leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 28 Packages Projects Releases Wiki Activity

Release v4.0.0 (#2294)

Browse Source
This commit is contained in:
Kihiro Bando
2025-05-13 15:55:29 -04:00
committed by GitHub
parent ad7b2f5e84
commit f115c3f854
299 changed files with 51495 additions and 4413 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cute/atom/copy_traits_sm100_tma.hpp
+2 -2
View File
@@ -104,7 +104,7 @@ struct Copy_Traits<SM100_TMA_2SM_LOAD, NumBitsPerTMA, AuxParams_>
auto
get_tma_tensor(GShape const& g_shape) const {
static_assert(is_congruent<decltype(g_shape), decltype(aux_params_.g_stride_)>::value);
return make_counting_tensor(make_layout(g_shape, aux_params_.g_stride_));
return make_coord_tensor(make_layout(g_shape, aux_params_.g_stride_));
}
// Don't try to execute a copy with SM100_TMA_2SM_LOAD before calling .with()
@@ -192,7 +192,7 @@ struct Copy_Traits<SM100_TMA_2SM_LOAD_MULTICAST, NumBitsPerTMA, AuxParams_>
auto
get_tma_tensor(GShape const& g_shape) const {
static_assert(is_congruent<decltype(g_shape), decltype(aux_params_.g_stride_)>::value);
return make_counting_tensor(make_layout(g_shape, aux_params_.g_stride_));
return make_coord_tensor(make_layout(g_shape, aux_params_.g_stride_));
}
// Don't try to execute a copy with SM100_TMA_2SM_LOAD_MULTICAST_OP before calling .with()
include/cute/atom/copy_traits_sm90_tma.hpp
+4 -4
View File
@@ -146,7 +146,7 @@ struct Copy_Traits<SM90_TMA_LOAD, NumBitsPerTMA, AuxParams_>
auto
get_tma_tensor(GShape const& g_shape) const {
static_assert(is_congruent<decltype(g_shape), decltype(aux_params_.g_stride_)>::value);
return make_counting_tensor(make_layout(g_shape, aux_params_.g_stride_));
return make_coord_tensor(make_layout(g_shape, aux_params_.g_stride_));
}
// Don't try to execute a copy with SM90_TMA_LOAD before calling .with()
@@ -276,7 +276,7 @@ struct Copy_Traits<SM90_TMA_LOAD_MULTICAST, NumBitsPerTMA, AuxParams_>
auto
get_tma_tensor(GShape const& g_shape) const {
static_assert(is_congruent<decltype(g_shape), decltype(aux_params_.g_stride_)>::value);
return make_counting_tensor(make_layout(g_shape, aux_params_.g_stride_));
return make_coord_tensor(make_layout(g_shape, aux_params_.g_stride_));
}
// Don't try to execute a copy with SM90_TMA_LOAD_MULTICAST before calling .with()
@@ -350,7 +350,7 @@ struct Copy_Traits<SM90_TMA_STORE, NumBitsPerTMA, AuxParams_>
auto
get_tma_tensor(GShape const& g_shape) const {
static_assert(is_congruent<decltype(g_shape), decltype(aux_params_.g_stride_)>::value);
return make_counting_tensor(make_layout(g_shape, aux_params_.g_stride_));
return make_coord_tensor(make_layout(g_shape, aux_params_.g_stride_));
}
// Construct new TMA_STORE with (unsafe) swapped out TMA descriptor ptr (for grouped gemm/ptr array gemm)
@@ -463,7 +463,7 @@ struct Copy_Traits<SM90_TMA_REDUCE_ADD, NumBitsPerTMA, AuxParams_>
auto
get_tma_tensor(GShape const& g_shape) const {
static_assert(is_congruent<decltype(g_shape), decltype(aux_params_.g_stride_)>::value);
return make_counting_tensor(make_layout(g_shape, aux_params_.g_stride_));
return make_coord_tensor(make_layout(g_shape, aux_params_.g_stride_));
}
template <class Coord, int... Is>