mirror of
https://github.com/NVIDIA/cutlass.git
synced 2026-05-11 17:00:05 +00:00
96dab34ad9
CUTLASS 2.1 contributes: - BLAS-style host-side API added to CUTLASS Library - Planar Complex GEMM kernels targeting Volta and Turing Tensor Cores - Minor enhancements and bug fixes
119 lines
4.9 KiB
C++
119 lines
4.9 KiB
C++
/***************************************************************************************************
|
|
* 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.
|
|
*
|
|
**************************************************************************************************/
|
|
|
|
/*! \file
|
|
\brief
|
|
Default kernel-level GEMM definitions combine threadblock-scoped matrix multiply-add with
|
|
the appropriate threadblock-scoped epilogue.
|
|
|
|
Note, CUTLASS epilogues universally target row-major outputs. Column-major outputs are
|
|
accommodated by exchanging A and B operands and assuming transposed layouts. Partial
|
|
specializations here choose 'device::GemmTransposed' to implement this functionality.
|
|
|
|
*/
|
|
|
|
#pragma once
|
|
|
|
#include "cutlass/cutlass.h"
|
|
|
|
#include "cutlass/layout/matrix.h"
|
|
#include "cutlass/numeric_types.h"
|
|
|
|
#include "cutlass/epilogue/threadblock/epilogue.h"
|
|
#include "cutlass/epilogue/thread/linear_combination.h"
|
|
|
|
#include "cutlass/gemm/gemm.h"
|
|
#include "cutlass/gemm/kernel/gemm.h"
|
|
#include "cutlass/gemm/kernel/gemm_pipelined.h"
|
|
#include "cutlass/gemm/threadblock/default_mma_core_sm75.h"
|
|
#include "cutlass/gemm/threadblock/default_mma_core_sm70.h"
|
|
#include "cutlass/gemm/threadblock/default_mma.h"
|
|
#include "cutlass/gemm/threadblock/default_mma_core_simt.h"
|
|
#include "cutlass/gemm/threadblock/threadblock_swizzle.h"
|
|
#include "cutlass/epilogue/threadblock/default_epilogue_complex_tensor_op.h"
|
|
#include "cutlass/epilogue/threadblock/default_epilogue_simt.h"
|
|
|
|
#include "cutlass/transform/threadblock/predicated_tile_iterator.h"
|
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
namespace cutlass {
|
|
namespace gemm {
|
|
namespace kernel {
|
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
template <
|
|
/// Element type for A matrix operand
|
|
typename ElementA_,
|
|
/// Layout type for A matrix operand
|
|
typename LayoutA_,
|
|
/// Element type for B matrix operand
|
|
typename ElementB_,
|
|
/// Layout type for B matrix operand
|
|
typename LayoutB_,
|
|
/// Element type for C and D matrix operands
|
|
typename ElementC_,
|
|
/// Layout type for C and D matrix operands
|
|
typename LayoutC_,
|
|
/// Element type for internal accumulation
|
|
typename ElementAccumulator,
|
|
/// Operator class tag
|
|
typename OperatorClass,
|
|
/// Tag indicating architecture to tune for
|
|
typename ArchTag,
|
|
/// Threadblock-level tile size (concept: GemmShape)
|
|
typename ThreadblockShape,
|
|
/// Warp-level tile size (concept: GemmShape)
|
|
typename WarpShape,
|
|
/// Warp-level tile size (concept: GemmShape)
|
|
typename InstructionShape,
|
|
/// Epilogue output operator
|
|
typename EpilogueOutputOp,
|
|
/// Threadblock-level swizzling operator
|
|
typename ThreadblockSwizzle,
|
|
/// Number of stages used in the pipelined mainloop
|
|
int Stages,
|
|
/// Complex elementwise transformation on A operand
|
|
ComplexTransform TransformA,
|
|
/// Complex elementwise transformation on B operand
|
|
ComplexTransform TransformB,
|
|
/// Multiply-add operator
|
|
typename Operator,
|
|
/// If true, kernel is configured to support serial reduction in the epilogue
|
|
bool SplitKSerial
|
|
>
|
|
struct DefaultGemmComplex;
|
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
} // namespace kernel
|
|
} // namespace gemm
|
|
} // namespace cutlass
|
|
|
|
////////////////////////////////////////////////////////////////////////////////
|