Announcement
Collapse
No announcement yet.
why do the nativ linux clind of Heroes of Newearth not work with the radeon driver?
Collapse
X
-
Originally posted by Pontostroy View Postpentium g620
opensuse + my repo with git mesa, kernel, xf86-drivers, etc.
Shader caching and llvm shader compiler are enabled.
I use old( 5-6 months) non updated verson of HoN.
yes maybe they fix this in the future.Phantom circuit Sequence Reducer Dyslexia
Comment
-
Originally posted by Pontostroy View Postpentium g620
opensuse + my repo with git mesa, kernel, xf86-drivers, etc.
Shader caching and llvm shader compiler are enabled.
I use old( 5-6 months) non updated verson of HoN.
Comment
-
Originally posted by bridgman View PostPontostroy, did you see much performance difference from enabling the LLVM shader compiler (not the shader cacheing) ?
But all other games shows difference about 1-2% not in favor of LLVM shader compiler.
http://openbenchmarking.org/result/1...BY-R600LLVM811
Comment
-
Originally posted by Pontostroy View PostIn HoN at start position - 31-32 fps, with R600_LLVM=0 - 27 fps. (mesa without shader caching patch )
But all other games shows difference about 1-2% not in favor of LLVM shader compiler.
http://openbenchmarking.org/result/1...BY-R600LLVM811
Is the table below correct ?
- without LLVM, without shader cache : 27 fps
- without LLVM, with shader cache : ???
- with LLVM, without shader cache : 31-32 fps
- with LLVM, with shader cache : 38-45 fps (from your earlier post)
Comment
-
Originally posted by bridgman View PostThanks. The PTS numbers were more in line with what I had been expecting, but it's nice to find at least one app that runs faster with LLVM.
Is the table below correct ?
- without LLVM, without shader cache : 27 fps
- without LLVM, with shader cache : ???
- with LLVM, without shader cache : 31-32 fps
- with LLVM, with shader cache : 38-45 fps (from your earlier post)
- without LLVM, without shader cache : 27 fps
- without LLVM, with shader cache : 31-32 fps
- with LLVM, without shader cache : ??? (there is no benchmark result for this case)
- with LLVM, with shader cache : 38-45 fps (from your earlier post)
more advance explanation :
Question:Pontostroy, did you see much performance difference from enabling the LLVM shader compiler? (not the shader cacheing) ?
Logic analyses result- (FPS without LLVM )and (without shader caching)
answer to this part: (did you see much performance difference from enabling the LLVM shader compiler?) In HoN at start position - 31-32 fps, with R600_LLVM=0 -
and then the logical part for " (without shader caching)" 27 fps. (mesa without shader caching patch )Last edited by Qaridarium; 06-14-2012, 07:03 AM.Phantom circuit Sequence Reducer Dyslexia
Comment
-
Originally posted by bridgman View PostThanks. The PTS numbers were more in line with what I had been expecting, but it's nice to find at least one app that runs faster with LLVM.
Is the table below correct ?
- without LLVM, without shader cache : 27 fps
- without LLVM, with shader cache : ???
- with LLVM, without shader cache : 31-32 fps
- with LLVM, with shader cache : 38-45 fps (from your earlier post)
The HoN hard to benchmark, at the starting point:
- without LLVM, without shader cache : 27 fps
- without LLVM, with shader cache : 32-33 fps
- with LLVM, without shader cache : 31-32 fps
- with LLVM, with shader cache : 31-32 fps
38-45 fps the average values in different places.
Comment
-
Originally posted by Pontostroy View PostThe HoN hard to benchmark, at the starting point:
- without LLVM, without shader cache : 27 fps
- without LLVM, with shader cache : 32-33 fps
- with LLVM, without shader cache : 31-32 fps
- with LLVM, with shader cache : 31-32 fps
38-45 fps the average values in different places.
and the expected fps for my card with your driver in use should be 25,6fps.
something is wrong in my systemPhantom circuit Sequence Reducer Dyslexia
Comment
-
Q, are you still running at 1920x1200 vs Pontostroy's 1600x900 ? If so that's almost 1.6x the pixel count right there. Factor in 1.25x for shader differences and 1.06x for clock differences and you're at maybe a 2.1x difference, so your frame rates would be sub-15 fps if everything scaled linearly (which doesn't actually happen but it's a start).
Maybe match resolution then match HON version as a next step ?
Comment
-
Also note, as I mentioned in another thread, that the llvm backend doesn't support indirect addressing yet. If the shader uses indirect addressing, the driver will use the non-llvm backend. A lot of games use indirect addressing.
Comment
Comment