astcenc: Update to 4.7.0

[akien-mga]

The 4.5.0 release is a minor release with minor image quality improvements,
and multiple build system quality-of-life improvements.

https://github.com/ARM-software/astc-encoder/releases/tag/4.5.0

Draft as we need to review the compile flags for the "invariant" mode, which seems to be what we use (since we don't define ASTCENC_NO_INVARIANCE:

https://github.com/ARM-software/astc-encoder/blob/008d43f1e57944fd6c5cb663ef2c21c4884835dd/Source/cmake_core.cmake#L231-L244

See all the changelog lines talking about invariance. I'm assuming this is something we want to ensure if possible @fire? (IIRC, this means having reproducible encodings on different systems, at the cost of some performance.)

[fire]

The cmake appears to be for msvc enable strict for older and precise for newer. There’s also a clang cl change.

For floating point math.

[akien-mga]

The cmake appears to be for msvc enable strict for older and precise for newer. There’s also a clang cl change.

For floating point math.

Yeah, and also for regular Clang and GCC:

                $<$<AND:${is_clang},$<VERSION_GREATER_EQUAL:$<CXX_COMPILER_VERSION>,10.0.0>>:-ffp-model=precise>
                $<${is_gnu_fe}:-ffp-contract=off>)

(is_gnu_fe is for GCC and Clang IIUC)

It's pretty fine grained, I'm not sure we want to reproduce all this exactly, but I'll see what can be done easily.

[fire]

Since I don’t have guidance, we can try reproducing the code.

[fire]

I slept on this.

This is trying to make the exported ASTC conversions stable across different machines with respect to math.

[akien-mga]

I added the same flags as upstream, aside from the ones for ClangCL as we don't officially support it yet, and the tooling we have in methods.py doesn't let me detect it easily.

[Chubercik]

What's the state of this PR? There have been 3 releases since 4.5.0, currently sitting at 4.7.0.

[aaronfranke]

Shouldn't this refactor be in a separate PR from the astcenc library update? Perhaps as a prerequisite?

[akien-mga]

It is :D #80719
But I'm not finding the time to polish it properly.

I might update this PR and just drop the compiler-specific tweaks, to add later.

[akien-mga]

Will look into this later, but if someone can test on MSVC and confirm that astc encoding still seems to behave ok, I think this can be merged already.

[akien-mga]

Removed all the compiler flag stuff for now, here's the old code for future reference:

# Compiler options to ensure invariant build.
# Sync with upstream `Source/cmake_core.cmake` options for ASTCENC_INVARIANCE.
if env.msvc:
    pass
    if env.compiler_version >= (19, 30, 0):
       env_thirdparty.AppendUnique(CCFLAGS=["/fp:precise"])
   else:
       env_thirdparty.AppendUnique(CCFLAGS=["/fp:strict"])
else:
    from methods import using_clang

    if using_clang(env) and env.compiler_version >= (10, 0, 0):
        env_thirdparty.AppendUnique(CCFLAGS=["-ffp-model=precise"])
    env_thirdparty.AppendUnique(CCFLAGS=["-ffp-contract=off"])

[fire]

You only live once merge?