Re: Optimised in compiler
As I recall, the impetus for EPIC was a soon discredited research paper implying that there was a huge amount of unexploited parallelism in existing source code that was missed by compiler peephole optimization and RISC runtime reordering.
I tested this with a DEC Alpha in 1993 running OSF/1. The compilers had the option to do deep/wide optimization across the entire set of source files for an application.
I compiled some large biomedical imaging and genomic applications and found the difference in runtime performance was about 3% or less - not worth the effort.
Given the lack of runtime optimization and the nondeterministic nature of cache misses and memory access, the only way to get Itanium to work was to put the entire working set in cache.
It would have been interesting to see what the highly regarded PA-RISC design team could have done has they not been displaced.
Of course the same goes for Alpha.