Related
I have a large collection of data that is read into memory - temporarily, but necessary for the system.
I have been checking the performance of std::vector as well as std::unordered_map.
For std::vector I used a struct of type:
struct information{
std::string name;
unsigned int offset;
}
For std::unordered_map I used the std::string for the key and the unsigned int offset for the value.
If, let's say, 2 000 000 of these are loaded into memory, I tried the following and got these results:
std::vector:
On random string, never really larger than 32 characters if a reserve was called on the vector.
std::vector<information> vec;
vec.reserve(2500000);
The insertion
vec.push_back({dataName, offset});
is quite fast. Trying to find data is very slow though. The find was implemented like this:
auto it = std::find_if(vec.begin(), vec.end(), [&name](information &info) -> bool {return info.name == name); });
Which makes sense seeing that it is a large vector and the correct struct is found on a name compare. But it was extremely poor performance. The memory used was fine - I assume a part of the memory growth was due to the std::string size resizing.
My question on the vector implementation is: Is there a way to increase the look up time? I know that a vector can be sorted to increase your look-up time, but then you lose time in sorting the vector. Especially on a vector of this size.
std::unordered_map:
The insertion
std::unordered_map<std::string, unsigned int> unordMap;
unordMap.reserve(2500000);
unordMap.emplace(name, offset);
takes a very long time. When reserving space beforehand in an attempt to shorten the insertion time the following happens:
The memory at the end of insertion is a lot more when not calling reserve, without reserve the memory is still a lot more than the vector implementation. The reserve doesn't really improve insertion time.
Of course the look up is very fast. My question about the std::unordered_map is can the insertion time and memory usage be improved?
If neither of these can be done, then my next question will probably be quite obvious. Is there a way to get a result in between these two data structures? What is the best for large amounts of data?
struct information{
std::string name;
unsigned int offset;
information(information const&)=default;
information(information&&)=default;
information(std::string n, unsigned o):name(std::move(n)),offset(o),hash(std::hash<std::string>()(name)) {};
information():information("",0) {};
bool operator<( information const& o ) const {
return tie() < o.tie();
}
std::tuple<std::size_t, std::string const&> tie() const { return std::tie(hash, name); }
private:
std::size_t hash;
};
Use the above structure for your std::vector.
After adding all the data, std::sort it.
To find something matching name do this:
struct information_searcher {
struct helper {
std::tuple<std::size_t, std::string const&> data;
helper( std::string const& o ):data(std::hash<std::string>()(o), o) {};
helper( helper const& o ) = default;
helper( information const& o ):data(o.tie()) {}
bool operator<( helper const& o ) const { return data < o.data; }
};
bool operator()( helper lhs, helper rhs ) const { return lhs < rhs; }
};
information* get_info_by_name( std::string const& name ) {
auto range = std::equal_range( vec.begin(), vec.end(), information_searcher::helper(name), information_searcher{} );
if (range.first == range.second) {
return nullptr;
} else {
return &*range.first;
}
}
which is a nearly zero-overhead lookup.
What we do here is hash the strings (for fast comparison), falling back on std::string comparison if we have a collision.
information_searcher is a class that lets us search the data without having to create an information (which would require a wasteful allocation).
get_info_by_name returns a pointer -- nullptr if not found, and a pointer to the first element with that name otherwise.
Changing information.name is impolite, and makes the hash field incorrect.
This may use moderately more memory than the naive std::vector version.
In general, if your work consists of 'add a bunch of stuff to a table' then 'do a bunch of lookups', your best bet is to build a std::vector, sort it in a fast way, then use equal_range to do the lookups. map and unordered_map are optimized for lots of mixed inserts/deletes/etc.
vector is usually implemented as 'dynamic array' and should be the most memory-efficient.
with good reservation strategy it can have insert of O(1) = fast. Searching is O(n) = very bad.
You can help vector by sorting it (and if you first load then search than I think it would be best - std::sort + std::binary_search).
You can as well implement something like insert-sort using std::lower_bound. insert = O(log n) = good, search = O(log n) = good
map (ordered) can actually do the same work, but may as well be implemented using tree = less memory efficient, access as good as sorted vector (but maybe less reallocation, but in your case, sorted vector is still best)
unordered_map is usually imlemented using hash tables = some memory overhead but fast operations (insertion cannot be that fast as in unsorted vector, but should still be pretty fast). The problen with hashing is that it can be fast and even fastest, but may as well be the worst solution (in extreme conditions). The above structures (sorted vector and map/tree are stable, always behave the same - logaritmic complexity).
