### Table 15.1. Characterization of ASTB applications. Video pixel processing combines many video-improvement algorithms in a chain. We show typical numbers; however, they depend on the number of (temporal) processing stages.

2004

Cited by 10

### Table 3: Performance on Image and Video Algorithms

"... In PAGE 4: ... In addition to the DSP kernels, the Frio core was designed to allow efficient implementations of typical image and video processing applications such as are found in next-generation wireless applications. Table3 summarizes the performance of the Frio core on the kernels of typical image and video processing benchmarks. V.... ..."

### Table 1. The set of 21 video algorithms.

1999

"... In PAGE 5: ... Here, we focus on the computational part. The set of algorithms we use is a collection of video processing algorithms for TV applications, that all have been implemented on VSPs, see Table1 . In this table the processing requirements of the algorithms are expressed as the number of operations per sample.... In PAGE 6: ...able 1. The set of 21 video algorithms. that are executed by the PEs. So, we first have to determine these functions, for different cluster- sizes, for the algorithms given in Table1 . We can use the following two approaches.... In PAGE 8: ...raphs. Therefore, the constraints have to be verified for every possible step. The local search algorithm used is called Variable Depth Search [15],[16],[21], and is a hill- climbing variant of local search. For each of the 21 algorithms of Table1 we have made seventeen clustered graphs with different cluster-sizes. Table 3 shows the number of clusters found for each cluster-size, for the entire set of algorithms.... In PAGE 11: ...5 and 15.5) results in the most efficient instance of the architecture template, in terms of silicon area, to support the set of algorithms of Table1 . Compared to the instance with the largest area (cluster-size 80, number of... ..."

Cited by 3

### Table 1. The set of 21 video algorithms.

1999

"... In PAGE 5: ... Here, we focus on the computational part. The set of algorithms we use is a collection of video processing algorithms for TV applications, that all have been implemented on VSPs, see Table1 . In this table the processing requirements of the algorithms are expressed as the number of operations per sample.... In PAGE 6: ...able 1. The set of 21 video algorithms. that are executed by the PEs. So, we first have to determine these functions, for different cluster- sizes, for the algorithms given in Table1 . We can use the following two approaches.... In PAGE 8: ... The local search algorithm used is called Variable Depth Search [15],[16],[21], and is a hill- climbing variant of local search. For each of the 21 algorithms of Table1 we have made seventeen clustered graphs with different cluster-sizes. Table 3 shows the number of clusters found for each cluster-size, for the entire set of algorithms.... In PAGE 11: ...5 and 15.5) results in the most efficient instance of the architecture template, in terms of silicon area, to support the set of algorithms of Table1 . Compared to the instance with the largest area (cluster-size 80, number of PEs 40, grain- size 10), we can achieve a gain of two.... ..."

Cited by 3

### Table 3: Video Textures, Consumer Preloaded, Time to Complete Processing (seconds)

in Towards

"... In PAGE 14: ...Figure 8: Tracker Pipeline: See Figure 9 for Outer DPS application in the absence of any communication. The results are summarized in Table3 . As might be expected, for a xed number of address spaces, adding more consumer threads increases the application performance asymptotically.... In PAGE 14: ... Ideally, the processing should scale linearly as the number of address spaces or as the number of threads per address space increases. However, the results in Table3 show that this is not the case. 5.... ..."

### Table 2: Performance on DSP Algorithms

"... In PAGE 4: ... This reduces the need for loop unrolling to only one iteration, improves code size, and allows simple assembly programming. Table2 summarizes the performance of the Frio core on the kernels of typical DSP benchmarks. In addition to the DSP kernels, the Frio core was designed to allow efficient implementations of typical image and video processing applications such as are found in next-generation wireless applications.... ..."

### Table 1. The used set of 21 video algorithms and their computational requirements.

1999

"... In PAGE 5: ... Here, we focus on the computational part. The set of algorithms we use is a collection of video processing algorithms for TV applications, which all have been implemented on VSPs; see Table1 . In this table the computational requirements of the algorithms are expressed as the average number of operations per sample.... In PAGE 7: ... The local search algorithm used is called variable depth search [1, 6, 11], and is a hill-climbing variant of local search. For each of the 21 algorithms of Table1 we have made clustered graphs with seventeen different cluster- sizes. Table 3 shows the number of clusters found for each cluster-size, for the entire set of algorithms.... In PAGE 10: ...5 and 15.5) results in the most efficient instances of the architecture template, in terms of silicon area, to support the set of algorithms of Table1 . Compared to the instance with the largest area (cluster-size 80, number of PEs 40, grain-size 10), we can achieve a gain factor of two.... ..."

Cited by 3

### Table 1. The used set of 21 video algorithms and their computational requirements.

in E.a. De Kock

1999

"... In PAGE 5: ... Here, we focus on the computational part. The set of algorithms we use is a collection of video processing algorithms for TV applications, which all have been implemented on VSPs; see Table1 . In this table the computational requirementsof the algorithms are expressed as the average number of operations per sample.... In PAGE 7: ... The local search algorithm used is called variable depth search [1, 6, 11], and is a hill-climbing variant of local search. For each of the 21 algorithms of Table1 we have made clustered graphs with seventeen different cluster- sizes. Table 3 shows the number of clusters found for each cluster-size, for the entire set of algorithms.... In PAGE 10: ...5 and 15.5) results in the most efficient instances of the architecture template, in terms of silicon area, to support the set of algorithms of Table1 . Compared to the instance with the largest area (cluster-size 80, number of PEs 40, grain-size #19 10), we can achieve a gain factor of two.... ..."

Cited by 3