All pull requests

Pull requests

• feat(ir): add opt-in rotating-slot software pipelines

  #2950·Hzfengsy opened Oct 3, 2026Member·

  ·
  7

• test(spmd): retire the mixed-kernel xfail and cover documented flags

  #2946·georgebisbas opened Sep 29, 2026Contributor·

  ·
  1

• test(distributed): add AllToAllV benchmark harness for RFC #2521

  #2945·georgebisbas opened Sep 29, 2026Contributor·

  ·
  27

• fix(codegen): isolate communication buffers on 64-byte cache lines

  #2942·hashiqiqixian opened Sep 29, 2026Contributor·

  ·
  1

• fix(ir): reuse gathered Mat storage in auto vector split

  #2941·tonibohnlein opened Sep 28, 2026Contributor·

  ·
  1

  6

• feat(distributed): chunk the InCore reduce_scatter lowering and accept a runtime SIZE

  #2939·georgebisbas opened Sep 28, 2026Contributor·

  ·
  8

• chore(toolchain): Update PTOAS to v0.66

  #2925·Little-oil opened Sep 28, 2026Contributor·

  ·
  15

• feat(passes): rotate slots by the iteration counter, not the induction variable

  #2922·georgebisbas opened Sep 27, 2026Contributor·

  ·
  4

• fix(passes): fold provably-false init_cond predicate after replication

  #2917·georgebisbas opened Sep 26, 2026Contributor·

  ·
  14

• fix(ir): preserve deferred full-box tile views through roundtrip

  #2911·Hzfengsy opened Sep 26, 2026Member·

  ·
  15

• perf(jit): use build identities for fast persistent cache hits

  #2909·Hzfengsy opened Sep 25, 2026Member·

  ·
  37

• fix(passes): narrow the InCore allgather stage to a runtime transfer extent

  #2908·georgebisbas opened Sep 25, 2026Contributor·

  ·
  6

• docs: document the pl.unroll 1024-iteration trip-count cap

  #2905·georgebisbas opened Sep 25, 2026Contributor·

  ·
  1

• docs(distributed): fix and refresh the existing collectives ladder

  #2904·georgebisbas opened Sep 25, 2026Contributor·2 tasks done·

  ·
  5

• fix(jit): generalize @pl.jit metadata dispatch past depth-1 attributes

  #2901·georgebisbas opened Sep 24, 2026Contributor·

  ·
  1

  10

• feat(collectives): multi-AIV lane-mapped all_to_all_v launches (RFC #2521 K3)

  #2900·georgebisbas opened Sep 24, 2026Contributor·

  ·
  14

• feat(buffer): lower slices, assembles and runtime valid extents

  #2898·lyfne123 opened Sep 24, 2026Collaborator·

  ·
  6

• ci: review with gpt-6-astra at medium effort

  #2893·lyfne123 opened Sep 24, 2026Collaborator·

  ·
  1

• feat(ir): verify AssignTypeSymmetry before and after every pass (#1285)

  Review effort 4/5
  #2888·lyfne123 opened Sep 23, 2026Collaborator·

  ·
  1

  21

• fix(pass): normalize Vec layouts before automatic V2C pushes

  Review effort 3/5
  #2883·YunjiQin opened Sep 23, 2026Collaborator·

  ·
  1

  3

• fix(codegen): keep an NZ-blocked host slice's logical scalar index

  Review effort 3/5
  #2875·zhangqi-chen opened Sep 22, 2026Collaborator·3 tasks done·

  ·
  20

• feat(fp4): allow FP4E2M1X2↔UINT8/INT8 reinterpret_view

  Review effort 2/5
  #2868·yanghaoran29 opened Sep 22, 2026Contributor·3 of 4 tasks·

  ·
  12

• feat(pass): pack frontend FP4 into FP4E2M1X2 for PTOAS

  Review effort 2/5
  #2848·yanghaoran29 opened Sep 21, 2026Contributor·4 tasks·

  ·
  13

• test(st): default the system-test memory planner to ptoas

  #2847·luohuan19 opened Sep 21, 2026Collaborator·5 of 6 tasks·

  ·
  1

• test(buffer): default-on integration branch for staged main repairs

  #2842·lyfne123 opened Sep 20, 2026Collaborator·

  ·
  1