[webnn] Add tests for concat operation. #36782

BruceDai · 2022-11-02T16:28:36Z

@fdwr This pr only has concat operation tests, this would make you review easily. I moved tests and test data onto resources/test_data/concat.json JSON file which will make maintain test smoothly, and I also addressed some feedbacks which you commented on #34287. Please help review it, thanks a lot.

@dontcallmedom @Honry @huningxin Please also take a review, thanks.

BruceDai · 2022-11-02T16:31:02Z

webnn/resources/test_data/concat.json

+    }
+  },
+  "inputsData": {
+    "float64Data": [


These inputsData and below expectedData were automatically generated by this scripts of WebNN-Baseline project.

fdwr · 2022-11-08T08:29:04Z

webnn/concat.https.any.js

+    graph = await builder.build({outputOperand});
+    await context.compute(graph, namedInputs, outputs);
+  }
+  assert_array_approx_equals_ulp(outputs.outputOperand, expectedShapeValue.data, PrecisionMetrics.concat.ULP[operandType], operandType);


[minor] Hmm, why does this function name use snake case whereas nearly every other function and variable uses camelCase? (e.g. sizeOfShape, computeSync, loadTestData)?

The assertion function assert_array_approx_equals_ulp() was intended to add into testharness.js previously, so its' name followed like function assert_array_equals() of testharness.js.
Since assert_array_approx_equals_ulp() is especial for WebNN API tests, I'm OK to use camelCase style.

I see. So there was some external convention.

Yes.
I go through some wpt tests of others features, there are common problems for variable/function name with snake case or camelCase ... :(
So what's your suggestion for this assertion function? Thanks.

Oh, I generally lean toward internal consistency with whatever code is immediately around it, but there's also the consistency argument of matching other "assert_" functions. I'm fine with whichever you decide - you answered my question either way. ✅

I leave this assert to be consistent with other "assert_" functions in testharness.js, thanks.

fdwr · 2022-11-08T08:29:37Z

webnn/resources/utils.js

+    distance = actualBitwise - expectedBitwise;
+    distance = distance >= 0 ? distance : -distance;
+    assert_true(distance <= nulp,
+                `actual ${actual[i]} should be close enough to expected ${expected[i]} by the acceptable ${nulp} ULP distance, while they have ${distance} ULP distance`);


Propose "while" -> "but".

Updated, thanks.

fdwr · 2022-11-08T08:30:23Z

webnn/resources/utils.js


+// Refer to precision metrics on https://github.com/webmachinelearning/webnn/issues/265#issuecomment-1256242643
+const PrecisionMetrics = {
+  concat: {ULP: {float32: 0, float16: 0}},


Nice, thanks for the order update.

fdwr · 2022-11-08T08:41:08Z

webnn/concat.https.any.js

+        }
+        const expectedShapeValue = {shape: test.expected.shape, data: expectedData[expectedDataCategory]};
+        await testConcat(operandType, isSync, inputShapeValues, test.axis, expectedShapeValue);
+      }, `${test.name} / ${operandType} / ${deviceType} / ${executionType}`);


[ignorable] I notice you usually put a new line after most }'s, and so at first glance, it looked like a typo here, as I couldn't tell if the brace was in the wrong place for the for loop, not realizing it was a second parameter to the promise call. So new line after the },? (note I'm not familiar with Javascript common style conventions, but some I find easier to visually parse than others 🤷)

fdwr · 2022-11-08T08:44:29Z

webnn/concat.https.any.js

+        for (const shape of inputShapes) {
+          const size = sizeOfShape(shape);
+          inputShapeValues.push({shape, data: inputsData[inputDataCategory].slice(pos, pos + size)});
+          pos += size;


How about position or offset? (I just like full words :))

Updated, thanks.

fdwr · 2022-11-08T08:44:58Z

webnn/resources/utils.js

  }
+
+  const json = loadJSON(file);
+  return JSON.parse(json.replace(/\\"|"(?:\\"|[^"])*"|(\/\/.*|\/\*[\s\S]*?\*\/)/g, (m, g) => g ? "" : m));


Nice, this way we can have comments in the JSON.

Honry

Generally LGTM, thanks @BruceDai !

webnn/concat.https.any.js

BruceDai

@fdwr @Honry Thanks for your reviewing, I've updated PR to address your comments, please take another look, thanks.

BruceDai · 2022-11-09T06:08:10Z

webnn/concat.https.any.js

+    graph = await builder.build({outputOperand});
+    await context.compute(graph, namedInputs, outputs);
+  }
+  assert_array_approx_equals_ulp(outputs.outputOperand, expectedShapeValue.data, PrecisionMetrics.concat.ULP[operandType], operandType);


I leave this assert to be consistent with other "assert_" functions in testharness.js, thanks.

BruceDai · 2022-11-09T06:08:55Z

webnn/resources/utils.js

+    distance = actualBitwise - expectedBitwise;
+    distance = distance >= 0 ? distance : -distance;
+    assert_true(distance <= nulp,
+                `actual ${actual[i]} should be close enough to expected ${expected[i]} by the acceptable ${nulp} ULP distance, while they have ${distance} ULP distance`);


Updated, thanks.

BruceDai · 2022-11-09T06:09:11Z

webnn/concat.https.any.js

+        for (const shape of inputShapes) {
+          const size = sizeOfShape(shape);
+          inputShapeValues.push({shape, data: inputsData[inputDataCategory].slice(pos, pos + size)});
+          pos += size;


Updated, thanks.

webnn/concat.https.any.js

BruceDai · 2022-11-10T06:15:40Z

Hi @huningxin, according to your suggestion on webmachinelearning/webnn-polyfill#172, I updated this PR by invoking sync functions and async functions clearly not using syncFlag flag, please have a review, thanks.

BruceDai · 2022-11-16T09:38:22Z

webnn/concat.https.any.js

+  return [{outputOperand}, inputs, outputs];
+};
+
+testWebNNOperation('concat', concatTests(), buildGraph);


Hi @fdwr, @Honry and @huningxin,
I refined code to reuse for later adding other operations tests. Now each test file named <operation-name>.https.any.js will have <operation-name>Tests(), buildGraph() two functions, and run test by invoking testWebNNOperation() function, please take another look, thanks.

BruceDai · 2022-11-16T09:44:48Z

webnn/resources/utils.js

+            }
+            tolerance++;
+          }
+          break;


@fdwr I created this getPrecisonTolerance() function, and add detail implementation to get a ULP tolerance for these operations which need compute by IEPOE, please take another look, thanks.

Would it be simpler to just include the function name in the table, right beside all the other values, rather than needing to fill in a switch statement (because then there are two separate places to look for values and update, a table and a code switch)? e.g.

function getGemmPrecisionTolerance(shapeA, options) { const shapeA = resources.inputA.shape; options = resources.options; ... } ... const PrecisionMetrics = { batchNormalization: {ULP: {float32: 6, float16: 6}}, ... gemm: {ULP: {float32: getGemmPrecisionTolerance, float16: getGemmPrecisionTolerance}} ... maxPool2d: {ULP: {float32: 0, float16: 0}}, ... }; ... function getPrecisonTolerance(opName, metricType, tensorDataType, resources) { let tolerance; if (metricType === 'ULP') { tolerance = PrecisionMetrics[opName].ULP[tensorDataType]; } else if (metricType === 'ATOL') { tolerance = PrecisionMetrics[opName].ATOL[operandType]; } // If the tolerance is dynamic, then evaluate the function to get the value. if (x instanceof Function) { tolerance = tolerance(resources); } return tolerance; }

(but I'm also fine with the current code)

Updated, thanks.

Honry · 2022-11-17T01:42:52Z

webnn/resources/utils.js

+   * @param {Boolean} [options.bTranspose]
+   * @returns {Number} A tolerance number
+   */
+  function getGemmPrecisionTolerance(shapeA, options) {


Is there any reason that we have to put these functions inside the testWebNNOperation? This would make code not easy to read.

Hi @Honry, these inner functions are only invoked by the implementation of testWebNNOperation() function, for this purpose, I define them inside, thanks.

Yeah, nested functions are harder to read and discern where the beginning and end are (even if only called by that one function).

Thanks for your suggestion, moved these functions outside.

fdwr · 2022-11-18T04:55:04Z

webnn/resources/utils.js

-
 const TypedArrayDict = {
  float32: Float32Array,
 };


(minor) We might as well fill in a few more now, given we know the other data types already (no reason to wait, right?).

int32: Int32Array, uint32: Uint32Array, int8: Int8Array, uint8: Uint8Array,

(obviously float16 is harder and can come later, after deciding how to handle it)

Added these code, thanks.

fdwr · 2022-11-18T04:59:50Z

webnn/resources/utils.js

+   * @param {Boolean} [options.bTranspose]
+   * @returns {Number} A tolerance number
+   */
+  function getGemmPrecisionTolerance(shapeA, options) {


Yeah, nested functions are harder to read and discern where the beginning and end are (even if only called by that one function).

fdwr · 2022-11-18T05:11:45Z

webnn/resources/utils.js

+            }
+            tolerance++;
+          }
+          break;


Would it be simpler to just include the function name in the table, right beside all the other values, rather than needing to fill in a switch statement (because then there are two separate places to look for values and update, a table and a code switch)? e.g.

function getGemmPrecisionTolerance(shapeA, options) { const shapeA = resources.inputA.shape; options = resources.options; ... } ... const PrecisionMetrics = { batchNormalization: {ULP: {float32: 6, float16: 6}}, ... gemm: {ULP: {float32: getGemmPrecisionTolerance, float16: getGemmPrecisionTolerance}} ... maxPool2d: {ULP: {float32: 0, float16: 0}}, ... }; ... function getPrecisonTolerance(opName, metricType, tensorDataType, resources) { let tolerance; if (metricType === 'ULP') { tolerance = PrecisionMetrics[opName].ULP[tensorDataType]; } else if (metricType === 'ATOL') { tolerance = PrecisionMetrics[opName].ATOL[operandType]; } // If the tolerance is dynamic, then evaluate the function to get the value. if (x instanceof Function) { tolerance = tolerance(resources); } return tolerance; }

(but I'm also fine with the current code)

fdwr · 2022-11-18T05:19:21Z

webnn/resources/utils.js

+  const checkResults = (opName, operandType, namedOutputOperands, outputs, resources) => {
+    const metricType = PrecisionMetrics[opName] ? Object.keys(PrecisionMetrics[opName])[0] : 'ULP';
+    const description = `test ${opName} ${operandType}`;
+    const tolerance = getPrecisonTolerance(opName, metricType, operandType, resources);


It's possible someday that WebNN will support operators which return multiple outputs, and each output could have its own tolerance. It's not an issue yet for WebNN v1, and so I'm not worried about it, but we should keep that in mind when writing the code. The only operators I can think of right now are MAX_POOL and TOP_K, but those both expect exact results, and so both outputs have the same tolerance anyway.

Yes, thanks for your insightful reminder.

BruceDai

Thanks @Honry @fdwr, I've updated the commit to address your comments, please take another look.

BruceDai · 2022-11-18T07:11:25Z

webnn/resources/utils.js

-
 const TypedArrayDict = {
  float32: Float32Array,
 };


Added these code, thanks.

BruceDai · 2022-11-18T07:12:09Z

webnn/resources/utils.js

+   * @param {Boolean} [options.bTranspose]
+   * @returns {Number} A tolerance number
+   */
+  function getGemmPrecisionTolerance(shapeA, options) {


Thanks for your suggestion, moved these functions outside.

BruceDai · 2022-11-18T07:12:31Z

webnn/resources/utils.js

+            }
+            tolerance++;
+          }
+          break;


Updated, thanks.

BruceDai · 2022-11-18T07:17:20Z

webnn/resources/utils.js

+  const checkResults = (opName, operandType, namedOutputOperands, outputs, resources) => {
+    const metricType = PrecisionMetrics[opName] ? Object.keys(PrecisionMetrics[opName])[0] : 'ULP';
+    const description = `test ${opName} ${operandType}`;
+    const tolerance = getPrecisonTolerance(opName, metricType, operandType, resources);


Yes, thanks for your insightful reminder.

BruceDai · 2022-11-21T01:20:26Z

@fdwr Sorry. Updated, thanks.

fdwr

Looks nice. Thanks.

huningxin

Looks good with some nits, thanks!

huningxin · 2022-11-21T06:54:23Z

webnn/concat.https.any.js

+
+// https://webmachinelearning.github.io/webnn/#api-mlgraphbuilder-concat
+
+const concatTests = () => {


concatTests looks common except the url of resources json file. Could it be moved to utils.js with a parameter for json file url?

Hi @huningxin, thanks for your reviewing and suggestions.
Since there're different implementations (L#16 to L#29) to set targetTests for WebNN operations, for example the implementation for gemmTests likes:

for (const test of tests) { const gemmOptions = {}; const options = test.options; if (options !== undefined) { if (options.c !== undefined) { gemmOptions.c = {data: inputsData[options.c.data]}; if (options.c.shape !== undefined) { gemmOptions.c['shape'] = options.c.shape; } } if (options.alpha !== undefined) { gemmOptions.alpha = inputsData[options.alpha.data]; } if (options.beta !== undefined) { gemmOptions.beta = inputsData[options.beta.data]; } if (options.aTranspose !== undefined) { gemmOptions.aTranspose = options.aTranspose; } if (options.bTranspose !== undefined) { gemmOptions.bTranspose = options.bTranspose; } } targetTests.push( { name: test.name, operandType: test.type, inputA: {shape: test.inputs.a.shape, data: inputsData[test.inputs.a.data]}, inputB: {shape: test.inputs.b.shape, data: inputsData[test.inputs.b.data]}, expected: {shape: test.expected.shape, data: {outputOperand: expectedData[test.expected.data]}}, options: gemmOptions } ); }

and implementation for clampTests likes:

for (const test of tests) { const expectedDataSource = test.expected.data; const inputDataSource = test.input.data; targetTests.push( { name: test.name, operandType: test.type, input: {shape: test.input.shape, data: inputsData[inputDataSource]}, expected: {shape: test.expected.shape, data: {outputOperand: expectedData[expectedDataSource]}}, options: test.options } ); }

so I implemented each <operation-name>Tests function for each operation, thanks.

@huningxin I just submitted tests for clamp (#37067) / gemm (#37068) / matmul (#37069) / reshape (#37070) / relu (#37071) / slice (#37072) operations, PTAL, thanks.

This function looks like mainly setting up the test object for targetTests. It seems to me that all information and configuration of each test object come from the JOSN file. I am wondering: could we just define all these info in the JSON file and load it into resources. I suppose we don't need different JS implementation to setup test object for different ops.

Did I miss anything?

@huningxin Do you suggest add actual test data for "data"? For example

wpt/webnn/resources/test_data/concat.json

Line 8 in 32290b2

"data": "float64Data"

my purpose of using the index string key of input / expected data is to reuse test data, thanks.

@huningxin I've updated this PR for more reuse code, please take another look. And there's a reusable tests demo on my own repo for your quickly review, thanks.

huningxin · 2022-11-21T06:58:33Z

webnn/concat.https.any.js

+  for (let i = 0; i < inputShapeValues.length; i++) {
+    inputOperands.push(builder.input('input' + i, {type: operandType, dimensions: inputShapeValues[i].shape}));
+    inputs['input' + i] = new TestTypedArray(inputShapeValues[i].data);
+  }


The code that sets inputOperands and inputs look common for other operator tests. Could they be extract to utils.js?

@huningxin Thanks for your suggestions, I will try to optimize setting inputOperands, inputs and outputs in the next PR.

Building inputs from JSON looks good.

huningxin · 2022-11-21T06:59:09Z

webnn/concat.https.any.js

+    inputs['input' + i] = new TestTypedArray(inputShapeValues[i].data);
+  }
+  const outputOperand = builder.concat(inputOperands, resources.axis);
+  const outputs = {'outputOperand': new TestTypedArray(sizeOfShape(resources.expected.shape))};


Ditto. The code that sets outputs looks common for other operator tests.

huningxin · 2022-11-21T07:08:01Z

webnn/concat.https.any.js

+  return [{outputOperand}, inputs, outputs];
+};
+
+testWebNNOperation('concat', concatTests(), buildGraph);


Could this be something like

testWebNNOperation('concat', '/webnn/resources/test_data/concat.json', (inputOperands, resources) => { return {outputOperand: builder.concat(inputOperands, resources.axis)}; });

BruceDai

@huningxin Thanks for your reviewing and suggestions, I will follow up in next PR, thanks.

BruceDai · 2022-11-21T07:14:07Z

webnn/concat.https.any.js

+
+// https://webmachinelearning.github.io/webnn/#api-mlgraphbuilder-concat
+
+const concatTests = () => {


Hi @huningxin, thanks for your reviewing and suggestions.
Since there're different implementations (L#16 to L#29) to set targetTests for WebNN operations, for example the implementation for gemmTests likes:

for (const test of tests) { const gemmOptions = {}; const options = test.options; if (options !== undefined) { if (options.c !== undefined) { gemmOptions.c = {data: inputsData[options.c.data]}; if (options.c.shape !== undefined) { gemmOptions.c['shape'] = options.c.shape; } } if (options.alpha !== undefined) { gemmOptions.alpha = inputsData[options.alpha.data]; } if (options.beta !== undefined) { gemmOptions.beta = inputsData[options.beta.data]; } if (options.aTranspose !== undefined) { gemmOptions.aTranspose = options.aTranspose; } if (options.bTranspose !== undefined) { gemmOptions.bTranspose = options.bTranspose; } } targetTests.push( { name: test.name, operandType: test.type, inputA: {shape: test.inputs.a.shape, data: inputsData[test.inputs.a.data]}, inputB: {shape: test.inputs.b.shape, data: inputsData[test.inputs.b.data]}, expected: {shape: test.expected.shape, data: {outputOperand: expectedData[test.expected.data]}}, options: gemmOptions } ); }

and implementation for clampTests likes:

for (const test of tests) { const expectedDataSource = test.expected.data; const inputDataSource = test.input.data; targetTests.push( { name: test.name, operandType: test.type, input: {shape: test.input.shape, data: inputsData[inputDataSource]}, expected: {shape: test.expected.shape, data: {outputOperand: expectedData[expectedDataSource]}}, options: test.options } ); }

so I implemented each <operation-name>Tests function for each operation, thanks.

BruceDai · 2022-11-21T07:34:31Z

webnn/concat.https.any.js

+    let position = 0;
+    for (const shape of inputShapes) {
+      const size = sizeOfShape(shape);
+      inputShapeValues.push({shape, data: inputsData[inputDataSource].slice(position, position + size)});


This code is especial to concat op. To inputs of others operations, it's simply to set data. @huningxin

BruceDai · 2022-11-21T07:38:24Z

webnn/concat.https.any.js

+  for (let i = 0; i < inputShapeValues.length; i++) {
+    inputOperands.push(builder.input('input' + i, {type: operandType, dimensions: inputShapeValues[i].shape}));
+    inputs['input' + i] = new TestTypedArray(inputShapeValues[i].data);
+  }


@huningxin Thanks for your suggestions, I will try to optimize setting inputOperands, inputs and outputs in the next PR.

BruceDai

@huningxin Thanks for your reviewing and suggestions.

BruceDai · 2022-11-26T14:48:44Z

webnn/concat.https.any.js

-  const inputShapeValues = resources.inputShapeValues;
-  const operandType = resources.operandType;
-  const TestTypedArray = TypedArrayDict[operandType];
+const computeConcat = (builder, resources) => {


Thanks. I will update it to buildConcat.

BruceDai · 2022-11-26T14:57:08Z

webnn/resources/utils.js

+    aTranspose: false,
+    bTranspose: false
+  }
+  const options = resources.options ? resources.options : defaultOptions;


Is there any method to get this 'MLGemmOptions' from browser? Thanks.

BruceDai · 2022-11-26T15:05:14Z

webnn/resources/utils.js

+ * @param {String[]} nameArray
+ * @returns {MLOperand[]} Input operands
+ */
+const createInputOperands = (builder, resources, nameArray) => {


Since the inputs of concat is a sequence, it's special to other ops.
I form its inputs, like

"inputs": [ { "name": "input1", "shape": [12], "data": [ -0.39444134019222243, ... ] }, { "name": "input2", "shape": [12], "data": [ 0.4918989118791477, .... ] } ],

while for inputs of clamp, I form it like

"inputs": { "x": { // use "x" for input operand name "shape": [12], "data": [ -0.39444134019222243, ... - ] } }

so this createInputOperands is for other ops, please see the usage on demo tests .

BruceDai · 2022-11-26T15:05:37Z

webnn/resources/utils.js

+ * Build a graph.
+ * @param {MLGraphBuilder} builder - A ML graph builder
+ * @param {Object} resources - Resources used for building a graph
+ * @param {Function} computeFunc - A compute function


Thanks, I will update it.

BruceDai · 2022-11-26T15:21:59Z

webnn/resources/test_data/concat.json

+          0.1211843192577362
+        ]
      },
      "type": "float32"


Thanks, I will update it. And is it ok to still keep "type" here to indicate that current test is for float32 precision?

BruceDai · 2022-11-28T03:52:46Z

@huningxin I've updated PR to address your comments, please take another look, thanks.

webnn/resources/utils.js

webnn/concat.https.any.js

BruceDai · 2022-11-28T08:03:31Z

@huningxin I updated to fix your last comments, PTAL, thanks.

webnn/concat.https.any.js

webnn/resources/utils.js

BruceDai · 2022-11-29T05:27:14Z

@huningxin Updated, please take another look, thanks.

huningxin

Looks good, thanks @BruceDai !

createContext() / createContextSync() build() / buildSync() compute() / computeSync()

inputDataCategory -> inputDataSource expectedDataCategory -> expectedDataSource

… for loop.

BruceDai · 2023-01-06T05:43:57Z

@dontcallmedom @Honry, I rebased with latest code, all checks have passed, PTAL, thanks.

wpt-pr-bot added webnn wg-s_webmachinelearning labels Nov 2, 2022

wpt-pr-bot assigned dontcallmedom Nov 2, 2022

wpt-pr-bot requested review from Honry and dontcallmedom November 2, 2022 16:28

BruceDai commented Nov 2, 2022

View reviewed changes

BruceDai mentioned this pull request Nov 3, 2022

[WIP][tool] Add scripts to generate test data for WPT WebNN operations tests webmachinelearning/webnn-baseline#22

Closed

BruceDai force-pushed the add_webnn_concat_tests branch 4 times, most recently from 822cb63 to 16985d3 Compare November 3, 2022 13:38

BruceDai mentioned this pull request Nov 3, 2022

WPT tests tracker webmachinelearning/webnn#265

Closed

BruceDai force-pushed the add_webnn_concat_tests branch 2 times, most recently from 7617c56 to a34c3cd Compare November 7, 2022 06:55

fdwr approved these changes Nov 8, 2022

View reviewed changes

Honry approved these changes Nov 9, 2022

View reviewed changes

webnn/concat.https.any.js Outdated Show resolved Hide resolved

BruceDai commented Nov 9, 2022

View reviewed changes

fdwr approved these changes Nov 10, 2022

View reviewed changes

BruceDai force-pushed the add_webnn_concat_tests branch from 7eefda2 to fe02442 Compare November 11, 2022 01:40

BruceDai commented Nov 16, 2022

View reviewed changes

BruceDai force-pushed the add_webnn_concat_tests branch 2 times, most recently from 05e0375 to 9024b21 Compare November 16, 2022 15:59

Honry reviewed Nov 17, 2022

View reviewed changes

fdwr approved these changes Nov 18, 2022

View reviewed changes

BruceDai commented Nov 18, 2022

View reviewed changes

fdwr approved these changes Nov 21, 2022

View reviewed changes

huningxin reviewed Nov 21, 2022

View reviewed changes

BruceDai commented Nov 21, 2022

View reviewed changes

BruceDai commented Nov 26, 2022

View reviewed changes

BruceDai force-pushed the add_webnn_concat_tests branch from 36c1f89 to c9737fc Compare November 28, 2022 03:50

BruceDai force-pushed the add_webnn_concat_tests branch from c9737fc to dabd6b0 Compare November 28, 2022 05:09

huningxin reviewed Nov 28, 2022

View reviewed changes

BruceDai force-pushed the add_webnn_concat_tests branch from dabd6b0 to 75094a4 Compare November 28, 2022 07:52

BruceDai force-pushed the add_webnn_concat_tests branch from 75094a4 to e0ab7ca Compare November 28, 2022 08:35

huningxin reviewed Nov 29, 2022

View reviewed changes

webnn/concat.https.any.js Outdated Show resolved Hide resolved

webnn/resources/utils.js Outdated Show resolved Hide resolved

fdwr approved these changes Nov 29, 2022

View reviewed changes

BruceDai force-pushed the add_webnn_concat_tests branch from e0ab7ca to d17dab8 Compare November 29, 2022 05:21

huningxin approved these changes Nov 29, 2022

View reviewed changes

BruceDai force-pushed the add_webnn_concat_tests branch 2 times, most recently from 35f9b64 to 7767126 Compare November 30, 2022 04:12

BruceDai mentioned this pull request Nov 30, 2022

Add webnn clamp tests #37067

Merged

BruceDai added 9 commits January 6, 2023 13:20

[webnn] Add tests for concat operation.

0784ced

[webnn] Updated concat tests to align with latest WebNN API Spec

afca329

createContext() / createContextSync() build() / buildSync() compute() / computeSync()

[webnn] Optimize concat tests by addressing comments.

b72f3e3

[webnn] Align the usage for sync and async functions.

6bb4588

[webnn] Optimize concat tests with renaming variables.

5d85aef

inputDataCategory -> inputDataSource expectedDataCategory -> expectedDataSource

[webnn] Optimized concat tests by moving if condition checking out of…

23c3f45

… for loop.

[webnn] Refined code to reuse for other operation tests.

ecaf2a3

[webnn] Address comments to move nested functions outside.

f9c75c5

[webnn] More refined code to reuse for other operation tests.

de07f07

BruceDai force-pushed the add_webnn_concat_tests branch from 7767126 to de07f07 Compare January 6, 2023 05:20

Honry merged commit ffcdfb5 into web-platform-tests:master Jan 6, 2023

moz-wptsync-bot added mozilla:gecko-blocked and removed mozilla:gecko-blocked labels Jan 6, 2023


		// https://webmachinelearning.github.io/webnn/#api-mlgraphbuilder-concat

		const concatTests = () => {

[webnn] Add tests for concat operation. #36782

[webnn] Add tests for concat operation. #36782

Uh oh!

Conversation

BruceDai commented Nov 2, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

BruceDai Nov 9, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

fdwr Nov 8, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

fdwr Nov 8, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Honry left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

BruceDai left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

BruceDai commented Nov 10, 2022

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

BruceDai commented Nov 2, 2022 •

edited

Loading

BruceDai Nov 9, 2022 •

edited

Loading

fdwr Nov 8, 2022 •

edited

Loading

fdwr Nov 8, 2022 •

edited

Loading

huningxin left a comment •

edited

Loading