squall/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity
e5d51840e8
cutlass/examples/03_visualize_layout/register_layout.cu

118 lines
5.7 KiB
Plaintext
Raw Normal View History

CUTLASS 2.0 (#62) CUTLASS 2.0 Substantially refactored for - Better performance, particularly for native Turing Tensor Cores - Robust and durable templates spanning the design space - Encapsulated functionality embodying modern C++11 programming techniques - Optimized containers and data types for efficient, generic, portable device code Updates to: - Quick start guide - Documentation - Utilities - CUTLASS Profiler Native Turing Tensor Cores - Efficient GEMM kernels targeting Turing Tensor Cores - Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands Coverage of existing CUTLASS functionality: - GEMM kernels targeting CUDA and Tensor Cores in NVIDIA GPUs - Volta Tensor Cores through native mma.sync and through WMMA API - Optimizations such as parallel reductions, threadblock rasterization, and intra-threadblock reductions - Batched GEMM operations - Complex-valued GEMMs Note: this commit and all that follow require a host compiler supporting C++11 or greater.
2019-11-20 08:55:34 +08:00
/***************************************************************************************************
CUTLASS 2.5
2021-02-26 22:58:26 +08:00
* Copyright (c) 2017-2021, NVIDIA CORPORATION. All rights reserved.
CUTLASS 2.0 (#62) CUTLASS 2.0 Substantially refactored for - Better performance, particularly for native Turing Tensor Cores - Robust and durable templates spanning the design space - Encapsulated functionality embodying modern C++11 programming techniques - Optimized containers and data types for efficient, generic, portable device code Updates to: - Quick start guide - Documentation - Utilities - CUTLASS Profiler Native Turing Tensor Cores - Efficient GEMM kernels targeting Turing Tensor Cores - Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands Coverage of existing CUTLASS functionality: - GEMM kernels targeting CUDA and Tensor Cores in NVIDIA GPUs - Volta Tensor Cores through native mma.sync and through WMMA API - Optimizations such as parallel reductions, threadblock rasterization, and intra-threadblock reductions - Batched GEMM operations - Complex-valued GEMMs Note: this commit and all that follow require a host compiler supporting C++11 or greater.
2019-11-20 08:55:34 +08:00
*
* 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 CUTLASS layout visualization example
*/
#include <map>
#include <memory>
#include "cutlass/layout/matrix.h"
#include "cutlass/layout/pitch_linear.h"
#include "cutlass/layout/tensor_op_multiplicand_sm70.h"
#include "cutlass/layout/tensor_op_multiplicand_sm75.h"
CUTLASS 2.2 (#96) Adds support for NVIDIA Ampere Architecture features. CUDA 11 Toolkit recommended.
2020-06-09 07:17:35 +08:00
#include "cutlass/layout/tensor_op_multiplicand_sm80.h"
CUTLASS 2.0 (#62) CUTLASS 2.0 Substantially refactored for - Better performance, particularly for native Turing Tensor Cores - Robust and durable templates spanning the design space - Encapsulated functionality embodying modern C++11 programming techniques - Optimized containers and data types for efficient, generic, portable device code Updates to: - Quick start guide - Documentation - Utilities - CUTLASS Profiler Native Turing Tensor Cores - Efficient GEMM kernels targeting Turing Tensor Cores - Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands Coverage of existing CUTLASS functionality: - GEMM kernels targeting CUDA and Tensor Cores in NVIDIA GPUs - Volta Tensor Cores through native mma.sync and through WMMA API - Optimizations such as parallel reductions, threadblock rasterization, and intra-threadblock reductions - Batched GEMM operations - Complex-valued GEMMs Note: this commit and all that follow require a host compiler supporting C++11 or greater.
