cli progress bar benchmark
Gábor Csárdi
2023-11-03
Source:vignettes/progress-benchmark.Rmd
progress-benchmark.Rmd
Introduction
We make sure that the timer is not TRUE
, by setting it
to ten hours.
library(cli)
# 10 hours
cli:::cli_tick_set(10 * 60 * 60 * 1000)
cli_tick_reset()
#> NULL
`__cli_update_due`
#> [1] FALSE
R benchmarks
The timer
fun <- function() NULL
ben_st <- bench::mark(
`__cli_update_due`,
fun(),
.Call(ccli_tick_reset),
interactive(),
check = FALSE
)
ben_st
#> # A tibble: 4 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 __cli_update_due 0 0 1059553087. 0B 0
#> 2 fun() 99ns 100ns 5086252. 0B 0
#> 3 .Call(ccli_tick_reset) 99ns 100ns 7910867. 0B 0
#> 4 interactive() 0 0 998807203. 0B 0
ben_st2 <- bench::mark(
if (`__cli_update_due`) foobar()
)
ben_st2
#> # A tibble: 1 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch> <bch:> <dbl> <bch:byt> <dbl>
#> 1 if (`__cli_update_due`) fooba… 0 0 37658699. 0B 0
cli_progress_along()
seq <- 1:100000
ta <- cli_progress_along(seq)
bench::mark(seq[[1]], ta[[1]])
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 seq[[1]] 99ns 100ns 6657415. 0B 0
#> 2 ta[[1]] 99ns 200ns 6073915. 0B 0
for
loop
This is the baseline:
f0 <- function(n = 1e5) {
x <- 0
seq <- 1:n
for (i in seq) {
x <- x + i %% 2
}
x
}
With progress bars:
fp <- function(n = 1e5) {
x <- 0
seq <- 1:n
for (i in cli_progress_along(seq)) {
x <- x + seq[[i]] %% 2
}
x
}
Overhead per iteration:
ben_taf <- bench::mark(f0(), fp())
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
ben_taf
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 f0() 15.7ms 18.4ms 52.5 25.2KB 50.6
#> 2 fp() 20.9ms 21.2ms 45.8 83.2KB 45.8
(ben_taf$median[2] - ben_taf$median[1]) / 1e5
#> [1] 28.3ns
ben_taf2 <- bench::mark(f0(1e6), fp(1e6))
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
ben_taf2
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 f0(1e+06) 181ms 186ms 4.95 0B 47.8
#> 2 fp(1e+06) 198ms 198ms 5.01 1.88KB 46.7
(ben_taf2$median[2] - ben_taf2$median[1]) / 1e6
#> [1] 11.7ns
ben_taf3 <- bench::mark(f0(1e7), fp(1e7))
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
ben_taf3
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 f0(1e+07) 1.79s 1.79s 0.560 0B 35.8
#> 2 fp(1e+07) 1.94s 1.94s 0.517 1.88KB 27.9
(ben_taf3$median[2] - ben_taf3$median[1]) / 1e7
#> [1] 14.8ns
ben_taf4 <- bench::mark(f0(1e8), fp(1e8))
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
ben_taf4
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 f0(1e+08) 17.6s 17.6s 0.0568 0B 35.7
#> 2 fp(1e+08) 19.7s 19.7s 0.0507 1.88KB 30.7
(ben_taf4$median[2] - ben_taf4$median[1]) / 1e8
#> [1] 21.1ns
Mapping with lapply()
This is the baseline:
With an index vector:
f01 <- function(n = 1e5) {
seq <- 1:n
ret <- lapply(seq_along(seq), function(i) {
seq[[i]] %% 2
})
invisible(ret)
}
With progress bars:
fp <- function(n = 1e5) {
seq <- 1:n
ret <- lapply(cli_progress_along(seq), function(i) {
seq[[i]] %% 2
})
invisible(ret)
}
