Re: RFC: Speeding up libstdc++.so with --dynamic-list-data

["H. J. Lu" <hjl at lucon dot org>]

On Mon, Jan 08, 2007 at 08:09:59PM -0800, Andrew Pinski wrote:
> On Mon, 2007-01-08 at 18:25 -0800, H. J. Lu wrote:
> > I am enclosing a patch to implement a new linker swicth,
> > --dynamic-list-data. It is -Bsymbolic for function symbols only.
> > I tried it with C, C++, Java and Fortran on Linux/ia32, Linux/x86-64
> > and Linux/ia64. There are only a few regressions. The function calls
> > within the new resulting DSOs will bind locally. It speeds up
> > the enclosed C++ testcase by
> > 
> > 1. 23% on ia64.
> > 2. 6% on ia32.
> > 3. 3% on x86-64.
> > 
> > Should we consider such optimization?
> 
> The real question is, does this work with operator new?
> 
> In that if I a C++ developer provides a seperate operator new (and
> delete), does libstdc++ use that one as required by the C++ standard?

These are the regressions I was talking about. I can support them with
a new linker switch. So far, we have

  --dynamic-list-cpp-typeinfo Use C++ typeinfo dynamic list
  --dynamic-list FILE         Read dynamic list

My current proposal adds

  --dynamic-list-data         Add data symbols to dynamic list

I can add a new one

  --dynamic-list-cpp-new     Add C++ new/delete to dynamic list

Then we can build libstdc++ with

--dynamic-list-data --dynamic-list-cpp-new


H.J.