The ISPD 2006 Placement Contest and Benchmark Suite Gi-Joon Nam, - - PowerPoint PPT Presentation

▶
the ispd 2006 placement contest and benchmark suite
SMART_READER_LITE
LIVE PREVIEW

The ISPD 2006 Placement Contest and Benchmark Suite Gi-Joon Nam, - - PowerPoint PPT Presentation

Jan 07, 2023 358 likes •593 views

The ISPD 2006 Placement Contest and Benchmark Suite Gi-Joon Nam, Charles J. Alpert, Paul G. Villarrubia IBM Corp. 1 Summary of ISPD 2005 Placement Contest 9 academic placement tools participated Good coverage of placement tools

slide-1
SLIDE 1

1

The ISPD 2006 Placement Contest and Benchmark Suite

Gi-Joon Nam, Charles J. Alpert, Paul G. Villarrubia IBM Corp.

slide-2
SLIDE 2

2

Summary of ISPD 2005 Placement Contest

  • 9 academic placement tools participated
  • Good coverage of placement tools
  • 8 new placement benchmarks were released.
  • All were derived from real industrial ASIC designs
  • Extensively being used in placement research
  • HPWL was used as sole quality metric
  • No routability estimation
  • No timing analysis
  • No runtime measurement
  • Analytic placement tools dominated
slide-3
SLIDE 3

3

A bit of Criticism

  • “The contest, however, evaluated legality and wire length,

not routability, which is a key concern for commercial placement tools”… EETimes 04/ 06/ 2005

  • Rather high free space in benchmarks (i.e., low utilization)
  • Sort of favors analytic placement algorithm
slide-4
SLIDE 4

4

ISPD 2006 Placement Contest

  • 9 teams again
  • APlace3, Capo, DPlace, Dragon, FastPlace, Kraftwerk,

mFAR, mPL6, Ntuplace

  • Provide another suite of real placement benchmarks
  • More advanced form of quality of metric
  • Legality
  • HPWL
  • Routability estimation via density target
  • Runtime
  • Contestants submit executables and administrator runs them
  • n new benchmarks
slide-5
SLIDE 5

5

Placement Solution Scoring Function

HPWL* (1 + Scaled_overflow_factor + CPU_factor)

slide-6
SLIDE 6

6

  • Impose a bin grid (10 circuit row height, width)
  • Measure the overflow
  • BOF = Σmovable_area_bin – bin_free_space* density_target
  • TOF = ΣBOF
  • Scale the total overflow TOF
  • Scaled_overflow_factor = TOF * bin_area * density_target /

Σmovable_objects_area

  • Scaled_overflow_factor2 is used

Placement Solution Scoring Function

HPWL* (1 + Scaled_overflow_factor + CPU_factor)

slide-7
SLIDE 7

7

Placement Solution Scoring Function

  • CPU_factor =

0.04* LOG2(your_CPU / median_CPU)

  • 2x slower ~ 4% penalty
  • 4x slower ~ 8% penalty
  • max 10% penalty

HPWL* (1 + Scaled_overflow_factor + CPU_factor)

slide-8
SLIDE 8

8

ISPD 2006 Benchmark Suite

4 9 .3 1 3 8 .7 8 4 9 .5 6 4 6 .4 5 2 6 .3 1 6 1 .6 6 8 3 .2 0 4 9 .9 8 Utilization % 8 0 8 0 5 0 5 0 8 0 9 0 8 0 5 0 Density Target% 7 6 .4 6 2 6 3 6 8 2 0 2 6 5 8 2 2 4 8 1 3 7 2 2 5 0 7 9 5 4 new blue7 5 9 .2 7 1 2 8 8 4 4 3 6 8 8 9 1 2 4 8 1 5 0 1 2 5 5 0 3 9 new blue6 7 4 .5 4 1 2 8 4 2 5 1 4 8 8 1 1 2 2 8 1 7 7 1 2 3 3 0 5 8 new blue5 6 5 .7 2 6 3 7 0 5 1 3 4 2 2 6 4 2 7 1 7 6 4 6 1 3 9 new blue4 8 4 .7 0 5 5 2 1 9 9 1 1 1 7 8 4 8 2 8 3 3 4 9 4 0 1 1 new blue3 8 6 .1 4 4 6 5 2 1 9 1 2 7 7 3 3 0 2 3 9 4 4 1 5 1 6 new blue2 8 5 .7 3 3 3 8 9 0 1 3 3 7 3 3 0 1 3 7 3 3 0 4 7 4 new blue1 7 8 .6 4 8 6 7 7 9 8 6 4 6 8 4 2 4 8 2 8 4 3 1 2 8 adaptec5 Density % # Nets # Fixed # Movs # Objs Nam e

slide-9
SLIDE 9

9

adaptec5

  • 843K objects
  • Density 79% , Utilization 50%
  • Density target 50%
slide-10
SLIDE 10

10

newblue1

  • 330K objects
  • Lots of large movable macros
  • Density 86% , Utilization 83%
  • Density target 80%
slide-11
SLIDE 11

11

newblue2

  • 442K objects
  • All standard cells were inflated by 2x
  • 3.7K small movable macros (a few circuit row height)
  • Density 86% , Utilization 62%
  • Density target 90%
slide-12
SLIDE 12

12

newblue3

  • 494K objects
  • Interesting floorplan
  • Density 85% , Utilization 26%
  • Density target 80%
slide-13
SLIDE 13

13

newblue4

  • 646K objects
  • Density 66% , Utilization 46%
  • Density target 50%
slide-14
SLIDE 14

14

newblue5

  • 1233K objects
  • Density 75% , Utilization 50%
  • Density target 50%
slide-15
SLIDE 15

15

newblue6

  • 1255K objects
  • Density 60% , Utilization 39%
  • Density target 80%
slide-16
SLIDE 16

16

newblue7

  • 2508K objects
  • Density 76% , Utilization 49%
  • Density target 80%
slide-17
SLIDE 17

17

Contest Results

1.32 1.23 1.04 1.03 1.06 1.03 1.04 1.00 1.00 nb6 1.28 1.35 1.76 1.13 1.21 1.13 1.00 1.02 1.04 nb5 1.39 1.36 1.33 1.24 1.16 1.11 1.05 1.04 1.03 Avg. 1.46 1.22 1.44 1.64 1.57 1.16 place9 1.25 1.14 1.36 1.77* 1.55 1.26 place8 1.05 1.35 1.37 1.02 1.22 1.82 place7 1.23 1.05 1.90 1.29 1.21 1.08 place6 1.05 1.35 1.13 1.05 1.20 1.26 place5 1.04 1.09 1.16 1.09 1.23 1.09 place4 1.07 1.03 1.16 1.07 1.00 1.02 place3 1.00 1.00 1.17 1.07 1.06 1.00 place2 1.00 1.01 1.00 1.00 1.19 1.01 place1 nb7 nb4 nb3 nb2 nb1 ad5

slide-18
SLIDE 18

18

Contest Results

1.32 1.23 1.04 1.03 1.06 1.03 1.04 1.00 1.00 nb6 1.28 1.35 1.76 1.13 1.21 1.13 1.00 1.02 1.04 nb5 1.39 1.36 1.33 1.24 1.16 1.11 1.05 1.04 1.03 Avg. 1.46 1.22 1.44 1.64 1.57 1.16 place9 1.25 1.14 1.36 1.77* 1.55 1.26 place8 1.05 1.35 1.37 1.02 1.22 1.82 place7 1.23 1.05 1.90 1.29 1.21 1.08 place6 1.05 1.35 1.13 1.05 1.20 1.26 place5 1.04 1.09 1.16 1.09 1.23 1.09 place4 1.07 1.03 1.16 1.07 1.00 1.02 ntuplace 1.00 1.00 1.17 1.07 1.06 1.00 place2 1.00 1.01 1.00 1.00 1.19 1.01 place1 nb7 nb4 nb3 nb2 nb1 ad5

