[CI/Build] Enable FP8 on NVIDIA Thor (#39712)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-04-29 09:48:52 -07:00
committed by GitHub
parent a80d6f150c
commit 4a42aba380
4 changed files with 13 additions and 38 deletions
@@ -141,7 +141,7 @@ struct cutlass_3x_gemm_sm100 {
sizeof(typename CollectiveEpilogue::SharedStorage))>,
KernelSchedule>::CollectiveOp;
using GemmKernel = enable_sm100f_only<cutlass::gemm::kernel::GemmUniversal<
using GemmKernel = enable_sm100_to_sm120<cutlass::gemm::kernel::GemmUniversal<
Shape<int, int, int, int>, CollectiveMainloop, CollectiveEpilogue, void>>;
};
@@ -125,7 +125,7 @@ struct cutlass_3x_gemm_fp8_blockwise {
MainloopScheduler
>::CollectiveOp>;
using KernelType = enable_sm100f_only<cutlass::gemm::kernel::GemmUniversal<
using KernelType = enable_sm100_to_sm120<cutlass::gemm::kernel::GemmUniversal<
Shape<int, int, int, int>, CollectiveMainloop, CollectiveEpilogue>>;
struct GemmKernel : public KernelType {};
@@ -92,7 +92,7 @@ struct cutlass_3x_gemm_sm100_fp8 {
// -----------------------------------------------------------
// Kernel definition
// -----------------------------------------------------------
using GemmKernel = enable_sm100f_only<cutlass::gemm::kernel::GemmUniversal<
using GemmKernel = enable_sm100_to_sm120<cutlass::gemm::kernel::GemmUniversal<
Shape<int, int, int, int>, CollectiveMainloop, CollectiveEpilogue, void>>;
};