diff --git a/apps/benchmark/App.tsx b/apps/benchmark/App.tsx
index c974c3f..27228b5 100644
--- a/apps/benchmark/App.tsx
+++ b/apps/benchmark/App.tsx
@@ -2,7 +2,7 @@
// whole matrix by itself and prints one `[bench] tick ...` line per second to
// logcat; bench/run.mjs pairs those with host-side CPU and PSS samples.
import React, { useEffect, useMemo, useRef, useState } from 'react';
-import { Linking, StyleSheet, Text as RNText, View } from 'react-native';
+import { Linking, PixelRatio, StyleSheet, Text as RNText, View } from 'react-native';
import {
Easing,
cancelAnimation,
@@ -22,11 +22,13 @@ import {
Rect as TgfxRect,
Shader as TgfxShader,
Text as TgfxText,
+ Vertices as TgfxVertices,
TgfxImage,
renderToImage,
} from 'react-native-tgfx';
import {
Canvas as SkCanvas,
+ Circle as SkCircle,
Fill as SkFill,
Group as SkGroup,
Image as SkImageView,
@@ -47,7 +49,10 @@ import { BG, CX, CY, H, PALETTE, SHAPES, W } from './layout';
// react-native-skia 3.x has one canvas, on Graphite.
const ENGINES = ['tgfx', 'skia'] as const;
-const SCENES = ['shapes', 'images', 'shader', 'pattern', 'paths', 'text', ...ANIM_SCENES] as const;
+// `shapes` pits each engine's fastest static path against the other: tgfx nodes,
+// skia's ShapeBlob. `shapes-nodes` and `shapes-atlas` hold the technique equal:
+// 20k declarative nodes on both, or one sprite atlas and one tinted mesh on both.
+const SCENES = ['shapes', 'shapes-nodes', 'shapes-atlas', 'images', 'shader', 'pattern', 'paths', 'text', ...ANIM_SCENES] as const;
// surface: tgfx composite="layer", skia backingView SurfaceView.
// texture: tgfx composite="texture", skia backingView TextureView (its default).
const VIEWS = ['surface', 'texture'] as const;
@@ -198,6 +203,47 @@ function tgfxTextureScene(i: number, size: number) {
);
}
+// `shapes-atlas` for tgfx: ShapeBlob's technique through the public API. One
+// white circle and one white rect, rasterized once at the device scale, side by
+// side in a two-cell atlas; every shape is a quad tinted by its vertex colour.
+const ATLAS_PX = SHAPES[0].size * PixelRatio.get();
+const ATLAS_CELL = Math.ceil(ATLAS_PX) + 2;
+
+function tgfxAtlasScene() {
+ const r = ATLAS_PX / 2;
+ return (
+ <>
+
+
+ >
+ );
+}
+
+// Two triangles per shape, padded by one device pixel on each side so the
+// atlas cell's antialiased edge is inside the quad.
+let atlasMesh: { vertices: { x: number; y: number }[]; textures: { x: number; y: number }[]; colors: string[] } | null = null;
+function tgfxAtlasMesh() {
+ if (atlasMesh != null) return atlasMesh;
+ const pad = 1 / PixelRatio.get();
+ const vertices: { x: number; y: number }[] = [];
+ const textures: { x: number; y: number }[] = [];
+ const colors: string[] = [];
+ for (const s of SHAPES) {
+ const x0 = s.x - pad;
+ const y0 = s.y - pad;
+ const x1 = s.x + s.size + pad;
+ const y1 = s.y + s.size + pad;
+ const u0 = s.circle ? 0 : ATLAS_CELL;
+ const u1 = u0 + ATLAS_PX + 2;
+ const v1 = ATLAS_PX + 2;
+ vertices.push({ x: x0, y: y0 }, { x: x1, y: y0 }, { x: x1, y: y1 }, { x: x0, y: y0 }, { x: x1, y: y1 }, { x: x0, y: y1 });
+ textures.push({ x: u0, y: 0 }, { x: u1, y: 0 }, { x: u1, y: v1 }, { x: u0, y: 0 }, { x: u1, y: v1 }, { x: u0, y: v1 });
+ for (let k = 0; k < 6; k += 1) colors.push(s.color);
+ }
+ atlasMesh = { vertices, textures, colors };
+ return atlasMesh;
+}
+
// ---- scenes ----------------------------------------------------------------
function TgfxScene({
@@ -205,14 +251,27 @@ function TgfxScene({
t,
images,
tile,
+ atlas,
}: {
scene: string;
t: ReturnType>;
images: TgfxImage[];
tile: TgfxImage | null;
+ atlas: TgfxImage | null;
}) {
const origin = { x: CX, y: CY };
- if (scene === 'shapes') {
+ if (scene === 'shapes-atlas') {
+ if (atlas == null) return null;
+ const mesh = tgfxAtlasMesh();
+ return (
+
+
+
+
+
+ );
+ }
+ if (scene === 'shapes' || scene === 'shapes-nodes') {
return (
{SHAPES.map((s, i) =>
@@ -291,6 +350,19 @@ function SkiaScene({
const patternTransform = useDerivedValue(() => [{ rotate: t.value }, { scale: 0.03 }]);
const uniforms = useDerivedValue(() => ({ uTime: t.value, uSize: [W, H] }));
const origin = vec(CX, CY);
+ if (scene === 'shapes-nodes') {
+ return (
+
+ {SHAPES.map((s, i) =>
+ s.circle ? (
+
+ ) : (
+
+ )
+ )}
+
+ );
+ }
if (scene === 'images') {
if (images.length < 8) return null;
return (
@@ -358,6 +430,7 @@ export default function App() {
const [combo, setCombo] = useState(0);
const [tgfxImages, setTgfxImages] = useState([]);
const [tgfxTile, setTgfxTile] = useState(null);
+ const [tgfxAtlas, setTgfxAtlas] = useState(null);
const [plan, setPlan] = useState(null);
const [gapMs, setGapMs] = useState(0);
// anim-imp for tgfx: which thread writes the 20k handles (bench://run?...&imp=js).
@@ -396,9 +469,12 @@ export default function App() {
built.push(await renderToImage(tgfxTextureScene(i, 128), { width: 128, height: 128 }));
}
const builtTile = await renderToImage(tgfxTextureScene(2, 64), { width: 64, height: 64 });
+ // Here, not in the scene: renderToImage needs no canvas mounted around it.
+ const builtAtlas = await renderToImage(tgfxAtlasScene(), { width: ATLAS_CELL * 2, height: ATLAS_CELL });
if (!alive) return;
setTgfxImages(built);
setTgfxTile(builtTile);
+ setTgfxAtlas(builtAtlas);
})();
return () => {
alive = false;
@@ -519,7 +595,7 @@ export default function App() {
{isAnim(scene) ? (
) : (
-
+
)}
)
@@ -533,7 +609,7 @@ export default function App() {
deviceIndependentFonts={scene === 'text' && dif}
>
- {scene === 'shapes' ? (
+ {scene === 'shapes' || scene === 'shapes-atlas' ? (
) : isAnim(scene) ? (
diff --git a/apps/benchmark/README.md b/apps/benchmark/README.md
index 8dd9711..9505db7 100644
--- a/apps/benchmark/README.md
+++ b/apps/benchmark/README.md
@@ -12,12 +12,21 @@ Static scenes: 20,000 shapes, 3,600 images, a 40-octave full-screen shader, a
repeating image pattern, 10,000 cubic paths (half stroked) and 6,000 texts, each
under one animated rotation driven by a shared value. `paths` runs any other
path count (`paths900` is the scene the tables before 2026-11-10 used; both
-engines sit at 120 fps on it on a current flagship).
+engines sit at 120 fps on it on a current flagship). The default matrix runs
+`paths900`, `paths2500`, `paths5000` and `paths` (10,000), so a table shows
+where each engine falls off.
Skia draws the static shapes through `Skia.ShapeBlob` + `drawShapes` (3.10),
recorded once into a picture: its explicit batched-shape API, the counterpart of
the batching tgfx does internally for plain ``/`` nodes. The
-unbatched Skia path is the `shapes-picture` row.
