webnn: Replace MLActivation with MLRecurrentNetworkActivation

Implements the changes from this spec PR:
webmachinelearning/webnn#718

Bug: 351866391
Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel,win11-blink-rel
Change-Id: I163c85af63468a2ad4166509196ac3734d379df2

webnn/idlharness.https.any.js

@@ -22,7 +22,6 @@ idl_test(
       ML: ['navigator.ml'],
       MLContext: ['context'],
       MLOperand: ['input', 'constant', 'output'],
-      MLActivation: ['relu'],
       MLGraphBuilder: ['builder'],
       MLGraph: ['graph']
     });
@@ -35,9 +34,6 @@ idl_test(
         {dataType: 'float32', dimensions: [2, 3]},
         new Float32Array(2 * 3).fill(1));
 
-    // Create an activation which won't be used in the graph.
-    self.relu = builder.relu();
-
     self.output = builder.add(input, constant);
 
     self.graph = await builder.build({output});

webnn/resources/utils_validation.js

@@ -407,11 +407,8 @@ function validateOptionsAxes(operationName) {
  * @param {String} operationName - An operation name
  * @param {Array} supportedDataTypes - Test building with these data types
  *     succeeds and test building with all other data types fails
- * @param {Boolean} alsoBuildActivation - If test building this operation as an
- *     activation
  */
-function validateUnaryOperation(
-    operationName, supportedDataTypes, alsoBuildActivation = false) {
+function validateUnaryOperation(operationName, supportedDataTypes) {
   promise_test(async t => {
     const builder = new MLGraphBuilder(context);
     for (let dataType of supportedDataTypes) {
@@ -451,23 +448,13 @@ function validateUnaryOperation(
       }
     }
   }, `[${operationName}] Throw if the dataType is not supported for an unary operator.`);
-
-  if (alsoBuildActivation) {
-    promise_test(async t => {
-      const builder = new MLGraphBuilder(context);
-      builder[operationName]();
-    }, `[${operationName}] Test building an activation`);
-  }
 }
 
 /**
  * Validate a single input operation
  * @param {String} operationName - An operation name
- * @param {Boolean} alsoBuildActivation - If test building this operation as an
- *     activation
  */
-function validateSingleInputOperation(
-    operationName, alsoBuildActivation = false) {
+function validateSingleInputOperation(operationName) {
   promise_test(async t => {
     const builder = new MLGraphBuilder(context);
     const supportedDataTypes =
@@ -503,13 +490,6 @@ function validateSingleInputOperation(
       }
     }
   }, `[${operationName}] Throw if the data type is not supported for the operator.`);
-
-  if (alsoBuildActivation) {
-    promise_test(async t => {
-      const builder = new MLGraphBuilder(context);
-      builder[operationName]();
-    }, `[${operationName}] Test building an activation.`);
-  }
 }
 
 /**

webnn/validation_tests/clamp.https.any.js

@@ -23,7 +23,7 @@ promise_test(async t => {
   const output = builder.clamp(input, options);
   assert_equals(output.dataType(), 'uint32');
   assert_array_equals(output.shape(), [1, 2, 3]);
-}, '[clamp] Test building an operator with options');
+}, '[clamp] Build with options');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
@@ -40,7 +40,7 @@ promise_test(async t => {
   const output = builder.clamp(input, options);
   assert_equals(output.dataType(), 'int32');
   assert_array_equals(output.shape(), [1, 2, 3, 4]);
-}, '[clamp] Test building an operator with options.minValue == options.maxValue');
+}, '[clamp] Build with options.minValue == options.maxValue');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
@@ -55,7 +55,7 @@ promise_test(async t => {
   const input =
       builder.input('input', {dataType: 'uint8', dimensions: [1, 2, 3]});
   assert_throws_js(TypeError, () => builder.clamp(input, options));
-}, '[clamp] Throw if options.minValue > options.maxValue when building an operator');
+}, '[clamp] Throw if options.minValue > options.maxValue');
 
