WebGPU front_end

[wez]

I just pushed preliminary support for using wgpu.rs to power an rendering front end in wezterm.

You can enable it by adding front_end = "WebGpu" to your configuration.

When using WebGPU, you will have access to a variety of different rendering backends, depending on your system:

• Vulkan
• Metal
• DirectX 12 (or 11)
• OpenGL
• Cpu

The default behavior is to select the best supported low-power (integrated) GPU available to your system.

You can explicitly select a discrete GPU if you wish by setting webgpu_power_preference = "HighPerformance".

You can explicit request a fallback/software based backend (not recommended, as it will be slow!) by setting webgpu_force_fallback_adapter = true.

You can list the available GPUs using the debug overlay; call the wezterm.gui.enumerate_gpus() function:

You can then force the use of one of those if you wish:

return {
   webgpu_preferred_adapter={
      backend="Vulkan",
      device_type="Cpu",
      name="llvmpipe (LLVM 14.0.0, 256 bits)"
   }
}

This is a new backend and likely has some issues that are waiting to be discovered: I'd love to hear about them!

[luisiacc]

using the Dx12 on nvidia makes the font looks weird:

Edit: also happened with internal Intel graphics, just enabling the front_end makes the font looks wider, like in the picture.

with it:

without it:

[wez]

@luisiacc: I'm seeing the same thing here on linux with nvidia/vulkan. It is reminiscent of this earlier issue:

• https://github.com/wez/wezterm/issues/1025

I'm continuing to dig into this.

[Suyashtnt]

It seems to just not want to create the adapter at all.

Config

local wezterm = require "wezterm"

return {
    color_scheme = "Catppuccin Mocha",
    font = wezterm.font_with_fallback {
        'JetBrainsMono Nerd Font Mono', -- Comic code breaks the font
        'Noto Color Emoji',
    },
    font_size = 13,
    line_height = 1.6,
    harfbuzz_features = {
        "cv06=1",
        "cv14=1",
        "cv32=1",
        "ss04=1",
        "ss07=1",
        "ss09=1",
    },
    enable_wayland = false, -- https://github.com/wez/wezterm/issues/1701
    front_end = "WebGpu",
    webgpu_force_fallback_adapater = true,
    cursor_blink_ease_in = "EaseIn",
    cursor_blink_ease_out = "EaseOut",
    animation_fps = 120,
    cursor_blink_rate = 800,
    cursor_thickness = 1.0,
    default_cursor_style = "BlinkingBar",
    initial_rows = 17,
    initial_cols = 70,
    window_padding = {
        left = 12,
        right = 12,
        top = 12,
        bottom = 12,
    },
    window_frame = {
        inactive_titlebar_bg = '#171723',
        active_titlebar_bg = '#171723',
        inactive_titlebar_fg = '#9399b2',
        active_titlebar_fg = '#cdd6f4',
        font = wezterm.font { family = 'Inter', weight = 'Bold' },
    },
    colors = {
        tab_bar = {
            active_tab = {
                bg_color = '#1e1e2e',
                fg_color = '#cdd6f4',
                intensity = 'Bold',
            },
            inactive_tab = {
                bg_color = '#313244',
                fg_color = '#9399b2',
            },
                        inactive_tab_hover = {
                                bg_color = '#11111b',
                                fg_color = '#cdd6f4',
                        },
            new_tab = {
                bg_color = '#181825',
                fg_color = '#cdd6f4',
            },
            new_tab_hover = {
                                bg_color = '#1e1e2e',
                                fg_color = '#cdd6f4',
                        },
        },
    },
}

[wez]

@Suyashtnt start wezterm using wezterm -n then press CTRL-SHIFT-L to open the Debug Overlay, then type wezterm.gui.enumerate_gpus() and press enter; please copy and paste the output from the terminal here.

