[Bug rtl-optimization/43653] [4.3/4.4/4.5/4.6 Regression] ICE at reload1.c:1188 with -O1 -ftree-vectorize and empty struct

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43653

--- Comment #10 from Uros Bizjak <ubizjak at gmail dot com> 2011-02-15 22:41:40 UTC ---
(In reply to comment #9)
> I think an x86 maintainer is going to need to take a look at this.

> Which simply isn't going to work and it's all downhill from there, including a
> surprise secondary memory allocation which ultimately triggers the ICE.
> 
> I was able to hack things to work by enabling a secondary reload when copying
> (plus (sp) (whatever)) into SSE_REGS, but that may not be the best solution. 
> Someone with more x86 internal knowledge should look at this.

Hm, indeed. SSE can't handle non-vector PLUS insn, so reload needs a bit of
help.

Following patch fixes the testcase:

Index: i386.c
===================================================================
--- i386.c    (revision 170189)
+++ i386.c    (working copy)
@@ -28313,6 +28313,11 @@ ix86_secondary_reload (bool in_p, rtx x,
     return Q_REGS;
     }

+  if (in_p && GET_CODE (x) == PLUS
+      && !VECTOR_MODE_P (mode)
+      && rclass == SSE_REGS)
+    return GENERAL_REGS;
+
   return NO_REGS;
 }


And generates:

    movq    .LC1(%rip), %rax
    leaq    (%rsp,%rax), %rax
    movq    %rax, -112(%rsp)
    movq    -112(%rsp), %xmm1
    punpcklqdq    %xmm1, %xmm1