Danylo Lytovchenko
390 rivejä
14 KiB
C++
390 rivejä
14 KiB
C++
/* Copyright (c) 2010 - 2021 Advanced Micro Devices, Inc.
|
|
|
|
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
of this software and associated documentation files (the "Software"), to deal
|
|
in the Software without restriction, including without limitation the rights
|
|
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
copies of the Software, and to permit persons to whom the Software is
|
|
furnished to do so, subject to the following conditions:
|
|
|
|
The above copyright notice and this permission notice shall be included in
|
|
all copies or substantial portions of the Software.
|
|
|
|
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
|
THE SOFTWARE. */
|
|
|
|
#include "OCLPerfMemLatency.h"
|
|
|
|
#include <assert.h>
|
|
#include <stdio.h>
|
|
#include <string.h>
|
|
|
|
#include "CL/cl.h"
|
|
#include "Timer.h"
|
|
|
|
static const unsigned int NUM_SIZES = 16;
|
|
// 2k up to 64MB
|
|
static const unsigned int Sizes[NUM_SIZES] = {
|
|
2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
|
|
524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, 67108864};
|
|
// Quiet pesky warnings
|
|
#ifdef WIN_OS
|
|
#define SNPRINTF sprintf_s
|
|
#else
|
|
#define SNPRINTF snprintf
|
|
#endif
|
|
void OCLPerfMemLatency::genShader() {
|
|
shader_.clear();
|
|
|
|
// DO NOT PUBLISH
|
|
// Adopted from SiSoft Sandra 2013's memory latency test
|
|
shader_ +=
|
|
"#ifdef MAKEVOLATILE\n"
|
|
"#define VOLATILE volatile\n"
|
|
"#else\n"
|
|
"#define VOLATILE\n"
|
|
"#endif\n"
|
|
"__kernel\n"
|
|
//"__attribute__((work_group_size_hint(1, 1, 1)))\n"
|
|
"void MemWalker(\n"
|
|
" global VOLATILE uint * restrict input,\n"
|
|
" __global uint * restrict output,\n"
|
|
" const uint uCount, const uint uSize,\n"
|
|
" const uint uOffset, const int bMem, const uint repeats)\n"
|
|
"{\n"
|
|
" uint o = uOffset;\n"
|
|
" uint lid = 0;//get_local_id(0)*o;\n"
|
|
" uint x = lid;\n"
|
|
"\n"
|
|
" for (uint loop = 0; loop < repeats; loop++) {\n"
|
|
" uint i = uCount;\n"
|
|
" while (i--) {\n"
|
|
" x = input[x] /* + o*/;\n"
|
|
" }\n"
|
|
" }\n"
|
|
"\n"
|
|
"#ifdef MAKERW\n"
|
|
" input[0] = x;\n"
|
|
"#endif\n"
|
|
" output[0] = x;\n"
|
|
"}\n";
|
|
|
|
// printf("shader:\n%s\n", shader_.c_str());
|
|
shader_ += "\n\n";
|
|
shader_ +=
|
|
"__kernel\n"
|
|
//"__attribute__((work_group_size_hint(1, 1, 1)))\n"
|
|
"void Overhead(\n"
|
|
" __global uint * restrict input,\n"
|
|
" __global uint * restrict output,\n"
|
|
" const uint uCount, const uint uSize,\n"
|
|
" const uint uOffset, const int bMem, const uint repeats)\n"
|
|
"{\n"
|
|
"#ifdef USE_FLOAT\n"
|
|
" float x = (float)input[0];\n"
|
|
