cutlass.h

Go to the documentation of this file.

/***************************************************************************************************
 * Copyright (c) 2017-2019, NVIDIA CORPORATION.  All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without modification, are permitted
 * provided that the following conditions are met:
 *     * Redistributions of source code must retain the above copyright notice, this list of
 *       conditions and the following disclaimer.
 *     * Redistributions in binary form must reproduce the above copyright notice, this list of
 *       conditions and the following disclaimer in the documentation and/or other materials
 *       provided with the distribution.
 *     * Neither the name of the NVIDIA CORPORATION nor the names of its contributors may be used
 *       to endorse or promote products derived from this software without specific prior written
 *       permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
 * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
 * STRICT LIABILITY, OR TOR (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 **************************************************************************************************/

#pragma once


namespace cutlass {


enum class Status {
  kSuccess,                    
  kErrorMisalignedOperand,     
  kErrorInvalidLayout,         
  kErrorInvalidProblem,        
  kErrorNotSupported,          
  kErrorWorkspaceNull,         
  kErrorInternal,              
  kInvalid                     
};

static inline char const* cutlassGetStatusString(cutlass::Status status) {
  switch (status) {
    case cutlass::Status::kSuccess:
      return "Success";
    case cutlass::Status::kErrorMisalignedOperand:
      return "Error Misaligned Operand";
    case cutlass::Status::kErrorInvalidLayout:
      return "Error Invalid Layout";
    case cutlass::Status::kErrorInvalidProblem:
      return "Error Invalid Problem";
    case cutlass::Status::kErrorNotSupported:
      return "Error Not Supported";
    case cutlass::Status::kErrorWorkspaceNull:
      return "Error Workspace Null";
    case cutlass::Status::kErrorInternal:
      return "Error Internal";
    case cutlass::Status::kInvalid: break;
  }

  return "Invalid status";
}


// CUDA 10.1 introduces the mma instruction
#if !defined(CUTLASS_ENABLE_TENSOR_CORE_MMA)
#define CUTLASS_ENABLE_TENSOR_CORE_MMA 0
#endif


#if defined(__NVCC__) || (defined(__clang__) && defined(__CUDA__))
#define CUTLASS_HOST_DEVICE __forceinline__ __device__ __host__
#define CUTLASS_DEVICE __forceinline__ __device__
#elif defined(__CUDACC_RTC__)
#define CUTLASS_HOST_DEVICE __forceinline__ __device__
#define CUTLASS_DEVICE __forceinline__ __device__
#else
#define CUTLASS_HOST_DEVICE inline
#endif

#define CUTLASS_ASSERT(x) assert(x)


// CUTLASS_PRAGMA_(UNROLL|NO_UNROLL) optimization directives for the CUDA compiler.
#if defined(__CUDA_ARCH__)
  #if defined(__CUDACC_RTC__) || (defined(__clang__) && defined(__CUDA__))
    #define CUTLASS_PRAGMA_UNROLL _Pragma("unroll")
    #define CUTLASS_PRAGMA_NO_UNROLL _Pragma("unroll 1")
  #else
    #define CUTLASS_PRAGMA_UNROLL #pragma unroll
    #define CUTLASS_PRAGMA_NO_UNROLL #pragma unroll 1
  #endif

  #define CUTLASS_GEMM_LOOP CUTLASS_PRAGMA_NO_UNROLL

#else

    #define CUTLASS_PRAGMA_UNROLL
    #define CUTLASS_PRAGMA_NO_UNROLL
    #define CUTLASS_GEMM_LOOP

#endif



static const int NUM_THREADS_PER_WARP = 32;
static const int NUM_THREADS_PER_HALF_WARP = NUM_THREADS_PER_WARP / 2;
static const int NUM_THREADS_PER_QUAD = 4;
static const int NUM_THREADS_PER_QUAD_PAIR = NUM_THREADS_PER_QUAD * 2;

#if defined(__NVCC__) || (defined(__clang__) && defined(__CUDA__))

CUTLASS_DEVICE
int LaneId() {
  int ret; 
  asm ("mov.u32 %0, %%laneid;" : "=r"(ret));
  return ret;
}

#endif


}  // namespace cutlass