Overhead per iteration:
ben_tam <- bench::mark(f0(), f01(), fp())
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
ben_tam
#> # A tibble: 3 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 f0() 90.1ms 97.7ms 7.93 781KB 13.9
#> 2 f01() 105ms 111ms 8.67 781KB 10.4
#> 3 fp() 113.5ms 128.6ms 7.20 783KB 14.4
(ben_tam$median[3] - ben_tam$median[1]) / 1e5
#> [1] 309ns
ben_tam2 <- bench::mark(f0(1e6), f01(1e6), fp(1e6))
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
ben_tam2
#> # A tibble: 3 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 f0(1e+06) 879.41ms 879.41ms 1.14 7.63MB 3.41
#> 2 f01(1e+06) 1.15s 1.15s 0.873 7.63MB 5.24
#> 3 fp(1e+06) 1.39s 1.39s 0.718 7.63MB 3.59
(ben_tam2$median[3] - ben_tam2$median[1]) / 1e6
#> [1] 513ns
(ben_tam2$median[3] - ben_tam2$median[2]) / 1e6
#> [1] 247ns
Mapping with purrr
This is the baseline:
f0 <- function(n = 1e5) {
seq <- 1:n
ret <- purrr::map(seq, function(x) {
x %% 2
})
invisible(ret)
}
With index vector:
f01 <- function(n = 1e5) {
seq <- 1:n
ret <- purrr::map(seq_along(seq), function(i) {
seq[[i]] %% 2
})
invisible(ret)
}
With progress bars:
fp <- function(n = 1e5) {
seq <- 1:n
ret <- purrr::map(cli_progress_along(seq), function(i) {
seq[[i]] %% 2
})
invisible(ret)
}
Overhead per iteration:
ben_pur <- bench::mark(f0(), f01(), fp())
ben_pur
#> # A tibble: 3 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 f0() 77.3ms 77.9ms 12.8 884KB 9.60
#> 2 f01() 92.6ms 93.2ms 10.7 781KB 10.7
#> 3 fp() 93.1ms 100.7ms 10.1 783KB 6.73
(ben_pur$median[3] - ben_pur$median[1]) / 1e5
#> [1] 228ns
(ben_pur$median[3] - ben_pur$median[2]) / 1e5
#> [1] 74.2ns
ben_pur2 <- bench::mark(f0(1e6), f01(1e6), fp(1e6))
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
ben_pur2
#> # A tibble: 3 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 f0(1e+06) 895.89ms 895.89ms 1.12 7.63MB 3.35
#> 2 f01(1e+06) 1.11s 1.11s 0.897 7.63MB 2.69
#> 3 fp(1e+06) 1.48s 1.48s 0.678 7.63MB 2.71
(ben_pur2$median[3] - ben_pur2$median[1]) / 1e6
#> [1] 579ns
(ben_pur2$median[3] - ben_pur2$median[2]) / 1e6
#> [1] 361ns
ticking()
f0 <- function(n = 1e5) {
i <- 0
x <- 0
while (i < n) {
x <- x + i %% 2
i <- i + 1
}
x
}
ben_tk <- bench::mark(f0(), fp())
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
ben_tk
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 f0() 18.18ms 22.36ms 44.7 40.3KB 1.94
#> 2 fp() 5.28s 5.28s 0.189 100.8KB 2.46
(ben_tk$median[2] - ben_tk$median[1]) / 1e5
#> [1] 52.6µs
Traditional API
f0 <- function(n = 1e5) {
x <- 0
for (i in 1:n) {
x <- x + i %% 2
}
x
}
fp <- function(n = 1e5) {
cli_progress_bar(total = n)
x <- 0
for (i in 1:n) {
x <- x + i %% 2
cli_progress_update()
}
x
}
ff <- function(n = 1e5) {
cli_progress_bar(total = n)
x <- 0
for (i in 1:n) {
x <- x + i %% 2
if (`__cli_update_due`) cli_progress_update()
}
x
}
ben_api <- bench::mark(f0(), ff(), fp())
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
ben_api
