https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91419
Bug ID: 91419
Summary: [10 Regression]: gcc.dg/tree-ssa/pr91091-2.c,
ssa-fre-61.c, ssa-fre-61.c with r273232
Product: gcc
Version: 10.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: hp at gcc dot gnu.org
CC: rguenth at gcc dot gnu.org
Target Milestone: ---
Target: cris-elf, pru-elf, m68k-unknown-linux-gnu
Commit r273232 (of 2019-07-08) introduced these regressions for (at least)
cris-elf, pru-elf and m68k-unknown-linux-gnu:
FAIL: gcc.dg/tree-ssa/pr91091-2.c scan-tree-dump-times fre1 "x = " 1
FAIL: gcc.dg/tree-ssa/ssa-fre-61.c (test for excess errors)
FAIL: gcc.dg/tree-ssa/ssa-fre-61.c scan-tree-dump-times fre1 "Replaced \\*p" 3
FAIL: gcc.dg/tree-ssa/ssa-fre-77.c scan-tree-dump fre1 "return 1;"
For cris-elf, the "excess errors" are (from gcc.log):
Excess errors:
ssa-fre-61.c:(.text+0x16): undefined reference to `link_error'
ssa-fre-61.c:(.text+0x30): undefined reference to `link_error'
ssa-fre-61.c:(.text+0x4a): undefined reference to `link_error'
The revision is confirmed for cris-elf by own testing. For
pru-elf and m68k-unknown-linux-gnu entries before and after
r273232 on gcc-testresults@ were compared, for example:
Before:
<https://gcc.gnu.org/ml/gcc-testresults/2019-07/msg00954.html>
(Results for
10.0.020190708(experimental)[trunkrevision273226](GCC) testsuite
on pru-unknown-elf)
<https://gcc.gnu.org/ml/gcc-testresults/2019-07/msg00932.html>
(Results for 10.0.0 20190707 (experimental) [trunk revision
273184] (GCC) testsuite on m68k-unknown-linux-gnu)
After:
<https://gcc.gnu.org/ml/gcc-testresults/2019-07/msg01192.html>
(Results for
10.0.020190710(experimental)[trunkrevision273328](GCC) testsuite
on pru-unknown-elf)
<https://gcc.gnu.org/ml/gcc-testresults/2019-07/msg01052.html>
(Results for 10.0.0 20190708 (experimental) [trunk revision
273247] (GCC) testsuite on m68k-unknown-linux-gnu)
Also, more recent gcc-testresults entries confirms that the regressions
remain.
I'm guessing that some target-ABI-structure-related knob has been overlooked,
common to these targets but different to more mainstream targets. One that
seems to match that criteria is PCC_BITFIELD_TYPE_MATTERS.
FWIW, the new test ssa-fre-78.c passes.
