Intel Corporation

Oracle International Corporation

Bharat Daga

Ajit Singh

Pradeep Janedula

Michael Behar

Aravind Babu Srinivasan

Ambili Vengallur

Bijoy Pazhanimala

Zoran Radovic

Graham Ricketson Murphy

Oren Agam

Krishnakumar Nair

Dmitri Vainbrand

Saurabh Dhoble

Embodiments described herein provide a processing apparatus comprising compute circuitry to generate neural network data for a convolutional neural network (CNN) and write the neural network data to a memory buffer. The compute circuitry additionally includes a direct memory access (DMA) controller including a hardware codec having encode circuitry and a decode circuitry. The DMA controller reads the neural network data from the memory buffer, encode the neural network data via the encode circuit, writes encoded neural network data to a memory device coupled with the processing apparatus, writes metadata for the encoded neural network data to the memory device coupled with the processing apparatus, and decodes encoded neural network data via the decode circuit in response to a request from the compute circuitry.

Compression for deep learning in case of sparse values mapped to non-zero value

An apparatus to facilitate execution of non-linear functions operations is disclosed. The apparatus comprises accelerator circuitry including a compute grid having a plurality of processing elements to execute neural network computations, store values resulting from the neural network computations, and perform piecewise linear (PWL) approximations of one or more non-linear functions using the stored values as input data.

Mechanism to perform non-linear functions in a machine learning accelerator

Hardware accelerators for accelerated grouped convolution operations. A first buffer of a hardware accelerator may receive a first row of an input feature map (IFM) from a memory. A first group comprising a plurality of tiles may receive a first row of the IFM. A plurality of processing elements of the first group may compute a portion of a first row of an output feature map (OFM) based on the first row of the IFM and a kernel. A second buffer of the accelerator may receive a third row of the IFM from the memory. A second group comprising a plurality of tiles may receive the third row of the IFM. A plurality of processing elements of the second group may compute a portion of a third row of the OFM based on the third row of the IFM and the kernel as part of a grouped convolution operation.

Efficient hardware architecture for accelerating grouped convolutions

Embodiments described herein provide a processing apparatus comprising compute logic to generate neural network data for a convolutional neural network (CNN) and write the neural network data to a memory buffer. The compute logic additionally includes a direct memory access (DMA) controller including a hardware codec having an encode unit and a decode unit, the DMA controller to read the neural network data from the memory buffer, encode the neural network data via the encode unit, write encoded neural network data to a memory device coupled with the processing apparatus, write metadata for the encoded neural network data to the memory device coupled with the processing apparatus, and decode encoded neural network data via the decode unit in response to a request from the compute logic.

One embodiment provides a compute apparatus to perform machine learning operations, the compute apparatus comprising a hardware accelerator including a compute unit to perform a Winograd convolution, the compute unit configurable to perform the Winograd convolution for a first kernel size using a transform associated with a second kernel size.

System and method for an optimized winograd convolution accelerator

An apparatus to facilitate acceleration of machine learning operations is disclosed. The apparatus comprises accelerator circuitry including a first set of processing elements to perform first computations including matrix multiplication operations, a second set of processing elements to perform second computations including sum of elements of weights and offset multiply operations and a third set of processing elements to perform third computations including sum of elements of inputs and offset multiply operations, wherein the second and third computations are performed in parallel with the first computations.

Machine learning accelerator architecture

Embodiments described herein provide a processing apparatus comprising compute logic to generate output feature map data for a convolutional neural network (CNN) and write the feature map data to a memory buffer; a direct memory access (DMA) controller including a feature map encoder, the DMA controller to read the feature map data from the memory buffer, encode the feature map data using one of multiple encode algorithms, and write encoded feature map data to memory coupled with the processing apparatus; and wherein the compute logic is to read the encoded feature map data from the memory in an encoded format and decode the encoded feature map data while reading the encoded feature map data.

System and method of encoding and decoding feature maps and weights for a convolutional neural network

A mechanism is described for facilitating efficient memory layout for enabling smart data compression in machine learning environments. A method of embodiments, as described herein, includes facilitating dividing an initial tile representing an image into primary multiple tiles such that each tile of the primary multiple tiles is regarded as an independent image as processed by one or more processors of a computing device. The method may further include computing the primary multiple tiles into secondary multiple tiles compatible in size of a local buffer. The method may further include merging the multiple secondary multiple tiles into a final tile representing the image, and compressing the final tile.

Efficient memory layout for enabling smart data compression in machine learning environments

Systems and methods for efficient memory corruption detection in a processor. A processor detects a first data structure is to be allocated in a physical memory. The physical memory may be a DRAM with a spare bank of memory reserved for a hardware failover mechanism. Either the processor or an operating system (OS) determines a first version number corresponding to the first data structure. During initialization of the first data structure, the first version number may be stored in a first location in the spare bank. The processor receives from the OS a pointer holding the first version number. When the processor executes memory access operations targeting the first data structure, the processor compares the first version number with a third version number stored in a location in the physical memory indicated by the memory access address. The processor may set a trap in response to determining a mismatch.

Growing community of inventors

Bengaluru, India

Bharat Daga