Crystal is much slower in recent releases

[renich]

Please, check this one out: https://gitlab.com/renich/benchmarks/-/jobs/493556257#L127. Crystal's test starts at line: 142. It used to be much faster than even C when printing 1_000_000 lines. Now, it's not.

My latest (local) result:

gcc -o one_million_c one_million.c > /dev/null
time ./one_million_c > /dev/null

real	0m0.068s
user	0m0.066s
sys	0m0.002s
echo

g++ -o one_million_cpp one_million.cpp > /dev/null
time ./one_million_cpp > /dev/null

real	0m0.115s
user	0m0.113s
sys	0m0.002s
echo

crystal build --release -o one_million_cr one_million.cr
time ./one_million_cr > /dev/null

real	0m0.353s
user	0m0.158s
sys	0m0.196s
echo

This started with v0.34.0... I think. It could've been v0.33.00.

In any case, I don't even know if this is a bug or not. Just wanted to report it.

I hope you fix it... otherwise, the days of bragging about this are gone: https://twitter.com/Renich/status/1138838599397519360 😢

Even @asterite commented on it! ;D

[j8r]

This are real-world benchmarks! 😄
Try setting STDOUT.flush_on_newline = false before printing.

[asterite]

Hi! @j8r is right. Please see #8935

[renich]

Right! We're back at beating C! ;D

[renich@introdesk one_million]$ make
gcc -o one_million_c one_million.c > /dev/null
time ./one_million_c > /dev/null

real	0m0.068s
user	0m0.067s
sys	0m0.001s
echo

g++ -o one_million_cpp one_million.cpp > /dev/null
time ./one_million_cpp > /dev/null

real	0m0.112s
user	0m0.112s
sys	0m0.000s
echo

crystal build --release -o one_million_cr one_million.cr
time ./one_million_cr > /dev/null

real	0m0.041s
user	0m0.040s
sys	0m0.003s
echo

go build -o one_million_go one_million.go
time ./one_million_go > /dev/null

real	0m0.356s
user	0m0.267s
sys	0m0.091s
echo

javac one_million.java
time java OneMillion > /dev/null

real	0m0.476s
user	0m0.562s
sys	0m0.139s
echo

time node one_million.js > /dev/null

real	0m2.821s
user	0m2.678s
sys	0m0.154s
echo

time perl one_million.pl > /dev/null

real	0m0.119s
user	0m0.111s
sys	0m0.004s
echo

time php one_million.phps > /dev/null

real	0m0.233s
user	0m0.134s
sys	0m0.088s
echo

time python2 one_million.py > /dev/null

real	0m0.663s
user	0m0.634s
sys	0m0.015s
echo

time python3 one_million.py3 > /dev/null

real	0m0.521s
user	0m0.518s
sys	0m0.002s
echo

time perl6 one_million.p6 > /dev/null

real	0m2.605s
user	0m2.608s
sys	0m0.025s
time ruby one_million.rb > /dev/null

real	0m0.799s
user	0m0.774s
sys	0m0.010s
echo

rustc -o one_million_rust one_million.rs > /dev/null
time ./one_million_rust > /dev/null

real	0m0.330s
user	0m0.248s
sys	0m0.081s

Yeah, I know it's not much of a benchmark. But it does some basic dereferencing and a loop. Pretty much what my coding, at least, is all about. 😅

[jhass]

So... issue closed? :D

[renich]

Well, yeah. I think. Thanks.