Combine info and bin_data buffers

We need to reduce the number of buffer bindings to 8 so it can run on
all WebGPU devices. The best candidate was to combine info and bin_data,
which are written by two different stages (draw_leaf and binning,
respectively), both read by coarse, and are unstructured, so the only
real shader change needed is to add the offset for the binning data.

I thought I was going to have to do a blit to make it fit, but
fortunately that wasn't needed.

Progress to #202

piet-wgsl/shader/binning.wgsl

@@ -146,7 +146,7 @@ fn main(
                 let count_packed = sh_count[count_ix / 2u][bin_ix];
                 idx += (count_packed >> (16u * (count_ix & 1u))) & 0xffffu;
             }
-            let offset = sh_chunk_offset[bin_ix];
+            let offset = config.bin_data_start + sh_chunk_offset[bin_ix];
             bin_data[offset + idx] = element_ix;
         }
         x += 1;

piet-wgsl/shader/coarse.wgsl

@@ -27,7 +27,7 @@ struct BinHeader {
 var<storage> bin_headers: array<BinHeader>;
 
 @group(0) @binding(4)
-var<storage> bin_data: array<u32>;
+var<storage> info_bin_data: array<u32>;
 
 @group(0) @binding(5)
 var<storage> paths: array<Path>;
@@ -36,12 +36,9 @@ var<storage> paths: array<Path>;
 var<storage> tiles: array<Tile>;
 
 @group(0) @binding(7)
-var<storage> info: array<u32>;
-
-@group(0) @binding(8)
 var<storage, read_write> bump: BumpAllocators;
 
-@group(0) @binding(9)
+@group(0) @binding(8)
 var<storage, read_write> ptcl: array<u32>;
 
 
@@ -208,8 +205,8 @@ fn main(
                     }
                 }
                 ix -= select(part_start_ix, sh_part_count[part_ix - 1u], part_ix > 0u);
-                let offset = sh_part_offsets[part_ix];
-                sh_drawobj_ix[local_id.x] = bin_data[offset + ix];
+                let offset = config.bin_data_start + sh_part_offsets[part_ix];
+                sh_drawobj_ix[local_id.x] = info_bin_data[offset + ix];
             }
             wr_ix = min(rd_ix + N_TILE, ready_ix);
             if wr_ix - rd_ix >= N_TILE || (wr_ix >= ready_ix && partition_ix >= n_partitions) {
@@ -326,14 +323,14 @@ fn main(
                 switch drawtag {
                     // DRAWTAG_FILL_COLOR
                     case 0x44u: {
-                        let linewidth = bitcast<f32>(info[di]);
+                        let linewidth = bitcast<f32>(info_bin_data[di]);
                         write_path(tile, linewidth);
                         let rgba_color = scene[dd];
                         write_color(CmdColor(rgba_color));
                     }
                     // DRAWTAG_FILL_LIN_GRADIENT
                     case 0x114u: {
-                        let linewidth = bitcast<f32>(info[di]);
+                        let linewidth = bitcast<f32>(info_bin_data[di]);
                         write_path(tile, linewidth);
                         let index = scene[dd];
                         let info_offset = di + 1u;
@@ -341,7 +338,7 @@ fn main(
                     }
                     // DRAWTAG_FILL_RAD_GRADIENT
                     case 0x2dcu: {
-                        let linewidth = bitcast<f32>(info[di]);
+                        let linewidth = bitcast<f32>(info_bin_data[di]);
                         write_path(tile, linewidth);
                         let index = scene[dd];
                         let info_offset = di + 1u;

piet-wgsl/shader/shared/config.wgsl

@@ -11,6 +11,10 @@ struct Config {
     n_path: u32,
     n_clip: u32,
 
+    // To reduce the number of bindings, info and bin data are combined
+    // into one buffer.
+    bin_data_start: u32,
+
     // offsets within scene buffer (in u32 units)
     // Note: this is a difference from piet-gpu, which is in bytes
     pathtag_base: u32,

