/**
* Compute a0 for a constant-valued coefficient (GL_FLAT shading).
*/
-static void constant_coef( struct lp_rast_triangle *tri,
+static void constant_coef( struct lp_setup_context *setup,
+ struct lp_rast_triangle *tri,
unsigned slot,
const float value,
unsigned i )
* Compute a0, dadx and dady for a linearly interpolated coefficient,
* for a triangle.
*/
-static void linear_coef( struct lp_rast_triangle *tri,
+static void linear_coef( struct lp_setup_context *setup,
+ struct lp_rast_triangle *tri,
float oneoverarea,
unsigned slot,
const float (*v1)[4],
* to define a0 as the sample at a pixel center somewhere near vmin
* instead - i'll switch to this later.
*/
- tri->inputs.a0[slot][i] = (v1[vert_attr][i] -
- (dadx * (v1[0][0] - 0.5f) +
- dady * (v1[0][1] - 0.5f)));
+ tri->inputs.a0[slot][i] = (a1 -
+ (dadx * (v1[0][0] - setup->pixel_offset) +
+ dady * (v1[0][1] - setup->pixel_offset)));
}
* Later, when we compute the value at a particular fragment position we'll
* divide the interpolated value by the interpolated W at that fragment.
*/
-static void perspective_coef( struct lp_rast_triangle *tri,
+static void perspective_coef( struct lp_setup_context *setup,
+ struct lp_rast_triangle *tri,
float oneoverarea,
unsigned slot,
const float (*v1)[4],
tri->inputs.dadx[slot][i] = dadx;
tri->inputs.dady[slot][i] = dady;
tri->inputs.a0[slot][i] = (a1 -
- (dadx * (v1[0][0] - 0.5f) +
- dady * (v1[0][1] - 0.5f)));
+ (dadx * (v1[0][0] - setup->pixel_offset) +
+ dady * (v1[0][1] - setup->pixel_offset)));
}
* We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
*/
static void
-setup_fragcoord_coef(struct lp_rast_triangle *tri,
+setup_fragcoord_coef(struct lp_setup_context *setup,
+ struct lp_rast_triangle *tri,
float oneoverarea,
unsigned slot,
const float (*v1)[4],
tri->inputs.dadx[slot][1] = 0.0;
tri->inputs.dady[slot][1] = 1.0;
/*Z*/
- linear_coef(tri, oneoverarea, slot, v1, v2, v3, 0, 2);
+ linear_coef(setup, tri, oneoverarea, slot, v1, v2, v3, 0, 2);
/*W*/
- linear_coef(tri, oneoverarea, slot, v1, v2, v3, 0, 3);
+ linear_coef(setup, tri, oneoverarea, slot, v1, v2, v3, 0, 3);
}
-static void setup_facing_coef( struct lp_rast_triangle *tri,
+static void setup_facing_coef( struct lp_setup_context *setup,
+ struct lp_rast_triangle *tri,
unsigned slot,
boolean frontface )
{
- constant_coef( tri, slot, 1.0f - frontface, 0 );
- constant_coef( tri, slot, 0.0f, 1 ); /* wasted */
- constant_coef( tri, slot, 0.0f, 2 ); /* wasted */
- constant_coef( tri, slot, 0.0f, 3 ); /* wasted */
+ constant_coef( setup, tri, slot, 1.0f - frontface, 0 );
+ constant_coef( setup, tri, slot, 0.0f, 1 ); /* wasted */
+ constant_coef( setup, tri, slot, 0.0f, 2 ); /* wasted */
+ constant_coef( setup, tri, slot, 0.0f, 3 ); /* wasted */
}
/**
* Compute the tri->coef[] array dadx, dady, a0 values.
*/
-static void setup_tri_coefficients( struct setup_context *setup,
+static void setup_tri_coefficients( struct lp_setup_context *setup,
struct lp_rast_triangle *tri,
float oneoverarea,
const float (*v1)[4],
const float (*v3)[4],
boolean frontface)
{
- struct lp_scene *scene = lp_setup_get_current_scene(setup);
unsigned slot;
- /* Allocate space for the a0, dadx and dady arrays
- */
- {
- unsigned bytes = (setup->fs.nr_inputs + 1) * 4 * sizeof(float);
- tri->inputs.a0 = lp_scene_alloc_aligned( scene, bytes, 16 );
- tri->inputs.dadx = lp_scene_alloc_aligned( scene, bytes, 16 );
- tri->inputs.dady = lp_scene_alloc_aligned( scene, bytes, 16 );
- }
-
/* The internal position input is in slot zero:
*/
- setup_fragcoord_coef(tri, oneoverarea, 0, v1, v2, v3);
+ setup_fragcoord_coef(setup, tri, oneoverarea, 0, v1, v2, v3);
/* setup interpolation for all the remaining attributes:
*/
switch (setup->fs.input[slot].interp) {
case LP_INTERP_CONSTANT:
for (i = 0; i < NUM_CHANNELS; i++)
- constant_coef(tri, slot+1, v3[vert_attr][i], i);
+ constant_coef(setup, tri, slot+1, v3[vert_attr][i], i);
break;
case LP_INTERP_LINEAR:
for (i = 0; i < NUM_CHANNELS; i++)
- linear_coef(tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i);
+ linear_coef(setup, tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i);
break;
case LP_INTERP_PERSPECTIVE:
for (i = 0; i < NUM_CHANNELS; i++)
- perspective_coef(tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i);
+ perspective_coef(setup, tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i);
break;
case LP_INTERP_POSITION:
/* XXX: fix me - duplicates the values in slot zero.
