alex/vello
1
0
Fork 0
mirror of https://github.com/italicsjenga/vello.git synced 2025-01-10 12:41:30 +11:00
Code Issues Packages Projects Releases Wiki Activity

elements.comp: use shared array of structs directly

Browse source 
The NVIDIA shader compiler bug that forced splitting of the state struct
into primitive types is now fixed.

Signed-off-by: Elias Naur <mail@eliasnaur.com>
This commit is contained in:
Elias Naur 2021-03-13 15:06:19 +01:00
parent fd746ea7a6
commit ad444f615c
3 changed files with 5 additions and 42 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

47
piet-gpu/shader/elements.comp
View file

@ -157,16 +157,7 @@ vec2 get_linewidth(State st) {
return 0.5 * st.linewidth * vec2(length(st.mat.xz), length(st.mat.yw)); return 0.5 * st.linewidth * vec2(length(st.mat.xz), length(st.mat.yw));
} }
// We should be able to use an array of structs but the NV shader compiler shared State sh_state[WG_SIZE];
// doesn't seem to like it :/
//shared State sh_state[WG_SIZE];
shared vec4 sh_mat[WG_SIZE];
shared vec2 sh_translate[WG_SIZE];
shared vec4 sh_bbox[WG_SIZE];
shared float sh_width[WG_SIZE];
shared uint sh_flags[WG_SIZE];
shared uint sh_path_count[WG_SIZE];
shared uint sh_pathseg_count[WG_SIZE];
shared uint sh_part_ix; shared uint sh_part_ix;
shared State sh_prefix; shared State sh_prefix;
@ -195,35 +186,15 @@ void main() {
th_state[i] = combine_state(th_state[i - 1], map_element(Element_index(ref, i))); th_state[i] = combine_state(th_state[i - 1], map_element(Element_index(ref, i)));
} }
State agg = th_state[N_ROWS - 1]; State agg = th_state[N_ROWS - 1];
sh_mat[gl_LocalInvocationID.x] = agg.mat; sh_state[gl_LocalInvocationID.x] = agg;
sh_translate[gl_LocalInvocationID.x] = agg.translate;
sh_bbox[gl_LocalInvocationID.x] = agg.bbox;
sh_width[gl_LocalInvocationID.x] = agg.linewidth;
sh_flags[gl_LocalInvocationID.x] = agg.flags;
sh_path_count[gl_LocalInvocationID.x] = agg.path_count;
sh_pathseg_count[gl_LocalInvocationID.x] = agg.pathseg_count;
for (uint i = 0; i < LG_WG_SIZE; i++) { for (uint i = 0; i < LG_WG_SIZE; i++) {
barrier(); barrier();
if (gl_LocalInvocationID.x >= (1 << i)) { if (gl_LocalInvocationID.x >= (1 << i)) {
State other; State other = sh_state[gl_LocalInvocationID.x - (1 << i)];
uint ix = gl_LocalInvocationID.x - (1 << i);
other.mat = sh_mat[ix];
other.translate = sh_translate[ix];
other.bbox = sh_bbox[ix];
other.linewidth = sh_width[ix];
other.flags = sh_flags[ix];
other.path_count = sh_path_count[ix];
other.pathseg_count = sh_pathseg_count[ix];
agg = combine_state(other, agg); agg = combine_state(other, agg);
} }
barrier(); barrier();
sh_mat[gl_LocalInvocationID.x] = agg.mat; sh_state[gl_LocalInvocationID.x] = agg;
sh_translate[gl_LocalInvocationID.x] = agg.translate;
sh_bbox[gl_LocalInvocationID.x] = agg.bbox;
sh_width[gl_LocalInvocationID.x] = agg.linewidth;
sh_flags[gl_LocalInvocationID.x] = agg.flags;
sh_path_count[gl_LocalInvocationID.x] = agg.path_count;
sh_pathseg_count[gl_LocalInvocationID.x] = agg.pathseg_count;
} }
State exclusive; State exclusive;
@ -304,15 +275,7 @@ void main() {
State row = exclusive; State row = exclusive;
if (gl_LocalInvocationID.x > 0) { if (gl_LocalInvocationID.x > 0) {
uint ix = gl_LocalInvocationID.x - 1; State other = sh_state[gl_LocalInvocationID.x - 1];
State other;
other.mat = sh_mat[ix];
other.translate = sh_translate[ix];
other.bbox = sh_bbox[ix];
other.linewidth = sh_width[ix];
other.flags = sh_flags[ix];
other.path_count = sh_path_count[ix];
other.pathseg_count = sh_pathseg_count[ix];
row = combine_state(row, other); row = combine_state(row, other);
} }
for (uint i = 0; i < N_ROWS; i++) { for (uint i = 0; i < N_ROWS; i++) {

BIN
piet-gpu/shader/elements.spv
View file

Binary file not shown.

BIN
piet-gpu/shader/path_coarse.spv
View file

Binary file not shown.