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

driver basic test and demo test refactoring

Browse Source
This commit is contained in:
Blaise Tine
2020-06-19 09:12:07 -07:00
parent e2e1b63e14
commit 68d9fc9a75
55 changed files with 1006 additions and 1205 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
driver/opae/Makefile
View File

@@ -18,7 +18,10 @@ CXXFLAGS +=-fstack-protector
CXXFLAGS += -fPIC
# Enable scope analyzer
#CXXFLAGS += -DSCOPE
CXXFLAGS += -DSCOPE
# config parameters
CXXFLAGS += -DNUM_WARPS=2 -DNUM_THREADS=2
LDFLAGS += -luuid

55
driver/opae/scope.cpp
View File

@@ -25,31 +25,34 @@ struct scope_signal_t {
const char* name;
};
constexpr int ilog2(int n) {
return (n > 1) ? 1 + ilog2(n >> 1) : 0;
}
static constexpr int NW_BITS = ilog2(NUM_WARPS);
static const scope_signal_t scope_signals[] = {
{ 2, "icache_req_warp_num" },
{ NW_BITS, "icache_req_warp_num" },
{ 32, "icache_req_addr" },
{ 2, "icache_req_tag" },
{ NW_BITS, "icache_req_tag" },
{ 32, "icache_rsp_data" },
{ 2, "icache_rsp_tag" },
{ NW_BITS, "icache_rsp_tag" },
{ 2, "dcache_req_warp_num" },
{ NW_BITS, "dcache_req_warp_num" },
{ 32, "dcache_req_curr_PC" },
{ 32, "dcache_req_addr" },
{ 1, "dcache_req_rw" },
{ 4, "dcache_req_byteen" },
{ 32, "dcache_req_data" },
{ 2, "dcache_req_tag" },
{ NW_BITS, "dcache_req_tag" },
{ 32, "dcache_rsp_data" },
{ 2 , "dcache_rsp_tag" },
{ NW_BITS, "dcache_rsp_tag" },
{ 32, "dram_req_addr" },
{ 1, "dram_req_rw" },
{ 16, "dram_req_byteen" },
{ 32, "dram_req_data" },
{ 29, "dram_req_tag" },
{ 32, "dram_rsp_data" },
{ 29, "dram_rsp_tag" },
@@ -58,30 +61,32 @@ static const scope_signal_t scope_signals[] = {
{ 16, "snp_req_tag" },
{ 16, "snp_rsp_tag" },
{ 2, "decode_warp_num" },
{ NW_BITS, "decode_warp_num" },
{ 32, "decode_curr_PC" },
{ 1, "decode_is_jal" },
{ 5, "decode_rs1" },
{ 5, "decode_rs2" },
{ 1, "decode_is_jal" },
{ 5, "decode_rs1" },
{ 5, "decode_rs2" },
{ 2, "execute_warp_num" },
{ NW_BITS, "execute_warp_num" },
{ 5, "execute_rd" },
{ 32, "execute_a" },
{ 32, "execute_b" },
{ 2, "writeback_warp_num" },
{ NW_BITS, "writeback_warp_num" },
{ 2, "writeback_wb" },
{ 5, "writeback_rd" },
{ 32, "writeback_data" },
///////////////////////////////////////////////////////////////////////////
{ 1, "icache_req_valid" },
{ 1, "icache_req_ready" },
{ 1, "icache_rsp_valid" },
{ 1, "icache_rsp_ready" },
{ 4, "dcache_req_valid" },
{ NUM_THREADS, "dcache_req_valid" },
{ 1, "dcache_req_ready" },
{ 4, "dcache_rsp_valid" },
{ NUM_THREADS, "dcache_rsp_valid" },
{ 1, "dcache_rsp_ready" },
{ 1, "dram_req_valid" },
@@ -94,14 +99,19 @@ static const scope_signal_t scope_signals[] = {
{ 1, "snp_rsp_valid" },
{ 1, "snp_rsp_ready" },
{ 4, "decode_valid" },
{ 4, "execute_valid" },
{ 4, "writeback_valid" },
{ NUM_THREADS, "decode_valid" },
{ NUM_THREADS, "execute_valid" },
{ NUM_THREADS, "writeback_valid" },
{ 1, "schedule_delay" },
{ 1, "memory_delay" },
{ 1, "exec_delay" },
{ 1, "gpr_stage_delay" },
{ 1, "busy" },
{ 1, "idram_req_valid" },
{ 1, "idram_req_ready" },
{ 1, "idram_rsp_valid" },
{ 1, "idram_rsp_ready" },
};
static const int num_signals = sizeof(scope_signals) / sizeof(scope_signal_t);
@@ -161,7 +171,10 @@ int vx_scope_stop(fpga_handle hfpga, uint64_t delay) {
CHECK_RES(fpgaWriteMMIO64(hfpga, 0, MMIO_CSR_SCOPE_CMD, 1));
assert(fwidth == (int)frame_width);
if (fwidth != (int)frame_width) {
std::cerr << "invalid frame_width: expecting " << std::dec << fwidth << "!" << std::endl;
std::abort();
}
std::vector<char> signal_data(frame_width+1);
uint64_t frame_offset = 0;

