Vortex 2.0 changes:
+ Microarchitecture optimizations + 64-bit support + Xilinx FPGA support + LLVM-16 support + Refactoring and quality control fixes
This commit is contained in:
Blaise Tine
@@ -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 = spmv
|
||||
|
||||
SRCS = main.cc parboil_opencl.c args.c gpu_info.c file.c convert_dataset.c mmio.c ocl.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
|
||||
|
||||
@@ -8,6 +8,10 @@
|
||||
#ifndef __FILEH__
|
||||
#define __FILEH__
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
void inputData(char* fName, int* len, int* depth, int* dim,int *nzcnt_len,int *pad,
|
||||
float** h_data, int** h_indices, int** h_ptr,
|
||||
int** h_perm, int** h_nzcnt);
|
||||
@@ -15,4 +19,8 @@ void inputData(char* fName, int* len, int* depth, int* dim,int *nzcnt_len,int *p
|
||||
void input_vec(char* fNanme, float *h_vec,int dim);
|
||||
void outputData(char* fName, float *h_Ax_vector,int dim);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
|
||||
#endif
|
||||
@@ -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
|
||||
|
||||
Binary file not shown.
@@ -17,6 +17,26 @@
|
||||
#include "gpu_info.h"
|
||||
#include "ocl.h"
|
||||
|
||||
static int read_kernel_file(const char* filename, uint8_t** data, size_t* size) {
|
||||
if (nullptr == filename || nullptr == 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;
|
||||
}
|
||||
|
||||
static int generate_vector(float *x_vector, int dim) {
|
||||
srand(54321);
|
||||
@@ -43,8 +63,6 @@ int main(int argc, char **argv) {
|
||||
strncpy(parameters->inpFiles[0], "1138_bus.mtx", 100);
|
||||
strncpy(parameters->inpFiles[1], "vector.bin", 100);
|
||||
|
||||
printf("OK\n");
|
||||
|
||||
if ((parameters->inpFiles[0] == NULL) || (parameters->inpFiles[1] == NULL)) {
|
||||
fprintf(stderr, "Expecting one input filename\n");
|
||||
exit(-1);
|
||||
@@ -63,8 +81,6 @@ int main(int argc, char **argv) {
|
||||
return -1;
|
||||
}
|
||||
|
||||
printf("OK\n");
|
||||
|
||||
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;
|
||||
@@ -72,19 +88,20 @@ int main(int argc, char **argv) {
|
||||
clContext, clDevice, CL_QUEUE_PROFILING_ENABLE, &clStatus);
|
||||
CHECK_ERROR("clCreateCommandQueue")
|
||||
|
||||
printf("OK\n");
|
||||
|
||||
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, "spmv_jds_naive", &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")
|
||||
|
||||
printf("OK\n");
|
||||
|
||||
char clOptions[50];
|
||||
sprintf(clOptions, "");
|
||||
clStatus = clBuildProgram(clProgram, 1, &clDevice, clOptions, NULL, NULL);
|
||||
@@ -93,8 +110,6 @@ int main(int argc, char **argv) {
|
||||
cl_kernel clKernel = clCreateKernel(clProgram, "spmv_jds_naive", &clStatus);
|
||||
CHECK_ERROR("clCreateKernel")
|
||||
|
||||
printf("OK\n");
|
||||
|
||||
int len;
|
||||
int depth;
|
||||
int dim;
|
||||
|
||||
@@ -1,6 +1,10 @@
|
||||
#ifndef __OCLH__
|
||||
#define __OCLH__
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
typedef struct {
|
||||
cl_uint major;
|
||||
cl_uint minor;
|
||||
@@ -18,4 +22,8 @@ char* readFile(const char*);
|
||||
exit(1); \
|
||||
}
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
|
||||
#endif
|
||||
|
||||
Reference in New Issue
Block a user