This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] Fix MEM_IN_STRUCT_P/MEM_SCALAR_P during expansion (PR rtl-optimization/47866)
- From: Jakub Jelinek <jakub at redhat dot com>
- To: gcc-patches at gcc dot gnu dot org
- Cc: Jeff Law <law at redhat dot com>, Mark Mitchell <mark at codesourcery dot com>
- Date: Tue, 8 Mar 2011 23:16:06 +0100
- Subject: [PATCH] Fix MEM_IN_STRUCT_P/MEM_SCALAR_P during expansion (PR rtl-optimization/47866)
- Reply-to: Jakub Jelinek <jakub at redhat dot com>
Hi!
torture/vector-2.c testcase is miscompiled on ia64, apparently because
a store uses a result of POST_MODIFY, which during sched1 without cselib
is considered variable and is marked as MEM_IN_STRUCT_P while all other
memory stores/loads for that variable are MEM_SCALAR_P and
fixed_scalar_and_varying_struct_p in alias.c thus says that
a fixed scalar can't alias with a varying struct.
I believe the bug is in saying that the store is MEM_IN_STRUCT_P, if all
other stores/loads from that area are MEM_SCALAR_P (the variable
is a 16 byte vector, i.e. TImode variable on ia64), then this store
should be MEM_SCALAR_P too.
In *.optimized dump this is:
vector(4) int t;
vector(4) int t;
vector(4) int a0;
int D.2001;
int D.2000;
int D.1999;
int D.1998;
<bb 2>:
t = { 0, 0, 0, 0 };
BIT_FIELD_REF <t, 32, 0> = 1;
a0_18 = t;
D.1998_3 = BIT_FIELD_REF <a0_18, 32, 0>;
D.1999_4 = BIT_FIELD_REF <a0_18, 32, 32>;
D.2000_5 = BIT_FIELD_REF <a0_18, 32, 64>;
D.2001_6 = BIT_FIELD_REF <a0_18, 32, 96>;
printf ("%d %d %d %d\n", D.1998_3, D.1999_4, D.2000_5, D.2001_6);
t = { 0, 0, 0, 0 };
BIT_FIELD_REF <t, 32, 32> = 1;
a0_19 = t;
D.1998_8 = BIT_FIELD_REF <a0_19, 32, 0>;
D.1999_9 = BIT_FIELD_REF <a0_19, 32, 32>;
D.2000_10 = BIT_FIELD_REF <a0_19, 32, 64>;
D.2001_11 = BIT_FIELD_REF <a0_19, 32, 96>;
printf ("%d %d %d %d\n", D.1998_8, D.1999_9, D.2000_10, D.2001_11);
and as t isn't AGGREGATE_TYPE nor COMPLEX_TYPE and is a decl,
it is marked MEM_SCALAR_P and e.g. set_mem_attributes_minus_bitpos
once MEM_SCALAR_P is set doesn't change it to MEM_IN_STRUCT_P
because of BIT_FIELD_REF etc. The BIT_FIELD_REF <t, 32, *> = 1
stores are done through store_field though, and for some reason
the code sets MEM_IN_STRUCT_P in that codepath unconditionally.
Changing this fixes the testcase, bootstrapped/regtested on x86_64-linux and
i686-linux. I'll try to test this on ia64-linux tomorrow.
2011-03-08 Jakub Jelinek <jakub@redhat.com>
PR rtl-optimization/47866
* expr.c (store_field): If MEM_SCALAR_P (target), don't use
MEM_SET_IN_STRUCT_P (to_rtx, 1), just set MEM_IN_STRUCT_P (to_rtx)
if target wasn't scalar.
--- gcc/expr.c.jj 2011-02-04 16:45:02.000000000 +0100
+++ gcc/expr.c 2011-03-08 20:49:19.531545778 +0100
@@ -5924,7 +5924,8 @@ store_field (rtx target, HOST_WIDE_INT b
if (to_rtx == target)
to_rtx = copy_rtx (to_rtx);
- MEM_SET_IN_STRUCT_P (to_rtx, 1);
+ if (!MEM_SCALAR_P (to_rtx))
+ MEM_IN_STRUCT_P (to_rtx) = 1;
if (!MEM_KEEP_ALIAS_SET_P (to_rtx) && MEM_ALIAS_SET (to_rtx) != 0)
set_mem_alias_set (to_rtx, alias_set);
Jakub