fft_fpga.c

Go to the documentation of this file.

/*----------------------------------------------------------------------------*/
/*  CP2K: A general program to perform molecular dynamics simulations         */
/*  Copyright 2000-2024 CP2K developers group <https://cp2k.org>              */
/*                                                                            */
/*  SPDX-License-Identifier: GPL-2.0-or-later                                 */
/*----------------------------------------------------------------------------*/
 
#if defined(__PW_FPGA)
 
// global dependencies
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
 
// common dependencies
#include "CL/opencl.h"
 
// local dependencies
#include "fft_fpga.h"
#include "opencl_utils.h"
 
// Function prototypes
int init();
void cleanup();
static void cleanup_program();
static void init_program(int N[3], char *data_path);
static void queue_setup();
void queue_cleanup();
static void fftfpga_run_3d(int inverse, int N[3], cmplx *c_in);
 
// --- CODE -------------------------------------------------------------------
 
int pw_fpga_initialize_() { return init(); }
 
void pw_fpga_final_() { cleanup(); }
 
/*******************************************************************************
 * \brief  check whether FFT3d can be computed on the FPGA or not. This depends
 *         on the availability of bitstreams whose sizes are for now listed here
 *         If the fft sizes are found and the FPGA is not setup before,
 *         it is done
 * \param  data_path - path to the data directory
 * \param  N - integer pointer to the size of the FFT3d
 * \retval true if fft3d size supported
 ******************************************************************************/
int pw_fpga_check_bitstream_(char *data_path, int N[3]) {
  static int fft_size[3] = {0, 0, 0};
 
  // check the supported sizes
  if ((N[0] == 16 && N[1] == 16 && N[2] == 16) ||
      (N[0] == 32 && N[1] == 32 && N[2] == 32) ||
      (N[0] == 64 && N[1] == 64 && N[2] == 64)) {
 
    // if first time
    if (fft_size[0] == 0 && fft_size[1] == 0 && fft_size[2] == 0) {
      fft_size[0] = N[0];
      fft_size[1] = N[1];
      fft_size[2] = N[2];
 
      init_program(fft_size, data_path);
    } else if (fft_size[0] == N[0] && fft_size[1] == N[1] &&
               fft_size[2] == N[2]) {
      // if same fft size as previous
      // dont do anything
    } else {
      // else if different fft size as previous
      // cleanup and initialize
      fft_size[0] = N[0];
      fft_size[1] = N[1];
      fft_size[2] = N[2];
 
      cleanup_program();
      init_program(fft_size, data_path);
    }
 
    return 1;
  } else {
    return 0;
  }
}
 
/*******************************************************************************
 * \brief   compute an in-place single precision complex 3D-FFT on the FPGA
 * \param   direction : direction - 1/forward, otherwise/backward FFT3d
 * \param   N   : integer pointer to size of FFT3d
 * \param   din : complex input/output single precision data pointer
 ******************************************************************************/
void pw_fpga_fft3d_sp_(int direction, int N[3], cmplx *din) {
  // setup device specific constructs
  if (direction == 1) {
    fftfpga_run_3d(0, N, din);
  } else {
    fftfpga_run_3d(1, N, din);
  }
}
 
/*******************************************************************************
 * \brief   compute an in-place double precision complex 3D-FFT on the FPGA
 * \param   direction : direction - 1/forward, otherwise/backward FFT3d
 * \param   N   : integer pointer to size of FFT3d
 * \param   din : complex input/output single precision data pointer
 ******************************************************************************/
void pw_fpga_fft3d_dp_(int direction, int N[3], cmplx *din) {
  // setup device specific constructs
  if (direction == 1) {
    fftfpga_run_3d(0, N, din);
  } else {
    fftfpga_run_3d(1, N, din);
  }
}
 
/*******************************************************************************
 * \brief   Execute a single precision complex FFT3d
 * \param   inverse : int
 * \param   N       : integer pointer to size of FFT3d
 * \param   din     : complex input/output single precision data pointer
 ******************************************************************************/
void fftfpga_run_3d(int inverse, int N[3], cmplx *c_in) {
  cl_int status = 0;
  int inverse_int = inverse;
  cl_kernel fft_kernel = NULL, fft_kernel_2 = NULL;
  cl_kernel fetch_kernel = NULL, transpose_kernel = NULL,
            transpose_kernel_2 = NULL;
 
