All pull requests

Pull requests

• perf(cell): materialize cell inverses once per group

  area: corearea: performance
  #401·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(fused): keep the cell table in the kernel precision

  area: corearea: performance
  #400·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(scheduler): record acceptance history elementwise

  area: corearea: performance
  #399·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(fused): evaluate whole GPU stencils in one pass

  area: corearea: performance
  #398·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(fused): larger GPU query chunks for minimum-image pairs

  area: performance
  #397·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(mcmc): select the chosen scheduler update on accelerators

  area: corearea: performance
  #395·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(cell_list): cheaper rank and capacity check for cell insertions

  area: corearea: performance
  #394·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(fused): evaluate minimum-image candidates without the pair pipeline

  area: corearea: performance
  #393·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(math): scalar triangular products on every backend

  area: corearea: performance
  #392·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(cell_list,fused): skip empty updates and chunks only on CPU

  area: corearea: performance
  #391·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(ewald): optional float32 structure factors

  area: corearea: performance
  #390·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(moves): pick random groups with segment maxima instead of a sort

  area: corearea: performance
  #389·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(ewald): cache the reciprocal prefactor with the structure factors

  area: corearea: performance
  #388·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(ops): prefix-sum compaction of first true positions off CPU

  area: corearea: performance
  #387·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(fused): optional float32 pair kernels

  area: corearea: performance
  #386·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• fix(lj): keep species-selected mixing values in the parameter dtype

  area: coretype: bug
  #385·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(fused): evaluate full energies through a chunked cell table

  area: corearea: performance
  #384·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• perf(cell_list): bisect all levels in one scanned sort

  area: corearea: performance
  #383·n-gao opened Oct 15, 2026Collaborator·

  ·
  1

• chore(deps): bump the github-actions group across 1 directory with 3 updates

  dependenciesgithub_actions
  #382·dependabot[bot] opened Oct 14, 2026Bot·

  ·

• fix(ewald): pass per-system cutoffs to the one-shot Ewald evaluation

  area: coretype: bug
  #378·n-gao opened Oct 13, 2026Collaborator·

  ·
  1

• refactor(index): make Index indices private

  area: apiarea: core
  #377·n-gao opened Oct 12, 2026Collaborator·

  ·
  1

• test(ewald,neighborlist): share compiled programs across cases

  area: infraarea: performance
  #376·n-gao opened Oct 12, 2026Collaborator·

  ·
  1

• test(fused): share compiled programs in the fused and cell-list tests

  area: corearea: performance
  #375·n-gao opened Oct 12, 2026Collaborator·

  ·
  1

• test(application): compile fewer programs in the simulation tests

  area: infra
  #374·n-gao opened Oct 12, 2026Collaborator·

  ·
  1

• test: cache every compiled program and trace coverage with sysmon

  area: infraarea: performancegithub_actions
  #373·n-gao opened Oct 12, 2026Collaborator·

  ·
  1