23
driver/rtlsim/Makefile
View File

@@ -13,18 +13,19 @@ DBG_PRINT_FLAGS = -DDBG_PRINT_CORE_ICACHE \
-DDBG_PRINT_WB \
-DDBG_PRINT_OPAE
#DBG_PRINT=$(DBG_PRINT_FLAGS)
DBG_PRINT=$(DBG_PRINT_FLAGS)
#MULTICORE += -DNUM_CLUSTERS=2 -DNUM_CORES=4
#MULTICORE += -DNUM_CLUSTERS=1 -DNUM_CORES=4
#MULTICORE += -DNUM_CLUSTERS=1 -DNUM_CORES=2
#CONFIGS += -DNUM_CLUSTERS=2 -DNUM_CORES=4
#CONFIGS += -DNUM_CLUSTERS=1 -DNUM_CORES=4
#CONFIGS += -DNUM_CLUSTERS=1 -DNUM_CORES=2
CONFIGS += -DNUM_WARPS=2 -DNUM_THREADS=2
#DEBUG=1
#AFU=1
DEBUG=1
AFU=1
CFLAGS += -fPIC
CFLAGS += -DUSE_RTLSIM $(MULTICORE)
CFLAGS += -DUSE_RTLSIM $(CONFIGS)
LDFLAGS += -shared -pthread
# LDFLAGS += -dynamiclib -pthread
@@ -35,7 +36,7 @@ SRCS = vortex.cpp ../common/vx_utils.cpp ../../hw/simulate/simulator.cpp
RTL_INCLUDE = -I../../hw/rtl -I../../hw/rtl/libs -I../../hw/rtl/interfaces -I../../hw/rtl/pipe_regs -I../../hw/rtl/cache
VL_FLAGS += --language 1800-2009 --assert -Wall -Wpedantic $(MULTICORE)
VL_FLAGS += --language 1800-2009 --assert -Wall -Wpedantic $(CONFIGS)
VL_FLAGS += -Wno-DECLFILENAME
VL_FLAGS += --x-initial unique
VL_FLAGS += --x-assign unique
@@ -47,9 +48,11 @@ VL_FLAGS += --x-assign unique
# Debugigng
ifdef DEBUG
VL_FLAGS += --trace -DVCD_OUTPUT $(DBG_PRINT)
CFLAGS += -DVCD_OUTPUT $(DBG_PRINT)
CFLAGS += -DVCD_OUTPUT $(DBG_PRINT)
#VL_FLAGS += -DDBG_CORE_REQ_INFO
#CFLAGS += -DDBG_CORE_REQ_INFO
else
CFLAGS += -DNDEBUG
CFLAGS += -DNDEBUG
VL_FLAGS += -DNDEBUG
endif