  // Device memory buffers
  cl_mem d_inData, d_outData;
 
  // Create the kernel - name passed in here must match kernel name in the
  // original CL file, that was compiled into an AOCX file using the AOC tool
  fft_kernel = clCreateKernel(program, "fft3da", &status);
  checkError(status, "Failed to create fft3da kernel");
  fft_kernel_2 = clCreateKernel(program, "fft3db", &status);
  checkError(status, "Failed to create fft3db kernel");
  fetch_kernel = clCreateKernel(program, "fetch", &status);
  checkError(status, "Failed to create fetch kernel");
  transpose_kernel = clCreateKernel(program, "transpose", &status);
  checkError(status, "Failed to create transpose kernel");
  transpose_kernel_2 = clCreateKernel(program, "transpose3d", &status);
  checkError(status, "Failed to create transpose3d kernel");
 
  d_inData = clCreateBuffer(context, CL_MEM_READ_WRITE,
                            sizeof(cmplx) * N[0] * N[1] * N[2], NULL, &status);
  checkError(status, "Failed to allocate input device buffer\n");
  d_outData = clCreateBuffer(context, CL_MEM_READ_WRITE,
                             sizeof(cmplx) * N[0] * N[1] * N[2], NULL, &status);
  checkError(status, "Failed to allocate output device buffer\n");
 
  cmplx *h_inData = (cmplx *)alignedMalloc(sizeof(cmplx) * N[0] * N[1] * N[2]);
  if (h_inData == NULL) {
    printf("Unable to allocate host memory\n");
    exit(1);
  }
  cmplx *h_outData = (cmplx *)alignedMalloc(sizeof(cmplx) * N[0] * N[1] * N[2]);
  if (h_outData == NULL) {
    printf("Unable to allocate host memory\n");
    exit(1);
  }
 
  memcpy(h_inData, c_in, sizeof(cmplx) * N[0] * N[1] * N[2]);
 
  queue_setup();
 
  // Copy data from host to device
  status = clEnqueueWriteBuffer(queue6, d_inData, CL_TRUE, 0,
                                sizeof(cmplx) * N[0] * N[1] * N[2], h_inData, 0,
                                NULL, NULL);
  checkError(status, "Failed to copy data to device");
 
  status = clFinish(queue6);
  checkError(status, "failed to finish");
 
  status = clSetKernelArg(fetch_kernel, 0, sizeof(cl_mem), (void *)&d_inData);
  checkError(status, "Failed to set kernel arg 0");
  status = clSetKernelArg(fft_kernel, 0, sizeof(cl_int), (void *)&inverse_int);
  checkError(status, "Failed to set kernel arg 1");
  status =
      clSetKernelArg(transpose_kernel, 0, sizeof(cl_mem), (void *)&d_outData);
  checkError(status, "Failed to set kernel arg 2");
  status =
      clSetKernelArg(fft_kernel_2, 0, sizeof(cl_int), (void *)&inverse_int);
  checkError(status, "Failed to set kernel arg 3");
 
  status = clEnqueueTask(queue1, fetch_kernel, 0, NULL, NULL);
  checkError(status, "Failed to launch fetch kernel");
 
  // Launch the fft kernel - we launch a single work item hence enqueue a task
  status = clEnqueueTask(queue2, fft_kernel, 0, NULL, NULL);
  checkError(status, "Failed to launch fft kernel");
 
  status = clEnqueueTask(queue3, transpose_kernel, 0, NULL, NULL);
  checkError(status, "Failed to launch transpose kernel");
 
  status = clEnqueueTask(queue4, fft_kernel_2, 0, NULL, NULL);
  checkError(status, "Failed to launch second fft kernel");
 
  status = clEnqueueTask(queue5, transpose_kernel_2, 0, NULL, NULL);
  checkError(status, "Failed to launch second transpose kernel");
 
  // Wait for all command queues to complete pending events
  status = clFinish(queue1);
  checkError(status, "failed to finish");
  status = clFinish(queue2);
  checkError(status, "failed to finish");
  status = clFinish(queue3);
  checkError(status, "failed to finish");
  status = clFinish(queue4);
  checkError(status, "failed to finish");
  status = clFinish(queue5);
  checkError(status, "failed to finish");
 
