Improve x86 CI coverage
For x86, we should run the full test suite for all supported instruction sets on 32- and 64-bit versions of *NIX and Windows (Wine). Ideally with different stack alignments too. Maybe even throw more compilers (notably MSVC) into the mix as well.
We should also add an option for selecting the cpu mask used for running the test suite, right now it will always pick the fastest one available making testing different instruction sets complicated.
There's a lot of combinations, so this would possibly require having more x86 hardware available for CI.
Edit: List of jobs and their current state, copied from @another's comment:
-
instruction sets -
SSE2 -
x86-32 -
x86-64 -
Windows -
Linux (!974 (merged))
-
-
-
SSSE3 -
x86-32 -
x86-64 -
Windows -
Linux (!974 (merged))
-
-
-
SSE4 -
x86-32 -
x86-64 -
Windows -
Linux (!974 (merged))
-
-
-
AVX2 -
Windows -
Linux (!974 (merged))
-
-
AVX512 -
Windows -
Linux
-
-
-
stack alignment -
x86-32 -
4, 16, (32, 64)
-
-
x86-64 -
16, (32), 64
-
-
-
compilers -
MSVC
-
-
multi-threading (!974 (merged)) -
1 tt, 2 ft -
2 tt, 1 ft -
2 tt, 2 ft
-
-
OS -
Windows -
Linux -
*BSD -
OSX
-
Edited by Marvin Scholz