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

project tests refactoring

Browse Source
This commit is contained in:
Blaise Tine
2021-06-13 17:42:04 -07:00
parent 47c3234659
commit 03406c0a3f
631 changed files with 394471 additions and 653511 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
tests/regression/dogfood/Makefile Normal file
View File

@@ -0,0 +1,69 @@
RISCV_TOOLCHAIN_PATH ?= /opt/riscv-gnu-toolchain
VORTEX_RT_PATH ?= $(wildcard ../../../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_LDFLAGS += -lm
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../../include -I../../../hw
PROJECT = dogfood
SRCS = main.cpp
all: $(PROJECT) kernel.bin kernel.dump
kernel.dump: kernel.elf
$(VX_DP) -d -r -t 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) -L../../stub -lvortex -o $@
run-fpga: $(PROJECT)
LD_LIBRARY_PATH=../../opae:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
run-asesim: $(PROJECT)
ASE_LOG=0 LD_LIBRARY_PATH=../../opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
run-vlsim: $(PROJECT)
ASE_LOG=0 LD_LIBRARY_PATH=../../opae/vlsim:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
run-rtlsim: $(PROJECT)
LD_LIBRARY_PATH=../../rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT) $(OPTS)
run-simx: $(PROJECT)
LD_LIBRARY_PATH=../../simx:$(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

15
tests/regression/dogfood/common.h Normal file
View File

@@ -0,0 +1,15 @@
#ifndef _COMMON_H_
#define _COMMON_H_
#define KERNEL_ARG_DEV_MEM_ADDR 0x7ffff000
struct kernel_arg_t {
uint32_t testid;
uint32_t num_tasks;
uint32_t task_size;
uint32_t src0_ptr;
uint32_t src1_ptr;
uint32_t dst_ptr;
};
#endif

BIN
tests/regression/dogfood/kernel.bin Executable file
View File

Binary file not shown.

