[LLVMdev] loop vectorizer misses opportunity, exploit

[Frank Winter]

A quite small but yet complete example function which all vectorization 
passes fail to optimize:

#include <cstdint>
#include <iostream>

void bar(std::uint64_t start, std::uint64_t end, float * __restrict__  
c, float * __restrict__ a, float * __restrict__ b)
{
   for ( std::uint64_t i = start ; i < end ; i += 4 ) {
     {
       const std::uint64_t ir0 = (i+0)%4 + 8*((i+0)/4);
       c[ ir0 ]         = a[ ir0 ]         + b[ ir0 ];
     }
     {
       const std::uint64_t ir0 = (i+1)%4 + 8*((i+1)/4);
       c[ ir0 ]         = a[ ir0 ]         + b[ ir0 ];
     }
     {
       const std::uint64_t ir0 = (i+2)%4 + 8*((i+2)/4);
       c[ ir0 ]         = a[ ir0 ]         + b[ ir0 ];
     }
     {
       const std::uint64_t ir0 = (i+3)%4 + 8*((i+3)/4);
       c[ ir0 ]         = a[ ir0 ]         + b[ ir0 ];
     }
   }
}

The loop index and array accesses for the first 4 iterations:

i 0:     0 1 2 3
i 4:     8 9 10 11
i 8:     16 17 18 19
i 12:     24 25 26 27

For example on an x86 processor with SSE (128 bit SIMD vectors) the loop body could be vectorized into 2 SIMD reads, 1 SIMD add and 1 SIMD store.

With current trunk I tried the following on the above example:

clang++ -emit-llvm -S loop_minimal.cc -std=c++11
opt -O3 -vectorize-slp -S loop_minimal.ll
opt -O3 -loop-vectorize -S loop_minimal.ll
opt -O3 -bb-vectorize -S loop_minimal.ll

All optimization passes miss the opportunity. It seems the SCEV AA pass doesn't understand modulo arithmetic.

How can the SCEV AA pass be extended to handle this type of arithmetic?

Frank

On 31/10/13 02:21, Renato Golin wrote:

> On 30 October 2013 18:40, Frank Winter <fwinter at jlab.org 
> <mailto:fwinter at jlab.org>> wrote:
>
>           const std::uint64_t ir0 = (i+0)%4;  // not working
>
>
> I thought this would be the case when I saw the original expression. 
> Maybe we need to teach module arithmetic to SCEV?
>
> --renato


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20131031/602f127b/attachment.html>