct
+
+all: $(PROJECT).dump $(PROJECT).hex
+
+lib$(PROJECT).a: DotProduct.cl
+	POCL_DEBUG=all POCL_DEBUG_LLVM_PASSES=1 LD_LIBRARY_PATH=$(RISCV_TOOLCHAIN_PATH)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -o lib$(PROJECT).a kernel.cl
+
+$(PROJECT).elf: main.cc lib$(PROJECT).a
+	$(CXX) $(CXXFLAGS) $(VX_CFLAGS) $(VX_SRCS) main.cc $(VX_LIBS) -o $(PROJECT).elf
+
+$(PROJECT).qemu: main.cc lib$(PROJECT).a
+	$(CXX) $(CXXFLAGS) main.cc $(QEMU_LIBS) -o $(PROJECT).qemu
+
+$(PROJECT).hex: $(PROJECT).elf
+	$(HEX) -O ihex $(PROJECT).elf $(PROJECT).hex
+
+$(PROJECT).dump: $(PROJECT).elf
+	$(DMP) -D $(PROJECT).elf > $(PROJECT).dump
+
+run: $(PROJECT).hex
+	POCL_DEBUG=all $(VX_SIMX_PATH)/Vcache_simX -E -a rv32i --core $(PROJECT).hex -s -b 1> emulator.debug
+
+qemu: $(PROJECT).qemu
+	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -d in_asm -D debug.log $(PROJECT).qemu
+
+gdb-s: $(PROJECT).qemu
+	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -g 1234 -d in_asm -D debug.log $(PROJECT).qemu
+
+gdb-c: $(PROJECT).qemu
+	$(GDB) $(PROJECT).qemu
+
+clean:
+	rm -rf *.elf *.dump *.hex

tests/opencl/DotProduct/main.cc

@@ -0,0 +1,270 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+
+// *********************************************************************
+// oclDotProduct Notes:  
+//
+// A simple OpenCL API demo application that implements a
+// vector dot product computation between 2 float arrays. 
+//
+// Runs computations with OpenCL on the GPU device and then checks results 
+// against basic host CPU/C++ computation.
+//
+// Uses 'shr' and 'ocl' functions from oclUtils and shrUtils libraries for compactness. 
+// But these are NOT required libs for OpenCL developement in general.
+// *********************************************************************
+
+// standard utilities and systems includes
+#include <oclUtils.h>
+#include <shrQATest.h>
+
+// Name of the file with the source code for the computation kernel
+// *********************************************************************
+const char* cSourceFile = "DotProduct.cl";
+
+// Host buffers for demo
+// *********************************************************************
+void *srcA, *srcB, *dst;        // Host buffers for OpenCL test
+void* Golden;                   // Host buffer for host golden processing cross check
+
+// OpenCL Vars
+cl_platform_id cpPlatform;      // OpenCL platform
+cl_device_id   *cdDevices;      // OpenCL device
+cl_context cxGPUContext;        // OpenCL context
+cl_command_queue cqCommandQueue;// OpenCL command que
+cl_program program;           // OpenCL program
+cl_kernel ckKernel;             // OpenCL kernel
+cl_mem cmDevSrcA;               // OpenCL device source buffer A
+cl_mem cmDevSrcB;               // OpenCL device source buffer B 
+cl_mem cmDevDst;                // OpenCL device destination buffer 
+size_t szGlobalWorkSize;        // Total # of work items in the 1D range
+size_t szLocalWorkSize;		    // # of work items in the 1D work group	
+size_t szParmDataBytes;			// Byte size of context information
+size_t szKernelLength;			// Byte size of kernel code
+cl_int ciErrNum;			    // Error code var
+char* cPathAndName = NULL;      // var for full paths to data, src, etc.
+char* cSourceCL = NULL;         // Buffer to hold source for compilation 
+const char* cExecutableName = NULL;
+
+// demo config vars
+int iNumElements= 1277944;	    // Length of float arrays to process (odd # for illustration)
+shrBOOL bNoPrompt = shrFALSE;  
+
+// Forward Declarations
+// *********************************************************************
+void DotProductHost(const float* pfData1, const float* pfData2, float* pfResult, int iNumElements);
+void Cleanup (int iExitCode);
+void (*pCleanup)(int) = &Cleanup;
+
+int *gp_argc = NULL;
+char ***gp_argv = NULL;
+
+// Main function 
+// *********************************************************************
+int main(int argc, char **argv)
+{
+    gp_argc = &argc;
+    gp_argv = &argv;
+
+    shrQAStart(argc, argv);
+
+    // Get the NVIDIA platform
+    ciErrNum = oclGetPlatformID(&cpPlatform);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
+    shrLog("clGetPlatformID...\n"); 
+
+    // Get the NVIDIA platform
+    ciErrNum = oclGetPlatformID(&cpPlatform);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
+    shrLog("clGetPlatformID...\n"); 
+
+    //Get all the devices
+    cl_uint uiNumDevices = 0;           // Number of devices available
+    cl_uint uiTargetDevice = 0;	        // Default Device to compute on
+    cl_uint uiNumComputeUnits;          // Number of compute units (SM's on NV GPU)
+    shrLog("Get the Device info and select Device...\n");
+    ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_DEFAULT, 0, NULL, &uiNumDevices);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
+    cdDevices = (cl_device_id *)malloc(uiNumDevices * sizeof(cl_device_id) );
+    ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_DEFAULT, uiNumDevices, cdDevices, NULL);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
+
+    // Get command line device options and config accordingly
+    shrLog("  # of Devices Available = %u\n", uiNumDevices); 
+    if(shrGetCmdLineArgumentu(argc, (const char**)argv, "device", &uiTargetDevice)== shrTRUE) 
+    {
+        uiTargetDevice = CLAMP(uiTargetDevice, 0, (uiNumDevices - 1));
+    }
+    shrLog("  Using Device %u: ", uiTargetDevice); 
+    oclPrintDevName(LOGBOTH, cdDevices[uiTargetDevice]);
+    ciErrNum = clGetDeviceInfo(cdDevices[uiTargetDevice], CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(uiNumComputeUnits), &uiNumComputeUnits, NULL);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
+    shrLog("\n  # of Compute Units = %u\n", uiNumComputeUnits); 
+
+    // get command line arg for quick test, if provided
+    bNoPrompt = shrCheckCmdLineFlag(argc, (const char**)argv, "noprompt");
+
+    // start logs
+	cExecutableName = argv[0];
+    shrSetLogFileName ("oclDotProduct.txt");
+    shrLog("%s Starting...\n\n# of float elements per Array \t= %u\n", argv[0], iNumElements); 
+
+    // set and log Global and Local work size dimensions
+    szLocalWorkSize = 256;
+    szGlobalWorkSize = shrRoundUp((int)szLocalWorkSize, iNumElements);  // rounded up to the nearest multiple of the LocalWorkSize
+    shrLog("Global Work Size \t\t= %u\nLocal Work Size \t\t= %u\n# of Work Groups \t\t= %u\n\n", 
+           szGlobalWorkSize, szLocalWorkSize, (szGlobalWorkSize % szLocalWorkSize + szGlobalWorkSize/szLocalWorkSize)); 
+
+    // Allocate and initialize host arrays
+    shrLog( "Allocate and Init Host Mem...\n"); 
+    srcA = (void *)malloc(sizeof(cl_float4) * szGlobalWorkSize);
+    srcB = (void *)malloc(sizeof(cl_float4) * szGlobalWorkSize);
+    dst = (void *)malloc(sizeof(cl_float) * szGlobalWorkSize);
+    Golden = (void *)malloc(sizeof(cl_float) * iNumElements);
+    shrFillArray((float*)srcA, 4 * iNumElements);
+    shrFillArray((float*)srcB, 4 * iNumElements);
+
+    // Get the NVIDIA platform
+    ciErrNum = oclGetPlatformID(&cpPlatform);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+    // Get a GPU device
+    ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_DEFAULT, 1, &cdDevices[uiTargetDevice], NULL);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+    // Create the context
+    cxGPUContext = clCreateContext(0, 1, &cdDevices[uiTargetDevice], NULL, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+    // Create a command-queue
+    shrLog("clCreateCommandQueue...\n"); 
+    cqCommandQueue = clCreateCommandQueue(cxGPUContext, cdDevices[uiTargetDevice], 0, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+    // Allocate the OpenCL buffer memory objects for source and result on the device GMEM
+    shrLog("clCreateBuffer (SrcA, SrcB and Dst in Device GMEM)...\n"); 
+    cmDevSrcA = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, sizeof(cl_float) * szGlobalWorkSize * 4, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    cmDevSrcB = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, sizeof(cl_float) * szGlobalWorkSize * 4, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    cmDevDst = clCreateBuffer(cxGPUContext, CL_MEM_WRITE_ONLY, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+    // Read the OpenCL kernel in from source file
+    shrLog("oclLoadProgSource (%s)...\n", cSourceFile); 
+    cPathAndName = shrFindFilePath(cSourceFile, argv[0]);
+    //oclCheckErrorEX(cPathAndName != NULL, shrTRUE, pCleanup);
+    cSourceCL = oclLoadProgSource(cPathAndName, "", &szKernelLength);
+    //oclCheckErrorEX(cSourceCL != NULL, shrTRUE, pCleanup);
+
+    // Create the program
+    shrLog("clCreateProgramWithSource...\n"); 
+    //program = clCreateProgramWithSource(cxGPUContext, 1, (const char **)&cSourceCL, &szKernelLength, &ciErrNum);
+    cl_program program =
+      clCreateProgramWithBuiltInKernels(context, 1, &device_id, "sgemm", NULL);
+        // Build the program with 'mad' Optimization option
+    #ifdef MAC
+        char* flags = "-cl-fast-relaxed-math -DMAC";
+    #else
+        char* flags = "-cl-fast-relaxed-math";
+    #endif
+    shrLog("clBuildProgram...\n"); 
+    ciErrNum = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
+    if (ciErrNum != CL_SUCCESS)
+    {
+        // write out standard error, Build Log and PTX, then cleanup and exit
+        shrLogEx(LOGBOTH | ERRORMSG, ciErrNum, STDERROR);
+        oclLogBuildInfo(program, oclGetFirstDev(cxGPUContext));
+        oclLogPtx(program, oclGetFirstDev(cxGPUContext), "oclDotProduct.ptx");
+        Cleanup(EXIT_FAILURE); 
+    }
+
+    // Create the kernel
+    shrLog("clCreateKernel (DotProduct)...\n"); 
+    ckKernel = clCreateKernel(program, "DotProduct", &ciErrNum);
+
+    // Set the Argument values
+    shrLog("clSetKernelArg 0 - 3...\n\n"); 
+    ciErrNum = clSetKernelArg(ckKernel, 0, sizeof(cl_mem), (void*)&cmDevSrcA);
+    ciErrNum |= clSetKernelArg(ckKernel, 1, sizeof(cl_mem), (void*)&cmDevSrcB);
+    ciErrNum |= clSetKernelArg(ckKernel, 2, sizeof(cl_mem), (void*)&cmDevDst);
+    ciErrNum |= clSetKernelArg(ckKernel, 3, sizeof(cl_int), (void*)&iNumElements);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+    // --------------------------------------------------------
+    // Core sequence... copy input data to GPU, compute, copy results back
+
+    // Asynchronous write of data to GPU device
+    shrLog("clEnqueueWriteBuffer (SrcA and SrcB)...\n"); 
+    ciErrNum = clEnqueueWriteBuffer(cqCommandQueue, cmDevSrcA, CL_FALSE, 0, sizeof(cl_float) * szGlobalWorkSize * 4, srcA, 0, NULL, NULL);
+    ciErrNum |= clEnqueueWriteBuffer(cqCommandQueue, cmDevSrcB, CL_FALSE, 0, sizeof(cl_float) * szGlobalWorkSize * 4, srcB, 0, NULL, NULL);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+    // Launch kernel
+    shrLog("clEnqueueNDRangeKernel (DotProduct)...\n"); 
+    ciErrNum = clEnqueueNDRangeKernel(cqCommandQueue, ckKernel, 1, NULL, &szGlobalWorkSize, &szLocalWorkSize, 0, NULL, NULL);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+    // Read back results and check accumulated errors
+    shrLog("clEnqueueReadBuffer (Dst)...\n\n"); 
+    ciErrNum = clEnqueueReadBuffer(cqCommandQueue, cmDevDst, CL_TRUE, 0, sizeof(cl_float) * szGlobalWorkSize, dst, 0, NULL, NULL);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+    // Compute and compare results for golden-host and report errors and pass/fail
+    shrLog("Comparing against Host/C++ computation...\n\n"); 
+    DotProductHost ((const float*)srcA, (const float*)srcB, (float*)Golden, iNumElements);
+    shrBOOL bMatch = shrComparefet((const float*)Golden, (const float*)dst, (unsigned int)iNumElements, 0.0f, 0);
+
+    // Cleanup and leave
+    Cleanup (EXIT_SUCCESS);
+}
+
+// "Golden" Host processing dot product function for comparison purposes
+// *********************************************************************
+void DotProductHost(const float* pfData1, const float* pfData2, float* pfResult, int iNumElements)
+{
+    int i, j, k;
+    for (i = 0, j = 0; i < iNumElements; i++) 
+    {
+        pfResult[i] = 0.0f;
+        for (k = 0; k < 4; k++, j++) 
+        {
+            pfResult[i] += pfData1[j] * pfData2[j]; 
+        } 
+    }
+}
+
+// Cleanup and exit code
+// *********************************************************************
+void Cleanup(int iExitCode)
+{
+    // Cleanup allocated objects
+    shrLog("Starting Cleanup...\n\n");
+    if(cPathAndName)free(cPathAndName);
+    if(cSourceCL)free(cSourceCL);
+	if(ckKernel)clReleaseKernel(ckKernel);  
+    if(program)clReleaseProgram(program);
+    if(cqCommandQueue)clReleaseCommandQueue(cqCommandQueue);
+    if(cxGPUContext)clReleaseContext(cxGPUContext);
+    if (cmDevSrcA)clReleaseMemObject(cmDevSrcA);
+    if (cmDevSrcB)clReleaseMemObject(cmDevSrcB);
+    if (cmDevDst)clReleaseMemObject(cmDevDst);
+
+    // Free host memory
+    free(srcA); 
+    free(srcB);
+    free (dst);
+    free(Golden);
+
+    if (cdDevices) free(cdDevices);
+
+    shrQAFinishExit(*gp_argc, (const char **)*gp_argv, (iExitCode == EXIT_SUCCESS) ? QA_PASSED : QA_FAILED);
+}

tests/opencl/DotProduct/oclUtils.h

@@ -0,0 +1,198 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+ 
+#ifndef OCL_UTILS_H
+#define OCL_UTILS_H
+
+// *********************************************************************
+// Utilities specific to OpenCL samples in NVIDIA GPU Computing SDK 
+// *********************************************************************
+
+// Common headers:  Cross-API utililties and OpenCL header
+#include <shrUtils.h>
+
+// All OpenCL headers
+#if defined (__APPLE__) || defined(MACOSX)
+    #include <OpenCL/opencl.h>
+#else
+    #include <CL/opencl.h>
+#endif 
+
+// Includes
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+
+// For systems with CL_EXT that are not updated with these extensions, we copied these
+// extensions from <CL/cl_ext.h>
+#ifndef CL_DEVICE_COMPUTE_CAPABILITY_MAJOR_NV
+  /* cl_nv_device_attribute_query extension - no extension #define since it has no functions */
+  #define CL_DEVICE_COMPUTE_CAPABILITY_MAJOR_NV       0x4000
+  #define CL_DEVICE_COMPUTE_CAPABILITY_MINOR_NV       0x4001
+  #define CL_DEVICE_REGISTERS_PER_BLOCK_NV            0x4002
+  #define CL_DEVICE_WARP_SIZE_NV                      0x4003
+  #define CL_DEVICE_GPU_OVERLAP_NV                    0x4004
+  #define CL_DEVICE_KERNEL_EXEC_TIMEOUT_NV            0x4005
+  #define CL_DEVICE_INTEGRATED_MEMORY_NV              0x4006
+#endif
+
+// reminders for build output window and log
+#ifdef _WIN32
+    #pragma message ("Note: including shrUtils.h")
+    #pragma message ("Note: including opencl.h")
+#endif
+
+// SDK Revision #
+#define OCL_SDKREVISION "7027912"
+
+// Error and Exit Handling Macros... 
+// *********************************************************************
+// Full error handling macro with Cleanup() callback (if supplied)... 
+// (Companion Inline Function lower on page)
+#define oclCheckErrorEX(a, b, c) __oclCheckErrorEX(a, b, c, __FILE__ , __LINE__) 
+
+// Short version without Cleanup() callback pointer
+// Both Input (a) and Reference (b) are specified as args
+#define oclCheckError(a, b) oclCheckErrorEX(a, b, 0) 
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the platform ID for NVIDIA if available, otherwise default to platform 0
+//!
+//! @return the id 
+//! @param clSelectedPlatformID         OpenCL platform ID
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_int oclGetPlatformID(cl_platform_id* clSelectedPlatformID);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Print info about the device
+//!
+//! @param iLogMode       enum LOGBOTH, LOGCONSOLE, LOGFILE
+//! @param device         OpenCL id of the device
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclPrintDevInfo(int iLogMode, cl_device_id device);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get and return device capability
+//!
+//! @return the 2 digit integer representation of device Cap (major minor). return -1 if NA 
+//! @param device         OpenCL id of the device
+//////////////////////////////////////////////////////////////////////////////
+extern "C" int oclGetDevCap(cl_device_id device);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Print the device name
+//!
+//! @param iLogMode       enum LOGBOTH, LOGCONSOLE, LOGFILE
+//! @param device         OpenCL id of the device
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclPrintDevName(int iLogMode, cl_device_id device);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the id of the first device from the context
+//!
+//! @return the id 
+//! @param cxGPUContext         OpenCL context
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_device_id oclGetFirstDev(cl_context cxGPUContext);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the id of the nth device from the context
+//!
+//! @return the id or -1 when out of range
+//! @param cxGPUContext         OpenCL context
+//! @param device_idx            index of the device of interest
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_device_id oclGetDev(cl_context cxGPUContext, unsigned int device_idx);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the id of device with maximal FLOPS from the context
+//!
+//! @return the id 
+//! @param cxGPUContext         OpenCL context
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_device_id oclGetMaxFlopsDev(cl_context cxGPUContext);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Loads a Program file and prepends the cPreamble to the code.
+//!
+//! @return the source string if succeeded, 0 otherwise
+//! @param cFilename        program filename
+//! @param cPreamble        code that is prepended to the loaded file, typically a set of #defines or a header
+//! @param szFinalLength    returned length of the code string
+//////////////////////////////////////////////////////////////////////////////
+extern "C" char* oclLoadProgSource(const char* cFilename, const char* cPreamble, size_t* szFinalLength);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get the binary (PTX) of the program associated with the device
+//!
+//! @param cpProgram    OpenCL program
+//! @param cdDevice     device of interest
+//! @param binary       returned code
+//! @param length       length of returned code
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclGetProgBinary( cl_program cpProgram, cl_device_id cdDevice, char** binary, size_t* length);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get and log the binary (PTX) from the OpenCL compiler for the requested program & device
+//!
+//! @param cpProgram                   OpenCL program
+//! @param cdDevice                    device of interest
+//! @param const char*  cPtxFileName   optional PTX file name
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclLogPtx(cl_program cpProgram, cl_device_id cdDevice, const char* cPtxFileName);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get and log the Build Log from the OpenCL compiler for the requested program & device
+//!
+//! @param cpProgram    OpenCL program
+//! @param cdDevice     device of interest
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclLogBuildInfo(cl_program cpProgram, cl_device_id cdDevice);
+
+// Helper function for De-allocating cl objects
+// *********************************************************************
+extern "C" void oclDeleteMemObjs(cl_mem* cmMemObjs, int iNumObjs);
+
+// Helper function to get OpenCL error string from constant
+// *********************************************************************
+extern "C" const char* oclErrorString(cl_int error);
+
+// Helper function to get OpenCL image format string (channel order and type) from constant
+// *********************************************************************
+extern "C" const char* oclImageFormatString(cl_uint uiImageFormat);
+
+// companion inline function for error checking and exit on error WITH Cleanup Callback (if supplied)
+// *********************************************************************
+inline void __oclCheckErrorEX(cl_int iSample, cl_int iReference, void (*pCleanup)(int), const char* cFile, const int iLine)
+{
+    // An error condition is defined by the sample/test value not equal to the reference
+    if (iReference != iSample)
+    {
+        // If the sample/test value isn't equal to the ref, it's an error by defnition, so override 0 sample/test value
+        iSample = (iSample == 0) ? -9999 : iSample; 
+
+        // Log the error info
+        shrLog("\n !!! Error # %i (%s) at line %i , in file %s !!!\n\n", iSample, oclErrorString(iSample), iLine, cFile);
+
+        // Cleanup and exit, or just exit if no cleanup function pointer provided.  Use iSample (error code in this case) as process exit code.
+        if (pCleanup != NULL)
+        {
+            pCleanup(iSample);
+        }
+        else 
+        {
+            shrLogEx(LOGBOTH | CLOSELOG, 0, "Exiting...\n");
+            exit(iSample);
+        }
+    }
+}
+
+#endif

tests/opencl/DotProduct/shrQATest.h

@@ -0,0 +1,238 @@
+/*
+* Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+*
+* Please refer to the NVIDIA end user license agreement (EULA) associated
+* with this source code for terms and conditions that govern your use of
+* this software. Any use, reproduction, disclosure, or distribution of
+* this software and related documentation outside the terms of the EULA
+* is strictly prohibited.
+*
+*/
+
+#ifndef SHR_QATEST_H
+#define SHR_QATEST_H
+
+// *********************************************************************
+// Generic utilities for NVIDIA GPU Computing SDK 
+// *********************************************************************
+
+// OS dependent includes
+#ifdef _WIN32
+    #pragma message ("Note: including windows.h")
+    #pragma message ("Note: including math.h")
+    #pragma message ("Note: including assert.h")
+    #pragma message ("Note: including time.h")
+
+// Headers needed for Windows
+    #include <windows.h>
+	#include <time.h>
+#else
+    // Headers needed for Linux
+    #include <sys/stat.h>
+    #include <sys/types.h>
+    #include <sys/time.h>
+    #include <stdio.h>
+    #include <stdlib.h>
+    #include <string.h>
+    #include <stdarg.h>
+    #include <unistd.h>
+    #include <time.h>
+#endif
+
+#ifndef STRCASECMP
+#ifdef _WIN32
+#define STRCASECMP _stricmp
+#else
+#define STRCASECMP strcasecmp
+#endif
+#endif
+
+#ifndef STRNCASECMP
+#ifdef _WIN32
+#define STRNCASECMP _strnicmp
+#else
+#define STRNCASECMP strncasecmp
+#endif
+#endif
+
+
+// Standardized QA Start/Finish for CUDA SDK tests
+#define shrQAStart(a, b)      __shrQAStart(a, b)
+#define shrQAFinish(a, b, c)  __shrQAFinish(a, b, c)
+#define shrQAFinish2(a, b, c, d) __shrQAFinish2(a, b, c, d)
+
+inline int findExeNameStart(const char *exec_name)
+{
+    int exename_start = (int)strlen(exec_name);
+
+    while( (exename_start > 0) && 
+            (exec_name[exename_start] != '\\') && 
+            (exec_name[exename_start] != '/') )
+    {
+        exename_start--;
+    }
+    if (exec_name[exename_start] == '\\' || 
+        exec_name[exename_start] == '/')
+    {
+        return exename_start+1;
+    } else {
+        return exename_start;
+    }
+}
+
+inline int __shrQAStart(int argc, char **argv)
+{
+    bool bQATest = false;
+    // First clear the output buffer
+    fflush(stdout);
+    fflush(stdout);
+
+    for (int i=1; i < argc; i++) {
+        int string_start = 0;
+        while (argv[i][string_start] == '-')
+           string_start++;
+        char *string_argv = &argv[i][string_start];
+
+        if (!STRCASECMP(string_argv, "qatest")) {
+           bQATest = true;
+        }
+    }
+    
+    // We don't want to print the entire path, so we search for the first 
+    int exename_start = findExeNameStart(argv[0]);
+    if (bQATest) {
+        fprintf(stdout, "&&&& RUNNING %s", &(argv[0][exename_start]));
+        for (int i=1; i < argc; i++) fprintf(stdout, " %s", argv[i]);
+        fprintf(stdout, "\n");
+    } else {
+        fprintf(stdout, "[%s] starting...\n", &(argv[0][exename_start]));
+    }
+    fflush(stdout);
+    printf("\n"); fflush(stdout);
+    return exename_start;
+}
+
+enum eQAstatus {
+    QA_FAILED = 0,
+    QA_PASSED = 1,
+    QA_WAIVED = 2
+};
+
+inline void __ExitInTime(int seconds)
+{
+    fprintf(stdout, "> exiting in %d seconds: ", seconds);
+    fflush(stdout);
+    time_t t;
+    int count;
+    for (t=time(0)+seconds, count=seconds; time(0) < t; count--) {
+        fprintf(stdout, "%d...", count);
+#ifdef WIN32
+        Sleep(1000);
+#else
+        sleep(1);
+#endif
+    }
+    fprintf(stdout,"done!\n\n"); 
+	fflush(stdout);
+}
+
+
+inline void __shrQAFinish(int argc, const char **argv, int iStatus)
+{
+    // By default QATest is disabled and NoPrompt is Enabled (times out at seconds passed into __ExitInTime() )
+    bool bQATest = false, bNoPrompt = true, bQuitInTime = true;
+    const char *sStatus[] = { "FAILED", "PASSED", "WAIVED", NULL };
+	
+    for (int i=1; i < argc; i++) {
+        int string_start = 0;
+        while (argv[i][string_start] == '-')
+           string_start++;
+
+        const char *string_argv = &argv[i][string_start];
+        if (!STRCASECMP(string_argv, "qatest")) {
+           bQATest = true;
+        }	
+        // For SDK individual samples that don't specify -noprompt or -prompt, 
+        // a 3 second delay will happen before exiting, giving a user time to view results
+        if (!STRCASECMP(string_argv, "noprompt") || !STRCASECMP(string_argv, "help")) {
+            bNoPrompt = true;
+            bQuitInTime = false;
+        }
+        if (!STRCASECMP(string_argv, "prompt")) {
+            bNoPrompt = false;
+            bQuitInTime = false;
+        }
+    }
+
+    int exename_start = findExeNameStart(argv[0]);
+    if (bQATest) {
+        fprintf(stdout, "&&&& %s %s", sStatus[iStatus], &(argv[0][exename_start]));
+        for (int i=1; i < argc; i++) fprintf(stdout, " %s", argv[i]);
+        fprintf(stdout, "\n");
+    } else {
+        fprintf(stdout, "[%s] test results...\n%s\n", &(argv[0][exename_start]), sStatus[iStatus]);
+    }
+    fflush(stdout);
+    printf("\n"); fflush(stdout);
+    if (bQuitInTime) {
+        __ExitInTime(3);
+    } else {
+        if (!bNoPrompt) {
+            fprintf(stdout, "\nPress <Enter> to exit...\n");
+            fflush(stdout);
+            getchar();
+        }
+    }
+}
+
+inline void __shrQAFinish2(bool bQATest, int argc, const char **argv, int iStatus)
+{
+    bool bQuitInTime = true;
+    const char *sStatus[] = { "FAILED", "PASSED", "WAIVED", NULL };
+	
+    for (int i=1; i < argc; i++) {
+        int string_start = 0;
+        while (argv[i][string_start] == '-')
+           string_start++;
+
+        const char *string_argv = &argv[i][string_start];
+        // For SDK individual samples that don't specify -noprompt or -prompt, 
+        // a 3 second delay will happen before exiting, giving a user time to view results
+        if (!STRCASECMP(string_argv, "noprompt") || !STRCASECMP(string_argv, "help")) {
+            bQuitInTime = false;
+        }
+        if (!STRCASECMP(string_argv, "prompt")) {
+            bQuitInTime = false;
+        }
+    }
+
+    int exename_start = findExeNameStart(argv[0]);
+    if (bQATest) {
+        fprintf(stdout, "&&&& %s %s", sStatus[iStatus], &(argv[0][exename_start]));
+        for (int i=1; i < argc; i++) fprintf(stdout, " %s", argv[i]);
+        fprintf(stdout, "\n");
+    } else {
+        fprintf(stdout, "[%s] test results...\n%s\n", &(argv[0][exename_start]), sStatus[iStatus]);
+    }
+    fflush(stdout);
+    
+    if (bQuitInTime) {
+        __ExitInTime(3);
+    }
+}
+
+inline void shrQAFinishExit(int argc, const char **argv, int iStatus)
+{
+    __shrQAFinish(argc, argv, iStatus);
+
+    exit(iStatus ? EXIT_SUCCESS : EXIT_FAILURE); 
+}
+
+inline void shrQAFinishExit2(bool bQAtest, int argc, const char **argv, int iStatus)
+{
+    __shrQAFinish2(bQAtest, argc, argv, iStatus);
+
+    exit(iStatus ? EXIT_SUCCESS : EXIT_FAILURE);
+}
+
+#endif

tests/opencl/DotProduct/shrUtils.h

@@ -0,0 +1,642 @@
+/*
+* Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+*
+* Please refer to the NVIDIA end user license agreement (EULA) associated
+* with this source code for terms and conditions that govern your use of
+* this software. Any use, reproduction, disclosure, or distribution of
+* this software and related documentation outside the terms of the EULA
+* is strictly prohibited.
+*
+*/
+
+#ifndef SHR_UTILS_H
+#define SHR_UTILS_H
+
+// *********************************************************************
+// Generic utilities for NVIDIA GPU Computing SDK 
+// *********************************************************************
+
+// reminders for output window and build log
+#ifdef _WIN32
+    #pragma message ("Note: including windows.h")
+    #pragma message ("Note: including math.h")
+    #pragma message ("Note: including assert.h")
+#endif
+
+// OS dependent includes
+#ifdef _WIN32
+    // Headers needed for Windows
+    #include <windows.h>
+#else
+    // Headers needed for Linux
+    #include <sys/stat.h>
+    #include <sys/types.h>
+    #include <sys/time.h>
+    #include <stdio.h>
+    #include <stdlib.h>
+    #include <string.h>
+    #include <stdarg.h>
+#endif
+
+// Other headers needed for both Windows and Linux
+#include <math.h>
+#include <assert.h>
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+
+// Un-comment the following #define to enable profiling code in SDK apps
+//#define GPU_PROFILING
+
+// Beginning of GPU Architecture definitions
+inline int ConvertSMVer2Cores(int major, int minor)
+{
+	// Defines for GPU Architecture types (using the SM version to determine the # of cores per SM
+	typedef struct {
+		int SM; // 0xMm (hexidecimal notation), M = SM Major version, and m = SM minor version
+		int Cores;
+	} sSMtoCores;
+
+	sSMtoCores nGpuArchCoresPerSM[] = 
+	{ { 0x10,  8 }, // Tesla Generation (SM 1.0) G80 class
+	  { 0x11,  8 }, // Tesla Generation (SM 1.1) G8x class
+	  { 0x12,  8 }, // Tesla Generation (SM 1.2) G9x class
+	  { 0x13,  8 }, // Tesla Generation (SM 1.3) GT200 class
+	  { 0x20, 32 }, // Fermi Generation (SM 2.0) GF100 class
+	  { 0x21, 48 }, // Fermi Generation (SM 2.1) GF10x class
+	  { 0x30, 192}, // Fermi Generation (SM 3.0) GK10x class
+	  {   -1, -1 }
+	};
+
+	int index = 0;
+	while (nGpuArchCoresPerSM[index].SM != -1) {
+		if (nGpuArchCoresPerSM[index].SM == ((major << 4) + minor) ) {
+			return nGpuArchCoresPerSM[index].Cores;
+		}
+		index++;
+	}
+	printf("MapSMtoCores SM %d.%d is undefined (please update to the latest SDK)!\n", major, minor);
+	return -1;
+}
+// end of GPU Architecture definitions
+
+
+// Defines and enum for use with logging functions
+// *********************************************************************
+#define DEFAULTLOGFILE "SdkConsoleLog.txt"
+#define MASTERLOGFILE "SdkMasterLog.csv"
+enum LOGMODES 
+{
+    LOGCONSOLE = 1, // bit to signal "log to console" 
+    LOGFILE    = 2, // bit to signal "log to file" 
+    LOGBOTH    = 3, // convenience union of first 2 bits to signal "log to both"
+    APPENDMODE = 4, // bit to set "file append" mode instead of "replace mode" on open
+    MASTER     = 8, // bit to signal master .csv log output
+    ERRORMSG   = 16, // bit to signal "pre-pend Error" 
+    CLOSELOG   = 32  // bit to close log file, if open, after any requested file write
+};
+#define HDASHLINE "-----------------------------------------------------------\n"
+
+// Standardized boolean
+enum shrBOOL
+{
+    shrFALSE = 0,
+    shrTRUE = 1
+};
+
+// Standardized MAX, MIN and CLAMP
+#define MAX(a, b) ((a > b) ? a : b)
+#define MIN(a, b) ((a < b) ? a : b)
+#define CLAMP(a, b, c) MIN(MAX(a, b), c)    // double sided clip of input a
+#define TOPCLAMP(a, b) (a < b ? a:b)	    // single top side clip of input a
+
+// Error and Exit Handling Macros... 
+// *********************************************************************
+// Full error handling macro with Cleanup() callback (if supplied)... 
+// (Companion Inline Function lower on page)
+#define shrCheckErrorEX(a, b, c) __shrCheckErrorEX(a, b, c, __FILE__ , __LINE__) 
+
+// Short version without Cleanup() callback pointer
+// Both Input (a) and Reference (b) are specified as args
+#define shrCheckError(a, b) shrCheckErrorEX(a, b, 0) 
+
+// Standardized Exit Macro for leaving main()... extended version
+// (Companion Inline Function lower on page)
+#define shrExitEX(a, b, c) __shrExitEX(a, b, c)
+
+// Standardized Exit Macro for leaving main()... short version
+// (Companion Inline Function lower on page)
+#define shrEXIT(a, b)        __shrExitEX(a, b, EXIT_SUCCESS)
+
+// Simple argument checker macro
+#define ARGCHECK(a) if((a) != shrTRUE)return shrFALSE 
+
+// Define for user-customized error handling
+#define STDERROR "file %s, line %i\n\n" , __FILE__ , __LINE__
+
+// Function to deallocate memory allocated within shrUtils
+// *********************************************************************
+extern "C" void shrFree(void* ptr);
+
+// *********************************************************************
+// Helper function to log standardized information to Console, to File or to both
+//! Examples: shrLogEx(LOGBOTH, 0, "Function A\n"); 
+//!         : shrLogEx(LOGBOTH | ERRORMSG, ciErrNum, STDERROR);
+//! 
+//! Automatically opens file and stores handle if needed and not done yet
+//! Closes file and nulls handle on request
+//! 
+//! @param 0 iLogMode: LOGCONSOLE, LOGFILE, LOGBOTH, APPENDMODE, MASTER, ERRORMSG, CLOSELOG.  
+//!          LOGFILE and LOGBOTH may be | 'd  with APPENDMODE to select file append mode instead of overwrite mode 
+//!          LOGFILE and LOGBOTH may be | 'd  with CLOSELOG to "write and close" 
+//!          First 3 options may be | 'd  with MASTER to enable independent write to master data log file
+//!          First 3 options may be | 'd  with ERRORMSG to start line with standard error message
+//! @param 2 dValue:    
+//!          Positive val = double value for time in secs to be formatted to 6 decimals. 
+//!          Negative val is an error code and this give error preformatting.
+//! @param 3 cFormatString: String with formatting specifiers like printf or fprintf.  
+//!          ALL printf flags, width, precision and type specifiers are supported with this exception: 
+//!              Wide char type specifiers intended for wprintf (%S and %C) are NOT supported
+//!              Single byte char type specifiers (%s and %c) ARE supported 
+//! @param 4... variable args: like printf or fprintf.  Must match format specifer type above.  
+//! @return 0 if OK, negative value on error or if error occurs or was passed in. 
+// *********************************************************************
+extern "C" int shrLogEx(int iLogMode, int iErrNum, const char* cFormatString, ...);
+
+// Short version of shrLogEx defaulting to shrLogEx(LOGBOTH, 0, 
+// *********************************************************************
+extern "C" int shrLog(const char* cFormatString, ...);
+
+// *********************************************************************
+// Delta timer function for up to 3 independent timers using host high performance counters 
+// Maintains state for 3 independent counters
+//! Example: double dElapsedTime = shrDeltaTime(0);
+//! 
+//! @param 0 iCounterID: Which timer to check/reset. (0, 1, 2)
+//! @return delta time of specified counter since last call in seconds.  Otherwise -9999.0 if error
+// *********************************************************************
+extern "C" double shrDeltaT(int iCounterID);
+
+// Optional LogFileNameOverride function
+// *********************************************************************
+extern "C" void shrSetLogFileName (const char* cOverRideName);
+
+// Helper function to init data arrays 
+// *********************************************************************
+extern "C" void shrFillArray(float* pfData, int iSize);
+
+// Helper function to print data arrays 
+// *********************************************************************
+extern "C" void shrPrintArray(float* pfData, int iSize);
+
+////////////////////////////////////////////////////////////////////////////
+//! Find the path for a filename
+//! @return the path if succeeded, otherwise 0
+//! @param filename        name of the file
+//! @param executablePath  optional absolute path of the executable
+////////////////////////////////////////////////////////////////////////////
+extern "C" char* shrFindFilePath(const char* filename, const char* executablePath);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing single precision floating point data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFilef( const char* filename, float** data, unsigned int* len, 
+              bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing double precision floating point data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFiled( const char* filename, double** data, unsigned int* len, 
+              bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing integer data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFilei( const char* filename, int** data, unsigned int* len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing unsigned integer data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is 
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFileui( const char* filename, unsigned int** data, 
+               unsigned int* len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing char / byte data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is 
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFileb( const char* filename, char** data, unsigned int* len, 
+              bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing unsigned char / byte data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFileub( const char* filename, unsigned char** data, 
+               unsigned int* len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing single precision floating point 
+//! data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+//! @param epsilon  epsilon for comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFilef( const char* filename, const float* data, unsigned int len,
+               const float epsilon, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing double precision floating point 
+//! data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+//! @param epsilon  epsilon for comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFiled( const char* filename, const float* data, unsigned int len,
+               const double epsilon, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing integer data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFilei( const char* filename, const int* data, unsigned int len,
+               bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing unsigned integer data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFileui( const char* filename, const unsigned int* data, 
+                unsigned int len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing char / byte data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFileb( const char* filename, const char* data, unsigned int len, 
+               bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing unsigned char / byte data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFileub( const char* filename, const unsigned char* data,
+                unsigned int len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Load PPM image file (with unsigned char as data element type), padding 
+//! 4th component
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param OutData  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+//! 
+//! Note: If *OutData is NULL this function allocates buffer that must be freed by caller
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrLoadPPM4ub(const char* file, unsigned char** OutData, 
+                             unsigned int *w, unsigned int *h);
+
+////////////////////////////////////////////////////////////////////////////
+//! Save PPM image file (with unsigned char as data element type, padded to 
+//! 4 bytes)
+//! @return shrTRUE if saving the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param data  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrSavePPM4ub( const char* file, unsigned char *data, 
+               unsigned int w, unsigned int h);
+
+////////////////////////////////////////////////////////////////////////////////
+//! Save PGM image file (with unsigned char as data element type)
+//! @return shrTRUE if saving the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param data  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrSavePGMub( const char* file, unsigned char *data, 
+              unsigned int w, unsigned int h); 
+
+////////////////////////////////////////////////////////////////////////////
+//! Load PGM image file (with unsigned char as data element type)
+//! @return shrTRUE if saving the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param data  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrLoadPGMub( const char* file, unsigned char** data,
+                  unsigned int *w,unsigned int *h);
+
+////////////////////////////////////////////////////////////////////////////
+// Command line arguments: General notes
+// * All command line arguments begin with '--' followed by the token; 
+//   token and value are seperated by '='; example --samples=50
+// * Arrays have the form --model=[one.obj,two.obj,three.obj] 
+//   (without whitespaces)
+////////////////////////////////////////////////////////////////////////////
+
+////////////////////////////////////////////////////////////////////////////
+//! Check if command line argument \a flag-name is given
+//! @return shrTRUE if command line argument \a flag_name has been given, 
+//!         otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param flag_name  name of command line flag
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCheckCmdLineFlag( const int argc, const char** argv, 
+                     const char* flag_name);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type int
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumenti( const int argc, const char** argv, 
+                        const char* arg_name, int* val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type unsigned int
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentu( const int argc, const char** argv, 
+                        const char* arg_name, unsigned int* val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type float
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentf( const int argc, const char** argv, 
+                        const char* arg_name, float* val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type string
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentstr( const int argc, const char** argv, 
+                          const char* arg_name, char** val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument list those element are strings
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  command line argument list
+//! @param len  length of the list / number of elements
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentListstr( const int argc, const char** argv, 
+                              const char* arg_name, char** val, 
+                              unsigned int* len);
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparef( const float* reference, const float* data,
+             const unsigned int len);
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two integer arrays
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparei( const int* reference, const int* data, 
+             const unsigned int len ); 
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two unsigned integer arrays, with epsilon and threshold
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param threshold  tolerance % # of comparison errors (0.15f = 15%)
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareuit( const unsigned int* reference, const unsigned int* data,
+            const unsigned int len, const float epsilon, const float threshold );
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two unsigned char arrays
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareub( const unsigned char* reference, const unsigned char* data,
+              const unsigned int len ); 
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two integers with a tolernance for # of byte errors
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+//! @param threshold  tolerance % # of comparison errors (0.15f = 15%)
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareubt( const unsigned char* reference, const unsigned char* data,
+             const unsigned int len, const float epsilon, const float threshold );
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two integer arrays witha n epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareube( const unsigned char* reference, const unsigned char* data,
+             const unsigned int len, const float epsilon );
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays with an epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparefe( const float* reference, const float* data,
+              const unsigned int len, const float epsilon );
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays with an epsilon tolerance for equality and a 
+//!     threshold for # pixel errors
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparefet( const float* reference, const float* data,
+             const unsigned int len, const float epsilon, const float threshold );
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays using L2-norm with an epsilon tolerance for 
+//! equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareL2fe( const float* reference, const float* data,
+                const unsigned int len, const float epsilon );
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two PPM image files with an epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param src_file   filename for the image to be compared
+//! @param data       filename for the reference data / gold image
+//! @param epsilon    epsilon to use for the comparison
+//! @param threshold  threshold of pixels that can still mismatch to pass (i.e. 0.15f = 15% must pass)
+//! $param verboseErrors output details of image mismatch to std::err
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparePPM( const char *src_file, const char *ref_file, const float epsilon, const float threshold);
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two PGM image files with an epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param src_file   filename for the image to be compared
+//! @param data       filename for the reference data / gold image
+//! @param epsilon    epsilon to use for the comparison
+//! @param threshold  threshold of pixels that can still mismatch to pass (i.e. 0.15f = 15% must pass)
+//! $param verboseErrors output details of image mismatch to std::err
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparePGM( const char *src_file, const char *ref_file, const float epsilon, const float threshold);
+
+extern "C" unsigned char* shrLoadRawFile(const char* filename, size_t size);
+
+extern "C" size_t shrRoundUp(int group_size, int global_size);
+
+// companion inline function for error checking and exit on error WITH Cleanup Callback (if supplied)
+// *********************************************************************
+inline void __shrCheckErrorEX(int iSample, int iReference, void (*pCleanup)(int), const char* cFile, const int iLine)
+{
+    if (iReference != iSample)
+    {
+        shrLogEx(LOGBOTH | ERRORMSG, iSample, "line %i , in file %s !!!\n\n" , iLine, cFile); 
+        if (pCleanup != NULL)
+        {
+            pCleanup(EXIT_FAILURE);
+        }
+        else 
+        {
+            shrLogEx(LOGBOTH | CLOSELOG, 0, "Exiting...\n");
+            exit(EXIT_FAILURE);
+        }
+    }
+}
+
+// Standardized Exit
+// *********************************************************************
+inline void __shrExitEX(int argc, const char** argv, int iExitCode)
+{
+#ifdef WIN32
+    if (!shrCheckCmdLineFlag(argc, argv, "noprompt") && !shrCheckCmdLineFlag(argc, argv, "qatest")) 
+#else 
+    if (shrCheckCmdLineFlag(argc, argv, "prompt") && !shrCheckCmdLineFlag(argc, argv, "qatest")) 
+#endif
+    {
+        shrLogEx(LOGBOTH | CLOSELOG, 0, "\nPress <Enter> to Quit...\n");                  
+        getchar();                                                           
+    }       
+    else 
+    {
+        shrLogEx(LOGBOTH | CLOSELOG, 0, "%s Exiting...\n", argv[0]); 
+    }
+    fflush(stderr);                                                         
+    exit(iExitCode);
+}
+
+#endif

tests/opencl/Makefile

@@ -0,0 +1,31 @@
+all:
+	$(MAKE) -C vecadd
+	$(MAKE) -C sgemm
+	$(MAKE) -C saxpy
+	$(MAKE) -C sfilter
+	$(MAKE) -C nearn
+	$(MAKE) -C guassian
+	
+run:
+	$(MAKE) -C vecadd run-vlsim 
+	$(MAKE) -C sgemm run-vlsim
+	$(MAKE) -C saxpy run-vlsim
+	$(MAKE) -C sfilter run-vlsim
+	$(MAKE) -C nearn run-vlsim
+	$(MAKE) -C guassian run-vlsim
+
+clean:
+	$(MAKE) -C vecadd clean
+	$(MAKE) -C sgemm clean
+	$(MAKE) -C saxpy clean
+	$(MAKE) -C sfilter clean
+	$(MAKE) -C nearn clean
+	$(MAKE) -C guassian clean
+
+clean-all:
+	$(MAKE) -C vecadd clean-all
+	$(MAKE) -C sgemm clean-all
+	$(MAKE) -C saxpy clean-all
+	$(MAKE) -C sfilter clean-all
+	$(MAKE) -C nearn clean-all
+	$(MAKE) -C guassian clean-all

tests/opencl/VectorHypot/Makefile

@@ -0,0 +1,65 @@
+RISCV_TOOLCHAIN_PATH ?= $(wildcard ../../../../riscv-gnu-toolchain/drops)
+POCL_CC_PATH ?= $(wildcard ../../../../pocl/drops_riscv_cc)
+POCL_INC_PATH ?= $(wildcard ../include)
+POCL_LIB_PATH ?= $(wildcard ../lib)
+VORTEX_RT_PATH ?= $(wildcard ../../../runtime)
+VX_SIMX_PATH ?= $(wildcard ../../../simX/obj_dir)
+
+CC  = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gcc
+CXX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-g++
+DMP = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objdump
+HEX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objcopy
+GDB = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gdb
+
+VX_SRCS =  $(VORTEX_RT_PATH)/newlib/newlib.c
+VX_SRCS += $(VORTEX_RT_PATH)/startup/vx_start.S
+VX_SRCS += $(VORTEX_RT_PATH)/intrinsics/vx_intrinsics.S
+VX_SRCS += $(VORTEX_RT_PATH)/io/vx_io.S $(VORTEX_RT_PATH)/io/vx_io.c
+VX_SRCS += $(VORTEX_RT_PATH)/fileio/fileio.S
+VX_SRCS += $(VORTEX_RT_PATH)/tests/tests.c
+VX_SRCS += $(VORTEX_RT_PATH)/vx_api/vx_api.c
+
+VX_CFLAGS = -nostartfiles -Wl,-Bstatic,-T,$(VORTEX_RT_PATH)/startup/vx_link.ld
+
+CXXFLAGS =  -g -O0 -march=rv32im -mabi=ilp32 
+CXXFLAGS += -ffreestanding # program may not begin at main()
+CXXFLAGS += -Wl,--gc-sections # enable garbage collection of unused input sections
+CXXFLAGS += -fno-rtti -fno-non-call-exceptions # disable RTTI and exceptions
+CXXFLAGS += -I$(POCL_INC_PATH) -I.
+
+VX_LIBS = -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
+QEMU_LIBS = $(VORTEX_RT_PATH)/qemu/vx_api.c -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
+
+PROJECT=VectorHypot
+
+all: $(PROJECT).dump $(PROJECT).hex
+
+lib$(PROJECT).a: VectorHypot.cl
+	POCL_DEBUG=all POCL_DEBUG_LLVM_PASSES=1 LD_LIBRARY_PATH=$(RISCV_TOOLCHAIN_PATH)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -o lib$(PROJECT).a kernel.cl
+
+$(PROJECT).elf: main.cc lib$(PROJECT).a
+	$(CXX) $(CXXFLAGS) $(VX_CFLAGS) $(VX_SRCS) main.cc $(VX_LIBS) -o $(PROJECT).elf
+
+$(PROJECT).qemu: main.cc lib$(PROJECT).a
+	$(CXX) $(CXXFLAGS) main.cc $(QEMU_LIBS) -o $(PROJECT).qemu
+
+$(PROJECT).hex: $(PROJECT).elf
+	$(HEX) -O ihex $(PROJECT).elf $(PROJECT).hex
+
+$(PROJECT).dump: $(PROJECT).elf
+	$(DMP) -D $(PROJECT).elf > $(PROJECT).dump
+
+run: $(PROJECT).hex
+	POCL_DEBUG=all $(VX_SIMX_PATH)/Vcache_simX -E -a rv32i --core $(PROJECT).hex -s -b 1> emulator.debug
+
+qemu: $(PROJECT).qemu
+	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -d in_asm -D debug.log $(PROJECT).qemu
+
+gdb-s: $(PROJECT).qemu
+	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -g 1234 -d in_asm -D debug.log $(PROJECT).qemu
+
+gdb-c: $(PROJECT).qemu
+	$(GDB) $(PROJECT).qemu
+
+clean:
+	rm -rf *.elf *.dump *.hex

tests/opencl/VectorHypot/VectorHypot.cl

@@ -0,0 +1,41 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+ 
+// OpenCL Kernel Function Naive Implementation for hyptenuse
+__kernel void VectorHypot(__global float4* fg4A, __global float4* fg4B, __global float4* fg4Hypot, unsigned int uiOffset, int iInnerLoopCount, unsigned int uiNumElements)
+{
+    // get index into global data array
+    size_t szGlobalOffset = get_global_id(0) + uiOffset;
+
+    // bound check 
+    if (szGlobalOffset >= uiNumElements)
+    {   
+        return; 
+    }
+
+    // Processing 4 elements per work item, so read fgA and fgB source values from GMEM
+    float4 f4A = fg4A[szGlobalOffset];
+    float4 f4B = fg4B[szGlobalOffset];
+    float4 f4H = (float4)0.0f;
+     
+    // Get the hypotenuses the vectors of 'legs', but exaggerate the time needed with loop  
+    for (int i = 0; i < iInnerLoopCount; i++)  
+    {
+        // compute the 4 hypotenuses using built-in function
+        f4H.x = hypot (f4A.x, f4B.x);
+        f4H.y = hypot (f4A.y, f4B.y);
+        f4H.z = hypot (f4A.z, f4B.z);
+        f4H.w = hypot (f4A.w, f4B.w);
+    }
+    
+    // Write 4 result values back out to GMEM
+    fg4Hypot[szGlobalOffset] = f4H;
+}

