leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 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
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

22
include/cutlass/transform/threadblock/predicated_tile_access_iterator.h
View File

@@ -245,10 +245,10 @@ class PredicatedTileAccessIteratorPredicates {
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() {
void clear_mask(bool enable = true) {
CUTLASS_PRAGMA_UNROLL
for (int i = 0; i < kPredicateWordCount; ++i) {
predicates_[i] = 0u;
predicates_[i] = enable ? 0u : predicates_[i];
}
}
@@ -551,8 +551,8 @@ class PredicatedTileAccessIterator<Shape_, Element_, layout::PitchLinear,
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() {
the_predicates.clear_mask();
void clear_mask(bool enable = true) {
the_predicates.clear_mask(enable);
}
/// Clears the predicate set efficiently
@@ -741,7 +741,7 @@ class PredicatedTileAccessIterator<Shape_, Element_, layout::ColumnMajor,
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { iterator_.clear_mask(); }
void clear_mask(bool enable = true) { iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
@@ -922,7 +922,7 @@ class PredicatedTileAccessIterator<Shape_, Element_, layout::RowMajor,
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { iterator_.clear_mask(); }
void clear_mask(bool enable = true) { iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
@@ -1224,7 +1224,7 @@ class PredicatedTileAccessIterator<Shape_, Element_, layout::AffineRankN<2>,
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { the_predicates.clear_mask(); }
void clear_mask(bool enable = true) { the_predicates.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
@@ -1401,7 +1401,7 @@ class PredicatedTileAccessIterator<Shape_, Element_, layout::AffineRank2ColumnMa
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { iterator_.clear_mask(); }
void clear_mask(bool enable = true) { iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
@@ -1578,7 +1578,7 @@ class PredicatedTileAccessIterator<Shape_, Element_, layout::AffineRank2RowMajor
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { iterator_.clear_mask(); }
void clear_mask(bool enable = true) { iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
@@ -1764,7 +1764,7 @@ class PredicatedTileAccessIterator<Shape_, Element_,
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { iterator_.clear_mask(); }
void clear_mask(bool enable = true) { iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
@@ -1948,7 +1948,7 @@ class PredicatedTileAccessIterator<Shape_, Element_,
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { iterator_.clear_mask(); }
void clear_mask(bool enable = true) { iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE

8
include/cutlass/transform/threadblock/predicated_tile_access_iterator_2dthreadtile.h
View File

@@ -403,10 +403,10 @@ class PredicatedTileAccessIterator2dThreadTile<Shape_, Element_, layout::PitchLi
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() {
void clear_mask(bool enable = true) {
CUTLASS_PRAGMA_UNROLL
for (int i = 0; i < kPredicateWordCount; ++i) {
predicates_[i] = 0u;
predicates_[i] = enable ? 0u : predicates_[i];
}
}
@@ -617,7 +617,7 @@ class PredicatedTileAccessIterator2dThreadTile<Shape_, Element_, layout::ColumnM
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { iterator_.clear_mask(); }
void clear_mask(bool enable = true) { iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
@@ -796,7 +796,7 @@ class PredicatedTileAccessIterator2dThreadTile<Shape_, Element_, layout::RowMajo
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { iterator_.clear_mask(); }
void clear_mask(bool enable = true) { iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE

24
include/cutlass/transform/threadblock/predicated_tile_iterator.h
View File

@@ -288,7 +288,7 @@ class PredicatedTileIterator<Shape_, Element_, layout::PitchLinear, AdvanceRank,
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { address_iterator_.clear_mask(); }
void clear_mask(bool enable = true) { address_iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
@@ -530,8 +530,8 @@ public:
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() {
iterator_.clear_mask();
void clear_mask(bool enable = true) {
iterator_.clear_mask(enable);
}
/// Clears the predicate set efficiently
@@ -738,8 +738,8 @@ public:
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() {
iterator_.clear_mask();
void clear_mask(bool enable = true) {
iterator_.clear_mask(enable);
}
/// Clears the predicate set efficiently
@@ -946,7 +946,7 @@ class PredicatedTileIterator<Shape_, Element_, layout::AffineRankN<2>, AdvanceRa
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { address_iterator_.clear_mask(); }
void clear_mask(bool enable = true) { address_iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
@@ -1184,8 +1184,8 @@ public:
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() {
iterator_.clear_mask();
void clear_mask(bool enable = true) {
iterator_.clear_mask(enable);
}
/// Clears the predicate set efficiently
@@ -1388,8 +1388,8 @@ public:
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() {
iterator_.clear_mask();
void clear_mask(bool enable = true) {
iterator_.clear_mask(enable);
}
/// Clears the predicate set efficiently
@@ -1600,7 +1600,7 @@ class PredicatedTileIterator<Shape_, Element_,
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { iterator_.clear_mask(); }
void clear_mask(bool enable = true) { iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
@@ -1785,7 +1785,7 @@ class PredicatedTileIterator<Shape_, Element_,
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { iterator_.clear_mask(); }
void clear_mask(bool enable = true) { iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE

10
include/cutlass/transform/threadblock/predicated_tile_iterator_2dthreadtile.h
View File

@@ -293,7 +293,7 @@ class PredicatedTileIterator2dThreadTile<Shape_, Element_, layout::PitchLinear,
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() { address_iterator_.clear_mask(); }
void clear_mask(bool enable = true) { address_iterator_.clear_mask(enable); }
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
@@ -525,8 +525,8 @@ public:
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() {
iterator_.clear_mask();
void clear_mask(bool enable = true) {
iterator_.clear_mask(enable);
}
/// Clears the predicate set efficiently
@@ -721,8 +721,8 @@ public:
/// Clears the predicate set efficiently
CUTLASS_HOST_DEVICE
void clear_mask() {
iterator_.clear_mask();
void clear_mask(bool enable = true) {
iterator_.clear_mask(enable);
}
/// Clears the predicate set efficiently