On Tue, 6 Jun 2023 at 15:14, Matthias Kretz via Libstdc++ < libstd...@gcc.gnu.org> wrote:
> Last part to resolve test failure introduced with PR109822. > > OK for master and backports (gcc-11 doesn't have __builtin_shufflevector, > though)? > Yep, OK. > > Tested on x86_64-pc-linux-gnu and powerpc64le-linux-gnu > > ----- >8 ----- > > Fix SFINAE on __is_intrinsic_type for 64-bit element types on non-VSX > POWER targets. > > Replace __extract_part implementation (which was the only user of > __shift_elements_right) by relying on __builtin_shufflevector (not > available in GCC 11). This removes another cast to 64-bit element type, > which breaks on non-VSX POWER. > > Signed-off-by: Matthias Kretz <m.kr...@gsi.de> > > libstdc++-v3/ChangeLog: > > PR libstdc++/109822 > * include/experimental/bits/simd.h: Include <bits/stl_algo.h> > for min(initializer_list). > Define __intrinsic_type_impl for all vectorizable types, but > without type member, if the target doesn't support it. > * include/experimental/bits/simd_builtin.h (__idx_permute): New. > (__shift_elements_right): Removed. > (__extract_part): Rewrite using __idx_permute. > (_S_reduce): Avoid unconditional cast to 64-bit element type > when the element type is known to be smaller. > --- > libstdc++-v3/include/experimental/bits/simd.h | 34 ++- > .../include/experimental/bits/simd_builtin.h | 250 +++++------------- > 2 files changed, 96 insertions(+), 188 deletions(-) > > > -- > ────────────────────────────────────────────────────────────────────────── > Dr. Matthias Kretz https://mattkretz.github.io > GSI Helmholtz Centre for Heavy Ion Research https://gsi.de > stdₓ::simd > ──────────────────────────────────────────────────────────────────────────
