Hi all,
This is V3 of a fix for PR117850. V2 wasn't picked up or pinged by me as my
work changed quickly, so I've taken the opportunity here to fix it up a bit.
The major differences from V1 are based on the feedback given there.
There are two things that I'd like to note:
* This fold is more general than that required to fix PR117850, in that
it chooses the highpart builtin whenever all of the interesting
arguments
are the highparts of vector varialbles. This case is already (mostly)
handled by combine, so it might not be necessary to do this here. The
real improvement that this fold makes is where we have vector highparts
and vector constants (cases that look like the PR).
* This code does not handle VOPs, which, IIUC, would need to be done
when folding builtins which satisfy aarch64_modifies_global_state_p or
aarch64_reads_global_state_p. An example of where this might be useful
is for extending floating-point conversions. But, as I say, I'm not
sure if this is worth it given combine can usually do this (the only
counter-example I've run into being vcvt_f32_bf16).
V1:
https://inbox.sourceware.org/gcc-patches/20250217190739.1680451-2-spencer.ab...@arm.com/
PR117850: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117850
Bootstrapped & regtested on aarch64-linux-gnu. OK for master?
Thanks,
Spencer
Spencer Abson (1):
aarch64: Fold builtins with highpart args to highpart equivalent
[PR117850]
gcc/config/aarch64/aarch64-builtin-pairs.def | 73 ++
gcc/config/aarch64/aarch64-builtins.cc | 183 +++++
.../aarch64/simd/fold_to_highpart_1.c | 717 ++++++++++++++++++
.../aarch64/simd/fold_to_highpart_2.c | 89 +++
.../aarch64/simd/fold_to_highpart_3.c | 83 ++
.../aarch64/simd/fold_to_highpart_4.c | 38 +
.../aarch64/simd/fold_to_highpart_5.c | 92 +++
.../aarch64/simd/fold_to_highpart_6.c | 37 +
.../gcc.target/aarch64/simd/vabal_combine.c | 72 --
9 files changed, 1312 insertions(+), 72 deletions(-)
create mode 100644 gcc/config/aarch64/aarch64-builtin-pairs.def
create mode 100644 gcc/testsuite/gcc.target/aarch64/simd/fold_to_highpart_1.c
create mode 100644 gcc/testsuite/gcc.target/aarch64/simd/fold_to_highpart_2.c
create mode 100644 gcc/testsuite/gcc.target/aarch64/simd/fold_to_highpart_3.c
create mode 100644 gcc/testsuite/gcc.target/aarch64/simd/fold_to_highpart_4.c
create mode 100644 gcc/testsuite/gcc.target/aarch64/simd/fold_to_highpart_5.c
create mode 100644 gcc/testsuite/gcc.target/aarch64/simd/fold_to_highpart_6.c
delete mode 100644 gcc/testsuite/gcc.target/aarch64/simd/vabal_combine.c
--
2.34.1
