https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121606
--- Comment #13 from Haochen Jiang <haochen.jiang at intel dot com> --- (In reply to rguent...@suse.de from comment #10) > > I wonder how at least the warning about using -mavx10.1 makes sense > given it says it enables 512bits which is the same behavior as in > GCC 16? The same diagnostic is in -mavx10.1-256 which is even more > confusing? I belive we should make -mavx10.1-512 not an Alias > and put the diagnostic there, not on -mavx10.1 uses? We could change the warning message. However, we should keep -mavx10.1 warning message there since it has changed from 256 bit to 512 bit. That is a critical change. > > For the diagnostic about -mno-evex512 the confusing thing is that > this option wasn't specified. The behavior of -mno-avx512fp16 > shouldn't change, and specifically adding -mavx10.1 (_not_ > -mavx10.1-256) should simply re-enable it. With > -mno-avx512fp16 -mavx10.1-256 it should, in 15, re-enable only > the 256 bit part of avx512fp16 obviously. -mno-avx512fp16 -mavx10.1 in GCC14/15 timeframe is a user-confusing option combination since at that time we keep orthogonal between AVX10.1 and AVX512FP16. User might not expect AVX512FP16 is enabled by AVX10.1 at that time. As the story goes, it changed to a normal imply in GCC 16, making it awkward for now. But the boat has sailed for GCC14/15, maybe we still need to keep that reminder.
