Move STATIC_BMI2 define to portability_macros.h

[pps83]

This change makes sure that STATIC_BMI and DYNAMIC_BMI are defined close to each other, and not in other headers.
Also, it reorders __BMI2__ check:

• if __BMI2__ defined, then set STATIC_BMI2 for all compilers
• use defined(_MSC_VER) && defined(__AVX2__) as fallback for ms compiler

[pps83]

Note, portability_macros.h is included only in 3 places:

1. compiler.h
2. zstd_cwksp.h
3. huf_decompress_amd64.S

compiler.h does include portability_macros.h.
zstd_cwksp.h includes compiler.h right after portability_macros.h
huf_decompress_amd64.S is asm file, doesn't check for STATIC_BMI

[Cyan4973]

I'm supportive of this change.

I'm also wondering if DYNAMIC_BMI2 is impacted by this welcome ordering change.
I would have expected DYNAMIC_BMI2 to depend on the state of STATIC_BMI2.

[Cyan4973]

Also, cc @terrelln , since it might be relevant to kernel code and assembly units

[pps83]

I would have expected DYNAMIC_BMI2 to depend on the state of STATIC_BMI2.

yep, but then I get more and more changes on top :)
in short, my understanding: DYNAMIC_BMI2 should only be set to 1 if STATIC_BMI2 is 0. But then, looking into the code, it seems it would affect it, as the checks do not seem to be consistent. And in some cases, code is quite questionable.

[pps83]

@terrelln ping

[pps83]

yep, but then I get more and more changes on top :) in short, my understanding: DYNAMIC_BMI2 should only be set to 1 if STATIC_BMI2 is 0. But then, looking into the code, it seems it would affect it, as the checks do not seem to be consistent. And in some cases, code is quite questionable.

After the change, step by step changes:

• there are only a few places where __BMI2__ is used, it should be replaced with STATIC_BMI2 (including this one)
• there is a place with the check if (defined(__BMI__) || defined(__BMI2__)) && defined(__GNUC__). Perhaps, checking for __BMI__ is enough: if __BMI2__ is enabled, that means that __BMI__ is also enabled.