tests/opencl/VectorHypot/main.cc

@@ -0,0 +1,686 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+
+// *********************************************************************
+// oclCopyComputeOverlap Notes:  
+//
+// OpenCL API demo application for NVIDIA CUDA GPU's that implements a
+// element by element vector hyptenuse computation using 2 input float arrays
+// and 1 output float array.
+//
+// Demonstrates host->GPU and GPU->host copies that are asynchronous/overlapped
+// with respect to GPU computation (and with respect to host thread). 
+//
+// Because the overlap acheivable for this computation and data set on a given system depends upon the GPU being used and the
+// GPU/Host bandwidth, the sample adjust the computation duration to test the most ideal case and test against a consistent standard.
+// This sample should be able to achieve up to 30% overlap on GPU's arch 1.2 and 1.3, and up to 50% on arch 2.0+ (Fermi) GPU's.
+//
+// After setup, warmup and calibration to the system, the sample runs 4 scenarios:  
+//      A) Computations with 2 command queues on GPU
+//         A multiple-cycle sequence is executed, timed and compared against the host
+//      B) Computations with 1 command queue on GPU
+//         A multiple-cycle sequence is executed, timed and compared against the host
+//
+//      The 2-command queue approach ought to be substantially faster
+//
+// For developmental purposes, the "iInnerLoopCount" variable passes into kernel and independently 
+// increases compute time without increasing data size (via a loop inside the kernel)
+//
+//      At some value of iInnerLoopCount, # of elements, workgroup size, etc the Overlap percentage should reach 30%:
+//      (This ~naively assumes time H2D bandwidth is the same as D2H bandwidth, but this is close on most systems)
+//
+//      If we name the time to copy single input vector H2D (or outpute vector D2H) as "T", then the optimum comparison case is:
+//        
+//          Single Queue with all the data and all the work  
+//             Ttot (serial)        = 4T + 4T + 2T      = 10T    
+//
+//          Dual Queue, where each queue has 1/2 the data and 1/2 the work 
+//             Tq0  (overlap)       = 2T + 2T + T .... 
+//             Tq1  (overlap)       = .... 2T + 2T + T
+//
+//             Ttot (elapsed, wall) = 2T + 2T + 2T + T  = 7T
+//
+//          Best Overlap % = 100.0 * (10T - 7T)/10T = 30.0 %	(Tesla arch 1.2 or 1.3, single copy engine)
+//
+//			For multiple independent cycles using arch >= 2.0 with 2 copy engines, input and output copies can also be overlapped.
+//			This doesn't help for the first cycle, but theoretically can lead to 50% overlap over many independent cycles.			
+// *********************************************************************
+
+// common SDK header for standard utilities and system libs 
+#include <oclUtils.h>
+#include <shrQATest.h>
+
+// Best possible and Min ratio of compute/copy overlap timing benefit to pass the test
+// values greater than 0.0f represent a speed-up relative to non-overlapped
+#define EXPECTED_OVERLAP 30.0f
+#define EXPECTED_OVERLAP_FERMI 45.0f
+#define PASS_FACTOR 0.60f
+#define RETRIES_ON_FAILURE 1
+
+// Base sizes for parameters manipulated dynamically or on the command line
+#define BASE_WORK_ITEMS 64
+#define BASE_ARRAY_LENGTH 40000
+#define BASE_LOOP_COUNT 32
+
+// Vars
+// *********************************************************************
+cl_platform_id cpPlatform;                          // OpenCL platform
+cl_context cxGPUContext;                            // OpenCL context
+cl_command_queue cqCommandQueue[2];                 // OpenCL command queues
+cl_device_id* cdDevices;                            // OpenCL device list  
+cl_program cpProgram;                               // OpenCL program
+cl_kernel ckKernel[2];                              // OpenCL kernel, 1 per queue
+cl_mem cmPinnedSrcA;                                // OpenCL pinned host source buffer A
+cl_mem cmPinnedSrcB;                                // OpenCL pinned host source buffer B 
+cl_mem cmPinnedResult;                              // OpenCL pinned host result buffer 
+float* fSourceA = NULL;                             // Mapped pointer for pinned Host source A buffer
+float* fSourceB = NULL;                             // Mapped pointer for pinned Host source B buffer
+float* fResult = NULL;                              // Mapped pointer for pinned Host result buffer 
+cl_mem cmDevSrcA;                                   // OpenCL device source buffer A
+cl_mem cmDevSrcB;                                   // OpenCL device source buffer B 
+cl_mem cmDevResult;                                 // OpenCL device result buffer 
+size_t szBuffBytes;                                 // Size of main buffers
+size_t szGlobalWorkSize;                            // 1D var for Total # of work items in the launched ND range
+size_t szLocalWorkSize = BASE_WORK_ITEMS;           // initial # of work items in the work group	
+cl_int ciErrNum;			                        // Error code var
+char* cPathAndName = NULL;                          // Var for full paths to data, src, etc.
+char* cSourceCL = NULL;                             // Buffer to hold source for compilation 
+const char* cExecutableName = NULL;
+
+// demo config vars
+const char* cSourceFile = "VectorHypot.cl";         // OpenCL computation kernel source code
+float* Golden = NULL;                               // temp buffer to hold golden results for cross check    
+bool bNoPrompt = false;                             // Command line switch to skip exit prompt
+bool bQATest = false;                               // Command line switch to test
+
+// Forward Declarations
+// *********************************************************************
+double DualQueueSequence(int iCycles, unsigned int uiNumElements, bool bShowConfig);
+double OneQueueSequence(int iCycles, unsigned int uiNumElements, bool bShowConfig);
+int AdjustCompute(cl_device_id cdTargetDevice, unsigned int uiNumElements, int iInitialLoopCount, int iCycles); 
+void VectorHypotHost(const float* pfData1, const float* pfData2, float* pfResult, unsigned int uiNumElements, int iInnerLoopCount);
+void Cleanup (int iExitCode);
+void (*pCleanup)(int) = &Cleanup;
+
+int *gp_argc = 0;
+const char *** gp_argv = NULL;
+
+// Main function 
+// *********************************************************************
+int main(int argc, const char **argv)
+{
+    //Locals
+    size_t szKernelLength;                      // Byte size of kernel code
+    double dBuildTime;                          // Compile time
+    cl_uint uiTargetDevice = 0;	                // Default Device to compute on
+    cl_uint uiNumDevsUsed = 1;                  // Number of devices used in this sample   
+    cl_uint uiNumDevices;                       // Number of devices available 
+    int iDevCap = -1;                           // Capability of device
+    int iInnerLoopCount = BASE_LOOP_COUNT;      // Varies "compute intensity" per data within the kernel 
+    const int iTestCycles = 10;                 // How many times to run the external test loop 
+    const int iWarmupCycles = 8;                // How many times to run the warmup sequence 
+    cl_uint uiWorkGroupMultiple = 4;            // Command line var (using "workgroupmult=<n>") to optionally increase workgroup size
+    cl_uint uiNumElements = BASE_ARRAY_LENGTH;  // initial # of elements per array to process (note: procesing 4 per work item)
+    cl_uint uiSizeMultiple = 4;                 // Command line var (using "sizemult=<n>") to optionally increase vector sizes
+    bool bPassFlag = false;                     // Var to accumulate test pass/fail
+    shrBOOL bMatch = shrFALSE;                  // Cross check result
+	shrBOOL bTestOverlap = shrFALSE;
+	double dAvgGPUTime[2] = {0.0, 0.0};         // Average time of iTestCycles calls for 2-Queue and 1-Queue test
+    double dHostTime[2] = {0.0, 0.0};           // Host computation time (2nd test is redundant but a good stability indicator)
+    float fMinPassCriteria[2] = {0.0f, 0.0f};	// Test pass cireria, adjusted dependant on GPU arch
+
+    gp_argc = &argc;
+    gp_argv = &argv;
+
+    shrQAStart(argc, (char **)argv);
+
+    // start logs 
+	cExecutableName = argv[0];
+    shrSetLogFileName ("oclCopyComputeOverlap.txt");
+    shrLog("%s Starting...\n\n", argv[0]); 
+
+    // get basic command line args 
+    bNoPrompt = (shrTRUE == shrCheckCmdLineFlag(argc, argv, "noprompt"));
+    bQATest   = (shrTRUE == shrCheckCmdLineFlag(argc, argv, "qatest"));
+    shrGetCmdLineArgumentu(argc, argv, "device", &uiTargetDevice);
+
+    // Optional Command-line multiplier for vector size 
+    //   Default val of 4 gives 10.24 million float elements per vector
+    //   Range of 3 - 16 (7.68 to 40.96 million floats) is reasonable range (if system and GPU have enough memory)
+    shrGetCmdLineArgumentu(argc, argv, "sizemult", &uiSizeMultiple);
+    uiSizeMultiple = CLAMP(uiSizeMultiple, 1, 50);  
+    uiNumElements = uiSizeMultiple * BASE_ARRAY_LENGTH * BASE_WORK_ITEMS;
+    shrLog("Array sizes = %u float elements\n", uiNumElements); 
+
+    // Optional Command-line multiplier for workgroup size (x 64 work items)
+    //   Default val of 4 gives szLocalWorkSize of 256.
+    //   Range of 1 - 8 (resulting in workgroup sizes of 64 to 512) is reasonable range
+    shrGetCmdLineArgumentu(argc, argv, "workgroupmult", &uiWorkGroupMultiple);
+    uiWorkGroupMultiple = CLAMP(uiWorkGroupMultiple, 1, 10); 
+    szLocalWorkSize = uiWorkGroupMultiple * BASE_WORK_ITEMS;
+    shrLog("Workgroup Size = %u\n\n", szLocalWorkSize); 
+
+    // Get the NVIDIA platform if available, otherwise use default
+    shrLog("Get the Platform ID...\n\n");
+    ciErrNum = oclGetPlatformID(&cpPlatform);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+    // Get OpenCL platform name and version
+    char cBuffer[256];
+    ciErrNum = clGetPlatformInfo (cpPlatform, CL_PLATFORM_NAME, sizeof(cBuffer), cBuffer, NULL);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    shrLog("Platform Name = %s\n\n", cBuffer);
+
+    // Get all the devices
+    shrLog("Get the Device info and select Device...\n");
+    ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_DEFAULT, 0, NULL, &uiNumDevices);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    cdDevices = (cl_device_id*)malloc(uiNumDevices * sizeof(cl_device_id));
+
+    // Ethans changes
+    CL_CHECK(clGetPlatformIDs(1, &platform_id, NULL));
+    CL_CHECK(clGetDeviceIDs(platform_id, CL_DEVICE_TYPE_DEFAULT, 1, &device_id, NULL));
+    //ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, uiNumDevices, cdDevices, NULL);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+    // Set target device and check capabilities 
+    shrLog(" # of Devices Available = %u\n", uiNumDevices); 
+    uiTargetDevice = CLAMP(uiTargetDevice, 0, (uiNumDevices - 1));
+    shrLog(" Using Device %u, ", uiTargetDevice); 
+    oclPrintDevName(LOGBOTH, cdDevices[uiTargetDevice]);  
+    iDevCap = oclGetDevCap(cdDevices[uiTargetDevice]);
+    if (iDevCap > 0) {
+       shrLog(", Capability = %d.%d\n\n", iDevCap/10, iDevCap%10);
+    } else {
+       shrLog("\n\n", iDevCap); 
+    }
+    if (strstr(cBuffer, "NVIDIA") != NULL)
+    {
+        if (iDevCap < 12)
+        {
+            shrLog("Device doesn't have overlap capability.  Skipping test...\n"); 
+            Cleanup (EXIT_SUCCESS);
+        }
+
+		// Device and Platform eligible for overlap testing
+		bTestOverlap = shrTRUE;
+
+        // If device has overlap capability, proceed
+        fMinPassCriteria[0] = PASS_FACTOR * EXPECTED_OVERLAP;               // 1st cycle overlap is same for 1 or 2 copy engines
+        if (iDevCap != 20) 
+        {
+            // Single copy engine
+            fMinPassCriteria[1] = PASS_FACTOR * EXPECTED_OVERLAP;            // avg of many cycles
+        }
+        else 
+        {
+            char cDevName[1024];
+            clGetDeviceInfo(cdDevices[uiTargetDevice], CL_DEVICE_NAME, sizeof(cDevName), &cDevName, NULL);
+            if(strstr(cDevName, "Quadro")!=0 || strstr(cDevName, "Tesla")!=0)
+            {
+                // Tesla or Quadro (arch = 2.0) ... Dual copy engine 
+                fMinPassCriteria[1] = PASS_FACTOR * EXPECTED_OVERLAP_FERMI;  // average of many cycles
+            }
+            else
+            {
+                // Geforce ... Single copy engine
+                fMinPassCriteria[1] = PASS_FACTOR * EXPECTED_OVERLAP;        // average of many cycles
+            }
+        }
+    }   
+
+    // Create the context
+    shrLog("clCreateContext...\n"); 
+    cxGPUContext = clCreateContext(0, uiNumDevsUsed, &cdDevices[uiTargetDevice], NULL, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    
+    // Create 2 command-queues
+    cqCommandQueue[0] = clCreateCommandQueue(cxGPUContext, cdDevices[uiTargetDevice], 0, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    shrLog("clCreateCommandQueue [0]...\n"); 
+    cqCommandQueue[1] = clCreateCommandQueue(cxGPUContext, cdDevices[uiTargetDevice], 0, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    shrLog("clCreateCommandQueue [1]...\n"); 
+
+    // Allocate the OpenCL source and result buffer memory objects on GPU device GMEM
+    szBuffBytes = sizeof(cl_float) * uiNumElements;
+    cmDevSrcA = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, szBuffBytes, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    cmDevSrcB = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, szBuffBytes, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    cmDevResult = clCreateBuffer(cxGPUContext, CL_MEM_WRITE_ONLY, szBuffBytes, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    shrLog("clCreateBuffer (Src A, Src B and Result GPU Device GMEM, 3 x %u floats) ...\n", uiNumElements); 
+
+    // Allocate pinned source and result host buffers:  
+    //   Note: Pinned (Page Locked) memory is needed for async host<->GPU memory copy operations ***
+    cmPinnedSrcA = clCreateBuffer(cxGPUContext, CL_MEM_READ_WRITE | CL_MEM_ALLOC_HOST_PTR, szBuffBytes, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    cmPinnedSrcB = clCreateBuffer(cxGPUContext, CL_MEM_READ_WRITE | CL_MEM_ALLOC_HOST_PTR, szBuffBytes, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    cmPinnedResult = clCreateBuffer(cxGPUContext, CL_MEM_READ_WRITE | CL_MEM_ALLOC_HOST_PTR, szBuffBytes, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    shrLog("clCreateBuffer (Src A, Src B and Result Pinned Host buffers, 3 x %u floats)...\n\n", uiNumElements); 
+
+    // Get mapped pointers to pinned input host buffers
+    //   Note:  This allows general (non-OpenCL) host functions to access pinned buffers using standard pointers
+    fSourceA = (cl_float*)clEnqueueMapBuffer(cqCommandQueue[0], cmPinnedSrcA, CL_TRUE, CL_MAP_WRITE, 0, szBuffBytes, 0, NULL, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    fSourceB = (cl_float*)clEnqueueMapBuffer(cqCommandQueue[0], cmPinnedSrcB, CL_TRUE, CL_MAP_WRITE, 0, szBuffBytes, 0, NULL, NULL, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    fResult = (cl_float*)clEnqueueMapBuffer(cqCommandQueue[0], cmPinnedResult, CL_TRUE, CL_MAP_READ, 0, szBuffBytes, 0, NULL, NULL, &ciErrNum);
+    //oclCheckErrorEX (ciErrNum, CL_SUCCESS, pCleanup);
+    shrLog("clEnqueueMapBuffer (Pointers to 3 pinned host buffers)...\n"); 
+
+    // Alloc temp golden buffer for cross checks
+    Golden = (float*)malloc(szBuffBytes);
+    //oclCheckErrorEX(Golden != NULL, shrTRUE, pCleanup);
+
+    // Read the OpenCL kernel in from source file
+    cPathAndName = shrFindFilePath(cSourceFile, argv[0]);
+    //oclCheckError(cPathAndName != NULL, shrTRUE);
+    cSourceCL = oclLoadProgSource(cPathAndName, "", &szKernelLength);
+   // oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    shrLog("oclLoadProgSource (%s)...\n", cSourceFile); 
+
+    // Create the program object
+    //cpProgram = clCreateProgramWithSource(cxGPUContext, 1, (const char **)&cSourceCL, &szKernelLength, &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    shrLog("clCreateProgramWithSource...\n"); 
+    cl_program program =
+      clCreateProgramWithBuiltInKernels(context, 1, &device_id, "VectorHypot", NULL);
+    // Build the program for the target device
+    clFinish(cqCommandQueue[0]);
+    shrDeltaT(0);
+    ciErrNum = clBuildProgram(program, uiNumDevsUsed, &cdDevices[uiTargetDevice], "-cl-fast-relaxed-math", NULL, NULL);
+    shrLog("clBuildProgram..."); 
+    if (ciErrNum != CL_SUCCESS)
+    {
+        // write out standard error, Build Log and PTX, then cleanup and exit
+        shrLogEx(LOGBOTH | ERRORMSG, (double)ciErrNum, STDERROR);
+        oclLogBuildInfo(program, oclGetFirstDev(cxGPUContext));
+        oclLogPtx(program, oclGetFirstDev(cxGPUContext), "VectorHypot.ptx");
+        Cleanup(EXIT_FAILURE);
+    }
+    dBuildTime = shrDeltaT(0);
+
+    // Ethan - Kernel Addition
+
+    if (program == NULL) {
+        std::cerr << "Failed to write program binary" << std::endl;
+        Cleanup(context, queue, program, kernel, memObjects);
+        return 1;
+    } else {
+        std::cout << "Read program from binary." << std::endl;
+    }
+
+    // Create the kernel
+    ckKernel[0] = clCreateKernel(program, "VectorHypot", &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    ckKernel[1] = clCreateKernel(program, "VectorHypot", &ciErrNum);
+    //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    shrLog("clCreateKernel (ckKernel[2])...\n"); 
+
+    // Offsets for 2 queues
+    cl_uint uiOffset[2] = {0, uiNumElements / (2 * 4)};
+
+    // Set the Argument values for the 1st kernel instance (queue 0)
+    ciErrNum = clSetKernelArg(ckKernel[0], 0, sizeof(cl_mem), (void*)&cmDevSrcA);
+    ciErrNum |= clSetKernelArg(ckKernel[0], 1, sizeof(cl_mem), (void*)&cmDevSrcB);
+    ciErrNum |= clSetKernelArg(ckKernel[0], 2, sizeof(cl_mem), (void*)&cmDevResult);
+    ciErrNum |= clSetKernelArg(ckKernel[0], 3, sizeof(cl_uint), (void*)&uiOffset[0]);
+    ciErrNum |= clSetKernelArg(ckKernel[0], 4, sizeof(cl_int), (void*)&iInnerLoopCount);
+    ciErrNum |= clSetKernelArg(ckKernel[0], 5, sizeof(cl_uint), (void*)&uiNumElements);
+    shrLog("clSetKernelArg ckKernel[0] args 0 - 5...\n"); 
+
+    // Set the Argument values for the 2d kernel instance (queue 1)
+    ciErrNum |= clSetKernelArg(ckKernel[1], 0, sizeof(cl_mem), (void*)&cmDevSrcA);
+    ciErrNum |= clSetKernelArg(ckKernel[1], 1, sizeof(cl_mem), (void*)&cmDevSrcB);
+    ciErrNum |= clSetKernelArg(ckKernel[1], 2, sizeof(cl_mem), (void*)&cmDevResult);
+    ciErrNum |= clSetKernelArg(ckKernel[1], 3, sizeof(cl_uint), (void*)&uiOffset[1]);
+    ciErrNum |= clSetKernelArg(ckKernel[1], 4, sizeof(cl_int), (void*)&iInnerLoopCount);
+    ciErrNum |= clSetKernelArg(ckKernel[1], 5, sizeof(cl_uint), (void*)&uiNumElements);
+    oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    shrLog("clSetKernelArg ckKernel[1] args 0 - 5...\n\n"); 
+
+    //*******************************************
+    // Warmup the driver with dual queue sequence
+    //*******************************************
+
+    // Warmup with dual queue sequence for iTestCycles
+    shrLog("Warmup with 2-Queue sequence, %d cycles...\n", iWarmupCycles);
+    DualQueueSequence(iWarmupCycles, uiNumElements, false);
+
+    // Use single queue config to adjust compute intensity 
+    shrLog("Adjust compute for GPU / system...\n");
+    iInnerLoopCount = AdjustCompute(cdDevices[uiTargetDevice], uiNumElements, iInnerLoopCount, iTestCycles); 
+    shrLog("  Kernel inner loop count = %d\n", iInnerLoopCount); 
+
+    //*******************************************
+    // Run and time with 2 command-queues
+    //*******************************************
+	for( int iRun =0; iRun <= RETRIES_ON_FAILURE; ++iRun ) {
+	
+	// Run the sequence iTestCycles times
+    dAvgGPUTime[0] = DualQueueSequence(iTestCycles, uiNumElements, false);
+
+    // Warmup then Compute on host iTestCycles times (using mapped standard pointer to pinned host cl_mem buffer) 
+    shrLog("  Device vs Host Result Comparison\t: "); 
+    VectorHypotHost(fSourceA, fSourceB, Golden, uiNumElements, iInnerLoopCount);   
+    shrDeltaT(0);
+    for (int i = 0; i < iTestCycles; i++)
+    {
+        VectorHypotHost (fSourceA, fSourceB, Golden, uiNumElements, iInnerLoopCount);   
+    }
+    dHostTime[0] = shrDeltaT(0)/iTestCycles;
+
+    // Compare host and GPU results (using mapped standard pointer to pinned host cl_mem buffer) 
+    bMatch = shrComparefet(Golden, fResult, uiNumElements, 0.0f, 0);
+    shrLog("gpu %s cpu\n", (bMatch == shrTRUE) ? "MATCHES" : "DOESN'T MATCH"); 
+    bPassFlag = (bMatch == shrTRUE);
+
+    //*******************************************
+    // Run and time with 1 command queue
+    //*******************************************
+    // Run the sequence iTestCycles times
+    dAvgGPUTime[1] = OneQueueSequence(iTestCycles, uiNumElements, false);
+
+    // Compute on host iTestCycles times (using mapped standard pointer to pinned host cl_mem buffer) 
+    shrLog("  Device vs Host Result Comparison\t: "); 
+    shrDeltaT(0);
+    for (int i = 0; i < iTestCycles; i++)
+    {
+        VectorHypotHost(fSourceA, fSourceB, Golden, (int)uiNumElements, iInnerLoopCount);   
+    }
+    dHostTime[1] = shrDeltaT(0)/iTestCycles;
+
+    // Compare host and GPU results (using mapped standard pointer to pinned host cl_mem buffer) 
+    bMatch = shrComparefet(Golden, fResult, uiNumElements, 0.0f, 0);
+    shrLog("gpu %s cpu\n", (bMatch == shrTRUE) ? "MATCHES" : "DOESN'T MATCH"); 
+    bPassFlag &= (bMatch == shrTRUE);
+
+    //*******************************************
+
+    // Compare Single and Dual queue timing 
+    shrLog("\nResult Summary:\n"); 
+
+    // Log GPU and CPU Time for 2-queue scenario
+    shrLog("  Avg GPU Elapsed Time for 2-Queues\t= %.5f s\n", dAvgGPUTime[0]);
+    shrLog("  Avg Host Elapsed Time\t\t\t= %.5f s\n\n", dHostTime[0]);
+
+    // Log GPU and CPU Time for 1-queue scenario
+    shrLog("  Avg GPU Elapsed Time for 1-Queue\t= %.5f s\n", dAvgGPUTime[1]);
+    shrLog("  Avg Host Elapsed Time\t\t\t= %.5f s\n\n", dHostTime[1]);
+
+    // Log overlap % for GPU (comparison of 2-queue and 1 queue scenarios) and status
+    double dAvgOverlap = 100.0 * (1.0 - dAvgGPUTime[0]/dAvgGPUTime[1]);
+    
+	if( bTestOverlap ) {
+		bool bAvgOverlapOK = (dAvgOverlap >= fMinPassCriteria[1]);
+		if( iRun == RETRIES_ON_FAILURE || bAvgOverlapOK ) {
+			shrLog("  Measured and (Acceptable) Avg Overlap\t= %.1f %% (%.1f %%)  -> Measured Overlap is %s\n\n", dAvgOverlap, fMinPassCriteria[1], bAvgOverlapOK ? "Acceptable" : "NOT Acceptable");
+
+			// Log info to master log in standard format
+			shrLogEx(LOGBOTH | MASTER, 0, "oclCopyComputeOverlap-Avg, Throughput = %.4f OverlapPercent, Time = %.5f s, Size = %u Elements, NumDevsUsed = %u, Workgroup = %u\n", 
+			dAvgOverlap, dAvgGPUTime[0], uiNumElements, uiNumDevsUsed, szLocalWorkSize); 
+			
+			bPassFlag &= bAvgOverlapOK;
+			break;
+		}
+	} 
+
+		shrLog("  Measured and (Acceptable) Avg Overlap\t= %.1f %% (%.1f %%)  -> Retry %d more time(s)...\n\n", dAvgOverlap, fMinPassCriteria[1], RETRIES_ON_FAILURE - iRun);
+	}
+
+
+    //*******************************************
+    // Report pass/fail, cleanup and exit
+    Cleanup (bPassFlag ? EXIT_SUCCESS : EXIT_FAILURE);
+}
+
+// Run 1 queue sequence for n cycles
+// *********************************************************************
+double OneQueueSequence(int iCycles, unsigned int uiNumElements, bool bShowConfig)
+{
+    // Use fresh source Data: (re)initialize pinned host array buffers (using mapped standard pointer to pinned host cl_mem buffer) 
+    shrFillArray(fSourceA, (int)uiNumElements);
+    shrFillArray(fSourceB, (int)uiNumElements);
+
+    // Reset Global work size for 1 command-queue, and log work sizes & dimensions
+    szGlobalWorkSize = shrRoundUp((int)szLocalWorkSize, (int)(uiNumElements/4));
+
+    // *** Make sure queues are empty and then start timer 
+    double dAvgTime = 0.0;
+    clFinish(cqCommandQueue[0]);
+    clFinish(cqCommandQueue[1]);  
+    shrDeltaT(0);
+
+    // Run the sequence iCycles times
+    for (int i = 0; i < iCycles; i++)
+    {
+        // Nonblocking Write of all of input data from host to device in command-queue 0 
+        ciErrNum = clEnqueueWriteBuffer(cqCommandQueue[0], cmDevSrcA, CL_FALSE, 0, szBuffBytes, (void*)&fSourceA[0], 0, NULL, NULL);
+        ciErrNum |= clEnqueueWriteBuffer(cqCommandQueue[0], cmDevSrcB, CL_FALSE, 0, szBuffBytes, (void*)&fSourceB[0], 0, NULL, NULL);
+        shrCheckError(ciErrNum, CL_SUCCESS);
+
+        // Launch kernel computation, command-queue 0 
+        ciErrNum = clEnqueueNDRangeKernel(cqCommandQueue[0], ckKernel[0], 1, NULL, &szGlobalWorkSize, &szLocalWorkSize, 0, NULL, NULL);
+        //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+        // Non Blocking Read of output data from device to host, command-queue 0 
+        ciErrNum = clEnqueueReadBuffer(cqCommandQueue[0], cmDevResult, CL_FALSE, 0, szBuffBytes, (void*)&fResult[0], 0, NULL, NULL);
+        shrCheckError(ciErrNum, CL_SUCCESS);
+
+        // Flush sequence to device (may not be necessary on Linux or WinXP or when using the NVIDIA Tesla Computing Cluster driver)
+        clFlush(cqCommandQueue[0]);
+    }
+    
+    // *** Assure sync to host and return average sequence time
+    clFinish(cqCommandQueue[0]);
+    dAvgTime = shrDeltaT(0)/(double)iCycles;
+
+    // Log config if asked for
+    if (bShowConfig)
+    {
+        shrLog("\n1-Queue sequence Configuration:\n");
+        shrLog("  Global Work Size (per command-queue)\t= %u\n  Local Work Size \t\t\t= %u\n  # of Work Groups (per command-queue)\t= %u\n  # of command-queues\t\t\t= 1\n", 
+           szGlobalWorkSize, szLocalWorkSize, szGlobalWorkSize/szLocalWorkSize); 
+    }
+    return dAvgTime;
+}
+
+// Run 2 queue sequence for n cycles
+// *********************************************************************
+double DualQueueSequence(int iCycles, unsigned int uiNumElements, bool bShowConfig)
+{
+    // Locals
+    size_t szHalfBuffer = szBuffBytes / 2;
+    size_t szHalfOffset = szHalfBuffer / sizeof(float);
+    double dAvgTime = 0.0;
+
+    // Use fresh source Data: (re)initialize pinned host array buffers (using mapped standard pointer to pinned host cl_mem buffer) 
+    shrFillArray(fSourceA, (int)uiNumElements);
+    shrFillArray(fSourceB, (int)uiNumElements);
+
+    // Set Global work size for 2 command-queues, and log work sizes & dimensions
+    szGlobalWorkSize = shrRoundUp((int)szLocalWorkSize, (int)(uiNumElements/(2 * 4)));
+
+    // Make sure queues are empty and then start timer 
+    clFinish(cqCommandQueue[0]);
+    clFinish(cqCommandQueue[1]);
+    shrDeltaT(0);
+
+    for (int i = 0; i < iCycles; i++)
+    {
+        // Mid Phase 0 
+        // Nonblocking Write of 1st half of input data from host to device in command-queue 0 
+        ciErrNum = clEnqueueWriteBuffer(cqCommandQueue[0], cmDevSrcA, CL_FALSE, 0, szHalfBuffer, (void*)&fSourceA[0], 0, NULL, NULL);
+        ciErrNum |= clEnqueueWriteBuffer(cqCommandQueue[0], cmDevSrcB, CL_FALSE, 0, szHalfBuffer, (void*)&fSourceB[0], 0, NULL, NULL);
+        shrCheckError(ciErrNum, CL_SUCCESS);
+
+        // Push out the write for queue 0 (and prior read from queue 1 at end of loop) to the driver 
+        // (not necessary on Linux, Mac OSX or WinXP)
+        clFlush(cqCommandQueue[0]);
+        clFlush(cqCommandQueue[1]);
+
+        // Start Phase 1 ***********************************
+
+        // Launch kernel computation, command-queue 0
+        // (Note:  The order MATTERS here on Fermi !  THE KERNEL IN THIS PHASE SHOULD BE LAUNCHED BEFORE THE WRITE)
+        ciErrNum = clEnqueueNDRangeKernel(cqCommandQueue[0], ckKernel[0], 1, NULL, &szGlobalWorkSize, &szLocalWorkSize, 0, NULL, NULL);
+        oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+        // Nonblocking Write of 2nd half of input data from host to device in command-queue 1 
+        // (Note:  The order MATTERS here on Fermi !  THE KERNEL IN THIS PHASE SHOULD BE LAUNCHED BEFORE THE WRITE)
+        ciErrNum = clEnqueueWriteBuffer(cqCommandQueue[1], cmDevSrcA, CL_FALSE, szHalfBuffer, szHalfBuffer, (void*)&fSourceA[szHalfOffset], 0, NULL, NULL);
+        ciErrNum |= clEnqueueWriteBuffer(cqCommandQueue[1], cmDevSrcB, CL_FALSE, szHalfBuffer, szHalfBuffer, (void*)&fSourceB[szHalfOffset], 0, NULL, NULL);
+        shrCheckError(ciErrNum, CL_SUCCESS);
+
+        // Push out the compute for queue 0 and write for queue 1 to the driver
+        // (not necessary on Linux, Mac OSX or WinXP)
+        clFlush(cqCommandQueue[0]);
+        clFlush(cqCommandQueue[1]);
+
+        // Start Phase 2 ***********************************
+
+        // Launch kernel computation, command-queue 1 
+        ciErrNum = clEnqueueNDRangeKernel(cqCommandQueue[1], ckKernel[1], 1, NULL, &szGlobalWorkSize, &szLocalWorkSize, 0, NULL, NULL);
+        //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+
+        // Non Blocking Read of 1st half of output data from device to host, command-queue 0 
+        ciErrNum = clEnqueueReadBuffer(cqCommandQueue[0], cmDevResult, CL_FALSE, 0, szHalfBuffer, (void*)&fResult[0], 0, NULL, NULL);
+        shrCheckError(ciErrNum, CL_SUCCESS);
+
+        // Push out the compute for queue 1 and the read for queue 0 to the driver 
+        // (not necessary on Linux, Mac OSX or WinXP)
+        clFlush(cqCommandQueue[0]);
+        clFlush(cqCommandQueue[1]);
+
+        // Start Phase 0 (Rolls over) ***********************************
+
+        // Non Blocking Read of 2nd half of output data from device to host, command-queue 1 
+        ciErrNum = clEnqueueReadBuffer(cqCommandQueue[1], cmDevResult, CL_FALSE, szHalfBuffer, szHalfBuffer, (void*)&fResult[szHalfOffset], 0, NULL, NULL);
+        shrCheckError(ciErrNum, CL_SUCCESS);
+    }
+
+    // *** Sync to host and get average sequence time
+    clFinish(cqCommandQueue[0]);
+    clFinish(cqCommandQueue[1]);    
+    dAvgTime = shrDeltaT(0)/(double)iCycles;
+
+    // Log config if asked for
+    if (bShowConfig)
+    {
+        shrLog("\n2-Queue sequence Configuration:\n");
+        shrLog("  Global Work Size (per command-queue)\t= %u\n  Local Work Size \t\t\t= %u\n  # of Work Groups (per command-queue)\t= %u\n  # of command-queues\t\t\t= 2\n", 
+           szGlobalWorkSize, szLocalWorkSize, szGlobalWorkSize/szLocalWorkSize); 
+    }
+
+    return dAvgTime;
+}
+
+// Function to adjust compute task according to device capability
+// This allows a consistent overlap % across a wide variety of GPU's for test purposes
+// It also implitly illustrates the relationship between compute capability and overlap at fixed work size
+// *********************************************************************
+int AdjustCompute(cl_device_id cdTargetDevice, unsigned int uiNumElements, int iInitLoopCount, int iCycles)
+{
+    // Locals
+    double dCopyTime, dComputeTime;
+    int iComputedLoopCount; 
+
+    // Change Source Data
+    shrFillArray(fSourceA, (int)uiNumElements);
+    shrFillArray(fSourceB, (int)uiNumElements);
+
+    // Reset Global work size for 1 command-queue, and log work sizes & dimensions
+    szGlobalWorkSize = shrRoundUp((int)szLocalWorkSize, (int)(uiNumElements/4));
+
+    // *** Make sure queues are empty and then start timer 
+    clFinish(cqCommandQueue[0]);
+    clFinish(cqCommandQueue[1]);  
+    shrDeltaT(0);
+
+    // Run the copy iCycles times and measure copy time on this system
+    for (int i = 0; i < iCycles; i++)
+    {
+        // Nonblocking Write of all of input data from host to device in command-queue 0 
+        ciErrNum = clEnqueueWriteBuffer(cqCommandQueue[0], cmDevSrcA, CL_FALSE, 0, szBuffBytes, (void*)&fSourceA[0], 0, NULL, NULL);
+        ciErrNum |= clEnqueueWriteBuffer(cqCommandQueue[0], cmDevSrcB, CL_FALSE, 0, szBuffBytes, (void*)&fSourceB[0], 0, NULL, NULL);
+        ciErrNum |= clFlush(cqCommandQueue[0]);
+        shrCheckError(ciErrNum, CL_SUCCESS);
+    }
+    clFinish(cqCommandQueue[0]);
+    dCopyTime = shrDeltaT(0);
+
+    // Run the compute iCycles times and measure compute time on this system
+    for (int i = 0; i < iCycles; i++)
+    {
+        // Launch kernel computation, command-queue 0 
+        ciErrNum = clEnqueueNDRangeKernel(cqCommandQueue[0], ckKernel[0], 1, NULL, &szGlobalWorkSize, &szLocalWorkSize, 0, NULL, NULL);
+        ciErrNum |= clFlush(cqCommandQueue[0]);
+        oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    }
+    clFinish(cqCommandQueue[0]);
+    dComputeTime = shrDeltaT(0);
+
+    // Determine number of core loop cycles proportional to copy/compute time ratio
+    dComputeTime = MAX(dComputeTime, 1.0e-6);
+    iComputedLoopCount = CLAMP(2, (int)((dCopyTime/dComputeTime) * (double)iInitLoopCount), (iInitLoopCount * 4));
+    ciErrNum |= clSetKernelArg(ckKernel[0], 4, sizeof(cl_int), (void*)&iComputedLoopCount);
+    ciErrNum |= clSetKernelArg(ckKernel[1], 4, sizeof(cl_int), (void*)&iComputedLoopCount);
+    oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
+    return (iComputedLoopCount);
+} 
+
+// Cleanup/Exit function 
+// *********************************************************************
+void Cleanup (int iExitCode)
+{
+    // Cleanup allocated objects
+    shrLog("Starting Cleanup...\n\n");
+    if(cPathAndName)free(cPathAndName);
+    if(cSourceCL)free(cSourceCL);
+    if(Golden)free(Golden);
+    if(ckKernel[0])clReleaseKernel(ckKernel[0]);
+    if(ckKernel[1])clReleaseKernel(ckKernel[1]);
+    if(program)clReleaseProgram(program);
+    if(fSourceA)clEnqueueUnmapMemObject(cqCommandQueue[0], cmPinnedSrcA, (void*)fSourceA, 0, NULL, NULL);
+    if(fSourceB)clEnqueueUnmapMemObject(cqCommandQueue[0], cmPinnedSrcB, (void*)fSourceB, 0, NULL, NULL);
+    if(fResult)clEnqueueUnmapMemObject(cqCommandQueue[0], cmPinnedResult, (void*)fResult, 0, NULL, NULL);
+    if(cmDevSrcA)clReleaseMemObject(cmDevSrcA);
+    if(cmDevSrcB)clReleaseMemObject(cmDevSrcB);
+    if(cmDevResult)clReleaseMemObject(cmDevResult);
+    if(cmPinnedSrcA)clReleaseMemObject(cmPinnedSrcA);
+    if(cmPinnedSrcB)clReleaseMemObject(cmPinnedSrcB);
+    if(cmPinnedResult)clReleaseMemObject(cmPinnedResult);
+    if(cqCommandQueue[0])clReleaseCommandQueue(cqCommandQueue[0]);
+    if(cqCommandQueue[1])clReleaseCommandQueue(cqCommandQueue[1]);
+    if(cxGPUContext)clReleaseContext(cxGPUContext);
+    if(cdDevices)free(cdDevices);
+
+    // Master status Pass/Fail (all tests)
+    shrQAFinishExit( *gp_argc, (const char **)*gp_argv, (iExitCode == EXIT_SUCCESS) ? QA_PASSED : QA_FAILED );
+}
+
+// "Golden" Host processing vector hyptenuse function for comparison purposes
+// *********************************************************************
+void VectorHypotHost(const float* pfData1, const float* pfData2, float* pfResult, unsigned int uiNumElements, int iInnerLoopCount)
+{
+    for (unsigned int i = 0; i < uiNumElements; i++) 
+    {
+        float fA = pfData1[i];
+        float fB = pfData2[i];
+        float fC = sqrtf(fA * fA + fB * fB);
+
+        pfResult[i] = fC;
+    }
+}

tests/opencl/VectorHypot/oclUtils.h

@@ -0,0 +1,198 @@
+/*
+ * Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+ 
+#ifndef OCL_UTILS_H
+#define OCL_UTILS_H
+
+// *********************************************************************
+// Utilities specific to OpenCL samples in NVIDIA GPU Computing SDK 
+// *********************************************************************
+
+// Common headers:  Cross-API utililties and OpenCL header
+#include <shrUtils.h>
+
+// All OpenCL headers
+#if defined (__APPLE__) || defined(MACOSX)
+    #include <OpenCL/opencl.h>
+#else
+    #include <CL/opencl.h>
+#endif 
+
+// Includes
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+
+// For systems with CL_EXT that are not updated with these extensions, we copied these
+// extensions from <CL/cl_ext.h>
+#ifndef CL_DEVICE_COMPUTE_CAPABILITY_MAJOR_NV
+  /* cl_nv_device_attribute_query extension - no extension #define since it has no functions */
+  #define CL_DEVICE_COMPUTE_CAPABILITY_MAJOR_NV       0x4000
+  #define CL_DEVICE_COMPUTE_CAPABILITY_MINOR_NV       0x4001
+  #define CL_DEVICE_REGISTERS_PER_BLOCK_NV            0x4002
+  #define CL_DEVICE_WARP_SIZE_NV                      0x4003
+  #define CL_DEVICE_GPU_OVERLAP_NV                    0x4004
+  #define CL_DEVICE_KERNEL_EXEC_TIMEOUT_NV            0x4005
+  #define CL_DEVICE_INTEGRATED_MEMORY_NV              0x4006
+#endif
+
+// reminders for build output window and log
+#ifdef _WIN32
+    #pragma message ("Note: including shrUtils.h")
+    #pragma message ("Note: including opencl.h")
+#endif
+
+// SDK Revision #
+#define OCL_SDKREVISION "7027912"
+
+// Error and Exit Handling Macros... 
+// *********************************************************************
+// Full error handling macro with Cleanup() callback (if supplied)... 
+// (Companion Inline Function lower on page)
+#define oclCheckErrorEX(a, b, c) __oclCheckErrorEX(a, b, c, __FILE__ , __LINE__) 
+
+// Short version without Cleanup() callback pointer
+// Both Input (a) and Reference (b) are specified as args
+#define oclCheckError(a, b) oclCheckErrorEX(a, b, 0) 
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the platform ID for NVIDIA if available, otherwise default to platform 0
+//!
+//! @return the id 
+//! @param clSelectedPlatformID         OpenCL platform ID
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_int oclGetPlatformID(cl_platform_id* clSelectedPlatformID);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Print info about the device
+//!
+//! @param iLogMode       enum LOGBOTH, LOGCONSOLE, LOGFILE
+//! @param device         OpenCL id of the device
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclPrintDevInfo(int iLogMode, cl_device_id device);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get and return device capability
+//!
+//! @return the 2 digit integer representation of device Cap (major minor). return -1 if NA 
+//! @param device         OpenCL id of the device
+//////////////////////////////////////////////////////////////////////////////
+extern "C" int oclGetDevCap(cl_device_id device);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Print the device name
+//!
+//! @param iLogMode       enum LOGBOTH, LOGCONSOLE, LOGFILE
+//! @param device         OpenCL id of the device
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclPrintDevName(int iLogMode, cl_device_id device);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the id of the first device from the context
+//!
+//! @return the id 
+//! @param cxGPUContext         OpenCL context
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_device_id oclGetFirstDev(cl_context cxGPUContext);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the id of the nth device from the context
+//!
+//! @return the id or -1 when out of range
+//! @param cxGPUContext         OpenCL context
+//! @param device_idx            index of the device of interest
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_device_id oclGetDev(cl_context cxGPUContext, unsigned int device_idx);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Gets the id of device with maximal FLOPS from the context
+//!
+//! @return the id 
+//! @param cxGPUContext         OpenCL context
+//////////////////////////////////////////////////////////////////////////////
+extern "C" cl_device_id oclGetMaxFlopsDev(cl_context cxGPUContext);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Loads a Program file and prepends the cPreamble to the code.
+//!
+//! @return the source string if succeeded, 0 otherwise
+//! @param cFilename        program filename
+//! @param cPreamble        code that is prepended to the loaded file, typically a set of #defines or a header
+//! @param szFinalLength    returned length of the code string
+//////////////////////////////////////////////////////////////////////////////
+extern "C" char* oclLoadProgSource(const char* cFilename, const char* cPreamble, size_t* szFinalLength);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get the binary (PTX) of the program associated with the device
+//!
+//! @param cpProgram    OpenCL program
+//! @param cdDevice     device of interest
+//! @param binary       returned code
+//! @param length       length of returned code
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclGetProgBinary( cl_program cpProgram, cl_device_id cdDevice, char** binary, size_t* length);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get and log the binary (PTX) from the OpenCL compiler for the requested program & device
+//!
+//! @param cpProgram                   OpenCL program
+//! @param cdDevice                    device of interest
+//! @param const char*  cPtxFileName   optional PTX file name
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclLogPtx(cl_program cpProgram, cl_device_id cdDevice, const char* cPtxFileName);
+
+//////////////////////////////////////////////////////////////////////////////
+//! Get and log the Build Log from the OpenCL compiler for the requested program & device
+//!
+//! @param cpProgram    OpenCL program
+//! @param cdDevice     device of interest
+//////////////////////////////////////////////////////////////////////////////
+extern "C" void oclLogBuildInfo(cl_program cpProgram, cl_device_id cdDevice);
+
+// Helper function for De-allocating cl objects
+// *********************************************************************
+extern "C" void oclDeleteMemObjs(cl_mem* cmMemObjs, int iNumObjs);
+
+// Helper function to get OpenCL error string from constant
+// *********************************************************************
+extern "C" const char* oclErrorString(cl_int error);
+
+// Helper function to get OpenCL image format string (channel order and type) from constant
+// *********************************************************************
+extern "C" const char* oclImageFormatString(cl_uint uiImageFormat);
+
+// companion inline function for error checking and exit on error WITH Cleanup Callback (if supplied)
+// *********************************************************************
+inline void __oclCheckErrorEX(cl_int iSample, cl_int iReference, void (*pCleanup)(int), const char* cFile, const int iLine)
+{
+    // An error condition is defined by the sample/test value not equal to the reference
+    if (iReference != iSample)
+    {
+        // If the sample/test value isn't equal to the ref, it's an error by defnition, so override 0 sample/test value
+        iSample = (iSample == 0) ? -9999 : iSample; 
+
+        // Log the error info
+        shrLog("\n !!! Error # %i (%s) at line %i , in file %s !!!\n\n", iSample, oclErrorString(iSample), iLine, cFile);
+
+        // Cleanup and exit, or just exit if no cleanup function pointer provided.  Use iSample (error code in this case) as process exit code.
+        if (pCleanup != NULL)
+        {
+            pCleanup(iSample);
+        }
+        else 
+        {
+            shrLogEx(LOGBOTH | CLOSELOG, 0, "Exiting...\n");
+            exit(iSample);
+        }
+    }
+}
+
+#endif

tests/opencl/VectorHypot/shrQATest.h

@@ -0,0 +1,238 @@
+/*
+* Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+*
+* Please refer to the NVIDIA end user license agreement (EULA) associated
+* with this source code for terms and conditions that govern your use of
+* this software. Any use, reproduction, disclosure, or distribution of
+* this software and related documentation outside the terms of the EULA
+* is strictly prohibited.
+*
+*/
+
+#ifndef SHR_QATEST_H
+#define SHR_QATEST_H
+
+// *********************************************************************
+// Generic utilities for NVIDIA GPU Computing SDK 
+// *********************************************************************
+
+// OS dependent includes
+#ifdef _WIN32
+    #pragma message ("Note: including windows.h")
+    #pragma message ("Note: including math.h")
+    #pragma message ("Note: including assert.h")
+    #pragma message ("Note: including time.h")
+
+// Headers needed for Windows
+    #include <windows.h>
+	#include <time.h>
+#else
+    // Headers needed for Linux
+    #include <sys/stat.h>
+    #include <sys/types.h>
+    #include <sys/time.h>
+    #include <stdio.h>
+    #include <stdlib.h>
+    #include <string.h>
+    #include <stdarg.h>
+    #include <unistd.h>
+    #include <time.h>
+#endif
+
+#ifndef STRCASECMP
+#ifdef _WIN32
+#define STRCASECMP _stricmp
+#else
+#define STRCASECMP strcasecmp
+#endif
+#endif
+
+#ifndef STRNCASECMP
+#ifdef _WIN32
+#define STRNCASECMP _strnicmp
+#else
+#define STRNCASECMP strncasecmp
+#endif
+#endif
+
+
+// Standardized QA Start/Finish for CUDA SDK tests
+#define shrQAStart(a, b)      __shrQAStart(a, b)
+#define shrQAFinish(a, b, c)  __shrQAFinish(a, b, c)
+#define shrQAFinish2(a, b, c, d) __shrQAFinish2(a, b, c, d)
+
+inline int findExeNameStart(const char *exec_name)
+{
+    int exename_start = (int)strlen(exec_name);
+
+    while( (exename_start > 0) && 
+            (exec_name[exename_start] != '\\') && 
+            (exec_name[exename_start] != '/') )
+    {
+        exename_start--;
+    }
+    if (exec_name[exename_start] == '\\' || 
+        exec_name[exename_start] == '/')
+    {
+        return exename_start+1;
+    } else {
+        return exename_start;
+    }
+}
+
+inline int __shrQAStart(int argc, char **argv)
+{
+    bool bQATest = false;
+    // First clear the output buffer
+    fflush(stdout);
+    fflush(stdout);
+
+    for (int i=1; i < argc; i++) {
+        int string_start = 0;
+        while (argv[i][string_start] == '-')
+           string_start++;
+        char *string_argv = &argv[i][string_start];
+
+        if (!STRCASECMP(string_argv, "qatest")) {
+           bQATest = true;
+        }
+    }
+    
+    // We don't want to print the entire path, so we search for the first 
+    int exename_start = findExeNameStart(argv[0]);
+    if (bQATest) {
+        fprintf(stdout, "&&&& RUNNING %s", &(argv[0][exename_start]));
+        for (int i=1; i < argc; i++) fprintf(stdout, " %s", argv[i]);
+        fprintf(stdout, "\n");
+    } else {
+        fprintf(stdout, "[%s] starting...\n", &(argv[0][exename_start]));
+    }
+    fflush(stdout);
+    printf("\n"); fflush(stdout);
+    return exename_start;
+}
+
+enum eQAstatus {
+    QA_FAILED = 0,
+    QA_PASSED = 1,
+    QA_WAIVED = 2
+};
+
+inline void __ExitInTime(int seconds)
+{
+    fprintf(stdout, "> exiting in %d seconds: ", seconds);
+    fflush(stdout);
+    time_t t;
+    int count;
+    for (t=time(0)+seconds, count=seconds; time(0) < t; count--) {
+        fprintf(stdout, "%d...", count);
+#ifdef WIN32
+        Sleep(1000);
+#else
+        sleep(1);
+#endif
+    }
+    fprintf(stdout,"done!\n\n"); 
+	fflush(stdout);
+}
+
+
+inline void __shrQAFinish(int argc, const char **argv, int iStatus)
+{
+    // By default QATest is disabled and NoPrompt is Enabled (times out at seconds passed into __ExitInTime() )
+    bool bQATest = false, bNoPrompt = true, bQuitInTime = true;
+    const char *sStatus[] = { "FAILED", "PASSED", "WAIVED", NULL };
+	
+    for (int i=1; i < argc; i++) {
+        int string_start = 0;
+        while (argv[i][string_start] == '-')
+           string_start++;
+
+        const char *string_argv = &argv[i][string_start];
+        if (!STRCASECMP(string_argv, "qatest")) {
+           bQATest = true;
+        }	
+        // For SDK individual samples that don't specify -noprompt or -prompt, 
+        // a 3 second delay will happen before exiting, giving a user time to view results
+        if (!STRCASECMP(string_argv, "noprompt") || !STRCASECMP(string_argv, "help")) {
+            bNoPrompt = true;
+            bQuitInTime = false;
+        }
+        if (!STRCASECMP(string_argv, "prompt")) {
+            bNoPrompt = false;
+            bQuitInTime = false;
+        }
+    }
+
+    int exename_start = findExeNameStart(argv[0]);
+    if (bQATest) {
+        fprintf(stdout, "&&&& %s %s", sStatus[iStatus], &(argv[0][exename_start]));
+        for (int i=1; i < argc; i++) fprintf(stdout, " %s", argv[i]);
+        fprintf(stdout, "\n");
+    } else {
+        fprintf(stdout, "[%s] test results...\n%s\n", &(argv[0][exename_start]), sStatus[iStatus]);
+    }
+    fflush(stdout);
+    printf("\n"); fflush(stdout);
+    if (bQuitInTime) {
+        __ExitInTime(3);
+    } else {
+        if (!bNoPrompt) {
+            fprintf(stdout, "\nPress <Enter> to exit...\n");
+            fflush(stdout);
+            getchar();
+        }
+    }
+}
+
+inline void __shrQAFinish2(bool bQATest, int argc, const char **argv, int iStatus)
+{
+    bool bQuitInTime = true;
+    const char *sStatus[] = { "FAILED", "PASSED", "WAIVED", NULL };
+	
+    for (int i=1; i < argc; i++) {
+        int string_start = 0;
+        while (argv[i][string_start] == '-')
+           string_start++;
+
+        const char *string_argv = &argv[i][string_start];
+        // For SDK individual samples that don't specify -noprompt or -prompt, 
+        // a 3 second delay will happen before exiting, giving a user time to view results
+        if (!STRCASECMP(string_argv, "noprompt") || !STRCASECMP(string_argv, "help")) {
+            bQuitInTime = false;
+        }
+        if (!STRCASECMP(string_argv, "prompt")) {
+            bQuitInTime = false;
+        }
+    }
+
+    int exename_start = findExeNameStart(argv[0]);
+    if (bQATest) {
+        fprintf(stdout, "&&&& %s %s", sStatus[iStatus], &(argv[0][exename_start]));
+        for (int i=1; i < argc; i++) fprintf(stdout, " %s", argv[i]);
+        fprintf(stdout, "\n");
+    } else {
+        fprintf(stdout, "[%s] test results...\n%s\n", &(argv[0][exename_start]), sStatus[iStatus]);
+    }
+    fflush(stdout);
+    
+    if (bQuitInTime) {
+        __ExitInTime(3);
+    }
+}
+
+inline void shrQAFinishExit(int argc, const char **argv, int iStatus)
+{
+    __shrQAFinish(argc, argv, iStatus);
+
+    exit(iStatus ? EXIT_SUCCESS : EXIT_FAILURE); 
+}
+
+inline void shrQAFinishExit2(bool bQAtest, int argc, const char **argv, int iStatus)
+{
+    __shrQAFinish2(bQAtest, argc, argv, iStatus);
+
+    exit(iStatus ? EXIT_SUCCESS : EXIT_FAILURE);
+}
+
+#endif

