Hi!
Apparently a recent change (2011-10-1[23] or so) regressed
gcc.dg/guality/asm-1.c test on x86_64-linux on the 4.6 branch.
This turned out to be a bug in loc_descriptor which has been passing wrong
mode on the recursive call, so when we had var_location note containing
SImode subreg of some DImode expression, the inner loc_descriptor would
just give up because of the mode mismatch.
Fixed thusly (here the trunk version of the patch, attached 4.6 version),
bootstrapped/regtested on x86_64-linux and i686-linux (both branches in each
case), ok for trunk/4.6?
2011-10-18 Jakub Jelinek <ja...@redhat.com>
* dwarf2out.c (loc_descriptor): For SUBREG pass SUBREG_REG's mode as
second argument instead of mode.
--- gcc/dwarf2out.c.jj 2011-09-08 11:21:10.000000000 +0200
+++ gcc/dwarf2out.c 2011-10-18 14:09:46.000000000 +0200
@@ -12502,7 +12502,8 @@ loc_descriptor (rtx rtl, enum machine_mo
legitimate to make the Dwarf info refer to the whole register which
contains the given subreg. */
if (REG_P (SUBREG_REG (rtl)) && subreg_lowpart_p (rtl))
- loc_result = loc_descriptor (SUBREG_REG (rtl), mode, initialized);
+ loc_result = loc_descriptor (SUBREG_REG (rtl),
+ GET_MODE (SUBREG_REG (rtl)), initialized);
else
goto do_default;
break;
Jakub
2011-10-18 Jakub Jelinek <ja...@redhat.com>
* dwarf2out.c (loc_descriptor): For SUBREG pass SUBREG_REG's mode as
second argument instead of mode.
--- gcc/dwarf2out.c.jj 2011-08-02 10:42:23.000000000 +0200
+++ gcc/dwarf2out.c 2011-10-18 11:21:38.000000000 +0200
@@ -14608,7 +14608,8 @@ loc_descriptor (rtx rtl, enum machine_mo
up an entire register. For now, just assume that it is
legitimate to make the Dwarf info refer to the whole register which
contains the given subreg. */
- loc_result = loc_descriptor (SUBREG_REG (rtl), mode, initialized);
+ loc_result = loc_descriptor (SUBREG_REG (rtl),
+ GET_MODE (SUBREG_REG (rtl)), initialized);
break;
case REG:
