#include "mat_mul.h"

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

#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[MAX_DEV];
static cl_kernel kernel[MAX_DEV];
static cl_mem a_d[MAX_DEV], b_d[MAX_DEV], c_d[MAX_DEV];

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

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 j = 0; j < ndev; j++){
      int delta = j == (ndev -1) ? (M -( M/ndev*(ndev -1))) : M/ndev;
      // Setup kernel arguments
      err = clSetKernelArg(kernel[j], 0, sizeof(cl_mem), &a_d[j]);
      CHECK_ERROR(err);
      err = clSetKernelArg(kernel[j], 1, sizeof(cl_mem), &b_d[j]);
      CHECK_ERROR(err);
      err = clSetKernelArg(kernel[j], 2, sizeof(cl_mem), &c_d[j]);
      CHECK_ERROR(err);
      err = clSetKernelArg(kernel[j], 3, sizeof(int), &delta);
      CHECK_ERROR(err);
      err = clSetKernelArg(kernel[j], 4, sizeof(int), &N);
      CHECK_ERROR(err);
      err = clSetKernelArg(kernel[j], 5, sizeof(int), &K);
      CHECK_ERROR(err);

      // Setup global work size and local work size
      size_t gws[2] = {(size_t)(delta+7)/8, (size_t)N}, lws[2] = {4,32};
      
      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[j], kernel[j], 2, NULL, gws, lws, 0, NULL, NULL);
      CHECK_ERROR(err);
  }

  for(int j=0; j<ndev; j++){
      // DO NOT REMOVE; NEEDED FOR TIME MEASURE
      err = clFinish(queue[j]);
      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) {
  int j, delta;
  
  // 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);
  for(j=0; j< ndev; j++){
  print_device_info(device[j]);
  }
  // Create OpenCL context
  context = clCreateContext(NULL, ndev, device, NULL, NULL, &err);
  CHECK_ERROR(err);
  
  for(j=0; j< ndev; j++){
      // Create OpenCL command queue
      queue[j] = clCreateCommandQueue(context, device[j], 0, &err);
      CHECK_ERROR(err);
  
      // Compile program from "kernel.cl"
      program[j] = create_and_build_program_with_source(context, device[j], "kernel.cl");
  
      // Extract kernel from compiled program
      kernel[j] = clCreateKernel(program[j], "sgemm", &err); 
      CHECK_ERROR(err);
  
      delta = j == (ndev -1) ?(M - (M/ndev*(ndev-1))) : M/ndev;
      // Create GPU buffers
      a_d[j] = clCreateBuffer(context, CL_MEM_READ_WRITE, delta * K * sizeof(float), NULL, &err);
      CHECK_ERROR(err);
      b_d[j] = clCreateBuffer(context, CL_MEM_READ_WRITE, K * N * sizeof(float), NULL, &err);
      CHECK_ERROR(err);
      c_d[j] = clCreateBuffer(context, CL_MEM_READ_WRITE, delta * N * sizeof(float), NULL, &err);
      CHECK_ERROR(err);
  
      // Write to GPU; A (cpu) -> a_d (gpu), B (cpu) -> b_d (gpu)
      int offset = M/ndev*j*K;
      err = clEnqueueWriteBuffer(queue[j], a_d[j], CL_FALSE, 0 , delta*K*sizeof(float), &A[offset], 0, NULL, NULL);
      CHECK_ERROR(err);
  
      err = clEnqueueWriteBuffer(queue[j], b_d[j], CL_FALSE, 0, K * N * sizeof(float), B, 0, NULL, NULL);
      CHECK_ERROR(err);
  }

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

void mat_mul_final(float *A, float *B, float *C, int M, int N, int K) {
   int j, delta, offset;
  // Read from GPU; c_d (gpu) -> C (cpu)
  for(j=0; j<ndev; j++){
      delta = (j == ndev -1) ? (M - (M/ndev*(ndev-1))) : M/ndev;
      offset = M/ndev*j*N;

      err = clEnqueueReadBuffer(queue[j], c_d[j], CL_TRUE, 0, delta * N * sizeof(float), &C[offset], 0, NULL, NULL);
      CHECK_ERROR(err);
  }

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