int setupCL(const char *filename)
{
//connect to a compute device
err = clGetDeviceIDs(NULL,CL_DEVICE_TYPE_CPU, 1, &devices, NULL);
//err = clGetDeviceIDs(NULL,CL_DEVICE_TYPE_GPU, 1, &devices, NULL);
//get info about device
size_t returned_size = 0;
cl_char vendor_name[1024] = {0};
cl_char device_name[1024] = {0};
err = clGetDeviceInfo(devices, CL_DEVICE_VENDOR, sizeof(vendor_name), vendor_name, &returned_size);
err|= clGetDeviceInfo(devices, CL_DEVICE_NAME, sizeof(device_name), device_name, &returned_size);
printf("Connecting to %s %s...\n", vendor_name, device_name);
//read the program
printf("Loading program '%s'\n\n", filename);
char *program_source = loadProgramSource(filename);
//create the context and command queue
context = clCreateContext(0, 1, &devices, NULL, NULL, &err);
cmd_queue = clCreateCommandQueue(context, devices, 0, NULL);
//create program from .cl file
program = clCreateProgramWithSource(context,1, (const char**)&program_source, NULL, &err);
//build the kernel program
err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
printf("Error: %d\n", err);
char build[2048];
clGetProgramBuildInfo(program, devices, CL_PROGRAM_BUILD_LOG, 2048, build, NULL);
printf("Build Log:\n%s\n",build);
//create the kernel
kernel = clCreateKernel(program, "rhs", &err);
//get size of buffers
size_t w_n_buffer_size = sizeof(float)*KSIZE*X*Y;
size_t K_buffer_size = sizeof(float)*KSIZE*X*Y;
size_t p_buffer_size = sizeof(float)*X*Y;
size_t H0_buffer_size = sizeof(float)*X*Y;
//create buffers
w_n_mem = clCreateBuffer(context, CL_MEM_READ_ONLY, w_n_buffer_size, NULL, NULL);
p_mem = clCreateBuffer(context, CL_MEM_READ_ONLY, p_buffer_size, NULL, NULL);
H0_mem = clCreateBuffer(context, CL_MEM_READ_ONLY, H0_buffer_size, NULL, NULL);
K_mem = clCreateBuffer(context, CL_MEM_READ_WRITE, K_buffer_size, NULL, NULL);
//set kernel arguments
err = clSetKernelArg(kernel, 0, sizeof(cl_mem), &w_n_mem);
err |= clSetKernelArg(kernel, 1, sizeof(cl_mem), &K_mem);
err |= clSetKernelArg(kernel, 2, sizeof(cl_mem), &p_mem);
err |= clSetKernelArg(kernel, 3, sizeof(cl_mem), &H0_mem);
err |= clSetKernelArg(kernel, 4, sizeof(int), &Y);
err |= clSetKernelArg(kernel, 5, sizeof(int), &X);
err |= clSetKernelArg(kernel, 6, sizeof(float), &dy);
err |= clSetKernelArg(kernel, 7, sizeof(float), &dx);
err |= clSetKernelArg(kernel, 8, sizeof(float), &alpha);
return CL_SUCCESS;
}
/*------------------------------------------------------------------------------------------------------------
* 指定したソースコードファイルからプログラムオブジェクとを作成する
* filename:カーネルのソースファイル
*/
void ClHelper::preloadProgram(const char* filename)
{
cl_int status;
char *source = 0;
// ファイルからプログラムを読み込む
try {
source = loadProgramSource(filename);
} catch (MyError err) {
fprintf(stderr, "Error: %s\n", err.cstr());
throw MyError("failed to load compute program from file.", __FUNCTION__);
}
// プログラムオブジェクトを作成する
mProgram = clCreateProgramWithSource(mContext,
1,
(const char **)&source,
NULL,
&status);
if (mProgram ==(cl_program)0) {
printError(status);
delete [] source;
throw MyError("failed to create program object", __FUNCTION__);
}
// プログラムをビルドする
cl_device_id devices[1];
devices[0] = mDevaiceId;
status = clBuildProgram(mProgram, 1, devices, NULL, NULL, NULL);
if (status != CL_SUCCESS) {
printError(status);
showBuildingLog(mProgram, devices[0]);
delete [] source;
throw MyError("failed to build program object.", __FUNCTION__);
}
delete [] source;
}
int main()
{
cl_platform_id platform_id = NULL;
cl_device_id device_id = NULL;
cl_context context = NULL;
cl_command_queue command_queue = NULL;
cl_mem objA = NULL;
cl_mem objB = NULL;
cl_mem objC = NULL;
cl_program program = NULL;
cl_kernel kernel = NULL;
cl_uint ret_num_devices;
