Luke Gorrie
Luke Gorrie
I am finally writing the code to decode RaptorJIT logs as text in the spirit of `-jdump`. The code is on a feature branch of Studio (studio/studio#122) and I previously...
[Incremental Dynamic Code Generation with Trace Trees](https://www.cs.montana.edu/ross/classes/fall2009/cs550/resources/Tracemonkey-01.pdf) is relevant literature from tracing JIT research at Mozilla on TraceMonkey. The claim is that they solve "the big one" of efficiently compiling...
The error message could be more helpful when you write a program that exhausts the number of 16-bit FFI ctype-id values. Currently says `table overflow` but perhaps better with e.g....
@alexandergall reported a hardcode JIT optimization issue over at https://github.com/snabbco/snabb/issues/1393. He sees the issue in a real non-trivial Snabb application. But the underlying issue seems to be very basic and...
The JIT specializes FFI code using exact CTYPE IDs. This means that each FFI value in the final trace is required to have one specific FFI type. A consequence of...
What is the `D` operand of the `FUNCC` bytecode, really? Seems to me like the [assembler VM interprets this operand](https://github.com/LuaJIT/LuaJIT/blob/v2.1/src/vm_x86.dasc#L5401-L5411) as being `args+1` and uses it to calculate the stack...
I would like to make it easy to extract fine-grained performance metrics for the virtual machine. For example, the distribution of time in nanoseconds for events like a GC step;...
How can we effectively test a port of the bytecode interpreter? We do need to port this to C and I am wondering how we can test the port effectively....
[RaptorJIT 1.0.0 is released](https://github.com/raptorjit/raptorjit/releases/tag/v1.0.0) as of this moment!
This is an idea to solve #107 by optimizing side traces more efficiently. The problem is that side traces only inherit snapshot values from their parents. The solution is for...