[CK_TILE] Fix mock token id, support g1u1/g1u0 through same inline code block (#1808)

* fix mock token id

* prepare host for g1u1

* reformat inline-asm

* restructure uk_0

* restructure gate_up

* done

* change default to init=1

* update readme

* fix a bug in interleave pipeline

* rcp for silu
This commit is contained in:
carlushuang
2025-01-16 17:51:10 +08:00
committed by GitHub
parent 8c29e06f3c
commit 1ff50e78c6
20 changed files with 1927 additions and 1413 deletions

View File

@@ -1,3 +1,10 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
// clang-format off
// define the CK_TILE_** macro before include this file to change kernel variation
// we will undef everything defined in this file
#ifndef CK_TILE_FLATMM_UK_MFMA
#define CK_TILE_FLATMM_UK_MFMA CK_TILE_FLATMM_UK_MFMA_BF16
#endif
@@ -816,3 +823,5 @@
#undef _UK_MFMA_
#undef _UK_PK_CVT_
#undef _UK_ATOMIC_ADD_
#undef CK_TILE_FLATMM_UK_MFMA
// clang-format on