Proof aggregation api #1395

cryptonemo · 2021-01-15T21:06:22Z

No description provided.

cryptonemo · 2021-01-15T21:07:06Z

Not ready yet, but somewhat presentable at this point.

cryptonemo · 2021-02-17T13:33:48Z

TODO: When API is out of draft status, update FIP details filecoin-project/FIPs#72

ozhtdong · 2021-02-18T09:36:20Z

I guess it won't work for 32Gb/64Gb sectors (yet), since the number of proofs can't be both multiples of 10 and power of 2.

cryptonemo · 2021-02-18T12:44:53Z

I guess it won't work for 32Gb/64Gb sectors (yet), since the number of proofs can't be both multiples of 10 and power of 2.

There are ways to address that -- the simplest of which is to pad by duplicating some proof a number of times to meet the pow2 requirement (which the tests do for now). Other options are being investigated.

cryptonemo · 2021-03-05T14:05:26Z

SRS Publication to IPFS can now be done as follows:

cargo run --release --bin srspublish -- --ipfs-bin `which ipfs`

SRS file download can now be done like this:

cargo run --release --bin paramfetch -- -j srs-inner-product.json

filecoin-proofs/src/api/seal.rs

nikkolasg · 2021-03-29T13:35:41Z

filecoin-proofs/src/api/seal.rs

+    );
+
+    let num_inputs = commit_inputs.len();
+    let num_inputs_per_proof = num_inputs / aggregated_proofs_len;


isn't this supposed to be fixed ? i.e. shouldn't we read that info from .. porep config or stg ?
What we want to do is twofold:

ensure there are always X public inputs per proof -> this is different than checking there are X * Y public inputs (where Y is the number of proofs) in total

ensure the number of proofs is a power of two, if it's not, we duplicate X public inputs as much as necessary

It feels 1. and 2. are mixed here, and 1. is checking the total number of public inputs rather than the number of public inputs per proofs. Hence the 2. check can be "biased".

filecoin-proofs/tests/api.rs

cryptonemo · 2021-03-29T19:11:50Z

In memory caching of srs prover and verifier keys (per aggregated proof count) is now separate (instead of always keeping the pair together).

cryptonemo · 2021-04-06T15:41:20Z

Rebased against master

Note: sha2raw tests are not building

fil-proofs-param/src/bin/srspublish.rs

filecoin-proofs/src/api/seal.rs

dignifiedquire · 2021-04-12T10:19:39Z

filecoin-proofs/src/api/seal.rs

+                    &commit_output.proof[..],
+                    &verifying_key,
+                )
+                .expect("failed to construct multi proof from bytes")


this should not panic

fix: remove no longer needed seal_commit_phase2_for_aggregation method

Previously, both prover and verifier keys were cached together, but since access depends on the user/caller, it's beneficial to cache the mseparately in case both aren't needed in RAM

feat: re-use input padding code where needed

docs: fix a comment typo

cryptonemo force-pushed the feat-aggregation branch from c2a7298 to 9c23a6a Compare January 29, 2021 14:17

cryptonemo force-pushed the feat-aggregation branch 2 times, most recently from b8410f0 to 8208be6 Compare February 15, 2021 17:22

cryptonemo force-pushed the feat-aggregation branch from 8208be6 to 38b2420 Compare February 18, 2021 14:10

cryptonemo mentioned this pull request Feb 22, 2021

Add API for Proofs aggregation #1417

Closed

cryptonemo force-pushed the feat-aggregation branch from 087eee1 to 38b2420 Compare February 26, 2021 12:59

cryptonemo force-pushed the feat-aggregation branch from 194f842 to 4b96992 Compare March 5, 2021 14:02

cryptonemo force-pushed the feat-aggregation branch 3 times, most recently from 0a43459 to 6893b1d Compare March 10, 2021 16:54

cryptonemo force-pushed the feat-aggregation branch 2 times, most recently from e0746bf to 4596099 Compare March 23, 2021 20:09

nikkolasg reviewed Mar 29, 2021

View reviewed changes

cryptonemo force-pushed the feat-aggregation branch from 5e0bae5 to 36b3d91 Compare April 6, 2021 15:40

cryptonemo force-pushed the feat-aggregation branch from 36b3d91 to 766dfc5 Compare April 6, 2021 17:10

cryptonemo marked this pull request as ready for review April 6, 2021 19:42

cryptonemo requested a review from dignifiedquire as a code owner April 6, 2021 19:42