2023-07-25 21:19:20 +00:00
|
|
|
// Copyright (c) Microsoft Corporation. All rights reserved.
|
|
|
|
|
// Licensed under the MIT License.
|
|
|
|
|
|
2024-08-14 23:51:22 +00:00
|
|
|
import { DataType } from '../../../wasm-common';
|
|
|
|
|
import { TensorView } from '../../tensor-view';
|
|
|
|
|
import { ShapeUtil } from '../../util';
|
|
|
|
|
import { AttributeWithCacheKey, createAttributeWithCacheKey } from '../attribute-with-cache-key';
|
|
|
|
|
import { ComputeContext, ProgramInfo, ProgramUniform, TensorInfo } from '../types';
|
|
|
|
|
|
|
|
|
|
import {
|
|
|
|
|
createTensorShapeVariables,
|
|
|
|
|
getElementAt,
|
|
|
|
|
IndicesHelper,
|
|
|
|
|
inputVariable,
|
|
|
|
|
outputVariable,
|
|
|
|
|
ShaderHelper,
|
|
|
|
|
UniformsArrayType,
|
|
|
|
|
} from './common';
|
2023-07-25 21:19:20 +00:00
|
|
|
|
|
|
|
|
export interface SliceAttributes extends AttributeWithCacheKey {
|
|
|
|
|
readonly starts: number[];
|
|
|
|
|
readonly ends: number[];
|
|
|
|
|
readonly axes: number[];
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
const validateInputs = (inputs: readonly TensorView[], attributes: SliceAttributes): void => {
|
|
|
|
|
if (!inputs || inputs.length < 1) {
|
|
|
|
|
throw new Error('too few inputs');
|
|
|
|
|
}
|
|
|
|
|
if (attributes.axes.length !== 0) {
|
|
|
|
|
if (attributes.axes.length !== attributes.starts.length || attributes.axes.length !== attributes.ends.length) {
|
|
|
|
|
throw new Error('axes, starts and ends must have the same length');
|
|
|
|
|
}
|
|
|
|
|
} else if (attributes.starts.length !== attributes.ends.length) {
|
|
|
|
|
throw new Error('starts and ends must have the same length');
|
|
|
|
|
}
|
|
|
|
|
inputs.slice(1).forEach((_, idx) => {
|
|
|
|
|
if (inputs[idx + 1].dataType !== DataType.int32 && inputs[idx + 1].dataType !== DataType.int64) {
|
|
|
|
|
throw new Error(`Input ${idx} must be an array of int32 or int64`);
|
|
|
|
|
}
|
|
|
|
|
});
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
const readInput = (inputs: readonly TensorView[], idx: number): number[] => {
|
|
|
|
|
const input: number[] = [];
|
|
|
|
|
if (inputs.length > idx) {
|
|
|
|
|
if (inputs[idx].dataType === DataType.int64) {
|
2024-08-14 23:51:22 +00:00
|
|
|
inputs[idx].getBigInt64Array().forEach((v) => input.push(Number(v)));
|
2023-08-03 22:48:47 +00:00
|
|
|
} else if (inputs[idx].dataType === DataType.int32) {
|
2024-08-14 23:51:22 +00:00
|
|
|
inputs[idx].getInt32Array().forEach((v) => input.push(Number(v)));
|
2023-07-25 21:19:20 +00:00
|
|
|
} else {
|
|
|
|
|
throw new Error(`Input ${idx} must be an array of int32 or int64`);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
return input;
|
|
|
|
|
};
|
|
|
|
|
|
2024-08-14 23:51:22 +00:00
|
|
|
const createSliceAttributesFromInputs = (
|
|
|
|
|
inputs: readonly TensorView[],
|
|
|
|
|
attributes: SliceAttributes,
|
|
|
|
|
): SliceAttributes => {
|
|
|
|
|
if (inputs.length > 1) {
|
|
|
|
|