[Suyashtnt]

> wezterm.gui.enumerate_gpus()
[
    {
        "backend": "Gl",
        "device": 0,
        "device_type": "Other",
        "name": "NVIDIA GeForce GTX 1660 SUPER/PCIe/SSE2",
        "vendor": 4318,
    },
]

[cairijun]

WebGpu (with Vulkan) seems noticeably smoother than OpenGL on my Windows, though no much difference on my another Linux x11 laptop. That's quite impressive!

Some font rendering issues noticed:

1. selected glyphs look faint, especially when freetype_render_target = 'Normal'
2. un-selected glyphs get a little bit thicker on WebGpu than OpenGL, especially when freetype_render_target = 'HorizontalLcd'

Windows 10 on name=Intel(R) UHD Graphics 630, device_type=IntegratedGpu, backend=Vulkan, driver=Intel Corporation, driver_info=Intel driver, vendor=32902, device=16024

[jokeyrhyme]

Okay, I have what might be a weird hardware combination: AMD Ryzan 7950X (integrated GPU) + AMD Radeon RX 6800XT (discrete GPU) + monitor plugged into 6800XT (oops, forgot to say Archlinux btw, with wezterm packaged from this commit: https://github.com/wez/wezterm/commit/49b9839fdb41eff4d2aae3bb91e6a4befb9cd20b )

> wezterm.gui.enumerate_gpus()
[
    {
        "backend": "Vulkan",
        "device": 29631,
        "device_type": "DiscreteGpu",
        "driver": "radv",
        "driver_info": "Mesa 22.2.3",
        "name": "AMD Radeon RX 6800 XT (RADV NAVI21)",
        "vendor": 4098,
    },
    {
        "backend": "Vulkan",
        "device": 5710,
        "device_type": "IntegratedGpu",
        "driver": "radv",
        "driver_info": "Mesa 22.2.3",
        "name": "AMD Radeon Graphics (RADV GFX1036)",
        "vendor": 4098,
    },
    {
        "backend": "Gl",
        "device": 0,
        "device_type": "Other",
        "name": "AMD Radeon RX 6800 XT (navi21, LLVM 14.0.6, DRM 3.48, 6.0.9-arch1-1)",
        "vendor": 4098,
    },
]
>

Tests with `AMD_VULKAN_ICD=RADV wezterm` (and with compositor launched with RADV):

- with just `front_end = "WebGpu"` -> > ERROR wezterm_gui > running message loop: Error while flushing display: Broken pipe (os error 32); terminating - with `front_end = "WebGpu"` and `webgpu_power_preference = "HighPerformance"` - > wezterm window appears with my shell, but is either non-responsive or isn't updating to show my input - with `front_end = "WebGpu"` and `webgpu_force_fallback_adapter = true` -> > ERROR wezterm_gui::frontend > Failed to create window: no compatible adapter found. Available: name=AMD Radeon RX 6800 XT (RADV NAVI21), device_type=DiscreteGpu, backend=Vulkan, driver=radv, driver_info=Mesa 22.2.3, vendor=4098, device=29631, compatible=yes > name=AMD Radeon Graphics (RADV GFX1036), device_type=IntegratedGpu, backend=Vulkan, driver=radv, driver_info=Mesa 22.2.3, vendor=4098, device=5710, compatible=yes > name=AMD Radeon RX 6800 XT (navi21, LLVM 14.0.6, DRM 3.48, 6.0.9-arch1-1), device_type=Other, backend=Gl, vendor=4098, device=0, compatible=yes

Tests with `AMD_VULKAN_ICD=AMDVLK wezterm` (but with compositor launched with RADV):

