Attached is dlmalloc compiled to JS. Results: m-c -m -j -p 9.273 jm -m -j -p -n 4.139 v8 2.739 TI is much better than current spidermonkey, but still significantly behind v8. I think this is fairly important since dlmalloc is real-world code, not an artificial microbenchmark. One difference from microbenchmarks is the size of the code. In bug 678939 Jan mentioned that not compiling IFEQX might be a factor in such scripts, and that should be now done on jm but not m-c, which can potentially explain some of the slowness on m-c (but not that jm is still slower than v8).
Forgot to say, those numbers are with arguments 400 400
You need to log in before you can comment on or make changes to this bug.