Ivo Kubjas

> Cursory look, there are a lot of reasons not to use ASN1. The standard is complex, implementations are old and clunky. There even has been instances of malicious code...

> @ivokub can you update status on this one? All is done. The range check gadget is agnostic to the builder if it either implements `Committer` or `Rangechecker` interfaces. If...

@gbotrel, @ThomasPiellard, @tomasandroil I implemented following benchmark (in groth recursion package): ```go func BenchmarkBN254Commitment(b *testing.B) { assert := test.NewAssert(b) innerCcs, innerVK, innerWitness, innerProof := getInnerCommitment(assert, ecc.BN254.ScalarField(), ecc.BN254.ScalarField()) assert.Equal(len(innerCcs.GetCommitments().CommitmentIndexes()), 1) //...

Acknowledged. I cannot give deadlines when we fix it though, it is a bit of work.

Hmm, I think ICICLE in principle supports other curves. I don't recall why I didn't add the support back then. EIther ICICLE at that time didn't support other curves, it...

> I can test it when ready. Thanks for the offer - I did a quick run to see if the interfaces make sense: https://github.com/Consensys/gnark/tree/feat/allcurve-accelerated First of all - I...

Thanks for testing it out. 9 seconds for solving seems quite a lot. When I'm solving directly with the constraint system without proving, then a 4M-ish circuit takes me 1.9...

Yup, witness computation is not GPU accelerated right now and I'm not sure it would be worth it to implement. We have optimized quite for CPU proving and have different...

It could be, I also recall there used to be strange errors when GPU ran out of memory. Sometimes I got non verifying proofs, later OOM, and sometimes machine crash....

Did you also verify the proofs? I recall at some point when I was doing tests that proving passed correctly, but at some point the proofs were incorrect.