d04047cc29bbc9d1f87a9346d1601e3dd87b6ca0 broken MSVC builds

[Mike Hommey]

On Sun, Jun 01, 2014 at 10:07:22PM -0700, Jason Evans wrote:
> On May 29, 2014, at 12:44 AM, Mike Hommey <mh+jemalloc at glandium.org> wrote:
> > d04047cc29bbc9d1f87a9346d1601e3dd87b6ca0 makes rtree.exe assert:
> > <jemalloc>: z:\jemalloc-dev\include\jemalloc/internal/arena.h:553:
> > Failed assertion: "ret == small_size2bin_compute(size)"
> 
> Fixed:
> 
> 	https://github.com/jemalloc/jemalloc/commit/0b5c92213fbafc52c5b5a5dc84e91eacc812ae0b
> 
> I apparently neglected to test the fallback implementations of lg_floor(), and perhaps unsurprisingly, both of them were broken.

Confirmed to fix it.

Which leaves us with the following problems on Windows:
- test/unit/mq.c doesn't build because of nanosleep()
- test/unit/SFMT.c crashes in gen_rand_array
- test/unit/tsd.c fails because of lack of initialization (See
  http://jemalloc.net/mailman/jemalloc-discuss/2014-May/000838.html)
- lg_floor fails to build on win64 because while LG_SIZEOF_PTR is 3,
  LG_SIZEOF_PTR and LG_SIZEOF_LONG are different.

Cheers,

Mike