cl_uint ret_num_platforms;
cl_int ret;
cl_event event1;
int i, j;
float *A;
float *B;
float *C;
A = (float *)malloc(4*4*sizeof(float));
B = (float *)malloc(4*4*sizeof(float));
C = (float *)malloc(4*4*sizeof(float));
/* Initialize input data */
for (i=0; i<4; i++) {
for (j=0; j<4; j++) {
A[i*4+j] = i*4+j+1;
B[i*4+j] = j*4+i+1;
}
}
/* Get Platform/Device Information*/
ret = clGetPlatformIDs(1, &platform_id, &ret_num_platforms);
ret = clGetDeviceIDs( platform_id, CL_DEVICE_TYPE_DEFAULT, 1, &device_id, &ret_num_devices);
/* Create OpenCL Context */
context = clCreateContext( NULL, 1, &device_id, NULL, NULL, &ret);
/* Create command queue */
command_queue = clCreateCommandQueue(context, device_id, 0, &ret);
/* Create Buffer Object */
objA = clCreateBuffer(context, CL_MEM_READ_WRITE, 4*4*sizeof(float), NULL, &ret);
objB = clCreateBuffer(context, CL_MEM_READ_WRITE, 4*4*sizeof(float), NULL, &ret);
objC = clCreateBuffer(context, CL_MEM_READ_WRITE, 4*4*sizeof(float), NULL, &ret);
/*
* Creating an user event
* As a user event is created, its execution status is set to be CL_SUBMITTED
* and we tag the event to a callback so when event reaches CL_COMPLETE, it will
* execute postProcess
*/
event1 = clCreateUserEvent(context, &ret);
clSetEventCallback(event1, CL_COMPLETE, &postProcess, "Looks like its done.");
/* Copy input data to the memory buffer */
ret = clEnqueueWriteBuffer(command_queue, objA, CL_TRUE, 0, 4*4*sizeof(float), A, 0, NULL, NULL );
printf("A has been written\n");
/* The next command will wait for event1 according to its status*/
ret = clEnqueueWriteBuffer(command_queue, objB, CL_TRUE, 0, 4*4*sizeof(float), B, 1, &event1, NULL);
printf("B has been written\n");
/* Tell event1 to complete */
clSetUserEventStatus(event1, CL_COMPLETE);
const char *file_names[] = {"sample_kernel.cl"};
const int NUMBER_OF_FILES = 1;
char* buffer[NUMBER_OF_FILES];
size_t sizes[NUMBER_OF_FILES];
loadProgramSource(file_names, NUMBER_OF_FILES, buffer, sizes);
/* Create kernel program from source file*/
program = clCreateProgramWithSource(context, 1, (const char **)buffer, sizes, &ret);
ret = clBuildProgram(program, 1, &device_id, NULL, NULL, NULL);
/* Create data parallel OpenCL kernel */
kernel = clCreateKernel(program, "sample", &ret);
/* Set OpenCL kernel arguments */
ret = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&objA);
ret = clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *)&objB);
ret = clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *)&objC);
size_t global_item_size = 4;
size_t local_item_size = 1;
/* Execute OpenCL kernel as data parallel */
ret = clEnqueueNDRangeKernel(command_queue, kernel, 1, NULL,
&global_item_size, &local_item_size, 0, NULL, NULL);
/* Transfer result to host */
ret = clEnqueueReadBuffer(command_queue, objC, CL_TRUE, 0, 4*4*sizeof(float), C, 0, NULL, NULL);
/* Display Results */
for (i=0; i<4; i++) {
for (j=0; j<4; j++) {
printf("%7.2f ", C[i*4+j]);
}
printf("\n");
}
/* Finalization */
ret = clFlush(command_queue);
ret = clFinish(command_queue);
ret = clReleaseKernel(kernel);
ret = clReleaseProgram(program);
ret = clReleaseMemObject(objA);
ret = clReleaseMemObject(objB);
ret = clReleaseMemObject(objC);
ret = clReleaseCommandQueue(command_queue);
ret = clReleaseContext(context);
free(A);
free(B);
free(C);
return 0;
}
int main(int argc, char** argv) {
/* OpenCL 1.1 data structures */
cl_platform_id* platforms;
cl_program program;
cl_device_id device;
cl_context context;
/* OpenCL 1.1 scalar data types */
cl_uint numOfPlatforms;
cl_int error;
/*
Get the number of platforms
Remember that for each vendor's SDK installed on the computer,
the number of available platform also increased.
