rsx: Enable multisampling #6055

kd-11 · 2019-06-06T13:32:51Z

Adds MSAA support (vulkan only for now since almost all testers will run vulkan)
Completely refactors out D3D12 as it was holding back progress with shared code.
Minor improvement to fp decompiler input modifiers.

Xcedf · 2019-06-06T14:19:12Z

Resistance 2
AA Auto
E {RSX [0x6c0e4b4]} RSX: Resolve shader:
#version 420
#extension GL_ARB_separate_shader_objects : enable
#extension GL_ARB_shader_stencil_export : enable

layout(std140, set=0, binding=0) uniform static_data{ ivec4 regs[8]; };
layout(set=0, binding=1) uniform sampler2DMS fs0;
layout(set=0, binding=2) uniform usampler2DMS fs1;
layout(pixel_center_integer) in vec4 gl_FragCoord;

void main()
{
ivec2 out_coord = ivec2(gl_FragCoord.xy);
ivec2 in_coord = (out_coord / regs[0].xy);
ivec2 sample_loc = out_coord % ivec2(regs[0].xy);
int sample_index = sample_loc.x + (sample_loc.y * regs[0].y);
float frag_depth = texelFetch(fs0, in_coord, sample_index).x;
uint frag_stencil = texelFetch(fs1, in_coord, sample_index).x;
gl_FragDepth = frag_depth;
gl_FragStencilRefARB = int(frag_stencil);
}

F {RSX [0x7075ab4]} RSX: class std::runtime_error thrown: Verification failed (e=0x7f):
(in file c:\projects\rpcs3\rpcs3\emu\rsx\vk\vkrendertargets.h:334)

AA Disabled
just

F {RSX [0x7075ab4]} RSX: class std::runtime_error thrown: Verification failed (e=0x7f):
(in file c:\projects\rpcs3\rpcs3\emu\rsx\vk\vkrendertargets.h:334)

Same for Killzone 2

kd-11 · 2019-06-06T14:21:40Z

Resolve shader message is unrelated. I just left it as log_error in case of compilation errors.

Xcedf · 2019-06-06T14:30:47Z

Well it's just verification fail then, and even if i disable it the game still ends with some error

kd-11 · 2019-06-06T14:44:12Z

Yea, looks like some files on my end are corrupted; I'm reinstalling some demos and re-running more tests. The verification crash at least is reproducible.

arabek · 2019-06-06T14:51:56Z

Killzone 2 fails for me too, even after nuking all the caches.

kd-11 · 2019-06-06T15:05:00Z

Fixed the VkRenderTargets.h crash caused by only testing MSAA x2 during development. MSAA x4 was completely broken which is what killzone 2 is using.

kd-11 · 2019-06-06T15:18:23Z

Several other issues identified by testers and under investigation (avoid repeating these):

~~Performance slowdown in ni no kuni (even compared to early MSAA-enabled builds)~~ Fixed.
~~Graphics corruption in ni no kuni when MSAA is disabled.~~ Fixed.

Xcedf · 2019-06-06T16:01:21Z

Now that's what i always expected
Resistance 2 (Native res)
master

MarioSonic2987 · 2019-06-06T16:08:53Z

Comparisons:

God of War III (E3 2009 Demo) - Minor improvement

Midnight Club: Los Angeles (200% scaling)

Racedriver: Grid (200% scaling)

Red Dead Redemption (200% scaling)

Tekken Tag Tournament 2

Uncharted 2: Among Thieves

Kravickas · 2019-06-06T16:38:37Z

Crysis Approve 👍

stride21 · 2019-06-06T17:48:33Z

Master:

PR:

Master:

PR:

scribam · 2019-06-06T18:41:52Z

rpcs3/Emu/RSX/GL/GLRenderTargets.cpp

@@ -234,17 +218,12 @@ void GLGSRender::init_buffers(rsx::framebuffer_creation_context context, bool sk

 	const u8 color_bpp = get_format_block_size_in_bytes(layout.color_format);
 	const u8 depth_bpp = (layout.depth_format == rsx::surface_depth_format::z16 ? 2 : 4);


