1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
|
COMMENT(see examples/vector*.cc, exprtempl.*)
Assume we are processing tt(std::vector) objects. Vectors may be assigned to
each other, but that's about it. We've seen (cf. section ref(VECTOR)) that its
member functions tend to operate on the current vector, but arithmetic
operations like addition, subtraction, multiplication and the like cannot
be applied to pairs of vectors.
Implementing the, e.g., addition operator for vectors is not difficult. If
tt(VecType) is our vector type, then implementing free functions like
tt(VecType &&operator+(VecType const &lhs, VecType const &rhs)) and
tt(VecType &&operator+(VecType &&lhs, VecType const &rhs))
performing the additions is a simple exercise (cf. chapter
ref(OVERLOADING)).
Now consider an expression like tt(one + two + three + four). It takes four
steps to compute this sum: first, tt(tmp = one) is computed, creating the
eventual return value. The vector tt(tmp) becomes the eventual return
value. Once it is available tt(tmp += two) is computed, followed by tt(tmp +=
three), and finally by tt(tmp += four) (of course
we shouldn't implement tt(std::vector::operator+=) as the std namespace is
off-limits to us, and we shouldn't derive a class from tt(std::vector)
offering tt(operator+=) according to Liskov's Substitution Principle
(cf. section ref(INHERITWHY)), but we could get around that.
COMMENT(by deriving a class offering operator+= in an anonymous namespace
and performing downcasts to the base class vectorthere.)
Here we simply assume tt(operator+=) is available).
Here's how we might implement tt(operator+=) for tt(VecType):
verb( VecType &VecType::operator+=(VecType const &rhs)
{
for (size_t idx = 0, end = size(); idx != end; ++idx)
(*this)[idx] += rhs[idx];
return *this;
})
Consider this implementation: once we add tt(VecType) objects and such
objects have tt(N) elements then we have to perform tt(2 * N) index
evaluations. When adding tt(k VecType) objects this adds up to tt(2 * N * k)
index expression evaluations (as eventually we also have to assign the
elements of the resulting temporary object to the destination object): lots
of index expression evaluations.
If instead we could manage to perform the evaluations `row by row', we
would only have to access each vector element only once (which in particular
applies to the temporary object). In that case, when adding k objects,
assigning the sums of their respective elements to a destination vector we
have to compute tt(N * (k + 1)) index expressions (tt(`k') for each of the
vectors, tt(`1') for the destination vector).
For tt(k == 1) the two methods are equally efficient in terms of index
computations. But that's not addition, that is assignment. So when adding any
number of vectors, assigning their sum to a destination vector using expression
templates
hi(expression template)hi(template: expression template)
is more efficient than the ordinary implementation of the addition
operator. We'll have a look at the design and implementation of expression
templates in the coming sections.
|
