[gpu] hb_gpu_paint: implement PUSH_GROUP / POP_GROUP compositing

Register push_group/pop_group paint callbacks, emit 1-texel
PUSH_GROUP / POP_GROUP ops, and handle them in all four paint
fragment shaders.

Each shader keeps an HB_GPU_PAINT_GROUP_DEPTH (=4) stack of vec4
accumulators.  PUSH_GROUP pushes the current `acc` and resets to
(0,0,0,0) so the enclosed subtree composites in isolation;
POP_GROUP pops the saved `acc` and composites the subtree result
into it using the mode stored in the op's aux slot (values match
hb_paint_composite_mode_t).

Implemented composite modes: CLEAR, SRC, DST, SRC_OVER (default),
DST_OVER, SRC_IN, DST_IN, SRC_OUT, DST_OUT, SRC_ATOP, DST_ATOP,
XOR, PLUS, SCREEN, DARKEN, LIGHTEN, MULTIPLY.  The CSS/PDF
separable blend modes (OVERLAY, COLOR_DODGE, COLOR_BURN,
HARD_LIGHT, SOFT_LIGHT, DIFFERENCE, EXCLUSION) and all four HSL
modes fall back to SRC_OVER for now.

Noto-COLRv1 happens not to emit any PaintComposite nodes, so this
doesn't change rendering for that font; fonts that do use it
(e.g. test_glyphs-glyf_colr_1.ttf) will pick this up.

Assisted-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Behdad Esfahbod committed 2026-04-13 21:45:22 -06:00
1 parent 51318f2608
commit e9f4be3ff2
9 files changed
+536 -8

No files matched your search

+70 -1
View File
@@ -169,6 +169,53 @@ static const char *hb_gpu_paint_fragment_glsl =
" return _hb_gpu_eval_stops (grad_base + 1, stop_count, t, foreground);\n"
"}\n"
"\n"
"/* Composite two premultiplied RGBA layers using one of the COLRv1\n"
" * compositing modes. Unsupported modes fall back to SRC_OVER.\n"
" * Values match hb_paint_composite_mode_t. */\n"
"vec4 _hb_gpu_composite (vec4 src, vec4 dst, int mode)\n"
"{\n"
" vec4 r = src + dst * (1.0 - src.a); /* SRC_OVER default */\n"
"\n"
" if (mode == 0) r = vec4 (0.0); /* CLEAR */\n"
" else if (mode == 1) r = src; /* SRC */\n"
" else if (mode == 2) r = dst; /* DST */\n"
" else if (mode == 4) r = dst + src * (1.0 - dst.a); /* DST_OVER */\n"
" else if (mode == 5) r = src * dst.a; /* SRC_IN */\n"
" else if (mode == 6) r = dst * src.a; /* DST_IN */\n"
" else if (mode == 7) r = src * (1.0 - dst.a); /* SRC_OUT */\n"
" else if (mode == 8) r = dst * (1.0 - src.a); /* DST_OUT */\n"
" else if (mode == 9) /* SRC_ATOP */\n"
" r = src * dst.a + dst * (1.0 - src.a);\n"
" else if (mode == 10) /* DST_ATOP */\n"
" r = dst * src.a + src * (1.0 - dst.a);\n"
" else if (mode == 11) /* XOR */\n"
" r = src * (1.0 - dst.a) + dst * (1.0 - src.a);\n"
" else if (mode == 12) /* PLUS */\n"
" r = min (src + dst, vec4 (1.0));\n"
" else if (mode == 13) { /* SCREEN (premul) */\n"
" r.rgb = src.rgb + dst.rgb - src.rgb * dst.rgb;\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" else if (mode == 15) { /* DARKEN */\n"
" r.rgb = min (src.rgb * dst.a, dst.rgb * src.a)\n"
" + src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" else if (mode == 16) { /* LIGHTEN */\n"
" r.rgb = max (src.rgb * dst.a, dst.rgb * src.a)\n"
" + src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" else if (mode == 23) { /* MULTIPLY (premul) */\n"
" r.rgb = src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a)\n"
" + src.rgb * dst.rgb;\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" /* SRC_OVER (3) and any unsupported mode use the default `r`. */\n"