353
tests/regression/dogfood/kernel.c Normal file
View File

@@ -0,0 +1,353 @@
#include <stdint.h>
#include <math.h>
#include <vx_intrinsics.h>
#include <vx_spawn.h>
#include "common.h"
typedef void (*PFN_Kernel)(int task_id, void* arg);
inline float __ieee754_sqrtf (float x) {
asm ("fsqrt.s %0, %1" : "=f" (x) : "f" (x));
return x;
}
void kernel_iadd(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct 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) {
int32_t a = src0_ptr[offset+i];
int32_t b = src1_ptr[offset+i];
int32_t c = a + b;
dst_ptr[offset+i] = c;
}
}
void kernel_imul(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct 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) {
int32_t a = src0_ptr[offset+i];
int32_t b = src1_ptr[offset+i];
int32_t c = a * b;
dst_ptr[offset+i] = c;
}
}
void kernel_idiv(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct 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) {
int32_t a = src0_ptr[offset+i];
int32_t b = src1_ptr[offset+i];
int32_t c = a / b;
dst_ptr[offset+i] = c;
}
}
void kernel_idiv_mul(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct 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) {
int32_t a = src0_ptr[offset+i];
int32_t b = src1_ptr[offset+i];
int32_t c = a / b;
int32_t d = a * b;
int32_t e = c + d;
dst_ptr[offset+i] = e;
}
}
void kernel_fadd(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c = a + b;
dst_ptr[offset+i] = c;
}
}
void kernel_fsub(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c = a - b;
dst_ptr[offset+i] = c;
}
}
void kernel_fmul(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c = a * b;
dst_ptr[offset+i] = c;
}
}
void kernel_fmadd(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c = a * b + b;
dst_ptr[offset+i] = c;
}
}
void kernel_fmsub(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c = a * b - b;
dst_ptr[offset+i] = c;
}
}
void kernel_fnmadd(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c =-a * b - b;
dst_ptr[offset+i] = c;
}
}
void kernel_fnmsub(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c =-a * b + b;
dst_ptr[offset+i] = c;
}
}
void kernel_fnmadd_madd(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c =-a * b - b;
float d = a * b + b;
float e = c + d;
dst_ptr[offset+i] = e;
}
}
void kernel_fdiv(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c = a / b;
dst_ptr[offset+i] = c;
}
}
void kernel_fdiv2(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c = a / b;
float d = b / a;
float e = c + d;
dst_ptr[offset+i] = e;
}
}
void kernel_fsqrt(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c = __ieee754_sqrtf(a * b);
dst_ptr[offset+i] = c;
}
}
void kernel_ftoi(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_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) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c = a + b;
int32_t d = (int32_t)c;
dst_ptr[offset+i] = d;
}
}
void kernel_ftou(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct kernel_arg_t*)(arg);
uint32_t count = _arg->task_size;
float* src0_ptr = (float*)_arg->src0_ptr;
float* src1_ptr = (float*)_arg->src1_ptr;
uint32_t* dst_ptr = (uint32_t*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
float a = src0_ptr[offset+i];
float b = src1_ptr[offset+i];
float c = a + b;
uint32_t d = (uint32_t)c;
dst_ptr[offset+i] = d;
}
}
void kernel_itof(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct 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;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
int32_t a = src0_ptr[offset+i];
int32_t b = src1_ptr[offset+i];
int32_t c = a + b;
float d = (float)c;
dst_ptr[offset+i] = d;
}
}
void kernel_utof(int task_id, void* arg) {
struct kernel_arg_t* _arg = (struct 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;
float* dst_ptr = (float*)_arg->dst_ptr;
uint32_t offset = task_id * count;
for (uint32_t i = 0; i < count; ++i) {
int32_t a = src0_ptr[offset+i];
int32_t b = src1_ptr[offset+i];
uint32_t c = a + b;
float d = (float)c;
dst_ptr[offset+i] = d;
}
}
static const PFN_Kernel sc_tests[] = {
kernel_iadd,
kernel_imul,
kernel_idiv,
kernel_idiv_mul,
kernel_fadd,
kernel_fsub,
kernel_fmul,
kernel_fmadd,
kernel_fmsub,
kernel_fnmadd,
kernel_fnmsub,
kernel_fnmadd_madd,
kernel_fdiv,
kernel_fdiv2,
kernel_fsqrt,
kernel_ftoi,
kernel_ftou,
kernel_itof,
kernel_utof,
};
void main() {
struct kernel_arg_t* arg = (struct kernel_arg_t*)KERNEL_ARG_DEV_MEM_ADDR;
vx_spawn_tasks(arg->num_tasks, sc_tests[arg->testid], arg);
}

1149
tests/regression/dogfood/kernel.dump Normal file
View File

File diff suppressed because it is too large Load Diff

BIN
tests/regression/dogfood/kernel.elf Executable file
View File

Binary file not shown.

