Dynamic Trade-Off Management for CPS Francesca Palumbo 1 , Claudio - - PowerPoint PPT Presentation

▶
dynamic trade off management for cps
SMART_READER_LITE
LIVE PREVIEW

Dynamic Trade-Off Management for CPS Francesca Palumbo 1 , Claudio - - PowerPoint PPT Presentation

Jun 06, 2023 547 likes •999 views

2nd Italian Workshop on Embedded Systems IWES 2017 Dynamic Trade-Off Management for CPS Francesca Palumbo 1 , Claudio Rubattu 1 , Carlo Sau 2 , Tiziana Fanni 2 , Paolo Meloni 2 , Luigi Raffo 2 1 Universit degli Studi di Sassari, PolComIng 2

slide-1
SLIDE 1

2nd Italian Workshop on Embedded Systems

IWES 2017

Dynamic Trade-Off Management for CPS

Francesca Palumbo1, Claudio Rubattu1, Carlo Sau2, Tiziana Fanni2, Paolo Meloni2, Luigi Raffo2

1Università degli Studi di Sassari, PolComIng 2Università degli Studi di Cagliari, DIEE

7-8 September 2017 – Rome (IT)

slide-2
SLIDE 2

Outline

  • Introduction
  • CERBERO and Cyber Physical Systems
  • HEVC Codec and Software Approximate Computing
  • Approximate HEVC interpolators
  • Coarse-Grained Reconfiguration
  • From CG HEVC Interpolators to CGR HEVC Interpolators
  • Results
  • Achieved Adaptivity
  • Comparison with the State of the Art
  • Conclusions

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-3
SLIDE 3

Outline

  • Introduction
  • CERBERO and Cyber Physical Systems
  • HEVC Codec and Software Approximate Computing
  • Approximate HEVC interpolators
  • Coarse-Grained Reconfiguration
  • From CG HEVC Interpolators to CGR HEVC Interpolators
  • Results
  • Achieved Adaptivity
  • Comparison with the State of the Art
  • Conclusions

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-4
SLIDE 4

CERBERO project

Cross-layer modEl-based framewoRk for multi-oBjective dEsign of Reconfigurable systems in unceRtain hybRid envirOnments  continuous design environment for Cyber-Physical Systems (CPS) including modelling, deployment and verification

http://www.cerbero-h2020.eu/

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-5
SLIDE 5

CERBERO project

Cross-layer modEl-based framewoRk for multi-oBjective dEsign of Reconfigurable systems in unceRtain hybRid envirOnments  continuous design environment for Cyber-Physical Systems (CPS) including modelling, deployment and verification

Self-healing system for planetary exploration

http://www.cerbero-h2020.eu/

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-6
SLIDE 6

CERBERO project

Cross-layer modEl-based framewoRk for multi-oBjective dEsign of Reconfigurable systems in unceRtain hybRid envirOnments  continuous design environment for Cyber-Physical Systems (CPS) including modelling, deployment and verification

Self-healing system for planetary exploration Smart Travelling for Electric Vehicle

http://www.cerbero-h2020.eu/

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-7
SLIDE 7

CERBERO project

Cross-layer modEl-based framewoRk for multi-oBjective dEsign of Reconfigurable systems in unceRtain hybRid envirOnments  continuous design environment for Cyber-Physical Systems (CPS) including modelling, deployment and verification

Self-healing system for planetary exploration Oceans Monitoring Smart Travelling for Electric Vehicle

http://www.cerbero-h2020.eu/

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-8
SLIDE 8

Cyber Physical Systems (CPS)

Complex systems with different interacting and deeply intertwined components, providing multiple and distinct behavioral modalities potentially changing over time, that contribute concurrently to determine the behavior of the system as a whole.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-9
SLIDE 9

Cyber Physical Systems (CPS)

Complex systems with different interacting and deeply intertwined components, providing multiple and distinct behavioral modalities potentially changing over time, that contribute concurrently to determine the behavior of the system as a whole. Layers (dominat aspects):

