# Using Compile with vector expressions

Consider a vector

vec[x_,y_,z_]={x^0.5*y^0.1,y^0.3*z^0.1,z^2*x*y};


I need to calculate, say, a table with vector values for the coordinates

tabscord=RandomReal[{0, 1}, {10^7, 3}];


I would like to use Compile. This is my realization:

tabcompiled =
Hold@Compile[{{tabscord, _Real, 2}},
Table[vec[tabscord[[i]][], tabscord[[i]][],
tabscord[[i]][]], {i, 1, Length[tabscord], 1}],
CompilationTarget -> "C", RuntimeOptions -> "Speed"] /.
DownValues@vec//ReleaseHold


However, it says

CompiledFunction::cfse: Compiled expression {0.815274,0.188035,0.433407} should be a machine-size real number.

To avoid this problem, I need to define first the coordinates of the vector, say,

vecx[x_,y_,z_]=vec[x,y,z][];
vecy[x_,y_,z_]=vec[x,y,z][];
vecz[x_,y_,z_]=vec[x,y,z][];

And then

tabcompiled =
Hold@Compile[{{tabscord, _Real, 2}},
Table[{vecx[tabscord[[i]][], tabscord[[i]][],
tabscord[[i]][]],vecy[tabscord[[i]][], tabscord[[i]][],
tabscord[[i]][]],vecz[tabscord[[i]][], tabscord[[i]][],
tabscord[[i]][]]}, {i, 1, Length[tabscord], 1}],
CompilationTarget -> "C", RuntimeOptions -> "Speed"] /.
DownValues@vecx/.DownValues@vecy/.DownValues@vecz//ReleaseHold


Could you please tell me whether it is possible to avoid defining separate coordinates of vec if using Compile?

Using a pure function

vec = Function[{x,y,z},{x^0.5*y^0.1,y^0.3*z^0.1,z^2*x*y}];


and inlining works:

f = Compile[{{x,_Real,2}},
Table[vec[x[[i,1]],x[[i,2]],x[[i,3]]],{i,1,Length[tabscord]}],
CompilationTarget->"C",RuntimeOptions->"Speed",
CompilationOptions->{"InlineExternalDefinitions"->True}
];

tabscord=RandomReal[{0,1},{10^7,3}];
f[tabscord]
(* takes about 3 seconds *)


See this answer for inlining definitions.

I understand that the question is about Compile, but this is a case where it is easy to get fast code without compiling: Assuming OP's definitions

vec[x_,y_,z_]={x^0.5*y^0.1,y^0.3*z^0.1,z^2*x*y};
tabscord=RandomReal[{0,1},{10^7,3}];


one can use

Transpose[vec@@Transpose[tabscord]]
(* takes about 1.3 seconds *)


• The inner transpose turns the three big columns into three big rows, and vec@@ then means that vec is called with x equal to the entire first column of tabscord, y equal to the second column, z the third column.
• Therefore, z^2*x*y is called with x, y, z equal to vectors of length $$10^7$$ which we can be sure is already optimized and we do not have to compile.