(Graphics.Shader): Handle EmitSuatom constant dests and EmitSuld zero dest reg.

[LDj3SNuD]

Contributes to Ryujinx/Ryujinx-Games-List#2828.

Now the game should go ingame.

Reference: #5502 (comment)

[github-actions]

Download the artifacts for this pull request:

• ryujinx-Release-1.1.0+e7e4c47-linux_x64
• ryujinx-Release-1.1.0+e7e4c47-win_x64

Experimental GUI (Avalonia)

• ava-ryujinx-Release-1.1.0+e7e4c47-linux_x64
• ava-ryujinx-Release-1.1.0+e7e4c47-macos_universal
• ava-ryujinx-Release-1.1.0+e7e4c47-win_x64

GUI-less (SDL2)

• sdl2-ryujinx-headless-Release-1.1.0+e7e4c47-linux_x64
• sdl2-ryujinx-headless-Release-1.1.0+e7e4c47-win_x64

Only for Developers

• ava-ryujinx-Debug-1.1.0+e7e4c47-linux_x64
• ava-ryujinx-Debug-1.1.0+e7e4c47-macos_universal
• ava-ryujinx-Debug-1.1.0+e7e4c47-win_x64
• ryujinx-Debug-1.1.0+e7e4c47-linux_x64
• ryujinx-Debug-1.1.0+e7e4c47-win_x64
• sdl2-ryujinx-headless-Debug-1.1.0+e7e4c47-linux_x64
• sdl2-ryujinx-headless-Debug-1.1.0+e7e4c47-win_x64

[gdkchan]

Can you fix EmitSuld too? It has the same issue. But in that case, dests array will be empty if srcB is equal to RegisterZeroIndex. There are places that accesses dests[0] so it is assuming that the size will be at least 1. Also I'm not sure what the effect of passing an empty array on the Operation would be in that case. For the EmitSuld case I think just doing if (srcB == RegisterConsts.RegisterZeroIndex) return; (properly formatted) would be fine since unlike SUATOM, SULD with a zero destination register should be a no-op.

[LDj3SNuD]

surely, in a few hours

[LDj3SNuD]

okay, I will correct, I have no experience in GPU

[LDj3SNuD]

Let me know if this is okay; and if I need to increment any flags.
I confirm that the game works with those changes.

[gdkchan]

lgtm, thanks.

[gdkchan]

Let me know if this is okay; and if I need to increment any flags. I confirm that the game works with those changes.

It looks correct now (please test with Jurassic Park to be sure, with shader cache disabled or purged, otherwise it might be using the shaders from your previous test).

There is no need to increment the shader cache version since the bug makes it crash before the shader is generated, so the emulator wouldn't have a chance to save the broken shader in the cache.

[LDj3SNuD]

Ok thanks, I always retest with the caches disabled or cleared.
About EmitSuld, the game would seem to call that method only with 2, 3 and 38 for srcB; until ingame at least.

[riperiperi]

LGTM