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

Vortex 2.0 changes:

Browse Source 
+ Microarchitecture optimizations
+ 64-bit support
+ Xilinx FPGA support
+ LLVM-16 support
+ Refactoring and quality control fixes
This commit is contained in:
Blaise Tine
2023-10-19 20:51:22 -07:00
parent d69a64c32c
commit d47cccc157
1300 changed files with 247321 additions and 311189 deletions
Download Patch File Download Diff File Expand all files Collapse all files

197
tests/regression/dogfood/main.cpp
View File

@@ -1,109 +1,49 @@
#include <iostream>
#include <vector>
#include <unordered_set>
#include <unistd.h>
#include <string.h>
#include <vector>
#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;
TestSuite* testSuite = nullptr;
const char* kernel_file = "kernel.bin";
int count = 0;
int count = 0;
std::unordered_set<int> included;
std::unordered_set<int> excluded;
int testid_s = 0;
int testid_e = (testMngr.size() - 1);
int testid_e = 0;
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;
kernel_arg_t kernel_arg;
vx_device_h device = nullptr;
std::vector<uint8_t> arg_buf;
std::vector<uint8_t> src1_buf;
std::vector<uint8_t> src2_buf;
std::vector<uint8_t> dst_buf;
kernel_arg_t kernel_arg = {};
static void show_usage() {
std::cout << "Vortex Test." << std::endl;
std::cout << "Usage: [-t:testid] [-s:testid] [-e:testid] [-k: kernel] [-n words] [-c] [-h: help]" << std::endl;
std::cout << "Usage: [-t<testid>: selected test] [-s<testid>: start test] [-e<testid>: end test] [-x<testid>: excluded tests]" << std::endl;
std::cout << " [-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) {
while ((c = getopt(argc, argv, "n:t:x:s:e:k:ch?")) != -1) {
switch (c) {
case 'n':
count = atoi(optarg);
break;
case 't':
testid_s = atoi(optarg);
testid_e = atoi(optarg);
included.insert(atoi(optarg));
break;
case 'x':
excluded.insert(atoi(optarg));
break;
case 's':
testid_s = atoi(optarg);
@@ -130,17 +70,8 @@ static void parse_args(int argc, char **argv) {
}
void cleanup() {
if (arg_buf) {
vx_buf_free(arg_buf);
}
if (src1_buf) {
vx_buf_free(src1_buf);
}
if (src2_buf) {
vx_buf_free(src2_buf);
}
if (dst_buf) {
vx_buf_free(dst_buf);
if (testSuite) {
delete testSuite;
}
if (device) {
vx_mem_free(device, kernel_arg.src0_addr);
@@ -152,7 +83,6 @@ void cleanup() {
int main(int argc, char *argv[]) {
int exitcode = 0;
size_t value;
// parse command arguments
parse_args(argc, argv);
@@ -171,12 +101,12 @@ int main(int argc, char *argv[]) {
std::cout << "open device connection" << std::endl;
RT_CHECK(vx_dev_open(&device));
uint64_t 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));
uint64_t num_cores, num_warps, num_threads;
RT_CHECK(vx_dev_caps(device, VX_CAPS_NUM_CORES, &num_cores));
RT_CHECK(vx_dev_caps(device, VX_CAPS_NUM_WARPS, &num_warps));
RT_CHECK(vx_dev_caps(device, VX_CAPS_NUM_THREADS, &num_threads));
int num_tasks = max_cores * max_warps * max_threads;
int num_tasks = num_cores * num_warps * num_threads;
int num_points = count * num_tasks;
size_t buf_size = num_points * sizeof(uint32_t);
@@ -188,59 +118,69 @@ int main(int argc, char *argv[]) {
RT_CHECK(vx_upload_kernel_file(device, kernel_file));
// allocate device memory
std::cout << "allocate device memory" << std::endl;
RT_CHECK(vx_mem_alloc(device, buf_size, &value));
kernel_arg.src0_addr = value;
RT_CHECK(vx_mem_alloc(device, buf_size, &value));
kernel_arg.src1_addr = value;
RT_CHECK(vx_mem_alloc(device, buf_size, &value));
kernel_arg.dst_addr = value;
std::cout << "allocate device memory" << std::endl;
RT_CHECK(vx_mem_alloc(device, buf_size, VX_MEM_TYPE_GLOBAL, &kernel_arg.src0_addr));
