我處於OpenCl的學習階段。內核沒有給出每個核心的預期結果
問題:創建兩個數組:輸入和輸出50個元素,初始化值爲「-1」。
我將兩個數組作爲輸入和輸出緩衝區傳遞給GPU並解決(遞增數組的每個元素)5次迭代。因此, 所有元素的值應該增加到4.如,我除以10個核心此工作通過
size_t global_work_size[1] = {10};
和我的內核是根據get_global_id解決所有陣列(0)值,這意味着,解決每個核心中的每個5個元素。
所以,我最後的輸出應該是:
output[0] = 4.0000
output[1] = 4.0000
...
...
output[48] = 4.0000
output[49] = 4.0000
但是,我的輸出來了:
output[0] = 9.0000
output[1] = 9.0000
output[2] = 9.0000
output[3] = 9.0000
output[4] = 9.0000
output[5] = 4.0000
output[6] = 4.0000
output[7] = 4.0000
output[8] = 4.0000
output[9] = 4.0000
...
...
output[48] = 4.0000
output[49] = 4.0000
我們可以看到,在輸出差是前10個元素。這可能是因爲我的第一個內核運行兩次。 但是,我不明白確切的原因。請建議我,我哪裏錯了?可能是邏輯錯誤,或者我錯過了OpenCl中的一件大事?
program.c文件:
#include <stdio.h>
#include <stdlib.h>
#ifdef __APPLE__
#include <OpenCL/opencl.h>
#else
#include <CL/cl.h>
#endif
#define MEM_SIZE (50)
#define MAX_SOURCE_SIZE (0x100000)
int main() {
float input[MEM_SIZE], output[MEM_SIZE];
int go, i;
for (i = 0; i < MEM_SIZE; i++) {
input[i] = -1.0;
output[i] = -1.0;
}
FILE *fp;
cl_device_id device_id = NULL;
cl_context context = NULL;
cl_command_queue command_queue = NULL;
cl_program program = NULL;
cl_kernel kernel = NULL;
cl_platform_id platform_id = NULL;
cl_uint ret_num_devices;
cl_uint ret_num_platforms;
cl_int ret;
size_t source_size;
cl_event event;
char *source_str;
fp = fopen("algebra.cl", "r");
if (!fp) {
fprintf(stderr, "Failed to load kernel.\n");
exit(1);
}
source_str = (char*)malloc(MAX_SOURCE_SIZE);
source_size = fread(source_str, 1, MAX_SOURCE_SIZE, fp);
fclose(fp);
/*Initialization*/
/* Get Platform and Device Info */
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);
/*Initialization complete*/
cl_mem inputBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY|CL_MEM_COPY_HOST_PTR, MEM_SIZE * sizeof(float),(void *) input, NULL);
cl_mem outputBuffer = clCreateBuffer(context, CL_MEM_WRITE_ONLY , MEM_SIZE * sizeof(float), NULL, NULL);
ret = clEnqueueWriteBuffer(command_queue,
inputBuffer,
CL_FALSE,
0,
MEM_SIZE * sizeof(float),
input,
0,
NULL,
&event);
ret = clWaitForEvents(1, &event);
clReleaseEvent(event);
ret = clEnqueueWriteBuffer(command_queue,
outputBuffer,
CL_FALSE,
0,
MEM_SIZE * sizeof(float),
output,
0,
NULL,
&event);
ret = clWaitForEvents(1, &event);
clReleaseEvent(event);
/* Create Kernel Program from the source */
program = clCreateProgramWithSource(context, 1, (const char **)&source_str,(const size_t *)&source_size, &ret);
/* Build Kernel Program */
ret = clBuildProgram(program, 1, &device_id, NULL, NULL, NULL);
/* Create OpenCL Kernel */
kernel = clCreateKernel(program, "calc", &ret);
/* Set OpenCL Kernel Parameters */
ret = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&inputBuffer);
ret = clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *)&outputBuffer);
/* Execute OpenCL Kernel*/
ret = clEnqueueTask(command_queue, kernel, 0, NULL,NULL);
double io;
size_t global_work_size[1] = {10};
inputBuffer = clCreateBuffer(context, CL_MEM_READ_ONLY|CL_MEM_COPY_HOST_PTR, MEM_SIZE * sizeof(float),(void *) output, NULL);
ret = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&inputBuffer);
ret = clEnqueueNDRangeKernel(command_queue, kernel, 1, NULL, global_work_size, NULL, 0, NULL, NULL);
ret = clEnqueueReadBuffer(command_queue, outputBuffer, CL_TRUE, 0, MEM_SIZE * sizeof(float), output, 0, NULL, NULL);
for (go = 0; go < MEM_SIZE; go++) {
printf("output[%d] = %f\n", go, output[go]);
}
printf("\n\n");
/* Finalization */
ret = clFlush(command_queue);
ret = clFinish(command_queue);
ret = clReleaseKernel(kernel);
ret = clReleaseProgram(program);
ret = clReleaseMemObject(inputBuffer);
ret = clReleaseMemObject(outputBuffer);
ret = clReleaseCommandQueue(command_queue);
ret = clReleaseContext(context);
return 1;
}
algebra.cl(內核文件):
__kernel void euler(__global float* in, __global float* out)
{
int idx = get_global_id(0) * 5;
int end_idx = idx + 5;
double j = 0;
int i;
while (j < 5.0) {
j++;
for (i = idx; i < end_idx; i++) {
out[i] = out[i] + 1;
}
}
}
GCC編譯:
gcc program.c -o test -l OpenCL -I /usr/local/opencl/AMD-APP-SDK-v2.9-RC-lnx32/include/ -L /usr/local/opencl/AMD-APP-SDK-v2.9-RC-lnx32/lib/x86
這句話是正確的out [i] = out [i] + 1; ?還有什麼是10000用的。 – Meluha
感謝您指出@SagarKotecha。我發佈了錯誤的代碼。更新正確的一個。請檢查。 –