feat(server): fully accelerated nvenc

docker/hwaccel.ml.yml

@@ -1,9 +1,7 @@
-version: "3.8"
-
 # Configurations for hardware-accelerated machine learning
 
 # If using Unraid or another platform that doesn't allow multiple Compose files,
-# you can inline the config for a backend by copying its contents 
+# you can inline the config for a backend by copying its contents
 # into the immich-machine-learning service in the docker-compose.yml file.
 
 # See https://immich.app/docs/features/ml-hardware-acceleration for info on usage.
@@ -30,7 +28,7 @@ services:
 
   openvino:
     device_cgroup_rules:
-      - "c 189:* rmw"
+      - 'c 189:* rmw'
     devices:
       - /dev/dri:/dev/dri
     volumes:

docker/hwaccel.transcoding.yml

@@ -1,5 +1,3 @@
-version: "3.8"
-
 # Configurations for hardware-accelerated transcoding
 
 # If using Unraid or another platform that doesn't allow multiple Compose files,

docs/docs/features/hardware-transcoding.md

@@ -22,7 +22,8 @@ You do not need to redo any transcoding jobs after enabling hardware acceleratio
 - WSL2 does not support Quick Sync.
 - Raspberry Pi is currently not supported.
 - Two-pass mode is only supported for NVENC. Other APIs will ignore this setting.
-- Only encoding is currently hardware accelerated, so the CPU is still used for software decoding and tone-mapping.
+- By default, only encoding is currently hardware accelerated. This means the CPU is still used for software decoding and tone-mapping.
+  - NVENC and RKMPP can be fully accelerated by enabling hardware decoding in the video transcoding settings.
 - Hardware dependent
   - Codec support varies, but H.264 and HEVC are usually supported.
     - Notably, NVIDIA and AMD GPUs do not support VP9 encoding.
@@ -65,6 +66,7 @@ For RKMPP to work:
 
 3. Redeploy the `immich-microservices` container with these updated settings.
 4. In the Admin page under `Video transcoding settings`, change the hardware acceleration setting to the appropriate option and save.
+5. (Optional) If using a compatible backend, you may enable hardware decoding for optimal performance.
 
 #### Single Compose File
 

open-api/immich-openapi-specs.json

@@ -9999,6 +9999,9 @@
           "accel": {
             "$ref": "#/components/schemas/TranscodeHWAccel"
           },
+          "accelDecode": {
+            "type": "boolean"
+          },
           "acceptedAudioCodecs": {
             "items": {
               "$ref": "#/components/schemas/AudioCodec"
@@ -10074,6 +10077,7 @@
         },
         "required": [
           "accel",
+          "accelDecode",
           "acceptedAudioCodecs",
           "acceptedVideoCodecs",
           "bframes",

open-api/typescript-sdk/src/fetch-client.ts

@@ -861,6 +861,7 @@ export type AssetFullSyncDto = {
 };
 export type SystemConfigFFmpegDto = {
     accel: TranscodeHWAccel;
+    accelDecode: boolean;
     acceptedAudioCodecs: AudioCodec[];
     acceptedVideoCodecs: VideoCodec[];
     bframes: number;

server/src/config.ts

@@ -97,6 +97,7 @@ export interface SystemConfig {
     preferredHwDevice: string;
     transcode: TranscodePolicy;
     accel: TranscodeHWAccel;
+    accelDecode: boolean;
     tonemap: ToneMapping;
   };
   job: Record<ConcurrentQueueName, { concurrency: number }>;
@@ -224,6 +225,7 @@ export const defaults = Object.freeze<SystemConfig>({
     transcode: TranscodePolicy.REQUIRED,
     tonemap: ToneMapping.HABLE,
     accel: TranscodeHWAccel.DISABLED,
+    accelDecode: false,
   },
   job: {
     [QueueName.BACKGROUND_TASK]: { concurrency: 5 },

server/src/dtos/system-config.dto.ts

@@ -132,6 +132,9 @@ export class SystemConfigFFmpegDto {
   @ApiProperty({ enumName: 'TranscodeHWAccel', enum: TranscodeHWAccel })
   accel!: TranscodeHWAccel;
 
+  @ValidateBoolean()
+  accelDecode!: boolean;
+
   @IsEnum(ToneMapping)
   @ApiProperty({ enumName: 'ToneMapping', enum: ToneMapping })
   tonemap!: ToneMapping;

server/src/services/media.service.spec.ts

@@ -1335,6 +1335,51 @@ describe(MediaService.name, () => {
       );
     });
 
+    it('should use hardware decoding for nvenc if enabled', async () => {
+      mediaMock.probe.mockResolvedValue(probeStub.matroskaContainer);
+      systemMock.get.mockResolvedValue({
+        ffmpeg: { accel: TranscodeHWAccel.NVENC, accelDecode: true },
+      });
+      assetMock.getByIds.mockResolvedValue([assetStub.video]);
+      await sut.handleVideoConversion({ id: assetStub.video.id });
+      expect(mediaMock.transcode).toHaveBeenCalledWith(
+        '/original/path.ext',
+        'upload/encoded-video/user-id/as/se/asset-id.mp4',
+        {
+          inputOptions: expect.arrayContaining([
+            '-hwaccel cuda',
+            '-hwaccel_output_format cuda',
+            '-noautorotate',
+            '-threads 1',
+          ]),
+          outputOptions: expect.arrayContaining([expect.stringContaining('scale_cuda=-2:720:format=nv12')]),
+          twoPass: false,
+        },
+      );
+    });
+
+    it('should use hardware tone-mapping for nvenc if hardware decoding is enabled and should tone map', async () => {
+      mediaMock.probe.mockResolvedValue(probeStub.videoStreamHDR);
+      systemMock.get.mockResolvedValue({
+        ffmpeg: { accel: TranscodeHWAccel.NVENC, accelDecode: true },
+      });
+      assetMock.getByIds.mockResolvedValue([assetStub.video]);
+      await sut.handleVideoConversion({ id: assetStub.video.id });
+      expect(mediaMock.transcode).toHaveBeenCalledWith(
+        '/original/path.ext',
+        'upload/encoded-video/user-id/as/se/asset-id.mp4',
+        {
+          inputOptions: expect.arrayContaining(['-hwaccel cuda', '-hwaccel_output_format cuda']),
+          outputOptions: expect.arrayContaining([
+            expect.stringContaining(
+              'tonemap_cuda=desat=0:matrix=bt709:primaries=bt709:range=pc:tonemap=hable:transfer=bt709:format=nv12',
+            ),
+          ]),
+          twoPass: false,
+        },
+      );
+    });
+
     it('should set options for qsv', async () => {
       storageMock.readdir.mockResolvedValue(['renderD128']);
       mediaMock.probe.mockResolvedValue(probeStub.matroskaContainer);
@@ -1633,8 +1678,9 @@ describe(MediaService.name, () => {
 
     it('should set options for rkmpp', async () => {
       storageMock.readdir.mockResolvedValue(['renderD128']);
+      storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => true, isCharacterDevice: () => true });
       mediaMock.probe.mockResolvedValue(probeStub.matroskaContainer);
-      systemMock.get.mockResolvedValue({ ffmpeg: { accel: TranscodeHWAccel.RKMPP } });
+      systemMock.get.mockResolvedValue({ ffmpeg: { accel: TranscodeHWAccel.RKMPP, accelDecode: true } });
       assetMock.getByIds.mockResolvedValue([assetStub.video]);
       await sut.handleVideoConversion({ id: assetStub.video.id });
       expect(mediaMock.transcode).toHaveBeenCalledWith(
@@ -1663,10 +1709,12 @@ describe(MediaService.name, () => {
 
     it('should set vbr options for rkmpp when max bitrate is enabled', async () => {
       storageMock.readdir.mockResolvedValue(['renderD128']);
+      storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => true, isCharacterDevice: () => true });
       mediaMock.probe.mockResolvedValue(probeStub.videoStreamVp9);
       systemMock.get.mockResolvedValue({
         ffmpeg: {
           accel: TranscodeHWAccel.RKMPP,
+          accelDecode: true,
           maxBitrate: '10000k',
           targetVideoCodec: VideoCodec.HEVC,
         },
@@ -1686,9 +1734,10 @@ describe(MediaService.name, () => {
 
