[Bug target/53724] New: ICE when using the 'd' asm operand modifier

[Bug target/53724] New: ICE when using the 'd' asm operand modifier

[svfuerst at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53724

             Bug #: 53724
           Summary: ICE when using the 'd' asm operand modifier
    Classification: Unclassified
           Product: gcc
           Version: 4.7.1
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: svfuerst@gmail.com


The following gives:
internal compiler error: in print_reg, at config/i386/i386.c:13692

typedef double ymmd __attribute__ ((vector_size (32)));
ymmd func(ymmd p1, ymmd p2)
{
    asm ("vfmadd132pd %1, %d0"
        : "+x" (p1) : "x" (p2));
    return p1;
}

when compiled with "gcc-4.7 -mavx -O2 gcc_asm.c -S -o gcc_asm.S"
using gcc-4.7.real (Debian 4.7.1-1) 4.7.0 on x86_64

[Bug target/53724] ICE when using the 'd' asm operand modifier

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53724

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hjl.tools at gmail dot com

--- Comment #1 from Uros Bizjak <ubizjak at gmail dot com> 2012-06-20 07:45:35 UTC ---
(In reply to comment #0)
> The following gives:
> internal compiler error: in print_reg, at config/i386/i386.c:13692
> 
> typedef double ymmd __attribute__ ((vector_size (32)));
> ymmd func(ymmd p1, ymmd p2)
> {
>     asm ("vfmadd132pd %1, %d0"
>         : "+x" (p1) : "x" (p2));
>     return p1;
> }

For some reason we assert that ymm registers can't be duplicated. H.J.?

[Bug target/53724] ICE when using the 'd' asm operand modifier

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53724

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-*-*, i?86-*-*
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-06-20
     Ever Confirmed|0                           |1
           Severity|enhancement                 |normal

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-06-20 09:11:42 UTC ---
Confirmed.

[Bug target/53724] ICE when using the 'd' asm operand modifier

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53724

--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> 2012-06-20 12:25:30 UTC ---
'd' is used to encode SSE instructions with VEX
encoding.  There are no SSE instructions with YMM
registers.

[Bug target/53724] ICE when using the 'd' asm operand modifier

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53724

--- Comment #4 from H.J. Lu <hjl.tools at gmail dot com> 2012-06-20 12:38:12 UTC ---
Something like this

diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index e2f5740..0daf601 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -13879,7 +13879,8 @@ print_reg (rtx x, int code, FILE *file)
     case 32:
       if (SSE_REG_P (x))
     {
-      gcc_assert (!duplicated);
+      if (duplicated)
+        error ("unsupported `d' specifier on AVX register");
       putc ('y', file);
       fputs (hi_reg_name[REGNO (x)] + 1, file);
       return;