[Bug rtl-optimization/37273] [4.4 Regression] IRA does not re-materializes addresses (loads from the TOC)

2008-12-31 Thread pinskia at gcc dot gnu dot org 


--- Comment #2 from pinskia at gcc dot gnu dot org  2008-12-31 21:30 ---
Note the testcase is from gcc.dg/20020103-1.c.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37273

[Bug rtl-optimization/37273] [4.4 Regression] IRA does not re-materializes addresses (loads from the TOC)

2008-09-09 Thread jsm28 at gcc dot gnu dot org 


-- 

jsm28 at gcc dot gnu dot org changed:

   What|Removed |Added

   Priority|P3  |P2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37273

[Bug rtl-optimization/37273] [4.4 Regression] IRA does not re-materializes addresses (loads from the TOC)

2008-08-28 Thread pinskia at gcc dot gnu dot org 


--- Comment #1 from pinskia at gcc dot gnu dot org  2008-08-29 05:27 ---
This spilling also happens on x86-darwin.  Note x86-darwin uses -fPIC by
default.
For PPC it looks like IRA is not looking into REG_EQUAL:
We have in the dump right before IRA:
(insn:HI 29 26 30 2 gcc/gcc/testsuite/gcc.dg/20020103-1.c:38 (set (reg:DI 3 3)
(reg/f:DI 119)) 352 {*movdi_internal64} (expr_list:REG_DEAD (reg/f:DI
119)
(expr_list:REG_EQUAL (symbol_ref:DI ("ext") [flags 0xc0] )
(nil

So we had spilled reg 119 instead of the re-materialize the address from the
REG_EQUAL.

x86-darwin looks like a different issue though, but it looks like IRA is not
calling targetm.delegitimize_address to get the original address to
re-materialize it after the asm.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37273

[Bug rtl-optimization/37273] [4.4 Regression] IRA does not re-materializes addresses (loads from the TOC)

2008-08-28 Thread pinskia at gcc dot gnu dot org 


-- 

pinskia at gcc dot gnu dot org changed:

   What|Removed |Added

   Target Milestone|--- |4.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37273