1
0
mirror of https://github.com/vlang/v.git synced 2023-08-10 21:13:21 +03:00
v/vlib/benchmark/benchmark.v

225 lines
5.4 KiB
V

module benchmark
import time
import term
/*
Example usage of this module:
```
import benchmark
mut bmark := benchmark.new_benchmark()
// by default the benchmark will be verbose, i.e. it will include timing information
// if you want it to be silent, set bmark.verbose = false
for {
bmark.step() // call this when you want to advance the benchmark.
// The timing info in bmark.step_message will be measured starting from the last call to bmark.step
....
//bmark.fail() // call this if the step failed
//bmark.step_message(('failed')
bmark.ok() // call this when the step succeeded
println( bmark.step_message('ok')
}
bmark.stop() // call when you want to finalize the benchmark
println( bmark.total_message('remarks about the benchmark') )
```
benchmark.start() and b.measure() are convenience methods,
intended to be used in combination. Their goal is to make
benchmarking of small snippets of code as *short*, easy to
write, and then to read and analyze the results, as possible.
Example:
```v
import benchmark
b := benchmark.start()
// your code 1 ...
b.measure('code_1')
// your code 2 ...
b.measure('code_2')
```
... which will produce on stdout something like this:
SPENT 17 ms in code_1
SPENT 462 ms in code_2
*/
const (
BOK = term.ok_message('OK ')
BFAIL = term.fail_message('FAIL')
BSKIP = term.warn_message('SKIP')
BSPENT = term.ok_message('SPENT')
)
pub struct Benchmark {
pub mut:
bench_timer time.StopWatch
step_timer time.StopWatch
ntotal int
nok int
nfail int
nskip int
verbose bool
nexpected_steps int
cstep int
no_cstep bool
bok string
bfail string
}
pub fn new_benchmark() Benchmark {
return Benchmark{
bench_timer: time.new_stopwatch()
verbose: true
}
}
pub fn new_benchmark_no_cstep() Benchmark {
return Benchmark{
bench_timer: time.new_stopwatch()
verbose: true
no_cstep: true
}
}
pub fn new_benchmark_pointer() &Benchmark {
return &Benchmark{
bench_timer: time.new_stopwatch()
verbose: true
}
}
pub fn (b mut Benchmark) set_total_expected_steps(n int) {
b.nexpected_steps = n
}
pub fn (b mut Benchmark) stop() {
b.bench_timer.stop()
}
pub fn (b mut Benchmark) step() {
b.step_timer.restart()
if !b.no_cstep {
b.cstep++
}
}
pub fn (b mut Benchmark) fail() {
b.step_timer.stop()
b.ntotal++
b.nfail++
}
pub fn (b mut Benchmark) ok() {
b.step_timer.stop()
b.ntotal++
b.nok++
}
pub fn (b mut Benchmark) skip() {
b.step_timer.stop()
b.ntotal++
b.nskip++
}
pub fn (b mut Benchmark) fail_many(n int) {
b.step_timer.stop()
b.ntotal += n
b.nfail += n
}
pub fn (b mut Benchmark) ok_many(n int) {
b.step_timer.stop()
b.ntotal += n
b.nok += n
}
pub fn (b mut Benchmark) neither_fail_nor_ok() {
b.step_timer.stop()
}
pub fn start() Benchmark {
mut b := new_benchmark()
b.step()
return b
}
pub fn (b mut Benchmark) measure(label string) i64 {
b.ok()
res := b.step_timer.elapsed().microseconds()
println(b.step_message_with_label(BSPENT, 'in $label'))
b.step()
return res
}
pub fn (b &Benchmark) step_message_with_label(label string, msg string) string {
timed_line := b.tdiff_in_ms(msg, b.step_timer.elapsed().microseconds())
if b.nexpected_steps > 1 {
mut sprogress := ''
if b.nexpected_steps < 10 {
sprogress = if b.no_cstep { 'TMP1/${b.nexpected_steps:1d}' } else {
'${b.cstep:1d}/${b.nexpected_steps:1d}'
}
} else if b.nexpected_steps >= 10 && b.nexpected_steps < 100 {
sprogress = if b.no_cstep { 'TMP2/${b.nexpected_steps:2d}' } else {
'${b.cstep:2d}/${b.nexpected_steps:2d}'
}
} else if b.nexpected_steps >= 100 && b.nexpected_steps < 1000 {
sprogress = if b.no_cstep { 'TMP3/${b.nexpected_steps:3d}' } else {
'${b.cstep:3d}/${b.nexpected_steps:3d}'
}
} else {
sprogress = if b.no_cstep { 'TMP4/${b.nexpected_steps:4d}' } else {
'${b.cstep:4d}/${b.nexpected_steps:4d}'
}
}
return '${label:-5s} [${sprogress}] ${timed_line}'
}
return '${label:-5s}${timed_line}'
}
pub fn (b &Benchmark) step_message(msg string) string {
return b.step_message_with_label('', msg)
}
pub fn (b &Benchmark) step_message_ok(msg string) string {
return b.step_message_with_label(BOK, msg)
}
pub fn (b &Benchmark) step_message_fail(msg string) string {
return b.step_message_with_label(BFAIL, msg)
}
pub fn (b &Benchmark) step_message_skip(msg string) string {
return b.step_message_with_label(BSKIP, msg)
}
pub fn (b &Benchmark) total_message(msg string) string {
mut tmsg := '${msg}\n ok, fail, skip, total = ' + term.ok_message('${b.nok:5d}') + ', ' + if b.nfail > 0 { term.red('${b.nfail:5d}') } else { '${b.nfail:5d}' } + ', ' + if b.nskip > 0 { term.bright_yellow('${b.nskip:5d}') } else { '${b.nskip:5d}' } + ', ' + '${b.ntotal:5d}'
if b.verbose {
tmsg = '<=== total time spent $tmsg'
}
mut spaces := ' '
if b.nexpected_steps > 1 {
// NB: the formula below accounts for the progress bar [step/total]
str_steps := '$b.nexpected_steps'
x := 4 + str_steps.len * 2 + 5
spaces = ' '.repeat(x)
}
return spaces + b.tdiff_in_ms(tmsg, b.bench_timer.elapsed().microseconds())
}
// .total_duration - returns the duration in ms
pub fn (b &Benchmark) total_duration() i64 {
return b.bench_timer.elapsed().milliseconds()
}
// //////////////////////////////////////////////////////////////////
fn (b &Benchmark) tdiff_in_ms(s string, tdiff i64) string {
if b.verbose {
return '${f64(tdiff)/1000.0:9.3f} ms $s'
}
return s
}