slide-19
SLIDE 19

19

Contest Results

1.32 1.23 1.04 1.03 1.06 1.03 1.04 1.00 1.00 nb6 1.28 1.35 1.76 1.13 1.21 1.13 1.00 1.02 1.04 nb5 1.39 1.36 1.33 1.24 1.16 1.11 1.05 1.04 1.03 Avg. 1.46 1.22 1.44 1.64 1.57 1.16 place9 1.25 1.14 1.36 1.77* 1.55 1.26 place8 1.05 1.35 1.37 1.02 1.22 1.82 place7 1.23 1.05 1.90 1.29 1.21 1.08 place6 1.05 1.35 1.13 1.05 1.20 1.26 place5 1.04 1.09 1.16 1.09 1.23 1.09 place4 1.07 1.03 1.16 1.07 1.00 1.02 ntuplace 1.00 1.00 1.17 1.07 1.06 1.00 mPL6 1.00 1.01 1.00 1.00 1.19 1.01 place1 nb7 nb4 nb3 nb2 nb1 ad5

slide-20
SLIDE 20

20

Contest Results

1.32 1.23 1.04 1.03 1.06 1.03 1.04 1.00 1.00 nb6 1.28 1.35 1.76 1.13 1.21 1.13 1.00 1.02 1.04 nb5 1.39 1.36 1.33 1.24 1.16 1.11 1.05 1.04 1.03 Avg. 1.46 1.22 1.44 1.64 1.57 1.16 Capo 1.25 1.14 1.36 1.77* 1.55 1.26 DPlace 1.05 1.35 1.37 1.02 1.22 1.82 FastPlace 1.23 1.05 1.90 1.29 1.21 1.08 Dragon 1.05 1.35 1.13 1.05 1.20 1.26 APlace3 1.04 1.09 1.16 1.09 1.23 1.09 mFAR 1.07 1.03 1.16 1.07 1.00 1.02 ntuplace 1.00 1.00 1.17 1.07 1.06 1.00 mPL6 1.00 1.01 1.00 1.00 1.19 1.01 kraftwerk nb7 nb4 nb3 nb2 nb1 ad5

* Illegal solution with few overlaps on AMD platform, Legal solution on Intel platform

slide-21
SLIDE 21

21

Contest Results

1.38 1.34 1.18 1.33 1.10 1.11 1.02 1.03 1.09

  • Avg. WL

0.32 9.32 22.09 0.12 3.82 2.71 4.10 1.36 1.68

  • Avg. Overflow

Penalty% 2.69

  • 4.54
  • 5.62
  • 5.90

5.31

  • 0.12

1.66 1.58

  • 5.04
  • Avg. CPU

Factor% Capo DPlace FastPlace Dragon APlace3 mFAR ntuplace mPL6 kraftwerk

slide-22
SLIDE 22

22

Results: What if CPU_factor is not included… .

1.32 1.11 1.29 1.10 1.03 1.01 1.05 1.03 1.00 nb6 1.45 1.86 1.27 1.26 1.16 1.19 1.10 1.00 1.04 nb5 1.42 1.39 1.34 1.31 1.11 1.11 1.08 1.03 1.02 Avg. 1.33 1.22 1.39 1.66 1.62 1.33 DPlace 1.14 1.43 1.33 1.07 1.33 1.87 FastPlace 1.40 1.21 1.32 1.56 1.55 1.15 Capo 1.30 1.14 1.92 1.32 1.27 1.16 Dragon 1.07 1.08 1.11 1.07 1.22 1.10 mFAR 1.01 1.28 1.00 1.00 1.15 1.21 APlace3 1.08 1.05 1.03 1.05 1.24 1.06 kraftwerk 1.09 1.02 1.06 1.03 1.00 1.00 ntuplace 1.00 1.00 1.05 1.01 1.06 1.00 mPL6 nb7 nb4 nb3 nb2 nb1 ad5

slide-23
SLIDE 23

23

Conclusion

  • Total 16 new placement benchmarks
  • All derived from real ASIC designs
  • Variety of floorplans
  • 5 benchmarks with more than million objects
  • ISPD 2006 Contest
  • Indirectly address routability issue
  • Turn-around time
  • Improvements from ISPD 2005 results
  • Can we include timing analysis into this flow?