"\n"
" return r;\n"
"}\n"
"\n"
"/* Walks the paint blob's flat op stream and returns a\n"
" * premultiplied RGBA coverage value for the current fragment.\n"
" *\n"
@@ -176,6 +223,8 @@ static const char *hb_gpu_paint_fragment_glsl =
" * foreground: caller-supplied foreground color, used when an op\n"
" * sets the is_foreground flag.\n"
" */\n"
"#define HB_GPU_PAINT_GROUP_DEPTH 4\n"
"\n"
"vec4 hb_gpu_paint (vec2 renderCoord, uint glyphLoc, vec4 foreground)\n"
"{\n"
" /* fwidth once, at uniform control flow: every per-layer\n"
@@ -190,6 +239,8 @@ static const char *hb_gpu_paint_fragment_glsl =
" int cursor = base + h2.r;\n"
"\n"
" vec4 acc = vec4 (0.0);\n"
" vec4 group_stack[HB_GPU_PAINT_GROUP_DEPTH];\n"
" int sp = 0;\n"
"\n"
" for (int i = 0; i < num_ops; i++)\n"
" {\n"
@@ -242,9 +293,27 @@ static const char *hb_gpu_paint_fragment_glsl =
"\n"
" cursor += 2;\n"
" }\n"
" else if (op_type == 2) /* PUSH_GROUP */\n"
" {\n"
" if (sp < HB_GPU_PAINT_GROUP_DEPTH) {\n"
" group_stack[sp] = acc;\n"
" sp++;\n"
" }\n"
" acc = vec4 (0.0);\n"
" cursor += 1;\n"
" }\n"
" else if (op_type == 3) /* POP_GROUP */\n"
" {\n"
" if (sp > 0) {\n"
" sp--;\n"
" vec4 src = acc;\n"
" vec4 dst = group_stack[sp];\n"
" acc = _hb_gpu_composite (src, dst, aux);\n"
" }\n"
" cursor += 1;\n"
" }\n"
" else\n"
" {\n"
" /* PUSH_GROUP, POP_GROUP not yet handled. */\n"
" break;\n"
" }\n"
" }\n"
+61 -1
View File
@@ -143,6 +143,46 @@ static const char *hb_gpu_paint_fragment_hlsl =
" return _hb_gpu_eval_stops (grad_base + 1, stop_count, t, foreground);\n"
"}\n"
"\n"
"float4 _hb_gpu_composite (float4 src, float4 dst, int mode)\n"
"{\n"
" float4 r = src + dst * (1.0 - src.a);\n"
"\n"
" if (mode == 0) r = float4 (0.0, 0.0, 0.0, 0.0);\n"
" else if (mode == 1) r = src;\n"
" else if (mode == 2) r = dst;\n"
" else if (mode == 4) r = dst + src * (1.0 - dst.a);\n"
" else if (mode == 5) r = src * dst.a;\n"
" else if (mode == 6) r = dst * src.a;\n"
" else if (mode == 7) r = src * (1.0 - dst.a);\n"
" else if (mode == 8) r = dst * (1.0 - src.a);\n"
" else if (mode == 9) r = src * dst.a + dst * (1.0 - src.a);\n"
" else if (mode == 10) r = dst * src.a + src * (1.0 - dst.a);\n"
" else if (mode == 11) r = src * (1.0 - dst.a) + dst * (1.0 - src.a);\n"
" else if (mode == 12) r = min (src + dst, float4 (1.0, 1.0, 1.0, 1.0));\n"
" else if (mode == 13) {\n"
" r.rgb = src.rgb + dst.rgb - src.rgb * dst.rgb;\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" else if (mode == 15) {\n"
" r.rgb = min (src.rgb * dst.a, dst.rgb * src.a)\n"
" + src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" else if (mode == 16) {\n"
" r.rgb = max (src.rgb * dst.a, dst.rgb * src.a)\n"
" + src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" else if (mode == 23) {\n"
" r.rgb = src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a)\n"
" + src.rgb * dst.rgb;\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" return r;\n"
"}\n"
"\n"
"#define HB_GPU_PAINT_GROUP_DEPTH 4\n"
"\n"
"float4 hb_gpu_paint (float2 renderCoord, uint glyphLoc_, float4 foreground)\n"
"{\n"
" /* fwidth once, at uniform control flow. */\n"
@@ -155,6 +195,8 @@ static const char *hb_gpu_paint_fragment_hlsl =
