joyeamd 94d47b1680 fmha hdim256 vectorize improve (#2086) ...

For hdim 256, will not have vectorized buffer load when seqlen % 256 != 0 and hdim % 256 = 0; this commit tries to solve this condition.

2025-04-16 09:21:04 +08:00

..

01_fmha

fmha hdim256 vectorize improve (#2086)

2025-04-16 09:21:04 +08:00

02_layernorm2d

[CK_TILE] Improve RMS/Layer Normalization 2 Pass Pipeline Performance (#1861)

2025-03-25 20:09:45 +08:00

03_gemm

[CK_TILE] Add 2:4 structured sparsity support for fp16 gemm (#1957)

2025-04-11 12:18:26 +02:00

04_img2col

[CK_TILE] Update example README files & fix script compatibility issue (#1548)

2024-10-08 10:45:12 +08:00

05_reduce

[CK_TILE] Adjust kBlockSize of reduce example for better perf (#1779)

2025-01-12 20:50:32 -08:00

06_permute

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

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

09_topk_softmax

topk_softmax (#1592)

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

10_rmsnorm2d

[CK_TILE] Improve RMS/Layer Normalization 2 Pass Pipeline Performance (#1861)

2025-03-25 20:09:45 +08:00

11_add_rmsnorm2d_rdquant

[A8W8 GEMM] Optimized weight-preshuffled implementation & add quantization datatype for CK TILE rms_norm (#1862)

2025-02-20 14:00:27 -08:00

12_smoothquant

[CK_TILE] Add Various Fusion Functions to RMSNorm (#1802)

2025-01-15 10:23:48 +08:00

13_moe_sorting

hotfix fix sorting int64 (#2025)

2025-03-28 11:31:52 +08:00

14_moe_smoothquant

add fp8 as dst (#1830)

2025-01-22 17:34:27 +08:00

15_fused_moe

hotfix fix sorting int64 (#2025)

2025-03-28 11:31:52 +08:00

16_batched_gemm

[CK_TILE] Switch to universal gemm for batched and grouped gemms (#1919)

2025-03-20 11:17:04 +01:00

17_grouped_gemm

[CK_TILE] Switch to universal gemm for batched and grouped gemms (#1919)

2025-03-20 11:17:04 +01:00

35_batched_transpose

add batched_transpose implement (#1660)

2025-01-28 16:22:02 -08:00

CMakeLists.txt

add batched_transpose implement (#1660)

2025-01-28 16:22:02 -08:00

remod.py

introducing ck_tile! (#1216)

2024-04-15 19:27:12 -05:00