Back to Cutlass

CUTLASS: gemm_pipelined.h Source File

docs/gemm__pipelined_8h_source.html

4.4.29.7 KB
Original Source

| | CUTLASS

CUDA Templates for Linear Algebra Subroutines and Solvers |

gemm_pipelined.h

Go to the documentation of this file.

1 /***************************************************************************************************

2 * Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.

3 *

4 * Redistribution and use in source and binary forms, with or without modification, are permitted

5 * provided that the following conditions are met:

6 * * Redistributions of source code must retain the above copyright notice, this list of

7 * conditions and the following disclaimer.

8 * * Redistributions in binary form must reproduce the above copyright notice, this list of

9 * conditions and the following disclaimer in the documentation and/or other materials

10 * provided with the distribution.

11 * * Neither the name of the NVIDIA CORPORATION nor the names of its contributors may be used

12 * to endorse or promote products derived from this software without specific prior written

13 * permission.

14 *

15 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR

16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND

17 * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE

18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,

19 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;

20 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,

21 * STRICT LIABILITY, OR TOR (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE

22 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

23 *

24 **************************************************************************************************/

29 #pragma once

30

31 #include "cutlass/cutlass.h"

32

33 #include "cutlass/aligned_buffer.h"

34 #include "cutlass/array.h"

35

36 #include "cutlass/numeric_types.h"

37 #include "cutlass/matrix_shape.h"

38

39 #include "cutlass/gemm/gemm.h"

40

42

43 namespace cutlass {

44 namespace gemm {

45 namespace kernel {

46

48

49 template <typename Mma, typename Epilogue, typename ThreadblockSwizzle>

50 __global__ void GemmPipelined(

51cutlass::gemm::GemmCoord problem_size,

52cutlass::gemm::GemmCoord grid_tiled_shape,

53typename Mma::IteratorA::Params params_A,

54typename Mma::IteratorA::TensorRef ref_A,

55typename Mma::IteratorB::Params params_B,

56typename Mma::IteratorB::TensorRef ref_B,

57typename Epilogue::Params params_epilogue

58 ) {

59

60// Shared storage needed by threadblock-scoped matrix multiply-accumulate

61 __shared__ union {

62typename Mma::SharedStorage main_loop;

63typename Epilogue::SharedStorage epilogue;

64 } shared_storage;

65

66// Compute threadblock location

67 ThreadblockSwizzle threadblock_swizzle;

68

69cutlass::gemm::GemmCoord tb_tile_offset = threadblock_swizzle.get_tile_offset();

70

71if (grid_tiled_shape.m() <= tb_tile_offset.m() ||

72 grid_tiled_shape.n() <= tb_tile_offset.n()) {

73

74return;

75 }

76

77// Compute initial location in logical coordinates

78cutlass::MatrixCoord tb_offset_A{

79 tb_tile_offset.m() * Mma::Shape::kM,

80 tb_tile_offset.k()

81 };

82

83cutlass::MatrixCoord tb_offset_B{

84 tb_tile_offset.k(),

85 tb_tile_offset.n() * Mma::Shape::kN

86 };

87

88// Compute position within threadblock

89int tb_thread_id = threadIdx.x;

90

91// Construct iterators to A and B operands

92typename Mma::IteratorA iterator_A(

93 params_A,

94 ref_A.data(),

95 {problem_size.m(), problem_size.k()},

96 tb_thread_id,

97 tb_offset_A);

98

99typename Mma::IteratorB iterator_B(

100 params_B,

101 ref_B.data(),

102 {problem_size.k(), problem_size.n()},

103 tb_thread_id,

104 tb_offset_B);

105

106int warp_id = __shfl_sync(0xffffffff, threadIdx.x / 32, 0);

107int lane_id = threadIdx.x % 32;

108

109//

110// Main loop

111//

112

113// Construct thread-scoped matrix multiply

114 Mma mma(shared_storage.main_loop, tb_thread_id, warp_id, lane_id);

115

116typename Mma::FragmentC accumulators;

117

118 accumulators.clear();

119

120// Compute threadblock-scoped matrix multiply-add

121 mma(problem_size, accumulators, iterator_A, iterator_B, accumulators);

122

123//

124// Epilogue

125//

126

127 Epilogue epilogue(

128 params_epilogue,

129 shared_storage.epilogue,

130 tb_thread_id,

131 warp_id,

132 lane_id);

133

134 tb_tile_offset = threadblock_swizzle.get_tile_offset();

135

136//assume identity swizzle

137MatrixCoord threadblock_offset(

138 tb_tile_offset.m() * Mma::Shape::kM,

139 tb_tile_offset.n() * Mma::Shape::kN

140 );

141

142// run efficient epilogue

143 epilogue({problem_size.m(), problem_size.n()}, accumulators, threadblock_offset);

144 }

145

147

148 } // namespace kernel

149 } // namespace gemm

150 } // namespace cutlass

cutlass

Definition: aligned_buffer.h:35

cutlass::gemm::GemmCoord

Definition: include/cutlass/gemm/gemm.h:94

gemm.h

Defines common types used for all GEMM-like operators.

cutlass::gemm::GemmCoord::n

CUTLASS_HOST_DEVICE Index const & n() const

Returns the GEMM N coordinate.

Definition: include/cutlass/gemm/gemm.h:137

cutlass::gemm::kernel::GemmPipelined

__global__ void GemmPipelined(cutlass::gemm::GemmCoord problem_size, cutlass::gemm::GemmCoord grid_tiled_shape, typename Mma::IteratorA::Params params_A, typename Mma::IteratorA::TensorRef ref_A, typename Mma::IteratorB::Params params_B, typename Mma::IteratorB::TensorRef ref_B, typename Epilogue::Params params_epilogue)

Definition: gemm_pipelined.h:50

cutlass::gemm::GemmCoord::k

CUTLASS_HOST_DEVICE Index const & k() const

Returns the GEMM K coordinate.

Definition: include/cutlass/gemm/gemm.h:145

array.h

Statically sized array of elements that accommodates all CUTLASS-supported numeric types and is safe ...

matrix_shape.h

Defines a Shape template for matrix tiles.

aligned_buffer.h

AlignedBuffer is a container for trivially copyable elements suitable for use in unions and shared me...

numeric_types.h

Top-level include for all CUTLASS numeric types.

cutlass::gemm::GemmCoord::m

CUTLASS_HOST_DEVICE Index const & m() const

Returns the GEMM M coordinate.

Definition: include/cutlass/gemm/gemm.h:129

cutlass.h

Basic include for CUTLASS.

cutlass::MatrixCoord

Definition: matrix_coord.h:39


Generated by 1.8.11