[AUDIO_WORKET] Add support for setting the render quantum size (#25747)

The `renderSizeHint` is now available in Chrome Canary
(`--enable-features=WebAudioConfigurableRenderQuantum`) so this can
finally be tested. Most of the work was already done in preparation, so
this required few changes.

Test with:
```
test/runner interactive.test_audio_worklet_params_mixing
```
Which has been extended to request a large `2048` sample audio frame
(which in an older browser will fallback to the default 128 samples).

Note for me: look at why and since how long we can bust out of the stack
without asserting. The stackAlloc() no longer appears to trip when
requesting more bytes than `wwParams.stackSize`.

Author: cwoffenden

src/audio_worklet.js

@@ -50,7 +50,7 @@ function createWasmAudioWorkletProcessor() {
       // 64 frames, for the case where a multi-MB stack is passed.
       this.outputViews = new Array(Math.min(((wwParams.stackSize - {{{ STACK_ALIGN }}}) / this.bytesPerChannel) | 0, /*sensible limit*/ 64));
 #if ASSERTIONS
-      console.assert(this.outputViews.length > 0, `AudioWorklet needs more stack allocating (at least ${this.bytesPerChannel})`);
+      assert(this.outputViews.length > 0, `AudioWorklet needs more stack allocating (at least ${this.bytesPerChannel})`);
 #endif
       this.createOutputViews();
 
@@ -138,8 +138,8 @@ function createWasmAudioWorkletProcessor() {
 #endif
       var oldStackPtr = stackSave();
 #if ASSERTIONS
-      console.assert(oldStackPtr == this.ctorOldStackPtr, 'AudioWorklet stack address has unexpectedly moved');
-      console.assert(outputViewsNeeded <= this.outputViews.length, `Too many AudioWorklet outputs (need ${outputViewsNeeded} but have stack space for ${this.outputViews.length})`);
+      assert(oldStackPtr == this.ctorOldStackPtr, 'AudioWorklet stack address has unexpectedly moved');
+      assert(outputViewsNeeded <= this.outputViews.length, `Too many AudioWorklet outputs (need ${outputViewsNeeded} but have stack space for ${this.outputViews.length})`);
 #endif
 
       // Allocate the necessary stack space. All pointer variables are in bytes;
@@ -154,6 +154,10 @@ function createWasmAudioWorkletProcessor() {
       var stackMemoryAligned = (stackMemoryStruct + stackMemoryData + 15) & ~15;
       var structPtr = stackAlloc(stackMemoryAligned);
       var dataPtr = structPtr + (stackMemoryAligned - stackMemoryData);
+#if ASSERTIONS
+      // TODO: look at why stackAlloc isn't tripping the assertions
+      assert(stackMemoryAligned <= wwParams.stackSize, `Not enough stack allocated to the AudioWorklet (need ${stackMemoryAligned}, got ${wwParams.stackSize})`);
+#endif
 
       // Copy input audio descriptor structs and data to Wasm (recall, structs
       // first, audio data after). 'inputsPtr' is the start of the C callback's
@@ -221,7 +225,7 @@ function createWasmAudioWorkletProcessor() {
         // And that the views' size match the passed in output buffers
         for (entry of outputList) {
           for (subentry of entry) {
-            console.assert(subentry.byteLength == this.bytesPerChannel, `AudioWorklet unexpected output buffer size (expected ${this.bytesPerChannel} got ${subentry.byteLength})`);
+            assert(subentry.byteLength == this.bytesPerChannel, `AudioWorklet unexpected output buffer size (expected ${this.bytesPerChannel} got ${subentry.byteLength})`);
           }
         }
       }

src/lib/libwebaudio.js

@@ -77,12 +77,7 @@ var LibraryWebAudio = {
 
   // Performs the work of getting the AudioContext's render quantum size.
   $emscriptenGetContextQuantumSize: (contextHandle) => {
-    // TODO: in a future release this will be something like:
-    //   return EmAudio[contextHandle].renderQuantumSize || 128;
-    // It comes two caveats: it needs the hint when generating the context adding to
-    // emscripten_create_audio_context(), and altering the quantum requires a secure
-    // context and fallback implementing. Until then we simply use the 1.0 API value:
-    return 128;
+    return EmAudio[contextHandle]['renderQuantumSize'] || 128;
   },
 
   // emscripten_create_audio_context() does not itself use the
@@ -100,9 +95,15 @@ var LibraryWebAudio = {
 #endif
 #endif
 
