starred/shadPS4

Fork 0

mirror of https://github.com/shadps4-emu/shadPS4.git synced 2026-04-26 00:05:58 +03:00

[GH-ISSUE #296] [CUSA03173] [01.09] Sliders in Bloodborne are drawn incorrectly #45

New issue

Closed

opened 2026-02-27 21:04:18 +03:00 by kerem · 13 comments

kerem commented

2026-02-27 21:04:18 +03:00

Owner

Originally created by @roamic on GitHub (Jul 16, 2024).
Original GitHub issue: https://github.com/shadps4-emu/shadPS4/issues/296

The bar near 7 should be partly filled.
It seems to me that it is missing some constant buffer updates for some reason.

This shader code is responsible for the positioning:

layout(set = 0, binding = 2, scalar) uniform vs_cbuf_block_f32
{
    float data[16];
} cbuf_104;

gl_Position.x = fma(cbuf_104.data[0u], vs_in_attr1.x, fma(cbuf_104.data[1u], vs_in_attr1.y, fma(cbuf_104.data[2u], 0.0, cbuf_104.data[3u] * 1.0)));
gl_Position.y = fma(cbuf_104.data[4u], vs_in_attr1.x, fma(cbuf_104.data[5u], vs_in_attr1.y, fma(cbuf_104.data[6u], 0.0, cbuf_104.data[7u] * 1.0)));

It looks like a transform matrix multiplication.

This is the buffer content at the time of rendering:

Originally created by @roamic on GitHub (Jul 16, 2024). Original GitHub issue: https://github.com/shadps4-emu/shadPS4/issues/296 ![image](https://github.com/user-attachments/assets/05995cda-6848-4563-bacf-57d75158d268) The bar near 7 should be partly filled. It seems to me that it is missing some constant buffer updates for some reason. This shader code is responsible for the positioning: ``` layout(set = 0, binding = 2, scalar) uniform vs_cbuf_block_f32 { float data[16]; } cbuf_104; gl_Position.x = fma(cbuf_104.data[0u], vs_in_attr1.x, fma(cbuf_104.data[1u], vs_in_attr1.y, fma(cbuf_104.data[2u], 0.0, cbuf_104.data[3u] * 1.0))); gl_Position.y = fma(cbuf_104.data[4u], vs_in_attr1.x, fma(cbuf_104.data[5u], vs_in_attr1.y, fma(cbuf_104.data[6u], 0.0, cbuf_104.data[7u] * 1.0))); ``` It looks like a transform matrix multiplication. This is the buffer content at the time of rendering: ![image](https://github.com/user-attachments/assets/33f6269c-4196-4cfb-8714-902c736aa545)

kerem

2026-02-27 21:04:18 +03:00

closed this issue
added the
bug

Bloodborne
labels

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@raphaelthegreat commented on GitHub (Jul 18, 2024):

Can you check if it uses inline constant buffers by any chance? Resource tracking pass has the pattern for them in gcn assembly.

@raphaelthegreat commented on GitHub (Jul 18, 2024): Can you check if it uses inline constant buffers by any chance? Resource tracking pass has the pattern for them in gcn assembly.

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@roamic commented on GitHub (Jul 18, 2024):

@raphaelthegreat no, doesn't seem so

/*000000000000*/ s_mov_b32       vcc_hi, lit(20)
/*000000000008*/ s_swappc_b64    s[0:1], s[0:1]
/*00000000000c*/ s_buffer_load_dwordx16 s[0:15], s[8:11], 0x0

@roamic commented on GitHub (Jul 18, 2024): @raphaelthegreat no, doesn't seem so ``` /*000000000000*/ s_mov_b32 vcc_hi, lit(20) /*000000000008*/ s_swappc_b64 s[0:1], s[0:1] /*00000000000c*/ s_buffer_load_dwordx16 s[0:15], s[8:11], 0x0 ```

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@raphaelthegreat commented on GitHub (Jul 18, 2024):

Go to core/memory.cpp and comment out the return; statements in the vulkan functions. Then apply this patch github.com/raphaelthegreat/shadPS4@a7c69c1512/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp (L346)

@raphaelthegreat commented on GitHub (Jul 18, 2024): Go to core/memory.cpp and comment out the return; statements in the vulkan functions. Then apply this patch https://github.com/raphaelthegreat/shadPS4/blob/a7c69c1512f26896a4a6301a8c6e6e67effa5036/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp#L346

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@raphaelthegreat commented on GitHub (Jul 18, 2024):

Also the same in ComputePipeline just in case

@raphaelthegreat commented on GitHub (Jul 18, 2024): Also the same in ComputePipeline just in case

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@roamic commented on GitHub (Jul 18, 2024):

With texture_cache.OnCpuWrite(address); it goes OutOfDeviceMemory before even showing the logos. Without it — it does nothing.

