Home Placement Contest (pdf txt) Final Program Hotel&Travel Call-for-papers (pdf doc txt)


ISPD & IEEE/CEDA Placement Contest Announcements

Contest benchmarks and Density target

Sample benchmarks and Script
  • Density target checking perl script: gzipped check_density_target perl script
    • Usage: check_density_target.pl node-file solution-pl-file scl-file density-target
    • example: %check_density_target.pl adaptec1.nodes adaptec1-yourplacer.pl adaptec1.scl 0.5
    • Simply superimpose a bin grid (120 x 120, i.e., 10 circuit row heights) over placement solution to calculate bin utilization. The bin utilization is defined as the movable cell usage divided by the free space (i.e., bin area - fixed usage in bin). The script reports the number of violation bins, the average overflow etc.
    • Pay extra attention to "Scaled Overflow per bin" value in the last line. This is the one we're going to use for score.
    • For only newblue1, you need a special check_density_target script and an input file: gzipped check_density_target_for_nw1 perl script and transformed.list
  • Legality checking perl script: gzipped check_legality perl script
    • Usage: check_legality.pl node-file input-pl-file solution-pl-file input-scl window-row-num
    • Same script from ISPD2005 contest
  • HPWL measuring perl script: gzipped hpwl perl script
    • Usage: hpwl.pl node-file input-pl-file solution-pl-file input-net-file
    • Same script from ISPD2005 contest
  • Placement runs with target density 60% on ISPD2005 benchmark suite
  • Inflated ISPD2005 benchmark suite
  • How to score you placement solution
    • Your placement solution must be legal.
    • The scaled HPWL score will be YOUR_PLACER_HPWL * (1 + scaled_overflow_per_bin/100). In other words, your scaled_overflow_per_bin from check_density_target.pl is the HPWL penalty in percentage form. Of course, if you have no violation, you'll get no penaly in HPWL.
    • We're still figuring out how to factor in CPU time. Stay tuned for this.

Frequently Asked Questions
  • How to calculate the design density (utilization): The design density (or utilization) is a design characteristics and it is different from the density target density which is a constraint for your placer. The design density is defined as follows:
    • design density = sum of movable object areas - available free space
    • available free space = placement area (sum of row areas) - sum of fixed object areas
  • Density Target value will be always higher than design density (utilization). For example, when the design density is 80%, the density target will be some value between 80% to 100% because it doesn't make sense to force placer to use 70% density target.
  • How the check_density_target.pl script calculate the bin usage: The script visits each object one by one (either fixed or movable) and identifies a set of bins which overlap with the object under consideration. For each overlapping bin, only the intersecting area between the bin and the object is added to bin's usage field, either in MUSAGE (movalbe usage) or FUSAGE (fixed usage), based on the type of the object. Suppose we have a object with area 100, and it overlaps with four bins, bin A, B, C, D with overlapping area 50, 20, 15 and 15 respectively. For this object, only 50 units of area is added to bin A's usage, 20 to bin B's etc.
  • No rotation/mirroring/flipping cells will be allowed. Pretty much the similar rules from last year.
  • Machine configuration for real contest: CPU: Two AMD Opteron(tm) Processor 252 with 1024 KB cache each. 2592.636 MHz Memory: 7980976 kB (8G) OS: Linux version 2.4.21-37.ELsmp (bhcompile@dolly.build.redhat.com) (gcc version 3.2.3 20030502 (Red Hat Linux 3.2.3-53))
  • Quality Metric = HPWL(1 + CPU_time_factor + Density_Target_penalty_factor) where CPU_time_factor = 0.04*log_2(your_cpu/median_cpu) and Density_target_penalty_factor = scaled_overflow_per_bin/100. CPU_time_factor will range from -0.1 to 0.1 (i.e. max 10% HPWL advantage/disadvantage), and scaled_overflow_per_bin is the value from check_density_target.pl script.

Participating Teams
  • APlace3: A.B. Kahng, Q. Wang (UCSD)
  • Capo: J. Roy, D. Papa, A. Ng, I. Markov (U of Michigan)
  • Dark Horse: A. R. Agnihotri, S. Ono, P. Madden (SUNY Binghamton)
  • DPlace: T. Luo, D. Pan (U of Texas)
  • Dragon: T. Taghavi, X. Yang, B.-K. Choi, M. Wang, M. Sarrafzadeh (UCLA)
  • FastPlace: N. Viswanathan, M. Pan, C. Chu (Iowa State U)
  • KraftWerk: P. Spindler, F. Johannes (Technical University of Munich)
  • mFAR: B. Hu, M. Marek-Sadowska (UCSB)
  • mPL: T. Chan, J. Cong, J. Shinnerl, K. Sze, M. Xie (UCLA)
  • NTUPlace: T.-C. Chen, Z.-W. Jiang, T.-C. Hsu, H.-C. Chen, Y.-W. Chang (National Taiwan University)

Important Dates
  • January 22, 2006: Registration deadline
  • January 30, 2006: A few sample benchmarks and perl scripts will be available
  • February 6, 2006: Test placer executable/run script submission deadline
  • March 27, 2006: Contest placer execuable/run script submission deadline

For any question or comments, please contat Gi-Joon Nam (gnam@us.ibm.com)