V_CMP_U64 semantics

[w1naenator]

Make Knack 2 menu fully operational.

[w1naenator]

[georgemoralis]

one more time cmp , @raphaelthegreat should check that :D

[raphaelthegreat]

I don't think this is correct, CMP_U64 is most often a thread-bit compare that will not work with U64 src. Did you confirm this is what the shader uses?

[w1naenator]

I don't think this is correct, CMP_U64 is most often a thread-bit compare that will not work with U64 src. Did you confirm this is what the shader uses?

AMD GCN3 ISA: https://docs.amd.com/v/u/en-US/gcn3-instruction-set-architecture
Vulkan subgroup docs: https://docs.vulkan.org/guide/latest/subgroups.html

AMD's GCN3 ISA documents V_CMP_U64 as a VOPC compare on unsigned 64-bit integers (PDF p. 228 / printed p. 12-122), and section Vector Compares: VCC and VCCZ defines vector compares as per-lane writes to VCC[n] = EXEC[n] & test (PDF p. 32 / printed p. 3-8). The ISA also notes that VCC physically resides in the SGPR register file and counts against SGPR sourcing limits (PDF p. 32), which explains why thread-bit idioms can happen, but does not make V_CMP_U64 itself a special thread-bit compare instruction.

So for the ISA semantics, the generic U64 compare lowering is justified. However, your question about the specific Knack 2 shader is fair: I do not yet have shader-specific proof from a dump/disassembly showing whether this instance is using a full 64-bit compare or a VCC/thread-bit idiom.

[Parotaku]

Once again, fixing V_CMP_U64 helps NO STRAIGHT ROADS too, showing videos & going further...

[tochnonechelovek]

KNACK™ 2 Demo - CUSA09687 goes ingame, just doesn't show graphics properly. Menus work as intended.
CUSA09687.log

RESIDENT EVIL 3 "Raccoon City Demo" - CUSA17925 also requires V_CMP_U64, but still crashes at launch, now on this error:
[Debug] <Critical> (GameThread) signals.cpp:70 SignalHandler: Unhandled Exception code 0xc0000005 at 0x802ad2a5b
CUSA17925.log

[gandalfthewhite19890404]

[Debug] <Critical> (GameThread) signals.cpp:70 SignalHandler... means nothing, honestly

[w1naenator]

KNACK™ 2 Demo - CUSA09687 goes ingame, just doesn't show graphics properly. Menus work as intended. CUSA09687.log

RESIDENT EVIL 3 "Raccoon City Demo" - CUSA17925 also requires V_CMP_U64, but still crashes at launch, now on this error: [Debug] <Critical> (GameThread) signals.cpp:70 SignalHandler: Unhandled Exception code 0xc0000005 at 0x802ad2a5b CUSA17925.log

Good, I'll check!

[Parotaku]

Resident Evil 4 (remake) also required V_CMP_U64 so now it crashes with:
[Debug] (shadPS4:GpuCommandProcessor) resource_tracking_pass.cpp:411 FindSharpSources: Unreachable code!
Bindless sharp access detected pc=0x14b8

shad_log.txt

[Parotaku]

Same error with Street Fighter 6 now:

shad_log.zip

[w1naenator]

Same error with Street Fighter 6 now:

shad_log.zip

Did it work on main branch binary?

[Parotaku]

No, I meant both crash on V_CMP_U64 error on Main (that's the reason I tested them on this PR), so I guess that a progress: now they crash at another type of error... ^^

[w1naenator]

No, I meant both crash on V_CMP_U64 error on Main (that's the reason I tested them on this PR), so I guess that a progress: now they crash at another type of error... ^^

I agree!

[w1naenator]

@Parotaku I do not have RE3 Demo, can you please test with this?
https://github.com/w1naenator/shadPS4/actions/runs/28056599939

[tochnonechelovek]

@Parotaku I do not have RE3 Demo, can you please test with this? https://github.com/w1naenator/shadPS4/actions/runs/28056599939

Not sure if Parotaku has it, but my tests aren't different.
CUSA17925.log

[Parotaku]

@Parotaku I do not have RE3 Demo, can you please test with this? https://github.com/w1naenator/shadPS4/actions/runs/28056599939

Not sure if Parotaku has it, but my tests aren't different. CUSA17925.log

Yeah, sorry... I don't have RE3 demo... but I tested with RE4 remake & Street Fighter 6 and there are no changes... :-/
shad_log.txt
shad_log.zip

[Hog185]

no diff in PT

[Randomuser8219]

KNACK 2 still works with the rebase.

[Parotaku]

With your update from Main, now No Straight Roads goes ingame everytime...

[Parotaku]

I don't know if this is still WIP, but now would be a good time to update it from Main (because of all the fixes for UE & others...). It might help new games go further?

[w1naenator]

I don't know if this is still WIP, but now would be a good time to update it from Main (because of all the fixes for UE & others...). It might help new games go further?

Yes, a lot of games need this.

[tochnonechelovek]

Tested on the latest version of this PR, it's the same as the previous report.

RESIDENT EVIL 3 "Raccoon City Demo":
CUSA17925.log
[Debug] <Critical> (GameThread) signals.cpp:70 SignalHandler: Unhandled Exception code 0xc0000005 at 0x802ad2a5b

KNACK™ 2 Demo:
CUSA09687.log

[w1naenator]

Tested on the latest version of this PR, it's the same as the previous report.

RESIDENT EVIL 3 "Raccoon City Demo": CUSA17925.log [Debug] <Critical> (GameThread) signals.cpp:70 SignalHandler: Unhandled Exception code 0xc0000005 at 0x802ad2a5b

KNACK™ 2 Demo: CUSA09687.log

I will check later.
I’m currently working on memory tracking. It seems to fix the vertex explosions in Ratchet & Clank on Intel 10th-generation CPUs and newer.

[georgemoralis]

rebased

[Gih-pt]

This should also improve Resident Evil Village, Resident Evil Resistance, and STAR WARS Jedi: Fallen Order.

shadps4-compatibility/shadps4-game-compatibility#2597
shadps4-compatibility/shadps4-game-compatibility#314
shadps4-compatibility/shadps4-game-compatibility#1303