This is a quick and dirty set of files to test Julia (and other languages) against C on the Mandelbrot set (and on the N-Body problem). I started out with a C mandelbrot version passed to me and attempted to implement the equivalent structure in Julia. Both versions will produce an output on standard output. Both programs also measure the execution time of just generating the mandelbrot set, not the file output time, and output it to standard error.
Running the computation loop only 100 times produced the following results on my crummy laptop (specs below)
| Number Trials | Minimum Execution Time | |
|---|---|---|
mandel.jl |
100 | 0.339978 |
mandel.c |
100 | 0.516530 |
The above results are using the "fast" version of the C program, with more aggressive compilation flags:
gcc -o mandel_fast -pipe -Wall -O3 -ffast-math -ffinite-math-only -march=native -mfpmath=sse -msse3 mandel.c
For the assembly generation the mandelbrot function was ensured to not be inlined and I added -ggdb3 to the compilation flags.
The Julia assembly language generation was done from the REPL after importing the file and calling code_native(mandelbrot, Tuple{Array{UInt8,2}}). When the Julia version of the program was called there were no special compilation flags, it was called like julia mandel.jl > mandel_julia.pbm.
| Field | Value |
|---|---|
| processor | (4 entries with values of 0-3) |
| vendor_id | GenuineIntel |
| cpu family | 6 |
| model | 78 |
| model name | Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz |
| stepping | 3 |
| microcode | 0x88 |
| cpu MHz | 2663.818 |
| cache size | 3072 KB |
| physical id | 0 |
| siblings | 4 |
| core id | 1 |
| cpu cores | 2 |
| apicid | 3 |
| initial apicid | 3 |
| fpu | yes |
| fpu_exception | yes |
| cpuid level | 22 |
| wp | yes |
| flags | fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch epb intel_pt tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid mpx rdseed adx smap clflushopt xsaveopt xsavec xgetbv1 xsaves dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp |
| bugs | |
| bogomips | 4800.00 |
| clflush size | 64 |
| cache_alignment | 64 |
| address sizes | 39 bits physical, 48 bits virtual |
| power management |
This program models the orbit of Jovian planets using a symplectic integrator. For most of the programs I used the code at the Computer Language Benchmarks Game site. Note that in the end my Julia program performed horrendously --- basically unusable. I went online to some Julia forums where they were discussing this exact problem and website and downloaded and ran the "best" version there to get these numbers. I implemented the Cython version myself, with some tips from co-workers as to some of the best places to look for optimization. This time the results are presented in roughly the multiple of the C execution speed (approximately 2 significant digits).
| Language | Execution Time |
|---|---|
| C | 1x |
| C++ | ~1x |
| Python (CPython) | ~100-200x |
| Python (pypy) | ~10-20x |
| Cython | 1.4x |
| Julia (best) | 1.6x |
| Julia (clean) | 3x |
| Julia (naïve) | >1000x |
| Java | 1.6x |