keep dimensions

code/samples/matmul.js

@@ -2,8 +2,16 @@
 const context = await navigator.ml.createContext({deviceType: 'gpu'});
 const builder = new MLGraphBuilder(context);
 // Step 1: Create a computational graph calculating `c = a * b`.
-const a = builder.input('a', {dataType: 'float32', shape: [3, 4]});
-const b = builder.input('b', {dataType: 'float32', shape: [4, 3]});
+const a = builder.input('a', {
+  dataType: 'float32',
+  dimensions: [3, 4],
+  shape: [3, 4],
+});
+const b = builder.input('b', {
+  dataType: 'float32',
+  dimensions: [4, 3],
+  shape: [4, 3],
+});
 const c = builder.matmul(a, b);
 // Step 2: Compile it into an executable graph.
 const graph = await builder.build({c});

code/samples/mul_add.js

@@ -1,4 +1,4 @@
-const operandType = {dataType: 'float32', shape: [2, 2]};
+const operandType = {dataType: 'float32', dimensions: [2, 2], shape: [2, 2]};
 const context = await navigator.ml.createContext();
 const builder = new MLGraphBuilder(context);
 // 1. Create a computational graph 'C = 0.2 * A + B'.

code/samples/simple_graph.js

@@ -18,7 +18,7 @@ const TENSOR_SIZE = 8;
 const builder = new MLGraphBuilder(context);
 
 // Create MLOperandDescriptor object.
-const desc = {dataType: 'float32', shape: TENSOR_DIMS};
+const desc = {dataType: 'float32', dimensions: TENSOR_DIMS, shape: TENSOR_DIMS};
 
 // constant1 is a constant MLOperand with the value 0.5.
 const constantBuffer1 = new Float32Array(TENSOR_SIZE).fill(0.5);

common/utils.js

@@ -121,7 +121,8 @@ export async function buildConstantByNpy(builder, url, targetType = 'float32') {
     throw new Error(`Conversion from ${npArray.dataType} ` +
         `to ${targetType} is not supported.`);
   }
-  return builder.constant({dataType: type, shape}, typedArray);
+  return builder.constant(
+      {dataType: type, dimensions: shape, shape}, typedArray);
 }
 
 // Convert video frame to a canvas element

face_recognition/facenet_nchw.js

@@ -140,6 +140,7 @@ export class FaceNetNchw {
     this.builder_ = new MLGraphBuilder(this.context_);
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
 

face_recognition/facenet_nhwc.js

@@ -141,6 +141,7 @@ export class FaceNetNhwc {
     this.builder_ = new MLGraphBuilder(this.context_);
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
 

facial_landmark_detection/face_landmark_nchw.js

@@ -71,6 +71,7 @@ export class FaceLandmarkNchw {
     this.builder_ = new MLGraphBuilder(this.context_);
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
 

facial_landmark_detection/face_landmark_nhwc.js

@@ -72,6 +72,7 @@ export class FaceLandmarkNhwc {
     this.builder_ = new MLGraphBuilder(this.context_);
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
 

facial_landmark_detection/ssd_mobilenetv2_face_nchw.js

@@ -115,6 +115,7 @@ ${nameArray[1]}`;
     this.builder_ = new MLGraphBuilder(this.context_);
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
 

facial_landmark_detection/ssd_mobilenetv2_face_nhwc.js

@@ -127,6 +127,7 @@ ${nameArray[1]}`;
     this.builder_ = new MLGraphBuilder(this.context_);
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
 

image_classification/efficientnet_fp16_nchw.js

@@ -77,6 +77,7 @@ export class EfficientNetFP16Nchw {
     this.builder_ = new MLGraphBuilder(this.context_);
     let data = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     data = this.builder_.cast(data, 'float16');

image_classification/mobilenet_nchw.js

@@ -91,6 +91,7 @@ export class MobileNetV2Nchw {
     this.builder_ = new MLGraphBuilder(this.context_);
     let data = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     if (this.dataType_ === 'float16') {

image_classification/mobilenet_nhwc.js

@@ -89,6 +89,7 @@ export class MobileNetV2Nhwc {
     const filterLayout = 'ohwi';
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     const conv0 = this.buildConv_(

image_classification/resnet50v1_fp16_nchw.js

@@ -78,6 +78,7 @@ export class ResNet50V1FP16Nchw {
     this.builder_ = new MLGraphBuilder(this.context_);
     let data = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     data = this.builder_.cast(data, 'float16');

