args nobs
if "`nobs'" == "" local nobs 10000000

capture program drop bench
program bench
    gettoken timer call: 0,    p(:)
    gettoken colon call: call, p(:)
    cap timer clear `timer'
    timer on `timer'
    `call'
    timer off `timer'
    qui timer list
    c_local r`timer' `=r(t`timer')'
end

clear
set obs `nobs'
gen groups = int(runiform() * 1000)
gen rsort  = rnormal()
gen rvar   = rnormal()
gen ix     = _n
sort rsort
local nprocessors = c(processors)

timer clear

preserve
bench 11: gcollapse (sum) rvar (mean) mean = rvar, by(groups)
restore, preserve
bench 10: collapse (sum) rvar (mean) mean = rvar, by(groups)
restore

preserve
bench 16: gcollapse (sd) sd = rvar (median) med = rvar, by(groups)
restore, preserve
bench 15: collapse (sd) sd = rvar (median) med = rvar, by(groups)
restore

* Back to wide would have an unfair advantage if I did it right after
* long because the data in memory would be sorted already! greshape
* allows going to long w/o sorting or checking dups, etc. We use the
* option -nochecks-, which runs very fast, to go to long and _then_
* benchmark wide.

preserve
    rename (rsort rvar) (r1 r2)
    bench 21: greshape long r, i(ix) j(j)
restore, preserve
    rename (rsort rvar) (r1 r2)
    greshape long r, i(ix) j(j) nochecks
    bench 26: greshape wide r, i(ix) j(j)
restore, preserve
    rename (rsort rvar) (r1 r2)
    bench 20: reshape long r, i(ix) j(j)
restore, preserve
    rename (rsort rvar) (r1 r2)
    greshape long r, i(ix) j(j) nochecks
    bench 25: reshape wide r, i(ix) j(j)
restore

* The relative gains are very comparable here, but other reshape
* solutions fare much better comparatibely when j is a number.  I also
* do think that numeric j is probably more common than string j.

* preserve
*     bench 21: greshape long r, i(ix) j(j) string
*     bench 26: greshape wide r, i(ix) j(j) string
* restore, preserve
*     bench 20: reshape long r, i(ix) j(j) string
*     bench 25: reshape wide r, i(ix) j(j) string
* restore

bench 31: gquantiles g_xtile = rvar, nq(10) xtile
bench 30: xtile s_xtile = rvar, nq(10)

bench 36: gquantiles g_pctile = rvar, nq(10) pctile
bench 35: pctile s_pctile = rvar, nq(10)

bench 41: gegen g_id = group(groups)
bench 40: egen  s_id = group(groups)

preserve
bench 46: gcontract groups
restore, preserve
bench 45: contract groups
restore

bench 51: gisid ix
bench 50: isid ix

preserve
bench 56: gduplicates drop groups, force
restore, preserve
bench 55: duplicates drop groups, force
restore

bench 61: qui glevelsof groups
bench 60: qui levelsof groups

bench 66: qui gdistinct groups
bench 65: qui distinct groups

bench 71: qui gstats winsor rvar, s(_wg)
bench 70: qui winsor2 groups

bench 76: qui gstats sum rvar
bench 75: qui sum rvar, detail

gen smallg = mod(groups, 10)

bench 81: qui gstats tab rvar, by(smallg) s(n mean min max)
bench 80: qui tabstat rvar,    by(smallg) s(n mean min max)

bench 86: qui gstats range (mean) g_rm = rvar, by(groups) interval(-10 10 ix)
bench 85: qui rangestat (mean) r_rm = rvar, by(groups) interval(ix -10 10)

local commands   ///
      collapse   ///
      collapse   ///
      reshape    ///
      reshape    ///
      xtile      ///
      pctile     ///
      egen       ///
      contract   ///
      isid       ///
      duplicates ///
      levelsof   ///
      distinct   ///
      winsor     ///
      sum_detail ///
      tabstat    ///
      range_stat

local bench_table `"     Versus | Native | gtools | % faster "'
local bench_table `"`bench_table'"' _n(1) `" ---------- | ------ | ------ | -------- "'
forvalues i = 10(5)85 {
    gettoken cmd commands: commands
    local pct      "`:disp %7.2f  100 * (`r`i'' - `r`=`i'+1'') / `r`i'''"
    local dnative  "`:disp %6.2f `r`i'''"
    local dgtools  "`:disp %6.2f `r`=`i'+1'''"
    local cmd      `"`:disp %10s "`cmd'"'"'
    local bench_table `"`bench_table'"' _n(1) `" `cmd' | `dnative' | `dgtools' | `pct'% "'
}
disp _n(1) `"`bench_table'"'