leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 28 Packages Projects Releases Wiki Activity

CUTLASS 2.5

Browse Source
This commit is contained in:
Andrew Kerr
2021-02-26 09:58:26 -05:00
parent ccb697bac7
commit 0e13748649
771 changed files with 15474 additions and 1715 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cutlass/layout/layout.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2020, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2021, 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:
include/cutlass/layout/matrix.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2020, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2021, 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:
include/cutlass/layout/pitch_linear.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2020, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2021, 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:
include/cutlass/layout/tensor.h
+29 -5
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2020, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2021, 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:
@@ -149,11 +149,11 @@ public:
fast_divmod(w, tmp, w, int(stride_[0]), c_mul, c_shr);
#else
n = int(index / (stride_[0] * stride_[1] * stride_[2]));
LongIndex residual = index % (stride_[0] * stride_[1] * stride_[2]);
n = int(index / stride_[2]);
LongIndex residual = index % stride_[2];
h = int(residual / (stride_[0] * stride_[1]));
residual = (residual % (stride_[0] * stride_[1]));
h = int(residual / stride_[1]);
residual = (residual % stride_[1]);
w = int(residual / stride_[0]);
c = int(residual % stride_[0]);
@@ -314,6 +314,15 @@ public:
CUTLASS_HOST_DEVICE
TensorNCxHWx(Stride const &stride = Stride(0)): stride_(stride) { }
/// Constructor
CUTLASS_HOST_DEVICE
TensorNCxHWx(
typename Stride::Index stride_w, ///< number of elements between adjacent W coordinates
typename Stride::Index stride_h, ///< number of elements between adjacent H coordinates
typename Stride::Index stride_n ///< number of elements between adjacent N coordinates
):
stride_(make_Coord(stride_w, stride_h, stride_n)) { }
/// Helper returns a layout to a tightly packed tensor
CUTLASS_HOST_DEVICE
static TensorNCxHWx packed(TensorCoord const &extent) {
@@ -404,6 +413,15 @@ public:
CUTLASS_HOST_DEVICE
TensorCxRSKx(Stride const &stride = Stride(0)): stride_(stride) { }
/// Constructor
CUTLASS_HOST_DEVICE
TensorCxRSKx(
typename Stride::Index stride_w, ///< number of elements between adjacent W coordinates
typename Stride::Index stride_h, ///< number of elements between adjacent H coordinates
typename Stride::Index stride_n ///< number of elements between adjacent N coordinates
):
stride_(make_Coord(stride_w, stride_h, stride_n)) { }
/// Helper returns a layout to a tightly packed tensor
CUTLASS_HOST_DEVICE
static TensorCxRSKx packed(TensorCoord const &extent) {
@@ -529,6 +547,12 @@ public:
LongIndex(stride_[3] * coord.n());
}
/// Returns the offset of a pitchlinear coordinate in linear memory.
CUTLASS_HOST_DEVICE
LongIndex operator()(PitchLinearCoord coord) const {
return coord.contiguous() + LongIndex(coord.strided() * stride_[3]);
}
/// Returns the stride of the layout
CUTLASS_HOST_DEVICE
Stride stride() const {
include/cutlass/layout/tensor_op_multiplicand_sm70.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2020, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2021, 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:
include/cutlass/layout/tensor_op_multiplicand_sm75.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2020, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2021, 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:
include/cutlass/layout/tensor_op_multiplicand_sm80.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2020, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2021, 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:
include/cutlass/layout/vector.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2020, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2021, 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: