Default Branch

cb859854a7 · [rocm-libraries] ROCm/rocm-libraries#6526 (commit 3e01710) · Updated 2026-07-13 10:44:45 +00:00

Branches

86ed46ca60 · remove invalid instance for MI300 · Updated 2026-04-20 10:59:09 +00:00

401
161

3e3cb36c7a · [fmha-bwd] Fix dK/dV left uninitialized for zero-length Q batches in group persistent · Updated 2026-04-17 07:54:52 +00:00

3455
4047

7182f6d139 · Add missing gfx1033 to gfx103 group definition · Updated 2026-04-09 14:17:41 +00:00

526
1

966302f00c · Update CMakeLists.txt · Updated 2026-04-08 18:56:25 +00:00

401
167

5e1846914a · Add missing gfx1033 to gfx103 group definition · Updated 2026-04-07 19:15:32 +00:00

526
1

cdf6573171 · Add missing gfx1033 to gfx103 group definition · Updated 2026-04-07 18:50:34 +00:00

1317
2

a586a1f8bd · [rocm-libraries] ROCm/rocm-libraries#6135 (commit 91f0518) · Updated 2026-04-03 22:08:33 +00:00

401
156

a586a1f8bd · [rocm-libraries] ROCm/rocm-libraries#6135 (commit 91f0518) · Updated 2026-04-03 22:08:33 +00:00

401
156

d9ecf29860 · Add float sink_v to the operator() parameters for qs_ks_vs and whole_k_prefetch pipelines · Updated 2026-04-01 08:50:21 +00:00

3455
4057

357a140e7b · [CK_TILE] Add pooling in tile_engine (#4469) · Updated 2026-04-01 07:31:46 +00:00

3455
4021

f6e7493055 · Single wave WarpGemmMfmaF16F16F32M32N32K16 tutorial · Updated 2026-03-31 20:12:17 +00:00

877
19

4af7e472a3 · Add unified attention kernel on top of CK develop · Updated 2026-03-30 17:37:12 +00:00

401
122

86f7ebcf27 · Add bs32 narrow tier: 2-warp 16x16 MFMA kBlockM=32 for decode · Updated 2026-03-30 17:04:38 +00:00

535
120

6004d045ea · docs(fmha): explain the difference between fp8, fp8bf16 and fp8fp32 precision modes · Updated 2026-03-30 14:43:36 +00:00

401
138

e6b8094f94 · [rocm-libraries] ROCm/rocm-libraries#5921 (commit 032ac1b) · Updated 2026-03-30 14:20:20 +00:00

401
136

7988f0a3ac · feat(tutorial): add 01_naive_gemm tutorial example · Updated 2026-03-27 20:25:03 +00:00

1459
8

bbc748defe · Add unified attention d64/GQA-8 kernel instances and fix BLOCK_SIZE for small head dims · Updated 2026-03-27 14:41:10 +00:00

535
107

42768785a4 · Update rocm-docs-core to 1.33.1 · Updated 2026-03-23 11:54:02 +00:00

849
5

a797c6ee1a · fix: address GPU memory segmentation fault caused by int32 overflow · Updated 2026-03-13 16:54:22 +00:00

732
1

a075f2f2e8 · Fix nan when seqq=0 · Updated 2026-03-12 08:33:26 +00:00

401
62