I thought a few of you may be interested in some timings that I've just made... with the first working ARM-coded interpreter + JIT compiler version....
The version written in C which we demoed at the RISC OS South East Show gives takes 19.4 seconds to render Apple4 centred on the screen at 1024x768x16million in ArtWorks.
Well the latest build completes the same test in 11.6 seconds so I've almost doubled the speed just by recoding in assembler. I should point out that there's still a lot of work to do on the JIT phase so this shouldn't be taken as representative of the final speed.
Rendering Apple4 is about the worst thing we can ask Aemulor to do, because it spends almost no time at all in the OS and thus everything is running under emulation. It's a very harsh performance test which is why I like it!
As an aside, Martin Wuerthner has been kind enough to send me some beta 32-bit rendering modules and this drops the rendering time down to 7.5 seconds... we're still emulating the main ArtWorks application code of course, but saving quite a bit of effort on the code that actually writes to the screen.
Right, time to sort out the JIT phase now, and then try out a few more of the possible optimisations on my (very) long list
Informal forum to keep you all informed on development progress
1 post • Page 1 of 1