ICL consumes 3.5 GB of RAM and reports "internal error: backend signals"

[support1]

ICL version: 10.1.013
Machine: Quad Core CPU (QX6850 @ 3.00GHz), 4GB RAM, RAID0 HDD
OS: Windows 2003 Server (x64)

After linker starts, it works for about 30 seconds at the same level, memory-wise, and then within 2 minutes goes through 3.5GB of RAM (Available Bytes just plunges to zero in perfmon). Shortly after it hits the paging file, the linker fails with these error messages:

1>Linking... (Intel C++ Environment)
1>(0): internal error: backend signals
1>
1>xilink: error error_during_IPO_compilation: problem during multi-file optimization compilation (code 4)
1>xilink: error error_during_IPO_compilation: problem during multi-file optimization compilation (code 4)

Is that just a memory leak in the linker or there's a limitation on project size?

Andre

[TimP]

Try removing IPO (e.g. /GL-). If you would like it investigated, and are willing to submit the case, file a problem report on premier.intel.com.

[JenniferJ]

Like Tim suggested, removing /GL likely solve this issue but may affect your performance.

There's no existing trackers on this. Could you report to PremierSupport?

If you OpenMP, try the compatibility libs in 10.1.

[support1]

No, there's no OpenMP used. Turning off Whole Program Optimizations and IPO (/Qipo-) didn't help. Only when I reduced the optimization level to /O2, I could compile the project. However, then I ran into another internal error that looks like this:

1>(0): internal error: 0_0

I had to disable optimizations for some source files in order to get past this problem.

Seems that ICL got buggier since I tested one a few months ago.

Andre

[John_O_Intel]

If you can provide a testcase, please submit an issue to https://premier.intel.comso we can investigate the internal compiler errors. You can also try compiling problem routines with /O1 or /Od.
_|ohnO

[Iliyan_Georgiev]

Out of curiosity (but not just), do you heavily use templates in your code? I got a few internal errors and large memory consumption a few times with my project which has templates and __forceinline functions everywhere... and IPO takes forever.