[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
--- Comment #9 from matz at gcc dot gnu dot org 2010-07-26 15:06 ---
Here's a testcase (nicer in the sense of not requiring inlining) that shows
the current compiler botching the nrv - retslot optimizations:
struct S {int x, y, makemelarge[5];};
S __attribute__((noinline)) f (S s) {
S r;
r.x = s.y;
r.y = s.x;
return r;
}
int __attribute__((noinline)) glob (int a, int b)
{
S local = { a, b };
local = f (local);
return local.y;
}
extern C void abort (void);
int main (void)
{
if (glob (1, 3) != 1)
abort ();
return 0;
}
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
--- Comment #10 from rguenth at gcc dot gnu dot org 2010-07-26 16:02 --- Subject: Bug 43784 Author: rguenth Date: Mon Jul 26 16:01:55 2010 New Revision: 162539 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=162539 Log: 2010-07-26 Richard Guenther rguent...@suse.de PR tree-optimization/43784 * tree-nrv.c (dest_safe_for_nrv_p): It's not safe to NRV if the destination is used by the call. * gcc.c-torture/execute/pr43784.c: New testcase. * g++.dg/torture/pr43784.C: Likewise. Added: trunk/gcc/testsuite/g++.dg/torture/pr43784.C trunk/gcc/testsuite/gcc.c-torture/execute/pr43784.c Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog trunk/gcc/tree-nrv.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
--- Comment #11 from rguenth at gcc dot gnu dot org 2010-07-26 16:02 --- Fixed. -- rguenth at gcc dot gnu dot org changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution||FIXED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
--- Comment #8 from joseph at codesourcery dot com 2010-07-24 19:49 --- Subject: Re: [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution On Fri, 23 Jul 2010, rguenth at gcc dot gnu dot org wrote: Joseph, this is really triggering undefined behavior in C, does it? I don't see anything undefined here. Remember that a return is not an assignment and it's only direct assignments of overlapping values that are disallowed (in N1256 this is 6.8.6.4 paragraph 4 and footnote 139; this was added in C90 TC1 following DR#001). -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
--- Comment #5 from rguenth at gcc dot gnu dot org 2010-07-23 11:20 ---
Confirmed. But I think this might trigger undefined behavior according to
the C standard as you are doing
v.e.b = v.d.b;
which has overlapping lhs/rhs. And we end up with
rp:
.LFB0:
.cfi_startproc
movq%rdi, %rax
movl$v, %esi
movl$64, %ecx
rep movsl
ret
.cfi_endproc
compared to 4.5 we end up with return slot optimization for the call to rp().
Testcase that also reproduces with -O[123]:
struct s {
unsigned char a[256];
};
union u {
struct { struct s b; int c; } d;
struct { int c; struct s b; } e;
};
static union u v;
static struct s *p = v.d.b;
static struct s *q = v.e.b;
static struct s __attribute__((noinline)) rp(void)
{
return *p;
}
static void qp(void)
{
*q = rp();
}
int main()
{
int i;
for (i = 0; i 256; i++)
p-a[i] = i;
qp();
for (i = 0; i 256; i++)
if (q-a[i] != i)
__builtin_abort();
return 0;
}
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
Status|UNCONFIRMED |NEW
Ever Confirmed|0 |1
Priority|P3 |P1
Last reconfirmed|-00-00 00:00:00 |2010-07-23 11:20:14
date||
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
--- Comment #6 from rguenth at gcc dot gnu dot org 2010-07-23 11:36 --- We still need to deal with it in the middle-end. I have a patch. -- rguenth at gcc dot gnu dot org changed: What|Removed |Added AssignedTo|unassigned at gcc dot gnu |rguenth at gcc dot gnu dot |dot org |org Status|NEW |ASSIGNED Last reconfirmed|2010-07-23 11:20:14 |2010-07-23 11:36:24 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
--- Comment #7 from rguenth at gcc dot gnu dot org 2010-07-23 11:40 --- Hm, no. I think tree-nrv.c is correct. I was playing with Index: gcc/tree-nrv.c === --- gcc/tree-nrv.c (revision 162450) +++ gcc/tree-nrv.c (working copy) @@ -310,7 +310,8 @@ dest_safe_for_nrv_p (gimple call) if (TREE_CODE (dest) == SSA_NAME) return true; - if (call_may_clobber_ref_p (call, dest)) + if (call_may_clobber_ref_p (call, dest) + || ref_maybe_used_by_stmt_p (call, dest)) return false; return true; but that's overly restrictive for something that is undefined even on the gimple level. Joseph, this is really triggering undefined behavior in C, does it? -- rguenth at gcc dot gnu dot org changed: What|Removed |Added CC||jsm28 at gcc dot gnu dot org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
--- Comment #4 from zsojka at seznam dot cz 2010-06-11 21:02 --- When rp() is declared with __attribute__((pure)), it fails even with -fno-ipa-pure-const. (still fails in r160527) I don't know if this is a bug or if this behaviour is fine according to standard. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
--- Comment #2 from zsojka at seznam dot cz 2010-05-22 17:11 --- Fails with -m32 too (as of r159696) -fno-ipa-profile doesn't help (that flag is new in r158969 compared to r158095) The problem is that temporary struct isn't created, and direct copy *q = *p is done. (rp is called with output parameter (rdi) == q, not a pointer to temporary data on stack) Furthermore, qp() isn't emitted in the object file (in neither r158095 or r158969), even though -fkeep-inline-functions was specified. (I am not sure this is a bug) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
--- Comment #3 from zsojka at seznam dot cz 2010-05-22 17:32 --- -fno-ipa-pure-const prevents the problem (the code is the same as when compiled by r158095) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
--- Comment #1 from zsojka at seznam dot cz 2010-04-18 16:26 --- Created an attachment (id=20409) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20409action=view) reduced testcase (from gcc.c-torture/execute/builtins/pr22237.c) Command line: gcc -Os -fkeep-inline-functions pr43784.c ./a.out -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
[Bug tree-optimization/43784] [4.6 Regression] -Os -fkeep-inline-functions causes FAIL: gcc.c-torture/execute/builtins/pr22237.c execution
-- rguenth at gcc dot gnu dot org changed: What|Removed |Added Target Milestone|--- |4.6.0 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43784
