Archives

  • 2018-07
  • 2018-10
  • 2018-11
  • 2019-04
  • 2019-05
  • 2019-06
  • 2019-07
  • 2019-08
  • 2019-09
  • 2019-10
  • 2019-11
  • 2019-12
  • 2020-01
  • 2020-02
  • 2020-03
  • 2020-04
  • 2020-05
  • 2020-06
  • 2020-07
  • 2020-08
  • 2020-09
  • 2020-10
  • 2020-11
  • 2020-12
  • 2021-01
  • 2021-02
  • 2021-03
  • 2021-04
  • 2021-05
  • 2021-06
  • 2021-07
  • 2021-08
  • 2021-09
  • 2021-10
  • 2021-11
  • 2021-12
  • 2022-01
  • 2022-02
  • 2022-03
  • 2022-04
  • 2022-05
  • 2022-06
  • 2022-07
  • 2022-08
  • 2022-09
  • 2022-10
  • 2022-11
  • 2022-12
  • 2023-01
  • 2023-02
  • 2023-03
  • 2023-04
  • 2023-05
  • 2023-06
  • 2023-07
  • 2023-08
  • 2023-09
  • 2023-10
  • 2023-11
  • 2023-12
  • 2024-01
  • 2024-02
  • 2024-03
  • 2024-04

    • The plots in Figs and tend

    2018-11-15

    The plots in Figs. 2 and 3 tend to cup-like shapes that demonstrate that the cache block sizes may have a striking influence on the performance. As shown in the figures, the energy metabolism of the processing time may be as dramatic as around 50% with respect to the poorest case (see top left panels). Therefore, tuning of the cache block sizes is paramount for an optimal execution. In case not all cores are to be used, there is a wide range of values that turns out to be optimal for the LLC block size, as pointed by the flatter performance curves (see right column in Figs. 2 and 3). If all cores are used, the optimal values concentrate in a narrower range. We have found that the formula LLC_size/(1.6×T), with LLC_size denoting the size of the LLC cache memory and T referring to the number of running threads, provides a good generic value for the LLC block size. As far as the L1 block size is concerned, the plots clearly show that the rows of the slices have to be broken into smaller rows that should take up to half the L1 size. Values of 8kB and 16kB turn out to be optimal whereas lower values (4kB) do not allow full advantage of the cache memory. If the block is set to the whole L1 cache memory (32kB), the effect may still be beneficial for large problem sizes (4K dataset, Fig. 3), when compared to the behaviour of the native row size. However, it may be negligible or unfavourable for smaller problem sizes (2K dataset, Fig. 2), due to the penalty derived from the increased number of iterations associated with the corresponding loop of the algorithm [1]. In conclusion, and based on these results, the formulas for quasi-optimal values of the LLC and L1 block sizes in AVX-vectorized and multithreaded tomographic reconstruction are given by LLC_size/(1.6×T) and L1_size/2, respectively, with LLC_size and L1_size denoting the size of the cache memory at that level and T referring to the number of running threads.
    Experimental design, materials and methods The data are from a 1995 telephone survey conducted by the East Carolina University Survey Research Laboratory. The survey used a random digit dialing sampling scheme. The sample was purchased from Survey Sampling, Inc. and interviews were computer assisted. Of the households that were contacted, 1077 respondents provided data for an overall response rate of 75% [10]. Revealed preference and stated preference (i.e., contingent behavior) outdoor recreation participation was next elicited with the question: “Now I would like to ask you about any outdoor recreational activities you may have done on the Pamlico Sound. By recreational activities, I mean fishing, hunting, swimming, boating, skiing, windsurfing, birdwatching, camping, and so on. Did you participate in any recreational activities on or near the Pamlico Sound during the past 12 months?” Respondents who did participate (20%) were then asked “About how many trips did you take during the past 12 months?” and “About how many trips do you think you will take during the next 12 months?” Of those who participate, the average number of trips last year is 10 and 12 for next year. Respondents who did not participate were asked “Do you plan to participate in any recreational activities on or near the Pamlico Sound during the next 12 months?” and “About how many trips do you think you will take during the next 12 months?” Of those responding yes (n=49), the average number of trips is 3. Respondents in the Pamlico Sound version were asked about any other trips cervix took: “Other than at the Pamlico Sound, did you participate in any outdoor recreational activities during the past 12 months?” and “Where did you go for these trips?” Only 4% went to the Albemarle Sound while 59% went to the ocean/beach, 14% went to the mountains, 14% went to lakes, 13% went to rivers, and 18% went to other places. These numbers sum to greater than 100% due to multiple answers. In the Albemarle–Pamlico version of the survey, if respondents had participated in recreation they were asked about current and future trips with the question “Where did you go for these trips, the Albemarle Sound (27%), the Pamlico Sound (41%), or both (32%)?” Respondents who did participate (20%) were then asked “About how many trips did you take during the past 12 months?” and “About how many trips do you think you will take during the next 12 months?” Of those who participate, the average number of trips last year is 13 and 17 for next year. All respondents were asked: “Other than at the Albemarle and Pamlico Sounds, did you participate in any outdoor recreational activities during the past 12 months?” Due to space limitations, only those respondents who had not participated in recreation on the Albemarle–Pamlico Sounds were asked where these trips took place (n=400), only 2% went to the Albemarle Sound, 64% went to the ocean/beach, 2% went to the mountains, 10% went to lakes, 9% went to rivers, and 14% went to other places. Respondents who did not participate were then asked “Do you plan to participate in any recreational activities on or near the Pamlico Sound during the next 12 months?” and “About how many trips do you think you will take during the next 12 months?” Of those responding yes (n=53), the average number of trips is 2.5.