RT_CHECK(vx_mem_alloc(device, buf_size, VX_MEM_TYPE_GLOBAL, &kernel_arg.src1_addr));
RT_CHECK(vx_mem_alloc(device, buf_size, VX_MEM_TYPE_GLOBAL, &kernel_arg.dst_addr));
kernel_arg.num_tasks = num_tasks;
kernel_arg.task_size = count;
std::cout << "dev_src0=" << std::hex << kernel_arg.src0_addr << std::dec << std::endl;
std::cout << "dev_src1=" << std::hex << kernel_arg.src1_addr << std::dec << std::endl;
std::cout << "dev_dst=" << std::hex << kernel_arg.dst_addr << std::dec << std::endl;
std::cout << "dev_src0=0x" << std::hex << kernel_arg.src0_addr << std::dec << std::endl;
std::cout << "dev_src1=0x" << std::hex << kernel_arg.src1_addr << std::dec << std::endl;
std::cout << "dev_dst=0x" << std::hex << kernel_arg.dst_addr << std::dec << std::endl;
// allocate shared memory
std::cout << "allocate shared memory" << std::endl;
RT_CHECK(vx_buf_alloc(device, sizeof(kernel_arg_t), &arg_buf));
RT_CHECK(vx_buf_alloc(device, buf_size, &src1_buf));
RT_CHECK(vx_buf_alloc(device, buf_size, &src2_buf));
RT_CHECK(vx_buf_alloc(device, buf_size, &dst_buf));
// allocate staging buffer
std::cout << "allocate staging buffer" << std::endl;
arg_buf.resize(sizeof(kernel_arg_t));
src1_buf.resize(buf_size);
src2_buf.resize(buf_size);
dst_buf.resize(buf_size);
// allocate test suite
testSuite = new TestSuite(device);
if (testid_e == 0) {
testid_e = (testSuite->size() - 1);
}
// execute tests
for (int t = testid_s; t <= testid_e; ++t) {
auto name = testMngr.get_name(t);
auto test = testMngr.get_test(t);
if (!included.empty()) {
if (included.count(t) == 0)
continue;
}
if (!excluded.empty()) {
if (excluded.count(t) != 0)
continue;
}
auto test = testSuite->get_test(t);
auto name = test->name();
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));
memcpy(arg_buf.data(), &kernel_arg, sizeof(kernel_arg_t));
RT_CHECK(vx_copy_to_dev(device, KERNEL_ARG_DEV_MEM_ADDR, arg_buf.data(), sizeof(kernel_arg_t)));
// 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));
RT_CHECK(test->setup(num_points, (void*)src1_buf.data(), (void*)src2_buf.data()));
// upload source buffer0
std::cout << "upload source buffer0" << std::endl;
RT_CHECK(vx_copy_to_dev(src1_buf, kernel_arg.src0_addr, buf_size, 0));
RT_CHECK(vx_copy_to_dev(device, kernel_arg.src0_addr, src1_buf.data(), buf_size));
// upload source buffer1
std::cout << "upload source buffer1" << std::endl;
RT_CHECK(vx_copy_to_dev(src2_buf, kernel_arg.src1_addr, buf_size, 0));
std::cout << "upload source buffer1" << std::endl;
RT_CHECK(vx_copy_to_dev(device, kernel_arg.src1_addr, src2_buf.data(), buf_size));
// 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;
((uint32_t*)dst_buf.data())[i] = 0xdeadbeef;
}
RT_CHECK(vx_copy_to_dev(dst_buf, kernel_arg.dst_addr, buf_size, 0));
RT_CHECK(vx_copy_to_dev(device, kernel_arg.dst_addr, dst_buf.data(), buf_size));
// start device
std::cout << "start device" << std::endl;
@@ -248,18 +188,15 @@ int main(int argc, char *argv[]) {
// wait for completion
std::cout << "wait for completion" << std::endl;
RT_CHECK(vx_ready_wait(device, MAX_TIMEOUT));
RT_CHECK(vx_ready_wait(device, VX_MAX_TIMEOUT));
// download destination buffer
std::cout << "download destination buffer" << std::endl;
RT_CHECK(vx_copy_from_dev(dst_buf, kernel_arg.dst_addr, buf_size, 0));
RT_CHECK(vx_copy_from_dev(device, dst_buf.data(), kernel_arg.dst_addr, buf_size));
// 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));
int errors = test->verify(num_points, dst_buf.data(), src1_buf.data(), src2_buf.data());
if (errors != 0) {
std::cout << "found " << std::dec << errors << " errors!" << std::endl;
std::cout << "Test" << t << "-" << name << " FAILED!" << std::endl << std::flush;