Source code
Revision control
Copy as Markdown
Other Tools
Test Info:
- This WPT test may be referenced by the following Test IDs:
- /webnn/conformance_tests/cos.https.any.html?cpu - WPT Dashboard Interop Dashboard
- /webnn/conformance_tests/cos.https.any.html?gpu - WPT Dashboard Interop Dashboard
- /webnn/conformance_tests/cos.https.any.html?npu - WPT Dashboard Interop Dashboard
- /webnn/conformance_tests/cos.https.any.worker.html?cpu - WPT Dashboard Interop Dashboard
- /webnn/conformance_tests/cos.https.any.worker.html?gpu - WPT Dashboard Interop Dashboard
- /webnn/conformance_tests/cos.https.any.worker.html?npu - WPT Dashboard Interop Dashboard
// META: title=test WebNN API element-wise cos operation
// META: global=window,dedicatedworker
// META: variant=?cpu
// META: variant=?gpu
// META: variant=?npu
// META: script=../resources/utils.js
// META: timeout=long
'use strict';
// Compute the cosine of the input tensor, element-wise.
//
// MLOperand cos(MLOperand input);
const getCosPrecisionTolerance = (graphResources) => {
const toleranceValueDict = {float32: 1 / 1024, float16: 1 / 512};
const expectedDataType =
getExpectedDataTypeOfSingleOutput(graphResources.expectedOutputs);
return {metricType: 'ATOL', value: toleranceValueDict[expectedDataType]};
};
const cosTests = [
{
'name': 'cos float32 0D scalar',
'graph': {
'inputs': {
'cosInput': {
'data': [85.56369018554688],
'descriptor': {shape: [], dataType: 'float32'}
}
},
'operators': [{
'name': 'cos',
'arguments': [{'input': 'cosInput'}],
'outputs': 'cosOutput'
}],
'expectedOutputs': {
'cosOutput': {
'data': [-0.7380040884017944],
'descriptor': {shape: [], dataType: 'float32'}
}
}
}
},
{
'name': 'cos float32 1D constant tensor',
'graph': {
'inputs': {
'cosInput': {
'data': [
85.56369018554688, -45.09983825683594, -94.67750549316406,
83.49029541015625, -31.367904663085938, 70.18042755126953,
-90.36229705810547, -83.00758361816406, 61.51649475097656,
-32.51877975463867, -48.3765869140625, -58.02735900878906,
89.79197692871094, -84.53326416015625, -58.23252487182617,
-76.14168548583984, -59.058876037597656, 77.38546752929688,
-98.67289733886719, -63.6115608215332, 26.85724639892578,
83.70417022705078, 76.56607055664062, -47.83436584472656
],
'descriptor': {shape: [24], dataType: 'float32'},
'constant': true
}
},
'operators': [{
'name': 'cos',
'arguments': [{'input': 'cosInput'}],
'outputs': 'cosOutput'
}],
'expectedOutputs': {
'cosOutput': {
'data': [
-0.7380040884017944, 0.43789437413215637, 0.9090799689292908,
-0.23584702610969543, 0.9988471865653992, 0.48416373133659363,
-0.7358400821685791, 0.24218930304050446, 0.25266921520233154,
0.4510514736175537, -0.31276169419288635, 0.09197491407394409,
-0.2537800967693329, -0.9583188891410828, -0.11282006651163101,
0.736129879951477, -0.80721116065979, -0.4045141637325287,
-0.283336341381073, 0.7111190557479858, -0.1531042903661728,
-0.43673399090766907, 0.39213326573371887, -0.7580515146255493
],
'descriptor': {shape: [24], dataType: 'float32'}
}
}
}
},
{
'name': 'cos float32 1D tensor',
'graph': {
'inputs': {
'cosInput': {
'data': [
85.56369018554688, -45.09983825683594, -94.67750549316406,
83.49029541015625, -31.367904663085938, 70.18042755126953,
-90.36229705810547, -83.00758361816406, 61.51649475097656,
