kernel4: separate area from alpha in clip stack

This change prepares for kernel4 to output alpha. No functional changes.

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

piet-gpu/shader/coarse.comp

@@ -418,8 +418,8 @@ void main() {
     if (bin_tile_x + tile_x < conf.width_in_tiles && bin_tile_y + tile_y < conf.height_in_tiles) {
         Cmd_End_write(cmd_alloc, cmd_ref);
         if (num_begin_slots > 0) {
-            // Write scratch allocation: one word per BeginClip per rasterizer chunk.
+            // Write scratch allocation: one state per BeginClip per rasterizer chunk.
-            uint scratch_size = num_begin_slots * TILE_WIDTH_PX * TILE_HEIGHT_PX * 4;
+            uint scratch_size = num_begin_slots * TILE_WIDTH_PX * TILE_HEIGHT_PX * CLIP_STATE_SIZE * 4;
             MallocResult scratch = malloc(scratch_size);
             // Ignore scratch.failed; we don't use the allocation and kernel4
             // checks for memory overflow before using it.

piet-gpu/shader/kernel4.comp

@@ -201,25 +201,28 @@ void main() {
             cmd_ref.offset += 4 + CmdImage_size;
             break;
         case Cmd_BeginClip:
-            uint base_ix = (scratch_alloc.offset >> 2) + clip_depth * TILE_WIDTH_PX * TILE_HEIGHT_PX +
+            uint base_ix = (scratch_alloc.offset >> 2) + CLIP_STATE_SIZE * (clip_depth * TILE_WIDTH_PX * TILE_HEIGHT_PX +
-                gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_LocalInvocationID.y;
+                gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_LocalInvocationID.y);
             for (uint k = 0; k < CHUNK; k++) {
                 uvec2 offset = chunk_offset(k);
-                uint state = packsRGB(vec4(rgb[k], clamp(abs(area[k]), 0.0, 1.0)));
+                uint srgb = packsRGB(vec4(rgb[k], 1.0));
-                write_mem(scratch_alloc, base_ix + offset.x + offset.y * TILE_WIDTH_PX, state);
+                float alpha = clamp(abs(area[k]), 0.0, 1.0);
+                write_mem(scratch_alloc, base_ix + 0 + CLIP_STATE_SIZE * (offset.x + offset.y * TILE_WIDTH_PX), srgb);
+                write_mem(scratch_alloc, base_ix + 1 + CLIP_STATE_SIZE * (offset.x + offset.y * TILE_WIDTH_PX), floatBitsToUint(alpha));
             }
             clip_depth++;
             cmd_ref.offset += 4;
             break;
         case Cmd_EndClip:
             clip_depth--;
-            base_ix = (scratch_alloc.offset >> 2) + clip_depth * TILE_WIDTH_PX * TILE_HEIGHT_PX +
+            base_ix = (scratch_alloc.offset >> 2) + CLIP_STATE_SIZE * (clip_depth * TILE_WIDTH_PX * TILE_HEIGHT_PX +
-                gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_LocalInvocationID.y;
+                gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_LocalInvocationID.y);
             for (uint k = 0; k < CHUNK; k++) {
                 uvec2 offset = chunk_offset(k);
-                uint state = read_mem(scratch_alloc, base_ix + offset.x + offset.y * TILE_WIDTH_PX);
+                uint srgb = read_mem(scratch_alloc, base_ix + 0 + CLIP_STATE_SIZE * (offset.x + offset.y * TILE_WIDTH_PX));
-                vec4 rgba = unpacksRGB(state);
+                uint alpha = read_mem(scratch_alloc, base_ix + 1 + CLIP_STATE_SIZE * (offset.x + offset.y * TILE_WIDTH_PX));
-                rgb[k] = mix(rgba.rgb, rgb[k], area[k] * rgba.a);
+                vec4 rgba = unpacksRGB(srgb);
+                rgb[k] = mix(rgba.rgb, rgb[k], area[k] * uintBitsToFloat(alpha));
             }
             cmd_ref.offset += 4;
             break;

piet-gpu/shader/setup.h

@@ -44,6 +44,9 @@ struct Config {
 #define MODE_NONZERO 0
 #define MODE_STROKE 1
 
+// Size of kernel4 clip state, in words.
+#define CLIP_STATE_SIZE 2
+
 // fill_mode_from_flags extracts the fill mode from tag flags.
 uint fill_mode_from_flags(uint flags) {
     return flags & 0x1;