Hexagon - How to implement dotnew semantics in SLEIGH

[CUB3D]

Hi, I'm currently working on a plugin to add support for the Qualcomm Hexagon architecture (docs) to ghidra.

I've made pretty good progress on dissassembly, but this architecture has many quirks that don't fit well into the sleigh model, but handling of dotnew registers is one I can't see a solution for.

Instructions in Hexagon are grouped into "packets" of up to 4 concurrently executing instructions, when referencing a register the value will be the value that the register held before the execution of the packet (the old value). If an instruction needs to reference the value produced by the current packet then a dotnew can be used. As an example:

Before: r0 = 0, P0=0
{ P0 = cmp.eq( r0 , #0 )
  if (P0) jump:t 0x1000 }
After: P0=0xFF, jump is not taken

Before: r0 = 0, P0=0
{ P0 = cmp.eq( r0 , #0 )
  if (P0.new) jump:t 0x1000 }
After: P0=0xFF, jump is *taken*

This is possible to implement in SLEIGH by having a second set of registers and some pre/post packet logic to shuffle the values around. However that only works if the use of the dotnew comes after its generation. The following is an example from real code where that doesn't hold:

{          if (P0.new) jump:nt 0x6d38 ; r0 = 0x16 ; P0=cmp.gtu( r16 , 0x3f ) }

As you can see the value is generated two sub-instructions after it is used, so that strategy can't be used. I can't really see any other feasible ways to implement this in Ghidra, so I'm hoping someone might have an idea how this could be done (ideally with as little re-writing of sleigh as possible)?