tests/opencl/VectorHypot/shrUtils.h

@@ -0,0 +1,642 @@
+/*
+* Copyright 1993-2010 NVIDIA Corporation.  All rights reserved.
+*
+* Please refer to the NVIDIA end user license agreement (EULA) associated
+* with this source code for terms and conditions that govern your use of
+* this software. Any use, reproduction, disclosure, or distribution of
+* this software and related documentation outside the terms of the EULA
+* is strictly prohibited.
+*
+*/
+
+#ifndef SHR_UTILS_H
+#define SHR_UTILS_H
+
+// *********************************************************************
+// Generic utilities for NVIDIA GPU Computing SDK 
+// *********************************************************************
+
+// reminders for output window and build log
+#ifdef _WIN32
+    #pragma message ("Note: including windows.h")
+    #pragma message ("Note: including math.h")
+    #pragma message ("Note: including assert.h")
+#endif
+
+// OS dependent includes
+#ifdef _WIN32
+    // Headers needed for Windows
+    #include <windows.h>
+#else
+    // Headers needed for Linux
+    #include <sys/stat.h>
+    #include <sys/types.h>
+    #include <sys/time.h>
+    #include <stdio.h>
+    #include <stdlib.h>
+    #include <string.h>
+    #include <stdarg.h>
+#endif
+
+// Other headers needed for both Windows and Linux
+#include <math.h>
+#include <assert.h>
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+
+// Un-comment the following #define to enable profiling code in SDK apps
+//#define GPU_PROFILING
+
+// Beginning of GPU Architecture definitions
+inline int ConvertSMVer2Cores(int major, int minor)
+{
+	// Defines for GPU Architecture types (using the SM version to determine the # of cores per SM
+	typedef struct {
+		int SM; // 0xMm (hexidecimal notation), M = SM Major version, and m = SM minor version
+		int Cores;
+	} sSMtoCores;
+
+	sSMtoCores nGpuArchCoresPerSM[] = 
+	{ { 0x10,  8 }, // Tesla Generation (SM 1.0) G80 class
+	  { 0x11,  8 }, // Tesla Generation (SM 1.1) G8x class
+	  { 0x12,  8 }, // Tesla Generation (SM 1.2) G9x class
+	  { 0x13,  8 }, // Tesla Generation (SM 1.3) GT200 class
+	  { 0x20, 32 }, // Fermi Generation (SM 2.0) GF100 class
+	  { 0x21, 48 }, // Fermi Generation (SM 2.1) GF10x class
+	  { 0x30, 192}, // Fermi Generation (SM 3.0) GK10x class
+	  {   -1, -1 }
+	};
+
+	int index = 0;
+	while (nGpuArchCoresPerSM[index].SM != -1) {
+		if (nGpuArchCoresPerSM[index].SM == ((major << 4) + minor) ) {
+			return nGpuArchCoresPerSM[index].Cores;
+		}
+		index++;
+	}
+	printf("MapSMtoCores SM %d.%d is undefined (please update to the latest SDK)!\n", major, minor);
+	return -1;
+}
+// end of GPU Architecture definitions
+
+
+// Defines and enum for use with logging functions
+// *********************************************************************
+#define DEFAULTLOGFILE "SdkConsoleLog.txt"
+#define MASTERLOGFILE "SdkMasterLog.csv"
+enum LOGMODES 
+{
+    LOGCONSOLE = 1, // bit to signal "log to console" 
+    LOGFILE    = 2, // bit to signal "log to file" 
+    LOGBOTH    = 3, // convenience union of first 2 bits to signal "log to both"
+    APPENDMODE = 4, // bit to set "file append" mode instead of "replace mode" on open
+    MASTER     = 8, // bit to signal master .csv log output
+    ERRORMSG   = 16, // bit to signal "pre-pend Error" 
+    CLOSELOG   = 32  // bit to close log file, if open, after any requested file write
+};
+#define HDASHLINE "-----------------------------------------------------------\n"
+
+// Standardized boolean
+enum shrBOOL
+{
+    shrFALSE = 0,
+    shrTRUE = 1
+};
+
+// Standardized MAX, MIN and CLAMP
+#define MAX(a, b) ((a > b) ? a : b)
+#define MIN(a, b) ((a < b) ? a : b)
+#define CLAMP(a, b, c) MIN(MAX(a, b), c)    // double sided clip of input a
+#define TOPCLAMP(a, b) (a < b ? a:b)	    // single top side clip of input a
+
+// Error and Exit Handling Macros... 
+// *********************************************************************
+// Full error handling macro with Cleanup() callback (if supplied)... 
+// (Companion Inline Function lower on page)
+#define shrCheckErrorEX(a, b, c) __shrCheckErrorEX(a, b, c, __FILE__ , __LINE__) 
+
+// Short version without Cleanup() callback pointer
+// Both Input (a) and Reference (b) are specified as args
+#define shrCheckError(a, b) shrCheckErrorEX(a, b, 0) 
+
+// Standardized Exit Macro for leaving main()... extended version
+// (Companion Inline Function lower on page)
+#define shrExitEX(a, b, c) __shrExitEX(a, b, c)
+
+// Standardized Exit Macro for leaving main()... short version
+// (Companion Inline Function lower on page)
+#define shrEXIT(a, b)        __shrExitEX(a, b, EXIT_SUCCESS)
+
+// Simple argument checker macro
+#define ARGCHECK(a) if((a) != shrTRUE)return shrFALSE 
+
+// Define for user-customized error handling
+#define STDERROR "file %s, line %i\n\n" , __FILE__ , __LINE__
+
+// Function to deallocate memory allocated within shrUtils
+// *********************************************************************
+extern "C" void shrFree(void* ptr);
+
+// *********************************************************************
+// Helper function to log standardized information to Console, to File or to both
+//! Examples: shrLogEx(LOGBOTH, 0, "Function A\n"); 
+//!         : shrLogEx(LOGBOTH | ERRORMSG, ciErrNum, STDERROR);
+//! 
+//! Automatically opens file and stores handle if needed and not done yet
+//! Closes file and nulls handle on request
+//! 
+//! @param 0 iLogMode: LOGCONSOLE, LOGFILE, LOGBOTH, APPENDMODE, MASTER, ERRORMSG, CLOSELOG.  
+//!          LOGFILE and LOGBOTH may be | 'd  with APPENDMODE to select file append mode instead of overwrite mode 
+//!          LOGFILE and LOGBOTH may be | 'd  with CLOSELOG to "write and close" 
+//!          First 3 options may be | 'd  with MASTER to enable independent write to master data log file
+//!          First 3 options may be | 'd  with ERRORMSG to start line with standard error message
+//! @param 2 dValue:    
+//!          Positive val = double value for time in secs to be formatted to 6 decimals. 
+//!          Negative val is an error code and this give error preformatting.
+//! @param 3 cFormatString: String with formatting specifiers like printf or fprintf.  
+//!          ALL printf flags, width, precision and type specifiers are supported with this exception: 
+//!              Wide char type specifiers intended for wprintf (%S and %C) are NOT supported
+//!              Single byte char type specifiers (%s and %c) ARE supported 
+//! @param 4... variable args: like printf or fprintf.  Must match format specifer type above.  
+//! @return 0 if OK, negative value on error or if error occurs or was passed in. 
+// *********************************************************************
+extern "C" int shrLogEx(int iLogMode, int iErrNum, const char* cFormatString, ...);
+
+// Short version of shrLogEx defaulting to shrLogEx(LOGBOTH, 0, 
+// *********************************************************************
+extern "C" int shrLog(const char* cFormatString, ...);
+
+// *********************************************************************
+// Delta timer function for up to 3 independent timers using host high performance counters 
+// Maintains state for 3 independent counters
+//! Example: double dElapsedTime = shrDeltaTime(0);
+//! 
+//! @param 0 iCounterID: Which timer to check/reset. (0, 1, 2)
+//! @return delta time of specified counter since last call in seconds.  Otherwise -9999.0 if error
+// *********************************************************************
+extern "C" double shrDeltaT(int iCounterID);
+
+// Optional LogFileNameOverride function
+// *********************************************************************
+extern "C" void shrSetLogFileName (const char* cOverRideName);
+
+// Helper function to init data arrays 
+// *********************************************************************
+extern "C" void shrFillArray(float* pfData, int iSize);
+
+// Helper function to print data arrays 
+// *********************************************************************
+extern "C" void shrPrintArray(float* pfData, int iSize);
+
+////////////////////////////////////////////////////////////////////////////
+//! Find the path for a filename
+//! @return the path if succeeded, otherwise 0
+//! @param filename        name of the file
+//! @param executablePath  optional absolute path of the executable
+////////////////////////////////////////////////////////////////////////////
+extern "C" char* shrFindFilePath(const char* filename, const char* executablePath);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing single precision floating point data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFilef( const char* filename, float** data, unsigned int* len, 
+              bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing double precision floating point data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFiled( const char* filename, double** data, unsigned int* len, 
+              bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing integer data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFilei( const char* filename, int** data, unsigned int* len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing unsigned integer data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is 
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFileui( const char* filename, unsigned int** data, 
+               unsigned int* len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing char / byte data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is 
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFileb( const char* filename, char** data, unsigned int* len, 
+              bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Read file \filename containing unsigned char / byte data
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param filename name of the source file
+//! @param data  uninitialized pointer, returned initialized and pointing to
+//!        the data read
+//! @param len  number of data elements in data, -1 on error
+//! @note If a NULL pointer is passed to this function and it is
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrReadFileub( const char* filename, unsigned char** data, 
+               unsigned int* len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing single precision floating point 
+//! data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+//! @param epsilon  epsilon for comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFilef( const char* filename, const float* data, unsigned int len,
+               const float epsilon, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing double precision floating point 
+//! data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+//! @param epsilon  epsilon for comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFiled( const char* filename, const float* data, unsigned int len,
+               const double epsilon, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing integer data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFilei( const char* filename, const int* data, unsigned int len,
+               bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing unsigned integer data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFileui( const char* filename, const unsigned int* data, 
+                unsigned int len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing char / byte data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFileb( const char* filename, const char* data, unsigned int len, 
+               bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Write a data file \filename containing unsigned char / byte data
+//! @return shrTRUE if writing the file succeeded, otherwise shrFALSE
+//! @param filename name of the file to write
+//! @param data  pointer to data to write
+//! @param len  number of data elements in data, -1 on error
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrWriteFileub( const char* filename, const unsigned char* data,
+                unsigned int len, bool verbose = false);
+
+////////////////////////////////////////////////////////////////////////////
+//! Load PPM image file (with unsigned char as data element type), padding 
+//! 4th component
+//! @return shrTRUE if reading the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param OutData  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+//! 
+//! Note: If *OutData is NULL this function allocates buffer that must be freed by caller
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrLoadPPM4ub(const char* file, unsigned char** OutData, 
+                             unsigned int *w, unsigned int *h);
+
+////////////////////////////////////////////////////////////////////////////
+//! Save PPM image file (with unsigned char as data element type, padded to 
+//! 4 bytes)
+//! @return shrTRUE if saving the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param data  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrSavePPM4ub( const char* file, unsigned char *data, 
+               unsigned int w, unsigned int h);
+
+////////////////////////////////////////////////////////////////////////////////
+//! Save PGM image file (with unsigned char as data element type)
+//! @return shrTRUE if saving the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param data  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrSavePGMub( const char* file, unsigned char *data, 
+              unsigned int w, unsigned int h); 
+
+////////////////////////////////////////////////////////////////////////////
+//! Load PGM image file (with unsigned char as data element type)
+//! @return shrTRUE if saving the file succeeded, otherwise shrFALSE
+//! @param file  name of the image file
+//! @param data  handle to the data read
+//! @param w     width of the image
+//! @param h     height of the image
+//! @note If a NULL pointer is passed to this function and it is initialized 
+//!       within shrUtils, then free() has to be used to deallocate the memory
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrLoadPGMub( const char* file, unsigned char** data,
+                  unsigned int *w,unsigned int *h);
+
+////////////////////////////////////////////////////////////////////////////
+// Command line arguments: General notes
+// * All command line arguments begin with '--' followed by the token; 
+//   token and value are seperated by '='; example --samples=50
+// * Arrays have the form --model=[one.obj,two.obj,three.obj] 
+//   (without whitespaces)
+////////////////////////////////////////////////////////////////////////////
+
+////////////////////////////////////////////////////////////////////////////
+//! Check if command line argument \a flag-name is given
+//! @return shrTRUE if command line argument \a flag_name has been given, 
+//!         otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param flag_name  name of command line flag
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCheckCmdLineFlag( const int argc, const char** argv, 
+                     const char* flag_name);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type int
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumenti( const int argc, const char** argv, 
+                        const char* arg_name, int* val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type unsigned int
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentu( const int argc, const char** argv, 
+                        const char* arg_name, unsigned int* val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type float
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentf( const int argc, const char** argv, 
+                        const char* arg_name, float* val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument of type string
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  value of the command line argument
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentstr( const int argc, const char** argv, 
+                          const char* arg_name, char** val);
+
+////////////////////////////////////////////////////////////////////////////
+//! Get the value of a command line argument list those element are strings
+//! @return shrTRUE if command line argument \a arg_name has been given and
+//!         is of the requested type, otherwise shrFALSE
+//! @param argc  argc as passed to main()
+//! @param argv  argv as passed to main()
+//! @param arg_name  name of the command line argument
+//! @param val  command line argument list
+//! @param len  length of the list / number of elements
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrGetCmdLineArgumentListstr( const int argc, const char** argv, 
+                              const char* arg_name, char** val, 
+                              unsigned int* len);
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparef( const float* reference, const float* data,
+             const unsigned int len);
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two integer arrays
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparei( const int* reference, const int* data, 
+             const unsigned int len ); 
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two unsigned integer arrays, with epsilon and threshold
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param threshold  tolerance % # of comparison errors (0.15f = 15%)
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareuit( const unsigned int* reference, const unsigned int* data,
+            const unsigned int len, const float epsilon, const float threshold );
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two unsigned char arrays
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareub( const unsigned char* reference, const unsigned char* data,
+              const unsigned int len ); 
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two integers with a tolernance for # of byte errors
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+//! @param threshold  tolerance % # of comparison errors (0.15f = 15%)
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareubt( const unsigned char* reference, const unsigned char* data,
+             const unsigned int len, const float epsilon, const float threshold );
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two integer arrays witha n epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareube( const unsigned char* reference, const unsigned char* data,
+             const unsigned int len, const float epsilon );
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays with an epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparefe( const float* reference, const float* data,
+              const unsigned int len, const float epsilon );
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays with an epsilon tolerance for equality and a 
+//!     threshold for # pixel errors
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparefet( const float* reference, const float* data,
+             const unsigned int len, const float epsilon, const float threshold );
+
+////////////////////////////////////////////////////////////////////////////
+//! Compare two float arrays using L2-norm with an epsilon tolerance for 
+//! equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param reference  handle to the reference data / gold image
+//! @param data       handle to the computed data
+//! @param len        number of elements in reference and data
+//! @param epsilon    epsilon to use for the comparison
+////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrCompareL2fe( const float* reference, const float* data,
+                const unsigned int len, const float epsilon );
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two PPM image files with an epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param src_file   filename for the image to be compared
+//! @param data       filename for the reference data / gold image
+//! @param epsilon    epsilon to use for the comparison
+//! @param threshold  threshold of pixels that can still mismatch to pass (i.e. 0.15f = 15% must pass)
+//! $param verboseErrors output details of image mismatch to std::err
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparePPM( const char *src_file, const char *ref_file, const float epsilon, const float threshold);
+
+////////////////////////////////////////////////////////////////////////////////
+//! Compare two PGM image files with an epsilon tolerance for equality
+//! @return shrTRUEif \a reference and \a data are identical, otherwise shrFALSE
+//! @param src_file   filename for the image to be compared
+//! @param data       filename for the reference data / gold image
+//! @param epsilon    epsilon to use for the comparison
+//! @param threshold  threshold of pixels that can still mismatch to pass (i.e. 0.15f = 15% must pass)
+//! $param verboseErrors output details of image mismatch to std::err
+////////////////////////////////////////////////////////////////////////////////
+extern "C" shrBOOL shrComparePGM( const char *src_file, const char *ref_file, const float epsilon, const float threshold);
+
+extern "C" unsigned char* shrLoadRawFile(const char* filename, size_t size);
+
+extern "C" size_t shrRoundUp(int group_size, int global_size);
+
+// companion inline function for error checking and exit on error WITH Cleanup Callback (if supplied)
+// *********************************************************************
+inline void __shrCheckErrorEX(int iSample, int iReference, void (*pCleanup)(int), const char* cFile, const int iLine)
+{
+    if (iReference != iSample)
+    {
+        shrLogEx(LOGBOTH | ERRORMSG, iSample, "line %i , in file %s !!!\n\n" , iLine, cFile); 
+        if (pCleanup != NULL)
+        {
+            pCleanup(EXIT_FAILURE);
+        }
+        else 
+        {
+            shrLogEx(LOGBOTH | CLOSELOG, 0, "Exiting...\n");
+            exit(EXIT_FAILURE);
+        }
+    }
+}
+
+// Standardized Exit
+// *********************************************************************
+inline void __shrExitEX(int argc, const char** argv, int iExitCode)
+{
+#ifdef WIN32
+    if (!shrCheckCmdLineFlag(argc, argv, "noprompt") && !shrCheckCmdLineFlag(argc, argv, "qatest")) 
+#else 
+    if (shrCheckCmdLineFlag(argc, argv, "prompt") && !shrCheckCmdLineFlag(argc, argv, "qatest")) 
+#endif
+    {
+        shrLogEx(LOGBOTH | CLOSELOG, 0, "\nPress <Enter> to Quit...\n");                  
+        getchar();                                                           
+    }       
+    else 
+    {
+        shrLogEx(LOGBOTH | CLOSELOG, 0, "%s Exiting...\n", argv[0]); 
+    }
+    fflush(stderr);                                                         
+    exit(iExitCode);
+}
+
+#endif

tests/opencl/bfs/CLHelper.h

@@ -0,0 +1,859 @@
+//------------------------------------------
+//--cambine:helper function for OpenCL
+//--programmer:	Jianbin Fang
+//--date:	27/12/2010
+//------------------------------------------
+#ifndef _CL_HELPER_
+#define _CL_HELPER_
+
+#include <CL/cl.h>
+#include <fstream>
+#include <iostream>
+#include <string>
+#include <vector>
+
+
+using std::string;
+using std::ifstream;
+using std::cerr;
+using std::endl;
+using std::cout;
+//#pragma OPENCL EXTENSION cl_nv_compiler_options:enable
+#define WORK_DIM 2 // work-items dimensions
+
+struct oclHandleStruct {
+  cl_context context;
+  cl_device_id *devices;
+  cl_command_queue queue;
+  cl_program program;
+  cl_int cl_status;
+  std::string error_str;
+  std::vector<cl_kernel> kernel;
+};
+
+struct oclHandleStruct oclHandles;
+
+char kernel_file[100] = "Kernels.cl";
+int total_kernels = 2;
+string kernel_names[2] = {"BFS_1", "BFS_2"};
+int work_group_size = 512;
+int device_id_inused = 0; // deviced id used (default : 0)
+
+int read_kernel_file(const char* filename, uint8_t** data, size_t* size) {
+  if (nullptr == filename || nullptr == data || 0 == size)
+    return -1;
+
+  FILE* fp = fopen(filename, "r");
+  if (NULL == fp) {
+    fprintf(stderr, "Failed to load kernel.");
+    return -1;
+  }
+  fseek(fp , 0 , SEEK_END);
+  long fsize = ftell(fp);
+  rewind(fp);
+
+  *data = (uint8_t*)malloc(fsize);
+  *size = fread(*data, 1, fsize, fp);
+  
+  fclose(fp);
+  
+  return 0;
+}
+
+/*
+ * Converts the contents of a file into a string
+ */
+string FileToString(const string fileName) {
+  ifstream f(fileName.c_str(), ifstream::in | ifstream::binary);
+
+  try {
+    size_t size;
+    char *str;
+    string s;
+
+    if (f.is_open()) {
+      size_t fileSize;
+      f.seekg(0, ifstream::end);
+      size = fileSize = f.tellg();
+      f.seekg(0, ifstream::beg);
+
+      str = new char[size + 1];
+      if (!str)
+        throw(string("Could not allocate memory"));
+
+      f.read(str, fileSize);
+      f.close();
+      str[size] = '\0';
+
+      s = str;
+      delete[] str;
+      return s;
+    }
+  } catch (std::string msg) {
+    cerr << "Exception caught in FileToString(): " << msg << endl;
+    if (f.is_open())
+      f.close();
+  } catch (...) {
+    cerr << "Exception caught in FileToString()" << endl;
+    if (f.is_open())
+      f.close();
+  }
+  string errorMsg = "FileToString()::Error: Unable to open file " + fileName;
+  throw(errorMsg);
+}
+//---------------------------------------
+// Read command line parameters
+//
+void _clCmdParams(int argc, char *argv[]) {
+  for (int i = 0; i < argc; ++i) {
+    switch (argv[i][1]) {
+    case 'g': //--g stands for size of work group
+      if (++i < argc) {
+        sscanf(argv[i], "%u", &work_group_size);
+      } else {
+        std::cerr << "Could not read argument after option " << argv[i - 1]
+                  << std::endl;
+        throw;
+      }
+      break;
+    case 'd': //--d stands for device id used in computaion
+      if (++i < argc) {
+        sscanf(argv[i], "%u", &device_id_inused);
+      } else {
+        std::cerr << "Could not read argument after option " << argv[i - 1]
+                  << std::endl;
+        throw;
+      }
+      break;
+    default:;
+    }
+  }
+}
+
+//---------------------------------------
+// Initlize CL objects
+//--description: there are 5 steps to initialize all the OpenCL objects needed
+//--revised on 04/01/2011: get the number of devices  and
+//  devices have no relationship with context
+void _clInit() {
+  printf("_clInit()\n");
+
+  int DEVICE_ID_INUSED = device_id_inused;
+  cl_int resultCL;
+
+  oclHandles.context = NULL;
+  oclHandles.devices = NULL;
+  oclHandles.queue = NULL;
+  oclHandles.program = NULL;
+
+  cl_uint deviceListSize;
+
+  //-----------------------------------------------
+  //--cambine-1: find the available platforms and select one
+
+  cl_uint numPlatforms = 1;
+  cl_platform_id targetPlatform = NULL;
+
+  cl_platform_id *allPlatforms =
+      (cl_platform_id *)malloc(numPlatforms * sizeof(cl_platform_id));
+
+  resultCL = clGetPlatformIDs(numPlatforms, allPlatforms, NULL);
+  if (resultCL != CL_SUCCESS)
+    throw(string("InitCL()::Error: Getting platform ids (clGetPlatformIDs)"));
+
+  // Select the target platform. Default: first platform
+  targetPlatform = allPlatforms[0];
+
+  /*for (int i = 0; i < numPlatforms; i++)
+{
+char pbuff[128];
+resultCL = clGetPlatformInfo( allPlatforms[i],
+                          CL_PLATFORM_VENDOR,
+                          sizeof(pbuff),
+                          pbuff,
+                          NULL);
+if (resultCL != CL_SUCCESS)
+throw (string("InitCL()::Error: Getting platform info (clGetPlatformInfo)"));
+
+  //printf("vedor is %s\n",pbuff);
+
+}
+free(allPlatforms);*/
+
+  //-----------------------------------------------
+  //--cambine-2: create an OpenCL context
+  /*cl_context_properties cprops[3] = { CL_CONTEXT_PLATFORM,
+ (cl_context_properties)targetPlatform, 0 };
+  oclHandles.context = clCreateContextFromType(cprops,
+                                              CL_DEVICE_TYPE_GPU,
+                                              NULL,
+                                              NULL,
+                                              &resultCL);
+
+  if ((resultCL != CL_SUCCESS) || (oclHandles.context == NULL))
+      throw (string("InitCL()::Error: Creating Context
+ (clCreateContextFromType)"));
+
+        //-----------------------------------------------
+  //--cambine-3: detect OpenCL devices
+  // First, get the size of device list
+ oclHandles.cl_status = clGetDeviceIDs(targetPlatform, CL_DEVICE_TYPE_GPU, 0,
+ NULL, &deviceListSize);
+ if(oclHandles.cl_status!=CL_SUCCESS){
+      throw(string("exception in _clInit -> clGetDeviceIDs"));
+ }
+ if (deviceListSize == 0)
+      throw(string("InitCL()::Error: No devices found."));
+
+              printf("OK1()\n");
+
+  //std::cout<<"device number:"<<deviceListSize<<std::endl;*/
+
+  // Now, allocate the device list
+  deviceListSize = 1;
+  oclHandles.devices =
+      (cl_device_id *)malloc(deviceListSize * sizeof(cl_device_id));
+  if (oclHandles.devices == 0)
+    throw(string("InitCL()::Error: Could not allocate memory."));
+
+  //* Next, get the device list data
+  oclHandles.cl_status =
+      clGetDeviceIDs(targetPlatform, CL_DEVICE_TYPE_DEFAULT, deviceListSize,
+                     oclHandles.devices, NULL);
+  if (oclHandles.cl_status != CL_SUCCESS) {
+    throw(string("exception in _clInit -> clGetDeviceIDs-2"));
+  }
+
+  oclHandles.context = clCreateContext(NULL, deviceListSize, oclHandles.devices,
+                                       NULL, NULL, &resultCL);
+  if ((resultCL != CL_SUCCESS) || (oclHandles.context == NULL))
+    throw(string("InitCL()::Error: Creating Context (clCreateContext)"));
+
+  //-----------------------------------------------
+  //--cambine-4: Create an OpenCL command queue
+  oclHandles.queue = clCreateCommandQueue(
+      oclHandles.context, oclHandles.devices[DEVICE_ID_INUSED], 0, &resultCL);
+  //printf("resultCL=%d, queue=0x%x\n", resultCL, oclHandles.queue);
+
+  if ((resultCL != CL_SUCCESS) || (oclHandles.queue == NULL))
+    throw(string("InitCL()::Creating Command Queue. (clCreateCommandQueue)"));
+  //-----------------------------------------------
+  //--cambine-5: Load CL file, build CL program object, create CL kernel object
+  /*std::string  source_str = FileToString(kernel_file);
+  const char * source    = source_str.c_str();
+  size_t sourceSize[]    = { source_str.length() };*/
+
+  //oclHandles.program = clCreateProgramWithBuiltInKernels(
+  //    oclHandles.context, 1, &oclHandles.devices[DEVICE_ID_INUSED],
+  //   "BFS_1;BFS_2", &resultCL);
+  /*oclHandles.program = clCreateProgramWithSource(oclHandles.context,
+                                                  1,
+                                                  &source,
+                                                  sourceSize,
+                                                  &resultCL);*/
+  // read kernel binary from file  
+  uint8_t *kernel_bin = NULL;
+  size_t kernel_size;
+  cl_int binary_status = 0;
+  if (0 != read_kernel_file("kernel.pocl", &kernel_bin, &kernel_size))
+    std::abort();
+
+ oclHandles.program = clCreateProgramWithBinary(
+    oclHandles.context, 1, &oclHandles.devices[DEVICE_ID_INUSED], &kernel_size, (const uint8_t**)&kernel_bin, &binary_status, &resultCL);
+  free(kernel_bin);
+
+  if ((resultCL != CL_SUCCESS) || (oclHandles.program == NULL))
+    throw(string("InitCL()::Error: Loading Binary into cl_program. "
+                 "(clCreateProgramWithBinary)"));
+
+  // insert debug information
+  // std::string options= "-cl-nv-verbose"; //Doesn't work on AMD machines
+  // options += " -cl-nv-opt-level=3";
+  resultCL = clBuildProgram(oclHandles.program, deviceListSize,
+                            oclHandles.devices, NULL, NULL, NULL);
+  if ((resultCL != CL_SUCCESS) || (oclHandles.program == NULL)) {
+    cerr << "InitCL()::Error: In clBuildProgram" << endl;
+
+    size_t length;
+    resultCL = clGetProgramBuildInfo(oclHandles.program,
+                                     oclHandles.devices[DEVICE_ID_INUSED],
+                                     CL_PROGRAM_BUILD_LOG, 0, NULL, &length);
+    if (resultCL != CL_SUCCESS)
+      throw(string("InitCL()::Error: Getting Program build "
+                   "info(clGetProgramBuildInfo)"));
+
+    char *buffer = (char *)malloc(length);
+    resultCL = clGetProgramBuildInfo(
+        oclHandles.program, oclHandles.devices[DEVICE_ID_INUSED],
+        CL_PROGRAM_BUILD_LOG, length, buffer, NULL);
+    if (resultCL != CL_SUCCESS)
+      throw(string("InitCL()::Error: Getting Program build "
+                   "info(clGetProgramBuildInfo)"));
+
+    cerr << buffer << endl;
+    free(buffer);
+
+    throw(string("InitCL()::Error: Building Program (clBuildProgram)"));
+  }
+
+// get program information in intermediate representation
+#ifdef PTX_MSG
+  size_t binary_sizes[deviceListSize];
+  char *binaries[deviceListSize];
+  // figure out number of devices and the sizes of the binary for each device.
+  oclHandles.cl_status =
+      clGetProgramInfo(oclHandles.program, CL_PROGRAM_BINARY_SIZES,
+                       sizeof(size_t) * deviceListSize, &binary_sizes, NULL);
+  if (oclHandles.cl_status != CL_SUCCESS) {
+    throw(string("--cambine:exception in _InitCL -> clGetProgramInfo-2"));
+  }
+
+  std::cout << "--cambine:" << binary_sizes << std::endl;
+  // copy over all of the generated binaries.
+  for (int i = 0; i < deviceListSize; i++)
+    binaries[i] = (char *)malloc(sizeof(char) * (binary_sizes[i] + 1));
+  oclHandles.cl_status =
+      clGetProgramInfo(oclHandles.program, CL_PROGRAM_BINARIES,
+                       sizeof(char *) * deviceListSize, binaries, NULL);
+  if (oclHandles.cl_status != CL_SUCCESS) {
+    throw(string("--cambine:exception in _InitCL -> clGetProgramInfo-3"));
+  }
+  for (int i = 0; i < deviceListSize; i++)
+    binaries[i][binary_sizes[i]] = '\0';
+  std::cout << "--cambine:writing ptd information..." << std::endl;
+  FILE *ptx_file = fopen("cl.ptx", "w");
+  if (ptx_file == NULL) {
+    throw(string("exceptions in allocate ptx file."));
+  }
+  fprintf(ptx_file, "%s", binaries[DEVICE_ID_INUSED]);
+  fclose(ptx_file);
+  std::cout << "--cambine:writing ptd information done." << std::endl;
+  for (int i = 0; i < deviceListSize; i++)
+    free(binaries[i]);
+#endif
+
+  for (int nKernel = 0; nKernel < total_kernels; nKernel++) {
+    /* get a kernel object handle for a kernel with the given name */
+    cl_kernel kernel = clCreateKernel(
+        oclHandles.program, (kernel_names[nKernel]).c_str(), &resultCL);
+
+    if ((resultCL != CL_SUCCESS) || (kernel == NULL)) {
+      string errorMsg = "InitCL()::Error: Creating Kernel (clCreateKernel) \"" +
+                        kernel_names[nKernel] + "\"";
+      throw(errorMsg);
+    }
+
+    oclHandles.kernel.push_back(kernel);
+  }
+// get resource alocation information
+#ifdef RES_MSG
+  char *build_log;
+  size_t ret_val_size;
+  oclHandles.cl_status = clGetProgramBuildInfo(
+      oclHandles.program, oclHandles.devices[DEVICE_ID_INUSED],
+      CL_PROGRAM_BUILD_LOG, 0, NULL, &ret_val_size);
+  if (oclHandles.cl_status != CL_SUCCESS) {
+    throw(string("exceptions in _InitCL -> getting resource information"));
+  }
+
+  build_log = (char *)malloc(ret_val_size + 1);
+  oclHandles.cl_status = clGetProgramBuildInfo(
+      oclHandles.program, oclHandles.devices[DEVICE_ID_INUSED],
+      CL_PROGRAM_BUILD_LOG, ret_val_size, build_log, NULL);
+  if (oclHandles.cl_status != CL_SUCCESS) {
+    throw(string(
+        "exceptions in _InitCL -> getting resources allocation information-2"));
+  }
+  build_log[ret_val_size] = '\0';
+  std::cout << "--cambine:" << build_log << std::endl;
+  free(build_log);
+#endif
+}
+
+//---------------------------------------
+// release CL objects
+void _clRelease() {
+  char errorFlag = false;
+
+  for (int nKernel = 0; nKernel < oclHandles.kernel.size(); nKernel++) {
+    if (oclHandles.kernel[nKernel] != NULL) {
+      cl_int resultCL = clReleaseKernel(oclHandles.kernel[nKernel]);
+      if (resultCL != CL_SUCCESS) {
+        cerr << "ReleaseCL()::Error: In clReleaseKernel" << endl;
+        errorFlag = true;
+      }
+      oclHandles.kernel[nKernel] = NULL;
+      printf("clReleaseKernel()\n");
+    }
+  }
+
+  if (oclHandles.program != NULL) {
+    cl_int resultCL = clReleaseProgram(oclHandles.program);
+    if (resultCL != CL_SUCCESS) {
+      cerr << "ReleaseCL()::Error: In clReleaseProgram" << endl;
+      errorFlag = true;
+    }
+    oclHandles.program = NULL;
+    printf("clReleaseProgram()\n");
+  }
+
+  if (oclHandles.queue != NULL) {
+    cl_int resultCL = clReleaseCommandQueue(oclHandles.queue);
+    if (resultCL != CL_SUCCESS) {
+      cerr << "ReleaseCL()::Error: In clReleaseCommandQueue" << endl;
+      errorFlag = true;
+    }
+    oclHandles.queue = NULL;
+    printf("clReleaseCommandQueue()\n");
+  }
+
+  if (oclHandles.context != NULL) {
+    cl_int resultCL = clReleaseContext(oclHandles.context);
+    if (resultCL != CL_SUCCESS) {
+      cerr << "ReleaseCL()::Error: In clReleaseContext" << endl;
+      errorFlag = true;
+    }
+    oclHandles.context = NULL;
+    printf("clReleaseContext()\n");
+  }
+
+  if (oclHandles.devices != NULL) {
+    cl_int resultCL = clReleaseDevice(oclHandles.devices[0]);
+    if (resultCL != CL_SUCCESS) {
+      cerr << "ReleaseCL()::Error: In clReleaseDevice" << endl;
+      errorFlag = true;
+    }
+    free(oclHandles.devices);
+    printf("clReleaseDevice()\n");
+  }
+
+  if (errorFlag)
+    throw(string("ReleaseCL()::Error encountered."));
+}
+//--------------------------------------------------------
+//--cambine:create buffer and then copy data from host to device
+cl_mem _clCreateAndCpyMem(int size, void *h_mem_source) throw(string) {
+  cl_mem d_mem;
+  d_mem = clCreateBuffer(oclHandles.context,
+                         CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, size,
+                         h_mem_source, &oclHandles.cl_status);
+#ifdef ERRMSG
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(string("excpetion in _clCreateAndCpyMem()"));
+#endif
+  return d_mem;
+}
+//-------------------------------------------------------
+//--cambine:	create read only  buffer for devices
+//--date:	17/01/2011
+cl_mem _clMallocRW(int size, void *h_mem_ptr) throw(string) {
+  cl_mem d_mem;
+  d_mem = clCreateBuffer(oclHandles.context,
+                         CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, size,
+                         h_mem_ptr, &oclHandles.cl_status);
+#ifdef ERRMSG
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(string("excpetion in _clMallocRW"));
+#endif
+  return d_mem;
+}
+//-------------------------------------------------------
+//--cambine:	create read and write buffer for devices
+//--date:	17/01/2011
+cl_mem _clMalloc(int size, void *h_mem_ptr) throw(string) {
+  cl_mem d_mem;
+  d_mem = clCreateBuffer(oclHandles.context,
+                         CL_MEM_WRITE_ONLY | CL_MEM_COPY_HOST_PTR, size,
+                         h_mem_ptr, &oclHandles.cl_status);
+#ifdef ERRMSG
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(string("excpetion in _clMalloc"));
+#endif
+  return d_mem;
+}
+
+//-------------------------------------------------------
+//--cambine:	transfer data from host to device
+//--date:	17/01/2011
+void _clMemcpyH2D(cl_mem d_mem, int size, const void *h_mem_ptr) throw(string) {
+  oclHandles.cl_status = clEnqueueWriteBuffer(
+      oclHandles.queue, d_mem, CL_TRUE, 0, size, h_mem_ptr, 0, NULL, NULL);
+#ifdef ERRMSG
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(string("excpetion in _clMemcpyH2D"));
+#endif
+}
+//--------------------------------------------------------
+//--cambine:create buffer and then copy data from host to device with pinned
+// memory
+cl_mem _clCreateAndCpyPinnedMem(int size, float *h_mem_source) throw(string) {
+  cl_mem d_mem, d_mem_pinned;
+  float *h_mem_pinned = NULL;
+  d_mem_pinned = clCreateBuffer(oclHandles.context,
+                                CL_MEM_READ_ONLY | CL_MEM_ALLOC_HOST_PTR, size,
+                                NULL, &oclHandles.cl_status);
+#ifdef ERRMSG
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(string("excpetion in _clCreateAndCpyMem()->d_mem_pinned"));
+#endif
+  //------------
+  d_mem = clCreateBuffer(oclHandles.context, CL_MEM_READ_ONLY, size, NULL,
+                         &oclHandles.cl_status);
+#ifdef ERRMSG
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(string("excpetion in _clCreateAndCpyMem() -> d_mem "));
+#endif
+  //----------
+  h_mem_pinned = (cl_float *)clEnqueueMapBuffer(
+      oclHandles.queue, d_mem_pinned, CL_TRUE, CL_MAP_WRITE, 0, size, 0, NULL,
+      NULL, &oclHandles.cl_status);
+#ifdef ERRMSG
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(string("excpetion in _clCreateAndCpyMem() -> clEnqueueMapBuffer"));
+#endif
+  int element_number = size / sizeof(float);
+#pragma omp parallel for
+  for (int i = 0; i < element_number; i++) {
+    h_mem_pinned[i] = h_mem_source[i];
+  }
+  //----------
+  oclHandles.cl_status = clEnqueueWriteBuffer(
+      oclHandles.queue, d_mem, CL_TRUE, 0, size, h_mem_pinned, 0, NULL, NULL);
+#ifdef ERRMSG
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(string("excpetion in _clCreateAndCpyMem() -> clEnqueueWriteBuffer"));
+#endif
+
+  return d_mem;
+}
+
+//--------------------------------------------------------
+//--cambine:create write only buffer on device
+cl_mem _clMallocWO(int size) throw(string) {
+  cl_mem d_mem;
+  d_mem = clCreateBuffer(oclHandles.context, CL_MEM_WRITE_ONLY, size, 0,
+                         &oclHandles.cl_status);
+#ifdef ERRMSG
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(string("excpetion in _clCreateMem()"));
+#endif
+  return d_mem;
+}
+
+//--------------------------------------------------------
+// transfer data from device to host
+void _clMemcpyD2H(cl_mem d_mem, int size, void *h_mem) throw(string) {
+  oclHandles.cl_status = clEnqueueReadBuffer(oclHandles.queue, d_mem, CL_TRUE,
+                                             0, size, h_mem, 0, 0, 0);
+#ifdef ERRMSG
+  oclHandles.error_str = "excpetion in _clCpyMemD2H -> ";
+  switch (oclHandles.cl_status) {
+  case CL_INVALID_COMMAND_QUEUE:
+    oclHandles.error_str += "CL_INVALID_COMMAND_QUEUE";
+    break;
+  case CL_INVALID_CONTEXT:
+    oclHandles.error_str += "CL_INVALID_CONTEXT";
+    break;
+  case CL_INVALID_MEM_OBJECT:
+    oclHandles.error_str += "CL_INVALID_MEM_OBJECT";
+    break;
+  case CL_INVALID_VALUE:
+    oclHandles.error_str += "CL_INVALID_VALUE";
+    break;
+  case CL_INVALID_EVENT_WAIT_LIST:
+    oclHandles.error_str += "CL_INVALID_EVENT_WAIT_LIST";
+    break;
+  case CL_MEM_OBJECT_ALLOCATION_FAILURE:
+    oclHandles.error_str += "CL_MEM_OBJECT_ALLOCATION_FAILURE";
+    break;
+  case CL_OUT_OF_HOST_MEMORY:
+    oclHandles.error_str += "CL_OUT_OF_HOST_MEMORY";
+    break;
+  default:
+    oclHandles.error_str += "Unknown reason";
+    break;
+  }
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(oclHandles.error_str);
+#endif
+}
+
+//--------------------------------------------------------
+// set kernel arguments
+void _clSetArgs(int kernel_id, int arg_idx, void *d_mem,
+                int size = 0) throw(string) {
+  if (!size) {
+    oclHandles.cl_status = clSetKernelArg(oclHandles.kernel[kernel_id], arg_idx,
+                                          sizeof(d_mem), &d_mem);
+#ifdef ERRMSG
+    oclHandles.error_str = "excpetion in _clSetKernelArg() ";
+    switch (oclHandles.cl_status) {
+    case CL_INVALID_KERNEL:
+      oclHandles.error_str += "CL_INVALID_KERNEL";
+      break;
+    case CL_INVALID_ARG_INDEX:
+      oclHandles.error_str += "CL_INVALID_ARG_INDEX";
+      break;
+    case CL_INVALID_ARG_VALUE:
+      oclHandles.error_str += "CL_INVALID_ARG_VALUE";
+      break;
+    case CL_INVALID_MEM_OBJECT:
+      oclHandles.error_str += "CL_INVALID_MEM_OBJECT";
+      break;
+    case CL_INVALID_SAMPLER:
+      oclHandles.error_str += "CL_INVALID_SAMPLER";
+      break;
+    case CL_INVALID_ARG_SIZE:
+      oclHandles.error_str += "CL_INVALID_ARG_SIZE";
+      break;
+    case CL_OUT_OF_RESOURCES:
+      oclHandles.error_str += "CL_OUT_OF_RESOURCES";
+      break;
+    case CL_OUT_OF_HOST_MEMORY:
+      oclHandles.error_str += "CL_OUT_OF_HOST_MEMORY";
+      break;
+    default:
+      oclHandles.error_str += "Unknown reason";
+      break;
+    }
+    if (oclHandles.cl_status != CL_SUCCESS)
+      throw(oclHandles.error_str);
+#endif
+  } else {
+    oclHandles.cl_status =
+        clSetKernelArg(oclHandles.kernel[kernel_id], arg_idx, size, d_mem);
+#ifdef ERRMSG
+    oclHandles.error_str = "excpetion in _clSetKernelArg() ";
+    switch (oclHandles.cl_status) {
+    case CL_INVALID_KERNEL:
+      oclHandles.error_str += "CL_INVALID_KERNEL";
+      break;
+    case CL_INVALID_ARG_INDEX:
+      oclHandles.error_str += "CL_INVALID_ARG_INDEX";
+      break;
+    case CL_INVALID_ARG_VALUE:
+      oclHandles.error_str += "CL_INVALID_ARG_VALUE";
+      break;
+    case CL_INVALID_MEM_OBJECT:
+      oclHandles.error_str += "CL_INVALID_MEM_OBJECT";
+      break;
+    case CL_INVALID_SAMPLER:
+      oclHandles.error_str += "CL_INVALID_SAMPLER";
+      break;
+    case CL_INVALID_ARG_SIZE:
+      oclHandles.error_str += "CL_INVALID_ARG_SIZE";
+      break;
+    case CL_OUT_OF_RESOURCES:
+      oclHandles.error_str += "CL_OUT_OF_RESOURCES";
+      break;
+    case CL_OUT_OF_HOST_MEMORY:
+      oclHandles.error_str += "CL_OUT_OF_HOST_MEMORY";
+      break;
+    default:
+      oclHandles.error_str += "Unknown reason";
+      break;
+    }
+    if (oclHandles.cl_status != CL_SUCCESS)
+      throw(oclHandles.error_str);
+#endif
+  }
+}
+void _clFinish() throw(string) {
+  oclHandles.cl_status = clFinish(oclHandles.queue);
+#ifdef ERRMSG
+  oclHandles.error_str = "excpetion in _clFinish";
+  switch (oclHandles.cl_status) {
+  case CL_INVALID_COMMAND_QUEUE:
+    oclHandles.error_str += "CL_INVALID_COMMAND_QUEUE";
+    break;
+  case CL_OUT_OF_RESOURCES:
+    oclHandles.error_str += "CL_OUT_OF_RESOURCES";
+    break;
+  case CL_OUT_OF_HOST_MEMORY:
+    oclHandles.error_str += "CL_OUT_OF_HOST_MEMORY";
+    break;
+  default:
+    oclHandles.error_str += "Unknown reasons";
+    break;
+  }
+  if (oclHandles.cl_status != CL_SUCCESS) {
+    throw(oclHandles.error_str);
+  }
+#endif
+}
+//--------------------------------------------------------
+//--cambine:enqueue kernel
+void _clInvokeKernel(int kernel_id, int work_items,
+                     int work_group_size) throw(string) {
+  cl_uint work_dim = WORK_DIM;
+  //cl_event e[1];
+  if (work_items % work_group_size != 0) // process situations that work_items
+                                         // cannot be divided by work_group_size
+    work_items =
+        work_items + (work_group_size - (work_items % work_group_size));
+  size_t local_work_size[] = {work_group_size, 1};
+  size_t global_work_size[] = {work_items, 1};
+  oclHandles.cl_status = clEnqueueNDRangeKernel(
+      oclHandles.queue, oclHandles.kernel[kernel_id], work_dim, 0,
+      global_work_size, local_work_size, 0, 0, NULL);
+#ifdef ERRMSG
+  oclHandles.error_str = "excpetion in _clInvokeKernel() -> ";
+  switch (oclHandles.cl_status) {
+  case CL_INVALID_PROGRAM_EXECUTABLE:
+    oclHandles.error_str += "CL_INVALID_PROGRAM_EXECUTABLE";
+    break;
+  case CL_INVALID_COMMAND_QUEUE:
+    oclHandles.error_str += "CL_INVALID_COMMAND_QUEUE";
+    break;
+  case CL_INVALID_KERNEL:
+    oclHandles.error_str += "CL_INVALID_KERNEL";
+    break;
+  case CL_INVALID_CONTEXT:
+    oclHandles.error_str += "CL_INVALID_CONTEXT";
+    break;
+  case CL_INVALID_KERNEL_ARGS:
+    oclHandles.error_str += "CL_INVALID_KERNEL_ARGS";
+    break;
+  case CL_INVALID_WORK_DIMENSION:
+    oclHandles.error_str += "CL_INVALID_WORK_DIMENSION";
+    break;
+  case CL_INVALID_GLOBAL_WORK_SIZE:
+    oclHandles.error_str += "CL_INVALID_GLOBAL_WORK_SIZE";
+    break;
+  case CL_INVALID_WORK_GROUP_SIZE:
+    oclHandles.error_str += "CL_INVALID_WORK_GROUP_SIZE";
+    break;
+  case CL_INVALID_WORK_ITEM_SIZE:
+    oclHandles.error_str += "CL_INVALID_WORK_ITEM_SIZE";
+    break;
+  case CL_INVALID_GLOBAL_OFFSET:
+    oclHandles.error_str += "CL_INVALID_GLOBAL_OFFSET";
+    break;
+  case CL_OUT_OF_RESOURCES:
+    oclHandles.error_str += "CL_OUT_OF_RESOURCES";
+    break;
+  case CL_MEM_OBJECT_ALLOCATION_FAILURE:
+    oclHandles.error_str += "CL_MEM_OBJECT_ALLOCATION_FAILURE";
+    break;
+  case CL_INVALID_EVENT_WAIT_LIST:
+    oclHandles.error_str += "CL_INVALID_EVENT_WAIT_LIST";
+    break;
+  case CL_OUT_OF_HOST_MEMORY:
+    oclHandles.error_str += "CL_OUT_OF_HOST_MEMORY";
+    break;
+  default:
+    oclHandles.error_str += "Unkown reseason";
+    break;
+  }
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(oclHandles.error_str);
+#endif
+  //_clFinish();
+  // oclHandles.cl_status = clWaitForEvents(1, &e[0]);
+  // #ifdef ERRMSG
+  // if (oclHandles.cl_status!= CL_SUCCESS)
+  //     throw(string("excpetion in _clEnqueueNDRange() -> clWaitForEvents"));
+  // #endif
+}
+void _clInvokeKernel2D(int kernel_id, int range_x, int range_y, int group_x,
+                       int group_y) throw(string) {
+  cl_uint work_dim = WORK_DIM;
+  size_t local_work_size[] = {group_x, group_y};
+  size_t global_work_size[] = {range_x, range_y};
+  //cl_event e[1];
+  /*if(work_items%work_group_size != 0)	//process situations that work_items
+    cannot be divided by work_group_size
+    work_items = work_items + (work_group_size-(work_items%work_group_size));*/
+  oclHandles.cl_status = clEnqueueNDRangeKernel(
+      oclHandles.queue, oclHandles.kernel[kernel_id], work_dim, 0,
+      global_work_size, local_work_size, 0, 0, NULL);
+#ifdef ERRMSG
+  oclHandles.error_str = "excpetion in _clInvokeKernel() -> ";
+  switch (oclHandles.cl_status) {
+  case CL_INVALID_PROGRAM_EXECUTABLE:
+    oclHandles.error_str += "CL_INVALID_PROGRAM_EXECUTABLE";
+    break;
+  case CL_INVALID_COMMAND_QUEUE:
+    oclHandles.error_str += "CL_INVALID_COMMAND_QUEUE";
+    break;
+  case CL_INVALID_KERNEL:
+    oclHandles.error_str += "CL_INVALID_KERNEL";
+    break;
+  case CL_INVALID_CONTEXT:
+    oclHandles.error_str += "CL_INVALID_CONTEXT";
+    break;
+  case CL_INVALID_KERNEL_ARGS:
+    oclHandles.error_str += "CL_INVALID_KERNEL_ARGS";
+    break;
+  case CL_INVALID_WORK_DIMENSION:
+    oclHandles.error_str += "CL_INVALID_WORK_DIMENSION";
+    break;
+  case CL_INVALID_GLOBAL_WORK_SIZE:
+    oclHandles.error_str += "CL_INVALID_GLOBAL_WORK_SIZE";
+    break;
+  case CL_INVALID_WORK_GROUP_SIZE:
+    oclHandles.error_str += "CL_INVALID_WORK_GROUP_SIZE";
+    break;
+  case CL_INVALID_WORK_ITEM_SIZE:
+    oclHandles.error_str += "CL_INVALID_WORK_ITEM_SIZE";
+    break;
+  case CL_INVALID_GLOBAL_OFFSET:
+    oclHandles.error_str += "CL_INVALID_GLOBAL_OFFSET";
+    break;
+  case CL_OUT_OF_RESOURCES:
+    oclHandles.error_str += "CL_OUT_OF_RESOURCES";
+    break;
+  case CL_MEM_OBJECT_ALLOCATION_FAILURE:
+    oclHandles.error_str += "CL_MEM_OBJECT_ALLOCATION_FAILURE";
+    break;
+  case CL_INVALID_EVENT_WAIT_LIST:
+    oclHandles.error_str += "CL_INVALID_EVENT_WAIT_LIST";
+    break;
+  case CL_OUT_OF_HOST_MEMORY:
+    oclHandles.error_str += "CL_OUT_OF_HOST_MEMORY";
+    break;
+  default:
+    oclHandles.error_str += "Unkown reseason";
+    break;
+  }
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(oclHandles.error_str);
+#endif
+  //_clFinish();
+  /*oclHandles.cl_status = clWaitForEvents(1, &e[0]);
+
+  #ifdef ERRMSG
+
+  if (oclHandles.cl_status!= CL_SUCCESS)
+
+      throw(string("excpetion in _clEnqueueNDRange() -> clWaitForEvents"));
+
+  #endif*/
+}
+
+//--------------------------------------------------------
+// release OpenCL objects
+void _clFree(cl_mem ob) throw(string) {
+  if (ob != NULL)
+    oclHandles.cl_status = clReleaseMemObject(ob);
+#ifdef ERRMSG
+  oclHandles.error_str = "excpetion in _clFree() ->";
+  switch (oclHandles.cl_status) {
+  case CL_INVALID_MEM_OBJECT:
+    oclHandles.error_str += "CL_INVALID_MEM_OBJECT";
+    break;
+  case CL_OUT_OF_RESOURCES:
+    oclHandles.error_str += "CL_OUT_OF_RESOURCES";
+    break;
+  case CL_OUT_OF_HOST_MEMORY:
+    oclHandles.error_str += "CL_OUT_OF_HOST_MEMORY";
+    break;
+  default:
+    oclHandles.error_str += "Unkown reseason";
+    break;
+  }
+  if (oclHandles.cl_status != CL_SUCCESS)
+    throw(oclHandles.error_str);
+#endif
+}
+#endif //_CL_HELPER_

tests/opencl/bfs/Makefile

@@ -0,0 +1,59 @@
+LLVM_PREFIX ?= /opt/llvm-riscv
+RISCV_TOOLCHAIN_PATH ?= /opt/riscv-gnu-toolchain
+SYSROOT ?= $(RISCV_TOOLCHAIN_PATH)/riscv32-unknown-elf
+POCL_CC_PATH ?= /opt/pocl/compiler
+POCL_RT_PATH ?= /opt/pocl/runtime
+
+VORTEX_DRV_PATH ?= $(realpath ../../../driver)
+VORTEX_RT_PATH ?= $(realpath ../../../runtime)
+
+K_LLCFLAGS += "-O3 -march=riscv32 -target-abi=ilp32f -mcpu=generic-rv32 -mattr=+m,+f -float-abi=hard -code-model=small"
+K_CFLAGS   += "-v -O3 -Wstack-usage=1024 --sysroot=$(SYSROOT) --gcc-toolchain=$(RISCV_TOOLCHAIN_PATH) -march=rv32imf -mabi=ilp32f -I$(VORTEX_RT_PATH)/include -fno-rtti -fno-exceptions -ffreestanding -nostartfiles -fdata-sections -ffunction-sections"
+K_LDFLAGS  += "-Wl,-Bstatic,-T$(VORTEX_RT_PATH)/linker/vx_link.ld -Wl,--gc-sections $(VORTEX_RT_PATH)/libvortexrt.a -lm"
+
+CXXFLAGS += -std=c++11 -O2 -Wall -Wextra -pedantic -Wfatal-errors
+#CXXFLAGS += -std=c++11 -O0 -g -Wall -Wextra -pedantic -Wfatal-errors
+
+CXXFLAGS += -I$(POCL_RT_PATH)/include
+
+LDFLAGS += -L$(POCL_RT_PATH)/lib -L$(VORTEX_DRV_PATH)/stub -lOpenCL -lvortex
+
+PROJECT = bfs
+
+SRCS = main.cc
+
+all: $(PROJECT) kernel.pocl
+
+kernel.pocl: kernel.cl
+	LLVM_PREFIX=$(LLVM_PREFIX) POCL_DEBUG=all LD_LIBRARY_PATH=$(LLVM_PREFIX)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -LLCFLAGS $(K_LLCFLAGS) -CFLAGS $(K_CFLAGS) -LDFLAGS $(K_LDFLAGS) -o kernel.pocl kernel.cl
+ 
+$(PROJECT): $(SRCS)
+	$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -o $@
+
+run-fpga: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-asesim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-vlsim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae/vlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-simx: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/simx:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-rtlsim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+.depend: $(SRCS)
+	$(CXX) $(CXXFLAGS) -MM $^ > .depend;
+
+clean:
+	rm -rf $(PROJECT) *.o .depend 
+
+clean-all: clean
+	rm -rf *.pocl *.dump
+
+ifneq ($(MAKECMDGOALS),clean)
+    -include .depend
+endif

tests/opencl/bfs/kernel.cl

@@ -0,0 +1,53 @@
+/* ============================================================
+//--cambine: kernel funtion of Breadth-First-Search
+//--author:	created by Jianbin Fang
+//--date:	06/12/2010
+============================================================ */
+
+//#pragma OPENCL EXTENSION cl_khr_byte_addressable_store: enable
+
+//Structure to hold a node information
+typedef struct{
+	int starting;
+	int no_of_edges;
+} Node;
+
+//--7 parameters
+__kernel void BFS_1( const __global Node* g_graph_nodes,
+					const __global int* g_graph_edges, 
+					__global char* g_graph_mask, 
+					__global char* g_updating_graph_mask, 
+					__global char* g_graph_visited, 
+					__global int* g_cost, 
+					const  int no_of_nodes){
+	int tid = get_global_id(0);
+	if( tid<no_of_nodes && g_graph_mask[tid]){
+		g_graph_mask[tid]=false;
+		for(int i=g_graph_nodes[tid].starting; i<(g_graph_nodes[tid].no_of_edges + g_graph_nodes[tid].starting); i++){
+			int id = g_graph_edges[i];
+			if(!g_graph_visited[id]){
+				g_cost[id]=g_cost[tid]+1;
+				g_updating_graph_mask[id]=true;
+				}
+			}
+	}	
+}
+
+//--5 parameters
+__kernel void BFS_2(__global char* g_graph_mask, 
+					__global char* g_updating_graph_mask, 
+					__global char* g_graph_visited, 
+					__global char* g_over,
+					const  int no_of_nodes
+					) {
+	int tid = get_global_id(0);
+	if( tid<no_of_nodes && g_updating_graph_mask[tid]){
+
+		g_graph_mask[tid]=true;
+		g_graph_visited[tid]=true;
+		*g_over=true;
+		g_updating_graph_mask[tid]=false;
+	}
+}
+
+

tests/opencl/bfs/main.cc

@@ -0,0 +1,297 @@
+//--by Jianbin Fang
+
+#include <cstdlib>
+#include <cstring>
+#include <iostream>
+#include <string>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/stat.h>
+
+#ifdef PROFILING
+#include "timer.h"
+#endif
+
+#include "CLHelper.h"
+#include "util.h"
+
+#define MAX_THREADS_PER_BLOCK 256
+
+// Structure to hold a node information
+struct Node {
+  int starting;
+  int no_of_edges;
+};
+
+//----------------------------------------------------------
+//--bfs on cpu
+//--programmer:	jianbin
+//--date:	26/01/2011
+//--note: width is changed to the new_width
+//----------------------------------------------------------
+void run_bfs_cpu(int no_of_nodes, Node *h_graph_nodes, int edge_list_size,
+                 int *h_graph_edges, char *h_graph_mask,
+                 char *h_updating_graph_mask, char *h_graph_visited,
+                 int *h_cost_ref) {
+  char stop;
+  int k = 0;
+  do {
+    // if no thread changes this value then the loop stops
+    stop = false;
+    for (int tid = 0; tid < no_of_nodes; tid++) {
+      if (h_graph_mask[tid] == true) {
+        h_graph_mask[tid] = false;
+        for (int i = h_graph_nodes[tid].starting;
+             i < (h_graph_nodes[tid].no_of_edges + h_graph_nodes[tid].starting);
+             i++) {
+          int id =
+              h_graph_edges[i]; //--cambine: node id is connected with node tid
+          if (!h_graph_visited[id]) { //--cambine: if node id has not been
+                                      //visited, enter the body below
+            h_cost_ref[id] = h_cost_ref[tid] + 1;
+            h_updating_graph_mask[id] = true;
+          }
+        }
+      }
+    }
+
+    for (int tid = 0; tid < no_of_nodes; tid++) {
+      if (h_updating_graph_mask[tid] == true) {
+        h_graph_mask[tid] = true;
+        h_graph_visited[tid] = true;
+        stop = true;
+        h_updating_graph_mask[tid] = false;
+      }
+    }
+    k++;
+  } while (stop);
+}
+//----------------------------------------------------------
+//--breadth first search on GPUs
+//----------------------------------------------------------
+void run_bfs_gpu(int no_of_nodes, Node *h_graph_nodes, int edge_list_size,
+                 int *h_graph_edges, char *h_graph_mask,
+                 char *h_updating_graph_mask, char *h_graph_visited,
+                 int *h_cost) throw(std::string) {
+
+  // int number_elements = height*width;
+  char h_over;
+  cl_mem d_graph_nodes, d_graph_edges, d_graph_mask, d_updating_graph_mask,
+      d_graph_visited, d_cost, d_over;
+
+  try {
+    //--1 transfer data from host to device
+    _clInit();
+
+    d_graph_nodes = _clMalloc(no_of_nodes * sizeof(Node), h_graph_nodes);
+    d_graph_edges = _clMalloc(edge_list_size * sizeof(int), h_graph_edges);
+    d_graph_mask = _clMallocRW(no_of_nodes * sizeof(char), h_graph_mask);
+    d_updating_graph_mask = _clMallocRW(no_of_nodes * sizeof(char), h_updating_graph_mask);
+    d_graph_visited = _clMallocRW(no_of_nodes * sizeof(char), h_graph_visited);
+
+    d_cost = _clMallocRW(no_of_nodes * sizeof(int), h_cost);
+    d_over = _clMallocRW(sizeof(char), &h_over);
+
+    _clMemcpyH2D(d_graph_nodes, no_of_nodes * sizeof(Node), h_graph_nodes);
+    _clMemcpyH2D(d_graph_edges, edge_list_size * sizeof(int), h_graph_edges);
+    _clMemcpyH2D(d_graph_mask, no_of_nodes * sizeof(char), h_graph_mask);
+    _clMemcpyH2D(d_updating_graph_mask, no_of_nodes * sizeof(char), h_updating_graph_mask);
+    _clMemcpyH2D(d_graph_visited, no_of_nodes * sizeof(char), h_graph_visited);
+    _clMemcpyH2D(d_cost, no_of_nodes * sizeof(int), h_cost);
+
+//--2 invoke kernel
+#ifdef PROFILING
+    timer kernel_timer;
+    double kernel_time = 0.0;
+    kernel_timer.reset();
+    kernel_timer.start();
+#endif
+
+    do {
+      h_over = false;
+      _clMemcpyH2D(d_over, sizeof(char), &h_over);
+      //--kernel 0
+      int kernel_id = 0;
+      int kernel_idx = 0;
+      _clSetArgs(kernel_id, kernel_idx++, d_graph_nodes);
+      _clSetArgs(kernel_id, kernel_idx++, d_graph_edges);
+      _clSetArgs(kernel_id, kernel_idx++, d_graph_mask);
+      _clSetArgs(kernel_id, kernel_idx++, d_updating_graph_mask);
+      _clSetArgs(kernel_id, kernel_idx++, d_graph_visited);
+      _clSetArgs(kernel_id, kernel_idx++, d_cost);
+      _clSetArgs(kernel_id, kernel_idx++, &no_of_nodes, sizeof(int));
+
+      // int work_items = no_of_nodes;
+      _clInvokeKernel(kernel_id, no_of_nodes, work_group_size);
+
+      //--kernel 1
+      kernel_id = 1;
+      kernel_idx = 0;
+      _clSetArgs(kernel_id, kernel_idx++, d_graph_mask);
+      _clSetArgs(kernel_id, kernel_idx++, d_updating_graph_mask);
+      _clSetArgs(kernel_id, kernel_idx++, d_graph_visited);
+      _clSetArgs(kernel_id, kernel_idx++, d_over);
+      _clSetArgs(kernel_id, kernel_idx++, &no_of_nodes, sizeof(int));
+
+      // work_items = no_of_nodes;
+      _clInvokeKernel(kernel_id, no_of_nodes, work_group_size);
+
+      _clMemcpyD2H(d_over, sizeof(char), &h_over);
+    } while (h_over);    
+
+#ifdef PROFILING
+    kernel_timer.stop();
+    kernel_time = kernel_timer.getTimeInSeconds();
+#endif
+    //--3 transfer data from device to host
+    _clMemcpyD2H(d_cost, no_of_nodes * sizeof(int), h_cost);
+//--statistics
+#ifdef PROFILING
+    std::cout << "kernel time(s):" << kernel_time << std::endl;
+#endif
+    //--4 release cl resources.
+    _clFree(d_graph_nodes);
+    _clFree(d_graph_edges);
+    _clFree(d_graph_mask);
+    _clFree(d_updating_graph_mask);
+    _clFree(d_graph_visited);
+    _clFree(d_cost);
+    _clFree(d_over);
+    _clRelease();
+  } catch (std::string msg) {
+    _clFree(d_graph_nodes);
+    _clFree(d_graph_edges);
+    _clFree(d_graph_mask);
+    _clFree(d_updating_graph_mask);
+    _clFree(d_graph_visited);
+    _clFree(d_cost);
+    _clFree(d_over);
+    _clRelease();
+    std::string e_str = "in run_transpose_gpu -> ";
+    e_str += msg;
+    throw(e_str);
+  }
+  return;
+}
+
+//----------------------------------------------------------
+//--cambine:	main function
+//--author:		created by Jianbin Fang
+//--date:		25/01/2011
+//----------------------------------------------------------
+int main(int argc, char *argv[]) {
+	printf("enter demo main\n");
+
+  int no_of_nodes;
+  int edge_list_size;
+  FILE *fp;
+  Node *h_graph_nodes;
+  char *h_graph_mask, *h_updating_graph_mask, *h_graph_visited;
+  
+  try {
+    char *input_f = "graph4096.txt";
+    printf("Reading File\n");
+    // Read in Graph from a file
+    fp = fopen(input_f, "r");
+    if (!fp) {
+      printf("Error Reading graph file\n");
+      return 0;
+    }
+
+    printf("Reading File completed!\n");
+
+    int source = 0;
+
+    fscanf(fp, "%d", &no_of_nodes);
+
+    int num_of_blocks = 1;
+    int num_of_threads_per_block = no_of_nodes;
+
+    // Make execution Parameters according to the number of nodes
+    // Distribute threads across multiple Blocks if necessary
+    if (no_of_nodes > MAX_THREADS_PER_BLOCK) {
+      num_of_blocks = (int)ceil(no_of_nodes / (double)MAX_THREADS_PER_BLOCK);
+      num_of_threads_per_block = MAX_THREADS_PER_BLOCK;
+    }
+    work_group_size = num_of_threads_per_block;
+    // allocate host memory
+    h_graph_nodes = (Node *)malloc(sizeof(Node) * no_of_nodes);
+    h_graph_mask = (char *)malloc(sizeof(char) * no_of_nodes);
+    h_updating_graph_mask = (char *)malloc(sizeof(char) * no_of_nodes);
+    h_graph_visited = (char *)malloc(sizeof(char) * no_of_nodes);
+
+    int start, edgeno;
+    // initalize the memory
+    for (int i = 0; i < no_of_nodes; i++) {
+      fscanf(fp, "%d %d", &start, &edgeno);
+      h_graph_nodes[i].starting = start;
+      h_graph_nodes[i].no_of_edges = edgeno;
+      h_graph_mask[i] = false;
+      h_updating_graph_mask[i] = false;
+      h_graph_visited[i] = false;
+    }
+    // read the source node from the file
+    fscanf(fp, "%d", &source);
+    source = 0;
+    // set the source node as true in the mask
+    h_graph_mask[source] = true;
+    h_graph_visited[source] = true;
+    fscanf(fp, "%d", &edge_list_size);
+    int id, cost;
+    int *h_graph_edges = (int *)malloc(sizeof(int) * edge_list_size);
+    for (int i = 0; i < edge_list_size; i++) {
+      fscanf(fp, "%d", &id);
+      fscanf(fp, "%d", &cost);
+      h_graph_edges[i] = id;
+    }
+
+    if (fp)
+      fclose(fp);
+    // allocate mem for the result on host side
+    int *h_cost = (int *)malloc(sizeof(int) * no_of_nodes);
+    int *h_cost_ref = (int *)malloc(sizeof(int) * no_of_nodes);
+    for (int i = 0; i < no_of_nodes; i++) {
+      h_cost[i] = -1;
+      h_cost_ref[i] = -1;
+    }
+    h_cost[source] = 0;
+    h_cost_ref[source] = 0;
+    //---------------------------------------------------------
+    //--gpu entry
+    run_bfs_gpu(no_of_nodes, h_graph_nodes, edge_list_size, h_graph_edges,
+                h_graph_mask, h_updating_graph_mask, h_graph_visited, h_cost);
+    //---------------------------------------------------------
+    //--cpu entry
+    // initalize the memory again
+    for (int i = 0; i < no_of_nodes; i++) {
+      h_graph_mask[i] = false;
+      h_updating_graph_mask[i] = false;
+      h_graph_visited[i] = false;
+    }
+    // set the source node as true in the mask
+    source = 0;
+    h_graph_mask[source] = true;
+    h_graph_visited[source] = true;
+    run_bfs_cpu(no_of_nodes, h_graph_nodes, edge_list_size, h_graph_edges,
+                h_graph_mask, h_updating_graph_mask, h_graph_visited,
+                h_cost_ref);
+    //---------------------------------------------------------
+    //--result varification
+    compare_results<int>(h_cost_ref, h_cost, no_of_nodes);
+    // release host memory
+    free(h_graph_nodes);
+    free(h_graph_mask);
+    free(h_updating_graph_mask);
+    free(h_graph_visited);
+
+  } catch (std::string msg) {
+    std::cout << "--cambine: exception in main ->" << msg << std::endl;
+    // release host memory
+    free(h_graph_nodes);
+    free(h_graph_mask);
+    free(h_updating_graph_mask);
+    free(h_graph_visited);
+  }
+  printf("Passed!\n");
+  return 0;
+}

tests/opencl/bfs/run

@@ -0,0 +1 @@
+./bfs ../../data/bfs/graph1MW_6.txt

tests/opencl/bfs/timer.cc

@@ -0,0 +1,78 @@
+#include <cstdlib>
+#include <cstring>
+#include <fstream>
+#include <iomanip>
+
+#include "timer.h"
+
+
+using namespace std;
+
+double timer::CPU_speed_in_MHz = timer::get_CPU_speed_in_MHz();
+
+
+double timer::get_CPU_speed_in_MHz()
+{
+#if defined __linux__
+    ifstream infile("/proc/cpuinfo");
+    char     buffer[256], *colon;
+
+    while (infile.good()) {
+	infile.getline(buffer, 256);
+
+	if (strncmp("cpu MHz", buffer, 7) == 0 && (colon = strchr(buffer, ':')) != 0)
+	    return atof(colon + 2);
+    }
+#endif
+
+    return 0.0;
+}
+
+
+void timer::print_time(ostream &str, const char *which, double time) const
+{
+    static const char *units[] = { " ns", " us", " ms", "  s", " ks", 0 };
+    const char	      **unit   = units;
+
+    time = 1000.0 * time / CPU_speed_in_MHz;
+
+    while (time >= 999.5 && unit[1] != 0) {
+	time /= 1000.0;
+	++ unit;
+    }
+
+    str << which << " = " << setprecision(3) << setw(4) << time << *unit;
+}
+
+
+ostream &timer::print(ostream &str)
+{
+    str << left << setw(25) << (name != 0 ? name : "timer") << ": " << right;
+
+    if (CPU_speed_in_MHz == 0)
+	str << "could not determine CPU speed\n";
+    else if (count > 0) {
+	double total = static_cast<double>(total_time);
+
+	print_time(str, "avg", total / static_cast<double>(count));
+	print_time(str, ", total", total);
+	str << ", count = " << setw(9) << count << '\n';
+    }
+    else
+	str << "not used\n";
+
+    return str;
+}
+
+
+ostream &operator << (ostream &str, class timer &timer)
+{
+    return timer.print(str);
+}
+
+double timer::getTimeInSeconds()
+{
+    double total = static_cast<double>(total_time);
+    double res = (total / 1000000.0) / CPU_speed_in_MHz;
+    return res;
+}

tests/opencl/bfs/timer.h

@@ -0,0 +1,101 @@
+#ifndef timer_h
+#define timer_h
+
+#include <iostream>
+
+class timer {
+public:
+  timer(const char *name = 0);
+  timer(const char *name, std::ostream &write_on_exit);
+
+  ~timer();
+
+  void start(), stop();
+  void reset();
+  std::ostream &print(std::ostream &);
+
+  double getTimeInSeconds();
+
+private:
+  void print_time(std::ostream &, const char *which, double time) const;
+
+  union {
+    long long total_time;
+    struct {
+#if defined __PPC__
+      int high, low;
+#else
+      int low, high;
+#endif
+    };
+  };
+
+  unsigned long long count;
+  const char *const name;
+  std::ostream *const write_on_exit;
+
+  static double CPU_speed_in_MHz, get_CPU_speed_in_MHz();
+};
+
+std::ostream &operator<<(std::ostream &, class timer &);
+
+inline void timer::reset() {
+  total_time = 0;
+  count = 0;
+}
+
+inline timer::timer(const char *name) : name(name), write_on_exit(0) {
+  reset();
+}
+
+inline timer::timer(const char *name, std::ostream &write_on_exit)
+    : name(name), write_on_exit(&write_on_exit) {
+  reset();
+}
+
+inline timer::~timer() {
+  if (write_on_exit != 0)
+    print(*write_on_exit);
+}
+
+inline void timer::start() {
+#if (defined __PATHSCALE__) && (defined __i386 || defined __x86_64)
+  unsigned eax, edx;
+
+  asm volatile("rdtsc" : "=a"(eax), "=d"(edx));
+
+  total_time -= ((unsigned long long)edx << 32) + eax;
+#elif (defined __GNUC__ || defined __INTEL_COMPILER) &&                        \
+    (defined __i386 || defined __x86_64)
+  asm volatile("rdtsc\n\t"
+               "subl %%eax, %0\n\t"
+               "sbbl %%edx, %1"
+               : "+m"(low), "+m"(high)
+               :
+               : "eax", "edx");
+#else
+#error Compiler/Architecture not recognized
+#endif
+}
+
+inline void timer::stop() {
+#if (defined __PATHSCALE__) && (defined __i386 || defined __x86_64)
+  unsigned eax, edx;
+
+  asm volatile("rdtsc" : "=a"(eax), "=d"(edx));
+
+  total_time += ((unsigned long long)edx << 32) + eax;
+#elif (defined __GNUC__ || defined __INTEL_COMPILER) &&                        \
+    (defined __i386 || defined __x86_64)
+  asm volatile("rdtsc\n\t"
+               "addl %%eax, %0\n\t"
+               "adcl %%edx, %1"
+               : "+m"(low), "+m"(high)
+               :
+               : "eax", "edx");
+#endif
+
+  ++count;
+}
+
+#endif

tests/opencl/bfs/util.h

@@ -0,0 +1,72 @@
+#ifndef _C_UTIL_
+#define _C_UTIL_
+#include <math.h>
+#include <iostream>
+
+//-------------------------------------------------------------------
+//--initialize array with maximum limit
+//-------------------------------------------------------------------
+template<typename datatype>
+void fill(datatype *A, const int n, const datatype maxi){
+    for (int j = 0; j < n; j++) 
+    {
+        A[j] = ((datatype) maxi * (rand() / (RAND_MAX + 1.0f)));
+    }
+}
+
+//--print matrix
+template<typename datatype>
+void print_matrix(datatype *A, int height, int width){
+	for(int i=0; i<height; i++){
+		for(int j=0; j<width; j++){
+			int idx = i*width + j;
+			std::cout<<A[idx]<<" ";
+		}
+		std::cout<<std::endl;
+	}
+
+	return;
+}
+//-------------------------------------------------------------------
+//--verify results
+//-------------------------------------------------------------------
+#define MAX_RELATIVE_ERROR  .002
+template<typename datatype>
+void verify_array(const datatype *cpuResults, const datatype *gpuResults, const int size){
+
+    char passed = true; 
+#pragma omp parallel for
+    for (int i=0; i<size; i++){
+      if (fabs(cpuResults[i] - gpuResults[i]) / cpuResults[i] > MAX_RELATIVE_ERROR){
+         passed = false; 
+      }
+    }
+    if (passed){
+        std::cout << "--cambine:passed:-)" << endl;
+    }
+    else{
+        std::cout << "--cambine: failed:-(" << endl;
+    }
+    return ;
+}
+template<typename datatype>
+void compare_results(const datatype *cpu_results, const datatype *gpu_results, const int size){
+
+    char passed = true; 
+//#pragma omp parallel for
+    for (int i=0; i<size; i++){
+      if (cpu_results[i]!=gpu_results[i]){
+         passed = false; 
+      }
+    }
+    if (passed){
+        std::cout << "--cambine: passed: -)" << endl;
+    }
+    else{
+        std::cout << "--cambine: failed :-(" << endl;
+    }
+    return ;
+}
+
+#endif
+

tests/opencl/convolution/.gitignore

@@ -0,0 +1 @@
+convolution

tests/opencl/convolution/Makefile

@@ -0,0 +1,59 @@
+LLVM_PREFIX ?= /opt/llvm-riscv
+RISCV_TOOLCHAIN_PATH ?= /opt/riscv-gnu-toolchain
+SYSROOT ?= $(RISCV_TOOLCHAIN_PATH)/riscv32-unknown-elf
+POCL_CC_PATH ?= /opt/pocl/compiler
+POCL_RT_PATH ?= /opt/pocl/runtime
+
+VORTEX_DRV_PATH ?= $(realpath ../../../driver)
+VORTEX_RT_PATH ?= $(realpath ../../../runtime)
+
+K_LLCFLAGS += "-O3 -march=riscv32 -target-abi=ilp32f -mcpu=generic-rv32 -mattr=+m,+f -float-abi=hard -code-model=small"
+K_CFLAGS   += "-v -O3 --sysroot=$(SYSROOT) --gcc-toolchain=$(RISCV_TOOLCHAIN_PATH) -march=rv32imf -mabi=ilp32f -I$(VORTEX_RT_PATH)/include -fno-rtti -fno-exceptions -ffreestanding -nostartfiles -fdata-sections -ffunction-sections"
+K_LDFLAGS  += "-Wl,-Bstatic,-T$(VORTEX_RT_PATH)/linker/vx_link.ld -Wl,--gc-sections $(VORTEX_RT_PATH)/libvortexrt.a -lm"
+
+CXXFLAGS += -std=c++11 -O2 -Wall -Wextra -pedantic -Wfatal-errors
+#CXXFLAGS += -std=c++11 -O0 -g -Wall -Wextra -pedantic -Wfatal-errors
+
+CXXFLAGS += -I$(POCL_RT_PATH)/include
+
+LDFLAGS += -L$(POCL_RT_PATH)/lib -L$(VORTEX_DRV_PATH)/simx -lOpenCL -lvortex
+
+PROJECT = convolution
+
+SRCS = main.cpp utils.cpp
+
+all: $(PROJECT) kernel.pocl
+
+kernel.pocl: kernel.cl
+	LLVM_PREFIX=$(LLVM_PREFIX) POCL_DEBUG=all LD_LIBRARY_PATH=$(LLVM_PREFIX)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -LLCFLAGS $(K_LLCFLAGS) -CFLAGS $(K_CFLAGS) -LDFLAGS $(K_LDFLAGS) -o kernel.pocl kernel.cl
+ 
+$(PROJECT): $(SRCS)
+	$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -o $@
+
+run-fpga: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-asesim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT)
+	
+run-vlsim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae/vlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-simx: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/simx:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-rtlsim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+.depend: $(SRCS)
+	$(CXX) $(CXXFLAGS) -MM $^ > .depend;
+
+clean:
+	rm -rf $(PROJECT) *.o .depend
+
+clean-all: clean
+	rm -rf *.pocl *.dump
+
+ifneq ($(MAKECMDGOALS),clean)
+    -include .depend
+endif

tests/opencl/convolution/kernel.cl

@@ -0,0 +1,54 @@
+__kernel
+void convolution(
+   __read_only  image2d_t  sourceImage,
+   __write_only image2d_t  outputImage, 
+   int rows,
+   int cols,
+   __constant float* filter, 
+   int filterWidth,
+   sampler_t sampler) 
+{
+   // Store each work-item’s unique row and column
+   int column = get_global_id(0);
+   int row    = get_global_id(1);
+
+   // Half the width of the filter is needed for indexing 
+   // memory later
+   int halfWidth = (int)(filterWidth/2);
+
+   // All accesses to images return data as four-element vector 
+   // (i.e., float4), although only the 'x' component will contain 
+   // meaningful data in this code
+   float4 sum = {0.0f, 0.0f, 0.0f, 0.0f};
+    
+   // Iterator for the filter
+   int filterIdx = 0;
+    
+   // Each work-item iterates around its local area based on the 
+   // size of the filter
+   int2 coords;  // Coordinates for accessing the image
+   // Iterate the filter rows
+   for(int i = -halfWidth; i <= halfWidth; i++) {
+      coords.y = row + i;
+
+      // Iterate over the filter columns
+      for(int j = -halfWidth; j <= halfWidth; j++) {
+         coords.x = column + j;
+
+         float4 pixel;
+         // Read a pixel from the image.  A single channel image 
+         // stores the pixel in the 'x' coordinate of the returned
+         // vector.
+         pixel = read_imagef(sourceImage, sampler, coords);
+         sum.x += pixel.x * filter[filterIdx++];
+      }
+   }
+    
+   // Copy the data to the output image if the
+   // work-item is in bounds
+   if(row < rows && column < cols) {
+      coords.x = column;
+      coords.y = row;
+      write_imagef(outputImage, coords, sum);
+   } 
+}

tests/opencl/convolution/main.cpp

@@ -0,0 +1,261 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <CL/cl.h>
+
+#include "utils.h"
+
+// This function takes a positive integer and rounds it up to
+// the nearest multiple of another provided integer
+unsigned int roundUp(unsigned int value, unsigned int multiple) {
+
+  // Determine how far past the nearest multiple the value is
+  unsigned int remainder = value % multiple;
+
+  // Add the difference to make the value a multiple
+  if(remainder != 0) {
+          value += (multiple-remainder);
+  }
+
+  return value;
+}
+
+// This function reads in a text file and stores it as a char pointer
+char* readSource(char* kernelPath) {
+
+   cl_int status;
+   FILE *fp;
+   char *source;
+   long int size;
+
+   printf("Program file is: %s\n", kernelPath);
+
+   fp = fopen(kernelPath, "rb");
+   if(!fp) {
+      printf("Could not open kernel file\n");
+      exit(-1);
+   }
+   status = fseek(fp, 0, SEEK_END);
+   if(status != 0) {
+      printf("Error seeking to end of file\n");
+      exit(-1);
+   }
+   size = ftell(fp);
+   if(size < 0) {
+      printf("Error getting file position\n");
+      exit(-1);
+   }
+
+   rewind(fp);
+
+   source = (char *)malloc(size + 1);
+
+   int i;
+   for (i = 0; i < size+1; i++) {
+      source[i]='\0';
+   }
+
+   if(source == NULL) {
+      printf("Error allocating space for the kernel source\n");
+      exit(-1);
+   }
+
+   fread(source, 1, size, fp);
+   source[size] = '\0';
+
+   return source;
+}
+
+void chk(cl_int status, const char* cmd) {
+
+   if(status != CL_SUCCESS) {
+      printf("%s failed (%d)\n", cmd, status);
+      exit(-1);
+   }
+}
+
+int main() {
+
+   int i, j, k, l;
+
+   // Rows and columns in the input image
+   int imageHeight;
+   int imageWidth;
+
+   const char* inputFile = "input.bmp";
+   const char* outputFile = "output.bmp";
+
+   // Homegrown function to read a BMP from file
+   float* inputImage = readImage(inputFile, &imageWidth,
+      &imageHeight);
+
+   // Size of the input and output images on the host
+   int dataSize = imageHeight*imageWidth*sizeof(float);
+
+   // Output image on the host
+   float* outputImage = NULL;
+   outputImage = (float*)malloc(dataSize);
+   float* refImage = NULL;
+   refImage = (float*)malloc(dataSize);
+
+   // 45 degree motion blur
+   float filter[49] =
+      {0,      0,      0,      0,      0,      0,      0,
+       0,      0,      0,      0,      0,      0,      0,
+       0,      0,     -1,      0,      1,      0,      0,
+       0,      0,     -2,      0,      2,      0,      0,
+       0,      0,     -1,      0,      1,      0,      0,
+       0,      0,      0,      0,      0,      0,      0,
+       0,      0,      0,      0,      0,      0,      0};
+
+   // The convolution filter is 7x7
+   int filterWidth = 7;  
+   int filterSize  = filterWidth*filterWidth;  // Assume a square kernel
+
+   // Set up the OpenCL environment
+   cl_int status;
+
+   // Discovery platform
+   cl_platform_id platform;
+   status = clGetPlatformIDs(1, &platform, NULL);
+   chk(status, "clGetPlatformIDs");
+
+   // Discover device
+   cl_device_id device;
+   clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, 1, &device, NULL);
+   chk(status, "clGetDeviceIDs");
+
+   // Create context
+   cl_context_properties props[3] = {CL_CONTEXT_PLATFORM,
+       (cl_context_properties)(platform), 0};
+   cl_context context;
+   context = clCreateContext(props, 1, &device, NULL, NULL, &status);
+   chk(status, "clCreateContext");
+
+   // Create command queue
+   cl_command_queue queue;
+   queue = clCreateCommandQueue(context, device, 0, &status);
+   chk(status, "clCreateCommandQueue");
+
+   // The image format describes how the data will be stored in memory
+   cl_image_format format;
+   format.image_channel_order     = CL_R;     // single channel
+   format.image_channel_data_type = CL_FLOAT; // float data type
+
+   // Create space for the source image on the device
+   cl_mem d_inputImage = clCreateImage2D(context, 0, &format, imageWidth, 
+      imageHeight, 0, NULL, &status);
+   chk(status, "clCreateImage2D");
+
+   // Create space for the output image on the device
+   cl_mem d_outputImage = clCreateImage2D(context, 0, &format, imageWidth, 
+      imageHeight, 0, NULL, &status);
+   chk(status, "clCreateImage2D");
+
+   // Create space for the 7x7 filter on the device
+   cl_mem d_filter = clCreateBuffer(context, 0, filterSize*sizeof(float), 
+      NULL, &status);
+   chk(status, "clCreateBuffer");
+
+   // Copy the source image to the device
+   size_t origin[3] = {0, 0, 0};  // Offset within the image to copy from
+   size_t region[3] = {imageWidth, imageHeight, 1}; // Elements to per dimension
+   status = clEnqueueWriteImage(queue, d_inputImage, CL_FALSE, origin, region, 
+      0, 0, inputImage, 0, NULL, NULL);
+   chk(status, "clEnqueueWriteImage");
+    
+   // Copy the 7x7 filter to the device
+   status = clEnqueueWriteBuffer(queue, d_filter, CL_FALSE, 0, 
+      filterSize*sizeof(float), filter, 0, NULL, NULL);
+   chk(status, "clEnqueueWriteBuffer");
+
+   // Create the image sampler
+   cl_sampler sampler = clCreateSampler(context, CL_FALSE, 
+      CL_ADDRESS_CLAMP_TO_EDGE, CL_FILTER_NEAREST, &status);
+   chk(status, "clCreateSampler");
+
+   const char* source = readSource("kernel.cl");
+
+   // Create a program object with source and build it
+   cl_program program;
+   program = clCreateProgramWithSource(context, 1, &source, NULL, NULL);
+   chk(status, "clCreateProgramWithSource");
+   status = clBuildProgram(program, 1, &device, NULL, NULL, NULL);
+   chk(status, "clBuildProgram");
+ 
+   // Create the kernel object
+   cl_kernel kernel;
+   kernel = clCreateKernel(program, "convolution", &status);
+   chk(status, "clCreateKernel");
+
+   // Set the kernel arguments
+   status  = clSetKernelArg(kernel, 0, sizeof(cl_mem), &d_inputImage);
+   status |= clSetKernelArg(kernel, 1, sizeof(cl_mem), &d_outputImage);
+   status |= clSetKernelArg(kernel, 2, sizeof(int), &imageHeight);
+   status |= clSetKernelArg(kernel, 3, sizeof(int), &imageWidth);
+   status |= clSetKernelArg(kernel, 4, sizeof(cl_mem), &d_filter);
+   status |= clSetKernelArg(kernel, 5, sizeof(int), &filterWidth);
+   status |= clSetKernelArg(kernel, 6, sizeof(cl_sampler), &sampler);
+   chk(status, "clSetKernelArg");
+
+   // Set the work item dimensions
+   size_t globalSize[2] = {imageWidth, imageHeight};
+   status = clEnqueueNDRangeKernel(queue, kernel, 2, NULL, globalSize, NULL, 0,
+      NULL, NULL);
+   chk(status, "clEnqueueNDRange");
+
+   // Read the image back to the host
+   status = clEnqueueReadImage(queue, d_outputImage, CL_TRUE, origin, 
+      region, 0, 0, outputImage, 0, NULL, NULL); 
+   chk(status, "clEnqueueReadImage");
+
+   // Write the output image to file
+   storeImage(outputImage, outputFile, imageHeight, imageWidth, inputFile);
+
+   // Compute the reference image
+   for(i = 0; i < imageHeight; i++) {
+      for(j = 0; j < imageWidth; j++) {
+         refImage[i*imageWidth+j] = 0;
+      }
+   }
+
+   // Iterate over the rows of the source image
+   int halfFilterWidth = filterWidth/2;
+   float sum;
+   for(i = 0; i < imageHeight; i++) {
+      // Iterate over the columns of the source image
+      for(j = 0; j < imageWidth; j++) {
+         sum = 0; // Reset sum for new source pixel
+         // Apply the filter to the neighborhood
+         for(k = - halfFilterWidth; k <= halfFilterWidth; k++) {
+            for(l = - halfFilterWidth; l <= halfFilterWidth; l++) {
+               if(i+k >= 0 && i+k < imageHeight && 
+                  j+l >= 0 && j+l < imageWidth) {
+                  sum += inputImage[(i+k)*imageWidth + j+l] * 
+                         filter[(k+halfFilterWidth)*filterWidth + 
+                            l+halfFilterWidth];
+               }
+            }
+         }
+         refImage[i*imageWidth+j] = sum;
+      }
+   }
+
+   int failed = 0;
+   for(i = 0; i < imageHeight; i++) {
+      for(j = 0; j < imageWidth; j++) {
+         if(abs(outputImage[i*imageWidth+j]-refImage[i*imageWidth+j]) > 0.01) {
+            printf("Results are INCORRECT\n");
+            printf("Pixel mismatch at <%d,%d> (%f vs. %f)\n", i, j,
+               outputImage[i*imageWidth+j], refImage[i*imageWidth+j]);
+            failed = 1;
+         }
+         if(failed) break;
+      }
+      if(failed) break;
+   }
+   if(!failed) {
+      printf("Results are correct\n");
+   }
+             
+   return 0;
+}

tests/opencl/convolution/utils.cpp

@@ -0,0 +1,180 @@
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "utils.h"
+
+void storeImage(float *imageOut, 
+                const char *filename, 
+                int rows, 
+                int cols, 
+                const char* refFilename) {
+
+   FILE *ifp, *ofp;
+   unsigned char tmp;
+   int offset;
+   unsigned char *buffer;
+   int i, j;
+
+   int bytes;
+
+   int height, width;
+
+   ifp = fopen(refFilename, "rb");
+   if(ifp == NULL) {
+      perror(filename);
+      exit(-1);
+   }
+
+   fseek(ifp, 10, SEEK_SET);
+   fread(&offset, 4, 1, ifp);
+
+   fseek(ifp, 18, SEEK_SET);
+   fread(&width, 4, 1, ifp);
+   fread(&height, 4, 1, ifp);
+
+   fseek(ifp, 0, SEEK_SET);
+
+   buffer = (unsigned char *)malloc(offset);
+   if(buffer == NULL) {
+      perror("malloc");
+      exit(-1);
+   }
+
+   fread(buffer, 1, offset, ifp);
+
+   printf("Writing output image to %s\n", filename);
+   ofp = fopen(filename, "wb");
+   if(ofp == NULL) {
+      perror("opening output file");
+      exit(-1);
+   }
+   bytes = fwrite(buffer, 1, offset, ofp);
+   if(bytes != offset) {
+      printf("error writing header!\n");
+      exit(-1);
+   }
+
+   // NOTE bmp formats store data in reverse raster order (see comment in
+   // readImage function), so we need to flip it upside down here.  
+   int mod = width % 4;
+   if(mod != 0) {
+      mod = 4 - mod;
+   }
+   //   printf("mod = %d\n", mod);
+   for(i = height-1; i >= 0; i--) {
+      for(j = 0; j < width; j++) {
+         tmp = (unsigned char)imageOut[i*cols+j];
+         fwrite(&tmp, sizeof(char), 1, ofp);
+      }
+      // In bmp format, rows must be a multiple of 4-bytes.  
+      // So if we're not at a multiple of 4, add junk padding.
+      for(j = 0; j < mod; j++) {
+         fwrite(&tmp, sizeof(char), 1, ofp);
+      }
+   } 
+
+   fclose(ofp);
+   fclose(ifp);
+
+   free(buffer);
+}
+
+/*
+ * Read bmp image and convert to byte array. Also output the width and height
+ */
+float* readImage(const char *filename, int* widthOut, int* heightOut) {
+
+   uchar* imageData;
+
+   int height, width;
+   uchar tmp;
+   int offset;
+   int i, j;
+
+   printf("Reading input image from %s\n", filename);
+   FILE *fp = fopen(filename, "rb");
+   if(fp == NULL) {
+       perror(filename);
+       exit(-1);
+   }
+
+   fseek(fp, 10, SEEK_SET);
+   fread(&offset, 4, 1, fp);
+
+   fseek(fp, 18, SEEK_SET);
+   fread(&width, 4, 1, fp);
+   fread(&height, 4, 1, fp);
+
+   printf("width = %d\n", width);
+   printf("height = %d\n", height);
+
+   *widthOut = width;
+   *heightOut = height;    
+
+   imageData = (uchar*)malloc(width*height);
+   if(imageData == NULL) {
+       perror("malloc");
+       exit(-1);
+   }
+
+   fseek(fp, offset, SEEK_SET);
+   fflush(NULL);
+
+   int mod = width % 4;
+   if(mod != 0) {
+       mod = 4 - mod;
+   }
+
+   // NOTE bitmaps are stored in upside-down raster order.  So we begin
+   // reading from the bottom left pixel, then going from left-to-right, 
+   // read from the bottom to the top of the image.  For image analysis, 
+   // we want the image to be right-side up, so we'll modify it here.
+
+   // First we read the image in upside-down
+
+   // Read in the actual image
+   for(i = 0; i < height; i++) {
+
+      // add actual data to the image
+      for(j = 0; j < width; j++) {
+         fread(&tmp, sizeof(char), 1, fp);
+         imageData[i*width + j] = tmp;
+      }
+      // For the bmp format, each row has to be a multiple of 4, 
+      // so I need to read in the junk data and throw it away
+      for(j = 0; j < mod; j++) {
+         fread(&tmp, sizeof(char), 1, fp);
+      }
+   }
+
+   // Then we flip it over
+   int flipRow;
+   for(i = 0; i < height/2; i++) {
+      flipRow = height - (i+1);
+      for(j = 0; j < width; j++) {
+         tmp = imageData[i*width+j];
+         imageData[i*width+j] = imageData[flipRow*width+j];
+         imageData[flipRow*width+j] = tmp;
+      }
+   }
+
+   fclose(fp);
+
+   // Input image on the host
+   float* floatImage = NULL;
+   floatImage = (float*)malloc(sizeof(float)*width*height);
+   if(floatImage == NULL) {
+      perror("malloc");
+      exit(-1);
+   }
+
+   // Convert the BMP image to float (not required)
+   for(i = 0; i < height; i++) {
+      for(j = 0; j < width; j++) {
+         floatImage[i*width+j] = (float)imageData[i*width+j];
+      }
+   }
+
+   free(imageData);
+   return floatImage;
+}

tests/opencl/convolution/utils.h

@@ -0,0 +1,11 @@
+#ifndef __UTILS__
+#define __UTILS__
+
+typedef unsigned char uchar;
+
+float* readImage(const char *filename, int* widthOut, int* heightOut);
+
+void storeImage(float *imageOut, const char *filename, int rows, int cols, 
+                const char* refFilename);
+
+#endif

tests/opencl/cutcp/Makefile

@@ -0,0 +1,67 @@
+RISCV_TOOLCHAIN_PATH ?= $(wildcard ../../../../riscv-gnu-toolchain/drops)
+POCL_CC_PATH ?= $(wildcard ../../../../pocl/drops_riscv_cc)
+POCL_INC_PATH ?= $(wildcard ../include)
+POCL_LIB_PATH ?= $(wildcard ../lib)
+VORTEX_RT_PATH ?= $(wildcard ../../../runtime)
+VX_SIMX_PATH ?= $(wildcard ../../../simX/obj_dir)
+
+CC  = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gcc
+CXX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-g++
+DMP = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objdump
+HEX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objcopy
+GDB = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gdb
+
+VX_SRCS =  $(VORTEX_RT_PATH)/newlib/newlib.c
+VX_SRCS += $(VORTEX_RT_PATH)/startup/vx_start.S
+VX_SRCS += $(VORTEX_RT_PATH)/intrinsics/vx_intrinsics.S
+VX_SRCS += $(VORTEX_RT_PATH)/io/vx_io.S $(VORTEX_RT_PATH)/io/vx_io.c
+VX_SRCS += $(VORTEX_RT_PATH)/fileio/fileio.S
+VX_SRCS += $(VORTEX_RT_PATH)/tests/tests.c
+VX_SRCS += $(VORTEX_RT_PATH)/vx_api/vx_api.c
+
+VX_CFLAGS = -nostartfiles -Wl,-Bstatic,-T,$(VORTEX_RT_PATH)/startup/vx_link.ld
+
+CXXFLAGS =  -g -O0 -march=rv32im -mabi=ilp32 
+CXXFLAGS += -ffreestanding # program may not begin at main()
+CXXFLAGS += -Wl,--gc-sections # enable garbage collection of unused input sections
+CXXFLAGS += -fno-rtti -fno-non-call-exceptions # disable RTTI and exceptions
+CXXFLAGS += -I$(POCL_INC_PATH) -I.
+
+VX_LIBS = -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
+QEMU_LIBS = $(VORTEX_RT_PATH)/qemu/vx_api.c -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
+
+PROJECT = cutcp
+
+SRCS = main.cc args.c parboil_opencl.c ocl.c gpu_info.c cutoff.c cutcpu.c output.c readatom.c excl.c
+
+all: $(PROJECT).dump $(PROJECT).hex
+
+lib$(PROJECT).a: kernel.cl
+	POCL_DEBUG=all POCL_DEBUG_LLVM_PASSES=1 LD_LIBRARY_PATH=$(RISCV_TOOLCHAIN_PATH)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -o lib$(PROJECT).a kernel.cl
+
+$(PROJECT).elf: $(SRCS) lib$(PROJECT).a
+	$(CXX) $(CXXFLAGS) $(VX_CFLAGS) $(VX_SRCS) $(SRCS) $(VX_LIBS) -o $(PROJECT).elf
+
+$(PROJECT).qemu: $(SRCS) lib$(PROJECT).a
+	$(CXX) $(CXXFLAGS) $(SRCS) $(QEMU_LIBS) -o $(PROJECT).qemu
+
+$(PROJECT).hex: $(PROJECT).elf
+	$(HEX) -O ihex $(PROJECT).elf $(PROJECT).hex
+
+$(PROJECT).dump: $(PROJECT).elf
+	$(DMP) -D $(PROJECT).elf > $(PROJECT).dump
+
+run: $(PROJECT).hex
+	POCL_DEBUG=all $(VX_SIMX_PATH)/Vcache_simX -E -a rv32i --core $(PROJECT).hex -s -b 1> emulator.debug
+
+qemu: $(PROJECT).qemu
+	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -d in_asm -D debug.log $(PROJECT).qemu
+
+gdb-s: $(PROJECT).qemu
+	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -g 1234 -d in_asm -D debug.log $(PROJECT).qemu
+
+gdb-c: $(PROJECT).qemu
+	$(GDB) $(PROJECT).qemu
+
+clean:
+	rm -rf *.o *.elf *.dump *.hex *.qemu *.log *.debug

