implement variable output sizing

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

piet-gpu/shader/coarse.comp

@@ -83,7 +83,7 @@ void main() {
     // Per-tile state
     uint tile_x = gl_LocalInvocationID.x % N_TILE_X;
     uint tile_y = gl_LocalInvocationID.x / N_TILE_X;
-    uint this_tile_ix = (bin_tile_y + tile_y) * WIDTH_IN_TILES + bin_tile_x + tile_x;
+    uint this_tile_ix = (bin_tile_y + tile_y) * conf.width_in_tiles + bin_tile_x + tile_x;
     CmdRef cmd_ref = CmdRef(conf.ptcl_base + this_tile_ix * PTCL_INITIAL_ALLOC);
     uint cmd_limit = cmd_ref.offset + PTCL_INITIAL_ALLOC - 2 * Cmd_size;
     // The nesting depth of the clip stack
@@ -370,7 +370,7 @@ void main() {
         rd_ix += N_TILE;
         if (rd_ix >= ready_ix && partition_ix >= n_partitions) break;
     }
-    if (bin_tile_x + tile_x < WIDTH_IN_TILES && bin_tile_y + tile_y < HEIGHT_IN_TILES) {
+    if (bin_tile_x + tile_x < conf.width_in_tiles && bin_tile_y + tile_y < conf.height_in_tiles) {
         Cmd_End_write(cmd_ref);
     }
 }

piet-gpu/shader/kernel4.comp

@@ -91,7 +91,7 @@ void main() {
         return;
     }
 
-    uint tile_ix = gl_WorkGroupID.y * WIDTH_IN_TILES + gl_WorkGroupID.x;
+    uint tile_ix = gl_WorkGroupID.y * conf.width_in_tiles + gl_WorkGroupID.x;
     CmdRef cmd_ref = CmdRef(conf.ptcl_base + tile_ix * PTCL_INITIAL_ALLOC);
 
     uvec2 xy_uint = uvec2(gl_GlobalInvocationID.x, gl_LocalInvocationID.y + TILE_HEIGHT_PX * gl_WorkGroupID.y);

piet-gpu/shader/setup.h

@@ -10,10 +10,6 @@
 #define LG_WG_FACTOR 1
 #define WG_FACTOR (1<<LG_WG_FACTOR)
 
-// TODO: compute all these
-
-#define WIDTH_IN_TILES 128
-#define HEIGHT_IN_TILES 96
 #define TILE_WIDTH_PX 16
 #define TILE_HEIGHT_PX 16
 
@@ -32,6 +28,8 @@
 struct Config {
     uint n_elements; // paths
     uint n_pathseg;
+    uint width_in_tiles;
+    uint height_in_tiles;
     uint tile_base;
     uint bin_base;
     uint ptcl_base;

piet-gpu/shader/tile_alloc.comp

@@ -56,10 +56,10 @@ void main() {
         y1 = int(ceil(fill.bbox.w * SY));
         break;
     }
-    x0 = clamp(x0, 0, WIDTH_IN_TILES);
-    y0 = clamp(y0, 0, HEIGHT_IN_TILES);
-    x1 = clamp(x1, 0, WIDTH_IN_TILES);
-    y1 = clamp(y1, 0, HEIGHT_IN_TILES);
+    x0 = clamp(x0, 0, int(conf.width_in_tiles));
+    y0 = clamp(y0, 0, int(conf.height_in_tiles));
+    x1 = clamp(x1, 0, int(conf.width_in_tiles));
+    y1 = clamp(y1, 0, int(conf.height_in_tiles));
 
     Path path;
     path.bbox = uvec4(x0, y0, x1, y1);

piet-gpu/src/lib.rs

@@ -222,8 +222,8 @@ impl Renderer {
         let state_buf = session.create_buffer(1 * 1024 * 1024, dev)?;
         let image_dev = session.create_image2d(WIDTH as u32, HEIGHT as u32, dev)?;
 
-        let mut config_buf_host = session.create_buffer(7*4, host)?;
-        let config_buf_dev = session.create_buffer(7*4, dev)?;
+        let mut config_buf_host = session.create_buffer(9*4, host)?;
+        let config_buf_dev = session.create_buffer(9*4, dev)?;
 
         // TODO: constants
         const PATH_SIZE: usize = 12;
@@ -241,7 +241,7 @@ impl Renderer {
         alloc += (n_pathseg * PATHSEG_SIZE + 3) & !3;
         let anno_base = alloc;
         alloc += (n_paths * ANNO_SIZE + 3) & !3;
-        config_buf_host.write(&[n_paths as u32, n_pathseg as u32, tile_base as u32, bin_base as u32, ptcl_base as u32, pathseg_base as u32, anno_base as u32])?;
+        config_buf_host.write(&[n_paths as u32, n_pathseg as u32, WIDTH_IN_TILES as u32, HEIGHT_IN_TILES as u32, tile_base as u32, bin_base as u32, ptcl_base as u32, pathseg_base as u32, anno_base as u32])?;
 
         let mut memory_buf_host = session.create_buffer(2*4, host)?;
         let memory_buf_dev = session.create_buffer(128 * 1024 * 1024, dev)?;