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

Function local_partition

include/cute/tensor_impl.hpp:1077–1087  ·  view source on GitHub ↗

Source from the content-addressed store, hash-verified

1075template <class Tensor, class LShape, class LStride, class Index,
1076 __CUTE_REQUIRES(is_tensor<remove_cvref_t<Tensor>>::value)>
1077CUTE_HOST_DEVICE
1078auto
1079local_partition(Tensor && tensor,
1080 Layout<LShape,LStride> const& tile, // coord -> index
1081 Index const& index) // index to slice for
1082{
1083 static_assert(is_integral<Index>::value);
1084 return outer_partition(static_cast<Tensor&&>(tensor),
1085 product_each(shape(tile)),
1086 tile.get_flat_coord(index));
1087}
1088
1089// Same as above, but with a projection parameter to strip out unwanted tiling modes for convenience
1090// when using projections of the same tiler.

Callers 2

dequantize_kernelFunction · 0.50
cooperative_clearFunction · 0.50

Calls 5

outer_partitionFunction · 0.85
product_eachFunction · 0.70
shapeFunction · 0.70
diceFunction · 0.70
get_flat_coordMethod · 0.45

Tested by

no test coverage detected