*/
error = clGetPlatformIDs(0, NULL, &numOfPlatforms);
if(error != CL_SUCCESS) {
perror("Unable to find any OpenCL platforms");
exit(1);
}
platforms = (cl_platform_id*) alloca(sizeof(cl_platform_id) * numOfPlatforms);
printf("Number of OpenCL platforms found: %d\n", numOfPlatforms);
error = clGetPlatformIDs(numOfPlatforms, platforms, NULL);
if(error != CL_SUCCESS) {
perror("Unable to find any OpenCL platforms");
exit(1);
}
// Search for a CPU/GPU device through the installed platforms
// Build a OpenCL program and do not run it.
for(cl_uint i = 0; i < numOfPlatforms; i++ ) {
// Get the GPU device
error = clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_GPU, 1, &device, NULL);
if(error != CL_SUCCESS) {
// Otherwise, get the CPU
error = clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_CPU, 1, &device, NULL);
}
if(error != CL_SUCCESS) {
perror("Can't locate any OpenCL compliant device");
exit(1);
}
/* Create a context */
context = clCreateContext(NULL, 1, &device, NULL, NULL, &error);
if(error != CL_SUCCESS) {
perror("Can't create a valid OpenCL context");
exit(1);
}
/* Load the two source files into temporary datastores */
const char *file_names[] = {"simple.cl", "simple_2.cl"};
const int NUMBER_OF_FILES = 2;
char* buffer[NUMBER_OF_FILES];
size_t sizes[NUMBER_OF_FILES];
loadProgramSource(file_names, NUMBER_OF_FILES, buffer, sizes);
/* Create the OpenCL program object */
program = clCreateProgramWithSource(context, NUMBER_OF_FILES, (const char**)buffer, sizes, &error);
if(error != CL_SUCCESS) {
perror("Can't create the OpenCL program object");
exit(1);
}
/* Build OpenCL program object and dump the error message, if any */
char *program_log;
const char options[] = "-cl-finite-math-only -cl-no-signed-zeros";
size_t log_size;
//error = clBuildProgram(program, 1, &device, argv[1], NULL, NULL);
// Uncomment the line below, comment the line above; re-build the program to use build options statically
error = clBuildProgram(program, 1, &device, options, NULL, NULL);
if(error != CL_SUCCESS) {
// If there's an error whilst building the program, dump the log
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, 0, NULL, &log_size);
program_log = (char*) malloc(log_size+1);
program_log[log_size] = '\0';
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG,
log_size+1, program_log, NULL);
printf("\n=== ERROR ===\n\n%s\n=============\n", program_log);
free(program_log);
exit(1);
}
/* Query the program as to how many kernels were detected */
cl_uint numOfKernels;
error = clCreateKernelsInProgram(program, 0, NULL, &numOfKernels);
if (error != CL_SUCCESS) {
perror("Unable to retrieve kernel count from program");
exit(1);
}
cl_kernel* kernels = (cl_kernel*) alloca(sizeof(cl_kernel) * numOfKernels);
error = clCreateKernelsInProgram(program, numOfKernels, kernels, NULL);
for(cl_uint i = 0; i < numOfKernels; i++) {
char kernelName[32];
cl_uint argCnt;
clGetKernelInfo(kernels[i], CL_KERNEL_FUNCTION_NAME, sizeof(kernelName), kernelName, NULL);
clGetKernelInfo(kernels[i], CL_KERNEL_NUM_ARGS, sizeof(argCnt), &argCnt, NULL);
printf("Kernel name: %s with arity: %d\n", kernelName, argCnt);
}
/* Clean up */
for(cl_uint i = 0; i < numOfKernels; i++) { clReleaseKernel(kernels[i]); }
for(i=0; i< NUMBER_OF_FILES; i++) { free(buffer[i]); }
clReleaseProgram(program);
clReleaseContext(context);
}
}
int main(int argc, char** argv) {
/* OpenCL 1.1 data structures */
cl_platform_id* platforms;
cl_program program;
cl_device_id device;
cl_context context;
cl_command_queue queue;
cl_uint numOfPlatforms;
cl_int error;
cl_mem matrixAMemObj; // input matrix A mem buffer
cl_mem matrixBMemObj; // input matrix B mem buffer
cl_mem matrixCMemObj; // input matrix C mem buffer
cl_int* matrixA; // input matrix A
cl_int* matrixB; // input matrix B
cl_int* matrixC; // input matrix C
cl_uint widthA = WIDTH_G;
cl_uint heightA = HEIGHT_G;
cl_uint widthB = WIDTH_G;
cl_uint heightB = HEIGHT_G;
{
// allocate memory for input and output matrices
// based on whatever matrix theory i know.
