It's not really very inefficient. Compare:
julia> f(x, y, z) = return x^2, y^3, z^4;
julia> g(a, b, c) = f(a, b, c)[3];
julia> @code_llvm g(1.0, 1.0, 1.0)
define double @"julia_g;767954"(double, double, double) {
top:
%3 = call double @pow(double %2, double 4.000000e+00), !dbg !525
ret double %3, !dbg !525
}
julia> function g(a, b, c)
_, _, v = f(a, b, c)
v
end;
julia> @code_llvm g(1.0, 1.0, 1.0)
define double @"julia_g;767955"(double, double, double) {
top:
%3 = call double @pow(double %2, double 4.000000e+00), !dbg !528
ret double %3, !dbg !530
}
If the return values are bits types, the optimizer is smart enough that
there's no overhead to pulling unused variables out of tuples. (In this
case, f was inlined, so g doesn't even compute x^2 or y^3.) At present, _,
a, b = f(x, y, z) will actually be more efficient than a, b = f(x, y,
z)[2:3], since the latter will allocate a new tuple on the heap, although
that could probably be improved.
If the return values are not bits types, or if the return type of f is not
inferrable, there will be slight overhead to pulling an unused value out of
a tuple, but in the former case the time to allocate the unused return
values will be vastly greater, and in the latter case the type instability
would be a performance bottleneck no matter what.
Simon
On Wednesday, October 15, 2014 11:23:20 AM UTC-4, Gray Calhoun wrote:
>
> Hi everyone, if a function returns multiple values, is there an idiomatic
> way to discard all but (say) the 3rd value returned? For example:
>
> f(x, y, z) = return x^2, y^3, z^4
> _, _, v = f(1, 2, 3)
>
> expresses the intent to only use the third value returned, but surely is
> inefficient.
>
> Thanks!
>
