AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SWDEV-470698 - fix formatting, add format check workflow (#657)

Browse Source
This commit is contained in:
Danylo Lytovchenko
2025-08-20 16:28:06 +02:00
committed by GitHub
parent 5840940caa
commit f7338717ae
1574 changed files with 162972 additions and 199346 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/opencl/tests/ocltst/module/perf/OCLPerfDoubleDMASeq.cpp
+44 -59
View File
@@ -48,7 +48,7 @@ static const int WindowWidth = 80;
const size_t MaxQueues = 3;
static const char *strKernel =
static const char* strKernel =
"__kernel void dummy(__global float4* out) \n"
"{ \n"
" uint id = get_global_id(0); \n"
@@ -68,15 +68,15 @@ OCLPerfDoubleDMASeq::OCLPerfDoubleDMASeq() {
OCLPerfDoubleDMASeq::~OCLPerfDoubleDMASeq() {}
void OCLPerfDoubleDMASeq::open(unsigned int test, char *units,
double &conversion, unsigned int deviceId) {
void OCLPerfDoubleDMASeq::open(unsigned int test, char* units, double& conversion,
unsigned int deviceId) {
_deviceId = deviceId;
OCLTestImp::open(test, units, conversion, deviceId);
CHECK_RESULT((error_ != CL_SUCCESS), "Error opening test");
test_ = test;
cl_device_type deviceType;
error_ = _wrapper->clGetDeviceInfo(devices_[deviceId], CL_DEVICE_TYPE,
sizeof(deviceType), &deviceType, NULL);
error_ = _wrapper->clGetDeviceInfo(devices_[deviceId], CL_DEVICE_TYPE, sizeof(deviceType),
&deviceType, NULL);
CHECK_RESULT((error_ != CL_SUCCESS), "CL_DEVICE_TYPE failed");
if (!(deviceType & CL_DEVICE_TYPE_GPU)) {
@@ -84,15 +84,13 @@ void OCLPerfDoubleDMASeq::open(unsigned int test, char *units,
failed_ = true;
return;
}
program_ = _wrapper->clCreateProgramWithSource(context_, 1, &strKernel, NULL,
&error_);
program_ = _wrapper->clCreateProgramWithSource(context_, 1, &strKernel, NULL, &error_);
CHECK_RESULT((error_ != CL_SUCCESS), "clCreateProgramWithSource() failed");
error_ = _wrapper->clBuildProgram(program_, 1, &devices_[deviceId], NULL,
NULL, NULL);
error_ = _wrapper->clBuildProgram(program_, 1, &devices_[deviceId], NULL, NULL, NULL);
if (error_ != CL_SUCCESS) {
char programLog[1024];
_wrapper->clGetProgramBuildInfo(program_, devices_[deviceId],
CL_PROGRAM_BUILD_LOG, 1024, programLog, 0);
_wrapper->clGetProgramBuildInfo(program_, devices_[deviceId], CL_PROGRAM_BUILD_LOG, 1024,
programLog, 0);
printf("\n%s\n", programLog);
fflush(stdout);
}
@@ -106,22 +104,19 @@ void OCLPerfDoubleDMASeq::open(unsigned int test, char *units,
events_ = ((test / MaxQueues) == 0) ? false : true;
size_t numBufs = (test_ % MaxQueues) + 1;
for (size_t b = 0; b < numBufs; ++b) {
buffer = _wrapper->clCreateBuffer(context_, CL_MEM_READ_WRITE, bufSize,
NULL, &error_);
buffer = _wrapper->clCreateBuffer(context_, CL_MEM_READ_WRITE, bufSize, NULL, &error_);
CHECK_RESULT((error_ != CL_SUCCESS), "clCreateBuffer() failed");
buffers_.push_back(buffer);
}
buffer = _wrapper->clCreateBuffer(context_,
CL_MEM_READ_WRITE | CL_MEM_ALLOC_HOST_PTR,
size_S, NULL, &error_);
buffer = _wrapper->clCreateBuffer(context_, CL_MEM_READ_WRITE | CL_MEM_ALLOC_HOST_PTR, size_S,
NULL, &error_);
CHECK_RESULT((error_ != CL_SUCCESS), "clCreateBuffer() failed");
buffers_.push_back(buffer);
}
static void CL_CALLBACK notify_callback(const char *errinfo,
const void *private_info, size_t cb,
void *user_data) {}
static void CL_CALLBACK notify_callback(const char* errinfo, const void* private_info, size_t cb,
void* user_data) {}
void OCLPerfDoubleDMASeq::run(void) {
if (failed_) {
@@ -135,32 +130,29 @@ void OCLPerfDoubleDMASeq::run(void) {
int q;
cl_command_queue_properties qProp = 0;
for (q = 0; q < numQueues; ++q) {
cl_command_queue cmdQueue = _wrapper->clCreateCommandQueue(
context_, devices_[_deviceId], qProp, &error_);
cl_command_queue cmdQueue =
_wrapper->clCreateCommandQueue(context_, devices_[_deviceId], qProp, &error_);
CHECK_RESULT((error_ != CL_SUCCESS), "clCreateCommandQueue() failed");
cmdQueues[q] = cmdQueue;
}
CHECK_RESULT_NO_RETURN((error_ != CL_SUCCESS), "Execution failed");
float *Data_s = (float *)_wrapper->clEnqueueMapBuffer(
cmdQueues[0], buffers_[numBufs], CL_TRUE, CL_MAP_READ | CL_MAP_WRITE, 0,
size_S, 0, NULL, NULL, &error_);
float* Data_s = (float*)_wrapper->clEnqueueMapBuffer(cmdQueues[0], buffers_[numBufs], CL_TRUE,
CL_MAP_READ | CL_MAP_WRITE, 0, size_S, 0,
NULL, NULL, &error_);
size_t gws[1] = {size_s / (4 * sizeof(float))};
size_t lws[1] = {256};
// Warm-up
for (q = 0; q < numQueues; ++q) {
error_ |= _wrapper->clEnqueueWriteBuffer(cmdQueues[q], buffers_[q], CL_FALSE, 0, size_s,
(char*)Data_s, 0, NULL, NULL);
error_ |= _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem), (void*)&buffers_[q]);
error_ |=
_wrapper->clEnqueueWriteBuffer(cmdQueues[q], buffers_[q], CL_FALSE, 0,
size_s, (char *)Data_s, 0, NULL, NULL);
error_ |= _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem),
(void *)&buffers_[q]);
error_ |= _wrapper->clEnqueueNDRangeKernel(cmdQueues[q], kernel_, 1, NULL,
gws, lws, 0, NULL, NULL);
error_ |=
_wrapper->clEnqueueReadBuffer(cmdQueues[q], buffers_[q], CL_FALSE, 0,
size_s, (char *)Data_s, 0, NULL, NULL);
_wrapper->clEnqueueNDRangeKernel(cmdQueues[q], kernel_, 1, NULL, gws, lws, 0, NULL, NULL);
error_ |= _wrapper->clEnqueueReadBuffer(cmdQueues[q], buffers_[q], CL_FALSE, 0, size_s,
(char*)Data_s, 0, NULL, NULL);
error_ |= _wrapper->clFinish(cmdQueues[q]);
}
@@ -180,13 +172,11 @@ void OCLPerfDoubleDMASeq::run(void) {
size_t iter = 0;
if (events_) {
while (1) {
error_ |= _wrapper->clEnqueueWriteBuffer(
cmdQueues[q], buffers_[idx], CL_FALSE, 0, size_s,
(char *)Data_s + s_done, 0, NULL, NULL);
error_ |= _wrapper->clEnqueueWriteBuffer(cmdQueues[q], buffers_[idx], CL_FALSE, 0, size_s,
(char*)Data_s + s_done, 0, NULL, NULL);
// Implicit flush of DMA engine on kernel start, because memory dependency
error_ |= _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem),
(void *)&buffers_[idx]);
error_ |= _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem), (void*)&buffers_[idx]);
int prevQ;
if (q == 0) {
prevQ = numQueues - 1;
@@ -194,13 +184,13 @@ void OCLPerfDoubleDMASeq::run(void) {
prevQ = q - 1;
}
if ((x[prevQ] != NULL) && (numQueues != 1)) {
error_ |= _wrapper->clEnqueueNDRangeKernel(
cmdQueues[q], kernel_, 1, NULL, gws, lws, 1, &x[prevQ], &x[q]);
error_ |= _wrapper->clEnqueueNDRangeKernel(cmdQueues[q], kernel_, 1, NULL, gws, lws, 1,
&x[prevQ], &x[q]);
error_ |= _wrapper->clReleaseEvent(x[prevQ]);
x[prevQ] = NULL;
} else {
error_ |= _wrapper->clEnqueueNDRangeKernel(
cmdQueues[q], kernel_, 1, NULL, gws, lws, 0, NULL, &x[q]);
error_ |= _wrapper->clEnqueueNDRangeKernel(cmdQueues[q], kernel_, 1, NULL, gws, lws, 0,
NULL, &x[q]);
if (numQueues == 1) {
error_ |= _wrapper->clReleaseEvent(x[q]);
x[q] = NULL;
@@ -209,9 +199,8 @@ void OCLPerfDoubleDMASeq::run(void) {
error_ |= _wrapper->clFlush(cmdQueues[q]);
// Change the queue
error_ |= _wrapper->clEnqueueReadBuffer(
cmdQueues[q], buffers_[idx], CL_FALSE, 0, size_s,
(char *)Data_s + s_done, 0, NULL, NULL);
error_ |= _wrapper->clEnqueueReadBuffer(cmdQueues[q], buffers_[idx], CL_FALSE, 0, size_s,
(char*)Data_s + s_done, 0, NULL, NULL);
if ((s_done += size_s) >= size_S) {
break;
@@ -229,20 +218,17 @@ void OCLPerfDoubleDMASeq::run(void) {
}
} else {
while (1) {
error_ |= _wrapper->clEnqueueWriteBuffer(
cmdQueues[q], buffers_[idx], CL_FALSE, 0, size_s,
(char *)Data_s + s_done, 0, NULL, NULL);
error_ |= _wrapper->clEnqueueWriteBuffer(cmdQueues[q], buffers_[idx], CL_FALSE, 0, size_s,
(char*)Data_s + s_done, 0, NULL, NULL);
// Implicit flush of DMA engine on kernel start, because memory dependency
error_ |= _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem),
(void *)&buffers_[idx]);
error_ |= _wrapper->clEnqueueNDRangeKernel(cmdQueues[q], kernel_, 1, NULL,
gws, lws, 0, NULL, NULL);
error_ |= _wrapper->clSetKernelArg(kernel_, 0, sizeof(cl_mem), (void*)&buffers_[idx]);
error_ |=
_wrapper->clEnqueueNDRangeKernel(cmdQueues[q], kernel_, 1, NULL, gws, lws, 0, NULL, NULL);
// Change the queue
error_ |= _wrapper->clEnqueueReadBuffer(
cmdQueues[q], buffers_[idx], CL_FALSE, 0, size_s,
(char *)Data_s + s_done, 0, NULL, NULL);
error_ |= _wrapper->clEnqueueReadBuffer(cmdQueues[q], buffers_[idx], CL_FALSE, 0, size_s,
(char*)Data_s + s_done, 0, NULL, NULL);
if ((s_done += size_s) >= size_S) {
break;
@@ -260,16 +246,15 @@ void OCLPerfDoubleDMASeq::run(void) {
}
timer.Stop();
error_ |= _wrapper->clEnqueueUnmapMemObject(cmdQueues[0], buffers_[numBufs],
Data_s, 0, NULL, NULL);
error_ |=
_wrapper->clEnqueueUnmapMemObject(cmdQueues[0], buffers_[numBufs], Data_s, 0, NULL, NULL);
error_ |= _wrapper->clFinish(cmdQueues[0]);
CHECK_RESULT_NO_RETURN((error_ != CL_SUCCESS), "Execution failed");
for (q = 0; q < numQueues; ++q) {
error_ = _wrapper->clReleaseCommandQueue(cmdQueues[q]);
CHECK_RESULT_NO_RETURN((error_ != CL_SUCCESS),
"clReleaseCommandQueue() failed");
CHECK_RESULT_NO_RETURN((error_ != CL_SUCCESS), "clReleaseCommandQueue() failed");
}
double GBytes = (double)(2 * size_S) / (double)(1000 * 1000 * 1000);