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

Re-enable all alignments for int accumulators (#807)

Browse Source
This commit is contained in:
Jack Kosaian 2023-02-06 22:01:15 -05:00 committed by GitHub
parent add4ba622f
commit 5921043981
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 6 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
include/cutlass/epilogue/threadblock/default_epilogue_tensor_op.h
View File

@ -136,14 +136,15 @@ struct DefaultIteratorsTensorOp<float, float, 4, ThreadblockShape, WarpShape, In
static int const kFragmentsPerIteration = 2;
};
/// Partial specialization for int32_t <= int32_t x 4
/// Partial specialization for int32_t <= int32_t
template <
int ElementsPerAccess,
typename ThreadblockShape,
typename WarpShape,
typename InstructionShape,
typename ThreadMap
>
struct DefaultIteratorsTensorOp<int32_t, int32_t, 4, ThreadblockShape, WarpShape, InstructionShape, ThreadMap> {
struct DefaultIteratorsTensorOp<int32_t, int32_t, ElementsPerAccess, ThreadblockShape, WarpShape, InstructionShape, ThreadMap> {
using WarpTileIterator = cutlass::epilogue::warp::TileIteratorTensorOp<
WarpShape,
@ -160,14 +161,15 @@ struct DefaultIteratorsTensorOp<int32_t, int32_t, 4, ThreadblockShape, WarpShape
static int const kFragmentsPerIteration = 1;
};
/// Partial specialization for float <= int32_t x 4
/// Partial specialization for float <= int32_t
template <
int ElementsPerAccess,
typename ThreadblockShape,
typename WarpShape,
typename InstructionShape,
typename ThreadMap
>
struct DefaultIteratorsTensorOp<float, int32_t, 4, ThreadblockShape, WarpShape, InstructionShape, ThreadMap> {
struct DefaultIteratorsTensorOp<float, int32_t, ElementsPerAccess, ThreadblockShape, WarpShape, InstructionShape, ThreadMap> {
using WarpTileIterator = cutlass::epilogue::warp::TileIteratorTensorOp<
WarpShape,