" for (uint loop = 0; loop < repeats; loop++) {\n"
|
|
" uint i = uCount;\n"
|
|
" x = (float)uOffset*x;\n"
|
|
" while (i--) {\n"
|
|
" x += (float)i;\n"
|
|
" }\n"
|
|
" }\n"
|
|
" output[0] = (uint)x;\n"
|
|
"#else\n"
|
|
" uint x = input[0];\n"
|
|
" for (uint loop = 0; loop < repeats; loop++) {\n"
|
|
" uint i = uCount;\n"
|
|
" x = x*uOffset;\n"
|
|
" while (i--) {\n"
|
|
" x += i;\n"
|
|
" }\n"
|
|
" }\n"
|
|
" output[0] = x;\n"
|
|
"#endif\n"
|
|
"}\n";
|
|
}
|
|
|
|
static void CL_CALLBACK notify_callback(const char* errinfo, const void* private_info, size_t cb,
|
|
void* user_data) {}
|
|
|
|
OCLPerfMemLatency::OCLPerfMemLatency() {
|
|
_numSubTests = NUM_SIZES * 6;
|
|
maxSize_ = Sizes[NUM_SIZES - 1];
|
|
}
|
|
|
|
OCLPerfMemLatency::~OCLPerfMemLatency() {}
|
|
|
|
void OCLPerfMemLatency::setData(cl_mem buffer, unsigned int val) {
|
|
void* ptr = _wrapper->clEnqueueMapBuffer(cmd_queue_, buffer, true, CL_MAP_WRITE, 0, width_, 0,
|
|
NULL, NULL, &error_);
|
|
unsigned int* data = (unsigned int*)ptr;
|
|
for (unsigned int i = 0; i < bufSizeDW_; i++) {
|
|
data[(i * (1024 + 17)) % bufSizeDW_] = ((i + 1) * (1024 + 17)) % bufSizeDW_;
|
|
}
|
|
error_ = _wrapper->clEnqueueUnmapMemObject(cmd_queue_, buffer, ptr, 0, NULL, NULL);
|
|
clFinish(cmd_queue_);
|
|
}
|
|
|
|
void OCLPerfMemLatency::checkData(cl_mem buffer) {
|
|
void* ptr = _wrapper->clEnqueueMapBuffer(cmd_queue_, buffer, true, CL_MAP_READ, 0,
|
|
sizeof(cl_uint), 0, NULL, NULL, &error_);
|
|
|
|
unsigned int* data = (unsigned int*)ptr;
|
|
if (data[0] != 0) {
|
|
printf("OutData= 0x%08x\n", data[0]);
|
|
CHECK_RESULT_NO_RETURN(data[0] != 0, "Data validation failed!\n");
|
|
}
|
|
error_ = _wrapper->clEnqueueUnmapMemObject(cmd_queue_, buffer, ptr, 0, NULL, NULL);
|
|
}
|
|
|
|
void OCLPerfMemLatency::open(unsigned int test, char* units, double& conversion,
|
|
unsigned int deviceId) {
|
|
cl_uint numPlatforms;
|
|
cl_platform_id platform = NULL;
|
|
cl_uint num_devices = 0;
|
|
cl_device_id* devices = NULL;
|
|
cl_device_id device = NULL;
|
|
_crcword = 0;
|
|
conversion = 1.0f;
|
|
_deviceId = deviceId;
|
|
moreThreads = false;
|
|
|
|
context_ = 0;
|
|
cmd_queue_ = 0;
|
|
program_ = 0;
|
|
kernel_ = 0;
|
|
inBuffer_ = 0;
|
|
outBuffer_ = 0;
|
|
_errorFlag = false; // Reset error code so a single error doesn't prevent
|
|
// other subtests from running
|
|
_errorMsg = "";
|
|
isAMD_ = false;
|
|
|
|
error_ = _wrapper->clGetPlatformIDs(0, NULL, &numPlatforms);
|
|
CHECK_RESULT(error_ != CL_SUCCESS, "clGetPlatformIDs failed");
|
|
if (0 < numPlatforms) {
|
|
cl_platform_id* platforms = new cl_platform_id[numPlatforms];
|
|
error_ = _wrapper->clGetPlatformIDs(numPlatforms, platforms, NULL);
|
|