- with just `front_end = "WebGpu"` -> > ERROR wgpu::backend::direct > Error in Adapter::request_device: not enough memory left > ERROR wezterm_gui::frontend > Failed to create window: Requesting a device failed ( probably https://github.com/GPUOpen-Drivers/AMDVLK/issues/305 ) - with `front_end = "WebGpu"` and `webgpu_power_preference = "HighPerformance"` - > > ERROR wgpu::backend::direct > Error in Adapter::request_device: not enough memory left > ERROR wezterm_gui::frontend > Failed to create window: Requesting a device failed - with `front_end = "WebGpu"` and `webgpu_force_fallback_adapter = true` -> > ERROR wezterm_gui::frontend > Failed to create window: no compatible adapter found. Available: > name=Null hardware (RADV NAVI10), device_type=DiscreteGpu, backend=Vulkan, driver=radv, driver_info=Mesa 22.2.3, vendor=4098, device=29456, compatible=yes > name=AMD Radeon RX 6800 XT (navi21, LLVM 14.0.6, DRM 3.48, 6.0.9-arch1-1), device_type=Other, backend=Gl, vendor=4098, device=0, compatible=yes

Tests with `AMD_VULKAN_ICD=AMDVLK wezterm` (and with compositor launched with AMDVLK):

- with just `front_end = "WebGpu"` -> > ERROR wgpu::backend::direct > Error in Adapter::request_device: not enough memory left > ERROR wezterm_gui::frontend > Failed to create window: Requesting a device failed ( probably https://github.com/GPUOpen-Drivers/AMDVLK/issues/305 ) - with `front_end = "WebGpu"` and `webgpu_power_preference = "HighPerformance"` - > > ERROR wgpu::backend::direct > Error in Adapter::request_device: not enough memory left > ERROR wezterm_gui::frontend > Failed to create window: Requesting a device failed - with `front_end = "WebGpu"` and `webgpu_force_fallback_adapter = true` -> > ERROR wezterm_gui::frontend > Failed to create window: no compatible adapter found. Available: > name=Null hardware (RADV NAVI10), device_type=DiscreteGpu, backend=Vulkan, driver=radv, driver_info=Mesa 22.2.3, vendor=4098, device=29456, compatible=yes > name=AMD Radeon RX 6800 XT (navi21, LLVM 14.0.6, DRM 3.48, 6.0.9-arch1-1), device_type=Other, backend=Gl, vendor=4098, device=0, compatible=yes

It's interesting to me that even when everything is using AMDVLK, that output still says "driver=radv", but maybe this is normal?

[wez]

especially when freetype_render_target = 'HorizontalLcd'

@cairijun: HorizontalLcd doesn't currently have correct logic for subpixel AA in the shader in WebGPU mode. It's a bit of a PITA to make it work correctly with alpha blending, requiring multiple render passes. In the OpenGL front-end we use a feature called dual source blending, but that isn't supported in wgpu. On modern high density displays you don't really need it anyway.

[wez]

@jokeyrhyme re: 'no compatible adapter found' messages, followed by a list that shows that all adapters are compatible. That is shown after trying to match the power preference and force-fallback preference using a wgpu built-in function. I'll do something to refine the error message.

For more flexible selection of the underlying adapter, it might make sense to put something like this into your config file:

local adapter = nil
local front_end = nil

for _, gpu in wezterm.gui.enumerate_gpus() do
   if gpu.backend == "Vulkan" and gpu.device_type == "Integrated" then
     adapter = gpu
     front_end = "WebGpu"
     break
   end
end

return {
   webgpu_preferred_adapter = adapter,
   front_end = front_end
}

[wez]

@Suyashtnt: could you try putting this in your config file:

local gpus = wezterm.gui.enumerate_gpus()
return {
  webgpu_preferred_adapter = gpus[1]
  front_end = "WebGpu",
}

[Suyashtnt]

@Suyashtnt: could you try putting this in your config file:

local gpus = wezterm.gui.enumerate_gpus()
return {
  webgpu_preferred_adapter = gpus[1]
  front_end = "WebGpu",
}

Will try in a few hours (I checked this issue at the right time damn)

[jokeyrhyme]

For more flexible selection of the underlying adapter... -- https://github.com/wez/wezterm/issues/2756#issuecomment-1322215192

Thanks, that's a handy snippet

I was mostly testing webgpu_force_fallback_adapter to see if it made anything better/worse

I use the same wezterm.lua file on multiple machines, so I'll review my decisions about what GPUs are plugged into which monitors and see if I can boil that down into Lua code that works on all of them

