mc4.htm

一个非常适合初学者入门的有关c++的文档

 More Effective C++ | Chapter 4: Efficiency Back to Item 15: Understand the costs of exception handling
Continue to Item 16: Remember the 80-20 rule.
Efficiency
I harbor a suspicion that someone has performed secret Pavlovian experiments on C++ software developers. How else can one explain the fact that when the word "efficiency" is mentioned, scores of programmers start to drool?
In fact, efficiency is no laughing matter. Programs that are too big or too slow fail to find acceptance, no matter how compelling their merits. This is perhaps as it should be. Software is supposed to help us do things better, and it's difficult to argue that slower is better, that demanding 32 megabytes of memory is better than requiring a mere 16, that chewing up 100 megabytes of disk space is better than swallowing only 50. Furthermore, though some programs take longer and use more memory because they perform more ambitious computations, too many programs can blame their sorry pace and bloated footprint on nothing more than bad design and slipshod programming.
Writing efficient programs in C++ starts with the recognition that C++ may well have nothing to do with any performance problems you've been having. If you want to write an efficient C++ program, you must first be able to write an efficient program. Too many developers overlook this simple truth. Yes, loops may be unrolled by hand and multiplications may be replaced by shift operations, but such micro-tuning leads nowhere if the higher-level algorithms you employ are inherently inefficient. Do you use quadratic algorithms when linear ones are available? Do you compute the same value over and over? Do you squander opportunities to reduce the average cost of expensive operations? If so, you can hardly be surprised if your programs are described like second-rate tourist attractions: worth a look, but only if you've got some extra time.
The material in this chapter attacks the topic of efficiency from two angles. The first is language-independent, focusing on things you can do in any programming language. C++ provides a particularly appealing implementation medium for these ideas, because its strong support for encapsulation makes it possible to replace inefficient class implementations with better algorithms and data structures that support the same interface.
The second focus is on C++ itself. High-performance algorithms and data structures are great, but sloppy implementation practices can reduce their effectiveness considerably. The most insidious mistake is both simple to make and hard to recognize: creating and destroying too many objects. Superfluous object constructions and destructions act like a hemorrhage on your program's performance, with precious clock-ticks bleeding away each time an unnecessary object is created and destroyed. This problem is so pervasive in C++ programs, I devote four separate items to describing where these objects come from and how you can eliminate them without compromising the correctness of your code.
Programs don't get big and slow only by creating too many objects. Other potholes on the road to high performance include library selection and implementations of language features. In the items that follow, I address these issues, too.
After reading the material in this chapter, you'll be familiar with several principles that can improve the performance of virtually any program you write, you'll know exactly how to prevent unnecessary objects from creeping into your software, and you'll have a keener awareness of how your compilers behave when generating executables.
It's been said that forewarned is forearmed. If so, think of the information that follows as preparation for battle. Back to Efficiency
Continue to Item 17: Consider using lazy evaluation
Item 16: Remember the 80-20 rule.
The 80-20 rule states that 80 percent of a program's resources are used by about 20 percent of the code: 80 percent of the runtime is spent in approximately 20 percent of the code; 80 percent of the memory is used by some 20 percent of the code; 80 percent of the disk accesses are performed for about 20 percent of the code; 80 percent of the maintenance effort is devoted to around 20 percent of the code. The rule has been repeatedly verified through examinations of countless machines, operating systems, and applications. The 80-20 rule is more than just a catchy phrase; it's a guideline about system performance that has both wide applicability and a solid empirical basis.
When considering the 80-20 rule, it's important not to get too hung up on numbers. Some people favor the more stringent 90-10 rule, and there's experimental evidence to back that, too. Whatever the precise numbers, the fundamental point is this: the overall performance of your software is almost always determined by a small part of its constituent code.
As a programmer striving to maximize your software's performance, the 80-20 rule both simplifies and complicates your life. On one hand, the 80-20 rule implies that most of the time you can produce code whose performance is, frankly, rather mediocre, because 80 percent of the time its efficiency doesn't affect the overall performance of the system you're working on. That may not do much for your ego, but it should reduce your stress level a little. On the other hand, the rule implies that if your software has a performance problem, you've got a tough job ahead of you, because you not only have to locate the small pockets of code that are causing the problem, you have to find ways to increase their performance dramatically. Of these tasks, the more troublesome is generally locating the bottlenecks. There are two fundamentally different ways to approach the matter: the way most people do it and the right way.
The way most people locate bottlenecks is to guess. Using experience, intuition, tarot cards and Ouija boards, rumors or worse, developer after developer solemnly proclaims that a program's efficiency problems can be traced to network delays, improperly tuned memory allocators, compilers that don't optimize aggressively enough, or some bonehead manager's refusal to permit assembly language for crucial inner loops. Such assessments are generally delivered with a condescending sneer, and usually both the sneerers and their prognostications are flat-out wrong.
Most programmers have lousy intuition about the performance characteristics of their programs, because program performance characteristics tend to be highly unintuitive. As a result, untold effort is poured into improving the efficiency of parts of programs that will never have a noticeable effect on their overall behavior. For example, fancy algorithms and data structures that minimize computation may be added to a program, but it's all for naught if the program is I/O-bound. Souped-up I/O libraries (see Item 23) may be substituted for the ones shipped with compilers, but there's not much point if the programs using them are CPU-bound.
That being the case, what do you do if you're faced with a slow program or one that uses too much memory? The 80-20 rule means that improving random parts of the program is unlikely to help very much. The fact that programs tend to have unintuitive performance characteristics means that trying to guess the causes of performance bottlenecks is unlikely to be much better than just improving random parts of your program. What, then, will work?
What will work is to empirically identify the 20 percent of your program that is causing you heartache, and the way to identify that horrid 20 percent is to use a program profiler. Not just any profiler will do, however. You want one that directly measures the resources you are interested in. For example, if your program is too slow, you want a profiler that tells you how much time is being spent in different parts of the program. That way you can focus on those places where a significant improvement in local efficiency will also yield a significant improvement in overall efficiency.
Profilers that tell you how many times each statement is executed or how many times each function is called are of limited utility. From a performance point of view, you do not care how many times a statement is executed or a function is called. It is, after all, rather rare to encounter a user of a program or a client of a library who complains that too many statements are being executed or too many functions are being called. If your software is fast enough, nobody cares how many statements are executed, and if it's too slow, nobody cares how few. All they care about is that they hate to wait, and if your program is making them do it, they hate you, too.
Still, knowing how often statements are executed or functions are called can sometimes yield insight into what your software is doing. If, for example, you think you're creating about a hundred objects of a particular type, it would certainly be worthwhile to discover that you're calling constructors in that class thousands of times. Furthermore, statement and function call counts can indirectly help you understand facets of your software's behavior you can't directly measure. If you have no direct way of measuring dynamic memory usage, for example, it may be helpful to know at least how often memory allocation and deallocation functions (e.g., operators new, new[], delete, and delete[] see Item 8) are called.
Of course, even the best of profilers is hostage to the data it's given to process. If you profile your program while it's processing unrepresentative input data, you're in no position to complain if the profiler leads you to fine-tune parts of your software the parts making up some 80 percent of it that have no bearing on its usual performance. Remember that a profiler can only tell you how a program behaved on a particular run (or set of runs), so if you profile a program using input data that is unrepresentative, you're going to get back a profile that is equally unrepresentative. That, in turn, is likely to lead to you to optimize your software's behavior for uncommon uses, and the overall impact on common uses may even be negative.
The best way to guard against these kinds of pathological results is to profile your software using as many data sets as possible. Moreover, you must ensure that each data set is representative of how the software is used by its clients (or at least its most important clients). It is usually easy to acquire representative data sets, because many clients are happy to let you use their data when profiling. After all, you'll then be tuning your software to meet their needs, and that can only be good for both of you. Back to Item 16: Remember the 80-20 rule
Continue to Item 18: Amortize the cost of expected computations
Item 17: Consider using lazy evaluation.
From the perspective of efficiency, the best computations are those you never perform at all. That's fine, but if you don't need to do something, why would you put code in your program to do it in the first place? And if you do need to do something, how can you possibly avoid executing the code that does it?
The key is to be lazy.
Remember when you were a child and your parents told you to clean your room? If you were anything like me, you'd say "Okay," then promptly go back to what you were doing. You would not clean your room. In fact, cleaning your room would be the last thing on your mind until you heard your parents coming down the hall to confirm that your room had, in fact, been cleaned. Then you'd sprint to your room and get to work as fast as you possibly could. If you were lucky, your parents would never check, and you'd avoid all the work cleaning your room normally entails.
It turns out that the same delay tactics that work for a five year old work for a C++ programmer. In Computer Science, however, we dignify such procrastination with the name lazy evaluation. When you employ lazy evaluation, you write your classes in such a way that they defer computations until the results of those computations are required. If the results are never required, the computations are never performed, and neither your software's clients nor your parents are any the wiser.
Perhaps you're wondering exactly what I'm talking about. Perhaps an example would help. Well, lazy evaluation is applicable in an enormous variety of application areas, so I'll describe four.
Reference Counting
Consider this code: 
class String { ... };                        // a string class (the standard
                                             // string type may be implemented
                                             // as described below, but it
                                             // doesn't have to be)

String s1 = "Hello";

