the source(from https://code.google.com/p/float-cl/ with some modifly ):
/*
This file is part of floatCL.
floatCL is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
floatCL is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with floatCL. If not, see <http://www.gnu.org/licenses/>.
Author: Aditya Vijayakumar
Date: Apr 10, 2011
*/
#include "floatCL.h"
cl_int error;
cl_platform_id platform;
cl_device_id device;
cl_uint platforms, devices;
cl_program prog;
cl_context context;
cl_command_queue cq;
cl_kernel k_copy, k_zero_mem, k_find_maxA, k_find_maxB, k_sort_fill,
k_reset_sort;
cl_mem ref_buf, ref_tmp, misc;
void init_custom_cl(int buf_size)
{
FILE * fp;
char cl_src[CL_SRC_SIZE];
const char * const_src[] = {cl_src};
size_t cl_src_size;
error = clGetPlatformIDs(1, &platform, &platforms);
error = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, &devices);
cl_context_properties properties[] = {CL_CONTEXT_PLATFORM,
(cl_context_properties)platform, 0};
context = clCreateContext(properties, 1, &device, NULL, NULL, &error);
cq = clCreateCommandQueue(context, device, 0, &error);
fp = fopen(CL_FILE_NAME, "r");
cl_src_size = fread(cl_src, 1, CL_SRC_SIZE, fp);
prog =
clCreateProgramWithSource(context, 1, const_src, &cl_src_size, &error);
error = clBuildProgram(prog, 0, NULL, "", NULL, NULL);
if (error != CL_SUCCESS)
{
puts("CL: RT compile error(parallel.cl)");
return;
}
k_copy = clCreateKernel(prog, KERNEL_COPY, &error);
k_zero_mem = clCreateKernel(prog, KERNEL_ZERO_MEM, &error);
k_find_maxA = clCreateKernel(prog, KERNEL2D_FIND_MAXA, &error);
k_find_maxB = clCreateKernel(prog, KERNEL_FIND_MAXB, &error);
k_sort_fill = clCreateKernel(prog, SKERNEL_SORT_FILL, &error);
k_reset_sort = clCreateKernel(prog, SKERNEL_RESET_SORT, &error);
ref_buf = clCreateBuffer(context, CL_MEM_READ_WRITE,
sizeof(float) * buf_size, NULL, &error);
ref_tmp = clCreateBuffer(context, CL_MEM_READ_WRITE,
sizeof(float) * buf_size, NULL, &error);
misc = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(float) * N_MISC,
NULL, &error);
}
void clean_custom_cl()
{
clReleaseKernel(k_copy);
clReleaseKernel(k_zero_mem);
clReleaseKernel(k_find_maxA);
clReleaseKernel(k_find_maxB);
clReleaseKernel(k_sort_fill);
clReleaseKernel(k_reset_sort);
clReleaseCommandQueue(cq);
clReleaseContext(context);
clReleaseMemObject(ref_buf);
clReleaseMemObject(ref_tmp);
clReleaseMemObject(misc);
}
cl_int call_sort(cl_mem vals, cl_mem index, int no)
{
int i;
int j;
float todel[100];
call_copy(ref_tmp, vals, no); // used as a copier
call_reset_sort();
for (i = 0; i < no; i++)
{
call_max(ref_tmp, misc, no);
call_sort_fill(ref_tmp, index);
}
}
cl_int call_reset_sort()
{
const size_t worksize = 1;
clSetKernelArg(k_reset_sort, 0, sizeof(misc), &misc);
error = clEnqueueNDRangeKernel(cq, k_reset_sort, 1, NULL, &worksize, NULL,
0, NULL, NULL);
error = clFinish(cq);
}
cl_int call_sort_fill(cl_mem vals, cl_mem index)
{
const size_t worksize = 1;
clSetKernelArg(k_sort_fill, 0, sizeof(vals), &vals);
clSetKernelArg(k_sort_fill, 1, sizeof(misc), &misc);
clSetKernelArg(k_sort_fill, 2, sizeof(index), &index);
error = clEnqueueNDRangeKernel(cq, k_sort_fill, 1, NULL, &worksize, NULL, 0,
NULL, NULL);
error = clFinish(cq);
}
cl_int call_max(cl_mem vals, cl_mem ret, int no)
{
const size_t worksize2d[] = {no, no};
const size_t worksize = no;
call_zero_mem(ref_buf, no);
clSetKernelArg(k_find_maxA, 0, sizeof(vals), &vals);
clSetKernelArg(k_find_maxA, 1, sizeof(ref_buf), &ref_buf);
error = clEnqueueNDRangeKernel(cq, k_find_maxA, 2, NULL, worksize2d, NULL,
0, NULL, NULL);
error = clFinish(cq);
clSetKernelArg(k_find_maxB, 0, sizeof(ref_buf), &ref_buf);
clSetKernelArg(k_find_maxB, 1, sizeof(ret), &ret);
error = clEnqueueNDRangeKernel(cq, k_find_maxB, 1, NULL, &worksize, NULL, 0,
