I have the tokenizer and it's test suite completely rewritten in Fan!
Now I also have a good real world application to use for benchmarking between Fan and native Java code. Tokenizing should exhibit about the worst possible performance for Fan because it's all about high throughput character IO, and Fan takes a significant hit with the boxed Ints. I was hoping to start with a 3 to 4 times performance degradation, but I was disappointed to see a 7.5 times degradation - time to tokenize all of the Fan code in sysTest:
Java: 95ms
Fan: 708ms
So my step is to buckle down and make sure I can optimize the performance to a more reasonable delta.
brian Thu 1 Jun 2006
I have the tokenizer and it's test suite completely rewritten in Fan!
Now I also have a good real world application to use for benchmarking between Fan and native Java code. Tokenizing should exhibit about the worst possible performance for Fan because it's all about high throughput character IO, and Fan takes a significant hit with the boxed Ints. I was hoping to start with a 3 to 4 times performance degradation, but I was disappointed to see a 7.5 times degradation - time to tokenize all of the Fan code in sysTest:
So my step is to buckle down and make sure I can optimize the performance to a more reasonable delta.