xref: /libCEED/backends/hip/ceed-hip-compile.cpp (revision 30f4f45fbb110fe49bb316a924b47463b018a077)

// Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC.
// Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707.
// All Rights reserved. See files LICENSE and NOTICE for details.
//
// This file is part of CEED, a collection of benchmarks, miniapps, software
// libraries and APIs for efficient high-order finite element and spectral
// element discretizations for exascale applications. For more information and
// source code availability see http://github.com/ceed.
//
// The CEED research is supported by the Exascale Computing Project 17-SC-20-SC,
// a collaborative effort of two U.S. Department of Energy organizations (Office
// of Science and the National Nuclear Security Administration) responsible for
// the planning and preparation of a capable exascale ecosystem, including
// software, applications, hardware, advanced system engineering and early
// testbed platforms, in support of the nation's exascale computing imperative.

#include <ceed-backend.h>
#include <string.h>
#include <sstream>
#include <stdarg.h>
#include <hip/hiprtc.h>
#include "ceed-hip.h"
#include "ceed-hip-compile.h"

#define CeedChk_hiprtc(ceed, x) \
do { \
  hiprtcResult result = static_cast<hiprtcResult>(x); \
  if (result != HIPRTC_SUCCESS) \
    return CeedError((ceed), x, hiprtcGetErrorString(result)); \
} while (0)

//------------------------------------------------------------------------------
// Compile HIP kernel
//------------------------------------------------------------------------------
int CeedCompileHip(Ceed ceed, const char *source, hipModule_t *module,
                    const CeedInt numopts, ...) {
  int ierr;
  hipFree(0); // Make sure a Context exists for hiprtc
  hiprtcProgram prog;

  // Add hip runtime include to string for generation
  std::ostringstream code;
  code << "\n#include <hip/hip_runtime.h>\n";

  // Macro definitions
  // Get kernel specific options, such as kernel constants
  const int optslen = 32;
  const int optsextra = 2;
  const char *opts[optsextra];
  if (numopts > 0) {
    va_list args;
    va_start(args, numopts);
    char *name;
    int val;
    for (int i = 0; i < numopts; i++) {
      name = va_arg(args, char *);
      val = va_arg(args, int);
      code << "#define " << name << " " << val << "\n";
    }
    va_end(args);
  }

  // Standard backend options
  code << "#define CeedScalar double\n#define CeedInt int\n\n";

  // Non-macro options
  opts[0] = "-default-device";
  struct hipDeviceProp_t prop;
  Ceed_Hip *ceed_data;
  ierr = CeedGetData(ceed, (void **)&ceed_data); CeedChk(ierr);
  CeedChk_Hip(ceed, hipGetDeviceProperties(&prop, ceed_data->deviceId));
  char buff[optslen];
  std::string gfxName = "gfx" + std::to_string(prop.gcnArch);
  std::string archArg = "--gpu-architecture="  + gfxName;
  snprintf(buff, optslen, archArg.c_str());
  opts[1] = buff;

  // Add string source argument provided in call
  code << source;

  // Create Program
  CeedChk_hiprtc(ceed, hiprtcCreateProgram(&prog, code.str().c_str(), NULL, 0, NULL, NULL));

  // Compile kernel
  hiprtcResult result = hiprtcCompileProgram(prog, numopts + optsextra, opts);
  if (result != HIPRTC_SUCCESS) {
    size_t logsize;
    CeedChk_hiprtc(ceed, hiprtcGetProgramLogSize(prog, &logsize));
    char *log;
    ierr = CeedMalloc(logsize, &log); CeedChk(ierr);
    CeedChk_hiprtc(ceed, hiprtcGetProgramLog(prog, log));
    return CeedError(ceed, (int)result, "%s\n%s", hiprtcGetErrorString(result), log);
  }

  size_t ptxsize;
  CeedChk_hiprtc(ceed, hiprtcGetCodeSize(prog, &ptxsize));
  char *ptx;
  ierr = CeedMalloc(ptxsize, &ptx); CeedChk(ierr);
  CeedChk_hiprtc(ceed, hiprtcGetCode(prog, ptx));
  CeedChk_hiprtc(ceed, hiprtcDestroyProgram(&prog));

  CeedChk_Hip(ceed, hipModuleLoadData(module, ptx));
  ierr = CeedFree(&ptx); CeedChk(ierr);

  return 0;
}

//------------------------------------------------------------------------------
// Get HIP kernel
//------------------------------------------------------------------------------
int CeedGetKernelHip(Ceed ceed, hipModule_t module, const char *name,
                      hipFunction_t *kernel) {

  CeedChk_Hip(ceed, hipModuleGetFunction(kernel, module, name));
  return 0;
}

//------------------------------------------------------------------------------
// Run HIP kernel
//------------------------------------------------------------------------------
int CeedRunKernelHip(Ceed ceed, hipFunction_t kernel, const int gridSize,
                      const int blockSize, void **args) {
  CeedChk_Hip(ceed, hipModuleLaunchKernel(kernel, gridSize, 1, 1, blockSize, 1,
                                  1, 0, NULL, args, NULL));
  return 0;
}

//------------------------------------------------------------------------------
// Run HIP kernel for spatial dimension
//------------------------------------------------------------------------------
int CeedRunKernelDimHip(Ceed ceed, hipFunction_t kernel, const int gridSize,
                         const int blockSizeX, const int blockSizeY,
                         const int blockSizeZ, void **args) {
  CeedChk_Hip(ceed, hipModuleLaunchKernel(kernel, gridSize, 1, 1,
                                  blockSizeX, blockSizeY, blockSizeZ,
                                  0, NULL, args, NULL));
  return 0;
}

//------------------------------------------------------------------------------
// Run HIP kernel for spatial dimension with sharde memory
//------------------------------------------------------------------------------
int CeedRunKernelDimSharedHip(Ceed ceed, hipFunction_t kernel, const int gridSize,
                               const int blockSizeX, const int blockSizeY,
                               const int blockSizeZ, const int sharedMemSize,
                               void **args) {
  CeedChk_Hip(ceed, hipModuleLaunchKernel(kernel, gridSize, 1, 1,
                                  blockSizeX, blockSizeY, blockSizeZ,
                                  sharedMemSize, NULL, args, NULL));
  return 0;
}