Re: C++ exception code size / recommended practice

Re: C++ exception code size / recommended practice

[Benjamin Kosnik]

Hmm. Your data is surprising to me. The last time I measured this was
when RTH was putting the call frame bits in, and I found that the
overhead in libstdc++ between -fno-exceptions and the default was less
than 30%. I have not done this recently.

Some questions:

Are you on x86? Have you tried the gcc-3.2.x series, and is it comparable?

-benjamin

Re: C++ exception code size / recommended practice

[Tim Jansen]

On Thursday 10 April 2003 18:43, you wrote:
> Are you on x86? Have you tried the gcc-3.2.x series, and is it comparable?

On x86. I have compiled it with gcc 3.2.2 and a newer 3.3 snapshot, it's a 
somewhat smaller:

gcc 3.2.2, without exceptions 343964 bytes
gcc 3.2.2, with exceptions 485164 bytes
gcc 3.3 prerelease 20030407, without exceptions 353400 bytes
gcc 3.3 prerelease 20030407, with exceptions 507208 bytes

bye...

Re: C++ exception code size / recommended practice

[Tim Jansen]

On Thursday 10 April 2003 18:42, Joe Buck wrote:
> Actually, the extra size is, for the most part, not code at all; rather,
> it is tables (for specifying how the stack is to be unwound if an
> exception occurs). 

Yes, ca 50%. Here are some values (the same app as in my first mail, gcc 3.3 
prerelease on x86, but now with -O2):

Without exceptions:
Filesize: 350092 bytes
.text : 298400
.eh_frame: 4
.gcc_except_table: does not exist
.eh_frame_hdr: does not exist

With exceptions:
Filesize: 524732 bytes
.text: 387936
.eh_frame: 45340
.gcc_except_table: 31280
.eh_frame_hdr: 7948

With exceptions, -fno-enforce-eh-specs:
Filesize: 523836 bytes
.text: 385856
.eh_frame: 45364
.gcc_except_table: 30002
.eh_frame_hdr: 7948
(there are only a few throw() declarations)

bye...

Re: C++ exception code size / recommended practice

[Joe Buck]

On Thu, Apr 10, 2003 at 01:20:32AM +0200, Tim Jansen wrote:
> Hi...
> 
> I wonder what are the recommended practices for decreasing code size of C++ 
> applications with exceptions. I switched a project from -fno-exceptions to 
> regular exception support. This increased the size of the two binaries, after 
> stripping, from 370572 and 39660 bytes to 581308 and 58408, respectively. 
> Thus the increase in code size is > 50% (compiler: GCC 3.3 20030226 from SuSE 
> 8.2). 

Actually, the extra size is, for the most part, not code at all; rather,
it is tables (for specifying how the stack is to be unwound if an
exception occurs).  Since the tables land in a separate section, in most
cases they will not occupy RAM until an exception occurs, at which point
they are paged in from disk.  What this means is that your executable
consumes extra disk space, but the increase in RAM is far less.

Re: C++ exception code size / recommended practice

[Jason Merrill]

On Thu, 10 Apr 2003 01:20:32 +0200, Tim Jansen <ml@tjansen.de> wrote:

> Are there any hopes to make the overhead smaller on 3.x gccs? Is a throws() 
> clause in the declaration recommended or not? It would be desirable for API 
> documentation purposes and a paper claimed that empty throws() decreases the 
> code size in VisualC++, but with gcc I noticed another small increase.
> Or, possibly, could the reason for the increase be that the app uses a 
> -fno-exceptions library (Qt) extensively?

The -fno-enforce-exception-specs flag should improve the code size with
throw() specs.

Jason

C++ exception code size / recommended practice

[Tim Jansen]

Hi...

I wonder what are the recommended practices for decreasing code size of C++ 
applications with exceptions. I switched a project from -fno-exceptions to 
regular exception support. This increased the size of the two binaries, after 
stripping, from 370572 and 39660 bytes to 581308 and 58408, respectively. 
Thus the increase in code size is > 50% (compiler: GCC 3.3 20030226 from SuSE 
8.2). 

Are there any hopes to make the overhead smaller on 3.x gccs? Is a throws() 
clause in the declaration recommended or not? It would be desirable for API 
documentation purposes and a paper claimed that empty throws() decreases the 
code size in VisualC++, but with gcc I noticed another small increase.
Or, possibly, could the reason for the increase be that the app uses a 
-fno-exceptions library (Qt) extensively?

I would appreciate any suggestions.

bye...