matrixA = (cl_int*)malloc(widthA * heightA * sizeof(cl_int));
matrixB = (cl_int*)malloc(widthB * heightB * sizeof(cl_int));
matrixC = (cl_int*)malloc(widthB * heightA * sizeof(cl_int));
memset(matrixA, 0, widthA * heightA * sizeof(cl_int));
memset(matrixB, 0, widthB * heightB * sizeof(cl_int));
memset(matrixC, 0, widthB * heightA * sizeof(cl_int));
fillRandom(matrixA, widthA, heightA, 643);
fillRandom(matrixB, widthB, heightB, 991);
}
/*
Get the number of platforms
Remember that for each vendor's SDK installed on the computer,
the number of available platform also increased.
*/
error = clGetPlatformIDs(0, NULL, &numOfPlatforms);
if(error != CL_SUCCESS) {
perror("Unable to find any OpenCL platforms");
exit(1);
}
platforms = (cl_platform_id*) alloca(sizeof(cl_platform_id) * numOfPlatforms);
printf("Number of OpenCL platforms found: %d\n", numOfPlatforms);
error = clGetPlatformIDs(numOfPlatforms, platforms, NULL);
if(error != CL_SUCCESS) {
perror("Unable to find any OpenCL platforms");
exit(1);
}
// Search for a GPU device through the installed platforms
// Build a OpenCL program and do not run it.
for(cl_int i = 0; i < numOfPlatforms; i++ ) {
// Get the GPU device
error = clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_GPU, 1, &device, NULL);
if(error != CL_SUCCESS) {
perror("Can't locate a OpenCL compliant device i.e. GPU");
exit(1);
}
/* Create a context */
context = clCreateContext(NULL, 1, &device, NULL, NULL, &error);
if(error != CL_SUCCESS) {
perror("Can't create a valid OpenCL context");
exit(1);
}
/* Load the two source files into temporary datastores */
const char *file_names[] = {"simple_mm_mult.cl"};
const int NUMBER_OF_FILES = 1;
char* buffer[NUMBER_OF_FILES];
size_t sizes[NUMBER_OF_FILES];
loadProgramSource(file_names, NUMBER_OF_FILES, buffer, sizes);
/* Create the OpenCL program object */
program = clCreateProgramWithSource(context, NUMBER_OF_FILES, (const char**)buffer, sizes, &error);
if(error != CL_SUCCESS) {
perror("Can't create the OpenCL program object");
exit(1);
}
/* Build OpenCL program object and dump the error message, if any */
char *program_log;
const char options[] = "";
size_t log_size;
error = clBuildProgram(program, 1, &device, options, NULL, NULL);
if(error != CL_SUCCESS) {
// If there's an error whilst building the program, dump the log
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, 0, NULL, &log_size);
program_log = (char*) malloc(log_size+1);
program_log[log_size] = '\0';
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG,
log_size+1, program_log, NULL);
printf("\n=== ERROR ===\n\n%s\n=============\n", program_log);
free(program_log);
exit(1);
}
// Queue is created with profiling enabled
cl_command_queue_properties props;
props |= CL_QUEUE_PROFILING_ENABLE;
queue = clCreateCommandQueue(context, device, props, &error);