  • functional
  • physical
  • communication

Subjected to Functional (F) and Non-Functional (NF) requirements variation in time.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-10
SLIDE 10

Cyber Physical Systems (CPS)

Complex systems with different interacting and deeply intertwined components, providing multiple and distinct behavioral modalities potentially changing over time, that contribute concurrently to determine the behavior of the system as a whole. Layers (dominat aspects):

  • functional
  • physical
  • communication

Subjected to Functional (F) and Non-Functional (NF) requirements variation in time.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-11
SLIDE 11

High Efficiency Video Coding (HEVC)

Recent video codec developed by the Joint Collaboration Team on Video Coding (VCEG and MPEG). It provides up to 50% bit rate reduction at the same subjective video quality with respect to previous standards (H.264).

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-12
SLIDE 12

High Efficiency Video Coding (HEVC)

Recent video codec developed by the Joint Collaboration Team on Video Coding (VCEG and MPEG). It provides up to 50% bit rate reduction at the same subjective video quality with respect to previous standards (H.264).

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-13
SLIDE 13

High Efficiency Video Coding (HEVC)

Recent video codec developed by the Joint Collaboration Team on Video Coding (VCEG and MPEG). It provides up to 50% bit rate reduction at the same subjective video quality with respect to previous standards (H.264).

TB MV

P-frame

MB SW

earlier reference frame

Computational Intensive and Power Hungry Step: Motion Estimation and Compensation

TB = target block SW = search window MB = matching block MV = motion vector

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-14
SLIDE 14

Approximate HEVC Interpolation in Software

A00 A01 A10 A11 a00 b00 c00 d00 e00 f00 g00 i00 l00 m00 n00 o00 p00 q00 h00

1/4 1/4 1/2 3/4 3/4 1/2

With high frame rates the motion vector could be composed of fractional pixel values. In these cases an interpolation (FIR filtering) of the matching block is necessary.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-15
SLIDE 15

Approximate HEVC Interpolation in Software

A00 A01 A10 A11 a00 b00 c00 d00 e00 f00 g00 i00 l00 m00 n00 o00 p00 q00 h00

1/4 1/4 1/2 3/4 3/4 1/2

With high frame rates the motion vector could be composed of fractional pixel values. In these cases an interpolation (FIR filtering) of the matching block is necessary.

luma MV α legacy approximate [1] 8/7 tap 7 tap 5 tap 3 tap 1 tap 1/4, 3/4

  • 1, 4, -10, 58, 17, -5, 1
  • 1, 4, -10, 58, 17, -5, 1

1, -6, 20, 54, -5

  • 4, 20, 48

64 1/2

  • 1, 4, -11, 40, 40, -11, 4, -1
  • 1, 4, 11, 40, 40, -11, 3

2, -9, 40, 40, -9

  • 9, 41, 32

64 chroma MV α legacy approximate [1] 4 tap 3 tap 2 tap 1 tap 1/8, 7/8

  • 2, 58, 10, -2
  • 3, 62, 5

58, 7 64 1/4, 3/4

  • 4, 54, 16, -2
  • 5, 58, 11

50, 15 64 3/8, 5/8

  • 6, 46, 28, -4
  • 7, 51, 20

41, 23 64 1/2

  • 4, 36, 36, -4
  • 6, 42, 28

32, 32 64 [1] E. Nogues et al., “Algorithmic-level approximate computing applied to energy efficient hevc decoding,” IEEE Trans. On Emerging Topics in Computing, 2016.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-16
SLIDE 16

Approximate HEVC Interpolation in Software

A00 A01 A10 A11 a00 b00 c00 d00 e00 f00 g00 i00 l00 m00 n00 o00 p00 q00 h00

1/4 1/4 1/2 3/4 3/4 1/2

With high frame rates the motion vector could be composed of fractional pixel values. In these cases an interpolation (FIR filtering) of the matching block is necessary.

