Re: [Discuss-gnuradio] Segfault with volk on 32 bit AMD

[Frederick Stevens]

Well, after a few restarts, here is my output.  I did a fresh pull from git because I was getting some errors with missing *.h files in gruel/src/swig or something like that.  Hope this helps!

RUN_VOLK_TESTS: volk_32fc_32f_multiply_32fc_a

Program received signal SIGSEGV, Segmentation fault.
0xb7edbb74 in volk_32fc_32f_multiply_32fc_a_generic (cVector=0xb7448008,
    aVector=0xb7768008, bVector=0xb78f8008, num_points=204600)
    at /home/fred/extras/gnuradio/gnuradio/volk/include/volk/volk_32fc_32f_multiply_32fc_a.h:74
74          *cPtr++ = (*aPtr++) * (*bPtr++);
(gdb) bt
#0  0xb7edbb74 in volk_32fc_32f_multiply_32fc_a_generic (cVector=0xb7448008,
    aVector=0xb7768008, bVector=0xb78f8008, num_points=204600)
    at /home/fred/extras/gnuradio/gnuradio/volk/include/volk/volk_32fc_32f_multiply_32fc_a.h:74
#1  0xb7ed4d68 in volk_32fc_32f_multiply_32fc_a_manual (cVector=0xb7448008,
    aVector=0xb7768008, bVector=0xb78f8008, num_points=204600,
    arch=0x8079ac4 "generic")
    at /home/fred/extras/gnuradio/gnuradio/build/volk/lib/volk.c:749
#2  0x08064533 in run_cast_test3 (
    func=0x80595c0 <address@hidden>, buffs=...,
    vlen=204600, iter=999, arch=...)
    at /home/fred/extras/gnuradio/gnuradio/volk/lib/qa_utils.cc:182
#3  0x08062770 in run_volk_tests (desc=...,
    manual_func=0x80595c0 <address@hidden>,
    name=..., tol=9.99999975e-05, scalar=..., vlen=204600, iter=1000,
    best_arch_vector=0xbfffe714)
    at /home/fred/extras/gnuradio/gnuradio/volk/lib/qa_utils.cc:351
#4  0x0805b3d3 in main (argc=1, argv=0xbffff204)
    at /home/fred/extras/gnuradio/gnuradio/volk/apps/volk_profile.cc:38
(gdb)