+unbatched Skia path is the `shapes-picture` row. Two more rows hold the
+technique equal:
+
+| scene | tgfx | skia |
+|---|---|---|
+| `shapes` | 20,000 ``/`` nodes | one ShapeBlob |
+| `shapes-nodes` | 20,000 ``/`` nodes | 20,000 ``/`` nodes |
+| `shapes-atlas` | a white circle and rect rasterized once into an atlas, 20,000 tinted quads in one `` | one ShapeBlob (the same technique, built in) |
Per-shape scenes (`AnimScenes.tsx`): every shape orbits its own home on its own
phase and speed, so no transform can be shared and tgfx's draw reordering
@@ -56,7 +65,7 @@ app and the rows stop.
| flag | default | |
|---|---|---|
| `--engines` | `tgfx,skia` | engine order; `skia,tgfx` runs skia first |
-| `--scenes` | all nine | subset, e.g. `paths,anim-imp`, or `paths2500` for another path count |
+| `--scenes` | all fourteen | subset, e.g. `paths,anim-imp`, or `paths7500` for another path count |
| `--views` | `surface,texture` | outer loop of the matrix |
| `--skip` | none | `engine:scene` pairs to leave out, e.g. `tgfx:anim-imp` |
| `--order` | `sequential` | `alternate` runs every engine per scene |
diff --git a/apps/benchmark/bench/results-2026-11-11-samsung-90hz-skia310-raw.md b/apps/benchmark/bench/results-2026-11-11-samsung-90hz-skia310-raw.md
new file mode 100644
index 0000000..15f0b74
--- /dev/null
+++ b/apps/benchmark/bench/results-2026-11-11-samsung-90hz-skia310-raw.md
@@ -0,0 +1,135 @@
+# Raw: Samsung SM-E146B, 90 Hz, react-native-skia 3.10.0
+
+`node bench/run.mjs --order alternate --reps 2 --gap 10 --cool 0.95 --cool-timeout 240 --skip tgfx:anim-imp --views `, once per view.
+
+## SurfaceView
+
+| engine · scene · view | Present FPS | UI FPS | JS FPS | CPU % (1 core = 100) | PSS MB | thermal | CPU cap % | ticks |
+|---|---|---|---|---|---|---|---|---|
+| tgfx · shapes · surface | 61.3 | 90 | 90 | 189 | 389.5 | 2 | 68 | 53 |
+| skia · shapes · surface | 89.4 | 91 | 91 | 70 | 322.4 | 2 | 100 | 53 |
+| tgfx · shapes-nodes · surface | 62.3 | 91 | 90 | 184 | 389.6 | 2 | 80 | 53 |
+| skia · shapes-nodes · surface | 26.1 | 91 | 90 | 95 | 454.1 | 2 | 100 | 53 |
+| tgfx · shapes-atlas · surface | 90.0 | 91 | 91 | 48 | 369.3 | 2 | 100 | 52 |
+| skia · shapes-atlas · surface | 89.4 | 91 | 91 | 70 | 320.7 | 2 | 100 | 52 |
+| tgfx · images · surface | 90.1 | 91 | 91 | 78 | 335.7 | 2 | 100 | 53 |
+| skia · images · surface | 88.1 | 91 | 91 | 111 | 344.7 | 2 | 92 | 53 |
+| tgfx · shader · surface | 15.0 | 91 | 91 | 25 | 329.4 | 2 | 100 | 53 |
+| skia · shader · surface | 14.8 | 91 | 91 | 28 | 305.6 | 2 | 100 | 53 |
+| tgfx · pattern · surface | 90.1 | 91 | 91 | 46 | 325.9 | 1 | 100 | 53 |
+| skia · pattern · surface | 89.6 | 91 | 91 | 58 | 300.7 | 1 | 100 | 53 |
+| tgfx · paths900 · surface | 90.3 | 91 | 91 | 76 | 348.6 | 0 | 100 | 53 |
+| skia · paths900 · surface | 27.2 | 91 | 90 | 52 | 405.5 | 0 | 100 | 53 |
+| tgfx · paths2500 · surface | 82.6 | 91 | 90 | 113 | 380.4 | 0 | 98 | 52 |
+| skia · paths2500 · surface | 27.1 | 91 | 90 | 60 | 413.6 | 0 | 100 | 53 |
+| tgfx · paths5000 · surface | 43.8 | 91 | 90 | 111 | 489.8 | 0 | 100 | 53 |
+| skia · paths5000 · surface | 17.8 | 91 | 90 | 53 | 430.1 | 0 | 100 | 53 |
+| tgfx · paths · surface | 8.8 | 91 | 90 | 167 | 521.9 | 0 | 100 | 53 |
+| skia · paths · surface | 10.3 | 91 | 91 | 47 | 453.7 | 0 | 100 | 53 |
+| tgfx · text · surface | 73.0 | 91 | 90 | 128 | 362.0 | 0 | 100 | 53 |
+| skia · text · surface | 32.4 | 91 | 91 | 122 | 402.5 | 0 | 100 | 53 |
+| tgfx · anim-decl · surface | 24.6 | 34 | 17 | 224 | 500.1 | 0 | 92 | 51 |
+| skia · anim-decl · surface | 32.3 | 32 | 16 | 143 | 474.6 | 0 | 98 | 51 |
+| skia · anim-imp · surface | 54.9 | 79 | 40 | 162 | 358.6 | 0 | 95 | 51 |
+| tgfx · shapes-picture · surface | 83.1 | 91 | 90 | 145 | 392.3 | 0 | 98 | 53 |
+| skia · shapes-picture · surface | 26.1 | 91 | 90 | 84 | 420.4 | 0 | 100 | 53 |
+
+`thermal` is the highest Android thermal status (0 none … 3 severe) seen at the start of the combo's runs. `CPU cap %` is the lowest scaling_max_freq / cpuinfo_max_freq seen during them: below 100 the kernel was throttling, whatever the status says.