278
tests/regression/dogfood/main.cpp Normal file
View File

@@ -0,0 +1,278 @@
#include <iostream>
#include <vector>
#include <unistd.h>
#include <string.h>
#include <vortex.h>
#include <VX_config.h>
#include "testcases.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)
///////////////////////////////////////////////////////////////////////////////
class TestMngr {
public:
TestMngr() {
this->add_test("iadd", new Test_IADD());
this->add_test("imul", new Test_IMUL());
this->add_test("idiv", new Test_IDIV());
this->add_test("idiv-mul", new Test_IDIV_MUL());
#ifdef EXT_F_ENABLE
this->add_test("fadd", new Test_FADD());
this->add_test("fsub", new Test_FSUB());
this->add_test("fmul", new Test_FMUL());
this->add_test("fmadd", new Test_FMADD());
this->add_test("fmsub", new Test_FMSUB());
this->add_test("fnmadd", new Test_FNMADD());
this->add_test("fnmsub", new Test_FNMSUB());
this->add_test("fnmadd-madd", new Test_FNMADD_MADD());
this->add_test("fdiv", new Test_FDIV());
this->add_test("fdiv2", new Test_FDIV2());
this->add_test("fsqrt", new Test_FSQRT());
this->add_test("ftoi", new Test_FTOI());
this->add_test("ftou", new Test_FTOU());
this->add_test("itof", new Test_ITOF());
this->add_test("utof", new Test_UTOF());
#endif
}
~TestMngr() {
for (size_t i = 0; i < _tests.size(); ++i) {
delete _tests[i];
}
}
const std::string& get_name(int testid) const {
return _names.at(testid);
}
ITestCase* get_test(int testid) const {
return _tests.at(testid);
}
void add_test(const char* name, ITestCase* test) {
_names.push_back(name);
_tests.push_back(test);
}
size_t size() const {
return _tests.size();
}
private:
std::vector<std::string> _names;
std::vector<ITestCase*> _tests;
};
///////////////////////////////////////////////////////////////////////////////
TestMngr testMngr;
const char* kernel_file = "kernel.bin";
int count = 0;
int testid_s = 0;
int testid_e = (testMngr.size() - 1);
bool stop_on_error = true;
vx_device_h device = nullptr;
vx_buffer_h arg_buf = nullptr;
vx_buffer_h src1_buf = nullptr;
vx_buffer_h src2_buf = nullptr;
vx_buffer_h dst_buf = nullptr;
static void show_usage() {
std::cout << "Vortex Driver Test." << std::endl;
std::cout << "Usage: [-t:testid] [-s:testid] [-e:testid] [-k: kernel] [-n words] [-c] [-h: help]" << std::endl;
}
static void parse_args(int argc, char **argv) {
int c;
while ((c = getopt(argc, argv, "n:t:s:e:k:ch?")) != -1) {
switch (c) {
case 'n':
count = atoi(optarg);
break;
case 't':
testid_s = atoi(optarg);
testid_e = atoi(optarg);
break;
case 's':
testid_s = atoi(optarg);
break;
case 'e':
testid_e = atoi(optarg);
break;
case 'k':
kernel_file = optarg;
break;
case 'c':
stop_on_error = false;
break;
case 'h':
case '?': {
show_usage();
exit(0);
} break;
default:
show_usage();
exit(-1);
}
}
}
void cleanup() {
if (arg_buf) {
vx_buf_release(arg_buf);
}
if (src1_buf) {
vx_buf_release(src1_buf);
}
if (src2_buf) {
vx_buf_release(src2_buf);
}
if (dst_buf) {
vx_buf_release(dst_buf);
}
if (device) {
vx_dev_close(device);
}
}
int main(int argc, char *argv[]) {
int exitcode = 0;
size_t value;
kernel_arg_t kernel_arg;