     it('should set cqp options for rkmpp when max bitrate is disabled', async () => {
       storageMock.readdir.mockResolvedValue(['renderD128']);
+      storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => true, isCharacterDevice: () => true });
       mediaMock.probe.mockResolvedValue(probeStub.matroskaContainer);
       systemMock.get.mockResolvedValue({
-        ffmpeg: { accel: TranscodeHWAccel.RKMPP, crf: 30, maxBitrate: '0' },
+        ffmpeg: { accel: TranscodeHWAccel.RKMPP, accelDecode: true, crf: 30, maxBitrate: '0' },
       });
       assetMock.getByIds.mockResolvedValue([assetStub.video]);
       await sut.handleVideoConversion({ id: assetStub.video.id });
@@ -1707,7 +1756,9 @@ describe(MediaService.name, () => {
       storageMock.readdir.mockResolvedValue(['renderD128']);
       storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => true, isCharacterDevice: () => true });
       mediaMock.probe.mockResolvedValue(probeStub.videoStreamHDR);
-      systemMock.get.mockResolvedValue({ ffmpeg: { accel: TranscodeHWAccel.RKMPP, crf: 30, maxBitrate: '0' } });
+      systemMock.get.mockResolvedValue({
+        ffmpeg: { accel: TranscodeHWAccel.RKMPP, accelDecode: true, crf: 30, maxBitrate: '0' },
+      });
       assetMock.getByIds.mockResolvedValue([assetStub.video]);
       await sut.handleVideoConversion({ id: assetStub.video.id });
       expect(mediaMock.transcode).toHaveBeenCalledWith(
@@ -1724,6 +1775,54 @@ describe(MediaService.name, () => {
         },
       );
     });
+
+    it('should use software decoding and tone-mapping if hardware decoding is disabled', async () => {
+      storageMock.readdir.mockResolvedValue(['renderD128']);
+      storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => true, isCharacterDevice: () => true });
+      mediaMock.probe.mockResolvedValue(probeStub.videoStreamHDR);
+      systemMock.get.mockResolvedValue({
+        ffmpeg: { accel: TranscodeHWAccel.RKMPP, accelDecode: false, crf: 30, maxBitrate: '0' },
+      });
+      assetMock.getByIds.mockResolvedValue([assetStub.video]);
+      await sut.handleVideoConversion({ id: assetStub.video.id });
+      expect(mediaMock.transcode).toHaveBeenCalledWith(
+        '/original/path.ext',
+        'upload/encoded-video/user-id/as/se/asset-id.mp4',
+        {
+          inputOptions: [],
+          outputOptions: expect.arrayContaining([
+            expect.stringContaining(
+              'zscale=t=linear:npl=100,tonemap=hable:desat=0,zscale=p=bt709:t=bt709:m=bt709:range=pc,format=yuv420p',
+            ),
+          ]),
+          twoPass: false,
+        },
+      );
+    });
+
+    it('should use software decoding and tone-mapping if opencl is not available', async () => {
+      storageMock.readdir.mockResolvedValue(['renderD128']);
+      storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => false, isCharacterDevice: () => false });
+      mediaMock.probe.mockResolvedValue(probeStub.videoStreamHDR);
+      systemMock.get.mockResolvedValue({
+        ffmpeg: { accel: TranscodeHWAccel.RKMPP, accelDecode: true, crf: 30, maxBitrate: '0' },
+      });
+      assetMock.getByIds.mockResolvedValue([assetStub.video]);
+      await sut.handleVideoConversion({ id: assetStub.video.id });
+      expect(mediaMock.transcode).toHaveBeenCalledWith(
+        '/original/path.ext',
+        'upload/encoded-video/user-id/as/se/asset-id.mp4',
+        {
+          inputOptions: [],
+          outputOptions: expect.arrayContaining([
+            expect.stringContaining(
+              'zscale=t=linear:npl=100,tonemap=hable:desat=0,zscale=p=bt709:t=bt709:m=bt709:range=pc,format=yuv420p',
+            ),
+          ]),
+          twoPass: false,
+        },
+      );
+    });
   });
 
   it('should tonemap when policy is required and video is hdr', async () => {

server/src/services/media.service.ts

@@ -36,9 +36,11 @@ import {
   AV1Config,
   H264Config,
   HEVCConfig,
-  NVENCConfig,
+  NvencHwDecodeConfig,
+  NvencSwDecodeConfig,
   QSVConfig,
-  RKMPPConfig,
+  RkmppHwDecodeConfig,
+  RkmppSwDecodeConfig,
   ThumbnailConfig,
   VAAPIConfig,
   VP9Config,
@@ -360,8 +362,7 @@ export class MediaService {
           `Error occurred during transcoding. Retrying with ${config.accel.toUpperCase()} acceleration disabled.`,
         );
       }
-      config.accel = TranscodeHWAccel.DISABLED;
-      transcodeOptions = await this.getCodecConfig(config).then((c) =>
+      transcodeOptions = await this.getCodecConfig({ ...config, accel: TranscodeHWAccel.DISABLED }).then((c) =>
         c.getOptions(target, mainVideoStream, mainAudioStream),
       );
       await this.mediaRepository.transcode(input, output, transcodeOptions);
@@ -494,7 +495,7 @@ export class MediaService {
     let handler: VideoCodecHWConfig;
     switch (config.accel) {
       case TranscodeHWAccel.NVENC: {
-        handler = new NVENCConfig(config);
+        handler = config.accelDecode ? new NvencHwDecodeConfig(config) : new NvencSwDecodeConfig(config);
         break;
       }
       case TranscodeHWAccel.QSV: {
@@ -506,7 +507,10 @@ export class MediaService {
         break;
       }
       case TranscodeHWAccel.RKMPP: {
-        handler = new RKMPPConfig(config, await this.getDevices(), await this.hasOpenCL());
+        handler =
+          config.accelDecode && (await this.hasOpenCL())
+            ? new RkmppHwDecodeConfig(config, await this.getDevices())
+            : new RkmppSwDecodeConfig(config, await this.getDevices());
         break;
       }
       default: {

server/src/services/system-config.service.spec.ts

@@ -66,6 +66,7 @@ const updatedConfig = Object.freeze<SystemConfig>({
     preferredHwDevice: 'auto',
     transcode: TranscodePolicy.REQUIRED,
     accel: TranscodeHWAccel.DISABLED,
+    accelDecode: false,
     tonemap: ToneMapping.HABLE,
   },
   logging: {