+
+### Where the CPU and memory go
+
+| engine · scene · view | UI thread % | JS thread % | busiest other threads % | Graphics MB | Native MB | Java MB |
+|---|---|---|---|---|---|---|
+| tgfx · shapes · surface | 7 | 2 | TgfxRender 150, nitro-thread 18, binder 6 | 161.1 | 142.2 | 4.0 |
+| skia · shapes · surface | 8 | 2 | RNSkia Present 24, RNSkia Render 16, binder 7 | 134.9 | 106.5 | 4.8 |
+| tgfx · shapes-nodes · surface | 7 | 2 | TgfxRender 148, nitro-thread 18, binder 6 | 161.1 | 141.7 | 4.4 |
+| skia · shapes-nodes · surface | 6 | 1 | RNSkia Render 57, RNSkia Present 22, binder 4 | 239.4 | 118.5 | 4.6 |
+| tgfx · shapes-atlas · surface | 10 | 2 | TgfxRender 17, binder 7, mali-cmar-backe 7 | 149.4 | 124.9 | 4.4 |
+| skia · shapes-atlas · surface | 8 | 2 | RNSkia Present 23, RNSkia Render 16, mali-cmar-backe 7 | 134.9 | 105.9 | 4.8 |
+| tgfx · images · surface | 7 | 1 | TgfxRender 54, binder 5, mali-cmar-backe 5 | 143.4 | 111.7 | 4.4 |
+| skia · images · surface | 6 | 1 | RNSkia Render 58, RNSkia Present 37, binder 5 | 150.4 | 110.8 | 4.8 |
+| tgfx · shader · surface | 12 | 3 | binder 4, TgfxRender 4, mali-cmar-backe 1 | 145.1 | 104.7 | 4.4 |
+| skia · shader · surface | 12 | 3 | RNSkia Present 5, binder 3, mali-cmar-backe 1 | 119.8 | 103.5 | 4.5 |
+| tgfx · pattern · surface | 10 | 2 | TgfxRender 15, binder 7, mali-cmar-backe 6 | 143.3 | 104.3 | 4.4 |
+| skia · pattern · surface | 9 | 2 | RNSkia Present 24, binder 7, mali-cmar-backe 7 | 118.1 | 102.1 | 4.8 |
+| tgfx · paths900 · surface | 7 | 1 | TgfxRender 53, binder 5, mali-cmar-backe 4 | 158.0 | 112.3 | 4.4 |
+| skia · paths900 · surface | 12 | 3 | RNSkia Render 13, RNSkia Present 11, mali-cmar-backe 5 | 219.4 | 103.6 | 4.6 |
+| tgfx · paths2500 · surface | 6 | 1 | TgfxRender 95, binder 6, mali-cmar-backe 3 | 175.8 | 123.2 | 4.4 |
+| skia · paths2500 · surface | 11 | 3 | RNSkia Render 27, RNSkia Present 9, mali-cmar-backe 4 | 224.4 | 104.6 | 4.6 |
+| tgfx · paths5000 · surface | 6 | 1 | TgfxRender 98, binder 5, mali-cmar-backe 2 | 259.1 | 148.6 | 4.4 |
+| skia · paths5000 · surface | 11 | 3 | RNSkia Render 25, RNSkia Present 6, mali-cmar-backe 4 | 230.5 | 109.5 | 4.5 |
+| tgfx · paths · surface | 6 | 1 | TgfxRender 144, nitro-thread 13, binder 4 | 241.5 | 186.6 | 4.4 |
+| skia · paths · surface | 11 | 3 | RNSkia Render 22, RNSkia Present 4, mali-cmar-backe 4 | 251.4 | 109.7 | 4.5 |
+| tgfx · text · surface | 6 | 1 | TgfxRender 101, nitro-thread 11, binder 5 | 151.5 | 128.8 | 4.4 |
+| skia · text · surface | 5 | 1 | RNSkia Render 93, RNSkia Present 17, binder 4 | 195.3 | 118.3 | 4.6 |
+| tgfx · anim-decl · surface | 97 | 0 | TgfxRender 98, nitro-thread 15, binder 4 | 152.6 | 189.0 | 4.4 |
+| skia · anim-decl · surface | 98 | 0 | RNSkia Render 27, RNSkia Present 15, binder 4 | 141.3 | 177.1 | 4.5 |