" int cursor = base + h2.r;\n"
"\n"
" float4 acc = float4 (0.0, 0.0, 0.0, 0.0);\n"
" float4 group_stack[HB_GPU_PAINT_GROUP_DEPTH];\n"
" int sp = 0;\n"
"\n"
" for (int i = 0; i < num_ops; i++)\n"
" {\n"
@@ -205,9 +247,27 @@ static const char *hb_gpu_paint_fragment_hlsl =
"\n"
" cursor += 2;\n"
" }\n"
" else if (op_type == 2)\n"
" {\n"
" if (sp < HB_GPU_PAINT_GROUP_DEPTH) {\n"
" group_stack[sp] = acc;\n"
" sp++;\n"
" }\n"
" acc = float4 (0.0, 0.0, 0.0, 0.0);\n"
" cursor += 1;\n"
" }\n"
" else if (op_type == 3)\n"
" {\n"
" if (sp > 0) {\n"
" sp--;\n"
" float4 src = acc;\n"
" float4 dst = group_stack[sp];\n"
" acc = _hb_gpu_composite (src, dst, aux);\n"
" }\n"
" cursor += 1;\n"
" }\n"
" else\n"
" {\n"
" /* PUSH_GROUP, POP_GROUP not yet handled. */\n"
" break;\n"
" }\n"
" }\n"
+61 -1
View File
@@ -153,6 +153,46 @@ static const char *hb_gpu_paint_fragment_msl =
" return _hb_gpu_eval_stops (hb_gpu_atlas, grad_base + 1, stop_count, t, foreground);\n"
"}\n"
"\n"
"static float4 _hb_gpu_composite (float4 src, float4 dst, int mode)\n"
"{\n"
" float4 r = src + dst * (1.0 - src.a); /* SRC_OVER default */\n"
"\n"
" if (mode == 0) r = float4 (0.0);\n"
" else if (mode == 1) r = src;\n"
" else if (mode == 2) r = dst;\n"
" else if (mode == 4) r = dst + src * (1.0 - dst.a);\n"
" else if (mode == 5) r = src * dst.a;\n"
" else if (mode == 6) r = dst * src.a;\n"
" else if (mode == 7) r = src * (1.0 - dst.a);\n"
" else if (mode == 8) r = dst * (1.0 - src.a);\n"
" else if (mode == 9) r = src * dst.a + dst * (1.0 - src.a);\n"
" else if (mode == 10) r = dst * src.a + src * (1.0 - dst.a);\n"
" else if (mode == 11) r = src * (1.0 - dst.a) + dst * (1.0 - src.a);\n"
" else if (mode == 12) r = min (src + dst, float4 (1.0));\n"
" else if (mode == 13) {\n"
" r.rgb = src.rgb + dst.rgb - src.rgb * dst.rgb;\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" else if (mode == 15) {\n"
" r.rgb = min (src.rgb * dst.a, dst.rgb * src.a)\n"
" + src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" else if (mode == 16) {\n"
" r.rgb = max (src.rgb * dst.a, dst.rgb * src.a)\n"
" + src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" else if (mode == 23) {\n"
" r.rgb = src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a)\n"
" + src.rgb * dst.rgb;\n"
" r.a = src.a + dst.a - src.a * dst.a;\n"
" }\n"
" return r;\n"
"}\n"
"\n"
"#define HB_GPU_PAINT_GROUP_DEPTH 4\n"
"\n"
"float4 hb_gpu_paint (float2 renderCoord, uint glyphLoc_, float4 foreground,\n"
" device const short4* hb_gpu_atlas)\n"
"{\n"
@@ -166,6 +206,8 @@ static const char *hb_gpu_paint_fragment_msl =
" int cursor = base + h2.r;\n"
"\n"
" float4 acc = float4 (0.0);\n"
" float4 group_stack[HB_GPU_PAINT_GROUP_DEPTH];\n"
" int sp = 0;\n"
"\n"
" for (int i = 0; i < num_ops; i++)\n"
" {\n"
@@ -219,9 +261,27 @@ static const char *hb_gpu_paint_fragment_msl =
"\n"
" cursor += 2;\n"
" }\n"
" else if (op_type == 2) /* PUSH_GROUP */\n"
" {\n"
" if (sp < HB_GPU_PAINT_GROUP_DEPTH) {\n"
" group_stack[sp] = acc;\n"
" sp++;\n"
" }\n"
" acc = float4 (0.0);\n"
" cursor += 1;\n"
" }\n"
" else if (op_type == 3) /* POP_GROUP */\n"
" {\n"
" if (sp > 0) {\n"
" sp--;\n"
" float4 src = acc;\n"
" float4 dst = group_stack[sp];\n"
" acc = _hb_gpu_composite (src, dst, aux);\n"
