Add Register Allocator

[drcjt]

No register allocation is done during compilation. The intermediate representation is stack based and this is converted into stack based operations of the Z80.

There are a couple of complications with adding register allocation support:

• The Z80 does not have many registers
• Numeric data types in C# are typically 32 bit based, e.g int, which require a pair of 16 bit registers on the Z80 making the lack of registers even worse.

Using a typical solution for register allocation may result in a lot of spilling due to the above resulting in potentially worse performance. Some experimentation will be required to determine if register allocation is worth it.

Note that the compiler will pass parameters in registers in certain limited cases such as when calling intrinsic methods.

[drcjt]

Dotnet runtime uses Linear Scan Register Allocator - https://github.com/dotnet/runtime/blob/main/docs/design/coreclr/jit/lsra-detail.md

SDCC uses Optimal Register allocator (in polynomial time) - https://link.springer.com/content/pdf/10.1007/978-3-642-37051-9_1.pdf

Some notes from the latter paper on Z80 architecture:

• Nine 8-bit registers are assigned by the allocator (A, B, C, D, E, H, L, IYL, IYH). IYL and IYH can only be used together as 16-bit register IY;
• There are instructions that treat BC, DE or HL as 16-bit registers
• Many 8-bit instructions can only use A as the left operand, while many 16-bit instrutions can only use HL as the left operand.
• There are some complex instructions, like djnz, a decrement-and-jump-if-not-zero instruction that always uses B as its operand, or ldir, which essentially implements memcpy() with the pointer to the destination in DE, source pointer in HL and number of bytes to copy in BC.