8
driver/tests/basic/Makefile
View File

@@ -44,16 +44,16 @@ $(PROJECT): $(SRCS)
$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -L../../stub -lvortex -o $@
run-fpga: $(PROJECT)
LD_LIBRARY_PATH=../../opae:$(LD_LIBRARY_PATH) ./$(PROJECT)
LD_LIBRARY_PATH=../../opae:$(LD_LIBRARY_PATH) ./$(PROJECT) -n 32
run-ase: $(PROJECT)
ASE_LOG=0 LD_LIBRARY_PATH=../../opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT)
ASE_LOG=0 LD_LIBRARY_PATH=../../opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT) -n 16
run-rtlsim: $(PROJECT)
LD_LIBRARY_PATH=../../rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
LD_LIBRARY_PATH=../../rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT) -n 16
run-simx: $(PROJECT)
LD_LIBRARY_PATH=../../simx:$(LD_LIBRARY_PATH) ./$(PROJECT)
LD_LIBRARY_PATH=../../simx:$(LD_LIBRARY_PATH) ./$(PROJECT) -n 16
.depend: $(SRCS)
$(CXX) $(CXXFLAGS) -MM $^ > .depend;

190
driver/tests/basic/basic.cpp
View File

@@ -1,10 +1,9 @@
#include <iostream>
#include <unistd.h>
#include <string.h>
#include <vortex.h>
#include "common.h"
int test = -1;
#define RT_CHECK(_expr) \
do { \
int _ret = _expr; \
@@ -15,79 +14,84 @@ int test = -1;
exit(-1); \
} while (false)
const char* kernel_file = "kernel.bin";
int test = -1;
uint32_t count = 0;
vx_device_h device = nullptr;
vx_buffer_h buffer = nullptr;
static void show_usage() {
std::cout << "Vortex Driver Test." << std::endl;
std::cout << "Usage: [-t testno][-k: kernel][-n words][-h: help]" << std::endl;
}
static void parse_args(int argc, char **argv) {
int c;
while ((c = getopt(argc, argv, "t:h?")) != -1) {
while ((c = getopt(argc, argv, "n:t:k:h?")) != -1) {
switch (c) {
case 't': {
case 'n':
count = atoi(optarg);
break;
case 't':
test = atoi(optarg);
} break;
break;
case 'k':
kernel_file = optarg;
break;
case 'h':
case '?': {
std::cout << "Test." << std::endl;
std::cout << "Usage: [-t testno][-h: help]" << std::endl;
show_usage();
exit(0);
} break;
default:
show_usage();
exit(-1);
}
}
}
void cleanup() {
if (buffer) {
vx_buf_release(buffer);
}
if (device) {
vx_dev_close(device);
}
}
uint64_t shuffle(int i, uint64_t value) {
return (value << i) | (value & ((1 << i)-1));;
}
vx_device_h device = nullptr;
vx_buffer_h sbuf = nullptr;
vx_buffer_h dbuf = nullptr;
int total_blocks = NUM_BLOCKS;
void cleanup() {
if (sbuf) {
vx_buf_release(sbuf);
}
if (dbuf) {
vx_buf_release(dbuf);
}
if (device) {
vx_dev_close(device);
}
}
int run_memcopy_test(vx_buffer_h sbuf,
vx_buffer_h dbuf,
uint32_t address,
uint64_t value,
int num_blocks) {
int run_memcopy_test(uint32_t dev_addr, uint64_t value, int num_blocks) {
int errors = 0;
// write sbuf data
// update source buffer
for (int i = 0; i < (64 * num_blocks) / 8; ++i) {
((uint64_t*)vx_host_ptr(sbuf))[i] = shuffle(i, value);
}
// clear dbuf data
for (int i = 0; i < (64 * num_blocks) / 8; ++i) {
((uint64_t*)vx_host_ptr(dbuf))[i] = 0;
((uint64_t*)vx_host_ptr(buffer))[i] = shuffle(i, value);
}
// write buffer to local memory
std::cout << "write buffer to local memory" << std::endl;
RT_CHECK(vx_copy_to_dev(sbuf, address, 64 * num_blocks, 0));
RT_CHECK(vx_copy_to_dev(buffer, dev_addr, 64 * num_blocks, 0));
// clear destination buffer
for (int i = 0; i < (64 * num_blocks) / 8; ++i) {
((uint64_t*)vx_host_ptr(buffer))[i] = 0;
}
// read buffer from local memory
std::cout << "read buffer from local memory" << std::endl;
RT_CHECK(vx_copy_from_dev(dbuf, address, 64 * num_blocks, 0));
RT_CHECK(vx_copy_from_dev(buffer, dev_addr, 64 * num_blocks, 0));
// verify result
std::cout << "verify result" << std::endl;
for (int i = 0; i < (64 * num_blocks) / 8; ++i) {
auto curr = ((uint64_t*)vx_host_ptr(dbuf))[i];
auto curr = ((uint64_t*)vx_host_ptr(buffer))[i];
auto ref = shuffle(i, value);
if (curr != ref) {
std::cout << "error at 0x" << std::hex << (address + 8 * i)
std::cout << "error at 0x" << std::hex << (dev_addr + 8 * i)
<< ": actual 0x" << curr << ", expected 0x" << ref << std::endl;
++errors;
}
@@ -102,35 +106,19 @@ int run_memcopy_test(vx_buffer_h sbuf,
return 0;
}
int run_kernel_test(vx_device_h device,
vx_buffer_h sbuf,
vx_buffer_h dbuf,
const char* program,
int num_blocks) {
int errors = 0;
uint64_t seed = 0x0badf00d40ff40ff;
int run_kernel_test(const kernel_arg_t& kernel_arg,
uint32_t buf_size,
uint32_t num_points) {