const starts: number[] = readInput(inputs, 1);
|
|
|
|
|
const ends: number[] = readInput(inputs, 2);
|
|
|
|
|
let axes: number[] = readInput(inputs, 3);
|
|
|
|
|
if (axes.length === 0) {
|
|
|
|
|
axes = [...Array(inputs[0].dims.length).keys()];
|
|
|
|
|
}
|
|
|
|
|
return createAttributeWithCacheKey({ starts, ends, axes });
|
|
|
|
|
} else {
|
|
|
|
|
return attributes;
|
|
|
|
|
}
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
const fixStartEndValues = (
|
|
|
|
|
value: number,
|
|
|
|
|
index: number,
|
|
|
|
|
inputShape: readonly number[],
|
|
|
|
|
axes: readonly number[],
|
|
|
|
|
steps: readonly number[],
|
|
|
|
|
): number => {
|
|
|
|
|
let newValue = value;
|
|
|
|
|
if (value < 0) {
|
|
|
|
|
newValue += inputShape[axes[index]];
|
|
|
|
|
}
|
|
|
|
|
if (steps[index] < 0) {
|
|
|
|
|
return Math.max(0, Math.min(newValue, inputShape[axes[index]] - 1));
|
|
|
|
|
} else {
|
|
|
|
|
return Math.max(0, Math.min(newValue, inputShape[axes[index]]));
|
|
|
|
|
}
|
|
|
|
|
};
|
2023-07-25 21:19:20 +00:00
|
|
|
|
2024-08-14 23:51:22 +00:00
|
|
|
const calculateInputIndicesImpl = (
|
|
|
|
|
input: IndicesHelper,
|
|
|
|
|
output: IndicesHelper,
|
|
|
|
|
inputShape: readonly number[],
|
|
|
|
|
): string =>
|
|
|
|
|
`fn calculateInputIndices(output_indices: ${output.type.indices}) -> ${input.type.indices} {
|
2023-12-12 19:12:23 +00:00
|
|
|
var input_indices: ${input.type.indices};
|
2023-07-25 21:19:20 +00:00
|
|
|
var carry = 0u;
|
|
|
|
|
for (var i = ${inputShape.length}; i >= 0; i--) {
|
2023-12-01 23:35:35 +00:00
|
|
|
let input_shape_i = ${getElementAt('uniforms.input_shape', 'i', inputShape.length)};
|
|
|
|
|
let steps_i = ${getElementAt('uniforms.steps', 'i', inputShape.length)};
|
|
|
|
|
let signs_i = ${getElementAt('uniforms.signs', 'i', inputShape.length)};
|
|
|
|
|
let starts_i = ${getElementAt('uniforms.starts', 'i', inputShape.length)};
|
2023-12-12 19:12:23 +00:00
|
|
|
var output_index = ${output.indicesGet('output_indices', 'i')};
|
|
|
|
|
var input_index = output_index * steps_i + starts_i + carry;
|
|
|
|
|
carry = input_index / input_shape_i;
|
|
|
|
|
input_index = input_index % input_shape_i;
|
2023-11-16 17:44:13 +00:00
|
|
|
if (signs_i < 0) {
|
2023-12-12 19:12:23 +00:00
|
|
|
input_index = input_shape_i - input_index - 1u + starts_i;
|
2023-07-25 21:19:20 +00:00
|
|
|
}
|
2023-12-12 19:12:23 +00:00
|
|
|
${input.indicesSet('input_indices', 'i', 'input_index')};
|
2023-07-25 21:19:20 +00:00
|
|
|
}
|
2023-12-12 19:12:23 +00:00
|
|
|
return input_indices;
|
2023-07-25 21:19:20 +00:00
|
|
|
}`;
|
|
|
|
|
|
2023-10-10 07:31:12 +00:00
|
|
|
const createSliceProgramInfo = (inputs: readonly TensorView[], attributes: SliceAttributes): ProgramInfo => {
|
|
|
|
|
const inputShape = inputs[0].dims;
|
|
|
|
|
const inputSize = ShapeUtil.size(inputShape);
|
2024-08-14 23:51:22 +00:00
|
|
|
const axes =
|
|
|
|
|
attributes.axes.length > 0
|
|
|
|
|
? ShapeUtil.normalizeAxes(attributes.axes, inputShape.length)
