Does anyone know compilation details of glibc distributed with mpss ?
Was icc used and with what flags ?
I am seeing particularly slow execution of functions like strtod, fprintf, malloc, etc.
glibc distributed with the MPSS was compiled with GCC. The most important flag that used was -O2. Are you seeing slow downs compared to Intel Xeon processors?
Yes. So far the best solution is to rewrite the code to avoid using libc functions. This unfortunately implies using custom memory allocation pool instead of new/delete..
I have tried to compile glibc and uClibc with intel compiler, but ran into problems. Seems like at least parts of the code use assembler inlines, and, perhaps, gnu assembler syntax.
In theory it should be possible to recompile only parts of C library with Intel compiler, but I have not tried that yet.