vectors adding error by opencl on i.mx6d

zhengfudi — Thu, 11 Aug 2016 03:30:39 GMT

Hi all,

I am learning opencl on i.mx6d, I wrote kernel code like this:

__kernel void helloworld ( __global uchar *input1,

__global uchar *input2,

__global uchar *output,

int width,

int height,

)

{

int y = get_global_id (0);

int x = get_global_id (1);

int id = (y * width) + x;

output[id] = input1[id]; // ok

// output[id] = input2[id]; // ok

// output[id] = input1[id] + input2[id]; // error

// output[id] = 0x01; // error

}

and host code like this:

#include <stdio.h>

#include <stdlib.h>

#include <assert.h>

#include "../include/CL/cl.h"

#define FSLCL_ERROR -1

#define FSLCL_SUCCESS CL_SUCCESS

#define MAX_SOURCE_SIZE (0x100000)

cl_mem buffer_input = NULL;

cl_mem buffer_input2 = NULL;

cl_mem buffer_output = NULL;

cl_kernel buffer_kernel = NULL;

size_t buffer_size = 0;

int buffer_width = 1024;

int buffer_height = 1024;

cl_platform_id platform_id;

cl_device_id device_id;

cl_context context;

cl_command_queue cq;

cl_program program;

cl_kernel kernel;

struct fsl_kernel_src_str

{

char *src;

long size;

};

typedef struct fsl_kernel_src_str fsl_kernel_src;

cl_int FSLCL_LoadKernelSource (char *filename, fsl_kernel_src *kernel)

{

FILE *fp = NULL;

fp = fopen (filename, "rb");

if (fp == NULL)

{

printf ("\nFailed to open: %s\n", filename);

return FSLCL_ERROR;

}

fseek (fp, 0, SEEK_END);

kernel->size = ftell (fp);

rewind (fp);

kernel->src = (char *) malloc (10 + sizeof (char) * kernel->size);

if (! kernel->src)

{

printf ("\nError Allocating memory to load CL program");

return FSLCL_ERROR;

}

kernel->size = fread (kernel->src, 1, MAX_SOURCE_SIZE, fp);

if (kernel->size < 1) {

printf("read file(%s) fail, ret=%d.\n", filename, kernel->size);

}

kernel->src[kernel->size] = '\0';

fclose (fp);

fp = NULL;

printf("close fp.\n");

return FSLCL_SUCCESS;

}

int main (int argc, char **argv)

{

int dimension = 2;

size_t global[2] = {buffer_width, buffer_height};

size_t local[2] = {4, 16};

int size_2d = buffer_width * buffer_height;

cl_int ret;

cl_int platforms;

char **data;

char **data2;

char **data0;

int i, j;

data0= (char **) malloc (buffer_width * sizeof (char *));

data = (char **) malloc (buffer_width * sizeof (char *));

data2 = (char **) malloc (buffer_width * sizeof (char *));

if(NULL == data0 || NULL == data || NULL == data2) {

printf("malloc1 err: %x, %x, %x\n", data0, data, data2);

return -1;

}

for (i = 0; i < buffer_width; i++)

{

data0[i] = (char *) malloc (buffer_height * sizeof (char));

data[i] = (char *) malloc (buffer_height * sizeof (char));

data2[i] = (char *) malloc (buffer_height * sizeof (char));

if(NULL == data0[i] || NULL == data[i] || NULL == data2[i]) {

printf("malloc2 err in i%d: %x, %x, %x\n", i, data0[i], data[i], data2[i]);

return -1;

}

for (i = 0; i < buffer_width; i++)

{

for (j = 0; j < buffer_height; j++)

{

data0[i][j] = 0;

data[i][j] = 0;

data2[i][j] = 0;

}

ret = clGetPlatformIDs (1, &platform_id, &platforms );

printf("clGetPlatformIDs ret = %d\n", ret);

assert (ret == CL_SUCCESS);

cl_int devices;

ret = clGetDeviceIDs (platform_id,CL_DEVICE_TYPE_GPU,1,&device_id, &devices);

assert (ret == CL_SUCCESS);

cl_context_properties properties[] = {CL_CONTEXT_PLATFORM, (cl_context_properties)platform_id, 0};

context = clCreateContext(properties, devices, &device_id, NULL,NULL,&ret);

assert (ret == CL_SUCCESS);

printf("clCreateContext ret =%d\n", ret);

cq = clCreateCommandQueue(context, device_id, 0, &ret);

assert (ret == CL_SUCCESS);

printf("clCreateCommandQueue ret =%d\n", ret);

fsl_kernel_src app_kernel;

ret = FSLCL_LoadKernelSource ((char *)"helloworld.cl", &app_kernel);

printf("FSLCL_LoadKernelSource ret =%d\n", ret);

// Submit the source code of the kernel to OpenCL

program = clCreateProgramWithSource (context, 1, (const char **)&app_kernel.src, 0,&ret);

printf("clCreateProgramWithSource ret =%d\n", ret);

