A matrix math facility for Power ISA(TM) processors
Authors:
José E. Moreira,
Kit Barton,
Steven Battle,
Peter Bergner,
Ramon Bertran,
Puneeth Bhat,
Pedro Caldeira,
David Edelsohn,
Gordon Fossum,
Brad Frey,
Nemanja Ivanovic,
Chip Kerchner,
Vincent Lim,
Shakti Kapoor,
Tulio Machado Filho,
Silvia Melitta Mueller,
Brett Olsson,
Satish Sadasivam,
Baptiste Saleil,
Bill Schmidt,
Rajalakshmi Srinivasaraghavan,
Shricharan Srivatsan,
Brian Thompto,
Andreas Wagner,
Nelson Wu
Abstract:
Power ISA(TM) Version 3.1 has introduced a new family of matrix math instructions, collectively known as the Matrix-Multiply Assist (MMA) facility. The instructions in this facility implement numerical linear algebra operations on small matrices and are meant to accelerate computation-intensive kernels, such as matrix multiplication, convolution and discrete Fourier transform. These instructions h…
▽ More
Power ISA(TM) Version 3.1 has introduced a new family of matrix math instructions, collectively known as the Matrix-Multiply Assist (MMA) facility. The instructions in this facility implement numerical linear algebra operations on small matrices and are meant to accelerate computation-intensive kernels, such as matrix multiplication, convolution and discrete Fourier transform. These instructions have led to a power- and area-efficient implementation of a high throughput math engine in the future POWER10 processor. Performance per core is 4 times better, at constant frequency, than the previous generation POWER9 processor. We also advocate the use of compiler built-ins as the preferred way of leveraging these instructions, which we illustrate through case studies covering matrix multiplication and convolution.
△ Less
Submitted 7 April, 2021;
originally announced April 2021.