CHECK_RESULT(error_ != CL_SUCCESS, "clGetPlatformIDs failed");
|
|
|
|
platform = platforms[_platformIndex];
|
|
char pbuf[100];
|
|
error_ = _wrapper->clGetPlatformInfo(platforms[_platformIndex], CL_PLATFORM_VENDOR,
|
|
sizeof(pbuf), pbuf, NULL);
|
|
num_devices = 0;
|
|
/* Get the number of requested devices */
|
|
error_ = _wrapper->clGetDeviceIDs(platforms[_platformIndex], type_, 0, NULL, &num_devices);
|
|
// Runtime returns an error when no GPU devices are present instead of just
|
|
// returning 0 devices
|
|
// CHECK_RESULT(error_ != CL_SUCCESS, "clGetDeviceIDs failed");
|
|
// Choose platform with GPU devices
|
|
if (num_devices > 0) {
|
|
if (!strcmp(pbuf, "Advanced Micro Devices, Inc.")) {
|
|
isAMD_ = true;
|
|
}
|
|
}
|
|
|
|
delete platforms;
|
|
}
|
|
|
|
width_ = Sizes[test % NUM_SIZES];
|
|
|
|
bufSizeDW_ = width_ / sizeof(cl_uint);
|
|
moreThreads = ((test / NUM_SIZES) % 2) ? true : false;
|
|
makeVolatile = (test >= 2 * NUM_SIZES) ? true : false;
|
|
makeRW = (test >= 4 * NUM_SIZES) ? true : false;
|
|
|
|
CHECK_RESULT(platform == 0, "Couldn't find OpenCL platform, cannot proceed");
|
|
|
|
devices = (cl_device_id*)malloc(num_devices * sizeof(cl_device_id));
|
|
CHECK_RESULT(devices == 0, "Failed to allocate devices");
|
|
|
|
/* Get the requested device */
|
|
error_ = _wrapper->clGetDeviceIDs(platform, type_, num_devices, devices, NULL);
|
|
CHECK_RESULT(error_ != CL_SUCCESS, "clGetDeviceIDs failed");
|
|
|
|
device = devices[0];
|
|
|
|
free(devices);
|
|
devices = NULL;
|
|
context_ = _wrapper->clCreateContext(NULL, 1, &device, notify_callback, NULL, &error_);
|
|
CHECK_RESULT(context_ == 0, "clCreateContext failed");
|
|
|
|
cmd_queue_ = _wrapper->clCreateCommandQueue(context_, device, 0, NULL);
|
|
CHECK_RESULT(cmd_queue_ == 0, "clCreateCommandQueue failed");
|
|
|
|
inBuffer_ = _wrapper->clCreateBuffer(context_, 0, width_, NULL, &error_);
|
|
CHECK_RESULT(inBuffer_ == 0, "clCreateBuffer(inBuffer) failed");
|
|
|
|
outBuffer_ = _wrapper->clCreateBuffer(context_, 0, 1 * sizeof(cl_uint), NULL, &error_);
|
|
CHECK_RESULT(outBuffer_ == 0, "clCreateBuffer(outBuffer) failed");
|
|
|
|
genShader();
|
|
char* tmp = (char*)shader_.c_str();
|
|
program_ = _wrapper->clCreateProgramWithSource(context_, 1, (const char**)&tmp, NULL, &error_);
|
|
CHECK_RESULT(program_ == 0, "clCreateProgramWithSource failed");
|
|
|
|
std::string args;
|
|
args.clear();
|
|
if (isAMD_) args += " -D USE_FLOAT";
|
|
if (makeVolatile) args += " -D MAKEVOLATILE";
|
|
if (makeRW) args += " -D MAKERW";
|
|
|
|
error_ = _wrapper->clBuildProgram(program_, 1, &device, args.c_str(), NULL, NULL);
|
|
if (error_ != CL_SUCCESS) {
|
|
cl_int intError;
|
|
char log[16384];
|
|
