On 05/05/2017 09:09 AM, Martin Sebor wrote:
On 05/05/2017 08:53 AM, Jeff Law wrote:
On 05/05/2017 08:03 AM, Xi Ruoyao wrote:
Hi all,
I'm testing GCC 7.1.0 and I've got some strange "UNRESOLVED" results.
The corresponding fragment of gcc.sum:
... ...
PASS: gcc.dg/tree-prof/wcoverage-mismatch.c compilation, -g
UNSUPPORTED:
/home/xry111/gcc/gcc-svn-git/gcc/testsuite/gcc.dg/tree-prof/wcoverage-mismatch.c
-fauto-profile: cannot run
create_gcov
Running
/home/xry111/gcc/gcc-svn-git/gcc/testsuite/gcc.dg/tree-ssa/tree-ssa.exp
...
PASS: gcc.dg/tree-ssa/20030530-2.c (test for excess errors)
UNRESOLVED: gcc.dg/tree-prof/wcoverage-mismatch.c scan-tree-dump-times
dom2 "first_gp_reg_save" 1
UNRESOLVED: gcc.dg/tree-prof/wcoverage-mismatch.c scan-tree-dump-times
dom2 "\\+" 1
PASS: gcc.dg/tree-ssa/20030611-1.c (test for excess errors)
... ...
It seems that some tests in tree-prof "leaked" into tree-ssa (I don't
have
the tool create_gcov). Then tree-ssa.exp can't handle them and produce
"UNRESOLVED" result.
Is this a known issue or a bug?
I'm using --enable-checking=release and LTO bootstrapping. Could them
be guilty?
I think there's something broken in the auto profiling dejagnu harness.
Test results have been unstable ever since that code was added to the
tree. I remove gcc/config/i386/gcc-auto-profile to get consistent
results, but we really need to figure out what's wrong with the dejagnu
support.
Many of the profiling tests routinely fail for me in parallel
builds due to bug 77684. A workaround (solution) in comment
#1 makes them pass.
Given the level of pain this causes and the inability for a mere mortal
user to change the value, I'd suggest somehow forcing the profiling
tests to run serially would be wise. Not sure how best to do that,
particularly with the auto-adjusting parallel support we have in the
testsuite these days.
jeff
