On 30/11/2018 14:19, Kyrill Tkachov wrote: > > On 23/11/18 17:01, Sam Tebbs wrote: >> Hi all, >> >> Currently on AArch32, invoking with -march=armv8.2-a+dotprod -mfpu=neon >> incorrectly enables armv7 dotproduct. This patch restricts dotproduct >> to armv8 >> to correct the issue. >> >> When using a float ABI different from that of the host platform, >> including >> /usr/include/gnu/stubs.h errors due to a non-existent stubs-soft.h >> file, so >> an inclusion of a standard header file is required to compile >> successfully and >> find the correct float ABI to select in >> check_effective_target_arm_v8_2a_dotprod_neon_ok_nocache. >> >> Bootstrapped and regression tested on arm-none-linux-gnueabihf with no >> regressions. >> >> OK for trunk? >> > > Ok. I'd hope it's a niche usecase as the -mfpu=auto is the recommended > way these days. >
Hmm, this should work in the same way as removing the fp16 or fp16fml extensions. So why doesn't specifying -mfpu=neon cause that bit to be cleared from the feature sets? R. > Thanks, > Kyrill > >> gcc/ChangeLog: >> >> 2018-11-23 Sam Tebbs<sam.te...@arm.com> >> >> * config/arm/arm.h (TARGET_DOTPROD): Add TARGET_VFP5 constraint. >> >> gcc/testsuite/ChangeLog: >> >> 2018-11-23 Sam Tebbs<sam.te...@arm.com> >> >> * gcc.target/arm/neon-dotprod-restriction.c: New file. >> * lib/target-supports.exp >> (check_effective_target_arm_v8_2a_dotprod_neon_ok_nocache): >> Include stdint.h. >> >
