Another checkpoint

piet-wgsl/shader/coarse.wgsl

@@ -71,7 +71,8 @@ var<storage> tiles: array<Tile>;
 // Much of this code assumes WG_SIZE == N_TILE. If these diverge, then
 // a fair amount of fixup is needed.
 let WG_SIZE = 256u;
-let N_SLICE = WG_SIZE / 32u;
+//let N_SLICE = WG_SIZE / 32u;
+let N_SLICE = 8u;
 
 var<workgroup> sh_bitmaps: array<array<atomic<u32>, N_TILE>, N_SLICE>;
 var<workgroup> sh_part_count: array<u32, WG_SIZE>;
@@ -174,7 +175,7 @@ fn main(
                     sh_part_count[local_id.x] = count;
                     workgroupBarrier();
                     if local_id.x >= (1u << i) {
-                        count += sh_part_count[local_id - (1u << i)];
+                        count += sh_part_count[local_id.x - (1u << i)];
                     }
                     workgroupBarrier();
                 }
@@ -235,7 +236,7 @@ fn main(
 
         // Prefix sum of tile counts
         sh_tile_count[local_id.x] = tile_count;
-        for (var i = 0; i < firstTrailingBit(N_TILE); i += 1u) {
+        for (var i = 0u; i < firstTrailingBit(N_TILE); i += 1u) {
             workgroupBarrier();
             if local_id.x >= (1u << i) {
                 tile_count += sh_tile_count[local_id.x - (1u << i)];
@@ -298,11 +299,13 @@ fn main(
             let drawtag = scene[config.drawtag_base + drawobj_ix];
             let dm = draw_monoids[drawobj_ix];
             let dd = config.drawdata_base + dm.scene_offset;
+            // TODO: set up draw info from monoid
             if clip_zero_depth == 0u {
                 let tile_ix = sh_tile_base[el_ix] + sh_tile_stride[el_ix] * tile_y + tile_x;
                 let tile = tiles[tile_ix];
                 switch drawtag {
-                    case DRAWTAG_FILL_COLOR: {
+                    // DRAWTAG_FILL_COLOR
+                    case 0x44u: {
                         // TODO: get linewidth from draw object
                         let linewidth = -1.0;
                         let rgba_color = scene[dd];

piet-wgsl/shader/draw_leaf.wgsl

@@ -50,7 +50,7 @@ fn main(
 ) {
     let ix = global_id.x;
     let tag_word = scene[config.drawtag_base + ix];
-    let agg = map_draw_tag(tag_word);
+    var agg = map_draw_tag(tag_word);
     sh_scratch[local_id.x] = agg;
     for (var i = 0u; i < firstTrailingBit(WG_SIZE); i += 1u) {
         workgroupBarrier();
@@ -64,7 +64,7 @@ fn main(
     workgroupBarrier();
     var m = draw_monoid_identity();
     if wg_id.x > 0u {
-        m = parent[wg_id.x - 1u];
+        m = reduced[wg_id.x - 1u];
     }
     if local_id.x > 0u {
         m = combine_draw_monoid(m, sh_scratch[local_id.x - 1u]);
@@ -84,7 +84,7 @@ fn main(
         let y1 = f32(bbox.y1) - 32768.0;
         let bbox_f = vec4(x0, y0, x1, y1);
         let fill_mode = u32(bbox.linewidth >= 0.0);
-        var mat: vec4<f32>;
+        var matrx: vec4<f32>;
         var translate: vec2<f32>;
         var linewidth = bbox.linewidth;
         if linewidth >= 0.0 || tag_word == DRAWTAG_FILL_LIN_GRADIENT || tag_word == DRAWTAG_FILL_RAD_GRADIENT {
@@ -92,18 +92,20 @@ fn main(
         }
         if linewidth >= 0.0 {
             // Note: doesn't deal with anisotropic case
-            linewidth *= sqrt(abs(mat.x * mat.w - mat.y * mat.z)); 
+            linewidth *= sqrt(abs(matrx.x * matrx.w - matrx.y * matrx.z)); 
         }
         switch tag_word {
-            case DRAWTAG_FILL_COLOR, DRAWTAG_FILL_IMAGE: {
+            // DRAWTAG_FILL_COLOR, DRAWTAG_FILL_IMAGE
+            case 0x44u, 0x48u: {
                 info[di] = bitcast<u32>(linewidth);
             }
-            case DRAWTAG_FILL_LIN_GRADIENT: {
+            // DRAWTAG_FILL_LIN_GRADIENT
+            case 0x114u: {
                 info[di] = bitcast<u32>(linewidth);
                 var p0 = bitcast<vec2<f32>>(vec2(scene[dd + 1u], scene[dd + 2u]));
                 var p1 = bitcast<vec2<f32>>(vec2(scene[dd + 3u], scene[dd + 4u]));
-                p0 = mat.xy * p0.x + mat.zw * p0.y + translate;
+                p0 = matrx.xy * p0.x + matrx.zw * p0.y + translate;
-                p1 = mat.xy * p1.x + mat.zw * p1.y + translate;
+                p1 = matrx.xy * p1.x + matrx.zw * p1.y + translate;
                 let dxy = p1 - p0;
                 let scale = 1.0 / dot(dxy, dxy);
                 let line_xy = dxy * scale;
@@ -112,14 +114,15 @@ fn main(
                 info[di + 2u] = bitcast<u32>(line_xy.y);
                 info[di + 3u] = bitcast<u32>(line_c);
             }
-            case DRAWTAG_FILL_RAD_GRADIENT: {
+            // DRAWTAG_FILL_RAD_GRADIENT
+            case 0x2dcu: {
                 info[di] = bitcast<u32>(linewidth);
                 var p0 = bitcast<vec2<f32>>(vec2(scene[dd + 1u], scene[dd + 2u]));
                 var p1 = bitcast<vec2<f32>>(vec2(scene[dd + 3u], scene[dd + 4u]));
                 let r0 = bitcast<f32>(scene[dd + 5u]);
                 let r1 = bitcast<f32>(scene[dd + 6u]);
-                let inv_det = 1.0 / (mat.x * mat.w - mat.y * mat.z);
+                let inv_det = 1.0 / (matrx.x * matrx.w - matrx.y * matrx.z);
-                let inv_mat = inv_det * vec4<f32>(mat.w, -mat.y, -mat.z, mat.x);
+                let inv_mat = inv_det * vec4<f32>(matrx.w, -matrx.y, -matrx.z, matrx.x);
                 var inv_tr = inv_mat.xz * translate.x + inv_mat.yw * translate.y;
                 inv_tr += p0;
                 let center1 = p1 - p0;

