Make the WHIR path's hash a parameter, and add an RPX256 instantiation - #992

Draft

MauroToscano wants to merge 28 commits into

multilinear-whirfrom

whir/rpx

Commits on Sep 15, 2026

fix(prover): pass the disk-spill argument at the four bench call sites
MauroToscano
committed
f57a71a
feat(multilinear): a hash parameter on the WHIR path, with keccak as the default
MauroToscano
committed
bcdd3dd

Commits on Sep 16, 2026

feat(crypto,multilinear): RPX256 on the WHIR path, host side
MauroToscano
committed
29fbb45
chore(crypto): gate the hash-metrics feature in make lint and make test
MauroToscano
committed
c9c452c
test(prover): the WHIR byte gate, over a canonically sorted EQ trace
MauroToscano
committed
60b6765
feat(math-cuda,multilinear): RPX256 on the WHIR device path
MauroToscano
committed
0502406
fix(multilinear): the four clippy findings in gpu.rs, so the cuda lint pass runs
MauroToscano
committed
140e269
feat(multilinear): derive the WHIR query count in integers, not f64
MauroToscano
committed
2e23db6
feat(prover): LAMBDA_VM_WHIR_HASH selects the hash, keccak by default
MauroToscano
committed
f2d93e1
fix(prover): make the commit-phase bench follow the hash knob, and make `phases` refuse
MauroToscano
committed
43a1cb7
feat(crypto,math-cuda): dispatch the RPX grind to the device
MauroToscano
committed
390909c
bench(prover): print the device commit and grind counts after each continuation arm
MauroToscano
committed
a1c338b
fix(prover): the byte gate must go through the hash dispatch, and assert
MauroToscano
committed
0cbc962
perf(math-cuda): a commitment keeps the tree it built, so its leaves hash once
MauroToscano
committed
3e38c9d
fix(math-cuda): route `paths` through the tree cache, and count per codeword
MauroToscano
committed
e14f63c
test(math-cuda): the reservation test owns the globals it asserts on
MauroToscano
committed
8f13724
perf(math-cuda): a commitment does not keep its tree — the retention is O(chains) by construction
MauroToscano
committed
2997b97
bench(prover): count host fallbacks beside the device commits
MauroToscano
committed
107bde9
test(math-cuda): the group guard's margin has to survive the allocator
MauroToscano
committed
070749e
feat(crypto): the RPX transcript hands out canonical felts, keccak unchanged
MauroToscano
committed
3e8c6fa
feat(crypto): count Fiat-Shamir absorbs and squeezes on BOTH arms
MauroToscano
committed
3f75574
feat(stark,prover): the WHIR transcript is the configuration's, checked by the compiler
MauroToscano
committed
a783f2d
fix(crypto,prover): count state reads apart from squeezes, and both sides of the proof
MauroToscano
committed
2ab421a
fix(prover): DECODE rows go in pc order, so the root is a function of the ELF
MauroToscano
committed
a850dd2
test(prover): pin the transcript pair, guarded by the guest ELF's sha256
MauroToscano
committed
c73568f

Navigation Menu

Make the WHIR path's hash a parameter, and add an RPX256 instantiation - #992

Make the WHIR path's hash a parameter, and add an RPX256 instantiation#992
MauroToscano wants to merge 28 commits into
multilinear-whirfrom
whir/rpx

Commits on Sep 15, 2026

fix(prover): pass the disk-spill argument at the four bench call sites

feat(multilinear): a hash parameter on the WHIR path, with keccak as the default

Commits on Sep 16, 2026

feat(crypto,multilinear): RPX256 on the WHIR path, host side

chore(crypto): gate the hash-metrics feature in make lint and make test

test(prover): the WHIR byte gate, over a canonically sorted EQ trace

feat(math-cuda,multilinear): RPX256 on the WHIR device path

fix(multilinear): the four clippy findings in gpu.rs, so the cuda lint pass runs

feat(multilinear): derive the WHIR query count in integers, not f64

feat(prover): LAMBDA_VM_WHIR_HASH selects the hash, keccak by default

fix(prover): make the commit-phase bench follow the hash knob, and make `phases` refuse

feat(crypto,math-cuda): dispatch the RPX grind to the device

bench(prover): print the device commit and grind counts after each continuation arm

fix(prover): the byte gate must go through the hash dispatch, and assert

perf(math-cuda): a commitment keeps the tree it built, so its leaves hash once

fix(math-cuda): route `paths` through the tree cache, and count per codeword

test(math-cuda): the reservation test owns the globals it asserts on

perf(math-cuda): a commitment does not keep its tree — the retention is O(chains) by construction

bench(prover): count host fallbacks beside the device commits

test(math-cuda): the group guard's margin has to survive the allocator

feat(crypto): the RPX transcript hands out canonical felts, keccak unchanged

feat(crypto): count Fiat-Shamir absorbs and squeezes on BOTH arms

feat(stark,prover): the WHIR transcript is the configuration's, checked by the compiler

fix(crypto,prover): count state reads apart from squeezes, and both sides of the proof

fix(prover): DECODE rows go in pc order, so the root is a function of the ELF

test(prover): pin the transcript pair, guarded by the guest ELF's sha256

Commits on Sep 17, 2026

fix(prover): the pinned ELF sha is the measured one, and the skip line shows both in full

feat(prover): pad a WHIR statement to a field element boundary

refactor(prover): the transcript pair pin is a base plus a per-branch term