Dump of assembler code for function volk_32fc_32f_multiply_32fc_a_generic:
   0xb7edbb39 <+0>:     push   %ebp
   0xb7edbb3a <+1>:     mov    %esp,%ebp
   0xb7edbb3c <+3>:     sub    $0x14,%esp
   0xb7edbb3f <+6>:     mov    0x8(%ebp),%eax
   0xb7edbb42 <+9>:     mov    %eax,-0x4(%ebp)
   0xb7edbb45 <+12>:    mov    0xc(%ebp),%eax
   0xb7edbb48 <+15>:    mov    %eax,-0x8(%ebp)
   0xb7edbb4b <+18>:    mov    0x10(%ebp),%eax
   0xb7edbb4e <+21>:    mov    %eax,-0xc(%ebp)
   0xb7edbb51 <+24>:    movl   $0x0,-0x10(%ebp)
   0xb7edbb58 <+31>:    movl   $0x0,-0x10(%ebp)
   0xb7edbb5f <+38>:    jmp    0xb7edbbae <volk_32fc_32f_multiply_32fc_a_generic+117>
   0xb7edbb61 <+40>:    mov    -0x8(%ebp),%eax
   0xb7edbb64 <+43>:    mov    (%eax),%ecx
   0xb7edbb66 <+45>:    mov    0x4(%eax),%edx
   0xb7edbb69 <+48>:    mov    %ecx,%eax
   0xb7edbb6b <+50>:    mov    %eax,-0x14(%ebp)
   0xb7edbb6e <+53>:    flds   -0x14(%ebp)
   0xb7edbb71 <+56>:    mov    -0xc(%ebp),%eax
=> 0xb7edbb74 <+59>:    flds   (%eax)
   0xb7edbb76 <+61>:    fmulp  %st,%st(1)
   0xb7edbb78 <+63>:    mov    %edx,-0x14(%ebp)
   0xb7edbb7b <+66>:    flds   -0x14(%ebp)
   0xb7edbb7e <+69>:    mov    -0xc(%ebp),%eax
   0xb7edbb81 <+72>:    flds   (%eax)
   0xb7edbb83 <+74>:    fmulp  %st,%st(1)
   0xb7edbb85 <+76>:    fxch   %st(1)
   0xb7edbb87 <+78>:    fstps  -0x14(%ebp)
   0xb7edbb8a <+81>:    mov    -0x14(%ebp),%ecx
   0xb7edbb8d <+84>:    fstps  -0x14(%ebp)
   0xb7edbb90 <+87>:    mov    -0x14(%ebp),%edx
   0xb7edbb93 <+90>:    mov    -0x4(%ebp),%eax
   0xb7edbb96 <+93>:    mov    %ecx,(%eax)
   0xb7edbb98 <+95>:    mov    %edx,0x4(%eax)
   0xb7edbb9b <+98>:    addl   $0x8,-0x4(%ebp)
   0xb7edbb9f <+102>:   addl   $0x8,-0x8(%ebp)
   0xb7edbba3 <+106>:   addl   $0x4,-0xc(%ebp)
   0xb7edbba7 <+110>:   addl   $0x4,-0xc(%ebp)
   0xb7edbbab <+114>:   incl   -0x10(%ebp)
   0xb7edbbae <+117>:   mov    -0x10(%ebp),%eax  
   0xb7edbb78 <+63>:    mov    %edx,-0x14(%ebp)
   0xb7edbb7b <+66>:    flds   -0x14(%ebp)
   0xb7edbb7e <+69>:    mov    -0xc(%ebp),%eax
   0xb7edbb81 <+72>:    flds   (%eax)
   0xb7edbb83 <+74>:    fmulp  %st,%st(1)
   0xb7edbb85 <+76>:    fxch   %st(1)
   0xb7edbb87 <+78>:    fstps  -0x14(%ebp)
   0xb7edbb8a <+81>:    mov    -0x14(%ebp),%ecx
   0xb7edbb8d <+84>:    fstps  -0x14(%ebp)
   0xb7edbb90 <+87>:    mov    -0x14(%ebp),%edx
   0xb7edbb93 <+90>:    mov    -0x4(%ebp),%eax
   0xb7edbb96 <+93>:    mov    %ecx,(%eax)
   0xb7edbb98 <+95>:    mov    %edx,0x4(%eax)
   0xb7edbb9b <+98>:    addl   $0x8,-0x4(%ebp)
   0xb7edbb9f <+102>:   addl   $0x8,-0x8(%ebp)
   0xb7edbba3 <+106>:   addl   $0x4,-0xc(%ebp)
   0xb7edbba7 <+110>:   addl   $0x4,-0xc(%ebp)
   0xb7edbbab <+114>:   incl   -0x10(%ebp)
   0xb7edbbae <+117>:   mov    -0x10(%ebp),%eax
   0xb7edbbb1 <+120>:   cmp    0x14(%ebp),%eax
   0xb7edbbb4 <+123>:   jb     0xb7edbb61 <volk_32fc_32f_multiply_32fc_a_generic+40>
   0xb7edbbb6 <+125>:   leave 
   0xb7edbbb7 <+126>:   ret   
End of assembler dump.

Cheers,

Fred
On 03/16/2012 02:15 PM, Tom Rondeau wrote:

On Fri, Mar 16, 2012 at 3:11 PM, Frederick Stevens <address@hidden> wrote:

I've added your suggested changes.  It will be an hour or so until compilation is finished. Slackware CFLAGS or CXXFLAGS are usually -O2 by default.

Cheers,

Fred

Thanks. GNU Radio resets these flag to -O3 right now, though.

Also, can't you run 'make -jN' to speed up the compilation?

Tom

 

On 03/16/2012 02:09 PM, Tom Rondeau wrote:

On Fri, Mar 16, 2012 at 3:03 PM, Frederick Stevens <address@hidden> wrote:

Here is the gcc info.  It will take me a bit to re compile gnuradio but I will turn on debugging in cmake.

gcc (GCC) 4.5.2
Copyright (C) 2010 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

I am running kernel 3.2.7 from slackware-current and swig 2.0.4 everything else is from slackbuilds.  I will have to get versions for you if you need them.