 // To be removed once infinite `minValue` is allowed. Tracked in
 // https://github.com/webmachinelearning/webnn/pull/647.
@@ -64,4 +64,4 @@ promise_test(async t => {
   const options = {minValue: -Infinity};
   const input = builder.input('input', {dataType: 'float16', dimensions: []});
   assert_throws_js(TypeError, () => builder.clamp(input, options));
-}, '[clamp] Throw if options.minValue is -Infinity when building an operator');
+}, '[clamp] Throw if options.minValue is -Infinity');

webnn/validation_tests/elu.https.any.js

@@ -6,7 +6,7 @@
 
 validateInputFromAnotherBuilder('elu');
 
-validateSingleInputOperation('elu', /*alsoBuildActivation=*/ true);
+validateSingleInputOperation('elu');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
@@ -16,37 +16,33 @@ promise_test(async t => {
   const output = builder.elu(input, options);
   assert_equals(output.dataType(), 'float32');
   assert_array_equals(output.shape(), [1, 2, 3]);
-}, '[elu] Test building an operator with options');
-
-promise_test(async t => {
-  const builder = new MLGraphBuilder(context);
-  const options = {alpha: 1.5};
-  builder.elu(options);
-}, '[elu] Test building an activation with options');
+}, '[elu] Build with options');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
   const options = {alpha: -1.0};
   const input =
       builder.input('input', {dataType: 'float32', dimensions: [1, 2, 3]});
   assert_throws_js(TypeError, () => builder.elu(input, options));
-}, '[elu] Throw if options.alpha <= 0 when building an operator');
+}, '[elu] Throw if options.alpha < 0');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
-  const options = {alpha: NaN};
-  const input = builder.input('input', {dataType: 'float16', dimensions: []});
+  const options = {alpha: 0};
+  const input = builder.input('input', {dataType: 'float32', dimensions: [1]});
   assert_throws_js(TypeError, () => builder.elu(input, options));
-}, '[elu] Throw if options.alpha is NaN when building an operator');
+}, '[elu] Throw if options.alpha == 0');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
-  const options = {alpha: 0};
-  assert_throws_js(TypeError, () => builder.elu(options));
-}, '[elu] Throw if options.alpha <= 0 when building an activation');
+  const options = {alpha: NaN};
+  const input = builder.input('input', {dataType: 'float16', dimensions: []});
+  assert_throws_js(TypeError, () => builder.elu(input, options));
+}, '[elu] Throw if options.alpha is NaN');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
   const options = {alpha: Infinity};
-  assert_throws_js(TypeError, () => builder.elu(options));
-}, '[elu] Throw if options.alpha is Infinity when building an activation');
+  const input = builder.input('input', {dataType: 'float32', dimensions: [1]});
+  assert_throws_js(TypeError, () => builder.elu(input, options));
+}, '[elu] Throw if options.alpha is Infinity');

webnn/validation_tests/gelu.https.any.js

@@ -6,4 +6,4 @@
 
 validateInputFromAnotherBuilder('gelu');
 
-validateSingleInputOperation('gelu', /*alsoBuildActivation=*/ true);
+validateSingleInputOperation('gelu');

webnn/validation_tests/gru.https.any.js

@@ -323,9 +323,7 @@ tests.forEach(
           options.layout = test.options.layout;
         }
         if (test.options.activations) {
-          options.activations = [];
-          test.options.activations.forEach(
-              activation => options.activations.push(builder[activation]()));
+          options.activations = test.options.activations;
         }
       }
 
@@ -429,18 +427,3 @@ multi_builder_test(async (t, builder, otherBuilder) => {
       () => builder.gru(
           input, weight, recurrentWeight, steps, hiddenSize, options));
 }, '[gru] throw if initialHiddenState option is from another builder');
-
-multi_builder_test(async (t, builder, otherBuilder) => {
-  const activation = builder.relu();
-  const activationFromOtherBuilder = otherBuilder.relu();
-  const options = {activations: [activation, activationFromOtherBuilder]};
-
-  const input = builder.input('input', kExampleInputDescriptor);
-  const weight = builder.input('weight', kExampleWeightDescriptor);
-  const recurrentWeight =
-      builder.input('recurrentWeight', kExampleRecurrentWeightDescriptor);
-  assert_throws_js(
-      TypeError,
-      () => builder.gru(
-          input, weight, recurrentWeight, steps, hiddenSize, options));
-}, '[gru] throw if any activation option is from another builder');

webnn/validation_tests/gruCell.https.any.js

@@ -340,10 +340,7 @@ tests.forEach(
               options.layout = test.options.layout;
             }
             if (test.options.activations) {
-              options.activations = [];
-              test.options.activations.forEach(
-                  activation =>
-                      options.activations.push(builder[activation]()));
+              options.activations = test.options.activations;
             }
           }
 
