From 63ba45c774d55434bcb70f9c002de89febab7072 Mon Sep 17 00:00:00 2001
From: Raph Levien <raph.levien@gmail.com>
Date: Wed, 3 Jun 2020 15:32:58 -0700
Subject: [PATCH] Fix performance issues

Use larger workgroup for tile initialization (utilization was poor).
Provide correct element count to coarse rasterizer.
---
 piet-gpu/shader/tile_alloc.comp |   2 +-
 piet-gpu/shader/tile_alloc.spv  | Bin 8752 -> 8736 bytes
 piet-gpu/src/lib.rs             |   4 ++--
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/piet-gpu/shader/tile_alloc.comp b/piet-gpu/shader/tile_alloc.comp
index e1dff63..0d25274 100644
--- a/piet-gpu/shader/tile_alloc.comp
+++ b/piet-gpu/shader/tile_alloc.comp
@@ -5,7 +5,7 @@
 
 #include "setup.h"
 
-#define LG_TILE_ALLOC_WG 5
+#define LG_TILE_ALLOC_WG 8
 #define TILE_ALLOC_WG (1 << LG_TILE_ALLOC_WG)
 
 layout(local_size_x = TILE_ALLOC_WG, local_size_y = 1) in;
diff --git a/piet-gpu/shader/tile_alloc.spv b/piet-gpu/shader/tile_alloc.spv
index 204cc903c46ce5145057abaf6954277ec7c912ae..37276476010468de9292b5620d8e2973b61fb095 100644
GIT binary patch
delta 1298
zcmZ9KOHWf#6ovO~Z3}7Ygn>zoq9MwopkgiZ5TsgPC<urz5SvOI)Ksfb5*^hT6Gsm8
zCOR_l2N;9t*oi;Hi8FtI4-x|FYwt;}w%OTht^GLroVzy)9}B5yq$^?OS)2ZM-=vlz
z=6!jv7+cPGW~EA{x>8)Ll~!vO(|<r(tz5ZZTwA|aeri&7*`&0`rPb0yYw<Jx@R*&W
zrdlgiiYwLiN3A|n{=s;MH{Hm^v!1uyC@0=Wb*vvr&UkJ9T(Z;K_Map>dJm}mgvvFa
z?UQB1yXDVI+m`RQxcTzSjjze9S7%~U$CP$Ro@vAUScvacoXK6hQ=G{ir=H2feCj_w
z9q_!~;hkBdntI(n*ekI|)-L<0^0*rH33tmzg|jkpCM;@)<@@}BJ(+sH5Db-Tz{hnB
zs*ec2mH(t;KFc2!N2Akk<sTR3x2b$$gR+ydKNe|X*C#8DJ}=aVz9+=dG*#WWgO<Q@
zvaKNK3Woxn;dn#%Up41o4&g8tW2kV99M@-9n0j*gHt<%gZZ;yq&0X@SFi9_iX~%?l
zQ}pM7c1oB!VzatEJ2Ebe&V|^W)3P7Z7i1g^Cx>&=((=L|bbd3)Mj#N}FYT-_#T2p~
z@DV*;L|Q?(OP(5H?gcLh^MWj?yUaylba$Cc!f2Dqqj#NELv#g~g?Wu{!A-9SH?^RE
zRYY_7Gs1jZnz5E}KEqXcJ`w(xE;T1VFXKo2S}X3qN~h`z!qmE+*M*sDLo={<H-yPS
zXV>7fGJ0|b*wx+?W=6+v3Detsx!c0zpy&0VId*VS<e3iYk&(b<(A-WjI0xZf<0av6
i<2&MLo9f{n11<a*47{afAC?X2a!KXxw#N7Ld}0S(4T32E

delta 1310
zcmZ9KTT7Ht6o&UV<{`u@FM`4}q_oBkHd&f!T6VNd%gQtzCh5YMvOGXtO+>d{)IePY
z{eTK%enI_}?)m{81fl2AZ^P6KoAti$I_$Ofw?2%ljWpF{+Ur9|LIeN&>!z7ZTln%|
zx^}iJ387FdF3#tdR|-ojp_cz4w3UV8{rvLky@jU%Wm_0Vdt6v5JPcd>>_5CVoHu82
zrBKYzFRnh?sxw;NQ`ek~ReI`jNitVisDD?(wdG9XSW;V_XlzYd%TF4c`?8d;qwIc4
zG+^7{r@{0aS~&VUuouu;v0X%$v2&GAjk%=c1cvq#wij#n)x5ik@51jS(Qi2?iQIfA
zspg~q@v)Q-^Db{gsN|boZSOvS*vWB*AEY}^7Tm!(3C?kh84NXHA2wTV*^w<B1j$q#
zLq3qc#$oUqvOaRnXY3KUnw5SLp9A|_CtlxCPA}(o$ZX<R=NOuGUQoxbdf;kwRSM3)
zQiS82jno+vPo%od`G)d86!&5;@v`W%scN4wu}&Y@d}IAK@-1J9E+9@GgZse-y-d43
z1@=wVpQqXY*gSpXOz(~ig4G?!-5KJ1N1xQk#q!M<;&-B*fxqYeO{W{7U=zF1M!+VU
z=yu5ac)SeSIdD7H9DVTxFM@qRmoyG@39KH6xeQhtBF@?+S4}Y{TmkzU-_l91g4MEV
zf~%+&<FA4Jwia_O<$i{7te;5xlcBC-Z*ctSpWO17HbE?Q6VW1(-vZm<8qK(Xx537!
zyL9pij>Wx99CN3@HXHdJu(jhSoCZ%NyuXfNp0<YzX28#|og4$4N-Zvz6L^s}7QPFv
bE<6iYTc<{R6}9SDvEdE2zPIvqXTJU~kBWe!

diff --git a/piet-gpu/src/lib.rs b/piet-gpu/src/lib.rs
index b568827..63d80fa 100644
--- a/piet-gpu/src/lib.rs
+++ b/piet-gpu/src/lib.rs
@@ -247,7 +247,7 @@ impl<D: Device> Renderer<D> {
         let coarse_alloc_start = WIDTH_IN_TILES * HEIGHT_IN_TILES * PTCL_INITIAL_ALLOC;
         device.write_buffer(
             &coarse_alloc_buf_host,
-            &[n_elements as u32, coarse_alloc_start as u32],
+            &[n_paths as u32, coarse_alloc_start as u32],
         )?;
         let coarse_code = include_bytes!("../shader/coarse.spv");
         let coarse_pipeline = device.create_simple_compute_pipeline(coarse_code, 5, 0)?;
@@ -323,7 +323,7 @@ impl<D: Device> Renderer<D> {
         cmd_buf.dispatch(
             &self.tile_pipeline,
             &self.tile_ds,
-            (((self.n_paths + 31) / 32) as u32, 1, 1),
+            (((self.n_paths + 255) / 256) as u32, 1, 1),
         );
         cmd_buf.write_timestamp(&query_pool, 2);
         cmd_buf.memory_barrier();