CMakeListx.txt file for OpenCL on OS X - ios

Question

I would like to make as automatic as possible the compilation and linking of my code projects using OpenCL on OS X, I know how to do it for C++ but I am experiencing problems for OpenCL. This is the code that I am using as an example:
main.cpp:
#include <stdio.h>
#include <stdlib.h>
#ifdef __APPLE__ //Mac OSX has a different name for the header file
#include <OpenCL/opencl.h>
#else
#include <CL/cl.h>
#endif
#define MEM_SIZE (128)//suppose we have a vector with 128 elements
#define MAX_SOURCE_SIZE (0x100000)
int main()
{
//In general Intel CPU and NV/AMD's GPU are in different platforms
//But in Mac OSX, all the OpenCL devices are in the platform "Apple"
cl_platform_id platform_id = NULL;
cl_device_id device_id = NULL;
cl_context context = NULL;
cl_command_queue command_queue = NULL; //"stream" in CUDA
cl_mem memobj = NULL;//device memory
cl_program program = NULL; //cl_prgram is a program executable created from the source or binary
cl_kernel kernel = NULL; //kernel function
cl_uint ret_num_devices;
cl_uint ret_num_platforms;
cl_int ret; //accepts return values for APIs
float mem[MEM_SIZE]; //alloc memory on host(CPU) ram
//OpenCL source can be placed in the source code as text strings or read from another file.
FILE *fp;
const char fileName[] = "./kernel.cl";
size_t source_size;
char *source_str;
cl_int i;
// read the kernel file into ram
fp = fopen(fileName, "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 );
//initialize the mem with 1,2,3...,n
for( i = 0; i < MEM_SIZE; i++ ) {
mem[i] = i;
}
//get the 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 context on the specified device
context = clCreateContext( NULL, 1, &device_id, NULL, NULL, &ret);
//create the command_queue (stream)
command_queue = clCreateCommandQueue(context, device_id, 0, &ret);
//alloc mem on the device with the read/write flag
memobj = clCreateBuffer(context, CL_MEM_READ_WRITE, MEM_SIZE * sizeof(float), NULL, &ret);
//copy the memory from host to device, CL_TRUE means blocking write/read
ret = clEnqueueWriteBuffer(command_queue, memobj, CL_TRUE, 0, MEM_SIZE * sizeof(float), mem, 0, NULL, NULL);
//create a program object for a context
//load the source code specified by the text strings into the program object
program = clCreateProgramWithSource(context, 1, (const char **)&source_str, (const size_t *)&source_size, &ret);
//build (compiles and links) a program executable from the program source or binary
ret = clBuildProgram(program, 1, &device_id, NULL, NULL, NULL);
//create a kernel object with specified name
kernel = clCreateKernel(program, "vecAdd", &ret);
//set the argument value for a specific argument of a kernel
ret = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&memobj);
//define the global size and local size (grid size and block size in CUDA)
size_t global_work_size[3] = {MEM_SIZE, 0, 0};
size_t local_work_size[3] = {MEM_SIZE, 0, 0};
//Enqueue a command to execute a kernel on a device ("1" indicates 1-dim work)
ret = clEnqueueNDRangeKernel(command_queue, kernel, 1, NULL, global_work_size, local_work_size, 0, NULL, NULL);
//copy memory from device to host
ret = clEnqueueReadBuffer(command_queue, memobj, CL_TRUE, 0, MEM_SIZE * sizeof(float), mem, 0, NULL, NULL);
//print out the result
for(i=0; i<MEM_SIZE; i++) {
printf("mem[%d] : %.2f\n", i, mem[i]);
}
//clFlush only guarantees that all queued commands to command_queue get issued to the appropriate device
//There is no guarantee that they will be complete after clFlush returns
ret = clFlush(command_queue);
//clFinish blocks until all previously queued OpenCL commands in command_queue are issued to the associated device and have completed.
ret = clFinish(command_queue);
ret = clReleaseKernel(kernel);
ret = clReleaseProgram(program);
ret = clReleaseMemObject(memobj);//free memory on device
ret = clReleaseCommandQueue(command_queue);
ret = clReleaseContext(context);
free(source_str);//free memory on host
return 0;
}
kernel.cl:
__kernel void vecAdd(__global float* a)
{
int gid = get_global_id(0);// in CUDA = blockIdx.x * blockDim.x + threadIdx.x
a[gid] += a[gid];
}
and this is my CMakelists.txt so far:
#Minimal OpenCL CMakeLists.txt by StreamHPC
cmake_minimum_required (VERSION 3.1)
project(GreatProject)
# Handle OpenCL
find_package(OpenCL REQUIRED)
include_directories(${OpenCL_INCLUDE_DIRS})
link_directories(${OpenCL_LIBRARY})
add_executable (main main.cpp)
target_include_directories (main PUBLIC ${CMAKE_CURRENT_SOURCE_DIR})
target_link_libraries (main ${OpenCL_LIBRARY})
Apparently it compiles, but when I run the executable I get the error:
Failed to load kernel.
I compiled successfully the code by hand following this answer, but my project is willing to have various kernels and various C++ files and headers, therefore I would like to use CMake in order to automatize the compilation of the project.
How should I modify my CMakeLists.txt script?
NOTE:
I guess that the file kernel.cl is not being compiled, I don't what is a proper way to guarantee having a CMakeLists.txt that always compiles all the *.cl files in the project directory in addition to all the *.cpp. Would be even better if it is posible to linking agains MKL.

Answer 1

For mac opencl is used as a framework you need to do the following to link libraries from the framework.
cmake_minimum_required (VERSION 2.6)
project (montecarlo_cl)
find_package(OpenCL REQUIRED)
include_directories( ${OPENCL_INCLUDE_DIR})
set (montecarlo_cl_src montecarlo_ocl.c)
add_executable (montecarlo_cl ${montecarlo_cl_src})
target_link_libraries(montecarlo_cl "-framework OpenCL" )