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

v4.5.1 update. (#3237)

Browse Source
This commit is contained in:
Junkai-Wu
2026-05-19 10:35:08 +08:00
committed by GitHub
parent e406c186f5
commit 982cb9e718
42 changed files with 6487 additions and 336 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
include/cute/arch/mma_sm100_umma.hpp
View File

@@ -1207,6 +1207,10 @@ struct SM100_MMA_S8_2x1SM_SS_SPARSE
}
};
template <class a_type, class b_type, class c_type, int M, int N,
UMMA::Major a_major, UMMA::Major b_major,
UMMA::ScaleIn a_neg = UMMA::ScaleIn::One,
UMMA::ScaleIn b_neg = UMMA::ScaleIn::One>
struct SM100_MMA_F8F6F4_SS
{
using DRegisters = void;
@@ -1452,6 +1456,10 @@ struct SM100_MMA_MXF8F6F4_SS_SPARSE
}
};
template <class a_type, class b_type, class c_type, int M, int N,
UMMA::Major a_major, UMMA::Major b_major,
UMMA::ScaleIn a_neg = UMMA::ScaleIn::One,
UMMA::ScaleIn b_neg = UMMA::ScaleIn::One>
struct SM100_MMA_F8F6F4_2x1SM_SS
{
using DRegisters = void;

26
include/cute/atom/mma_traits_sm100.hpp
View File

@@ -3327,12 +3327,9 @@ struct MMA_Traits<SM100_MMA_S8_2x1SM_SS_SPARSE<a_type, b_type, c_type,
template <class a_type, class b_type, class c_type,
int M, int N, UMMA::Major a_major, UMMA::Major b_major,
UMMA::ScaleIn a_neg, UMMA::ScaleIn b_neg>
struct MMA_Traits<SM100_MMA_F8F6F4_SS, a_type, b_type, c_type,
cute::C<M>, cute::C<N>,
cute::integral_constant<UMMA::Major, a_major>,
cute::integral_constant<UMMA::Major, b_major>,
cute::integral_constant<UMMA::ScaleIn, a_neg>,
cute::integral_constant<UMMA::ScaleIn, b_neg>>
struct MMA_Traits<SM100_MMA_F8F6F4_SS<a_type, b_type, c_type,
M, N, a_major, b_major,
a_neg, b_neg>>
{
using ValTypeD = c_type;
using ValTypeA = a_type;
@@ -3390,7 +3387,9 @@ struct MMA_Traits<SM100_MMA_F8F6F4_SS, a_type, b_type, c_type,
uint32_t tmem_c = raw_pointer_cast(D.data());
uint64_t idesc = UMMA::make_runtime_instr_desc<>(traits.idesc_);
SM100_MMA_F8F6F4_SS::fma(desc_a, desc_b, tmem_c, uint32_t(traits.accumulate_), idesc);
SM100_MMA_F8F6F4_SS<a_type, b_type, c_type,
M, N, a_major, b_major,
a_neg, b_neg>::fma(desc_a, desc_b, tmem_c, uint32_t(traits.accumulate_), idesc);
}
};
@@ -3745,12 +3744,9 @@ struct MMA_Traits<SM100_MMA_F8F6F4_SS_SPARSE<a_type, b_type, c_type,
template <class a_type, class b_type, class c_type,
int M, int N, UMMA::Major a_major, UMMA::Major b_major,
UMMA::ScaleIn a_neg, UMMA::ScaleIn b_neg>
struct MMA_Traits<SM100_MMA_F8F6F4_2x1SM_SS, a_type, b_type, c_type,
cute::C<M>, cute::C<N>,
cute::integral_constant<UMMA::Major, a_major>,
cute::integral_constant<UMMA::Major, b_major>,
cute::integral_constant<UMMA::ScaleIn, a_neg>,
cute::integral_constant<UMMA::ScaleIn, b_neg>>
struct MMA_Traits<SM100_MMA_F8F6F4_2x1SM_SS<a_type, b_type, c_type,
M, N, a_major, b_major,
a_neg, b_neg>>
{
using ValTypeD = c_type;
using ValTypeA = a_type;
@@ -3808,7 +3804,9 @@ struct MMA_Traits<SM100_MMA_F8F6F4_2x1SM_SS, a_type, b_type, c_type,
uint32_t tmem_c = raw_pointer_cast(D.data());
uint64_t idesc = UMMA::make_runtime_instr_desc<>(traits.idesc_);
SM100_MMA_F8F6F4_2x1SM_SS::fma(desc_a, desc_b, tmem_c, uint32_t(traits.accumulate_), idesc);
SM100_MMA_F8F6F4_2x1SM_SS<a_type, b_type, c_type,
M, N, a_major, b_major,
a_neg, b_neg>::fma(desc_a, desc_b, tmem_c, uint32_t(traits.accumulate_), idesc);
}
};