[EmulGL] Unwind loops in gl_util

gl_util contains lots of unneeded if checks and its matrix-matrix, vector-vector multiplication is done in a loop way that can be unwounded. With JDK 18 and Vector API it would yield even more improvement.

I.e.:

public static float[] mulMatrix41(float a[], float b[]) {
    // assume a = 4x4, b = 4x1
    float temp[] = new float[4];
    int i, j, x;
    for (i = 0; i < 4; i++) {
      for (j = 0; j < 4; j++) {
        x = j << 2 | i;
        if (a[x] != 0 && b[j] != 0) {
          if (a[x] == 1) {
            temp[i] += b[j];
          } else if (b[j] == 1) {
            temp[i] += a[x];
          } else {
            temp[i] += a[x] * b[j];
          }
        }
      }
    }
    return temp;
  }

Can be unwound to:

  float temp[] = new float[4];
  temp[0] += a[0] * b[0];
  temp[0] += a[4] * b[1];
  temp[0] += a[8] * b[2];
  temp[0] += a[12] * b[3];
  temp[1] += a[1] * b[0];
  temp[1] += a[5] * b[1];
  temp[1] += a[9] * b[2];
  temp[1] += a[13] * b[3];
  temp[2] += a[2] * b[0];
  temp[2] += a[6] * b[1];
  temp[2] += a[10] * b[2];
  temp[2] += a[14] * b[3];
  temp[3] += a[3] * b[0];
  temp[3] += a[7] * b[1];
  temp[3] += a[11] * b[2];
  temp[3] += a[15] * b[3];
  return temp;

[valb3r]

Basically new version outperforms 3-4 times old version

[jzy3d]

Useful info related to unwinding loops : I went into GLU to debug a 3D to 2D projection stuff and noticed a lot of for loops for matrix operations in the GLU class. For the record they’re used mainly in gluProject / gluUnProject. These primitives are used for 3D text position on screen and mouse selection (hitting 3D world from a mouse click), so not that often, but worth knowing.