gpu,op/clip: [compute] get rid of stroke vs fill flags

The fill mode is now controlled by a SetFillMode command, not by flags
on each path segment and fill command.

Signed-off-by: Elias Naur <mail@eliasnaur.com>
This commit is contained in:
Elias Naur
2021-03-19 18:40:31 +01:00
parent 60bab15164
commit 13da40f601
17 changed files with 584 additions and 764 deletions
+75 -69
View File
@@ -63,6 +63,11 @@ uint state_flag_index(uint partition_ix) {
#define FLAG_SET_LINEWIDTH 1
#define FLAG_SET_BBOX 2
#define FLAG_RESET_BBOX 4
#define FLAG_SET_FILL_MODE 8
// Fill modes take up the next bit. Non-zero fill is 0, stroke is 1.
#define LG_FILL_MODE 4
#define FILL_MODE_BITS 1
#define FILL_MODE_MASK (FILL_MODE_BITS << LG_FILL_MODE)
// This is almost like a monoid (the interaction between transformation and
// bounding boxes is approximate)
@@ -88,8 +93,11 @@ State combine_state(State a, State b) {
c.translate.x = a.mat.x * b.translate.x + a.mat.z * b.translate.y + a.translate.x;
c.translate.y = a.mat.y * b.translate.x + a.mat.w * b.translate.y + a.translate.y;
c.linewidth = (b.flags & FLAG_SET_LINEWIDTH) == 0 ? a.linewidth : b.linewidth;
c.flags = (a.flags & (FLAG_SET_LINEWIDTH | FLAG_SET_BBOX)) | b.flags;
c.flags = (a.flags & (FLAG_SET_LINEWIDTH | FLAG_SET_BBOX | FLAG_SET_FILL_MODE)) | b.flags;
c.flags |= (a.flags & FLAG_RESET_BBOX) >> 1;
uint fill_mode = (b.flags & FLAG_SET_FILL_MODE) == 0 ? a.flags : b.flags;
fill_mode &= FILL_MODE_MASK;
c.flags = (c.flags & ~FILL_MODE_MASK) | fill_mode;
c.path_count = a.path_count + b.path_count;
c.pathseg_count = a.pathseg_count + b.pathseg_count;
c.trans_count = a.trans_count + b.trans_count;
@@ -99,7 +107,7 @@ State combine_state(State a, State b) {
State map_element(ElementRef ref) {
// TODO: it would *probably* be more efficient to make the memory read patterns less
// divergent, though it would be more wasted memory.
uint tag = Element_tag(ref);
uint tag = Element_tag(ref).tag;
State c;
c.bbox = vec4(0.0, 0.0, 0.0, 0.0);
c.mat = vec4(1.0, 0.0, 0.0, 1.0);
@@ -110,30 +118,26 @@ State map_element(ElementRef ref) {
c.pathseg_count = 0;
c.trans_count = 0;
switch (tag) {
case Element_FillLine:
case Element_StrokeLine:
LineSeg line = Element_FillLine_read(ref);
case Element_Line:
LineSeg line = Element_Line_read(ref);
c.bbox.xy = min(line.p0, line.p1);
c.bbox.zw = max(line.p0, line.p1);
c.pathseg_count = 1;
break;
case Element_FillQuad:
case Element_StrokeQuad:
QuadSeg quad = Element_FillQuad_read(ref);
case Element_Quad:
QuadSeg quad = Element_Quad_read(ref);
c.bbox.xy = min(min(quad.p0, quad.p1), quad.p2);
c.bbox.zw = max(max(quad.p0, quad.p1), quad.p2);
c.pathseg_count = 1;
break;
case Element_FillCubic:
case Element_StrokeCubic:
CubicSeg cubic = Element_FillCubic_read(ref);
case Element_Cubic:
CubicSeg cubic = Element_Cubic_read(ref);
c.bbox.xy = min(min(cubic.p0, cubic.p1), min(cubic.p2, cubic.p3));
c.bbox.zw = max(max(cubic.p0, cubic.p1), max(cubic.p2, cubic.p3));
c.pathseg_count = 1;
break;
case Element_Fill:
case Element_FillColor:
case Element_FillImage:
case Element_Stroke:
case Element_BeginClip:
c.flags = FLAG_RESET_BBOX;
c.path_count = 1;
@@ -152,6 +156,10 @@ State map_element(ElementRef ref) {
c.translate = t.translate;
c.trans_count = 1;
break;
case Element_SetFillMode:
SetFillMode fm = Element_SetFillMode_read(ref);
c.flags = FLAG_SET_FILL_MODE | (fm.fill_mode << LG_FILL_MODE);
break;
}
return c;
}
@@ -291,111 +299,109 @@ void main() {
// gains to be had from stashing in shared memory or possibly
// registers (though register pressure is an issue).
ElementRef this_ref = Element_index(ref, i);
uint tag = Element_tag(this_ref);
switch (tag) {
case Element_FillLine:
case Element_StrokeLine:
LineSeg line = Element_StrokeLine_read(this_ref);
PathStrokeCubic path_cubic;
ElementTag tag = Element_tag(this_ref);
uint fill_mode = fill_mode_from_flags(st.flags >> LG_FILL_MODE);
bool is_stroke = fill_mode == MODE_STROKE;
switch (tag.tag) {
case Element_Line:
LineSeg line = Element_Line_read(this_ref);
PathCubic path_cubic;
path_cubic.p0 = line.p0;
path_cubic.p1 = mix(line.p0, line.p1, 1.0 / 3.0);
path_cubic.p2 = mix(line.p1, line.p0, 1.0 / 3.0);
path_cubic.p3 = line.p1;
path_cubic.path_ix = st.path_count;
path_cubic.trans_ix = st.trans_count;
if (tag == Element_StrokeLine) {
if (is_stroke) {
path_cubic.stroke = get_linewidth(st);
} else {
path_cubic.stroke = vec2(0.0);
}
// We do encoding a bit by hand to minimize divergence. Another approach
// would be to have a fill/stroke bool.
PathSegRef path_out_ref = PathSegRef(conf.pathseg_alloc.offset + (st.pathseg_count - 1) * PathSeg_size);
uint out_tag = tag == Element_FillLine ? PathSeg_FillCubic : PathSeg_StrokeCubic;
write_mem(conf.pathseg_alloc, path_out_ref.offset >> 2, out_tag);
PathStrokeCubic_write(conf.pathseg_alloc, PathStrokeCubicRef(path_out_ref.offset + 4), path_cubic);
PathSeg_Cubic_write(conf.pathseg_alloc, path_out_ref, fill_mode, path_cubic);
break;
case Element_FillQuad:
case Element_StrokeQuad:
QuadSeg quad = Element_StrokeQuad_read(this_ref);
case Element_Quad:
QuadSeg quad = Element_Quad_read(this_ref);
path_cubic.p0 = quad.p0;
path_cubic.p1 = mix(quad.p1, quad.p0, 1.0 / 3.0);
path_cubic.p2 = mix(quad.p1, quad.p2, 1.0 / 3.0);
path_cubic.p3 = quad.p2;
path_cubic.path_ix = st.path_count;
path_cubic.trans_ix = st.trans_count;
if (tag == Element_StrokeQuad) {
if (is_stroke) {
path_cubic.stroke = get_linewidth(st);
} else {
path_cubic.stroke = vec2(0.0);
}
// We do encoding a bit by hand to minimize divergence. Another approach
// would be to have a fill/stroke bool.
path_out_ref = PathSegRef(conf.pathseg_alloc.offset + (st.pathseg_count - 1) * PathSeg_size);
out_tag = tag == Element_FillQuad ? PathSeg_FillCubic : PathSeg_StrokeCubic;
write_mem(conf.pathseg_alloc, path_out_ref.offset >> 2, out_tag);
PathStrokeCubic_write(conf.pathseg_alloc, PathStrokeCubicRef(path_out_ref.offset + 4), path_cubic);
PathSeg_Cubic_write(conf.pathseg_alloc, path_out_ref, fill_mode, path_cubic);
break;
case Element_FillCubic:
case Element_StrokeCubic:
CubicSeg cubic = Element_StrokeCubic_read(this_ref);
case Element_Cubic:
CubicSeg cubic = Element_Cubic_read(this_ref);
path_cubic.p0 = cubic.p0;
path_cubic.p1 = cubic.p1;
path_cubic.p2 = cubic.p2;
path_cubic.p3 = cubic.p3;
path_cubic.path_ix = st.path_count;
path_cubic.trans_ix = st.trans_count;
if (tag == Element_StrokeCubic) {
if (is_stroke) {
path_cubic.stroke = get_linewidth(st);
} else {
path_cubic.stroke = vec2(0.0);
}
// We do encoding a bit by hand to minimize divergence. Another approach
// would be to have a fill/stroke bool.
