-
Notifications
You must be signed in to change notification settings - Fork 11
Performance Tests
The test cases are took mainly from two sources:
- Project Euler, featuring interesting mathematical and programming problems. The problems can often achieve significant performance improvements from compilation.
- The computer language benchmarks game, featuring toy benchmark programs. The benchmark problems are closer to the use cases for compilation in real life.
- Operating System: Ubuntu 18.04
- Compiler: GCC 7.4
- Processor: Intel Coffee Lake @ ~4.3GHz
- Mathematica: v12.0
The functions are implemented following a few guidelines:
- Use built-in functions, unless the function is not compilable.
- Minimize the use of
For
andWhile
. - The code for the built-in compiler should not invoke
MainEvaluate
.
In addition, parallelization is not used in the implementations.
The overhead of each function call is nearly uniform across different functions (180±10 nanoseconds), and it is subtracted from the timing.
Problem | Not compiled | Built-in compiler | MathCompile |
---|---|---|---|
Mandelbrot set | 254 ms | 17 ms (15x) | 3.2 ms (79x) |
fannkuch-redux | 36 μs | 1.38 μs (26x) | 0.135 μs (270x) |
Multiples of 3 and 5 | 620 μs | 23.3 μs (27x) | 1.38 μs (450x) |
Sum square difference | 14.1 μs | 0.16 μs (87x) | 0.071 μs (200x) |
Largest product in a series | 356 μs | 107 μs (3x) | 25.9 μs (14x) |
Special Pythagorean triplet | 316 ms | 780 μs (400x) | 197 μs (1600x) |
Longest Collatz sequence | 6.4 s | 3.5 s (1.8x) | 0.185 s (35x) |
Digit fifth powers | 1.52 s | 0.190 s (8x) | 0.035 s (43x) |
Import a numerical array | 1.22 s | 0.544 s (2.2x) | |
Export a numerical array | 4.23 s | 0.537 s (8x) |
-
f0
: the function that is not compiled; -
f1
: the function compiled by MathCompile; -
f2
: the function compiled by the built-in compiler,Compile
is always called with optionsCompilationTarget->"C"
andRuntimeOptions->"Speed"
, even if they are not included below.
Generate a 200×200 array of Mandelbrot set between -1.5-i and 0.5+i, the the maximum number of iteration is 50.
f0=Function[{},Table[Module[{g=0.I,n=0},
While[n<50&&Abs@g<2,++n;g=g*g+(i+j I)];Boole[Abs@g<2]],{i,-1.5,0.5,0.01},{j,-1.,1.,0.01}]];
f1=CompileToBinary[f0];
f2=Compile[{},Table[Module[{g=0.I,n=0},
While[n<50&&Abs@g<2,++n;g=g*g+(i+j I)];Boole[Abs@g<2]],{i,-1.5,0.5,0.01},{j,-1.,1.,0.01}]];
Source: The computer language benchmarks game
Given a permutation p of {1,…,n}, flip the first i=p[0] elements of p repeatedly, until p[0]=1. Count the number of flips.
The functions are called on all permutations of {1,…,8}, taking the average execution time for each permutation.
f0=Function[{Typed[x,{Integer,1}]},Module[{n=0},
NestWhile[Module[{y=#,i=#[[1]]},++n;y[[;;i]]=Reverse@y[[;;i]];y]&,x,#[[1]]!=1&];n]];
f1=CompileToBinary[f0];
f2=Compile[{{x,_Integer,1}},Module[{n=0},
NestWhile[Module[{y=#,i=#[[1]]},++n;y[[;;i]]=Reverse@y[[;;i]];y]&,x,#[[1]]!=1&];n]];
Source: The computer language benchmarks game
Find the sum of all the multiples of 3 or 5 below 1000.
f0=Function[{Typed[x,Integer]},Total@Select[Range[x],Divisible[#,3]||Divisible[#,5]&]];
f1=CompileToBinary[f0];
f2=Compile[{{x,_Integer}},Sum[If[Mod[i,3]==0||Mod[i,5]==0,i,0],{i,x}]];
Source: Project Euler #1
Find the difference between the sum of the squares of the first one hundred natural numbers and the square of the sum.
f0=Function[{Typed[x,Integer]},Sum[i,{i,x}]^2-Sum[i^2,{i,x}]];
f1=CompileToBinary[f0];
f2=Compile[{{x,_Integer}},Sum[i,{i,x}]^2-Sum[i^2,{i,x}]];
Source: Project Euler #6
Find the thirteen adjacent digits in the 1000-digit number that have the greatest product.
f0=Function[{Typed[x,{Integer,1}]},Max@Table[Times@@x[[i;;i+12]],{i,1,Length[x]-12}]];
f1=CompileToBinary[f0];
f2=Compile[{{x,_Integer,1}},Max@Table[Times@@x[[i;;i+12]],{i,1,Length[x]-12}]];
Note: the 1000-digit number is represented as a list of integers.
Source: Project Euler #8
There exists exactly one Pythagorean triplet for which a+b+c=1000. Find the product abc.
f0=Function[{},Module[{r=0,k=0},
Do[k=1000-i-j;If[k>0&&i^2+j^2==k^2,r=i j k;Break[]],{i,1,999},{j,1,999}];r]];
f1=CompileToBinary[f0];
f2=Compile[{},Module[{r=0,k=0},
Do[k=1000-i-j;If[k>0&&i^2+j^2==k^2,r=i j k;Break[]],{i,1,999},{j,1,999}];r]];
Source: Project Euler #9
Which starting number, under one million, produces the longest chain of Collatz sequence?
f0=Function[{Typed[x,Integer]},
Ordering[Table[Module[{n=1,g=i},While[g>1,++n;g=If[OddQ[g],3g+1,Quotient[g,2]];];n],{i,x}],-1]];
f1=CompileToBinary[f0];
f2=Ordering[#,-1]&@*Evaluate[Compile[{{x,_Integer}},
Table[Module[{n=1,g=i},While[g>1,++n;g=If[OddQ[g],3g+1,Quotient[g,2]];];n],{i,x}]]];
Note: all three implementations do not use dynamic programming.
Source: Project Euler #14
Find the sum of all the numbers that can be written as the sum of fifth powers of their digits.
f0 = Function[{},Total@Select[Range[2,999999],Total[IntegerDigits[#]^5]==#&]];
f1=CompileToBinary[f0];
f2=Compile[{},Total@Table[Module[{sum=0,g=i},
While[g>0,sum+=Mod[g,10]^5;g=Quotient[g,10];];If[i==sum,i,0]],{i,2,999999}]];
Note: IntegerDigits
is not compilable by Compile
, so its implementation is different.
Source: Project Euler #30
Import a 1000x1000 numerical array in TSV format. The elements are double-precision numbers distributed uniformly on log scale between 10-15 and 1015.
f0 = Function[{file},Import[file,"TSV"]];
f1 = CompileToBinary[Function[{Typed[file,String]},Import[file,{"TSV","Real64"}]]];
Export a 1000x1000 numerical array in TSV format.
f0 = Function[{Typed[file,String]},Export[file,10.^RandomReal[{-15.,15.},{1000,1000}],"TSV"]];
f1 = CompileToBinary[f0];