The problem with a large vector is the lookup time when you don't know the index of objects you want. One way to improve it is, as you stated, to keep an ordered vector and do binary search on it. That way, lookup time will not be of linear complexity but rather of logarithmic complexity, which saves up quite a lot of time with very large containers. This is the lookup used in std::map (the ordered one). You can do a similar binary search using std::lower_bound or std::equal_range on your std::vector.
The problem with a large unordered map is completely different : this kind of container use a hash function and modulus calculation in order to place the elements according to their keys in a standard array. So when you have a n elements in a std::unordered_map, it is very unlikely that you only need an n-elements-long array, because some indices will not be filled. You will use at least the greatest index produced by the hash-and-modulo. One way to improve memory usage as well as insertion time is to write your own hash function. But it might be hard depending on what kind of strings you are using.
Well, the optimal solution here would be to create std::map, which is logarithmic in complexity both in insertion and in lookup. Although, I don't see any reason why you wouldn't use std::vector. It is pretty fast when using quick sort to sort even 2M elements, especially if you do it once. std::binary_search is then really fast. Think it over if you need to make a lot of lookups between insertions.
I have an std::vector of floats that I want to not contain duplicates but the math that populates the vector isn't 100% precise. The vector has values that differ by a few hundredths but should be treated as the same point. For example here's some values in one of them:
...
X: -43.094505
X: -43.094501
X: -43.094498
...
What would be the best/most efficient way to remove duplicates from a vector like this.
First sort your vector using std::sort. Then use std::unique with a custom predicate to remove the duplicates.
std::unique(v.begin(), v.end(),
[](double l, double r) { return std::abs(l - r) < 0.01; });
// treats any numbers that differ by less than 0.01 as equal
Live demo
Sorting is always a good first step. Use std::sort().
Remove not sufficiently unique elements: std::unique().
Last step, call resize() and maybe also shrink_to_fit().
If you want to preserve the order, do the previous 3 steps on a copy (omit shrinking though).
Then use std::remove_if with a lambda, checking for existence of the element in the copy (binary search) (don't forget to remove it if found), and only retain elements if found in the copy.
I say std::sort() it, then go through it one by one and remove the values within certain margin.
You can have a separate write iterator to the same vector and one resize operation at the end - instead of calling erase() for each removed element or having another destination copy for increased performance and smaller memory usage.
If your vector cannot contain duplicates, it may be more appropriate to use an std::set. You can then use a custom comparison object to consider small changes as being inconsequential.
Hi you could comprare like this
bool isAlmostEquals(const double &f1, const double &f2)
{
double allowedDif = xxxx;
return (abs(f1 - f2) <= allowedDif);
}
but it depends of your compare range and the double precision is not on your side
if your vector is sorted you could use std::unique with the function as predicate
I would do the following:
Create a set<double>
go through your vector in a loop or using a functor
Round each element and insert into the set
Then you can swap your vector with an empty vector
Copy all elements from the set to the empty vector
The complexity of this approach will be n * log(n) but it's simpler and can be done in a few lines of code. The memory consumption will double from just storing the vector. In addition set consumes slightly more memory per each element than vector. However, you will destroy it after using.
std::vector<double> v;
v.push_back(-43.094505);
v.push_back(-43.094501);
v.push_back(-43.094498);
v.push_back(-45.093435);
std::set<double> s;
std::vector<double>::const_iterator it = v.begin();
for(;it != v.end(); ++it)
s.insert(floor(*it));
v.swap(std::vector<double>());
v.resize(s.size());
std::copy(s.begin(), s.end(), v.begin());
The problem with most answers so far is that you have an unusual "equality". If A and B are similar but not identical, you want to treat them as equal. Basically, A and A+epsilon still compare as equal, but A+2*epsilon does not (for some unspecified epsilon). Or, depending on your algorithm, A*(1+epsilon) does and A*(1+2*epsilon) does not.
That does mean that A+epsilon compares equal to A+2*epsilon. Thus A = B and B = C does not imply A = C. This breaks common assumptions in <algorithm>.
