2020 LLVM Developers’ Meeting: F. Hahn “Matrix Support in LLVM and Clang”
Автор: LLVM
Загружено: 2020-10-26
Просмотров: 939
Описание:
https://llvm.org/devmtg/2020-09/
—
Matrix Support in LLVM and Clang - Florian Hahn
Slides: https://llvm.org/devmtg/2020-09/slide...
—
Fast matrix operations are the key to the performance of linear algebra algorithms, which serve as engines of machine learning networks and AR applications. We recently added support for key matrix operations to Clang and LLVM, which is available in the 11.0 release. We will start by introducing examples on the C++ language level. Next we will discuss aspects of the implementation and design in Clang. Then we will show how matrix operations are lowered to vector operations in LLVM. Finally we will compare the performance of the matrix types extension to implementations in specialized libraries like Eigen. This will also include a sketch of how specialized instructions, like AArch64’s udot instruction, can be used to achieve the best performance for a given target. We will also share our experiences with collaborating with both the LLVM and Clang communities through RFCs and reviews, how this shaped the design and implementation and enabled more uses, for example in MLIR.
—
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