2019-11-20 08:55:34 +08:00
#include "visualize_layout.h"
#include "register_layout.h"
/////////////////////////////////////////////////////////////////////////////////////////////////
void RegisterLayouts(std::map<std::string, std::unique_ptr<VisualizeLayoutBase> > &layouts) {
struct {
char const *name;
VisualizeLayoutBase *ptr;
} layout_pairs[] = {
{"PitchLinear", new VisualizeLayout<cutlass::layout::PitchLinear>},
{"ColumnMajor", new VisualizeLayout<cutlass::layout::ColumnMajor>},
{"RowMajor", new VisualizeLayout<cutlass::layout::RowMajor>},
{"ColumnMajorInterleaved<4>",
new VisualizeLayout<cutlass::layout::ColumnMajorInterleaved<4>>},
{"RowMajorInterleaved<4>",
new VisualizeLayout<cutlass::layout::RowMajorInterleaved<4>>},
// Integer matrix multiply.int4 8832 Interleaved-64
{"TensorOpMultiplicand<4,64>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<4, 64>>},
// Integer matrix multiply.int4 8832 TN kblock128
{"TensorOpMultiplicand<4,128>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<4, 128>>},
CUTLASS 2.2 (#96) Adds support for NVIDIA Ampere Architecture features. CUDA 11 Toolkit recommended.
2020-06-09 07:17:35 +08:00
// Integer matrix multiply.int4 16864 TN kblock256
{"TensorOpMultiplicand<4,256>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<4, 256>>},
CUTLASS 2.0 (#62) CUTLASS 2.0 Substantially refactored for - Better performance, particularly for native Turing Tensor Cores - Robust and durable templates spanning the design space - Encapsulated functionality embodying modern C++11 programming techniques - Optimized containers and data types for efficient, generic, portable device code Updates to: - Quick start guide - Documentation - Utilities - CUTLASS Profiler Native Turing Tensor Cores - Efficient GEMM kernels targeting Turing Tensor Cores - Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands Coverage of existing CUTLASS functionality: - GEMM kernels targeting CUDA and Tensor Cores in NVIDIA GPUs - Volta Tensor Cores through native mma.sync and through WMMA API - Optimizations such as parallel reductions, threadblock rasterization, and intra-threadblock reductions - Batched GEMM operations - Complex-valued GEMMs Note: this commit and all that follow require a host compiler supporting C++11 or greater.
2019-11-20 08:55:34 +08:00
// Integer matrix multiply 8816 Interleaved-32
{"TensorOpMultiplicand<8,32>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<8, 32>>},
// Integer matrix multiply 8816 TN kblock64
{"TensorOpMultiplicand<8,64>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<8, 64>>},
CUTLASS 2.2 (#96) Adds support for NVIDIA Ampere Architecture features. CUDA 11 Toolkit recommended.
2020-06-09 07:17:35 +08:00
{"TensorOpMultiplicand<8,128>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<8, 128>>},
CUTLASS 2.0 (#62) CUTLASS 2.0 Substantially refactored for - Better performance, particularly for native Turing Tensor Cores - Robust and durable templates spanning the design space - Encapsulated functionality embodying modern C++11 programming techniques - Optimized containers and data types for efficient, generic, portable device code Updates to: - Quick start guide - Documentation - Utilities - CUTLASS Profiler Native Turing Tensor Cores - Efficient GEMM kernels targeting Turing Tensor Cores - Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands Coverage of existing CUTLASS functionality: - GEMM kernels targeting CUDA and Tensor Cores in NVIDIA GPUs - Volta Tensor Cores through native mma.sync and through WMMA API - Optimizations such as parallel reductions, threadblock rasterization, and intra-threadblock reductions - Batched GEMM operations - Complex-valued GEMMs Note: this commit and all that follow require a host compiler supporting C++11 or greater.
2019-11-20 08:55:34 +08:00
// Matrix Multiply 1688 TN kblock32
{"TensorOpMultiplicand<16,32>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<16, 32>>},
// Matrix multiply 1688 NT
{"TensorOpMultiplicand<16,64>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<16, 64>>},
CUTLASS 2.2 (#96) Adds support for NVIDIA Ampere Architecture features. CUDA 11 Toolkit recommended.
