wu-arch/vortex
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update

Browse Source
This commit is contained in:
Blaise Tine
2019-11-24 00:43:03 -05:00
parent cec065b6fd
commit f04c2e0be8
1 changed files with 702 additions and 704 deletions
Download Patch File Download Diff File Expand all files Collapse all files

312
benchmarks/opencl/bfs/CLHelper.h
View File

@@ -7,10 +7,11 @@
#define _CL_HELPER_
#include <CL/cl.h>
#include <vector>
#include <iostream>
#include <fstream>
#include <iostream>
#include <string>
#include <vector>
using std::string;
using std::ifstream;
@@ -20,8 +21,7 @@ using std::cout;
//#pragma OPENCL EXTENSION cl_nv_compiler_options:enable
#define WORK_DIM 2 // work-items dimensions
struct oclHandleStruct
{
struct oclHandleStruct {
cl_context context;
cl_device_id *devices;
cl_command_queue queue;
@@ -42,25 +42,23 @@ int device_id_inused = 0; //deviced id used (default : 0)
/*
* Converts the contents of a file into a string
*/
string FileToString(const string fileName)
{
string FileToString(const string fileName) {
ifstream f(fileName.c_str(), ifstream::in | ifstream::binary);
try
{
try {
size_t size;
char *str;
string s;
if(f.is_open())
{
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"));
if (!str)
throw(string("Could not allocate memory"));
f.read(str, fileSize);
f.close();
@@ -70,58 +68,45 @@ string FileToString(const string fileName)
delete[] str;
return s;
}
}
catch(std::string msg)
{
} catch (std::string msg) {
cerr << "Exception caught in FileToString(): " << msg << endl;
if (f.is_open())
f.close();
}
catch(...)
{
} catch (...) {
cerr << "Exception caught in FileToString()" << endl;
if (f.is_open())
f.close();
}
string errorMsg = "FileToString()::Error: Unable to open file "
+ fileName;
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])
{
for (int i = 0; i < argc; ++i) {
switch (argv[i][1]) {
case 'g': //--g stands for size of work group
if (++i < argc)
{
if (++i < argc) {
sscanf(argv[i], "%u", &work_group_size);
}
else
{
std::cerr << "Could not read argument after option " << argv[i-1] << std::endl;
} 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)
{
if (++i < argc) {
sscanf(argv[i], "%u", &device_id_inused);
}
else
{
std::cerr << "Could not read argument after option " << argv[i-1] << std::endl;
} else {
std::cerr << "Could not read argument after option " << argv[i - 1]
<< std::endl;
throw;
}
break;
default:
;
default:;
}
}
}
//---------------------------------------
@@ -129,9 +114,9 @@ void _clCmdParams(int argc, char* argv[]){
//--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()
{
void _clInit() {
printf("_clInit()\n");
int DEVICE_ID_INUSED = device_id_inused;
cl_int resultCL;
@@ -145,26 +130,20 @@ void _clInit()
//-----------------------------------------------
//--cambine-1: find the available platforms and select one
cl_uint numPlatforms;
cl_uint numPlatforms = 1;
cl_platform_id targetPlatform = NULL;
resultCL = clGetPlatformIDs(0, NULL, &numPlatforms);
if (resultCL != CL_SUCCESS)
throw (string("InitCL()::Error: Getting number of platforms (clGetPlatformIDs)"));
//printf("number of platforms:%d\n",numPlatforms); //by cambine
if (!(numPlatforms > 0))
throw (string("InitCL()::Error: No platforms found (clGetPlatformIDs)"));
cl_platform_id* allPlatforms = (cl_platform_id*) malloc(numPlatforms * sizeof(cl_platform_id));
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 */
// Select the target platform. Default: first platform
targetPlatform = allPlatforms[0];
for (int i = 0; i < numPlatforms; i++)
/*for (int i = 0; i < numPlatforms; i++)
{
char pbuff[128];
resultCL = clGetPlatformInfo( allPlatforms[i],
@@ -178,11 +157,12 @@ void _clInit()
//printf("vedor is %s\n",pbuff);
}
free(allPlatforms);
free(allPlatforms);*/
//-----------------------------------------------
//--cambine-2: create an OpenCL context
cl_context_properties cprops[3] = { CL_CONTEXT_PLATFORM, (cl_context_properties)targetPlatform, 0 };
/*cl_context_properties cprops[3] = { CL_CONTEXT_PLATFORM,
(cl_context_properties)targetPlatform, 0 };
oclHandles.context = clCreateContextFromType(cprops,
CL_DEVICE_TYPE_GPU,
NULL,
@@ -190,37 +170,49 @@ void _clInit()
&resultCL);
if ((resultCL != CL_SUCCESS) || (oclHandles.context == NULL))
