http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49226
Summary: problematic _Complex long double argument passing
Product: gcc
Version: 4.7.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: ebotca...@gcc.gnu.org
Blocks: 48830
Target: ia64-*-*
rtlanal.c:simplify_subreg_regno has these lines at the end:
/* See whether (reg:YMODE YREGNO) is valid.
??? We allow invalid registers if (reg:XMODE XREGNO) is also invalid.
This is a kludge to work around how float/complex arguments are passed
on 32-bit SPARC and should be fixed. */
if (!HARD_REGNO_MODE_OK (yregno, ymode)
&& HARD_REGNO_MODE_OK (xregno, xmode))
return -1;
It turns out that they are outdated, removing the kludge as follows:
/* See whether (reg:YMODE YREGNO) is valid. */
if (!HARD_REGNO_MODE_OK (yregno, ymode))
return -1;
works fine on the SPARC (both 32-bit and 64-bit). Now there is a problem on
IA-64. For the testcase to be attached, the function is invoked on:
#0 simplify_subreg_regno (xregno=122, xmode=XCmode, offset=0, ymode=XFmode)
at /home/eric/svn/gcc/gcc/rtlanal.c:3461
3461 return (int) yregno;
and must return a valid regno, although everything is invalid according to the
HARD_REGNO_MODE_OK macro, otherwise wrong code is generated:
--- scalar-by-value-3_0.s 2011-05-29 15:26:12.174486000 -0400
+++ scalar-by-value-3.s 2011-05-29 15:26:29.694284000 -0400
@@ -453,10 +453,10 @@
stfe [r14] = f6
mov r14 = r34
;;
- ld8 r40 = [r14]
+ ld8 r38 = [r14]
adds r14 = 8, r34
;;
- ld8 r41 = [r14]
+ ld8 r39 = [r14]
br.call.sptk.many b0 = testvacld#
mov r1 = r35
mov r36 = r0
This is on the way of PR target/48830.