2020-06-09 07:17:35 +08:00
// Matrix multiply 1688.TF32 TN kblock16
{"TensorOpMultiplicand<32,16>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<32, 16>>},
// Matrix multiply 1688.TF32 TN kblock32
{"TensorOpMultiplicand<32,32>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<32, 32>>},
// Matrix multiply 1688 NT
{"TensorOpMultiplicandCongruous<32,32>",
new VisualizeLayout<
cutlass::layout::TensorOpMultiplicandCongruous<32, 32>>},
// Matrix multiply 884 NT
{"TensorOpMultiplicandCongruous<64,16>",
new VisualizeLayout<
cutlass::layout::TensorOpMultiplicandCongruous<64, 16>>},
// Matrix multiply 884 TN
{"TensorOpMultiplicand64bCrosswise",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand64bCrosswise>},
CUTLASS 2.0 (#62) CUTLASS 2.0 Substantially refactored for - Better performance, particularly for native Turing Tensor Cores - Robust and durable templates spanning the design space - Encapsulated functionality embodying modern C++11 programming techniques - Optimized containers and data types for efficient, generic, portable device code Updates to: - Quick start guide - Documentation - Utilities - CUTLASS Profiler Native Turing Tensor Cores - Efficient GEMM kernels targeting Turing Tensor Cores - Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands Coverage of existing CUTLASS functionality: - GEMM kernels targeting CUDA and Tensor Cores in NVIDIA GPUs - Volta Tensor Cores through native mma.sync and through WMMA API - Optimizations such as parallel reductions, threadblock rasterization, and intra-threadblock reductions - Batched GEMM operations - Complex-valued GEMMs Note: this commit and all that follow require a host compiler supporting C++11 or greater.
2019-11-20 08:55:34 +08:00
{"TensorOpMultiplicandCongruous<128,4>",
new VisualizeLayout<
cutlass::layout::TensorOpMultiplicandCongruous<128, 4>>},
{"TensorOpMultiplicandCrosswise<128,4>",
new VisualizeLayout<
cutlass::layout::TensorOpMultiplicandCrosswise<128, 4>>},
{"VoltaTensorOpMultiplicandCongruous<16>",
new VisualizeLayout<
cutlass::layout::VoltaTensorOpMultiplicandCongruous<16>>},
{"VoltaTensorOpMultiplicandCrosswise<16,32>",
new VisualizeLayout<
CUTLASS 2.2 (#96) Adds support for NVIDIA Ampere Architecture features. CUDA 11 Toolkit recommended.
2020-06-09 07:17:35 +08:00
cutlass::layout::VoltaTensorOpMultiplicandCrosswise<16, 32>>}
CUTLASS 2.0 (#62) CUTLASS 2.0 Substantially refactored for - Better performance, particularly for native Turing Tensor Cores - Robust and durable templates spanning the design space - Encapsulated functionality embodying modern C++11 programming techniques - Optimized containers and data types for efficient, generic, portable device code Updates to: - Quick start guide - Documentation - Utilities - CUTLASS Profiler Native Turing Tensor Cores - Efficient GEMM kernels targeting Turing Tensor Cores - Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands Coverage of existing CUTLASS functionality: - GEMM kernels targeting CUDA and Tensor Cores in NVIDIA GPUs - Volta Tensor Cores through native mma.sync and through WMMA API - Optimizations such as parallel reductions, threadblock rasterization, and intra-threadblock reductions - Batched GEMM operations - Complex-valued GEMMs Note: this commit and all that follow require a host compiler supporting C++11 or greater.
2019-11-20 08:55:34 +08:00
};
for (auto layout : layout_pairs) {
layouts.emplace(std::string(layout.name), std::unique_ptr<VisualizeLayoutBase>(layout.ptr));
}
}
/////////////////////////////////////////////////////////////////////////////////////////////////
Reference in New Issue Copy Permalink