#> # A tibble: 3 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 f0() 17.73ms 26.55ms 34.7 19.9KB 5.79
#> 2 ff() 28.96ms 33.81ms 28.5 28.6KB 1.90
#> 3 fp() 2.91s 2.91s 0.344 26.1KB 2.41
(ben_api$median[3] - ben_api$median[1]) / 1e5
#> [1] 28.8µs
(ben_api$median[2] - ben_api$median[1]) / 1e5
#> [1] 72.6ns
ben_api2 <- bench::mark(f0(1e6), ff(1e6), fp(1e6))
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
ben_api2
#> # A tibble: 3 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 f0(1e+06) 161.5ms 161.8ms 6.17 0B 6.17
#> 2 ff(1e+06) 228.8ms 228.9ms 4.37 1.88KB 4.37
#> 3 fp(1e+06) 26.8s 26.8s 0.0373 1.88KB 2.20
(ben_api2$median[3] - ben_api2$median[1]) / 1e6
#> [1] 26.7µs
(ben_api2$median[2] - ben_api2$median[1]) / 1e6
#> [1] 67.1ns
C benchmarks
Baseline function:
() {
SEXP test_baselineint i;
int res = 0;
for (i = 0; i < 2000000000; i++) {
+= i % 2;
res }
return ScalarInteger(res);
}
Switch + modulo check:
(SEXP progress) {
SEXP test_moduloint i;
int res = 0;
int progress_ = LOGICAL(progress)[0];
for (i = 0; i < 2000000000; i++) {
if (i % 10000 == 0 && progress_) cli_progress_set(R_NilValue, i);
+= i % 2;
res }
return ScalarInteger(res);
}
cli progress bar API:
() {
SEXP test_cliint i;
int res = 0;
= PROTECT(cli_progress_bar(2000000000, NULL));
SEXP bar for (i = 0; i < 2000000000; i++) {
if (CLI_SHOULD_TICK) cli_progress_set(bar, i);
+= i % 2;
res }
(bar);
cli_progress_done(1);
UNPROTECTreturn ScalarInteger(res);
}
() {
SEXP test_cli_unrollint i = 0;
int res = 0;
= PROTECT(cli_progress_bar(2000000000, NULL));
SEXP bar int s, final, step = 2000000000 / 100000;
for (s = 0; s < 100000; s++) {
if (CLI_SHOULD_TICK) cli_progress_set(bar, i);
= (s + 1) * step;
final for (i = s * step; i < final; i++) {
+= i % 2;
res }
}
(bar);
cli_progress_done(1);
UNPROTECTreturn ScalarInteger(res);
}
library(progresstest)
ben_c <- bench::mark(
test_baseline(),
test_modulo(),
test_cli(),
test_cli_unroll()
)
ben_c
#> # A tibble: 4 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 test_baseline() 1.28s 1.28s 0.782 2.08KB 0
#> 2 test_modulo() 2.17s 2.17s 0.462 2.24KB 0
#> 3 test_cli() 1.61s 1.61s 0.622 23.69KB 0
#> 4 test_cli_unroll() 1.28s 1.28s 0.782 3.35KB 0
(ben_c$median[3] - ben_c$median[1]) / 2000000000
#> [1] 1ns
Display update
We only update the display a fixed number of times per second. (Currently maximum five times per second.)
Let’s measure how long a single update takes.
Iterator with a bar
cli_progress_bar(total = 100000)
bench::mark(cli_progress_update(force = TRUE), max_iterations = 10000)
#> ■ 0% | ETA: 6m
#> ■ 0% | ETA: 2h
#> ■ 0% | ETA: 2h
#> ■ 0% | ETA: 1h
#> ■ 0% | ETA: 1h
#> ■ 0% | ETA: 1h
#> ■ 0% | ETA: 1h
#> ■ 0% | ETA: 46m
#> ■ 0% | ETA: 42m
#> ■ 0% | ETA: 39m
#> ■ 0% | ETA: 37m
#> ■ 0% | ETA: 35m
#> ■ 0% | ETA: 33m
#> ■ 0% | ETA: 32m
#> ■ 0% | ETA: 30m
#> ■ 0% | ETA: 29m
#> ■ 0% | ETA: 28m
#> ■ 0% | ETA: 27m
#> ■ 0% | ETA: 27m
#> ■ 0% | ETA: 26m
#> ■ 0% | ETA: 25m
#> ■ 0% | ETA: 25m
#> ■ 0% | ETA: 24m
#> ■ 0% | ETA: 24m
