Browse Source

Initial benchmark: comparing haskell's Text with icu on EN corpus

Run `make all` to run the benchmark.
master
Félix Baylac-Jacqué 11 months ago
commit
d4488c5096
  1. 22
      Makefile
  2. 24
      default.nix
  3. 7
      haskell/HaskellReadUTF8.hs
  4. 11
      icu/icu-read-utf8.c
  5. 50
      readme.md

22
Makefile

@ -0,0 +1,22 @@
CFLAGS=-O2 -Wall $(shell pkgconf -cflags icu-io)
LDFLAGS=$(shell pkgconf -libs icu-io)
CC=gcc
all: haskell-read-utf8 icu-read-utf8 bench
.PHONY: all
bench: haskell-read-utf8 icu-read-utf8
hyperfine ./haskell-read-utf8 ./icu-read-utf8
.PHONE: bench
haskell-read-utf8: ./haskell/HaskellReadUTF8.hs
ghc -o ./haskell-read-utf8 -O ./haskell/HaskellReadUTF8.hs
icu-read-utf8: ./icu/icu-read-utf8.c
$(CC) $(CFLAGS) $(LDFLAGS) ./icu/icu-read-utf8.c -o ./icu-read-utf8
clean:
rm -f ./haskell/*.{o,hi}
rm -f ./haskell-read-utf8
rm -f ./icu-read-utf8
.PHONY: clean

24
default.nix

@ -0,0 +1,24 @@
{ pkgs ? import <nixpkgs> {} }:
let ghcClosure =
pkgs.haskellPackages.ghcWithPackages
(p:[
p.text
]);
in pkgs.stdenv.mkDerivation {
pname = "bench-my-utf8";
version = "0.0.1";
installPhase = ''
mkdir -p $out/bin
mv haskell-read-utf8 $out/bin
'';
nativeBuildInputs = [
pkgs.gnumake
ghcClosure
pkgs.gcc
pkgs.icu
pkgs.hyperfine
pkgs.pkgconf
];
}

7
haskell/HaskellReadUTF8.hs

@ -0,0 +1,7 @@
module Main where
import qualified Data.Text.IO as TIO
import qualified Data.Text as T
main :: IO ()
main = TIO.readFile "text-test-data/english.txt" >> pure ()

11
icu/icu-read-utf8.c

@ -0,0 +1,11 @@
#include <unicode/ustdio.h>
#include <stdlib.h>
#include <unicode/umachine.h>
int main(void)
{
UFILE *in = u_fopen("text-test-data/english.txt", "r", NULL, "UTF-8");
// Using a 10 MB buffer for now. It fits all the corpus I want to test agaisnt.
UChar *charBuff = malloc(10000000 * sizeof(UChar));
int32_t i = u_file_read(charBuff, 10000000, in);
}

50
readme.md

@ -0,0 +1,50 @@
# UTF-8 Benchmark
The idea is to benchmark several utf-8 aspects of the Haskell Text package. Namely utf-8 encoding/decoding and various unicode casing operations. Hopefully, we'll find ways to improve text's performance.
For the time being, we're comparing the Text implementation with the C ICU one. In the future I also plan to test it against C++ Boost and the Rust stdlib.
# Implemented so far
- [x] UTF-8 decoding from file.
- [x] English
- [ ] Chinese
- [ ] French
- [ ] Russian
- [ ] UTF-8 encoding to file.
- [ ] English
- [ ] Chinese
- [ ] French
- [ ] Russian
- [ ] UTF-8 encoding to file.
- [ ] English
- [ ] Chinese
- [ ] French
- [ ] Russian
# Usage
```bash
nix-shell
make all
```
# Findings
## UTF-8 Decoding
```
hyperfine ./haskell-read-utf8 ./icu-read-utf8
Benchmark #1: ./haskell-read-utf8
Time (mean ± σ): 23.3 ms ± 0.9 ms [User: 14.9 ms, System: 8.3 ms]
Range (min … max): 22.0 ms … 26.1 ms 111 runs
Benchmark #2: ./icu-read-utf8
Time (mean ± σ): 12.5 ms ± 0.8 ms [User: 7.6 ms, System: 4.9 ms]
Range (min … max): 11.5 ms … 16.1 ms 176 runs
Summary
'./icu-read-utf8' ran
1.85 ± 0.14 times faster than './haskell-read-utf8'
```
Loading…
Cancel
Save