Oolite Win32/64 users: Please test

[another_commander]

The link below will allow you to download an SSE2-optimized version of the Oolite for Windows deployment executable. Apart from SSE2 instructions, this build was created with compiler optimizations set to the maximum possible. Any feedback on its performance compared to the official 1.77 executable would be much appreciated. You need to download and install the latest nightly for Windows and just drop this exe in oolite.app to test it.

This is an experimental file and cannot be considered an official release. We are just trying to see to what point we can push performance and whether publishing SSE2-enabled versions of the game is something that is worth the effort, so your input will be valuable.

Thanks for your time testing this.
https://docs.google.com/file/d/0BwG6R5Q ... sp=sharing

[Tricky]

I've been using that setup for sometime, infact I added this to the end of ADDITIONAL_CFLAGS and ADDITIONAL_OBJCFLAGS

Code: Select all

-O3 -march=core2 -mtune=core2 -msse2

I think there might be some problems with some of the math code (seen unusual behaviour of ships not quite reaching their target, especially whilst docking, always stuck in GO_TO_COORDS), however using -O2 I have not seen this.

It definately screws up if I use -mssse3

[Cody]

Is this the same as the optimised version I already have, Admiral? (I've installed it anyway, into last night's nightly.)

[another_commander]

Is this the same as the optimised version I already have, Admiral? (I've installed it anyway, into last night's nightly.)

Yup, you don't need to re-download.

[another_commander]

I've been using that setup for sometime, infact I added this to the end of ADDITIONAL_CFLAGS and ADDITIONAL_OBJCFLAGS

Code: Select all

-O3 -march=core2 -mtune=core2 -msse2

I think there might be some problems with some of the math code (seen unusual behaviour of ships not quite reaching their target, especially whilst docking, always stuck in GO_TO_COORDS), however using -O2 I have not seen this.

It definately screws up if I use -mssse3

For reference, the set of (added) flags used here is:

Code: Select all

-Ofast -mfpmath=sse -msse2 -mtune=generic -mstackrealign

I have very often seen that mtune=generic generates better code than any march=whatever, plus the executables can run on any processor, with optimal performance on the newer ones.

If the unusual behaviours with docking etc. are confirmed also with this exe, then I guess we may have to remain with what we have. Let's give it a few days and see what we get, though.

[Tricky]

Those added flags don't cause any problems with performFlyToRangeFromDestination that I have seen so far. Strangely -O3 does cause it on a regular basis.

-O2 is probably a safer bet as it won't turn on -ffast-math or any of the other -O3 optimizations.

[Tricky]

I've ended up using -O0 (force no optimization) and left the SSE2 stuff in. Stops 99% of the stutters I was seeing. FPS max with no objects in view, goes down to ~60 FPS with the planet in view and ~30 FPS if viewing a ship/station at close range. That is with full shaders on.

• Vista Home Premium (32 bit) SP2
• Core2 Duo T7250 @ 2.0GHz
• 2Gb RAM
• GeForce 8600M GS/PCIe/SSE2

[JensAyton]

-O3 doesn’t imply -ffast-math. This is the main difference between -O3 and -Ofast.

[Rorschachhamster]

It seemed to work better and almost without stuttering here, too, on Windows 7 Ultimate 64 with a 2,4 GHz double core Intel board, 4GB and GeForce 8800 GT.
Just a little nervous about not having the latest version of trunk.

[JazHaz]

What does this mean? A 64 bit version of Oolite for Windows?

I don't know what SSE2 is. Looked on Wikipedia but I'm not very technically minded, and its over the top of my head lol. Plain English please!

I do have a supporting processor (Intel Core i7).

[another_commander]

What does this mean? A 64 bit version of Oolite for Windows?

I don't know what SSE2 is. Looked on Wikipedia but I'm not very technically minded, and its over the top of my head lol. Plain English please!

No, it means just a version of Oolite for Windows compiled with plenty of "go-faster" switches turned on, at the expense of not being able to run on older processors which don't support the SSE2 instruction set.

I do have a supporting processor (Intel Core i7).

Good, then you will be able to test this executable for us if you want.

[JazHaz]

I do have a supporting processor (Intel Core i7).

Good, then you will be able to test this executable for us if you want.

Already downloaded, along with last night's trunk.

[Cody]

Just to confirm - the latest trunk nightly (r5678) is not SSE2-optimised, yes?

[another_commander]

Just to confirm - the latest trunk nightly (r5678) is not SSE2-optimised, yes?

Correct. The nightlies are always configured as standard test release builds.

[Cody]

You've already had my feedback on the SSE2-optimised version.
As regards lag, I found r5678 to be the worst* so far, and have reverted to the SSE2 version.

(*only a dumb pilot's impression - I can only play my normal game and see how it works.)