int errors = 0;
int src_dev_addr = DEV_MEM_SRC_ADDR;
int dest_dev_addr = DEV_MEM_DST_ADDR;
// write sbuf data
for (int i = 0; i < (64 * num_blocks) / 8; ++i) {
((uint64_t*)vx_host_ptr(sbuf))[i] = shuffle(i, seed);
}
// clear dbuf data
for (int i = 0; i < (64 * num_blocks) / 8; ++i) {
((uint64_t*)vx_host_ptr(dbuf))[i] = 0;
// update source buffer
for (uint32_t i = 0; i < num_points; ++i) {
((int32_t*)vx_host_ptr(buffer))[i] = i;
}
// write buffer to local memory
std::cout << "write buffer to local memory" << std::endl;
RT_CHECK(vx_copy_to_dev(sbuf, src_dev_addr, 64 * num_blocks, 0));
// upload program
std::cout << "upload program" << std::endl;
RT_CHECK(vx_upload_kernel_file(device, program));
RT_CHECK(vx_copy_to_dev(buffer, kernel_arg.src_ptr, buf_size, 0));
// start device
std::cout << "start device" << std::endl;
@@ -142,19 +130,24 @@ int run_kernel_test(vx_device_h device,
// flush the caches
std::cout << "flush the caches" << std::endl;
RT_CHECK(vx_flush_caches(device, dest_dev_addr, 64 * num_blocks));
RT_CHECK(vx_flush_caches(device, kernel_arg.dst_ptr, buf_size));
// clear destination buffer
for (uint32_t i = 0; i < num_points; ++i) {
((int32_t*)vx_host_ptr(buffer))[i] = 0;
}
// read buffer from local memory
std::cout << "read buffer from local memory" << std::endl;
RT_CHECK(vx_copy_from_dev(dbuf, dest_dev_addr, 64 * num_blocks, 0));
RT_CHECK(vx_copy_from_dev(buffer, kernel_arg.dst_ptr, buf_size, 0));
// verify result
std::cout << "verify result" << std::endl;
for (int i = 0; i < (64 * num_blocks) / 8; ++i) {
auto curr = ((uint64_t*)vx_host_ptr(dbuf))[i];
auto ref = shuffle(i, seed);
for (uint32_t i = 0; i < num_points; ++i) {
int32_t curr = ((int32_t*)vx_host_ptr(buffer))[i];
int32_t ref = i;
if (curr != ref) {
std::cout << "error at 0x" << std::hex << (dest_dev_addr + 8 * i)
std::cout << "error at value " << i
<< ": actual 0x" << curr << ", expected 0x" << ref << std::endl;
++errors;
}
@@ -170,33 +163,66 @@ int run_kernel_test(vx_device_h device,
}
int main(int argc, char *argv[]) {
size_t value;
kernel_arg_t kernel_arg;
// parse command arguments
parse_args(argc, argv);
std::cout << "total blocks: " << total_blocks << std::endl;
if (count == 0) {
count = 1;
}
uint32_t max_cores = vx_dev_caps(VX_CAPS_MAX_CORES);
uint32_t num_points = max_cores * count;
uint32_t num_blocks = (num_points * sizeof(uint32_t) + 63) / 64;
uint32_t buf_size = num_blocks * 64;
std::cout << "number of points: " << num_points << std::endl;
std::cout << "buffer size: " << buf_size << " bytes" << std::endl;
// open device connection
std::cout << "open device connection" << std::endl;
RT_CHECK(vx_dev_open(&device));
// create source buffer
std::cout << "create source buffer" << std::endl;
RT_CHECK(vx_alloc_shared_mem(device, total_blocks * 64, &sbuf));
// create destination buffer
std::cout << "create destination buffer" << std::endl;
RT_CHECK(vx_alloc_shared_mem(device, total_blocks * 64, &dbuf));
// allocate device memory
RT_CHECK(vx_alloc_dev_mem(device, buf_size, &value));
kernel_arg.src_ptr = value;
RT_CHECK(vx_alloc_dev_mem(device, buf_size, &value));
kernel_arg.dst_ptr = value;
kernel_arg.count = count;
std::cout << "dev_src=" << std::hex << kernel_arg.src_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, &buffer));
// run tests
if (0 == test || -1 == test) {
std::cout << "run memcopy test" << std::endl;
RT_CHECK(run_memcopy_test(sbuf, dbuf, DEV_MEM_SRC_ADDR, 0x0badf00d00ff00ff, 1));
RT_CHECK(run_memcopy_test(sbuf, dbuf, DEV_MEM_SRC_ADDR, 0x0badf00d40ff40ff, total_blocks));
RT_CHECK(run_memcopy_test(kernel_arg.src_ptr, 0x0badf00d00ff00ff, 1));
RT_CHECK(run_memcopy_test(kernel_arg.src_ptr, 0x0badf00d40ff40ff, num_blocks));
}
if (1 == test || -1 == test) {
// upload program
std::cout << "upload program" << std::endl;
RT_CHECK(vx_upload_kernel_file(device, kernel_file));
// upload kernel argument
std::cout << "upload kernel argument" << std::endl;
{
auto buf_ptr = (void*)vx_host_ptr(buffer);
memcpy(buf_ptr, &kernel_arg, sizeof(kernel_arg_t));
RT_CHECK(vx_copy_to_dev(buffer, KERNEL_ARG_DEV_MEM_ADDR, sizeof(kernel_arg_t), 0));
}
std::cout << "run kernel test" << std::endl;
RT_CHECK(run_kernel_test(device, sbuf, dbuf, "kernel.bin", total_blocks));
RT_CHECK(run_kernel_test(kernel_arg, buf_size, num_points));
}
// cleanup

