Add exhaustive/extensive tests

[tgross35]

Add a generator that will test all inputs for input spaces u32::MAX or smaller (e.g. single-argument f32 routines). For anything larger, still run approximately u32::MAX tests, but distribute inputs evenly across the function domain.

Since we often only want to run one of these tests at a time, this implementation parallelizes within each test using rayon. A custom test runner is used so a progress bar is possible.

Specific tests must be enabled by setting the LIBM_EXTENSIVE_TESTS environment variable, e.g.

LIBM_EXTENSIVE_TESTS=all_f16,cos,cosf cargo run ...

Testing on a recent machine, most tests take about two minutes or less. The Bessel functions are quite slow and take closer to 10 minutes, and FMA is increased to run for about the same.

[tgross35]

A start at "extensive" tests that run up to u32::MAX inputs, only on requested functions. Requested tests are set via env with a few different options:

# Run one or more exhaustive test
LIBM_EXTENSIVE_TESTS=sinf,cosf cargo t --features test-multiprecision --release -- extensive
# Run all `f32` or `f64` tests
LIBM_EXTENSIVE_TESTS=all_f32 cargo t --features test-multiprecision --release -- extensive
# Run everything
LIBM_EXTENSIVE_TESTS=all cargo t --features test-multiprecision --release -- extensive

For single-arg f32 it tests with all possible values in the domain. For f64, it runs u32::MAX tests with a logspace. For functions with N > 1 arguments, it currently does a logspace for each arg, each with u32::MAX^(1/N) steps.

I ran a handful of tests on my 24-core machine, the current setup with Rayon ranges about 10m-1h depending on the test. That can probably be improved, I haven't looked at this from a perf perspective at all (though one thing I noticed quickly from htop is there is a lot of time spent in kthreads, maybe I need to double check if Rayon is doing batching).

After I get this a bit better sorted out, I am going to hook up CI to run relevant tests when source files change. Current state is very scratchy.

@beetrees you may be interested since I know you suggested this a few times.

[tgross35]

I'm using the mimic harness so I can skip tests at runtime and show a progress bar, sample output.

running 90 tests
test mp_extensive_acos       ... ignored, extensive tests are only run if specified in RUST_LIBM_EXTENSIVE
test mp_extensive_acosf      ... starting extensive tests for `acosf`
[9m  100%] #################### acosf 4,294,967,295/4,294,967,295: 7,672,836.6102/s eta 0s
ok
test mp_extensive_acosh      ... ignored, extensive tests are only run if specified in RUST_LIBM_EXTENSIVE
test mp_extensive_acoshf     ... starting extensive tests for `acoshf`
[5m  73 %] ###############----- acoshf 3,139,620,000/4,294,967,295: 3,544,521.3401/s eta 5m (1.6715916e36,)

[tgross35]

(though one thing I noticed quickly from htop is there is a lot of time spent in kthreads, maybe I need to double check if Rayon is doing batching)

Way faster with chunking, acosf (sample test with a small domain) completes in two minutes on my machine, asinhf (sample with an unbounded domain) completes in 5 minutes.

[tgross35]

One last thing I need to do here is figure out how to reduce the scope for jnf / ynf, on my machine MPFR hits GNU MP: Cannot allocate memory (size=3458764513820540808).

[tgross35]

I wanted to move this to a rust-lang branch so I can point other PRs at it. Reopened as #388.