Adding noise on Nvidia only darkens the window background and creates dotted diagonal lines

[MathisP75]

Hyprland Version

Hyprland, built from branch main at commit 751d285 dirty (props: bump ver to 0.32.0).

Bug or Regression?

Bug

Description

noise = 0.0

noise = 1.0

How to reproduce

On an Nvidia GPU, set noise to a value higher than 0 and see the darkness increase as you increase the value. No noise is being generated. The dotted lines also become more visible.

Crash reports, logs, images, videos

No response

[MathisP75]

My card is a 1660ti. Applies on nvidia-dkmsfor both 535 and 545.

[MathisP75]

Another user reproduced the issue on a 3090

This is his result for a pacman -Qs nvidia:

local/cuda 12.3.0-1
    NVIDIA's GPU programming toolkit
local/egl-wayland 2:1.1.13-1
    EGLStream-based Wayland external platform
local/ffnvcodec-headers 12.1.14.0-1
    FFmpeg version of headers required to interface with Nvidias codec APIs
local/lib32-nvidia-utils 535.113.01-1
    NVIDIA drivers utilities (32-bit)
local/libva-nvidia-driver-git 0.0.11.r1.gea6d862-1
    VA-API implementation that uses NVDEC as a backend (git version)
local/libvdpau 1.5-2
    Nvidia VDPAU library
local/libxnvctrl 545.29.02-1
    NVIDIA NV-CONTROL X extension
local/nvidia-open-dkms 535.113.01-8
    NVIDIA open kernel modules
local/nvidia-utils 535.113.01-2
    NVIDIA drivers utilities
local/opencl-nvidia 535.104.05-1
    OpenCL implemention for NVIDIA

[MathisP75]

This is mine:

local/egl-wayland 2:1.1.13-1
    EGLStream-based Wayland external platform
local/ffnvcodec-headers 12.1.14.0-1
    FFmpeg version of headers required to interface with Nvidias codec APIs
local/hyprland-nvidia-git 0.32.0.r5.g8dd02eb5-1
    A dynamic tiling Wayland compositor based on wlroots that doesn't sacrifice on its looks. (NVIDIA patch)
local/lib32-libvdpau 1.5-2
    Nvidia VDPAU library
local/lib32-nvidia-utils 535.113.01-1
    NVIDIA drivers utilities (32-bit)
local/libva-nvidia-driver-git 0.0.11.r1.gea6d862-1
    VA-API implementation that uses NVDEC as a backend (git version)
local/libvdpau 1.5-2
    Nvidia VDPAU library
local/libxnvctrl 535.113.01-1
    NVIDIA NV-CONTROL X extension
local/nvidia-dkms 535.113.01-2
    NVIDIA drivers - module sources
local/nvidia-settings 535.113.01-1
    Tool for configuring the NVIDIA graphics driver
local/nvidia-utils 535.113.01-2
    NVIDIA drivers utilities
local/nvtop 3.0.2-1
    GPUs process monitoring for AMD, Intel and NVIDIA

[MathisP75]

The issue also occurs on a default Hyprland config on a separate profile.

[MathisP75]

Other users encounter the same issue. This is with an Nvidia GPU, where there is no noise but the dotted lines are visible:

This is the same user running on an iGPU, where the noise works:

The lines might be there too, but they could be harder to see if it's the case.

[MathisP75]

Another user has the noise working on GPU: NVIDIA GeForce GTX 1060 6GB, GPU Driver: NVIDIA 545.29.02, but the lines are present (3 passes, 0.5 noise and size 8):

Tweaked colors for better visibility:

[vaxerski]

with all due respect, I don't give a shit. Why should I be the one responsible for nvidia's shader compiler having a tantrum?

Feel free to experiment with the noise generator for yourself:

Hyprland/src/render/shaders/Textures.hpp

Lines 386 to 413 in b8a615f

	inline const std::string FRAGBLURFINISH = R"#(
	precision mediump float;
	varying vec2 v_texcoord; // is in 0-1
	uniform sampler2D tex;
	uniform float noise;
	uniform float brightness;
	float hash(vec2 p) {
	return fract(sin(dot(p, vec2(12.9898, 78.233))) * 43758.5453);
	}
	void main() {
	vec4 pixColor = texture2D(tex, v_texcoord);
	// noise
	float noiseHash = hash(v_texcoord);
	float noiseAmount = (mod(noiseHash, 1.0) - 0.5);
	pixColor.rgb += noiseAmount * noise;
	// brightness
	if (brightness < 1.0) {
	pixColor.rgb *= brightness;
	}
	gl_FragColor = pixColor;
	}
	)#";

Closing. If you find a fix, feel free to MR, but I won't be the one hunting for it.

[Epikastema]

Have a fix working for the lines on my machine (RTX 4090). Would appreciate a test on different hardware to confirm. Having some weird issues pushing to my fork atm.

0001.patch

[vaxerski]

feel free to make a MR, but will need some further testing for sure.