luma MV α legacy approximate [1] 8/7 tap 7 tap 5 tap 3 tap 1 tap 1/4, 3/4

  • 1, 4, -10, 58, 17, -5, 1
  • 1, 4, -10, 58, 17, -5, 1

1, -6, 20, 54, -5

  • 4, 20, 48

64 1/2

  • 1, 4, -11, 40, 40, -11, 4, -1
  • 1, 4, 11, 40, 40, -11, 3

2, -9, 40, 40, -9

  • 9, 41, 32

64 chroma MV α legacy approximate [1] 4 tap 3 tap 2 tap 1 tap 1/8, 7/8

  • 2, 58, 10, -2
  • 3, 62, 5

58, 7 64 1/4, 3/4

  • 4, 54, 16, -2
  • 5, 58, 11

50, 15 64 3/8, 5/8

  • 6, 46, 28, -4
  • 7, 51, 20

41, 23 64 1/2

  • 4, 36, 36, -4
  • 6, 42, 28

32, 32 64 [1] E. Nogues et al., “Algorithmic-level approximate computing applied to energy efficient hevc decoding,” IEEE Trans. On Emerging Topics in Computing, 2016.

 up to 28% energy saving with a small degradation of decoding quality on an ARM big.LITTLE SoC

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-17
SLIDE 17

Outline

  • Introduction
  • CERBERO and Cyber Physical Systems
  • HEVC Codec and Software Approximate Computing
  • Approximate HEVC interpolators
  • Coarse-Grained Reconfiguration
  • From CG HEVC Interpolators to CGR HEVC Interpolators
  • Results
  • Achieved Adaptivity
  • Comparison with the State of the Art
  • Conclusions

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-18
SLIDE 18

DSP ASIC GPU CPU GP

Flexibility Efficiency

CG RECONF FG

Coarse-Grained Reconfiguration (CGR)

Reconfigurable computing provides a trade-off between execution efficiency typical of ASICs and flexibility mainly exhibited by GP devices.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-19
SLIDE 19

DSP ASIC GPU CPU GP

Flexibility Efficiency

CG RECONF FG

Coarse-Grained Reconfiguration (CGR)

Fine-Grained (FG) Coarse-Grained (CG) bit-level word-level flexibility   speed   memory  

Reconfigurable computing provides a trade-off between execution efficiency typical of ASICs and flexibility mainly exhibited by GP devices.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-20
SLIDE 20

CG HEVC interpolators

filtered pixel

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63

X8x8

example N=4

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-21
SLIDE 21

CG HEVC interpolators

filtered pixel

serial horizontal 8-taps FIR

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63

  • 1

2 3 4 5 9 10 11 12 13 17 18 19 20 21 25 26 27 28 29 33 34 35 36 37 41 42 43 44 45 49 50 51 52 53 57 58 59 60 61

  • y1=c0x0+c1x1+c2x2+c3x3

y2=c0x1+c1x2+c2x3+c3x4 y3=c0x2+c1x3+c2x4+c3x5

 horizontal Filtering (N-1 cols)

X8x8 Y8x5

example N=4

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-22
SLIDE 22

CG HEVC interpolators

filtered pixel

serial horizontal 8-taps FIR horizontally filtered rows

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63

  • 1

2 3 4 5 9 10 11 12 13 17 18 19 20 21 25 26 27 28 29 33 34 35 36 37 41 42 43 44 45 49 50 51 52 53 57 58 59 60 61

  • y1=c0x0+c1x1+c2x2+c3x3

y2=c0x1+c1x2+c2x3+c3x4 y3=c0x2+c1x3+c2x4+c3x5

 horizontal Filtering (N-1 cols)

X8x8 Y8x5

example N=4

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-23
SLIDE 23
  • 9

10 11

  • 12

13 17 18 19 20 21 25 26 27 28 29 33 34 35 36 37

  • 41
  • 42
  • 43
  • 44
  • 45
  • CG HEVC interpolators

filtered pixel

serial horizontal 8-taps FIR parallel vertical 8-taps FIR horizontally filtered rows

