thread-specific gmonparam structures and then merging them before dumping
the profile buffers.
Written by Nathan J. Williams, contributed by Wasabi Systems, Inc.
_ASM_LABEL(cerror) and _ASM_LABEL(curbrk) to _C_LABEL(__cerror) and
_C_LABEL(__curbrk) (or their respective architecture-specific equivalents) to
avoid possible name clashes with identifiers used in user applications.
* Do the same for minbrk on all architectures to avoid a GCC-specific (and
on ELF architectures effectively useless) symbol reference renaming in MI code.
> Prevent recursive invocation of mcount() while in kernel by setting profiling
> state to BUSY while in mcount(). This allows profiling to work when mcount()
> calls a profiled routine (for example, udiv on the SPARC).
using HASHFRACTION directly. in time-critical calculations, if HASHFRACTION
is a power of two, check that p->hashfraction == HASHFRACTION and if so do
the calculation with the compiled-in value so that the compiler can optimize
out (potentially) expensive divisions. if p->hashfraction != HASHFRACTION,
actually do the division. This has the result that on machines with slow
division, the division can be optimized out of the common case, but that
if HASHFRACTION changes from the compiled-in value (for whatever reason),
profiling will still work. Changes suggested by Chris Torek.