@@ -457,20 +454,3 @@ multi_builder_test(async (t, builder, otherBuilder) => {
       () => builder.gruCell(
           input, weight, recurrentWeight, hiddenState, hiddenSize, options));
 }, '[gruCell] throw if recurrentBias option is from another builder');
-
-multi_builder_test(async (t, builder, otherBuilder) => {
-  const activation = builder.relu();
-  const activationFromOtherBuilder = otherBuilder.relu();
-  const options = {activations: [activation, activationFromOtherBuilder]};
-
-  const input = builder.input('input', kExampleInputDescriptor);
-  const weight = builder.input('weight', kExampleWeightDescriptor);
-  const recurrentWeight =
-      builder.input('recurrentWeight', kExampleRecurrentWeightDescriptor);
-  const hiddenState =
-      builder.input('hiddenState', kExampleHiddenStateDescriptor);
-  assert_throws_js(
-      TypeError,
-      () => builder.gruCell(
-          input, weight, recurrentWeight, hiddenState, hiddenSize, options));
-}, '[gruCell] throw if any activation option is from another builder');

webnn/validation_tests/hardSigmoid.https.any.js

@@ -6,8 +6,7 @@
 
 validateInputFromAnotherBuilder('hardSigmoid');
 
-validateUnaryOperation(
-    'hardSigmoid', floatingPointTypes, /*alsoBuildActivation=*/ true);
+validateUnaryOperation('hardSigmoid', floatingPointTypes);
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
@@ -17,23 +16,18 @@ promise_test(async t => {
   const output = builder.hardSigmoid(input, options);
   assert_equals(output.dataType(), 'float16');
   assert_array_equals(output.shape(), [1, 2, 3]);
-}, '[hardSigmoid] Test building an operator with options');
-
-promise_test(async t => {
-  const builder = new MLGraphBuilder(context);
-  const options = {alpha: 0.2};
-  builder.hardSigmoid(options);
-}, '[hardSigmoid] Test building an activation with options');
+}, '[hardSigmoid] Test building with options');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
   const options = {beta: NaN};
   const input = builder.input('input', {dataType: 'float32', dimensions: []});
   assert_throws_js(TypeError, () => builder.hardSigmoid(input, options));
-}, '[hardSigmoid] Throw if options.beta is NaN when building an operator');
+}, '[hardSigmoid] Throw if options.beta is NaN');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
   const options = {alpha: Infinity};
-  assert_throws_js(TypeError, () => builder.hardSigmoid(options));
-}, '[hardSigmoid] Throw if options.alpha is Infinity when building an activation');
+  const input = builder.input('input', {dataType: 'float32', dimensions: [1]});
+  assert_throws_js(TypeError, () => builder.hardSigmoid(input, options));
+}, '[hardSigmoid] Throw if options.alpha is Infinity');

webnn/validation_tests/hardSwish.https.any.js

@@ -6,5 +6,4 @@
 
 validateInputFromAnotherBuilder('hardSwish');
 
-validateUnaryOperation(
-    'hardSwish', floatingPointTypes, /*alsoBuildActivation=*/ true);
+validateUnaryOperation('hardSwish', floatingPointTypes);

webnn/validation_tests/leakyRelu.https.any.js

@@ -6,7 +6,7 @@
 
 validateInputFromAnotherBuilder('leakyRelu');
 
