[Bug other/59691] New: cilk-plus failure on PENTIUM2

[Bug other/59691] New: cilk-plus failure on PENTIUM2

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

            Bug ID: 59691
           Summary: cilk-plus failure on PENTIUM2
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: other
          Assignee: unassigned at gcc dot gnu.org
          Reporter: bernd.edlinger at hotmail dot de

Almost all cilk-plus execution tests fail due to invalid instruction
on PENTIUM2.

Core was generated by `./fib.exe'.
Program terminated with signal 4, Illegal instruction.
#0  0x0036237a in sysdep_save_fp_ctrl_state () from
/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libcilkrts/.libs/libcilkrts.so.5
(gdb) disass
Dump of assembler code for function sysdep_save_fp_ctrl_state:
   0x00362370 <+0>:     mov    0x4(%esp),%eax
   0x00362374 <+4>:     cmpb   $0x0,0x3(%eax)
   0x00362378 <+8>:     je     0x362381 <sysdep_save_fp_ctrl_state+17>
=> 0x0036237a <+10>:    stmxcsr 0x28(%eax)
   0x0036237e <+14>:    fnstsw 0x2c(%eax)
   0x00362381 <+17>:    repz ret
End of assembler dump.
(gdb) bt
#0  0x0036237a in sysdep_save_fp_ctrl_state () from
/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libcilkrts/.libs/libcilkrts.so.5
#1  0x00362c5b in __cilkrts_save_fp_ctrl_state () from
/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libcilkrts/.libs/libcilkrts.so.5
#2  0x08048818 in fib ()
#3  0x0804868b in main ()

cat /proc/cpuinfo
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 6
model           : 5
model name      : Pentium II (Deschutes)
stepping        : 2
microcode       : 0x2a
cpu MHz         : 448.819
cache size      : 512 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 2
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov
pse36 mmx fxsr
bogomips        : 897.63
clflush size    : 32
cache_alignment : 32
address sizes   : 36 bits physical, 32 bits virtual
processor       : 1
vendor_id       : GenuineIntel
cpu family      : 6
model           : 5
model name      : Pentium II (Deschutes)
stepping        : 2
microcode       : 0x2a
cpu MHz         : 448.819
cache size      : 512 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 2
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov
pse36 mmx fxsr
bogomips        : 897.67
clflush size    : 32
cache_alignment : 32
address sizes   : 36 bits physical, 32 bits virtual

[Bug target/59691] cilk-plus run failures on non-sse processors

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |i?86-*-*
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-01-06
          Component|other                       |target
   Target Milestone|---                         |4.9.0
            Summary|cilk-plus failure on        |cilk-plus run failures on
                   |PENTIUM2                    |non-sse processors
     Ever confirmed|0                           |1

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
This is an SSE only instruction.  The code unconditionally uses this
instruction.

Confirmed.

        __asm__ ("stmxcsr %0" : "=m" (sf->mxcsr));

[Bug target/59691] cilk-plus run failures on non-sse processors

[bviyer at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

--- Comment #2 from Balaji V. Iyer <bviyer at gmail dot com> ---
Hello Bernd,
    I don't have a pentium 2 machine readily available. Can you try this patch
and see if it works for you?

Thanks,

Balaji V. Iyer.

diff --git a/libcilkrts/runtime/config/x86/os-unix-sysdep.c
b/libcilkrts/runtime
/config/x86/os-unix-sysdep.c
index 881bc3f..8d5642c 100644
--- a/libcilkrts/runtime/config/x86/os-unix-sysdep.c
+++ b/libcilkrts/runtime/config/x86/os-unix-sysdep.c
@@ -47,7 +47,7 @@
 // save the FP state (rounding mode and the like) before calling setjmp.  We
 // will need to restore that state when we resume.
 #ifndef __MIC__
-# if defined(__i386__) || defined(__x86_64)
+# if (defined(__i386__) || defined(__x86_64)) && defined (__SSE__)
 #   define RESTORE_X86_FP_STATE
 # endif // defined(__i386__) || defined(__x86_64)
 #endif  // __MIC__

[Bug target/59691] cilk-plus run failures on non-sse processors

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Balaji V. Iyer from comment #2)
> Hello Bernd,
>     I don't have a pentium 2 machine readily available. Can you try this
> patch and see if it works for you?

This is not correct, there should be a runtime check rather than a compile time
one.

> 
> Thanks,
> 
> Balaji V. Iyer.
> 
> diff --git a/libcilkrts/runtime/config/x86/os-unix-sysdep.c
> b/libcilkrts/runtime
> /config/x86/os-unix-sysdep.c
> index 881bc3f..8d5642c 100644
> --- a/libcilkrts/runtime/config/x86/os-unix-sysdep.c
> +++ b/libcilkrts/runtime/config/x86/os-unix-sysdep.c
> @@ -47,7 +47,7 @@
>  // save the FP state (rounding mode and the like) before calling setjmp.  We
>  // will need to restore that state when we resume.
>  #ifndef __MIC__
> -# if defined(__i386__) || defined(__x86_64)
> +# if (defined(__i386__) || defined(__x86_64)) && defined (__SSE__)
>  #   define RESTORE_X86_FP_STATE
>  # endif // defined(__i386__) || defined(__x86_64)
>  #endif  // __MIC__

[Bug target/59691] cilk-plus run failures on non-sse processors

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

--- Comment #4 from Jeffrey A. Law <law at redhat dot com> ---
On 01/22/14 10:45, pinskia at gcc dot gnu.org wrote:
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691
>
> --- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
> (In reply to Balaji V. Iyer from comment #2)
>> Hello Bernd,
>>      I don't have a pentium 2 machine readily available. Can you try this
>> patch and see if it works for you?
>
> This is not correct, there should be a runtime check rather than a compile time
> one.
Also note that you can usually change the underlying processor in VM 
environments.  That's obviously useful for testing this kind of thing. 
VMs are, umm, goodness.

jeff

[Bug target/59691] cilk-plus run failures on non-sse processors

[bviyer at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

--- Comment #5 from Balaji V. Iyer <bviyer at gmail dot com> ---
Created attachment 31938
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=31938&action=edit
Possible Fix.