10
driver/tests/basic/common.h
View File

@@ -1,8 +1,12 @@
#ifndef _COMMON_H_
#define _COMMON_H_
#define DEV_MEM_SRC_ADDR 0x10000040
#define DEV_MEM_DST_ADDR 0x20000080
#define NUM_BLOCKS 16
#define KERNEL_ARG_DEV_MEM_ADDR 0x7ffff000
struct kernel_arg_t {
uint32_t count;
uint32_t src_ptr;
uint32_t dst_ptr;
};
#endif

BIN
driver/tests/basic/kernel.bin
View File

Binary file not shown.

19
driver/tests/basic/kernel.c
View File

@@ -4,17 +4,14 @@
#include "common.h"
void main() {
int64_t* x = (int64_t*)DEV_MEM_SRC_ADDR;
int64_t* y = (int64_t*)DEV_MEM_DST_ADDR;
int num_words = (NUM_BLOCKS * 64) / 8;
int core_id = vx_core_id();
int num_cores = vx_num_cores();
int num_words_per_core = num_words / num_cores;
int offset = core_id * num_words_per_core;
struct kernel_arg_t* arg = (struct kernel_arg_t*)KERNEL_ARG_DEV_MEM_ADDR;
uint32_t count = arg->count;
int32_t* src_ptr = (int32_t*)arg->src_ptr;
int32_t* dst_ptr = (int32_t*)arg->dst_ptr;
for (int i = 0; i < num_words_per_core; ++i) {
y[offset + i] = x[offset + i];
uint32_t offset = vx_core_id() * count;
for (uint32_t i = 0; i < count; ++i) {
dst_ptr[offset + i] = src_ptr[offset + i];
}
}