You can still sort the values, that is a sane thing to do. But you have to consider what to do with a long range of similar values in the result. If the range is long enough, the difference between the first and last can still be large. There's no simple answer.
I've recently been porting a Python application to C++, but am now at a loss as to how I can port a specific function. Here's the corresponding Python code:
def foo(a, b): # Where `a' is a list of strings, as is `b'
for x in a:
if not x in b:
return False
return True
I wish to have a similar function:
bool
foo (char* a[], char* b[])
{
// ...
}
What's the easiest way to do this? I've tried working with the STL algorithms, but can't seem to get them to work. For example, I currently have this (using the glib types):
gboolean
foo (gchar* a[], gchar* b[])
{
gboolean result;
std::sort (a, (a + (sizeof (a) / sizeof (*a))); // The second argument corresponds to the size of the array.
std::sort (b, (b + (sizeof (b) / sizeof (*b)));
result = std::includes (b, (b + (sizeof (b) / sizeof (*b))),
a, (a + (sizeof (a) / sizeof (*a))));
return result;
}
I'm more than willing to use features of C++11.
I'm just going to add a few comments to what others have stressed and give a better algorithm for what you want.
Do not use pointers here. Using pointers doesn't make it c++, it makes it bad coding. If you have a book that taught you c++ this way, throw it out. Just because a language has a feature, does not mean it is proper to use it anywhere you can. If you want to become a professional programmer, you need to learn to use the appropriate parts of your languages for any given action. When you need a data structure, use the one appropriate to your activity. Pointers aren't data structures, they are reference types used when you need an object with state lifetime - i.e. when an object is created on one asynchronous event and destroyed on another. If an object lives it's lifetime without any asynchronous wait, it can be modeled as a stack object and should be. Pointers should never be exposed to application code without being wrapped in an object, because standard operations (like new) throw exceptions, and pointers do not clean themselves up. In other words, pointers should always be used only inside classes and only when necessary to respond with dynamic created objects to external events to the class (which may be asynchronous).
Do not use arrays here. Arrays are simple homogeneous collection data types of stack lifetime of size known at compiletime. They are not meant for iteration. If you need an object that allows iteration, there are types that have built in facilities for this. To do it with an array, though, means you are keeping track of a size variable external to the array. It also means you are enforcing external to the array that the iteration will not extend past the last element using a newly formed condition each iteration (note this is different than just managing size - it is managing an invariant, the reason you make classes in the first place). You do not get to reuse standard algorithms, are fighting decay-to-pointer, and generally are making brittle code. Arrays are (again) useful only if they are encapsulated and used where the only requirement is random access into a simple type, without iteration.
Do not sort a vector here. This one is just odd, because it is not a good translation from your original problem, and I'm not sure where it came from. Don't optimise early, but don't pessimise early by choosing a bad algorithm either. The requirement here is to look for each string inside another collection of strings. A sorted vector is an invariant (so, again, think something that needs to be encapsulated) - you can use existing classes from libraries like boost or roll your own. However, a little bit better on average is to use a hash table. With amortised O(N) lookup (with N the size of a lookup string - remember it's amortised O(1) number of hash-compares, and for strings this O(N)), a natural first way to translate "look up a string" is to make an unordered_set<string> be your b in the algorithm. This changes the complexity of the algorithm from O(NM log P) (with N now the average size of strings in a, M the size of collection a and P the size of collection b), to O(NM). If the collection b grows large, this can be quite a savings.
In other words
gboolean foo(vector<string> const& a, unordered_set<string> const& b)
Note, you can now pass constant to the function. If you build your collections with their use in mind, then you often have potential extra savings down the line.
The point with this response is that you really should never get in the habit of writing code like that posted. It is a shame that there are a few really (really) bad books out there that teach coding with strings like this, and it is a real shame because there is no need to ever have code look that horrible. It fosters the idea that c++ is a tough language, when it has some really nice abstractions that do this easier and with better performance than many standard idioms in other languages. An example of a good book that teaches you how to use the power of the language up front, so you don't build bad habits, is "Accelerated C++" by Koenig and Moo.
