gbenson.net

Today I got my second method working, String.hashCode(). Now I have conditional and unconditional branching, field access, array loads, a whole bunch of integer operators, and returning with a result implemented and (somewhat) tested. The bytecode coverage chart says I’m 50% done, but I don’t believe it.

Well, it’s taken a month and a half — and over 2000 lines of code — but I finally got a method out of Shark.

I made a chart showing which bytecodes are implemented, which I’ll keep updated as I progress. The estimated total coverage of 18% is slightly fanciful as it treats all bytecodes as equally complex, with nop having the same weight as new for example. Some codes are marked as complete but untested too. The way the compiler is structured means that in simple cases I can copy and paste whole blocks of bytecodes from the server compiler, so where I was doing one bytecode in a block I’ve copied the lot across. Most of them ought to be fine, but a couple are dubious. I’m still shuffling things around to try and make things less so.

Onwards…

Shark just JITted its first bytecode:

load i32* %local0               ; :15 [#uses=1]
inttoptr i32 %sp to i32*        ; :16 [#uses=1]
store i32 %15, i32* %16
%sp1 = sub i32 %sp, 4           ;  [#uses=2]

Ladies and gentlemen, it is aload_0!

Yesterday I had a mini-milestone in that I got enough bits and pieces written to have HotSpot call my JIT’s compile_method() method. The next trick will be to get a piece of code back into HotSpot without it being relocated on the way.

Of course, the zero interpreter doesn’t have profiling, so it only works if you force it with -Xcomp at the moment.

Also, I disabled InlineIntrinsics, which reminded me that there was some kind of fast accessor thing for JNI that I disabled in the original ppc port. I should fix that at some point.

Towards the end of last week I started trying to figure out how to slot an LLVM-based JIT into IcedTea. I had a vague idea that I could simply port C1 to it, but I’m not sure that would work so well. Both existing JITs expect to manage register allocation, for example, which LLVM will have to do. So I started hacking at the build system to have it build an entirely new JIT. I called it “shark” — I was going to call it C0 what with the other JITs being called C1 and C2, but I didn’t like the mild confusion that having them named and numbered causes, and shark was too good a name to waste on a one-off dummy architecture.

And speaking of one-off dummy architectures, my “if I can build on a platform that doesn’t even exist then I can build on anything” proved short lived — it doesn’t help on architectures that do exist but have really stupid unames that need truncating. I’m talking about ARM, of course, uname freak of the cosmos. So I’m going to try and get that working this week too.

As of about an hour ago it should be possible to build icedtea6 on any Linux system with no more effort than ./configure and make.

I’m in the final stages of a build system for zero that will work on any architecture. For kicks, I hacked my /bin/uname to say “shark” instead of “ppc”. I figure if I can build on a platform that doesn’t even exist then I can pretty much build on anything.

Xerxes Rånby sent me this photo today.

I got two emails with zero patches yesterday. One I was expecting, from Yi Zhan of Intel who has it working on ia64; the other was totally out of the blue, from Ruediger Oertel of SuSE who has it building on s390 and s390x. I always hoped it would Just Work™ but I never quite believed it! Now all I need is the time to integrate it all!

Which I may just have. While I was poking around trying to fix the IcedTea 7 stack overflow I discovered that a) we mark a lot of stack unusable with 64kb pages, and b) the stack region locator doesn’t seem to work, at least for the cases with guard pages, or for ia64, or possibly with 64kb pages when the Java thread is an initial thread. I spent some time looking into it and rewriting it, and I think I have it sussed. (If you ever wanted to know where ia64 puts its guard pages then you need this!) Now it all seems to work I’m going to spend the afternoon writing some emails to the HotSpot list, then next week I intend to hole up and implement my holy grail of a zero build system that does not require any patching to build on a new architecture.

So don’t expect any action from me until that’s done :)

The 64kb pages issue turned out to be really simple. HotSpot reserves a bunch of pages at the end of the stack for overflow recovery. There’s one “red” page, two “yellow” pages, and three “shadow zone” pages. I don’t know what any of these pages actually do, but I do know that a 512k stack of eight 64k pages less six reserved pages and one glibc guard page leaves… 64k for everything else. It was throwing a StackOverflowError because it was actually running out of stack.