* [PATCH] Fix SCCVN aggregate copy code some more
@ 2011-08-11 16:50 Richard Guenther
2012-09-06 12:23 ` H.J. Lu
0 siblings, 1 reply; 2+ messages in thread
From: Richard Guenther @ 2011-08-11 16:50 UTC (permalink / raw)
To: gcc-patches
When adjusting the C++ FE code generation for base type copies
I noticed the XFAILs on g++.dg/tree-ssa/pr41186.C and investigated
things a bit. We can fix the testcase easily iff the frontend
emits the MEM_REF variants, thus the following patch.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk.
Richard.
2011-08-11 Richard Guenther <rguenther@suse.de>
* tree-ssa-sccvn.c (vn_reference_lookup_3): Avoid redundant
lookups, make looking through aggregate copies stronger.
Index: gcc/tree-ssa-sccvn.c
===================================================================
*** gcc/tree-ssa-sccvn.c (revision 177649)
--- gcc/tree-ssa-sccvn.c (working copy)
*************** vn_reference_lookup_3 (ao_ref *ref, tree
*** 1351,1367 ****
{
VEC (vn_reference_op_s, heap) *tem;
tree lhs = gimple_assign_lhs (def_stmt);
/* Avoid re-allocation overhead. */
VEC_truncate (vn_reference_op_s, lhs_ops, 0);
copy_reference_ops_from_ref (lhs, &lhs_ops);
tem = lhs_ops;
! lhs_ops = valueize_refs (lhs_ops);
gcc_assert (lhs_ops == tem);
! lhs_ref_ok = ao_ref_init_from_vn_reference (&lhs_ref, get_alias_set (lhs),
! TREE_TYPE (lhs), lhs_ops);
! if (lhs_ref_ok
! && !refs_may_alias_p_1 (ref, &lhs_ref, true))
! return NULL;
}
base = ao_ref_base (ref);
--- 1351,1377 ----
{
VEC (vn_reference_op_s, heap) *tem;
tree lhs = gimple_assign_lhs (def_stmt);
+ bool valueized_anything = false;
/* Avoid re-allocation overhead. */
VEC_truncate (vn_reference_op_s, lhs_ops, 0);
copy_reference_ops_from_ref (lhs, &lhs_ops);
tem = lhs_ops;
! lhs_ops = valueize_refs_1 (lhs_ops, &valueized_anything);
gcc_assert (lhs_ops == tem);
! if (valueized_anything)
! {
! lhs_ref_ok = ao_ref_init_from_vn_reference (&lhs_ref,
! get_alias_set (lhs),
! TREE_TYPE (lhs), lhs_ops);
! if (lhs_ref_ok
! && !refs_may_alias_p_1 (ref, &lhs_ref, true))
! return NULL;
! }
! else
! {
! ao_ref_init (&lhs_ref, lhs);
! lhs_ref_ok = true;
! }
}
base = ao_ref_base (ref);
*************** vn_reference_lookup_3 (ao_ref *ref, tree
*** 1469,1474 ****
--- 1479,1498 ----
j--;
}
+ /* ??? The innermost op should always be a MEM_REF and we already
+ checked that the assignment to the lhs kills vr. Thus for
+ aggregate copies using char[] types the vn_reference_op_eq
+ may fail when comparing types for compatibility. But we really
+ don't care here - further lookups with the rewritten operands
+ will simply fail if we messed up types too badly. */
+ if (j == 0 && i == 0
+ && VEC_index (vn_reference_op_s, lhs_ops, 0)->opcode == MEM_REF
+ && VEC_index (vn_reference_op_s, vr->operands, i)->opcode == MEM_REF
+ && tree_int_cst_equal
+ (VEC_index (vn_reference_op_s, lhs_ops, 0)->op0,
+ VEC_index (vn_reference_op_s, vr->operands, i)->op0))
+ i--, j--;
+
/* i now points to the first additional op.
??? LHS may not be completely contained in VR, one or more
VIEW_CONVERT_EXPRs could be in its way. We could at least
Index: gcc/testsuite/g++.dg/tree-ssa/pr41186.C
===================================================================
*** gcc/testsuite/g++.dg/tree-ssa/pr41186.C (revision 177649)
--- gcc/testsuite/g++.dg/tree-ssa/pr41186.C (working copy)
***************
*** 1,5 ****
/* { dg-do compile } */
! /* { dg-options "-O -fdump-tree-fre1-details" } */
struct Foo {
Foo() {};
--- 1,5 ----
/* { dg-do compile } */
! /* { dg-options "-O -fno-tree-sra -fdump-tree-fre1-details" } */
struct Foo {
Foo() {};
*************** int main()
*** 30,35 ****
}
/* { dg-final { scan-tree-dump "Replaced b1.b with 1" "fre1" } } */
! /* { dg-final { scan-tree-dump "Replaced b1.i with 0" "fre1" { xfail *-*-* } } } */
! /* { dg-final { scan-tree-dump "Replaced b1.f with 1" "fre1" { xfail *-*-* } } } */
/* { dg-final { cleanup-tree-dump "fre1" } } */
--- 30,35 ----
}
/* { dg-final { scan-tree-dump "Replaced b1.b with 1" "fre1" } } */
! /* { dg-final { scan-tree-dump "Replaced b1.D.\[0-9\]*.i with 0" "fre1" } } */
! /* { dg-final { scan-tree-dump "Replaced b1.D.\[0-9\]*.f with 1" "fre1" } } */
/* { dg-final { cleanup-tree-dump "fre1" } } */
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Fix SCCVN aggregate copy code some more
2011-08-11 16:50 [PATCH] Fix SCCVN aggregate copy code some more Richard Guenther
@ 2012-09-06 12:23 ` H.J. Lu
0 siblings, 0 replies; 2+ messages in thread
From: H.J. Lu @ 2012-09-06 12:23 UTC (permalink / raw)
To: Richard Guenther; +Cc: gcc-patches
On Thu, Aug 11, 2011 at 8:31 AM, Richard Guenther <rguenther@suse.de> wrote:
>
> When adjusting the C++ FE code generation for base type copies
> I noticed the XFAILs on g++.dg/tree-ssa/pr41186.C and investigated
> things a bit. We can fix the testcase easily iff the frontend
> emits the MEM_REF variants, thus the following patch.
>
> Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk.
>
> Richard.
>
> 2011-08-11 Richard Guenther <rguenther@suse.de>
>
> * tree-ssa-sccvn.c (vn_reference_lookup_3): Avoid redundant
> lookups, make looking through aggregate copies stronger.
>
This caused:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54498
--
H.J.
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2012-09-06 12:23 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2011-08-11 16:50 [PATCH] Fix SCCVN aggregate copy code some more Richard Guenther
2012-09-06 12:23 ` H.J. Lu
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).