8
driver/tests/demo/Makefile
View File

@@ -41,16 +41,16 @@ $(PROJECT): $(SRCS)
$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -L../../stub -lvortex -o $@
run-fpga: $(PROJECT)
LD_LIBRARY_PATH=../../opae:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin -n 16
LD_LIBRARY_PATH=../../opae:$(LD_LIBRARY_PATH) ./$(PROJECT) -n 16
run-ase: $(PROJECT)
ASE_LOG=0 LD_LIBRARY_PATH=../../opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin -n 16
ASE_LOG=0 LD_LIBRARY_PATH=../../opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT) -n 16
run-rtlsim: $(PROJECT)
LD_LIBRARY_PATH=../../rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin -n 16
LD_LIBRARY_PATH=../../rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT) -n 16
run-simx: $(PROJECT)
LD_LIBRARY_PATH=../../simx:$(LD_LIBRARY_PATH) ./$(PROJECT) -f kernel.bin -n 16
LD_LIBRARY_PATH=../../simx:$(LD_LIBRARY_PATH) ./$(PROJECT) -n 16
.depend: $(SRCS)
$(CXX) $(CXXFLAGS) -MM $^ > .depend;

2
driver/tests/demo/common.h
View File

@@ -4,7 +4,7 @@
#define KERNEL_ARG_DEV_MEM_ADDR 0x7ffff000
struct kernel_arg_t {
uint32_t stride;
uint32_t count;
uint32_t src0_ptr;
uint32_t src1_ptr;
uint32_t dst_ptr;

