Comment # 72 on bug 1212101 from Richard Biener

(In reply to Martin Sirringhaus from comment #71)
> (In reply to Richard Biener from comment #69)
> > (In reply to Martin Sirringhaus from comment #68)
> > > Created attachment 867688 [details]
> > > Preprocessed SkOpts_avx
> > > 
> > > Here is the preprocessed SkOpts_avx with -flto. 
> > > Although, fwiw I suspect this file is a red herring.
> > 
> > A quick check doesn't show any symbols instantiated besides the obviously
> > desired ones when using -O3 -mavx.
> > 
> > As said there are other CUs that use ISA flags upon compilation so we'd have
> > to double check all their object files, like do readelf -s on them and grep
> > for
> > VecStorage for example.
> 
> I used the build-log that is attached here.
> 
> No hits:
> 
> /var/tmp/build-root/openSUSE_Factory-x86_64 ❯ for objf in $(grep mavx
> .build.log | split --column='-1' | sed 's|\.c.*$|\.o|g'); do
> OBJFILE="$(basename $objf)"; echo $OBJFILE; fd $OBJFILE
> "./home/abuild/rpmbuild/BUILD/obj/" -x readelf -s | grep VecStorage; done

That's the LTO built objects or the ones with LTO disabled?  With LTO
objects you have to resort to gcc-nm instead (that will inspect the LTO
symbol table).

> SIMD_avx2.o
> ConvolutionFilterAVX2.o
> SwizzleAVX2.o
> SkOpts_avx.o
> SkOpts_hsw.o
> SkOpts_skx.o
> aom_subpixel_8t_intrin_avx2.o
> blend_a64_mask_avx2.o
> fft_avx2.o
> highbd_convolve_avx2.o
> highbd_loopfilter_avx2.o
> intrapred_avx2.o
> cdef_block_avx2.o
> av1_inv_txfm_avx2.o
> cfl_avx2.o
> convolve_2d_avx2.o
> convolve_avx2.o
> highbd_convolve_2d_avx2.o
> highbd_inv_txfm_avx2.o
> highbd_jnt_convolve_avx2.o
> highbd_wiener_convolve_avx2.o
> jnt_convolve_avx2.o
> reconinter_avx2.o
> selfguided_avx2.o
> wiener_convolve_avx2.o
> vp9_diamond_search_sad_avx.o
> vp9_error_avx2.o
> vp9_quantize_avx2.o
> avg_intrin_avx2.o
> fwd_txfm_avx2.o
> loopfilter_avx2.o
> quantize_avx.o
> quantize_avx2.o
> sad4d_avx2.o
> sad_avx2.o
> subtract_avx2.o
> variance_avx2.o
> vpx_subpixel_8t_intrin_avx2.o
> Unified_cpp_common_audio_avx2_gn0.o
> Unified_cpp_aec3_aec3_avx2_gn0.o
> Unified_cpp_vector_math_avx2_gn0.o