Aharrypotter pushed to main at Aharrypotter/zl_bench 2026-09-10 20:01:49 +08:00
2786eb9886 bench: orchestrate a multi-card campaign without touching the measurement path
03fdcbab28 bench: partition operators across cards by measured cost, and rescore in place
c6a7f834cb bench: ignore the switch's own annotations when fingerprinting the IR
Compare 3 commits »
Aharrypotter pushed to main at Aharrypotter/zl_bench 2026-09-10 18:59:43 +08:00
4ac2a9acd8 bench: report when a foreign vendor's FlagGems implementation gets loaded
2e456bdc74 bench: let the verdict govern the report, and put the caliber in the table
a11863399a bench: bound each screening pair and record what it cost
1fd3abd56d bench: rename NO_KERNEL to NO_TRITON and say what it rules out
4d69320139 bench: let ab_rlc_test.sh decide which operators are worth six rounds
Compare 12 commits »
Aharrypotter pushed to main at Aharrypotter/zl_bench 2026-09-10 14:36:10 +08:00
c95474a432 bench: refuse to report a comparison whose two sides have different calibers
aa147c938e feat: repeat an A/B over rounds and report whether the verdict holds
9665d36a7c feat: allow the direct timer to repeat and take the median
Compare 3 commits »
Aharrypotter pushed to main at Aharrypotter/zl_bench 2026-09-10 10:47:21 +08:00
d7cb7efc94 fix: have the campaign worker ask zl_bench for its plugin list
75405cc555 Merge the S5000 MUSA RLC product campaign
29af59be9b feat: absorb the MThreads compat plugin and the A/B range report
c9f9b7885f feat: check the run's preconditions, and record which tuner cache it used
e970ed94c2 docs: document the MUSA RLC campaign
Compare 9 commits »
Aharrypotter pushed to main at Aharrypotter/zl_bench 2026-09-09 17:57:36 +08:00
ac818291a6 feat: turn on graph capture for MUSA
6fa5f5d7a2 refactor: let any backend own graph capture and its failure modes
Compare 2 commits »
Aharrypotter pushed to main at Aharrypotter/zl_bench 2026-09-09 17:04:06 +08:00
783c0655bc fix: never let the parallel sweep cover only part of the shape set
92dae82c7f feat: expose the upstream benchmark level as BENCH_LEVEL
c8ea586d0f feat: make the harness backend-agnostic
35a928a516 chore: pin shipped text formats to LF
Compare 4 commits »
Aharrypotter pushed to feature/musa-benchmark-support at Aharrypotter/zl_bench 2026-09-02 12:19:41 +08:00
abf9475ff3 feat: report a per-operator speedup range across two runs
7522143747 feat: drop stale CUDA-only skip guards on MThreads
3c9529dab2 fix: take graph-capture timing from Triton's device interface
64baa11a23 chore: pin shipped text formats to LF
Compare 4 commits »
Aharrypotter created branch feature/musa-benchmark-support in Aharrypotter/zl_bench 2026-09-02 12:19:41 +08:00
Aharrypotter pushed to feature/musa-rlc-product-harness at Aharrypotter/zl_bench 2026-09-02 11:57:05 +08:00
e970ed94c2 docs: document the MUSA RLC campaign
8a685638bb feat: re-account a sealed MUSA RLC run without re-running it
8bbea472bb feat: add the S5000 MUSA RLC product campaign
22139e4b87 chore: pin shipped text formats to LF
3b54f458e6 feat: add a MThreads benchmark compatibility plugin
Compare 5 commits »
Aharrypotter created branch feature/musa-rlc-product-harness in Aharrypotter/zl_bench 2026-09-02 11:57:05 +08:00
Aharrypotter created repository Aharrypotter/zl_bench 2026-09-02 11:56:08 +08:00