Add build and compute buttons to the conv2d demo

reillyeon · Jul 13, 2024 · 591e88b · 591e88b
1 parent 8acd727
commit 591e88b
Showing 1 changed file with 59 additions and 18 deletions.
diff --git a/webnn-conv2d.html b/webnn-conv2d.html
@@ -36,11 +36,19 @@
         <td><canvas id="output" width="500" height="500"></canvas></td>
       </tr>
     </table>
+    <p>
+      <button id="build" disabled>Build</button> <button disabled id="compute">Compute</button>
+    </p>
     <pre id="status"></pre>
     <script>
       const channels = 4;
       let inputData;
 
+      let context;
+      let graph;
+      let outputWidth;
+      let outputHeight;
+
       async function createBlurGraph(context) {
         const inputLayout = inputLayoutElement.value;
         const builder = new MLGraphBuilder(context);
@@ -139,31 +147,46 @@
       const outputCtx = outputCanvas.getContext('2d');
       const statusSpan = document.getElementById('status');
 
-      async function run() {
+      async function computeGraph() {
         statusSpan.textContent = '';
 
         try {
           outputCtx.clearRect(0, 0, outputCanvas.width, outputCanvas.height);
 
-          const context = await navigator.ml.createContext({deviceType: deviceTypeElement.value});
-
-          const buildStart = performance.now();
-          const {graph, outputWidth, outputHeight} =
-              filterTypeElement.value == 'blur' ?
-                await createBlurGraph(context) : await createGrayscaleGraph(context);
-          const buildEnd = performance.now();
-
           const input = imageDataToTensor(inputData);
           const output = new Float32Array(outputWidth * outputHeight * channels);
 
-          const computeStart = performance.now();
+          performance.mark('compute-start');
           const {inputs, outputs} = await context.compute(graph, {input}, {output});
-          const computeEnd = performance.now();
+          performance.mark('compute-end');
 
           const outputData = tensorToImageData(outputs.output, outputWidth, outputHeight);
           outputCtx.putImageData(outputData, 0, 0);
 
-          statusSpan.textContent = `Build took ${(buildEnd - buildStart).toFixed(1)}ms. Compute took ${(computeEnd - computeStart).toFixed(1)}ms.`;
+          const computeMeasure = performance.measure('compute-measure', 'compute-start', 'compute-end');
+          statusSpan.textContent = `Compute took ${(computeMeasure.duration).toFixed(1)}ms.`;
+        } catch (e) {
+          statusSpan.textContent = e.stack;
+        }
+      }
+
+      async function buildGraph() {
+        statusSpan.textContent = '';
+
+        try {
+          context = await navigator.ml.createContext({deviceType: deviceTypeElement.value});
+
+          performance.mark('build-start');
+          ({graph, outputWidth, outputHeight} =
+              filterTypeElement.value == 'blur' ?
+                await createBlurGraph(context) : await createGrayscaleGraph(context));
+          performance.mark('build-end');
+
+          buildButton.disabled = true;
+          computeButton.disabled = false;
+
+          const buildMeasure = performance.measure('build-duration', 'build-start', 'build-end');
+          statusSpan.textContent = `Build took ${(buildMeasure.duration).toFixed(1)}ms.`;
         } catch (e) {
           statusSpan.textContent = e.stack;
         }
@@ -178,33 +201,51 @@
           inputData = inputCtx.getImageData(0, 0, image.width, image.height);
 
           if (!('ml' in navigator)) {
-            statusSpan.textContent = 'WebNN not supported in your browser.';
+            statusSpan.textContent = 'WebNN is not supported in your browser.';
             return;
           }
 
           deviceTypeElement.disabled = false;
           filterTypeElement.disabled = false;
           blurRadiusElement.disabled = false;
           inputLayoutElement.disabled = false;
-          run();
+          buildButton.disabled = false;
         };
         image.src = 'photo.jpg';
       }
 
+      function contextOptionsChanged() {
+        buildButton.disabled = false;
+        context = null;
+        graphOptionsChanged();
+      }
+
+      function graphOptionsChanged() {
+        buildButton.disabled = false;
+        computeButton.disabled = true;
+        graph = null;
+      }
+
       const deviceTypeElement = document.getElementById('deviceType');
-      deviceTypeElement.onchange = run
+      deviceTypeElement.onchange = contextOptionsChanged
 
       const filterTypeElement = document.getElementById('filterType');
       filterTypeElement.onchange = () => {
         blurRadiusElement.disabled = filterTypeElement.value != 'blur';
-        run();
+        graphOptionsChanged();
       };
 
       const blurRadiusElement = document.getElementById('blurRadius');
-      blurRadiusElement.onchange = run
+      blurRadiusElement.onchange = graphOptionsChanged
 
       const inputLayoutElement = document.getElementById('inputLayout');
-      inputLayoutElement.onchange = run
+      inputLayoutElement.onchange = graphOptionsChanged
+
+      const buildButton = document.getElementById('build');
+      buildButton.onclick = buildGraph;
+
+      const computeButton = document.getElementById('compute');
+      computeButton.onclick = computeGraph;
 
       loadInput();
     </script>