[Bug c/51357] New: Simple program crash when enabling AVX

[Bug c/51357] New: Simple program crash when enabling AVX

[cdubout at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51357

             Bug #: 51357
           Summary: Simple program crash when enabling AVX
    Classification: Unclassified
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: critical
          Priority: P3
         Component: c
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: cdubout@gmail.com


Created attachment 25953
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25953
Preprocessed version

This simple program (I attached the preprocessed version):

#include <stdlib.h>
#include <stdio.h>

void conv(const float* x, const int m,
          const float* y, const int n,
          float* z)
{
    int i, j;
    float sum;

    for (i = 0; i < m - n + 1; ++i) {
        sum = 0.0f;

        for (j = 0; j < n - 3; j += 4) {
            sum += x[i + j + 0] * y[j + 0];
            sum += x[i + j + 1] * y[j + 1];
            sum += x[i + j + 2] * y[j + 2];
            sum += x[i + j + 3] * y[j + 3];
        }

        z[i] = sum;
    }
}

int main()
{
    const int m = 128000;
    const int n = 64;

    float* x = (float*)malloc(m * sizeof(float));
    float* y = (float*)malloc(n * sizeof(float));
    float* z = (float*)malloc((m - n + 1) * sizeof(float));

    conv(x, m, y, n, z);

    printf("%f\n", z[0]);

    free(x);
    free(y);
    free(z);
}

Crash if I compile it with gcc 4.6.2 (and gcc 4.6.1) optimizations turned on
(-O2) and if I enable AVX (-march=corei7-avx).
It works without optimizations (or with only -Os) or without AVX
(-march=corei7).
I use the latest Apple clang assembler (included with XCode 4.2.1).

Here is the output of gcc -v:

MacBook-Pro-2:main charles$ gcc-4.6 -v -save-temps -march=corei7-avx -O2
main.cpp
Using built-in specs.
COLLECT_GCC=gcc-4.6
COLLECT_LTO_WRAPPER=/usr/local/Cellar/gcc/4.6.2/gcc/libexec/gcc/x86_64-apple-darwin11.2.0/4.6.2/lto-wrapper
Target: x86_64-apple-darwin11.2.0
Configured with: ../configure
--enable-languages=c,c++,fortran,java,objc,obj-c++
--prefix=/usr/local/Cellar/gcc/4.6.2/gcc
--datarootdir=/usr/local/Cellar/gcc/4.6.2/share
--bindir=/usr/local/Cellar/gcc/4.6.2/bin --program-suffix=-4.6
--with-gmp=/usr/local/Cellar/gmp/5.0.2 --with-mpfr=/usr/local/Cellar/mpfr/3.1.0
--with-mpc=/usr/local/Cellar/libmpc/0.9 --with-system-zlib
--enable-stage1-checking --enable-plugin --enable-lto --disable-nls
--disable-fully-dynamic-string
Thread model: posix
gcc version 4.6.2 (GCC) 
COLLECT_GCC_OPTIONS='-mmacosx-version-min=10.7.2' '-v' '-save-temps'
'-march=corei7-avx' '-O2'

/usr/local/Cellar/gcc/4.6.2/gcc/libexec/gcc/x86_64-apple-darwin11.2.0/4.6.2/cc1plus
-E -quiet -v -D__DYNAMIC__ main.cpp -fPIC -mmacosx-version-min=10.7.2
-march=corei7-avx -O2 -fpch-preprocess -o main.ii
ignoring nonexistent directory
"/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2/../../../../x86_64-apple-darwin11.2.0/include"
#include "..." search starts here:
#include <...> search starts here:

/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2/../../../../include/c++/4.6.2

/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2/../../../../include/c++/4.6.2/x86_64-apple-darwin11.2.0

/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2/../../../../include/c++/4.6.2/backward

/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2/include
 /usr/local/include
 /usr/local/Cellar/gcc/4.6.2/gcc/include

/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2/include-fixed
 /usr/include
 /System/Library/Frameworks
 /Library/Frameworks
End of search list.
COLLECT_GCC_OPTIONS='-mmacosx-version-min=10.7.2' '-v' '-save-temps'
'-march=corei7-avx' '-O2'