image_classification/resnet50v2_nchw.js

@@ -100,6 +100,7 @@ export class ResNet50V2Nchw {
     this.builder_ = new MLGraphBuilder(this.context_);
     const data = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     const bn1 = this.buildBatchNorm_(data, '0', '', false);

image_classification/resnet50v2_nhwc.js

@@ -122,6 +122,7 @@ export class ResNet50V2Nhwc {
     this.builder_ = new MLGraphBuilder(this.context_);
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     const conv1 = await this.buildConv_(

image_classification/squeezenet_nchw.js

@@ -45,6 +45,7 @@ export class SqueezeNetNchw {
     this.builder_ = new MLGraphBuilder(this.context_);
     const data = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     const conv0 = this.buildConv_(data, 'conv0', {strides: [2, 2]});

image_classification/squeezenet_nhwc.js

@@ -56,6 +56,7 @@ export class SqueezeNetNhwc {
     const layout = 'nhwc';
     const placeholder = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     const conv1 = this.buildConv_(

lenet/lenet.js

@@ -27,6 +27,7 @@ export class LeNet {
     const inputShape = /* nchw */ [1, 1, 28, 28];
     let input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: inputShape,
       shape: inputShape,
     });
 
@@ -50,7 +51,11 @@ export class LeNet {
               conv1FilterData, conv1FitlerShape, this.oihwToOhwiPermutation_);
     }
     const conv1Filter = this.builder_.constant(
-        {dataType: 'float32', shape: conv1FitlerShape},
+        {
+          dataType: 'float32',
+          dimensions: conv1FitlerShape,
+          shape: conv1FitlerShape,
+        },
         conv1FilterData);
     byteOffset +=
         sizeOfShape(conv1FitlerShape) * Float32Array.BYTES_PER_ELEMENT;
@@ -59,7 +64,7 @@ export class LeNet {
     const add1BiasData =
         new Float32Array(arrayBuffer, byteOffset, sizeOfShape(add1BiasShape));
     const add1Bias = this.builder_.constant(
-        {dataType: 'float32', shape: add1BiasShape},
+        {dataType: 'float32', dimensions: add1BiasShape, shape: add1BiasShape},
         add1BiasData,
     );
     byteOffset += sizeOfShape(add1BiasShape) * Float32Array.BYTES_PER_ELEMENT;
@@ -87,14 +92,17 @@ export class LeNet {
               conv2FilterData, conv2FilterShape, this.oihwToOhwiPermutation_);
     }
     const conv2Filter = this.builder_.constant(
-        {dataType: 'float32', shape: conv2FilterShape},
+        {
+          dataType: 'float32',
+          dimensions: conv2FilterShape,
+          shape: conv2FilterShape},
         conv2FilterData);
     byteOffset +=
         sizeOfShape(conv2FilterShape) * Float32Array.BYTES_PER_ELEMENT;
 
     const add2BiasShape = [50];
     const add2Bias = this.builder_.constant(
-        {dataType: 'float32', shape: add2BiasShape},
+        {dataType: 'float32', dimensions: add2BiasShape, shape: add2BiasShape},
         new Float32Array(arrayBuffer, byteOffset, sizeOfShape(add2BiasShape)));
     byteOffset += sizeOfShape(add2BiasShape) * Float32Array.BYTES_PER_ELEMENT;
     conv2Options.bias = add2Bias;
@@ -120,15 +128,15 @@ export class LeNet {
 
     const matmul1Shape = [500, 800];
     const matmul1Weights = this.builder_.constant(
-        {dataType: 'float32', shape: matmul1Shape},
+        {dataType: 'float32', dimensions: matmul1Shape, shape: matmul1Shape},
         new Float32Array(arrayBuffer, byteOffset, sizeOfShape(matmul1Shape)));
     byteOffset += sizeOfShape(matmul1Shape) * Float32Array.BYTES_PER_ELEMENT;
     const matmul1WeightsTransposed = this.builder_.transpose(matmul1Weights);
     const matmul1 = this.builder_.gemm(reshape1, matmul1WeightsTransposed);
 