Possible fix.

[Bug target/59691] cilk-plus run failures on non-sse processors

[bviyer at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

--- Comment #6 from Balaji V. Iyer <bviyer at gmail dot com> ---
Hi Jeff, Andrew and Bernd,
    Can you please try out the "possible fix" and see if that works for you?

Thanks,

Balaji V. Iyer.

[Bug target/59691] cilk-plus run failures on non-sse processors

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

--- Comment #7 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
(In reply to Balaji V. Iyer from comment #6)
> Hi Jeff, Andrew and Bernd,
>    Can you please try out the "possible fix" and
> see if that works for you?
>
> Thanks,
> 
> Balaji V. Iyer.

Hi the patch works,

but there is still one test case failing:

spawn /home/ed/gnu/gcc-build/gcc/testsuite/g++/../../xg++
-B/home/ed/gnu/gcc-build/gcc/testsuite/g++/../../
/home/ed/gnu/gcc-4.9-trunk/gcc/testsuite/g++.dg/cilk-plus/CK/catch_exc.cc
-fno-diagnostics-show-caret -fdiagnostics-color=never -nostdinc++
-I/home/ed/gnu/gcc-build/i686-pc-linux-gnu/libstdc++-v3/include/i686-pc-linux-gnu
-I/home/ed/gnu/gcc-build/i686-pc-linux-gnu/libstdc++-v3/include
-I/home/ed/gnu/gcc-4.9-trunk/libstdc++-v3/libsupc++
-I/home/ed/gnu/gcc-4.9-trunk/libstdc++-v3/include/backward
-I/home/ed/gnu/gcc-4.9-trunk/libstdc++-v3/testsuite/util -fmessage-length=0 -O1
-fcilkplus -fcilkplus -lcilkrts
-L/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libstdc++-v3/src/.libs
-B/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libstdc++-v3/src/.libs
-L/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libstdc++-v3/src/.libs
-L/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libcilkrts/.libs -lm -o
./catch_exc.exe^M
PASS: g++.dg/cilk-plus/CK/catch_exc.cc  -O1 -fcilkplus (test for excess errors)
Setting LD_LIBRARY_PATH to
.:/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libstdc++-v3/src/.libs:/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libstdc++-v3/src/.libs:/home/ed/gnu/gcc-build/gcc:/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libcilkrts/.libs:.:/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libstdc++-v3/src/.libs:/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libstdc++-v3/src/.libs:/home/ed/gnu/gcc-build/gcc:/home/ed/gnu/gcc-build/i686-pc-linux-gnu/./libcilkrts/.libs:/home/ed/gnu/gcc-build/./gmp/.libs:/home/ed/gnu/gcc-build/./prev-gmp/.libs:/home/ed/gnu/gcc-build/./mpfr/.libs:/home/ed/gnu/gcc-build/./prev-mpfr/.libs:/home/ed/gnu/gcc-build/./mpc/.libs:/home/ed/gnu/gcc-build/./prev-mpc/.libs
spawn [open ...]^M
terminate called after throwing an instance of 'int'
FAIL: g++.dg/cilk-plus/CK/catch_exc.cc  -O1 -fcilkplus execution test

[Bug target/59691] cilk-plus run failures on non-sse processors

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

--- Comment #8 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Balaji V. Iyer from comment #5)
> Created attachment 31938 [details]
> Possible Fix.
> 
> Possible fix.

You can replace cpu_supports_sse with __builtin_cpu_supports
since __builtin_cpu_supports("sse") is compiled into:

movl    __cpu_model+12(%rip), %eax

by GCC.

[Bug target/59691] cilk-plus run failures on non-sse processors

[bviyer at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

--- Comment #9 from Balaji V. Iyer <bviyer at gmail dot com> ---
Hi Bernd and H. J.,
    The catch_exc.cc bug is a different issue. It is failing in most 32 bit
machines and we are currently looking into it.

Thanks,

Balaji V. Iyer.

[Bug target/59691] cilk-plus run failures on non-sse processors

[bviyer at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

--- Comment #10 from bviyer at gcc dot gnu.org ---
Author: bviyer
Date: Mon Feb 10 16:56:54 2014
New Revision: 207664

URL: http://gcc.gnu.org/viewcvs?rev=207664&root=gcc&view=rev
Log:
Fix for PR target/59691.
+2014-02-10  Balaji V. Iyer  <balaji.v.iyer@intel.com>
+
+       PR target/59691
+       * runtime/config/x86/os-unix-sysdep.c (__builtin_cpu_supports): New
+       function.
+       (restore_x86_fp_state): Added a check if the cpu supports the
+       instruction before emitting it.
+       (sysdep_save_fp_ctrl_state): Likewise.
+


Modified:
    trunk/libcilkrts/ChangeLog
    trunk/libcilkrts/runtime/config/x86/os-unix-sysdep.c

[Bug target/59691] cilk-plus run failures on non-sse processors

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59691

Jeffrey A. Law <law at redhat dot com> changed:

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

--- Comment #11 from Jeffrey A. Law <law at redhat dot com> ---
Fixed on trunk.