#define CL_TARGET_OPENCL_VERSION 300

#include <CL/cl.h>
#include <stdio.h>

int main()
{
	// Connect to a compute device
	//
	int gpu = 1;
	cl_device_id deviceID;
	cl_int err = clGetDeviceIDs(NULL, gpu ? CL_DEVICE_TYPE_GPU : CL_DEVICE_TYPE_CPU, 1, &deviceID, NULL);
	if (err != CL_SUCCESS) {
		printf("Error: Failed to create a device group!\n");
		return 1;
	}

	FILE *file = fopen("src/test.cl", "r");
	fseek(file, 0, SEEK_END);
	size_t length = ftell(file);
	fseek(file, 0, SEEK_SET);
	char *buffer = calloc(1, length + 1);
	fread(buffer, length, 1, file);

	printf("%s", buffer);

	cl_context context = clCreateContext(NULL, 1, &deviceID, NULL, NULL, &err);
	if(!context) {
		printf("Failed to create context\n");
		return 1;
	}

	cl_command_queue queue = clCreateCommandQueueWithProperties(context, deviceID, NULL, &err);
	if(!queue) {
		printf("Failed to create command queue\n");
		return 1;
	}

	cl_program program = clCreateProgramWithSource(context, 1, (const char **) &buffer, &length, &err);
	if(!program) {
		printf("Failed to create program\n");
		return 1;
	}

	err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
	if(err != CL_SUCCESS) {
		printf("Failed to build program\n");
		// clGetProgramBuildInfo...
		return 1;
	}

	cl_kernel kernel = clCreateKernel(program, "do_stuff", &err);
	if(!kernel) {
		printf("Failed to create kernel\n");
		return 1;
	}

	unsigned int inputSize = 256000000;
	float *inputData = calloc(inputSize, sizeof(float));
	for(unsigned int i = 0; i < inputSize; i++) {
		inputData[i] = i;
	}

	cl_mem input = clCreateBuffer(context, CL_MEM_READ_ONLY, sizeof(float) * inputSize, NULL, &err);
	cl_mem output = clCreateBuffer(context, CL_MEM_WRITE_ONLY, sizeof(float) * inputSize, NULL, &err);
	if(!input || !output) {
		printf("Failed to allocate input/output buffer\n");
		return 1;
	}

	err = clEnqueueWriteBuffer(queue, input, CL_TRUE, 0, sizeof(float) * inputSize, inputData, 0, NULL, NULL);
	if(err != CL_SUCCESS) {
		printf("Failed to write to buffer\n");
		return 1;
	}

	err = 0;
	err = clSetKernelArg(kernel, 0, sizeof(input), &input);
	err |= clSetKernelArg(kernel, 1, sizeof(output), &output);
	err |= clSetKernelArg(kernel, 2, sizeof(unsigned int), &inputSize);
	if(err != CL_SUCCESS) {
		printf("Failed to set kernel args\n");
		return 1;
	}

	/*char *info = calloc(1, 1024);
	clGetProgramInfo(program, CL_PROGRAM_STRING_DEBUG_INFO, 1024, info, NULL);
	printf("INFO: %s\n", info);*/

	size_t local;
	err = clGetKernelWorkGroupInfo(kernel, deviceID, CL_KERNEL_WORK_GROUP_SIZE, sizeof(local), &local, NULL);
	if(err != CL_SUCCESS) {
		printf("Failed to get work group size\n");
		return 1;
	}

	printf("Group size is %zu\n", local);

	size_t global = inputSize;
	err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0, NULL, NULL);
	if(err != CL_SUCCESS) {
		printf("Failed to enqueue\n");
		return 1;
	}

	clFinish(queue);

	float *outputData = calloc(inputSize, sizeof(float));
	err = clEnqueueReadBuffer(queue, output, CL_TRUE, 0, sizeof(float) * inputSize, outputData, 0, NULL, NULL);
	if(err != CL_SUCCESS) {
		printf("Failed to read from buffer\n");
		return 1;
	}

	for(unsigned int i = 0; i < inputSize; i++) {
		if(i % 1000 != 0) continue;
		printf("%f: %f\n", inputData[i], outputData[i]);
	}
}