Slide 1

Slide 1 text

Energy-Efficient 360-Degree Video Rendering on FPGA via Algorithm- Architecture Co-Design Qiuyue Sun Amir Taherin Yawo Siatitse Yuhao Zhu

Slide 2

Slide 2 text

Virtual Reality

Slide 3

Slide 3 text

No content

Slide 4

Slide 4 text

No content

Slide 5

Slide 5 text

360-Degree Video Delivery Pipeline

Slide 6

Slide 6 text

360-Degree Video Delivery Pipeline Original Frame

Slide 7

Slide 7 text

360-Degree Video Delivery Pipeline Rendering Original Frame

Slide 8

Slide 8 text

360-Degree Video Delivery Pipeline Rendering Original Frame

Slide 9

Slide 9 text

360-Degree Video Delivery Pipeline Rendering Field of View (FOV) Original Frame

Slide 10

Slide 10 text

360-Degree Video Delivery Pipeline Rendering Field of View (FOV) Consumes over 4 W power Exceeds TDP of typical mobile devices Original Frame

Slide 11

Slide 11 text

Rendering 4 Current Rendering Algorithm

Slide 12

Slide 12 text

Rendering 4 Current Rendering Algorithm Mapping Perspective Update Filtering

Slide 13

Slide 13 text

Rendering 4 Current Rendering Algorithm Mapping Perspective Update Filtering Matrix Multiplication

Slide 14

Slide 14 text

Rendering 4 Current Rendering Algorithm Mapping Perspective Update Filtering Matrix Multiplication Cartesian Coordinates

Slide 15

Slide 15 text

Rendering 4 Current Rendering Algorithm Mapping Perspective Update Filtering Matrix Multiplication Cartesian Coordinates Linear Interpolation

Slide 16

Slide 16 text

Current Implementation Field of View (FOV) Original Frame

Slide 17

Slide 17 text

Current Implementation (x, y) (x’, y’) Field of View (FOV) Original Frame

Slide 18

Slide 18 text

Current Implementation (x, y) (x’, y’) Field of View (FOV) Original Frame

Slide 19

Slide 19 text

Challenges: Memory Accesses

Slide 20

Slide 20 text

Challenges: Memory Accesses ▸ Irregular Access Pattern

Slide 21

Slide 21 text

Challenges: Memory Accesses ▸ Irregular Access Pattern ▹Accesses are not sequential

Slide 22

Slide 22 text

Challenges: Memory Accesses ▸ Irregular Access Pattern ▹Accesses are not sequential

Slide 23

Slide 23 text

Challenges: Memory Accesses ▸ Irregular Access Pattern ▹Accesses are not sequential ▹Severely hurts the efficiency of hardware acceleration

Slide 24

Slide 24 text

Challenges: Memory Accesses ▸ Irregular Access Pattern ▹Accesses are not sequential ▹Severely hurts the efficiency of hardware acceleration ▸ Large Footprint

Slide 25

Slide 25 text

Challenges: Memory Accesses ▸ Irregular Access Pattern ▹Accesses are not sequential ▹Severely hurts the efficiency of hardware acceleration ▸ Large Footprint ▹1080P is ~5.9 MB and 4K is ~23.7 MB

Slide 26

Slide 26 text

Challenges: Memory Accesses ▸ Irregular Access Pattern ▹Accesses are not sequential ▹Severely hurts the efficiency of hardware acceleration ▸ Large Footprint ▹1080P is ~5.9 MB and 4K is ~23.7 MB ▹Cannot be fully captured by a typical on-chip memory

Slide 27

Slide 27 text

Our Design (x’, y’) (x, y)

Slide 28

Slide 28 text

Our Design ▸ Enforce a streaming data access (x’, y’) (x, y)

Slide 29

Slide 29 text

Our Design ▸ Enforce a streaming data access (x’, y’) (x, y)

Slide 30

Slide 30 text

Our Design ▸ Enforce a streaming data access (x’, y’) (x, y)

Slide 31

Slide 31 text

Our Design ▸ Enforce a streaming data access ▸ Reduce unnecessary computations (x’, y’) (x, y)

Slide 32

Slide 32 text

Our Design ▸ Enforce a streaming data access ▸ Reduce unnecessary computations ▹Perform boundary checking

Slide 33

Slide 33 text

Our Design ▸ Enforce a streaming data access ▸ Reduce unnecessary computations ▹Perform boundary checking ▸ Fully pipeline pixel rendering

Slide 34

Slide 34 text

Setup and Evaluation 8

Slide 35

Slide 35 text

Setup and Evaluation 8 ▸ Xilinx Zynq UltraScale+ MPSoC ZCU104

Slide 36

Slide 36 text

Setup and Evaluation 8 ▸ Xilinx Zynq UltraScale+ MPSoC ZCU104 ▸ Pascal GPU on the Nvidia Jetson TX2

Slide 37

Slide 37 text

Setup and Evaluation 8 ▸ Xilinx Zynq UltraScale+ MPSoC ZCU104 ▸ Pascal GPU on the Nvidia Jetson TX2 ▸ Real User Trace Evaluation

Slide 38

Slide 38 text

Setup and Evaluation 8 ▸ Xilinx Zynq UltraScale+ MPSoC ZCU104 ▸ Pascal GPU on the Nvidia Jetson TX2 ▸ Real User Trace Evaluation ▸ Baseline: Original algorithm implemented on GPU and FPGA

Slide 39

Slide 39 text

Setup and Evaluation 8 Energy Savings(%) 0 20 40 60 RC Elephant NYC Rhino Paris Venice Saving over FPGA Savings over GPU ▸ Xilinx Zynq UltraScale+ MPSoC ZCU104 ▸ Pascal GPU on the Nvidia Jetson TX2 ▸ Real User Trace Evaluation ▸ Baseline: Original algorithm implemented on GPU and FPGA

Slide 40

Slide 40 text

Setup and Evaluation 8 Energy Savings(%) 0 20 40 60 RC Elephant NYC Rhino Paris Venice Saving over FPGA Savings over GPU ▸ Xilinx Zynq UltraScale+ MPSoC ZCU104 ▸ Pascal GPU on the Nvidia Jetson TX2 ▸ Real User Trace Evaluation ▸ Baseline: Original algorithm implemented on GPU and FPGA

Slide 41

Slide 41 text

Setup and Evaluation 8 Energy Savings(%) 0 20 40 60 RC Elephant NYC Rhino Paris Venice Saving over FPGA Savings over GPU ▸ Xilinx Zynq UltraScale+ MPSoC ZCU104 ▸ Pascal GPU on the Nvidia Jetson TX2 ▸ Real User Trace Evaluation ▸ Baseline: Original algorithm implemented on GPU and FPGA

Slide 42

Slide 42 text

Conclusion 9

Slide 43

Slide 43 text

Conclusion 9 ▸ Virtual reality popularity is growing rapidly

Slide 44

Slide 44 text

Conclusion 9 ▸ 360-degree video rendering consumes excessive power ▸ Virtual reality popularity is growing rapidly

Slide 45

Slide 45 text

Conclusion 9 ▸ 360-degree video rendering consumes excessive power ▸ Our co-design achieves on average 26.4% and 40.0% energy savings over baselines ▸ Virtual reality popularity is growing rapidly