### Table 3: The static reference counts of Time-Reads generated by di erent compiler algorithms. The data in parentheses represent the results of a scalar version of each algorithm. For ALG3, the average gure shows the averaged statistics only from 3 benchmarks: FLO52, MDG, and QCD.

2000

"... In PAGE 24: ...ection 4.2.). Benchmarks and static reference statistics Table3 shows the number of potentially stale data references (Time-Reads) marked at compile time using the 3 compiler algorithms. For comparison, we have implemented both array and scalar data- ow analysis versions of each algorithm and show their corresponding results.... In PAGE 24: ... In our compiler implementation, all the variables are treated as shared variables. Therefore, the number of the potentially stale data references are over-estimated in Table3 . However, during simulation, we issue normal memory read operations for all private read references.... ..."

Cited by 2

### Table 3: The static reference counts of Time-Reads generated by di erent compiler algorithms. The data in parentheses represent the results of a scalar version of each algorithm. For ALG3, the average gure shows the averaged statistics only from 3 benchmarks: FLO52, MDG, and QCD.

"... In PAGE 24: ...ection 4.2.). Benchmarks and static reference statistics Table3 shows the number of potentially stale data references (Time-Reads) marked at compile time using the 3 compiler algorithms. For comparison, we have implemented both array and scalar data- ow analysis versions of each algorithm and show their corresponding results.... In PAGE 24: ... In our compiler implementation, all the variables are treated as shared variables. Therefore, the number of the potentially stale data references are over-estimated in Table3 . However, during simulation, we issue normal memory read operations for all private read references.... ..."

### Table 4.11: The compilation time (in seconds) of the algorithm in various phases on a notebook computer.

### Table II confirms the increased efficiency of CAM with constraints. Finally, as an illustrative example, we present the performance of CAM and CAMC on a real-world global optimization problem, taken from the field of computational chemistry. The problem consists in finding a stable conformation of a small protein (met-enkaephalin), which minimizes its potential energy as a function of dihedral angles [15]. This is a benchmark problem for testing optimization algorithms in chemistry. Earlier we reported our numerical experiments with CAM on this challenging problem [14]. As the global variables we took 10 backbone dihedral angles, with all otheranglestreated aslocal variables.The objectivefunction has in order of 1011 local minimizers. The values of the potential energy were computed using ECEPP/3 software package. The detailed configuration and system setup are described in Ref. [14].

2003

### Table II. Run Times of Benchmarks, as a Function of the Register Allocation Algorithm Used when Compiling Them

1999

Cited by 87

### Table 4.2: The compilation time (in seconds) of my algorithm in various phases on a simulator.

### Table 1. Fmax (performance) and compile time results for full and partial TDC algorithm

2001

"... In PAGE 8: ... To model this, we pre-process each design with a random assignment of pins. Table1 shows the overall results. On average, the full implementation of our algorithm shows a 38.... ..."

Cited by 6

### Table 1: Executable sizes (in bytes) and compilation time (in seconds) for di erent algorithms variants.

"... In PAGE 5: ... The parallel mode features two kinds of parallel quicksort, unbalanced and balanced, as well as a the multiway mergesort. The quicksort variants work completely in place, and are quite small in executable size (see Table1 ). On the downside, they do not support stable sort, and they require nested parallel regions.... ..."

### Table 4.7: The compilation time (in seconds) of the multi-region algorithm in various phases on a simulator.

### Table 2. Genetic algorithm (optimized power)/(initial power) for fixed compile time

2000

"... In PAGE 13: ... For the largest graph, the fixed simulation time was not long enough to make much improvement, but the best result occurred for , where the simulations are less frequent. Table2 summarizes the power reduction for the genetic algorithm with and without using the period graph, with a fixed compile time of one hour. 9 Conclusion This paper has explored a period graph model that enables efficient voltage scaling optimization for self-timed implementations of iterative applications.... ..."

Cited by 2