Gcc4

[rmrfchik]

The question for Laze and Sashz.
I can do now to try to build new gcc, new lib, new qt, but want to know: are you already did/doing this?

And, in general, what is your plans for this software?

[omro]

This might be a dense question, lol. What is the benefit of doing this?

[rmrfchik]

This might be a dense question, lol. What is the benefit of doing this?
[div align=\"right\"][a href=\"index.php?act=findpost&pid=93088\"][{POST_SNAPBACK}][/a][/div]

Gcc4 support "visibility" keyword, so qt4 will be linked much fatser at runtime.

[omro]

Would this be for KDE and KDE based apps or just for general use?

[rmrfchik]

Would this be for KDE and KDE based apps or just for general use?
[div align=\"right\"][a href=\"index.php?act=findpost&pid=93101\"][{POST_SNAPBACK}][/a][/div]

And for kde too

[omro]

Would this be for KDE and KDE based apps or just for general use?
[div align=\"right\"][a href=\"index.php?act=findpost&pid=93101\"][{POST_SNAPBACK}][/a][/div]

And for kde too
[div align=\"right\"][a href=\"index.php?act=findpost&pid=93117\"][{POST_SNAPBACK}][/a][/div]

If gnome is considered too slow for the Z, surely KDE would be too? Plus isn't KDE a bit lairy and cluttered and iconified for such a small screen?

[adf]

most of the people on the kde thread seem to think it is a bit heavy. There are some advantages to installing kde, though.  Mainly the ability to run a recent konqueror in a another wm (like xfwm, or the default wm), make use of kde apps like kate and konsole in the same way.  Some of this would be true for gnome, as well, of course.

kde looks fine on the z, and is useable, though it lags a little sometimes.

Re gcc4  read some tests run by someone on the project (I think..it was a while ago and I was actually looking for somthing else at the time) I got the impression that it won't make anything faster or smaller.

[omro]

Which would actually run better on the Z? Gnome or KDE?

[adf]

no idea.  you might look into benchmarks on desktops if anyone has done them.  I expect them to be about the same, or pretty close.

[Da_Blitz]

i wonder if they have the autovectorisation anbled for the iwmmx instructions, i bet we would see a speed improvment on the C1K and c3x00's then,

apart from the the new optimization framework looks like it will give much better resaults in the next version or 2, for those who turn optimisation flags on

ethier way i can see the biggest speed advantage comming from using the autovectorize stuff to utilize the iwmmx instructions automatically without the coder specfically writing the program with mmx code in it, however code generated would not work on any other model except the c1K/c3x00  which is a shame because i expect we would see some great resaults

oh well guess its time for me to get my feet wet and start up a seperate ipkg tree

[adf]

are you sure iwmmx stuff wouldn't run?  mmx and sse optimized desktop stuff seems to run on processors without those features.  I'd gain nothing on my 6k, but I'd be happy to test something to verify this, if necessary.

[Da_Blitz]

it depends wether your code is written to provide fall back for cases where thos features are unavalible (try disabling runtime cpu detection in mplayer and enabling 3dNow then run that binary on a Intel chip)

the only cases where this would work is where the instruction would rasie and interupt and the kernel has software routines to hadel this (slower than just coding without vectorisation as you now have os overhead)

however under the xscales the iwmmx is imlemnted as a coprocessor and not as instructions (ie it uses an instruction which sends some bytes to the coporcessor that the coprocessor understands as an instruction (sort of like a psedo instruction))

this coprocessor only exsists under the pxa27x series, i belive that under the pxa25x pxa26x there may be a diffrent coprocessor or none at all and therefore you ethier get an error or it does nothing resualting in a hard to debug error

as a side note you need an iwmmx enabled kernel otherwise the program will be killed by the kernel (as it would be classed as an "illegal instruction") when it does somthing it shouldnt do (like use an iwmmx instruction) you would get somthing like seg fault on the command line