Loop Vectorization 01

[Royi]

Hello,

I have a Vectorization optimization problem.
I have a struct pDst which have 3 fields named: 'red', 'green' and 'blue'.
The type might be 'Char', 'Short' or 'Float'.This is given and can not be altered.
We have another array pSrc which represents an image [RGB] - Namely an array of 3 pointers which every one of them point to a layer of an image.Each layer is built using IPP plane oriented image (Namely, Each plane is formed independently - 'ippiMalloc_32f_C1'):http://software.intel.com/sites/products/documentation/hpc/ipp/ippi/ippi_ch3/functn_Malloc.html

We would like to copy it as described in the following code:

for(int y = 0; y < imageHeight; ++y)
{
    for(int x = 0; x < imageWidth; ++x)
    {
        pDst[x + y * pDstRowStep].red     = pSrc[0][x + y * pSrcRowStep];
        pDst[x + y * pDstRowStep].green = pSrc[1][x + y * pSrcRowStep];
        pDst[x + y * pDstRowStep].blue    = pSrc[2][x + y * pSrcRowStep];
    }
} 

Yet, in this form the compiler can't vectorize the code.
At first it says: "loop was not vectorized: existence of vector dependence.".
When I use the  #pragma ivdep to help the compiler (Since there's no dependence) I get the following error:"loop was not vectorized: dereference too complex.".

Anyone has an idea how to allow vectorization?

I have Intel Compiler 13.0.

Thanks.

[JenniferJ]

Check out the pragma simd feature. The online doc is here: http://software.intel.com/sites/products/documentation/studio/composer/en-us/2011Update/compiler_c/cref_cls/common/cppref_pragma_simd.htm Please read this article first - http://software.intel.com/en-us/articles/requirements-for-vectorizing-loops-with-pragma-simd/ Jennifer

[Royi]

Hi Jennifer, You referred me to an article where it says how to enforce vectorization. Where I want to understand why isn't it automatically vectorized. Maybe there is a different form I should arrange the data to yield the needed vectorization. Thank You.

[Mark_S_Intel1]

The compiler does not vectorize the code because it believes vectorizing it would be inefficient mostly due to use of Array of Structure access (AOS) which requires generations of gather/scatter instructions that are slow relative to use of linear access instructions: struct X { float red, green, blue; }; struct X *restrict pDst; float *restrict pSrc[3]; void foo(int imageHeight, int imageWidth, int pDstRowStep, int pSrcRowStep){ int x, y; for (y = 0; y < imageHeight; y++){ for (x = 0; x < imageWidth; x++){ pDst[x + y * pDstRowStep].red = pSrc[0][x + y * pSrcRowStep]; pDst[x + y * pDstRowStep].green = pSrc[1][x + y * pSrcRowStep]; pDst[x + y * pDstRowStep].blue = pSrc[2][x + y * pSrcRowStep]; } } } $ icc -vec-report2 -c -restrict vec10.cpp -V Intel(R) C Intel(R) 64 Compiler XE for applications running on Intel(R) 64, Version 13.0.0.079 Build 20120731 Copyright (C) 1985-2012 Intel Corporation. All rights reserved. vec10.cpp(11): (col. 4) remark: loop was not vectorized: vectorization possible but seems inefficient. vec10.cpp(10): (col. 3) remark: loop was not vectorized: not inner loop. Section 5.3 about SoA vs AoS at http://software.intel.com/en-us/articles/a-guide-to-auto-vectorization-with-intel-c-compilers/ might give some helpful information about this. You can bypass the compiler's cost benefit analysis and have it vectorize the loop by using the -vec-threshold0 option, but the code may run slow for default vectorization target which is SSE2: $ icc -vec-report2 -c -restrict vec10.cpp -V -vec-threshold0 Intel(R) C Intel(R) 64 Compiler XE for applications running on Intel(R) 64, Version 13.0.0.079 Build 20120731 vec10.cpp(11): (col. 4) remark: LOOP WAS VECTORIZED. vec10.cpp(10): (col. 3) remark: loop was not vectorized: not inner loop. The compiler can vectorize the code for AVX without the use of -vec-threshold0 option but I am not sure if it will give much speed up compare to the non-vectorized version: $ icc -vec-report2 -c -restrict vec10.cpp -V -xAVX Intel(R) C Intel(R) 64 Compiler XE for applications running on Intel(R) 64, Version 13.0.0.079 Build 20120731 vec10.cpp(11): (col. 4) remark: LOOP WAS VECTORIZED. vec10.cpp(10): (col. 3) remark: loop was not vectorized: not inner loop.