All pull requests

Pull requests

• [CI] update ci notify

  #1990·Fanerly opened Nov 5, 2026Collaborator·

  ·
  7

• Feat/upgrade catlass v2.2.0

  #1989·bingmeiyou opened Nov 5, 2026Contributor·21 of 28 tasks·

  ·
  3

• [build] Abort at the shmem build failure instead of a later path

  #1987·173787247 opened Nov 4, 2026·

  ·
  1

  5

• [Bugfix] Validate bitwise_xor dtypes to prevent compiler crashes

  #1986·Fanerly opened Nov 4, 2026Collaborator·10 of 30 tasks·

  ·
  5

• [Ascend][Bugfix] Keep materialized 1D broadcast operands vectorized

  #1985·173787247 opened Nov 4, 2026·

  ·
  1

  8

• [Fix] AscendVidReduction: halve sigmoid/round/clamp counts under threads=2 (#1897)

  #1983·2Sheldonlaw opened Nov 4, 2026Contributor·

  ·
  1

  9

• [Fix] Correct copy_pipe_to_l1 auto-sync pipeline annotation: A2/A3 MTE2 GM->L1 load, A5 MTE1 bind (#1780)

  #1981·2Sheldonlaw opened Nov 3, 2026Contributor·

  ·
  1

  5

• [Bugfix] Route first-Parallel-axis broadcast operands through scalar reads to fix VEC UB alignment trap （#1885）

  #1979·2Sheldonlaw opened Nov 2, 2026Contributor·

  ·
  1

  7

• [Fix] AscendVidReduction wrongly halves tile-op size for whole-block [1,N] UB in row-split loops (#1800)

  #1978·2Sheldonlaw opened Nov 2, 2026Contributor·

  ·
  1

  9

• [Refactor] Port TileLang-Ascend onto TileLang main

  #1975·liggest opened Oct 31, 2026Contributor·

  ·
  2

• Fix column slice stride

  #1973·liuwenda4 opened Oct 31, 2026Contributor·10 of 18 tasks·

  ·
  3

• [Example] Add experimental Gated MLA decode with prefix mask vectorization

  #1970·dlaerbinjwy opened Oct 30, 2026Contributor·10 of 29 tasks·

  ·
  2

• [Testing] Clarify pytest requirements for operator contributions

  #1969·dlaerbinjwy opened Oct 30, 2026Contributor·5 tasks done·

  ·
  2

• [Bugfix] AscendVidReduction: halve sigmoid/round count for threads=2 (#1897)

  #1968·zwj-sep opened Oct 30, 2026Contributor·

  ·
  1

  10

• [Perf] Cache symbolic shape resolution in cython wrapper to cut per-call dispatch overhead (#1898)

  #1967·2Sheldonlaw opened Oct 30, 2026Contributor·

  ·
  1

  5

• [Example] Add GLM-5.3 kpool fp8_mqa_logits operator for Ascend NPU

  #1966·zwh1025 opened Oct 30, 2026Contributor·

  ·
  5

• [Refactor] insert sync support z3 migration from tillelang master

  #1965·erhsh opened Oct 30, 2026Contributor·4 of 29 tasks·

  ·
  5

• [AscendNPU-IR][Expert] Preserve local UB view descriptors

  #1962·guodong2005 opened Oct 29, 2026·

  ·
  1

• [AscendNPU-IR][Expert] Infer unary broadcast from accessed regions

  #1961·guodong2005 opened Oct 29, 2026·

  ·
  1

• [AscendNPU-IR][Language] Preserve source offsets in T.view

  #1960·guodong2005 opened Oct 29, 2026·

  ·
  1

• [Bugfix] Fix PTO bitwise_xor uint8 workspace view shape

  #1959·Fanerly opened Oct 29, 2026Collaborator·16 of 29 tasks·

  ·
  1

  7

• [Docs] clarify T.Tensor usage and deprecate T.Buffer DSL interface

  #1957·txy-ucas opened Oct 28, 2026·4 of 28 tasks·

  ·
  1

• fix(carver): use the selected SDK for A3 resource budgets

  #1955·tsukiyokai opened Oct 28, 2026·

  ·
  1

• [Example] Add fla_chunk_o operator and test

  #1953·stepjzy opened Oct 28, 2026Contributor·

  ·
  10

• [Example] Add fla_chunk_intra_token_parallel operator and test

  #1952·stepjzy opened Oct 28, 2026Contributor·

  ·
  7