120
driver/tests/demo/demo.cpp
View File

@@ -14,23 +14,26 @@
exit(-1); \
} while (false)
const char* program_file = "kernel.bin";
uint32_t data_stride = 0;
const char* kernel_file = "kernel.bin";
uint32_t count = 0;
vx_device_h device = nullptr;
vx_buffer_h buffer = nullptr;
static void show_usage() {
std::cout << "Vortex Driver Test." << std::endl;
std::cout << "Usage: [-f: program] [-n stride] [-h: help]" << 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:f:h?")) != -1) {
while ((c = getopt(argc, argv, "n:k:h?")) != -1) {
switch (c) {
case 'n':
data_stride = atoi(optarg);
count = atoi(optarg);
break;
case 'f':
program_file = optarg;
case 'k':
kernel_file = optarg;
break;
case 'h':
case '?': {
@@ -42,16 +45,8 @@ static void parse_args(int argc, char **argv) {
exit(-1);
}
}
if (nullptr == program_file) {
show_usage();
exit(-1);
}
}
vx_device_h device = nullptr;
vx_buffer_h buffer = nullptr;
void cleanup() {
if (buffer) {
vx_buf_release(buffer);
@@ -61,9 +56,7 @@ void cleanup() {
}
}
int run_test(vx_device_h device,
vx_buffer_h buffer,
const kernel_arg_t& kernel_arg,
int run_test(const kernel_arg_t& kernel_arg,
uint32_t buf_size,
uint32_t num_points) {
// start device
@@ -86,13 +79,13 @@ int run_test(vx_device_h device,
std::cout << "verify result" << std::endl;
{
int errors = 0;
auto buf_ptr = (int*)vx_host_ptr(buffer);
auto buf_ptr = (int32_t*)vx_host_ptr(buffer);
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 value " << i
<< ": actual 0x" << cur << ", expected 0x" << ref << std::endl;
<< ": actual 0x" << cur << ", expected 0x" << ref << std::endl;
++errors;
}
}
@@ -113,21 +106,18 @@ int main(int argc, char *argv[]) {
// parse command arguments
parse_args(argc, argv);
if (count == 0) {
count = 1;
}
uint32_t max_cores = vx_dev_caps(VX_CAPS_MAX_CORES);
uint32_t max_warps = vx_dev_caps(VX_CAPS_MAX_WARPS);
uint32_t max_threads = vx_dev_caps(VX_CAPS_MAX_THREADS);
if (data_stride == 0) {
data_stride = 1;
}
uint32_t num_points = count * max_cores * max_warps * max_threads;
uint32_t buf_size = num_points * sizeof(uint32_t);
kernel_arg.stride = data_stride;
uint32_t num_points = max_cores * max_warps * max_threads;
uint32_t buf_size = num_points * data_stride * sizeof(uint32_t);
std::cout << "number of workitems: " << num_points << std::endl;
std::cout << "workitem size: " << data_stride * sizeof(uint32_t) << " bytes" << std::endl;
std::cout << "number of points: " << num_points << std::endl;
std::cout << "buffer size: " << buf_size << " bytes" << std::endl;
// open device connection
@@ -136,55 +126,29 @@ int main(int argc, char *argv[]) {
// upload program
std::cout << "upload program" << std::endl;
RT_CHECK(vx_upload_kernel_file(device, program_file));
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.count = 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, &buffer));
// populate source buffer0 values
std::cout << "populate source buffer0 values" << std::endl;
{
auto buf_ptr = (int*)vx_host_ptr(buffer);
for (uint32_t i = 0; i < num_points; ++i) {
buf_ptr[i] = i-1;
}
}
// upload source buffer0
std::cout << "upload source buffer0" << std::endl;
RT_CHECK(vx_copy_to_dev(buffer, kernel_arg.src0_ptr, buf_size, 0));
// populate source buffer1 values
std::cout << "populate source buffer1 values" << std::endl;
{
auto buf_ptr = (int*)vx_host_ptr(buffer);
for (uint32_t i = 0; i < num_points; ++i) {
buf_ptr[i] = i+1;
}
}
// upload source buffer1
std::cout << "upload source buffer1" << std::endl;
RT_CHECK(vx_copy_to_dev(buffer, kernel_arg.src1_ptr, buf_size, 0));
// upload kernel argument
std::cout << "upload kernel argument" << std::endl;
{
@@ -193,9 +157,41 @@ int main(int argc, char *argv[]) {
RT_CHECK(vx_copy_to_dev(buffer, KERNEL_ARG_DEV_MEM_ADDR, sizeof(kernel_arg_t), 0));
}
// upload source buffer0
{
auto buf_ptr = (int32_t*)vx_host_ptr(buffer);
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(buffer, kernel_arg.src0_ptr, buf_size, 0));
// upload source buffer1
{
auto buf_ptr = (int32_t*)vx_host_ptr(buffer);
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(buffer, kernel_arg.src1_ptr, buf_size, 0));
// clear destination buffer
{
auto buf_ptr = (int32_t*)vx_host_ptr(buffer);
for (uint32_t i = 0; i < num_points; ++i) {
buf_ptr[i] = 0;
}
}
std::cout << "clear destination buffer" << std::endl;
RT_CHECK(vx_copy_to_dev(buffer, kernel_arg.dst_ptr, buf_size, 0));
// run tests
std::cout << "run tests" << std::endl;
RT_CHECK(run_test(device, buffer, kernel_arg, buf_size, num_points));
RT_CHECK(run_test(kernel_arg, buf_size, num_points));
// cleanup
std::cout << "cleanup" << std::endl;

BIN
driver/tests/demo/kernel.bin Executable file → Normal file
View File

Binary file not shown.

13
driver/tests/demo/kernel.c
View File

@@ -6,13 +6,14 @@
void kernel_body(void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
int* src0_ptr = (int*)_arg->src0_ptr;
int* src1_ptr = (int*)_arg->src1_ptr;
int* dst_ptr = (int*)_arg->dst_ptr;
uint32_t count = _arg->count;
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 = vx_thread_gid() * count;
unsigned offset = vx_thread_gid() * _arg->stride;
for (unsigned i = 0; i < _arg->stride; ++i) {
for (uint32_t i = 0; i < count; ++i) {
dst_ptr[offset+i] = src0_ptr[offset+i] + src1_ptr[offset+i];
}
}