[Bug tree-optimization/34437] New: several test failures for gcc.dg/vect/no-scevccp-*

[Bug tree-optimization/34437] New: several test failures for gcc.dg/vect/no-scevccp-*

[janis at gcc dot gnu dot org]

The following test failures on powerpc64-linux with -m32:

    FAIL: gcc.dg/vect/no-scevccp-outer-10a.c execution test
    FAIL: gcc.dg/vect/no-scevccp-outer-10b.c execution test
    FAIL: gcc.dg/vect/no-scevccp-outer-11.c execution test
    FAIL: gcc.dg/vect/no-scevccp-outer-12.c execution test
    FAIL: gcc.dg/vect/no-scevccp-outer-15.c execution test
    FAIL: gcc.dg/vect/no-scevccp-outer-22.c execution test
    FAIL: gcc.dg/vect/no-scevccp-outer-7.c execution test
    FAIL: gcc.dg/vect/no-scevccp-noreassoc-outer-3.c execution test

began with this patch:

    http://gcc.gnu.org/viewcvs?view=rev&rev=129920

    r129920 | jakub | 2007-11-06 08:28:37 +0000 (Tue, 06 Nov 2007)

The same tests pass when run with -m64.


-- 
           Summary: several test failures for gcc.dg/vect/no-scevccp-*
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: janis at gcc dot gnu dot org
GCC target triplet: powerpc-unknown-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34437

[Bug tree-optimization/34437] several test failures for gcc.dg/vect/no-scevccp-*

[jakub at gcc dot gnu dot org]

------- Comment #1 from jakub at gcc dot gnu dot org  2007-12-11 23:00 -------
See http://gcc.gnu.org/ml/gcc-patches/2007-11/msg00340.html
Apparently -m32 -maltivec has been always badly broken.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34437

[Bug tree-optimization/34437] several test failures for gcc.dg/vect/no-scevccp-*

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2007-12-11 23:05 -------
No the autovectorizer testsuite uses -mabi=altivec also.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34437

[Bug tree-optimization/34437] several test failures for gcc.dg/vect/no-scevccp-*

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2007-12-11 23:08 -------
Or someone messed up vect.exp :(.

Anyways try this patch:
Index: testsuite/gcc.dg/vect/vect.exp
===================================================================
--- testsuite/gcc.dg/vect/vect.exp      (revision 130629)
+++ testsuite/gcc.dg/vect/vect.exp      (working copy)
@@ -49,7 +49,7 @@ if  [istarget "powerpc-*paired*"]  {
        return
     }

-    lappend DEFAULT_VECTCFLAGS "-maltivec"
+    lappend DEFAULT_VECTCFLAGS "-maltivec -mabi=altivec"
     if [check_vmx_hw_available] {
        set dg-do-what-default run
     } else {


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34437

[Bug tree-optimization/34437] several test failures for gcc.dg/vect/no-scevccp-*

[janis at gcc dot gnu dot org]

------- Comment #4 from janis at gcc dot gnu dot org  2007-12-11 23:27 -------
The patch in comment #3 works, except that there need to be quotes around each
option: "-maltivec" "-mabi=altivec".


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34437

[Bug tree-optimization/34437] several test failures for gcc.dg/vect/no-scevccp-*

[janis at gcc dot gnu dot org]

------- Comment #5 from janis at gcc dot gnu dot org  2007-12-12 17:44 -------
It's legitimate to use -maltivec without -mabi=altivec unless linking against
objects that were compiled with -mabi=altivec, so adding the option to vect.exp
is merely hiding the problem.  Is there a reason why only these particular
tests would need the AltiVec ABI, and only with -fno-tree-scev-cprop?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34437

[Bug tree-optimization/34437] several test failures for gcc.dg/vect/no-scevccp-*

[jakub at gcc dot gnu dot org]

------- Comment #6 from jakub at gcc dot gnu dot org  2007-12-12 17:51 -------
As -m32 -maltivec without -mabi=altivec has the VMX regs effectively neither
fixed, nor call saved, nor call used, it simply can't work reliably.
Registers really need to have some rules for using them, while -m32 -maltivec
means complete anarchy.  The compiler is told calls won't clobber values stored
in them, but nothing saves the registers, so things work only as long as no
calls are crossed with live VMX registers, or with pure luck that the callee
clobbers different VMX registers than the parent has live accross the call.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34437

[Bug tree-optimization/34437] several test failures for gcc.dg/vect/no-scevccp-*

[janis at gcc dot gnu dot org]

------- Comment #7 from janis at gcc dot gnu dot org  2007-12-12 23:28 -------
I had assumed that -maltivec -mabi=no-altivec was supposed to work in the
general case, then had some discussions about it on #gcc today and learned that
it doesn't and is a known problem.  Given that, we ought to add "-mabi=altivec"
to the options used in vect.exp.  I've got some other testsuite fixes to test,
I'll test that at the same time.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34437

[Bug tree-optimization/34437] several test failures for gcc.dg/vect/no-scevccp-*

[janis at gcc dot gnu dot org]

------- Comment #8 from janis at gcc dot gnu dot org  2009-07-08 19:48 -------
Fixed awhile ago by changing -mabi=altivec to the default for powerpc*-linux.


-- 

janis at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34437