tests/opencl/cutcp/args.c

@@ -0,0 +1,617 @@
+
+#include <parboil.h>
+#include <errno.h>
+#include <limits.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+
+/*****************************************************************************/
+/* Memory management routines */
+
+/* Free an array of owned strings. */
+void
+pb_FreeStringArray(char **string_array)
+{
+  char **p;
+
+  if (!string_array) return;
+  for (p = string_array; *p; p++) free(*p);
+  free(string_array);
+}
+
+struct pb_PlatformParam *
+pb_PlatformParam(char *name, char *version)
+{
+  if (name == NULL) {
+    fprintf(stderr, "pb_PlatformParam: Invalid argument\n");
+    exit(-1);
+  }
+
+  struct pb_PlatformParam *ret =
+    (struct pb_PlatformParam *)malloc(sizeof (struct pb_PlatformParam));
+
+  ret->name = name;
+  ret->version = version;
+  return ret;
+}
+
+void
+pb_FreePlatformParam(struct pb_PlatformParam *p)
+{
+  if (p == NULL) return;
+
+  free(p->name);
+  free(p->version);
+  free(p);
+}
+
+struct pb_DeviceParam *
+pb_DeviceParam_index(int index)
+{
+  struct pb_DeviceParam *ret =
+    (struct pb_DeviceParam *)malloc(sizeof (struct pb_DeviceParam));
+  ret->criterion = pb_Device_INDEX;
+  ret->index = index;
+  return ret;
+}
+
+struct pb_DeviceParam *
+pb_DeviceParam_cpu(void)
+{
+  struct pb_DeviceParam *ret =
+    (struct pb_DeviceParam *)malloc(sizeof (struct pb_DeviceParam));
+  ret->criterion = pb_Device_CPU;
+  return ret;
+}
+
+struct pb_DeviceParam *
+pb_DeviceParam_gpu(void)
+{
+  struct pb_DeviceParam *ret =
+    (struct pb_DeviceParam *)malloc(sizeof (struct pb_DeviceParam));
+  ret->criterion = pb_Device_GPU;
+  return ret;
+}
+
+struct pb_DeviceParam *
+pb_DeviceParam_accelerator(void)
+{
+  struct pb_DeviceParam *ret =
+    (struct pb_DeviceParam *)malloc(sizeof (struct pb_DeviceParam));
+  ret->criterion = pb_Device_ACCELERATOR;
+  return ret;
+}
+
+struct pb_DeviceParam *
+pb_DeviceParam_name(char *name)
+{
+  struct pb_DeviceParam *ret =
+    (struct pb_DeviceParam *)malloc(sizeof (struct pb_DeviceParam));
+  ret->criterion = pb_Device_NAME;
+  ret->name = name;
+  return ret;
+}
+
+void
+pb_FreeDeviceParam(struct pb_DeviceParam *p)
+{
+  if (p == NULL) return;
+
+  switch(p->criterion) {
+  case pb_Device_NAME:
+    free(p->name);
+    break;
+  case pb_Device_INDEX:
+  case pb_Device_CPU:
+  case pb_Device_ACCELERATOR:
+    break;
+  default:
+    fprintf(stderr, "pb_FreeDeviceParam: Invalid argument\n");
+    exit(-1);
+  }
+}
+
+void
+pb_FreeParameters(struct pb_Parameters *p)
+{
+  free(p->outFile);
+  pb_FreeStringArray(p->inpFiles);
+  pb_FreePlatformParam(p->platform);
+  pb_FreeDeviceParam(p->device);
+  free(p);
+}
+
+/*****************************************************************************/
+
+/* Parse a comma-delimited list of strings into an
+ * array of strings. */
+static char ** 
+read_string_array(char *in)
+{
+  char **ret;
+  int i;
+  int count;			/* Number of items in the input */
+  char *substring;		/* Current substring within 'in' */
+
+  /* Count the number of items in the string */
+  count = 1;
+  for (i = 0; in[i]; i++) if (in[i] == ',') count++;
+
+  /* Allocate storage */
+  ret = (char **)malloc((count + 1) * sizeof(char *));
+
+  /* Create copies of the strings from the list */
+  substring = in;
+  for (i = 0; i < count; i++) {
+    char *substring_end;
+    int substring_length;
+
+    /* Find length of substring */
+    for (substring_end = substring;
+	 (*substring_end != ',') && (*substring_end != 0);
+	 substring_end++);
+
+    substring_length = substring_end - substring;
+
+    /* Allocate memory and copy the substring */
+    ret[i] = (char *)malloc(substring_length + 1);
+    memcpy(ret[i], substring, substring_length);
+    ret[i][substring_length] = 0;
+
+    /* go to next substring */
+    substring = substring_end + 1;
+  }
+  ret[i] = NULL;		/* Write the sentinel value */
+
+  return ret;
+}
+
+static void
+report_parse_error(const char *str)
+{
+  fputs(str, stderr);
+}
+
+/* Interpret a string as a 'pb_DeviceParam' value.
+ * Return a pointer to a new value, or NULL on failure.
+ */
+static struct pb_DeviceParam *
+read_device_param(char *str)
+{  
+  /* Try different ways of interpreting 'device_string' until one works */
+
+  /* If argument is an integer, then interpret it as a device index */
+  errno = 0;
+  char *end;
+  long device_int = strtol(str, &end, 10);
+  if (!errno) {
+    /* Negative numbers are not valid */
+    if (device_int < 0 || device_int > INT_MAX) return NULL;
+
+    return pb_DeviceParam_index(device_int);
+  }
+
+  /* Match against predefined strings */
+  if (strcmp(str, "CPU") == 0)
+    return pb_DeviceParam_cpu();
+  if (strcmp(str, "GPU") == 0)
+    return pb_DeviceParam_gpu();
+  if (strcmp(str, "ACCELERATOR") == 0)
+    return pb_DeviceParam_accelerator();
+
+  /* Assume any other string is a device name */
+  return pb_DeviceParam_name(strdup(str));
+}
+
+/* Interpret a string as a 'pb_PlatformParam' value.
+ * Return a pointer to a new value, or NULL on failure.
+ */
+static struct pb_PlatformParam *
+read_platform_param(char *str)
+{
+  int separator_index;          /* Index of the '-' character separating
+                                 * name and version number.  It's -1 if
+                                 * there's no '-' character. */
+
+  /* Find the last occurrence of '-' in 'str' */
+  {
+    char *cur;
+    separator_index = -1;
+    for (cur = str; *cur; cur++) {
+      if (*cur == '-') separator_index = cur - str;
+    }
+  }
+
+  /* The platform name is either the entire string, or all characters before
+   * the separator */
+  int name_length = separator_index == -1 ? strlen(str) : separator_index;
+  char *name_str = (char *)malloc(name_length + 1);
+  memcpy(name_str, str, name_length);
+  name_str[name_length] = 0;
+
+  /* The version is either NULL, or all characters after the separator */
+  char *version_str;
+  if (separator_index == -1) {
+    version_str = NULL;
+  }
+  else {
+    const char *version_input_str = str + separator_index + 1;
+    int version_length = strlen(version_input_str);
+
+    version_str = (char *)malloc(version_length + 1);
+    memcpy(version_str, version_input_str, version_length);
+    version_str[version_length] = 0;
+  }
+
+  /* Create output structure */
+  return pb_PlatformParam(name_str, version_str);
+}
+
+/****************************************************************************/
+/* Argument parsing state */
+
+/* Argument parsing state.
+ *
+ * Arguments that are interpreted by the argument parser are removed from
+ * the list.  Variables 'argc' and 'argn' do not count arguments that have
+ * been removed.
+ *
+ * During argument parsing, the array of arguments is compacted, overwriting
+ * the erased arguments.  Variable 'argv_put' points to the array element
+ * where the next argument will be written.  Variable 'argv_get' points to
+ * the array element where the next argument will be read from.
+ */
+struct argparse {
+  int argc;			/* Number of arguments.  Mutable. */
+  int argn;			/* Current argument index. */
+  char **argv_get;		/* Argument value being read. */
+  char **argv_put;		/* Argument value being written.
+				 * argv_put <= argv_get. */
+};
+
+static void
+initialize_argparse(struct argparse *ap, int argc, char **argv)
+{
+  ap->argc = argc;
+  ap->argn = 0;
+  ap->argv_get = ap->argv_put = argv;
+}
+
+/* Finish argument parsing, without processing the remaining arguments.
+ * Write new argument count into _argc. */
+static void
+finalize_argparse(struct argparse *ap, int *_argc, char **argv)
+{
+  /* Move the remaining arguments */
+  for(; ap->argn < ap->argc; ap->argn++)
+    *ap->argv_put++ = *ap->argv_get++;
+
+  /* Update the argument count */
+  *_argc = ap->argc;
+
+  /* Insert a terminating NULL */
+  argv[ap->argc] = NULL;
+}
+
+/* Delete the current argument.  The argument will not be visible
+ * when argument parsing is done. */
+static void
+delete_argument(struct argparse *ap)
+{
+  if (ap->argn >= ap->argc) {
+    fprintf(stderr, "delete_argument\n");
+  }
+  ap->argc--;
+  ap->argv_get++;
+}
+
+/* Go to the next argument.  Also, move the current argument to its
+ * final location in argv. */
+static void
+next_argument(struct argparse *ap)
+{
+  if (ap->argn >= ap->argc) {
+    fprintf(stderr, "next_argument\n");
+  }
+  /* Move argument to its new location. */
+  *ap->argv_put++ = *ap->argv_get++;
+  ap->argn++;
+}
+
+static int
+is_end_of_arguments(struct argparse *ap)
+{
+  return ap->argn == ap->argc;
+}
+
+/* Get the current argument */
+static char *
+get_argument(struct argparse *ap)
+{
+  return *ap->argv_get;
+}
+
+/* Get the current argument, and also delete it */
+static char *
+consume_argument(struct argparse *ap)
+{
+  char *ret = get_argument(ap);
+  delete_argument(ap);
+  return ret;
+}
+
+/****************************************************************************/
+
+/* The result of parsing a command-line argument */
+typedef enum {
+  ARGPARSE_OK,                  /* Success */
+  ARGPARSE_ERROR,               /* Error */
+  ARGPARSE_DONE                 /* Success, and do not continue parsing */
+} result;
+
+typedef result parse_action(struct argparse *ap, struct pb_Parameters *params);
+
+
+/* A command-line option */
+struct option {
+  char short_name;              /* If not 0, the one-character
+                                 * name of this option */
+  const char *long_name;        /* If not NULL, the long name of this option */
+  parse_action *action;       /* What to do when this option occurs.
+                               * Sentinel value is NULL.
+                               */
+};
+
+/* Output file
+ *
+ * -o FILE
+ */
+static result
+parse_output_file(struct argparse *ap, struct pb_Parameters *params)
+{
+  if (is_end_of_arguments(ap))
+    {
+      report_parse_error("Expecting file name after '-o'\n");
+      return ARGPARSE_ERROR;
+    }
+
+  /* Replace the output file name */
+  free(params->outFile);
+  params->outFile = strdup(consume_argument(ap));
+
+  return ARGPARSE_OK;
+}
+
+/* Input files
+ *
+ * -i FILE,FILE,...
+ */
+static result
+parse_input_files(struct argparse *ap, struct pb_Parameters *params)
+{
+  if (is_end_of_arguments(ap))
+    {
+      report_parse_error("Expecting file name after '-i'\n");
+      return ARGPARSE_ERROR;
+    }
+
+  /* Replace the input file list */
+  pb_FreeStringArray(params->inpFiles);
+  params->inpFiles = read_string_array(consume_argument(ap));
+  return ARGPARSE_OK;
+}
+
+/* End of options
+ *
+ * --
+ */
+
+static result
+parse_end_options(struct argparse *ap, struct pb_Parameters *params)
+{
+  return ARGPARSE_DONE;
+}
+
+/* OpenCL device
+ *
+ * --device X
+ */
+
+static result
+parse_device(struct argparse *ap, struct pb_Parameters *params)
+{
+  /* Read the next argument, which specifies a device */
+
+  if (is_end_of_arguments(ap))
+    {
+      report_parse_error("Expecting device specification after '--device'\n");
+      return ARGPARSE_ERROR;
+    }
+
+  char *device_string = consume_argument(ap);
+  struct pb_DeviceParam *device_param = read_device_param(device_string);
+
+  if (!device_param) {
+    report_parse_error("Unrecognized device specification format on command line\n");
+    return ARGPARSE_ERROR;
+  }
+
+  /* Save the result */
+  pb_FreeDeviceParam(params->device);
+  params->device = device_param;
+
+  return ARGPARSE_OK;
+}
+
+static result
+parse_platform(struct argparse *ap, struct pb_Parameters *params)
+{
+  /* Read the next argument, which specifies a platform */
+
+  if (is_end_of_arguments(ap))
+    {
+      report_parse_error("Expecting device specification after '--platform'\n");
+      return ARGPARSE_ERROR;
+    }
+
+  char *platform_string = consume_argument(ap);
+  struct pb_PlatformParam *platform_param = read_platform_param(platform_string);
+
+  if (!platform_param) {
+    report_parse_error("Unrecognized platform specification format on command line\n");
+    return ARGPARSE_ERROR;
+  }
+
+  /* Save the result */
+  pb_FreePlatformParam(params->platform);
+  params->platform = platform_param;
+
+  return ARGPARSE_OK;
+}
+  
+
+static struct option options[] = {
+  { 'o', NULL, &parse_output_file },
+  { 'i', NULL, &parse_input_files },
+  { '-', NULL, &parse_end_options },
+  { 0, "device", &parse_device },
+  { 0, "platform", &parse_platform },
+  { 0, NULL, NULL }
+};
+
+static int
+is_last_option(struct option *op)
+{
+  return op->action == NULL;
+}
+
+/****************************************************************************/
+
+/* Parse command-line parameters.
+ * Return zero on error, nonzero otherwise.
+ * On error, the other outputs may be invalid.
+ *
+ * The information collected from parameters is used to update
+ * 'ret'.  'ret' should be initialized.
+ *
+ * '_argc' and 'argv' are updated to contain only the unprocessed arguments.
+ */
+static int
+pb_ParseParameters (struct pb_Parameters *ret, int *_argc, char **argv)
+{
+  char *err_message;
+  struct argparse ap;
+
+  /* Each argument */
+  initialize_argparse(&ap, *_argc, argv);
+  while(!is_end_of_arguments(&ap)) {
+    result arg_result;          /* Result of parsing this option */
+    char *arg = get_argument(&ap);
+
+    /* Process this argument */
+    if (arg[0] == '-') {
+      /* Single-character flag */
+      if ((arg[1] != 0) && (arg[2] == 0)) {
+        delete_argument(&ap);	/* This argument is consumed here */
+
+        /* Find a matching short option */
+        struct option *op;
+        for (op = options; !is_last_option(op); op++) {
+          if (op->short_name == arg[1]) {
+            arg_result = (*op->action)(&ap, ret);
+            goto option_was_processed;
+          }
+        }
+
+        /* No option matches */
+        report_parse_error("Unexpected command-line parameter\n");
+        arg_result = ARGPARSE_ERROR;
+        goto option_was_processed;
+      }
+
+      /* Long flag */
+      if (arg[1] == '-') {
+        delete_argument(&ap);	/* This argument is consumed here */
+
+        /* Find a matching long option */
+        struct option *op;
+        for (op = options; !is_last_option(op); op++) {
+          if (op->long_name && strcmp(&arg[2], op->long_name) == 0) {
+            arg_result = (*op->action)(&ap, ret);
+            goto option_was_processed;
+          }
+        }
+
+        /* No option matches */
+        report_parse_error("Unexpected command-line parameter\n");
+        arg_result = ARGPARSE_ERROR;
+        goto option_was_processed;
+      }
+    }
+    else {
+      /* Other arguments are ignored */
+      next_argument(&ap);
+      arg_result = ARGPARSE_OK;
+      goto option_was_processed;
+    }
+
+  option_was_processed:
+    /* Decide what to do next based on 'arg_result' */
+    switch(arg_result) {
+    case ARGPARSE_OK:
+      /* Continue processing */
+      break;
+
+    case ARGPARSE_ERROR:
+      /* Error exit from the function */
+      return 0;
+
+    case ARGPARSE_DONE:
+      /* Normal exit from the argument parsing loop */
+      goto end_of_options;
+    }
+  } /* end for each argument */
+
+  /* If all arguments were processed, then normal exit from the loop */
+
+ end_of_options:
+  finalize_argparse(&ap, _argc, argv);
+  return 1;
+}
+
+/*****************************************************************************/
+/* Other exported functions */
+
+struct pb_Parameters *
+pb_ReadParameters(int *_argc, char **argv)
+{
+  struct pb_Parameters *ret =
+    (struct pb_Parameters *)malloc(sizeof(struct pb_Parameters));
+
+  /* Initialize the parameters structure */
+  ret->outFile = NULL;
+  ret->inpFiles = (char **)malloc(sizeof(char *));
+  ret->inpFiles[0] = NULL;
+  ret->platform = NULL;
+  ret->device = NULL;
+
+  /* Read parameters and update _argc, argv */
+  if (!pb_ParseParameters(ret, _argc, argv)) {
+    /* Parse error */
+    pb_FreeParameters(ret);
+    return NULL;
+  }
+
+  return ret;
+}
+
+int
+pb_Parameters_CountInputs(struct pb_Parameters *p)
+{
+  int n;
+
+  for (n = 0; p->inpFiles[n]; n++);
+  return n;
+}
+

tests/opencl/cutcp/atom.h

@@ -0,0 +1,37 @@
+/***************************************************************************
+ *cr
+ *cr            (C) Copyright 2008-2010 The Board of Trustees of the
+ *cr                        University of Illinois
+ *cr                         All Rights Reserved
+ *cr
+ ***************************************************************************/
+
+#ifndef ATOM_H
+#define ATOM_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+  typedef struct Atom_t {
+    float x, y, z, q;
+  } Atom;
+
+  typedef struct Atoms_t {
+    Atom *atoms;
+    int size;
+  } Atoms;
+
+  typedef struct Vec3_t {
+    float x, y, z;
+  } Vec3;
+
+  Atoms *read_atom_file(const char *fname);
+  void free_atom(Atoms *atom);
+  void get_atom_extent(Vec3 *lo, Vec3 *hi, Atoms *atom);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* ATOM_H */

tests/opencl/cutcp/cutcpu.c

@@ -0,0 +1,195 @@
+/***************************************************************************
+ *cr
+ *cr            (C) Copyright 2008-2010 The Board of Trustees of the
+ *cr                        University of Illinois
+ *cr                         All Rights Reserved
+ *cr
+ ***************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <parboil.h>
+#include "atom.h"
+#include "cutoff.h"
+
+#undef DEBUG_PASS_RATE
+#define CHECK_CYLINDER_CPU
+
+#define CELLEN      4.f
+#define INV_CELLEN  (1.f/CELLEN)
+
+extern int cpu_compute_cutoff_potential_lattice(
+    Lattice *lattice,                  /* the lattice */
+    float cutoff,                      /* cutoff distance */
+    Atoms *atoms                       /* array of atoms */
+    )
+{
+  int nx = lattice->dim.nx;
+  int ny = lattice->dim.ny;
+  int nz = lattice->dim.nz;
+  float xlo = lattice->dim.lo.x;
+  float ylo = lattice->dim.lo.y;
+  float zlo = lattice->dim.lo.z;
+  float gridspacing = lattice->dim.h;
+  int natoms = atoms->size;
+  Atom *atom = atoms->atoms;
+
+  const float a2 = cutoff * cutoff;
+  const float inv_a2 = 1.f / a2;
+  float s;
+  const float inv_gridspacing = 1.f / gridspacing;
+  const int radius = (int) ceilf(cutoff * inv_gridspacing) - 1;
+    /* lattice point radius about each atom */
+
+  int n;
+  int i, j, k;
+  int ia, ib, ic;
+  int ja, jb, jc;
+  int ka, kb, kc;
+  int index;
+  int koff, jkoff;
+
+  float x, y, z, q;
+  float dx, dy, dz;
+  float dz2, dydz2, r2;
+  float e;
+  float xstart, ystart;
+
+  float *pg;
+
+  int gindex;
+  int ncell, nxcell, nycell, nzcell;
+  int *first, *next;
+  float inv_cellen = INV_CELLEN;
+  Vec3 minext, maxext;		/* Extent of atom bounding box */
+  float xmin, ymin, zmin;
+  float xmax, ymax, zmax;
+
+#if DEBUG_PASS_RATE
+  unsigned long long pass_count = 0;
+  unsigned long long fail_count = 0;
+#endif
+
+  /* find min and max extent */
+  get_atom_extent(&minext, &maxext, atoms);
+
+  /* number of cells in each dimension */
+  nxcell = (int) floorf((maxext.x-minext.x) * inv_cellen) + 1;
+  nycell = (int) floorf((maxext.y-minext.y) * inv_cellen) + 1;
+  nzcell = (int) floorf((maxext.z-minext.z) * inv_cellen) + 1;
+  ncell = nxcell * nycell * nzcell;
+
+  /* allocate for cursor link list implementation */
+  first = (int *) malloc(ncell * sizeof(int));
+  for (gindex = 0;  gindex < ncell;  gindex++) {
+    first[gindex] = -1;
+  }
+  next = (int *) malloc(natoms * sizeof(int));
+  for (n = 0;  n < natoms;  n++) {
+    next[n] = -1;
+  }
+
+  /* geometric hashing */
+  for (n = 0;  n < natoms;  n++) {
+    if (0==atom[n].q) continue;  /* skip any non-contributing atoms */
+    i = (int) floorf((atom[n].x - minext.x) * inv_cellen);
+    j = (int) floorf((atom[n].y - minext.y) * inv_cellen);
+    k = (int) floorf((atom[n].z - minext.z) * inv_cellen);
+    gindex = (k*nycell + j)*nxcell + i;
+    next[n] = first[gindex];
+    first[gindex] = n;
+  }
+
+  /* traverse the grid cells */
+  for (gindex = 0;  gindex < ncell;  gindex++) {
+    for (n = first[gindex];  n != -1;  n = next[n]) {
+      x = atom[n].x - xlo;
+      y = atom[n].y - ylo;
+      z = atom[n].z - zlo;
+      q = atom[n].q;
+
+      /* find closest grid point with position less than or equal to atom */
+      ic = (int) (x * inv_gridspacing);
+      jc = (int) (y * inv_gridspacing);
+      kc = (int) (z * inv_gridspacing);
+
+      /* find extent of surrounding box of grid points */
+      ia = ic - radius;
+      ib = ic + radius + 1;
+      ja = jc - radius;
+      jb = jc + radius + 1;
+      ka = kc - radius;
+      kb = kc + radius + 1;
+
+      /* trim box edges so that they are within grid point lattice */
+      if (ia < 0)   ia = 0;
+      if (ib >= nx) ib = nx-1;
+      if (ja < 0)   ja = 0;
+      if (jb >= ny) jb = ny-1;
+      if (ka < 0)   ka = 0;
+      if (kb >= nz) kb = nz-1;
+
+      /* loop over surrounding grid points */
+      xstart = ia*gridspacing - x;
+      ystart = ja*gridspacing - y;
+      dz = ka*gridspacing - z;
+      for (k = ka;  k <= kb;  k++, dz += gridspacing) {
+        koff = k*ny;
+        dz2 = dz*dz;
+        dy = ystart;
+        for (j = ja;  j <= jb;  j++, dy += gridspacing) {
+          jkoff = (koff + j)*nx;
+          dydz2 = dy*dy + dz2;
+#ifdef CHECK_CYLINDER_CPU
+          if (dydz2 >= a2) continue;
+#endif
+
+          dx = xstart;
+          index = jkoff + ia;
+          pg = lattice->lattice + index;
+
+#if defined(__INTEL_COMPILER)
+          for (i = ia;  i <= ib;  i++, pg++, dx += gridspacing) {
+            r2 = dx*dx + dydz2;
+            s = (1.f - r2 * inv_a2) * (1.f - r2 * inv_a2);
+            e = q * (1/sqrtf(r2)) * s;
+            *pg += (r2 < a2 ? e : 0);  /* LOOP VECTORIZED!! */
+          }
+#else
+          for (i = ia;  i <= ib;  i++, pg++, dx += gridspacing) {
+            r2 = dx*dx + dydz2;
+            if (r2 >= a2)
+		{
+#ifdef DEBUG_PASS_RATE
+		  fail_count++;
+#endif
+		  continue;
+		}
+#ifdef DEBUG_PASS_RATE
+	    pass_count++;
+#endif
+            s = (1.f - r2 * inv_a2);
+            e = q * (1/sqrtf(r2)) * s * s;
+            *pg += e;
+          }
+#endif
+        }
+      } /* end loop over surrounding grid points */
+
+    } /* end loop over atoms in a gridcell */
+  } /* end loop over gridcells */
+
+  /* free memory */
+  free(next);
+  free(first);
+
+  /* For debugging: print the number of times that the test passed/failed */
+#ifdef DEBUG_PASS_RATE
+  printf ("Pass :%lld\n", pass_count);
+  printf ("Fail :%lld\n", fail_count);
+#endif
+
+  return 0;
+}

tests/opencl/cutcp/cutoff.c

@@ -0,0 +1,499 @@
+/***************************************************************************
+ *cr
+ *cr            (C) Copyright 2008-2010 The Board of Trustees of the
+ *cr                        University of Illinois
+ *cr                         All Rights Reserved
+ *cr
+ ***************************************************************************/
+
+#include <CL/cl.h>
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <parboil.h>
+
+#include "atom.h"
+#include "cutoff.h"
+#include "macros.h"
+#include "ocl.h"
+
+// OpenCL 1.1 support for int3 is not uniform on all implementations, so
+// we use int4 instead.  Only the 'x', 'y', and 'z' fields of xyz are used.
+typedef cl_int4 xyz;
+
+//extern "C" int gpu_compute_cutoff_potential_lattice(
+int gpu_compute_cutoff_potential_lattice(
+    struct pb_TimerSet *timers,
+    Lattice *lattice,                  /* the lattice */
+    float cutoff,                      /* cutoff distance */
+    Atoms *atoms,                      /* array of atoms */
+    int verbose,                        /* print info/debug messages */
+    struct pb_Parameters *parameters
+    )
+{
+  int nx = lattice->dim.nx;
+  int ny = lattice->dim.ny;
+  int nz = lattice->dim.nz;
+  float xlo = lattice->dim.lo.x;
+  float ylo = lattice->dim.lo.y;
+  float zlo = lattice->dim.lo.z;
+  float h = lattice->dim.h;
+  int natoms = atoms->size;
+  Atom *atom = atoms->atoms;
+
+  xyz nbrlist[NBRLIST_MAXLEN];
+  int nbrlistlen = 0;
+
+  int binHistoFull[BIN_DEPTH+1] = { 0 };   /* clear every array element */
+  int binHistoCover[BIN_DEPTH+1] = { 0 };  /* clear every array element */
+  int num_excluded = 0;
+
+  int xRegionDim, yRegionDim, zRegionDim;
+  int xRegionIndex, yRegionIndex, zRegionIndex;
+  int xOffset, yOffset, zOffset;
+  int lnx, lny, lnz, lnall;
+  float *regionZeroAddr, *thisRegion;
+  cl_mem regionZeroCl;
+  int index, indexRegion;
+
+  int c;
+  xyz binDim;
+  int nbins;
+  cl_float4 *binBaseAddr, *binZeroAddr;
+  cl_mem binBaseCl, binZeroCl;
+  int *bincntBaseAddr, *bincntZeroAddr;
+  Atoms *extra = NULL;
+
+  cl_mem NbrListLen;
+  cl_mem NbrList;
+
+  int i, j, k, n;
+  int sum, total;
+
+  float avgFillFull, avgFillCover;
+  const float cutoff2 = cutoff * cutoff;
+  const float inv_cutoff2 = 1.f / cutoff2;
+
+  size_t gridDim[3], blockDim[3];
+
+  // The "compute" timer should be active upon entry to this function
+
+  /* pad lattice to be factor of 8 in each dimension */
+  xRegionDim = (int) ceilf(nx/8.f);
+  yRegionDim = (int) ceilf(ny/8.f);
+  zRegionDim = (int) ceilf(nz/8.f);
+
+  lnx = 8 * xRegionDim;
+  lny = 8 * yRegionDim;
+  lnz = 8 * zRegionDim;
+  lnall = lnx * lny * lnz;
+
+  /* will receive energies from OpenCL */
+  regionZeroAddr = (float *) malloc(lnall * sizeof(float));
+
+  /* create bins */
+  c = (int) ceil(cutoff * BIN_INVLEN);  /* count extra bins around lattice */
+  binDim.x = (int) ceil(lnx * h * BIN_INVLEN) + 2*c;
+  binDim.y = (int) ceil(lny * h * BIN_INVLEN) + 2*c;
+  binDim.z = (int) ceil(lnz * h * BIN_INVLEN) + 2*c;
+  nbins = binDim.x * binDim.y * binDim.z;
+  binBaseAddr = (cl_float4 *) calloc(nbins * BIN_DEPTH, sizeof(cl_float4));
+  binZeroAddr = binBaseAddr + ((c * binDim.y + c) * binDim.x + c) * BIN_DEPTH;
+
+  bincntBaseAddr = (int *) calloc(nbins, sizeof(int));
+  bincntZeroAddr = bincntBaseAddr + (c * binDim.y + c) * binDim.x + c;
+
+  /* create neighbor list */
+  if (ceilf(BIN_LENGTH / (8*h)) == floorf(BIN_LENGTH / (8*h))) {
+    float s = sqrtf(3);
+    float r2 = (cutoff + s*BIN_LENGTH) * (cutoff + s*BIN_LENGTH);
+    int cnt = 0;
+    /* develop neighbor list around 1 cell */
+    if (2*c + 1 > NBRLIST_DIM) {
+      fprintf(stderr, "must have cutoff <= %f\n",
+          (NBRLIST_DIM-1)/2 * BIN_LENGTH);
+      return -1;
+    }
+    for (k = -c;  k <= c;  k++) {
+      for (j = -c;  j <= c;  j++) {
+        for (i = -c;  i <= c;  i++) {
+          if ((i*i + j*j + k*k)*BIN_LENGTH*BIN_LENGTH >= r2) continue;
+          nbrlist[cnt].x = i;
+          nbrlist[cnt].y = j;
+          nbrlist[cnt].z = k;
+          cnt++;
+        }
+      }
+    }
+    nbrlistlen = cnt;
+  }
+  else if (8*h <= 2*BIN_LENGTH) {
+    float s = 2.f*sqrtf(3);
+    float r2 = (cutoff + s*BIN_LENGTH) * (cutoff + s*BIN_LENGTH);
+    int cnt = 0;
+    /* develop neighbor list around 3-cube of cells */
+    if (2*c + 3 > NBRLIST_DIM) {
+      fprintf(stderr, "must have cutoff <= %f\n",
+          (NBRLIST_DIM-3)/2 * BIN_LENGTH);
+      return -1;
+    }
+    for (k = -c;  k <= c;  k++) {
+      for (j = -c;  j <= c;  j++) {
+        for (i = -c;  i <= c;  i++) {
+          if ((i*i + j*j + k*k)*BIN_LENGTH*BIN_LENGTH >= r2) continue;
+          nbrlist[cnt].x = i;
+          nbrlist[cnt].y = j;
+          nbrlist[cnt].z = k;
+          cnt++;
+        }
+      }
+    }
+    nbrlistlen = cnt;
+  }
+  else {
+    fprintf(stderr, "must have h <= %f\n", 0.25 * BIN_LENGTH);
+    return -1;
+  }
+
+  /* perform geometric hashing of atoms into bins */
+  {
+    /* array of extra atoms, permit average of one extra per bin */
+    Atom *extra_atoms = (Atom *) calloc(nbins, sizeof(Atom));
+    int extra_len = 0;
+    
+    for (n = 0;  n < natoms;  n++) {
+      cl_float4 p;
+      p.x = atom[n].x - xlo;
+      p.y = atom[n].y - ylo;
+      p.z = atom[n].z - zlo;
+      p.w = atom[n].q;
+      i = (int) floorf(p.x * BIN_INVLEN);
+      j = (int) floorf(p.y * BIN_INVLEN);
+      k = (int) floorf(p.z * BIN_INVLEN);
+      if (i >= -c && i < binDim.x - c &&
+	  j >= -c && j < binDim.y - c &&
+	  k >= -c && k < binDim.z - c &&
+	  atom[n].q != 0) {
+	int index = (k * binDim.y + j) * binDim.x + i;
+	cl_float4 *bin = binZeroAddr + index * BIN_DEPTH;
+	int bindex = bincntZeroAddr[index];
+	if (bindex < BIN_DEPTH) {
+	  /* copy atom into bin and increase counter for this bin */
+	  bin[bindex] = p;
+	  bincntZeroAddr[index]++;
+	}
+	else {
+	  /* add index to array of extra atoms to be computed with CPU */
+	  if (extra_len >= nbins) {
+	    fprintf(stderr, "exceeded space for storing extra atoms\n");
+	    return -1;
+	  }
+	  extra_atoms[extra_len] = atom[n];
+	  extra_len++;
+	}
+      }
+      else {
+	/* excluded atoms are either outside bins or neutrally charged */
+	num_excluded++;
+      }
+    }
+
+    /* Save result */
+    extra = (Atoms *)malloc(sizeof(Atoms));
+    extra->atoms = extra_atoms;
+    extra->size = extra_len;
+  }
+
+  /* bin stats */
+  sum = total = 0;
+  for (n = 0;  n < nbins;  n++) {
+    binHistoFull[ bincntBaseAddr[n] ]++;
+    sum += bincntBaseAddr[n];
+    total += BIN_DEPTH;
+  }
+  avgFillFull = sum / (float) total;
+  sum = total = 0;
+  for (k = 0;  k < binDim.z - 2*c;  k++) {
+    for (j = 0;  j < binDim.y - 2*c;  j++) {
+      for (i = 0;  i < binDim.x - 2*c;  i++) {
+        int index = (k * binDim.y + j) * binDim.x + i;
+        binHistoCover[ bincntZeroAddr[index] ]++;
+        sum += bincntZeroAddr[index];
+        total += BIN_DEPTH;
+      }
+    }
+  }
+  avgFillCover = sum / (float) total;
+
+  if (verbose) {
+    /* report */
+    printf("number of atoms = %d\n", natoms);
+    printf("lattice spacing = %g\n", h);
+    printf("cutoff distance = %g\n", cutoff);
+    printf("\n");
+    printf("requested lattice dimensions = %d %d %d\n", nx, ny, nz);
+    printf("requested space dimensions = %g %g %g\n", nx*h, ny*h, nz*h);
+    printf("expanded lattice dimensions = %d %d %d\n", lnx, lny, lnz);
+    printf("expanded space dimensions = %g %g %g\n", lnx*h, lny*h, lnz*h);
+    printf("number of bytes for lattice data = %u\n", (unsigned int) (lnall*sizeof(float)));
+    printf("\n");
+    printf("bin padding thickness = %d\n", c);
+    printf("bin cover dimensions = %d %d %d\n",
+        binDim.x - 2*c, binDim.y - 2*c, binDim.z - 2*c);
+    printf("bin full dimensions = %d %d %d\n", binDim.x, binDim.y, binDim.z);
+    printf("number of bins = %d\n", nbins);
+    printf("total number of atom slots = %d\n", nbins * BIN_DEPTH);
+    printf("%% overhead space = %g\n",
+        (natoms / (double) (nbins * BIN_DEPTH)) * 100);
+    printf("number of bytes for bin data = %u\n",
+        (unsigned int)(nbins * BIN_DEPTH * sizeof(cl_float4)));
+    printf("\n");
+    printf("bin histogram with padding:\n");
+    sum = 0;
+    for (n = 0;  n <= BIN_DEPTH;  n++) {
+      printf("     number of bins with %d atoms:  %d\n", n, binHistoFull[n]);
+      sum += binHistoFull[n];
+    }
+    printf("     total number of bins:  %d\n", sum);
+    printf("     %% average fill:  %g\n", avgFillFull * 100);
+    printf("\n");
+    printf("bin histogram excluding padding:\n");
+    sum = 0;
+    for (n = 0;  n <= BIN_DEPTH;  n++) {
+      printf("     number of bins with %d atoms:  %d\n", n, binHistoCover[n]);
+      sum += binHistoCover[n];
+    }
+    printf("     total number of bins:  %d\n", sum);
+    printf("     %% average fill:  %g\n", avgFillCover * 100);
+    printf("\n");
+    printf("number of extra atoms = %d\n", extra->size);
+    printf("%% atoms that are extra = %g\n", (extra->size / (double) natoms) * 100);
+    printf("\n");
+
+    /* sanity check on bins */
+    sum = 0;
+    for (n = 0;  n <= BIN_DEPTH;  n++) {
+      sum += n * binHistoFull[n];
+    }
+    sum += extra->size + num_excluded;
+    printf("sanity check on bin histogram with edges:  "
+        "sum + others = %d\n", sum);
+    sum = 0;
+    for (n = 0;  n <= BIN_DEPTH;  n++) {
+      sum += n * binHistoCover[n];
+    }
+    sum += extra->size + num_excluded;
+    printf("sanity check on bin histogram excluding edges:  "
+        "sum + others = %d\n", sum);
+    printf("\n");
+
+    /* neighbor list */
+    printf("neighbor list length = %d\n", nbrlistlen);
+    printf("\n");
+  }
+
+  printf("Ok!\n");
+
+  pb_Context* pb_context;
+  pb_context = pb_InitOpenCLContext(parameters);
+  if (pb_context == NULL) {
+    fprintf (stderr, "Error: No OpenCL platform/device can be found."); 
+    return -1;
+  }
+
+  printf("Ok!\n");
+
+  cl_int clStatus;
+  cl_device_id clDevice = (cl_device_id) pb_context->clDeviceId;
+  cl_platform_id clPlatform = (cl_platform_id) pb_context->clPlatformId;
+  cl_context clContext = (cl_context) pb_context->clContext;
+
+  cl_command_queue clCommandQueue = clCreateCommandQueue(clContext,clDevice,CL_QUEUE_PROFILING_ENABLE,&clStatus);
+  CHECK_ERROR("clCreateCommandQueue")
+
+  pb_SetOpenCL(&clContext, &clCommandQueue);
+  
+  //const char* clSource[] = {readFile("src/opencl_base/kernel.cl")};
+  //cl_program clProgram = clCreateProgramWithSource(clContext,1,clSource,NULL,&clStatus);
+  cl_program clProgram = clCreateProgramWithBuiltInKernels(
+      clContext, 1, &clDevice, "opencl_cutoff_potential_lattice", &clStatus);
+  CHECK_ERROR("clCreateProgramWithSource")
+
+  char clOptions[50];
+  sprintf(clOptions,"-I src/opencl_base");  //-cl-nv-verbose
+
+  clStatus = clBuildProgram(clProgram,1,&clDevice,clOptions,NULL,NULL);
+  if (clStatus != CL_SUCCESS) {
+    size_t string_size = 0;
+    clGetProgramBuildInfo(clProgram, clDevice, CL_PROGRAM_BUILD_LOG, 
+                          0, NULL, &string_size);
+    char* string = (char*)malloc(string_size*sizeof(char));
+    clGetProgramBuildInfo(clProgram, clDevice, CL_PROGRAM_BUILD_LOG, 
+                          string_size, string, NULL);
+    puts(string);
+  }
+
+  CHECK_ERROR("clBuildProgram")
+
+  cl_kernel clKernel = clCreateKernel(clProgram,"opencl_cutoff_potential_lattice",&clStatus);
+  CHECK_ERROR("clCreateKernel")
+
+  /* setup OpenCL kernel parameters */
+  blockDim[0] = 8;
+  blockDim[1] = 8;
+  blockDim[2] = 2;
+  gridDim[0] = 4 * xRegionDim * blockDim[0];
+  gridDim[1] = yRegionDim * blockDim[1];
+  gridDim[2] = 1 * blockDim[2];
+
+  /* allocate and initialize memory on OpenCL device */
+  pb_SwitchToTimer(timers, pb_TimerID_COPY);
+  if (verbose) {
+    printf("Allocating %.2fMB on OpenCL device for potentials\n",
+           lnall * sizeof(float) / (double) (1024*1024));
+  }
+
+  regionZeroCl = clCreateBuffer(clContext,CL_MEM_WRITE_ONLY,lnall*sizeof(float),NULL,&clStatus);
+  CHECK_ERROR("clCreateBuffer")
+
+  // clMemSet(clCommandQueue,regionZeroCl,0,lnall*sizeof(float));
+
+  if (verbose) {
+    printf("Allocating %.2fMB on OpenCL device for atom bins\n",
+           nbins * BIN_DEPTH * sizeof(cl_float4) / (double) (1024*1024));
+  }
+
+  binBaseCl = clCreateBuffer(clContext,CL_MEM_READ_ONLY,nbins*BIN_DEPTH*sizeof(cl_float4),NULL,&clStatus);
+  CHECK_ERROR("clCreateBuffer")
+ 
+  clStatus = clEnqueueWriteBuffer(clCommandQueue,binBaseCl,CL_TRUE,0,nbins*BIN_DEPTH*sizeof(cl_float4),binBaseAddr,0,NULL,NULL);
+  CHECK_ERROR("clEnqueueWriteBuffer")
+
+  //Sub buffers are not supported in OpenCL v1.0
+  int offset = ((c * binDim.y + c) * binDim.x + c) * BIN_DEPTH;  
+
+  NbrListLen = clCreateBuffer(clContext,CL_MEM_READ_ONLY,sizeof(int),NULL,&clStatus);
+  CHECK_ERROR("clCreateBuffer")
+  clStatus = clEnqueueWriteBuffer(clCommandQueue,NbrListLen,CL_TRUE,0,sizeof(int),&nbrlistlen,0,NULL,NULL);
+  CHECK_ERROR("clEnqueueWriteBuffer")
+
+  NbrList = clCreateBuffer(clContext,CL_MEM_READ_ONLY,NBRLIST_MAXLEN*sizeof(xyz),NULL,&clStatus);
+  CHECK_ERROR("clCreateBuffer")
+  clStatus = clEnqueueWriteBuffer(clCommandQueue,NbrList,CL_TRUE,0,nbrlistlen*sizeof(xyz),nbrlist,0,NULL,NULL);
+  CHECK_ERROR("clEnqueueWriteBuffer")
+
+  if (verbose) 
+    printf("\n");
+
+  clStatus = clSetKernelArg(clKernel,0,sizeof(int),&(binDim.x));
+  clStatus = clSetKernelArg(clKernel,1,sizeof(int),&(binDim.y));
+  clStatus = clSetKernelArg(clKernel,2,sizeof(cl_mem),&binBaseCl);
+  clStatus = clSetKernelArg(clKernel,3,sizeof(int),&offset);
+  clStatus = clSetKernelArg(clKernel,4,sizeof(float),&h);
+  clStatus = clSetKernelArg(clKernel,5,sizeof(float),&cutoff2);
+  clStatus = clSetKernelArg(clKernel,6,sizeof(float),&inv_cutoff2);
+  clStatus = clSetKernelArg(clKernel,7,sizeof(cl_mem),&regionZeroCl);
+  clStatus = clSetKernelArg(clKernel,9,sizeof(cl_mem),&NbrListLen);
+  clStatus = clSetKernelArg(clKernel,10,sizeof(cl_mem),&NbrList);
+  CHECK_ERROR("clSetKernelArg")
+
+  printf("Ok!!\n");
+
+
+  /* loop over z-dimension, invoke OpenCL kernel for each x-y plane */
+  pb_SwitchToTimer(timers, pb_TimerID_KERNEL);
+  printf("Invoking OpenCL kernel on %d region planes...\n", zRegionDim);
+  for (zRegionIndex = 0;  zRegionIndex < zRegionDim;  zRegionIndex++) {
+    printf("  computing plane %d\r", zRegionIndex);
+    fflush(stdout);
+
+    clStatus = clSetKernelArg(clKernel,8,sizeof(int),&zRegionIndex);
+    CHECK_ERROR("clSetKernelArg")
+
+    printf("Ok**!2\n");
+
+    clStatus = clEnqueueNDRangeKernel(clCommandQueue,clKernel,3,NULL,gridDim,blockDim,0,NULL,NULL);
+
+    printf("Ok**!2\n");
+
+    CHECK_ERROR("clEnqueueNDRangeKernel")
+
+    printf("Ok**!2\n");
+
+    clStatus = clFinish(clCommandQueue);
+
+    printf("Ok**!2\n");
+
+    CHECK_ERROR("clFinish")
+  }
+
+  printf("Ok++!\n");
+
+  printf("Finished OpenCL kernel calls                        \n");
+
+  /* copy result regions from OpenCL device */
+  pb_SwitchToTimer(timers, pb_TimerID_COPY);
+  clStatus = clEnqueueReadBuffer(clCommandQueue,regionZeroCl,CL_TRUE,0,lnall*sizeof(float),regionZeroAddr,0,NULL,NULL);
+  CHECK_ERROR("clEnqueueReadBuffer")
+
+  /* free OpenCL memory allocations */
+  clStatus = clReleaseMemObject(regionZeroCl);
+  clStatus = clReleaseMemObject(binBaseCl);
+  clStatus = clReleaseMemObject(NbrListLen);
+  clStatus = clReleaseMemObject(NbrList);
+  CHECK_ERROR("clReleaseMemObject")
+
+  clStatus = clReleaseKernel(clKernel);
+  clStatus = clReleaseProgram(clProgram);
+  clStatus = clReleaseCommandQueue(clCommandQueue);
+  clStatus = clReleaseContext(clContext);
+
+  //free((void*)clSource[0]);
+
+  /* transpose regions back into lattice */
+  pb_SwitchToTimer(timers, pb_TimerID_COMPUTE);
+  for (k = 0;  k < nz;  k++) {
+    zRegionIndex = (k >> 3);
+    zOffset = (k & 7);
+
+    for (j = 0;  j < ny;  j++) {
+      yRegionIndex = (j >> 3);
+      yOffset = (j & 7);
+
+      for (i = 0;  i < nx;  i++) {
+        xRegionIndex = (i >> 3);
+        xOffset = (i & 7);
+
+        thisRegion = regionZeroAddr
+          + ((zRegionIndex * yRegionDim + yRegionIndex) * xRegionDim
+              + xRegionIndex) * REGION_SIZE;
+
+        indexRegion = (zOffset * 8 + yOffset) * 8 + xOffset;
+        index = (k * ny + j) * nx + i;
+
+        lattice->lattice[index] = thisRegion[indexRegion];
+      }
+    }
+  }
+
+  /* handle extra atoms */
+  if (extra->size > 0) {
+    printf("computing extra atoms on CPU\n");
+    if (cpu_compute_cutoff_potential_lattice(lattice, cutoff, extra)) {
+      fprintf(stderr, "cpu_compute_cutoff_potential_lattice() failed "
+          "for extra atoms\n");
+      return -1;
+    }
+    printf("\n");
+  }
+
+  /* cleanup memory allocations */
+  free(regionZeroAddr);
+  free(binBaseAddr);
+  free(bincntBaseAddr);
+  free_atom(extra);
+
+  return 0;
+}

tests/opencl/cutcp/cutoff.h

@@ -0,0 +1,72 @@
+/***************************************************************************
+ *cr
+ *cr            (C) Copyright 2008-2010 The Board of Trustees of the
+ *cr                        University of Illinois
+ *cr                         All Rights Reserved
+ *cr
+ ***************************************************************************/
+
+#ifndef CUTOFF_H
+#define CUTOFF_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define SHIFTED
+
+  /* A structure to record how points in 3D space map to array
+     elements.  Array element (z, y, x)
+     where 0 <= x < nx, 0 <= y < ny, 0 <= z < nz
+     maps to coordinate (xlo, ylo, zlo) + h * (x, y, z).
+  */
+  typedef struct LatticeDim_t {
+    /* Number of lattice points in x, y, z dimensions */
+    int nx, ny, nz;
+
+    /* Lowest corner of lattice */
+    Vec3 lo;
+
+    /* Lattice spacing */
+    float h;
+  } LatticeDim;
+
+  /* An electric potential field sampled on a regular grid.  The
+     lattice size and grid point positions are specified by 'dim'.
+  */
+  typedef struct Lattice_t {
+    LatticeDim dim;
+    float *lattice;
+  } Lattice;
+
+  LatticeDim lattice_from_bounding_box(Vec3 lo, Vec3 hi, float h);
+
+  Lattice *create_lattice(LatticeDim dim);
+  void destroy_lattice(Lattice *);
+
+  int gpu_compute_cutoff_potential_lattice(
+      struct pb_TimerSet *timers,
+      Lattice *lattice,
+      float cutoff,                      /* cutoff distance */
+      Atoms *atom,                       /* array of atoms */
+      int verbose,                        /* print info/debug messages */
+      struct pb_Parameters *parameters
+    );
+
+  int cpu_compute_cutoff_potential_lattice(
+      Lattice *lattice,                  /* the lattice */
+      float cutoff,                      /* cutoff distance */
+      Atoms *atoms                       /* array of atoms */
+    );
+
+  int remove_exclusions(
+      Lattice *lattice,                  /* the lattice */
+      float exclcutoff,                  /* exclusion cutoff distance */
+      Atoms *atom                        /* array of atoms */
+    );
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* CUTOFF_H */

tests/opencl/cutcp/excl.c

@@ -0,0 +1,157 @@
+/***************************************************************************
+ *cr
+ *cr            (C) Copyright 2008-2010 The Board of Trustees of the
+ *cr                        University of Illinois
+ *cr                         All Rights Reserved
+ *cr
+ ***************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <parboil.h>
+
+#include "atom.h"
+#include "cutoff.h"
+
+#define CELLEN      4.f
+#define INV_CELLEN  (1.f/CELLEN)
+
+extern int remove_exclusions(
+    Lattice *lattice,                  /* the lattice */
+    float cutoff,                      /* exclusion cutoff distance */
+    Atoms *atoms                       /* array of atoms */
+    )
+{
+  int nx = lattice->dim.nx;
+  int ny = lattice->dim.ny;
+  int nz = lattice->dim.nz;
+  float xlo = lattice->dim.lo.x;
+  float ylo = lattice->dim.lo.y;
+  float zlo = lattice->dim.lo.z;
+  float gridspacing = lattice->dim.h;
+  Atom *atom = atoms->atoms;
+
+  const float a2 = cutoff * cutoff;
+  const float inv_gridspacing = 1.f / gridspacing;
+  const int radius = (int) ceilf(cutoff * inv_gridspacing) - 1;
+    /* lattice point radius about each atom */
+
+  int n;
+  int i, j, k;
+  int ia, ib, ic;
+  int ja, jb, jc;
+  int ka, kb, kc;
+  int index;
+  int koff, jkoff;
+
+  float x, y, z, q;
+  float dx, dy, dz;
+  float dz2, dydz2, r2;
+  float e;
+  float xstart, ystart;
+
+  float *pg;
+
+  int gindex;
+  int ncell, nxcell, nycell, nzcell;
+  int *first, *next;
+  float inv_cellen = INV_CELLEN;
+  Vec3 minext, maxext;
+
+  /* find min and max extent */
+  get_atom_extent(&minext, &maxext, atoms);
+
+  /* number of cells in each dimension */
+  nxcell = (int) floorf((maxext.x-minext.x) * inv_cellen) + 1;
+  nycell = (int) floorf((maxext.y-minext.y) * inv_cellen) + 1;
+  nzcell = (int) floorf((maxext.z-minext.z) * inv_cellen) + 1;
+  ncell = nxcell * nycell * nzcell;
+
+  /* allocate for cursor link list implementation */
+  first = (int *) malloc(ncell * sizeof(int));
+  for (gindex = 0;  gindex < ncell;  gindex++) {
+    first[gindex] = -1;
+  }
+  next = (int *) malloc(atoms->size * sizeof(int));
+  for (n = 0;  n < atoms->size;  n++) {
+    next[n] = -1;
+  }
+
+  /* geometric hashing */
+  for (n = 0;  n < atoms->size;  n++) {
+    if (0==atom[n].q) continue;  /* skip any non-contributing atoms */
+    i = (int) floorf((atom[n].x - minext.x) * inv_cellen);
+    j = (int) floorf((atom[n].y - minext.y) * inv_cellen);
+    k = (int) floorf((atom[n].z - minext.z) * inv_cellen);
+    gindex = (k*nycell + j)*nxcell + i;
+    next[n] = first[gindex];
+    first[gindex] = n;
+  }
+
+  /* traverse the grid cells */
+  for (gindex = 0;  gindex < ncell;  gindex++) {
+    for (n = first[gindex];  n != -1;  n = next[n]) {
+      x = atom[n].x - xlo;
+      y = atom[n].y - ylo;
+      z = atom[n].z - zlo;
+      q = atom[n].q;
+
+      /* find closest grid point with position less than or equal to atom */
+      ic = (int) (x * inv_gridspacing);
+      jc = (int) (y * inv_gridspacing);
+      kc = (int) (z * inv_gridspacing);
+
+      /* find extent of surrounding box of grid points */
+      ia = ic - radius;
+      ib = ic + radius + 1;
+      ja = jc - radius;
+      jb = jc + radius + 1;
+      ka = kc - radius;
+      kb = kc + radius + 1;
+
+      /* trim box edges so that they are within grid point lattice */
+      if (ia < 0)   ia = 0;
+      if (ib >= nx) ib = nx-1;
+      if (ja < 0)   ja = 0;
+      if (jb >= ny) jb = ny-1;
+      if (ka < 0)   ka = 0;
+      if (kb >= nz) kb = nz-1;
+
+      /* loop over surrounding grid points */
+      xstart = ia*gridspacing - x;
+      ystart = ja*gridspacing - y;
+      dz = ka*gridspacing - z;
+      for (k = ka;  k <= kb;  k++, dz += gridspacing) {
+        koff = k*ny;
+        dz2 = dz*dz;
+
+        dy = ystart;
+        for (j = ja;  j <= jb;  j++, dy += gridspacing) {
+          jkoff = (koff + j)*nx;
+          dydz2 = dy*dy + dz2;
+
+          dx = xstart;
+          index = jkoff + ia;
+          pg = lattice->lattice + index;
+
+          for (i = ia;  i <= ib;  i++, pg++, dx += gridspacing) {
+            r2 = dx*dx + dydz2;
+
+	    /* If atom and lattice point are too close, set the lattice value
+	     * to zero */
+            if (r2 < a2) *pg = 0;
+          }
+        }
+      } /* end loop over surrounding grid points */
+
+    } /* end loop over atoms in a gridcell */
+  } /* end loop over gridcells */
+
+  /* free memory */
+  free(next);
+  free(first);
+
+  return 0;
+}

tests/opencl/cutcp/gpu_info.c

@@ -0,0 +1,55 @@
+/***************************************************************************
+ *cr
+ *cr            (C) Copyright 2010 The Board of Trustees of the
+ *cr                        University of Illinois
+ *cr                         All Rights Reserved
+ *cr
+ ***************************************************************************/
+//#include <endian.h>
+#include <stdlib.h>
+#include <malloc.h>
+#include <stdio.h>
+#include <inttypes.h>
+
+#include "gpu_info.h"
+
+void compute_active_thread(size_t *thread,
+			   size_t *grid,
+			   int task,
+			   int pad,
+			   int major,
+			   int minor,
+			   int sm)
+{
+	int max_thread;
+	int max_block=8;
+	if(major==1)
+	{
+		if(minor>=2)
+			max_thread=1024;
+		else
+			max_thread=768;
+	}
+	else if(major==2)
+		max_thread=1536;
+	else
+		//newer GPU  //keep using 2.0
+		max_thread=1536;
+	
+	int _grid;
+	int _thread;
+	
+	if(task*pad>sm*max_thread)
+	{
+		_thread=max_thread/max_block;
+		_grid = ((task*pad+_thread-1)/_thread)*_thread;
+	}
+	else
+	{
+		_thread=pad;
+		_grid=task*pad;
+	}
+
+	thread[0]=_thread;
+	grid[0]=_grid;
+}

tests/opencl/cutcp/gpu_info.h

@@ -0,0 +1,20 @@
+/***************************************************************************
+ *cr
+ *cr            (C) Copyright 2010 The Board of Trustees of the
+ *cr                        University of Illinois
+ *cr                         All Rights Reserved
+ *cr
+ ***************************************************************************/
+
+#ifndef __GPUINFOH__
+#define __GPUINFOH__
+
+void compute_active_thread(size_t *thread,
+			   size_t *grid,
+			   int task,
+			   int pad,
+			   int major,
+			   int minor,
+			   int sm);
+
+#endif

tests/opencl/cutcp/kernel.cl

@@ -0,0 +1,104 @@
+/*
+ * potential lattice is decomposed into size 8^3 lattice point "regions"
+ *
+ * THIS IMPLEMENTATION:  one thread per lattice point
+ * thread block size 128 gives 4 thread blocks per region
+ * kernel is invoked for each x-y plane of regions,
+ * where gridDim.x is 4*(x region dimension) so that blockIdx.x 
+ * can absorb the z sub-region index in its 2 lowest order bits
+ *
+ * Regions are stored contiguously in memory in row-major order
+ *
+ * The bins have to not only cover the region, but they need to surround
+ * the outer edges so that region sides and corners can still use
+ * neighbor list stencil.  The binZeroAddr is actually a shifted pointer into
+ * the bin array (binZeroAddr = binBaseAddr + (c*binDim_y + c)*binDim_x + c)
+ * where c = ceil(cutoff / binsize).  This allows for negative offsets to
+ * be added to myBinIndex.
+ *
+ * The (0,0,0) spatial origin corresponds to lower left corner of both
+ * regionZeroAddr and binZeroAddr.  The atom coordinates are translated
+ * during binning to enforce this assumption.
+ */
+
+#include "macros.h"
+
+// OpenCL 1.1 support for int3 is not uniform on all implementations, so
+// we use int4 instead.  Only the 'x', 'y', and 'z' fields of xyz are used.
+typedef int4 xyz;
+
+__kernel void opencl_cutoff_potential_lattice(
+    int binDim_x,
+    int binDim_y,
+    __global float4 *binBaseAddr,
+    int offset,
+    float h,                /* lattice spacing */
+    float cutoff2,          /* square of cutoff distance */
+    float inv_cutoff2,
+    __global float *regionZeroAddr,  /* address of lattice regions starting at origin */
+    int zRegionIndex,
+    __constant int *NbrListLen,
+    __constant xyz *NbrList
+    )
+{
+  __global float4* binZeroAddr = binBaseAddr + offset;
+
+  __global float *myRegionAddr;
+  int Bx, By, Bz;
+
+  /* thread id */
+  const int tid = (get_local_id(2)*get_local_size(1) + 
+                      get_local_id(1))*get_local_size(0) + get_local_id(0);
+
+  /* this is the start of the sub-region indexed by tid */
+  myRegionAddr = regionZeroAddr + ((zRegionIndex*get_num_groups(1)
+	+ get_group_id(1))*(get_num_groups(0)>>2) + (get_group_id(0)>>2))*REGION_SIZE
+	+ (get_group_id(0)&3)*SUB_REGION_SIZE;
+
+  /* spatial coordinate of this lattice point */
+  float x = (8 * (get_group_id(0) >> 2) + get_local_id(0)) * h;
+  float y = (8 * get_group_id(1) + get_local_id(1)) * h;
+  float z = (8 * zRegionIndex + 2*(get_group_id(0)&3) + get_local_id(2)) * h;
+
+  float dx;
+  float dy;
+  float dz;
+  float r2;
+  float s;
+
+  int totalbins = 0;
+
+  /* bin number determined by center of region */
+  Bx = (int) floor((8 * (get_group_id(0) >> 2) + 4) * h * BIN_INVLEN);
+  By = (int) floor((8 * get_group_id(1) + 4) * h * BIN_INVLEN);
+  Bz = (int) floor((8 * zRegionIndex + 4) * h * BIN_INVLEN);
+
+  float energy = 0.f;
+  int bincnt;
+  for (bincnt = 0; bincnt < *NbrListLen;  bincnt++) {
+    int i = Bx + NbrList[bincnt].x;
+    int j = By + NbrList[bincnt].y;
+    int k = Bz + NbrList[bincnt].z;
+
+  	__global float4* p_global = binZeroAddr + 
+  	                       (((k*binDim_y + j)*binDim_x + i) * BIN_DEPTH);
+
+    int m;
+    for (m = 0;  m < BIN_DEPTH;  m++) {
+    	float aq = p_global[m].w;
+      if (0.f != aq) {
+        dx = p_global[m].x - x;
+        dy = p_global[m].y - y;
+        dz = p_global[m].z - z;
+        r2 = dx*dx + dy*dy + dz*dz;
+        if (r2 < cutoff2) {
+          s = (1.f - r2 * inv_cutoff2);
+          energy += aq * rsqrt(r2) * s * s;
+        }
+      }
+    } /* end loop over atoms in bin */
+  } /* end loop over neighbor list */
+
+  /* store into global memory */
+  myRegionAddr[tid+0] = energy;
+}