Cheers,

Fred

While you're doing the rebuild, can you set the optimization flag to -O2? It's -O3 right now by default, and every now and then it can be a problem (haven't heard of one in a while, but it's been a thing).

It should be setting '-DCMAKE_CXXFLAGS="-O2"'.

Tom

 

On 03/16/2012 01:50 PM, Nick Foster wrote:

On Fri, Mar 16, 2012 at 11:42 AM, Frederick Stevens <address@hidden> wrote:

Rafael,

Here is the output from gdb:

RUN_VOLK_TESTS: volk_32fc_32f_multiply_32fc_a

Program received signal SIGSEGV, Segmentation fault.
0xb7f2c6e0 in volk_32fc_32f_multiply_32fc_a_generic ()
  from /home/fred/extras/gnuradio/gnuradio/build/volk/lib/libvolk.so.0.0.0



Dump of assembler code for function volk_32fc_32f_multiply_32fc_a_generic:
  0xb7f2c6c0 <+0>:     push   %ebp
  0xb7f2c6c1 <+1>:     mov    %esp,%ebp
  0xb7f2c6c3 <+3>:     push   %edi
  0xb7f2c6c4 <+4>:     push   %esi
  0xb7f2c6c5 <+5>:     mov    0x8(%ebp),%edx
  0xb7f2c6c8 <+8>:     mov    0xc(%ebp),%ecx
  0xb7f2c6cb <+11>:    mov    0x10(%ebp),%edi
  0xb7f2c6ce <+14>:    mov    0x14(%ebp),%esi
  0xb7f2c6d1 <+17>:    test   %esi,%esi
  0xb7f2c6d3 <+19>:    je     0xb7f2c6fc <volk_32fc_32f_multiply_32fc_a_generic+60>
  0xb7f2c6d5 <+21>:    xor    %eax,%eax
  0xb7f2c6d7 <+23>:    mov    %esi,%esi
  0xb7f2c6d9 <+25>:    lea    0x0(%edi,%eiz,1),%edi
=> 0xb7f2c6e0 <+32>:    flds   (%edi,%eax,8)
  0xb7f2c6e3 <+35>:    flds   (%ecx,%eax,8)
  0xb7f2c6e6 <+38>:    fmul   %st(1),%st
  0xb7f2c6e8 <+40>:    fxch   %st(1)
  0xb7f2c6ea <+42>:    fmuls  0x4(%ecx,%eax,8)
  0xb7f2c6ee <+46>:    fxch   %st(1)
  0xb7f2c6f0 <+48>:    fstps  (%edx,%eax,8)
  0xb7f2c6f3 <+51>:    fstps  0x4(%edx,%eax,8)
  0xb7f2c6f7 <+55>:    inc    %eax
  0xb7f2c6f8 <+56>:    cmp    %eax,%esi
  0xb7f2c6fa <+58>:    ja     0xb7f2c6e0 <volk_32fc_32f_multiply_32fc_a_generic+32>
  0xb7f2c6fc <+60>:    pop    %esi
  0xb7f2c6fd <+61>:    pop    %edi
  0xb7f2c6fe <+62>:    pop    %ebp
  0xb7f2c6ff <+63>:    ret
End of assembler dump.

I omitted the first part of the program execution since everything seemed to be working fine.  Hope this helps.  Let me know if you would like me to try something else.

Cheers,


Fred

OK, that's weird as hell. That's the generic implementation, which is just a std::complex multiply in a for loop. Can you give me your gcc version?

--n

 

On 03/16/2012 12:39 PM, Rafael Diniz wrote:

Also using the gdb "disassemble" command you could trace which instruction
raised the fault.

Cheers,
Rafael Diniz

Well, that doesn't make me happy at all. Especially since we just patched
the release yesterday...

Sounds like the AMD chip is handling something wrong (or differently) with
the alignment. All vectors passed using volk_profile are supposed to be
properly byte aligned.

Fred, can you just check for me what happens when you run it under gdb?
Just run "gdb volk_profile" and when it crashes, do a "bt" and post the
results of the backtrace here.

Thanks,
Tom

_______________________________________________
Discuss-gnuradio mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio

_______________________________________________
Discuss-gnuradio mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio

_______________________________________________
Discuss-gnuradio mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio

_______________________________________________
Discuss-gnuradio mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio