I installed PSXE 2018 and rebuilt one of my major projects in release (i.e. with optimizations, compiler options see below). I'm not happy to get an ICE in one 2000 lines routine, which builds and runs fine with all PSXE 2017 compilers and produces reasonable results. Debug compilation runs fine with PSXE 2018.
My goal is to derive a (simple) reproducer. I hate to be so unspecific here, sorry. Is there a chance to get an information in which part/line of code the ICE is triggered exactly. Any fortcom logs available?
I know, I could commit the whole code to OSC, but I don't like to do that by different reasons.
Any hints on tracking down the ICE are welcome.
PSXE 2018 initial release
Compiling with Intel(R) Visual Fortran Compiler 126.96.36.199 [Intel(R) 64]... ifort /nologo /MP /O3 /Qparallel /assume:buffered_io /Qopt-matmul /fpp /Qopenmp /warn:all /real_size:64 /Qinit:snan /Qinit:arrays /fp:source /Qfp-speculation=safe /module:"x64\Release\\" /object:"x64\Release\\" /Fd"x64\Release\vc140.pdb" /check:bounds /libs:static /threads /c /Qimf-precision:high /QaxCORE-AVX512 /nostandard-realloc-lhs /Qlocation,link,"C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\\bin\amd64" /Qm64 "D:\...\mod_calc_section_2D_BECAS.f90"
/QaxCORE-AVX512 from compiler options did the trick. The ICE is gone with no /QaxCORE-XXXX option set. /QaxCORE-AVX2 delivers also the ICE. /QxCORE-AVX2 works fine as well as /QxHost (Xeon E5-1650V3) and also /QaxAVX.
Has anyone tested the 'processor-optimized code path' compiler options /QaxCORE-AVX2 and /QaxCORE-AVX512 with PSXE 2018?