tests/opencl/cutcp/macros.h

@@ -0,0 +1,69 @@
+#ifndef __MACROSH__
+#define __MACROSH__
+
+#ifdef __DEVICE_EMULATION__
+#define DEBUG
+/* define which grid block and which thread to examine */
+#define BX  0
+#define BY  0
+#define TX  0
+#define TY  0
+#define TZ  0
+#define EMU(code) do { \
+  if (blockIdx.x==BX && blockIdx.y==BY && \
+      threadIdx.x==TX && threadIdx.y==TY && threadIdx.z==TZ) { \
+    code; \
+  } \
+} while (0)
+#define INT(n)    printf("%s = %d\n", #n, n)
+#define FLOAT(f)  printf("%s = %g\n", #f, (double)(f))
+#define INT3(n)   printf("%s = %d %d %d\n", #n, (n).x, (n).y, (n).z)
+#define FLOAT4(f) printf("%s = %g %g %g %g\n", #f, (double)(f).x, \
+    (double)(f).y, (double)(f).z, (double)(f).w)
+#else
+#define EMU(code)
+#define INT(n)
+#define FLOAT(f)
+#define INT3(n)
+#define FLOAT4(f)
+#endif
+
+/* report error from OpenCL */
+#define CHECK_ERROR(errorMessage)           \
+  if(clStatus != CL_SUCCESS)                \
+  {                                         \
+     printf("Error: %s!\n",errorMessage);   \
+     printf("Line: %d\n",__LINE__);         \
+     exit(1);                               \
+  }
+
+/*
+ * neighbor list:
+ * stored in constant memory as table of offsets
+ * flat index addressing is computed by kernel
+ *
+ * reserve enough memory for 11^3 stencil of grid cells
+ * this fits within 16K of memory
+ */
+#define NBRLIST_DIM  11
+#define NBRLIST_MAXLEN (NBRLIST_DIM * NBRLIST_DIM * NBRLIST_DIM)
+
+/*
+ * atom bins cached into shared memory for processing
+ *
+ * this reserves 4K of shared memory for 32 atom bins each containing 8 atoms,
+ * should permit scheduling of up to 3 thread blocks per SM
+ */
+#define BIN_DEPTH         8  /* max number of atoms per bin */
+#define BIN_SIZE         32  /* size of bin in floats */
+#define BIN_CACHE_MAXLEN 32  /* max number of atom bins to cache */
+
+#define BIN_LENGTH      4.f  /* spatial length in Angstroms */
+#define BIN_INVLEN  (1.f / BIN_LENGTH)
+/* assuming density of 1 atom / 10 A^3, expectation is 6.4 atoms per bin
+ * so that bin fill should be 80% (for non-empty regions of space) */
+
+#define REGION_SIZE     512  /* number of floats in lattice region */
+#define SUB_REGION_SIZE 128  /* number of floats in lattice sub-region */
+
+#endif

tests/opencl/cutcp/main.cc

@@ -0,0 +1,194 @@
+/***************************************************************************
+ *cr
+ *cr            (C) Copyright 2008-2010 The Board of Trustees of the
+ *cr                        University of Illinois
+ *cr                         All Rights Reserved
+ *cr
+ ***************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <parboil.h>
+
+#include "atom.h"
+#include "cutoff.h"
+#include "output.h"
+
+#define ERRTOL 1e-4f
+
+#define NOKERNELS             0
+#define CUTOFF1               1
+#define CUTOFF6              32
+#define CUTOFF6OVERLAP       64
+#define CUTOFFCPU         16384
+
+
+int appenddata(const char *filename, int size, double time) {
+  FILE *fp;
+  fp=fopen(filename, "a");
+  if (fp == NULL) {
+    printf("error appending to file %s..\n", filename);
+    return -1;
+  }
+  fprintf(fp, "%d  %.3f\n", size, time);
+  fclose(fp);
+  return 0;
+}
+
+LatticeDim
+lattice_from_bounding_box(Vec3 lo, Vec3 hi, float h)
+{
+  LatticeDim ret;
+
+  ret.nx = (int) floorf((hi.x-lo.x)/h) + 1;
+  ret.ny = (int) floorf((hi.y-lo.y)/h) + 1;
+  ret.nz = (int) floorf((hi.z-lo.z)/h) + 1;
+  ret.lo = lo;
+  ret.h = h;
+
+  return ret;
+}
+
+Lattice *
+create_lattice(LatticeDim dim)
+{
+  int size;
+  Lattice *lat = (Lattice *)malloc(sizeof(Lattice));
+
+  if (lat == NULL) {
+    fprintf(stderr, "Out of memory\n");
+    exit(1);
+  }
+
+  lat->dim = dim;
+
+  /* Round up the allocated size to a multiple of 8 */
+  size = ((dim.nx * dim.ny * dim.nz) + 7) & ~7;
+  lat->lattice = (float *)calloc(size, sizeof(float));
+
+  if (lat->lattice == NULL) {
+    fprintf(stderr, "Out of memory\n");
+    exit(1);
+  }
+
+  return lat;
+}
+
+
+void
+destroy_lattice(Lattice *lat)
+{
+  if (lat) {
+    free(lat->lattice);
+    free(lat);
+  }
+}
+
+int main(int argc, char *argv[]) {
+  Atoms *atom;
+
+  LatticeDim lattice_dim;
+  Lattice *gpu_lattice;
+  Vec3 min_ext, max_ext;	/* Bounding box of atoms */
+  Vec3 lo, hi;			/* Bounding box with padding  */
+
+  float h = 0.5f;		/* Lattice spacing */
+  float cutoff = 12.f;		/* Cutoff radius */
+  float exclcutoff = 1.f;	/* Radius for exclusion */
+  float padding = 0.5f;		/* Bounding box padding distance */
+
+  int n;
+
+  struct pb_Parameters *parameters;
+  struct pb_TimerSet timers;
+
+  /* Read input parameters */
+  parameters = pb_ReadParameters(&argc, argv);
+  if (parameters == NULL) {
+    exit(1);
+  }
+
+  parameters->inpFiles = (char **)malloc(sizeof(char *) * 2);
+  parameters->inpFiles[0] = (char *)malloc(100);
+  parameters->inpFiles[1] = NULL;
+  strncpy(parameters->inpFiles[0], "watbox.sl40.pqr", 100);
+
+  /* Expect one input file */
+  if (pb_Parameters_CountInputs(parameters) != 1) {
+    fprintf(stderr, "Expecting one input file\n");
+    exit(1);
+  }
+
+  pb_InitializeTimerSet(&timers);
+  pb_SwitchToTimer(&timers, pb_TimerID_IO);
+
+  printf("OK\n");
+
+  {
+    const char *pqrfilename = parameters->inpFiles[0];
+
+    if (!(atom = read_atom_file(pqrfilename))) {
+      fprintf(stderr, "read_atom_file() failed\n");
+      exit(1);
+    }
+    printf("read %d atoms from file '%s'\n", atom->size, pqrfilename);
+  }
+
+  printf("OK\n");
+
+  /* find extent of domain */
+  pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
+  get_atom_extent(&min_ext, &max_ext, atom);
+  printf("extent of domain is:\n");
+  printf("  minimum %g %g %g\n", min_ext.x, min_ext.y, min_ext.z);
+  printf("  maximum %g %g %g\n", max_ext.x, max_ext.y, max_ext.z);
+
+  printf("padding domain by %g Angstroms\n", padding);
+  lo = (Vec3) {min_ext.x - padding, min_ext.y - padding, min_ext.z - padding};
+  hi = (Vec3) {max_ext.x + padding, max_ext.y + padding, max_ext.z + padding};
+  printf("domain lengths are %g by %g by %g\n", hi.x-lo.x, hi.y-lo.y, hi.z-lo.z);
+
+  lattice_dim = lattice_from_bounding_box(lo, hi, h);
+  gpu_lattice = create_lattice(lattice_dim);
+  printf("\n");
+
+  /*
+   *  Run OpenCL kernel
+   *  (Begin and end with COMPUTE timer active)
+   */
+  if (gpu_compute_cutoff_potential_lattice(&timers, gpu_lattice, cutoff, atom, 0, parameters)) {
+    fprintf(stderr, "Computation failed\n");
+    exit(1);
+  }
+
+  /*
+   * Zero the lattice points that are too close to an atom.  This is
+   * necessary for numerical stability.
+   */
+  if (remove_exclusions(gpu_lattice, exclcutoff, atom)) {
+    fprintf(stderr, "remove_exclusions() failed for gpu lattice\n");
+    exit(1);
+  }
+
+  printf("\n");
+
+  pb_SwitchToTimer(&timers, pb_TimerID_IO);
+
+  /* Print output */
+  if (parameters->outFile) {
+    //write_lattice_summary(parameters->outFile, gpu_lattice);
+  }
+  pb_SwitchToTimer(&timers, pb_TimerID_COMPUTE);
+
+  /* Cleanup */
+  destroy_lattice(gpu_lattice);
+  free_atom(atom);
+
+  pb_SwitchToTimer(&timers, pb_TimerID_NONE);
+  pb_PrintTimerSet(&timers);
+  pb_FreeParameters(parameters);
+
+  return 0;
+}

tests/opencl/cutcp/ocl.c

@@ -0,0 +1,49 @@
+#include <CL/cl.h>
+#include <stdio.h>
+#include <string.h>
+#include "ocl.h"
+
+char* readFile(const char* fileName)
+{
+        FILE* fp;
+        fp = fopen(fileName,"r");
+        if(fp == NULL)
+        {
+                printf("Error 1!\n");
+                exit(1);
+        }
+
+        fseek(fp,0,SEEK_END);
+        long size = ftell(fp);
+        rewind(fp);
+
+        char* buffer = (char*)malloc(sizeof(char)*(size+1));
+        if(buffer  == NULL)
+        {
+                printf("Error 2!\n");
+                fclose(fp);
+                exit(1);
+        }
+
+        size_t res = fread(buffer,1,size,fp);
+        if(res != size)
+        {
+                printf("Error 3!\n");
+                fclose(fp);
+                exit(1);
+        }
+
+	buffer[size] = 0;
+        fclose(fp);
+        return buffer;
+}
+
+void clMemSet(cl_command_queue clCommandQueue, cl_mem buf, int val, size_t size)
+{
+	cl_int clStatus;
+	char* temp = (char*)malloc(size);
+	memset(temp,val,size);
+	clStatus = clEnqueueWriteBuffer(clCommandQueue,buf,CL_TRUE,0,size,temp,0,NULL,NULL);
+	CHECK_ERROR("clEnqueueWriteBuffer")
+	free(temp);
+}

tests/opencl/cutcp/ocl.h

@@ -0,0 +1,17 @@
+#ifndef __OCLH__
+#define __OCLH__
+
+#include <stdlib.h>
+
+void clMemSet(cl_command_queue, cl_mem, int, size_t);
+char* readFile(const char*);
+
+#define CHECK_ERROR(errorMessage)           \
+  if(clStatus != CL_SUCCESS)                \
+  {                                         \
+     printf("Error: %s!\n",errorMessage);   \
+     printf("Line: %d\n",__LINE__);         \
+     exit(1);                               \
+  }
+
+#endif

tests/opencl/cutcp/output.c

@@ -0,0 +1,67 @@
+/***************************************************************************
+ *cr
+ *cr            (C) Copyright 2008-2010 The Board of Trustees of the
+ *cr                        University of Illinois
+ *cr                         All Rights Reserved
+ *cr
+ ***************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <inttypes.h>
+#include <math.h>
+#include <parboil.h>
+
+#include "atom.h"
+#include "cutoff.h"
+
+void
+write_lattice_summary(const char *filename, Lattice *lattice)
+{
+  float *lattice_data = lattice->lattice;
+  int nx = lattice->dim.nx;
+  int ny = lattice->dim.ny;
+  int nz = lattice->dim.nz;
+
+  /* Open output file */
+  FILE *outfile = fopen(filename, "w");
+
+  if (outfile == NULL) {
+    fprintf(stderr, "Cannot open output file\n");
+    exit(1);
+  }
+
+  /* Write the sum of the the absolute values of all lattice potentials */
+  {
+    double abspotential = 0.0;
+    float tmp;
+    int i;
+
+    for (i = 0; i < nx * ny * nz; i++)
+      abspotential += fabs((double) lattice_data[i]);
+
+    tmp = (float) abspotential;
+
+    fwrite(&tmp, 1, sizeof(float), outfile);
+  }
+
+  /* Write the size of a lattice plane */
+  {
+    uint32_t tmp;
+
+    tmp = (uint32_t) (lattice->dim.nx * lattice->dim.ny);
+    fwrite(&tmp, 1, sizeof(uint32_t), outfile);
+  }
+
+  /* Write the plane of lattice data at z=0 and z = nz-1 */
+  {
+    int plane_size = nx * ny;
+
+    fwrite(lattice_data, plane_size, sizeof(float), outfile);
+    fwrite(lattice_data + (nz-1) * plane_size, plane_size, sizeof(float),
+	   outfile);
+  }
+
+  /* Cleanup */
+  fclose(outfile);
+}

tests/opencl/cutcp/output.h

@@ -0,0 +1,25 @@
+/***************************************************************************
+ *cr
+ *cr            (C) Copyright 2008-2010 The Board of Trustees of the
+ *cr                        University of Illinois
+ *cr                         All Rights Reserved
+ *cr
+ ***************************************************************************/
+
+#ifndef OUTPUT_H
+#define OUTPUT_H
+
+#include "cutoff.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void
+write_lattice_summary(const char *filename, Lattice *lattice);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

tests/opencl/cutcp/parboil.h

@@ -0,0 +1,348 @@
+/*
+ * (c) 2010 The Board of Trustees of the University of Illinois.
+ */
+#ifndef PARBOIL_HEADER
+#define PARBOIL_HEADER
+
+#include <stdio.h>
+#include <string.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <unistd.h>
+
+/* A platform as specified by the user on the command line */
+struct pb_PlatformParam {
+  char *name;                   /* The platform name.  This string is owned. */
+  char *version;                /* The platform version; may be NULL.
+                                 * This string is owned. */
+};
+
+/* Create a PlatformParam from the given strings.
+ * 'name' must not be NULL.  'version' may be NULL.
+ * If not NULL, the strings should have been allocated by malloc(),
+ * and they will be owned by the returned object.
+ */
+struct pb_PlatformParam *
+pb_PlatformParam(char *name, char *version);
+
+void
+pb_FreePlatformParam(struct pb_PlatformParam *);
+
+/* A criterion for how to select a device */
+enum pb_DeviceSelectionCriterion {
+  pb_Device_INDEX,             /* Enumerate the devices and select one
+                                * by its number */
+  pb_Device_CPU,                /* Select a CPU device */
+  pb_Device_GPU,                /* Select a GPU device  */
+  pb_Device_ACCELERATOR,        /* Select an accelerator device */
+  pb_Device_NAME                /* Select a device by name */
+};
+
+/* A device as specified by the user on the command line */
+struct pb_DeviceParam {
+  enum pb_DeviceSelectionCriterion criterion;
+  union {
+    int index;                  /* If criterion == pb_Device_INDEX,
+                                 * the index of the device */
+    char *name;                 /* If criterion == pb_Device_NAME,
+                                 * the name of the device.
+                                 * This string is owned. */
+  };
+};
+
+struct pb_DeviceParam *
+pb_DeviceParam_index(int index);
+
+struct pb_DeviceParam *
+pb_DeviceParam_cpu(void);
+
+struct pb_DeviceParam *
+pb_DeviceParam_gpu(void);
+
+struct pb_DeviceParam *
+pb_DeviceParam_accelerator(void);
+
+/* Create a by-name device selection criterion.
+ * The string should have been allocated by malloc(), and it will will be
+ * owned by the returned object.
+ */
+struct pb_DeviceParam *
+pb_DeviceParam_name(char *name);
+
+void
+pb_FreeDeviceParam(struct pb_DeviceParam *);
+
+/* Command line parameters for benchmarks */
+struct pb_Parameters {
+  char *outFile;		/* If not NULL, the raw output of the
+				 * computation should be saved to this
+				 * file. The string is owned. */
+  char **inpFiles;		/* A NULL-terminated array of strings
+				 * holding the input file(s) for the
+				 * computation.  The array and strings
+				 * are owned. */
+  struct pb_PlatformParam *platform; /* If not NULL, the platform
+                                      * specified on the command line. */
+  struct pb_DeviceParam *device; /* If not NULL, the device
+                                      * specified on the command line. */
+};
+
+/* Read command-line parameters.
+ *
+ * The argc and argv parameters to main are read, and any parameters
+ * interpreted by this function are removed from the argument list.
+ *
+ * A new instance of struct pb_Parameters is returned.
+ * If there is an error, then an error message is printed on stderr
+ * and NULL is returned.
+ */
+struct pb_Parameters *
+pb_ReadParameters(int *_argc, char **argv);
+
+/* Free an instance of struct pb_Parameters.
+ */
+void
+pb_FreeParameters(struct pb_Parameters *p);
+
+void
+pb_FreeStringArray(char **);
+
+/* Count the number of input files in a pb_Parameters instance.
+ */
+int
+pb_Parameters_CountInputs(struct pb_Parameters *p);
+
+/* A time or duration. */
+//#if _POSIX_VERSION >= 200112L
+typedef unsigned long long pb_Timestamp; /* time in microseconds */
+//#else
+//# error "Timestamps not implemented"
+//#endif
+
+enum pb_TimerState {
+  pb_Timer_STOPPED,
+  pb_Timer_RUNNING,
+};
+
+struct pb_Timer {
+  enum pb_TimerState state;
+  pb_Timestamp elapsed;		/* Amount of time elapsed so far */
+  pb_Timestamp init;		/* Beginning of the current time interval,
+				 * if state is RUNNING.  End of the last 
+				 * recorded time interfal otherwise.  */
+};
+
+/* Reset a timer.
+ * Use this to initialize a timer or to clear
+ * its elapsed time.  The reset timer is stopped.
+ */
+void
+pb_ResetTimer(struct pb_Timer *timer);
+
+/* Start a timer.  The timer is set to RUNNING mode and
+ * time elapsed while the timer is running is added to
+ * the timer.
+ * The timer should not already be running.
+ */
+void
+pb_StartTimer(struct pb_Timer *timer);
+
+/* Stop a timer.
+ * This stops adding elapsed time to the timer.
+ * The timer should not already be stopped.
+ */
+void
+pb_StopTimer(struct pb_Timer *timer);
+
+/* Get the elapsed time in seconds. */
+double
+pb_GetElapsedTime(struct pb_Timer *timer);
+
+/* Execution time is assigned to one of these categories. */
+enum pb_TimerID {
+  pb_TimerID_NONE = 0,
+  pb_TimerID_IO,		/* Time spent in input/output */
+  pb_TimerID_KERNEL,		/* Time spent computing on the device, 
+				 * recorded asynchronously */
+  pb_TimerID_COPY,		/* Time spent synchronously moving data 
+				 * to/from device and allocating/freeing 
+				 * memory on the device */
+  pb_TimerID_DRIVER,		/* Time spent in the host interacting with the 
+				 * driver, primarily for recording the time 
+                                 * spent queueing asynchronous operations */
+  pb_TimerID_COPY_ASYNC,	/* Time spent in asynchronous transfers */
+  pb_TimerID_COMPUTE,		/* Time for all program execution other
+				 * than parsing command line arguments,
+				 * I/O, kernel, and copy */
+  pb_TimerID_OVERLAP,		/* Time double-counted in asynchronous and 
+				 * host activity: automatically filled in, 
+				 * not intended for direct usage */
+  pb_TimerID_LAST		/* Number of timer IDs */
+};
+
+/* Dynamic list of asynchronously tracked times between events */
+struct pb_async_time_marker_list {
+  char *label; // actually just a pointer to a string
+  enum pb_TimerID timerID;	/* The ID to which the interval beginning 
+                                 * with this marker should be attributed */
+  void * marker; 
+  //cudaEvent_t marker; 		/* The driver event for this marker */
+  struct pb_async_time_marker_list *next; 
+};
+
+struct pb_SubTimer {
+  char *label;
+  struct pb_Timer timer;
+  struct pb_SubTimer *next;
+};
+
+struct pb_SubTimerList {
+  struct pb_SubTimer *current;
+  struct pb_SubTimer *subtimer_list;
+};
+
+/* A set of timers for recording execution times. */
+struct pb_TimerSet {
+  enum pb_TimerID current;
+  struct pb_async_time_marker_list* async_markers;
+  pb_Timestamp async_begin;
+  pb_Timestamp wall_begin;
+  struct pb_Timer timers[pb_TimerID_LAST];
+  struct pb_SubTimerList *sub_timer_list[pb_TimerID_LAST];
+};
+
+/* Reset all timers in the set. */
+void
+pb_InitializeTimerSet(struct pb_TimerSet *timers);
+
+void
+pb_AddSubTimer(struct pb_TimerSet *timers, char *label, enum pb_TimerID pb_Category);
+
+/* Select which timer the next interval of time should be accounted
+ * to. The selected timer is started and other timers are stopped.
+ * Using pb_TimerID_NONE stops all timers. */
+void
+pb_SwitchToTimer(struct pb_TimerSet *timers, enum pb_TimerID timer);
+
+void
+pb_SwitchToSubTimer(struct pb_TimerSet *timers, char *label, enum pb_TimerID category);
+
+/* Print timer values to standard output. */
+void
+pb_PrintTimerSet(struct pb_TimerSet *timers);
+
+/* Release timer resources */
+void
+pb_DestroyTimerSet(struct pb_TimerSet * timers);
+
+void
+pb_SetOpenCL(void *clContextPtr, void *clCommandQueuePtr);
+
+
+typedef struct pb_Device_tag {
+  char* name;
+  void* clDevice;
+  int id;
+  unsigned int in_use;
+  unsigned int available;
+} pb_Device;
+
+struct pb_Context_tag;
+typedef struct pb_Context_tag pb_Context;
+
+typedef struct pb_Platform_tag {
+  char* name;
+  char* version;
+  void* clPlatform;
+  unsigned int in_use;
+  pb_Context** contexts;
+  pb_Device** devices;
+} pb_Platform;
+
+struct pb_Context_tag {
+  void* clPlatformId;
+  void* clContext;
+  void* clDeviceId;
+  pb_Platform* pb_platform;
+  pb_Device* pb_device;
+};
+
+// verbosely print out list of platforms and their devices to the console.
+pb_Platform**
+pb_GetPlatforms();
+
+// Choose a platform according to the given platform specification
+pb_Platform*
+pb_GetPlatform(struct pb_PlatformParam *platform);
+
+// choose a platform: by name, name & version
+pb_Platform*
+pb_GetPlatformByName(const char* name);
+
+pb_Platform*
+pb_GetPlatformByNameAndVersion(const char* name, const char* version);
+
+// Choose a device according to the given device specification
+pb_Device*
+pb_GetDevice(pb_Platform* pb_platform, struct pb_DeviceParam *device);
+
+pb_Device**
+pb_GetDevices(pb_Platform* pb_platform);
+
+// choose a device by name.
+pb_Device*
+pb_GetDeviceByName(pb_Platform* pb_platform, const char* name);
+
+pb_Platform*
+pb_GetPlatformByEnvVars();
+
+pb_Context*
+pb_InitOpenCLContext(struct pb_Parameters* parameters);
+
+void
+pb_ReleasePlatforms();
+
+void
+pb_ReleaseContext(pb_Context* c);
+
+void
+pb_PrintPlatformInfo(pb_Context* c);
+
+void
+perf_init();
+
+//#define MEASURE_KERNEL_TIME
+
+#include <CL/cl.h>
+
+#ifdef MEASURE_KERNEL_TIME
+#define clEnqueueNDRangeKernel(q,k,d,o,dg,db,a,b,c) pb_clEnqueueNDRangeKernel((q), (k), (d), (o), (dg), (db), (a), (b), (c))
+cl_int
+pb_clEnqueueNDRangeKernel(cl_command_queue /* command_queue */,
+                       cl_kernel        /* kernel */,
+                       cl_uint          /* work_dim */,
+                       const size_t *   /* global_work_offset */,
+                       const size_t *   /* global_work_size */,
+                       const size_t *   /* local_work_size */,
+                       cl_uint          /* num_events_in_wait_list */,
+                       const cl_event * /* event_wait_list */,
+                       cl_event *       /* event */);
+#endif
+
+enum { T_FLOAT, T_DOUBLE, T_SHORT, T_INT, T_UCHAR };
+void pb_sig_float(char*, float*, int);
+void pb_sig_double(char*, double*, int);
+void pb_sig_short(char*, short*, int);
+void pb_sig_int(char*, int*, int);
+void pb_sig_uchar(char*, unsigned char*, unsigned int);
+void pb_sig_clmem(char*, cl_command_queue, cl_mem, int);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif //PARBOIL_HEADER
+

tests/opencl/cutcp/readatom.c

@@ -0,0 +1,139 @@
+/***************************************************************************
+ *cr
+ *cr            (C) Copyright 2008-2010 The Board of Trustees of the
+ *cr                        University of Illinois
+ *cr                         All Rights Reserved
+ *cr
+ ***************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include "atom.h"
+
+
+#define LINELEN 96
+#define INITLEN 20
+
+
+Atoms *read_atom_file(const char *fname)
+{
+  FILE *file;
+  char line[LINELEN];
+
+  Atom *atom;			/* Atom array */
+  int len = INITLEN;		/* Size of atom array */
+  int cnt = 0;			/* Number of atoms read */
+
+  /* allocate initial atom array */
+  atom = (Atom *) malloc(len * sizeof(Atom));
+  if (NULL==atom) {
+    fprintf(stderr, "can't allocate memory\n");
+    return NULL;
+  }
+
+  int i;
+  for (i = 0; i < len; ++i) {
+    atom[i].x = i+0;
+    atom[i].y = i+1;
+    atom[i].z = i+2;
+    atom[i].q = 1;
+  }
+
+#if 0
+  /* open atom "pqr" file */
+  file = fopen(fname, "r");
+  if (NULL==file) {
+    fprintf(stderr, "can't open file \"%s\" for reading\n", fname);
+    return NULL;
+  }
+
+  /* loop to read pqr file line by line */
+  while (fgets(line, LINELEN, file) != NULL) {
+
+    if (strncmp(line, "ATOM  ", 6) != 0 && strncmp(line, "HETATM", 6) != 0) {
+      continue;  /* skip anything that isn't an atom record */
+    }
+
+    if (cnt==len) {  /* extend atom array */
+      void *tmp = realloc(atom, 2*len*sizeof(Atom));
+      if (NULL==tmp) {
+        fprintf(stderr, "can't allocate more memory\n");
+        return NULL;
+      }
+      atom = (Atom *) tmp;
+      len *= 2;
+    }
+
+    /* read position coordinates and charge from atom record */
+    if (sscanf(line, "%*s %*d %*s %*s %*d %f %f %f %f", &(atom[cnt].x),
+          &(atom[cnt].y), &(atom[cnt].z), &(atom[cnt].q)) != 4) {
+      fprintf(stderr, "atom record %d does not have expected format\n", cnt+1);
+      return NULL;
+    }
+
+    cnt++;  /* count atoms as we store them */
+  }
+
+  /* verify EOF and close file */
+  if ( !feof(file) ) {
+    fprintf(stderr, "did not find EOF\n");
+    return NULL;
+  }
+  if (fclose(file)) {
+    fprintf(stderr, "can't close file\n");
+    return NULL;
+  }
+#endif
+
+  /* Build the output data structure */
+  {
+    Atoms *out = (Atoms *)malloc(sizeof(Atoms));
+
+    if (NULL == out) {
+      fprintf(stderr, "can't allocate memory\n");
+      return NULL;
+    }
+
+    out->size = cnt;
+    out->atoms = atom;
+
+    return out;
+  }
+}
+
+
+void free_atom(Atoms *atom)
+{
+  if (atom) {
+    free(atom->atoms);
+    free(atom);
+  }
+}
+
+void
+get_atom_extent(Vec3 *out_lo, Vec3 *out_hi, Atoms *atom)
+{
+  Atom *atoms = atom->atoms;
+  int natoms = atom->size;
+  Vec3 lo;
+  Vec3 hi;
+  int n;
+
+  hi.x = lo.x = atoms[0].x;
+  hi.y = lo.y = atoms[0].y;
+  hi.z = lo.z = atoms[0].z;
+
+  for (n = 1; n < natoms; n++) {
+    lo.x = fminf(lo.x, atoms[n].x);
+    hi.x = fmaxf(hi.x, atoms[n].x);
+    lo.y = fminf(lo.y, atoms[n].y);
+    hi.y = fmaxf(hi.y, atoms[n].y);
+    lo.z = fminf(lo.z, atoms[n].z);
+    hi.z = fmaxf(hi.z, atoms[n].z);
+  }
+
+  *out_lo = lo;
+  *out_hi = hi;
+}

tests/opencl/guassian/Makefile

@@ -0,0 +1,62 @@
+LLVM_PREFIX ?= /opt/llvm-riscv
+RISCV_TOOLCHAIN_PATH ?= /opt/riscv-gnu-toolchain
+SYSROOT ?= $(RISCV_TOOLCHAIN_PATH)/riscv32-unknown-elf
+POCL_CC_PATH ?= /opt/pocl/compiler
+POCL_RT_PATH ?= /opt/pocl/runtime
+
+VORTEX_DRV_PATH ?= $(realpath ../../../driver)
+VORTEX_RT_PATH ?= $(realpath ../../../runtime)
+
+K_LLCFLAGS += "-O3 -march=riscv32 -target-abi=ilp32f -mcpu=generic-rv32 -mattr=+m,+f -float-abi=hard -code-model=small"
+K_CFLAGS   += "-v -O3 --sysroot=$(SYSROOT) --gcc-toolchain=$(RISCV_TOOLCHAIN_PATH) -march=rv32imf -mabi=ilp32f -I$(VORTEX_RT_PATH)/include -fno-rtti -fno-exceptions -ffreestanding -nostartfiles -fdata-sections -ffunction-sections"
+K_LDFLAGS  += "-Wl,-Bstatic,-T$(VORTEX_RT_PATH)/linker/vx_link.ld -Wl,--gc-sections $(VORTEX_RT_PATH)/libvortexrt.a -lm"
+
+CXXFLAGS += -std=c++11 -O2 -Wall -Wextra -Wfatal-errors
+#CXXFLAGS += -std=c++11 -O0 -g -Wall -Wextra -Wfatal-errors
+
+CXXFLAGS += -Wno-deprecated-declarations -Wno-unused-parameter
+CXXFLAGS += -Wno-unused-variable -Wno-narrowing -Wno-unused-result -Wno-unused-but-set-variable
+
+CXXFLAGS += -I$(POCL_RT_PATH)/include
+
+LDFLAGS += -L$(POCL_RT_PATH)/lib -L$(VORTEX_DRV_PATH)/stub -lOpenCL -lvortex
+
+PROJECT = guassian
+
+SRCS = main.cc clutils.cpp utils.cpp
+
+all: $(PROJECT) kernel.pocl
+
+kernel.pocl: kernel.cl
+	LLVM_PREFIX=$(LLVM_PREFIX) POCL_DEBUG=all LD_LIBRARY_PATH=$(LLVM_PREFIX)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -LLCFLAGS $(K_LLCFLAGS) -CFLAGS $(K_CFLAGS) -LDFLAGS $(K_LDFLAGS) -o kernel.pocl kernel.cl
+ 
+$(PROJECT): $(SRCS)
+	$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -o $@
+
+run-fpga: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-asesim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT)
+	
+run-vlsim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae/vlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-simx: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/simx:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-rtlsim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+.depend: $(SRCS)
+	$(CXX) $(CXXFLAGS) -MM $^ > .depend;
+
+clean:
+	rm -rf $(PROJECT) *.o .depend
+
+clean-all: clean
+	rm -rf *.pocl *.dump
+
+ifneq ($(MAKECMDGOALS),clean)
+    -include .depend
+endif

tests/opencl/guassian/OriginalParallel.c

@@ -0,0 +1,241 @@
+/*-----------------------------------------------------------
+** ge_p.c -- The program is to solve a linear system Ax = b
+**   by using Gaussian Elimination. The algorithm on page 101
+**   ("Foundations of Parallel Programming") is used.  
+**   The sequential version is ge_s.c.  This parallel 
+**   implementation converts three independent for() loops 
+**   into three Fans.  Use the data file ge_3.dat to verify 
+**   the correction of the output. 
+**
+** Written by Andreas Kura, 02/15/95
+** Modified by Chong-wei Xu, /04/20/95
+**-----------------------------------------------------------
+*/
+#include <us.h>
+#include <stdio.h>
+
+int Size, t;
+float **a, *b;
+BEGIN_SHARED_DECL
+  float **m;
+END_SHARED_DECL;
+FILE *fp;
+
+void InitProblemOnce();
+void InitPerRun();
+void ForwardSub();
+void Fan1();
+void Fan2();
+void Fan3();
+void InitMat();
+void InitAry();
+void PrintMat();
+void PrintAry();
+
+main ()
+{
+  InitializeUs();
+  MakeSharedVariables;  /* to make SHARED m */
+
+  InitProblemOnce();
+  InitPerRun();
+  ForwardSub();
+
+  printf("The result of matrix m is: \n");
+  PrintMat(SHARED m, Size, Size);
+  printf("The result of matrix a is: \n");
+  PrintMat(a, Size, Size);
+  printf("The result of array b is: \n");
+  PrintAry(b, Size);
+}
+
+/*------------------------------------------------------
+** InitProblemOnce -- Initialize all of matrices and
+** vectors by opening a data file specified by the user.
+**
+** We used dynamic array **a, *b, and **m to allocate
+** the memory storages.
+**------------------------------------------------------
+*/
+void InitProblemOnce()
+{
+  char filename[30];
+ 
+  printf("Enter the data file name: ");
+  scanf("%s", filename);
+  printf("The file name is: %s\n", filename);
+ 
+  fp = fopen(filename, "r");
+ 
+  fscanf(fp, "%d", &Size);
+  a = (float **) UsAllocScatterMatrix(Size, Size, sizeof(float));
+  /* 
+  a = (float **) malloc(Size * sizeof(float *));
+  for (i=0; i<Size; i++) {
+    a[i] = (float *) malloc(Size * sizeof(float));
+  }
+  */
+  InitMat(a, Size, Size);
+  printf("The input matrix a is:\n");
+  PrintMat(a, Size, Size);
+ 
+  b = (float *) UsAlloc(Size * sizeof(float));
+  /* 
+  b = (float *) malloc(Size * sizeof(float));
+  */
+  InitAry(b, Size);
+  printf("The input array b is:\n");
+  PrintAry(b, Size);
+ 
+  SHARED m = (float **) UsAllocScatterMatrix(Size, Size, sizeof(float));
+  /*
+  m = (float **) malloc(Size * sizeof(float *));
+  for (i=0; i<Size; i++) {
+    m[i] = (float *) malloc(Size * sizeof(float));
+  }
+  */
+
+  Share(&Size);
+  Share(&a);
+  Share(&b);
+}
+
+/*------------------------------------------------------
+** InitPerRun() -- Initialize the contents of the
+** multipier matrix **m
+**------------------------------------------------------
+*/
+void InitPerRun() 
+{
+  int i, j;
+
+  for (i=0; i<Size; i++)
+    for (j=0; j<Size; j++) 
+      SHARED m[i][j] = 0.0;
+}
+
+/*------------------------------------------------------
+** ForwardSub() -- Forward substitution of Gaussian
+** elimination.
+**------------------------------------------------------
+*/
+void ForwardSub()
+{
+  for (t=0; t<(Size-1); t++) {
+    Share(&t);
+    GenOnI(Fan1, Size-1-t);  /* t=0 to (Size-2), the range is
+                             ** Size-2-t+1 = Size-1-t
+                             */   
+    GenOnA(Fan2, Size-1-t, Size-t);
+    GenOnI(Fan3, Size-1-t);
+  }
+}
+
+/*-------------------------------------------------------
+** Fan1() -- Calculate multiplier matrix
+** Pay attention to the index.  Index i give the range
+** which starts from 0 to range-1.  The real values of
+** the index should be adjust and related with the value
+** of t which is defined on the ForwardSub().
+**-------------------------------------------------------
+*/
+void Fan1(dummy, i)
+int dummy, i;
+{   
+  /* Use these printf() to display the nodes and index */
+  printf("from node #%d\n", PhysProcToUsProc(Proc_Node));
+  SHARED m[i+t+1][t] = a[i+t+1][t] / a[t][t];
+  printf("i=%d, a[%d][%d]=%.2f, a[%d][%d]=%.2f, m[%d][%d]=%.2f\n",
+     (i+t+1),t,t,a[t][t],(i+t+1),t,a[i+t+1][t],(i+t+1),t,
+     SHARED m[i+t+1][t]);
+}
+
+/*-------------------------------------------------------
+** Fan2() -- Modify the matrix A into LUD
+**-------------------------------------------------------
+*/ 
+void Fan2(dummy, i, j)
+int dummy, i, j;
+{
+  a[i+1+t][j+t] -= SHARED m[i+1+t][t] * a[t][j+t];
+  Share (&a);
+}
+
+/*-------------------------------------------------------
+** Fan3() -- Modify the array b
+**-------------------------------------------------------
+*/
+void Fan3(dummy, i)
+int dummy, i;
+{
+  b[i+1+t] -= SHARED m[i+1+t][t] * b[t];
+}
+
+/*------------------------------------------------------
+** InitMat() -- Initialize the matrix by reading data
+** from the data file
+**------------------------------------------------------
+*/
+void InitMat(ary, nrow, ncol)
+float **ary;
+int nrow, ncol;
+{
+  int i, j;
+
+  for (i=0; i<nrow; i++) {
+    for (j=0; j<ncol; j++) {
+      fscanf(fp, "%f",  &ary[i][j]);
+    }
+  }  
+}
+
+/*------------------------------------------------------
+** PrintMat() -- Print the contents of the matrix
+**------------------------------------------------------
+*/
+void PrintMat(ary, nrow, ncol)
+float **ary;
+int nrow, ncol;
+{
+  int i, j;
+ 
+  for (i=0; i<nrow; i++) {
+    for (j=0; j<ncol; j++) {
+      printf("%8.2f ", ary[i][j]);
+    }
+    printf("\n");
+  }
+  printf("\n");
+}
+
+/*------------------------------------------------------
+** InitAry() -- Initialize the array (vector) by reading
+** data from the data file
+**------------------------------------------------------
+*/
+void InitAry(ary, ary_size)
+float *ary;
+int ary_size;
+{
+  int i;
+ 
+  for (i=0; i<ary_size; i++) {
+    fscanf(fp, "%f",  &ary[i]);
+  }
+}  
+ 
+/*------------------------------------------------------
+** PrintAry() -- Print the contents of the array (vector)
+**------------------------------------------------------
+*/
+void PrintAry(ary, ary_size)
+float *ary;
+int ary_size;
+{
+  int i;
+ 
+  for (i=0; i<ary_size; i++) {
+    printf("%.2f ", ary[i]);
+  }
+  printf("\n");
+}

tests/opencl/guassian/README.txt

@@ -0,0 +1,60 @@
+The Gaussian Elimination application solves systems of equations using the
+gaussian elimination method.
+
+The application analyzes an n x n matrix and an associated 1 x n vector to solve a 
+set of equations with n variables and n unknowns. The matrix and vector describe equations
+of the form:
+
+             a0x + b0y + c0z + d0w = e0
+             a1x + b1y + c1z + d1w = e1
+             a2x + b2y + c2z + d2w = e2
+             a3x + b3y + c3z + d3w = e3
+
+where in this case n=4.  The matrix for the above equations would be as follows:
+
+            [a0 b0 c0 d0]
+            [a1 b1 c1 d1]
+            [a2 b2 c2 d2]
+            [a3 b3 c3 d3]
+            
+and the vector would be:
+
+            [e0]
+            [e1]
+            [e2]
+            [e3]
+
+The application creates a solution vector:
+
+            [x]
+            [y]
+            [z]
+            [w]
+            
+
+The Makefile may need to be adjusted for different machines, but it was written for Mac OS X and
+Linux with either NVIDIA or AMD OpenCL SDKs.
+
+Additional input files can be created with the matrixGenerator.py file in the data folder.
+
+Gaussian Elimination Usage
+
+    gaussianElimination [filename] [-hqt] [-p [int] -d [int]]
+    
+    example:
+    $ ./gaussianElimination matrix4.txt
+    
+    filename     the filename that holds the matrix data
+    
+    -h, --help   Display the help file
+    -q           Quiet mode. Suppress all text output.
+    -t           Print timing information.
+    
+    -p [int]     Choose the platform (must choose both platform and device)
+    -d [int]     Choose the device (must choose both platform and device)
+    
+    
+    Notes: 1. The filename is required as the first parameter.
+           2. If you declare either the device or the platform,
+              you must declare both.
+

tests/opencl/guassian/clutils.h

@@ -0,0 +1,281 @@
+/****************************************************************************\ 
+ * Copyright (c) 2011, Advanced Micro Devices, Inc.                           *
+ * All rights reserved.                                                       *
+ *                                                                            *
+ * Redistribution and use in source and binary forms, with or without         *
+ * modification, are permitted provided that the following conditions         *
+ * are met:                                                                   *
+ *                                                                            *
+ * Redistributions of source code must retain the above copyright notice,     *
+ * this list of conditions and the following disclaimer.                      *
+ *                                                                            *
+ * Redistributions in binary form must reproduce the above copyright notice,  *
+ * this list of conditions and the following disclaimer in the documentation  *
+ * and/or other materials provided with the distribution.                     *
+ *                                                                            *
+ * Neither the name of the copyright holder nor the names of its contributors *
+ * may be used to endorse or promote products derived from this software      *
+ * without specific prior written permission.                                 *
+ *                                                                            *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS        *
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED  *
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR *
+ * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR          *
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,      *
+ * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,        *
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR         *
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF     *
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING       *
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS         *
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.               *
+ *                                                                            *
+ * If you use the software (in whole or in part), you shall adhere to all     *
+ * applicable U.S., European, and other export laws, including but not        *
+ * limited to the U.S. Export Administration Regulations (“EAR”), (15 C.F.R.  *
+ * Sections 730 through 774), and E.U. Council Regulation (EC) No 1334/2000   *
+ * of 22 June 2000.  Further, pursuant to Section 740.6 of the EAR, you       *
+ * hereby certify that, except pursuant to a license granted by the United    *
+ * States Department of Commerce Bureau of Industry and Security or as        *
+ * otherwise permitted pursuant to a License Exception under the U.S. Export  *
+ * Administration Regulations ("EAR"), you will not (1) export, re-export or  *
+ * release to a national of a country in Country Groups D:1, E:1 or E:2 any   *
+ * restricted technology, software, or source code you receive hereunder,     *
+ * or (2) export to Country Groups D:1, E:1 or E:2 the direct product of such *
+ * technology or software, if such foreign produced direct product is subject *
+ * to national security controls as identified on the Commerce Control List   *
+ *(currently found in Supplement 1 to Part 774 of EAR).  For the most current *
+ * Country Group listings, or for additional information about the EAR or     *
+ * your obligations under those regulations, please refer to the U.S. Bureau  *
+ * of Industry and Security’s website at http://www.bis.doc.gov/.             *
+ \****************************************************************************/
+
+#ifndef __CL_UTILS_H__
+#define __CL_UTILS_H__
+
+#include <CL/cl.h>
+
+// The cl_time type is OS specific
+#ifdef _WIN32
+#include <tchar.h>
+#include <Windows.h>
+typedef __int64 cl_time; 
+#else
+#include <sys/time.h>
+typedef double cl_time;
+#endif
+
+//-------------------------------------------------------
+// Initialization and Cleanup
+//-------------------------------------------------------
+
+// Detects platforms and devices, creates context and command queue
+cl_context cl_init(char devicePreference='\0');
+
+// Creates a context given a platform and a device
+cl_context cl_init_context(int platform,int dev,int quiet=0);
+
+// Releases resources used by clutils
+void    cl_cleanup();
+
+// Releases a kernel object
+void    cl_freeKernel(cl_kernel kernel);
+
+// Releases a memory object
+void    cl_freeMem(cl_mem mem);
+
+// Releases a program object
+void    cl_freeProgram(cl_program program);
+
+// Returns the global command queue
+cl_command_queue cl_getCommandQueue();
+
+
+//-------------------------------------------------------
+// Synchronization functions
+//-------------------------------------------------------
+
+// Performs a clFinish on the command queue
+void    cl_sync();
+
+
+//-------------------------------------------------------
+// Memory allocation
+//-------------------------------------------------------
+
+// Allocates a regular buffer on the device
+cl_mem  cl_allocBuffer(size_t mem_size, 
+            cl_mem_flags flags = CL_MEM_READ_WRITE);
+
+// XXX I don't think this does exactly what we want it to do
+// Allocates a read-only buffer and transfers the data
+cl_mem  cl_allocBufferConst(size_t mem_size, void* host_ptr);
+
+// Allocates pinned memory on the host
+cl_mem  cl_allocBufferPinned(size_t mem_size);
+
+// Allocates an image on the device
+cl_mem  cl_allocImage(size_t height, size_t width, char type, 
+            cl_mem_flags flags = CL_MEM_READ_WRITE);
+
+
+
+//-------------------------------------------------------
+// Data transfers
+//-------------------------------------------------------
+
+// Copies a buffer from the device to pinned memory on the host and 
+// maps it so it can be read
+void*   cl_copyAndMapBuffer(cl_mem dst, cl_mem src, size_t size); 
+
+// Copies from one buffer to another
+void    cl_copyBufferToBuffer(cl_mem dst, cl_mem src, size_t size);
+
+// Copies data to a buffer on the device
+void    cl_copyBufferToDevice(cl_mem dst, void *src, size_t mem_size, 
+            cl_bool blocking = CL_TRUE);
+
+// Copies data to an image on the device
+void cl_copyImageToDevice(cl_mem dst, void* src, size_t height, size_t width);
+
+// Copies an image from the device to the host
+void    cl_copyImageToHost(void* dst, cl_mem src, size_t height, size_t width);
+
+// Copies data from a device buffer to the host
+void    cl_copyBufferToHost(void *dst, cl_mem src, size_t mem_size, 
+            cl_bool blocking = CL_TRUE);
+
+// Copies data from a buffer on the device to an image on the device
+void    cl_copyBufferToImage(cl_mem src, cl_mem dst, int height, int width);
+
+// Maps a buffer
+void*   cl_mapBuffer(cl_mem mem, size_t mem_size, cl_mem_flags flags);
+
+// Unmaps a buffer
+void    cl_unmapBuffer(cl_mem mem, void *ptr);
+
+// Writes data to a zero-copy buffer on the device
+void    cl_writeToZCBuffer(cl_mem mem, void* data, size_t size);
+
+//-------------------------------------------------------
+// Program and kernels
+//-------------------------------------------------------
+
+// Compiles a program
+cl_program  cl_compileProgram(char* kernelPath, char* compileoptions, 
+                bool verboseoptions = 0);
+
+// Creates a kernel
+cl_kernel   cl_createKernel(cl_program program, const char* kernelName);
+
+
+// Sets a kernel argument
+void        cl_setKernelArg(cl_kernel kernel, unsigned int index, size_t size, 
+                void* data);
+
+
+//-------------------------------------------------------
+// Profiling/events
+//-------------------------------------------------------
+
+// Computes the execution time (start to end) for an event
+double  cl_computeExecTime(cl_event);
+
+// Compute the elapsed time between two CPU timer values
+double  cl_computeTime(cl_time start, cl_time end); 
+
+// Creates an event from CPU timers
+void    cl_createUserEvent(cl_time start, cl_time end, char* desc);
+
+// Disable logging of events
+void    cl_disableEvents();
+
+// Enable logging of events
+void    cl_enableEvents();
+
+// Query the current system time
+void    cl_getTime(cl_time* time); 
+
+// Calls a function which prints events to the terminal
+void    cl_printEvents();
+
+// Calls a function which writes the events to a file
+void    cl_writeEventsToFile(char* path);
+
+
+//-------------------------------------------------------
+// Error handling
+//-------------------------------------------------------
+
+// Compare a status value to CL_SUCCESS and optionally exit on error
+int     cl_errChk(const cl_int status, const char *msg, bool exitOnErr);
+
+// Queries the supported image formats for the device and prints
+// them to the screen
+void    printSupportedImageFormats();
+
+//-------------------------------------------------------
+// Platform and device information
+//-------------------------------------------------------
+
+bool    cl_deviceIsAMD(cl_device_id dev=NULL);
+bool    cl_deviceIsNVIDIA(cl_device_id dev=NULL);
+bool    cl_platformIsNVIDIA(cl_platform_id plat=NULL);
+char*   cl_getDeviceDriverVersion(cl_device_id dev=NULL);
+char*   cl_getDeviceName(cl_device_id dev=NULL);
+char*   cl_getDeviceVendor(cl_device_id dev=NULL);
+char*   cl_getDeviceVersion(cl_device_id dev=NULL);
+char*   cl_getPlatformName(cl_platform_id platform);
+char*   cl_getPlatformVendor(cl_platform_id platform);
+
+//-------------------------------------------------------
+// Utility functions
+//-------------------------------------------------------
+
+char* catStringWithInt(const char* str, int integer);
+
+char* itoa_portable(int value, char* result, int base);
+
+//-------------------------------------------------------
+// Data types
+//-------------------------------------------------------
+typedef struct{
+    int x;
+    int y;
+} int2;
+
+typedef struct{
+    float x;
+    float y;
+}float2;
+
+typedef struct{
+    float x;
+    float y;
+    float z;
+    float w;
+}float4;
+
+//-------------------------------------------------------
+// Defines
+//-------------------------------------------------------
+
+#define MAX_ERR_VAL 64
+
+#define NUM_PROGRAMS 7
+
+#define NUM_KERNELS 13
+#define KERNEL_INIT_DET 0 
+#define KERNEL_BUILD_DET 1 
+#define KERNEL_SURF_DESC 2
+#define KERNEL_NORM_DESC 3
+#define KERNEL_NON_MAX_SUP 4
+#define KERNEL_GET_ORIENT1 5
+#define KERNEL_GET_ORIENT2 6
+#define KERNEL_NN 7
+#define KERNEL_SCAN 8
+#define KERNEL_SCAN4 9
+#define KERNEL_TRANSPOSE 10
+#define KERNEL_SCANIMAGE 11
+#define KERNEL_TRANSPOSEIMAGE 12
+
+#endif

tests/opencl/guassian/gaussianElim.h

@@ -0,0 +1,40 @@
+#ifndef _GAUSSIANELIM
+#define _GAUSSIANELIM
+
+#include <iostream>
+#include <vector>
+#include <float.h>
+#include <string.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <algorithm>
+
+#include "clutils.h"
+
+// All OpenCL headers
+#if defined (__APPLE__) || defined(MACOSX)
+    #include <OpenCL/opencl.h>
+#else
+    #include <CL/opencl.h>
+#endif 
+
+float *OpenClGaussianElimination(
+	cl_context context,
+	int timing);
+
+void printUsage();
+int parseCommandline(int argc, char *argv[], char* filename,
+                     int *q, int *t, int *p, int *d);
+                     
+void InitPerRun(int size,float *m);
+void ForwardSub(cl_context context, float *a, float *b, float *m, int size,int timing);
+void BackSub(float *a, float *b, float *finalVec, int size);
+void Fan1(float *m, float *a, int Size, int t);
+void Fan2(float *m, float *a, float *b,int Size, int j1, int t);
+//void Fan3(float *m, float *b, int Size, int t);
+void InitMat(FILE *fp, int size, float *ary, int nrow, int ncol);
+void InitAry(FILE *fp, float *ary, int ary_size);
+void PrintMat(float *ary, int size, int nrow, int ncolumn);
+void PrintAry(float *ary, int ary_size);
+float eventTime(cl_event event,cl_command_queue command_queue);
+#endif

