All pull requests

Pull requests

• [FlagMega] Address paged-attention combine state through the result's coordinates

  COREfeature/flagmega
  #1434·zhangnju opened Oct 23, 2026Contributor·

  ·

• [TLE][CommonIR] Support explicit tile copy extents with targeted validation

  ascendCOREtletriton_v3.5.x
  #1433·togepro opened Oct 23, 2026Collaborator·

  ·

• [FIX][Ascend][XCS] Allow tl.where to take pointers as operands

  ascendCOREtriton_v3.5.x
  #1430·finneyyan opened Oct 23, 2026Collaborator·

  ·

• [TLE][LANGUAGE] Add pad_value to tle.gpu.copy

  COREmaintle
  #1429·i4oolish opened Oct 23, 2026Collaborator·

  ·

• [TLE][CommonIR] Add TensorView tile load/store lowering

  ascendCOREtletriton_v3.5.x
  #1426·togepro opened Oct 22, 2026Collaborator·

  ·

• [KMCompiler][TLERaw] Add support for passing tensor fragment arguments.

  COREiluvatarmainnvidiatle
  #1425·PixelLantern opened Oct 22, 2026·

  ·

• [TLE] Update is_tle_supported and require_tle api

  CI/CDCOREDOCmaintileirtle
  #1423·zhzhcookie opened Oct 22, 2026Collaborator·

  ·

• [SPEC][XPU] Specialize XPU Python sources

  mainxpu
  #1422·Stardep-lmc opened Oct 22, 2026Collaborator·

  ·

• [KMCompiler] support custom types in kernel(kernel arguments, field access, etc)

  amdCI/CDCOREDOCenflamehcuiluvatarmainmetaxmthreadsnvidiappurpuspacemitsunrisethrivetileirtsingmicroxpu
  #1421·zhongsanming opened Oct 22, 2026Contributor·

  ·

• [HCU] Add FP64 MMAC support

  hcumain
  #1419·ChunhaoSun opened Oct 21, 2026Collaborator·

  ·
  1

• Skip nvidia-sm90 runtime tests on non-SM90 devices

  COREfeature/flagmega
  #1418·zhangnju opened Oct 21, 2026Contributor·

  ·

• packaging(deb): build one package per backend from a backend table

  CI/CDDOCGITmain
  #1417·shiptux opened Oct 21, 2026Contributor·

  ·
  3

• [TLE][Ascend] Add raw cannshmem dialect path and tutorials

  ascendCOREtletriton_v3.5.x
  #1414·i3wanna2 opened Oct 20, 2026Collaborator·

  ·

• [Ascend][XCS] Fix lightning-indexer UB overflow

  ascendCI/CDCOREtletriton_v3.5.x
  #1413·finneyyan opened Oct 20, 2026Collaborator·

  ·

• [TLE][BACKEND][TEST] Add set_layout support for HCU

  hcumaintle
  #1410·YOOkoishi opened Oct 20, 2026·

  ·

• [TLE][BACKEND][TEST] Add set_layout support for Iluvatar

  CI/CDCOREiluvatarmaintle
  #1409·YOOkoishi opened Oct 20, 2026·

  ·

• [TLE][Ascend][XCS] Fix the precision error of swiglu

  ascendCI/CDtletriton_v3.5.x
  #1405·finneyyan opened Oct 19, 2026Collaborator·

  ·

• [FlagMega] Pass the pool-ABI bank row index as a tl value

  COREfeature/flagmega
  #1404·zhangnju opened Oct 19, 2026Contributor·

  ·
  1

• [FlagMega] Load source-rdata on ROCm via the HIP host-register API

  COREfeature/flagmega
  #1403·zhangnju opened Oct 19, 2026Contributor·

  ·
  1

• [QC][TLE][MTHREADS] Support mixed FP8/BF16 SQMMA pipelines on PH1

  COREmainmthreadstle
  #1401·July-h5kf3 opened Oct 19, 2026Contributor·

  ·

• [TLE][BACKEND] Fall back to an installed flagcx package when the wheel bundles none

  COREiluvatarmainnvidiatle
  #1393·tengqm opened Oct 18, 2026Contributor·

  ·
  1

  1

• [XPU] Stage liblaunch.a that references try_launch_table

  mainxpu
  #1391·Cross2pro opened Oct 16, 2026Contributor·2 of 3 tasks·

  ·
  1

• [HCU] Support fp64 MMAC (v_mmac_16x16x4_f64) via LLVM intrinsic

  hcumain
  #1384·w1120029931-bit opened Oct 15, 2026·

  ·
  1

• [BUILD] packaging: depend on the Python development headers at runtime

  CI/CDmain
  #1382·shiptux opened Oct 15, 2026Contributor·

  ·

• [FlagMega] Re-infer a slice dtype in the Qwen3 vLLM profile (fix prefill type mismatch)

  COREfeature/flagmega
  #1373·zhangnju opened Oct 14, 2026Contributor·2 tasks·

  ·