*/
- setup_fragcoord_coef(tri, oneoverarea, slot+1, v1, v2, v3);
+ setup_fragcoord_coef(setup, tri, oneoverarea, slot+1, v1, v2, v3);
break;
case LP_INTERP_FACING:
- setup_facing_coef(tri, slot+1, frontface);
+ setup_facing_coef(setup, tri, slot+1, frontface);
break;
default:
-static inline int subpixel_snap( float a )
+static INLINE int subpixel_snap( float a )
{
return util_iround(FIXED_ONE * a - (FIXED_ONE / 2));
}
+
+/**
+ * Alloc space for a new triangle plus the input.a0/dadx/dady arrays
+ * immediately after it.
+ * The memory is allocated from the per-scene pool, not per-tile.
+ * \param tri_size returns number of bytes allocated
+ * \param nr_inputs number of fragment shader inputs
+ * \return pointer to triangle space
+ */
+static INLINE struct lp_rast_triangle *
+alloc_triangle(struct lp_scene *scene, unsigned nr_inputs, unsigned *tri_size)
+{
+ unsigned input_array_sz = NUM_CHANNELS * (nr_inputs + 1) * sizeof(float);
+ struct lp_rast_triangle *tri;
+ unsigned bytes;
+ char *inputs;
+
+ assert(sizeof(*tri) % 16 == 0);
+
+ bytes = sizeof(*tri) + (3 * input_array_sz);
+
+ tri = lp_scene_alloc_aligned( scene, bytes, 16 );
+
+ inputs = (char *) (tri + 1);
+ tri->inputs.a0 = (float (*)[4]) inputs;
+ tri->inputs.dadx = (float (*)[4]) (inputs + input_array_sz);
+ tri->inputs.dady = (float (*)[4]) (inputs + 2 * input_array_sz);
+
+ *tri_size = bytes;
+
+ return tri;
+}
+
+
+
/**
* Do basic setup for triangle rasterization and determine which
* framebuffer tiles are touched. Put the triangle in the scene's
* bins for the tiles which we overlap.
*/
static void
-do_triangle_ccw(struct setup_context *setup,
+do_triangle_ccw(struct lp_setup_context *setup,
const float (*v1)[4],
const float (*v2)[4],
const float (*v3)[4],
boolean frontfacing )
{
/* x/y positions in fixed point */
- const int x1 = subpixel_snap(v1[0][0]);
- const int x2 = subpixel_snap(v2[0][0]);
- const int x3 = subpixel_snap(v3[0][0]);
- const int y1 = subpixel_snap(v1[0][1]);
- const int y2 = subpixel_snap(v2[0][1]);
- const int y3 = subpixel_snap(v3[0][1]);
+ const int x1 = subpixel_snap(v1[0][0] + 0.5 - setup->pixel_offset);
+ const int x2 = subpixel_snap(v2[0][0] + 0.5 - setup->pixel_offset);
+ const int x3 = subpixel_snap(v3[0][0] + 0.5 - setup->pixel_offset);
+ const int y1 = subpixel_snap(v1[0][1] + 0.5 - setup->pixel_offset);
+ const int y2 = subpixel_snap(v2[0][1] + 0.5 - setup->pixel_offset);
+ const int y3 = subpixel_snap(v3[0][1] + 0.5 - setup->pixel_offset);
struct lp_scene *scene = lp_setup_get_current_scene(setup);
- struct lp_rast_triangle *tri = lp_scene_alloc_aligned( scene, sizeof *tri, 16 );
- float area, oneoverarea;
+ struct lp_rast_triangle *tri;
+ int area;
+ float oneoverarea;
int minx, maxx, miny, maxy;
+ unsigned tri_bytes;
+
+ tri = alloc_triangle(scene, setup->fs.nr_inputs, &tri_bytes);
+
+#ifdef DEBUG
+ tri->v[0][0] = v1[0][0];
+ tri->v[1][0] = v2[0][0];
+ tri->v[2][0] = v3[0][0];
+ tri->v[0][1] = v1[0][1];
+ tri->v[1][1] = v2[0][1];
+ tri->v[2][1] = v3[0][1];
+#endif
tri->dx12 = x1 - x2;
tri->dx23 = x2 - x3;
tri->dy23 = y2 - y3;
tri->dy31 = y3 - y1;
- area = (tri->dx12 * tri->dy31 -
- tri->dx31 * tri->dy12);
+ area = (tri->dx12 * tri->dy31 - tri->dx31 * tri->dy12);
LP_COUNT(nr_tris);
*
* XXX: subject to overflow??
