This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[patch committed SH] Fix PR target/66591
- From: Kaz Kojima <kkojima at rr dot iij4u dot or dot jp>
- To: gcc-patches at gcc dot gnu dot org
- Date: Sat, 20 Jun 2015 08:03:04 +0900 (JST)
- Subject: [patch committed SH] Fix PR target/66591
- Authentication-results: sourceware.org; auth=none
I've committed the attached patch to fix PR target/66591 which is
an ice-on-valid-code with -mlra.
SH has only one register in INDEX_REGISTER_CLASS and if the index
term of a memory address is a subreg of DImode register, LRA can't
find appropriate hardware registers in the problematic situation.
It looks to be rare even on SH in the actual codes and seems that
pre-allocating R0 for that index term in sh.c:prepare_move_operands
is enough ATM. The patch is tested on sh4-unknown-linux-gnu with
and without -mlra. CSiBE shows no visible regression.
Regards,
kaz
--
2015-06-19 Kaz Kojima <kkojima@gcc.gnu.org>
PR target/66591
* config/sh/sh.c (prepare_move_operands): Replace subreg
index term with R0 for base and index addressing.
diff --git a/config/sh/sh.c b/config/sh/sh.c
index 8fc8197..b338a6a 100644
--- a/config/sh/sh.c
+++ b/config/sh/sh.c
@@ -1777,10 +1777,14 @@ prepare_move_operands (rtx operands[], machine_mode mode)
target/55212.
We split possible load/store to two move insns via r0 so as to
shorten R0 live range. It will make some codes worse but will
- win on avarage for LRA. */
+ win on average for LRA.
+ Also when base+index addressing is used and the index term is
+ a subreg, LRA assumes that more hard registers can be available
+ in some situation. It isn't the case for SH in the problematic
+ case. We can pre-allocate R0 for that index term to avoid
+ the issue. See PR target/66591. */
else if (sh_lra_p ()
&& TARGET_SH1 && ! TARGET_SH2A
- && (mode == QImode || mode == HImode)
&& ((REG_P (operands[0]) && MEM_P (operands[1]))
|| (REG_P (operands[1]) && MEM_P (operands[0]))))
{
@@ -1788,7 +1792,8 @@ prepare_move_operands (rtx operands[], machine_mode mode)
rtx reg = operands[load_p ? 0 : 1];
rtx adr = XEXP (operands[load_p ? 1 : 0], 0);
- if (REGNO (reg) >= FIRST_PSEUDO_REGISTER
+ if ((mode == QImode || mode == HImode)
+ && REGNO (reg) >= FIRST_PSEUDO_REGISTER
&& GET_CODE (adr) == PLUS
&& REG_P (XEXP (adr, 0))
&& (REGNO (XEXP (adr, 0)) >= FIRST_PSEUDO_REGISTER)
@@ -1800,6 +1805,17 @@ prepare_move_operands (rtx operands[], machine_mode mode)
emit_move_insn (r0_rtx, operands[1]);
operands[1] = r0_rtx;
}
+ if (REGNO (reg) >= FIRST_PSEUDO_REGISTER
+ && GET_CODE (adr) == PLUS
+ && REG_P (XEXP (adr, 0))
+ && (REGNO (XEXP (adr, 0)) >= FIRST_PSEUDO_REGISTER)
+ && SUBREG_P (XEXP (adr, 1))
+ && REG_P (SUBREG_REG (XEXP (adr, 1))))
+ {
+ rtx r0_rtx = gen_rtx_REG (GET_MODE (XEXP (adr, 1)), R0_REG);
+ emit_move_insn (r0_rtx, XEXP (adr, 1));
+ XEXP (adr, 1) = r0_rtx;
+ }
}
}