optimize tonemapping performance (#2037)

* add env var to disable vulkan

* tonemap after scaling

* vulkan tonemapping still needs to happen before scaling

CHANGELOG.md

@@ -26,8 +26,11 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
   - VAAPI may use hardware-accelerated tone mapping (when opencl accel is also available)
   - NVIDIA may use hardware-accelerated tone mapping (when vulkan accel and libplacebo filter are also available)
   - QSV may use hardware-accelerated tone mapping (when hardware decoding is used)
-  - In all other cases, HDR content will use a software pipeline and the clip algorithm
+  - In all other cases, HDR content will use a software pipeline and the linear algorithm
 - Use hardware-accelerated padding with VAAPI
+- Add environment variable `ETV_DISABLE_VULKAN`
+  - Any non-empty value will disable use of Vulkan acceleration and force software tonemapping
+  - This may be needed with misbehaving NVIDIA drivers on Windows
 
 ### Changed
 - Start to make UI minimally responsive (functional on smaller screens)

ErsatzTV.FFmpeg/Filter/TonemapFilter.cs

@@ -20,7 +20,7 @@ public class TonemapFilter : BaseFilter
             string pixelFormat = _currentState.PixelFormat.Match(pf => pf.FFmpegName, () => string.Empty);
 
             var tonemap =
-                $"zscale=transfer=linear,tonemap=clip,zscale=transfer=bt709,format={_desiredPixelFormat.FFmpegName}";
+                $"zscale=transfer=linear,tonemap=linear,zscale=transfer=bt709,format={_desiredPixelFormat.FFmpegName}";
 
             if (_currentState.FrameDataLocation == FrameDataLocation.Hardware)
             {

ErsatzTV.FFmpeg/Pipeline/NvidiaPipelineBuilder.cs

@@ -80,7 +80,8 @@ public class NvidiaPipelineBuilder : SoftwarePipelineBuilder
 
         bool isHdrTonemap = decodeCapability == FFmpegCapability.Hardware
                             && _ffmpegCapabilities.HasHardwareAcceleration(HardwareAccelerationMode.Vulkan)
-                            && videoStream.ColorParams.IsHdr;
+                            && videoStream.ColorParams.IsHdr
+                            && string.IsNullOrWhiteSpace(System.Environment.GetEnvironmentVariable("ETV_DISABLE_VULKAN"));
 
         if (decodeCapability == FFmpegCapability.Hardware || encodeCapability == FFmpegCapability.Hardware)
         {
@@ -181,9 +182,20 @@ public class NvidiaPipelineBuilder : SoftwarePipelineBuilder
         //     desiredState = desiredState with { PixelFormat = Some(pixelFormat) };
         // }
 
-        currentState = SetTonemap(videoInputFile, videoStream, ffmpegState, desiredState, currentState);
+        // vulkan scale doesn't seem to handle HDR, so we need to tonemap before scaling
+        if (ffmpegState.IsHdrTonemap)
+        {
+            currentState = SetTonemap(videoInputFile, videoStream, ffmpegState, desiredState, currentState);
+        }
+
         currentState = SetDeinterlace(videoInputFile, context, currentState);
         currentState = SetScale(videoInputFile, videoStream, context, ffmpegState, desiredState, currentState);
+
+        if (!ffmpegState.IsHdrTonemap)
+        {
+            currentState = SetTonemap(videoInputFile, videoStream, ffmpegState, desiredState, currentState);
+        }
+
         currentState = SetPad(videoInputFile, videoStream, desiredState, currentState);
         currentState = SetCrop(videoInputFile, desiredState, currentState);
         SetStillImageLoop(videoInputFile, videoStream, ffmpegState, desiredState, pipelineSteps);
@@ -673,19 +685,19 @@ public class NvidiaPipelineBuilder : SoftwarePipelineBuilder
             scaleStep = new ScaleCudaFilter(
                 currentState with
                 {
-                    PixelFormat = !context.Is10BitOutput && (context.HasWatermark ||
-                                                             context.HasSubtitleOverlay ||
-                                                             context.ShouldDeinterlace ||
-                                                             desiredState.ScaledSize != desiredState.PaddedSize ||
-                                                             context.HasSubtitleText ||
-                                                             ffmpegState is
-                                                             {
-                                                                 DecoderHardwareAccelerationMode:
-                                                                 HardwareAccelerationMode.Nvenc,
-                                                                 EncoderHardwareAccelerationMode:
-                                                                 HardwareAccelerationMode.None
-                                                             })
-                        ? desiredState.PixelFormat.Map(pf => (IPixelFormat)new PixelFormatNv12(pf.Name))
+                    PixelFormat = context is { IsHdr: false, Is10BitOutput: false } && (context.HasWatermark ||
+                        context.HasSubtitleOverlay ||
+                        context.ShouldDeinterlace ||
+                        desiredState.ScaledSize != desiredState.PaddedSize ||
+                        context.HasSubtitleText ||
+                        ffmpegState is
+                        {
+                            DecoderHardwareAccelerationMode:
+                            HardwareAccelerationMode.Nvenc,
+                            EncoderHardwareAccelerationMode:
+                            HardwareAccelerationMode.None
+                        })
+                        ? desiredState.PixelFormat.Map(IPixelFormat (pf) => new PixelFormatNv12(pf.Name))
                         : Option<IPixelFormat>.None
                 },
                 desiredState.ScaledSize,

