cutlass::gemm::thread::detail::Mma_HFMA2< Shape, LayoutA, LayoutB, layout::RowMajor, false > Struct Template Reference

#include <mma_sm60.h>

Public Types
using	FragmentA = Array< half_t, Shape::kMK >
	A operand storage. More...
using	FragmentB = Array< half_t, Shape::kKN >
	B operand storage. More...
using	FragmentC = Array< half_t, Shape::kMN >
	C operand storage. More...

Public Member Functions
CUTLASS_HOST_DEVICE void	operator() (FragmentC &D, FragmentA const &A, FragmentB const &B, FragmentC const &C)
	Computes a matrix product D = A * B + C. More...

Member Typedef Documentation

template<typename Shape , typename LayoutA , typename LayoutB >

using cutlass::gemm::thread::detail::Mma_HFMA2< Shape, LayoutA, LayoutB, layout::RowMajor, false >::FragmentA = Array<half_t, Shape::kMK>

template<typename Shape , typename LayoutA , typename LayoutB >

using cutlass::gemm::thread::detail::Mma_HFMA2< Shape, LayoutA, LayoutB, layout::RowMajor, false >::FragmentB = Array<half_t, Shape::kKN>

template<typename Shape , typename LayoutA , typename LayoutB >

using cutlass::gemm::thread::detail::Mma_HFMA2< Shape, LayoutA, LayoutB, layout::RowMajor, false >::FragmentC = Array<half_t, Shape::kMN>

Member Function Documentation

template<typename Shape , typename LayoutA , typename LayoutB >

CUTLASS_HOST_DEVICE void cutlass::gemm::thread::detail::Mma_HFMA2< Shape, LayoutA, LayoutB, layout::RowMajor, false >::operator() ( FragmentC &  D, FragmentA const &  A, FragmentB const &  B, FragmentC const &  C  )

inline

Initialize output with input

Use 1x1x2 HFMA2 sequence for bulk of computation

---

The documentation for this struct was generated from the following file:

• gemm/thread/mma_sm60.h