Skip to content

Latest commit

 

History

History
141 lines (96 loc) · 4.26 KB

benchmarks.md

File metadata and controls

141 lines (96 loc) · 4.26 KB

Graphs show speedup over single-threaded execution (higher is better)

Libraries

library("tidyverse")

## ── Attaching packages ─────────────────────────────────────── tidyverse 1.3.1 ──

## ✔ ggplot2 3.3.6     ✔ purrr   0.3.4
## ✔ tibble  3.1.8     ✔ dplyr   1.0.9
## ✔ tidyr   1.2.0     ✔ stringr 1.4.0
## ✔ readr   2.1.2     ✔ forcats 0.5.1

## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag()    masks stats::lag()

library("Rcpp")
library("RcppParallel")

## 
## Attaching package: 'RcppParallel'

## The following object is masked from 'package:Rcpp':
## 
##     LdFlags

library("RcppThread")

## 
## Attaching package: 'RcppThread'

## The following object is masked from 'package:RcppParallel':
## 
##     LdFlags

## The following object is masked from 'package:Rcpp':
## 
##     LdFlags

library("ggthemes")

Utilities

Rcpp::sourceCpp(here::here("benchmarks/benchmarks.cpp"))
wait_for <- 5

plot_df <- function(df, title = NULL) {
  p <- df %>%
    pivot_longer(-n, "call") %>%
    ggplot(aes(n, value, color = call)) +
    geom_line(size = 0.6) +
    expand_limits(y = 0) +
    labs(color = "", linetype = "") +
    naglr::theme_naglr(plot_title_size = 12) +
    theme(legend.margin = margin(1, 1, 1, 1)) +
    theme(legend.position = "bottom") +
    scale_x_log10() +
    ylab("speedup") +
    labs(title = title)
  print(p)
}

Benchmarks

empty threads

ns <- 10^(2:6)
res <- benchEmpty(rev(ns), wait_for)
df <- cbind(data.frame(n = rev(ns)), as.data.frame(res[, -1]))
plot_df(df, "empty jobs")


ggsave(here::here("benchmarks/benchEmptyThread.pdf"), width = 7.5, height = 3)

compute 1000 square roots per thread

ns <- 10^(2:6)
res <- benchSqrt(rev(ns), wait_for)
df <- cbind(data.frame(n = rev(ns)), as.data.frame(res[, -1]))
plot_df(df, "1000x sqrt")


ggsave(here::here("benchmarks/benchSqrt.pdf"), width = 7.5, height = 3)

compute and write 1000 square roots per thread

ns <- 10^(2:6)
res <- benchSqrtWrite(rev(ns), wait_for)
df <- cbind(data.frame(n = rev(ns)), as.data.frame(res[, -1]))
plot_df(df, "1000x sqrt modify inplace")


ggsave(here::here("benchmarks/benchSqrtWrite.pdf"), width = 7.5, height = 3)

kernel density estimation (d = 10)

ns <- 10^(2:5)
res <- benchKDE(rev(ns), 100, wait_for)
df <- cbind(data.frame(n = rev(ns)), as.data.frame(res[, -1]))
plot_df(df, "kernel density d = 10")


ggsave(here::here("benchmarks/benchKDE-10.pdf"), width = 7.5, height = 3)

kernel density estimation (d = 100)

ns <- 10^(2:5)
res <- benchKDE(rev(ns), 100, wait_for)
df <- cbind(data.frame(n = rev(ns)), as.data.frame(res[, -1]))
plot_df(df, "kernel density d = 100")


ggsave(here::here("benchmarks/benchKDE-100.pdf"), width = 7.5, height = 3)

Kendall matrix (d = 10, imbalanced)

ns <- 10^(2:4)
res <- benchKendall(rev(ns), 10, wait_for)
df <- cbind(data.frame(n = rev(ns)), as.data.frame(res[, -1]))
plot_df(df, "Kendall matrix (unbalanced) d = 10")


ggsave(here::here("benchmarks/benchKendall-10.pdf"), width = 7.5, height = 3)

Kendall matrix (d = 100, imbalanced)

ns <- 10^(2:4)
res <- benchKendall(rev(ns), 100, wait_for)
df <- cbind(data.frame(n = rev(ns)), as.data.frame(res[, -1]))
plot_df(df, "Kendall matrix (unbalanced) d = 100")


ggsave(here::here("benchmarks/benchKendall-100.pdf"), width = 7.5, height = 3)