Method and system for query processing over tensor runtimes

Patent metadata

Field	Value
Publication number	US-12468701-B2
Application number	US-202217587952-A
Country	US
Kind code	B2
Filing date	Jan 28, 2022
Priority date	Jan 28, 2022
Publication date	Nov 11, 2025
Grant date	Nov 11, 2025

Abstract

Example aspects include techniques for query processing over deep neural network runtimes. These techniques may include receiving a query including one or more query operators and determining a query representation based on the one or more query operators. In addition, the techniques may include determining a neural network program based on the query representation, the neural network program including one or more neural network operators for performing the query in a neural network runtime, generating a neural network data structure based on a dataset associated with the query, and executing the neural network program in the neural network runtime over the neural network data structure to generate a query result.

First claim

What is claimed is:

1 . A method comprising:
receiving a query including one or more query operators; determining a query representation based on the one or more query operators; generating, based on the query representation, a neural network program for executing in a neural network runtime at least in part by representing one or more of the query operators, for which a corresponding operator is not provided by the neural network runtime, as one or more neural network operators for performing the query in a neural network runtime, wherein the neural network runtime provides an executable environment to at least one of train neural network models during a training mode or evaluate the neural network models in a non-training mode; generating a neural network data structure based on a dataset associated with the query; and executing the query by executing the neural network program, including the one or more neural network operators, in the neural network runtime over the neural network data structure to generate a query result.

2 . The method of claim 1 , wherein the query representation is a query plan, and determining the query representation comprises generating the query plan via a query optimizer.

3 . The method of claim 1 , wherein determining the neural network program based on the query representation comprises:
identifying a query operator of the one or more query operators; and determining a neural network operator of the one or more neural network operators, the neural network operator configured to perform at least a function of the query operator.

4 . The method of claim 1 , wherein the neural network program includes a tensor program, the one or more neural network operations include a tensor operation, and the neural network runtime includes a tensor runtime.

5 . The method of claim 1 , wherein the dataset includes columnar data, and generating the neural network data structure based on the dataset comprises generating an n-dimensional array based at least in part on a data type of the columnar data.

6 . The method of claim 1 , wherein the one or more query operators includes a structured query language operator and the one or more neural network operators includes a transformation operator, a reduction operator, an arithmetic operator, or a logical operator.

7 . The method of claim 1 , wherein the neural network runtime is configured to compile the neural network program over a plurality of processing hardware.

8 . The method of claim 1 , wherein the query includes a machine learning operator executable within the neural network runtime.

9 . A non-transitory computer-readable device having instructions thereon that, when executed by at least one computing device, causes the at least one computing device to perform operations comprising:
receiving a query including one or more query operators; determining a query representation based on the one or more query operators; generating, based on the query representation, a neural network program for executing in a neural network runtime at least in part by representing one or more of the query operators, for which a corresponding operator is not provided by the neural network runtime, as one or more neural network operators for performing the query in a neural network runtime, wherein the neural network runtime provides an executable environment to at least one of train neural network models during a training mode or evaluate the neural network models in a non-training mode; generating a neural network data structure based on a dataset associated with the query; and executing the query by executing the neural network program, including the one or more neural network operators, in the neural network runtime over the neural network data structure to generate a query result.

10 . The non-transitory computer-readable device of claim 9 , wherein determining the neural network program based on the query representation comprises:
identifying a query operator of the one or more query operators; and determining a neural network operator of the one or more neural network operators, the neural network operator configured to perform at least a function of the query operator.

11 . The non-transitory computer-readable device of claim 9 , wherein the neural network program includes a tensor program, the one or more neural network operations include a tensor operation, and the neural network runtime includes a tensor runtime.

12 . The non-transitory computer-readable device of claim 9 , wherein the dataset includes columnar data, and generating the neural network data structure based on the data comprises generating an n-dimensional array based at least in part on a data type of the columnar data.

13 . The non-transitory computer-readable device of claim 9 , wherein the one or more query operators includes a structured query language operator and the one or more neural network operators includes a transformation operator, a reduction operator, an arithmetic operator, or a logical operator.

14 . The non-transitory computer-readable device of claim 9 , wherein the query includes a machine learning operator executable within the neural network runtime.

15 . A system comprising:
a memory storing instructions thereon; and at least one processor coupled with the memory and configured by the instructions to: receive a query including one or more query operators;
determine a query representation based on the one or more query operators;
generate, based on the query representation, a neural network program for executing in a neural network runtime at least in part by representing one or more of the query operators, for which a corresponding operator is not provided by the neural network runtime, as one or more neural network operators for performing the query in a neural network runtime, wherein the neural network runtime provides an executable environment to at least one of train neural network models during a training mode or evaluate the neural network models in a non-training mode;
generate a neural network data structure based on a dataset associated with the query; and execute the query by executing the neural network program, including the one or more neural network operators, in the neural network runtime over the neural network data structure to generate a query result.

16 . The system of claim 15 , wherein the query representation is a query plan, and to determine the query representation, the at least one processor is further configured by the instructions to generate the query plan via a query optimizer.

17 . The system of claim 15 , wherein to determine the neural network program based on the query representation, the at least one processor is further configured by the instructions to:
identify a query operator of the one or more query operators; and determine a neural network operator of the one or more neural network operators, the neural network operator configured to perform at least a function of the query operator.

18 . The system of claim 15 , wherein the neural network program includes a tensor program, the one or more neural network operations include a tensor operation, and the neural network runtime includes a tensor runtime.

19 . The system of claim 15 , wherein the dataset includes columnar data, and to generate the neural network data structure based on the data, the at least one processor is further configured by the instructions to generate an n-dimensional array based at least in part on a data type of the columna

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Interlandi Matteo
• Karanasos Konstantinos
• He Dong
• Banda Dalitso Hansini
• Camacho Rodriguez Jesus
• Sen Rathijit
• Nakandala Supun Chathurang

Classifications

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06F16/2458

  Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries · CPC title

• G06F16/24542 Primary

  Plan optimisation · CPC title

• G06F16/24569 Primary

  Query processing with adaptation to specific hardware, e.g. adapted for using GPUs or SSDs · CPC title