macOS: "acquire_image failed, re-creating"

[parasyte]

Running any example on macOS with RUST_LOG=info logs the following warnings, followed by a panic.

[2019-08-31T07:23:56Z INFO  shadow::framework] Initializing the window...
[2019-08-31T07:23:56Z INFO  wgpu_native::instance] Adapter Metal AdapterInfo { name: "Intel(R) HD Graphics 630", vendor: 0, device: 0, device_type: IntegratedGpu }
[2019-08-31T07:23:56Z INFO  wgpu_native::device] creating swap chain SwapChainDescriptor { usage: OUTPUT_ATTACHMENT, format: Bgra8UnormSrgb, width: 1600, height: 1200, present_mode: Vsync }
[2019-08-31T07:23:56Z INFO  gfx_backend_metal::window] build_swapchain SwapchainConfig { present_mode: Fifo, composite_alpha: OPAQUE, format: Bgra8Srgb, extent: Extent2D { width: 1600, height: 1200 }, image_count: 2, image_layers: 1, image_usage: COLOR_ATTACHMENT }
[2019-08-31T07:23:56Z INFO  shadow::framework] Initializing the example...
[2019-08-31T07:23:57Z INFO  gfx_backend_metal::device] Entry point EntryPoint { name: "main", execution_model: Vertex, work_group_size: WorkGroupSize { x: 0, y: 0, z: 0 } }
[2019-08-31T07:23:57Z INFO  gfx_backend_metal::device] Entry point EntryPoint { name: "main", execution_model: Fragment, work_group_size: WorkGroupSize { x: 0, y: 0, z: 0 } }
[2019-08-31T07:23:57Z INFO  gfx_backend_metal::device] Entry point EntryPoint { name: "main", execution_model: Vertex, work_group_size: WorkGroupSize { x: 0, y: 0, z: 0 } }
[2019-08-31T07:23:57Z INFO  gfx_backend_metal::device] Entry point EntryPoint { name: "main", execution_model: Fragment, work_group_size: WorkGroupSize { x: 0, y: 0, z: 0 } }
[2019-08-31T07:23:57Z INFO  shadow::framework] Entering render loop...
[2019-08-31T07:23:58Z WARN  gfx_backend_metal::window] Swapchain drawables are changed, unable to wait for 0
[2019-08-31T07:23:58Z WARN  gfx_backend_metal::window] Failed to get the drawable of frame 0
[2019-08-31T07:23:58Z WARN  wgpu_native::swap_chain] present failed: OutOfDate
[2019-08-31T07:23:58Z WARN  gfx_backend_metal::window] Swapchain drawables are changed, unable to wait for 1
[2019-08-31T07:23:58Z WARN  gfx_backend_metal::window] Failed to get the drawable of frame 1
[2019-08-31T07:23:58Z WARN  wgpu_native::swap_chain] present failed: OutOfDate
[2019-08-31T07:23:58Z WARN  gfx_backend_metal::window] No frame is available
[2019-08-31T07:23:58Z WARN  wgpu_native::swap_chain] acquire_image failed, re-creating
thread 'main' panicked at 'not yet implemented', /Users/parasyte/.cargo/git/checkouts/wgpu-53e70f8674b08dd4/b58c96e/wgpu-native/src/swap_chain.rs:132:13
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace.

I can get around the panic by updating the wgpu-native commit hash to gfx-rs/wgpu#314 But I still get the acquire_image failed warning quite often. Sometimes the renderer gives up and I get a seizure-inducing red/black flash at full frame rate. When it does work (rarely, meaning no acquire_image failed warning), the animation on the shadow example is much slower than it used to be (similar to #72). It's slow, but it doesn't use excessive amounts of CPU or anything.

FWIW I bisected this back to #70. Prior to this patch, everything is stable and solid. Haven't seen any warnings with this build at all.

[kvark]

The animation is slow because the timings were changed by a factor of 10: https://github.com/gfx-rs/wgpu-rs/pull/70/files#diff-46de19cddbd1eb0a33d7067f94f01bc7R261

The change in acquiring/resize behavior could be related to winit update. gfx-rs swapchain hasn't changed.

Also note that I'm currently seeing the swapchain resize in wgpu-rs (which happens immediately on start) to be wrong, so more fixes are coming.