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63

  • 1

2 3 4 5 9 10 11 12 13 17 18 19 20 21 25 26 27 28 29 33 34 35 36 37 41 42 43 44 45 49 50 51 52 53 57 58 59 60 61

  • y1=c0x0+c1x1+c2x2+c3x3

y2=c0x1+c1x2+c2x3+c3x4 y3=c0x2+c1x3+c2x4+c3x5 z9=c0y1+c1y9+c2y17+c3y25 z10=c0y2+c1y10+c2y18+c3y26 z11=c0y1+c1y9+c2y17+c3y25

 horizontal Filtering (N-1 cols)  vertical Filtering (N-1 rows)

X8x8 Y8x5 Z5x5

example N=4

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-24
SLIDE 24
  • 9

10 11

  • 12

13 17 18 19 20 21 25 26 27 28 29 33 34 35 36 37

  • 41
  • 42
  • 43
  • 44
  • 45
  • CG HEVC interpolators

filtered pixel

serial horizontal 8-taps FIR parallel vertical 8-taps FIR horizontally filtered rows final normalization step

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63

  • 1

2 3 4 5 9 10 11 12 13 17 18 19 20 21 25 26 27 28 29 33 34 35 36 37 41 42 43 44 45 49 50 51 52 53 57 58 59 60 61

  • y1=c0x0+c1x1+c2x2+c3x3

y2=c0x1+c1x2+c2x3+c3x4 y3=c0x2+c1x3+c2x4+c3x5 z9=c0y1+c1y9+c2y17+c3y25 z10=c0y2+c1y10+c2y18+c3y26 z11=c0y1+c1y9+c2y17+c3y25

 horizontal Filtering (N-1 cols)  vertical Filtering (N-1 rows)

X8x8 Y8x5 Z5x5

example N=4

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-25
SLIDE 25

CGR HEVC interpolators

profile HIGH MEDIUM LOW quality # taps energy quality # taps energy quality # taps energy luma  8/7   5   3  chroma 4 3 2

[2] F. Palumbo et al., “Runtime energy versus quality tuning in motion compensation filters for HEVC,” Proc. of the PDeS Conf., 2016.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-26
SLIDE 26

CGR HEVC interpolators

switching elements

profile HIGH MEDIUM LOW quality # taps energy quality # taps energy quality # taps energy luma  8/7   5   3  chroma 4 3 2

[2] F. Palumbo et al., “Runtime energy versus quality tuning in motion compensation filters for HEVC,” Proc. of the PDeS Conf., 2016.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-27
SLIDE 27

CGR HEVC interpolators

switching elements clock gating

profile HIGH MEDIUM LOW quality # taps energy quality # taps energy quality # taps energy luma  8/7   5   3  chroma 4 3 2

[2] F. Palumbo et al., “Runtime energy versus quality tuning in motion compensation filters for HEVC,” Proc. of the PDeS Conf., 2016.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-28
SLIDE 28

CGR HEVC interpolators

switching elements clock gating

  • perand

isolation

profile HIGH MEDIUM LOW quality # taps energy quality # taps energy quality # taps energy luma  8/7   5   3  chroma 4 3 2

[2] F. Palumbo et al., “Runtime energy versus quality tuning in motion compensation filters for HEVC,” Proc. of the PDeS Conf., 2016.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-29
SLIDE 29

CGR HEVC interpolators

switching elements clock gating

  • perand

isolation

profile HIGH MEDIUM LOW quality # taps energy quality # taps energy quality # taps energy luma  8/7   5   3  chroma 4 3 2

[2] F. Palumbo et al., “Runtime energy versus quality tuning in motion compensation filters for HEVC,” Proc. of the PDeS Conf., 2016.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-30
SLIDE 30

CGR HEVC interpolators

switching elements clock gating

  • perand

isolation

profile HIGH MEDIUM LOW quality # taps energy quality # taps energy quality # taps energy luma  8/7   5   3  chroma 4 3 2

[2] F. Palumbo et al., “Runtime energy versus quality tuning in motion compensation filters for HEVC,” Proc. of the PDeS Conf., 2016.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-31
SLIDE 31