     const add3BiasShape = [1, 500];
     const add3Bias = this.builder_.constant(
-        {dataType: 'float32', shape: add3BiasShape},
+        {dataType: 'float32', dimensions: add3BiasShape, shape: add3BiasShape},
         new Float32Array(arrayBuffer, byteOffset, sizeOfShape(add3BiasShape)));
     byteOffset += sizeOfShape(add3BiasShape) * Float32Array.BYTES_PER_ELEMENT;
     const add3 = this.builder_.add(matmul1, add3Bias);
@@ -140,15 +148,15 @@ export class LeNet {
 
     const matmul2Shape = [10, 500];
     const matmul2Weights = this.builder_.constant(
-        {dataType: 'float32', shape: matmul2Shape},
+        {dataType: 'float32', dimensions: matmul2Shape, shape: matmul2Shape},
         new Float32Array(arrayBuffer, byteOffset, sizeOfShape(matmul2Shape)));
     byteOffset += sizeOfShape(matmul2Shape) * Float32Array.BYTES_PER_ELEMENT;
     const matmul2WeightsTransposed = this.builder_.transpose(matmul2Weights);
     const matmul2 = this.builder_.gemm(reshape2, matmul2WeightsTransposed);
 
     const add4BiasShape = [1, 10];
     const add4Bias = this.builder_.constant(
-        {dataType: 'float32', shape: add4BiasShape},
+        {dataType: 'float32', dimensions: add4BiasShape, shape: add4BiasShape},
         new Float32Array(arrayBuffer, byteOffset, sizeOfShape(add4BiasShape)));
     const add4 = this.builder_.add(matmul2, add4Bias);
 

nnotepad/js/index.js

@@ -130,6 +130,7 @@ function explain(outputs) {
       .map((output) =>
         [
           'dataType: ' + output.dataType,
+          'dimensions: ' + Util.stringify(output.shape),
           'shape: ' + Util.stringify(output.shape),
           'tensor: ' + dumpTensor(output.shape, output.buffer, 8),
         ].join('\n'),

nnotepad/js/nnotepad.js

@@ -475,7 +475,8 @@ export class NNotepad {
         });
       }(tensor, 0));
       const ctor = WebNNUtil.dataTypeToBufferType(dataType);
-      return `_.constant({dataType: "${dataType}", shape: ${
+      return `_.constant({dataType: "${dataType}", dimensions: ${
+        Util.stringify(shape)}}, shape: ${
         Util.stringify(shape)}}, new ${ctor.name}([${
         elements.map((n) => Util.stringifyNumber(n, dataType)).join(',')}]))`;
     }
@@ -500,7 +501,8 @@ export class NNotepad {
         }
         const dims = shape.value.map((expr) => expr.value);
         const ctor = WebNNUtil.dataTypeToBufferType(dataType.value);
-        return `_.constant({dataType: "${dataType.value}", shape: ${
+        return `_.constant({dataType: "${dataType.value}", dimensions: ${
+          Util.stringify(dims)}}, shape: ${
           Util.stringify(dims)}}, new ${
           ctor.name}(await Util.loadBuffer(${Util.stringify(url.value)})))`;
       }
@@ -516,7 +518,8 @@ export class NNotepad {
         const dims = shape.value.map((expr) => expr.value);
         const ctor = WebNNUtil.dataTypeToBufferType(dataType.value);
         const len = dims.reduce((a, b) => a * b, 1);
-        return `_.constant({dataType: "${dataType.value}", shape: ${
+        return `_.constant({dataType: "${dataType.value}", dimensions: ${
+          Util.stringify(dims)}}, shape: ${
           Util.stringify(dims)}}, new ${
           ctor.name}(${len}))`;
       }
@@ -595,6 +598,7 @@ export class NNotepad {
     return outputOperands.map(
         (op, index) => ({
           dataType: op.dataType(),
+          dimensions: op.shape(),
           shape: op.shape(),
           buffer: maybeProxyForFloat16Array(result.outputs[`output-${index}`]),
         }));

