All pull requests

Pull requests

• [ASM] [HIP] [JIT] Clamp bf16 MLA decode kv_indices and move gfx950 kernels to opus/gfx950/

  ASMHIPJIT
  #7395·kaiyang-1 opened Nov 3, 2026Contributor·

  ·
  2

• MHAv4 gfx942 parity

  #7394·nsakkine opened Nov 3, 2026Contributor·

  ·
  2

• [CI] ci: avoid build runner for disabled wheel jobs

  CI
  #7392·gyohuangxin opened Nov 3, 2026Member·

  ·
  3

• [Triton/Gluon] gfx950 a16w16: Stream-K for compute-bound shapes (draft)

  #7391·jamesbowley opened Nov 3, 2026·

  ·

• [ASM] [HIP] Fold gfx1250 MLA v4.1 sparse prefill split-K in-kernel with workgroup clusters

  ASMHIP
  #7390·kaiyang-1 opened Nov 3, 2026Contributor·

  ·
  5

• [Config] [Perf] Add DeepSeek V4.1 EP16 A4W4 MoE tuning configs

  Config
  #7389·james-huang09 opened Nov 3, 2026Collaborator·

  ·
  2

• [FlyDSL] [Misc] pa_decode: key the tile kernel on FlyDSL dtypes, not dtype strings and flags

  #7388·fsx950223 opened Nov 3, 2026Contributor·4 of 10 tasks·

  ·

• [HIP] [OPUS] Add OPUS GEMM BlockScale_Mxfp8 GEMM for M <=1024 Shape.

  HIPOPUS
  #7387·Fyzyukk opened Nov 3, 2026Contributor·

  ·
  2

• [Bugfix] Allow gfx942 FlyDSL a16w4 and a16wi4 MoE tuning

  #7386·amd-oshkarav opened Nov 3, 2026·2 of 10 tasks·

  ·
  2

• [CI] ci: cache CK submodule checkout

  CI
  #7385·gyohuangxin opened Nov 3, 2026Member·

  ·
  5

• [FlyDSL] gfx1250 PA decode: BF16 wave-pair pipeline at D64, drop redundant D128 checks

  #7384·zhiding512 opened Nov 3, 2026Contributor·

  ·
  1

• Add gfx950 a8w8 blockscale GEMM tunings for Qwen3-14B-FP8

  #7383·xiaofei-zheng opened Nov 3, 2026Contributor·4 of 10 tasks·

  ·
  19

• [Triton] [Config] Add gfx1201 gemm_a16w16_gated config

  #7382·klin2024 opened Nov 3, 2026Contributor·

  ·
  2

• [Config] [Tune] GLM-5.3 a8w8 blockscale GEMM rows for gfx950

  #7381·amd-sriram opened Nov 3, 2026Contributor·3 of 6 tasks·

  ·
  2

• [Triton] gfx950 unified attention: do not specialize on num_seqs (+ reduce_segments)

  #7380·vorapolsiloai opened Nov 3, 2026Contributor·

  ·
  2

• [FlyDSL] [AOT] Build gfx1250-strict wheels from the gfx1250 kernel family

  FlyDSL
  #7379·zufayu opened Nov 3, 2026Contributor·

  ·
  4

• [Triton] [Config] Add gfx1201 gemm_a16w16 config

  #7378·klin2024 opened Nov 3, 2026Contributor·

  ·
  2

• Qsa index scorer/stage2

  #7377·demonsan opened Nov 3, 2026Contributor·

  ·
  2

• [HIP] [Bugfix] Skip grouped top-k launches with zero tokens

  HIP
  #7376·XinyuJiangCMU opened Nov 3, 2026Contributor·

  ·
  3

• [ASM] [HIP] [JIT] [gfx1250] asm mha fp8 fwd: add D192x128 per-tensor varlen kernel

  ASMHIPJIT
  #7375·shay-li77 opened Nov 3, 2026Contributor·6 of 11 tasks·

  ·
  10

• [HIP] [JIT] Add gfx950 K-pool indexer cache writers

  HIPJIT
  #7374·sumin-hong opened Nov 3, 2026·

  ·
  8

• [HIP] [Bugfix] sample_kernels: partial-vector and prologue-tail defects

  HIP
  #7373·yitingw1 opened Nov 3, 2026Contributor·1 task done·

  ·
  7

• [HIP] [FlyDSL] conv3d gfx950: kernel-only timing in tuner and tables, memoize fp32 bias

  #7372·huizzhan opened Nov 3, 2026Contributor·

  ·
  2

• [ASM] perf(pa): update gfx950 unified MTP code objects

  ASM
  #7371·fangche123 opened Nov 3, 2026Contributor·10 tasks·

  ·
  3

• [Triton/Gluon] [Config] Add gfx1201 RMSNorm large-M/small-N config

  #7369·klin2024 opened Nov 3, 2026Contributor·

  ·
  2