CGR HEVC interpolators

switching elements clock gating

  • perand

isolation

profile HIGH MEDIUM LOW quality # taps energy quality # taps energy quality # taps energy luma  8/7   5   3  chroma 4 3 2

[2] F. Palumbo et al., “Runtime energy versus quality tuning in motion compensation filters for HEVC,” Proc. of the PDeS Conf., 2016.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-32
SLIDE 32

Outline

  • Introduction
  • CERBERO and Cyber Physical Systems
  • HEVC Codec and Software Approximate Computing
  • Approximate HEVC interpolators
  • Coarse-Grained Reconfiguration
  • From CG HEVC Interpolators to CGR HEVC Interpolators
  • Results
  • Achieved Adaptivity
  • Comparison with the State of the Art
  • Conclusions

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-33
SLIDE 33

Achieved Adaptivity

design @200 MHz Xilinx XC7Z020 LUT FF BRAM DSP Fmax [MHz] tap dP (Vivado) [mW] dE [μJ] time per block [cycles] # interpolated pixels in a fixed time legacy_luma 212 37 4 16 213 8 11 0.248 460 57957 reconf_luma (vs legacy %) 582 (+175%) 85 (+130%) 4 (+0%) 16 (+0%) 200 (-6%) 8 12 (+9%) 0.270 (+9%) 460 (+0%) 57957 (+0%) 7 11 (+0%) 0.245 (-1%) 395 (-14%) 59033 (+2%) 5 10 (-9%) 0.217 (-12%) 265 (-42%) 61191 (+6%) 3 10 (-9%) 0.211 (-15%) 135 (-71%) 63357 (+9%) legacy_chroma 163 33 2 8 217 4 9 0.053 107 14753 reconf_chroma (vs legacy %) 383 (+135%) 65 (+97%) 2 (+0%) 8 (+0%) 200 (-12%) 4 9 (+0%) 0.053 (+0%) 107 (+0%) 14753 (+0%) 3 8 (-11%) 0.045 (-13%) 73 (-32%) 15293 (+4%) 2 6 (-33%) 0.033 (-37%) 39 (-64%) 15835 (+7%)

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-34
SLIDE 34

Achieved Adaptivity

design @200 MHz Xilinx XC7Z020 LUT FF BRAM DSP Fmax [MHz] tap dP (Vivado) [mW] dE [μJ] time per block [cycles] # interpolated pixels in a fixed time legacy_luma 212 37 4 16 213 8 11 0.248 460 57957 reconf_luma (vs legacy %) 582 (+175%) 85 (+130%) 4 (+0%) 16 (+0%) 200 (-6%) 8 12 (+9%) 0.270 (+9%) 460 (+0%) 57957 (+0%) 7 11 (+0%) 0.245 (-1%) 395 (-14%) 59033 (+2%) 5 10 (-9%) 0.217 (-12%) 265 (-42%) 61191 (+6%) 3 10 (-9%) 0.211 (-15%) 135 (-71%) 63357 (+9%) legacy_chroma 163 33 2 8 217 4 9 0.053 107 14753 reconf_chroma (vs legacy %) 383 (+135%) 65 (+97%) 2 (+0%) 8 (+0%) 200 (-12%) 4 9 (+0%) 0.053 (+0%) 107 (+0%) 14753 (+0%) 3 8 (-11%) 0.045 (-13%) 73 (-32%) 15293 (+4%) 2 6 (-33%) 0.033 (-37%) 39 (-64%) 15835 (+7%)