intError = _wrapper->clGetProgramBuildInfo(program_, device, CL_PROGRAM_BUILD_LOG,
|
|
16384 * sizeof(char), log, NULL);
|
|
printf("Build error -> %s\n", log);
|
|
|
|
CHECK_RESULT(0, "clBuildProgram failed");
|
|
}
|
|
kernel_ = _wrapper->clCreateKernel(program_, "MemWalker", &error_);
|
|
CHECK_RESULT(kernel_ == 0, "clCreateKernel(MemWalker) failed");
|
|
|
|
kernel2_ = _wrapper->clCreateKernel(program_, "Overhead", &error_);
|
|
CHECK_RESULT(kernel2_ == 0, "clCreateKernel(Overhead) failed");
|
|
|
|
error_ = _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem), (void*)&inBuffer_);
|
|
error_ = _wrapper->clSetKernelArg(kernel_, 1, sizeof(cl_mem), (void*)&outBuffer_);
|
|
error_ = _wrapper->clSetKernelArg(kernel_, 2, sizeof(cl_uint), (void*)&bufSizeDW_);
|
|
error_ = _wrapper->clSetKernelArg(kernel_, 3, sizeof(cl_uint), (void*)&bufSizeDW_);
|
|
unsigned int zero = 0;
|
|
error_ = _wrapper->clSetKernelArg(kernel_, 4, sizeof(cl_uint), (void*)&zero);
|
|
int bMem = 1;
|
|
error_ = _wrapper->clSetKernelArg(kernel_, 5, sizeof(cl_int), (void*)&bMem);
|
|
// Limit the repeats, large buffers will have more samples, but the test runs
|
|
// for a long time
|
|
repeats_ = std::max((maxSize_ >> 4) / bufSizeDW_, 1u);
|
|
error_ = _wrapper->clSetKernelArg(kernel_, 6, sizeof(cl_uint), (void*)&repeats_);
|
|
|
|
error_ = _wrapper->clSetKernelArg(kernel2_, 0, sizeof(cl_mem), (void*)&inBuffer_);
|
|
error_ = _wrapper->clSetKernelArg(kernel2_, 1, sizeof(cl_mem), (void*)&outBuffer_);
|
|
error_ = _wrapper->clSetKernelArg(kernel2_, 2, sizeof(cl_uint), (void*)&bufSizeDW_);
|
|
error_ = _wrapper->clSetKernelArg(kernel2_, 3, sizeof(cl_uint), (void*)&bufSizeDW_);
|
|
error_ = _wrapper->clSetKernelArg(kernel2_, 4, sizeof(cl_uint), (void*)&zero);
|
|
error_ = _wrapper->clSetKernelArg(kernel2_, 5, sizeof(cl_int), (void*)&bMem);
|
|
error_ = _wrapper->clSetKernelArg(kernel2_, 6, sizeof(cl_uint), (void*)&repeats_);
|
|
|
|
setData(inBuffer_, (int)1.0f);
|
|
}
|
|
|
|
void OCLPerfMemLatency::run(void) {
|
|
int global = 1;
|
|
int local = 1;
|
|
|
|
if (moreThreads) {
|
|
if (isAMD_) {
|
|
global *= 64;
|
|
local *= 64;
|
|
} else {
|
|
global *= 32;
|
|
local *= 32;
|
|
}
|
|
}
|
|
size_t global_work_size[1] = {(size_t)global};
|
|
size_t local_work_size[1] = {(size_t)local};
|
|
|
|
// Warm-up
|
|
unsigned int warmup = 128;
|
|
error_ = _wrapper->clSetKernelArg(kernel_, 2, sizeof(cl_uint), (void*)&warmup);
|
|
error_ = _wrapper->clEnqueueNDRangeKernel(cmd_queue_, kernel_, 1, NULL,
|
|
(const size_t*)global_work_size,
|
|
(const size_t*)local_work_size, 0, NULL, NULL);
|
|
|
|
CHECK_RESULT(error_, "clEnqueueNDRangeKernel failed");
|
|
error_ = _wrapper->clSetKernelArg(kernel_, 2, sizeof(cl_uint), (void*)&bufSizeDW_);
|
