Benchmarks

The Video Editing test use cases capture some common uses of video editing applications.

Implementation

The Video Editing test uses parts from PCMark 8 Video Editing and Media To Go tests.

Windows Media Foundation is used with its built-in codecs to transcode video. Hardware acceleration is allowed to be used if the system supports it and has the necessary Media Foundation setup done.

The Video Editing test uses FFmpeg on the sharpening and deshaking parts. The test uses pre-built FFmpeg binaries.

The Video editing test supports OpenCL. The benchmark application selects a preferred OpenCL device to use.

Part 1: on the go 

Stage 1: Fast downscaling

• Code path: x86/x64
• Uses Media Foundation Fast transcode feature to transcode video files to a format suitable for mobile use
• Code path: x86/x64 and whatever is the implementation with Media Foundation H.264 codecs installed on the system
• Similar to Video to go test in PCMark 8

Part 2: Sharpening

• Sharpens the 1080p H.264 video
• Uses publicly available executable FFmpeg.exe
• Command line: FFmpeg.exe -y -v 40 -i <input file> -vf scale=w=1920:h=1080:flags=bicubic,unsharp=opencl=%OCL%:lx=7:ly=7:la=0.56:cx=7:cy=7:ca=0.28 -strict -2 <output file> 
• Stage 2 - CPU:
  • Run always
  • Code path: x86/x64
• Stage 2 - OCL:
  • Condition: OpenCL device available
  • Code path: OpenCL
  • If the OpenCL of the Stage 2 test takes longer than the CPU version, the CPU result is used instead 

Part 3: Deshaking

• Uses publicly available executable FFmpeg.exe
• Video deshaked
• Command line: FFmpeg.exe -y -v 40 -i <input file> -vf deshake=opencl=%OCL%:search=1:blocksize=16,crop=in_w-384:in_h-216,scale=w=1920:h=1080 -strict -2 <output file>
• Stage 3 - CPU:
  • Run always
  • Code path: x86/x64
• Stage 3 - OpenCL:
  • Condition: OpenCL device available
  • Code path: OpenCL
  • If the OpenCL test takes longer than the CPU test, the CPU run time is used instead

Workloads

The Video Editing test simulates editing videos in a video editing program. The workloads perform the following tasks:

1. Fast downscaling, common for example in use with mobile devices
2. Sharpening the video
3. Deshaking filtering

The workloads measure the frames produced per second in the video editing program.

On the go    =    M_1 / M_2
 
Where: 
M_1    =    dbg_pcm10_video_go_frames
M_2    =    dbg_pcm10_video_go_time

Sharpening CPU    =    M_3 / M_4
 
Where: 
M_3    =    dbg_pcm10_video_downscale_frames
M_4    =    dbg_pcm10_video_downscale_time_cpu

Sharpening OCL    =    M_5 / Min(M_6,M_7)
 
Where: 
M_5    =    dbg_pcm10_video_downscale_frames
M_6    =    dbg_pcm10_video_downscale_time_ocl
M_7    =    dbg_pcm10_video_downscale_time_cpu

Deshaking CPU    =    M_8 / M_9
 
Where: 
M_8    =    dbg_pcm10_video_deshake_frames
M_9    =    dbg_pcm10_video_deshake_time_cpu

Deshaking OCL    =    M_10 / Min(M_11,M_12)
 
Where: 
M_10    =    dbg_pcm10_video_deshake_frames
M_11    =    dbg_pcm10_video_deshake_time_ocl
M_12    =    dbg_pcm10_video_deshake_time_cpu

Video Editing score

We use a weighted harmonic mean to calculate the Video Editing score from the workload scores.

Video Editing score    =    K × geomean(R_1,R_2,R_3)

Where: 
K      =    scoring coefficient    =    93
R_2    =     geomean(A_1,A_2)
R_3    =     geomean(A_3,A_4)