Profile ZKVM verifier and optimize cycles #44
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
yczhangsjtu
changed the title
yczhangsjtu
force-pushed
the
feat/basefold-profile-cycles
branch
from
bd58b18 to
9591fa3
Compare
yczhangsjtu
changed the title
yczhangsjtu
changed the title
kunxian-xia
added a commit
that referenced
this pull request
Aug 7, 2025
* Dense Matrix * Added hints * Finished get_base_codeword_dimensions * WIP mmcs * WIP mmcs * Update mmcs * WIP mmcs * Finished MMCS * Added dot_product * query_phase input * WIP query_phase * Fix bug in sorting * WIP query phase * WIP query phase * WIP query phase * WIP query phase * New Sorting Impl * WIP query phase * WIP query phase * WIP query phase * Finished query_phase encoding * Support serialized input * Implement the naive encode small method * Bug workaround * Temp store: starting to use openvm mmcs instruction * Temp store: index to bits * Temp store: clean up mmcs * Temp store: clear compilation errors * Fix compilation error * Fix hash variable reading bug * Use dyn array for dimensions * Fix multiplication between var and ext * Fix mmcs reading * Remove unnecessary witness stream * Add doc for generating mmcs test data * Try fixing mmcs * Try fixing mmcs * Try fixing mmcs * Add comment * Use the same poseidon2 constants as the test data * Specify branch for test data gen * MMCS test passes * Rewrite fold coeff according to current basefold code * Fix * Merge e2e modification * Fix compilation errors from merge * Fix compilation errors from merge * Fix compilation errors from merge * (WIP) Connecting e2e with query phase * WIP * (WIP) transform ceno query phase verifier input to current * (WIP) Fix query phase transform * (WIP) Fix query phase transform * (WIP) Fix query phase transform * (WIP) Fix query phase transform * Fix query phase transform * Comment out connecting code temporarily * Query phase compile successful * (WIP) Debugging query phase * Fix a bug in verifier query phase * Read additional hints from binary file * Fix: batch opening (#28) * comment out * wip * hash read/write unit test passed * wip2 * add inv_2 to input stream * Fix memory out of bound problem * update Cargo.lock * Avoid providing two-adic generators inverses by hint * Replace idx_bits by num2bits_f * Replace idx_len by max_num_vars + rate log * Change index bits to small endian * Try fixing new index check * Fix new index check * Sub one from index len * Identified the cause * Fix index out of bound error * Change comment * Add native verify test * Fix evals shape error * Fix mmcs verify failure * Remove some print lines * fmt * Supply all hints * Fix new index compute * Fix ext mmcs verify dimension * Slice idx bits in ext mmcs verify * Some small fixes * right shift by hint * Fix verifier_folding_coeffs_level * Successfully run to first checkpoint * Fails at last line * Identified unimplemented function build_eq_x_r_vec_sequential_with_offset * batch verifier query phase test passes * Print the cycle count --------- Co-authored-by: Yuncong Zhang <[email protected]> * Switch ceno reliance * Fix compilation errors due to out of date code * Update test query phase batch * Fix query opening proof * Implement basefold proof variable * Update query phase verifier input * Preparing test data for query phase with updated code * Implement basefold proof transform * Prepare query phase verifier input * Prepare query phase verifier input * Fix final message access * Switch ceno reliance to small field support * Create basefold verifier function * Check final message sizes * Fix final message size * Fix final message size * Check query opening proof len * Compute total number of polys * Sample batch coeffs * Compute max_num_var * Write sumcheck messages and commits to transcript * Write final message to transcript * Complete the code for batch verifier * Add verifier test * Try to fix some compilation errors in e2e * Connecting pcs with e2e * Simplify BaseFold verifier (#34) * [Upgrade] ZKVMProof Verifier Update (#31) * Remove index reversal * Add a cycle tracker * Delete a loop * Better casting * Change verifier logic * Finish opcdoe proof verification debugging * Finish debugging table proof verification * Debug verifier * Finish debugging updated verifier * Remove unnecessary table proof fields * Remove unnecessary parsing * Update Plonky3 * Migrate away from temporary build branch * Switch ceno reliance * Fix compilation errors due to out of date code * Update test query phase batch * Fix query opening proof * Implement basefold proof variable * Update query phase verifier input * Preparing test data for query phase with updated code * Implement basefold proof transform * Prepare query phase