@roamic commented on GitHub (Jul 18, 2024): With `texture_cache.OnCpuWrite(address);` it goes `OutOfDeviceMemory` before even showing the logos. Without it — it does nothing.

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@raphaelthegreat commented on GitHub (Jul 18, 2024):

Also add a check that address is not zero as I remember BB passes some invalid buffers

@raphaelthegreat commented on GitHub (Jul 18, 2024): Also add a check that address is not zero as I remember BB passes some invalid buffers

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@roamic commented on GitHub (Jul 18, 2024):

It doesn't go OOM anymore even without the address check, but the original issue persists.

@roamic commented on GitHub (Jul 18, 2024): It doesn't go OOM anymore even without the address check, but the original issue persists.

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@roamic commented on GitHub (Aug 16, 2024):

Just for the history I recap what Turtle has uncovered:
The game uses user registers to supply a parameter to the shader. Those need to be detected and implemented with push constants.

@roamic commented on GitHub (Aug 16, 2024): Just for the history I recap what Turtle has uncovered: The game uses user registers to supply a parameter to the shader. Those need to be detected and implemented with push constants.

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@roamic commented on GitHub (Aug 17, 2024):

These artifacts are caused by the same issue. They use the same vertex shader.
And it's not about user registers, it's about constant buffers being invalid. The supplied CB contains two 2x4 2D transform matrices for vertex position and texcoords. Those matrices always contain values that were set during their initialization rather than current values. So either game cannot update them for some reason, or the emulator misses the update.

@roamic commented on GitHub (Aug 17, 2024): ![image](https://github.com/user-attachments/assets/c8fd7bc3-108f-41ea-8a7c-709e6612f674) These artifacts are caused by the same issue. They use the same vertex shader. And it's not about user registers, it's about constant buffers being invalid. The supplied CB contains two 2x4 2D transform matrices for vertex position and texcoords. Those matrices always contain values that were set during their initialization rather than current values. So either game cannot update them for some reason, or the emulator misses the update.

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@roamic commented on GitHub (Aug 17, 2024):

One other possible cause is vertex buffers. Fragment shader has fragment alpha channel set to vs_attr_0.w that is directly copied from the VS input. And I can see that some values are 0. This even seems more likely than CBs.

@roamic commented on GitHub (Aug 17, 2024): One other possible cause is vertex buffers. Fragment shader has fragment alpha channel set to `vs_attr_0.w` that is directly copied from the VS input. And I can see that some values are `0`. This even seems more likely than CBs.

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@andrew101sanders commented on GitHub (Aug 17, 2024):

Ahh, I wish I had seen this issue before trying to track down the issue in renderdoc too, ha.

Just contributing the discussion that was had in the discord a few weeks back regarding it:

[7:26 PM]IndecisiveTurtle:
v_mac_f32 v2, s8, v8
[7:26 PM]IndecisiveTurtle: s8 doesn't appear to be set anywhere in shader so its a user data reg
[7:26 PM]IndecisiveTurtle: user data regs atm are hardcoded as compilation constants in the shader
[7:26 PM]IndecisiveTurtle: So if the game uses a UD value as a dynamic adjustment for slider it wont work and it will stay stuck
[7:27 PM]IndecisiveTurtle: UD regs need to be made into push constants

@andrew101sanders commented on GitHub (Aug 17, 2024): Ahh, I wish I had seen this issue before trying to track down the issue in renderdoc too, ha. Just contributing the discussion that was had in the discord a few weeks back regarding it: ![image](https://github.com/user-attachments/assets/8bad3175-2880-46de-b16a-814c127795a5) [7:26 PM]IndecisiveTurtle: v_mac_f32 v2, s8, v8 [7:26 PM]IndecisiveTurtle: s8 doesn't appear to be set anywhere in shader so its a user data reg [7:26 PM]IndecisiveTurtle: user data regs atm are hardcoded as compilation constants in the shader [7:26 PM]IndecisiveTurtle: So if the game uses a UD value as a dynamic adjustment for slider it wont work and it will stay stuck [7:27 PM]IndecisiveTurtle: UD regs need to be made into push constants

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@raphaelthegreat commented on GitHub (Aug 17, 2024):

This turned out to be incorrect, I missed the place where s8 was being written in the shader

@raphaelthegreat commented on GitHub (Aug 17, 2024): This turned out to be incorrect, I missed the place where s8 was being written in the shader

kerem commented

2026-02-27 21:04:18 +03:00

Author

Owner

@roamic commented on GitHub (Aug 19, 2024):

fixed with #464

@roamic commented on GitHub (Aug 19, 2024): fixed with #464

kerem referenced this issue

2026-02-27 21:11:45 +03:00

[PR #45] [MERGED] dummy libc printf call #1239