throw (string("InitCL()::Error: Creating Context (clCreateContextFromType)"));
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);
// 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."));
//std::cout<<"device number:"<<deviceListSize<<std::endl;
printf("OK1()\n");
/* Now, allocate the device list */
oclHandles.devices = (cl_device_id *)malloc(deviceListSize * sizeof(cl_device_id));
//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_GPU, deviceListSize, \
//* 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);
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)"));
@@ -230,45 +222,41 @@ void _clInit()
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 = clCreateProgramWithBuiltInKernels(
oclHandles.context, 1, &oclHandles.devices[DEVICE_ID_INUSED],
"BFS_1;BFS_2", &resultCL);
/*oclHandles.program = clCreateProgramWithSource(oclHandles.context,
1,
&source,
sourceSize,
&resultCL);*/
if ((resultCL != CL_SUCCESS) || (oclHandles.program == NULL))
throw(string("InitCL()::Error: Loading Binary into cl_program. (clCreateProgramWithBinary)"));
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))
{
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);
CL_PROGRAM_BUILD_LOG, 0, NULL, &length);
if (resultCL != CL_SUCCESS)
throw(string("InitCL()::Error: Getting Program build info(clGetProgramBuildInfo)"));
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);
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)"));
throw(string("InitCL()::Error: Getting Program build "
"info(clGetProgramBuildInfo)"));
cerr << buffer << endl;
free(buffer);
@@ -281,7 +269,9 @@ void _clInit()
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 );
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"));
}
@@ -290,7 +280,9 @@ void _clInit()
// 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 );
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"));
}
@@ -308,16 +300,14 @@ void _clInit()
free(binaries[i]);
#endif
for (int nKernel = 0; nKernel < total_kernels; nKernel++)
{
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);
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] + "\"";
if ((resultCL != CL_SUCCESS) || (kernel == NULL)) {
string errorMsg = "InitCL()::Error: Creating Kernel (clCreateKernel) \"" +
kernel_names[nKernel] + "\"";
throw(errorMsg);
}
@@ -327,15 +317,20 @@ void _clInit()
#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);
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);
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"));
throw(string(
"exceptions in _InitCL -> getting resources allocation information-2"));
}
build_log[ret_val_size] = '\0';
std::cout << "--cambine:" << build_log << std::endl;
@@ -345,17 +340,13 @@ void _clInit()
//---------------------------------------
// release CL objects
void _clRelease()
{
void _clRelease() {
char errorFlag = false;
for (int nKernel = 0; nKernel < oclHandles.kernel.size(); nKernel++)
{
if (oclHandles.kernel[nKernel] != NULL)
{
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)
{
if (resultCL != CL_SUCCESS) {
cerr << "ReleaseCL()::Error: In clReleaseKernel" << endl;
errorFlag = true;
}
@@ -364,22 +355,18 @@ void _clRelease()
oclHandles.kernel.clear();
}
if (oclHandles.program != NULL)
{
if (oclHandles.program != NULL) {
cl_int resultCL = clReleaseProgram(oclHandles.program);
if (resultCL != CL_SUCCESS)
{
if (resultCL != CL_SUCCESS) {
cerr << "ReleaseCL()::Error: In clReleaseProgram" << endl;
errorFlag = true;
}
oclHandles.program = NULL;
}
if (oclHandles.queue != NULL)
{
if (oclHandles.queue != NULL) {
cl_int resultCL = clReleaseCommandQueue(oclHandles.queue);
if (resultCL != CL_SUCCESS)
{
if (resultCL != CL_SUCCESS) {
cerr << "ReleaseCL()::Error: In clReleaseCommandQueue" << endl;
errorFlag = true;
}
@@ -388,25 +375,25 @@ void _clRelease()
free(oclHandles.devices);
if (oclHandles.context != NULL)
{
if (oclHandles.context != NULL) {
cl_int resultCL = clReleaseContext(oclHandles.context);
if (resultCL != CL_SUCCESS)
{
if (resultCL != CL_SUCCESS) {
cerr << "ReleaseCL()::Error: In clReleaseContext" << endl;
errorFlag = true;
}
oclHandles.context = NULL;
}
if (errorFlag) throw(string("ReleaseCL()::Error encountered."));
