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/reduce_product.https.any.html?cpu - WPT Dashboard Interop Dashboard
- /webnn/conformance_tests/reduce_product.https.any.html?gpu - WPT Dashboard Interop Dashboard
- /webnn/conformance_tests/reduce_product.https.any.html?npu - WPT Dashboard Interop Dashboard
- /webnn/conformance_tests/reduce_product.https.any.worker.html?cpu - WPT Dashboard Interop Dashboard
- /webnn/conformance_tests/reduce_product.https.any.worker.html?gpu - WPT Dashboard Interop Dashboard
- /webnn/conformance_tests/reduce_product.https.any.worker.html?npu - WPT Dashboard Interop Dashboard
// META: title=test WebNN API reduction operations
// META: global=window,dedicatedworker
// META: variant=?cpu
// META: variant=?gpu
// META: variant=?npu
// META: script=../resources/utils.js
// META: timeout=long
'use strict';
// Reduce the input tensor along all dimensions, or along the axes specified in
// the axes array parameter.
//
// dictionary MLReduceOptions {
// sequence<[EnforceRange] unsigned long> axes;
// boolean keepDimensions = false;
// };
//
// MLOperand reduceProduct(MLOperand input, optional MLReduceOptions options
// = {});
const getReductionOperatorsPrecisionTolerance = (graphResources) => {
return {
metricType: 'ULP',
value: getReducedElementCount(graphResources),
};
};
const reduceProductTests = [
{
'name': 'reduceProduct float32 0D constant tensor default options',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [-68.75911712646484],
'descriptor': {shape: [], dataType: 'float32'},
'constant': true
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [{'input': 'reduceProductInput'}],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': -68.75911712646484,
'descriptor': {shape: [], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 0D constant tensor empty axes',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [-68.75911712646484],
'descriptor': {shape: [], dataType: 'float32'},
'constant': true
}
},
'operators': [{
'name': 'reduceProduct',
'arguments':
[{'input': 'reduceProductInput'}, {'options': {'axes': []}}],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': -68.75911712646484,
'descriptor': {shape: [], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 1D constant tensor default options',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [24], dataType: 'float32'},
'constant': true
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [{'input': 'reduceProductInput'}],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': 1.5855958784642327e+37,
'descriptor': {shape: [], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 1D tensor default options',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [24], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [{'input': 'reduceProductInput'}],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': 1.5855958784642327e+37,
'descriptor': {shape: [], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 2D tensor default options',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [4, 6], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [{'input': 'reduceProductInput'}],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': 1.5855958784642327e+37,
'descriptor': {shape: [], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 3D tensor default options',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [2, 3, 4], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [{'input': 'reduceProductInput'}],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': 1.5855958784642327e+37,
'descriptor': {shape: [], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 4D tensor default options',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [2, 2, 2, 3], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [{'input': 'reduceProductInput'}],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': 1.5855958784642327e+37,
'descriptor': {shape: [], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 5D tensor default options',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [2, 1, 4, 1, 3], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [{'input': 'reduceProductInput'}],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': 1.5855958784642327e+37,
'descriptor': {shape: [], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 3D tensor options.axes',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [2, 3, 4], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments':
[{'input': 'reduceProductInput'}, {'options': {'axes': [2]}}],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': [
7519895, -1292816.375, 2441721.75, -110637.7734375, -7380313.5,
-818030.5
],
'descriptor': {shape: [2, 3], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 4D tensor options.axes',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [2, 2, 2, 3], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments':
[{'input': 'reduceProductInput'}, {'options': {'axes': [0, 2]}}],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': [
4227263.5, -446960.5625, 3811296.75, 1280298.5, -1343475.375,
1280118.75
],
'descriptor': {shape: [2, 3], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 3D tensor options.keepDimensions=false',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [2, 3, 4], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [
{'input': 'reduceProductInput'},
{'options': {'keepDimensions': false}}
],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': 1.5855958784642327e+37,
'descriptor': {shape: [], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 3D tensor options.keepDimensions=true',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [2, 3, 4], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [
{'input': 'reduceProductInput'}, {'options': {'keepDimensions': true}}
],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': [1.5855958784642327e+37],
'descriptor': {shape: [1, 1, 1], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 4D tensor options.keepDimensions=false',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [2, 2, 2, 3], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [
{'input': 'reduceProductInput'},
{'options': {'keepDimensions': false}}
],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': 1.5855958784642327e+37,
'descriptor': {shape: [], dataType: 'float32'}
}
}
}
},
{
'name': 'reduceProduct float32 4D tensor options.keepDimensions=true',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [2, 2, 2, 3], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [
{'input': 'reduceProductInput'}, {'options': {'keepDimensions': true}}
],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': [1.5855958784642327e+37],
'descriptor': {shape: [1, 1, 1, 1], dataType: 'float32'}
}
}
}
},
{
'name':
'reduceProduct float32 4D tensor options.axes with options.keepDimensions=false',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [2, 2, 2, 3], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [
{'input': 'reduceProductInput'},
{'options': {'axes': [1, 3], 'keepDimensions': false}}
],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': [-3638925568, 6523364352, -414643360, 1610916352],
'descriptor': {shape: [2, 2], dataType: 'float32'}
}
}
}
},
{
'name':
'reduceProduct float32 4D tensor options.axes with options.keepDimensions=true',
'graph': {
'inputs': {
'reduceProductInput': {
'data': [
-68.75911712646484, 99.44961547851562, 24.86055564880371,
-44.23515319824219, -22.699743270874023, 79.97555541992188,
14.4650239944458, 49.23109436035156, 30.058706283569336,
69.45106506347656, -20.15709686279297, -58.0255126953125,
51.896610260009766, -2.020799160003662, 39.392974853515625,
26.78073501586914, -97.97651672363281, 48.66154479980469,
-85.19523620605469, -18.16986083984375, 64.83759307861328,
-14.95883846282959, -74.50932312011719, -11.319679260253906
],
'descriptor': {shape: [2, 2, 2, 3], dataType: 'float32'}
}
},
'operators': [{
'name': 'reduceProduct',
'arguments': [
{'input': 'reduceProductInput'},
{'options': {'axes': [1, 3], 'keepDimensions': true}}
],
'outputs': 'reduceProductOutput'
}],
'expectedOutputs': {
'reduceProductOutput': {
'data': [-3638925568, 6523364352, -414643360, 1610916352],
'descriptor': {shape: [2, 1, 2, 1], dataType: 'float32'}
}
}
}
}
];
if (navigator.ml) {
reduceProductTests.forEach((test) => {
webnn_conformance_test(
buildAndExecuteGraph, getReductionOperatorsPrecisionTolerance, test);
});
} else {
test(() => assert_implements(navigator.ml, 'missing navigator.ml'));
}