Matrix multiplication not thread safe.

[xivaxy]

MRE: several copies of a matrix are multiplied by the same matrix several times independently. The result should be the same for all copies, but the function usually returns false for N>=4. Some multiplication are not performed correctly.

function testMM(T, R, N)
    m0 = rand(0:2, (N,N))
    m1 = rand(0:2, (N,N))
    m = [deepcopy(m0) for i=1:R]
    Threads.@threads for i=1:R
        for t=1:T
            m[i] *= m1
        end
    end
    return all(x->(x==m[1]),m)
end 

testMM(5,8,3)
# true

testMM(5,8,4)
# false - most likely

Threads.nthreads() is 4 for me. With a normal For-loop without threads the function always returns true.

The issue only happens with matrix multiplication, but not with addition, multiplying by a scalar or element-wise multiplication.

Julia Version 0.6.0
Commit 903644385b* (2017-06-19 13:05 UTC)
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Sandybridge)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, ivybridge)

[andreasnoack]

I'm unable to reproduce this with

Julia Version 0.6.0-rc2.0
Commit 68e911be53 (2017-05-18 02:31 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin13.4.0)
  CPU: Intel(R) Core(TM) i7-4870HQ CPU @ 2.50GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, haswell)

and

Julia Version 0.7.0-DEV.653
Commit 32aa9edad4* (2017-06-19 17:55 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin16.7.0)
  CPU: Intel(R) Core(TM) i7-4870HQ CPU @ 2.50GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT NO_AFFINITY HASWELL)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, haswell)
Environment:

[yuyichao]

And you did set the thread count?

[andreasnoack]

Oops. Typo in environment variable. Updated.

[yuyichao]

There are at least three global buffers in matmul.jl that's not thread safe.