update 3.8 v2 (#2112)

* update 3.8 v2

* update 3.8

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
This commit is contained in:
Yujia Zhai
2025-02-19 22:03:14 -05:00
committed by GitHub
co-authored by yuzhai
parent e9627ce55b
commit b84e9802d8
166 changed files with 3986 additions and 4037 deletions
-2
View File
@@ -86,5 +86,3 @@
#define CUTE_ARCH_FLOAT2_MATH_ENABLED
#endif
+1
View File
@@ -208,6 +208,7 @@ to_CUtensorMapDataType() {
if constexpr (is_same_v<T, uint8_t>) { return CU_TENSOR_MAP_DATA_TYPE_UINT8; } else
if constexpr (is_same_v<T, float_e4m3_t>) { return CU_TENSOR_MAP_DATA_TYPE_UINT8; } else
if constexpr (is_same_v<T, float_e5m2_t>) { return CU_TENSOR_MAP_DATA_TYPE_UINT8; } else
if constexpr (is_same_v<T, float_ue8m0_t>) { return CU_TENSOR_MAP_DATA_TYPE_UINT8; } else
if constexpr (is_same_v<T, type_erased_dynamic_float8_t>) { return CU_TENSOR_MAP_DATA_TYPE_UINT8;} else
if constexpr (is_same_v<T, uint16_t>) { return CU_TENSOR_MAP_DATA_TYPE_UINT16; } else
if constexpr (is_same_v<T, uint32_t>) { return CU_TENSOR_MAP_DATA_TYPE_UINT32; } else
+1 -1
View File
@@ -956,7 +956,7 @@ template <class a_type, class b_type, class c_type, class sf_type,
UMMA::ScaleIn a_neg = UMMA::ScaleIn::One, UMMA::ScaleIn b_neg = UMMA::ScaleIn::One>
struct SM100_MMA_MXF4_SS
{
static_assert(M == 128, "SM100_MMA_MXF4_SS M-mode size should be 128 for 1 CTA cluster OMMA.");
static_assert(M == 128, "SM100_MMA_MXF4_SS M-mode size should be 128 for 1 CTA cluster MMA.");
static_assert((N % 8 == 0) && (8 <= N) && (N <= 256), "SM100_MMA_MXF4_SS N-mode size should be a multiple of 8 between 8 and 256.");
static_assert((VS == 16) || (VS == 32), "SM100_MMA_MXF4_SS Vector size can only be 16 or 32.");