Hi Denis,
maybe you could also have a look at / approve the Binutils side:
https://sourceware.org/pipermail/binutils/2026-August/150970.html
and unrelated
https://sourceware.org/pipermail/binutils/2026-August/150805.html
Johann
Am 25.08.26 um 14:15 schrieb Denis Chertykov:
пн, 24 авг. 2026 г. в 14:37, Georg-Johann Lay <[email protected]>:
This patch outputs .gnu_attribute tags for the bit sizes
of long double (12 = Tag_GNU_AVR_BITS_LONG_DOUBLE) resp.
double (8 = Tag_GNU_AVR_BITS_DOUBLE).
The recently introduced gimple mini-pass avr_pass_has
scans the code for the presence of [long] double, and
avr_file_end performs similar scans on the varpool.
There are cases where libgcc shares multilibs when they
don't depend on -m[long-]double=X, so that just emitting
.gnu_attribute *.X according to -m[long-]double=X is too strict.
The patch uses the .gnu_attribute's as introduced in
https://sourceware.org/PR34557 (Binutils 2.48).
The Binutils patch is currently under review:
https://sourceware.org/pipermail/binutils/2026-August/150970.html
This GCC patch is atop of
https://gcc.gnu.org/pipermail/gcc-patches/2026-August/728536.html
which introduces avr_pass_has and does the configure work to
determine HAVE_AS_AVR_GNU_ATTRIBUTE.
The patch passes without new regressions.
Ok for trunk?
Ok. Please appy.
Denis
