leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 25 Packages Projects Releases Wiki Activity

CUTLASS 2.2 (#96)

Browse Source 
Adds support for NVIDIA Ampere Architecture features. CUDA 11 Toolkit recommended.
This commit is contained in:
Andrew Kerr
2020-06-08 16:17:35 -07:00
committed by GitHub
parent e33d90b361
commit 86931fef85
584 changed files with 51080 additions and 3373 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/03_visualize_layout/register_layout.cu
+26 -2
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2020, 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:
@@ -34,6 +34,8 @@
#include "cutlass/layout/pitch_linear.h"
#include "cutlass/layout/tensor_op_multiplicand_sm70.h"
#include "cutlass/layout/tensor_op_multiplicand_sm75.h"
#include "cutlass/layout/tensor_op_multiplicand_sm80.h"
#include "visualize_layout.h"
#include "register_layout.h"
@@ -59,18 +61,40 @@ void RegisterLayouts(std::map<std::string, std::unique_ptr<VisualizeLayoutBase>
// Integer matrix multiply.int4 8832 TN kblock128
{"TensorOpMultiplicand<4,128>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<4, 128>>},
// Integer matrix multiply.int4 16864 TN kblock256
{"TensorOpMultiplicand<4,256>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<4, 256>>},
// 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>>},
{"TensorOpMultiplicand<8,128>",
new VisualizeLayout<cutlass::layout::TensorOpMultiplicand<8, 128>>},
// 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>>},
// 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>},
{"TensorOpMultiplicandCongruous<128,4>",
new VisualizeLayout<
cutlass::layout::TensorOpMultiplicandCongruous<128, 4>>},
@@ -82,7 +106,7 @@ void RegisterLayouts(std::map<std::string, std::unique_ptr<VisualizeLayoutBase>
cutlass::layout::VoltaTensorOpMultiplicandCongruous<16>>},
{"VoltaTensorOpMultiplicandCrosswise<16,32>",
new VisualizeLayout<
cutlass::layout::VoltaTensorOpMultiplicandCrosswise<16, 32>>},
cutlass::layout::VoltaTensorOpMultiplicandCrosswise<16, 32>>}
};
for (auto layout : layout_pairs) {