Qianfeng 3d50f57f43 Update for fmha_fwd qs_ks_vs pipeline (#1810) ...

* Update for fmha_fwd qs_ks_vs pipeline

* Remove _builtin_amdgcn_sched_barrier(0)

* Move p_compute to p converting earlier for trying to increase vgprs re-using

* Enable GetQKBlockGemm to use WarpGemm-16x16x16 for QLoadOnce==false situation

* Re-add __builtin_amdgcn_sched_barrier(0)

---------

Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>

2025-01-13 12:43:05 +08:00

..

add_rmsnorm2d_rdquant

support max3 in smoothquant and add+ rmsnorm + rdquant (#1654)

2024-11-27 05:01:15 +08:00

common

[Ck tile] layernorm2d fwd optimize (#1637)

2024-11-08 12:28:23 +08:00

elementwise

[CK_TILE] fused-moe first version (#1634)

2024-11-26 11:14:56 +08:00

epilogue

[CK TILE] GEMM and Batched GEMM SplitK support (#1724)

2024-12-28 14:40:17 +01:00

flatmm/block

Enhance printing functionality (#1751)

2024-12-17 09:19:44 +01:00

fmha

Update for fmha_fwd qs_ks_vs pipeline (#1810)

2025-01-13 12:43:05 +08:00

fused_moe

[CK_TILE] optimize moe-sorting kernel (#1771)

2024-12-23 10:59:02 +08:00

gemm

[CK TILE] GEMM and Batched GEMM SplitK support (#1724)

2024-12-28 14:40:17 +01:00

image_to_column

[CK_TILE] Image to Column kernel (#1532)

2024-09-27 22:57:38 +02:00

layernorm2d

enable bias feature that add bias before adding residual (for rtpllm project) (#1741)

2025-01-08 17:51:06 +08:00

norm_reduce

Ck tile/layernorm: implement naive reduce, opt performance (#1784)

2025-01-03 14:28:59 +08:00

permute

Remove virtual destructors from unary ops (#1610)

2024-10-30 17:42:50 +01:00

reduce/block

[Ck_tile] smoothquant (#1617)

2024-11-01 13:51:56 +08:00

rmsnorm2d

[Ck_tile] smoothquant (#1617)

2024-11-01 13:51:56 +08:00

smoothquant

Ck tile/smoothquant out stride (#1742)

2024-12-13 11:53:52 +08:00

softmax/block

topk_softmax (#1592)

2024-10-26 23:52:49 +08:00

topk/block

topk_softmax (#1592)

2024-10-26 23:52:49 +08:00

topk_softmax

topk_softmax (#1592)

2024-10-26 23:52:49 +08:00

add_rmsnorm2d_rdquant.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

common.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

elementwise.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

epilogue.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

flatmm.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

fmha.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

fused_moe.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

gemm.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

image_to_column.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

layernorm2d.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

norm_reduce.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

permute.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

reduce.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

rmsnorm2d.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

smoothquant.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

softmax.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

topk_softmax.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00

topk.hpp

[CK_TILE] fmha fwd splitkv optimization for decode (seqlen_q=1) (#1789)

2025-01-07 18:49:24 +08:00