Finding the end of the catch clauses

Mark Mitchell
Sun Mar 19 09:41:00 GMT 2000

>>>>> "Martin" == Martin v Loewis <> writes:

    Martin> When compiling large C++ programs, significant time is
    Martin> spent in push_to_sequence, when switching to the
    Martin> catch_clauses insn sequence. In particular, the time is
    Martin> spent in iterating over the growing insn list, every time
    Martin> a handler is added to the list.

Yup.  I've become increasingly convinced that the whole notion of
catch_clauses is wrong; we should use flow analysis to find them
later, and to generic code motion of unlikely code to the end of the
function.  (Such a mechanism would allow us to use profiling feedback
easily as well, for example.)  But, for now, we've got what we've got.

    Martin> The full bootstrap and testsuite run was successfully
    Martin> performed on i586-pc-linux-gnu. If approved, the cp and
    Martin> java Changelog entries will go in their separate change
    Martin> logs, of course.

This is OK.  Thanks!

Mark Mitchell         
CodeSourcery, LLC     

More information about the Gcc-patches mailing list