#> ■ 0% | ETA: 23m
#> ■ 0% | ETA: 23m
#> ■ 0% | ETA: 22m
#> ■ 0% | ETA: 22m
#> ■ 0% | ETA: 22m
#> ■ 0% | ETA: 21m
#> ■ 0% | ETA: 21m
#> ■ 0% | ETA: 21m
#> ■ 0% | ETA: 21m
#> ■ 0% | ETA: 20m
#> ■ 0% | ETA: 20m
#> ■ 0% | ETA: 20m
#> ■ 0% | ETA: 20m
#> ■ 0% | ETA: 19m
#> ■ 0% | ETA: 19m
#> ■ 0% | ETA: 19m
#> ■ 0% | ETA: 19m
#> ■ 0% | ETA: 19m
#> ■ 0% | ETA: 19m
#> ■ 0% | ETA: 19m
#> ■ 0% | ETA: 18m
#> ■ 0% | ETA: 18m
#> ■ 0% | ETA: 18m
#> ■ 0% | ETA: 18m
#> ■ 0% | ETA: 18m
#> ■ 0% | ETA: 18m
#> ■ 0% | ETA: 18m
#> ■ 0% | ETA: 18m
#> ■ 0% | ETA: 18m
#> ■ 0% | ETA: 17m
#> ■ 0% | ETA: 18m
#> ■ 0% | ETA: 17m
#> ■ 0% | ETA: 17m
#> ■ 0% | ETA: 17m
#> ■ 0% | ETA: 17m
#> ■ 0% | ETA: 17m
#> ■ 0% | ETA: 17m
#> ■ 0% | ETA: 17m
#> ■ 0% | ETA: 17m
#> ■ 0% | ETA: 17m
#> ■ 0% | ETA: 17m
#> ■ 0% | ETA: 17m
#> # A tibble: 1 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:> <bch:> <dbl> <bch:byt> <dbl>
#> 1 cli_progress_update(force = … 7.28ms 7.5ms 131. 1.34MB 2.04
cli_progress_done()
Iterator without a bar
cli_progress_bar(total = NA)
bench::mark(cli_progress_update(force = TRUE), max_iterations = 10000)
#> ⠙ 1 done (335/s) | 4ms
#> ⠹ 2 done (56/s) | 36ms
#> ⠸ 3 done (67/s) | 45ms
#> ⠼ 4 done (74/s) | 54ms
#> ⠴ 5 done (80/s) | 63ms
#> ⠦ 6 done (84/s) | 72ms
#> ⠧ 7 done (87/s) | 81ms
#> ⠇ 8 done (90/s) | 90ms
#> ⠏ 9 done (92/s) | 98ms
#> ⠋ 10 done (94/s) | 107ms
#> ⠙ 11 done (95/s) | 116ms
#> ⠹ 12 done (97/s) | 125ms
#> ⠸ 13 done (98/s) | 133ms
#> ⠼ 14 done (99/s) | 142ms
#> ⠴ 15 done (100/s) | 151ms
#> ⠦ 16 done (101/s) | 160ms
#> ⠧ 17 done (101/s) | 169ms
#> ⠇ 18 done (102/s) | 177ms
#> ⠏ 19 done (102/s) | 186ms
#> ⠋ 20 done (103/s) | 195ms
#> ⠙ 21 done (103/s) | 204ms
#> ⠹ 22 done (104/s) | 212ms
#> ⠸ 23 done (104/s) | 221ms
#> ⠼ 24 done (105/s) | 230ms
#> ⠴ 25 done (105/s) | 238ms
#> ⠦ 26 done (105/s) | 247ms
#> ⠧ 27 done (106/s) | 256ms
#> ⠇ 28 done (106/s) | 265ms
#> ⠏ 29 done (106/s) | 273ms
#> ⠋ 30 done (107/s) | 282ms
#> ⠙ 31 done (107/s) | 291ms
#> ⠹ 32 done (107/s) | 300ms
#> ⠸ 33 done (107/s) | 308ms
#> ⠼ 34 done (107/s) | 317ms
#> ⠴ 35 done (108/s) | 326ms
#> ⠦ 36 done (108/s) | 335ms
#> ⠧ 37 done (108/s) | 343ms
#> ⠇ 38 done (108/s) | 352ms
#> ⠏ 39 done (108/s) | 361ms
#> ⠋ 40 done (108/s) | 370ms
#> ⠙ 41 done (109/s) | 378ms
#> ⠹ 42 done (109/s) | 387ms
#> ⠸ 43 done (109/s) | 396ms
#> ⠼ 44 done (109/s) | 404ms
#> ⠴ 45 done (109/s) | 413ms
#> ⠦ 46 done (109/s) | 422ms
#> ⠧ 47 done (109/s) | 431ms
#> ⠇ 48 done (109/s) | 439ms
#> ⠏ 49 done (109/s) | 448ms
#> ⠋ 50 done (108/s) | 462ms
#> ⠙ 51 done (109/s) | 471ms
#> ⠹ 52 done (109/s) | 479ms
#> ⠸ 53 done (109/s) | 488ms
#> ⠼ 54 done (109/s) | 496ms
#> ⠴ 55 done (109/s) | 505ms
#> ⠦ 56 done (109/s) | 514ms
#> ⠧ 57 done (109/s) | 523ms
#> ⠇ 58 done (109/s) | 532ms
#> # A tibble: 1 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:> <bch:> <dbl> <bch:byt> <dbl>
#> 1 cli_progress_update(force = … 8.54ms 8.74ms 114. 198KB 2.04
cli_progress_done()