+    // Converts AUDIO_CONTEXT_RENDER_SIZE_* into AudioContextRenderSizeCategory
+    // enums, otherwise returns a positive int value.
+    function readRenderSizeHint(val) {
+      return (val < 0) ? 'hardware' : (val || 'default');
+    }
     var opts = options ? {
       latencyHint: UTF8ToString({{{ makeGetValue('options', C_STRUCTS.EmscriptenWebAudioCreateAttributes.latencyHint, '*') }}}) || undefined,
-      sampleRate: {{{ makeGetValue('options', C_STRUCTS.EmscriptenWebAudioCreateAttributes.sampleRate, 'u32') }}} || undefined
+      sampleRate: {{{ makeGetValue('options', C_STRUCTS.EmscriptenWebAudioCreateAttributes.sampleRate, 'u32') }}} || undefined,
+      renderSizeHint: readRenderSizeHint({{{ makeGetValue('options', C_STRUCTS.EmscriptenWebAudioCreateAttributes.renderSizeHint, 'i32') }}})
     } : undefined;
 
 #if WEBAUDIO_DEBUG

src/struct_info.json

@@ -1269,7 +1269,8 @@
         "structs": {
             "EmscriptenWebAudioCreateAttributes": [
               "latencyHint",
-              "sampleRate"
+              "sampleRate",
+              "renderSizeHint"
             ],
             "WebAudioParamDescriptor": [
               "defaultValue",

src/struct_info_generated.json

@@ -700,8 +700,9 @@
             "visibilityState": 4
         },
         "EmscriptenWebAudioCreateAttributes": {
-            "__size__": 8,
+            "__size__": 12,
             "latencyHint": 0,
+            "renderSizeHint": 8,
             "sampleRate": 4
         },
         "EmscriptenWebGLContextAttributes": {

src/struct_info_generated_wasm64.json

@@ -702,6 +702,7 @@
         "EmscriptenWebAudioCreateAttributes": {
             "__size__": 16,
             "latencyHint": 0,
+            "renderSizeHint": 12,
             "sampleRate": 8
         },
         "EmscriptenWebGLContextAttributes": {

system/include/emscripten/webaudio.h

@@ -19,10 +19,16 @@ extern "C" {
 
 typedef int EMSCRIPTEN_WEBAUDIO_T;
 
+// Default render size of 128 frames
+#define AUDIO_CONTEXT_RENDER_SIZE_DEFAULT 0
+// Let the hardware determine the best render size
+#define AUDIO_CONTEXT_RENDER_SIZE_HARDWARE -1
+
 typedef struct EmscriptenWebAudioCreateAttributes
 {
 	const char *latencyHint; // Specify one of "balanced", "interactive" or "playback"
 	uint32_t sampleRate; // E.g. 44100 or 48000
+	int32_t renderSizeHint; // AUDIO_CONTEXT_RENDER_SIZE_* or number of samples
 } EmscriptenWebAudioCreateAttributes;
 
 // Creates a new Web Audio AudioContext, and returns a handle to it.

test/codesize/audio_worklet_wasm.expected.js

@@ -1,4 +1,4 @@
-var m = globalThis.Module || "undefined" != typeof Module ? Module : {}, r = !!globalThis.AudioWorkletGlobalScope, t = "em-ww" == globalThis.name || r, u, z, I, J, G, E, w, X, F, D, C, Y, A, Z;
+var m = globalThis.Module || "undefined" != typeof Module ? Module : {}, p = !!globalThis.AudioWorkletGlobalScope, t = "em-ww" == globalThis.name || p, u, z, I, J, G, E, w, X, F, D, C, Y, A, Z;
 
 function v(a) {
     u = a;
@@ -10,12 +10,12 @@ function v(a) {
     a.G = a.M = 0;
 }
 
-t && !r && (onmessage = a => {
+t && !p && (onmessage = a => {
     onmessage = null;
     v(a.data);
 });
 
