[d3d8] Fix up swizzle for all opcodes requiring a replicate swizzle

[WinterSnowfall]

Fixes #4113.

https://learn.microsoft.com/en-us/windows-hardware/drivers/ddi/d3d9types/ne-d3d9types-_d3dshader_instruction_opcode_type

The information in the docs isn't clear or complete, but it is at least clear that RSQ isn't the only opcode which requires the explicit use of a replicate swizzle for its inputs.

If the swizzle on inputs is somehow .xyzw / D3DVS_NOSWIZZLE (although that is technically illegal in the context of the affected opcodes) it appears native d3d8 does indeed impose a replicate swizzle, as @AlpyneDreams has already determined when looking at Indiana Jones And The Emperor's Tomb, and also based on empirical evidence it appears .w is the one that makes things look as expected, and in this case also fixes the missing fog in NFS Hot Pursuit 2.

Needs testing to ensure nothing regresses otherwise, but once that is done, I'll undraft.

[K0bin]

The shader bytecode came with the game and wasn't compiled at runtime, right?

[WinterSnowfall]

The shader bytecode came with the game and wasn't compiled at runtime, right?

Pretty sure that's the case for almost all d3d8 games, if not, in fact, all. Misunderstood your question. I'm not sure why d3d8to9 does the disassembly, but in d8vk we translate the declaration and function directly to a d3d9-style shader declaration and function (don't ask me any of the specifics here :P, what I do know at least is that they're encoded differently).

[WinterSnowfall]

Tested on the full complement of ~200 d3d8 games - found no regressions or other types of issues.