Performance Computing: The Brave GNU World

“Building a fast open-source compiler for IA-32 meant knowing
what to change and how.”

“Tuning a compiler is like trying to open a combination lock by
trial and error, except that you cannot progress through every
combination in an orderly fashion: when you change one tumbler, not
only can the others randomly change their values, but the lock’s
combination may change as well. You shouldn’t give up on
performance tuning because of its complexity, though, just
understand that any discussion about optimization is relative and
necessarily incomplete.”

“When developing the GNU compilers for the Intel Pentium II
processor, we confronted an advanced superscalar processor that had
the potential to deliver outstanding integer and floating-point
performance. While the Pentium II can run codes generated for the
386, 486, and Pentium unmodified, new compiler strategies are
required to get the full performance possible with the Pentium II.
Here, we will discuss three specific ways to maximize Pentium II

Complete Story