Re: [PATCH 0/6] Convert gimple to a C++ class hierarchy

[David Malcolm <dmalcolm at redhat dot com>]

On Fri, 2013-08-30 at 15:44 +0200, Michael Matz wrote:
> Hi,
> 
> On Thu, 29 Aug 2013, David Malcolm wrote:
> 
> > Successfully bootstrapped and tested on x86_64-unknown-linux-gnu: all
> > testcases show the same results as an unpatched build (relative to
> > r202029).
> 
> I'd like to see some statistics for cc1{,plus} codesize and for compile 
> time of something reasonably large (needing say 60 seconds to 
> compile normally), before/after patch series.

Here's the result of a pair of builds of r202029 without and with the
patches, configured with --enable-checking=release, running "make", then
stripping debuginfo [1]

# ll */build/gcc/cc1
-rwxr-xr-x. 1 root root 13230048 Aug 30 15:00 control/build/gcc/cc1
-rwxr-xr-x. 1 root root 13230144 Aug 30 15:00 experiment/build/gcc/cc1
(98 bytes difference)

# ll */build/gcc/cc1obj
-rwxr-xr-x. 1 root root 13426336 Aug 30 15:00 control/build/gcc/cc1obj
-rwxr-xr-x. 1 root root 13426432 Aug 30 15:00 experiment/build/gcc/cc1obj
(96 bytes diff)

# ll */build/gcc/cc1plus
-rwxr-xr-x. 1 root root 14328480 Aug 29 13:59 control/build/gcc/cc1plus
-rwxr-xr-x. 1 root root 14328608 Aug 29 13:59 experiment/build/gcc/cc1plus
(128 bytes diff)

# ll */build/gcc/f951
-rwxr-xr-x. 1 root root 13960728 Aug 30 15:05 control/build/gcc/f951
-rwxr-xr-x. 1 root root 13960856 Aug 30 15:05 experiment/build/gcc/f951
(128 bytes diff)

# ll */build/gcc/jc1
-rwxr-xr-x. 1 root root 12607704 Aug 30 15:17 control/build/gcc/jc1
-rwxr-xr-x. 1 root root 12607704 Aug 30 15:17 experiment/build/gcc/jc1
(the same size)

So the overall sizes of such binaries are essentially unchanged.

To dig a bit deeper, I extended my asmdiff tool [2] to compare sizes of
functions; I'm attaching the results of comparing cc1plus before/after.

Any suggestions on what to compile to compare performance?  By 60
seconds, do you mean 60s for one TU, or do you mean a large build e.g.
the linux kernel?

> And the manual GTY markers are so not maintainable in the long run, 
> gengtype or something else really needs to be taught to create them 
> automatically.

Apart from the GTY aspect, how do people feel about the patch series?
FWIW I have vague thoughts about doing something similar for tree -
doing so *might* give an easier route to the type vs expression
separation that Andrew spoke about at the Cauldron rearchitecture BoF.
(I started looking at doing a similar C++-ification of rtx, but...
gahhhhh)

Dave

[1] yes, I built as root; this was done on a throwaway provisioning of a
RHEL 6.4 x86_64 box.
[2] https://github.com/davidmalcolm/asmdiff

Attachment: diff.txt
Description: Text document