[Bug tree-optimization/55662] New: SLP vectorization of sqrt fails if in a loop

[vincenzo.innocente at cern dot ch]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55662



             Bug #: 55662

           Summary: SLP vectorization of sqrt fails if in a loop

    Classification: Unclassified

           Product: gcc

           Version: 4.8.0

            Status: UNCONFIRMED

          Severity: normal

          Priority: P3

         Component: tree-optimization

        AssignedTo: unassig...@gcc.gnu.org

        ReportedBy: vincenzo.innoce...@cern.ch

                CC: marc.gli...@ens.fr





given the code below

computeOne vectorizes, computeS does not

c++ -std=c++11 -Ofast -mavx2 -S sqrtV.cc; less sqrtV.s

gcc version 4.8.0 20121212 (experimental) [trunk revision 194442] (GCC) 



funny enough it works with arbitrary (inlined) functions of mine (instead of

sqrtf) see computeA 



computeL does "vectorize", not in the way I expected!

I really do not understand the code generated for computeAL



cat >> sqrtV.cc

#include<cmath>

#include<type_traits>



typedef float __attribute__( ( vector_size( 16 ) ) ) float32x4_t;

typedef float __attribute__( ( vector_size( 32 ) ) ) float32x8_t;

typedef int __attribute__( ( vector_size( 32 ) ) ) int32x8_t;



float32x8_t va[1024];

float32x8_t vb[1024];

float32x8_t vc[1024];



template<typename Vec, typename F> 

inline

Vec apply(Vec v, F f) {

  typedef typename std::remove_reference<decltype(v[0])>::type T;

  constexpr int N = sizeof(Vec)/sizeof(T);

  Vec ret;

  for (int i=0;i!=N;++i) ret[i] = f(v[i]);

  return ret;

}



void computeOne() {

    vb[0]=apply(va[0],sqrtf);

}



void computeS() {

  for (int i=0;i!=1024;++i)

    vb[i]=apply(va[i],sqrtf);

}



void computeL() {

  for (int i=0;i!=1024;++i)

    for (int j=0;j!=8;++j)

    vb[i][j]=sqrtf(va[i][j]);

}



template<typename Float>

inline

Float atanF(Float t) {

  constexpr float PIO4F = 0.7853981633974483096f;

  Float z= (t > 0.4142135623730950f) ? (t-1.0f)/(t+1.0f) : t;

  // if( t > 0.4142135623730950f ) // * tan pi/8 



  Float z2 = z * z;

  Float ret =

    ((( 8.05374449538e-2f * z2

    - 1.38776856032E-1f) * z2

      + 1.99777106478E-1f) * z2

     - 3.33329491539E-1f) * z2 * z

    + z;



  // move back in place

  return ( t > 0.4142135623730950f ) ? ret : ret + PIO4F;

  return ret;

}



void computeA() {

  for (int i=0;i!=1024;++i)

    vb[i]=apply(va[i],atanF<float>);

}



void computeAL() {

  for (int i=0;i!=1024;++i)

    for (int j=0;j!=8;++j)

      vb[i][j]=atanF(va[i][j]);

}