David Doria
David Doria

Reputation: 10273

SSE addition producing garbage

I am trying to compare SSE float[4] addition to standard float[4] addition. I tried this:

#include <iostream>
#include <vector>

struct Point4
{
  Point4()
  {
    data[0] = 0;
    data[1] = 0;
    data[2] = 0;
    data[3] = 0;
  }

  float data[4];
};

static float SumOfDifferences(const Point4& a, const Point4& b)
{
  // This function only returns the sum of the sum of the components
  float sumValues = 0.0f;
  for(unsigned int i = 0; i < 4; ++i)
  {
    sumValues += a.data[i] + b.data[i];
  }
  return sumValues;
}

void Standard()
{
  Point4 a;
  a.data[0] = 1;
  a.data[1] = 2;
  a.data[2] = 3;
  a.data[3] = 4;

  Point4 b;
  b.data[0] = 1;
  b.data[1] = 6;
  b.data[2] = 3;
  b.data[3] = 5;

  float total = 0.0f;
  for(unsigned int i = 0; i < 1e6; ++i)
  {
    total += SumOfDifferences(a, b);
  }

  std::cout << "total: " << total << std::endl;
}

void Vectorized()
{
  typedef int v4sf __attribute__ (( vector_size(4*sizeof(float)) ));

  v4sf a;
  float* aPointer = (float*)&a;
  aPointer[0] = 1; aPointer[1] = 2; aPointer[2] = 3; aPointer[3] = 4;

  v4sf b;
  float* bPointer = (float*)&b;
  bPointer[0] = 1; bPointer[1] = 2; bPointer[2] = 3; bPointer[3] = 4;

  float total = 0.0f;
  v4sf result;
  float* resultPointer = (float*)&result;

  for(unsigned int i = 0; i < 1e6; ++i)
  {
    result = a + b; // Vectorized operation

    // Sum the components of the result (this is done with the "total += " in the Standard() loop
    for(unsigned int component = 0; component < 4; ++component)
    {
      total += resultPointer[component];
    }
  }

  std::cout << "total: " << total << std::endl;
}

int main()
{

//  Standard();

  Vectorized();

  return 0;
}

but the output is 'inf' for the Vectorized() function. When I stepped through with a debugger, the values of 'result' seem to be garbage (i'd expect them to be (0, 4, 0, 1) ). Where am I going wrong here?

Upvotes: 1

Views: 140

Answers (1)

Michael
Michael

Reputation: 58487

Try typedef float v4sf __attribute__ (( vector_size(4*sizeof(float)) ));
I get 2e+07 as the result.

Upvotes: 1

Related Questions