path_out_ref = PathSegRef(conf.pathseg_alloc.offset + (st.pathseg_count - 1) * PathSeg_size);
out_tag = tag == Element_FillCubic ? PathSeg_FillCubic : PathSeg_StrokeCubic;
write_mem(conf.pathseg_alloc, path_out_ref.offset >> 2, out_tag);
PathStrokeCubic_write(conf.pathseg_alloc, PathStrokeCubicRef(path_out_ref.offset + 4), path_cubic);
PathSeg_Cubic_write(conf.pathseg_alloc, path_out_ref, fill_mode, path_cubic);
break;
case Element_Stroke:
Stroke stroke = Element_Stroke_read(this_ref);
AnnoStroke anno_stroke;
anno_stroke.rgba_color = stroke.rgba_color;
vec2 lw = get_linewidth(st);
anno_stroke.bbox = st.bbox + vec4(-lw, lw);
anno_stroke.linewidth = st.linewidth * sqrt(abs(st.mat.x * st.mat.w - st.mat.y * st.mat.z));
AnnotatedRef out_ref = AnnotatedRef(conf.anno_alloc.offset + (st.path_count - 1) * Annotated_size);
Annotated_Stroke_write(conf.anno_alloc, out_ref, anno_stroke);
break;
case Element_Fill:
Fill fill = Element_Fill_read(this_ref);
AnnoFill anno_fill;
case Element_FillColor:
FillColor fill = Element_FillColor_read(this_ref);
AnnoColor anno_fill;
anno_fill.rgba_color = fill.rgba_color;
anno_fill.bbox = st.bbox;
out_ref = AnnotatedRef(conf.anno_alloc.offset + (st.path_count - 1) * Annotated_size);
Annotated_Fill_write(conf.anno_alloc, out_ref, anno_fill);
if (is_stroke) {
vec2 lw = get_linewidth(st);
anno_fill.bbox = st.bbox + vec4(-lw, lw);
anno_fill.linewidth = st.linewidth * sqrt(abs(st.mat.x * st.mat.w - st.mat.y * st.mat.z));
} else {
anno_fill.bbox = st.bbox;
anno_fill.linewidth = 0.0;
}
AnnotatedRef out_ref = AnnotatedRef(conf.anno_alloc.offset + (st.path_count - 1) * Annotated_size);
Annotated_Color_write(conf.anno_alloc, out_ref, fill_mode, anno_fill);
break;
case Element_FillImage:
FillImage fill_img = Element_FillImage_read(this_ref);
AnnoFillImage anno_fill_img;
anno_fill_img.index = fill_img.index;
anno_fill_img.offset = fill_img.offset;
anno_fill_img.bbox = st.bbox;
AnnoImage anno_img;
anno_img.index = fill_img.index;
anno_img.offset = fill_img.offset;
if (is_stroke) {
vec2 lw = get_linewidth(st);
anno_img.bbox = st.bbox + vec4(-lw, lw);
anno_img.linewidth = st.linewidth * sqrt(abs(st.mat.x * st.mat.w - st.mat.y * st.mat.z));
} else {
anno_img.bbox = st.bbox;
anno_img.linewidth = 0.0;
}
out_ref = AnnotatedRef(conf.anno_alloc.offset + (st.path_count - 1) * Annotated_size);
Annotated_FillImage_write(conf.anno_alloc, out_ref, anno_fill_img);
Annotated_Image_write(conf.anno_alloc, out_ref, fill_mode, anno_img);
break;
case Element_BeginClip:
Clip begin_clip = Element_BeginClip_read(this_ref);
AnnoClip anno_begin_clip = AnnoClip(begin_clip.bbox);
// This is the absolute bbox, it's been transformed during encoding.
anno_begin_clip.bbox = begin_clip.bbox;
AnnoBeginClip anno_begin_clip;
if (is_stroke) {
vec2 lw = get_linewidth(st);
// This is the absolute bbox, it's been transformed during encoding.
anno_begin_clip.bbox = begin_clip.bbox + vec4(-lw, lw);
anno_begin_clip.linewidth = st.linewidth * sqrt(abs(st.mat.x * st.mat.w - st.mat.y * st.mat.z));
} else {
anno_begin_clip.bbox = begin_clip.bbox;
anno_fill.linewidth = 0.0;
}
out_ref = AnnotatedRef(conf.anno_alloc.offset + (st.path_count - 1) * Annotated_size);
Annotated_BeginClip_write(conf.anno_alloc, out_ref, anno_begin_clip);
Annotated_BeginClip_write(conf.anno_alloc, out_ref, fill_mode, anno_begin_clip);
break;
case Element_EndClip:
Clip end_clip = Element_EndClip_read(this_ref);
// This bbox is expected to be the same as the begin one.
AnnoClip anno_end_clip = AnnoClip(end_clip.bbox);
AnnoEndClip anno_end_clip = AnnoEndClip(end_clip.bbox);
out_ref = AnnotatedRef(conf.anno_alloc.offset + (st.path_count - 1) * Annotated_size);
Annotated_EndClip_write(conf.anno_alloc, out_ref, anno_end_clip);
break;