From c67696714b519f3c74957978caf0be72462ea2ac Mon Sep 17 00:00:00 2001
From: Elias Naur <mail@eliasnaur.com>
Date: Fri, 18 Dec 2020 00:29:25 +0100
Subject: [PATCH] coarse.comp: don't write Cmd_End to tiles out of bounds

If WIDTH_IN_TILES or HEIGHT_IN_TILES are not divisible by N_TILE_X or N_TILE_Y
respectively, the previously unconditional Cmd_End_write would write out of
bounds.

Signed-off-by: Elias Naur <mail@eliasnaur.com>
---
 piet-gpu/shader/coarse.comp |   4 +++-
 piet-gpu/shader/coarse.spv  | Bin 39376 -> 39668 bytes
 piet-gpu/src/lib.rs         |   2 +-
 3 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/piet-gpu/shader/coarse.comp b/piet-gpu/shader/coarse.comp
index a70318a..f835c2b 100644
--- a/piet-gpu/shader/coarse.comp
+++ b/piet-gpu/shader/coarse.comp
@@ -370,5 +370,7 @@ void main() {
         rd_ix += N_TILE;
         if (rd_ix >= ready_ix && partition_ix >= n_partitions) break;
     }
-    Cmd_End_write(cmd_ref);
+    if (bin_tile_x + tile_x < WIDTH_IN_TILES && bin_tile_y + tile_y < HEIGHT_IN_TILES) {
+        Cmd_End_write(cmd_ref);
+    }
 }
diff --git a/piet-gpu/shader/coarse.spv b/piet-gpu/shader/coarse.spv
index 260db69611077342550fa9a7e16b49fd67f0cc46..e0d89e5607670266959fadf3d08093111a327ae2 100644
GIT binary patch
delta 401
zcmYk1y)Hyi6ouD4!=1?_Q;UgCf&>qsKqC>o|KB^q)q8@GXh^gkz#EVgl|rFVl86V;
zs)?V<w=O4~<ea_sI%}`J_d^V?SD{r4!*wZBQw{B3wjSGAH$JzIvaF2d&}egTTz$$V
zZSfSu39|DbmGNvO=7<+22j+TKEb?6QGyPk!M1D(tsQ$`fncN!l7jO+~R>0=((Y>pt
zHL%YY9{yY9n)^_waPnS<yZP(Y(+0%dzh;}{7MV}Yw!!AT$|r`&*zryEYM>pk8^7RX
zy$A4M9by{Hnu2T2gBAJg&(_B%b|IfAo}s7<c0R?nP}&2Z63u&nWYCEx2lHd!bX7Ob

delta 107
zcmeyemFdD}CSGPHWp)MzMg}eh4u*)0yqBdILpDE@YGz~%-<&F=&c&(2z{aqUiGd+}
z^HR0VY(U9KU2zx94h9B>P&NjJt3Vk$24;p3ARENDWME|g3NeHM**!oyWb@i0cMSj-
CZ56Tr

diff --git a/piet-gpu/src/lib.rs b/piet-gpu/src/lib.rs
index 0272664..c1ab974 100644
--- a/piet-gpu/src/lib.rs
+++ b/piet-gpu/src/lib.rs
@@ -404,7 +404,7 @@ impl Renderer {
         cmd_buf.dispatch(
             &self.coarse_pipeline,
             &self.coarse_ds,
-            (WIDTH as u32 / 256, HEIGHT as u32 / 256, 1),
+            ((WIDTH as u32 + 255) / 256, (HEIGHT as u32 + 255) / 256, 1),
         );
         cmd_buf.write_timestamp(&query_pool, 6);
         cmd_buf.memory_barrier();