Very long instruction words

[marekjm]

Implement a VLIW architecture inside the VM. This will require major changes to the bytecode format, and assembly language syntax, but can lead to performance improvements.

Some more information about VLIW on IBM Research.

[marekjm]

Suggested syntax:

{ offset... } instruction operands...

where the { offset... } part is optional (empty offset list means that only one instruction is executed at a time). Many offsets can be specified. Assembler translates the local-instruction index offsets to absolute offsets.

Example code:

.function: main
    {1} istore 1 42
        istore 2 17
    {1} print 1
        iinc 2
    {1,2,3} print 2
        nop
        izero 1
        izero 0
    end
.end

[marekjm]

Closing due to changes in design. Implementing instruction level parallelism put on hold until some undefined moment in the future.