CUTLASS 2.10 (#615)

Co-authored-by: Aniket Shivam <ashivam@nvidia.com>
This commit is contained in:
ANIKET SHIVAM
2022-09-03 18:48:46 -04:00
committed by GitHub
co-authored by Aniket Shivam
parent ca23ff7924
commit b72cbf957d
289 changed files with 43708 additions and 2513 deletions
+13 -5
View File
@@ -45,6 +45,8 @@
#include "cutlass/gemm/kernel/default_gemm.h"
#include "cutlass/gemm/device/default_gemm_configuration.h"
#include "cutlass/layout/permute.h"
////////////////////////////////////////////////////////////////////////////////
namespace cutlass {
@@ -225,7 +227,9 @@ template <
/// Gather operand B by using an index array
bool GatherB = false,
/// Scatter result D by using an index array
bool ScatterD = false>
bool ScatterD = false,
/// Permute result D
typename PermuteDLayout = layout::NoPermute>
class Gemm {
public:
@@ -280,7 +284,8 @@ class Gemm {
SharedMemoryClearOption::kNone,
GatherA,
GatherB,
ScatterD
ScatterD,
PermuteDLayout
>::GemmKernel;
/// Argument structure
@@ -559,14 +564,16 @@ template <
/// Gather operand B by using an index array
bool GatherB,
/// Scatter result D by using an index array
bool ScatterD
bool ScatterD,
/// Permute result D
typename PermuteDLayout
>
class Gemm<ElementA_, LayoutA_, ElementB_, LayoutB_, ElementC_,
layout::ColumnMajor, // partially specialized on LayoutC
ElementAccumulator_, OperatorClass_, ArchTag_, ThreadblockShape_,
WarpShape_, InstructionShape_, EpilogueOutputOp_,
ThreadblockSwizzle_, Stages, AlignmentA, AlignmentB, SplitKSerial,
Operator_, GatherA, GatherB, ScatterD> {
Operator_, GatherA, GatherB, ScatterD, PermuteDLayout> {
public:
using ElementA = ElementA_;
@@ -617,7 +624,8 @@ class Gemm<ElementA_, LayoutA_, ElementB_, LayoutB_, ElementC_,
Operator,
GatherB,
GatherA,
ScatterD
ScatterD,
PermuteDLayout
>;
using UnderlyingArguments = typename UnderlyingOperator::Arguments;