-32.51877975463867, -48.3765869140625, -58.02735900878906,
89.79197692871094, -84.53326416015625, -58.23252487182617,
-76.14168548583984, -59.058876037597656, 77.38546752929688,
-98.67289733886719, -63.6115608215332, 26.85724639892578,
83.70417022705078, 76.56607055664062, -47.83436584472656
],
'descriptor': {shape: [24], dataType: 'float32'}
}
},
'operators': [{
'name': 'cos',
'arguments': [{'input': 'cosInput'}],
'outputs': 'cosOutput'
}],
'expectedOutputs': {
'cosOutput': {
'data': [
-0.7380040884017944, 0.43789437413215637, 0.9090799689292908,
-0.23584702610969543, 0.9988471865653992, 0.48416373133659363,
-0.7358400821685791, 0.24218930304050446, 0.25266921520233154,
0.4510514736175537, -0.31276169419288635, 0.09197491407394409,
-0.2537800967693329, -0.9583188891410828, -0.11282006651163101,
0.736129879951477, -0.80721116065979, -0.4045141637325287,
-0.283336341381073, 0.7111190557479858, -0.1531042903661728,
-0.43673399090766907, 0.39213326573371887, -0.7580515146255493
],
'descriptor': {shape: [24], dataType: 'float32'}
}
}
}
},
{
'name': 'cos float32 2D tensor',
'graph': {
'inputs': {
'cosInput': {
'data': [
85.56369018554688, -45.09983825683594, -94.67750549316406,
83.49029541015625, -31.367904663085938, 70.18042755126953,
-90.36229705810547, -83.00758361816406, 61.51649475097656,
-32.51877975463867, -48.3765869140625, -58.02735900878906,
89.79197692871094, -84.53326416015625, -58.23252487182617,
-76.14168548583984, -59.058876037597656, 77.38546752929688,
-98.67289733886719, -63.6115608215332, 26.85724639892578,
83.70417022705078, 76.56607055664062, -47.83436584472656
],
'descriptor': {shape: [4, 6], dataType: 'float32'}
}
},
'operators': [{
'name': 'cos',
'arguments': [{'input': 'cosInput'}],
'outputs': 'cosOutput'
}],
'expectedOutputs': {
'cosOutput': {
'data': [
-0.7380040884017944, 0.43789437413215637, 0.9090799689292908,
-0.23584702610969543, 0.9988471865653992, 0.48416373133659363,
-0.7358400821685791, 0.24218930304050446, 0.25266921520233154,
0.4510514736175537, -0.31276169419288635, 0.09197491407394409,
-0.2537800967693329, -0.9583188891410828, -0.11282006651163101,
0.736129879951477, -0.80721116065979, -0.4045141637325287,
-0.283336341381073, 0.7111190557479858, -0.1531042903661728,
-0.43673399090766907, 0.39213326573371887, -0.7580515146255493
],
'descriptor': {shape: [4, 6], dataType: 'float32'}
}
}
}
},
{
'name': 'cos float32 3D tensor',
'graph': {
'inputs': {
'cosInput': {
'data': [
85.56369018554688, -45.09983825683594, -94.67750549316406,
83.49029541015625, -31.367904663085938, 70.18042755126953,
-90.36229705810547, -83.00758361816406, 61.51649475097656,
-32.51877975463867, -48.3765869140625, -58.02735900878906,
89.79197692871094, -84.53326416015625, -58.23252487182617,
-76.14168548583984, -59.058876037597656, 77.38546752929688,
-98.67289733886719, -63.6115608215332, 26.85724639892578,
83.70417022705078, 76.56607055664062, -47.83436584472656
],
'descriptor': {shape: [2, 3, 4], dataType: 'float32'}
}
},
'operators': [{
'name': 'cos',
'arguments': [{'input': 'cosInput'}],
'outputs': 'cosOutput'
}],
'expectedOutputs': {
'cosOutput': {
'data': [
-0.7380040884017944, 0.43789437413215637, 0.9090799689292908,
-0.23584702610969543, 0.9988471865653992, 0.48416373133659363,
-0.7358400821685791, 0.24218930304050446, 0.25266921520233154,
