From ac3ac3ddff777ad4fd0ebcf9243137891fe29c0f Mon Sep 17 00:00:00 2001
From: Elias Naur <mail@eliasnaur.com>
Date: Sun, 13 Sep 2020 12:58:47 +0200
Subject: [PATCH] shader: introduce a crude setting for adjusting the maximum
 workgroup size

Both the Vulkan and OpenGL ES spec allow implementations to limit workgroups to
128 threads. Add a LG_WG_FACTOR setting for easy switching between 128 and 256
threads, with 256 being kept as the default setting.

Manually tested that LG_WG_FACTOR = 0 (128 threads) works as expected.

Signed-off-by: Elias Naur <mail@eliasnaur.com>
---
 piet-gpu/shader/backdrop.comp   |   2 +-
 piet-gpu/shader/coarse.comp     |   2 +-
 piet-gpu/shader/elements.spv    | Bin 52964 -> 52988 bytes
 piet-gpu/shader/setup.h         |   9 +++++++--
 piet-gpu/shader/tile_alloc.comp |   2 +-
 5 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/piet-gpu/shader/backdrop.comp b/piet-gpu/shader/backdrop.comp
index afe4d62..beba683 100644
--- a/piet-gpu/shader/backdrop.comp
+++ b/piet-gpu/shader/backdrop.comp
@@ -15,7 +15,7 @@
 
 #include "setup.h"
 
-#define LG_BACKDROP_WG 8
+#define LG_BACKDROP_WG (7 + LG_WG_FACTOR)
 #define BACKDROP_WG (1 << LG_BACKDROP_WG)
 
 layout(local_size_x = BACKDROP_WG, local_size_y = 1) in;
diff --git a/piet-gpu/shader/coarse.comp b/piet-gpu/shader/coarse.comp
index 56b3c07..57cbc8b 100644
--- a/piet-gpu/shader/coarse.comp
+++ b/piet-gpu/shader/coarse.comp
@@ -41,7 +41,7 @@ layout(set = 0, binding = 4) buffer PtclBuf {
 #include "tile.h"
 #include "ptcl.h"
 
-#define LG_N_PART_READ 8
+#define LG_N_PART_READ (7 + LG_WG_FACTOR)
 #define N_PART_READ (1 << LG_N_PART_READ)
 
 shared uint sh_elements[N_TILE];
diff --git a/piet-gpu/shader/elements.spv b/piet-gpu/shader/elements.spv
index 5bd0650bf9e65af7056d2d68ac333bd7d2dc744a..54c5cc0f645353bfc360465a3037aa6a2ef29745 100644
GIT binary patch
delta 982
zcmZ9K%}Z2K7{=f0ovTK=7}||SkOXN$;6_9gLTK$jAToj0z@#EYM8ZfoD@BCCh3TYy
zX_7XL*UWDrZPBt_+n`;m_6Gf)%RT7GaJbLMd*1V$_ss92e_HhG{i>!KGtX-IeJj`h
zX~?~-H5-2XsDE@|&;m=?JNQU?H2u&l;wu;U2l!wno5ha&wF*D55<ERLG~6E-spnor
zO~MY&c=)6-?|CQTzpMV`<ZJX+`nFoNEeB>NVX`?4qKKrnPFhXzRYb@0a7A={(lb*;
z$APAZj*r0=(Q)-tJjdlLp7<RuxEfg4bGz8ymq3`qs3ThvJP+n9yT*C}&QmOX0X)r;
zuLdQGir}ZL@A%#3M_hICFQF|K$`7vOmQWU1&rz#UUxB6l1W&};8@O~$s+Vz0;=Kh+
zS7*JZcvtK#gQdyS*u>>cu<t!sspK)jR3vn8bQR_*#tdI=S}FXv-;i4ewJ{czYog*E
z#JWk<NANwq(&g&~q`3=z0!!PcM#t-VWw>-#vjNu2s75`+RU`kGXLl`xv=Bg3RFf{g
zn`qMGBKr(hlfQ8jan-m(wxUUYLD^!tNpFLtouBj$T>AM*@4}@ko~A19CcO`Klhz+f
zT;4O19D<cf`}lLEBEe1i2xFEaZ8|9QWp3U03cAik$KyvohD&pol6M@gXW9q;1M?%5
Ac>n+a

delta 1011
zcmZ9KTT4`76o%J0j!lHYTvrH5P$2}}h=f7)2l@vxfyRJoND)ym70#6rfppPfI%jq?
z+a0E|Agzo3LAOD7U3C)$1wGGZzET-B`#r30?e|-2em9F>o5k*8NAH@5LX`P^nD73!
zq24S<j2}HRG%-99rRa>?@Wa)K>Y<37&#dD2;Ul%tQS97*u);6JV|aCJY<#F)WY5<t
zYC2=#KhKVBY!nJh*`H0tb(I_Fee~^%u9zuBEM=8#St*Z5>exlAv%HO%cmr<4#HR`o
zMob(SMoj!9+=z+W&v=RJ8&AH63*IY5wr@wXzutpz5u<@v0bd56<GI9R;T#{n0zS_3
z34eDKb8emfTC}U+?R0K+L=1A7F0^ZC>Sw5tUk9sw22ZxNo8_qIh6CEq!70pJV2u1j
zE>Uh%y*pqvLpUwD-Xz{#up#xFoP2;!lPPAw=XrZZ?-%M`0{@KZ9|zahA0q99deouH
z2gtaUBagrbc&qEX1hsVcIk4ISHR<k8;OePn9_(_gv4`9m{VxS(@D$<z9k5LXd1q+q
z?V@-Fw#oN)6S+0%BhSIBm+%5jJuTrSSZ(<dTIG6##_}b+g12}~>%iq{32(q@3I4_8
zdM60>7Hmk*`FE42CA<S)pvE(L|CTU)y4JJhJ<4WAaH{>^ScI#k>*_6LA8Y-?e*rnX
Bmze+n

diff --git a/piet-gpu/shader/setup.h b/piet-gpu/shader/setup.h
index 6c778ca..6bdde03 100644
--- a/piet-gpu/shader/setup.h
+++ b/piet-gpu/shader/setup.h
@@ -3,6 +3,11 @@
 // Much of this will be made dynamic in various ways, but for now it's easiest
 // to hardcode and keep all in one place.
 
+// A LG_WG_FACTOR of n scales workgroup sizes by 2^n. Use 0 for a
+// maximum workgroup size of 128, or 1 for a maximum size of 256.
+#define LG_WG_FACTOR 1
+#define WG_FACTOR (1<<LG_WG_FACTOR)
+
 // TODO: compute all these
 
 #define WIDTH_IN_TILES 128
@@ -17,7 +22,7 @@
 // of that kernel is equal to the number of bins, but should probably
 // be more flexible (it's 512 in the K&L paper).
 #define N_TILE_X 16
-#define N_TILE_Y 16
+#define N_TILE_Y (8 * WG_FACTOR)
 #define N_TILE (N_TILE_X * N_TILE_Y)
-#define LG_N_TILE 8
+#define LG_N_TILE (7 + LG_WG_FACTOR)
 #define N_SLICE (N_TILE / 32)
diff --git a/piet-gpu/shader/tile_alloc.comp b/piet-gpu/shader/tile_alloc.comp
index ee5d1db..b0c99a5 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 8
+#define LG_TILE_ALLOC_WG (7 + LG_WG_FACTOR)
 #define TILE_ALLOC_WG (1 << LG_TILE_ALLOC_WG)
 
 layout(local_size_x = TILE_ALLOC_WG, local_size_y = 1) in;