Both compilers in that 2004 paper were hardware vendor compilers, which are ofte...

cliffbean on Jan 9, 2012 | parent | context | favorite | on: The "C is Efficient" Language Fallacy (2006)

Both compilers in that 2004 paper were hardware vendor compilers, which are often heavily tuned for SPEC CPU2000. Also, both were tested with cross-translation-unit optimizations enabled; inlining in key places can make many pointer aliasing problems disappear. It's likely that these compilers already know how to do whatever is needed to get good SPEC scores even without restrict.

MichaelSalib on Jan 9, 2012 [–]

That's a really good point. I'd love to see the same analysis done with gcc/clang/icc.