+| skia · anim-imp · surface | 97 | 1 | RNSkia Render 33, RNSkia Present 26, binder 6 | 153.9 | 123.5 | 4.6 |
+| tgfx · shapes-picture · surface | 6 | 1 | TgfxRender 121, nitro-thread 13, binder 4 | 161.1 | 144.5 | 4.4 |
+| skia · shapes-picture · surface | 7 | 2 | RNSkia Render 44, RNSkia Present 24, binder 4 | 230.1 | 113.0 | 4.5 |
+
+## TextureView
+
+| engine · scene · view | Present FPS | UI FPS | JS FPS | CPU % (1 core = 100) | PSS MB | thermal | CPU cap % | ticks |
+|---|---|---|---|---|---|---|---|---|
+| tgfx · shapes · texture | 56.4 | 90 | 90 | 202 | 421.2 | 2 | 80 | 53 |
+| skia · shapes · texture | 89.1 | 91 | 91 | 87 | 337.7 | 2 | 100 | 53 |
+| tgfx · shapes-nodes · texture | 54.6 | 90 | 90 | 200 | 424.5 | 2 | 80 | 53 |
+| skia · shapes-nodes · texture | 25.4 | 91 | 90 | 101 | 491.6 | 2 | 98 | 52 |
+| tgfx · shapes-atlas · texture | 88.1 | 91 | 91 | 76 | 401.4 | 2 | 100 | 53 |
+| skia · shapes-atlas · texture | 89.1 | 91 | 91 | 88 | 338.6 | 2 | 98 | 53 |
+| tgfx · images · texture | 88.8 | 91 | 91 | 99 | 368.0 | 2 | 98 | 53 |
+| skia · images · texture | 85.1 | 91 | 90 | 134 | 364.6 | 2 | 84 | 53 |
+| tgfx · shader · texture | 12.6 | 45 | 44 | 22 | 364.7 | 2 | 100 | 51 |
+| skia · shader · texture | 14.3 | 33 | 31 | 23 | 326.0 | 2 | 100 | 51 |
+| tgfx · pattern · texture | 89.0 | 91 | 91 | 77 | 358.6 | 2 | 100 | 53 |
+| skia · pattern · texture | 88.9 | 91 | 91 | 77 | 318.5 | 2 | 76 | 53 |
+| tgfx · paths900 · texture | 88.4 | 91 | 90 | 95 | 378.9 | 2 | 92 | 53 |
+| skia · paths900 · texture | 26.6 | 66 | 65 | 52 | 426.3 | 2 | 100 | 51 |
+| tgfx · paths2500 · texture | 72.5 | 91 | 90 | 132 | 472.1 | 2 | 80 | 53 |
+| skia · paths2500 · texture | 26.3 | 68 | 66 | 62 | 436.2 | 2 | 100 | 51 |
+| tgfx · paths5000 · texture | 41.4 | 91 | 90 | 121 | 515.5 | 2 | 80 | 53 |
+| skia · paths5000 · texture | 17.6 | 72 | 70 | 53 | 448.2 | 2 | 100 | 51 |
+| tgfx · paths · texture | 4.8 | 90 | 90 | 204 | 545.9 | 2 | 80 | 53 |
+| skia · paths · texture | 10.1 | 79 | 78 | 47 | 477.6 | 2 | 100 | 52 |
+| tgfx · text · texture | 65.4 | 90 | 90 | 146 | 392.5 | 2 | 80 | 53 |
+| skia · text · texture | 30.3 | 88 | 86 | 131 | 420.0 | 2 | 92 | 51 |
+| tgfx · anim-decl · texture | 11.6 | 30 | 16 | 216 | 530.5 | 2 | 72 | 51 |
+| skia · anim-decl · texture | 9.4 | 32 | 16 | 116 | 484.1 | 2 | 88 | 51 |
+| skia · anim-imp · texture | 21.9 | 76 | 39 | 129 | 374.6 | 2 | 88 | 52 |
+| tgfx · shapes-picture · texture | 69.3 | 90 | 90 | 171 | 425.2 | 2 | 88 | 53 |
+| skia · shapes-picture · texture | 25.3 | 91 | 90 | 91 | 463.4 | 2 | 100 | 52 |
+
+`thermal` is the highest Android thermal status (0 none … 3 severe) seen at the start of the combo's runs. `CPU cap %` is the lowest scaling_max_freq / cpuinfo_max_freq seen during them: below 100 the kernel was throttling, whatever the status says.
