On Tue, 2024-01-09 at 11:55 -0700, Jeff Law wrote: > > > On 1/2/24 12:41, Ilya Leoshkevich wrote: > > GCC can emit code between the function label and the .LASANPC > > label, > > making the latter unaligned. Some architectures cannot load > > unaligned > > labels directly and require literal pool entries, which is > > inefficient. > > > > Move the invocation of asan_function_start to > > ASM_OUTPUT_FUNCTION_LABEL, which guarantees that no additional code > > is > > emitted. This allows setting the .LASANPC label alignment to the > > respective function alignment. > > --- > > gcc/asan.cc | 6 ++---- > > gcc/config/i386/i386.cc | 2 +- > > gcc/config/s390/s390.cc | 2 +- > > gcc/defaults.h | 2 +- > > gcc/final.cc | 3 --- > > gcc/output.h | 4 ++++ > > gcc/varasm.cc | 14 ++++++++++++++ > > 7 files changed, 23 insertions(+), 10 deletions(-) > So this needs a ChangeLog obviously. I assume you've tested on > s390[x]. > It should also be tested on x86 since it's the only other platform > that redefined ASM_OUTPUT_FUNCTION_LABEL. > > Assuming those tests pass without regression, then this is fine for > the > trunk. > > Thanks, > Jeff
Hi Jeff, Since Jakub already approved this 2/2, you approved 1/2, and x86_64/ppc64le/s390x regtests were successful, I've already pushed this series (with ChangeLogs). Unfortunately people discovered two regressions on i686 [1] and ppc64be [2]. The first one is already sorted out, I'm currently regtesting the fix for the second one and will push it as soon as it's done. Best regards, Ilya [1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113251 [2] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113284
