Hi!
When adding checks for PIC register in address cost calculation
(http://gcc.gnu.org/ml/gcc-cvs/2014-10/msg00411.html) it was meant to affect
only RTL passes.
Since !pic_offset_table_rtx is not enough for it (I see that
pic_offset_table_rtx enabled on GIMPLE level) following change explicitly adds
this restriction.
Bootstrapped and regtested with RUNTESTFLAGS=--target_board='unix{-m32,-fpic}'
Is it ok for trunk?
Thanks,
Igor
Changelog:
2014-12-12 Igor Zamyatin igor.zamya...@intel.com
* config/i386/i386.c (ix86_address_cost): Add explicit restriction
to RTL level for the check for PIC register.
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index fffddfc..799411c 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -12802,12 +12802,14 @@ ix86_address_cost (rtx x, machine_mode, addr_space_t,
bool)
Therefore only pic_offset_table_rtx could be hoisted out, which is not
profitable for x86. */
if (parts.base
- (!pic_offset_table_rtx
- || REGNO (pic_offset_table_rtx) != REGNO(parts.base))
+ (current_pass-type == GIMPLE_PASS
+ || (!pic_offset_table_rtx
+ || REGNO (pic_offset_table_rtx) != REGNO(parts.base)))
(!REG_P (parts.base) || REGNO (parts.base) = FIRST_PSEUDO_REGISTER)
parts.index
- (!pic_offset_table_rtx
- || REGNO (pic_offset_table_rtx) != REGNO(parts.index))
+ (current_pass-type == GIMPLE_PASS
+ || (!pic_offset_table_rtx
+ || REGNO (pic_offset_table_rtx) != REGNO(parts.index)))
(!REG_P (parts.index) || REGNO (parts.index) = FIRST_PSEUDO_REGISTER)
parts.base != parts.index)
cost++;