Vortex 2.0 changes:

+ Microarchitecture optimizations + 64-bit support + Xilinx FPGA support + LLVM-16 support + Refactoring and quality control fixes
2023-10-19 20:51:22 -07:00
parent d69a64c32c
commit d47cccc157
1300 changed files with 247321 additions and 311189 deletions
--- a/tests/opencl/cutcp/Makefile
+++ b/tests/opencl/cutcp/Makefile
@@ -1,69 +1,9 @@
-XLEN ?= 32
-
-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$(XLEN).ld
-
-CXXFLAGS =  -g -O0 -march=rv32im -mabi=ilp32 
-CXXFLAGS += -ffreestanding # program may not begin at main()
-CXXFLAGS += -Wl,--gc-sections # enable garbage collection of unused input sections
-CXXFLAGS += -fno-rtti -fno-non-call-exceptions # disable RTTI and exceptions
-CXXFLAGS += -I$(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
+CXXFLAGS += -I.

-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
+OPTS ?= 

-$(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 $(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
+include ../common.mk
--- a/tests/opencl/cutcp/cutoff.c
+++ b/tests/opencl/cutcp/cutoff.c
@@ -19,6 +19,27 @@
 #include "macros.h"
 #include "ocl.h"

+static int read_kernel_file(const char* filename, uint8_t** data, size_t* size) {
+  if (NULL == filename || NULL == data || 0 == size)
+    return CL_INVALID_VALUE;
+
+  FILE* fp = fopen(filename, "r");
+  if (NULL == fp) {
+    fprintf(stderr, "Failed to load kernel.");
+    return CL_INVALID_VALUE;
+  }
+  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 CL_SUCCESS;
+}
+
 // 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;
@@ -294,8 +315,6 @@ int gpu_compute_cutoff_potential_lattice(
    printf("\n");
  }

-  printf("Ok!\n");
-
  pb_Context* pb_context;
  pb_context = pb_InitOpenCLContext(parameters);
  if (pb_context == NULL) {
@@ -303,8 +322,6 @@ int gpu_compute_cutoff_potential_lattice(
    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;
@@ -317,8 +334,13 @@ int gpu_compute_cutoff_potential_lattice(
  
  //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);
+  uint8_t *kernel_bin = NULL;
+  size_t kernel_size;
+  cl_int binary_status = 0;  
+  clStatus = read_kernel_file("kernel.pocl", &kernel_bin, &kernel_size);
+  CHECK_ERROR("read_kernel_file")  
+	cl_program clProgram = clCreateProgramWithBinary(
+      clContext, 1, &clDevice, &kernel_size, (const uint8_t**)&kernel_bin, &binary_status, &clStatus);
  CHECK_ERROR("clCreateProgramWithSource")

  char clOptions[50];
@@ -399,9 +421,6 @@ int gpu_compute_cutoff_potential_lattice(
  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);
@@ -412,26 +431,16 @@ int gpu_compute_cutoff_potential_lattice(
    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");
+  printf("Finished OpenCL kernel calls\n");

  /* copy result regions from OpenCL device */
  pb_SwitchToTimer(timers, pb_TimerID_COPY);
--- a/tests/opencl/cutcp/gpu_info.h
+++ b/tests/opencl/cutcp/gpu_info.h
@@ -9,6 +9,10 @@
 #ifndef __GPUINFOH__
 #define __GPUINFOH__

+#ifdef __cplusplus
+extern "C" {
+#endif
+
 void compute_active_thread(size_t *thread,
 			   size_t *grid,
 			   int task,
@@ -17,4 +21,8 @@ void compute_active_thread(size_t *thread,
 			   int minor,
 			   int sm);

+#ifdef __cplusplus
+}
+#endif
+
 #endif
--- a/tests/opencl/cutcp/libcutcp.a
+++ b/tests/opencl/cutcp/libcutcp.a
--- a/tests/opencl/cutcp/main.cc
+++ b/tests/opencl/cutcp/main.cc
@@ -124,8 +124,6 @@ int main(int argc, char *argv[]) {
  pb_InitializeTimerSet(&timers);
  pb_SwitchToTimer(&timers, pb_TimerID_IO);

-  printf("OK\n");
-
  {
    const char *pqrfilename = parameters->inpFiles[0];

@@ -136,8 +134,6 @@ int main(int argc, char *argv[]) {
    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);
--- a/tests/opencl/cutcp/ocl.h
+++ b/tests/opencl/cutcp/ocl.h
@@ -3,6 +3,10 @@

 #include <stdlib.h>

+#ifdef __cplusplus
+extern "C" {
+#endif
+
 void clMemSet(cl_command_queue, cl_mem, int, size_t);
 char* readFile(const char*);

@@ -14,4 +18,8 @@ char* readFile(const char*);
     exit(1);                               \
  }

+#ifdef __cplusplus
+}
+#endif
+
 #endif