This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
fix SPE prologue
- From: Nathan Sidwell <nathan at codesourcery dot com>
- To: GCC Patches <gcc-patches at gcc dot gnu dot org>, David Edelsohn <dje at watson dot ibm dot com>
- Date: Tue, 11 Dec 2007 12:35:47 +0000
- Subject: fix SPE prologue
I've applied this obvious fix for an SPE prologue problem. We were generating
offsets relative to $sp, which is not correct when there's a frame pointer. We
should be unconditionally using frame_reg_rtx, which will either be r1 or r12 as
appropriate. Here's the assembly we get for a reduced compile/920501-4.i
testcase with -O1 -mcpu=8548 -mfloat-gprs=double -mspe=yes -mabi=spe -fPIC.
without patch:
mr 12,1
lis 0,0xffff
ori 0,0,32744
stwux 1,1,0
mflr 0
stw 0,4(12)
addi 11,1,-24 <-- 24 bytes below stack pointer
evstdd 30,0(11)
with patch:
mr 12,1
lis 0,0xffff
ori 0,0,32744
stwux 1,1,0
mflr 0
stw 0,4(12)
addi 11,12,-24 <-- 24 bytes below frame pointer
evstdd 30,0(11)
Epilogue generation was already correct.
nathan
--
Nathan Sidwell :: http://www.codesourcery.com :: CodeSourcery
2007-12-11 Nathan Sidwell <nathan@codesourcery.com>
* config/rs6000/rs6000.c (rs6000_emit_prologue): Use frame_reg_rtx
for SPE saves.
Index: config/rs6000/rs6000.c
===================================================================
--- config/rs6000/rs6000.c (revision 130773)
+++ config/rs6000/rs6000.c (working copy)
@@ -15694,7 +15694,7 @@ rs6000_emit_prologue (void)
if (spe_regs_addressable_via_sp)
{
- spe_save_area_ptr = sp_reg_rtx;
+ spe_save_area_ptr = frame_reg_rtx;
spe_offset = info->spe_gp_save_offset + sp_offset;
}
else
@@ -15715,7 +15715,7 @@ rs6000_emit_prologue (void)
}
spe_save_area_ptr = gen_rtx_REG (Pmode, 11);
- emit_insn (gen_addsi3 (spe_save_area_ptr, sp_reg_rtx,
+ emit_insn (gen_addsi3 (spe_save_area_ptr, frame_reg_rtx,
GEN_INT (info->spe_gp_save_offset + sp_offset)));
spe_offset = 0;