|
|
|
|
|
: [...Array(inputShape.length).keys()];
|
2023-10-10 07:31:12 +00:00
|
|
|
let steps = readInput(inputs, 4);
|
2024-08-14 23:51:22 +00:00
|
|
|
steps.forEach(
|
|
|
|
|
(step) =>
|
|
|
|
|
step !== 0 ||
|
|
|
|
|
(() => {
|
|
|
|
|
throw new Error('step cannot be 0');
|
|
|
|
|
}),
|
|
|
|
|
);
|
2023-10-10 07:31:12 +00:00
|
|
|
if (steps.length === 0) {
|
|
|
|
|
steps = Array(axes.length).fill(1);
|
|
|
|
|
}
|
|
|
|
|
const starts = attributes.starts.map((start, i) => fixStartEndValues(start, i, inputShape, axes, steps));
|
2023-07-25 21:19:20 +00:00
|
|
|
|
2023-10-10 07:31:12 +00:00
|
|
|
const ends = attributes.ends.map((end, i) => fixStartEndValues(end, i, inputShape, axes, steps));
|
2023-07-25 21:19:20 +00:00
|
|
|
|
2023-11-16 17:44:13 +00:00
|
|
|
if (axes.length !== starts.length || axes.length !== ends.length) {
|
|
|
|
|
throw new Error('start, ends and axes should have the same number of elements');
|
|
|
|
|
}
|
|
|
|
|
|
2023-10-10 07:31:12 +00:00
|
|
|
if (axes.length !== inputShape.length) {
|
|
|
|
|
for (let i = 0; i < inputShape.length; ++i) {
|
|
|
|
|
if (!axes.includes(i)) {
|
|
|
|
|
starts.splice(i, 0, 0);
|
|
|
|
|
ends.splice(i, 0, inputShape[i]);
|
|
|
|
|
steps.splice(i, 0, 1);
|
2023-07-25 21:19:20 +00:00
|
|
|
}
|
2023-10-10 07:31:12 +00:00
|
|
|
}
|
|
|
|
|
}
|
2024-08-14 23:51:22 +00:00
|
|
|
const signs = steps.map((step) => Math.sign(step));
|
2023-10-10 07:31:12 +00:00
|
|
|
// Convert negative steps to positive steps and reverse starts and ends
|
|
|
|
|
steps.forEach((step, i, array) => {
|
|
|
|
|
if (step < 0) {
|
|
|
|
|
const numSteps = (ends[i] - starts[i]) / step;
|
|
|
|
|
const newEnd = starts[i];
|
|
|
|
|
const newStart = newEnd + numSteps * steps[i];
|
|
|
|
|
starts[i] = newStart;
|
|
|
|
|
ends[i] = newEnd;
|
|
|
|
|
array[i] = -step;
|
|
|
|
|
}
|
|
|
|
|
});
|
2023-11-16 17:44:13 +00:00
|
|
|
// Output rank is expected to be less than or equal to the input rank.
|
2023-10-10 07:31:12 +00:00
|
|
|
const outputShape = inputShape.slice(0);
|
|
|
|
|
axes.forEach((axis, _) => {
|
|
|
|
|
outputShape[axis] = Math.ceil((ends[axis] - starts[axis]) / steps[axis]);
|
|
|
|
|
});
|
2024-08-14 23:51:22 +00:00
|
|
|
const outputTensorInfo: TensorInfo = { dims: outputShape, dataType: inputs[0].dataType };
|
2023-07-25 21:19:20 +00:00
|
|
|
|
2023-12-01 01:10:33 +00:00
|
|
|
const output = outputVariable('output', inputs[0].dataType, outputShape.length);
|
|
|
|
|
const input = inputVariable('input', inputs[0].dataType, inputs[0].dims.length);
|
2023-10-10 07:31:12 +00:00
|
|
|
const outputSize = ShapeUtil.size(outputShape);
|
2023-12-01 01:10:33 +00:00
|
|
|
const uniforms: UniformsArrayType = [
|
2024-08-14 23:51:22 +00:00
|
|
|
{ name: 'outputSize', type: 'u32' },
|
|
|
|
|
{ name: 'starts', type: 'u32', length: starts.length },
|
|
|
|
|
{ name: 'signs', type: 'i32', length: signs.length },
|
|
|
|
|
{ name: 'steps', type: 'u32', length: steps.length },