  // Copy results from device to host
  status = clEnqueueReadBuffer(queue3, d_outData, CL_TRUE, 0,
                               sizeof(cmplx) * N[0] * N[1] * N[2], h_outData, 0,
                               NULL, NULL);
  checkError(status, "Failed to read data from device");
 
  memcpy(c_in, h_outData, sizeof(cmplx) * N[0] * N[1] * N[2]);
 
  queue_cleanup();
 
  if (h_outData)
    free(h_outData);
  if (h_inData)
    free(h_inData);
 
  if (d_inData)
    clReleaseMemObject(d_inData);
  if (d_outData)
    clReleaseMemObject(d_outData);
 
  if (fetch_kernel)
    clReleaseKernel(fetch_kernel);
  if (fft_kernel)
    clReleaseKernel(fft_kernel);
  if (fft_kernel_2)
    clReleaseKernel(fft_kernel_2);
  if (transpose_kernel)
    clReleaseKernel(transpose_kernel);
  if (transpose_kernel_2)
    clReleaseKernel(transpose_kernel_2);
}
 
/*******************************************************************************
 * \brief   Initialize the program - select device, create context and program
 ******************************************************************************/
void init_program(int N[3], char *data_path) {
  cl_int status = 0;
 
  // use the first device.
  device = devices[0];
 
  // Create the context.
  context = clCreateContext(NULL, 1, &device, &openCLContextCallBackFxn, NULL,
                            &status);
  checkError(status, "Failed to create context");
 
  // Create the program.
  program = getProgramWithBinary(context, &device, 1, N, data_path);
  if (program == NULL) {
    printf("Failed to create program");
    exit(1);
  }
  // Build the program that was just created.
  status = clBuildProgram(program, 0, NULL, "", NULL, NULL);
  checkError(status, "Failed to build program");
}
 
/*******************************************************************************
 * \brief   Free resources allocated during program initialization
 ******************************************************************************/
void cleanup_program() {
  if (program)
    clReleaseProgram(program);
  if (context)
    clReleaseContext(context);
}
 
/*******************************************************************************
 * \brief   Initialize the OpenCL FPGA environment - platform and devices
 * \retval  true if error in initialization
 ******************************************************************************/
int init() {
  cl_int status = 0;
 
  // Get the OpenCL platform.
  platform = findPlatform("Intel(R) FPGA");
  if (platform == NULL) {
    printf("ERROR: Unable to find Intel(R) FPGA OpenCL platform\n");
    return 1;
  }
  // Query the available OpenCL devices.
  cl_uint num_devices;
  devices = getDevices(platform, CL_DEVICE_TYPE_ALL, &num_devices);
 
  return 0;
}
 
/*******************************************************************************
 * \brief   Free resources allocated during initialization - devices
 ******************************************************************************/
void cleanup() {
  cleanup_program();
  free(devices);
}
 
/*******************************************************************************
 * \brief   Create a command queue for each kernel
 ******************************************************************************/
void queue_setup() {
  cl_int status = 0;
  // Create one command queue for each kernel.
  queue1 =
      clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &status);
  checkError(status, "Failed to create command queue1");
  queue2 =
      clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &status);
  checkError(status, "Failed to create command queue2");
  queue3 =
      clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &status);
  checkError(status, "Failed to create command queue3");
  queue4 =
      clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &status);
  checkError(status, "Failed to create command queue4");
  queue5 =
      clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &status);
  checkError(status, "Failed to create command queue5");
  queue6 =
      clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &status);
  checkError(status, "Failed to create command queue6");
}
 
/*******************************************************************************
 * \brief   Release all command queues
 ******************************************************************************/
void queue_cleanup() {
  if (queue1)
    clReleaseCommandQueue(queue1);
  if (queue2)
    clReleaseCommandQueue(queue2);
  if (queue3)
    clReleaseCommandQueue(queue3);
  if (queue4)
    clReleaseCommandQueue(queue4);
  if (queue5)
    clReleaseCommandQueue(queue5);
  if (queue6)
    clReleaseCommandQueue(queue6);
}
 
#endif