alex/vello
1
0
Fork 0
mirror of https://github.com/italicsjenga/vello.git synced 2025-01-10 20:51:29 +11:00
Code Issues Packages Projects Releases Wiki Activity

Create chunks over the x axis in addition to y axis

Browse source 
This allows more coalescing with image loads/stores, since all of our images are stored with a tiled layout.
This commit is contained in:
Tatsuyuki Ishi 2021-03-23 20:54:49 +09:00
parent f0127812eb
commit 4864a7fe0f
2 changed files with 20 additions and 11 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

31
piet-gpu/shader/kernel4.comp
View file

@ -13,9 +13,12 @@
#include "mem.h" #include "mem.h"
#include "setup.h" #include "setup.h"
#define CHUNK 8 #define CHUNK_X 2
#define CHUNK_DY (TILE_HEIGHT_PX / CHUNK) #define CHUNK_Y 4
layout(local_size_x = TILE_WIDTH_PX, local_size_y = CHUNK_DY) in; #define CHUNK CHUNK_X * CHUNK_Y
#define CHUNK_DX (TILE_WIDTH_PX / CHUNK_X)
#define CHUNK_DY (TILE_HEIGHT_PX / CHUNK_Y)
layout(local_size_x = CHUNK_DX, local_size_y = CHUNK_DY) in;
layout(set = 0, binding = 1) readonly buffer ConfigBuf { layout(set = 0, binding = 1) readonly buffer ConfigBuf {
Config conf; Config conf;
@ -84,10 +87,14 @@ uint packsRGB(vec4 rgba) {
return packUnorm4x8(rgba.wzyx); return packUnorm4x8(rgba.wzyx);
} }
uvec2 chunk_offset(uint i) {
return uvec2(i % CHUNK_X * CHUNK_DX, i / CHUNK_X * CHUNK_DY);
}
vec4[CHUNK] fillImage(uvec2 xy, CmdImage cmd_img) { vec4[CHUNK] fillImage(uvec2 xy, CmdImage cmd_img) {
vec4 rgba[CHUNK]; vec4 rgba[CHUNK];
for (uint i = 0; i < CHUNK; i++) { for (uint i = 0; i < CHUNK; i++) {
ivec2 uv = ivec2(xy.x, xy.y + i * CHUNK_DY) + cmd_img.offset; ivec2 uv = ivec2(xy + chunk_offset(i)) + cmd_img.offset;
#ifdef ENABLE_IMAGE_INDICES #ifdef ENABLE_IMAGE_INDICES
vec4 fg_rgba = imageLoad(images[cmd_img.index], uv); vec4 fg_rgba = imageLoad(images[cmd_img.index], uv);
#else #else
@ -108,7 +115,7 @@ void main() {
Alloc cmd_alloc = slice_mem(conf.ptcl_alloc, tile_ix * PTCL_INITIAL_ALLOC, PTCL_INITIAL_ALLOC); Alloc cmd_alloc = slice_mem(conf.ptcl_alloc, tile_ix * PTCL_INITIAL_ALLOC, PTCL_INITIAL_ALLOC);
CmdRef cmd_ref = CmdRef(cmd_alloc.offset); CmdRef cmd_ref = CmdRef(cmd_alloc.offset);
uvec2 xy_uint = uvec2(gl_GlobalInvocationID.x, gl_LocalInvocationID.y + TILE_HEIGHT_PX * gl_WorkGroupID.y); uvec2 xy_uint = uvec2(gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_WorkGroupID.x, gl_LocalInvocationID.y + TILE_HEIGHT_PX * gl_WorkGroupID.y);
vec2 xy = vec2(xy_uint); vec2 xy = vec2(xy_uint);
vec3 rgb[CHUNK]; vec3 rgb[CHUNK];
float mask[CHUNK]; float mask[CHUNK];
@ -120,7 +127,7 @@ void main() {
rgb[i] = vec3(0.5); rgb[i] = vec3(0.5);
#ifdef ENABLE_IMAGE_INDICES #ifdef ENABLE_IMAGE_INDICES
if (xy_uint.x < 1024 && xy_uint.y < 1024) { if (xy_uint.x < 1024 && xy_uint.y < 1024) {
rgb[i] = imageLoad(images[gl_WorkGroupID.x / 64], ivec2(xy_uint.x, xy_uint.y + CHUNK_DY * i)/4).rgb; rgb[i] = imageLoad(images[gl_WorkGroupID.x / 64], ivec2(xy_uint + chunk_offset(i))/4).rgb;
} }
#endif #endif
