Compiling Fastest Binaries

[comp1]

Hi guys,

Lately I've been working on mastering (or improving) my compiling skills with simple sources.

I've been playing around with sources of Ilia Muraviev's programs since they are quite easy to compile and don't require any modification/cleaning of the code to compile successfully.

Attached are sources and my compiles for LZPXJ 1.2h, PX 1.0, and QUAD 1.12.

I wanted to get some feedback and to see if anyone else could compile even faster binaries. And if so, then I have more to learn. The idea is to compile faster Win32 executables.

Here are my quick results for compression only with timer 3.01 for ENWIK8:

LZPXJ 9

   ILIA | 26.536s
  COMP1 | 23.151s
SPEEDUP | 12.76%

PX

   ILIA | 51.246s
  COMP1 | 42.339s
SPEEDUP | 17.38%

QUAD -x

   ILIA | 16.084s
  COMP1 | 13.790s
SPEEDUP | 14.26%

Curious to see how much faster we can go from the expert compilers!

[encode]

These are quite old compiles. So no surprise to me...

[comp1]

These are quite old compiles. So no surprise to me...

I know the programs are old now but I was using them as a starting point. Size optimization is much easier than speed optimization. As an example, I was able to compile PX to 12.5 KB and only slightly slower than my speed compile.

[nikkho]

Do not know what compilation switches changes did, but probably using latest ICL, GCC or Visual C++ with PGO willl give a boost.

[joerg]

@comp1: you wrote you want compile fastest binaries - then i am very interested in this "older" program algorithms - but they are parallel algorithms ...

parallel BZIP2: https://launchpad.net/pbzip2/1.1/1.1...-1.1.13.tar.gz

in case of PBZIP2 in the past are visible big diffences: http://encode.su/threads/253-Paralle...ll=1#post11595

interesting seems too: parallel lzip: http://download.savannah.gnu.org/rel...1.5-rc1.tar.lz

best regards

[euaqa]

And, above me...

Fastest with/from... to...

Example: Compiler | target-platform | testet hardware.

[snowcat]

Can I change the source code?
My idea is remove c library, only use win api, so printf can't be use...

ps: much smaller and a little faster

[m^2]

Can I change the source code?
My idea is remove c library, only use win api, so printf can't be use...

ps: much smaller and a little faster

Faster? Show me benchmmark results.

[comp1]

Can I change the source code?
My idea is remove c library, only use win api, so printf can't be use...

ps: much smaller and a little faster

I apologize for replying so late.

No, the idea is not to change the source code.

But if you can get a significant performance improvement from your suggested changes, please share the binary and some benchmarks.

This is not a competition--just a thread to share knowledge and see what improvements can be made in terms of speed.