Interprocedural optimizaitons may be inlining more code. Turn IPO off then see what happens.
Jim Dempsey
The settings implicit in /Qip are set up and down on each compiler release to keep pace with various other compilers and platforms. /Qip- or more detailed use of the documented limits associated with /Qip will help control object size. In addition, /O1 usually makes much smaller code than the default; you forgot to tell us if you set /O1 in both cases.
Newer compilers also tend to do more auto-vectorization and multiple versioning. If you have a specific problem with multi-versioning hurting performance, you should file the example on premier.intel.com. You could compare the opt-report between the compiler versions to see whether these attempted optimizations are involved.
If a new vectorization is not useful for you, perhaps on account of your loop count not matching the assumed value of 100 (in the case where the compiler can't estimate it from the context), you should consider
!dir$ loop count(your expected value).