// parse command arguments
parse_args(argc, argv);
if (count == 0) {
count = 1;
}
std::cout << std::dec;
std::cout << "test ids: " << testid_s << " - " << testid_e << std::endl;
std::cout << "workitem size: " << count << std::endl;
std::cout << "using kernel: " << kernel_file << std::endl;
// 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));
int num_tasks = max_cores * max_warps * max_threads;
int num_points = count * num_tasks;
size_t buf_size = num_points * sizeof(uint32_t);
std::cout << "number of points: " << num_points << std::endl;
std::cout << "buffer size: " << buf_size << " bytes" << std::endl;
// upload program
std::cout << "upload kernel" << 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::dec << std::endl;
std::cout << "dev_src1=" << std::hex << kernel_arg.src1_ptr << std::dec << std::endl;
std::cout << "dev_dst=" << std::hex << kernel_arg.dst_ptr << std::dec << std::endl;
// allocate shared memory
std::cout << "allocate shared memory" << std::endl;
RT_CHECK(vx_alloc_shared_mem(device, sizeof(kernel_arg_t), &arg_buf));
RT_CHECK(vx_alloc_shared_mem(device, buf_size, &src1_buf));
RT_CHECK(vx_alloc_shared_mem(device, buf_size, &src2_buf));
RT_CHECK(vx_alloc_shared_mem(device, buf_size, &dst_buf));
for (int t = testid_s; t <= testid_e; ++t) {
auto name = testMngr.get_name(t);
auto test = testMngr.get_test(t);
std::cout << "Test" << t << ": " << name << std::endl;
// upload kernel argument
std::cout << "upload kernel argument" << std::endl;
kernel_arg.testid = t;
memcpy((void*)vx_host_ptr(arg_buf), &kernel_arg, sizeof(kernel_arg_t));
RT_CHECK(vx_copy_to_dev(arg_buf, KERNEL_ARG_DEV_MEM_ADDR, sizeof(kernel_arg_t), 0));
// get test arguments
std::cout << "get test arguments" << std::endl;
test->setup(num_points, (void*)vx_host_ptr(src1_buf), (void*)vx_host_ptr(src2_buf));
// upload source buffer0
std::cout << "upload source buffer0" << std::endl;
RT_CHECK(vx_copy_to_dev(src1_buf, kernel_arg.src0_ptr, buf_size, 0));
// upload source buffer1
std::cout << "upload source buffer1" << std::endl;
RT_CHECK(vx_copy_to_dev(src2_buf, kernel_arg.src1_ptr, buf_size, 0));
// clear destination buffer
std::cout << "clear destination buffer" << std::endl;
for (int i = 0; i < num_points; ++i) {
((uint32_t*)vx_host_ptr(dst_buf))[i] = 0xdeadbeef;
}
RT_CHECK(vx_copy_to_dev(dst_buf, kernel_arg.dst_ptr, buf_size, 0));
// 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(dst_buf, kernel_arg.dst_ptr, buf_size, 0));
// verify destination
std::cout << "verify test result" << std::endl;
int errors = test->verify(num_points,
(void*)vx_host_ptr(dst_buf),
(void*)vx_host_ptr(src1_buf),
(void*)vx_host_ptr(src2_buf));
if (errors != 0) {
std::cout << "found " << std::dec << errors << " errors!" << std::endl;
std::cout << "Test" << t << "-" << name << " FAILED!" << std::endl << std::flush;
if (stop_on_error) {
cleanup();
exit(1);
}
exitcode = 1;
} else {
std::cout << "Test" << t << "-" << name << " PASSED!" << std::endl << std::flush;
}
}
// cleanup
std::cout << "cleanup" << std::endl;
cleanup();
return exitcode;
}

