This Thursday (tomorrow), March 9th (9am California Time, 17:00 UTC), @VFerrari (UNICAMP) will present some work on codegen for convolutions.
Convolution is one of the most common linear algebra operations for machine-learning applications. With the increasingly important compiler integration into such frameworks, improving performance in code-generated implementations of convolution is key for the future. In this talk, Victor Ferrari (UNICAMP) presents a new convolution approach for MLIR which improves performance when compared to the traditional Im2Col + GEMM method used in the
linalg dialect and many other places.
Zoom Meeting Link is unchanged, the presentation will be recorded and posted here and on our talks page on the website as usual.
Meeting ID: 851 5109 0498