Cannot compile `main`

[leonardoalt]

When I try to compile (cargo check) the current main branch, I get

error[E0432]: unresolved import `crate::QuinticExtensionData`
  --> /home/leo/.cargo/git/checkouts/plonky3-aad943b14b27c697/87bad26/monty-31/src/x86_64_avx512/packing.rs:31:5
   |
31 |     QuinticExtensionData, RelativelyPrimePower, halve_avx512,
   |     ^^^^^^^^^^^^^^^^^^^^
   |     |
   |     no `QuinticExtensionData` in the root
   |     help: a similar name exists in the module: `BinomialExtensionData`

This seems to come from https://github.com/TomWambsgans/Plonky3/tree/lean-multisig#, which doesn't compile when AVX is enabled, though I don't know where leanMultisig enables AVX by default for P3.

RUSTFLAGS="-C target-feature=+avx512f" cargo +1.90 check
    Checking p3-field v0.3.0 (/home/leo/devel/plonky3-lean/field)
    Checking criterion v0.7.0
error[E0425]: cannot find value `WIDTH` in this scope
   --> field/src/packed/interleaves.rs:177:31
    |
177 |             transmute::<[u32; WIDTH], _>([
    |                               ^^^^^ not found in this scope
    |
help: you might be missing a const parameter
    |
175 |         const INTERLEAVE1_INDICES<const WIDTH: /* Type */>: __m512i = unsafe {
    |                                  +++++++++++++++++++++++++

For more information about this error, try `rustc --explain E0425`.
error: could not compile `p3-field` (lib) due to 1 previous error
warning: build failed, waiting for other jobs to finish...

Note that upstream Plonky3 does compile with the same command. I wanted to open an issue in the P3 fork instead, but it doesn't accept issues, so opening it here instead.

I assume the issue needs to be fixed in the P3 fork repo separately, but this repo could also wrap that logic so that the error doesn't propagate for users that do have AVX512.

OS: Arch Linux
Arch: x84_64 AVX512 AVX2

[TomWambsgans]

Hi, sorry about that. It should compile now. But I am not sure tests pass for avx512, so I will let the issue open and try to improve the CI in the future.

[leonardoalt]

Thanks!

I ran the tests on my machine that has AVX but without AVX enabled and got this:

running 2 tests
test tests::test_xmss_signature ... ok
test tests::test_aggregate_xmss_signature ... FAILED

failures:

---- tests::test_aggregate_xmss_signature stdout ----

thread 'tests::test_aggregate_xmss_signature' panicked at /home/leo/.cargo/git/checkouts/multilinear-toolkit-746c0f983ff94f8e/eeb6404/sumcheck/src/prove.rs:332:5:
assertion `left == right` failed
  left: QuinticExtensionField { value: [6237812, 1911683616, 68427917, 1399194245, 336217948] }
 right: QuinticExtensionField { value: [2039587920, 746130554, 86290116, 162889164, 1490823340] }
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    tests::test_aggregate_xmss_signature

Does the code require aarch for the tests to pass?
Would it be an option to allow the user to have neither neon nor avx?

[TomWambsgans]

Does the code require aarch for the tests to pass?
No, at least in theory

So that's good to know, thanks again for reporting these issues.

My 2 local machines are aarch (M4 mac) on which tests pass, and x86, on which tests also pass (with and without avx2).
For avx512 I would expect bugs (to be fixed soon), but without the feature enabled that's weird. The best explanation I have is that it's due to the .cargo/.config.toml file that enabled avx512 by default, so you would have to manually add something like RUSTFLAGS="-C target-feature=-avx512f" to really disable it (or delete the .cargo/.config.toml file). Does it make sens to you? If that's the explanation, do tests pass now (at least without avx512) ?