+
+### Where the CPU and memory go
+
+| engine · scene · view | UI thread % | JS thread % | busiest other threads % | Graphics MB | Native MB | Java MB |
+|---|---|---|---|---|---|---|
+| tgfx · shapes · texture | 10 | 2 | TgfxRender 148, nitro-thread 19, RenderThread 11 | 195.3 | 142.0 | 4.4 |
+| skia · shapes · texture | 11 | 2 | RenderThread 20, RNSkia Present 17, RNSkia Render 15 | 149.9 | 106.8 | 4.7 |
+| tgfx · shapes-nodes · texture | 10 | 2 | TgfxRender 144, nitro-thread 18, RenderThread 11 | 195.3 | 144.3 | 4.4 |
+| skia · shapes-nodes · texture | 8 | 1 | RNSkia Render 57, RNSkia Present 23, mali-cmar-backe 4 | 277.3 | 119.7 | 4.5 |
+| tgfx · shapes-atlas · texture | 13 | 2 | RenderThread 23, TgfxRender 15, mali-cmar-backe 13 | 183.6 | 125.4 | 4.4 |
+| skia · shapes-atlas · texture | 11 | 2 | RenderThread 20, RNSkia Present 17, RNSkia Render 15 | 155.9 | 106.6 | 4.7 |
+| tgfx · images · texture | 10 | 1 | TgfxRender 48, RenderThread 19, mali-cmar-backe 10 | 177.5 | 112.3 | 4.4 |
+| skia · images · texture | 9 | 1 | RNSkia Render 60, RNSkia Present 40, RenderThread 14 | 171.3 | 111.2 | 4.7 |
+| tgfx · shader · texture | 6 | 2 | RenderThread 4, TgfxRender 4, binder 4 | 183.2 | 105.2 | 4.4 |
+| skia · shader · texture | 4 | 1 | RNSkia Present 4, RenderThread 4, mali-cmar-backe 3 | 144.8 | 103.8 | 4.4 |
+| tgfx · pattern · texture | 12 | 2 | RenderThread 22, TgfxRender 14, mali-cmar-backe 13 | 177.4 | 105.1 | 4.4 |
+| skia · pattern · texture | 11 | 2 | RenderThread 20, RNSkia Present 17, mali-cmar-backe 11 | 138.9 | 102.4 | 4.7 |
+| tgfx · paths900 · texture | 10 | 2 | TgfxRender 52, RenderThread 17, mali-cmar-backe 9 | 191.2 | 112.3 | 4.4 |
+| skia · paths900 · texture | 9 | 2 | RNSkia Render 12, RenderThread 8, RNSkia Present 8 | 244.8 | 103.8 | 4.5 |
+| tgfx · paths2500 · texture | 10 | 1 | TgfxRender 95, RenderThread 13, mali-cmar-backe 7 | 266.2 | 124.1 | 4.4 |
+| skia · paths2500 · texture | 8 | 2 | RNSkia Render 26, RNSkia Present 7, RenderThread 7 | 249.7 | 105.6 | 4.5 |
+| tgfx · paths5000 · texture | 8 | 1 | TgfxRender 96, RenderThread 7, binder 4 | 291.7 | 143.3 | 4.4 |
+| skia · paths5000 · texture | 8 | 2 | RNSkia Render 25, mali-cmar-backe 5, RNSkia Present 5 | 256.1 | 110.1 | 4.5 |
+| tgfx · paths · texture | 7 | 1 | TgfxRender 151, nitro-thread 17, binder 4 | 268.8 | 185.9 | 4.4 |
+| skia · paths · texture | 9 | 2 | RNSkia Render 22, mali-cmar-backe 4, binder 3 | 278.6 | 110.7 | 4.4 |
+| tgfx · text · texture | 9 | 1 | TgfxRender 106, RenderThread 11, nitro-thread 9 | 185.7 | 129.0 | 4.4 |
+| skia · text · texture | 6 | 1 | RNSkia Render 95, RNSkia Present 16, RenderThread 5 | 219.0 | 118.4 | 4.5 |
+| tgfx · anim-decl · texture | 96 | 0 | TgfxRender 100, nitro-thread 9, RenderThread 3 | 186.6 | 189.0 | 4.4 |
