perf(prover): device-only preprocessed tables and GPU commits for mid-size tables - #888

Open

ColoCarletti wants to merge 54 commits into

mainfrom

gpu-opt-round4-residency

Commits on Jul 23, 2026

add profiling
ColoCarletti
committed
84a0872

Commits on Jul 24, 2026

fix(profiling): field fixes from the first sessions on the 5090 box
ColoCarletti
committed
743d156
docs(profiling): complete the toolkit README as a reference
ColoCarletti
committed
a48f5d4
perf(gpu): async pinned D2H + pre-created event pool + precomputed-tree cache
ColoCarletti
committed
530f423
perf(prover): pipeline + concurrent epoch proving in continuations
ColoCarletti
committed
f102818
perf(gpu): dim-split constraint interpreter with liveness-reused value slots
ColoCarletti
committed
58cb87e
perf(gpu): commit preprocessed tables through the fused GPU pipeline
ColoCarletti
committed
057d68c
perf(prover): overlap the global prove with the epoch proves' tail
ColoCarletti
committed
966d335
perf(prover): share per-ELF DECODE artifacts across continuation epochs
ColoCarletti
committed
0098e94
perf(prover): pipeline epoch trace builds onto a builder pool
ColoCarletti
committed
78d18b9
perf(prover): cache pre-captured AIR prototypes per table type
ColoCarletti
committed
7c54af7
profiling: re-home the toolkit spans onto the pipelined continuation flow
ColoCarletti
committed
0488420

Commits on Jul 27, 2026

Commits on Jul 28, 2026

Merge branch 'main' into gpu-opt-5090
diegokingston
authored
aa8788d

Commits on Jul 29, 2026

refactor(prover): prove continuation epochs on a single worker
ColoCarletti
committed
0d33073
chore: sync recursion bench lockfile with ecsm's num-integer dep
ColoCarletti
committed
62ba934
Merge branch 'main' into gpu-opt-5090
ColoCarletti
committed
3fd9bb8
new opt
ColoCarletti
committed
dc3632c
Merge branch 'main' into gpu-opt-round2-impl
ColoCarletti
authored
e53f484
fix(gpu): harden round-2 residency paths after review
ColoCarletti
committed
65f4d97
Merge branch 'main' into gpu-opt-round2-impl
ColoCarletti
committed
e9b01ad
perf(prover): replace table chunks with a VRAM-admitted per-table scheduler
ColoCarletti
committed
adbe561

Commits on Jul 31, 2026

perf(gpu): device-only preprocessed tables, lower LDE threshold, PCIe hygiene
ColoCarletti
committed
5d2cda3

Commits on Aug 5, 2026

Commits on Aug 6, 2026

Commits on Aug 7, 2026