#include "mat_mul.h"

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

#define CHECK_ERROR(err) \
  if (err != CL_SUCCESS) { \
    printf("[%s:%d] OpenCL error %d\n", __FILE__, __LINE__, err); \
    exit(EXIT_FAILURE); \
  }

static cl_int err;
static cl_platform_id platform;
//static cl_device_id device;
static cl_device_id devices[4];
//static cl_context context;
static cl_context contexts[4];
//static cl_command_queue queue;
static cl_command_queue queues[4];
//static cl_program program;
static cl_program programs[4];
//static cl_kernel kernel;
static cl_kernel kernels[4];
//static cl_mem a_d, b_d, c_d;
static cl_mem a_d[4], b_d[4], c_d[4];

static float *A, *B, *C;
static int M, N, K;
static int CORESIZE;


void mat_mul(float *_A, float *_B, float *_C, int _M, int _N, int _K) {
  A = _A, B = _B, C = _C;
  M = _M, N = _N, K = _K;

  // Split M for Multicore
  int M_split = (M / CORESIZE) + ((M % CORESIZE) > 0);

  // Setup kernel arguments
  //err = clSetKernelArg(kernel, 0, sizeof(cl_mem), &a_d);
  //CHECK_ERROR(err);
  //err = clSetKernelArg(kernel, 1, sizeof(cl_mem), &b_d);
  //CHECK_ERROR(err);
  //err = clSetKernelArg(kernel, 2, sizeof(cl_mem), &c_d);
  //CHECK_ERROR(err);
  //err = clSetKernelArg(kernel, 3, sizeof(int), &M);
  //CHECK_ERROR(err);
  //err = clSetKernelArg(kernel, 4, sizeof(int), &N);
  //CHECK_ERROR(err);
  //err = clSetKernelArg(kernel, 5, sizeof(int), &K);
  //CHECK_ERROR(err);
  for (int d = 0; d < CORESIZE; d++) {
    err = clSetKernelArg(kernels[d], 0, sizeof(cl_mem), &a_d[d]);
    CHECK_ERROR(err);
    err = clSetKernelArg(kernels[d], 1, sizeof(cl_mem), &b_d[d]);
    CHECK_ERROR(err);
    err = clSetKernelArg(kernels[d], 2, sizeof(cl_mem), &c_d[d]);
    CHECK_ERROR(err);
    err = clSetKernelArg(kernels[d], 3, sizeof(int), &M_split);
    CHECK_ERROR(err);
    err = clSetKernelArg(kernels[d], 4, sizeof(int), &N);
    CHECK_ERROR(err);
    err = clSetKernelArg(kernels[d], 5, sizeof(int), &K);
    CHECK_ERROR(err);
  }

  // Setup global work size and local work size
  //size_t gws[2] = {(size_t)M, (size_t)N}, lws[2] = {1, 1};
  size_t gws[2] = {(size_t)M_split, (size_t)N}, lws[2] = {2, 128};
  for (int i = 0; i < 2; ++i) {
    // By OpenCL spec, global work size should be MULTIPLE of local work size
    // Formula below achieve it
    // e.g., gws = 25, lws = 16, then (25 + 16 - 1) / 16 * 16 = 40 / 16 * 16 = 2 * 16 = 32
    gws[i] = (gws[i] + lws[i] - 1) / lws[i] * lws[i];
  }

  // Run kernel
  //err = clEnqueueNDRangeKernel(queue, kernel, 2, NULL, gws, lws, 0, NULL, NULL);
  //CHECK_ERROR(err);
  for (int d = 0; d < CORESIZE; d++) {
    err = clEnqueueNDRangeKernel(queues[d], kernels[d], 2, NULL, gws, lws, 0, NULL, NULL);
    CHECK_ERROR(err);
  }