-validateSingleInputOperation('leakyRelu', /*alsoBuildActivation=*/ true);
+validateSingleInputOperation('leakyRelu');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
@@ -16,23 +16,18 @@ promise_test(async t => {
   const output = builder.leakyRelu(input, options);
   assert_equals(output.dataType(), 'float32');
   assert_array_equals(output.shape(), [1, 2, 3]);
-}, '[leakyRelu] Test building an operator with options');
-
-promise_test(async t => {
-  const builder = new MLGraphBuilder(context);
-  const options = {alpha: 0.03};
-  builder.leakyRelu(options);
-}, '[leakyRelu] Test building an activation with options');
+}, '[leakyRelu] Build with options');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
   const options = {alpha: Infinity};
   const input = builder.input('input', {dataType: 'float16', dimensions: []});
   assert_throws_js(TypeError, () => builder.leakyRelu(input, options));
-}, '[leakyRelu] Throw if options.alpha is Infinity when building an operator');
+}, '[leakyRelu] Throw if options.alpha is Infinity');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
   const options = {alpha: -NaN};
-  assert_throws_js(TypeError, () => builder.leakyRelu(options));
-}, '[leakyRelu] Throw if options.alpha is -NaN when building an activation');
+  const input = builder.input('input', {dataType: 'float32', dimensions: [1]});
+  assert_throws_js(TypeError, () => builder.leakyRelu(input, options));
+}, '[leakyRelu] Throw if options.alpha is -NaN');

webnn/validation_tests/linear.https.any.js

@@ -6,8 +6,7 @@
 
 validateInputFromAnotherBuilder('linear');
 
-validateUnaryOperation(
-    'linear', floatingPointTypes, /*alsoBuildActivation=*/ true);
+validateUnaryOperation('linear', floatingPointTypes);
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
@@ -17,23 +16,18 @@ promise_test(async t => {
   const output = builder.linear(input, options);
   assert_equals(output.dataType(), 'float32');
   assert_array_equals(output.shape(), [1, 2, 3]);
-}, '[linear] Test building an operator with options');
-
-promise_test(async t => {
-  const builder = new MLGraphBuilder(context);
-  const options = {beta: 1.5};
-  builder.linear(options);
-}, '[linear] Test building an activation with options');
+}, '[linear] Build with options');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
   const options = {beta: -Infinity};
   const input = builder.input('input', {dataType: 'float16', dimensions: []});
   assert_throws_js(TypeError, () => builder.linear(input, options));
-}, '[linear] Throw if options.beta is -Infinity when building an operator');
+}, '[linear] Throw if options.beta is -Infinity');
 
 promise_test(async t => {
   const builder = new MLGraphBuilder(context);
   const options = {alpha: NaN};
-  assert_throws_js(TypeError, () => builder.linear(options));
-}, '[linear] Throw if options.alpha is NaN when building an activation');
+  const input = builder.input('input', {dataType: 'float32', dimensions: [1]});
+  assert_throws_js(TypeError, () => builder.linear(input, options));
+}, '[linear] Throw if options.alpha is NaN');

webnn/validation_tests/lstm.https.any.js

@@ -318,9 +318,7 @@ tests.forEach(
           options.layout = test.options.layout;
         }
         if (test.options.activations) {
-          options.activations = [];
-          test.options.activations.forEach(
-              activation => options.activations.push(builder[activation]()));
+          options.activations = test.options.activations;
         }
       }
 
@@ -455,18 +453,3 @@ multi_builder_test(async (t, builder, otherBuilder) => {
       () => builder.lstm(
           input, weight, recurrentWeight, steps, hiddenSize, options));
 }, '[lstm] throw if initialCellState option is from another builder');
-
-multi_builder_test(async (t, builder, otherBuilder) => {
-  const activation = builder.relu();
-  const activationFromOtherBuilder = otherBuilder.relu();
-  const options = {activations: [activation, activationFromOtherBuilder]};
-
-  const input = builder.input('input', kExampleInputDescriptor);
-  const weight = builder.input('weight', kExampleWeightDescriptor);
-  const recurrentWeight =
-      builder.input('recurrentWeight', kExampleRecurrentWeightDescriptor);
-  assert_throws_js(
-      TypeError,
-      () => builder.lstm(
-          input, weight, recurrentWeight, steps, hiddenSize, options));
-}, '[lstm] throw if any activation option is from another builder');