String s2 = s1;                              // call String copy ctor


A common implementation for the String copy constructor would result in s1 and s2 each having its own copy of "Hello" after s2 is initialized with s1. Such a copy constructor would incur a relatively large expense, because it would have to make a copy of s1's value to give to s2, and that would typically entail allocating heap memory via the new operator (see Item 8) and calling strcpy to copy the data in s1 into the memory allocated by s2. This is eager evaluation: making a copy of s1 and putting it into s2 just because the String copy constructor was called. At this point, however, there has been no real need for s2 to have a copy of the value, because s2 hasn't been used yet.
The lazy approach is a lot less work. Instead of giving s2 a copy of s1's value, we have s2 share s1's value. All we have to do is a little bookkeeping so we know who's sharing what, and in return we save the cost of a call to new and the expense of copying anything. The fact that s1 and s2 are sharing a data structure is transparent to clients, and it certainly makes no difference in statements like the following, because they only read values, they don't write them: 
cout << s1;                              // read s1's value

cout << s1 + s2;                         // read s1's and s2's values


In fact, the only time the sharing of values makes a difference is when one or the other string is modified; then it's important that only one string be changed, not both. In this statement, s2.convertToUpperCase();


it's crucial that only s2's value be changed, not s1's also.
To handle statements like this, we have to implement String's convertToUpperCase function so that it makes a copy of s2's value and makes that value private to s2 before modifying it. Inside convertToUpperCase, we can be lazy no longer: we have to make a copy of s2's (shared) value for s2's private use. On the other hand, if s2 is never modified, we never have to make a private copy of its value. It can continue to share a value as long as it exists. If we're lucky, s2 will never be modified, in which case we'll never have to expend the effort to give it its own value.
The details on making this kind of value sharing work (including all the code) are provided in Item 29, but the idea is lazy evaluation: don't bother to make a copy of something until you really need one. Instead, be lazy use someone else's copy as long as you can get away with it. In some application areas, you can often get away with it forever.
Distinguishing Reads from Writes
Pursuing the example of reference-counting strings a bit further, we come upon a second way in which lazy evaluation can help us. Consider this code: 
String s = "Homer's Iliad";                  // Assume s is a
                                             // reference-counted string
...

cout << s[3];                         // call operator[] to read s[3]
s[3] = 'x';                           // call operator[] to write s[3]


The first call to operator[] is to read part of a string, but the second call is to perform a write. We'd like to be able to distinguish the read call from the write, because reading a reference-counted string is cheap, but writing to such a string may require splitting off a new copy of the string's value prior to the write.
This puts us in a difficult implementation position. To achieve what we want, we need to do different things inside operator[] (depending on whether it's being called to perform a read or a write). How can we determine whether operator[] has been called in a read or a write context? The brutal truth is that we can't. By using lazy evaluation and proxy classes as described in Item 30, however, we can defer the decision on whether to take read actions or write actions until we can determine which is correct.
Lazy Fetching
As a third example of lazy evaluation, imagine you've got a program that uses large objects containing many constituent fields. Such objects must persist across program runs, so they're stored in a database. Each object has a unique object identifier that can be used to retrieve the object from the database: 
class LargeObject {                        // large persistent objects
public:
  LargeObject(ObjectID id);                // restore object from disk

  const string& field1() const;            // value of field 1
  int field2() const;                      // value of field 2
  double field3() const;                   // ...
  const string& field4() const;
  const string& field5() const;
  ...

};


Now consider the cost of restoring a LargeObject from disk: void restoreAndProcessObject(ObjectID id)
{
  LargeObject object(id);                  // restore object

  ...

}


Because LargeObject instances are big, getting all the data for such an object might be a costly database operation, especially if the data must be retrieved from a remote database and pushed across a network. In some cases, the cost of reading all that data would be unnecessary. For example, consider this kind of application: void restoreAndProcessObject(ObjectID id)
{
  LargeObject object(id);

  if (object.field2() == 0) {
    cout << "Object " << id << ": null field2.\n";
  }
}


Here only the value of field2 is required, so any effort spent setting up the other fields is wasted.
The lazy approach to this problem is to read no data from disk when a LargeObject object is created. Instead, only the "shell" of an object is created, and data is retrieved from the database only when that particular data is needed inside the object. Here's one way to implement this kind of "demand-paged" object initialization: class LargeObject {
public:
  LargeObject(ObjectID id);

  const string& field1() const;
  int field2() const;
  double field3() const;
  const string& field4() const;
  ...

private:
  ObjectID oid;

  mutable string *field1Value;               // see below for a
  mutable int *field2Value;                  // discussion of "mutable"
  mutable double *field3Value;
  mutable string *field4Value;
  ...

};