[CK_TILE] support hdim=192/128 pair for deepseekv3 (#1961)

* support hdim=192/128 pair

* remove useless print

* update
This commit is contained in:
carlushuang
2025-03-11 21:07:40 +08:00
committed by GitHub
parent 0db7c8f0b2
commit 7a93b16ff6
8 changed files with 35 additions and 8 deletions

View File

@@ -252,3 +252,11 @@ CK_TILE_DECLARE_ENV_VAR_BOOL(CK_TILE_LOGGING)
#else // for GPU code
#define CK_TILE_USE_OCP_FP8 0
#endif
#ifndef CK_TILE_USE_BUFFER_ADDRESSING_BUILTIN
#if __clang_major__ >= 20
#define CK_TILE_USE_BUFFER_ADDRESSING_BUILTIN 1
#else
#define CK_TILE_USE_BUFFER_ADDRESSING_BUILTIN 0
#endif
#endif