verifier input * Prepare query phase verifier input * Fix final message access * Switch ceno reliance to small field support * basefold verifier for one matrix (#35) * wip * wip2 * wip3 * fix test * fix * fmt * fri part of verifying basefold proof for 1 matrix passed * sumcheck part 1 * sumcheck part 2 * sumcheck part 3 * cleanup * more cleanups --------- Co-authored-by: Ray Gao <[email protected]> Co-authored-by: xkx <[email protected]> * Fix some issues after merge * Make compilation pass temporarily * Make test pass before query phase * add multiple matrices as inputs to the basefold's unit test * fix * unit test passed * support matrices that have same height * refactor tests * Supply the permutation and make the random case pass * Try fixing transcript inconsistency * Use bin to dec le * Add pow witness * Basefold verifier passes for simple case * Update dependency * Basefold verifier passes decreasing and random batches * update ceno dependencies * comment out patch * refactor * the computation of max_num_var is simplified * put perm to RoundVariable * Change the way for computing batch * Remove the unnecessary slice * Simplify code * Fix bug: the first mmcs batch verify passes * Initialize all zeros * remove debug routines * rename * clean * cleanup * ignore e2e test * Complete Basefold verifier and Basefold e2e integration (#36) * Switch ceno reliance * Fix compilation errors due to out of date code * Update test query phase batch * Fix query opening proof * Implement basefold proof variable * Update query phase verifier input * Preparing test data for query phase with updated code * Implement basefold proof transform * Prepare query phase verifier input * Prepare query phase verifier input * Fix final message access * Switch ceno reliance to small field support * Create basefold verifier function * Check final message sizes * Fix final message size * Fix final message size * Check query opening proof len * Compute total number of polys * Sample batch coeffs * Compute max_num_var * Write sumcheck messages and commits to transcript * Write final message to transcript * Complete the code for batch verifier * Add verifier test * Try to fix some compilation errors in e2e * Connecting pcs with e2e * Fix some issues after merge * Make compilation pass temporarily * Make test pass before query phase * Supply the permutation and make the random case pass * Try fixing transcript inconsistency * Use bin to dec le * Add pow witness * Basefold verifier passes for simple case * Update dependency * Basefold verifier passes decreasing and random batches * update ceno dependencies * comment out patch * refactor * the computation of max_num_var is simplified * put perm to RoundVariable * remove debug routines * rename * clean * cleanup * ignore e2e test --------- Co-authored-by: kunxian xia <[email protected]> * Fix * Fix one bug: alpha is batch coeffs 1 not 0 * Handle case when batch size is 1 * Allocate all zeros only once * Construct test multiple rounds and test with fibonacci e2e data * Add trace subscriber to profile cycles * Profile query phase * Profile query phase * Some small optimizations * Allocate opened values buffer outside query loop * Compute low high values buffer outside loop * Compute alpha offset outside loop * Remove batch coeffs offset * Compute all zero slices outside loop * Remove round openings from loop * Clean ups * Compute height outside loop * Compute permuted dimensions outside loop * Merge two loops over openings * Profile fri rounds * Fix compilation errors after merge * Dependencies moved to dev * Fix unit tests * chore remove unused dependency * Remove opening height from round context: duplicate of log2_height * Compute low and high values buffer in loop * Provide some comments * All zeros slice computed in loop * Fix small issues --------- Co-authored-by: Kunming Jiang <[email protected]> Co-authored-by: kunxian xia <[email protected]> Co-authored-by: Ray Gao <[email protected]>
kunxian-xia
pushed a commit
that referenced
this pull request
Aug 7, 2025
…56) * Optimize FRI round * Small improve * Small improve * Small improve * Remove unnecessary computing coeff * Fix small issues * Improve comment
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add cycle trackers and a trace subscriber to the tests to profile the basefold verifier.
Before this PR: cycle count approximately 1050000 (for the Fibonacci e2e case)
Identified issues: many query-independent computations are executed inside the loop over queries (there are 100 queries in the current setting).
(WIP) Extracting query-independent computations outside the loop and save in a
RoundContextVariablestruct.Current cycle count in this PR: approximately 890000
#45