All pull requests

Pull requests

• fix(jit): preserve bindings after tuple unpacking

  #3346·wangqin1723-max opened Nov 9, 2026Contributor·

  ·
  8

• feat(codegen): emit the URMA async transfer helper and request its engine (merge on top of #3343 after #3339)

  #3344·georgebisbas opened Nov 9, 2026Contributor·8 of 12 tasks·

  ·
  47

• feat(backend): add the URMA async transfer adapter (merge on top of #3339)

  #3343·georgebisbas opened Nov 9, 2026Contributor·6 of 7 tasks·

  ·
  47

• feat(ir): preserve async-transfer InOut->In and the destination alias

  #3339·georgebisbas opened Nov 9, 2026Contributor·9 of 12 tasks·

  ·
  13

• feat(distributed): initialize Host DDR communication windows

  #3338·YunjiQin opened Nov 9, 2026Collaborator·

  ·
  1

• chore(runtime): bump simpler to 82cf61e98 and size the TREM HP scratch

  #3337·Little-oil opened Nov 9, 2026Contributor·

  ·
  17

• feat(simt): add atomic updates and explicit synchronization

  #3336·lyfne123 opened Nov 9, 2026Collaborator·

  ·
  19

• feat(runtime): carry FP8E5M2 with its native runtime dtype

  #3335·Little-oil opened Nov 9, 2026Contributor·

  ·
  14

• feat(reg): support register scope operator adaptations

  #3334·yiyuan-123 opened Nov 9, 2026Contributor·

  ·
  10

• feat(reduction): support row_major out_layout for row reductions

  #3333·Little-oil opened Nov 9, 2026Contributor·4 of 5 tasks·

  ·
  1

  50

• feat(jit): normalize external calls and specialize call signatures

  #3332·lyfne123 opened Nov 9, 2026Collaborator·

  ·
  75

• Fix: compile replay-safe Graph metadata

  #3326·MaxwellF1 opened Nov 7, 2026Contributor·

  ·
  36

• Fix: preserve backing storage reads in parameter directions

  #3323·MaxwellF1 opened Nov 6, 2026Contributor·

  ·
  111

• Fix: specialize JIT submissions and preserve Graph identities

  #3321·MaxwellF1 opened Nov 6, 2026Contributor·

  ·
  80

• feat(ci): select DeepSeek and Qwen operator compile cases

  #3316·luohuan19 opened Nov 6, 2026Collaborator·

  ·
  13

• feat(ir): lower indexed gather and scatter to Buffer IR

  #3314·lyfne123 opened Nov 6, 2026Collaborator·

  ·
  3

• feat(mx): accept every FP8 pair in the tensor MX matmul

  #3313·Little-oil opened Nov 5, 2026Contributor·

  ·
  25

• feat(gemv): add the A5 MX GEMV family

  #3312·Little-oil opened Nov 5, 2026Contributor·

  ·
  12

• feat(a5): request an NZ destination layout on the tile cast

  #3311·Little-oil opened Nov 5, 2026Contributor·

  ·
  16

• feat(codegen): store an MX scale tile through its own ND view

  #3310·Little-oil opened Nov 5, 2026Contributor·

  ·
  38

• feat(gemv): accept FP8 and HF8 operand pairs on A5

  #3309·Little-oil opened Nov 5, 2026Contributor·

  ·
  26

• feat(mx): accept every FP8 pair in the tile MX matmul family

  #3308·Little-oil opened Nov 5, 2026Contributor·

  ·
  21

• feat(matmul): accept mixed FP8 operand pairs on A5

  #3307·Little-oil opened Nov 5, 2026Contributor·

  ·
  9

• test(quant): complete MXFP8 quantization coverage and refresh MX quant docs

  #3303·Little-oil opened Nov 5, 2026Contributor·

  ·
  26

• fix(ir): check raw MX quant workspaces by physical capacity

  #3302·Little-oil opened Nov 5, 2026Contributor·

  ·
  31