Add first party C tests

[Timmmm]

Add first party C tests that can be used for tests specific to this model. This uses Clang as the cross-compiler.

This includes a copy of nanoprinf for printing to the UART.

[jordancarlin]

I definitely prefer this to the Zig implementation, even if the CMake file is a bit messier. Ideally it would work with both gcc and clang, especially because a lot of other riscv projects point users very strongly towards gcc (like the riscv-arch-test documentation).

[jordancarlin]

Thanks for adding GCC. A few more comments.

[jordancarlin]

As our tests evolve we'll need a more comprehensive march string. We use rv64gcv_zcb_zfa_zba_zbb_zbc_zbs_zfh_zicboz_zicbop_zicbom_zicond_zbkb_zbkx_zknd_zkne_zknh for our tests. We really want to enable every extension so that any instruction can be used in a test file for the model.

[trdthg]

But some extensions are incompatible, such as Zcf and Zclsd, each test should be able to specify its own match string

One example is gcc's test framework, you can specify it in the comments (for reference)

gcc/testsuite/gcc.target/riscv/cmo-zicbom-1.c

/* { dg-do compile } */
/* { dg-options "-march=rv64gc_zicbom -mabi=lp64" } */
...

[jordancarlin]

That is a very good point. Not sure that we need that level of complexity yet in the first version of this PR (would rather get some testing in instead of waiting for it to be perfect and continue with almost no tests), but definitely something we will need to address eventually.

[trdthg]

completely agree, would be great to see this merged soon.

[Timmmm]

Yeah my plan was to embed a list of march and model configs inside the source files and then we would run each test for all of those. Probably worth waiting until after the new config system is done though.

[jordancarlin]

Seems reasonable. We should probably open an issue to track this after this PR is merged.

[arichardson]

I like this, just a few minor comments (and haven't managed to review all the asm code)

[Timmmm]

I figured out the lga thing - in the code I copied it from it was fine with just la but for some reason I was getting a relocation out of range error.

It was because I forgot to set __global_pointer$ in the linker script, and I guess it defaults to 0 or something, but the code starts at like 0x8000000000 so it was out of the 2GB range limit on la. Setting __global_pointer$ to be near .text means we can just use la.

I'll fix the other issues mentioned in a bit.

[github-actions]

Test Results

396 tests  +4   396 ✅ +4   1m 19s ⏱️ ±0s
  1 suites ±0     0 💤 ±0 
  1 files   ±0     0 ❌ ±0 

Results for commit 16a2f97. ± Comparison against base commit 91b55d3.

♻️ This comment has been updated with latest results.

[Timmmm]

I removed the max_pmp test. I'll add it with my PR that fixes the PMP code.

[jordancarlin]

LGTM at this point. There are still improvements to be made (especially with ISA strings), but those can be part of a follow up PR. This will be great to get in so new tests can easily be added for the model.

[arichardson]

LGTM, this is great and any improvements can be made in follow-up changes