#include "mat_mul.h"

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

#define TS  64
#define WPT 32
#define MAX_DEV 4

#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      [MAX_DEV]   ;
static cl_context       context                 ;
static cl_command_queue queue       [MAX_DEV]   ;
static cl_program       program                 ;
static cl_kernel        kernel      [MAX_DEV]   ;
static cl_mem           a_d         [MAX_DEV]   ;
static cl_mem           b_d         [MAX_DEV]   ;
static cl_mem           c_d         [MAX_DEV]   ;

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

static int ndev ; 
static int M_str[MAX_DEV];
static int M_siz[MAX_DEV];

int func_align(int A, int B){

    return  ( (A + B - 1) / B) * B ;

  //int C ;
  //C = A % B   ;
  //if (C == 0) return  A   ;
  //else {
  //    return (A / B + 1) * B  ;
  //}
}

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;

  for (int i = 0 ; i < ndev ; i++){  
      // Setup kernel arguments
      err = clSetKernelArg(kernel[i], 0, sizeof(cl_mem), &a_d[i]); 
      CHECK_ERROR(err);
      err = clSetKernelArg(kernel[i], 1, sizeof(cl_mem), &b_d[i]);
      CHECK_ERROR(err);
      err = clSetKernelArg(kernel[i], 2, sizeof(cl_mem), &c_d[i]);
      CHECK_ERROR(err);
      err = clSetKernelArg(kernel[i], 3, sizeof(int), &M_siz[i]);
      CHECK_ERROR(err);
      err = clSetKernelArg(kernel[i], 4, sizeof(int), &N);
      CHECK_ERROR(err);
      err = clSetKernelArg(kernel[i], 5, sizeof(int), &K);
      CHECK_ERROR(err);
    
      // Setup global work size and local work size
      size_t gws[2] = {(size_t)func_align(M_siz[i], TS)/WPT, (size_t)func_align(N, TS)}, lws[2] = {TS/WPT, TS};  // TODO Optimize POINT
        //  SW ---> Optimize nornally, then optimze kernel
        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[i], kernel[i], 2, NULL, gws, lws, 0, NULL, NULL); // Run at GPU.  kernel : program
      CHECK_ERROR(err);
  }
  for (int i = 0 ; i < ndev ; i++){
    
      // DO NOT REMOVE; NEEDED FOR TIME MEASURE
      err = clFinish(queue[i]);
      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, ndev, device, "", NULL, NULL);
  if (err == CL_BUILD_PROGRAM_FAILURE) {
      for (int i = 0 ; i < ndev ; i++){
        size_t log_size;
        CHECK_ERROR(clGetProgramBuildInfo(program, device[i], CL_PROGRAM_BUILD_LOG, 0, NULL, &log_size));
        char *log = (char*)malloc(log_size + 1);
        CHECK_ERROR(clGetProgramBuildInfo(program, device[i], 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) {
  int i;
  
  // 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, 0, NULL, (unsigned int*) &ndev);
  CHECK_ERROR(err);
  err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, ndev, device, NULL);
  CHECK_ERROR(err);

//ndev = 4;
  //    First Device Job
  M_str [0] =   0           ;
  M_siz [0] =   M / ndev    ;

  //    Assign Remained Device
  for (i = 1 ; i < ndev ; i++){
    M_str[i] = M_str[i - 1] + M_siz[i - 1];

    if (i == ndev - 1) {    //  Last Device
        M_siz[i] = M - M_str[i] ;
    } else {
        M_siz[i] = M / ndev                   ;
    }
  }
  for (i = 0 ; i < ndev ; i++){
    print_device_info(device[i]);
  }

  // Create OpenCL context
  context = clCreateContext(NULL, ndev, device, NULL, NULL, &err);
  CHECK_ERROR(err);

  // Create OpenCL command queue
  for (i = 0 ; i < ndev ; i++) {
    queue[i] = clCreateCommandQueue(context, device[i], 0, &err);
    CHECK_ERROR(err);
  }

  // Compile program from "kernel.cl"
  program = create_and_build_program_with_source(context, device, "kernel.cl");

  // Extract kernel from compiled program
  for (i = 0; i < ndev ; i++){
    kernel[i] = clCreateKernel(program, "sgemm", &err); 
    CHECK_ERROR(err);

  // Create GPU buffers
    a_d[i] = clCreateBuffer(context, CL_MEM_READ_WRITE, M_siz[i] * K * sizeof(float), NULL, &err);
    CHECK_ERROR(err);
    b_d[i] = clCreateBuffer(context, CL_MEM_READ_WRITE, K        * N * sizeof(float), NULL, &err);
    CHECK_ERROR(err);
    c_d[i] = clCreateBuffer(context, CL_MEM_READ_WRITE, M_siz[i] * N * sizeof(float), NULL, &err);
    CHECK_ERROR(err);

  // Write to GPU; A (cpu) -> a_d (gpu), B (cpu) -> b_d (gpu)
    err = clEnqueueWriteBuffer(queue[i], a_d[i], CL_TRUE, 0, M_siz[i] * K * sizeof(float), &A[M_str[i] * K], 0, NULL, NULL);
    CHECK_ERROR(err);
    err = clEnqueueWriteBuffer(queue[i], b_d[i], CL_TRUE, 0, K        * N * sizeof(float), B,                0, NULL, NULL);
    CHECK_ERROR(err);

  // DO NOT REMOVE; NEEDED FOR TIME MEASURE
    err = clFinish(queue[i]);
    CHECK_ERROR(err);
  }
}

void mat_mul_final(float *A, float *B, float *C, int M, int N, int K) {
  // Read from GPU; c_d (gpu) -> C (cpu)
  for (int i = 0 ; i < ndev ; i++) {
    err = clEnqueueReadBuffer(queue[i], c_d[i], CL_TRUE, 0, M_siz[i] * N * sizeof(float), &C[M_str[i] * N], 0, NULL, NULL);
    CHECK_ERROR(err);
  }

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

}