Signed distance field optimizations

[duvisit]

After trying demos and reading through source code of edt3aafunc.c I think there is some room for small optimizations.

• Use float instead of double in distance field computations.

• Remove lines from computegradient() function

---

#if 0
glength = gx[k]*gx[k] + gy[k]*gy[k];
if(glength > 0.0) { // Avoid division by zero
    glength = sqrt(glength);
    gx[k]=gx[k]/glength;
    gy[k]=gy[k]/glength;
}
#endif

---

and lines from edgedf() function

---

#if 0
    if(glength>0) {
        gx = gx/glength;
        gy = gy/glength;
    }
    /* Everything is symmetric wrt sign and transposition,
     * so move to first octant (gx>=0, gy>=0, gx>=gy) to
     * avoid handling all possible edge directions.
     */
    gx = fabs(gx);
    gy = fabs(gy);
#else
    gx = fabs(gx/glength);
    gy = fabs(gy/glength);
#endif

---

Division by zero is handled by the if (gx==0 || gy==0) few lines before.

• In edtaa3() function use:

---

  #if 0
    if(a > 1.0) a = 1.0;
    if(a < 0.0) a = 0.0; // Clip grayscale values outside the range [0,1]
    if(a == 0.0) return 1000000.0; // Not an object pixel, return "very far" ("don't know yet")
  #else
    if(a <= 0.0f) return 1000000.0f;
    if(a > 1.0f) a = 1.0f;
  #endif

---

Hope this doesn't break anything.

Great library, bye.

[rougier]

Thanks. the edt3aafunc.c comes from another library and mostly used for illustration. Not sure it's worth optimizing it and it may break things. Did you notice huge speedup with your patch?

[duvisit]

I get around 20% improvements. Probably mostly from double to float change.

[rougier]

Maybe not worth the change.

[duvisit]

I agree. PR #278 is better.