Use portable _Float128 instead of __float128. Fix ARM64 builds.

[N-Storm]

Hello,

( This fixes currently broken builds for ARM64 arch. )

From https://gcc.gnu.org/onlinedocs/gcc/Floating-Types.html:

__float128 is available on i386, x86_64, IA-64, LoongArch and hppa HP-UX, as well as on PowerPC GNU/Linux targets that enable the vector scalar (VSX) instruction set. __float128 supports the 128-bit floating type. On i386, x86_64, PowerPC, LoongArch and IA-64, other than HP-UX, __float128 is an alias for _Float128. On hppa and IA-64 HP-UX, __float128 is an alias for long double.

I.e. __float128 aren't portable, as it's not available on every platform. ARM64 aka aarch64 are the example platform without __float128. While _Float128 are standard and main platforms just alias __float128 to _Float128 anyways.

This fixes building for ARM64 with GCC. I've tested under Ubuntu 25.x / gcc-14.x and main branch for prpll.

In case there are some reasons to have __float128 here, I can change this patch to add conditional alias __float128 -> _Float128 only if aarch64 macro defined.

[tdulcet]

Maybe update the CI to test on ARM as well. For example, update this line for Linux:

gpuowl/.github/workflows/ci.yml

Line 16 in d12753e

os: [ubuntu-22.04, ubuntu-24.04]

to something like:

        os: [ubuntu-22.04, ubuntu-24.04, ubuntu-22.04-arm, ubuntu-24.04-arm]

and this line for macOS:

gpuowl/.github/workflows/ci.yml

Line 91 in d12753e

runs-on: macos-13

to something like:

    runs-on: ${{ matrix.os }}
    strategy:
      matrix:
        os: [macos-13, macos-latest]
      fail-fast: false

You may need to remove the hardcoded GCC 14 as well.

[Artoria2e5]

I don't think f128 is used anywhere currently. Perhaps it can be deleted?

~/gpuowl$ grep -r f128 .
./src/common.h:using f128 = __float128;

Use of f128 was supposedly introduced in 817a983, but I think it's been reverted somewhere down the line.