- Paolo Redaelli
- 2011-10-27 @ 08:56
I tried to profile the compiler itself;
compile commands/compile_to_c --boost --profile -o profiling-compile
mv profiling-compile ../../target/bin/
profiling-compile commands/compile_to_c --boost --profile -o
profiling-compile --jobs 4
I'm not sure I understand the profile.se I got; you may get it at
http://www.box.net/shared/gjho80qk3fnej44q0bks (too big for librelist)
My wild guess is that tiny gains in POINTER.append_in and STRING.extend
would have big impact, or that we shall found how they are used in the
compile itself and provide use-specific variant.
I tried to find any clue on which classes to parallelize looking for
high time spent in relatively few invocation a single feature but I
coulnd't identify a clear pattern.