github-mirrors/vortex
1
0
Fork 0
mirror of https://github.com/vortexgpgpu/vortex.git synced 2025-04-23 21:39:10 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Added prefetch test to commit

Browse source
This commit is contained in:
Santosh Raghav Srivatsan 2021-11-08 16:32:52 -05:00
parent 66e253ab95
commit ad633aa639
4 changed files with 310 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

70
tests/regression/prefetch/Makefile Normal file
View file

@ -0,0 +1,70 @@
RISCV_TOOLCHAIN_PATH ?= /opt/riscv-gnu-toolchain
VORTEX_DRV_PATH ?= $(realpath ../../../driver)
VORTEX_RT_PATH ?= $(realpath ../../../runtime)
OPTS ?= -n64
VX_CC = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gcc
VX_CXX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-g++
VX_DP = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objdump
VX_CP = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objcopy
VX_CFLAGS += -march=rv32imf -mabi=ilp32f -O3 -Wstack-usage=1024 -ffreestanding -nostartfiles -fdata-sections -ffunction-sections
VX_CFLAGS += -I$(VORTEX_RT_PATH)/include -I$(VORTEX_RT_PATH)/../hw
VX_LDFLAGS += -Wl,-Bstatic,-T,$(VORTEX_RT_PATH)/linker/vx_link.ld -Wl,--gc-sections $(VORTEX_RT_PATH)/libvortexrt.a
VX_SRCS = kernel.c
#CXXFLAGS += -std=c++11 -O2 -Wall -Wextra -pedantic -Wfatal-errors
CXXFLAGS += -std=c++11 -O0 -g -Wall -Wextra -pedantic -Wfatal-errors
CXXFLAGS += -I$(VORTEX_DRV_PATH)/include
LDFLAGS += -L$(VORTEX_DRV_PATH)/stub -lvortex
PROJECT = prefetch
SRCS = main.cpp
all: $(PROJECT) kernel.bin kernel.dump
kernel.dump: kernel.elf
$(VX_DP) -D kernel.elf > kernel.dump
kernel.bin: kernel.elf
$(VX_CP) -O binary kernel.elf kernel.bin
kernel.elf: $(VX_SRCS)
$(VX_CC) $(VX_CFLAGS) $(VX_SRCS) $(VX_LDFLAGS) -o kernel.elf
$(PROJECT): $(SRCS)
$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -o $@
run-simx: $(PROJECT) kernel.bin
LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/simx:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
run-fpga: $(PROJECT) kernel.bin
LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/fpga:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
run-asesim: $(PROJECT) kernel.bin
LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/asesim:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
run-vlsim: $(PROJECT) kernel.bin
LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/vlsim:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
run-rtlsim: $(PROJECT) kernel.bin
LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
.depend: $(SRCS)
$(CXX) $(CXXFLAGS) -MM $^ > .depend;
clean:
rm -rf $(PROJECT) *.o .depend
clean-all: clean
rm -rf *.elf *.bin *.dump
ifneq ($(MAKECMDGOALS),clean)
-include .depend
endif

14
tests/regression/prefetch/common.h Normal file
View file

@ -0,0 +1,14 @@
#ifndef _COMMON_H_
#define _COMMON_H_
#define KERNEL_ARG_DEV_MEM_ADDR 0x7ffff000
typedef struct {
uint32_t num_tasks;
uint32_t task_size;
uint32_t src0_ptr;
uint32_t src1_ptr;
uint32_t dst_ptr;
} kernel_arg_t;
#endif

24
tests/regression/prefetch/kernel.c Normal file
View file

@ -0,0 +1,24 @@
#include <stdint.h>
#include <vx_intrinsics.h>
#include <vx_spawn.h>
#include "common.h"
void kernel_body(int task_id, const kernel_arg_t* arg) {
uint32_t count = arg->task_size;
int32_t* src0_ptr = (int32_t*)arg->src0_ptr;
int32_t* src1_ptr = (int32_t*)arg->src1_ptr;
int32_t* dst_ptr = (int32_t*)arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
vx_prefetch((uint32_t)(src0_ptr) + offset + i);
vx_prefetch((uint32_t)(src1_ptr) + offset + i);
dst_ptr[offset+i] = src0_ptr[offset+i] + src1_ptr[offset+i];
}
}
void main() {
const kernel_arg_t* arg = (const kernel_arg_t*)KERNEL_ARG_DEV_MEM_ADDR;
vx_spawn_tasks(arg->num_tasks, kernel_body, arg);
}

