|
CUTLASS_HOST_DEVICE | LinearScalingDevicePtr (Params const &_params) |
| Ctor. More...
|
|
CUTLASS_DEVICE | LinearScaling () |
| Ctor. More...
|
|
CUTLASS_DEVICE | LinearScaling (Params const &_params) |
| Ctor. More...
|
|
CUTLASS_DEVICE bool | source_required () const |
|
template<typename FragmentA_ , typename FragmentB_ > |
CUTLASS_DEVICE void | evaluate (FragmentA_ const &accum, FragmentB_ &output) |
| Evaluate the functor. More...
|
|
template<typename ScalarAccum , typename ScalarOutput , int size> |
CUTLASS_DEVICE void | evaluate (ScalarAccum const *accum, ScalarOutput *output) |
| Evaluate the functor, without using fragment in the API. More...
|
|
template<typename FragmentA_ , typename FragmentB_ > |
CUTLASS_DEVICE void | evaluate (FragmentA_ const &accum, FragmentB_ const &old, FragmentB_ &output) |
| Evaluate the functor. More...
|
|
template<typename ScalarAccum , typename ScalarOutput , int size> |
CUTLASS_DEVICE void | evaluate (ScalarAccum const *accum, ScalarOutput const *old, ScalarOutput *output) |
| Evaluate the functor, without using fragment in the API. More...
|
|
template<typename Scalar_, typename FragmentMultiplyAdd_ = FragmentMultiplyAdd<Scalar_, Scalar_>>
struct cutlass::gemm::LinearScalingDevicePtr< Scalar_, FragmentMultiplyAdd_ >
Functor to compute linear combination of fragments. This is intended to support passing scalars either by value from the host or by reference to device-side scalar elements. This is inspired by cuBLAS's device pointer mode.