piet-wgsl/shader/path_coarse.wgsl

@@ -64,9 +64,6 @@ fn read_i16_point(ix: u32) -> vec2<f32> {
 }
 
 #ifndef cubics_out
-let TILE_WIDTH = 16u;
-let TILE_HEIGHT = 16u;
-
 struct SubdivResult {
     val: f32,
     a0: f32,

piet-wgsl/shader/shared/config.wgsl

@@ -28,20 +28,10 @@ struct Config {
 
 // Geometry of tiles and bins
 
-
 let TILE_WIDTH = 16u;
 let TILE_HEIGHT = 16u;
 // Number of tiles per bin
 let N_TILE_X = 16u;
 let N_TILE_Y = 16u;
-let N_TILE = N_TILE_X * N_TILE_Y;
+//let N_TILE = N_TILE_X * N_TILE_Y;
-
+let N_TILE = 256u;
-// Should ptcl stuff move to a separate import?
-
-// Layout of per-tile command list
-// Initial allocation, in u32's.
-let PTCL_INITIAL_ALLOC = 64u;
-let PTCL_INCREMENT = 256u;
-
-// Amount of space taken by jump
-let PTCL_HEADROOM = 2u;

piet-wgsl/shader/shared/drawtag.wgsl

@@ -41,12 +41,13 @@ fn draw_monoid_identity() -> DrawMonoid {
     return DrawMonoid();
 }
 
-fn combine_draw_monoid(a: DrawMonoid, b: DrawMonoid) {
+fn combine_draw_monoid(a: DrawMonoid, b: DrawMonoid) -> DrawMonoid {
     var c: DrawMonoid;
     c.path_ix = a.path_ix + b.path_ix;
     c.clip_ix = a.clip_ix + b.clip_ix;
     c.scene_offset = a.scene_offset + b.scene_offset;
     c.info_offset = a.info_offset + b.info_offset;
+    return c;
 }
 
 fn map_draw_tag(tag_word: u32) -> DrawMonoid {

piet-wgsl/shader/shared/ptcl.wgsl

@@ -14,6 +14,14 @@
 //
 // Also licensed under MIT license, at your choice.
 
+// Layout of per-tile command list
+// Initial allocation, in u32's.
+let PTCL_INITIAL_ALLOC = 64u;
+let PTCL_INCREMENT = 256u;
+
+// Amount of space taken by jump
+let PTCL_HEADROOM = 2u;
+
 // Tags for PTCL commands
 let CMD_END = 0u;
 let CMD_FILL = 1u;
@@ -30,7 +38,7 @@ struct CmdFill {
 }
 
 struct CmdJump {
-    target: u32,
+    new_ix: u32,
 }
 
 struct CmdColor {