Hi, On Sun, Dec 04, 2011 at 05:00:41PM +0100, Jan Hubicka wrote: > > > No to what? To the fact that HOST_WIDEST_INT is host-dependent > > > and thus should not be used to drive code generation? Or no to the > > > fact that we can (and do) use int64_t as host integer type? > > > > No to the fact that int64_t should be used (and the occurrences in the LTO > > code > > are OK). hwint.h is precisely supposed to insulate the compiler from the > > host > > (of course we all know that this isn't 100% true) and HOST_WIDEST_INT is > > the > > proper type to be used in this case, see existing examples. > > Yep, most of the profiling code (where gcov_type is HOST_WIDEST_INT) is not > safe when compiled on host with 32bit ints only. This was always considered > acceptable given that bootstrapped compiler will have long long and stage1 > compiler won't excercise the limits. This case seems same to me. So I would > preffer the existing practice using HOST_WIDEST_INT. There is no need to have > some code using int64_t and other HOST_WIDEST_INT for same reason. > > The actual patch is OK either with HOST_WIDEST_INT or in64_t based on where > this discussion will lead. >
I have no strong preference either and so far it seems that HOST_WIDEST_INT prevailed slightly so I will re-test the patch and commit it as it is if there are no new problems. Thanks, Martin
