| Message ID | 79a550e3-d3ab-6319-6ae5-0ef20a0286b8@ispras.ru |
|---|---|
| State | New |
| Headers | show |
| Series | Fix PR 85423 (Re: Fix PRs 80463, 83972, 83480) | expand |
On Mon, Apr 23, 2018 at 05:03:09PM +0300, Andrey Belevantsev wrote: > I have put the test in gcc.dg though it requires -march=nano. Do you want > me to create an extra machine-dependent test? If it is a compile time test, no need for that. Just add /* { dg-additional-options "-march=nano" { target i?86-*-* x86_64-*-* } } */ Jakub
On Mon, 23 Apr 2018, Andrey Belevantsev wrote: > Hello, > > So this PR shows that I have incorrectly mirrored the conditional from > sched-deps.c that creates the dependence from a debug insn on the previous > insn (see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80463#c3 for the > hunk). Thus we have incorrectly discarded some legitimate debug-debug > dependencies. The straightforward fix works for all of four PRs, tested on > x86-64. > > I have put the test in gcc.dg though it requires -march=nano. Do you want > me to create an extra machine-dependent test? I see Jakub addressed this question (thanks!) so please incorporate his suggestion. There's also a typo in the ChangeLog, OK with that fixed. Alexander > 2018-04-23 Andrey Belevantsev <abel@ispras.ru> > > PR rtl-optimization/85423 > > * sel-sched-ir.c (has_dependence_note_mem_dep): Only discard Should be "has_dependence_note_dep" (without the "mem_" ;) Alexander
diff --git a/gcc/sel-sched-ir.c b/gcc/sel-sched-ir.c index ee970522890..85ff5bd3eb4 100644 --- a/gcc/sel-sched-ir.c +++ b/gcc/sel-sched-ir.c @@ -3308,7 +3308,7 @@ has_dependence_note_dep (insn_t pro, ds_t ds ATTRIBUTE_UNUSED) that a bookkeeping copy should be movable as the original insn. Detect that here and allow that movement if we allowed it before in the first place. */ - if (DEBUG_INSN_P (real_con) + if (DEBUG_INSN_P (real_con) && !DEBUG_INSN_P (real_pro) && INSN_UID (NEXT_INSN (pro)) == INSN_UID (real_con)) return; diff --git a/gcc/testsuite/gcc.dg/pr85423.c b/gcc/testsuite/gcc.dg/pr85423.c new file mode 100644 index 00000000000..21d4a2eb4b9 --- /dev/null +++ b/gcc/testsuite/gcc.dg/pr85423.c @@ -0,0 +1,26 @@ +/* { dg-do compile { target powerpc*-*-* ia64-*-* i?86-*-* x86_64-*-* } } */ +/* { dg-options "-O2 -fselective-scheduling2 -fvar-tracking-assignments -fno-guess-branch-probability -fno-peephole2 -fno-ssa-phiopt -fno-tree-pre --param max-jump-thread-duplication-stmts=8 -w" } */ + +int vn, xm; + +void +i1 (int); + +void +mb (int *ap, int ev) +{ + while (vn < 1) + { + i1 (vn); + + ev += *ap && ++vn; + + while (xm < 1) + ++xm; + + if (*ap == 0) + *ap = ev; + + ++vn; + } +}
Hello, So this PR shows that I have incorrectly mirrored the conditional from sched-deps.c that creates the dependence from a debug insn on the previous insn (see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80463#c3 for the hunk). Thus we have incorrectly discarded some legitimate debug-debug dependencies. The straightforward fix works for all of four PRs, tested on x86-64. I have put the test in gcc.dg though it requires -march=nano. Do you want me to create an extra machine-dependent test? Best, Andrey 2018-04-23 Andrey Belevantsev <abel@ispras.ru> PR rtl-optimization/85423 * sel-sched-ir.c (has_dependence_note_mem_dep): Only discard dependencies to debug insns when previous insn is non-debug. * gcc.dg/pr85423.c: New test.