I think the variable depth_bpp can be removed unless you prefer to replace:

m_depth_surface_info.bpp = (layout.depth_format == rsx::surface_depth_format::z16? 2 : 4);

by

m_depth_surface_info.bpp = depth_bpp;

MsDarkLow · 2019-06-06T21:05:51Z

MSAA seems to work as intended in 4. However, I noticed a difference in colors in Yakuza 3, 4 and Ishin, they seem to show up incorrectly.

Yakuza 3

(Doesn't seem to support MSAA)

Master

PR

Yakuza 4

Master

PR MSAA

PR

Yakuza Ishin

(Doesn't seem to support MSAA???)

Master

PR MSAA

PR

e1489 · 2019-06-07T00:42:30Z

;=; how do I try this?

Artins90 · 2019-06-07T01:56:09Z

;=; how do I try this?

Click show all checks below.
Select details near appveryor
Open artifacts and dowload the 17MB file
The antialising option is in the GPU options section, select auto to activate it.

dbz400 · 2019-06-07T12:13:10Z

D3D12 renderer selection box will be removed as well ?

AniLeo · 2019-06-07T12:17:39Z

d3d12 is not yet fully removed

kd-11 · 2019-06-07T14:52:14Z

@MsDarkLow

However, I noticed a difference in colors in Yakuza 3, 4 and Ishin, they seem to show up incorrectly.

Maybe show the differences clearly? I cannot spot any differences since I'm not familiar with those games. There are some tweaks to the fragment decompiler based on hardware testing on a PS3, but I still have doubts as to the accuracy; if the commit is wrong compared to PS3 I can revert it from the PR and submit it separately at a later date.

anseesaw · 2019-06-07T19:13:48Z

What does this mean for d3d12，Improve or delete？

RainbowCookie32 · 2019-06-07T19:15:28Z

DX12 isn't being improved. I'd say that we are one step closer to removal with this PR (I think)

MarioSonic2987 · 2019-06-08T06:26:17Z

TTT2 still crashes in Fallen Garden with MSAA enabled:
Log TTT2 Broken 2

Killzone 2 has flickering shadows and blurred graphics.

Master:

This PR:

MsDarkLow · 2019-06-08T10:18:47Z

@kd-11

Maybe show the differences clearly? I cannot spot any differences since I'm not familiar with those games. There are some tweaks to the fragment decompiler based on hardware testing on a PS3, but I still have doubts as to the accuracy; if the commit is wrong compared to PS3 I can revert it from the PR and submit it separately at a later date.

I got a better scenario with Ishin which is way more clear. Took screenshots in the main menu including how it looks on a PS3.

Yakuza Ishin

PS3

Master

PR (both OpenGL & Vulkan look like this)

- Remove deprecated functionality - Refactor to share code between common routines

- Requires native_pitch value to take samples into account

- vk: Enable depth buffer resolve+unresolve - vk: Add AMD stenciling extension support - rsx: Temporarily disables MSAA-compatible hacks such as transparency AA - TODO: Add paths to optionally disable MSAA

…ies in the cache

- Resolve image first before performing any transfer operations

- TODO: Option to completely skip clamping in some architectures as it is not needed in most games - Mostly affects older GPUs that do not have access to native fp16

- Refactor overlays and resolve passes to support use of push constants instead of relying buffer map/unmap - Add support for nvidia resolve (NV is the only vendor not supporting shader_stencil_export)

- Arguments to the transform function are xxyy not xyxy

…henever the parent is written to - Fixes successive reads of an antialiased surface that is still bound between reads

…d by spec

- Missing ref increment when using framebuffer could lead to use-after-free. How master was not crashing is surprising

kd-11 · 2019-06-14T12:59:18Z

TTT2 crash is now fixed.

