schedule hootloop, change blockK to 128 for fp8

This commit is contained in:
AMD-dteng
2025-04-08 17:57:30 +08:00
parent b5f91e13bb
commit 855375c766
2 changed files with 45 additions and 2 deletions

View File

@@ -31,7 +31,7 @@ float flatmm_calc(const ck_tile::FlatmmHostArgs& args, const ck_tile::stream_con
// This part comes from the Codegen
constexpr ck_tile::index_t M_Tile = 128;
constexpr ck_tile::index_t N_Tile = 128;
constexpr ck_tile::index_t K_Tile = 64;
constexpr ck_tile::index_t K_Tile = 128;
constexpr ck_tile::index_t M_Warp = 1;
constexpr ck_tile::index_t N_Warp = 4;