[parasyte]

FWIW, I do not see any similar issues with gfx/examples dropping the swapchain. It just works:

$ RUST_LOG=info cargo run --bin quad --features metal
    Finished dev [unoptimized + debuginfo] target(s) in 0.23s
     Running `/Users/parasyte/other-projects/gfx/target/debug/quad`
AdapterInfo { name: "AMD Radeon Pro 560", vendor: 0, device: 0, device_type: DiscreteGpu }
AdapterInfo { name: "Intel(R) HD Graphics 630", vendor: 0, device: 0, device_type: IntegratedGpu }
Memory types: [MemoryType { properties: DEVICE_LOCAL, heap_index: 0 }, MemoryType { properties: CPU_VISIBLE | COHERENT, heap_index: 1 }, MemoryType { properties: DEVICE_LOCAL | CPU_VISIBLE, heap_index: 1 }, MemoryType { properties: DEVICE_LOCAL | CPU_VISIBLE | CPU_CACHED, heap_index: 1 }]
formats: Some([Bgra8Unorm, Bgra8Srgb, Rgba16Sfloat])
SwapchainConfig { present_mode: Fifo, composite_alpha: OPAQUE, format: Bgra8Srgb, extent: Extent2D { width: 1024, height: 768 }, image_count: 2, image_layers: 1, image_usage: COLOR_ATTACHMENT }
[2019-09-05T06:47:54Z INFO  gfx_backend_metal::window] build swapchain SwapchainConfig { present_mode: Fifo, composite_alpha: OPAQUE, format: Bgra8Srgb, extent: Extent2D { width: 1024, height: 768 }, image_count: 2, image_layers: 1, image_usage: COLOR_ATTACHMENT }
[2019-09-05T06:47:54Z INFO  gfx_backend_metal::device] Entry point EntryPoint { name: "main", execution_model: Vertex, work_group_size: WorkGroupSize { x: 0, y: 0, z: 0 } }
[2019-09-05T06:47:54Z INFO  gfx_backend_metal::device] Entry point EntryPoint { name: "main", execution_model: Fragment, work_group_size: WorkGroupSize { x: 0, y: 0, z: 0 } }
DROPPED!

[parasyte]

Ok, I take that back. :( Running the colour-uniform example clearly demonstrates the same issue. So this problem is in the metal backend. It wasn't entirely fixed by gfx-rs/gfx#2882 (though that did immensely help the quad example).

I suspect this is the same issue as reported in gfx-rs/gfx#2563

[parasyte]

Here's a gist showing the output from the gfx-hal colour-uniform example. https://gist.github.com/parasyte/a579dcfddba9f3ba639c3de0fe03663c

[kvark]

The gfx quad example was converted to use the new swapchain, other examples didn't.

…

On Sep 5, 2019, at 03:31, Jay Oster ***@***.***> wrote: Here's a gist showing the output from the gfx-hal quad example. https://gist.github.com/parasyte/a579dcfddba9f3ba639c3de0fe03663c — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[seivan]

@kvark Is the new swapchain stuff in gfx-rs master?
I assume it's not part of
gfx-hal = {git = "https://github.com/gfx-rs/gfx.git", branch = "hal-0.3" } ?

[seivan]

I get this flickering.

My swapchain management looks like

                match win_event {
                    Resized(_, _)
                    | SizeChanged(_, _)
                    | Maximized
                    | Hidden
                    | Minimized
                    | FocusGained
                    | FocusLost => {
                        println!("Recreating swapchain");
                        god_object.recreate_swap_chain = true;
                    }

I am not using Winit, but SDL2 - my code hasn't changed from when it worked to this, I just point to the master branch of wgpu exclusively.

This is with master branch wgpu-rs without any modification to the dependencies.
It feels like something broke or I've missed updating something...

[kvark]

It's in gfx master, but not in wgpu master. I'm working on an update

…

On Sep 5, 2019, at 08:39, Seivan Heidari ***@***.***> wrote: I get this flickering. My swapchain management looks like match win_event { Resized(_, _) | SizeChanged(_, _) | Maximized | Hidden | Minimized | FocusGained | FocusLost => { println!("Recreating swapchain"); god_object.recreate_swap_chain = true; } This is with master branch wgpu-rs without any modification to the dependencies. It feels like something broke or I've missed updating something... — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[lachlansneff]

@kvark Any updates on this?

[kvark]

We switched to gfx-rs from git yesterday. I'm working on the swapchain refactoring.

…

On Sep 6, 2019, at 19:50, Lachlan Sneff ***@***.***> wrote: @kvark Any updates on this? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[kvark]

gfx-rs/wgpu#330 already works on macOS, but there is still work to do before it can be merged.