MCPcopy Create free account
hub / github.com/NVIDIA/cutlass / copy_aligned

Function copy_aligned

include/cute/algorithm/copy.hpp:331–342  ·  view source on GitHub ↗

Source from the content-addressed store, hash-verified

329template <class SrcEngine, class SrcLayout,
330 class DstEngine, class DstLayout>
331CUTE_HOST_DEVICE
332void
333copy_aligned(Tensor<SrcEngine, SrcLayout> const& src,
334 Tensor<DstEngine, DstLayout> & dst)
335{
336 if constexpr (is_static<decltype(shape(src))>::value && is_static<decltype(shape(dst))>::value) {
337 // Tensors with static shapes can be filtered
338 return copy(AutoFilter(AutoVectorizingCopyWithAssumedAlignment<128>{}), src, dst);
339 } else {
340 return copy(AutoVectorizingCopyWithAssumedAlignment<128>{}, src, dst);
341 }
342}
343
344// Specializaton for Atom AutoVectorizingCopyAssumedAlignment
345template <int MaxVecBits, class... Args,

Callers 9

computeMethod · 0.85
softmaxMethod · 0.85
computeMethod · 0.85
reduceMethod · 0.85
reduceMethod · 0.85
begin_loopMethod · 0.85
reduceMethod · 0.85
reduceMethod · 0.85
previsitMethod · 0.85

Calls 3

AutoFilterClass · 0.85
copyFunction · 0.70
shapeFunction · 0.50

Tested by 9

computeMethod · 0.68
softmaxMethod · 0.68
computeMethod · 0.68
reduceMethod · 0.68
reduceMethod · 0.68
begin_loopMethod · 0.68
reduceMethod · 0.68
reduceMethod · 0.68
previsitMethod · 0.68