" }\n"
" cursor += 1;\n"
" }\n"
" else\n"
" {\n"
" /* PUSH_GROUP, POP_GROUP not yet handled. */\n"
" break;\n"
" }\n"
" }\n"
+57 -1
View File
@@ -155,6 +155,46 @@ static const char *hb_gpu_paint_fragment_wgsl =
" return _hb_gpu_eval_stops (hb_gpu_atlas, grad_base + 1, stop_count, t, foreground);\n"
"}\n"
"\n"
"fn _hb_gpu_composite (src: vec4f, dst: vec4f, mode: i32) -> vec4f\n"
"{\n"
" var r = src + dst * (1.0 - src.a);\n"
"\n"
" if (mode == 0) { r = vec4f (0.0); }\n"
" else if (mode == 1) { r = src; }\n"
" else if (mode == 2) { r = dst; }\n"
" else if (mode == 4) { r = dst + src * (1.0 - dst.a); }\n"
" else if (mode == 5) { r = src * dst.a; }\n"
" else if (mode == 6) { r = dst * src.a; }\n"
" else if (mode == 7) { r = src * (1.0 - dst.a); }\n"
" else if (mode == 8) { r = dst * (1.0 - src.a); }\n"
" else if (mode == 9) { r = src * dst.a + dst * (1.0 - src.a); }\n"
" else if (mode == 10) { r = dst * src.a + src * (1.0 - dst.a); }\n"
" else if (mode == 11) { r = src * (1.0 - dst.a) + dst * (1.0 - src.a); }\n"
" else if (mode == 12) { r = min (src + dst, vec4f (1.0)); }\n"
" else if (mode == 13) {\n"
" r = vec4f (src.rgb + dst.rgb - src.rgb * dst.rgb,\n"
" src.a + dst.a - src.a * dst.a);\n"
" }\n"
" else if (mode == 15) {\n"
" r = vec4f (min (src.rgb * dst.a, dst.rgb * src.a)\n"
" + src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a),\n"
" src.a + dst.a - src.a * dst.a);\n"
" }\n"
" else if (mode == 16) {\n"
" r = vec4f (max (src.rgb * dst.a, dst.rgb * src.a)\n"
" + src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a),\n"
" src.a + dst.a - src.a * dst.a);\n"
" }\n"
" else if (mode == 23) {\n"
" r = vec4f (src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a)\n"
" + src.rgb * dst.rgb,\n"
" src.a + dst.a - src.a * dst.a);\n"
" }\n"
" return r;\n"
"}\n"
"\n"
"const HB_GPU_PAINT_GROUP_DEPTH: i32 = 4;\n"
"\n"
"fn hb_gpu_paint (renderCoord: vec2f, glyphLoc_: u32, foreground: vec4f,\n"
" hb_gpu_atlas: ptr<storage, array<vec4<i32>>, read>) -> vec4f\n"
"{\n"
@@ -173,6 +213,8 @@ static const char *hb_gpu_paint_fragment_wgsl =
" var cursor = base + h2.r;\n"
"\n"
" var acc = vec4f (0.0);\n"
" var group_stack: array<vec4f, HB_GPU_PAINT_GROUP_DEPTH>;\n"
" var sp: i32 = 0;\n"
"\n"
" for (var i: i32 = 0; i < num_ops; i = i + 1)\n"
" {\n"
@@ -227,8 +269,22 @@ static const char *hb_gpu_paint_fragment_wgsl =
" acc = src + acc * (1.0 - src.a);\n"
"\n"
" cursor = cursor + 2;\n"
" } else if (op_type == 2) { // PUSH_GROUP\n"
" if (sp < HB_GPU_PAINT_GROUP_DEPTH) {\n"
" group_stack[sp] = acc;\n"
" sp = sp + 1;\n"
" }\n"
" acc = vec4f (0.0);\n"
" cursor = cursor + 1;\n"
" } else if (op_type == 3) { // POP_GROUP\n"
" if (sp > 0) {\n"
" sp = sp - 1;\n"
" let src = acc;\n"
" let dst = group_stack[sp];\n"
" acc = _hb_gpu_composite (src, dst, aux);\n"
" }\n"
" cursor = cursor + 1;\n"
" } else {\n"
" /* PUSH_GROUP, POP_GROUP not yet handled. */\n"
" break;\n"
" }\n"
" }\n"
+70 -1
View File
@@ -168,6 +168,53 @@ vec4 _hb_gpu_sample_sweep (vec2 renderCoord, int grad_base,
return _hb_gpu_eval_stops (grad_base + 1, stop_count, t, foreground);
}
/* Composite two premultiplied RGBA layers using one of the COLRv1
* compositing modes. Unsupported modes fall back to SRC_OVER.
