@christophv

Facilitate SIMD-Code-Generation in the Polyhedral Model by Hardware-aware Automatic Code-Transformation

, , , and . IMPACT 2013, (2013)

Abstract

(b) constantly sized blocks (like, eg, the default tiling into blocks of 32 iterations performed by PluTo) is unlikely to perform well for every application and every system. This is especially true for deeply nested loops with many state- ments within the innermost loop. As an example, if d is the depth of a loop nest, then the innermost statements of a tile using PluTo's default tiling are executed 32d times and the data accessed by these statements is unlikely to fit into a cache with increasing d. Nonetheless, this fits quite well for 'typical'loop nests with depth two or ...

Links and resources

Tags