|
2023-12-01 01:10:33 +00:00
|
|
|
];
|
|
|
|
|
|
|
|
|
|
const programUniforms: ProgramUniform[] = [
|
2024-08-14 23:51:22 +00:00
|
|
|
{ type: DataType.uint32, data: outputSize },
|
|
|
|
|
{ type: DataType.uint32, data: starts },
|
|
|
|
|
{ type: DataType.int32, data: signs },
|
|
|
|
|
{ type: DataType.uint32, data: steps },
|
|
|
|
|
...createTensorShapeVariables(inputs[0].dims, outputShape),
|
2023-12-01 01:10:33 +00:00
|
|
|
];
|
2023-07-25 21:19:20 +00:00
|
|
|
|
2023-10-10 07:31:12 +00:00
|
|
|
const getShaderSource = (shaderHelper: ShaderHelper) => `
|
2023-11-16 17:44:13 +00:00
|
|
|
${shaderHelper.registerUniforms(uniforms).declareVariables(input, output)}
|
2023-12-12 19:12:23 +00:00
|
|
|
${calculateInputIndicesImpl(input, output, inputShape)}
|
2023-07-25 21:19:20 +00:00
|
|
|
${shaderHelper.mainStart()}
|
2023-11-16 17:44:13 +00:00
|
|
|
${shaderHelper.guardAgainstOutOfBoundsWorkgroupSizes('uniforms.outputSize')}
|
2023-12-12 19:12:23 +00:00
|
|
|
let output_indices = ${output.offsetToIndices('global_idx')};
|
|
|
|
|
let input_indices = calculateInputIndices(output_indices);
|
|
|
|
|
${output.setByOffset('global_idx', input.getByIndices('input_indices'))}
|
2023-07-25 21:19:20 +00:00
|
|
|
}`;
|
2023-10-10 07:31:12 +00:00
|
|
|
return {
|
|
|
|
|
name: 'Slice',
|
2024-08-14 23:51:22 +00:00
|
|
|
shaderCache: { hint: `${signs.length}_${starts.length}_${steps.length}`, inputDependencies: ['rank'] },
|
2023-10-10 07:31:12 +00:00
|
|
|
getShaderSource,
|
|
|
|
|
getRunData: () => ({
|
|
|
|
|
outputs: [outputTensorInfo],
|
2024-08-14 23:51:22 +00:00
|
|
|
dispatchGroup: { x: Math.ceil(inputSize / 64 /* workgroup size */) },
|
|
|
|
|
programUniforms,
|
|
|
|
|
}),
|
2023-10-10 07:31:12 +00:00
|
|
|
};
|
|
|
|
|
};
|
2023-07-25 21:19:20 +00:00
|
|
|
|
|
|
|
|
export const slice = (context: ComputeContext, attributes: SliceAttributes): void => {
|
|
|
|
|
validateInputs(context.inputs, attributes);
|
2023-10-10 07:31:12 +00:00
|
|
|
const updatedAttributes = createSliceAttributesFromInputs(context.inputs, attributes);
|
2024-08-14 23:51:22 +00:00
|
|
|
context.compute(createSliceProgramInfo(context.inputs, updatedAttributes), { inputs: [0] });
|
2023-10-10 07:31:12 +00:00
|
|
|
// if (ShapeUtil.size(program.outputs[0].dims) > 0) {
|
|
|
|
|
// context.compute(programInfoLoader, {inputs: [0]});
|
|
|
|
|
// } else {
|
|
|
|
|
// // TODO: support empty output
|
|
|
|
|
// throw new Error('slice: output size is 0');
|
|
|
|
|
// }
|
2023-07-25 21:19:20 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
export const parseSliceAttributes = (attributes: Record<string, unknown>): SliceAttributes => {
|
|
|
|
|
const starts = attributes.starts as number[];
|
|
|
|
|
const ends = attributes.ends as number[];
|
|
|
|
|
const axes = attributes.axes as number[];
|
2024-08-14 23:51:22 +00:00
|
|
|
return createAttributeWithCacheKey({ starts, ends, axes });
|
2023-07-25 21:19:20 +00:00
|
|
|
};
|