// and compile it (after this we could extract the compiled version)

if (ret == CL_SUCCESS)

ret = clBuildProgram (program, 1, device_id, NULL, NULL, NULL);

if (ret < 0)

{

printf ("Failed\n");

printf ("\nReturn: %d\n", ret);

clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, app_kernel.size, app_kernel.src, NULL);

printf ("\n%s", app_kernel.src);

}

assert(ret == CL_SUCCESS);

buffer_input = clCreateBuffer (context, CL_MEM_READ_ONLY, size_2d, NULL, &ret);

printf("clCreateBuffer for input ret =%d\n", ret);

assert (ret == CL_SUCCESS);

buffer_input2 = clCreateBuffer (context, CL_MEM_READ_ONLY, size_2d, NULL, &ret);

printf("clCreateBuffer for input2 ret =%d\n", ret);

assert (ret == CL_SUCCESS);

buffer_output = clCreateBuffer (context, CL_MEM_WRITE_ONLY , size_2d, NULL, &ret);

assert (ret == CL_SUCCESS);

// get a handle and map parameters for the kernel

kernel = clCreateKernel(program, "helloworld", &ret);

printf("clCreateKernel ret =%d\n", ret);

usleep(100*1000);

assert (ret == CL_SUCCESS);

clSetKernelArg (kernel, 0, sizeof(cl_mem), &buffer_input);

clSetKernelArg (kernel, 1, sizeof(cl_mem), &buffer_input2);

clSetKernelArg (kernel, 2, sizeof(cl_mem), &buffer_output);

clSetKernelArg (kernel, 3, sizeof(int), &buffer_width);

clSetKernelArg (kernel, 4, sizeof(int), &buffer_height);

for (i = 0; i < buffer_width; i++)

{

for (j = 0; j < buffer_height; j++)

{

data0[i][j] = rand () % 10;

data[i][j] = rand () % 10;

}

ret = clEnqueueWriteBuffer (cq, buffer_input, CL_TRUE, 0, size_2d, data0, 0, NULL, NULL);

printf("clEnqueueWriteBuffer input1 ret =%d\n", ret);

assert (ret == CL_SUCCESS);

ret = clEnqueueWriteBuffer (cq, buffer_input2, CL_TRUE, 0, size_2d, data, 0, NULL, NULL);

printf("clEnqueueWriteBuffer input2 ret =%d\n", ret);

assert (ret == CL_SUCCESS);

ret = clEnqueueNDRangeKernel(cq, kernel, dimension, NULL, &global, &local, 0, NULL, NULL);

printf("clEnqueueNDRangeKernel ret =%d\n", ret);

assert (ret == CL_SUCCESS);

clFinish(cq);

ret = clEnqueueReadBuffer(cq, buffer_output, CL_TRUE, 0, size_2d, data2, 0, NULL, NULL);

printf("clEnqueueReadBuffer ret =%d\n", ret);

assert (ret == CL_SUCCESS);

printf ("\nResult:\n");

for (i = 0; i < buffer_width; i++)

{

for (j = 0; j < buffer_height; j++)

{

// printf ("\n%d , %d -- %d", data0[i][j], data[i][j], data2[i][j]);

printf("%d, ", data2[i][j]);

}

printf ("\n");

clFlush( cq);

clFinish(cq);

clReleaseMemObject (buffer_input);

clReleaseMemObject (buffer_input2);

clReleaseMemObject (buffer_output);

clReleaseContext(context);

clReleaseKernel(kernel);

clReleaseProgram(program);

clReleaseCommandQueue(cq);

for (i = 0; i < buffer_width; i++)

{

free(data0[i]);

free(data[i]);

free(data2[i]);

}

free(data0);

free(data);

free(data2);

return 0;

}

Question:

If the computing method in kernel code is "output[id] = input1[id];" or "output[id] = input2[id];", the whole program works well.

But, if the computing method in kernel code changed to "output[id] = input1[id]+input2[id];" or "output[id] = 0x01;", the program runs with an error output: segmentation fault. I have corfirmed that the line leads to this is "printf("%d, ", data2[i][j]);" in host code, but i am not sure what is the true cause. I will really appreciate it if someone can help me.

Re: vectors adding error by opencl on i.mx6d

igorpadykov — Thu, 11 Aug 2016 05:12:57 GMT

Hi zheng

please look at hello world example on

OpenCL Hello World | NXP Community

https://community.freescale.com/servlet/JiveServlet/previewBody/100694-102-1-15669/White%20Paper%20on%20get%20started%20…

Best regards

igor

-----------------------------------------------------------------------------------------------------------------------

Note: If this post answers your question, please click the Correct Answer button. Thank you!

-----------------------------------------------------------------------------------------------------------------------

topic vectors adding error by opencl on i.mx6d in i.MX Processors

vectors adding error by opencl on i.mx6d

Re: vectors adding error by opencl on i.mx6d