Abstract
Scientific programmers often turn to vendor-tuned Basic Linear Algebra Subprograms (BLAS) to obtain portable high performance. However, many numerical algorithms require several BLAS calls in sequence, and those successive calls result in suboptimal performance because the entire sequence should be optimized in concert. Alternatively, a programmer could start write their matrix operations directly in Fortran or C and use a state-of-the-art optimizing compiler. However, our experiments show that optimizing compilers often attain only one-quarter the performance of hand-optimized code. In this talk I present a domain-specific compiler for matrix algebra, the Build to Order BLAS (BTO), that reliably achieves high performance using a scalable search algorithm for choosing the best combination of loop fusion, array contraction, and multithreading for data parallelism. The BTO compiler generates code that is between 16% slower and 39% faster than hand-optimized code.