tree 9074de2d1e6608c9a146856593eb544d51700828
parent 950af8a0ebb3b9fd5d30f778af448f97adc380c8
author Jaeden Amero <jaeden@patater.com> 1775898911 +0000
committer LIU Hao <lh_mouse@126.com> 1776425905 +0800
gpgsig -----BEGIN SSH SIGNATURE-----
 U1NIU0lHAAAAAQAAADMAAAALc3NoLWVkMjU1MTkAAAAgpNkDDBuVPiz0dPd+zGoqB2ET19
 YkAU0wAtjm07QRvk4AAAADZ2l0AAAAAAAAAAZzaGE1MTIAAABTAAAAC3NzaC1lZDI1NTE5
 AAAAQPPtW2TO10GGCCoSe4TVtuIzglflZ8F+xSmw9ANSjswYWjs+Wxn68K0I0VAySj2csv
 ArTJhlKymaXF3eWPHUwgI=
 -----END SSH SIGNATURE-----

github: gcc: Test i386 Interlocked functions

Add a step in the testcases-gcc job that compiles and links the
t_intrinc test with -march=i386 using the i686 GCC cross-compiler, and
then runs the test.

Without kernel32 fallback in intrin-impl.h, GCC correctly refuses to
inline __sync_* builtins on i386 (no CMPXCHG or XADD) and instead emits
calls to library helpers that do not exist and fail to link.

Even with kernel32 fallback, the final linked binary contains i486+
instructions from the CRT startup code, which is built for i686.
Building t_intrinc with -march doesn't change this. However, this does
not prevent the test from running on the CI runner, which has a modern
x86 CPU that supports these instructions.

Signed-off-by: Jaeden Amero <jaeden@patater.com>
Signed-off-by: LIU Hao <lh_mouse@126.com>