  // DO NOT REMOVE; NEEDED FOR TIME MEASURE
  //err = clFinish(queue);
  //CHECK_ERROR(err);
  for (int d = 0; d < CORESIZE; d++) {
    err = clFinish(queues[d]);
    CHECK_ERROR(err);
  }
}

static void print_platform_info(cl_platform_id platform) {
  size_t sz;
  char *buf;
  CHECK_ERROR(clGetPlatformInfo(platform, CL_PLATFORM_NAME, 0, NULL, &sz));
  buf = (char*)malloc(sz);
  CHECK_ERROR(clGetPlatformInfo(platform, CL_PLATFORM_NAME, sz, buf, NULL));
  printf("Detected OpenCL platform: %s\n", buf);
  free(buf);
}

static void print_device_info(cl_device_id device) {
  size_t sz;
  char *buf;
  CHECK_ERROR(clGetDeviceInfo(device, CL_DEVICE_NAME, 0, NULL, &sz));
  buf = (char*)malloc(sz);
  CHECK_ERROR(clGetDeviceInfo(device, CL_DEVICE_NAME, sz, buf, NULL));
  printf("Detected OpenCL device: %s\n", buf);
  free(buf);
}

static cl_program create_and_build_program_with_source(cl_context context, cl_device_id device, const char *file_name) {
  FILE *file = fopen(file_name, "rb");
  if (file == NULL) {
    printf("Failed to open %s\n", file_name);
    exit(EXIT_FAILURE);
  }
  fseek(file, 0, SEEK_END);
  size_t source_size = ftell(file);
  rewind(file);
  char *source_code = (char*)malloc(source_size + 1);
  size_t ntotal = 0;
  while (ntotal < source_size) {
    int nread = fread(source_code, sizeof(char), source_size, file);
    ntotal += nread;
  }
  source_code[source_size] = '\0';
  fclose(file);
  cl_program program = clCreateProgramWithSource(context, 1, (const char **)&source_code, &source_size, &err);
  CHECK_ERROR(err);
  free(source_code);
  err = clBuildProgram(program, 1, &device, "", NULL, NULL);
  if (err == CL_BUILD_PROGRAM_FAILURE) {
    size_t log_size;
    CHECK_ERROR(clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, 0, NULL, &log_size));
    char *log = (char*)malloc(log_size + 1);
    CHECK_ERROR(clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, log_size, log, NULL));
    log[log_size] = 0;
    printf("Compile error:\n%s\n", log);
    free(log);
  }
  CHECK_ERROR(err);
  return program;
}

void mat_mul_init(float *A, float *B, float *C, int M, int N, int K) {

  // Set Core Size
  CORESIZE = 4;

  // Get OpenCL platform
  err = clGetPlatformIDs(1, &platform, NULL);
  CHECK_ERROR(err);
  print_platform_info(platform);

  // Get OpenCL device
  //err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL); // Device 1
  err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, CORESIZE, devices, NULL); // Multi device
  CHECK_ERROR(err);
  //print_device_info(device);
  for (int d = 0; d < CORESIZE; d++) {
    print_device_info(devices[d]);
  }

  // Create OpenCL context
  //context = clCreateContext(NULL, 1, &device, NULL, NULL, &err);
  //context = clCreateContext(NULL, CORESIZE, devices, NULL, NULL, &err); // Use 1 context
  for (int d = 0; d < CORESIZE; d++) {
    // Create OpenCL context
    contexts[d] = clCreateContext(NULL, 1, &devices[d], NULL, NULL, &err);
    CHECK_ERROR(err);
  }

  // Create OpenCL command queue
  //queue = clCreateCommandQueue(context, device, 0, &err);
  //CHECK_ERROR(err);
  for (int d = 0; d < CORESIZE; d++) {
    queues[d] = clCreateCommandQueue(contexts[d], devices[d], 0, &err);
    CHECK_ERROR(err);
  }

  // Compile program from "kernel.cl"
  //program = create_and_build_program_with_source(context, device, "kernel.cl");
  for (int d = 0; d < CORESIZE; d++) {
    programs[d] = create_and_build_program_with_source(contexts[d], devices[d], "kernel.cl");
  }

