leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 24 Packages Projects Releases Wiki Activity

CUTLASS 3.5.1 (#1623)

Browse Source 
* CUTLASS 3.5.1

* updates, optimizations, fixes
This commit is contained in:
Vijay Thakkar
2024-07-29 08:46:24 -04:00
committed by GitHub
parent 56b46e2d13
commit be60a0b272
312 changed files with 19793 additions and 6775 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cute/arch/copy.hpp
+8 -6
View File
@@ -68,7 +68,7 @@ struct UniversalCopy
//
// Placeholder for the copy algorithm's stronger auto-vectorizing behavior
// that assumes alignment of dynamic layouts up to MaxVecBits
// that assumes alignment of pointers and dynamic layouts up to MaxVecBits
//
template <int MaxVecBits = 128>
@@ -80,15 +80,17 @@ struct AutoVectorizingCopyWithAssumedAlignment
};
//
// Placeholder for the copy algorithm's default auto-vectorizing behavior
// that does not assume alignment of dynamic layouts
// AutoVectorizingCopy alias assumes maximal alignment of pointers and dynamic strides.
// If this is not the case then AutoVectorizingCopyWithAssumedAlignment should be used instead
//
using AutoVectorizingCopy = AutoVectorizingCopyWithAssumedAlignment<8>;
using AutoVectorizingCopy = AutoVectorizingCopyWithAssumedAlignment<128>;
// Alias
using DefaultCopy = AutoVectorizingCopy;
//
// DefaultCopy alias does not assume alignment of pointers or dynamic strides.
//
using DefaultCopy = AutoVectorizingCopyWithAssumedAlignment<8>;
//
// Global memory prefetch into L2