AizerMortenort · 2019-06-14T13:08:12Z

Hello anyone able to fix Dynasty Warriors Gundam Reborn crash ?

AizerMortenort · 2019-06-14T15:30:59Z

Whats so wrong into bringing the crash from Dynasty Warriors Gundam Reborn to a developer's attention? Game crashes for years with same error so i dont see any update fixing it. Thanks

RainbowCookie32 · 2019-06-14T15:31:38Z

Because it's completely unrelated to this PR

AniLeo · 2019-06-14T15:32:16Z

You're wasting a developer's time with a false regression report, it's completely unrelated to this pull request

AizerMortenort · 2019-06-14T15:32:17Z

Still this is not a reason for hate.

kd-11 force-pushed the msaa branch from 5b887a8 to 63755ec Compare June 6, 2019 13:36

kd-11 added Enhancement RSX labels Jun 6, 2019

kd-11 force-pushed the msaa branch from 63755ec to 7f56cc8 Compare June 6, 2019 15:04

kd-11 force-pushed the msaa branch from 7f56cc8 to 733480b Compare June 6, 2019 15:27

scribam reviewed Jun 6, 2019

View reviewed changes

kd-11 force-pushed the msaa branch from 733480b to f4a42c4 Compare June 7, 2019 16:35

kd-11 force-pushed the msaa branch 2 times, most recently from 13176cc to 12a0809 Compare June 8, 2019 09:27

kd-11 added 21 commits June 14, 2019 15:56

d3d12: Remove all shared code with other backends

0355616

rsx: Refactoring and cleanup after d3d12 separation

6a299c2

- Remove deprecated functionality - Refactor to share code between common routines

rsx: Fix AA coordinate transforms

5c47122

- Requires native_pitch value to take samples into account

rsx: Enable MSAA

6ecd27e

- vk: Enable depth buffer resolve+unresolve - vk: Add AMD stenciling extension support - rsx: Temporarily disables MSAA-compatible hacks such as transparency AA - TODO: Add paths to optionally disable MSAA

rsx: Fixup for resolution scaling support

6accc90

rsx: Re-enable optional sample-to-pixel transformation

3eae036

rsx: Add option to toggle MSAA

82ec474

rsx: Correctly count number of valid entries if there are broken entr…

27518da

…ies in the cache

vk: Clean up WCB readbacks when resource is multisampled

d04255d

- Resolve image first before performing any transfer operations

rsx/fp: Ease pressure on fragment shaders when emulating clamp16

60a4df0

- TODO: Option to completely skip clamping in some architectures as it is not needed in most games - Mostly affects older GPUs that do not have access to native fp16

vk: Improve overlay passes for resolve/unreolve

1db5b22

- Refactor overlays and resolve passes to support use of push constants instead of relying buffer map/unmap - Add support for nvidia resolve (NV is the only vendor not supporting shader_stencil_export)

rsx: Attempt to reduce stencil load overhead for nvidia cards

9929486

rsx: Fix typo

fe515cc

- Arguments to the transform function are xxyy not xyxy

vk: Add some missing barriers

d0b0870

vk: Fix WCB for antialiased memory

48b5773

rsx: Force invalidate of children by issuing a resolve notification w…

105b0ad

…henever the parent is written to - Fixes successive reads of an antialiased surface that is still bound between reads

rsx: Typo fix

2e0deb1

vk: Simplify WCB heuristics and fix out-of-bounds access

e9db748

vk: Avoid submitting wrong sample count in overlay passes

1a1698a

vk: Do not use pixel_center_origin as its use is explicitly restricte…

dd4eb26

…d by spec

vk: Fixup for missing resource reference

3a72472

- Missing ref increment when using framebuffer could lead to use-after-free. How master was not crashing is surprising

kd-11 force-pushed the msaa branch from 65de963 to 3a72472 Compare June 14, 2019 12:57

