[HECnet] Benchmarks - WHETSTONE.C

[Clem Cole]

I should point out, I speak for my self in all of this not for my  employer or previous ones.


On Tue, Jan 15, 2013 at 12:17 PM, Tim Sneddon <tim at sneddon.id.au> wrote:
It did do IA-32, IA-64, Alpha,  MIPS.

And INTEL*64 to an extent (although it was not released).   As I understand it, the project was show it could be done and it was fodder/part of the GEM vs Intel C war that GEM lost.    But as I understand it from my compiler friends, it was fairy easy to take the x32 tables and make them 64 bits and add new instructions.    The advantage was the IL was easier to do some of the cool things INTEL*64 needs but alas it was in BLISS and Intel was not going to base it's compiler's on BLISS (which I can understand from a business standpoint].

As Grove once told me, GEM was designed to be a compiler to last 25 years.   They knew it would have to support a number of features we now re having a heck of time dealing with in today's code generators and ILs.    No other suite so far has done as wide   job and able to handle the diversity of languages and architectures.


GCC has been made to work, and was the first FOSS compiler to come close.   But the difference between GCC's code generator for INTEL*64 compared to icc is not even close for real applications programs.

It will be interesting to see if LLVM is able to do as well as GEM did.   You have a lot of the thought leaders in compiler land betting on it.   Apple's moved to it.    Intel will do something with it for the Exascale machines because DOE wants it, as will IBM or any other firm that wants to "prime" a Supercomputer in the future I would bet (but do not know).