mask[i] = 1.0; mask[i] = 1.0;
@ -144,7 +151,7 @@ void main() {
vec2 line_vec = seg.vector; vec2 line_vec = seg.vector;
for (uint k = 0; k < CHUNK; k++) { for (uint k = 0; k < CHUNK; k++) {
vec2 dpos = xy + vec2(0.5, 0.5) - seg.origin; vec2 dpos = xy + vec2(0.5, 0.5) - seg.origin;
dpos.y += float(k * CHUNK_DY); dpos += vec2(chunk_offset(k));
float t = clamp(dot(line_vec, dpos) / dot(line_vec, line_vec), 0.0, 1.0); float t = clamp(dot(line_vec, dpos) / dot(line_vec, line_vec), 0.0, 1.0);
df[k] = min(df[k], length(line_vec * t - dpos)); df[k] = min(df[k], length(line_vec * t - dpos));
} }
@ -163,7 +170,7 @@ void main() {
do { do {
TileSeg seg = TileSeg_read(new_alloc(tile_seg_ref.offset, TileSeg_size), tile_seg_ref); TileSeg seg = TileSeg_read(new_alloc(tile_seg_ref.offset, TileSeg_size), tile_seg_ref);
for (uint k = 0; k < CHUNK; k++) { for (uint k = 0; k < CHUNK; k++) {
vec2 my_xy = vec2(xy.x, xy.y + float(k * CHUNK_DY)); vec2 my_xy = xy + vec2(chunk_offset(k));
vec2 start = seg.origin - my_xy; vec2 start = seg.origin - my_xy;
vec2 end = start + seg.vector; vec2 end = start + seg.vector;
vec2 window = clamp(vec2(start.y, end.y), 0.0, 1.0); vec2 window = clamp(vec2(start.y, end.y), 0.0, 1.0);
@ -227,7 +234,8 @@ void main() {
clip_tos = m.alloc; clip_tos = m.alloc;
uint base_ix = (clip_tos.offset >> 2) + gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_LocalInvocationID.y; uint base_ix = (clip_tos.offset >> 2) + gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_LocalInvocationID.y;
for (uint k = 0; k < CHUNK; k++) { for (uint k = 0; k < CHUNK; k++) {
write_mem(clip_tos, base_ix + k * TILE_WIDTH_PX * CHUNK_DY, blend_stack[blend_slot][k]); uvec2 offset = chunk_offset(k);
write_mem(clip_tos, base_ix + offset.x + offset.y * TILE_WIDTH_PX, blend_stack[blend_slot][k]);
} }
blend_spill++; blend_spill++;
} }
@ -242,7 +250,8 @@ void main() {
if (blend_sp == blend_spill) { if (blend_sp == blend_spill) {
uint base_ix = (clip_tos.offset >> 2) + gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_LocalInvocationID.y; uint base_ix = (clip_tos.offset >> 2) + gl_LocalInvocationID.x + TILE_WIDTH_PX * gl_LocalInvocationID.y;
for (uint k = 0; k < CHUNK; k++) { for (uint k = 0; k < CHUNK; k++) {
blend_stack[blend_slot][k] = read_mem(clip_tos, base_ix + k * TILE_WIDTH_PX * CHUNK_DY); uvec2 offset = chunk_offset(k);
blend_stack[blend_slot][k] = read_mem(clip_tos, base_ix + offset.x + offset.y * TILE_WIDTH_PX);
} }
clip_tos.offset = read_mem(clip_tos, (clip_tos.offset >> 2) + CLIP_LINK_OFFSET); clip_tos.offset = read_mem(clip_tos, (clip_tos.offset >> 2) + CLIP_LINK_OFFSET);
blend_spill--; blend_spill--;
@ -262,6 +271,6 @@ void main() {
} }
for (uint i = 0; i < CHUNK; i++) { for (uint i = 0; i < CHUNK; i++) {
imageStore(image, ivec2(xy_uint.x, xy_uint.y + CHUNK_DY * i), vec4(tosRGB(rgb[i]), 1.0)); imageStore(image, ivec2(xy_uint + chunk_offset(i)), vec4(tosRGB(rgb[i]), 1.0));
} }
} }

BIN
piet-gpu/shader/kernel4.spv
View file

Binary file not shown.