<div dir="ltr"><div>From: Clem Cole <<a href="mailto:clemc@ccc.com">clemc@ccc.com</a>><br><br></div>Hey Clem...thanks for the input. Definitely enjoying the last couple of weeks discourse with Bliss and Fortran. Good times they were.<br><div><br>>On Mon, Feb 5, 2018 at 4:27 PM, Kenneth Seefried <<a href="mailto:kjseefried@gmail.com">kjseefried@gmail.com</a>><br>
wrote:<br>
<br>>> Have you looked at the Intel Fortran compiler (<a href="https://software.intel.com/" rel="noreferrer" target="_blank">https://software.intel.com/</a><br>
>> en-us/fortran-compilers)? <br><br>
> Indeed and it has the added advantage of 'having the DEC compiler DNA<br>
> ground up and injected back in' to quote a retired Intel Fellow (Rich Grove<br>
> - who had lead the Gem compiler at DEC). A number of the same people are<br>
> still developing it (*i.e*. those that have not yet retired).<br>
<br></div><div>Yes, but didn't you say that (unfortunately) only a fraction of that DNA made it in, and that nothing like GEM level sophistication still exists?</div><div><br>
> It is the most popular of the Fortran compilers for the HPC community and<br>> the commercial ISV's that still have code in Fortran. <br><br></div><div>I wouldn't dispute that, per se, but the IBM Fortran compiler for POWER (collectively AIX & zOS) is a pretty formidable beast as well. If I had more access to it, I'd be interested in some compare and contrast. Power8 & Power9 still potentially have legs in the HPC world, I would think.<br><br></div><div>Speaking of...know anything about the NEC Fortran compiler targeting SX? I've got a friend in the geotechnical field who raves, but he can't share much (obviously). Interesting machine, the SX.<br><br>Since we're on the subject...wonder what happened to the compiler
tech from the CDC Cyber that evolved through
7x->170->180->180/990->203/205->ETA. We had a 990 at GaTech, and UGA ended up with a 205 for reasons that mystify to this day. Should research
that some day.</div><div><br>> There is great care to try to ensure old code from old systems 'just work' <br>> as well as bring in modern Fortran features. <br><br></div><div>Indeed. The code I referenced originally was work I did in '80s era Cray Fortran for the XMP/YMP boxes. Pre-co-arrays, I believe (at least I wasn't using them). The Intel compiler had no trouble with it. Just had to strip out the VMS-isms that the front-ends needed.<br></div><div><br>> Modern Fortran BTW, is from a syntax stand point, nothing like what <br>> I was taught in the late 60s/early 70s.<br><br></div><div>Ha! Nor what I was taught in the late 70s & 80s. Very different beast now, but not without charms. The parallel intrinsics are pretty nifty. <br>
<br>
> The point is that Intel takes Fortran seriously, since the codes that come<br>
> out of the compiler sells real silicon, which of course is what Intel<br>
> fundamentally does - sell expensive, hot rocks to run real codes.<br><br></div><div>So glad to hear that, and I've heard the IBM Power guys say the same thing about their Fortran efforts. Basically nothing really shows off what a SIMD instruction set can do that's accessible to 'normal' programmers like Fortran. Or so they say. Maybe we just have a soft spot.<br></div><div>
<br></div><div>Cheers.<br><br></div><div>KJ<br></div><div><br><br><br></div></div>