All pull requests

Pull requests

• Perfify: Introduce new HypothesisOp + Rename Hypothesis Argument Op

  #3670·javathunderman opened Oct 21, 2026Collaborator·

  ·

• xla targets: lower cudaDeviceSynchronize to reactantXLASync

  #3668·wsmoses opened Oct 21, 2026Member·

  ·

• Add command buffer compatible attribute on jit_call / kernel_call

  #3666·Pangoraw opened Oct 20, 2026Collaborator·

  ·

• Walk the body backwards for WhileDeadResults liveness

  #3662·Pangoraw opened Oct 20, 2026Collaborator·

  ·

• Index WhileDuplicateCarried candidates by init and yield

  #3661·Pangoraw opened Oct 20, 2026Collaborator·

  ·

• Convert LAPACK info to the op's result type

  #3658·kshyatt opened Oct 20, 2026Contributor·

  ·

• Take the orgqr reflector count from tau

  #3657·kshyatt opened Oct 20, 2026Contributor·

  ·
  2

• Size the LU permutation by rows

  #3656·kshyatt opened Oct 20, 2026Contributor·

  ·

• Conjugate V in the TPU gesvj lowering

  #3655·kshyatt opened Oct 20, 2026Contributor·

  ·

• Make Tessera passes work with func dialect

  #3649·jumerckx opened Oct 19, 2026Collaborator·

  ·

• Remove Breeze Reactant pin in CI

  #3637·Pangoraw opened Oct 16, 2026Collaborator·

  ·
  4

• reread_barrier: materialize a computed value read with expansion

  #3634·wsmoses opened Oct 16, 2026Member·

  ·

• avoid excessive JIT specialization of some LAPACK ops

  #3632·mofeing opened Oct 16, 2026Collaborator·

  ·
  2

• Small factorizations nits and fixes

  #3631·kshyatt opened Oct 16, 2026Contributor·

  ·
  2

• reduce_minor_dims: reduce a product of broadcasts along its trailing dimension

  #3627·wsmoses opened Oct 15, 2026Member·

  ·

• ScatterMaskedIndexSlice: cut a grid at the loop guard's last block

  #3626·wsmoses opened Oct 15, 2026Member·

  ·

• transposed_view_elementwise: compute in the layout of a transposed view

  #3624·wsmoses opened Oct 15, 2026Member·

  ·

• gather_operand_box: gathers that read only a box of a value read its slice

  #3617·wsmoses opened Oct 15, 2026Member·

  ·

• while_trip_count_from_guard: stop a loop where a monotone guard makes its body a no-op

  #3607·wsmoses opened Oct 15, 2026Member·

  ·

• Widen an affine.for only where it would raise per lane; keep carried loads of a recurring buffer

  #3603·wsmoses opened Oct 14, 2026Member·

  ·
  1

• Dot formation: a kept dimension only one operand broadcasts is a free dimension

  #3588·wsmoses opened Oct 14, 2026Member·

  ·

• broadcastindim_licm: keep expanding broadcasts inside while loops

  #3552·Pangoraw opened Oct 14, 2026Collaborator·

  ·
  2

• AffineCFG: a loop runs in lock step recursively, with the loops nested in it

  #3550·wsmoses opened Oct 14, 2026Member·

  ·
  2

• Fold comparisons of affine index expressions decided over their shape

  #3547·wsmoses opened Oct 14, 2026Member·

  ·
  1

• Raise affine-bounded parallel axes densely behind a guard

  #3545·wsmoses opened Oct 14, 2026Member·

  ·
  1