Wolfgang Lenerz wrote: >> Thanks. Actually QPC2 v3.21 will even be 10-15% faster. :-) > C'me on, don't let it hang like that, we (I) want more details!!!!
The m68k core code is basically very old and recently I have continued to clean it up, i.e. add some comments, change old comments that still were in German (I've switched to commenting in English ages ago, just shows how old some parts are) and so on. Then I started playing around with it. Since 3.20 I use a different assembler which gives me more control of the alignment of code, which can be an important speed factor. The bigger the alignment the better the processor can access the code, but bigger alignment also means bigger code, which in turn is bad for the cache... it's a delicate balance. Then I did re-evaluate some alternative implementations of some commands I did some years ago. Processors have changed since then, caches got bigger etc, so code that previously was slower can be faster now. Finally I did re-optimize the main emulation loop for more current processors (which is only 8 instructions big, but the instruction and memory access order is very important), which actually gave the biggest boost. Marcel _______________________________________________ QL-Users Mailing List http://www.quanta.org.uk/mailing.htm