*/
- if (area <= 0.0f) {
- lp_scene_putback_data( scene, sizeof *tri );
+ if (area <= 0) {
+ lp_scene_putback_data( scene, tri_bytes );
LP_COUNT(nr_culled_tris);
return;
}
if (miny == maxy ||
minx == maxx) {
- lp_scene_putback_data( scene, sizeof *tri );
+ lp_scene_putback_data( scene, tri_bytes );
LP_COUNT(nr_culled_tris);
return;
}
*/
setup_tri_coefficients( setup, tri, oneoverarea, v1, v2, v3, frontfacing );
+ tri->inputs.facing = frontfacing ? 1.0F : -1.0F;
+
/* half-edge constants, will be interated over the whole render target.
*/
tri->c1 = tri->dy12 * x1 - tri->dx12 * y1;
tri->ei2 = tri->dx23 - tri->dy23 - tri->eo2;
tri->ei3 = tri->dx31 - tri->dy31 - tri->eo3;
+ /* Fill in the inputs.step[][] arrays.
+ * We've manually unrolled some loops here.
+ */
{
const int xstep1 = -tri->dy12;
const int xstep2 = -tri->dy23;
const int xstep3 = -tri->dy31;
-
const int ystep1 = tri->dx12;
const int ystep2 = tri->dx23;
const int ystep3 = tri->dx31;
-
- int qx, qy, ix, iy;
- int i = 0;
-
- for (qy = 0; qy < 2; qy++) {
- for (qx = 0; qx < 2; qx++) {
- for (iy = 0; iy < 2; iy++) {
- for (ix = 0; ix < 2; ix++, i++) {
- int x = qx * 2 + ix;
- int y = qy * 2 + iy;
- tri->inputs.step[0][i] = x * xstep1 + y * ystep1;
- tri->inputs.step[1][i] = x * xstep2 + y * ystep2;
- tri->inputs.step[2][i] = x * xstep3 + y * ystep3;
- }
- }
- }
- }
+
+#define SETUP_STEP(i, x, y) \
+ do { \
+ tri->inputs.step[0][i] = x * xstep1 + y * ystep1; \
+ tri->inputs.step[1][i] = x * xstep2 + y * ystep2; \
+ tri->inputs.step[2][i] = x * xstep3 + y * ystep3; \
+ } while (0)
+
+ SETUP_STEP(0, 0, 0);
+ SETUP_STEP(1, 1, 0);
+ SETUP_STEP(2, 0, 1);
+ SETUP_STEP(3, 1, 1);
+
+ SETUP_STEP(4, 2, 0);
+ SETUP_STEP(5, 3, 0);
+ SETUP_STEP(6, 2, 1);
+ SETUP_STEP(7, 3, 1);
+
+ SETUP_STEP(8, 0, 2);
+ SETUP_STEP(9, 1, 2);
+ SETUP_STEP(10, 0, 3);
+ SETUP_STEP(11, 1, 3);
+
+ SETUP_STEP(12, 2, 2);
+ SETUP_STEP(13, 3, 2);
+ SETUP_STEP(14, 2, 3);
+ SETUP_STEP(15, 3, 3);
+#undef STEP
}
/*
maxx = maxx / TILE_SIZE;
maxy = maxy / TILE_SIZE;
- /* Clamp maxx, maxy to framebuffer size
+ /*
+ * Clamp to framebuffer size
*/
+ minx = MAX2(minx, 0);
+ miny = MAX2(miny, 0);
maxx = MIN2(maxx, scene->tiles_x - 1);
maxy = MIN2(maxy, scene->tiles_y - 1);
}
-static void triangle_cw( struct setup_context *setup,
+static void triangle_cw( struct lp_setup_context *setup,
const float (*v0)[4],
const float (*v1)[4],
const float (*v2)[4] )
}
-static void triangle_ccw( struct setup_context *setup,
+static void triangle_ccw( struct lp_setup_context *setup,
const float (*v0)[4],
const float (*v1)[4],
const float (*v2)[4] )
}
-static void triangle_both( struct setup_context *setup,
+static void triangle_both( struct lp_setup_context *setup,
const float (*v0)[4],
const float (*v1)[4],
const float (*v2)[4] )
}
-static void triangle_nop( struct setup_context *setup,
+static void triangle_nop( struct lp_setup_context *setup,
const float (*v0)[4],
const float (*v1)[4],
const float (*v2)[4] )
void
-lp_setup_choose_triangle( struct setup_context *setup )
+lp_setup_choose_triangle( struct lp_setup_context *setup )
{
switch (setup->cullmode) {
case PIPE_WINDING_NONE: