https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118956
Bug ID: 118956
Summary: [15 regression]
gcc.target/aarch64/sve/pred-not-gen-[14].c fail after
r15-268-g9dbff9c05520a74e
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Keywords: missed-optimization, testsuite-fail
Severity: normal
Priority: P3
Component: rtl-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: rsandifo at gcc dot gnu.org
Target Milestone: ---
Target: aarch64*-*-*
gcc.target/aarch64/sve/pred-not-gen-1.c and
gcc.target/aarch64/sve/pred-not-gen-4.c started failing with
r15-268-g9dbff9c05520a74e:
FAIL: gcc.target/aarch64/sve/pred-not-gen-1.c scan-assembler-not \\tbic\\t
FAIL: gcc.target/aarch64/sve/pred-not-gen-1.c scan-assembler-times
\\tnot\\tp[0-9]+\\.b, p[0-9]+/z, p[0-9]+\\.b\\n 1
FAIL: gcc.target/aarch64/sve/pred-not-gen-4.c scan-assembler-not \\tbic\\t
FAIL: gcc.target/aarch64/sve/pred-not-gen-4.c scan-assembler-times
\\tnot\\tp[0-9]+\\.b, p[0-9]+/z, p[0-9]+\\.b\\n 1
We can reclaim the BIC to NOT conversion in post-reload late-combine by
tweaking the patterns. I have a patch for that part. However, we then end up
with duplicate comparisons. We'd need a post-split cse patch, such as:
...
NEXT_PASS (pass_split_after_reload);
NEXT_PASS (pass_postreload_cse);
...
in order to remove the duplicates and recover the intended code. However,
adding a new pass seems too invasive for GCC 15.
Replacing one comparison and two logic ops with two comparisons and one logic
op isn't necessarily a win, although it does reduce the latency for uses of the
non-inverted comparison. I'm therefore not sure it's worth applying the
pattern patch without the follow-on CSE.
We might need to XFAIL these tests for GCC 15 and revisit for GCC 16.