tests/opencl/guassian/gettimeofday.cpp

@@ -0,0 +1,74 @@
+#include "stdio.h"
+#include <time.h>
+#include <windows.h>
+#include <iostream> 
+//using namespace System;
+using namespace std;
+ 
+#if defined(_MSC_VER) || defined(_MSC_EXTENSIONS)
+  #define DELTA_EPOCH_IN_MICROSECS  11644473600000000Ui64
+#else
+  #define DELTA_EPOCH_IN_MICROSECS  11644473600000000ULL
+#endif
+ 
+struct timezone
+{
+  int  tz_minuteswest; /* minutes W of Greenwich */
+  int  tz_dsttime;     /* type of dst correction */
+};
+ 
+
+// Definition of a gettimeofday function
+ int gettimeofday(struct timeval *tv, struct timezone *tz)
+{
+// Define a structure to receive the current Windows filetime
+  FILETIME ft;
+ 
+// Initialize the present time to 0 and the timezone to UTC
+  unsigned __int64 tmpres = 0;
+  static int tzflag = 0;
+ 
+  if (NULL != tv)
+  {
+    GetSystemTimeAsFileTime(&ft);
+ 
+// The GetSystemTimeAsFileTime returns the number of 100 nanosecond 
+// intervals since Jan 1, 1601 in a structure. Copy the high bits to 
+// the 64 bit tmpres, shift it left by 32 then or in the low 32 bits.
+    tmpres |= ft.dwHighDateTime;
+    tmpres <<= 32;
+    tmpres |= ft.dwLowDateTime;
+ 
+// Convert to microseconds by dividing by 10
+    tmpres /= 10;
+ 
+// The Unix epoch starts on Jan 1 1970.  Need to subtract the difference 
+// in seconds from Jan 1 1601.
+    tmpres -= DELTA_EPOCH_IN_MICROSECS;
+ 
+// Finally change microseconds to seconds and place in the seconds value. 
+// The modulus picks up the microseconds.
+    tv->tv_sec = (long)(tmpres / 1000000UL);
+    tv->tv_usec = (long)(tmpres % 1000000UL);
+  }
+ 
+  if (NULL != tz)
+  {
+    if (!tzflag)
+    {
+      _tzset();
+      tzflag++;
+    }
+  
+// Adjust for the timezone west of Greenwich
+	long seconds_diff;
+	_get_timezone(&seconds_diff);
+      tz->tz_minuteswest = seconds_diff / 60;
+	int hours_offset;
+	_get_daylight(&hours_offset);
+    tz->tz_dsttime = hours_offset;
+  }
+ 
+  return 0;
+}
+    

tests/opencl/guassian/gettimeofday.h

@@ -0,0 +1,17 @@
+
+#ifdef _WIN32
+#include <WinSock.h>
+/**
+Based on code seen at.
+
+http://www.winehq.org/pipermail/wine-devel/2003-June/018082.html
+
+http://msdn.microsoft.com/en-us/library/ms740560
+
+*/
+int gettimeofday(struct timeval *tv, struct timezone *tz);
+#else
+#include <sys/time.h>
+#endif
+
+

tests/opencl/guassian/kernel.cl

@@ -0,0 +1,49 @@
+//#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
+
+typedef struct latLong
+    {
+        float lat;
+        float lng;
+    } LatLong;
+
+__kernel void Fan1(__global float *m_dev,
+                  __global float *a_dev,
+                  __global float *b_dev,
+                  const int size,
+                  const int t) {
+    int globalId = get_global_id(0);
+                              
+    if (globalId < size-1-t) {
+         *(m_dev + size * (globalId + t + 1)+t) = *(a_dev + size * (globalId + t + 1) + t) / *(a_dev + size * t + t);    
+    }
+}
+
+
+__kernel void Fan2(__global float *m_dev,
+                  __global float *a_dev,
+                  __global float *b_dev,
+                  const int size,
+                  const int t) {
+	 int globalId = get_global_id(0);
+	 
+	 int globalIdx = get_global_id(0);
+	 int globalIdy = get_global_id(1);
+      if (globalIdx < size-1-t && globalIdy < size-t) {
+         a_dev[size*(globalIdx+1+t)+(globalIdy+t)] -= m_dev[size*(globalIdx+1+t)+t] * a_dev[size*t+(globalIdy+t)];
+ 	 
+ 	    if(globalIdy == 0){
+ 		   b_dev[globalIdx+1+t] -= m_dev[size*(globalIdx+1+t)+(globalIdy+t)] * b_dev[t];
+ 	    }
+ 	 }
+//   One dimensional
+// 	 int globalIdx = globalId % size;
+// 	 int globalIdy = globalId / size;
+// 	 
+// 	 if (globalIdx < size-1-t && globalIdy < size-t) {
+//          a_dev[size*(globalIdx+1+t)+(globalIdy+t)] -= m_dev[size*(globalIdx+1+t)+t] * a_dev[size*t+(globalIdy+t)];
+// 	 }
+// 	 if(globalIdy == 0){
+//  		   b_dev[globalIdx+1+t] -= m_dev[size*(globalIdx+1+t)+(globalIdy+t)] * b_dev[t];
+//      }
+    
+}

tests/opencl/guassian/main.cc

@@ -0,0 +1,422 @@
+#ifndef __GAUSSIAN_ELIMINATION__
+#define __GAUSSIAN_ELIMINATION__
+
+#include "gaussianElim.h"
+
+cl_context context = NULL;
+
+int main(int argc, char *argv[]) {
+  printf("enter demo main\n");
+  float *a = NULL, *b = NULL, *finalVec = NULL;
+  float *m = NULL;
+  int size;
+
+  FILE *fp;
+
+  // args
+  char filename[100];
+  int quiet = 0, timing = 0, platform = -1, device = -1;
+
+  // parse command line
+  if (parseCommandline(argc, argv, filename, &quiet, &timing, &platform,
+                       &device)) {
+    printUsage();
+    return 0;
+  }
+
+  context = cl_init_context(platform, device, quiet);
+
+  fp = fopen(filename, "r");
+  fscanf(fp, "%d", &size);
+
+  a = (float *)malloc(size * size * sizeof(float));
+
+  printf("OK\n");
+
+  InitMat(fp, size, a, size, size);
+  // printf("The input matrix a is:\n");
+  // PrintMat(a, size, size, size);
+  b = (float *)malloc(size * sizeof(float));
+
+  InitAry(fp, b, size);
+  // printf("The input array b is:\n");
+  // PrintAry(b, size);
+
+  // create the solution matrix
+  m = (float *)malloc(size * size * sizeof(float));
+
+  // create a new vector to hold the final answer
+  finalVec = (float *)malloc(size * sizeof(float));
+
+  InitPerRun(size, m);
+
+  // begin timing
+
+  // run kernels
+  ForwardSub(context, a, b, m, size, timing);
+
+  // end timing
+  if (!quiet) {
+    printf("The result of matrix m is: \n");
+
+    PrintMat(m, size, size, size);
+    printf("The result of matrix a is: \n");
+    PrintMat(a, size, size, size);
+    printf("The result of array b is: \n");
+    PrintAry(b, size);
+
+    BackSub(a, b, finalVec, size);
+    printf("The final solution is: \n");
+    PrintAry(finalVec, size);
+  }
+
+  fclose(fp);
+  free(m);
+  free(a);
+  free(b);
+  free(finalVec);
+  // OpenClGaussianElimination(context,timing);
+
+  cl_cleanup();
+
+  printf("Passed!\n");
+  return 0;
+}
+
+/*------------------------------------------------------
+ ** ForwardSub() -- Forward substitution of Gaussian
+ ** elimination.
+ **------------------------------------------------------
+ */
+void ForwardSub(cl_context context, float *a, float *b, float *m, int size,
+                int timing) {
+  // 1. set up kernels
+  cl_kernel fan1_kernel, fan2_kernel;
+  cl_int status = 0;
+  cl_program gaussianElim_program;
+  cl_event writeEvent, kernelEvent, readEvent;
+  float writeTime = 0, readTime = 0, kernelTime = 0;
+  float writeMB = 0, readMB = 0;
+  
+  gaussianElim_program = cl_compileProgram((char *)"gaussianElim_kernels.cl", NULL);
+  
+  fan1_kernel = clCreateKernel(gaussianElim_program, "Fan1", &status);
+  status = cl_errChk(status, (char *)"Error Creating Fan1 kernel", true);
+  if (status)
+    exit(1);
+
+  fan2_kernel = clCreateKernel(gaussianElim_program, "Fan2", &status);
+  status = cl_errChk(status, (char *)"Error Creating Fan2 kernel", true);
+  if (status)
+    exit(1);
+
+  // 2. set up memory on device and send ipts data to device
+
+  cl_mem a_dev, b_dev, m_dev;
+
+  cl_int error = 0;
+
+  a_dev = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                         sizeof(float) * size * size, NULL, &error);
+
+  b_dev = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(float) * size, NULL,
+                         &error);
+
+  m_dev = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                         sizeof(float) * size * size, NULL, &error);
+
+  cl_command_queue command_queue = cl_getCommandQueue();
+
+  error = clEnqueueWriteBuffer(command_queue, a_dev,
+                               1, // change to 0 for nonblocking write
+                               0, // offset
+                               sizeof(float) * size * size, a, 0, NULL,
+                               &writeEvent);
+
+  if (timing)
+    writeTime += eventTime(writeEvent, command_queue);
+  clReleaseEvent(writeEvent);
+
+  error = clEnqueueWriteBuffer(command_queue, b_dev,
+                               1, // change to 0 for nonblocking write
+                               0, // offset
+                               sizeof(float) * size, b, 0, NULL, &writeEvent);
+  if (timing)
+    writeTime += eventTime(writeEvent, command_queue);
+  clReleaseEvent(writeEvent);
+
+  error = clEnqueueWriteBuffer(command_queue, 
+                               m_dev,
+                               1, // change to 0 for nonblocking write
+                               0, // offset
+                               sizeof(float) * size * size, m, 0, NULL,
+                               &writeEvent);
+  if (timing)
+    writeTime += eventTime(writeEvent, command_queue);
+  clReleaseEvent(writeEvent);
+  writeMB = (float)(sizeof(float) * size * (size + size + 1) / 1e6);
+
+  // 3. Determine block sizes
+  size_t globalWorksizeFan1[1];
+  size_t globalWorksizeFan2[2];
+
+  globalWorksizeFan1[0] = size;
+  globalWorksizeFan2[0] = size;
+  globalWorksizeFan2[1] = size;
+
+  int t;
+  // 4. Setup and Run kernels
+  for (t = 0; t < (size - 1); t++) {
+    // kernel args
+    cl_int argchk;
+    argchk = clSetKernelArg(fan1_kernel, 0, sizeof(cl_mem), (void *)&m_dev);
+    argchk |= clSetKernelArg(fan1_kernel, 1, sizeof(cl_mem), (void *)&a_dev);
+    argchk |= clSetKernelArg(fan1_kernel, 2, sizeof(cl_mem), (void *)&b_dev);
+    argchk |= clSetKernelArg(fan1_kernel, 3, sizeof(int), (void *)&size);
+    argchk |= clSetKernelArg(fan1_kernel, 4, sizeof(int), (void *)&t);
+
+    cl_errChk(argchk, "ERROR in Setting Fan1 kernel args", true);
+
+    // launch kernel
+    error =
+        clEnqueueNDRangeKernel(command_queue, fan1_kernel, 1, 0,
+                               globalWorksizeFan1, NULL, 0, NULL, &kernelEvent);
+
+    cl_errChk(error, "ERROR in Executing Fan1 Kernel", true);
+    if (timing) {
+      //             printf("here1a\n");
+      kernelTime += eventTime(kernelEvent, command_queue);
+      //             printf("here1b\n");
+    }
+    clReleaseEvent(kernelEvent);
+    // Fan1<<<dimGrid,dimBlock>>>(m_cuda,a_cuda,Size,t);
+    // cudaThreadSynchronize();
+
+    // kernel args
+    argchk = clSetKernelArg(fan2_kernel, 0, sizeof(cl_mem), (void *)&m_dev);
+    argchk |= clSetKernelArg(fan2_kernel, 1, sizeof(cl_mem), (void *)&a_dev);
+    argchk |= clSetKernelArg(fan2_kernel, 2, sizeof(cl_mem), (void *)&b_dev);
+    argchk |= clSetKernelArg(fan2_kernel, 3, sizeof(int), (void *)&size);
+    argchk |= clSetKernelArg(fan2_kernel, 4, sizeof(int), (void *)&t);
+
+    cl_errChk(argchk, "ERROR in Setting Fan2 kernel args", true);
+
+    // launch kernel
+    error =
+        clEnqueueNDRangeKernel(command_queue, fan2_kernel, 2, 0,
+                               globalWorksizeFan2, NULL, 0, NULL, &kernelEvent);
+
+    cl_errChk(error, "ERROR in Executing Fan1 Kernel", true);
+    if (timing) {
+      //             printf("here2a\n");
+      kernelTime += eventTime(kernelEvent, command_queue);
+      //             printf("here2b\n");
+    }
+    clReleaseEvent(kernelEvent);
+    // Fan2<<<dimGridXY,dimBlockXY>>>(m_cuda,a_cuda,b_cuda,Size,Size-t,t);
+    // cudaThreadSynchronize();
+  }
+  // 5. transfer data off of device
+  error =
+      clEnqueueReadBuffer(command_queue, a_dev,
+                          1, // change to 0 for nonblocking write
+                          0, // offset
+                          sizeof(float) * size * size, a, 0, NULL, &readEvent);
+
+  cl_errChk(error, "ERROR with clEnqueueReadBuffer", true);
+  if (timing)
+    readTime += eventTime(readEvent, command_queue);
+  clReleaseEvent(readEvent);
+
+  error = clEnqueueReadBuffer(command_queue, b_dev,
+                              1, // change to 0 for nonblocking write
+                              0, // offset
+                              sizeof(float) * size, b, 0, NULL, &readEvent);
+  cl_errChk(error, "ERROR with clEnqueueReadBuffer", true);
+  if (timing)
+    readTime += eventTime(readEvent, command_queue);
+  clReleaseEvent(readEvent);
+
+  error =
+      clEnqueueReadBuffer(command_queue, m_dev,
+                          1, // change to 0 for nonblocking write
+                          0, // offset
+                          sizeof(float) * size * size, m, 0, NULL, &readEvent);
+
+  cl_errChk(error, "ERROR with clEnqueueReadBuffer", true);
+  if (timing)
+    readTime += eventTime(readEvent, command_queue);
+  clReleaseEvent(readEvent);
+  readMB = (float)(sizeof(float) * size * (size + size + 1) / 1e6);
+
+  if (timing) {
+    printf("Matrix Size\tWrite(s) [size]\t\tKernel(s)\tRead(s)  "
+           "[size]\t\tTotal(s)\n");
+    printf("%dx%d      \t", size, size);
+
+    printf("%f [%.2fMB]\t", writeTime, writeMB);
+
+    printf("%f\t", kernelTime);
+
+    printf("%f [%.2fMB]\t", readTime, readMB);
+
+    printf("%f\n\n", writeTime + kernelTime + readTime);
+  }
+  
+  cl_freeMem(a_dev);
+  cl_freeMem(b_dev);
+  cl_freeMem(m_dev);
+  cl_freeKernel(fan1_kernel);
+  cl_freeKernel(fan2_kernel);
+  cl_freeProgram(gaussianElim_program);
+}
+
+float eventTime(cl_event event, cl_command_queue command_queue) {
+  cl_int error = 0;
+  cl_ulong eventStart, eventEnd;
+  clFinish(command_queue);
+  error = clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_START,
+                                  sizeof(cl_ulong), &eventStart, NULL);
+  cl_errChk(error, "ERROR in Event Profiling.", true);
+  error = clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_END,
+                                  sizeof(cl_ulong), &eventEnd, NULL);
+  cl_errChk(error, "ERROR in Event Profiling.", true);
+
+  return (float)((eventEnd - eventStart) / 1e9);
+}
+
+int parseCommandline(int argc, char *argv[], char *filename, int *q, int *t,
+                     int *p, int *d) {
+  int i;
+  // if (argc < 2) return 1; // error
+  strncpy(filename, "matrix4.txt", 100);
+  char flag;
+
+  for (i = 1; i < argc; i++) {
+    if (argv[i][0] == '-') { // flag
+      flag = argv[i][1];
+      switch (flag) {
+      case 'h': // help
+        return 1;
+        break;
+      case 'q': // quiet
+        *q = 1;
+        break;
+      case 't': // timing
+        *t = 1;
+        break;
+      case 'p': // platform
+        i++;
+        *p = atoi(argv[i]);
+        break;
+      case 'd': // device
+        i++;
+        *d = atoi(argv[i]);
+        break;
+      }
+    }
+  }
+  if ((*d >= 0 && *p < 0) ||
+      (*p >= 0 &&
+       *d < 0)) // both p and d must be specified if either are specified
+    return 1;
+  return 0;
+}
+
+void printUsage() {
+  printf("Gaussian Elimination Usage\n");
+  printf("\n");
+  printf("gaussianElimination [filename] [-hqt] [-p [int] -d [int]]\n");
+  printf("\n");
+  printf("example:\n");
+  printf("$ ./gaussianElimination matrix4.txt\n");
+  printf("\n");
+  printf("filename     the filename that holds the matrix data\n");
+  printf("\n");
+  printf("-h           Display the help file\n");
+  printf("-q           Quiet mode. Suppress all text output.\n");
+  printf("-t           Print timing information.\n");
+  printf("\n");
+  printf("-p [int]     Choose the platform (must choose both platform and "
+         "device)\n");
+  printf("-d [int]     Choose the device (must choose both platform and "
+         "device)\n");
+  printf("\n");
+  printf("\n");
+  printf("Notes: 1. The filename is required as the first parameter.\n");
+  printf("       2. If you declare either the device or the platform,\n");
+  printf("          you must declare both.\n\n");
+}
+
+/*------------------------------------------------------
+ ** InitPerRun() -- Initialize the contents of the
+ ** multipier matrix **m
+ **------------------------------------------------------
+ */
+void InitPerRun(int size, float *m) {
+  int i;
+  for (i = 0; i < size * size; i++)
+    *(m + i) = 0.0;
+}
+void BackSub(float *a, float *b, float *finalVec, int size) {
+  // solve "bottom up"
+  int i, j;
+  for (i = 0; i < size; i++) {
+    finalVec[size - i - 1] = b[size - i - 1];
+    for (j = 0; j < i; j++) {
+      finalVec[size - i - 1] -= *(a + size * (size - i - 1) + (size - j - 1)) *
+                                finalVec[size - j - 1];
+    }
+    finalVec[size - i - 1] =
+        finalVec[size - i - 1] / *(a + size * (size - i - 1) + (size - i - 1));
+  }
+}
+void InitMat(FILE *fp, int size, float *ary, int nrow, int ncol) {
+  int i, j;
+
+  for (i = 0; i < nrow; i++) {
+    for (j = 0; j < ncol; j++) {
+      fscanf(fp, "%f", ary + size * i + j);
+    }
+  }
+}
+/*------------------------------------------------------
+ ** InitAry() -- Initialize the array (vector) by reading
+ ** data from the data file
+ **------------------------------------------------------
+ */
+void InitAry(FILE *fp, float *ary, int ary_size) {
+  int i;
+
+  for (i = 0; i < ary_size; i++) {
+    fscanf(fp, "%f", &ary[i]);
+  }
+}
+/*------------------------------------------------------
+ ** PrintMat() -- Print the contents of the matrix
+ **------------------------------------------------------
+ */
+void PrintMat(float *ary, int size, int nrow, int ncol) {
+  int i, j;
+
+  for (i = 0; i < nrow; i++) {
+    for (j = 0; j < ncol; j++) {
+      printf("%8.2f ", *(ary + size * i + j));
+    }
+    printf("\n");
+  }
+  printf("\n");
+}
+
+/*------------------------------------------------------
+ ** PrintAry() -- Print the contents of the array (vector)
+ **------------------------------------------------------
+ */
+void PrintAry(float *ary, int ary_size) {
+  int i;
+  for (i = 0; i < ary_size; i++) {
+    printf("%.2f ", ary[i]);
+  }
+  printf("\n\n");
+}
+#endif

tests/opencl/guassian/matrix4.txt

@@ -0,0 +1,11 @@
+4
+
+-0.6	-0.5	0.7	0.3	
+-0.3	-0.9	0.3	0.7	
+-0.4	-0.5	-0.3	-0.8	
+0.0	-0.1	0.2	0.9	
+
+-0.85	-0.68	0.24	-0.53	
+
+0.7	0.0	-0.4	-0.5	
+

tests/opencl/guassian/run

@@ -0,0 +1 @@
+./gaussian ../../data/gaussian/matrix4.txt

tests/opencl/guassian/utils.cpp

@@ -0,0 +1,204 @@
+/****************************************************************************\ 
+ * Copyright (c) 2011, Advanced Micro Devices, Inc.                           *
+ * All rights reserved.                                                       *
+ *                                                                            *
+ * Redistribution and use in source and binary forms, with or without         *
+ * modification, are permitted provided that the following conditions         *
+ * are met:                                                                   *
+ *                                                                            *
+ * Redistributions of source code must retain the above copyright notice,     *
+ * this list of conditions and the following disclaimer.                      *
+ *                                                                            *
+ * Redistributions in binary form must reproduce the above copyright notice,  *
+ * this list of conditions and the following disclaimer in the documentation  *
+ * and/or other materials provided with the distribution.                     *
+ *                                                                            *
+ * Neither the name of the copyright holder nor the names of its contributors *
+ * may be used to endorse or promote products derived from this software      *
+ * without specific prior written permission.                                 *
+ *                                                                            *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS        *
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED  *
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR *
+ * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR          *
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,      *
+ * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,        *
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR         *
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF     *
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING       *
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS         *
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.               *
+ *                                                                            *
+ * If you use the software (in whole or in part), you shall adhere to all     *
+ * applicable U.S., European, and other export laws, including but not        *
+ * limited to the U.S. Export Administration Regulations (“EAR”), (15 C.F.R.  *
+ * Sections 730 through 774), and E.U. Council Regulation (EC) No 1334/2000   *
+ * of 22 June 2000.  Further, pursuant to Section 740.6 of the EAR, you       *
+ * hereby certify that, except pursuant to a license granted by the United    *
+ * States Department of Commerce Bureau of Industry and Security or as        *
+ * otherwise permitted pursuant to a License Exception under the U.S. Export  *
+ * Administration Regulations ("EAR"), you will not (1) export, re-export or  *
+ * release to a national of a country in Country Groups D:1, E:1 or E:2 any   *
+ * restricted technology, software, or source code you receive hereunder,     *
+ * or (2) export to Country Groups D:1, E:1 or E:2 the direct product of such *
+ * technology or software, if such foreign produced direct product is subject *
+ * to national security controls as identified on the Commerce Control List   *
+ *(currently found in Supplement 1 to Part 774 of EAR).  For the most current *
+ * Country Group listings, or for additional information about the EAR or     *
+ * your obligations under those regulations, please refer to the U.S. Bureau  *
+ * of Industry and Security’s website at http://www.bis.doc.gov/.             *
+ \****************************************************************************/
+
+#include <stdio.h>
+#include <sys/stat.h>
+#include <string.h>
+#include <stdlib.h>
+
+#include "utils.h"
+
+static bool usingImages = true;
+
+//! A wrapper for malloc that checks the return value
+void* alloc(size_t size) {
+
+    void* ptr = NULL;
+    ptr = malloc(size);
+    if(ptr == NULL) {
+        perror("malloc");
+        exit(-1);
+    }
+
+    return ptr;
+}
+
+// This function checks to make sure a file exists before we open it
+void checkFile(char* filename) 
+{
+    
+    struct stat fileStatus;
+    if(stat(filename, &fileStatus) != 0) {
+        printf("Error opening file: %s\n", filename);
+        exit(-1);
+    }
+    else {
+        if(!(S_IFREG & fileStatus.st_mode)) {
+            printf("File %s is not a regular file\n", filename);
+            exit(-1);
+        }
+    }
+}
+
+
+// This function checks to make sure a directory exists 
+void checkDir(char* dirpath) 
+{
+    
+    struct stat fileStatus;
+    if(stat(dirpath, &fileStatus) != 0) {
+        printf("Directory does not exist: %s\n", dirpath);
+        exit(-1);
+    }
+    else {
+        if(!(S_IFDIR & fileStatus.st_mode)) {
+            printf("Directory was not provided: %s\n", dirpath);
+            exit(-1);
+        }
+    }
+}
+
+// Parse the command line arguments
+void parseArguments(int argc, char** argv, char** input, char** events, 
+    char** ipts, char* devicePref, bool* verifyResults) 
+{
+    
+    for(int i = 2; i < argc; i++) {
+        if(strcmp(argv[i], "-d") == 0) {   // Event dump found
+            if(i == argc-1) {
+                printf("Usage: -e Needs directory path\n");
+                exit(-1);
+            }
+            devicePref[0] = argv[i+1][0];
+            i++;
+            continue;
+        }
+        if(strcmp(argv[i], "-e") == 0) {   // Event dump found
+            if(i == argc-1) {
+                printf("Usage: -e Needs directory path\n");
+                exit(-1);
+            }
+            *events = argv[i+1];
+            i++;
+            continue;
+        }
+        if(strcmp(argv[i], "-i") == 0) {   // Input found
+            if(i == argc-1) {
+                printf("Usage: -i Needs directory path\n");
+                exit(-1);
+            }
+            *input = argv[i+1];
+            i++;
+            continue;
+        }
+        if(strcmp(argv[i], "-l") == 0) {   // Ipts dump found
+            if(i == argc-1) {
+                printf("Usage: -l Needs directory path\n");
+                exit(-1);
+            }
+            *ipts = argv[i+1];
+            i++;
+            continue;
+        }
+        if(strcmp(argv[i], "-n") == 0) {   // Don't use OpenCL images
+            setUsingImages(false);
+            continue;
+        }
+        if(strcmp(argv[i], "-v") == 0) {   // Verify results
+            *verifyResults = true;
+            continue;
+        }
+    }
+}
+
+
+// This function that takes a positive integer 'value' and returns
+// the nearest multiple of 'multiple' (used for padding columns)
+unsigned int roundUp(unsigned int value, unsigned int multiple) {
+
+   unsigned int remainder = value % multiple;
+   
+   // Make the value a multiple of multiple
+   if(remainder != 0) {
+      value += (multiple-remainder);
+   }
+
+   return value;
+}
+
+
+// Concatenate two strings and return a pointer to the new string
+char* smartStrcat(char* str1, char* str2) 
+{
+    char* newStr = NULL;
+
+    newStr = (char*)alloc((strlen(str1)+strlen(str2)+1)*sizeof(char));
+
+    strcpy(newStr, str1);
+    strcat(newStr, str2);
+
+    return newStr;
+}
+
+
+// Set the value of using images to true if they are being
+// used, or false if they are not
+void setUsingImages(bool val) 
+{
+    usingImages = val;
+}
+
+
+// Return whether or not images are being used
+bool isUsingImages() 
+{
+    return usingImages;
+}

tests/opencl/guassian/utils.h

@@ -0,0 +1,84 @@
+/****************************************************************************\ 
+ * Copyright (c) 2011, Advanced Micro Devices, Inc.                           *
+ * All rights reserved.                                                       *
+ *                                                                            *
+ * Redistribution and use in source and binary forms, with or without         *
+ * modification, are permitted provided that the following conditions         *
+ * are met:                                                                   *
+ *                                                                            *
+ * Redistributions of source code must retain the above copyright notice,     *
+ * this list of conditions and the following disclaimer.                      *
+ *                                                                            *
+ * Redistributions in binary form must reproduce the above copyright notice,  *
+ * this list of conditions and the following disclaimer in the documentation  *
+ * and/or other materials provided with the distribution.                     *
+ *                                                                            *
+ * Neither the name of the copyright holder nor the names of its contributors *
+ * may be used to endorse or promote products derived from this software      *
+ * without specific prior written permission.                                 *
+ *                                                                            *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS        *
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED  *
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR *
+ * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR          *
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,      *
+ * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,        *
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR         *
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF     *
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING       *
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS         *
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.               *
+ *                                                                            *
+ * If you use the software (in whole or in part), you shall adhere to all     *
+ * applicable U.S., European, and other export laws, including but not        *
+ * limited to the U.S. Export Administration Regulations (“EAR”), (15 C.F.R.  *
+ * Sections 730 through 774), and E.U. Council Regulation (EC) No 1334/2000   *
+ * of 22 June 2000.  Further, pursuant to Section 740.6 of the EAR, you       *
+ * hereby certify that, except pursuant to a license granted by the United    *
+ * States Department of Commerce Bureau of Industry and Security or as        *
+ * otherwise permitted pursuant to a License Exception under the U.S. Export  *
+ * Administration Regulations ("EAR"), you will not (1) export, re-export or  *
+ * release to a national of a country in Country Groups D:1, E:1 or E:2 any   *
+ * restricted technology, software, or source code you receive hereunder,     *
+ * or (2) export to Country Groups D:1, E:1 or E:2 the direct product of such *
+ * technology or software, if such foreign produced direct product is subject *
+ * to national security controls as identified on the Commerce Control List   *
+ *(currently found in Supplement 1 to Part 774 of EAR).  For the most current *
+ * Country Group listings, or for additional information about the EAR or     *
+ * your obligations under those regulations, please refer to the U.S. Bureau  *
+ * of Industry and Security’s website at http://www.bis.doc.gov/.             *
+ \****************************************************************************/
+
+#ifndef _UTILS_
+#define _UTILS_
+
+// Wrapper for malloc
+void* alloc(size_t size);
+
+// Checks for existence of directory
+void checkDir(char* dirpath);
+
+// Check for existence of file
+void checkFile(char* filename);
+
+// Parse the input command line options to the program
+void parseArguments(int argc, char** argv, char** input, char** events, 
+    char** ipts, char* devicePref, bool* verifyResults);
+
+
+// Print the program usage information
+void printUsage();
+
+// Rounds up size to the nearest multiple of multiple
+unsigned int roundUp(unsigned int value, unsigned int multiple);
+
+// Concatenate two strings, creating a new one
+char* smartStrcat(char* str1, char* str2);
+
+// Set the value of usingImages
+void setUsingImages(bool val);
+
+// Return whether or not images are being used
+bool isUsingImages(); 
+
+#endif

tests/opencl/kmeans/.gitignore

@@ -0,0 +1,2 @@
+kmeans
+

tests/opencl/kmeans/Makefile

@@ -0,0 +1,59 @@
+LLVM_PREFIX ?= /opt/llvm-riscv
+RISCV_TOOLCHAIN_PATH ?= /opt/riscv-gnu-toolchain
+SYSROOT ?= $(RISCV_TOOLCHAIN_PATH)/riscv32-unknown-elf
+POCL_CC_PATH ?= /opt/pocl/compiler
+POCL_RT_PATH ?= /opt/pocl/runtime
+
+VORTEX_DRV_PATH ?= $(realpath ../../../driver)
+VORTEX_RT_PATH ?= $(realpath ../../../runtime)
+
+K_LLCFLAGS += "-O3 -march=riscv32 -target-abi=ilp32f -mcpu=generic-rv32 -mattr=+m,+f -float-abi=hard -code-model=small"
+K_CFLAGS   += "-v -O3 --sysroot=$(SYSROOT) --gcc-toolchain=$(RISCV_TOOLCHAIN_PATH) -march=rv32imf -mabi=ilp32f -I$(VORTEX_RT_PATH)/include -fno-rtti -fno-exceptions -ffreestanding -nostartfiles -fdata-sections -ffunction-sections"
+K_LDFLAGS  += "-Wl,-Bstatic,-T$(VORTEX_RT_PATH)/linker/vx_link.ld -Wl,--gc-sections $(VORTEX_RT_PATH)/libvortexrt.a -lm"
+
+CXXFLAGS += -std=c++11 -O2 -Wall -Wextra -pedantic -Wfatal-errors
+#CXXFLAGS += -std=c++11 -O0 -g -Wall -Wextra -pedantic -Wfatal-errors
+
+CXXFLAGS += -I$(POCL_RT_PATH)/include
+
+LDFLAGS += -L$(POCL_RT_PATH)/lib -L$(VORTEX_DRV_PATH)/stub -lOpenCL -lvortex
+
+PROJECT = kmeans
+
+SRCS = main.cc read_input.c rmse.c kmeans_clustering.c cluster.c getopt.c
+
+all: $(PROJECT) kernel.pocl
+
+kernel.pocl: kernel.cl
+	LLVM_PREFIX=$(LLVM_PREFIX) POCL_DEBUG=all LD_LIBRARY_PATH=$(LLVM_PREFIX)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -LLCFLAGS $(K_LLCFLAGS) -CFLAGS $(K_CFLAGS) -LDFLAGS $(K_LDFLAGS) -o kernel.pocl kernel.cl
+ 
+$(PROJECT): $(SRCS)
+	$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -o $@
+
+run-fpga: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-asesim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT)
+	
+run-vlsim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae/vlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-simx: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/simx:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+run-rtlsim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
+
+.depend: $(SRCS)
+	$(CXX) $(CXXFLAGS) -MM $^ > .depend;
+
+clean:
+	rm -rf $(PROJECT) *.o .depend 
+
+clean-all: clean
+	rm -rf *.pocl *.dump
+
+ifneq ($(MAKECMDGOALS),clean)
+    -include .depend
+endif

tests/opencl/kmeans/cluster.c

@@ -0,0 +1,155 @@
+/*****************************************************************************/
+/*IMPORTANT:  READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.         */
+/*By downloading, copying, installing or using the software you agree        */
+/*to this license.  If you do not agree to this license, do not download,    */
+/*install, copy or use the software.                                         */
+/*                                                                           */
+/*                                                                           */
+/*Copyright (c) 2005 Northwestern University                                 */
+/*All rights reserved.                                                       */
+
+/*Redistribution of the software in source and binary forms,                 */
+/*with or without modification, is permitted provided that the               */
+/*following conditions are met:                                              */
+/*                                                                           */
+/*1       Redistributions of source code must retain the above copyright     */
+/*        notice, this list of conditions and the following disclaimer.      */
+/*                                                                           */
+/*2       Redistributions in binary form must reproduce the above copyright   */
+/*        notice, this list of conditions and the following disclaimer in the */
+/*        documentation and/or other materials provided with the distribution.*/ 
+/*                                                                            */
+/*3       Neither the name of Northwestern University nor the names of its    */
+/*        contributors may be used to endorse or promote products derived     */
+/*        from this software without specific prior written permission.       */
+/*                                                                            */
+/*THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ``AS    */
+/*IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED      */
+/*TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY, NON-INFRINGEMENT AND         */
+/*FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL          */
+/*NORTHWESTERN UNIVERSITY OR ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT,       */
+/*INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES          */
+/*(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR          */
+/*SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)          */
+/*HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,         */
+/*STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN    */
+/*ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE             */
+/*POSSIBILITY OF SUCH DAMAGE.                                                 */
+/******************************************************************************/
+
+/*************************************************************************/
+/**   File:         cluster.c                                           **/
+/**   Description:  Takes as input a file, containing 1 data point per  **/
+/**                 per line, and performs a fuzzy c-means clustering   **/
+/**                 on the data. Fuzzy clustering is performed using    **/
+/**                 min to max clusters and the clustering that gets    **/
+/**                 the best score according to a compactness and       **/
+/**                 separation criterion are returned.                  **/
+/**   Author:  Brendan McCane                                           **/
+/**            James Cook University of North Queensland.               **/
+/**            Australia. email: mccane@cs.jcu.edu.au                   **/
+/**                                                                     **/
+/**   Edited by: Jay Pisharath, Wei-keng Liao                           **/
+/**              Northwestern University.                               **/
+/**																		**/
+/**   ================================================================  **/
+/**																		**/
+/**   Edited by: Shuai Che, David Tarjan, Sang-Ha Lee					**/
+/**				 University of Virginia									**/
+/**																		**/
+/**   Description:	No longer supports fuzzy c-means clustering;	 	**/
+/**					only regular k-means clustering.					**/
+/**					No longer performs "validity" function to analyze	**/
+/**					compactness and separation crietria; instead		**/
+/**					calculate root mean squared error.					**/
+/**                                                                     **/
+/*************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <limits.h>
+#include <math.h>
+#include <float.h>
+#include "kmeans.h"
+
+float	min_rmse_ref = FLT_MAX;		
+extern double wtime(void);
+	/* reference min_rmse value */
+
+/*---< cluster() >-----------------------------------------------------------*/
+int cluster(int      npoints,				/* number of data points */
+            int      nfeatures,				/* number of attributes for each point */
+            float  **features,			/* array: [npoints][nfeatures] */                  
+            int      min_nclusters,			/* range of min to max number of clusters */
+			int		 max_nclusters,
+            float    threshold,				/* loop terminating factor */
+            int     *best_nclusters,		/* out: number between min and max with lowest RMSE */
+            float ***cluster_centres,		/* out: [best_nclusters][nfeatures] */
+			float	*min_rmse,				/* out: minimum RMSE */
+			int		 isRMSE,				/* calculate RMSE */
+			int		 nloops					/* number of iteration for each number of clusters */
+			)
+{    
+	int		nclusters;						/* number of clusters k */	
+	int		index =0;						/* number of iteration to reach the best RMSE */
+	int		rmse;							/* RMSE for each clustering */
+    int    *membership;						/* which cluster a data point belongs to */
+    float **tmp_cluster_centres;			/* hold coordinates of cluster centers */
+	int		i;
+
+	/* allocate memory for membership */
+    membership = (int*) malloc(npoints * sizeof(int));
+
+	/* sweep k from min to max_nclusters to find the best number of clusters */
+	for(nclusters = min_nclusters; nclusters <= max_nclusters; nclusters++)
+	{
+		if (nclusters > npoints) break;	/* cannot have more clusters than points */
+
+		/* allocate device memory, invert data array (@ kmeans_cuda.cu) */
+		allocate(npoints, nfeatures, nclusters, features);
+
+		/* iterate nloops times for each number of clusters */
+		for(i = 0; i < nloops; i++)
+		{
+			/* initialize initial cluster centers, CUDA calls (@ kmeans_cuda.cu) */
+			tmp_cluster_centres = kmeans_clustering(features,
+													nfeatures,
+													npoints,
+													nclusters,
+													threshold,
+													membership);
+
+			if (*cluster_centres) {
+				free((*cluster_centres)[0]);
+				free(*cluster_centres);
+			}
+			*cluster_centres = tmp_cluster_centres;
+	        
+					
+			/* find the number of clusters with the best RMSE */
+			if(isRMSE)
+			{
+				rmse = rms_err(features,
+							   nfeatures,
+							   npoints,
+							   tmp_cluster_centres,
+							   nclusters);
+				
+				if(rmse < min_rmse_ref){
+					min_rmse_ref = rmse;			//update reference min RMSE
+					*min_rmse = min_rmse_ref;		//update return min RMSE
+					*best_nclusters = nclusters;	//update optimum number of clusters
+					index = i;						//update number of iteration to reach best RMSE
+				}
+			}			
+		}
+		
+		deallocateMemory();							/* free device memory (@ kmeans_cuda.cu) */
+	}
+
+    free(membership);
+
+    return index;
+}
+

tests/opencl/kmeans/getopt.h

@@ -0,0 +1,191 @@
+
+
+/* getopt.h */
+/* Declarations for getopt.
+   Copyright (C) 1989-1994, 1996-1999, 2001 Free Software 
+   Foundation, Inc. This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute 
+   it and/or modify it under the terms of the GNU Lesser 
+   General Public License as published by the Free Software 
+   Foundation; either version 2.1 of the License, or 
+   (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will 
+   be useful, but WITHOUT ANY WARRANTY; without even the 
+   implied warranty of MERCHANTABILITY or FITNESS FOR A 
+   PARTICULAR PURPOSE.  See the GNU Lesser General Public 
+   License for more details.
+
+   You should have received a copy of the GNU Lesser General 
+   Public License along with the GNU C Library; if not, write 
+   to the Free Software Foundation, Inc., 59 Temple Place,
+   Suite 330, Boston, MA 02111-1307 USA.  */
+
+  
+
+  
+
+#ifndef _GETOPT_H
+
+#ifndef __need_getopt
+# define _GETOPT_H 1
+#endif
+
+/* If __GNU_LIBRARY__ is not already defined, either we are being used
+   standalone, or this is the first header included in the source file.
+   If we are being used with glibc, we need to include <features.h>, but
+   that does not exist if we are standalone.  So: if __GNU_LIBRARY__ is
+   not defined, include <ctype.h>, which will pull in <features.h> for us
+   if it's from glibc.  (Why ctype.h?  It's guaranteed to exist and it
+   doesn't flood the namespace with stuff the way some other headers do.)  */
+#if !defined __GNU_LIBRARY__
+# include <ctype.h>
+#endif
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+/* For communication from `getopt' to the caller.
+   When `getopt' finds an option that takes an argument,
+   the argument value is returned here.
+   Also, when `ordering' is RETURN_IN_ORDER,
+   each non-option ARGV-element is returned here.  */
+
+extern char *optarg;
+
+/* Index in ARGV of the next element to be scanned.
+   This is used for communication to and from the caller
+   and for communication between successive calls to `getopt'.
+
+   On entry to `getopt', zero means this is the first call; initialize.
+
+   When `getopt' returns -1, this is the index of the first of the
+   non-option elements that the caller should itself scan.
+
+   Otherwise, `optind' communicates from one call to the next
+   how much of ARGV has been scanned so far.  */
+
+extern int optind;
+
+/* Callers store zero here to inhibit the error message `getopt' prints
+   for unrecognized options.  */
+
+extern int opterr;
+
+/* Set to an option character which was unrecognized.  */
+
+extern int optopt;
+
+#ifndef __need_getopt
+/* Describe the long-named options requested by the application.
+   The LONG_OPTIONS argument to getopt_long or getopt_long_only is a vector
+   of `struct option' terminated by an element containing a name which is
+   zero.
+
+   The field `has_arg' is:
+   no_argument          (or 0) if the option does not take an argument,
+   required_argument    (or 1) if the option requires an argument,
+   optional_argument    (or 2) if the option takes an optional argument.
+
+   If the field `flag' is not NULL, it points to a variable that is set
+   to the value given in the field `val' when the option is found, but
+   left unchanged if the option is not found.
+
+   To have a long-named option do something other than set an `int' to
+   a compiled-in constant, such as set a value from `optarg', set the
+   option's `flag' field to zero and its `val' field to a nonzero
+   value (the equivalent single-letter option character, if there is
+   one).  For long options that have a zero `flag' field, `getopt'
+   returns the contents of the `val' field.  */
+
+struct option
+{
+# if (defined __STDC__ && __STDC__) || defined __cplusplus
+  const char *name;
+# else
+  char *name;
+# endif
+  /* has_arg can't be an enum because some compilers complain about
+     type mismatches in all the code that assumes it is an int.  */
+  int has_arg;
+  int *flag;
+  int val;
+};
+
+/* Names for the values of the `has_arg' field of `struct option'.  */
+
+# define no_argument            0
+# define required_argument      1
+# define optional_argument      2
+#endif  /* need getopt */
+
+
+/* Get definitions and prototypes for functions to process the
+   arguments in ARGV (ARGC of them, minus the program name) for
+   options given in OPTS.
+
+   Return the option character from OPTS just read.  Return -1 when
+   there are no more options.  For unrecognized options, or options
+   missing arguments, `optopt' is set to the option letter, and '?' is
+   returned.
+
+   The OPTS string is a list of characters which are recognized option
+   letters, optionally followed by colons, specifying that that letter
+   takes an argument, to be placed in `optarg'.
+
+   If a letter in OPTS is followed by two colons, its argument is
+   optional.  This behavior is specific to the GNU `getopt'.
+
+   The argument `--' causes premature termination of argument
+   scanning, explicitly telling `getopt' that there are no more
+   options.
+
+   If OPTS begins with `--', then non-option arguments are treated as
+   arguments to the option '\0'.  This behavior is specific to the GNU
+   `getopt'.  */
+
+#if (defined __STDC__ && __STDC__) || defined __cplusplus
+# ifdef __GNU_LIBRARY__
+/* Many other libraries have conflicting prototypes for getopt, with
+   differences in the consts, in stdlib.h.  To avoid compilation
+   errors, only prototype getopt for the GNU C library.  */
+extern int getopt (int ___argc, char *const *___argv, const char *__shortopts);
+# else /* not __GNU_LIBRARY__ */
+extern int getopt ();
+# endif /* __GNU_LIBRARY__ */
+
+# ifndef __need_getopt
+extern int getopt_long (int ___argc, char *const *___argv,
+                        const char *__shortopts,
+                        const struct option *__longopts, int *__longind);
+extern int getopt_long_only (int ___argc, char *const *___argv,
+                             const char *__shortopts,
+                             const struct option *__longopts, int *__longind);
+
+/* Internal only.  Users should not call this directly.  */
+extern int _getopt_internal (int ___argc, char *const *___argv,
+                             const char *__shortopts,
+                             const struct option *__longopts, int *__longind,
+                             int __long_only);
+# endif
+#else /* not __STDC__ */
+extern int getopt ();
+# ifndef __need_getopt
+extern int getopt_long ();
+extern int getopt_long_only ();
+
+extern int _getopt_internal ();
+# endif
+#endif /* __STDC__ */
+
+#ifdef  __cplusplus
+}
+#endif
+
+/* Make sure we later can get all the definitions and declarations.  */
+#undef __need_getopt
+
+#endif /* getopt.h */
+

tests/opencl/kmeans/kernel.cl

@@ -0,0 +1,61 @@
+#ifndef FLT_MAX
+#define FLT_MAX 3.40282347e+38
+#endif
+
+__kernel void
+kmeans_kernel_c(__global float  *feature,   
+			  __global float  *clusters,
+			  __global int    *membership,
+			    int     npoints,
+				int     nclusters,
+				int     nfeatures,
+				int		offset,
+				int		size
+			  ) 
+{
+	unsigned int point_id = get_global_id(0);
+    int index = 0;
+    //const unsigned int point_id = get_global_id(0);
+		if (point_id < npoints)
+		{
+			float min_dist=FLT_MAX;
+			for (int i=0; i < nclusters; i++) {
+				
+				float dist = 0;
+				float ans  = 0;
+				for (int l=0; l<nfeatures; l++){
+						ans += (feature[l * npoints + point_id]-clusters[i*nfeatures+l])* 
+							   (feature[l * npoints + point_id]-clusters[i*nfeatures+l]);
+				}
+
+				dist = ans;
+				if (dist < min_dist) {
+					min_dist = dist;
+					index    = i;
+					
+				}
+			}
+		  //printf("%d\n", index);
+		  membership[point_id] = index;
+		}	
+	
+	return;
+}
+
+__kernel void
+kmeans_swap(__global float  *feature,   
+			__global float  *feature_swap,
+			int     npoints,
+			int     nfeatures
+){
+
+	unsigned int tid = get_global_id(0);
+	//for(int i = 0; i <  nfeatures; i++)
+	//	feature_swap[i * npoints + tid] = feature[tid * nfeatures + i];
+    //Lingjie Zhang modificated at 11/05/2015
+    if (tid < npoints){
+	    for(int i = 0; i <  nfeatures; i++)
+		    feature_swap[i * npoints + tid] = feature[tid * nfeatures + i];
+    }
+    // end of Lingjie Zhang's modification
+} 

tests/opencl/kmeans/kmeans.h

@@ -0,0 +1,65 @@
+/*****************************************************************************/
+/*IMPORTANT:  READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.         */
+/*By downloading, copying, installing or using the software you agree        */
+/*to this license.  If you do not agree to this license, do not download,    */
+/*install, copy or use the software.                                         */
+/*                                                                           */
+/*                                                                           */
+/*Copyright (c) 2005 Northwestern University                                 */
+/*All rights reserved.                                                       */
+
+/*Redistribution of the software in source and binary forms,                 */
+/*with or without modification, is permitted provided that the               */
+/*following conditions are met:                                              */
+/*                                                                           */
+/*1       Redistributions of source code must retain the above copyright     */
+/*        notice, this list of conditions and the following disclaimer.      */
+/*                                                                           */
+/*2       Redistributions in binary form must reproduce the above copyright   */
+/*        notice, this list of conditions and the following disclaimer in the */
+/*        documentation and/or other materials provided with the distribution.*/ 
+/*                                                                            */
+/*3       Neither the name of Northwestern University nor the names of its    */
+/*        contributors may be used to endorse or promote products derived     */
+/*        from this software without specific prior written permission.       */
+/*                                                                            */
+/*THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ``AS    */
+/*IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED      */
+/*TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY, NON-INFRINGEMENT AND         */
+/*FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL          */
+/*NORTHWESTERN UNIVERSITY OR ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT,       */
+/*INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES          */
+/*(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR          */
+/*SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)          */
+/*HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,         */
+/*STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN    */
+/*ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE             */
+/*POSSIBILITY OF SUCH DAMAGE.                                                 */
+/******************************************************************************/
+
+#ifndef _H_FUZZY_KMEANS
+#define _H_FUZZY_KMEANS
+
+#ifndef FLT_MAX
+#define FLT_MAX 3.40282347e+38
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+float   euclid_dist_2        (float*, float*, int);
+int     find_nearest_point   (float* , int, float**, int);
+float	rms_err(float**, int, int, float**, int);
+int     cluster(int, int, float**, int, int, float, int*, float***, float*, int, int);
+int setup(int argc, char** argv);
+int allocate(int npoints, int nfeatures, int nclusters, float **feature);
+void deallocateMemory();
+int	kmeansOCL(float **feature, int nfeatures, int npoints, int nclusters, int *membership, float **clusters, int *new_centers_len, float  **new_centers);
+float** kmeans_clustering(float **feature, int nfeatures, int npoints, int nclusters, float threshold, int *membership); 
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