piet-wgsl/src/engine.rs

@@ -329,7 +329,7 @@ impl Engine {
                 }
                 Command::Clear(proxy, offset, size) => {
                     let buffer = bind_map.get_or_create(*proxy, device)?;
-                    encoder.clear_buffer(buffer, *offset, *size)
+                    encoder.clear_buffer(buffer, *offset, *size);
                 }
             }
         }

piet-wgsl/src/render.rs

@@ -33,6 +33,7 @@ struct Config {
     n_drawobj: u32,
     n_path: u32,
     n_clip: u32,
+    bin_data_start: u32,
     pathtag_base: u32,
     pathdata_base: u32,
     drawtag_base: u32,
@@ -209,6 +210,7 @@ pub fn render_full(
     // TODO: calculate for real when we do rectangles
     let n_drawobj = n_path;
     let n_clip = data.n_clip;
+    let bin_data_start = n_drawobj * MAX_DRAWINFO_SIZE as u32;
 
     let new_width = next_multiple_of(width, 16);
     let new_height = next_multiple_of(height, 16);
@@ -222,6 +224,7 @@ pub fn render_full(
         n_drawobj,
         n_path,
         n_clip,
+        bin_data_start,
         pathtag_base,
         pathdata_base,
         drawtag_base,
@@ -278,7 +281,7 @@ pub fn render_full(
         [config_buf, scene_buf, draw_reduced_buf],
     );
     let draw_monoid_buf = ResourceProxy::new_buf(n_drawobj as u64 * DRAWMONOID_SIZE);
-    let info_buf = ResourceProxy::new_buf(n_drawobj as u64 * MAX_DRAWINFO_SIZE);
+    let info_bin_data_buf = ResourceProxy::new_buf(1 << 20);
     let clip_inp_buf = ResourceProxy::new_buf(data.n_clip as u64 * CLIP_INP_SIZE);
     recording.dispatch(
         shaders.draw_leaf,
@@ -289,7 +292,7 @@ pub fn render_full(
             draw_reduced_buf,
             path_bbox_buf,
             draw_monoid_buf,
-            info_buf,
+            info_bin_data_buf,
             clip_inp_buf,
         ],
     );
@@ -329,7 +332,6 @@ pub fn render_full(
     }
     let draw_bbox_buf = ResourceProxy::new_buf(n_path as u64 * DRAW_BBOX_SIZE);
     let bump_buf = BufProxy::new(BUMP_SIZE);
-    let bin_data_buf = ResourceProxy::new_buf(1 << 20);
     let width_in_bins = (config.width_in_tiles + 15) / 16;
     let height_in_bins = (config.height_in_tiles + 15) / 16;
     let n_bins = width_in_bins * height_in_bins;
@@ -346,7 +348,7 @@ pub fn render_full(
             clip_bbox_buf,
             draw_bbox_buf,
             bump_buf,
-            bin_data_buf,
+            info_bin_data_buf,
             bin_header_buf,
         ],
     );
@@ -395,10 +397,9 @@ pub fn render_full(
             scene_buf,
             draw_monoid_buf,
             bin_header_buf,
-            bin_data_buf,
+            info_bin_data_buf,
             path_buf,
             tile_buf,
-            info_buf,
             bump_buf,
             ptcl_buf,
         ],
@@ -414,7 +415,7 @@ pub fn render_full(
             ResourceProxy::Image(out_image),
             ptcl_buf,
             gradient_image,
-            info_buf,
+            info_bin_data_buf,
         ],
     );
     (recording, ResourceProxy::Image(out_image))

piet-wgsl/src/shaders.rs

@@ -259,7 +259,6 @@ pub fn full_shaders(device: &Device, engine: &mut Engine) -> Result<FullShaders,
             BindType::BufReadOnly,
             BindType::BufReadOnly,
             BindType::BufReadOnly,
-            BindType::BufReadOnly,
             BindType::Buffer,
             BindType::Buffer,
         ],

piet-wgsl/src/util.rs

@@ -34,7 +34,6 @@ impl RenderContext {
         let adapter = instance.request_adapter(&Default::default()).await.unwrap();
         let features = adapter.features();
         let mut limits = Limits::default();
-        limits.max_storage_buffers_per_shader_stage = 16;
         let (device, queue) = adapter
             .request_device(
                 &wgpu::DeviceDescriptor {