-if (r) {
+if (p) {
     function a(b) {
         class h extends AudioWorkletProcessor {
             constructor(d) {
@@ -36,7 +36,7 @@ if (r) {
                 return b;
             }
             process(d, g, e) {
-                var l = d.length, p = g.length, f, q, k = 12 * (l + p), n = 0;
+                var l = d.length, q = g.length, f, r, k = 12 * (l + q), n = 0;
                 for (f of d) n += f.length;
                 n *= this.s;
                 var H = 0;
@@ -53,15 +53,15 @@ if (r) {
                     G[k + 4 >> 2] = this.u;
                     G[k + 8 >> 2] = n;
                     k += 12;
-                    for (q of f) E.set(q, n >> 2), n += this.s;
+                    for (r of f) E.set(r, n >> 2), n += this.s;
                 }
                 d = k;
-                for (f = 0; q = e[f++]; ) G[k >> 2] = q.length, G[k + 4 >> 2] = n, k += 8, E.set(q, n >> 2), 
-                n += 4 * q.length;
+                for (f = 0; r = e[f++]; ) G[k >> 2] = r.length, G[k + 4 >> 2] = n, k += 8, E.set(r, n >> 2), 
+                n += 4 * r.length;
                 e = k;
                 for (f of g) G[k >> 2] = f.length, G[k + 4 >> 2] = this.u, G[k + 8 >> 2] = n, k += 12, 
                 n += this.s * f.length;
-                if (l = this.v(l, B, p, e, N, d, this.A)) for (f of g) for (q of f) q.set(this.B[--H]);
+                if (l = this.v(l, B, q, e, N, d, this.A)) for (f of g) for (r of f) r.set(this.B[--H]);
                 F(U);
                 return !!l;
             }
@@ -129,9 +129,12 @@ var K = [], L = a => {
 }, T = a => {
     if (a) {
         var c = G[a >> 2];
+        c = (c ? S(c) : "") || void 0;
+        var b = J[a + 8 >> 2];
         a = {
-            latencyHint: (c ? S(c) : "") || void 0,
-            sampleRate: G[a + 4 >> 2] || void 0
+            latencyHint: c,
+            sampleRate: G[a + 4 >> 2] || void 0,
+            N: 0 > b ? "hardware" : b || "default"
         };
     } else a = void 0;
     a = new AudioContext(a);
@@ -140,10 +143,10 @@ var K = [], L = a => {
 }, V = (a, c, b, h, d) => {
     var g = b ? J[b + 4 >> 2] : 0;
     if (b) {
-        var e = J[b >> 2], l = G[b + 8 >> 2], p = g;
+        var e = J[b >> 2], l = G[b + 8 >> 2], q = g;
         if (l) {
             l >>= 2;
-            for (var f = []; p--; ) f.push(G[l++]);
+            for (var f = []; q--; ) f.push(G[l++]);
             l = f;
         } else l = void 0;
         b = {
@@ -156,7 +159,7 @@ var K = [], L = a => {
             processorOptions: {
                 v: h,
                 A: d,
-                u: 128
+                u: O[a].renderQuantumSize || 128
             }
         };
     } else b = void 0;
@@ -191,17 +194,17 @@ var K = [], L = a => {
     if (!e) return l();
     e.addModule(m.js).then((() => {
         e.port || (e.port = {
-            postMessage: p => {
-                p._boot ? (e.D = new AudioWorkletNode(g, "em-bootstrap", {
-                    processorOptions: p
+            postMessage: q => {
+                q._boot ? (e.D = new AudioWorkletNode(g, "em-bootstrap", {
+                    processorOptions: q
                 }), e.D.port.onmessage = f => {
                     e.port.onmessage(f);
-                }) : e.D.port.postMessage(p);
+                }) : e.D.port.postMessage(q);
             }
         });
         e.port.postMessage({
             _boot: 1,
-            N: ba++,
+            O: ba++,
             G: m.wasm,
             L: w,
             J: c,
@@ -243,7 +246,7 @@ function y() {
         C = a.n;
         Y = a.o;
         A = a.k;
-        t ? (Y(u.J, u.F), r || (removeEventListener("message", M), K = K.forEach(L), addEventListener("message", L))) : a.i();
+        t ? (Y(u.J, u.F), p || (removeEventListener("message", M), K = K.forEach(L), addEventListener("message", L))) : a.i();
         t || X();
     }));
 }

test/codesize/test_minimal_runtime_code_size_audio_worklet.json

@@ -1,10 +1,10 @@
 {
   "a.html": 519,
   "a.html.gz": 357,
-  "a.js": 4235,
-  "a.js.gz": 2170,
+  "a.js": 4309,
+  "a.js.gz": 2219,
   "a.wasm": 1329,
   "a.wasm.gz": 895,
-  "total": 6083,
-  "total_gz": 3422
+  "total": 6157,
+  "total_gz": 3471
 }

test/webaudio/audioworklet_params_mixing.c

@@ -9,10 +9,18 @@
 // create variable parameter data sizes, depending on the browser, it's also the
 // ideal to test audio worklets don't corrupt TLS variables.
 
+// Large render size (approx 42ms)
+#define RENDER_SIZE_HINT 2048
+
 // This needs to be big enough for the stereo output, 2x inputs, 2x params and
 // the worker stack. To note that different browsers have different stack size
 // requirement (see notes in process() plus the expansion of the params).
+#ifndef RENDER_SIZE_HINT
 #define AUDIO_STACK_SIZE 6144
+#else
+// float bytes * stereo * ins/outs + extra stack
+#define AUDIO_STACK_SIZE ((RENDER_SIZE_HINT * 4 * 2 * 5) + 1024)
+#endif
 
 // Shared file playback and bootstrap
 #include "audioworklet_test_shared.inc"

test/webaudio/audioworklet_test_shared.inc

@@ -95,8 +95,12 @@ int main(void) {
   assert(workletStack);
 
   // Set at least the latency hint to test the attribute setting
+  // (The render size we take from the calling code if set)
   EmscriptenWebAudioCreateAttributes attrs = {
-    .latencyHint = "balanced"
+    .latencyHint = "balanced",
+#ifdef RENDER_SIZE_HINT
+    .renderSizeHint = RENDER_SIZE_HINT
+#endif
   };
   EMSCRIPTEN_WEBAUDIO_T context = emscripten_create_audio_context(&attrs);
   emscripten_start_wasm_audio_worklet_thread_async(context, workletStack, AUDIO_STACK_SIZE, getStartCallback(), NULL);