]>

Table 4: HPWL improvement due to flow-based whitespace redistribution on the ICCAD 2004 IBM-MSwPins mixed-size benchmarks [27]. On average, the flows are able to reallocate whitespace and improve HPWL by nearly 3% while scaling well with increasing quantities of movable objects.

Benchmark Movables Runtime (s) Improvement ibm01 12506 11 2.41% ibm02 19342 23 2.60% ibm03 22853 27 3.81% ibm04 27220 38 2.96% ibm05 28146 46 2.27% ibm06 32332 44 2.87% ibm07 45639 69 3.17% ibm08 51023 64 2.82% ibm09 53110 69 4.23% ibm10 68685 129 1.91% ibm11 70152 102 4.42% ibm12 70439 135 1.87% ibm13 83709 127 3.83% ibm14 147088 262 2.64% ibm15 161187 342 3.67% ibm16 182980 404 2.62% ibm17 184752 446 2.54% ibm18 210341 338 2.47% Average 2.86%