| |
| <!doctype html> |
| <meta charset=utf-8> |
| <meta name="timeout" content="long"> |
| <title>test conv2d + relu subgraph</title> |
| <script src="/resources/testharness.js"></script> |
| <script src="/resources/testharnessreport.js"></script> |
| <script src="../resources/libs/numpy.js"></script> |
| <script src="../resources/utils.js"></script> |
| <script> |
| let context; |
| let builder; |
| |
| promise_setup(async () => { |
| context = await navigator.ml.createContext(); |
| builder = new MLGraphBuilder(context); |
| }, {explicit_timeout: true, test_timeout: 180000}); |
| |
| const tests = loadTests('/webnn/resources/test_data/conv2d_relu/conv2d_relu.json'); |
| for (let test of tests) { |
| promise_test(async () => { |
| console.time(`run ${test.name} test`); |
| console.time(`load and build for ${test.name} test`) |
| let inputInfo = await readFromNpy('/webnn/resources/test_data/' + test.inputs.input.data, test.inputs.input.type); |
| let input = builder.input( |
| 'input', {type: test.inputs.input.type, dimensions: test.inputs.input.shape}); |
| let weightsInfo = await readFromNpy('/webnn/resources/test_data/' + test.inputs.filter.data, test.inputs.filter.type); |
| let filter = builder.constant( |
| {type: test.inputs.filter.type, dimensions: test.inputs.filter.shape}, weightsInfo.buffer); |
| let conv2dOptions = {...test.options}; |
| let biasInfo = await readFromNpy('/webnn/resources/test_data/' + test.options.bias.data, test.options.bias.type); |
| let bias = builder.constant( |
| {type: test.options.bias.type, dimensions: test.options.bias.shape}, biasInfo.buffer); |
| conv2dOptions.bias = bias; |
| let middleOut = builder.conv2d(input, filter, conv2dOptions); |
| let out = builder.relu(middleOut); |
| console.time(`build for ${test.name} test`); |
| let graph = await builder.build({'output': out}); |
| console.timeEnd(`build for ${test.name} test`); |
| console.timeEnd(`load and build for ${test.name} test`); |
| |
| let expectedInfo = await readFromNpy('/webnn/resources/test_data/' + test.expected.data, test.expected.type); |
| let outputBuffer = new TypedArrayDict[test.expected.type](sizeOfShape(test.expected.shape)); |
| let inputs = {'input': inputInfo.buffer}; |
| let outputs = {'output': outputBuffer}; |
| console.time(`compute for ${test.name} test`) |
| const results = await context.compute(graph, inputs, outputs); |
| console.timeEnd(`compute for ${test.name} test`) |
| const tolerence = inputInfo.dimensions[3] * weightsInfo.dimensions[1] * weightsInfo.dimensions[2] * 2; |
| console.time(`compare output with baseline ${test.name} test`); |
| assert_array_approx_equals_ulp(results.outputs.output, expectedInfo.buffer, tolerence, test.expected.type, `test ${test.name}`); |
| console.timeEnd(`compare output with baseline ${test.name} test`); |
| console.timeEnd(`run ${test.name} test`); |
| }, test.name); |
| } |
| |
| </script> |