NULL, NULL);
error = clFinish(cq);
}
cl_int call_zero_mem(cl_mem vals, int no)
{
const size_t worksize = no;
clSetKernelArg(k_zero_mem, 0, sizeof(vals), &vals);
error = clEnqueueNDRangeKernel(cq, k_zero_mem, 1, NULL, &worksize, NULL, 0,
NULL, NULL);
error = clFinish(cq);
}
cl_int call_copy(cl_mem a, cl_mem b, int no)
{
const size_t worksize = no;
clSetKernelArg(k_copy, 0, sizeof(a), &a);
clSetKernelArg(k_copy, 1, sizeof(b), &b);
error = clEnqueueNDRangeKernel(cq, k_copy, 1, NULL, &worksize, NULL, 0,
NULL, NULL);
error = clFinish(cq);
}
cl_int read_from_GPURAM(cl_mem mem, float * a, int n)
{
error = clEnqueueReadBuffer(cq, mem, CL_TRUE, 0, sizeof(float) * n, a, 0,
NULL, NULL);
error = clFinish(cq);
return error;
}
cl_mem put_to_GPURAM(float * a, int n)
{
cl_mem mem;
mem = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(float) * n, NULL,
&error);
error = clEnqueueWriteBuffer(cq, mem, CL_TRUE, 0, sizeof(float) * n, a, 0,
NULL, NULL);
return mem;
}
cl_mem only_alloc_mem(int n)
{
cl_mem mem;
mem = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(float) * n, NULL,
&error);
return mem;
}
after apply source formatter(default settings, allman) of CB svn 9986 build:
/*
This file is part of floatCL.
This file is part of floatCL.
floatCL is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
floatCL is free software: you can redistribute it and/or modify
floatCL is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
the Free Software Foundation, either version 3 of the License, or
You should have received a copy of the GNU General Public License
along with floatCL. If not, see <http://www.gnu.org/licenses/>.
Author: Aditya Vijayakumar
Date: Apr 10, 2011
*/
but WITHOUT ANY WARRANTY; without even the implied warranty of
#include "floatCL.h"
GNU General Public License for more details.
cl_int error;
cl_platform_id platform;
cl_device_id device;
cl_uint platforms, devices;
cl_program prog;
cl_context context;
cl_command_queue cq;
You should have received a copy of the GNU General Public License
cl_kernel k_copy, k_zero_mem, k_find_maxA, k_find_maxB, k_sort_fill,
k_reset_sort;
cl_mem ref_buf, ref_tmp, misc;
Date: Apr 10, 2011
void init_custom_cl(int buf_size)
{
FILE * fp;
char cl_src[CL_SRC_SIZE];
const char * const_src[] = {cl_src};
size_t cl_src_size;
error = clGetPlatformIDs(1, &platform, &platforms);
error = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, &devices);
cl_platform_id platform;
cl_context_properties properties[] = {CL_CONTEXT_PLATFORM,
(cl_context_properties)platform, 0
};
context = clCreateContext(properties, 1, &device, NULL, NULL, &error);
cq = clCreateCommandQueue(context, device, 0, &error);
cl_uint platforms, devices;
fp = fopen(CL_FILE_NAME, "r");
cl_src_size = fread(cl_src, 1, CL_SRC_SIZE, fp);
cl_context context;
prog =
clCreateProgramWithSource(context, 1, const_src, &cl_src_size, &error);
error = clBuildProgram(prog, 0, NULL, "", NULL, NULL);
if (error != CL_SUCCESS)
{
puts("CL: RT compile error(parallel.cl)");
return;
}
k_copy = clCreateKernel(prog, KERNEL_COPY, &error);
k_zero_mem = clCreateKernel(prog, KERNEL_ZERO_MEM, &error);
k_find_maxA = clCreateKernel(prog, KERNEL2D_FIND_MAXA, &error);
k_find_maxB = clCreateKernel(prog, KERNEL_FIND_MAXB, &error);
k_sort_fill = clCreateKernel(prog, SKERNEL_SORT_FILL, &error);
k_reset_sort = clCreateKernel(prog, SKERNEL_RESET_SORT, &error);
k_reset_sort;
ref_buf = clCreateBuffer(context, CL_MEM_READ_WRITE,
sizeof(float) * buf_size, NULL, &error);
ref_tmp = clCreateBuffer(context, CL_MEM_READ_WRITE,
sizeof(float) * buf_size, NULL, &error);
misc = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(float) * N_MISC,
NULL, &error);
}
cl_mem ref_buf, ref_tmp, misc;
void clean_custom_cl()
{
clReleaseKernel(k_copy);
clReleaseKernel(k_zero_mem);