[Suyashtnt]

@wez colours are still bork

[Suyashtnt]

local wezterm = require "wezterm"
local gpus = wezterm.gui.enumerate_gpus()

return {
    color_scheme = "Catppuccin Mocha",
    font = wezterm.font_with_fallback {
        'JetBrainsMono Nerd Font Mono', -- Comic code breaks the font
        'Noto Color Emoji',
    },
    font_size = 13,
    line_height = 1.6,
    harfbuzz_features = {
        "cv06=1",
        "cv14=1",
        "cv32=1",
        "ss04=1",
        "ss07=1",
        "ss09=1",
    },
    webgpu_preferred_adapter = gpus[1],
    front_end = "WebGpu",
    enable_wayland = true, -- https://github.com/wez/wezterm/issues/1701
    cursor_blink_ease_in = "EaseIn",
    cursor_blink_ease_out = "EaseOut",
    animation_fps = 120,
    cursor_blink_rate = 800,
    cursor_thickness = 1.0,
    default_cursor_style = "BlinkingBar",
    initial_rows = 17,
    initial_cols = 70,
    window_padding = {
        left = 12,
        right = 12,
        top = 12,
        bottom = 12,
    },
    window_frame = {
        inactive_titlebar_bg = '#171723',
        active_titlebar_bg = '#171723',
        inactive_titlebar_fg = '#9399b2',
        active_titlebar_fg = '#cdd6f4',
        font = wezterm.font { family = 'Inter', weight = 'Bold' },
    },
    colors = {
        tab_bar = {
            active_tab = {
                bg_color = '#1e1e2e',
                fg_color = '#cdd6f4',
                intensity = 'Bold',
            },
            inactive_tab = {
                bg_color = '#313244',
                fg_color = '#9399b2',
            },
                        inactive_tab_hover = {
                                bg_color = '#11111b',
                                fg_color = '#cdd6f4',
                        },
            new_tab = {
                bg_color = '#181825',
                fg_color = '#cdd6f4',
            },
            new_tab_hover = {
                                bg_color = '#1e1e2e',
                                fg_color = '#cdd6f4',
                        },
        },
    },
}

[luisiacc]

don't know if anyone has tried this, but when entering fullscreen on windows 11 with the webgpu activated, I just get an entire black screen.

[colemickens]

I'm getting really terrible performance. Pasting the lua line in took 15 seconds to actualize:

wezterm version: 20221119-145034-49b9839f
OpenGL version: WebGPU: name=RENOIR (renoir, LLVM 14.0.6, DRM 3.48, 6.0.9), device_type=Other, backend=Gl, vendor=4098, device=0
Enter lua statements or expressions and hit Enter.
Press ESC or CTRL-D to exit
> wezterm.gui.enumerate_gpus()
[
    {
        "backend": "Gl",
        "device": 0,
        "device_type": "Other",
        "name": "RENOIR (renoir, LLVM 14.0.6, DRM 3.48, 6.0.9)",
        "vendor": 4098,
    },
]

What can I do to debug or understand what backends are available? For example, my compositor is running in OpenGL mode, but vulkaninfo seems to indicate Vulkan ought to be available.

[luisiacc]

don't know if anyone has tried this, but when entering fullscreen on windows 11 with the webgpu activated, I just get an entire black screen.

@wez this happens still on a lower scale but it's still pretty annoying, whenever I enter fullscreen with webGpu activated, I get a blackscreen entirely for like 2 seconds, once I'm able to see wezterm gui, I can write and all of that, if alt-tab it, or try to change to another program I get the same 2 seconds with blackscreen, and again when trying to alt-tab to wezterm again.

[wez]

@luisiacc please open a separate issue for that so that I can track it more effectively

[wez]

For the most part, this has seemed to have worked quite well. I've just pushed some docs for this, and I'm going to close out this issue. If you have a webgpu issue, please file a separate issue so that it can be tracked effectively.

[github-actions[bot]]

I'm going to lock this issue because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active issues. If you have found a problem that seems similar to this, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.