Intel Parallel Studio XE 2016 Update 1 for Windows - very slow compile time

[Morgan_L_]

I have tried compiling a large code under both Visual Studio and Cygwin environments using this FORTRAN compiler.  It is too slow to be usable.  Whereas ifort 12.1(8) took 1-3 seconds per routine this version takes ~5 minutes per routine.  I stopped the build after a few routines.  There is something very wrong with this behavior.  The install went normally after installing VS 2016 and Cygwin.

[Steven_L_Intel1]

We would need to have the sources to try a build ourselves to see what the problem is.

When you say "per routine" do you really mean "per source file"? Are you using a floating license or named-user license? When I look up your email domain I don't find any current licenses so I can't tell.

[Morgan_L_]

The code is MCNP 6.1.1.  The source is only available from RSICC officially.  It is about 500,000 lines of FORTRAN in many routines.  When I do the compile under VS or Cygwin it ticks off a subroutine in 1-3 seconds.  it takes about 10 minutes to compile and link the whole mess.  With ifort 2016 it is taking ~5 minutes for each subroutine.  I can't even imagine what it is doing - but it isn't pretty.  The cpu usage during the compile is like in a wait state - very low utilization - maybe 2%.

[Morgan_L_]

I am using the trail version.

[Steven_L_Intel1]

Again, is this one subroutine per file?  Five minutes per file is indeed absurdly long. Would you be willing to provide us one of the source files that takes so long to compile? We'd also need any INCLUDE files it references. You can use Intel Premier Support (as a trial user you should have access to this), or use Send Author a Message to send it to me directly. Please ZIP the files and attach the ZIP to the reply.

Please also tell me what compiler switches are used.

[Morgan_L_]

Steve - MCNP is a code with about 700,000 lines of fortran.  It is configured in Cygwin which sends each subroutine as a file to ifort.  DOE restrictions prevent releasing the source code in any form.  The subroutines tick off as compiled every 1-3 seconds using 12.1(8).

I did a build of a utility routine I wrote to process MCNP output files.  Under VS 1010 and ifort 12.1(8) it builds in maybe 5 seconds.  It takes almost a minute with VS 2015 and ifort 2016.  It I can send.

Thanks for the rapid responses.

[Morgan_L_]

Steve - I don't think it is the code - it is the setup.  Something got hosed.

[mecej4]

Morgan L. wrote:

Steve - I don't think it is the code - it is the setup.  Something got hosed.

Indeed, your guess seems to be correct.

I compiled your test program of #6 using IFort 16.0.1.139 on an aging low-end laptop with a 5400 RPM HDD, Pentium T4300 CPU and W10. With /Od, the time to compile and link in an IFort command window was 0.6 seconds; with the default option (/fast), the time went up to 1 second.

I suspect problems with the license manager used by the compiler.

[nichols__john]

on an aging laptop with a 5400 RPM HDD, Pentium T4300 CPU and W10.

The English Professor got herself a HP 360 Pavilion I think which has a Pentium T chip, a HDD and I put W10 on it for her.  I need to rip out the HDD and replace with an SSD and up the RAM.  She uses it to collect field data and it is quite light and good.  Had to fix Office today for her, TEAM Viewer for doing support long distance is wonderful.

At the moment the w10 preview and the w10 release appear to be the same. Preview is a lot of fun.

She runs the programs I compile with Intel Fortran -- now she asks for mods to make them run faster.  She dislikes screen writes - slows up the code -- fussy users.

Latest IFORT is good.

John