But also, you should always think about the points made here, independent of the language you are using. You should never try to enforce invariants outside of encapsulation - that was the biggest source of savings of reuse found in Object Oriented Design. And you should always choose your data structures appropriate for their actual use. And whenever possible, use the power of the language you are using to your advantage, to keep you from having to reinvent the wheel. C++ already has string management and compare built in, it already has efficient lookup data structures. It has the power to make many tasks that you can describe simply coded simply, if you give the problem a little thought.
Your first problem is related to the way arrays are (not) handled in C++. Arrays live a kind of very fragile shadow existence where, if you as much as look at them in a funny way, they are converted into pointers. Your function doesn't take two pointers-to-arrays as you expect. It takes two pointers to pointers.
In other words, you lose all information about the size of the arrays. sizeof(a) doesn't give you the size of the array. It gives you the size of a pointer to a pointer.
So you have two options: the quick and dirty ad-hoc solution is to pass the array sizes explicitly:
gboolean foo (gchar** a, int a_size, gchar** b, int b_size)
Alternatively, and much nicer, you can use vectors instead of arrays:
gboolean foo (const std::vector<gchar*>& a, const std::vector<gchar*>& b)
Vectors are dynamically sized arrays, and as such, they know their size. a.size() will give you the number of elements in a vector. But they also have two convenient member functions, begin() and end(), designed to work with the standard library algorithms.
So, to sort a vector:
std::sort(a.begin(), a.end());
And likewise for std::includes.
Your second problem is that you don't operate on strings, but on char pointers. In other words, std::sort will sort by pointer address, rather than by string contents.
Again, you have two options:
If you insist on using char pointers instead of strings, you can specify a custom comparer for std::sort (using a lambda because you mentioned you were ok with them in a comment)
std::sort(a.begin(), a.end(), [](gchar* lhs, gchar* rhs) { return strcmp(lhs, rhs) < 0; });
Likewise, std::includes takes an optional fifth parameter used to compare elements. The same lambda could be used there.
Alternatively, you simply use std::string instead of your char pointers. Then the default comparer works:
gboolean
foo (const std::vector<std::string>& a, const std::vector<std::string>& b)
{
gboolean result;
std::sort (a.begin(), a.end());
std::sort (b.begin(), b.end());
result = std::includes (b.begin(), b.end(),
a.begin(), a.end());
return result;
}
Simpler, cleaner and safer.
The sort in the C++ version isn't working because it's sorting the pointer values (comparing them with std::less as it does with everything else). You can get around this by supplying a proper comparison functor. But why aren't you actually using std::string in the C++ code? The Python strings are real strings, so it makes sense to port them as real strings.
In your sample snippet your use of std::includes is pointless since it will use operator< to compare your elements. Unless you are storing the same pointers in both your arrays the operation will not yield the result you are looking for.
Comparing adresses is not the same thing as comparing the true content of your c-style-strings.
You'll also have to supply std::sort with the neccessary comparator, preferrably std::strcmp (wrapped in a functor).
It's currently suffering from the same problem as your use of std::includes, it's comparing addresses instead of the contents of your c-style-strings.
This whole "problem" could have been avoided by using std::strings and std::vectors.
Example snippet
#include <iostream>
#include <algorithm>
#include <cstring>
typedef char gchar;
gchar const * a1[5] = {
"hello", "world", "stack", "overflow", "internet"
};
gchar const * a2[] = {
"world", "internet", "hello"
};
...
int
main (int argc, char *argv[])
{
auto Sorter = [](gchar const* lhs, gchar const* rhs) {
return std::strcmp (lhs, rhs) < 0 ? true : false;
};
std::sort (a1, a1 + 5, Sorter);
std::sort (a2, a2 + 3, Sorter);
if (std::includes (a1, a1 + 5, a2, a2 + 3, Sorter)) {
std::cerr << "all elements in a2 was found in a1!\n";
} else {
std::cerr << "all elements in a2 wasn't found in a1!\n";
}
}
output
all elements in a2 was found in a1!
A naive transcription of the python version would be:
bool foo(std::vector<std::string> const &a,std::vector<std::string> const &b) {
for(auto &s : a)
if(end(b) == std::find(begin(b),end(b),s))
return false;
return true;
}
It turns out that sorting the input is very slow. (And wrong in the face of duplicate elements.) Even the naive function is generally much faster. Just goes to show again that premature optimization is the root of all evil.