* Values match hb_paint_composite_mode_t. */
vec4 _hb_gpu_composite (vec4 src, vec4 dst, int mode)
{
vec4 r = src + dst * (1.0 - src.a); /* SRC_OVER default */
if (mode == 0) r = vec4 (0.0); /* CLEAR */
else if (mode == 1) r = src; /* SRC */
else if (mode == 2) r = dst; /* DST */
else if (mode == 4) r = dst + src * (1.0 - dst.a); /* DST_OVER */
else if (mode == 5) r = src * dst.a; /* SRC_IN */
else if (mode == 6) r = dst * src.a; /* DST_IN */
else if (mode == 7) r = src * (1.0 - dst.a); /* SRC_OUT */
else if (mode == 8) r = dst * (1.0 - src.a); /* DST_OUT */
else if (mode == 9) /* SRC_ATOP */
r = src * dst.a + dst * (1.0 - src.a);
else if (mode == 10) /* DST_ATOP */
r = dst * src.a + src * (1.0 - dst.a);
else if (mode == 11) /* XOR */
r = src * (1.0 - dst.a) + dst * (1.0 - src.a);
else if (mode == 12) /* PLUS */
r = min (src + dst, vec4 (1.0));
else if (mode == 13) { /* SCREEN (premul) */
r.rgb = src.rgb + dst.rgb - src.rgb * dst.rgb;
r.a = src.a + dst.a - src.a * dst.a;
}
else if (mode == 15) { /* DARKEN */
r.rgb = min (src.rgb * dst.a, dst.rgb * src.a)
+ src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);
r.a = src.a + dst.a - src.a * dst.a;
}
else if (mode == 16) { /* LIGHTEN */
r.rgb = max (src.rgb * dst.a, dst.rgb * src.a)
+ src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);
r.a = src.a + dst.a - src.a * dst.a;
}
else if (mode == 23) { /* MULTIPLY (premul) */
r.rgb = src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a)
+ src.rgb * dst.rgb;
r.a = src.a + dst.a - src.a * dst.a;
}
/* SRC_OVER (3) and any unsupported mode use the default `r`. */
return r;
}
/* Walks the paint blob's flat op stream and returns a
* premultiplied RGBA coverage value for the current fragment.
*
@@ -175,6 +222,8 @@ vec4 _hb_gpu_sample_sweep (vec2 renderCoord, int grad_base,
* foreground: caller-supplied foreground color, used when an op
* sets the is_foreground flag.
