coarse,binning: organize bins into width_in_bins x height_in_bins

The binning shader supports up to N_TILE bins. To efficiently cover wide or
tall viewports, convert the rigid N_TILE_X x N_TILE_Y bin layout to a variable
width_in_bins x height_in_bins layout.

Signed-off-by: Elias Naur <mail@eliasnaur.com>

piet-gpu/shader/binning.comp

@@ -78,16 +78,18 @@ void main() {
     // trying to keep divergence low.
     // Right now, it's just a bbox, but we'll get finer with
     // segments.
-    x0 = clamp(x0, 0, N_TILE_X);
-    x1 = clamp(x1, x0, N_TILE_X);
-    y0 = clamp(y0, 0, N_TILE_Y);
-    y1 = clamp(y1, y0, N_TILE_Y);
+    uint width_in_bins = (conf.width_in_tiles + N_TILE_X - 1)/N_TILE_X;
+    uint height_in_bins = (conf.height_in_tiles + N_TILE_Y - 1)/N_TILE_Y;
+    x0 = clamp(x0, 0, int(width_in_bins));
+    x1 = clamp(x1, x0, int(width_in_bins));
+    y0 = clamp(y0, 0, int(height_in_bins));
+    y1 = clamp(y1, y0, int(height_in_bins));
     if (x0 == x1) y1 = y0;
     int x = x0, y = y0;
     uint my_slice = gl_LocalInvocationID.x / 32;
     uint my_mask = 1 << (gl_LocalInvocationID.x & 31);
     while (y < y1) {
-        atomicOr(bitmaps[my_slice][y * N_TILE_X + x], my_mask);
+        atomicOr(bitmaps[my_slice][y * width_in_bins + x], my_mask);
         x++;
         if (x == x1) {
             x = x0;
@@ -128,7 +130,7 @@ void main() {
     x = x0;
     y = y0;
     while (y < y1) {
-        uint bin_ix = y * N_TILE_X + x;
+        uint bin_ix = y * width_in_bins + x;
         uint out_mask = bitmaps[my_slice][bin_ix];
         if ((out_mask & my_mask) != 0) {
             uint idx = bitCount(out_mask & (my_mask - 1));

piet-gpu/shader/coarse.comp

@@ -71,7 +71,8 @@ void main() {
 
     // Could use either linear or 2d layouts for both dispatch and
     // invocations within the workgroup. We'll use variables to abstract.
-    uint bin_ix = N_TILE_X * gl_WorkGroupID.y + gl_WorkGroupID.x;
+    uint width_in_bins = (conf.width_in_tiles + N_TILE_X - 1)/N_TILE_X;
+    uint bin_ix = width_in_bins * gl_WorkGroupID.y + gl_WorkGroupID.x;
     uint partition_ix = 0;
     uint n_partitions = (conf.n_elements + N_TILE - 1) / N_TILE;
     uint th_ix = gl_LocalInvocationID.x;