kd-11 merged commit e515d9b into RPCS3:master Jun 14, 2019

RPCS3 locked as off-topic and limited conversation to collaborators Jun 14, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

rsx: Enable multisampling #6055

rsx: Enable multisampling #6055

kd-11 commented Jun 6, 2019

Xcedf commented Jun 6, 2019

kd-11 commented Jun 6, 2019

Xcedf commented Jun 6, 2019

kd-11 commented Jun 6, 2019

arabek commented Jun 6, 2019

kd-11 commented Jun 6, 2019

kd-11 commented Jun 6, 2019 •

edited

Loading

Xcedf commented Jun 6, 2019

MarioSonic2987 commented Jun 6, 2019 •

edited

Loading

Kravickas commented Jun 6, 2019 •

edited

Loading

stride21 commented Jun 6, 2019

scribam Jun 6, 2019 •

edited

Loading

MsDarkLow commented Jun 6, 2019

e1489 commented Jun 7, 2019

Artins90 commented Jun 7, 2019

dbz400 commented Jun 7, 2019

AniLeo commented Jun 7, 2019

kd-11 commented Jun 7, 2019

anseesaw commented Jun 7, 2019

RainbowCookie32 commented Jun 7, 2019 •

edited

Loading

MarioSonic2987 commented Jun 8, 2019 •

edited

Loading

MsDarkLow commented Jun 8, 2019

kd-11 commented Jun 14, 2019

AizerMortenort commented Jun 14, 2019

AizerMortenort commented Jun 14, 2019

RainbowCookie32 commented Jun 14, 2019

AniLeo commented Jun 14, 2019

AizerMortenort commented Jun 14, 2019

		@@ -234,17 +218,12 @@ void GLGSRender::init_buffers(rsx::framebuffer_creation_context context, bool sk

		const u8 color_bpp = get_format_block_size_in_bytes(layout.color_format);
		const u8 depth_bpp = (layout.depth_format == rsx::surface_depth_format::z16 ? 2 : 4);

rsx: Enable multisampling #6055

rsx: Enable multisampling #6055

Conversation

kd-11 commented Jun 6, 2019

Xcedf commented Jun 6, 2019

kd-11 commented Jun 6, 2019

Xcedf commented Jun 6, 2019

kd-11 commented Jun 6, 2019

arabek commented Jun 6, 2019

kd-11 commented Jun 6, 2019

kd-11 commented Jun 6, 2019 • edited Loading

Xcedf commented Jun 6, 2019

MarioSonic2987 commented Jun 6, 2019 • edited Loading

Kravickas commented Jun 6, 2019 • edited Loading

stride21 commented Jun 6, 2019

scribam Jun 6, 2019 • edited Loading

Choose a reason for hiding this comment

MsDarkLow commented Jun 6, 2019

e1489 commented Jun 7, 2019

Artins90 commented Jun 7, 2019

dbz400 commented Jun 7, 2019

AniLeo commented Jun 7, 2019

kd-11 commented Jun 7, 2019

anseesaw commented Jun 7, 2019

RainbowCookie32 commented Jun 7, 2019 • edited Loading

MarioSonic2987 commented Jun 8, 2019 • edited Loading

MsDarkLow commented Jun 8, 2019

kd-11 commented Jun 14, 2019

AizerMortenort commented Jun 14, 2019

AizerMortenort commented Jun 14, 2019

RainbowCookie32 commented Jun 14, 2019

AniLeo commented Jun 14, 2019

AizerMortenort commented Jun 14, 2019

kd-11 commented Jun 6, 2019 •

edited

Loading

MarioSonic2987 commented Jun 6, 2019 •

edited

Loading

Kravickas commented Jun 6, 2019 •

edited

Loading

scribam Jun 6, 2019 •

edited

Loading

RainbowCookie32 commented Jun 7, 2019 •

edited

Loading

MarioSonic2987 commented Jun 8, 2019 •

edited

Loading