600
tests/regression/dogfood/testcases.h Normal file
View File

@@ -0,0 +1,600 @@
#pragma once
#include <iostream>
#include <math.h>
#include <limits>
union Float_t {
float f;
int i;
struct {
uint32_t man : 23;
uint32_t exp : 8;
uint32_t sign : 1;
} parts;
};
inline float fround(float x, int32_t precision = 8) {
auto power_of_10 = std::pow(10, precision);
return std::round(x * power_of_10) / power_of_10;
}
inline bool almost_equal_eps(float a, float b, int ulp = 128) {
auto eps = std::numeric_limits<float>::epsilon() * (std::max(fabs(a), fabs(b)) * ulp);
auto d = fabs(a - b);
if (d > eps) {
std::cout << "*** almost_equal_eps: d=" << d << ", eps=" << eps << std::endl;
return false;
}
return true;
}
inline bool almost_equal_ulp(float a, float b, int32_t ulp = 6) {
Float_t fa{a}, fb{b};
auto d = std::abs(fa.i - fb.i);
if (d > ulp) {
std::cout << "*** almost_equal_ulp: a=" << a << ", b=" << b << ", ulp=" << d << ", ia=" << std::hex << fa.i << ", ib=" << fb.i << std::endl;
return false;
}
return true;
}
inline bool almost_equal(float a, float b) {
if (a == b)
return true;
/*if (almost_equal_eps(a, b))
return true;*/
return almost_equal_ulp(a, b);
}
class ITestCase {
public:
ITestCase() {}
virtual ~ITestCase() {}
virtual void setup(int n, void* src1, void* src2) = 0;
virtual int verify(int n, void* dst, const void* src1, const void* src2) = 0;
};
class Test_IADD : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (int32_t*)src1;
auto b = (int32_t*)src2;
for (int i = 0; i < n; ++i) {
a[i] = n/2 - i;
b[i] = n/2 + i;
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (int32_t*)src1;
auto b = (int32_t*)src2;
auto c = (int32_t*)dst;
for (int i = 0; i < n; ++i) {
auto ref = a[i] + b[i];
if (c[i] != ref) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_IMUL : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (int32_t*)src1;
auto b = (int32_t*)src2;
for (int i = 0; i < n; ++i) {
a[i] = n/2 - i;
b[i] = n/2 + i;
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (int32_t*)src1;
auto b = (int32_t*)src2;
auto c = (int32_t*)dst;
for (int i = 0; i < n; ++i) {
auto ref = a[i] * b[i];
if (c[i] != ref) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_IDIV : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (int32_t*)src1;
auto b = (int32_t*)src2;
for (int i = 0; i < n; ++i) {
a[i] = n/2 - i;
b[i] = n/2 + i;
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (int32_t*)src1;
auto b = (int32_t*)src2;
auto c = (int32_t*)dst;
for (int i = 0; i < n; ++i) {
auto ref = a[i] / b[i];
if (c[i] != ref) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_IDIV_MUL : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (int32_t*)src1;
auto b = (int32_t*)src2;
for (int i = 0; i < n; ++i) {
a[i] = n/2 - i;
b[i] = n/2 + i;
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (int32_t*)src1;
auto b = (int32_t*)src2;
auto c = (int32_t*)dst;
for (int i = 0; i < n; ++i) {
auto x = a[i] / b[i];
auto y = a[i] * b[i];
auto ref = x + y;
if (c[i] != ref) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FADD : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround((n - i) * (1.0f/n));
b[i] = fround((n + i) * (1.0f/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto ref = a[i] + b[i];
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FSUB : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround((n - i) * (1.0f/n));
b[i] = fround((n + i) * (1.0f/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto ref = a[i] - b[i];
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FMUL : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround((n - i) * (1.0f/n));
b[i] = fround((n + i) * (1.0f/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto ref = a[i] * b[i];
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FMADD : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround((n - i) * (1.0f/n));
b[i] = fround((n + i) * (1.0f/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto ref = a[i] * b[i] + b[i];
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FMSUB : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround((n - i) * (1.0f/n));
b[i] = fround((n + i) * (1.0f/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto ref = a[i] * b[i] - b[i];
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FNMADD : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround((n - i) * (1.0f/n));
b[i] = fround((n + i) * (1.0f/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto ref = -a[i] * b[i] - b[i];
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FNMSUB : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround((n - i) * (1.0f/n));
b[i] = fround((n + i) * (1.0f/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto ref = -a[i] * b[i] + b[i];
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FNMADD_MADD : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround((n - i) * (1.0f/n));
b[i] = fround((n + i) * (1.0f/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto x = -a[i] * b[i] - b[i];
auto y = a[i] * b[i] + b[i];
auto ref = x + y;
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FDIV : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround((n - i) * (1.0f/n));
b[i] = fround((n + i) * (1.0f/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto ref = a[i] / b[i];
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FDIV2 : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround((n - i) * (1.0f/n));
b[i] = fround((n + i) * (1.0f/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto x = a[i] / b[i];
auto y = b[i] / a[i];
auto ref = x + y;
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FSQRT : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
float q = 1.0f + (i % 64);
a[i] = q;
b[i] = q;
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto ref = sqrt(a[i] * b[i]);
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FTOI : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround((n/2 - i) + (float(i)/n));
b[i] = fround((n/2 - i) + (float(i)/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (int32_t*)dst;
for (int i = 0; i < n; ++i) {
auto x = a[i] + b[i];
auto ref = (int32_t)x;
if (c[i] != ref) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_FTOU : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (float*)src1;
auto b = (float*)src2;
for (int i = 0; i < n; ++i) {
a[i] = fround(i + (float(i)/n));
b[i] = fround(i + (float(i)/n));
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (float*)src1;
auto b = (float*)src2;
auto c = (uint32_t*)dst;
for (int i = 0; i < n; ++i) {
auto x = a[i] + b[i];
auto ref = (uint32_t)x;
if (c[i] != ref) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_ITOF : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (int32_t*)src1;
auto b = (int32_t*)src2;
for (int i = 0; i < n; ++i) {
a[i] = n/2 - i;
b[i] = n/2 - i;
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (int32_t*)src1;
auto b = (int32_t*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto x = a[i] + b[i];
auto ref = (float)x;
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};
class Test_UTOF : public ITestCase {
public:
void setup(int n, void* src1, void* src2) override {
auto a = (uint32_t*)src1;
auto b = (uint32_t*)src2;
for (int i = 0; i < n; ++i) {
a[i] = i;
b[i] = i;
}
}
int verify(int n, void* dst, const void* src1, const void* src2) override {
int errors = 0;
auto a = (uint32_t*)src1;
auto b = (uint32_t*)src2;
auto c = (float*)dst;
for (int i = 0; i < n; ++i) {
auto x = a[i] + b[i];
auto ref = (float)x;
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected " << ref << ", actual " << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
}
}
return errors;
}
};