2017/10/19

10:58

I can do stack caching in a non-traditional way so to speak. There is a tread of variables. Basically my idea is similar to the initialization states for variables. One of the major things is that variables can be copied all over the place and my previous attempts at keeping trails of variables were quite very much messy. However, instead what I can do instead of having only the initialization flags is having a unique data specifier so to speak. So every variable has a value which can be virtually copied to others and such. But, instead of allocating based on the Java locals or stack, all allocations are against the data allocation set. So basically any values which are tossed will just lose their allocation and such. Traversing to other states will just use copy operations for the most part to make things match. Methods with smaller number of variables will work better. It would be a good optimization. Then when it comes to register and stack allocations, those single data allocations points are what are considered. Then with this, it will be easier to tell when data points no longer have any allocations to them.

11:05

This will be the HIL, but I can do virtualized register allocation with just a very large set of registers. Native code compilers which have a limited set of registers can just instead determine how points are allocated. So that way the allocation of unique points purely are managed in an independent way without going into any machine specific areas such as the stack. So essentially an infinite register set is managed. Additionally, the number of variables used can be used as an indicator as to how the stack should be setup. So for example if only 6 variables are used and if there are enough registers to fit more than 6 and many other things, then there would not need to be stack placement at all (unless for like method calls).

11:20

Then that means no temporary variables are needed at all, because I would have used that previously for temporary things.

12:50

Also, push and pop operations can instead use the unique data variables instead of using the variables themselves and just refer to the data indexes instead of using the variables. This way I do not need to keep track of values in variables and worry about pushing and popping. Instead the operations are done on the data indexes rather than just the variables.

13:33

I can also optimize operations on values too. Since I can know some constants which are available at compilation time statically, than can be used to write better code for cases which would never normally be reached for example. Since I do know enumerated types and their structures, I can optimize those objects and treat them somewhat special as enumerations. The same can be said for strings too.