On Sat, May 18, 2013 at 6:00 AM, Sriraman Tallam <tmsri...@google.com> wrote:
>>> > I don't really understand why you made the change to x86intrin.h instead >>> > of >>> > making it inside each *mmintrin.h header. The code would be the same size, >>> > it would let us include smmintrin.h directly if we wanted to, and >>> > x86intrin.h would also automatically work. >>> >>> Right, I should have done that instead! >> >> Yeah, definitely. For the standalone headers, which have currently >> __<FEATURE>__ guards inside of it, please replace it by the larger snippets >> involving #pragma, and in the x86intrin.h/immintrin.h headers include those >> unconditionally, instead of just if __<FEATURE>__ is defined. >> For the non-standalone headers (newer ones like avxintrin.h), replace >> the #ifdef __<FEATURE>__ in immintrin.h/x86intrin.h with larger snippets. > > > * I did mostly as suggested except that even for avx and avx2 headers > I did not see the harm in doing it in the header itself. AVX header > did not have the "#ifndef _AVXINTRIN_H_INCLUDED" which I added before > doing this. I have added test cases to show it is doing the right > thing for avx. > * I also found that when the caller to these intrinsics do not have > the right target attribute, an error is raised in -O2 mode but not in > -O0 mode. I have fixed this with a patch to ipa-inline.c, please see > if this is alright. Test case intrinsics_5.c checks if an error is > raised. > * LZCNT needed to be handled which is done now. * common/config/i386/i386-common.c: Handle LZCNT. The above part is OK for mainline and release patches. Please commit LZCNT part as a separate patch to mainline. Also, please get middle-end part reviewed and committed before we proceed with target-dependent part. Thanks, Uros.
