* CPU/Recompiler: Use rel32 call where possible for no-args
* JitCodeBuffer: Support using preallocated buffer
* CPU/Recompiler/AArch64: Use bl instead of blr for short branches
* CPU/CodeCache: Allocate recompiler buffer in program space
This means we don't need 64-bit moves for every call out of the
recompiler.
* GTE: Don't store as u16 and load as u32
* CPU/Recompiler: Add methods to emit global load/stores
* GTE: Convert class to namespace
* CPU/Recompiler: Call GTE functions directly
* Settings: Turn into a global variable
* GPU: Replace local pointers with global
* InterruptController: Turn into a global pointer
* System: Replace local pointers with global
* Timers: Turn into a global instance
* DMA: Turn into a global instance
* SPU: Turn into a global instance
* CDROM: Turn into a global instance
* MDEC: Turn into a global instance
* Pad: Turn into a global instance
* SIO: Turn into a global instance
* CDROM: Move audio FIFO to the heap
* CPU/Recompiler: Drop ASMFunctions
No longer needed since we have code in the same 4GB window.
* CPUCodeCache: Turn class into namespace
* Bus: Local pointer -> global pointers
* CPU: Turn class into namespace
* Bus: Turn into namespace
* GTE: Store registers in CPU state struct
Allows relative addressing on ARM.
* CPU/Recompiler: Align code storage to page size
* CPU/Recompiler: Fix relative branches on A64
* HostInterface: Local references to global
* System: Turn into a namespace, move events out
* Add guard pages
* Android: Fix build
While this has the potential to be slower, there seems to be quite a few
games which break with the larger FIFO. In some cases, they wait for the
DMA interrupt, then busy wait for the GPU to go idle again. If the FIFO
is larger, this wait time till be longer, breaking things.
This will cause a slight performance loss. I've left some knobs in which
can be tweaked to mitigate this, but the goal is to be compatible with
all games which require them.
