match.pd: Disallow side-effects in GENERIC for non-COND_EXPR to COND_EXPR simplificat...
authorJakub Jelinek <jakub@redhat.com>
Sat, 15 Feb 2020 11:53:44 +0000 (12:53 +0100)
committerJakub Jelinek <jakub@redhat.com>
Sat, 15 Feb 2020 11:53:44 +0000 (12:53 +0100)
commit187dd955dbee3939c3a2ca7b6839e7f709999125
tree02a3d390dc575f8d7179eabef999a485306e9544
parent55b00d14f4daf671b865550c119dafdeb3139672
match.pd: Disallow side-effects in GENERIC for non-COND_EXPR to COND_EXPR simplifications [PR93744]

As the following testcases show (the first one reported, last two
found by code inspection), we need to disallow side-effects
in simplifications that turn some unconditional expression into conditional
one.  From my little understanding of genmatch.c, it is able to
automatically disallow side effects if the same operand is used multiple
times in the match pattern, maybe if it is used multiple times in the
replacement pattern, and if it is used in conditional contexts in the match
pattern, could it be taught to handle this case too?  If yes, perhaps
just the first hunk could be usable for 8/9 backports (+ the testcases).

2020-02-15  Jakub Jelinek  <jakub@redhat.com>

PR tree-optimization/93744
* match.pd (((m1 >/</>=/<= m2) * d -> (m1 >/</>=/<= m2) ? d : 0,
A - ((A - B) & -(C cmp D)) -> (C cmp D) ? B : A,
A + ((B - A) & -(C cmp D)) -> (C cmp D) ? B : A): For GENERIC, make
sure @2 in the first and @1 in the other patterns has no side-effects.

* gcc.c-torture/execute/pr93744-1.c: New test.
* gcc.c-torture/execute/pr93744-2.c: New test.
* gcc.c-torture/execute/pr93744-3.c: New test.
gcc/ChangeLog
gcc/match.pd
gcc/testsuite/ChangeLog
gcc/testsuite/gcc.c-torture/execute/pr93744-1.c [new file with mode: 0644]
gcc/testsuite/gcc.c-torture/execute/pr93744-2.c [new file with mode: 0644]
gcc/testsuite/gcc.c-torture/execute/pr93744-3.c [new file with mode: 0644]