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

CUTLASS 2.6.1 - functional and performance enhancements to strided DGRAD, fixes, and tuning

Browse Source 
* cutlass 2.6 update

* remove debug prints

* cutlass 2.6.1 (minor update)

* Updated CHANGELOG.

* Minor edit to readme to indicate patch version.

* Minor edit to readme.

Co-authored-by:  Haicheng Wu <haichengw@nvidia.com>, Andrew Kerr <akerr@nvidia.com>
This commit is contained in:
Manish Gupta
2021-09-03 10:26:15 -07:00
committed by GitHub
co-authored by Haicheng Wu <haichengw@nvidia.com>, Andrew Kerr <akerr@nvidia.com>
parent a01feb93d9
commit 6c2f8f2fb8
55 changed files with 317 additions and 315 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cutlass/epilogue/threadblock/predicated_tile_iterator.h
+13 -5
View File
@@ -62,7 +62,8 @@ namespace threadblock {
///
template <
typename ThreadMap_, ///< Thread map (conept: OutputTileThreadMap)
typename Element_ ///< Element data type
typename Element_, ///< Element data type
bool UseCUDAStore = false
>
class PredicatedTileIterator {
public:
@@ -341,10 +342,17 @@ public:
bool guard = row_guard && mask_.predicates[column];
cutlass::arch::global_store<AccessType, sizeof(AccessType)>(
frag_ptr[frag_row_idx * ThreadMap::Iterations::kColumn + column],
(void *)&memory_pointer[column * ThreadMap::Delta::kColumn / kElementsPerAccess],
guard);
if (UseCUDAStore) {
if (guard) {
memory_pointer[column * ThreadMap::Delta::kColumn / kElementsPerAccess] =
frag_ptr[frag_row_idx * ThreadMap::Iterations::kColumn + column];
}
} else {
cutlass::arch::global_store<AccessType, sizeof(AccessType)>(
frag_ptr[frag_row_idx * ThreadMap::Iterations::kColumn + column],
(void *)&memory_pointer[column * ThreadMap::Delta::kColumn / kElementsPerAccess],
guard);
}
}
if (row + 1 < ThreadMap::Iterations::kRow) {