I am currently porting a large code base to multiple arm platforms using various tool chains provided by device manufactures. If I build and link the code base on linux it takes roughly 30 seconds to link. For a number of the cross compilers I have this time is about the same but for several of them this time exceeds 10+ minutes. I have tried a number of techniques to lower this but nothing helps (e.g. creating a super lib). The only thing that makes a noticeable different is linking with -0s which cuts the time by close to 40% but still leaves me with a 6 minute link and also is less then ideal during debugging since much of the code is optimized out.
What could make one cross compiler so much slower then the others? In general the charactastics of these compilers are nearly identaly (ARMv7, NEON support, gcc 4.x).
Any suggestions which might improve link time? I link to 3 static libraries, the rest is dynamic links to system libraries.