cl_kernel kernel = clCreateKernel(program, "mmmult", &error);
matrixAMemObj = clCreateBuffer(context,
CL_MEM_READ_ONLY|CL_MEM_COPY_HOST_PTR,
widthA * heightA * sizeof(cl_int),
matrixA,
&error);
matrixBMemObj = clCreateBuffer(context,
CL_MEM_READ_ONLY|CL_MEM_COPY_HOST_PTR,
widthB * heightB * sizeof(cl_int),
matrixB,
&error);
matrixCMemObj = clCreateBuffer(context,
CL_MEM_WRITE_ONLY|CL_MEM_ALLOC_HOST_PTR,
widthB * heightA * sizeof(cl_int),
0,
&error);
clSetKernelArg(kernel, 0, sizeof(cl_int),(void*)&widthB);
clSetKernelArg(kernel, 1, sizeof(cl_int),(void*)&heightA);
clSetKernelArg(kernel, 2, sizeof(cl_mem),(void*)&matrixAMemObj);
clSetKernelArg(kernel, 3, sizeof(cl_mem),(void*)&matrixBMemObj);
clSetKernelArg(kernel, 4, sizeof(cl_mem),(void*)&matrixCMemObj);
size_t globalThreads[] = {widthB, heightA};
cl_event exeEvt;
cl_ulong executionStart, executionEnd;
error = clEnqueueNDRangeKernel(queue,
kernel,
2,
NULL,
globalThreads,
NULL,
0,
NULL,
&exeEvt);
clWaitForEvents(1, &exeEvt);
if(error != CL_SUCCESS) {
printf("Kernel execution failure!\n");
exit(-22);
}
// let's understand how long it took?
clGetEventProfilingInfo(exeEvt, CL_PROFILING_COMMAND_START, sizeof(executionStart), &executionStart, NULL);
clGetEventProfilingInfo(exeEvt, CL_PROFILING_COMMAND_END, sizeof(executionEnd), &executionEnd, NULL);
clReleaseEvent(exeEvt);
// printf("Execution the matrix-matrix multiplication took %lu.%lu s\n", (executionEnd - executionStart)/1000000000, (executionEnd - executionStart)%1000000000);
printf("Execution the matrix-matrix multiplication took %lu s\n", (executionEnd - executionStart));
clEnqueueReadBuffer(queue,
matrixCMemObj,
CL_TRUE,
0,
widthB * heightA * sizeof(cl_int),
matrixC,
0,
NULL,
NULL);
if (compare(matrixC, matrixA, matrixB, heightA, widthA, widthB))
printf("Passed!\n");
else
printf("Failed!\n");
/* Clean up */
for(i=0; i< NUMBER_OF_FILES; i++) { free(buffer[i]); }
clReleaseProgram(program);
clReleaseContext(context);
clReleaseMemObject(matrixAMemObj);
clReleaseMemObject(matrixBMemObj);
clReleaseMemObject(matrixCMemObj);
}
free(matrixA);
free(matrixB);
free(matrixC);
}
int main(int argc, char** argv) {
/* OpenCL 1.1 data structures */
cl_platform_id* platforms;
cl_program program;
cl_device_id device;
cl_context context;
/* OpenCL 1.1 scalar data types */
cl_uint numOfPlatforms;
cl_int error;
/*
Prepare an array of UserData via dynamic memory allocation
*/
UserData* ud_in = (UserData*) malloc( sizeof(UserData) * DATA_SIZE); // input to device
UserData* ud_out = (UserData*) malloc( sizeof(UserData) * DATA_SIZE); // output from device
for( int i = 0; i < DATA_SIZE; ++i) {
(ud_in + i)->x = i;
(ud_in + i)->y = i;
(ud_in + i)->z = i;
(ud_in + i)->w = 3 * i;
}
/*
Get the number of platforms
Remember that for each vendor's SDK installed on the computer,
the number of available platform also increased.