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-35
SLIDE 35

Achieved Adaptivity

design @200 MHz Xilinx XC7Z020 LUT FF BRAM DSP Fmax [MHz] tap dP (Vivado) [mW] dE [μJ] time per block [cycles] # interpolated pixels in a fixed time legacy_luma 212 37 4 16 213 8 11 0.248 460 57957 reconf_luma (vs legacy %) 582 (+175%) 85 (+130%) 4 (+0%) 16 (+0%) 200 (-6%) 8 12 (+9%) 0.270 (+9%) 460 (+0%) 57957 (+0%) 7 11 (+0%) 0.245 (-1%) 395 (-14%) 59033 (+2%) 5 10 (-9%) 0.217 (-12%) 265 (-42%) 61191 (+6%) 3 10 (-9%) 0.211 (-15%) 135 (-71%) 63357 (+9%) legacy_chroma 163 33 2 8 217 4 9 0.053 107 14753 reconf_chroma (vs legacy %) 383 (+135%) 65 (+97%) 2 (+0%) 8 (+0%) 200 (-12%) 4 9 (+0%) 0.053 (+0%) 107 (+0%) 14753 (+0%) 3 8 (-11%) 0.045 (-13%) 73 (-32%) 15293 (+4%) 2 6 (-33%) 0.033 (-37%) 39 (-64%) 15835 (+7%)

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-36
SLIDE 36

Achieved Adaptivity

design @200 MHz Xilinx XC7Z020 LUT FF BRAM DSP Fmax [MHz] tap dP (Vivado) [mW] dE [μJ] time per block [cycles] # interpolated pixels in a fixed time legacy_luma 212 37 4 16 213 8 11 0.248 460 57957 reconf_luma (vs legacy %) 582 (+175%) 85 (+130%) 4 (+0%) 16 (+0%) 200 (-6%) 8 12 (+9%) 0.270 (+9%) 460 (+0%) 57957 (+0%) 7 11 (+0%) 0.245 (-1%) 395 (-14%) 59033 (+2%) 5 10 (-9%) 0.217 (-12%) 265 (-42%) 61191 (+6%) 3 10 (-9%) 0.211 (-15%) 135 (-71%) 63357 (+9%) legacy_chroma 163 33 2 8 217 4 9 0.053 107 14753 reconf_chroma (vs legacy %) 383 (+135%) 65 (+97%) 2 (+0%) 8 (+0%) 200 (-12%) 4 9 (+0%) 0.053 (+0%) 107 (+0%) 14753 (+0%) 3 8 (-11%) 0.045 (-13%) 73 (-32%) 15293 (+4%) 2 6 (-33%) 0.033 (-37%) 39 (-64%) 15835 (+7%)

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-37
SLIDE 37

Achieved Adaptivity

design @200 MHz Xilinx XC7Z020 LUT FF BRAM DSP Fmax [MHz] tap dP (Vivado) [mW] dE [μJ] time per block [cycles] # interpolated pixels in a fixed time legacy_luma 212 37 4 16 213 8 11 0.248 460 57957 reconf_luma (vs legacy %) 582 (+175%) 85 (+130%) 4 (+0%) 16 (+0%) 200 (-6%) 8 12 (+9%) 0.270 (+9%) 460 (+0%) 57957 (+0%) 7 11 (+0%) 0.245 (-1%) 395 (-14%) 59033 (+2%) 5 10 (-9%) 0.217 (-12%) 265 (-42%) 61191 (+6%) 3 10 (-9%) 0.211 (-15%) 135 (-71%) 63357 (+9%) legacy_chroma 163 33 2 8 217 4 9 0.053 107 14753 reconf_chroma (vs legacy %) 383 (+135%) 65 (+97%) 2 (+0%) 8 (+0%) 200 (-12%) 4 9 (+0%) 0.053 (+0%) 107 (+0%) 14753 (+0%) 3 8 (-11%) 0.045 (-13%) 73 (-32%) 15293 (+4%) 2 6 (-33%) 0.033 (-37%) 39 (-64%) 15835 (+7%)

