All codegen in JSC runs through the MacroAssembler interface, which prevents any of the JITs from making machine specific optimisations (with the exception of those bits of code that are inherently architecture/os specific like ABI, etc). In general this means that there are no architecture specific optimisations. There are a couple of caveats to this statement: * The ARM backends have a much more complex linking pass that uses the most efficient instructions possible for any given branch, whereas the other architectures use exclusively long jumps. It would be nice to have all the architectures use the more advanced linker pass if for no other reason than reduced code size (we generate a lot of 5 byte jumps in x86 code when a 2 byte jump would be sufficient) * In theory any particular backend _could_ do peephole optimisations. One could even just generate a more complex IR that it performed further optimisations on (an extreme example might be an LLVM IR backend for instance) although given the usage of most JS on the web I suspect such an approach would be too expensive to be worthwhile. However I'm not sure just how much of a win architecture specific optimisations would give us at the moment, as I feel that our overall codegen quality is relatively poor (although the DFG does improve it substantially), and we can probably get a long way with more high level optimisation work. --Oliver On Jan 7, 2012, at 12:01 AM, wingoog moon wrote:

HI!! I know that DFG JIT can do some general optimizatiopns such as CSE,OSR,dead code eliminations. But all these optimizations are machine independent. So what machine dependent optimizations DFG JIT can do???(for inatance for ARM platform) _______________________________________________ squirrelfish-dev mailing list squirrelfish-dev@lists.webkit.org http://lists.webkit.org/mailman/listinfo.cgi/squirrelfish-dev