rsx: Shader decompiler improvements by kd-11 · Pull Request #5860 · RPCS3/rpcs3

kd-11 · 2019-04-18T13:55:27Z

Fix fp16 emulation. Simple clamping is not enough as infinity and nan need preserving and nvidia cards clamp NaN to the lower bounds which is wrong.
Use native fp16 types if the hardware or driver supports it. In OpenGL most modern hardware will report this feature but for vulkan only Volta, Turing and Vega report this functionality. Should help lower the overhead of the fp16 emulation.
Fix shadow compare operations on vulkan. In OpenGL, GL_ARB_shadow specifies that compare_R_to_texture clamps R to (0-1) before comparing. Vulkan behaves more like the older GL_SGIX_shadow extension and R is not necessarily clamped. This is particularly true when the depth buffer is in floating point format such as D32F_S8. Fixes broken shadows when using AMD cards or NV cards when the 'High precision Z' option is enabled.
Fix poor attribute interpolation precision on Nvidia which causes spurious failure of floating point comparisons in fragment shaders and leads to visual noise.
Properly enforces IEEE-754 rules for all opcodes with the exception of DIVSQ which is provably broken on realhw as well and is now done per-channel to match realhw. LG2 and RSQ seem to ignore the sign bit of the input arguments but otherwise behave in an IEEE compliant way.

- Emulating f16 with f32 is not ideal and requires a lot of value clamping - Using native data type can significantly improve performance and accuracy - With openGL, check for the compatible extensions NV_gpu_shader5 and AMD_gpu_shader_half_float - With Vulkan, enable this functionality in the deviceFeatures if applicable. (VK_KHR_shader_float16_int8 extension) - Temporarily disable hw fp16 for vulkan

- Improve support for float16_t by minimizing mixed inputs to functions (ambiguous overloads) - Minimize amount of downcasts in code by using opcode flags - Re-enable float16_t support for vulkan

…rmats are used - The fixed-point D24S8 format does special Z clamping during compare which matches PS3 behaviour - D32S8 is a floating point format and comparison with Dref > 1 always fails causing black edges/borders

- The hw generates inaccurate values when doing perspective-correct interpolation of vertex output attributes and makes the comparison (a == b) fail even when they are a fixed constant value. - Increase equality tolerance when doing comparisons in fragment shaders for NV cards only to work around this issue. - Teepo fix

- While mul(0, nan) = nan and 0 / 0 = nan, 0 / sqrt(0) = 0 because of hw gremlins. normalize(0) is also nan so this behaviour does not work around that particular case either which makes it even more baffling.

MarioSonic2987 · 2019-04-18T15:23:52Z

One Piece: Pirate Warriors has regressed. Now shows black textures in Vulkan and throw a shader compilation error with OpenGL.

Master

This PR

PR OpenGL:

E {RSX [0x02192f0]} RSX: Failed to compile shader: 0(322) : error C7011: implicit cast from "int" to "float16_t"

F {RSX [0x02192f0]} RSX: class gl::glsl::link_exception thrown: linkage failed: 'Fragment info
-------------
0(322) : error C7011: implicit cast from "int" to "float16_t"
(0) : error C2003: incompatible options for link
'

PR Vulkan:

Log OPPW (OpenGL PR)

Shaderlog OPPW (PR)

In Grand Theft Auto IV, there's a weird circle around lights.

GTA IV

In Grand Theft Auto V, there's a weird shading on characters with Vulkan (with OpenGL it's the same as Vulkan in master).

GTA V

arcadee1977 · 2019-04-18T15:25:09Z

Thanks for your great work! All of Nvidia card's graphic issue have been fixed.

incognitoh · 2019-04-18T15:35:27Z

RR7 basically looks perfect on NVIDIA now. Nice

stride21 · 2019-04-18T16:04:43Z

This PR causes black squares to appear in Battlefield: Bad Company.

PR: OpenGL:

PR Vulkan:

Master OpenGL:

Master Vulkan:

RPCS3.log.gz

Motorstorm when using Vulkan also regressed in this PR. The black on the cars, trees and other objects seem to only happen when they're in the shade.
PR Vulkan:

Master Vulkan:

RPCS3.log.gz

jenci8888 · 2019-04-18T23:07:49Z

Progression and regression other games.