|
_wrapper->clFinish(cmd_queue_);
|
|
|
|
// Restore input buffer when finished as it may have been modified by RW test
|
|
setData(inBuffer_, (int)1.0f);
|
|
|
|
CPerfCounter timer, timer2;
|
|
|
|
timer.Reset();
|
|
timer.Start();
|
|
|
|
error_ = _wrapper->clEnqueueNDRangeKernel(cmd_queue_, kernel_, 1, NULL,
|
|
(const size_t*)global_work_size,
|
|
(const size_t*)local_work_size, 0, NULL, NULL);
|
|
|
|
CHECK_RESULT(error_, "clEnqueueNDRangeKernel failed");
|
|
|
|
_wrapper->clFinish(cmd_queue_);
|
|
|
|
timer.Stop();
|
|
|
|
checkData(outBuffer_);
|
|
|
|
timer2.Reset();
|
|
timer2.Start();
|
|
|
|
error_ = _wrapper->clEnqueueNDRangeKernel(cmd_queue_, kernel2_, 1, NULL,
|
|
(const size_t*)global_work_size,
|
|
(const size_t*)local_work_size, 0, NULL, NULL);
|
|
|
|
CHECK_RESULT(error_, "clEnqueueNDRangeKernel failed");
|
|
|
|
_wrapper->clFinish(cmd_queue_);
|
|
|
|
timer2.Stop();
|
|
double sec = timer.GetElapsedTime() - timer2.GetElapsedTime();
|
|
|
|
// Read latency in ns
|
|
double perf = sec * (double)(1e09) / ((double)bufSizeDW_ * (double)repeats_);
|
|
|
|
_perfInfo = (float)perf;
|
|
char buf[256];
|
|
char buf2[32];
|
|
if (makeRW)
|
|
SNPRINTF(buf2, sizeof(buf2), "volatileRW");
|
|
else if (makeVolatile)
|
|
SNPRINTF(buf2, sizeof(buf2), "volatile");
|
|
else
|
|
buf2[0] = '\0';
|
|
SNPRINTF(buf, sizeof(buf), "%10s %2d threads, %8d reads, %5d repeats (ns)", buf2, global,
|
|
bufSizeDW_, repeats_);
|
|
testDescString = buf;
|
|
}
|
|
|
|
unsigned int OCLPerfMemLatency::close(void) {
|
|
_wrapper->clFinish(cmd_queue_);
|
|
|
|
if (inBuffer_) {
|
|
error_ = _wrapper->clReleaseMemObject(inBuffer_);
|
|
CHECK_RESULT_NO_RETURN(error_ != CL_SUCCESS, "clReleaseMemObject(inBuffer_) failed");
|
|
}
|
|
if (outBuffer_) {
|
|
error_ = _wrapper->clReleaseMemObject(outBuffer_);
|
|
CHECK_RESULT_NO_RETURN(error_ != CL_SUCCESS, "clReleaseMemObject(outBuffer_) failed");
|
|
}
|
|
if (kernel_) {
|
|
error_ = _wrapper->clReleaseKernel(kernel_);
|
|
CHECK_RESULT_NO_RETURN(error_ != CL_SUCCESS, "clReleaseKernel failed");
|
|
}
|
|
if (kernel2_) {
|
|
error_ = _wrapper->clReleaseKernel(kernel2_);
|
|
CHECK_RESULT_NO_RETURN(error_ != CL_SUCCESS, "clReleaseKernel failed");
|
|
}
|
|
if (program_) {
|
|
error_ = _wrapper->clReleaseProgram(program_);
|
|
CHECK_RESULT_NO_RETURN(error_ != CL_SUCCESS, "clReleaseProgram failed");
|
|
}
|
|
if (cmd_queue_) {
|
|
error_ = _wrapper->clReleaseCommandQueue(cmd_queue_);
|
|
CHECK_RESULT_NO_RETURN(error_ != CL_SUCCESS, "clReleaseCommandQueue failed");
|
|
}
|
|
if (context_) {
|
|
error_ = _wrapper->clReleaseContext(context_);
|
|
CHECK_RESULT_NO_RETURN(error_ != CL_SUCCESS, "clReleaseContext failed");
|
|
}
|
|
|
|
return _crcword;
|
|
}
|