Lists

Terms and Conditions
Lists hosted on this site
Email the Postmaster
Tips for posting to public mailing lists

Re: [patch] massive libtool speedup for large libs

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [patch] massive libtool speedup for large libs

Subject: Re: [patch] massive libtool speedup for large libs
From: Jonas Maebe <email@hidden>
Date: Thu, 1 Mar 2007 13:13:56 +0100


On 1 mrt 2007, at 10:58, Olivier Tristan wrote:

Scott Wheeler wrote:
I reworked create_library() to skip the buffer stage and write directly to the target file. In the case above this knocks out about 60% of the runtime and leaves the system at least semi-responsive. There are other places that could be optimized, but the major bottleneck at this point is just writing the library to disk. I did binary diffs on the output of the system version and my own and the only different bytes are the time stamp.
It would be definitely great if this could be integrated in Xcode 3.0.
We suffer from the same problem and it looks very promising.

Note that judicious use of plain writes without internal caching can also have a quite detrimental effect on speed. E.g, Apple's ld64 does that (in 10.4 at least) and it's really a lot slower than the regular ld (which seems to write in blocks of at least 32KB). ld64 moreover even does a lot of zero-byte writes (from fs_usage):

13:03:24.539 pwrite F=3 B=0x0 O=0x00002230 0.000002 ld64 13:03:24.539 pwrite F=3 B=0x20 O=0x00002230 0.000006 ld64 13:03:24.539 pwrite F=3 B=0x0 O=0x00002250 0.000002 ld64 13:03:24.539 pwrite F=3 B=0x20 O=0x00002250 0.000006 ld64 13:03:24.539 pwrite F=3 B=0x0 O=0x00002270 0.000002 ld64 13:03:24.539 pwrite F=3 B=0x20 O=0x00002270 0.000006 ld64 13:03:24.539 pwrite F=3 B=0x0 O=0x00002290 0.000002 ld64 13:03:24.539 pwrite F=3 B=0x30 O=0x00002290 0.000006 ld64 13:03:24.539 pwrite F=3 B=0x0 O=0x000022c0 0.000002 ld64 13:03:24.539 pwrite F=3 B=0x20 O=0x000022c0 0.000006 ld64 13:03:24.539 pwrite F=3 B=0x0 O=0x000022e0 0.000002 ld64 13:03:24.539 pwrite F=3 B=0x20 O=0x000022e0 0.000006 ld64 13:03:24.539 pwrite F=3 B=0x0 O=0x00002300 0.000002 ld64 13:03:24.539 pwrite F=3 B=0x20 O=0x00002300 0.000006 ld64 13:03:24.539 pwrite F=3 B=0x0 O=0x00002320 0.000002 ld64 13:03:24.539 pwrite F=3 B=0x20 O=0x00002320 0.000006 ld64 13:03:24.539 pwrite F=3 B=0x0 O=0x00002340 0.000002 ld64 13:03:24.539 pwrite F=3 B=0x30 O=0x00002340 0.000006 ld64 13:03:24.539 pwrite F=3 B=0x40 O=0x00002370 0.000006 ld64 13:03:24.539 pwrite F=3 B=0x40 O=0x000023b0 0.000006 ld64

But even without the zero byte writes, performing a system call for writing anything less than 32KB or so is asking for slow performance.

Yes, I still have to file a radar for this (although I'm hesitant to spend time on it, since I suppose that ld64 will have gained significant improvements in 10.5 given that it's much more used there).


Jonas
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Xcode-users mailing list      (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden


Follow-Ups:

Re: [patch] massive libtool speedup for large libs
From: Laurence Harris <email@hidden>


References:  
  >Re: [patch] massive libtool speedup for large libs (From: Olivier Tristan <email@hidden>)




Prev by Date:
Re: [patch] massive libtool speedup for large libs

Next by Date:
Re: Is anyone still using CodeWarrior with 10.4., Universal?

Previous by thread:
Re: [patch] massive libtool speedup for large libs

Next by thread:
Re: [patch] massive libtool speedup for large libs

Index(es):

Date
Thread