ErsatzTV.FFmpeg/Pipeline/PipelineBuilderBase.cs

@@ -199,7 +199,8 @@ public abstract class PipelineBuilderBase : IPipelineBuilder
             _subtitleInputFile.Map(s => s is { IsImageBased: false, Method: SubtitleMethod.Burn }).IfNone(false),
             desiredState.Deinterlaced,
             desiredState.PixelFormat.Map(pf => pf.BitDepth).IfNone(8) == 10,
-            false);
+            false,
+            videoStream.ColorParams.IsHdr);
 
         SetThreadCount(ffmpegState, desiredState, pipelineSteps);
         SetSceneDetect(videoStream, ffmpegState, desiredState, pipelineSteps);

ErsatzTV.FFmpeg/Pipeline/PipelineContext.cs

@@ -7,4 +7,5 @@ public record PipelineContext(
     bool HasSubtitleText,
     bool ShouldDeinterlace,
     bool Is10BitOutput,
-    bool IsIntelVaapiOrQsv);
+    bool IsIntelVaapiOrQsv,
+    bool IsHdr);

ErsatzTV.FFmpeg/Pipeline/QsvPipelineBuilder.cs

@@ -169,11 +169,11 @@ public class QsvPipelineBuilder : SoftwarePipelineBuilder
         }
 
         // _logger.LogDebug("After decode: {PixelFormat}", currentState.PixelFormat);
-        currentState = SetTonemap(videoInputFile, videoStream, ffmpegState, desiredState, currentState);
         currentState = SetDeinterlace(videoInputFile, context, ffmpegState, currentState);
         // _logger.LogDebug("After deinterlace: {PixelFormat}", currentState.PixelFormat);
         currentState = SetScale(videoInputFile, videoStream, context, ffmpegState, desiredState, currentState);
         // _logger.LogDebug("After scale: {PixelFormat}", currentState.PixelFormat);
+        currentState = SetTonemap(videoInputFile, videoStream, ffmpegState, desiredState, currentState);
         currentState = SetPad(videoInputFile, videoStream, desiredState, currentState);
         // _logger.LogDebug("After pad: {PixelFormat}", currentState.PixelFormat);
         currentState = SetCrop(videoInputFile, desiredState, currentState);

ErsatzTV.FFmpeg/Pipeline/SoftwarePipelineBuilder.cs

@@ -104,8 +104,8 @@ public class SoftwarePipelineBuilder : PipelineBuilderBase
         {
             SetDeinterlace(videoInputFile, context, currentState);
 
-            currentState = SetTonemap(videoInputFile, videoStream, desiredState, currentState);
             currentState = SetScale(videoInputFile, videoStream, desiredState, currentState);
+            currentState = SetTonemap(videoInputFile, videoStream, desiredState, currentState);
             currentState = SetPad(videoInputFile, videoStream, desiredState, currentState);
             currentState = SetCrop(videoInputFile, desiredState, currentState);
             SetStillImageLoop(videoInputFile, videoStream, ffmpegState, desiredState, pipelineSteps);

ErsatzTV.FFmpeg/Pipeline/VaapiPipelineBuilder.cs

@@ -171,14 +171,14 @@ public class VaapiPipelineBuilder : SoftwarePipelineBuilder
 
         // _logger.LogDebug("After decode: {PixelFormat}", currentState.PixelFormat);
 
-        currentState = SetTonemap(videoInputFile, videoStream, ffmpegState, desiredState, currentState);
-
         currentState = SetDeinterlace(videoInputFile, context, ffmpegState, currentState);
         // _logger.LogDebug("After deinterlace: {PixelFormat}", currentState.PixelFormat);
 
         currentState = SetScale(videoInputFile, videoStream, context, ffmpegState, desiredState, currentState);
         // _logger.LogDebug("After scale: {PixelFormat}", currentState.PixelFormat);
 
+        currentState = SetTonemap(videoInputFile, videoStream, ffmpegState, desiredState, currentState);
+
         currentState = SetPad(videoInputFile, desiredState, currentState);
         // _logger.LogDebug("After pad: {PixelFormat}", currentState.PixelFormat);