202
tests/regression/prefetch/main.cpp Normal file
View file

@ -0,0 +1,202 @@
#include <iostream>
#include <unistd.h>
#include <string.h>
#include <vortex.h>
#include "common.h"
#define RT_CHECK(_expr) \
do { \
int _ret = _expr; \
if (0 == _ret) \
break; \
printf("Error: '%s' returned %d!\n", #_expr, (int)_ret); \
cleanup(); \
exit(-1); \
} while (false)
///////////////////////////////////////////////////////////////////////////////
const char* kernel_file = "kernel.bin";
uint32_t count = 0;
vx_device_h device = nullptr;
vx_buffer_h staging_buf = nullptr;
static void show_usage() {
std::cout << "Vortex Test." << std::endl;
std::cout << "Usage: [-k: kernel] [-n words] [-h: help]" << std::endl;
}
static void parse_args(int argc, char **argv) {
int c;
while ((c = getopt(argc, argv, "n:k:h?")) != -1) {
switch (c) {
case 'n':
count = atoi(optarg);
break;
case 'k':
kernel_file = optarg;
break;
case 'h':
case '?': {
show_usage();
exit(0);
} break;
default:
show_usage();
exit(-1);
}
}
}
void cleanup() {
if (staging_buf) {
vx_buf_release(staging_buf);
}
if (device) {
vx_dev_close(device);
}
}
int run_test(const kernel_arg_t& kernel_arg,
uint32_t buf_size,
uint32_t num_points) {
// start device
std::cout << "start device" << std::endl;
RT_CHECK(vx_start(device));
// wait for completion
std::cout << "wait for completion" << std::endl;
RT_CHECK(vx_ready_wait(device, -1));
// download destination buffer
std::cout << "download destination buffer" << std::endl;
RT_CHECK(vx_copy_from_dev(staging_buf, kernel_arg.dst_ptr, buf_size, 0));
// verify result
std::cout << "verify result" << std::endl;
{
int errors = 0;
auto buf_ptr = (int32_t*)vx_host_ptr(staging_buf);
for (uint32_t i = 0; i < num_points; ++i) {
int ref = i + i;
int cur = buf_ptr[i];
if (cur != ref) {
std::cout << "error at result #" << std::dec << i
<< std::hex << ": actual 0x" << cur << ", expected 0x" << ref << std::endl;
++errors;
}
}
if (errors != 0) {
std::cout << "Found " << std::dec << errors << " errors!" << std::endl;
std::cout << "FAILED!" << std::endl;
return 1;
}
}
return 0;
}
int main(int argc, char *argv[]) {
size_t value;
kernel_arg_t kernel_arg;
// parse command arguments
parse_args(argc, argv);
if (count == 0) {
count = 1;
}
// open device connection
std::cout << "open device connection" << std::endl;
RT_CHECK(vx_dev_open(&device));
unsigned max_cores, max_warps, max_threads;
RT_CHECK(vx_dev_caps(device, VX_CAPS_MAX_CORES, &max_cores));
RT_CHECK(vx_dev_caps(device, VX_CAPS_MAX_WARPS, &max_warps));
RT_CHECK(vx_dev_caps(device, VX_CAPS_MAX_THREADS, &max_threads));
uint32_t num_tasks = max_cores * max_warps * max_threads;
uint32_t num_points = count * num_tasks;
uint32_t buf_size = num_points * sizeof(int32_t);
std::cout << "number of points: " << num_points << std::endl;
std::cout << "buffer size: " << buf_size << " bytes" << std::endl;
// upload program
std::cout << "upload program" << std::endl;
RT_CHECK(vx_upload_kernel_file(device, kernel_file));
// allocate device memory
std::cout << "allocate device memory" << std::endl;
RT_CHECK(vx_alloc_dev_mem(device, buf_size, &value));
kernel_arg.src0_ptr = value;
RT_CHECK(vx_alloc_dev_mem(device, buf_size, &value));
kernel_arg.src1_ptr = value;
RT_CHECK(vx_alloc_dev_mem(device, buf_size, &value));
kernel_arg.dst_ptr = value;
kernel_arg.num_tasks = num_tasks;
kernel_arg.task_size = count;
std::cout << "dev_src0=" << std::hex << kernel_arg.src0_ptr << std::endl;
std::cout << "dev_src1=" << std::hex << kernel_arg.src1_ptr << std::endl;
std::cout << "dev_dst=" << std::hex << kernel_arg.dst_ptr << std::endl;
// allocate shared memory
std::cout << "allocate shared memory" << std::endl;
uint32_t alloc_size = std::max<uint32_t>(buf_size, sizeof(kernel_arg_t));
RT_CHECK(vx_alloc_shared_mem(device, alloc_size, &staging_buf));
// upload kernel argument
std::cout << "upload kernel argument" << std::endl;
{
auto buf_ptr = (int*)vx_host_ptr(staging_buf);
memcpy(buf_ptr, &kernel_arg, sizeof(kernel_arg_t));
RT_CHECK(vx_copy_to_dev(staging_buf, KERNEL_ARG_DEV_MEM_ADDR, sizeof(kernel_arg_t), 0));
}
// upload source buffer0
{
auto buf_ptr = (int32_t*)vx_host_ptr(staging_buf);
for (uint32_t i = 0; i < num_points; ++i) {
buf_ptr[i] = i-1;
}
}
std::cout << "upload source buffer0" << std::endl;
RT_CHECK(vx_copy_to_dev(staging_buf, kernel_arg.src0_ptr, buf_size, 0));
// upload source buffer1
{
auto buf_ptr = (int32_t*)vx_host_ptr(staging_buf);
for (uint32_t i = 0; i < num_points; ++i) {
buf_ptr[i] = i+1;
}
}
std::cout << "upload source buffer1" << std::endl;
RT_CHECK(vx_copy_to_dev(staging_buf, kernel_arg.src1_ptr, buf_size, 0));
// clear destination buffer
{
auto buf_ptr = (int32_t*)vx_host_ptr(staging_buf);
for (uint32_t i = 0; i < num_points; ++i) {
buf_ptr[i] = 0xdeadbeef;
}
}
std::cout << "clear destination buffer" << std::endl;
RT_CHECK(vx_copy_to_dev(staging_buf, kernel_arg.dst_ptr, buf_size, 0));
// run tests
std::cout << "run tests" << std::endl;
RT_CHECK(run_test(kernel_arg, buf_size, num_points));
// cleanup
std::cout << "cleanup" << std::endl;
cleanup();
std::cout << "PASSED!" << std::endl;
return 0;
}