v3.9 update (#2203)

* v3.9 update

* voidD

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
This commit is contained in:
Yujia Zhai
2025-04-02 15:11:18 -04:00
committed by GitHub
co-authored by yuzhai
parent 62750a2b75
commit 6f4921858b
129 changed files with 7719 additions and 2036 deletions
+43
View File
@@ -115,3 +115,46 @@ TEST(SM90_CuTe_Hopper, CooperativeGemmTilingF16) {
}
#endif
#if defined(CUTE_ARCH_STSM_SM90_ENABLED)
TEST(SM90_CuTe_Hopper, CooperativeGemmSTSM) {
constexpr uint32_t thread_block_size = 128;
constexpr int MaxVecBits = 128;
using TA = cute::half_t;
using TB = cute::half_t;
using TC = cute::half_t;
auto tiled_mma =
TiledMMA<
MMA_Atom<SM80_16x8x16_F16F16F16F16_TN>,
Layout<Shape<_2, _2, _1>, Stride<_1, _2, _0>>,
Tile<_32, _32, _16>
>{};
auto global_a_layout = make_layout(Shape<_64, _64>{}, LayoutRight{});
auto global_b_layout = make_layout(Shape<_64, _64>{}, LayoutRight{});
auto global_c_layout = make_layout(Shape<_64, _64>{}, LayoutRight{});
test_cooperative_gemm<thread_block_size,
MaxVecBits,
TA, TB, TC>
(global_a_layout,
global_b_layout,
global_c_layout,
global_a_layout,
global_b_layout,
global_c_layout,
tiled_mma,
identity{},
identity{},
identity{},
identity{},
SM75_U32x4_LDSM_N{},
SM75_U32x4_LDSM_N{},
SM75_U32x4_LDSM_N{},
SM90_U32x4_STSM_N{});
}
#endif