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-38
SLIDE 38

Comparison with the State of the Art

  • 55.09%
  • 76.23 %
  • 92-91%
  • 92.91%
  • 95.80%
  • 95.80%

200 400 600 800 1000 1200 1400 1600

[3] - 65 nm [4] - 40 nm [5] - 65 nm this work - 65 nm this work mF - 65 nm this work - 28 nm this work mF - 28 nm

dE per pixel [pJ] High (8/7 tap luma, 4 tap chroma) Medium (5 tap luma, 3 tap chroma) Low (3 tap luma, 2 tap chroma)

this work: 4 luma and 3 chroma parallel filters running at 150 MHz  UHD@60 fps

[3] V. Afonso et al., “Low cost and high throughput FME interpolation for the HEVC emerging video coding standard,” Proc. of the IEEE LASCAS Conf., 2013. [4] E. Kalali et al., “A reconfigurable HEVC sub pixel interpolation hardware,” Proc. of the IEEE ICCE Conf., 2013. [5] C. M. Diniz et al., “A reconfigurable hardware architecture for fractional pixel interpolation in high efficiency video coding,” IEEE Comput.-Aided Des. Integr. Circuits Syst., vol. 34, no. 2, pp. 238–251, 2015.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-39
SLIDE 39

Comparison with the State of the Art

  • 55.09%
  • 76.23 %
  • 92-91%
  • 92.91%
  • 95.80%
  • 95.80%

200 400 600 800 1000 1200 1400 1600

[3] - 65 nm [4] - 40 nm [5] - 65 nm this work - 65 nm this work mF - 65 nm this work - 28 nm this work mF - 28 nm

dE per pixel [pJ] High (8/7 tap luma, 4 tap chroma) Medium (5 tap luma, 3 tap chroma) Low (3 tap luma, 2 tap chroma)

this work: 4 luma and 3 chroma parallel filters running at 150 MHz  UHD@60 fps

[3] V. Afonso et al., “Low cost and high throughput FME interpolation for the HEVC emerging video coding standard,” Proc. of the IEEE LASCAS Conf., 2013. [4] E. Kalali et al., “A reconfigurable HEVC sub pixel interpolation hardware,” Proc. of the IEEE ICCE Conf., 2013. [5] C. M. Diniz et al., “A reconfigurable hardware architecture for fractional pixel interpolation in high efficiency video coding,” IEEE Comput.-Aided Des. Integr. Circuits Syst., vol. 34, no. 2, pp. 238–251, 2015.

  • 6%
  • 17%
  • 14%
  • 22%
  • 23%
  • 29%
  • 24%
  • 36%

50 100 150

this work - 65 nm this work mF

  • 65 nm

this work - 28 nm this work mF

  • 28 nm

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-40
SLIDE 40

Outline

  • Introduction
  • CERBERO and Cyber Physical Systems
  • HEVC Codec and Software Approximate Computing
  • Approximate HEVC interpolators
  • Coarse-Grained Reconfiguration
  • From CG HEVC Interpolators to CGR HEVC Interpolators
  • Results
  • Achieved Adaptivity
  • Comparison with the State of the Art
  • Conclusions

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-41
SLIDE 41

Conclusions

  • CERBERO: continuous design environment for Cyber-Physical Systems

(CPS)

  • run-time F/NF requirements driven adaptivity
  • HEVC power/energy hungry, latest video coding standard
  • Approximate computing on HEVC interpolators demonstrated to provide

energy versus quality trade-off

  • Coarse-Grained Reconfiguration (CGR) allows approximation of HEVC

interpolators in hardware

  • our solution challenges outperforms state of the art solutions

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-42
SLIDE 42

Future Directions

At the CPS physical level application specific efficient accelerators capable of providing flexibility and dynamic adaptation to changeable F/NF requirements.

adaptive accelerator

NF requirements: energy budget locally remotely F requirements: video quality demand

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-43
SLIDE 43

Acknowledgements

The CERBERO project has received funding from the EU Commission’s H2020 Programme under grant agreement No 732105.

Tiziana Fanni, University of Cagliari - Dynamic Trade-Off Management for CPS

slide-44
SLIDE 44

2nd Italian Workshop on Embedded Systems

IWES 2017

7-8 September 2017 – Rome (IT) Tiziana Fanni tiziana.fanni@diee.unica.it

Dynamic Trade-Off Management for CPS