-GT5/6 no more flickering on car anymore, even still present rainbow, but more or less.
-GT5 tries to load night tracks when gets Vulkan (out of memory) error.
-More unstable ingame GT6, It can hangs ingame/intro with Vulkan error. (the error same as master)
-NFS Most Wanted has regression black car (not including traffic vehicles)
-NFS Rivals tries to load main menu but I get this:
E {RSX [0x05648f0]} RSX: Unsupported depth conversion (0x9A) then,
F {RSX [0x055a8ec]} RSX: class std::runtime_error thrown: Assertion Failed! Vulkan API call failed with unrecoverable error: Device lost (Driver crashed with unspecified error or stopped responding and recovered) (VK_ERROR_DEVICE_LOST) (in file c:\projects\rpcs3\rpcs3\emu\rsx\vk\vkhelpers.h:964) (same as GT5)

Screenshots:

Log:
RPCS3-GT5.gz (crash when loads night tracks including out of memory error as well)
RPCS3.log.gz (NFS Rivals)

RagdollClash · 2019-04-18T23:08:43Z

Regression with Sonic's model in Sonic Unleashed

Image

RPCS3.log.gz

RainbowCookie32 · 2019-04-19T01:35:39Z

Beyond Two Souls is back to normal in this PR (seems like it broke at some point in master)

kd-11 · 2019-04-19T06:51:39Z

Need some kind of trace to debug any of these. RRC or RDC will do.

RagdollClash · 2019-04-19T07:23:45Z

If you think you could trace the issue with Unleashed with the RSX capture looking like this, then I can give you one.

Details

I believe this is an RSX capture only issue. Either that or something went wrong on my end when loading the capture.

MarioSonic2987 · 2019-04-19T07:58:29Z

RSX Capture (OPPW VK Master)

RSX Capture (OPPW VK PR)

stride21 · 2019-04-19T15:43:20Z

MotorStorm Vulkan RSX Captures

Battlefield Bad Company OpenGL RSX Captures

Battlefield Bad Company Vulkan RSX Captures

kd-11 · 2019-04-20T08:27:56Z

@slashiee Don't worry about RSX capture not looking correct in replay, it usually still draws stuff properly but breaks post-processing which often relies on Cell-side intervention to look right.

- Properly test for NaN and Inf when clamping down to fp16 - Optimize divsq a bit; mix(vec, vec, bvec) emits OpSelect which is what we want here, instead of component-wise selection which is much slower.

kd-11 · 2019-04-20T10:05:49Z

Retest, improved clamp16 and divsq a bit.
As for OpenGL refusing to compile some fp16-native code, it seems to be a driver bug when the driver is looking for a function override; investigating possible workarounds.

MarioSonic2987 · 2019-04-20T10:26:57Z

One Piece: Pirate Warriors and GTA V regressions are now fixed.

But in GTA IV it's the same:

…d implicit conversions

kd-11 · 2019-04-20T10:34:39Z

Check the OpenGL crash; should be fixed now. For GTA4 I'll need comparison RDC to debug further. Vulkan vs master should be enough.
The differences between OpenGL and Vulkan in some cases is because OpenGL supports float16 'natively'. In reality most vulkan drivers also do support float16 but they do not advertise this feature properly. The differences appear when the native float16 implementation is more accurate than the one we emulate with the shader. A config option "disable native float16 support" exists in config.yml to force use of the shader emulation path and should also 'break' OpenGL when enabled.

Xcedf · 2019-04-20T10:34:56Z

Sonic Unleashed regression fixed

MarioSonic2987 · 2019-04-20T10:51:08Z

RDC Vulkan GTA IV Master
RDC Vulkan GTA IV PR

OPPW crash with OpenGL is now fixed, by the way.

kd-11 · 2019-04-23T15:05:07Z

Added a workaround for the problematic instruction. It no longer behaves like real hardware but we cannot match exactly the PS3's TMU so I think going out of spec there is fine. Basically the texture filtering does not seem to match the PS3 exactly pixel-for-pixel which is expected. Since texturing is completely fixed function (non-programmable) a workaround is the only way to get around this. I can make the hotspots dance around the screen by altering the filtering even very slightly causes the hotspots to migrate. More research on this topic will be carried out at a later date, but likely only for academic curiosity rather than improving any games (NaNs are rather useless as they are often substituted with 0 anyway).

stride21 · 2019-04-23T16:21:03Z

Glowy things are fixed with this PR but now shooting with a weapon causes a fatal error.