Here's an unordered_set version that is usually somewhat faster than the naive version (or was for the values/usage patterns I tested):
bool foo(std::vector<std::string> const& a,std::unordered_set<std::string> const& b) {
for(auto &s:a)
if(b.count(s) < 1)
return false;
return true;
}
On the other hand, if the vectors are already sorted and b is relatively small ( less than around 200k for me ) then std::includes is very fast. So if you care about speed you just have to optimize for the data and usage pattern you're actually dealing with.
UPDATED:
I am working on a program whose performance is very critical. I have a vector of structs that are NOT sorted. I need to perform many search operations in this vector. So I decided to cache the vector data into a map like this:
std::map<long, int> myMap;
for (int i = 0; i < myVector.size(); ++i)
{
const Type& theType = myVector[i];
myMap[theType.key] = i;
}
When I search the map, the results of the rest of the program are much faster. However, the remaining bottleneck is the creation of the map itself (it is taking about 0.8 milliseconds on average to insert about 1,500 elements in it). I need to figure out a way to trim this time down. I am simply inserting a long as the key and an int as the value. I don't understand why it is taking this long.
Another idea I had was to create a copy of the vector (can't touch the original one) and somehow perform a faster sort than the std::sort (it takes way too long to sort it).
Edit:
Sorry everyone. I meant to say that I am creating a std::map where the key is a long and the value is an int. The long value is the struct's key value and the int is the index of the corresponding element in the vector.
Also, I did some more debugging and realized that the vector is not sorted at all. It's completely random. So doing something like a stable_sort isn't going to work out.
ANOTHER UPDATE:
Thanks everyone for the responses. I ended up creating a vector of pairs (std::vector of std::pair(long, int)). Then I sorted the vector by the long value. I created a custom comparator that only looked at the first part of the pair. Then I used lower_bound to search for the pair. Here's how I did it all:
typedef std::pair<long,int> Key2VectorIndexPairT;
typedef std::vector<Key2VectorIndexPairT> Key2VectorIndexPairVectorT;
bool Key2VectorIndexPairComparator(const Key2VectorIndexPairT& pair1, const Key2VectorIndexPairT& pair2)
{
return pair1.first < pair2.first;
}
...
Key2VectorIndexPairVectorT sortedVector;
sortedVector.reserve(originalVector.capacity());
// Assume "original" vector contains unsorted elements.
for (int i = 0; i < originalVector.size(); ++i)
{
const TheStruct& theStruct = originalVector[i];
sortedVector.insert(Key2VectorIndexPairT(theStruct.key, i));
}
std::sort(sortedVector.begin(), sortedVector.end(), Key2VectorIndexPairComparator);
...
const long keyToSearchFor = 20;
const Key2VectorIndexPairVectorT::const_iterator cItorKey2VectorIndexPairVector = std::lower_bound(sortedVector.begin(), sortedVector.end(), Key2VectorIndexPairT(keyToSearchFor, 0 /* Provide dummy index value for search */), Key2VectorIndexPairComparator);
if (cItorKey2VectorIndexPairVector->first == keyToSearchFor)
{
const int vectorIndex = cItorKey2VectorIndexPairVector->second;
const TheStruct& theStruct = originalVector[vectorIndex];
// Now do whatever you want...
}
else
{
// Could not find element...
}
This yielded a modest performance gain for me. Before the total time for my calculations were 3.75 milliseconds and now it is down to 2.5 milliseconds.
Both std::map and std::set are built on a binary tree and so adding items does dynamic memory allocation. If your map is largely static (i.e. initialized once at the start and then rarely or never has new items added or removed) you'd probably be better to use a sorted vector and a std::lower_bound to look up items using a binary search.
Maps take a lot of time for two reasons
You need to do a lot of memory allocation for your data storage
You need to perform O(n lg n) comparisons for the sort.
If you are just creating this as one batch, then throwing the whole map out, using a custom pool allocator may be a good idea here - eg, boost's pool_alloc. Custom allocators can also apply optimizations such as not actually deallocating any memory until the map's completely destroyed, etc.
Since your keys are integers, you may want to consider writing your own container based on a radix tree (on the bits of the key) as well. This may give you significantly improved performance, but since there is no STL implementation, you may need to write your own.