*/
#define HB_GPU_PAINT_GROUP_DEPTH 4
vec4 hb_gpu_paint (vec2 renderCoord, uint glyphLoc, vec4 foreground)
{
/* fwidth once, at uniform control flow: every per-layer
@@ -189,6 +238,8 @@ vec4 hb_gpu_paint (vec2 renderCoord, uint glyphLoc, vec4 foreground)
int cursor = base + h2.r;
vec4 acc = vec4 (0.0);
vec4 group_stack[HB_GPU_PAINT_GROUP_DEPTH];
int sp = 0;
for (int i = 0; i < num_ops; i++)
{
@@ -241,9 +292,27 @@ vec4 hb_gpu_paint (vec2 renderCoord, uint glyphLoc, vec4 foreground)
cursor += 2;
}
else if (op_type == 2) /* PUSH_GROUP */
{
if (sp < HB_GPU_PAINT_GROUP_DEPTH) {
group_stack[sp] = acc;
sp++;
}
acc = vec4 (0.0);
cursor += 1;
}
else if (op_type == 3) /* POP_GROUP */
{
if (sp > 0) {
sp--;
vec4 src = acc;
vec4 dst = group_stack[sp];
acc = _hb_gpu_composite (src, dst, aux);
}
cursor += 1;
}
else
{
/* PUSH_GROUP, POP_GROUP not yet handled. */
break;
}
}
+61 -1
View File
@@ -142,6 +142,46 @@ float4 _hb_gpu_sample_sweep (float2 renderCoord, int grad_base,
return _hb_gpu_eval_stops (grad_base + 1, stop_count, t, foreground);
}
float4 _hb_gpu_composite (float4 src, float4 dst, int mode)
{
float4 r = src + dst * (1.0 - src.a);
if (mode == 0) r = float4 (0.0, 0.0, 0.0, 0.0);
else if (mode == 1) r = src;
else if (mode == 2) r = dst;
else if (mode == 4) r = dst + src * (1.0 - dst.a);
else if (mode == 5) r = src * dst.a;
else if (mode == 6) r = dst * src.a;
else if (mode == 7) r = src * (1.0 - dst.a);
else if (mode == 8) r = dst * (1.0 - src.a);
else if (mode == 9) r = src * dst.a + dst * (1.0 - src.a);
else if (mode == 10) r = dst * src.a + src * (1.0 - dst.a);
else if (mode == 11) r = src * (1.0 - dst.a) + dst * (1.0 - src.a);
else if (mode == 12) r = min (src + dst, float4 (1.0, 1.0, 1.0, 1.0));
else if (mode == 13) {
r.rgb = src.rgb + dst.rgb - src.rgb * dst.rgb;
r.a = src.a + dst.a - src.a * dst.a;
}
else if (mode == 15) {
r.rgb = min (src.rgb * dst.a, dst.rgb * src.a)
+ src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);
r.a = src.a + dst.a - src.a * dst.a;
}
else if (mode == 16) {
r.rgb = max (src.rgb * dst.a, dst.rgb * src.a)
+ src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);
r.a = src.a + dst.a - src.a * dst.a;
}
else if (mode == 23) {
r.rgb = src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a)
+ src.rgb * dst.rgb;
r.a = src.a + dst.a - src.a * dst.a;
}
return r;
}
#define HB_GPU_PAINT_GROUP_DEPTH 4
float4 hb_gpu_paint (float2 renderCoord, uint glyphLoc_, float4 foreground)
{
/* fwidth once, at uniform control flow. */
@@ -154,6 +194,8 @@ float4 hb_gpu_paint (float2 renderCoord, uint glyphLoc_, float4 foreground)
int cursor = base + h2.r;
float4 acc = float4 (0.0, 0.0, 0.0, 0.0);
float4 group_stack[HB_GPU_PAINT_GROUP_DEPTH];
int sp = 0;
for (int i = 0; i < num_ops; i++)
{
@@ -204,9 +246,27 @@ float4 hb_gpu_paint (float2 renderCoord, uint glyphLoc_, float4 foreground)
cursor += 2;
}
else if (op_type == 2)
{
if (sp < HB_GPU_PAINT_GROUP_DEPTH) {
group_stack[sp] = acc;
sp++;
}
acc = float4 (0.0, 0.0, 0.0, 0.0);
cursor += 1;
}
else if (op_type == 3)
{
if (sp > 0) {
sp--;
float4 src = acc;
float4 dst = group_stack[sp];
acc = _hb_gpu_composite (src, dst, aux);
}
cursor += 1;
}
else
{
/* PUSH_GROUP, POP_GROUP not yet handled. */
break;
}
}
+61 -1
View File
