Which Env. will trigger R_X86_64_PC32 relocation issue

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

With PGI 11.8 and later, it’s still suffering the ‘overflow in R_X86_64_PC32 relocation’ issue on certain 64-bit Linux box if without spcifying -fPIC/mcmodel=medium for compile and link, but there is completely no this sort of issue on certain 64-bit Linux box even if without specifying -fPIC/mcmodel=medium for link and compile

I know this issue can be fixed by using -fPIC/mcmodel=medium, but tired always need to add/remove them by manual.

Just wondering

  1. Which sort of Env. will trigger R_X86_64_PC32 relocation issue?

  2. How to force to link with 64-bit offset based on the references instead 32-bit by default without using -fPIC/mcmodel=medium?

  3. Or how to change the system Env. settings to avoid R_X86_64_PC32 relocation issue?


    Thanks

Jerry

2

Hi Jerry,

By default, Linux uses the small memory model. This means that the static offset between the executable code and all the data in the data section is accessible within 32-bits.

Hence, you have one of two options:

Use -fpic/-mcmodel=medium so the base offset to the data is re-locatable.
or
Use less static data, typically by changing static arrays to be dynamically allocated.


Why this would occur on some Linux systems but not others, I’m not sure. I would guess however, that your static data is large but different assemblers and linkers will put in different static data and those difference could put you over the 32-bit offset. Another possibility is which libraries are being linked and if they are being linked statically or dynamically.

  • Mat