+| skia · anim-decl · texture | 98 | 0 | RNSkia Render 7, RNSkia Present 4, binder 3 | 148.7 | 177.9 | 4.4 |
+| skia · anim-imp · texture | 97 | 1 | RNSkia Render 12, RNSkia Present 8, RenderThread 4 | 173.3 | 124.6 | 4.5 |
+| tgfx · shapes-picture · texture | 10 | 2 | TgfxRender 122, nitro-thread 15, RenderThread 13 | 196.4 | 144.1 | 4.4 |
+| skia · shapes-picture · texture | 8 | 1 | RNSkia Render 43, RNSkia Present 25, mali-cmar-backe 5 | 273.6 | 113.6 | 4.5 |
diff --git a/apps/benchmark/bench/results-2026-11-11-samsung-90hz-skia310.md b/apps/benchmark/bench/results-2026-11-11-samsung-90hz-skia310.md
new file mode 100644
index 0000000..97f1b27
--- /dev/null
+++ b/apps/benchmark/bench/results-2026-11-11-samsung-90hz-skia310.md
@@ -0,0 +1,83 @@
+# tgfx vs react-native-skia 3.10.0 on a mid-range phone
+
+Samsung SM-E146B (Exynos 1330, Mali-G68, Android 15, 1080x2408, 90 Hz),
+release build, Hermes, new architecture. react-native-tgfx from source (tgfx
+submodule `1b2a429`), react-native-skia 3.10.0 from npm.
+
+`node bench/run.mjs --order alternate --reps 2 --gap 10 --cool 0.95 --cool-timeout 240 --skip tgfx:anim-imp`,
+one view at a time.
+
+Every combo ran twice, each time in a fresh process started once the CPU
+frequency cap was back above 95 %. Present FPS is the median of the four 8 s
+windows (two per run), counted at SurfaceFlinger. Raw output:
+`results-2026-11-11-samsung-90hz-skia310-raw.md`.
+
+## Present FPS
+
+| scene | tgfx surface | skia surface | tgfx texture | skia texture |
+|---|---|---|---|---|
+| shapes, 20k (tgfx nodes, skia ShapeBlob) | 61.3 | **89.4** | 56.4 | **89.1** |
+| shapes-nodes, 20k nodes on both | **62.3** | 26.1 | **54.6** | 25.4 |
+| shapes-atlas, one atlas on both | 90.0 | 89.4 | 88.1 | 89.1 |
+| shapes-picture, 20k draws recorded once | **83.1** | 26.1 | **69.3** | 25.3 |
+| images, 3,600 | 90.1 | 88.1 | 88.8 | 85.1 |
+| shader | 15.0 | 14.8 | 12.6 | 14.3 |
+| pattern | 90.1 | 89.6 | 89.0 | 88.9 |
+| paths, 900 | **90.3** | 27.2 | **88.4** | 26.6 |
+| paths, 2,500 | **82.6** | 27.1 | **72.5** | 26.3 |
+| paths, 5,000 | **43.8** | 17.8 | **41.4** | 17.6 |
+| paths, 10,000 | 8.8 | 10.3 | 4.8 | **10.1** |
+| text, 6,000 (skia `deviceIndependentFonts`) | **73.0** | 32.4 | **65.4** | 30.3 |
+| anim-decl, 5k shapes, two derived values each | 24.6 | **32.3** | 11.6 | 9.4 |
+| anim-imp, 5k shapes, one worklet | ~17 by hand | 54.9 | ~6 by hand | 21.9 |
+
+## CPU
+
+| scene | tgfx surface | skia surface | tgfx texture | skia texture |
+|---|---|---|---|---|
+| shapes | 189 % | 70 % | 202 % | 87 % |
+| shapes-nodes | 184 % | 95 % | 200 % | 101 % |
+| shapes-atlas | 48 % | 70 % | 76 % | 88 % |
+| paths, 5,000 | 111 % | 53 % | 121 % | 53 % |
+| paths, 10,000 | 167 % | 47 % | 204 % | 47 % |
+| text | 128 % | 122 % | 146 % | 131 % |
+
+Whole process, 100 = one core.
