"... In PAGE 7: ... The global power law is omitted as its overhead is negligible. According to Table1 , minskew is slightly faster to construct than power. However, as discussed in Section 2.... ..."

"... In PAGE 29: ...4 Overhead of pre-processing and estimation Apart from the global approach, the other methods require pre-computing necessary structures prior to performing estimation. Table2 evaluates the pre-processing time for all solutions and methods, where the memory size is equivalent to that of 100 patches for P-Patch (grid resolution set to 64). Minskew is the fastest to build, which, however, does not justify its poor estimation accuracy.... ..."

"... In PAGE 5: ... 4.3 Time Cost Table5 shows the construction time of CBBN and BMN classifiers for five se- lected data sets. The results show that CBBN is more computationally expensive than BMN.... ..."

"... In PAGE 14: ... The tables are given at the end of this report. The wall clock times are measured for the matrix construction ( Table1 and 4), preconditioner construction (Table 2 and 5), and preconditioner construction combined with linear system solution (Table 3 and 6). These measurements are given for both the momentum equations... In PAGE 16: ... One reason for this is the overhead of computation due to the auxiliary grid cells. Another reason is a small vector length, which leads to low Mega op rates on the RISC computers of the Cray T3D (compare Table1 0). The e ciency loss for the solution algorithm is more severe than that for the matrix construction.... In PAGE 18: ....3.2. The predicted rates ( Table1 7) show a good correspondence with the observed rates (Table 9). We also use the values of tl and tb from the inner product to predict the rate of the matrix vector product.... In PAGE 19: ... The results are given in Table 24. There is a good correspondence between the predictions (Table 24) and the measurements ( Table1 6). So the described model can be used to predict the e ciency of the proposed parallel method also for larger grid sizes and/or a larger number of processors.... In PAGE 27: ...1 4270.2 Table1 0: Inner product performance in Mega ops per second without communication... In PAGE 28: ...5 200.7 Table1 1: Momentum equation matrix construction; total time per variable in seconds #PE apos;s 1 2 4 8 16 32 64 128 256 32 8 24.15 44.... In PAGE 28: ...04 52.28 Table1 2: Momentum preconditioner construction; total time per variable in seconds #PE apos;s 1 2 4 8 16 32 64 128 256 32 8 247.5 576.... In PAGE 28: ...8 808.1 Table1 3: Momentum preconditioner construction and solution; total time per variable in seconds... In PAGE 29: ...8 211.2 Table1 4: Pressure equation matrix construction; total time per cell in seconds #PE apos;s 1 2 4 8 16 32 64 128 256 32 8 36.59 52.... In PAGE 29: ...29 82.01 Table1 5: Pressure preconditioner construction; total time per variable in seconds #PE apos;s 1 2 4 8 16 32 64 128 256 32 8 295.3 696.... In PAGE 29: ...3 804.2 Table1 6: Pressure preconditioner construction and solution; total time per cell in seconds... In PAGE 30: ...9 4231.5 Table1 7: Estimated inner product Mega op rates with communication #PE apos;s 1 2 4 8 16 32 64 128 256 32 8 9.8 19.... In PAGE 30: ...3 2537.2 Table1 8: Estimated matrix vector product Mega op rates (fm = 10, no overlap) #PE apos;s 1 2 4 8 16 32 64 128 256 32 8 4.9 9.... In PAGE 30: ...6 1274.3 Table1 9: Estimated matrix vector product Mega op rates (fm = 5, no overlap)... ..."

"... In PAGE 2: ... Thusn2c a binary opn2d eration has the time complexityOn28jFjn01jGjn29. The inn0duence of the CT is shown in Table1 . One can observe that only a minority of the ISCASn2785 n5b3n5d benchmarks can be constructed as BDDs in approprin2d ate time without a CT.... ..."

"... In PAGE 5: ... It has been run with a number of small scenes, up to 184 polygons with very encouraging results. Table1 shows a summary of the timings for the program running on a SUN SparcStation 20,... ..."

"... In PAGE 7: ... The tests shown are referred to combined color and shape retrieval. Table1 shows that building a single tree for multiple features requires less time than building a tree for every feature. Considering the queries, combined features (Table 3) are faster than separated features (Table 2) because in the... ..."

"... In PAGE 5: ... Model compression Although PSC gvspl are more general than PM vsplit transformations, they offer a surprisingly concise representation of ^ M. Table1 lists the average number of bits re- quired to encode each field of the gvspl records. Using arithmetic coding [30], the vertex id field faig requires log2i bits, and the boolean parameter midpi requires 0.... In PAGE 5: ... To account for this redistribution of area, we identify the principal simplex from which s receives its area by specifying its index in P01(star(faig)) Ki. The column labeled in Table1 sums the bits of each field of the gvspl records. Multiplying by the number n of vertices in ^ M gives the total number of bits for the PSC representation of the model (e.... In PAGE 6: ... This speeds up the optimization, improves modelcompression, andallowsustointroducenon-quadraticenergy terms like E area. 5 RESULTS Table1 gives quantitative results for the examples in the figures and in the video. Simplification times for our prototype are measured on an SGI Indigo2 Extreme (150MHz R4400).... ..."