#!/usr/bin/env bash
# The speed half of cost.sh, on programs that are long enough to time.
#
# Why this exists beside cost.sh rather than inside it: every program in
# test/programs runs in about two and a half milliseconds, of which nearly
# all is fork, exec and the dynamic loader. Best-of-five does not rescue a
# signal that is not there, and a table of 97 rows that all say "2.5ms vs
# 2.6ms" would be a measurement of execve. So the corpus answers the size
# question and these four answer the speed one, each written so that one
# suspected cost is most of what the program does.
#
# Four builds of each, and the third column is the one to read:
#
#   llvm      as shipped, -O2. Frequently the loop is simply gone; that is a
#             true number about the toolchain and a useless one about codegen.
#   llvm -O0  via --debug, which forces it. LLVM's instruction selection with
#             its optimiser off -- the fair comparison for a backend that has
#             no optimiser.
#   x86       this backend.
#   x86 nbc   --no-bounds-checks, for b2, where the difference is the check.
#
# And --dev on both sides, which is the one suspected cost the two backends
# share: every cross-function call goes through an indirection cell, so it is
# a load and an indirect call where a release build has a direct one. b1 is
# where that has to show.
#
# Usage: spike/x86/bench.sh [name-substring ...]
set -u
orig=$(pwd)
here=$(cd "$(dirname "$0")" && pwd)
root=$(cd "$here/../.." && pwd)
cd "$root" || exit 1

if [ -n "${FLAN:-}" ]; then
  case $FLAN in /*) flan=$FLAN;; *) flan=$orig/$FLAN;; esac
else
  dune build --root . bin/main.exe 2>&1 | head -30
  flan=$root/_build/default/bin/main.exe
fi
test -x "$flan" || { echo "build failed" >&2; exit 1; }

out=${COST_OUT:-${TMPDIR:-/tmp}/flan-bench.$$}
mkdir -p "$out" || exit 1
trap 'rm -rf "$out"' EXIT

REPS=${COST_REPS:-5}

best () {
  min=
  for i in $(seq "$REPS"); do
    t0=$(date +%s%N)
    timeout 120 "$1" >/dev/null 2>&1 </dev/null
    t1=$(date +%s%N)
    d=$(( (t1 - t0) / 1000 ))
    if [ -z "$min" ] || [ "$d" -lt "$min" ]; then min=$d; fi
  done
  echo "$min"
}

own () {
  nm --defined-only -S "$1" 2>/dev/null \
    | awk 'NF==4 && ($3=="T"||$3=="t") && $4 ~ /^flan\./ {n+=strtonum("0x"$2)} END{print n+0}'
}

printf 'name\tllvm_us\to0_us\tx86_us\tllvm_nbc_us\tx86_nbc_us\tllvm_dev_us\tx86_dev_us\tllvm_own\to0_own\tx86_own\tx86_dev_own\n'

for src in "$root"/spike/x86/bench/*.flan; do
  name=$(basename "$src" .flan)
  if [ $# -gt 0 ]; then
    want=0
    for pat in "$@"; do case "$name" in *"$pat"*) want=1;; esac; done
    [ $want = 1 ] || continue
  fi
  "$flan" build "$src" -o "$out/l" >/dev/null 2>&1 || { echo "$name: llvm build failed" >&2; continue; }
  "$flan" build "$src" --debug -o "$out/d" >/dev/null 2>&1 || { echo "$name: -O0 build failed" >&2; continue; }
  "$flan" build "$src" --x86 -o "$out/x" >/dev/null 2>&1 || { echo "$name: x86 build failed" >&2; continue; }
  "$flan" build "$src" --no-bounds-checks -o "$out/ln" >/dev/null 2>&1
  "$flan" build "$src" --x86 --no-bounds-checks -o "$out/xn" >/dev/null 2>&1
  "$flan" build "$src" --dev -o "$out/lv" >/dev/null 2>&1
  "$flan" build "$src" --x86 --dev -o "$out/xv" >/dev/null 2>&1
  printf '%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\n' "$name" \
    "$(best "$out/l")" "$(best "$out/d")" "$(best "$out/x")" \
    "$(best "$out/ln")" "$(best "$out/xn")" \
    "$(best "$out/lv")" "$(best "$out/xv")" \
    "$(own "$out/l")" "$(own "$out/d")" "$(own "$out/x")" "$(own "$out/xv")"
done