  // Extract kernel from compiled program
  //kernel = clCreateKernel(program, "sgemm", &err); 
  //CHECK_ERROR(err);
  for (int d = 0; d < CORESIZE; d++) {
    kernels[d] = clCreateKernel(programs[d], "sgemm", &err); 
    CHECK_ERROR(err);
  }

  // Split M for Multicore
  int M_split = (M / CORESIZE) + ((M % CORESIZE) > 0);

  // Create GPU buffers
  //a_d = clCreateBuffer(context, CL_MEM_READ_WRITE, M * K * sizeof(float), NULL, &err);
  //CHECK_ERROR(err);
  //b_d = clCreateBuffer(context, CL_MEM_READ_WRITE, M * K * sizeof(float), NULL, &err);
  //CHECK_ERROR(err);
  //c_d = clCreateBuffer(context, CL_MEM_READ_WRITE, M * N * sizeof(float), NULL, &err);
  //CHECK_ERROR(err);
  for (int d = 0; d < CORESIZE; d++) {
    a_d[d] = clCreateBuffer(contexts[d], CL_MEM_READ_ONLY, M_split * K * sizeof(float), NULL, &err);
    CHECK_ERROR(err);
    b_d[d] = clCreateBuffer(contexts[d], CL_MEM_READ_ONLY, K * N * sizeof(float), NULL, &err);
    CHECK_ERROR(err);
    c_d[d] = clCreateBuffer(contexts[d], CL_MEM_READ_WRITE, M_split * N * sizeof(float), NULL, &err);
    CHECK_ERROR(err);
  }

  // Write to GPU; A (cpu) -> a_d (gpu), B (cpu) -> b_d (gpu)
  //err = clEnqueueWriteBuffer(queue, a_d, CL_TRUE, 0, M * K * sizeof(float), A, 0, NULL, NULL);
  //CHECK_ERROR(err);
  //err = clEnqueueWriteBuffer(queue, b_d, CL_TRUE, 0, K * N * sizeof(float), B, 0, NULL, NULL);
  //CHECK_ERROR(err);
  for (int d = 0; d < CORESIZE; d++) {
    err = clEnqueueWriteBuffer(queues[d], a_d[d], CL_TRUE, 0, M_split * K * sizeof(float), &A[d * M_split * K], 0, NULL, NULL);
    CHECK_ERROR(err);
    err = clEnqueueWriteBuffer(queues[d], b_d[d], CL_TRUE, 0, K * N * sizeof(float), B, 0, NULL, NULL);
    CHECK_ERROR(err);
  }

  // DO NOT REMOVE; NEEDED FOR TIME MEASURE
  //err = clFinish(queue);
  //CHECK_ERROR(err);
  for (int d = 0; d < CORESIZE; d++) {
    err = clFinish(queues[d]);
    CHECK_ERROR(err);
  }
}

void mat_mul_final(float *A, float *B, float *C, int M, int N, int K) {
  // Split M for Multicore
  int M_split = (M / CORESIZE) + ((M % CORESIZE) > 0);

  // Read from GPU; c_d (gpu) -> C (cpu)
  //err = clEnqueueReadBuffer(queue, c_d, CL_TRUE, 0, M * N * sizeof(float), C, 0, NULL, NULL);
  //CHECK_ERROR(err);
  for (int d = 0; d < CORESIZE; d++) {
    if (d == CORESIZE - 1) {
      int M_last = M - (M_split * d);
      err = clEnqueueReadBuffer(queues[d], c_d[d], CL_TRUE, 0, M_last * N * sizeof(float), &C[d * M_split * N], 0, NULL, NULL);
    }
    else {
      err = clEnqueueReadBuffer(queues[d], c_d[d], CL_TRUE, 0, M_split * N * sizeof(float), &C[d * M_split * N], 0, NULL, NULL);
    }
    CHECK_ERROR(err);
  }

  // DO NOT REMOVE; NEEDED FOR TIME MEASURE
  //err = clFinish(queue);
  //CHECK_ERROR(err);
  for (int d = 0; d < CORESIZE; d++) {
    err = clFinish(queues[d]);
    CHECK_ERROR(err);
  }
}