0.4510514736175537, -0.31276169419288635, 0.09197491407394409,
-0.2537800967693329, -0.9583188891410828, -0.11282006651163101,
0.736129879951477, -0.80721116065979, -0.4045141637325287,
-0.283336341381073, 0.7111190557479858, -0.1531042903661728,
-0.43673399090766907, 0.39213326573371887, -0.7580515146255493
],
'descriptor': {shape: [2, 3, 4], dataType: 'float32'}
}
}
}
},
{
'name': 'cos float32 4D tensor',
'graph': {
'inputs': {
'cosInput': {
'data': [
85.56369018554688, -45.09983825683594, -94.67750549316406,
83.49029541015625, -31.367904663085938, 70.18042755126953,
-90.36229705810547, -83.00758361816406, 61.51649475097656,
-32.51877975463867, -48.3765869140625, -58.02735900878906,
89.79197692871094, -84.53326416015625, -58.23252487182617,
-76.14168548583984, -59.058876037597656, 77.38546752929688,
-98.67289733886719, -63.6115608215332, 26.85724639892578,
83.70417022705078, 76.56607055664062, -47.83436584472656
],
'descriptor': {shape: [2, 2, 2, 3], dataType: 'float32'}
}
},
'operators': [{
'name': 'cos',
'arguments': [{'input': 'cosInput'}],
'outputs': 'cosOutput'
}],
'expectedOutputs': {
'cosOutput': {
'data': [
-0.7380040884017944, 0.43789437413215637, 0.9090799689292908,
-0.23584702610969543, 0.9988471865653992, 0.48416373133659363,
-0.7358400821685791, 0.24218930304050446, 0.25266921520233154,
0.4510514736175537, -0.31276169419288635, 0.09197491407394409,
-0.2537800967693329, -0.9583188891410828, -0.11282006651163101,
0.736129879951477, -0.80721116065979, -0.4045141637325287,
-0.283336341381073, 0.7111190557479858, -0.1531042903661728,
-0.43673399090766907, 0.39213326573371887, -0.7580515146255493
],
'descriptor': {shape: [2, 2, 2, 3], dataType: 'float32'}
}
}
}
},
{
'name': 'cos float32 5D tensor',
'graph': {
'inputs': {
'cosInput': {
'data': [
85.56369018554688, -45.09983825683594, -94.67750549316406,
83.49029541015625, -31.367904663085938, 70.18042755126953,
-90.36229705810547, -83.00758361816406, 61.51649475097656,
-32.51877975463867, -48.3765869140625, -58.02735900878906,
89.79197692871094, -84.53326416015625, -58.23252487182617,
-76.14168548583984, -59.058876037597656, 77.38546752929688,
-98.67289733886719, -63.6115608215332, 26.85724639892578,
83.70417022705078, 76.56607055664062, -47.83436584472656
],
'descriptor': {shape: [2, 1, 4, 1, 3], dataType: 'float32'}
}
},
'operators': [{
'name': 'cos',
'arguments': [{'input': 'cosInput'}],
'outputs': 'cosOutput'
}],
'expectedOutputs': {
'cosOutput': {
'data': [
-0.7380040884017944, 0.43789437413215637, 0.9090799689292908,
-0.23584702610969543, 0.9988471865653992, 0.48416373133659363,
-0.7358400821685791, 0.24218930304050446, 0.25266921520233154,
0.4510514736175537, -0.31276169419288635, 0.09197491407394409,
-0.2537800967693329, -0.9583188891410828, -0.11282006651163101,
0.736129879951477, -0.80721116065979, -0.4045141637325287,
-0.283336341381073, 0.7111190557479858, -0.1531042903661728,
-0.43673399090766907, 0.39213326573371887, -0.7580515146255493
],
'descriptor': {shape: [2, 1, 4, 1, 3], dataType: 'float32'}
}
}
}
}
];
if (navigator.ml) {
cosTests.forEach((test) => {
webnn_conformance_test(
buildAndExecuteGraph, getCosPrecisionTolerance, test);
});
} else {
test(() => assert_implements(navigator.ml, 'missing navigator.ml'));
}