On 06/11/2012 10:16 AM, Eric Anholt wrote:
On Fri, 8 Jun 2012 14:31:16 -0600, Brian Paul<[email protected]> wrote:
diff --git a/tests/general/drawpix-z.c b/tests/general/drawpix-z.c
new file mode 100644
index 0000000..cdd6ac9
--- /dev/null
+++ b/tests/general/drawpix-z.c
The code would have made more sense to me if you had:
float ndc(float z)
{
return -1.0 + 2.0 * z;
}
+static void
+draw_z_gradient(GLfloat zLeft, GLfloat zRight)
+{
+ GLfloat verts[4][3];
+
+ verts[0][0] = -1.0; verts[0][1] = -1.0; verts[0][2] = zLeft;
+ verts[1][0] = 1.0; verts[1][1] = -1.0; verts[1][2] = zRight;
+ verts[2][0] = 1.0; verts[2][1] = 1.0; verts[2][2] = zRight;
+ verts[3][0] = -1.0; verts[3][1] = 1.0; verts[3][2] = zLeft;
verts[0][0] = -1.0; verts[0][1] = -1.0; verts[0][2] = ndc(zLeft);
verts[1][0] = 1.0; verts[1][1] = -1.0; verts[1][2] = ndc(zRight);
verts[2][0] = 1.0; verts[2][1] = 1.0; verts[2][2] = ndc(zRight);
verts[3][0] = -1.0; verts[3][1] = 1.0; verts[3][2] = ndc(zLeft);
+enum piglit_result
+piglit_display(void)
+{
float left = epsilon;
float right = 1.0 - epsilon;
+
+ /* create image of Z values increasing from left to right */
+ buf = (GLfloat *)
+ malloc(piglit_width * piglit_height * sizeof(GLfloat));
+ for (j = 0; j< piglit_height; j++) {
+ for (i = 0; i< piglit_width; i++) {
+ float z = i / (float) (piglit_width - 1);
+ z = epsilon + (z * (1.0 - 2.0 * epsilon));
z = left + i * (right - left);
+ buf[j * piglit_width + i] = z;
+ }
+ }
+
+ /* draw a red quad behind the Z gradient - it should not be visible */
+ glColor4fv(red);
+ draw_z_gradient(-1.0 + 2.0 * epsilon, 1.0);
draw_z_gradient(left + epsilon, right + epsilon);
+ draw_z_gradient(-1.0, 1.0 - 2.0 * epsilon);
draw_z_gradient(left - epsilon, right - epsilon);
With the code you had, if I'm reading this right, the epsilon in
drawpixels is on a different scale from the epsilon in draw_z_gradient,
so on the +/- (1.0 - 2.0 * epsilon) side, the zs are equal to the
z that was DrawPixels()ed, and basically you're getting lucky.
Since the Z ortho range was [-1,1] while zbuffer vals are [0,1] I was
off by a factor of two. My epsilon was large enough to hide that.
BTW, having a tight epsilon isn't a concern in this particular test.
Thanks for the review. I think I can simplify this further by simply
using an ortho projection such that Z is in [0,1] instead of [-1,1]...
-Brian
_______________________________________________
Piglit mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/piglit
