perf(bb): WebGPU MSM tree-reduce SMVP — parallel preamble + ebid timing fix by AztecBot · Pull Request #23349 · AztecProtocol/aztec-packages

feat(bb): BrowserStack-driven MSM-webgpu bench harness

AztecBot

committed

feat(bb): host-side partition for tree-reduce SMVP (Step 1)

AztecBot

committed

wip(bb): smvp_tree_phase1 WGSL shader + bench scaffolding (Step 2, unvalidated)

AztecBot

committed

chore(bb): remove WIP diagnostic listeners from phase1 bench page

AztecBot

committed

feat(bb): smvp_tree_phase1 correctness validated (Step 2 done)

AztecBot

committed

feat(bb): smvp_tree_phase2 recursion shader + bench (Step 3 done)

AztecBot

committed

wip(bb): smvp_tree.ts host orchestrator + end-to-end bench (Step 5, partial)

AztecBot

committed

fix(bb): orchestrator CPU reference parenthesizes like the GPU (Step 5 validated)

AztecBot

committed

chore(bb): TODO marker at smvp_batch_affine_gpu call for Step 6 swap

AztecBot

committed

chore(bb): document MAX_SLICE_ENTRIES scale-up attempt

AztecBot

committed

feat(bb): smvp_tree_phase1/2 v2 — precomputed rank_to_raw + scale to SWEET_B=1024

AztecBot

committed

feat(bb): eliminate CPU sort between phases — outputs already globally bucket-sorted

AztecBot

committed

feat(bb): scatter + scatter_init shaders bridging tree-reduce output to finalize pipeline

AztecBot

committed

feat(bb): smvp_tree_adapter — bridge from tree-reduce to existing finalize pipeline

AztecBot

committed

feat(bb): bench harness skew=heavy mode + bigger pops summary

AztecBot

committed

feat(bb): integrate tree-reduce into smvp_batch_affine_gpu (Step 6)

AztecBot

committed

chore(bb): debug readback for tree-reduce entry_bucket_id

AztecBot

committed

perf(bb): smvp_tree v3 — fully parallel preamble + skip y in Phase A

AztecBot

committed

fix(bb): tree-reduce ebid timing — read fresh CSR, not stale buffer

AztecBot

committed

chore(bb): noble-direct autorun + 5x determinism probe

AztecBot

committed

chore(bb): rule out stale-workspace as source of stock non-determinism

AztecBot

committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

perf(bb): WebGPU MSM tree-reduce SMVP — parallel preamble + ebid timing fix#23349

perf(bb): WebGPU MSM tree-reduce SMVP — parallel preamble + ebid timing fix#23349
AztecBot wants to merge 21 commits into
zw/msm-webgpu-mont-mul-benchfrom
cb/b7178f5b65e7

Commits on May 17, 2026

feat(bb): BrowserStack-driven MSM-webgpu bench harness

feat(bb): host-side partition for tree-reduce SMVP (Step 1)

wip(bb): smvp_tree_phase1 WGSL shader + bench scaffolding (Step 2, unvalidated)

chore(bb): remove WIP diagnostic listeners from phase1 bench page

feat(bb): smvp_tree_phase1 correctness validated (Step 2 done)

feat(bb): smvp_tree_phase2 recursion shader + bench (Step 3 done)

wip(bb): smvp_tree.ts host orchestrator + end-to-end bench (Step 5, partial)

fix(bb): orchestrator CPU reference parenthesizes like the GPU (Step 5 validated)

chore(bb): TODO marker at smvp_batch_affine_gpu call for Step 6 swap

chore(bb): document MAX_SLICE_ENTRIES scale-up attempt

feat(bb): smvp_tree_phase1/2 v2 — precomputed rank_to_raw + scale to SWEET_B=1024

feat(bb): eliminate CPU sort between phases — outputs already globally bucket-sorted

feat(bb): scatter + scatter_init shaders bridging tree-reduce output to finalize pipeline

feat(bb): smvp_tree_adapter — bridge from tree-reduce to existing finalize pipeline

feat(bb): bench harness skew=heavy mode + bigger pops summary

feat(bb): integrate tree-reduce into smvp_batch_affine_gpu (Step 6)

chore(bb): debug readback for tree-reduce entry_bucket_id

perf(bb): smvp_tree v3 — fully parallel preamble + skip y in Phase A

fix(bb): tree-reduce ebid timing — read fresh CSR, not stale buffer

chore(bb): noble-direct autorun + 5x determinism probe

chore(bb): rule out stale-workspace as source of stock non-determinism