+
+## Findings
+
+- **Shapes.** Each engine's fastest static path (`shapes`): skia's ShapeBlob
+ holds 89 fps; tgfx's 20k nodes reach 61 with `TgfxRender` saturated. With
+ the technique held equal the picture flips: 20k declarative nodes on
+ both, tgfx draws 2.2-2.4x skia (62 vs 26 surface); one atlas on both,
+ both sit at the ceiling and tgfx's `` mesh uses less CPU (48 vs
+ 70 %).
+- **Paths.** skia is GPU-bound on this Mali at every count: 27 fps at 900
+ and 2,500, 18 at 5,000, 10 at 10,000, with its render pool under 30 %.
+ tgfx holds 90 at 900 and 83 at 2,500, then `TgfxRender` saturates: 44 at
+ 5,000. At 10,000 it falls below skia, to 9 and 5, at 167 and 204 % CPU,
+ and the two runs disagree (9.7 and 8.2 surface, 10.7 and 4.3 texture): the
+ path geometry (about 86 MB) is larger than the canvas's GPU cache (64 MB),
+ so it is evicted and rebuilt every frame.
+- **Text and recorded shapes.** tgfx leads by 2.2x and 2.7-3.2x; skia's
+ `RNSkia Render` is the limit on text (93 %).
+- **Per-shape animation** is UI-thread bound for both (UI thread 96-98 %).
+ Declarative: skia ahead on SurfaceView, even on TextureView. tgfx
+ `anim-imp` is skipped by the runner; timed by hand inside the scene's
+ measured period (atrace, 8 s): about 17 fps surface, 6 texture.
+- **SurfaceView vs TextureView.** Static rows lose up to 17 % on TextureView
+ for tgfx and up to 7 % for skia; `anim-decl` halves or worse on both.
+
+## Conditions
+
+- Thermal status 0-2 (2 for the whole TextureView half). The CPU cap was back
+ above 95 % at the start of every combo; tgfx's CPU-bound rows (shapes,
+ shapes-nodes, text, paths from 2,500 on TextureView) dipped to 68-80 %
+ mid-combo, so read differences under ~10 % there as noise. skia's GPU-bound
+ rows matched across reps and between the two builds to within 2 %.
+- The notification shade was pulled down over the app for most of both
+ builds' runs (water on the screen while cooling it), so the runner's
+ parity stills show the shade. A back-to-back check with it closed and
+ open, eight combos each: skia's rows matched within 1 %; tgfx's
+ CPU-bound rows moved -5 to +8 %, in both directions, with the CPU cap.
+ Stills of the shapes scenes taken afterwards match across engines.
diff --git a/apps/benchmark/bench/run.mjs b/apps/benchmark/bench/run.mjs
index a6bf895..6542d3f 100644
--- a/apps/benchmark/bench/run.mjs
+++ b/apps/benchmark/bench/run.mjs
@@ -28,7 +28,7 @@ const PRESENT = arg('present', 'atrace');
// --gap idles the phone with no canvas between combos, so heat from one
// engine's run is not carried into the next one's.
const GAP = Number(arg('gap', '0'));
-const ALL_SCENES = 'shapes,images,shader,pattern,paths,text,anim-decl,anim-imp,shapes-picture';
+const ALL_SCENES = 'shapes,shapes-nodes,shapes-atlas,images,shader,pattern,paths900,paths2500,paths5000,paths,text,anim-decl,anim-imp,shapes-picture';
// surface: both engines on a SurfaceView; texture: both on a TextureView.
const VIEWS = arg('views', 'surface,texture');
// --skip tgfx:anim-imp drops engine:scene pairs (from every view).