If you don't need to sort the data, use a hash table, such as std::unordered_map; these avoid the significant overhead needed for sorting data, and also can reduce the amount of memory allocation needed.
Finally, depending on the overall design of the program, it may be helpful to simply reuse the same map instead of recreating it over and over. Just delete and add keys as needed, rather than building a new vector, then building a new map. Again, this may not be possible in the context of your program, but if it is, it would definitely help you.
I suspect it's the memory management and tree rebalancing that's costing you here.
Obviously profiling may be able to help you pinpoint the issue.
I would suggest as a general idea to just copy the long/int data you need into another vector and since you said it's almost sorted, use stable_sort on it to finish the ordering. Then use lower_bound to locate the items in the sorted vector.
std::find is a linear scan(it has to be since it works on unsorted data). If you can sort(std::sort guaranties n log(n) behavior) the data then you can use std::binary_search to get log(n) searches. But as pointed out by others it may be copy time is the problem.
If keys are solid and short, perhaps try std::hash_map instead. From MSDN's page on hash_map Class:
The main advantage of hashing over sorting is greater efficiency; a
successful hashing performs insertions, deletions, and finds in
constant average time as compared with a time proportional to the
logarithm of the number of elements in the container for sorting
techniques.
Map creation can be a performance bottleneck (in the sense that it takes a measurable amount of time) if you're creating a large map and you're copying large chunks of data into it. You're also using the obvious (but suboptimal) way of inserting elements into a std::map - if you use something like:
myMap.insert(std::make_pair(theType.key, theType));
this should improve the insertion speed, but it will result in a slight change in behaviour if you encounter duplicate keys - using insert will result in values for duplicate keys being dropped, whereas using your method, the last element with the duplicate key will be inserted into the map.
I would also look into avoiding a making a copy of the data (for example by storing a pointer to it instead) if your profiling results determine that it's the copying of the element that is expensive. But for that you'll have to profile the code, IME guesstimates tend to be wrong...
Also, as a side note, you might want to look into storing the data in a std::set using custom comparator as your contains the key already. That however will not really result in a big speed up as constructing a set in this case is likely to be as expensive as inserting it into a map.
I'm not a C++ expert, but it seems that your problem stems from copying the Type instances, instead of a reference/pointer to the Type instances.
std::map<Type> myMap; // <-- this is wrong, since std::map requires two template parameters, not one
If you add elements to the map and they're not pointers, then I believe the copy constructor is invoked and that will certainly cause delays with a large data structure. Use the pointer instead:
std::map<KeyType, ObjectType*> myMap;
Furthermore, your example is a little confusing since you "insert" a value of type int in the map when you're expecting a value of type Type. I think you should assign the reference to the item, not the index.
myMap[theType.key] = &myVector[i];
Update:
The more I look at your example, the more confused I get. If you're using the std::map, then it should take two template types:
map<T1,T2> aMap;
So what are you REALLY mapping? map<Type, int> or something else?
It seems that you're using the Type.key member field as a key to the map (it's a valid idea), but unless key is of the same type as Type, then you can't use it as the key to the map. So is key an instance of Type??
Furthermore, you're mapping the current vector index to the key in the map, which indicates that you're just want the index to the vector so you can later access that index location fast. Is that what you want to do?
Update 2.0:
After reading your answer it seems that you're using std::map<long,int> and in that case there is no copying of the structure involved. Furthermore, you don't need to make a local reference to the object in the vector. If you just need to access the key, then access it by calling myVector[i].key.
Your building a copy of the table from the broken example you give, and not just a reference.
Why Can't I store references in an STL map in C++?
Whatever you store in the map it relies on you not changing the vector.
Try a lookup map only.
typedef vector<Type> Stuff;
Stuff myVector;
typedef std::map<long, *Type> LookupMap;
LookupMap myMap;
LookupMap::iterator hint = myMap.begin();
for (Stuff::iterator it = myVector.begin(); myVector.end() != it; ++it)
{
hint = myMap.insert(hint, std::make_pair(it->key, &*it));
}
Or perhaps drop the vector and just store it in the map??
Since your vector is already partially ordered, you may want to instead create an auxiliary array referencing (indices of) the elements in your original vector. Then you can sort the auxiliary array using Timsort which has good performance for partially sorted data (such as yours).