@@ -152,6 +152,46 @@ static float4 _hb_gpu_sample_sweep (float2 renderCoord, int grad_base,
return _hb_gpu_eval_stops (hb_gpu_atlas, grad_base + 1, stop_count, t, foreground);
}
static float4 _hb_gpu_composite (float4 src, float4 dst, int mode)
{
float4 r = src + dst * (1.0 - src.a); /* SRC_OVER default */
if (mode == 0) r = float4 (0.0);
else if (mode == 1) r = src;
else if (mode == 2) r = dst;
else if (mode == 4) r = dst + src * (1.0 - dst.a);
else if (mode == 5) r = src * dst.a;
else if (mode == 6) r = dst * src.a;
else if (mode == 7) r = src * (1.0 - dst.a);
else if (mode == 8) r = dst * (1.0 - src.a);
else if (mode == 9) r = src * dst.a + dst * (1.0 - src.a);
else if (mode == 10) r = dst * src.a + src * (1.0 - dst.a);
else if (mode == 11) r = src * (1.0 - dst.a) + dst * (1.0 - src.a);
else if (mode == 12) r = min (src + dst, float4 (1.0));
else if (mode == 13) {
r.rgb = src.rgb + dst.rgb - src.rgb * dst.rgb;
r.a = src.a + dst.a - src.a * dst.a;
}
else if (mode == 15) {
r.rgb = min (src.rgb * dst.a, dst.rgb * src.a)
+ src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);
r.a = src.a + dst.a - src.a * dst.a;
}
else if (mode == 16) {
r.rgb = max (src.rgb * dst.a, dst.rgb * src.a)
+ src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a);
r.a = src.a + dst.a - src.a * dst.a;
}
else if (mode == 23) {
r.rgb = src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a)
+ src.rgb * dst.rgb;
r.a = src.a + dst.a - src.a * dst.a;
}
return r;
}
#define HB_GPU_PAINT_GROUP_DEPTH 4
float4 hb_gpu_paint (float2 renderCoord, uint glyphLoc_, float4 foreground,
device const short4* hb_gpu_atlas)
{
@@ -165,6 +205,8 @@ float4 hb_gpu_paint (float2 renderCoord, uint glyphLoc_, float4 foreground,
int cursor = base + h2.r;
float4 acc = float4 (0.0);
float4 group_stack[HB_GPU_PAINT_GROUP_DEPTH];
int sp = 0;
for (int i = 0; i < num_ops; i++)
{
@@ -218,9 +260,27 @@ float4 hb_gpu_paint (float2 renderCoord, uint glyphLoc_, float4 foreground,
cursor += 2;
}
else if (op_type == 2) /* PUSH_GROUP */
{
if (sp < HB_GPU_PAINT_GROUP_DEPTH) {
group_stack[sp] = acc;
sp++;
}
acc = float4 (0.0);
cursor += 1;
}
else if (op_type == 3) /* POP_GROUP */
{
if (sp > 0) {
sp--;
float4 src = acc;
float4 dst = group_stack[sp];
acc = _hb_gpu_composite (src, dst, aux);
}
cursor += 1;
}
else
{
/* PUSH_GROUP, POP_GROUP not yet handled. */
break;
}
}
+57 -1
View File
@@ -154,6 +154,46 @@ fn _hb_gpu_sample_sweep (renderCoord: vec2f, grad_base: i32,
return _hb_gpu_eval_stops (hb_gpu_atlas, grad_base + 1, stop_count, t, foreground);
}
fn _hb_gpu_composite (src: vec4f, dst: vec4f, mode: i32) -> vec4f
{
var r = src + dst * (1.0 - src.a);
if (mode == 0) { r = vec4f (0.0); }
else if (mode == 1) { r = src; }
else if (mode == 2) { r = dst; }
else if (mode == 4) { r = dst + src * (1.0 - dst.a); }
else if (mode == 5) { r = src * dst.a; }
else if (mode == 6) { r = dst * src.a; }
else if (mode == 7) { r = src * (1.0 - dst.a); }
else if (mode == 8) { r = dst * (1.0 - src.a); }
else if (mode == 9) { r = src * dst.a + dst * (1.0 - src.a); }
else if (mode == 10) { r = dst * src.a + src * (1.0 - dst.a); }
else if (mode == 11) { r = src * (1.0 - dst.a) + dst * (1.0 - src.a); }
else if (mode == 12) { r = min (src + dst, vec4f (1.0)); }
else if (mode == 13) {
