Refactoring to allow up to 64 registers

ReadyTalk / avian

[INACTIVE] Avian is a lightweight virtual machine and class library designed to provide a useful subset of Java's features, suitable for building self-contained applications.

https://readytalk.github.io/avian/

Other

1.22k stars 173 forks source link

Refactoring to allow up to 64 registers #377

Closed joshuawarner32 closed 9 years ago

joshuawarner32 commented 9 years ago

This does several things:

Split the single 64-bit mask type used in plan* functions into two 64-bit fields of the "RegisterMask" class
Make Register a class, replacing many previous "int"s and adding a bit of extra type safety
Rename lir::Register to lir::RegisterPair, for clarity
Other tangential refactorings

This compiles on:

x86_64 (linux, mac)
i386 (mac)
arm (linux and ios)

Additionally, all the tests pass on the above platforms, with the obvious exception of ios-arm (lacking a jail-broken device).

Future improvement:

I'm not a huge fan of the (int8_t) casts this sprinkles around, but I'm not sure of a better way to do that, while preserving the added type safety

joshuawarner32 commented 9 years ago

@dicej, in case my IM client wasn't working, I pushed a version of your in-progress work rebased on top of this (conflicts fixed) to https://github.com/joshuawarner32/avian/tree/dicej-master-rebased. I'm happy to do the same with any further work you might have done, but not pushed yet.

dicej commented 9 years ago

Looks good to me. Let me know when the tests have passed and I'll merge it.

joshuawarner32 commented 9 years ago

I force-pushed an update to my latest commit, to fix the cmake build.

The tests now pass for me on:

darwin-{x86_64,i386}-debug
linux-x86_64
linux-arm-debug
test/ci.sh on linux-x86_64

./test/ci.sh fails for me on darwin, but that's a separate issue (I the very latest java 1.8 installed, and it fails in the openjdk build): Assertion failed: (cl->fixedSize + (cl->arrayField ? cl->arrayField->elementSize : 0) < 32 * BytesPerWord), function typeObjectMask, file src/tools/type-generator/main.cpp, line 1332.

dicej commented 9 years ago

Travis still doesn't like it. Did you push to the right branch?

joshuawarner32 commented 9 years ago

Yeah, the new travis failure is just because travis apparently uses a slightly older version of gcc, which doesn't like the member begin() and end() methods for the enhanced-for loop. I just pushed an update for that.

joshuawarner32 commented 9 years ago

Travis seems to be taking an excessively long time to even start the build lately. I wonder if they're running into a resource wall and severely rate-limiting open-source projects.