Automatic Parallelization in GCC: for Research and for Real (Keynote Talk)
Abstract
Portability of performance has been the underlying assumption for the vast majority of software developments. Micro-architectures and run-time systems have been designed to hide the parallelism and non-uniformity of the hardware, but with diminishing returns in performance and poor power efficiency. Despite decades of successes with optimizing compilers, the complexity of modern hardware has incrementally destroyed this dream for most developers. Every day, more programmers are forced to resort to platformspecific optimizations, committing early on specific parallel implementations. This is a dramatic regression.
Domains
Hardware Architecture [cs.AR]
Origin : Files produced by the author(s)