-
Notifications
You must be signed in to change notification settings - Fork 540
Pull requests: ROCm/aiter
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[HIP] Fix rmsnorm_quant cross-row reads and writes for rows not a multiple of 4 bytes
HIP
#5290
opened Sep 5, 2026 by
rk9595
Loading…
review-pr: gates that can be checked, and 200 PRs of evidence about them
#5289
opened Sep 5, 2026 by
zufayu
Contributor
Loading…
[FlyDSL] Rewrite gfx950 HGEMM test to op_test standard
FlyDSL
#5287
opened Sep 5, 2026 by
xytpai
Contributor
Loading…
[FlyDSL] [DSv4] Bound the FP4 MQA-logits store with a window-sized V# instead of a compare
FlyDSL
#5285
opened Sep 5, 2026 by
valarLip
Collaborator
Loading…
3 of 4 tasks
[Config] tune(dsv4): gfx950 FP8 blockscale bpreshuffle for wq_b / wqkv_a
Config
#5283
opened Sep 5, 2026 by
lixiufei-leo
Loading…
1 task
[FlyDSL][DSv4] Fuse FP4 MQA scoring with streaming TopK
#5282
opened Sep 5, 2026 by
AMD-yanfeiwang
Contributor
•
Draft
[CK] [JIT] feat(cktile a8w8-bpreshuffle): add rowcol_wp_v2 as a selectable kernel type
CK
JIT
#5280
opened Sep 5, 2026 by
ThomasNing
Contributor
Loading…
[Config] [AMD][DSV4] [tune][gfx950] wo_b/wq_b a8w8 blockscale bpreshuffle configs
Config
#5279
opened Sep 4, 2026 by
karverma-amd
Contributor
Loading…
[Config] Add gfx950 a8w8_blockscale tunings for GLM-5.3 shapes (13-61% on low M)
Config
#5278
opened Sep 4, 2026 by
stefanskiasan
Loading…
[Config] [FMoE] Tuned bf16 MoE for K2 horizon 375B
Config
#5275
opened Sep 4, 2026 by
a-sidorova
Contributor
Loading…
1 task done
[FlyDSL] [gfx1250] Quantize each MoE source token once instead of once per route
FlyDSL
#5274
opened Sep 4, 2026 by
XingerZhu
Contributor
Loading…
1 task
[HIP] Perf/gfx1250 per group quant
HIP
#5273
opened Sep 4, 2026 by
yzhou103
Contributor
Loading…
1 task
[HIP] fea: add LL, LL128 proto into gfx9
ci:atom
HIP
#5272
opened Sep 4, 2026 by
TennyWang1223
Contributor
Loading…
1 task
fix(mega_moe): stop the reference clamping SwiGLU at swiglu_limit=0
#5271
opened Sep 4, 2026 by
JohnQinAMD
Contributor
Loading…
3 tasks done
fix(tune): default batch size silently disables the tuner's own checkpointing
#5270
opened Sep 4, 2026 by
ThomasNing
Contributor
Loading…
[HIP] [OPUS] [JIT] fix(jit): baton liveness is undecidable from pid+host under --network host
HIP
JIT
OPUS
#5269
opened Sep 3, 2026 by
ThomasNing
Contributor
Loading…
feat(tune): record untuned shapes for every GEMM family, not just bf16
#5268
opened Sep 3, 2026 by
ThomasNing
Contributor
Loading…
[HIP] [CK] perf(tuner): build the result frame once instead of concatenating per row
CK
HIP
#5262
opened Sep 3, 2026 by
ThomasNing
Contributor
Loading…
[Triton/Gluon] split K correct OOB afp4wfp4
Triton/Gluon
#5261
opened Sep 3, 2026 by
afriedri
Loading…
[FlyDSL] Add tuned BF16 GEMM configs for GLM-5.2 decode shapes (N=6144)
FlyDSL
#5260
opened Sep 3, 2026 by
nehaprakriya
Loading…
1 task done
[HIP] [CK] [MoE] Reject non-int32 index buffers in the topk kernels
CK
HIP
#5255
opened Sep 3, 2026 by
i-kosarev
Loading…
[Triton/Gluon] [FlyDSL] [CI] Group the delta-rule ops under a linear_attention family
CI
FlyDSL
Triton/Gluon
#5254
opened Sep 3, 2026 by
Boss2002n
Contributor
Loading…
[Triton/Gluon] chunk_kimi_delta_attn: accept a non-fp32 KDA state
Triton/Gluon
#5249
opened Sep 3, 2026 by
XiaobingSuper
Contributor
Loading…
[Triton/Gluon] [Conv2D] Add Conv2D configuration files for gfx1101 (RDNA3) and gfx1150 (RDNA3.5)
Triton/Gluon
#5246
opened Sep 3, 2026 by
Ragua1
Loading…
Previous Next
ProTip!
Filter pull requests by the default branch with base:main.