Synthetic Performance Test: GCC vs Intel ICC vs LuaJIT vs LuaJIT+FFI vs JavaScript

Hardware

Intel i7 2.3 GHz 4 cores (8 logical), 16GB RAM, SSD, macOS Mojave + Parallels 12;
VM Oracle Linux 7.6 configured with 4 cores, 8 GB RAM

C Test File

#include stdio.h
 
#define N 4000
#define S 1000
 
struct t {
        double a, b, f;
};
 
 
int main (int argc, char **argv) {
        int i, j;
        struct t t[N];
 
        for(i=0; i
                t[i].a = 0;
                t[i].b = 1;
                t[i].f = i * 0.25;
        };
 
        for(j=0; j
                for(i=0; i
                        t[i].a += t[i].b * t[i].f;
                        t[i].b -= t[i].a * t[i].f;
                }
                printf("%.6f\n", t[1].a);
        }
 
        return 0;
}

`GCC (4.8.5)`

gcc lua_perf.c -o lua_perf -O3 -Wall -march=native -ftree-parallelize-loops=4 -floop-parallelize-all -ftree-vectorize

time ./lua_perf > /dev/null

real 0m5.604s user 0m22.263s sys 0m0.118s

top 404 root 20 0 36468 4624 1356 R 400.0(%CPU) 0.1(%MEM) 0:09.08 lua_perf

GCC8 (8.3.1 from devtoolset-8)

real 0m5.695s user 0m22.632s sys 0m0.124s

`ICC (19.0.4.235)`

/opt/intel/system_studio_2019/bin/icc lua_perf.c -O2 -o lua_perf_icc -no-prec-div -ipo -xSSE4.2 -parallel

time ./lua_perf_icc > /dev/null

real 0m5.322s user 0m21.186s sys 0m0.074s

top 14344 root 20 0 247848 6588 3100 R 400.0 (%CPU) 0.1(%MEM) 0:14.55 lua_perf_icc

`Lua Test File`


local N = 4000 
local S = 1000 
local t = {} 
for i = 0, N do 
  t[i] = { a = 0, b = 1, f = i * 0.25 } 
end 
for j = 0, S-1 do 
  for i = 0, N-1 do 
    t[i].a = t[i].a + t[i].b * t[i].f 
   t[i].b = t[i].b - t[i].a * t[i].f 
  end 
  print(string.format("%.6f", t[1].a)) 
end

`LuaJIT`

time /usr/local/openresty/luajit/bin/luajit lua_perf.lua > /dev/null

real 3m4.680s user 3m4.612s sys 0m0.042s

15581 root 20 0 38572 28360 2000 R 100.0(%CPU) 0.4(%MEM) 0:04.08 luajit

`Lua+FFI Test File`


--collectgarbage('setpause', 2000)
local ffi = require("ffi")
ffi.cdef[[
typedef struct { double a, b, f; } table_elem;
]]
local N = 140000
local S = 110000
local t = ffi.new("table_elem[?]", N)
for i = 0, N-1 do
  t[i].a = 0.0
  t[i].b = 1.0
  t[i].f = i * 0.25
end

for j = 0, S-1 do
  for i = 0, N-1 do
    t[i].a = t[i].a + t[i].b * t[i].f
    t[i].b = t[i].b - t[i].a * t[i].f
  end
  print(string.format("%.6f", t[1].a))
end

`LuaJIT+FFI`

time /usr/local/openresty/luajit/bin/luajit lua_perf_ffi.lua > /dev/null

real 0m22.603s user 0m22.589s sys 0m0.012s

15625 root 20 0 17920 7508 1968 R 100.0(%CPU) 0.1(%MEM) 0:08.00 luajit

`JavaScript Test File`

class lua_perf {

public double a, b, f;

        static final int N=4000;
        static final int S=1000;
 
        public static void main (String[] argv) {
                int i, j;
                lua_perf[] t = new lua_perf[N];
                for(i=0; i
                        t[i] = new lua_perf();
                        t[i].a = 0;
                        t[i].b = 1;
                        t[i].f = i * 0.25;
                };
 
                for(j=0; j
                        for(i=0; i
                                t[i].a += t[i].b * t[i].f;
                                t[i].b -= t[i].a * t[i].f;
                        }
                        System.out.println(t[1].a);
                }
        }
}

`JavaScript (without any optimize in JVM)`

java version "1.8.0_201" Java(TM) SE Runtime Environment (build 1.8.0_201-b09) Java HotSpot(TM) 64-Bit Server VM (build 25.201-b09, mixed mode)


/u01/app/oracle/product/19.0.0/dbhome_1/jdk/bin/javac lua_perf.java


time /u01/app/oracle/product/19.0.0/dbhome_1/jdk/bin/java lua_perf > /dev/null

real 0m33.052s user 0m32.849s sys 0m0.425s

15516 root 20 0 4452640 37940 16056 S 100.7(%CPU) 0.6(%MEM) 0:07.33 java

`Files size`

-rwxr-xr-x. 1 root root 8.6K Aug 9 18:13 lua_perf -rw-r--r--. 1 root root 572 Aug 9 17:43 lua_perf.c -rw-r--r--. 1 root root 822 Aug 9 19:04 lua_perf.class -rw-r--r--. 1 root root 438 Aug 9 17:25 lua_perf_ffi.lua -rwxr-xr-x. 1 root root 8.3K Aug 9 18:26 lua_perf_gcc8 -rwxr-xr-x. 1 root root 28K Aug 9 18:28 lua_perf_icc -rw-r--r--. 1 root root 777 Aug 9 19:04 lua_perf.java -rw-r--r--. 1 root root 380 Aug 9 19:03 lua_perf.lua

`Conclusions`

GCC and ICC have similar performance; Intel is a little bit faster(1.07x) in this particular test. LuaJIT_FFI has C-like performance but needs parallelism for speed of C programs compiled with parallel option. LuaJIT (NOT FFI) has not bad performance for script language. JavaScript has good performance, but loves many RAM as usually. Pure Lua (NOT JIT) has not been considered as it will be a priory too slow in this test.

sdmrnv, 2019-08-09 [0.87ms, r]