clReleaseKernel(k_find_maxA);
clReleaseKernel(k_find_maxB);
clReleaseKernel(k_sort_fill);
clReleaseKernel(k_reset_sort);
void init_custom_cl(int buf_size)
clReleaseCommandQueue(cq);
clReleaseContext(context);
FILE * fp;
clReleaseMemObject(ref_buf);
clReleaseMemObject(ref_tmp);
clReleaseMemObject(misc);
}
const char * const_src[] = {cl_src};
cl_int call_sort(cl_mem vals, cl_mem index, int no)
{
int i;
int j;
float todel[100];
call_copy(ref_tmp, vals, no); // used as a copier
call_reset_sort();
for (i = 0; i < no; i++)
{
call_max(ref_tmp, misc, no);
call_sort_fill(ref_tmp, index);
}
}
cl_int call_reset_sort()
{
const size_t worksize = 1;
(cl_context_properties)platform, 0};
clSetKernelArg(k_reset_sort, 0, sizeof(misc), &misc);
cq = clCreateCommandQueue(context, device, 0, &error);
error = clEnqueueNDRangeKernel(cq, k_reset_sort, 1, NULL, &worksize, NULL,
0, NULL, NULL);
error = clFinish(cq);
}
fp = fopen(CL_FILE_NAME, "r");
cl_int call_sort_fill(cl_mem vals, cl_mem index)
{
const size_t worksize = 1;
clSetKernelArg(k_sort_fill, 0, sizeof(vals), &vals);
clSetKernelArg(k_sort_fill, 1, sizeof(misc), &misc);
clSetKernelArg(k_sort_fill, 2, sizeof(index), &index);
clCreateProgramWithSource(context, 1, const_src, &cl_src_size, &error);
error = clEnqueueNDRangeKernel(cq, k_sort_fill, 1, NULL, &worksize, NULL, 0,
NULL, NULL);
error = clFinish(cq);
}
if (error != CL_SUCCESS)
cl_int call_max(cl_mem vals, cl_mem ret, int no)
{
const size_t worksize2d[] = {no, no};
const size_t worksize = no;
puts("CL: RT compile error(parallel.cl)");
call_zero_mem(ref_buf, no);
}
clSetKernelArg(k_find_maxA, 0, sizeof(vals), &vals);
clSetKernelArg(k_find_maxA, 1, sizeof(ref_buf), &ref_buf);
k_copy = clCreateKernel(prog, KERNEL_COPY, &error);
error = clEnqueueNDRangeKernel(cq, k_find_maxA, 2, NULL, worksize2d, NULL,
0, NULL, NULL);
error = clFinish(cq);
k_find_maxA = clCreateKernel(prog, KERNEL2D_FIND_MAXA, &error);
clSetKernelArg(k_find_maxB, 0, sizeof(ref_buf), &ref_buf);
clSetKernelArg(k_find_maxB, 1, sizeof(ret), &ret);
k_sort_fill = clCreateKernel(prog, SKERNEL_SORT_FILL, &error);
error = clEnqueueNDRangeKernel(cq, k_find_maxB, 1, NULL, &worksize, NULL, 0,
NULL, NULL);
error = clFinish(cq);
}
cl_int call_zero_mem(cl_mem vals, int no)
{
const size_t worksize = no;
sizeof(float) * buf_size, NULL, &error);
clSetKernelArg(k_zero_mem, 0, sizeof(vals), &vals);
sizeof(float) * buf_size, NULL, &error);
error = clEnqueueNDRangeKernel(cq, k_zero_mem, 1, NULL, &worksize, NULL, 0,
NULL, NULL);
error = clFinish(cq);
}
NULL, &error);
cl_int call_copy(cl_mem a, cl_mem b, int no)
{
const size_t worksize = no;
clSetKernelArg(k_copy, 0, sizeof(a), &a);
clSetKernelArg(k_copy, 1, sizeof(b), &b);
{
error = clEnqueueNDRangeKernel(cq, k_copy, 1, NULL, &worksize, NULL, 0,
NULL, NULL);
error = clFinish(cq);
}
clReleaseKernel(k_zero_mem);
cl_int read_from_GPURAM(cl_mem mem, float * a, int n)
{
error = clEnqueueReadBuffer(cq, mem, CL_TRUE, 0, sizeof(float) * n, a, 0,
NULL, NULL);
error = clFinish(cq);
clReleaseKernel(k_find_maxB);
return error;
}
clReleaseKernel(k_reset_sort);
cl_mem put_to_GPURAM(float * a, int n)
{
cl_mem mem;
clReleaseCommandQueue(cq);
mem = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(float) * n, NULL,
&error);
error = clEnqueueWriteBuffer(cq, mem, CL_TRUE, 0, sizeof(float) * n, a, 0,
NULL, NULL);
return mem;
}
clReleaseMemObject(ref_tmp);
cl_mem only_alloc_mem(int n)
{
cl_mem mem;
}
mem = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(float) * n, NULL,
&error);
cl_int call_sort(cl_mem vals, cl_mem index, int no)
return mem;
}
int i;
update:
I have found out this problem maybe cause by some un-readable charter. It's ok after redo change line for each line.
Oh dear. I thought that was fixed a long time ago... :-[