machine: keep RP2 PLL calculations out of startup

[davecheney]

Fixes #5836

RP2040 and RP2350 PLL table generation exceeds interp's default loop limit, leaving clock calculations in firmware startup.

Split the per-product search so the calculations complete at compile time.

[dgryski]

I wonder if there's some sort of pragma or compile option we could add to be able to hunt these things out across all the platforms.

[davecheney]

I wonder if there's some sort of pragma or compile option we could add to be able to hunt these things out across all the platforms.

jinx; I have a sketch of a //go:compiletime pragma which will explode if the function is still present by link time.

[deadprogram]

@davecheney please see this error on Windows https://github.com/tinygo-org/tinygo/actions/runs/37575262876/job/113295674067?pr=5837#step:16:10

[davecheney]

@deadprogram PTAL. The windows failure was caused by git on windows converting testdata/ files from \n, to \r\n automagically. There are 11 existing workarounds for this in the codebase. I added a 12th. I'm going to take a separate pass tomorrow to try to fix the underlying problem.

[deadprogram]

Thanks for working on this @davecheney. I ran the tests on my local machine and took a good look. The following is edited from an automated review:

1. Now that all: check out testdata with LF line endings on all platforms #5872 is merged, the CRLF note in rp2_pll_test.go around the package clause is no longer needed. Perhaps rebase and drop that comment, and maybe use a simple bytes.CutPrefix for package main\n?

2. TestRP2PLLPreEvaluation alone fails on dev and passes with this change, so it already guards the regression. TestRP2PLLTable adds a lot of machinery (AST extraction, wasip1 build, wazero run) for a change that only moves the loop body into a helper. Should we keep just the pre-evaluation test?

3. The register checks in TestRP2PLLPreEvaluation hardcode the FBDIV and POSTDIV values for the current default clocks (200 MHz on pico, 150 MHz on pico2). If someone changes a default clock later this test fails for an unrelated reason. Maybe checking only that the machine.pdTable and machine.pllsys* symbols are gone is enough?

Sounds OK?

[davecheney]

@deadprogram PTAL. I've simplified the commit, dropping most of the testing and keeping the critical part - splitting genTable into two functions. The remaining test is simply double underpants that this doesn't regress. I'm working on a replacement that will add a //go:compiletime annotation which will remove the need for the test.

[deadprogram]

Looking good now thank you very much @davecheney for your fine efforts. Now merging.