Walter Bright:

>It turns out that dmd's runtime library function had a crummy implementation 
>of long division in it.<

In that case I have spent few hours narrowing the problem to a very small 
benchmark. Then I have left to you to spot and fix the precise cause of the low 
performance. And today I know x86 assembly a bit better :-)


> I know a person who has done it (looked at the JIT output) with Java using a 
> debugger,

I have done this often, many of my bug reports for LLVM compare the asm 
produced by LLVM with the asm produced by the Sun JavaVM:
http://weblogs.java.net/blog/2008/03/30/deep-dive-assembly-code-java
See for example:
http://llvm.org/bugs/show_bug.cgi?id=5501

Bye,
bearophile

Reply via email to