I think you've got some other problem. Creating a vector of 1500 <long, int> pairs, and sorting it based on the longs should take considerably less than 0.8 milliseconds (at least assuming we're talking about a reasonably modern, desktop/server type processor).
To try to get an idea of what we should see here, I did a quick bit of test code:
#include <vector>
#include <algorithm>
#include <time.h>
#include <iostream>
int main() {
const int size = 1500;
const int reps = 100;
std::vector<std::pair<long, int> > init;
std::vector<std::pair<long, int> > data;
long total = 0;
// Generate "original" array
for (int i=0; i<size; i++)
init.push_back(std::make_pair(rand(), i));
clock_t start = clock();
for (int i=0; i<reps; i++) {
// copy the original array
std::vector<std::pair<long, int> > data(init.begin(), init.end());
// sort the copy
std::sort(data.begin(), data.end());
// use data that depends on sort to prevent it being optimized away
total += data[10].first;
total += data[size-10].first;
}
clock_t stop = clock();
std::cout << "Ignore: " << total << "\n";
clock_t ticks = stop - start;
double seconds = ticks / (double)CLOCKS_PER_SEC;
double ms = seconds * 1000.0;
double ms_p_iter = ms / reps;
std::cout << ms_p_iter << " ms/iteration.";
return 0;
}
Running this on my somewhat "trailing edge" (~5 year-old) machine, I'm getting times around 0.1 ms/iteration. I'd expect searching in this (using std::lower_bound or std::upper_bound) to be somewhat faster than searching in an std::map as well (since the data in the vector is allocated contiguously, we can expect better locality of reference, leading to better cache usage).
Thanks everyone for the responses. I ended up creating a vector of pairs (std::vector of std::pair(long, int)). Then I sorted the vector by the long value. I created a custom comparator that only looked at the first part of the pair. Then I used lower_bound to search for the pair. Here's how I did it all:
typedef std::pair<long,int> Key2VectorIndexPairT;
typedef std::vector<Key2VectorIndexPairT> Key2VectorIndexPairVectorT;
bool Key2VectorIndexPairComparator(const Key2VectorIndexPairT& pair1, const Key2VectorIndexPairT& pair2)
{
return pair1.first < pair2.first;
}
...
Key2VectorIndexPairVectorT sortedVector;
sortedVector.reserve(originalVector.capacity());
// Assume "original" vector contains unsorted elements.
for (int i = 0; i < originalVector.size(); ++i)
{
const TheStruct& theStruct = originalVector[i];
sortedVector.insert(Key2VectorIndexPairT(theStruct.key, i));
}
std::sort(sortedVector.begin(), sortedVector.end(), Key2VectorIndexPairComparator);
...
const long keyToSearchFor = 20;
const Key2VectorIndexPairVectorT::const_iterator cItorKey2VectorIndexPairVector = std::lower_bound(sortedVector.begin(), sortedVector.end(), Key2VectorIndexPairT(keyToSearchFor, 0 /* Provide dummy index value for search */), Key2VectorIndexPairComparator);
if (cItorKey2VectorIndexPairVector->first == keyToSearchFor)
{
const int vectorIndex = cItorKey2VectorIndexPairVector->second;
const TheStruct& theStruct = originalVector[vectorIndex];
// Now do whatever you want...
}
else
{
// Could not find element...
}
This yielded a modest performance gain for me. Before the total time for my calculations were 3.75 milliseconds and now it is down to 2.5 milliseconds.
I have an array of 1000-2000 elements which are pointers to objects. I want to keep my array sorted and obviously I want to do this as quick as possible. They are sorted by a member and not allocated contiguously so assume a cache miss whenever I access the sort-by member.
Currently I'm sorting on-demand rather than on-add, but because of the cache misses and [presumably] non-inlining of the member access the inner loop of my quick sort is slow.
I'm doing tests and trying things now, (and see what the actual bottleneck is) but can anyone recommend a good alternative to speeding this up?
Should I do an insert-sort instead of quicksorting on-demand, or should I try and change my model to make the elements contigious and reduce cache misses?
OR, is there a sort algorithm I've not come accross which is good for data that is going to cache miss?