/usr/local/Cellar/gcc/4.6.2/gcc/libexec/gcc/x86_64-apple-darwin11.2.0/4.6.2/cc1plus
-fpreprocessed main.ii -fPIC -quiet -dumpbase main.cpp
-mmacosx-version-min=10.7.2 -march=corei7-avx -auxbase main -O2 -version -o
main.s
GNU C++ (GCC) version 4.6.2 (x86_64-apple-darwin11.2.0)
    compiled by GNU C version 4.6.2, GMP version 5.0.2, MPFR version 3.1.0, MPC
version 0.9
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
GNU C++ (GCC) version 4.6.2 (x86_64-apple-darwin11.2.0)
    compiled by GNU C version 4.6.2, GMP version 5.0.2, MPFR version 3.1.0, MPC
version 0.9
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
Compiler executable checksum: b1d97ab7cd7bb3a1ba06b8ac1d280faa
COLLECT_GCC_OPTIONS='-mmacosx-version-min=10.7.2' '-v' '-save-temps'
'-march=corei7-avx' '-O2'
 as -arch x86_64 -force_cpusubtype_ALL -o main.o main.s
COMPILER_PATH=/usr/local/Cellar/gcc/4.6.2/gcc/libexec/gcc/x86_64-apple-darwin11.2.0/4.6.2/:/usr/local/Cellar/gcc/4.6.2/gcc/libexec/gcc/x86_64-apple-darwin11.2.0/4.6.2/:/usr/local/Cellar/gcc/4.6.2/gcc/libexec/gcc/x86_64-apple-darwin11.2.0/:/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2/:/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/
LIBRARY_PATH=/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2/:/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2/../../../:/usr/lib/
COLLECT_GCC_OPTIONS='-mmacosx-version-min=10.7.2' '-v' '-save-temps'
'-march=corei7-avx' '-O2'

/usr/local/Cellar/gcc/4.6.2/gcc/libexec/gcc/x86_64-apple-darwin11.2.0/4.6.2/collect2
-dynamic -arch x86_64 -macosx_version_min 10.7.2 -weak_reference_mismatches
non-weak -o a.out -lcrt1.10.5.o
-L/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2
-L/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2/../../..
main.o -no_compact_unwind -lSystem -lgcc_ext.10.5 -lgcc -lSystem -v
collect2 version 4.6.2 (x86_64 Darwin)
/usr/bin/ld -dynamic -arch x86_64 -macosx_version_min 10.7.2
-weak_reference_mismatches non-weak -o a.out -lcrt1.10.5.o
-L/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2
-L/usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2/../../..
main.o -no_compact_unwind -lSystem -lgcc_ext.10.5 -lgcc -lSystem -v
@(#)PROGRAM:ld  PROJECT:ld64-127.2
Library search paths:
    /usr/local/Cellar/gcc/4.6.2/gcc/lib/gcc/x86_64-apple-darwin11.2.0/4.6.2
    /usr/local/Cellar/gcc/4.6.2/gcc/lib
    /usr/lib
    /usr/local/lib
Framework search paths:
    /Library/Frameworks/
    /System/Library/Frameworks/

It works if I do not unroll the inner loop, I guess because it is not using AVX
in that case.
Versions of the program using SSE or AVX intrinsics crash in the same
conditions.

[Bug c/51357] Simple program crash when enabling AVX

[cdubout at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51357

--- Comment #1 from Charles <cdubout at gmail dot com> 2011-11-30 15:23:54 UTC ---
Created attachment 25955
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25955
Generated assembly

[Bug c/51357] Simple program crash when enabling AVX

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51357

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-11-30 15:59:50 UTC ---
Doesn't crash for me on x86_64-linux. The testcase is buggy, because the x and
y heap arrays are uninitialized, could contain any values including signalling
NaNs etc.  Nothing is vectorized here (you are just compiling with -O2, not -O3
-ffast-math that would be needed for vectorization in this case).  I see
nothing wrong on the GCC side here, so either your OS doesn't support AVX, or
the assembler is buggy, ...

[Bug c/51357] Simple program crash when enabling AVX

[cdubout at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51357

Charles <cdubout at gmail dot com> changed:

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

--- Comment #3 from Charles <cdubout at gmail dot com> 2011-11-30 17:37:01 UTC ---
AVX indeed works on my machine, and initializing x, y, z does not change
anything.
I suppose the problem come from my assembler then.
Sorry to have wasted your time.