webnn/validation_tests/matmul.https.any.js

// META: title=validation tests for WebNN API matmul operation
// META: global=window
// META: variant=?cpu
// META: variant=?gpu
// META: variant=?npu
// META: script=../resources/utils_validation.js

'use strict';

const label = 'matmul_123';
validateTwoInputsFromMultipleBuilders('matmul');
validateTwoBroadcastableInputsTensorLimit('matmul', label);

const tests = [
  {
    name: '[matmul] Throw if first input\'s rank is less than 2',
    inputs: {
      a: {dataType: 'float32', shape: [2]},
      b: {dataType: 'float32', shape: [2, 2]}
    }
  },
  {
    name: '[matmul] Throw if second input\'s rank is less than 2',
    inputs: {
      a: {dataType: 'float32', shape: [2, 2]},
      b: {dataType: 'float32', shape: [2]}
    }
  },
  {
    name: '[matmul] Test with 2-D input and 4-D input',
    inputs: {
      a: {dataType: 'float32', shape: [1, 4]},
      b: {dataType: 'float32', shape: [2, 2, 4, 2]}
    },
    output: {dataType: 'float32', shape: [2, 2, 1, 2]}
  },
  {
    name: '[matmul] Test with 2-D input and 2-D input',
    inputs: {
      a: {dataType: 'float32', shape: [4, 2]},
      b: {dataType: 'float32', shape: [2, 3]}
    },
    output: {dataType: 'float32', shape: [4, 3]}
  },
  {
    // batchShape is a clone of inputShape with the spatial dimensions
    // (last 2 items) removed.
    name:
        '[matmul] Test with 3-D input and 3-D input of broadcastable batchShape',
    inputs: {
      a: {dataType: 'float32', shape: [2, 3, 4]},
      b: {dataType: 'float32', shape: [1, 4, 1]}
    },
    output: {dataType: 'float32', shape: [2, 3, 1]}
  },
  {
    // batchShape is a clone of inputShape with the spatial dimensions
    // (last 2 items) removed.
    name:
        '[matmul] Test with 4-D input and 3-D input of broadcastable batchShape',
    inputs: {
      a: {dataType: 'float32', shape: [2, 2, 3, 4]},
      b: {dataType: 'float32', shape: [1, 4, 5]}
    },
    output: {dataType: 'float32', shape: [2, 2, 3, 5]}
  },
  {
    name: '[matmul] Test with 3-D input and 3-D input',
    inputs: {
      a: {dataType: 'float32', shape: [2, 3, 4]},
      b: {dataType: 'float32', shape: [2, 4, 5]}
    },
    output: {dataType: 'float32', shape: [2, 3, 5]}
  },
  {
    name: '[matmul] Throw if the input data type is not floating point',
    inputs: {
      a: {dataType: 'uint32', shape: [2, 3, 4]},
      b: {dataType: 'uint32', shape: [2, 4, 5]}
    }
  },
  {
    name: '[matmul] Throw if data type of two inputs don\'t match',
    inputs: {
      a: {dataType: 'float32', shape: [2, 3, 4]},
      b: {dataType: 'float16', shape: [2, 4, 5]}
    }
  },
  {
    name:
        '[matmul] Throw if columns of first input\'s shape doesn\'t match the rows of second input\'s shape',
    inputs: {
      a: {dataType: 'float32', shape: /* [rows, columns] */[2, 3]},
      b: {dataType: 'float32', shape: /* [rows, columns] */[2, 4]}
    },
  },
  {
    // batchShape is a clone of inputShape with the spatial dimensions
    // (last 2 items) removed.
    name: '[matmul] Throw if batchShapes aren\'t bidirectionally broadcastable',
    inputs: {
      a: {dataType: 'float32', shape: [3, 3, 4]},
      b: {dataType: 'float32', shape: [2, 4, 1]}
    },
  },
];

tests.forEach(test => promise_test(async t => {
                const builder = new MLGraphBuilder(context);
                const inputA = builder.input('a', test.inputs.a);
                const inputB = builder.input('b', test.inputs.b);
                if (test.output) {
                  const output = builder.matmul(inputA, inputB);
                  assert_equals(output.dataType, test.output.dataType);
                  assert_array_equals(output.shape, test.output.shape);
                } else {
                  const options = {label};
                  const regrexp = new RegExp('\\[' + label + '\\]');
                  assert_throws_with_label(
                      () => builder.matmul(inputA, inputB, options), regrexp);
                }
              }, test.name));