I am looking for a way to simplify functions accelerated by math kernel libraries ...is there any way to make flair signature out of them to understand it better. For instance, I want to get the original data structure or math function for this code (e.g sin or cos or array in math kernel)

For instance:

        v18 = (__m128i)_mm_cmpneqps(
                           _mm_unpacklo_ps((__m128)*(_DWORD *)(v17 + 160), (__m128)*(_DWORD *)(v17 + 240)),
                           _mm_unpacklo_ps((__m128)*(_DWORD *)(v17 + 200), (__m128)*(_DWORD *)(v17 + 280))),
        v16 += 8LL;
        v19 = _mm_sub_epi32(
                             _mm_unpacklo_ps((__m128)*(_DWORD *)v17, (__m128)*(_DWORD *)(v17 + 80)),
                             _mm_unpacklo_ps((__m128)*(_DWORD *)(v17 + 40), (__m128)*(_DWORD *)(v17 + 120))),
        v17 += 320LL;
        v15 = _mm_add_epi32(_mm_add_epi32(v15, v19), _mm_sub_epi32(0LL, v18));
      while ( v16 < v14 );

This was created by libmmd.dll 11.1

  • What do you mean by simplify !! Be more specific when asking questions, otherwise it will either be closed or voted down ;)
    – yaspr
    Commented Jul 30, 2015 at 15:38
  • I mean a very similar structure to what was originally written before compiling using MKL. For instance, you may use sin or cos or eigen function in math kernel and it gets translated to these types of SSE dependent instructions which are not human readable (at least for me ) Commented Jul 30, 2015 at 16:10
  • find definition of v17, place cursor on it, press y, enter _DWORD* .This is the only simplification you can do in this specific code.
    – w s
    Commented Aug 2, 2015 at 10:11


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.