tests/opencl/kmeans/kmeans_clustering.c

@@ -0,0 +1,176 @@
+/*****************************************************************************/
+/*IMPORTANT:  READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.         */
+/*By downloading, copying, installing or using the software you agree        */
+/*to this license.  If you do not agree to this license, do not download,    */
+/*install, copy or use the software.                                         */
+/*                                                                           */
+/*                                                                           */
+/*Copyright (c) 2005 Northwestern University                                 */
+/*All rights reserved.                                                       */
+
+/*Redistribution of the software in source and binary forms,                 */
+/*with or without modification, is permitted provided that the               */
+/*following conditions are met:                                              */
+/*                                                                           */
+/*1       Redistributions of source code must retain the above copyright     */
+/*        notice, this list of conditions and the following disclaimer.      */
+/*                                                                           */
+/*2       Redistributions in binary form must reproduce the above copyright   */
+/*        notice, this list of conditions and the following disclaimer in the */
+/*        documentation and/or other materials provided with the distribution.*/ 
+/*                                                                            */
+/*3       Neither the name of Northwestern University nor the names of its    */
+/*        contributors may be used to endorse or promote products derived     */
+/*        from this software without specific prior written permission.       */
+/*                                                                            */
+/*THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ``AS    */
+/*IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED      */
+/*TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY, NON-INFRINGEMENT AND         */
+/*FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL          */
+/*NORTHWESTERN UNIVERSITY OR ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT,       */
+/*INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES          */
+/*(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR          */
+/*SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)          */
+/*HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,         */
+/*STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN    */
+/*ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE             */
+/*POSSIBILITY OF SUCH DAMAGE.                                                 */
+/******************************************************************************/
+
+/*************************************************************************/
+/**   File:         kmeans_clustering.c                                 **/
+/**   Description:  Implementation of regular k-means clustering        **/
+/**                 algorithm                                           **/
+/**   Author:  Wei-keng Liao                                            **/
+/**            ECE Department, Northwestern University                  **/
+/**            email: wkliao@ece.northwestern.edu                       **/
+/**                                                                     **/
+/**   Edited by: Jay Pisharath                                          **/
+/**              Northwestern University.                               **/
+/**                                                                     **/
+/**   ================================================================  **/
+/**																		**/
+/**   Edited by: Shuai Che, David Tarjan, Sang-Ha Lee					**/
+/**				 University of Virginia									**/
+/**																		**/
+/**   Description:	No longer supports fuzzy c-means clustering;	 	**/
+/**					only regular k-means clustering.					**/
+/**					No longer performs "validity" function to analyze	**/
+/**					compactness and separation crietria; instead		**/
+/**					calculate root mean squared error.					**/
+/**                                                                     **/
+/*************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <float.h>
+#include <math.h>
+#include "kmeans.h"
+
+#define RANDOM_MAX 2147483647
+
+extern double wtime(void);
+
+/*----< kmeans_clustering() >---------------------------------------------*/
+float** kmeans_clustering(float **feature,    /* in: [npoints][nfeatures] */
+                          int     nfeatures,
+                          int     npoints,
+                          int     nclusters,
+                          float   threshold,
+                          int    *membership) /* out: [npoints] */
+{    
+    int      i, j, n = 0;				/* counters */
+	int		 loop=0, temp;
+    int     *new_centers_len;	/* [nclusters]: no. of points in each cluster */
+    float    delta;				/* if the point moved */
+    float  **clusters;			/* out: [nclusters][nfeatures] */
+    float  **new_centers;		/* [nclusters][nfeatures] */
+
+	int     *initial;			/* used to hold the index of points not yet selected
+								   prevents the "birthday problem" of dual selection (?)
+								   considered holding initial cluster indices, but changed due to
+								   possible, though unlikely, infinite loops */
+	int      initial_points;
+	int		 c = 0;
+
+	/* nclusters should never be > npoints
+	   that would guarantee a cluster without points */
+	if (nclusters > npoints)
+		nclusters = npoints;
+
+    /* allocate space for and initialize returning variable clusters[] */
+    clusters    = (float**) malloc(nclusters *             sizeof(float*));
+    clusters[0] = (float*)  malloc(nclusters * nfeatures * sizeof(float));
+    for (i=1; i<nclusters; i++)
+        clusters[i] = clusters[i-1] + nfeatures;
+
+	/* initialize the random clusters */
+	initial = (int *) malloc (npoints * sizeof(int));
+	for (i = 0; i < npoints; i++)
+	{
+		initial[i] = i;
+	}
+	initial_points = npoints;
+
+    /* randomly pick cluster centers */
+    for (i=0; i<nclusters && initial_points >= 0; i++) {
+		//n = (int)rand() % initial_points;		
+		
+        for (j=0; j<nfeatures; j++)
+            clusters[i][j] = feature[initial[n]][j];	// remapped
+
+		/* swap the selected index to the end (not really necessary,
+		   could just move the end up) */
+		temp = initial[n];
+		initial[n] = initial[initial_points-1];
+		initial[initial_points-1] = temp;
+		initial_points--;
+		n++;
+    }
+
+	/* initialize the membership to -1 for all */
+    for (i=0; i < npoints; i++)
+	  membership[i] = -1;
+
+    /* allocate space for and initialize new_centers_len and new_centers */
+    new_centers_len = (int*) calloc(nclusters, sizeof(int));
+
+    new_centers    = (float**) malloc(nclusters *            sizeof(float*));
+    new_centers[0] = (float*)  calloc(nclusters * nfeatures, sizeof(float));
+    for (i=1; i<nclusters; i++)
+        new_centers[i] = new_centers[i-1] + nfeatures;
+
+	/* iterate until convergence */
+	do {
+        delta = 0.0;
+		// CUDA
+		delta = (float) kmeansOCL(feature,			/* in: [npoints][nfeatures] */
+								   nfeatures,		/* number of attributes for each point */
+								   npoints,			/* number of data points */
+								   nclusters,		/* number of clusters */
+								   membership,		/* which cluster the point belongs to */
+								   clusters,		/* out: [nclusters][nfeatures] */
+								   new_centers_len,	/* out: number of points in each cluster */
+								   new_centers		/* sum of points in each cluster */
+								   );
+
+		/* replace old cluster centers with new_centers */
+		/* CPU side of reduction */
+		for (i=0; i<nclusters; i++) {
+			for (j=0; j<nfeatures; j++) {
+				if (new_centers_len[i] > 0)
+					clusters[i][j] = new_centers[i][j] / new_centers_len[i];	/* take average i.e. sum/n */
+				new_centers[i][j] = 0.0;	/* set back to 0 */
+			}
+			new_centers_len[i] = 0;			/* set back to 0 */
+		}	 
+		c++;
+    } while ((delta > threshold) && (loop++ < 500));	/* makes sure loop terminates */
+	printf("iterated %d times\n", c);
+    free(new_centers[0]);
+    free(new_centers);
+    free(new_centers_len);
+
+    return clusters;
+}
+

tests/opencl/kmeans/main.cc

@@ -0,0 +1,382 @@
+#include "kmeans.h"
+#include <iostream>
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <string>
+
+#ifdef WIN
+#include <windows.h>
+#else
+#include <pthread.h>
+#include <sys/time.h>
+double gettime() {
+  struct timeval t;
+  gettimeofday(&t, NULL);
+  return t.tv_sec + t.tv_usec * 1e-6;
+}
+#endif
+
+#ifdef NV
+#include <oclUtils.h>
+#else
+#include <CL/cl.h>
+#endif
+
+#ifndef FLT_MAX
+#define FLT_MAX 3.40282347e+38
+#endif
+
+#ifdef RD_WG_SIZE_0_0
+#define BLOCK_SIZE RD_WG_SIZE_0_0
+#elif defined(RD_WG_SIZE_0)
+#define BLOCK_SIZE RD_WG_SIZE_0
+#elif defined(RD_WG_SIZE)
+#define BLOCK_SIZE RD_WG_SIZE
+#else
+#define BLOCK_SIZE 256
+#endif
+
+#ifdef RD_WG_SIZE_1_0
+#define BLOCK_SIZE2 RD_WG_SIZE_1_0
+#elif defined(RD_WG_SIZE_1)
+#define BLOCK_SIZE2 RD_WG_SIZE_1
+#elif defined(RD_WG_SIZE)
+#define BLOCK_SIZE2 RD_WG_SIZE
+#else
+#define BLOCK_SIZE2 256
+#endif
+
+// local variables
+static cl_context context;
+static cl_command_queue cmd_queue;
+static cl_device_type device_type;
+static cl_device_id *device_list;
+static cl_int num_devices;
+
+static int initialize(int use_gpu) {
+  cl_int result;
+  size_t size;
+
+  /*// create OpenCL context
+  cl_platform_id platform_id;
+  if (clGetPlatformIDs(1, &platform_id, NULL) != CL_SUCCESS) {
+    printf("ERROR: clGetPlatformIDs(1,*,0) failed\n");
+    return -1;
+  }
+  cl_context_properties ctxprop[] = {CL_CONTEXT_PLATFORM,
+                                     (cl_context_properties)platform_id, 0};
+  device_type = use_gpu ? CL_DEVICE_TYPE_GPU : CL_DEVICE_TYPE_CPU;
+  context = clCreateContextFromType(ctxprop, device_type, NULL, NULL, NULL);
+  if (!context) {
+    printf("ERROR: clCreateContextFromType(%s) failed\n",
+           use_gpu ? "GPU" : "CPU");
+    return -1;
+  }
+
+  // get the list of GPUs
+  result = clGetContextInfo(context, CL_CONTEXT_DEVICES, 0, NULL, &size);
+  num_devices = (int)(size / sizeof(cl_device_id));
+
+  if (result != CL_SUCCESS || num_devices < 1) {
+    printf("ERROR: clGetContextInfo() failed\n");
+    return -1;
+  }
+  device_list = new cl_device_id[num_devices];
+  if (!device_list) {
+    printf("ERROR: new cl_device_id[] failed\n");
+    return -1;
+  }
+  result =
+      clGetContextInfo(context, CL_CONTEXT_DEVICES, size, device_list, NULL);
+  if (result != CL_SUCCESS) {
+    printf("ERROR: clGetContextInfo() failed\n");
+    return -1;
+  }*/
+
+  cl_platform_id platform_id;
+  num_devices = 1;
+  device_list = new cl_device_id[num_devices];
+
+  result = clGetPlatformIDs(1, &platform_id, NULL);
+  result = clGetDeviceIDs(platform_id, CL_DEVICE_TYPE_DEFAULT, 1, device_list, NULL);
+  context = clCreateContext(NULL, 1, device_list, NULL, NULL,  &result);
+
+  // create command queue for the first device
+  cmd_queue = clCreateCommandQueue(context, device_list[0], 0, NULL);
+  if (!cmd_queue) {
+    printf("ERROR: clCreateCommandQueue() failed\n");
+    return -1;
+  }
+
+  return 0;
+}
+
+static int shutdown() {
+  // release resources
+  if (cmd_queue)
+    clReleaseCommandQueue(cmd_queue);
+  if (context)
+    clReleaseContext(context);
+  if (device_list)
+    delete device_list;
+
+  // reset all variables
+  cmd_queue = 0;
+  context = 0;
+  device_list = 0;
+  num_devices = 0;
+  device_type = 0;
+
+  return 0;
+}
+
+cl_mem d_feature;
+cl_mem d_feature_swap;
+cl_mem d_cluster;
+cl_mem d_membership;
+
+cl_kernel kernel;
+cl_kernel kernel_s;
+cl_kernel kernel2;
+
+int *membership_OCL;
+int *membership_d;
+float *feature_d;
+float *clusters_d;
+float *center_d;
+
+	
+static int read_kernel_file(const char* filename, uint8_t** data, size_t* size) {
+  if (nullptr == filename || nullptr == data || 0 == size)
+    return -1;
+
+  FILE* fp = fopen(filename, "r");
+  if (NULL == fp) {
+    fprintf(stderr, "Failed to load kernel.");
+    return -1;
+  }
+  fseek(fp , 0 , SEEK_END);
+  long fsize = ftell(fp);
+  rewind(fp);
+
+  *data = (uint8_t*)malloc(fsize);
+  *size = fread(*data, 1, fsize, fp);
+  
+  fclose(fp);
+  
+  return 0;
+}
+
+
+int allocate(int n_points, int n_features, int n_clusters, float **feature) {
+  /*int sourcesize = 1024 * 1024;
+  char *source = (char *)calloc(sourcesize, sizeof(char));
+  if (!source) {
+    printf("ERROR: calloc(%d) failed\n", sourcesize);
+    return -1;
+  }
+
+  // read the kernel core source
+  char *tempchar = "./kmeans.cl";
+  FILE *fp = fopen(tempchar, "rb");
+  if (!fp) {
+    printf("ERROR: unable to open '%s'\n", tempchar);
+    return -1;
+  }
+  fread(source + strlen(source), sourcesize, 1, fp);
+  fclose(fp);*/
+
+  // OpenCL initialization
+  int use_gpu = 1;
+  if (initialize(use_gpu))
+    return -1;
+
+  // compile kernel
+  cl_int err = 0;
+  //const char *slist[2] = {source, 0};
+  //cl_program prog = clCreateProgramWithSource(context, 1, slist, NULL, &err);
+  cl_program prog = clCreateProgramWithBuiltInKernels(context, 1, device_list, "kmeans_kernel_c;kmeans_swap", &err);       
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clCreateProgramWithSource() => %d\n", err);
+    return -1;
+  }
+  err = clBuildProgram(prog, 0, NULL, NULL, NULL, NULL);
+  { // show warnings/errors
+    //	static char log[65536]; memset(log, 0, sizeof(log));
+    //	cl_device_id device_id = 0;
+    //	err = clGetContextInfo(context, CL_CONTEXT_DEVICES, sizeof(device_id),
+    //&device_id, NULL);
+    //	clGetProgramBuildInfo(prog, device_id, CL_PROGRAM_BUILD_LOG,
+    // sizeof(log)-1, log, NULL);
+    //	if(err || strstr(log,"warning:") || strstr(log, "error:"))
+    // printf("<<<<\n%s\n>>>>\n", log);
+  }
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clBuildProgram() => %d\n", err);
+    return -1;
+  }
+
+  char *kernel_kmeans_c = "kmeans_kernel_c";
+  char *kernel_swap = "kmeans_swap";
+
+  kernel_s = clCreateKernel(prog, kernel_kmeans_c, &err);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clCreateKernel() 0 => %d\n", err);
+    return -1;
+  }
+  kernel2 = clCreateKernel(prog, kernel_swap, &err);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clCreateKernel() 0 => %d\n", err);
+    return -1;
+  }
+
+  clReleaseProgram(prog);
+
+  d_feature = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                             n_points * n_features * sizeof(float), NULL, &err);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clCreateBuffer d_feature (size:%d) => %d\n",
+           n_points * n_features, err);
+    return -1;
+  }
+  d_feature_swap =
+      clCreateBuffer(context, CL_MEM_READ_WRITE,
+                     n_points * n_features * sizeof(float), NULL, &err);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clCreateBuffer d_feature_swap (size:%d) => %d\n",
+           n_points * n_features, err);
+    return -1;
+  }
+  d_cluster =
+      clCreateBuffer(context, CL_MEM_READ_WRITE,
+                     n_clusters * n_features * sizeof(float), NULL, &err);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clCreateBuffer d_cluster (size:%d) => %d\n",
+           n_clusters * n_features, err);
+    return -1;
+  }
+  d_membership = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                                n_points * sizeof(int), NULL, &err);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clCreateBuffer d_membership (size:%d) => %d\n", n_points,
+           err);
+    return -1;
+  }
+
+  // write buffers
+  err = clEnqueueWriteBuffer(cmd_queue, d_feature, 1, 0,
+                             n_points * n_features * sizeof(float), feature[0],
+                             0, 0, 0);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clEnqueueWriteBuffer d_feature (size:%d) => %d\n",
+           n_points * n_features, err);
+    return -1;
+  }
+
+  clSetKernelArg(kernel2, 0, sizeof(void *), (void *)&d_feature);
+  clSetKernelArg(kernel2, 1, sizeof(void *), (void *)&d_feature_swap);
+  clSetKernelArg(kernel2, 2, sizeof(cl_int), (void *)&n_points);
+  clSetKernelArg(kernel2, 3, sizeof(cl_int), (void *)&n_features);
+
+  size_t global_work[3] = {n_points, 1, 1};
+  /// Ke Wang adjustable local group size 2013/08/07 10:37:33
+  size_t local_work_size = BLOCK_SIZE; // work group size is defined by
+                                       // RD_WG_SIZE_0 or RD_WG_SIZE_0_0
+                                       // 2014/06/10 17:00:51
+  if (global_work[0] % local_work_size != 0)
+    global_work[0] = (global_work[0] / local_work_size + 1) * local_work_size;
+
+  err = clEnqueueNDRangeKernel(cmd_queue, kernel2, 1, NULL, global_work,
+                               &local_work_size, 0, 0, 0);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clEnqueueNDRangeKernel()=>%d failed\n", err);
+    return -1;
+  }
+
+  membership_OCL = (int *)malloc(n_points * sizeof(int));
+}
+
+void deallocateMemory() {
+  clReleaseMemObject(d_feature);
+  clReleaseMemObject(d_feature_swap);
+  clReleaseMemObject(d_cluster);
+  clReleaseMemObject(d_membership);
+  free(membership_OCL);
+}
+
+int main(int argc, char **argv) {
+  printf("WG size of kernel_swap = %d, WG size of kernel_kmeans = %d \n",
+         BLOCK_SIZE, BLOCK_SIZE2);
+  setup(argc, argv);
+  shutdown();
+}
+
+int kmeansOCL(float **feature, /* in: [npoints][nfeatures] */
+              int n_features, int n_points, int n_clusters, int *membership,
+              float **clusters, int *new_centers_len, float **new_centers) {
+
+  int delta = 0;
+  int i, j, k;
+  cl_int err = 0;
+
+  size_t global_work[3] = {n_points, 1, 1};
+
+  /// Ke Wang adjustable local group size 2013/08/07 10:37:33
+  size_t local_work_size = BLOCK_SIZE2; // work group size is defined by
+                                        // RD_WG_SIZE_1 or RD_WG_SIZE_1_0
+                                        // 2014/06/10 17:00:41
+  if (global_work[0] % local_work_size != 0)
+    global_work[0] = (global_work[0] / local_work_size + 1) * local_work_size;
+
+  err = clEnqueueWriteBuffer(cmd_queue, d_cluster, 1, 0,
+                             n_clusters * n_features * sizeof(float),
+                             clusters[0], 0, 0, 0);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clEnqueueWriteBuffer d_cluster (size:%d) => %d\n", n_points,
+           err);
+    return -1;
+  }
+
+  int size = 0;
+  int offset = 0;
+
+  clSetKernelArg(kernel_s, 0, sizeof(void *), (void *)&d_feature_swap);
+  clSetKernelArg(kernel_s, 1, sizeof(void *), (void *)&d_cluster);
+  clSetKernelArg(kernel_s, 2, sizeof(void *), (void *)&d_membership);
+  clSetKernelArg(kernel_s, 3, sizeof(cl_int), (void *)&n_points);
+  clSetKernelArg(kernel_s, 4, sizeof(cl_int), (void *)&n_clusters);
+  clSetKernelArg(kernel_s, 5, sizeof(cl_int), (void *)&n_features);
+  clSetKernelArg(kernel_s, 6, sizeof(cl_int), (void *)&offset);
+  clSetKernelArg(kernel_s, 7, sizeof(cl_int), (void *)&size);
+
+  err = clEnqueueNDRangeKernel(cmd_queue, kernel_s, 1, NULL, global_work,
+                               &local_work_size, 0, 0, 0);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clEnqueueNDRangeKernel()=>%d failed\n", err);
+    return -1;
+  }
+  clFinish(cmd_queue);
+  err = clEnqueueReadBuffer(cmd_queue, d_membership, 1, 0,
+                            n_points * sizeof(int), membership_OCL, 0, 0, 0);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: Memcopy Out\n");
+    return -1;
+  }
+
+  delta = 0;
+  for (i = 0; i < n_points; i++) {
+    int cluster_id = membership_OCL[i];
+    new_centers_len[cluster_id]++;
+    if (membership_OCL[i] != membership[i]) {
+      delta++;
+      membership[i] = membership_OCL[i];
+    }
+    for (j = 0; j < n_features; j++) {
+      new_centers[cluster_id][j] += feature[i][j];
+    }
+  }
+
+  return delta;
+}

tests/opencl/kmeans/read_input.c

@@ -0,0 +1,338 @@
+/*****************************************************************************/
+/*IMPORTANT:  READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.         */
+/*By downloading, copying, installing or using the software you agree        */
+/*to this license.  If you do not agree to this license, do not download,    */
+/*install, copy or use the software.                                         */
+/*                                                                           */
+/*                                                                           */
+/*Copyright (c) 2005 Northwestern University                                 */
+/*All rights reserved.                                                       */
+
+/*Redistribution of the software in source and binary forms,                 */
+/*with or without modification, is permitted provided that the               */
+/*following conditions are met:                                              */
+/*                                                                           */
+/*1       Redistributions of source code must retain the above copyright     */
+/*        notice, this list of conditions and the following disclaimer.      */
+/*                                                                           */
+/*2       Redistributions in binary form must reproduce the above copyright   */
+/*        notice, this list of conditions and the following disclaimer in the */
+/*        documentation and/or other materials provided with the distribution.*/
+/*                                                                            */
+/*3       Neither the name of Northwestern University nor the names of its    */
+/*        contributors may be used to endorse or promote products derived     */
+/*        from this software without specific prior written permission.       */
+/*                                                                            */
+/*THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ``AS    */
+/*IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED      */
+/*TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY, NON-INFRINGEMENT AND         */
+/*FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL          */
+/*NORTHWESTERN UNIVERSITY OR ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT,       */
+/*INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES          */
+/*(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR          */
+/*SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)          */
+/*HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,         */
+/*STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN    */
+/*ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE             */
+/*POSSIBILITY OF SUCH DAMAGE.                                                 */
+/******************************************************************************/
+
+/*************************************************************************/
+/**   File:         example.c                                           **/
+/**   Description:  Takes as input a file:                              **/
+/**                 ascii  file: containing 1 data point per line       **/
+/**                 binary file: first int is the number of objects     **/
+/**                              2nd int is the no. of features of each **/
+/**                              object                                 **/
+/**                 This example performs a fuzzy c-means clustering    **/
+/**                 on the data. Fuzzy clustering is performed using    **/
+/**                 min to max clusters and the clustering that gets    **/
+/**                 the best score according to a compactness and       **/
+/**                 separation criterion are returned.                  **/
+/**   Author:  Wei-keng Liao                                            **/
+/**            ECE Department Northwestern University                   **/
+/**            email: wkliao@ece.northwestern.edu                       **/
+/**                                                                     **/
+/**   Edited by: Jay Pisharath                                          **/
+/**              Northwestern University.                               **/
+/**                                                                     **/
+/**   ================================================================  **/
+/**
+ * **/
+/**   Edited by: Shuai Che, David Tarjan, Sang-Ha Lee
+ * **/
+/**				 University of Virginia
+ * **/
+/**
+ * **/
+/**   Description:	No longer supports fuzzy c-means clustering;
+ * **/
+/**					only regular k-means clustering.
+ * **/
+/**					No longer performs "validity" function to
+ * analyze	**/
+/**					compactness and separation crietria; instead
+ * **/
+/**					calculate root mean squared error.
+ * **/
+/**                                                                     **/
+/*************************************************************************/
+#define _CRT_SECURE_NO_DEPRECATE 1
+
+#include "kmeans.h"
+#include <fcntl.h>
+#include <limits.h>
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+extern double wtime(void);
+
+/*---< usage() >------------------------------------------------------------*/
+void usage(char *argv0) {
+  char *help = "\nUsage: %s [switches] -i filename\n\n"
+               "    -i filename      :file containing data to be clustered\n"
+               "    -m max_nclusters :maximum number of clusters allowed    "
+               "[default=5]\n"
+               "    -n min_nclusters :minimum number of clusters allowed    "
+               "[default=5]\n"
+               "    -t threshold     :threshold value                       "
+               "[default=0.001]\n"
+               "    -l nloops        :iteration for each number of clusters "
+               "[default=1]\n"
+               "    -b               :input file is in binary format\n"
+               "    -r               :calculate RMSE                        "
+               "[default=off]\n"
+               "    -o               :output cluster center coordinates     "
+               "[default=off]\n";
+  fprintf(stderr, help, argv0);
+  exit(-1);
+}
+
+/*---< main() >-------------------------------------------------------------*/
+int setup(int argc, char **argv) {
+  int opt;
+  extern char *optarg;
+  char *filename = 0;
+  float *buf;
+  char line[1024];
+  int isBinaryFile = 0;
+
+  float threshold = 0.001; /* default value */
+  int max_nclusters = 5;   /* default value */
+  int min_nclusters = 5;   /* default value */
+  int best_nclusters = 0;
+  int nfeatures = 0;
+  int npoints = 0;
+  float len;
+
+  float **features;
+  float **cluster_centres = NULL;
+  int i, j, index;
+  int nloops = 1; /* default value */
+
+  int isRMSE = 0;
+  float rmse;
+
+  int isOutput = 0;
+  // float	cluster_timing, io_timing;
+
+  /* obtain command line arguments and change appropriate options */
+  while ((opt = getopt(argc, argv, "i:t:m:n:l:bro")) != EOF) {
+    switch (opt) {
+    case 'i':
+      filename = optarg;
+      break;
+    case 'b':
+      isBinaryFile = 1;
+      break;
+    case 't':
+      threshold = atof(optarg);
+      break;
+    case 'm':
+      max_nclusters = atoi(optarg);
+      break;
+    case 'n':
+      min_nclusters = atoi(optarg);
+      break;
+    case 'r':
+      isRMSE = 1;
+      break;
+    case 'o':
+      isOutput = 1;
+      break;
+    case 'l':
+      nloops = atoi(optarg);
+      break;
+    case '?':
+      usage(argv[0]);
+      break;
+    default:
+      usage(argv[0]);
+      break;
+    }
+  }
+
+  /* ============== I/O begin ==============*/
+  /* get nfeatures and npoints */
+  // io_timing = omp_get_wtime();
+
+  /*if (isBinaryFile) { // Binary file input
+    FILE *infile;
+    if ((infile = fopen("100", "r")) == NULL) {
+      fprintf(stderr, "Error: no such file (%s)\n", filename);
+      exit(1);
+    }
+    fread(&npoints, 1, sizeof(int), infile);
+    fread(&nfeatures, 1, sizeof(int), infile);
+
+    // allocate space for features[][] and read attributes of all objects
+    buf = (float *)malloc(npoints * nfeatures * sizeof(float));
+    features = (float **)malloc(npoints * sizeof(float *));
+    features[0] = (float *)malloc(npoints * nfeatures * sizeof(float));
+    for (i = 1; i < npoints; i++) {
+      features[i] = features[i - 1] + nfeatures;
+    }
+    fread(buf, 1, npoints * nfeatures * sizeof(float), infile);
+    fclose(infile);
+  } else {
+    FILE *infile;
+    if ((infile = fopen("100", "r")) == NULL) {
+      fprintf(stderr, "Error: no such file (%s)\n", filename);
+      exit(1);
+    }
+    while (fgets(line, 1024, infile) != NULL)
+      if (strtok(line, " \t\n") != 0) {
+        npoints++;
+      }
+    rewind(infile);
+    while (fgets(line, 1024, infile) != NULL) {
+      if (strtok(line, " \t\n") != 0) {
+        // ignore the id (first attribute): nfeatures = 1;
+        while (strtok(NULL, " ,\t\n") != NULL)
+          nfeatures++;
+        break;
+      }
+    }
+
+    // allocate space for features[] and read attributes of all objects
+    buf = (float *)malloc(npoints * nfeatures * sizeof(float));
+    features = (float **)malloc(npoints * sizeof(float *));
+    features[0] = (float *)malloc(npoints * nfeatures * sizeof(float));
+    for (i = 1; i < npoints; i++)
+      features[i] = features[i - 1] + nfeatures;
+    rewind(infile);
+    i = 0;
+    while (fgets(line, 1024, infile) != NULL) {
+      if (strtok(line, " \t\n") == NULL)
+        continue;
+      for (j = 0; j < nfeatures; j++) {
+        buf[i] = atof(strtok(NULL, " ,\t\n"));
+        i++;
+      }
+    }
+    fclose(infile);
+  }*/
+
+  npoints = 100;
+  nfeatures = 100;
+  buf = (float *)malloc(npoints * nfeatures * sizeof(float));
+  features = (float **)malloc(npoints * sizeof(float *));
+  features[0] = (float *)malloc(npoints * nfeatures * sizeof(float));
+  for (i = 1; i < npoints; i++) {
+    features[i] = features[i - 1] + nfeatures;
+  }
+  for (i = 0; i < npoints * nfeatures; ++i) {   
+    buf[i] = (i % 64);
+  }
+
+  // io_timing = omp_get_wtime() - io_timing;
+
+  printf("\nI/O completed\n");
+  printf("\nNumber of objects: %d\n", npoints);
+  printf("Number of features: %d\n", nfeatures);
+  /* ============== I/O end ==============*/
+
+  // error check for clusters
+  if (npoints < min_nclusters) {
+    printf("Error: min_nclusters(%d) > npoints(%d) -- cannot proceed\n",
+           min_nclusters, npoints);
+    exit(0);
+  }
+
+  srand(7); /* seed for future random number generator */
+  memcpy(
+      features[0], buf,
+      npoints * nfeatures *
+          sizeof(
+              float)); /* now features holds 2-dimensional array of features */
+  free(buf);
+
+  /* ======================= core of the clustering ===================*/
+
+  // cluster_timing = omp_get_wtime();		/* Total clustering time */
+  cluster_centres = NULL;
+  index = cluster(npoints,       /* number of data points */
+                  nfeatures,     /* number of features for each point */
+                  features,      /* array: [npoints][nfeatures] */
+                  min_nclusters, /* range of min to max number of clusters */
+                  max_nclusters, threshold, /* loop termination factor */
+                  &best_nclusters,  /* return: number between min and max */
+                  &cluster_centres, /* return: [best_nclusters][nfeatures] */
+                  &rmse,            /* Root Mean Squared Error */
+                  isRMSE,           /* calculate RMSE */
+                  nloops); /* number of iteration for each number of clusters */
+
+  // cluster_timing = omp_get_wtime() - cluster_timing;
+
+  /* =============== Command Line Output =============== */
+
+  /* cluster center coordinates
+     :displayed only for when k=1*/
+  if ((min_nclusters == max_nclusters) && (isOutput == 1)) {
+    printf("\n================= Centroid Coordinates =================\n");
+    for (i = 0; i < max_nclusters; i++) {
+      printf("%d:", i);
+      for (j = 0; j < nfeatures; j++) {
+        printf(" %.2f", cluster_centres[i][j]);
+      }
+      printf("\n\n");
+    }
+  }
+
+  len = (float)((max_nclusters - min_nclusters + 1) * nloops);
+
+  printf("Number of Iteration: %d\n", nloops);
+  // printf("Time for I/O: %.5fsec\n", io_timing);
+  // printf("Time for Entire Clustering: %.5fsec\n", cluster_timing);
+
+  if (min_nclusters != max_nclusters) {
+    if (nloops != 1) { // range of k, multiple iteration
+      // printf("Average Clustering Time: %fsec\n",
+      //		cluster_timing / len);
+      printf("Best number of clusters is %d\n", best_nclusters);
+    } else { // range of k, single iteration
+      // printf("Average Clustering Time: %fsec\n",
+      //		cluster_timing / len);
+      printf("Best number of clusters is %d\n", best_nclusters);
+    }
+  } else {
+    if (nloops != 1) { // single k, multiple iteration
+      // printf("Average Clustering Time: %.5fsec\n",
+      //		cluster_timing / nloops);
+      if (isRMSE) // if calculated RMSE
+        printf("Number of trials to approach the best RMSE of %.3f is %d\n",
+               rmse, index + 1);
+    } else {      // single k, single iteration
+      if (isRMSE) // if calculated RMSE
+        printf("Root Mean Squared Error: %.3f\n", rmse);
+    }
+  }
+
+  /* free up memory */
+  free(features[0]);
+  free(features);
+  return (0);
+}

tests/opencl/kmeans/rmse.c

@@ -0,0 +1,94 @@
+/*************************************************************************/
+/**   File:         rmse.c												**/
+/**   Description:  calculate root mean squared error of particular     **/
+/**                 clustering.											**/
+/**   Author:  Sang-Ha Lee												**/
+/**            University of Virginia.									**/
+/**																		**/
+/**   Note: euclid_dist_2() and find_nearest_point() adopted from       **/
+/**			Minebench code.												**/
+/**                                                                     **/
+/*************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <float.h>
+#include <math.h>
+
+#include "kmeans.h"
+
+extern double wtime(void);
+
+/*----< euclid_dist_2() >----------------------------------------------------*/
+/* multi-dimensional spatial Euclid distance square */
+__inline
+float euclid_dist_2(float *pt1,
+                    float *pt2,
+                    int    numdims)
+{
+    int i;
+    float ans=0.0;
+
+    for (i=0; i<numdims; i++)
+        ans += (pt1[i]-pt2[i]) * (pt1[i]-pt2[i]);
+
+    return(ans);
+}
+
+/*----< find_nearest_point() >-----------------------------------------------*/
+__inline
+int find_nearest_point(float  *pt,          /* [nfeatures] */
+                       int     nfeatures,
+                       float  **pts,         /* [npts][nfeatures] */
+                       int     npts)
+{
+    int index, i;
+    float max_dist=FLT_MAX;
+
+    /* find the cluster center id with min distance to pt */
+    for (i=0; i<npts; i++) {
+        float dist;
+        dist = euclid_dist_2(pt, pts[i], nfeatures);  /* no need square root */
+        if (dist < max_dist) {
+            max_dist = dist;
+            index    = i;
+        }
+    }
+    return(index);
+}
+
+/*----< rms_err(): calculates RMSE of clustering >-------------------------------------*/
+float rms_err	(float **feature,         /* [npoints][nfeatures] */
+                 int     nfeatures,
+                 int     npoints,
+                 float **cluster_centres, /* [nclusters][nfeatures] */
+                 int     nclusters)
+{
+    int    i;
+	int	   nearest_cluster_index;	/* cluster center id with min distance to pt */
+    float  sum_euclid = 0.0;		/* sum of Euclidean distance squares */
+    float  ret;						/* return value */
+    
+    /* calculate and sum the sqaure of euclidean distance*/	
+    #pragma omp parallel for \
+                shared(feature,cluster_centres) \
+                firstprivate(npoints,nfeatures,nclusters) \
+                private(i, nearest_cluster_index) \
+                schedule (static)	
+    for (i=0; i<npoints; i++) {
+        nearest_cluster_index = find_nearest_point(feature[i], 
+													nfeatures, 
+													cluster_centres, 
+													nclusters);
+
+		sum_euclid += euclid_dist_2(feature[i],
+									cluster_centres[nearest_cluster_index],
+									nfeatures);
+		
+    }	
+	/* divide by n, then take sqrt */
+	ret = sqrt(sum_euclid / npoints);
+
+    return(ret);
+}
+

tests/opencl/lbm/Makefile

@@ -0,0 +1,67 @@
+RISCV_TOOLCHAIN_PATH ?= $(wildcard ../../../../riscv-gnu-toolchain/drops)
+POCL_CC_PATH ?= $(wildcard ../../../../pocl/drops_riscv_cc)
+POCL_INC_PATH ?= $(wildcard ../include)
+POCL_LIB_PATH ?= $(wildcard ../lib)
+VORTEX_RT_PATH ?= $(wildcard ../../../runtime)
+VX_SIMX_PATH ?= $(wildcard ../../../simX/obj_dir)
+
+CC  = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gcc
+CXX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-g++
+DMP = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objdump
+HEX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objcopy
+GDB = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gdb
+
+VX_SRCS =  $(VORTEX_RT_PATH)/newlib/newlib.c
+VX_SRCS += $(VORTEX_RT_PATH)/startup/vx_start.S
+VX_SRCS += $(VORTEX_RT_PATH)/intrinsics/vx_intrinsics.S
+VX_SRCS += $(VORTEX_RT_PATH)/io/vx_io.S $(VORTEX_RT_PATH)/io/vx_io.c
+VX_SRCS += $(VORTEX_RT_PATH)/fileio/fileio.S
+VX_SRCS += $(VORTEX_RT_PATH)/tests/tests.c
+VX_SRCS += $(VORTEX_RT_PATH)/vx_api/vx_api.c
+
+VX_CFLAGS = -nostartfiles -Wl,-Bstatic,-T,$(VORTEX_RT_PATH)/startup/vx_link.ld
+
+CXXFLAGS =  -g -O0 -march=rv32im -mabi=ilp32 
+CXXFLAGS += -ffreestanding # program may not begin at main()
+CXXFLAGS += -Wl,--gc-sections # enable garbage collection of unused input sections
+CXXFLAGS += -fno-rtti -fno-non-call-exceptions # disable RTTI and exceptions
+CXXFLAGS += -I$(POCL_INC_PATH) -I.
+
+VX_LIBS = -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
+QEMU_LIBS = $(VORTEX_RT_PATH)/qemu/vx_api.c -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
+
+PROJECT = lbm
+ 
+SRCS = main.cc args.c parboil_opencl.c gpu_info.c lbm.c ocl.c
+
+all: $(PROJECT).dump $(PROJECT).hex
+
+lib$(PROJECT).a: kernel.cl
+	POCL_DEBUG=all POCL_DEBUG_LLVM_PASSES=1 LD_LIBRARY_PATH=$(RISCV_TOOLCHAIN_PATH)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -o lib$(PROJECT).a kernel.cl
+
+$(PROJECT).elf: $(SRCS) lib$(PROJECT).a
+	$(CXX) $(CXXFLAGS) $(VX_CFLAGS) $(VX_SRCS) $(SRCS) $(VX_LIBS) -o $(PROJECT).elf
+
+$(PROJECT).qemu: $(SRCS) lib$(PROJECT).a
+	$(CXX) $(CXXFLAGS) $(SRCS) $(QEMU_LIBS) -o $(PROJECT).qemu
+
+$(PROJECT).hex: $(PROJECT).elf
+	$(HEX) -O ihex $(PROJECT).elf $(PROJECT).hex
+
+$(PROJECT).dump: $(PROJECT).elf
+	$(DMP) -D $(PROJECT).elf > $(PROJECT).dump
+
+run: $(PROJECT).hex
+	POCL_DEBUG=all $(VX_SIMX_PATH)/Vcache_simX -E -a rv32i --core $(PROJECT).hex -s -b 1> emulator.debug
+
+qemu: $(PROJECT).qemu
+	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -d in_asm -D debug.log $(PROJECT).qemu
+
+gdb-s: $(PROJECT).qemu
+	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -g 1234 -d in_asm -D debug.log $(PROJECT).qemu
+
+gdb-c: $(PROJECT).qemu
+	$(GDB) $(PROJECT).qemu
+
+clean:
+	rm -rf *.o *.elf *.dump *.hex *.qemu *.log *.debug

tests/opencl/lbm/args.c

@@ -0,0 +1,617 @@
+
+#include <parboil.h>
+#include <errno.h>
+#include <limits.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+
+/*****************************************************************************/
+/* Memory management routines */
+
+/* Free an array of owned strings. */
+void
+pb_FreeStringArray(char **string_array)
+{
+  char **p;
+
+  if (!string_array) return;
+  for (p = string_array; *p; p++) free(*p);
+  free(string_array);
+}
+
+struct pb_PlatformParam *
+pb_PlatformParam(char *name, char *version)
+{
+  if (name == NULL) {
+    fprintf(stderr, "pb_PlatformParam: Invalid argument\n");
+    exit(-1);
+  }
+
+  struct pb_PlatformParam *ret =
+    (struct pb_PlatformParam *)malloc(sizeof (struct pb_PlatformParam));
+
+  ret->name = name;
+  ret->version = version;
+  return ret;
+}
+
+void
+pb_FreePlatformParam(struct pb_PlatformParam *p)
+{
+  if (p == NULL) return;
+
+  free(p->name);
+  free(p->version);
+  free(p);
+}
+
+struct pb_DeviceParam *
+pb_DeviceParam_index(int index)
+{
+  struct pb_DeviceParam *ret =
+    (struct pb_DeviceParam *)malloc(sizeof (struct pb_DeviceParam));
+  ret->criterion = pb_Device_INDEX;
+  ret->index = index;
+  return ret;
+}
+
+struct pb_DeviceParam *
+pb_DeviceParam_cpu(void)
+{
+  struct pb_DeviceParam *ret =
+    (struct pb_DeviceParam *)malloc(sizeof (struct pb_DeviceParam));
+  ret->criterion = pb_Device_CPU;
+  return ret;
+}
+
+struct pb_DeviceParam *
+pb_DeviceParam_gpu(void)
+{
+  struct pb_DeviceParam *ret =
+    (struct pb_DeviceParam *)malloc(sizeof (struct pb_DeviceParam));
+  ret->criterion = pb_Device_GPU;
+  return ret;
+}
+
+struct pb_DeviceParam *
+pb_DeviceParam_accelerator(void)
+{
+  struct pb_DeviceParam *ret =
+    (struct pb_DeviceParam *)malloc(sizeof (struct pb_DeviceParam));
+  ret->criterion = pb_Device_ACCELERATOR;
+  return ret;
+}
+
+struct pb_DeviceParam *
+pb_DeviceParam_name(char *name)
+{
+  struct pb_DeviceParam *ret =
+    (struct pb_DeviceParam *)malloc(sizeof (struct pb_DeviceParam));
+  ret->criterion = pb_Device_NAME;
+  ret->name = name;
+  return ret;
+}
+
+void
+pb_FreeDeviceParam(struct pb_DeviceParam *p)
+{
+  if (p == NULL) return;
+
+  switch(p->criterion) {
+  case pb_Device_NAME:
+    free(p->name);
+    break;
+  case pb_Device_INDEX:
+  case pb_Device_CPU:
+  case pb_Device_ACCELERATOR:
+    break;
+  default:
+    fprintf(stderr, "pb_FreeDeviceParam: Invalid argument\n");
+    exit(-1);
+  }
+}
+
+void
+pb_FreeParameters(struct pb_Parameters *p)
+{
+  free(p->outFile);
+  pb_FreeStringArray(p->inpFiles);
+  pb_FreePlatformParam(p->platform);
+  pb_FreeDeviceParam(p->device);
+  free(p);
+}
+
+/*****************************************************************************/
+
+/* Parse a comma-delimited list of strings into an
+ * array of strings. */
+static char ** 
+read_string_array(char *in)
+{
+  char **ret;
+  int i;
+  int count;			/* Number of items in the input */
+  char *substring;		/* Current substring within 'in' */
+
+  /* Count the number of items in the string */
+  count = 1;
+  for (i = 0; in[i]; i++) if (in[i] == ',') count++;
+
+  /* Allocate storage */
+  ret = (char **)malloc((count + 1) * sizeof(char *));
+
+  /* Create copies of the strings from the list */
+  substring = in;
+  for (i = 0; i < count; i++) {
+    char *substring_end;
+    int substring_length;
+
+    /* Find length of substring */
+    for (substring_end = substring;
+	 (*substring_end != ',') && (*substring_end != 0);
+	 substring_end++);
+
+    substring_length = substring_end - substring;
+
+    /* Allocate memory and copy the substring */
+    ret[i] = (char *)malloc(substring_length + 1);
+    memcpy(ret[i], substring, substring_length);
+    ret[i][substring_length] = 0;
+
+    /* go to next substring */
+    substring = substring_end + 1;
+  }
+  ret[i] = NULL;		/* Write the sentinel value */
+
+  return ret;
+}
+
+static void
+report_parse_error(const char *str)
+{
+  fputs(str, stderr);
+}
+
+/* Interpret a string as a 'pb_DeviceParam' value.
+ * Return a pointer to a new value, or NULL on failure.
+ */
+static struct pb_DeviceParam *
+read_device_param(char *str)
+{  
+  /* Try different ways of interpreting 'device_string' until one works */
+
+  /* If argument is an integer, then interpret it as a device index */
+  errno = 0;
+  char *end;
+  long device_int = strtol(str, &end, 10);
+  if (!errno) {
+    /* Negative numbers are not valid */
+    if (device_int < 0 || device_int > INT_MAX) return NULL;
+
+    return pb_DeviceParam_index(device_int);
+  }
+
+  /* Match against predefined strings */
+  if (strcmp(str, "CPU") == 0)
+    return pb_DeviceParam_cpu();
+  if (strcmp(str, "GPU") == 0)
+    return pb_DeviceParam_gpu();
+  if (strcmp(str, "ACCELERATOR") == 0)
+    return pb_DeviceParam_accelerator();
+
+  /* Assume any other string is a device name */
+  return pb_DeviceParam_name(strdup(str));
+}
+
+/* Interpret a string as a 'pb_PlatformParam' value.
+ * Return a pointer to a new value, or NULL on failure.
+ */
+static struct pb_PlatformParam *
+read_platform_param(char *str)
+{
+  int separator_index;          /* Index of the '-' character separating
+                                 * name and version number.  It's -1 if
+                                 * there's no '-' character. */
+
+  /* Find the last occurrence of '-' in 'str' */
+  {
+    char *cur;
+    separator_index = -1;
+    for (cur = str; *cur; cur++) {
+      if (*cur == '-') separator_index = cur - str;
+    }
+  }
+
+  /* The platform name is either the entire string, or all characters before
+   * the separator */
+  int name_length = separator_index == -1 ? strlen(str) : separator_index;
+  char *name_str = (char *)malloc(name_length + 1);
+  memcpy(name_str, str, name_length);
+  name_str[name_length] = 0;
+
+  /* The version is either NULL, or all characters after the separator */
+  char *version_str;
+  if (separator_index == -1) {
+    version_str = NULL;
+  }
+  else {
+    const char *version_input_str = str + separator_index + 1;
+    int version_length = strlen(version_input_str);
+
+    version_str = (char *)malloc(version_length + 1);
+    memcpy(version_str, version_input_str, version_length);
+    version_str[version_length] = 0;
+  }
+
+  /* Create output structure */
+  return pb_PlatformParam(name_str, version_str);
+}
+
+/****************************************************************************/
+/* Argument parsing state */
+
+/* Argument parsing state.
+ *
+ * Arguments that are interpreted by the argument parser are removed from
+ * the list.  Variables 'argc' and 'argn' do not count arguments that have
+ * been removed.
+ *
+ * During argument parsing, the array of arguments is compacted, overwriting
+ * the erased arguments.  Variable 'argv_put' points to the array element
+ * where the next argument will be written.  Variable 'argv_get' points to
+ * the array element where the next argument will be read from.
+ */
+struct argparse {
+  int argc;			/* Number of arguments.  Mutable. */
+  int argn;			/* Current argument index. */
+  char **argv_get;		/* Argument value being read. */
+  char **argv_put;		/* Argument value being written.
+				 * argv_put <= argv_get. */
+};
+
+static void
+initialize_argparse(struct argparse *ap, int argc, char **argv)
+{
+  ap->argc = argc;
+  ap->argn = 0;
+  ap->argv_get = ap->argv_put = argv;
+}
+
+/* Finish argument parsing, without processing the remaining arguments.
+ * Write new argument count into _argc. */
+static void
+finalize_argparse(struct argparse *ap, int *_argc, char **argv)
+{
+  /* Move the remaining arguments */
+  for(; ap->argn < ap->argc; ap->argn++)
+    *ap->argv_put++ = *ap->argv_get++;
+
+  /* Update the argument count */
+  *_argc = ap->argc;
+
+  /* Insert a terminating NULL */
+  argv[ap->argc] = NULL;
+}
+
+/* Delete the current argument.  The argument will not be visible
+ * when argument parsing is done. */
+static void
+delete_argument(struct argparse *ap)
+{
+  if (ap->argn >= ap->argc) {
+    fprintf(stderr, "delete_argument\n");
+  }
+  ap->argc--;
+  ap->argv_get++;
+}
+
+/* Go to the next argument.  Also, move the current argument to its
+ * final location in argv. */
+static void
+next_argument(struct argparse *ap)
+{
+  if (ap->argn >= ap->argc) {
+    fprintf(stderr, "next_argument\n");
+  }
+  /* Move argument to its new location. */
+  *ap->argv_put++ = *ap->argv_get++;
+  ap->argn++;
+}
+
+static int
+is_end_of_arguments(struct argparse *ap)
+{
+  return ap->argn == ap->argc;
+}
+
+/* Get the current argument */
+static char *
+get_argument(struct argparse *ap)
+{
+  return *ap->argv_get;
+}
+
+/* Get the current argument, and also delete it */
+static char *
+consume_argument(struct argparse *ap)
+{
+  char *ret = get_argument(ap);
+  delete_argument(ap);
+  return ret;
+}
+
+/****************************************************************************/
+
+/* The result of parsing a command-line argument */
+typedef enum {
+  ARGPARSE_OK,                  /* Success */
+  ARGPARSE_ERROR,               /* Error */
+  ARGPARSE_DONE                 /* Success, and do not continue parsing */
+} result;
+
+typedef result parse_action(struct argparse *ap, struct pb_Parameters *params);
+
+
+/* A command-line option */
+struct option {
+  char short_name;              /* If not 0, the one-character
+                                 * name of this option */
+  const char *long_name;        /* If not NULL, the long name of this option */
+  parse_action *action;       /* What to do when this option occurs.
+                               * Sentinel value is NULL.
+                               */
+};
+
+/* Output file
+ *
+ * -o FILE
+ */
+static result
+parse_output_file(struct argparse *ap, struct pb_Parameters *params)
+{
+  if (is_end_of_arguments(ap))
+    {
+      report_parse_error("Expecting file name after '-o'\n");
+      return ARGPARSE_ERROR;
+    }
+
+  /* Replace the output file name */
+  free(params->outFile);
+  params->outFile = strdup(consume_argument(ap));
+
+  return ARGPARSE_OK;
+}
+
+/* Input files
+ *
+ * -i FILE,FILE,...
+ */
+static result
+parse_input_files(struct argparse *ap, struct pb_Parameters *params)
+{
+  if (is_end_of_arguments(ap))
+    {
+      report_parse_error("Expecting file name after '-i'\n");
+      return ARGPARSE_ERROR;
+    }
+
+  /* Replace the input file list */
+  pb_FreeStringArray(params->inpFiles);
+  params->inpFiles = read_string_array(consume_argument(ap));
+  return ARGPARSE_OK;
+}
+
+/* End of options
+ *
+ * --
+ */
+
+static result
+parse_end_options(struct argparse *ap, struct pb_Parameters *params)
+{
+  return ARGPARSE_DONE;
+}
+
+/* OpenCL device
+ *
+ * --device X
+ */
+
+static result
+parse_device(struct argparse *ap, struct pb_Parameters *params)
+{
+  /* Read the next argument, which specifies a device */
+
+  if (is_end_of_arguments(ap))
+    {
+      report_parse_error("Expecting device specification after '--device'\n");
+      return ARGPARSE_ERROR;
+    }
+
+  char *device_string = consume_argument(ap);
+  struct pb_DeviceParam *device_param = read_device_param(device_string);
+
+  if (!device_param) {
+    report_parse_error("Unrecognized device specification format on command line\n");
+    return ARGPARSE_ERROR;
+  }
+
+  /* Save the result */
+  pb_FreeDeviceParam(params->device);
+  params->device = device_param;
+
+  return ARGPARSE_OK;
+}
+
+static result
+parse_platform(struct argparse *ap, struct pb_Parameters *params)
+{
+  /* Read the next argument, which specifies a platform */
+
+  if (is_end_of_arguments(ap))
+    {
+      report_parse_error("Expecting device specification after '--platform'\n");
+      return ARGPARSE_ERROR;
+    }
+
+  char *platform_string = consume_argument(ap);
+  struct pb_PlatformParam *platform_param = read_platform_param(platform_string);
+
+  if (!platform_param) {
+    report_parse_error("Unrecognized platform specification format on command line\n");
+    return ARGPARSE_ERROR;
+  }
+
+  /* Save the result */
+  pb_FreePlatformParam(params->platform);
+  params->platform = platform_param;
+
+  return ARGPARSE_OK;
+}
+  
+
+static struct option options[] = {
+  { 'o', NULL, &parse_output_file },
+  { 'i', NULL, &parse_input_files },
+  { '-', NULL, &parse_end_options },
+  { 0, "device", &parse_device },
+  { 0, "platform", &parse_platform },
+  { 0, NULL, NULL }
+};
+
+static int
+is_last_option(struct option *op)
+{
+  return op->action == NULL;
+}
+
+/****************************************************************************/
+
+/* Parse command-line parameters.
+ * Return zero on error, nonzero otherwise.
+ * On error, the other outputs may be invalid.
+ *
+ * The information collected from parameters is used to update
+ * 'ret'.  'ret' should be initialized.
+ *
+ * '_argc' and 'argv' are updated to contain only the unprocessed arguments.
+ */
+static int
+pb_ParseParameters (struct pb_Parameters *ret, int *_argc, char **argv)
+{
+  char *err_message;
+  struct argparse ap;
+
+  /* Each argument */
+  initialize_argparse(&ap, *_argc, argv);
+  while(!is_end_of_arguments(&ap)) {
+    result arg_result;          /* Result of parsing this option */
+    char *arg = get_argument(&ap);
+
+    /* Process this argument */
+    if (arg[0] == '-') {
+      /* Single-character flag */
+      if ((arg[1] != 0) && (arg[2] == 0)) {
+        delete_argument(&ap);	/* This argument is consumed here */
+
+        /* Find a matching short option */
+        struct option *op;
+        for (op = options; !is_last_option(op); op++) {
+          if (op->short_name == arg[1]) {
+            arg_result = (*op->action)(&ap, ret);
+            goto option_was_processed;
+          }
+        }
+
+        /* No option matches */
+        report_parse_error("Unexpected command-line parameter\n");
+        arg_result = ARGPARSE_ERROR;
+        goto option_was_processed;
+      }
+
+      /* Long flag */
+      if (arg[1] == '-') {
+        delete_argument(&ap);	/* This argument is consumed here */
+
+        /* Find a matching long option */
+        struct option *op;
+        for (op = options; !is_last_option(op); op++) {
+          if (op->long_name && strcmp(&arg[2], op->long_name) == 0) {
+            arg_result = (*op->action)(&ap, ret);
+            goto option_was_processed;
+          }
+        }
+
+        /* No option matches */
+        report_parse_error("Unexpected command-line parameter\n");
+        arg_result = ARGPARSE_ERROR;
+        goto option_was_processed;
+      }
+    }
+    else {
+      /* Other arguments are ignored */
+      next_argument(&ap);
+      arg_result = ARGPARSE_OK;
+      goto option_was_processed;
+    }
+
+  option_was_processed:
+    /* Decide what to do next based on 'arg_result' */
+    switch(arg_result) {
+    case ARGPARSE_OK:
+      /* Continue processing */
+      break;
+
+    case ARGPARSE_ERROR:
+      /* Error exit from the function */
+      return 0;
+
+    case ARGPARSE_DONE:
+      /* Normal exit from the argument parsing loop */
+      goto end_of_options;
+    }
+  } /* end for each argument */
+
+  /* If all arguments were processed, then normal exit from the loop */
+
+ end_of_options:
+  finalize_argparse(&ap, _argc, argv);
+  return 1;
+}
+
+/*****************************************************************************/
+/* Other exported functions */
+
+struct pb_Parameters *
+pb_ReadParameters(int *_argc, char **argv)
+{
+  struct pb_Parameters *ret =
+    (struct pb_Parameters *)malloc(sizeof(struct pb_Parameters));
+
+  /* Initialize the parameters structure */
+  ret->outFile = NULL;
+  ret->inpFiles = (char **)malloc(sizeof(char *));
+  ret->inpFiles[0] = NULL;
+  ret->platform = NULL;
+  ret->device = NULL;
+
+  /* Read parameters and update _argc, argv */
+  if (!pb_ParseParameters(ret, _argc, argv)) {
+    /* Parse error */
+    pb_FreeParameters(ret);
+    return NULL;
+  }
+
+  return ret;
+}
+
+int
+pb_Parameters_CountInputs(struct pb_Parameters *p)
+{
+  int n;
+
+  for (n = 0; p->inpFiles[n]; n++);
+  return n;
+}
+