[Valgrind-users] _mm_maskstore_ps issue

[Mark A. <mar...@gm...>]

Hi,

I was writing unit tests for some code that used the x86 compiler intrinsic
_mm_maskstore_ps (
https://software.intel.com/sites/products/documentation/doclib/iss/2013/compiler/cpp-lin/GUID-C32657F6-AAF5-4DCC-AC94-2BA02AEBADFB.htm).
I observed that Valgrind 3.10.1 thought the resulting instruction was
illegal. However, non-valgrind execution of the code works correctly (on an
avx machine, of course). Compiler gcc (Ubuntu 4.9.2-0ubuntu1~14.04) 4.9.2
(same with clang 3.5.0)

Code snippet to reproduce:

#include <immintrin.h>
#include <xmmintrin.h>
#include <iostream>

int main(int, char **)
{
    float dest[32];
    __m128 source = _mm_set1_ps(1.0);
    __m128i mask128 = _mm_set_epi32(0, -1, -1, -1);
    _mm_maskstore_ps(dest, mask128, source);

    std::cout << dest[0] << dest[1] << dest[2] << dest[3];
    // Writes "111?" to stdout
    return 0;
}

g++ source.cpp -g -mavx -lstdc++

valgrind ./a.out

==442== Memcheck, a memory error detector
==442== Copyright (C) 2002-2013, and GNU GPL'd, by Julian Seward et al.
==442== Using Valgrind-3.10.1 and LibVEX; rerun with -h for copyright info
==442== Command: ./a.out
==442==
vex amd64->IR: unhandled instruction bytes: 0xC4 0xE2 0x71 0x2E 0x0 0x8B
0x9D 0x6C
vex amd64->IR:   REX=0 REX.W=0 REX.R=0 REX.X=0 REX.B=0
vex amd64->IR:   VEX=1 VEX.L=0 VEX.nVVVV=0x1 ESC=0F38
vex amd64->IR:   PFX.66=1 PFX.F2=0 PFX.F3=0
==442== valgrind: Unrecognised instruction at address 0x400835.
==442==    at 0x400835: _mm_maskstore_ps (avxintrin.h:937)
==442==    by 0x400835: main (maskstore-bug.cpp:10)
==442== Your program just tried to execute an instruction that Valgrind
==442== did not recognise.  There are two possible reasons for this.
==442== 1. Your program has a bug and erroneously jumped to a non-code
==442==    location.  If you are running Memcheck and you just saw a
==442==    warning about a bad jump, it's probably your program's fault.
==442== 2. The instruction is legitimate but Valgrind doesn't handle it,
==442==    i.e. it's Valgrind's fault.  If you think this is the case or
==442==    you are not sure, please let us know and we'll try to fix it.
==442== Either way, Valgrind will now raise a SIGILL signal which will
==442== probably kill your program.
==442==
==442== Process terminating with default action of signal 4 (SIGILL)
==442==  Illegal opcode at address 0x400835
==442==    at 0x400835: _mm_maskstore_ps (avxintrin.h:937)
==442==    by 0x400835: main (maskstore-bug.cpp:10)
==442==
==442== HEAP SUMMARY:
==442==     in use at exit: 0 bytes in 0 blocks
==442==   total heap usage: 0 allocs, 0 frees, 0 bytes allocated
==442==
==442== All heap blocks were freed -- no leaks are possible
==442==
==442== For counts of detected and suppressed errors, rerun with: -v
==442== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
Illegal instruction (core dumped)

Further observations I made:
* use of _mm_maskload_ps was fine
* disassembly by DDT showed that a VMASKMOVPS was being generated, which is
correct
* different masks, or an aligned dest made no difference

Is this a known issue, or have I done something wrong, or?

Thanks!

Mark