*/
error = clGetPlatformIDs(0, NULL, &numOfPlatforms);
if(error != CL_SUCCESS ) {
perror("Unable to find any OpenCL platforms");
exit(1);
}
platforms = (cl_platform_id*) alloca(sizeof(cl_platform_id) * numOfPlatforms);
printf("Number of OpenCL platforms found: %d\n", numOfPlatforms);
error = clGetPlatformIDs(numOfPlatforms, platforms, NULL);
if(error != CL_SUCCESS ) {
perror("Unable to find any OpenCL platforms");
exit(1);
}
// Search for a CPU/GPU device through the installed platforms
// Build a OpenCL program and do not run it.
for(cl_uint i = 0; i < numOfPlatforms; i++ ) {
// Get the GPU device
error = clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_GPU, 1, &device, NULL);
if(error != CL_SUCCESS) {
// Otherwise, get the CPU
error = clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_CPU, 1, &device, NULL);
}
if(error != CL_SUCCESS) {
perror("Can't locate any OpenCL compliant device");
exit(1);
}
/* Create a context */
context = clCreateContext(NULL, 1, &device, NULL, NULL, &error);
if(error != CL_SUCCESS) {
perror("Can't create a valid OpenCL context");
exit(1);
}
/* Load the two source files into temporary datastores */
const char *file_names[] = {"user_test.cl"};
const int NUMBER_OF_FILES = 1;
char* buffer[NUMBER_OF_FILES];
size_t sizes[NUMBER_OF_FILES];
loadProgramSource(file_names, NUMBER_OF_FILES, buffer, sizes);
/* Create the OpenCL program object */
program = clCreateProgramWithSource(context, NUMBER_OF_FILES, (const char**)buffer, sizes, &error);
if(error != CL_SUCCESS) {
perror("Can't create the OpenCL program object");
exit(1);
}
/* Build OpenCL program object and dump the error message, if any */
char *program_log;
size_t log_size;
error = clBuildProgram(program, 1, &device, NULL, NULL, NULL);
if(error != CL_SUCCESS) {
// If there's an error whilst building the program, dump the log
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, 0, NULL, &log_size);
program_log = (char*) malloc(log_size+1);
program_log[log_size] = '\0';
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG,
log_size+1, program_log, NULL);
printf("\n=== ERROR ===\n\n%s\n=============\n", program_log);
free(program_log);
exit(1);
}
/* Query the program as to how many kernels were detected */
cl_uint numOfKernels;
error = clCreateKernelsInProgram(program, 0, NULL, &numOfKernels);
if (error != CL_SUCCESS) {
perror("Unable to retrieve kernel count from program");
exit(1);
}
cl_kernel* kernels = (cl_kernel*) alloca(sizeof(cl_kernel) * numOfKernels);
error = clCreateKernelsInProgram(program, numOfKernels, kernels, NULL);
for(cl_uint i = 0; i < numOfKernels; i++) {
char kernelName[32];
cl_uint argCnt;
clGetKernelInfo(kernels[i], CL_KERNEL_FUNCTION_NAME, sizeof(kernelName), kernelName, NULL);
clGetKernelInfo(kernels[i], CL_KERNEL_NUM_ARGS, sizeof(argCnt), &argCnt, NULL);
printf("Kernel name: %s with arity: %d\n", kernelName, argCnt);
printf("About to create command queue and enqueue this kernel...\n");
/* Create a command queue */
cl_command_queue cQ = clCreateCommandQueue(context, device, 0, &error);
if (error != CL_SUCCESS) {
perror("Unable to create command-queue");
exit(1);
}
/* Create a OpenCL buffer object */
cl_mem UDObj = clCreateBuffer(context, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR,
sizeof(UserData) * DATA_SIZE, ud_in, &error);
if (error != CL_SUCCESS) {
perror("Unable to create buffer object");
exit(1);
}
/* Let OpenCL know that the kernel is suppose to receive an argument */
error = clSetKernelArg(kernels[i], 0, sizeof(cl_mem), &UDObj);
if (error != CL_SUCCESS) {
perror("Unable to set buffer object as kernel argument");
exit(1);
}
/* Enqueue the kernel to the command queue */
error = clEnqueueTask(cQ, kernels[i], 0, NULL, NULL);
if (error != CL_SUCCESS) {
perror("Unable to enqueue task to command-queue");
exit(1);
}
printf("Task has been enqueued successfully!\n");
/* Enqueue the read-back from device to host */
error = clEnqueueReadBuffer(cQ, UDObj,
CL_TRUE, // blocking read
0, // write from the start
sizeof(UserData) * DATA_SIZE, // how much to copy
ud_out, 0, NULL, NULL);
if ( valuesOK(ud_in, ud_out) ) {
printf("Check passed!\n");
} else printf("Check failed!\n");
/* Release the command queue */
clReleaseCommandQueue(cQ);
clReleaseMemObject(UDObj);
}
/* Clean up */
for(cl_uint i = 0; i < numOfKernels; i++) { clReleaseKernel(kernels[i]); }
for(i=0; i< NUMBER_OF_FILES; i++) { free(buffer[i]); }
clReleaseProgram(program);
clReleaseContext(context);
}
free(ud_in);
free(ud_out);
}
int main(int argc, char** argv) {
/* OpenCL 1.1 data structures */
cl_platform_id* platforms;
cl_program program;
cl_context context;
/* OpenCL 1.1 scalar data types */
cl_uint numOfPlatforms;
cl_int error;
cl_float16* ud_in = (cl_float16*) malloc( sizeof(cl_float16) * DATA_SIZE); // input to device
cl_float16* ud_out = (cl_float16*) malloc( sizeof(cl_float16) * DATA_SIZE); // output from device
for( int i = 0; i < DATA_SIZE; ++i) {
ud_in[i] = (cl_float16){ (float)i,(float)i,(float)i,(float)i, (float)i,(float)i,(float)i,(float)i, (float)i,(float)i,(float)i,(float)i, (float)i,(float)i,(float)i,(float)i };
}
/*
Get the number of platforms
Remember that for each vendor's SDK installed on the computer,
the number of available platform also increased.
*/
error = clGetPlatformIDs(0, NULL, &numOfPlatforms);
if(error != CL_SUCCESS ) {
perror("Unable to find any OpenCL platforms");
exit(1);
}
platforms = (cl_platform_id*) alloca(sizeof(cl_platform_id) * numOfPlatforms);
printf("Number of OpenCL platforms found: %d\n", numOfPlatforms);
error = clGetPlatformIDs(numOfPlatforms, platforms, NULL);
if(error != CL_SUCCESS ) {
perror("Unable to find any OpenCL platforms");
exit(1);
}
// Search for a CPU/GPU device through the installed platforms
// Build a OpenCL program and do not run it.
for(cl_uint i = 0; i < numOfPlatforms; i++ ) {
cl_uint numOfDevices = 0;
/* Determine how many devices are connected to your platform */
error = clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_ALL, 0, NULL, &numOfDevices);
if (error != CL_SUCCESS ) {
perror("Unable to obtain any OpenCL compliant device info");
exit(1);
}
cl_device_id* devices = (cl_device_id*) alloca(sizeof(cl_device_id) * numOfDevices);
/* Load the information about your devices into the variable 'devices' */
error = clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_ALL, numOfDevices, devices, NULL);
if (error != CL_SUCCESS ) {
perror("Unable to obtain any OpenCL compliant device info");
exit(1);
}
printf("Number of detected OpenCL devices: %d\n", numOfDevices);
/* Create a context */
cl_context_properties ctx[3] = { CL_CONTEXT_PLATFORM, (cl_context_properties)platforms[i], 0 };
context = clCreateContext(ctx, numOfDevices, devices, NULL, NULL, &error);
if(error != CL_SUCCESS) {
perror("Can't create a valid OpenCL context");
exit(1);
}
/* For each device, create a buffer and partition that data among the devices for compute! */
cl_mem inobj = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
sizeof(cl_float16) * DATA_SIZE, ud_in, &error);
if(error != CL_SUCCESS) {
perror("Can't create a buffer");
exit(1);
}
int offset = 0;
for(int i = 0; i < numOfDevices; ++i, ++offset ) {
/* Load the two source files into temporary datastores */
const char *file_names[] = {"vector_load.cl"};
const int NUMBER_OF_FILES = 1;