Master:

PR:

PR error:

RPCS3 log.zip

stride21 · 2019-04-23T20:10:21Z

Battlefield Bad Company fatal error is now fixed.

Kravickas · 2019-04-23T21:00:57Z

Since u are working on shaders, I leave this here if u wanted to look in to that too even tho its regression from another time.
RPCS3.log.gz
Far Cry 2 when loading ingame and generating shaders

·E 0:04:32.978121 {RSX [0x008ad10]} RSX: ERROR: 0:330: '' :  syntax error, unexpected EQUAL
ERROR: 1 compilation errors.  No code generated.


·E 0:04:32.978128 {RSX [0x008ad10]} RSX: 
·F 0:04:32.978395 {RSX [0x008ad10]} RSX: class std::runtime_error thrown: Failed to compile vertex shader
(in file c:\projects\rpcs3\rpcs3\emu\rsx\vk\vkhelpers.h:2982)

Let it be if its completly different issue or hard to fix.

kd-11 · 2019-04-23T22:22:22Z

If you don't have the shaderlog then you have to submit an opengl log. Vulkan backend does not write any shader info in the log file.

Kravickas · 2019-04-23T22:51:19Z

OGL log RPCS3.log.gz

kd-11 · 2019-04-24T08:59:43Z

Enable 'Log shader programs' in debug settings and then after the crash (disable async shader compiler beforehand) attach shaderlog/vp_analyser.bin file.

TODO: Investigate the _s input modifier behaviour further, in case it can avoid generating zeroes from a MAD instruction. x = MAD(+ve, -ve, -ve) with _s input modifier in BFBC expects result to be Non-zero

Kravickas · 2019-04-24T12:35:32Z

OGL, strict enabled, async shaders disabled
vp_analyser.zip
i hope thats it

vulkan, strict enabled, async shaders disabled
vp_analyser.zip

- When moving to CC, the operation has VEC flag disabled and also temp regs disabled. Looks to be the catch-all ELSE in the selection logic.

kd-11 · 2019-04-24T15:15:47Z

Far Cry 2 crash should be fixed now.

EwoutH · 2019-04-24T15:20:35Z

Has someone benchmarked this PR on a Nvidia Turing (GTX 16 or RTX 20 series) or AMD Vega (Radeon Vega 56/64 or VII) yet? Those GPUs offer double the FP16 performance, would be interesting to see the gains.

kd-11 · 2019-04-24T15:25:15Z

The version of glslang we use to emit SPIR-V may not have the optimizations built in to properly support fp16 as it is a relatively recent thing in the vulkan core spec, although the older AMD extension VK_AMD_gpu_shader_half_float seems to work correctly (there should be no difference to the emitter in theory, but you never know). A Vulkan tools update is coming soon to bring us to newer SDK versions, but it could be a few weeks before that is done.

Kravickas · 2019-04-24T16:06:25Z

Far Cry 2 fixed, n1 thanks

Xcedf · 2019-04-24T17:29:19Z

Confirm Far Cry 2 fixed

kd-11 added 7 commits April 18, 2019 16:41

rsx: Implement strict clamp16 operation needed for NVIDIA cards

d00a15d

rsx: Shader decompiler cleanup and improvements

0bf2e9c

- Improve support for float16_t by minimizing mixed inputs to functions (ambiguous overloads) - Minimize amount of downcasts in code by using opcode flags - Re-enable float16_t support for vulkan

vk: Skip feature check if extension is not supported

e28d5b8

rsx: Proper workaround for broken DIVSQ instruction on realhw

b2e6439

- While mul(0, nan) = nan and 0 / 0 = nan, 0 / sqrt(0) = 0 because of hw gremlins. normalize(0) is also nan so this behaviour does not work around that particular case either which makes it even more baffling.

kd-11 mentioned this pull request Apr 18, 2019

Deadstorm Pirates: Enemies have transparency issues #5567

Closed

legend800 mentioned this pull request Apr 19, 2019

Ridge Racer 7[BCES00009] graphic issue #5731

Closed

Asinin3 mentioned this pull request Apr 19, 2019

Regression #5785 Ni no Kuni: Wrath of the White Witch [BLUS30947] #5848

Closed

rsx: Fragment decompiler fixups

c2c430b

- Properly test for NaN and Inf when clamping down to fp16 - Optimize divsq a bit; mix(vec, vec, bvec) emits OpSelect which is what we want here, instead of component-wise selection which is much slower.

rsx: Simplify use of some mixed input functions using OPFLAGS to avoi…

25c20d9

…d implicit conversions

kd-11 force-pushed the master branch from c856aa9 to 34496f6 Compare April 20, 2019 12:05

kd-11 force-pushed the master branch 2 times, most recently from 51007d7 to e939e3a Compare April 23, 2019 19:19

rsx: Fragment shader decompiler cleanup

e28ebd0

TODO: Investigate the _s input modifier behaviour further, in case it can avoid generating zeroes from a MAD instruction. x = MAD(+ve, -ve, -ve) with _s input modifier in BFBC expects result to be Non-zero

kd-11 force-pushed the master branch from e939e3a to e28ebd0 Compare April 24, 2019 12:23

rsx: Fix VP writes to CC with a MOV instruction

9a6a356

- When moving to CC, the operation has VEC flag disabled and also temp regs disabled. Looks to be the catch-all ELSE in the selection logic.

Kravickas mentioned this pull request Apr 24, 2019

Regressions after #5436 #5460

Closed

Merge branch 'master' into master

159e059

kd-11 merged commit 243df38 into RPCS3:master Apr 25, 2019

Whatcookie mentioned this pull request May 5, 2019

vk: Broken shadow on Ninja Gaiden Sigma #4913

Closed

This was referenced Jun 1, 2019

GL/Vulkan: Worms Ultimate Mayhem #1808

Closed

Vulkan: Worms 2: Armageddon/Worms Crazy Golf #2192

Closed

aqxa1 mentioned this pull request Jun 30, 2019

Persona 5 (Vulkan): native fp16 has graphical glitches on Vega (RADV and AMDVLK) #6150

Closed

connorwalks mentioned this pull request Feb 26, 2020

[Regression] MLB the Show 10-16 broken textures with Vulkan after PR #5860, might only affect Turing cards #7606

Closed

Uh oh!

Uh oh!

Conversation

kd-11 commented Apr 18, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

MarioSonic2987 commented Apr 18, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

arcadee1977 commented Apr 18, 2019

Uh oh!

incognitoh commented Apr 18, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stride21 commented Apr 18, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jenci8888 commented Apr 18, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

RagdollClash commented Apr 18, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

RainbowCookie32 commented Apr 19, 2019

Uh oh!

kd-11 commented Apr 19, 2019

Uh oh!

RagdollClash commented Apr 19, 2019

Uh oh!

MarioSonic2987 commented Apr 19, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stride21 commented Apr 19, 2019

Uh oh!

kd-11 commented Apr 20, 2019

Uh oh!

kd-11 commented Apr 20, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

MarioSonic2987 commented Apr 20, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kd-11 commented Apr 20, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Xcedf commented Apr 20, 2019

Uh oh!

MarioSonic2987 commented Apr 20, 2019

Uh oh!

kd-11 commented Apr 23, 2019

Uh oh!

stride21 commented Apr 23, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stride21 commented Apr 23, 2019

Uh oh!

Kravickas commented Apr 23, 2019

Uh oh!

kd-11 commented Apr 23, 2019

Uh oh!

Kravickas commented Apr 23, 2019

Uh oh!

kd-11 commented Apr 24, 2019

Uh oh!

Kravickas commented Apr 24, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kd-11 commented Apr 24, 2019

Uh oh!

EwoutH commented Apr 24, 2019

Uh oh!

kd-11 commented Apr 24, 2019

Uh oh!

Kravickas commented Apr 24, 2019

Uh oh!

Xcedf commented Apr 24, 2019

Uh oh!

Reviewers

Assignees

Labels

kd-11 commented Apr 18, 2019 •

edited

Loading

MarioSonic2987 commented Apr 18, 2019 •

edited

Loading

incognitoh commented Apr 18, 2019 •

edited

Loading

stride21 commented Apr 18, 2019 •

edited

Loading

jenci8888 commented Apr 18, 2019 •

edited

Loading

RagdollClash commented Apr 18, 2019 •

edited

Loading

MarioSonic2987 commented Apr 19, 2019 •

edited

Loading

kd-11 commented Apr 20, 2019 •

edited

Loading

MarioSonic2987 commented Apr 20, 2019 •

edited

Loading

kd-11 commented Apr 20, 2019 •

edited

Loading

stride21 commented Apr 23, 2019 •

edited

Loading

Kravickas commented Apr 24, 2019 •

edited

Loading