Re: [PATCH] [C++] save space in language_function

[Gabriel Dos Reis <gdr at integrable-solutions dot net>]

Andrew Pinski <pinskia@physics.uc.edu> writes:

| On Jan 5, 2004, at 03:39, Gerald Pfeifer wrote:
| 
| > On Sun, 4 Jan 2004, Andrew Pinski wrote:
| >> I should have said I am going to post a series of C++ patches which
| >> helps the memory usage on the famous PR 8361.
| >
| > This is highly appreciated!  Do you have concrete numbers on how these
| > patches (cumulatively and/or individually) improve the situation?
| 
| I had write wrote down the numbers some where but I cannot find it.
| But I remember it is around 24K savings, not much but a start.
| There is already a bug filed against 3.4 for memory usage when
| instantiations
| templates and I had thought I had a patch but I was warned by RTH to
| that I have to
| be careful with a patch in this area.

When I was working on compile-time improvement for Apple, I tried many
different versions of patches that reduce memory consumption by using
bitfields.  In fact, as ever, it was not obvious that one gets that huge
-memory- saving (that is why it would be good to send numbers as
reported by say -fmemory-report).  The issue is a time-space
optimization. If you use a bitfield, some acrhitectures might need
more instruction to extract or store information there.  So you might
end up with a speed slowdown.  So you will also need to send report on
speed (say by -ftime-report).

-- Gaby