Computational array microprocessor system using non-consecutive data formatting

Patent metadata

Field	Value
Publication number	US-11157441-B2
Application number	US-201815920173-A
Country	US
Kind code	B2
Filing date	Mar 13, 2018
Priority date	Jul 24, 2017
Publication date	Oct 26, 2021
Grant date	Oct 26, 2021

Abstract

A microprocessor system comprises a computational array and a hardware data formatter. The computational array includes a plurality of computation units that each operates on a corresponding value addressed from memory. The values operated by the computation units are synchronously provided together to the computational array as a group of values to be processed in parallel. The hardware data formatter is configured to gather the group of values, wherein the group of values includes a first subset of values located consecutively in memory and a second subset of values located consecutively in memory. The first subset of values is not required to be located consecutively in the memory from the second subset of values.

First claim

What is claimed is:

1. A microprocessor system, comprising:
a computational array that includes a plurality of computation units, wherein each of the plurality of computation units operates on a corresponding value addressed from memory and the values operated by the plurality of computation units are synchronously provided together to the computational array as a group of values to be processed in parallel, the group of values being utilized as a first input to the computational array; and
a hardware data formatter configured to gather the group of values based on a data formatting operation, the data formatting operation identifying at least a stride, and the hardware data formatter comprising a plurality of read buffers configured to store respective subsets of the values,
wherein each subset corresponds to values located consecutively in the memory, wherein a number of values from each subset is determined based on the stride, the number of values indicating values of each subset which are to be utilized for processing based on the stride, wherein remaining values of each subset are not utilized, wherein the group of values includes the values of each subset which are to be utilized and the remaining values of each subset which are not utilized,
wherein the group of values are provided, by the hardware data formatter, to the computational array, and wherein the computational array disables particular computation units corresponding to the remaining values of each subset which are not utilized.

2. The system of claim 1 , wherein the subsets of the values include a first subset of values located consecutively in the memory and a second subset of values located consecutively in the memory, wherein the first subset of values is not located consecutively in the memory from the second subset of values, and wherein a difference in memory address between the first subset of values and the second subset of values is based on the stride.

3. The system of claim 1 , wherein the computational array is configured to receive at least two vector input operands.

4. The system of claim 1 , wherein the computational array is configured to perform a dot-product component operation using the group of values in parallel.

5. The system of claim 1 , wherein each computation unit of the plurality of computation units includes an arithmetic logic unit, an accumulator, and a shadow register.

6. The system of claim 1 , wherein the first input corresponds to an input channel of vision data.

7. The system of claim 1 , wherein the first input corresponds to sensor data.

8. The system of claim 7 , wherein the sensor data is non-image sensor data.

9. The system of claim 8 , wherein the non-image sensor data includes ultrasonic, radar, or LiDAR data.

10. The system of claim 1 , wherein the first input corresponds to a convolution filter.

11. The system of claim 10 , wherein the convolution filter is constructed to identify features of a second input which corresponds to sensor data.

12. The system of claim 2 , wherein the first subset of values is retrieved from a cache using a single cache read.

13. The system of claim 2 , wherein the first subset of values and the second subset of values are retrieved from a single cache line.

14. The system of claim 1 , wherein the memory is configured to dynamically adjust an allocation between a first portion of the memory for a data input and a second portion of the memory for a weight input.

15. The system of claim 2 , wherein the hardware data formatter is configured to determine a corresponding start memory address for each of the first subset and the second sub set.

16. The system of claim 15 , wherein the hardware data formatter is configured to determine a corresponding end memory address for each of the first subset of values and the second subset of values.

17. The system of claim 15 , wherein a cache check is performed for each of the first subset and the second subset including by determining whether a value stored at the determined starting memory addresses for the first subset has been cached and determining whether a value stored at the determined starting memory addresses for the second subset has been cached.

18. The system of claim 2 , wherein a cache check is performed for the first subset including by determining whether a first value and a last value for the first subset are stored in a cache.

19. The system of claim 1 , wherein the data formatting operation further indicates a padding parameter.

20. A method comprising:
receiving a computational operation;
receiving a data formatting operation at a hardware data formatter, the data formatting operation indicating at least a stride, wherein the hardware data formatter comprises a plurality of read buffers;
retrieving a first group of values associated with an input data, wherein the first group of values includes a first subset of values located consecutively in a memory and a second subset of values located consecutively in the memory, and the first subset of values is not located consecutively in the memory from the second subset of values, wherein a number of values from the first subset is determined based on the stride, the number of values indicating values of the first subset which are to be utilized for the computational operation based on the stride, and wherein remaining values of the first subset are not utilized;
retrieving a second group of values associated with a weight data;
providing in parallel the first group of values and the second group of values to a computational array microprocessor comprising a plurality of computational units arranged as a matrix, wherein the computational array disables particular computation units corresponding to the remaining values of the first subset which are not utilized; and
processing the first group of values and the second group of values as operands in parallel using the computational array.

21. A microprocessor system, comprising:
a computational array that includes a plurality of computation units, wherein each of the plurality of computation units operates on a corresponding value addressed from memory and the values operated by the plurality of computation units are synchronously provided together to the computational array as a group of values, wherein the group of values includes at least 96 values and the group of values includes at least 12 subsets of values;
a hardware data formatter configured to gather the group of values based on a data formatting operation, the data formatting operation identifying at least a stride, and the hardware data formatter comprising a plurality of read buffers configured to store the at least 12 subsets,
wherein the group of values includes a first subset of values located consecutively in the memory and a second subset of values located consecutively in the memory, and the first subset of values is not required to be located consecutively in the memory from the second subset of values,
wherein a number of values from the first subset is determined based on the stride, the number of values indicating values of the first subset which are to be utilized for the computational operation, and wherein remaining values of the first subset are not utilized, wherein the group of values includes the values of the first subset which are to be utilized based on the stride and the remaining values of the first subset which are not utilized,
wherein the group of values are provided, by the hardware data formatter, to the computational

Assignees

• Tesla Inc

Inventors

• Talpes Emil
• Mcgee William
• Bannon Peter Joseph

Classifications

• G06N3/048

  Activation functions · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06F7/5443 Primary

  Sum of products (for applications thereof, see the relevant places, e.g. G06F17/10, H03H17/00) · CPC title

• G06F2207/4824

  Neural networks · CPC title