r = vec4f (src.rgb + dst.rgb - src.rgb * dst.rgb,
src.a + dst.a - src.a * dst.a);
}
else if (mode == 15) {
r = vec4f (min (src.rgb * dst.a, dst.rgb * src.a)
+ src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a),
src.a + dst.a - src.a * dst.a);
}
else if (mode == 16) {
r = vec4f (max (src.rgb * dst.a, dst.rgb * src.a)
+ src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a),
src.a + dst.a - src.a * dst.a);
}
else if (mode == 23) {
r = vec4f (src.rgb * (1.0 - dst.a) + dst.rgb * (1.0 - src.a)
+ src.rgb * dst.rgb,
src.a + dst.a - src.a * dst.a);
}
return r;
}
const HB_GPU_PAINT_GROUP_DEPTH: i32 = 4;
fn hb_gpu_paint (renderCoord: vec2f, glyphLoc_: u32, foreground: vec4f,
hb_gpu_atlas: ptr<storage, array<vec4<i32>>, read>) -> vec4f
{
@@ -172,6 +212,8 @@ fn hb_gpu_paint (renderCoord: vec2f, glyphLoc_: u32, foreground: vec4f,
var cursor = base + h2.r;
var acc = vec4f (0.0);
var group_stack: array<vec4f, HB_GPU_PAINT_GROUP_DEPTH>;
var sp: i32 = 0;
for (var i: i32 = 0; i < num_ops; i = i + 1)
{
@@ -226,8 +268,22 @@ fn hb_gpu_paint (renderCoord: vec2f, glyphLoc_: u32, foreground: vec4f,
acc = src + acc * (1.0 - src.a);
cursor = cursor + 2;
} else if (op_type == 2) { // PUSH_GROUP
if (sp < HB_GPU_PAINT_GROUP_DEPTH) {
group_stack[sp] = acc;
sp = sp + 1;
}
acc = vec4f (0.0);
cursor = cursor + 1;
} else if (op_type == 3) { // POP_GROUP
if (sp > 0) {
sp = sp - 1;
let src = acc;
let dst = group_stack[sp];
acc = _hb_gpu_composite (src, dst, aux);
}
cursor = cursor + 1;
} else {
/* PUSH_GROUP, POP_GROUP not yet handled. */
break;
}
}
+38
View File
@@ -81,6 +81,42 @@ hb_gpu_paint_pop_clip (hb_paint_funcs_t *funcs HB_UNUSED,
c->pending_clip = false;
}
/* Emit a 1-texel control op. Used for PUSH_GROUP / POP_GROUP. */
static void
emit_control_op (hb_gpu_paint_t *c, int16_t op_type, int16_t aux)
{
if (unlikely (!c->ops.resize (c->ops.length + 4)))
{
c->unsupported = true;
return;
}
int16_t *o = &c->ops.arrayZ[c->ops.length - 4];
o[0] = op_type;
o[1] = aux;
o[2] = 0;
o[3] = 0;
c->num_ops++;
}
static void
hb_gpu_paint_push_group (hb_paint_funcs_t *funcs HB_UNUSED,
void *paint_data,
void *user_data HB_UNUSED)
{
emit_control_op ((hb_gpu_paint_t *) paint_data,
HB_GPU_PAINT_OP_PUSH_GROUP, 0);
}
static void
hb_gpu_paint_pop_group (hb_paint_funcs_t *funcs HB_UNUSED,
void *paint_data,
hb_paint_composite_mode_t mode,
void *user_data HB_UNUSED)
{
emit_control_op ((hb_gpu_paint_t *) paint_data,
HB_GPU_PAINT_OP_POP_GROUP, (int16_t) mode);
}
/* Quantize unsigned byte 0..255 to signed Q15 0..32767. */
static inline int16_t
color_to_q15 (unsigned byte_value)
@@ -576,6 +612,8 @@ static struct hb_gpu_paint_funcs_lazy_loader_t
hb_paint_funcs_set_push_clip_glyph_func (funcs, hb_gpu_paint_push_clip_glyph, nullptr, nullptr);
hb_paint_funcs_set_pop_clip_func (funcs, hb_gpu_paint_pop_clip, nullptr, nullptr);
hb_paint_funcs_set_push_group_func (funcs, hb_gpu_paint_push_group, nullptr, nullptr);
hb_paint_funcs_set_pop_group_func (funcs, hb_gpu_paint_pop_group, nullptr, nullptr);
hb_paint_funcs_set_color_func (funcs, hb_gpu_paint_color, nullptr, nullptr);
hb_paint_funcs_set_linear_gradient_func (funcs, hb_gpu_paint_linear_gradient, nullptr, nullptr);
hb_paint_funcs_set_radial_gradient_func (funcs, hb_gpu_paint_radial_gradient, nullptr, nullptr);