ifort version 14.0.1.106 vs ifort 16. factor of 3 slower?

[jean-pierre_e_]

I use this command on a fortran.f90 program, with quite some use of mkl with version 14 ifort -O3 -xHost -prof-use -ftrapuv -o it -I${MKLROOT}/include/intel64/lp64 -I${MKLROOT}/include mutation_2side_pinch.f90 /opt/intel/compilers_and_libraries_2016.3.210/linux/mkl/lib/intel64_lin/libmkl_blas95_lp64.a /opt/intel/compilers_and_libraries_2016.3.210/linux/mkl/lib/intel64_lin/libmkl_lapack95_lp64.a -L${MKLROOT}/lib/intel64 -lmkl_intel_lp64 -lmkl_core -lmkl_sequential -lpthread -lm -ldl -opt-report 3 -opt-report-file=opt.lst -ansi-alias -check bounds -align all -fno-alias -fno-fnalias -fp-model fast=2 -align array64byte -IPF-fp-relaxed -sox -warn all -check-uninit

with version 16 /home/eckmann/intel16/compilers_and_libraries_2016/linux/bin/intel64/ifort -Ofast -O3 -xHost -ftrapuv -o it16 mutation_2side_pinch.f90 -lpthread -lm -ldl -mkl=sequential -I${MKLROOT16}/include/ -ansi-alias -align all -fno-alias -fno-fnalias -fp-model fast=2 -align array64byte -IPF-fp-relaxed -sox -warn all -L${MKLROOT16}/lib/intel64 -lmkl_intel_lp64 -lmkl_core -lmkl_sequential -lpthread -lm -ldl I am testing the new version to buy it. but it is 3 times slower. This is on an i7 chip. (I dont want to parallelize) The configuration was suggested by the tool you provide... Any hints?

[Xiaoping_D_Intel]

The main difference between your 14.0 and 16.0 compilation command lines is PGO (Profile Guided Optimization) was enabled in for 14.0 by option "-prof-use" but not for 16.0.

Can you also have a try to enable PGO for 16.0? Moreover, a test case will be helpful for further investigation.

Thanks,

Xiaoping Duan

Intel Customer Support

[jean-pierre_e_]

No, I did use -prof-gen, ran the program, and then used -pro-use. Is that what you mean

I attach the fortran program. on my machine, with ifort 14 about 0.3 seconds, with ifort16 1. second

(the proportion is the same for longer runs)

thanks for your help

[Xiaoping_D_Intel]

Yes, we need a test case to do further investigation on your performance issue. If the code can't be post on a public forum please open an issue through Intel Premier Support.

Thanks,

Xiaoping