char* buffer[NUMBER_OF_FILES];
size_t sizes[NUMBER_OF_FILES];
loadProgramSource(file_names, NUMBER_OF_FILES, buffer, sizes);
/* Create the OpenCL program object */
program = clCreateProgramWithSource(context, NUMBER_OF_FILES, (const char**)buffer, sizes, &error);
if(error != CL_SUCCESS) {
perror("Can't create the OpenCL program object");
exit(1);
}
/* Build OpenCL program object and dump the error message, if any */
char *program_log;
size_t log_size;
error = clBuildProgram(program, 1, &devices[i], NULL, NULL, NULL);
if(error != CL_SUCCESS) {
// If there's an error whilst building the program, dump the log
clGetProgramBuildInfo(program, devices[i], CL_PROGRAM_BUILD_LOG, 0, NULL, &log_size);
program_log = (char*) malloc(log_size+1);
program_log[log_size] = '\0';
clGetProgramBuildInfo(program, devices[i], CL_PROGRAM_BUILD_LOG,
log_size+1, program_log, NULL);
printf("\n=== ERROR ===\n\n%s\n=============\n", program_log);
free(program_log);
exit(1);
}
/* Query the program as to how many kernels were detected */
cl_uint numOfKernels;
error = clCreateKernelsInProgram(program, 0, NULL, &numOfKernels);
if (error != CL_SUCCESS) {
perror("Unable to retrieve kernel count from program");
exit(1);
}
cl_kernel* kernels = (cl_kernel*) alloca(sizeof(cl_kernel) * numOfKernels);
error = clCreateKernelsInProgram(program, numOfKernels, kernels, NULL);
/* Loop thru each kernel and execute on device */
for(cl_uint j = 0; j < numOfKernels; j++) {
char kernelName[32];
cl_uint argCnt;
clGetKernelInfo(kernels[j], CL_KERNEL_FUNCTION_NAME, sizeof(kernelName), kernelName, NULL);
clGetKernelInfo(kernels[j], CL_KERNEL_NUM_ARGS, sizeof(argCnt), &argCnt, NULL);
printf("Kernel name: %s with arity: %d\n", kernelName, argCnt);
printf("About to create command queue and enqueue this kernel...\n");
/* Create a command queue */
cl_command_queue cQ = clCreateCommandQueue(context, devices[i], 0, &error);
if (error != CL_SUCCESS) {
perror("Unable to create command-queue");
exit(1);
}
/* Create a buffer and copy the data from the main buffer */
cl_mem outobj = clCreateBuffer(context, CL_MEM_WRITE_ONLY,
sizeof(cl_float16) * DATA_SIZE, 0, &error);
if (error != CL_SUCCESS) {
perror("Unable to create sub-buffer object");
exit(1);
}
/* Let OpenCL know that the kernel is suppose to receive an argument */
error = clSetKernelArg(kernels[j], 0, sizeof(cl_mem), &inobj);
error = clSetKernelArg(kernels[j], 1, sizeof(cl_mem), &outobj);
if (error != CL_SUCCESS) {
perror("Unable to set buffer object in kernel");
exit(1);
}
/* Enqueue the kernel to the command queue */
size_t threadsPerGroup[] = {4};
size_t numOfGroups[] = { DATA_SIZE / threadsPerGroup[0] };
error = clEnqueueNDRangeKernel(cQ,
kernels[j],
1,
0,
numOfGroups,
threadsPerGroup,0, NULL, NULL);
if (error != CL_SUCCESS) {
perror("Unable to enqueue task to command-queue");
exit(1);
}
printf("Task has been enqueued successfully!\n");
/* Enqueue the read-back from device to host */
error = clEnqueueReadBuffer(cQ, outobj,
CL_TRUE, // blocking read
0, // read from the start
sizeof(cl_float16)*DATA_SIZE, // how much to copy
ud_out, 0, NULL, NULL);
/* Check the returned data */
if ( valuesOK(ud_in, ud_out, DATA_SIZE) ) {
printf("Check passed!\n");
} else printf("Check failed!\n");
/* Release the command queue */
clReleaseCommandQueue(cQ);
clReleaseMemObject(outobj);
}
/* Clean up */
for(cl_uint i = 0; i < numOfKernels; i++) { clReleaseKernel(kernels[i]); }
for(int i=0; i< NUMBER_OF_FILES; i++) { free(buffer[i]); }
clReleaseProgram(program);
}// end of device loop and execution
clReleaseMemObject(inobj);
clReleaseContext(context);
}// end of platform loop
free(ud_in);
free(ud_out);
}