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);
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()"));
@@ -418,7 +405,9 @@ cl_mem _clCreateAndCpyMem(int size, void * h_mem_source) throw(string){
//--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);
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"));
@@ -430,7 +419,9 @@ cl_mem _clMallocRW(int size, void * h_mem_ptr) throw(string){
//--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);
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"));
@@ -442,7 +433,8 @@ cl_mem _clMalloc(int size, void * h_mem_ptr) throw(string){
//--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);
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"));
@@ -454,22 +446,23 @@ void _clMemcpyH2D(cl_mem d_mem, int size, const void *h_mem_ptr) throw(string){
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);
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);
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, \
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)
@@ -481,9 +474,8 @@ cl_mem _clCreateAndCpyPinnedMem(int size, float* h_mem_source) throw(string){
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);
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"));
@@ -492,12 +484,12 @@ cl_mem _clCreateAndCpyPinnedMem(int size, float* h_mem_source) throw(string){
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);
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()"));
@@ -508,7 +500,8 @@ cl_mem _clMallocWO(int size) throw(string){
//--------------------------------------------------------
// 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);
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) {
@@ -544,9 +537,11 @@ void _clMemcpyD2H(cl_mem d_mem, int size, void * h_mem) throw(string){
//--------------------------------------------------------
// set kernel arguments
void _clSetArgs(int kernel_id, int arg_idx, void * d_mem, int size = 0) throw(string){
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);
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) {
@@ -581,9 +576,9 @@ void _clSetArgs(int kernel_id, int arg_idx, void * d_mem, int size = 0) throw(st
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);
} 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) {
@@ -637,7 +632,6 @@ void _clFinish() throw(string){
default:
oclHandles.error_str += "Unknown reasons";
break;
}
if (oclHandles.cl_status != CL_SUCCESS) {
throw(oclHandles.error_str);
@@ -646,19 +640,22 @@ void _clFinish() throw(string){
}
//--------------------------------------------------------
//--cambine:enqueue kernel
void _clInvokeKernel(int kernel_id, int work_items, int work_group_size) throw(string){
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));
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, \
oclHandles.cl_status = clEnqueueNDRangeKernel(
oclHandles.queue, oclHandles.kernel[kernel_id], work_dim, 0,
global_work_size, local_work_size, 0, 0, &(e[0]));
#ifdef ERRMSG
oclHandles.error_str = "excpetion in _clInvokeKernel() -> ";
switch(oclHandles.cl_status)
{
switch (oclHandles.cl_status) {
case CL_INVALID_PROGRAM_EXECUTABLE:
oclHandles.error_str += "CL_INVALID_PROGRAM_EXECUTABLE";
break;
@@ -715,19 +712,21 @@ void _clInvokeKernel(int kernel_id, int work_items, int work_group_size) throw(s
// 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){
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
/*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, \
oclHandles.cl_status = clEnqueueNDRangeKernel(
oclHandles.queue, oclHandles.kernel[kernel_id], work_dim, 0,
global_work_size, local_work_size, 0, 0, &(e[0]));
#ifdef ERRMSG
oclHandles.error_str = "excpetion in _clInvokeKernel() -> ";
switch(oclHandles.cl_status)
{
switch (oclHandles.cl_status) {
case CL_INVALID_PROGRAM_EXECUTABLE:
oclHandles.error_str += "CL_INVALID_PROGRAM_EXECUTABLE";
break;
@@ -796,8 +795,7 @@ void _clFree(cl_mem ob) throw(string){
oclHandles.cl_status = clReleaseMemObject(ob);
#ifdef ERRMSG
oclHandles.error_str = "excpetion in _clFree() ->";
switch(oclHandles.cl_status)
{
switch (oclHandles.cl_status) {
case CL_INVALID_MEM_OBJECT:
oclHandles.error_str += "CL_INVALID_MEM_OBJECT";
break;