nsnet2/nsnet2.js

@@ -38,12 +38,14 @@ export class NSNet2 {
     // Build up the network.
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: [batchSize, frames, this.frameSize],
       shape: [batchSize, frames, this.frameSize],
     });
     const relu20 = this.builder_.relu(this.builder_.add(this.builder_.matmul(input, weight172), biasFcIn0));
     const transpose31 = this.builder_.transpose(relu20, {permutation: [1, 0, 2]});
     const initialState92 = this.builder_.input('initialState92', {
       dataType: 'float32',
+      dimensions: [1, batchSize, this.hiddenSize],
       shape: [1, batchSize, this.hiddenSize],
     });
     const [gru94, gru93] = this.builder_.gru(transpose31, weight192, recurrentWeight193, frames, this.hiddenSize,
@@ -54,6 +56,7 @@ export class NSNet2 {
     const squeeze95 = this.builder_.reshape(gru93, squeeze95Shape);
     const initialState155 = this.builder_.input('initialState155', {
       dataType: 'float32',
+      dimensions: [1, batchSize, this.hiddenSize],
       shape: [1, batchSize, this.hiddenSize],
     });
     const [gru157, gru156] = this.builder_.gru(squeeze95, weight212, recurrentWeight213, frames, this.hiddenSize,

object_detection/ssd_mobilenetv1_nchw.js

@@ -81,6 +81,7 @@ ${nameArray[1]}_BatchNorm_batchnorm`;
     this.builder_ = new MLGraphBuilder(this.context_);
     let input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     if (this.targetDataType_ === 'float16') {

object_detection/ssd_mobilenetv1_nhwc.js

@@ -87,6 +87,7 @@ ${nameArray[1]}_BatchNorm_batchnorm`;
     this.builder_ = new MLGraphBuilder(this.context_);
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     const strides = [2, 2];

object_detection/tiny_yolov2_nchw.js

@@ -63,10 +63,11 @@ export class TinyYoloV2Nchw {
     this.builder_ = new MLGraphBuilder(this.context_);
     let image = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     let mulScale = this.builder_.constant(
-        {dataType: 'float32', shape: [1]},
+        {dataType: 'float32', dimensions: [1], shape: [1]},
         new Float32Array([0.003921568859368563]),
     );
     const poolOptions = {

object_detection/tiny_yolov2_nhwc.js

@@ -57,6 +57,7 @@ export class TinyYoloV2Nhwc {
     this.builder_ = new MLGraphBuilder(this.context_);
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
 

rnnoise/rnnoise.js

@@ -53,6 +53,7 @@ export class RNNoise {
     // Build up the network.
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: [this.batchSize_, this.frames_, this.featureSize],
       shape: [this.batchSize_, this.frames_, this.featureSize],
     });
     const inputDense0 = this.builder_.matmul(input, inputDenseKernel0);
@@ -68,6 +69,7 @@ export class RNNoise {
         [1, 3 * this.vadGruHiddenSize]);
     const vadGruInitialH = this.builder_.input('vadGruInitialH', {
       dataType: 'float32',
+      dimensions: [1, this.batchSize_, this.vadGruHiddenSize],
       shape: [1, this.batchSize_, this.vadGruHiddenSize],
     });
     const [vadGruYH, vadGruY] = this.builder_.gru(vadGruX,
@@ -95,6 +97,7 @@ export class RNNoise {
         [1, 3 * this.noiseGruHiddenSize]);
     const noiseGruInitialH = this.builder_.input('noiseGruInitialH', {
       dataType: 'float32',
+      dimensions: [1, this.batchSize_, this.noiseGruHiddenSize],
       shape: [1, this.batchSize_, this.noiseGruHiddenSize],
     });
     const [noiseGruYH, noiseGruY] = this.builder_.gru(noiseGruX,
@@ -122,6 +125,7 @@ export class RNNoise {
         [1, 3 * this.denoiseGruHiddenSize]);
     const denoiseGruInitialH = this.builder_.input('denoiseGruInitialH', {
       dataType: 'float32',
+      dimensions: [1, this.batchSize_, this.denoiseGruHiddenSize],
       shape: [1, this.batchSize_, this.denoiseGruHiddenSize],
     });
     const [denoiseGruYH, denoiseGruY] = this.builder_.gru(denoiseGruX,

semantic_segmentation/deeplabv3_mnv2_nchw.js

@@ -90,6 +90,7 @@ export class DeepLabV3MNV2Nchw {
 
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     const conv0 = this.buildConv_(

semantic_segmentation/deeplabv3_mnv2_nhwc.js

@@ -81,6 +81,7 @@ export class DeepLabV3MNV2Nhwc {
     const strides = [2, 2];
     const input = this.builder_.input('input', {
       dataType: 'float32',
+      dimensions: this.inputOptions.inputShape,
       shape: this.inputOptions.inputShape,
     });
     const conv0 = await this.buildConv_(