#include "mat_mul.h"

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

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

#define TS 32
#define WPT 8
#define WIDTH 8
#define MAX_DEV 4
#define RTS (TS/WPT)

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], b_d[MAX_DEV], c_d[MAX_DEV];

//static cl_uint num_plaforms;

static float *A, *B, *C;
static int M, N, K;
static int ndev;
static int CopyBuf = 0;

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;

  // Setup kernel arguments
  for(int i = 0; i < ndev; i++) {
    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);
    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)M/ndev, (size_t)N/WIDTH}, lws[2] = {TS, TS/WIDTH};
  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
    //printf("1.gws[%d]:%ld, lws[%d]:%ld\n", i, gws[i], i, lws[i]);
    gws[i] = (gws[i] + lws[i] - 1) / lws[i] * lws[i];
    //printf("2.gws[%d]:%ld, lws[%d]:%ld\n", i, gws[i], i, lws[i]);
  }

  // Run kernel
  for(int i = 0; i < ndev; i++){
    err = clEnqueueNDRangeKernel(queue[i], kernel[i], 2, NULL, gws, lws, 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);
  }
}

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);
  }
//    printf("Function : %s, Line : %d\n", __FUNCTION__, __LINE__); fflush(stdout);
  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) {
  // 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, NULL, (unsigned int*) &ndev);
  err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, ndev, device, NULL);
  CHECK_ERROR(err);
  for(int i = 0; i < ndev; i++){
    printf("GPU %d ", 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(int i = 0; i < ndev; i++){
    queue[i] = clCreateCommandQueue(context, device[i], 0, &err);
    CHECK_ERROR(err);
  }

  for(int i = 0; i < ndev; i++){
  // Compile program from "kernel.cl"
    program = create_and_build_program_with_source(context, device[i], "kernel.cl");
  // Extract kernel from compiled program
    kernel[i] = clCreateKernel(program, "sgemm", &err); 
    CHECK_ERROR(err);
  }
/////////////////////////////////////////////
  M = (_M & (TS-1))? _M + (TS - (_M & (TS-1))) : _M;
  N = (_N & (TS-1))? _N + (TS - (_N & (TS-1))) : _N;
  K = (_K & (TS-1))? _K + (TS - (_K & (TS-1))) : _K;
  CopyBuf = 0;
  if(M == _M && K == _K){
    A = _A;
  }
  else{
    alloc_mat(&A, M, K);

    for (int i = 0; i < M; i++) { 
      for (int j = 0; j < K; j++) {
        if(i<_M && j<_K)
          A[i * K + j] = _A[i * _K + j];
        else
          A[i * K + j] = 0;
      }
    }
    CopyBuf = 1;
  }

  if(K == _K && N == _N){
    B = _B;
  }
  else{
    alloc_mat(&B, K, N);

    for (int i = 0; i < K; i++) { 
      for (int j = 0; j < N; j++) {
        if(i<_K && j<_N)
          B[i * N + j] = _B[i * _N + j];
        else
          B[i * N + j] = 0;
      }
    }
    CopyBuf = 1;
  }
  
  if(M == _M && N == _N){
    C = _C;
  }
  else{
    alloc_mat(&C, M, N);
    zero_mat(C, M, N);
    CopyBuf = 1;
  }
////////////////////////////////////

  int size_A = M * K * sizeof(float)/ndev, size_C = M * N * sizeof(float)/ndev;

  for(int i = 0; i < ndev; i++){
    if(i == (ndev-1)){
      size_A = M * K * sizeof(float) - size_A * i;
      size_C = M * N * sizeof(float) - size_C * i;
    }
  // Create GPU buffers
    a_d[i] = clCreateBuffer(context, CL_MEM_READ_WRITE, size_A, 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, size_C, 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, size_A, &A[i*M*K/ndev], 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
  for(int i = 0; i < ndev; i++){
    err = clFinish(queue[i]);
    CHECK_ERROR(err);
  }
}

void mat_mul_final(float *_A, float *_B, float *_C, int _M, int _N, int _K) {
  int size_C = M * N * sizeof(float)/ndev;

  // Read from GPU; c_d (gpu) -> C (cpu)
  for(int i = 0; i < ndev; i++){
    if(i == (ndev-1)){
      size_C = M * N * sizeof(float) - size_C * i;
    }    
    err = clEnqueueReadBuffer(queue[i], c_d[i], CL_TRUE, 0, size_C, &C[i*M*N/ndev], 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);
  }
  
  if(CopyBuf == 1){
    for (int i = 0; i < _M; i++) { 
      for (int j = 0; j < _N; j++) {
          _C[i * _N + j] = C[i * N + j];
      }
    }
  }
}