Hi all, This patch reimplements the saturating-truncate-and-insert-into-high intrinsics using the appropriate RTL codes and builtins.
Bootstrapped on aarch64-none-linux-gnu and tested on aarch64_be-none-elf too.
Pushing to trunk.
Thanks,
Kyrill
gcc/
* config/aarch64/aarch64-simd.md (aarch64_<su>qxtn2<mode>_le): Define.
(aarch64_<su>qxtn2<mode>_be): Likewise.
(aarch64_<su>qxtn2<mode>): Likewise.
* config/aarch64/aarch64-simd-builtins.def (sqxtn2, uqxtn2): Define
builtins.
* config/aarch64/iterators.md (SAT_TRUNC): Define code_iterator.
(su): Handle ss_truncate and us_truncate.
* config/aarch64/arm_neon.h (vqmovn_high_s16): Reimplement using
builtin.
(vqmovn_high_s32): Likewise.
(vqmovn_high_s64): Likewise.
(vqmovn_high_u16): Likewise.
(vqmovn_high_u32): Likewise.
(vqmovn_high_u64): Likewise.
gcc/testsuite/
* gcc.target/aarch64/narrow_high-intrinsics.c: Update uqxtn2 and sqxtn2
scan-assembler-times.
qmovn-hi.patch
Description: qmovn-hi.patch
