Add and test Emscripten/WASM compilation

[agriyakhetarpal]

See #19 (comment)

This PR sets up BLIS compiled to WASM (using my patches in Pyodide) and CMocka compiled to WASM, then runs the test suite by compiling semicolon-lapack to WASM.

I see that you merge PRs with a merge commit instead of squashing/rebasing, so I will rebase my commits accordingly so as to have a cleaner commit history (this is mostly rough work with unclean commit history, which I committed after testing stuff out over the weekend).

xref pyodide/pyodide-recipes#604, pyodide/pyodide-recipes#609 and pyodide/pyodide-recipes#610, and flame/blis#491.

[agriyakhetarpal]

The 480 tests (and thousands of subtests) are all green! See https://github.com/agriyakhetarpal/semicolon-lapack/actions/runs/28874908379/job/85647289889?pr=1 🎉

[ilayn]

Ah very cool, thanks a lot for this. Finally all pieces are coming together.

I see that you merge PRs with a merge commit instead of squashing/rebasing, so I will rebase my commits accordingly so as to have a cleaner commit history

That's my bad, whatever that button is set up for, I just click it without thinking too much. I really don't have any preferences so please use whichever style you prefer. If you want to squash it for a single functionality that would be indeed better. But in general as long as it is kinda sorta clean I don't think I would care either way.

I also saw that during build process emscripten has an extension that leads to warnings, which is probably why it is busy printing tons of stuff like

 [980/3580] Compiling C object src/libsemilapack.a.p/s_strtrs.c.o
[981/3580] Compiling C object src/libsemilapack.a.p/s_strttp.c.o
[982/3580] Compiling C object src/libsemilapack.a.p/s_strttf.c.o
[983/3580] Compiling C object src/libsemilapack.a.p/z_dzsum1.c.o
[984/3580] Compiling C object src/libsemilapack.a.p/z_izmax1.c.o
[985/3580] Compiling C object src/libsemilapack.a.p/z_ilazlr.c.o
../src/z/ilazlr.c:33:23: warning: complex initialization specifying real and imaginary components is an extension [-Wcomplex-component-init]
   33 |     const c128 zero = CMPLX(0.0, 0.0);
      |                       ^~~~~~~~~~~~~~~
/home/runner/work/_temp/fd95dd82-078c-4694-be0f-9cba4b7fadf0/emsdk-main/upstream/emscripten/cache/sysroot/include/complex.h:125:21: note: expanded from macro 'CMPLX'
  125 | #define CMPLX(x, y) __CMPLX(x, y, double)
      |                     ^~~~~~~~~~~~~~~~~~~~~
/home/runner/work/_temp/fd95dd82-078c-4694-be0f-9cba4b7fadf0/emsdk-main/upstream/emscripten/cache/sysroot/include/complex.h:121:40: note: expanded from macro '__CMPLX'
  121 | #define __CMPLX(x, y, t) (+(_Complex t){ (t)(x), (t)(y) })
      |                                        ^~~~~~~~~~~~~~~~~~
1 warning generated.
[986/3580] Compiling C object src/libsemilapack.a.p/s_stzrzf.c.o
[987/3580] Compiling C object src/libsemilapack.a.p/z_ilazlc.c.o
../src/z/ilazlc.c:33:23: warning: complex initialization specifying real and imaginary components is an extension [-Wcomplex-component-init]
   33 |     const c128 zero = CMPLX(0.0, 0.0);
      |                       ^~~~~~~~~~~~~~~
/home/runner/work/_temp/fd95dd82-078c-4694-be0f-9cba4b7fadf0/emsdk-main/upstream/emscripten/cache/sysroot/include/complex.h:125:21: note: expanded from macro 'CMPLX'
  125 | #define CMPLX(x, y) __CMPLX(x, y, double)
      |                     ^~~~~~~~~~~~~~~~~~~~~
/home/runner/work/_temp/fd95dd82-078c-4694-be0f-9cba4b7fadf0/emsdk-main/upstream/emscripten/cache/sysroot/include/complex.h:121:40: note: expanded from macro '__CMPLX'
  121 | #define __CMPLX(x, y, t) (+(_Complex t){ (t)(x), (t)(y) })
      |                                        ^~~~~~~~~~~~~~~~~~
1 warning generated.
[988/3580] Compiling C object src/libsemilapack.a.p/z_zcposv.c.o
../src/z/zcposv.c:113:25: warning: complex initialization specifying real and imaginary components is an extension [-Wcomplex-component-init]
  113 |     const c128 NEGONE = CMPLX(-1.0, + 0.0);
      |                         ^~~~~~~~~~~~~~~~~~
/home/runner/work/_temp/fd95dd82-078c-4694-be0f-9cba4b7fadf0/emsdk-main/upstream/emscripten/cache/sysroot/include/complex.h:125:21: note: expanded from macro 'CMPLX'
  125 | #define CMPLX(x, y) __CMPLX(x, y, double)
      |                     ^~~~~~~~~~~~~~~~~~~~~
/home/runner/work/_temp/fd95dd82-078c-4694-be0f-9cba4b7fadf0/emsdk-main/upstream/emscripten/cache/sysroot/include/complex.h:

I did not check but is it possible to use a silencer for that like -Wno-complex-component-init or something?

[ilayn]

Looks good @agriyakhetarpal thanks again. There is one more flag that is lighting up but I can do it with the next PR anyways. So in it goes.

[agriyakhetarpal]

Thank you!

While looking at this locally, I also had a bit of a TIL. Leaving it here since it might be interesting (but I assume you already know all this)!

What I found is more about the Apple SDK than Emscripten. Since Emscripten is based on LLVM-Clang, I kind of expected Clang in a native build on my macOS machine to show these warnings as well. However, even with -Wcomplex-component-init and -Weverything, AppleClang did not emit anything. I did some digging about this, and found that the reason is more about the libc and not about the compiler :D

This is what macOS's libc does for the CMPLX definition – and quite cheekily too at the macro expansion site itself, in my opinion:

#define CMPLX(__real,__imag) \
  _Pragma("clang diagnostic push") \
  _Pragma("clang diagnostic ignored \"-Wcomplex-component-init\"") \
  (double _Complex){(__real),(__imag)} \
  _Pragma("clang diagnostic pop")

Emscripten uses MUSL, whose <complex.h> defines __CMPLX(x, y, t) (+(_Complex t){ (t)(x), (t)(y) }) with no such pragma guard.

So I thought it made sense not to limit this to Emscripten specifically, since theoretically a Linux/non-Apple Clang build could show these warnings against a more standard glibc/MUSL.

I can take a look at adding more flags (you merged as I was writing this comment, so thanks for raising that!).

[ilayn]

I should have guessed. In the early versions of this library I had so many issues with it that I was about to give up on MacOSX. So many "liberal" changes done over the years that is making things seriously difficult. But glad that it is not detrimental.