Edit: Maybe I worded this wrong :), I don't actually need my array sorted all the time (I'm not iterating through them sequentially for anything) I just need it sorted when I'm doing a binary chop to find a matching object, and doing that quicksort at that time (when I want to search) is currently my bottleneck, because of the cache misses and jumps (I'm using a < operator on my object, but I'm hoping that inlines in release)
Simple approach: insertion sort on every insert. Since your elements are not aligned in memory I'm guessing linked list. If so, then you could transform it into a linked list with jumps to the 10th element, the 100th and so on. This is kind of similar to the next suggestion.
Or you reorganize your container structure into a binary tree (or what every tree you like, B, B*, red-black, ...) and insert elements like you would insert them into a search tree.
Running a quicksort on each insertion is enormously inefficient. Doing a binary search and insert operation would likely be orders of magnitude faster. Using a binary search tree instead of a linear array would reduce the insert cost.
Edit: I missed that you were doing sort on extraction, not insert. Regardless, keeping things sorted amortizes sorting time over each insert, which almost has to be a win, unless you have a lot of inserts for each extraction.
If you want to keep the sort on-extract methodology, then maybe switch to merge sort, or another sort that has good performance for mostly-sorted data.
I think the best approach in your case would be changing your data structure to something logarithmic and rethinking your architecture. Because the bottleneck of your application is not that sorting thing, but the question why do you have to sort everything on each insert and try to compensate that by adding on-demand sort?.
Another thing you could try (that is based on your current implementation) is implementing an external pointer - something mapping table / function and sort those second keys, but I actually doubt it would benefit in this case.
Instead of the array of the pointers you may consider an array of structs which consist of both a pointer to your object and the sort criteria. That is:
Instead of
struct MyType {
// ...
int m_SomeField; // this is the sort criteria
};
std::vector<MyType*> arr;
You may do this:
strcut ArrayElement {
MyType* m_pObj; // the actual object
int m_SortCriteria; // should be always equal to the m_pObj->m_SomeField
};
std::vector<ArrayElement> arr;
You may also remove the m_SomeField field from your struct, if you only access your object via this array.
By such in order to sort your array you won't need to dereference m_pObj every iteration. Hence you'll utilize the cache.
Of course you must keep the m_SortCriteria always synchronized with m_SomeField of the object (in case you're editing it).
As you mention, you're going to have to do some profiling to determine if this is a bottleneck and if other approaches provide any relief.
Alternatives to using an array are std::set or std::multiset which are normally implemented as R-B binary trees, and so have good performance for most applications. You're going to have to weigh using them against the frequency of the sort-when-searched pattern you implemented.
In either case, I wouldn't recommend rolling-your-own sort or search unless you're interested in learning more about how it's done.
I would think that sorting on insertion would be better. We are talking O(log N) comparisons here, so say ceil( O(log N) ) + 1 retrieval of the data to sort with.
For 2000, it amounts to: 8
What's great about this is that you can buffer the data of the element to be inserted, that's how you only have 8 function calls to actually insert.
You may wish to look at some inlining, but do profile before you're sure THIS is the tight spot.
Nowadays you could use a set, either a std::set, if you have unique values in your structure member, or, std::multiset if you have duplicate values in you structure member.
One side note: The concept using pointers, is in general not advisable.
STL containers (if used correctly) give you nearly always an optimized performance.
Anyway. Please see some example code:
#include <iostream>
#include <array>
#include <algorithm>
#include <set>
#include <iterator>
// Demo data structure, whatever
struct Data {
int i{};
};
// -----------------------------------------------------------------------------------------
// All in the below section is executed during compile time. Not during runtime
// It will create an array to some thousands pointer
constexpr std::size_t DemoSize = 4000u;
using DemoPtrData = std::array<const Data*, DemoSize>;
using DemoData = std::array<Data, DemoSize>;
consteval DemoData createDemoData() {
DemoData dd{};
int k{};
for (Data& d : dd)
d.i = k++*2;
return dd;
}
constexpr DemoData demoData = createDemoData();
consteval DemoPtrData createDemoPtrData(const DemoData& dd) {
DemoPtrData dpd{};
for (std::size_t k{}; k < dpd.size(); ++k)
dpd[k] = &dd[k];
return dpd;
}
constexpr DemoPtrData dpd = createDemoPtrData(demoData);
// -----------------------------------------------------------------------------------------
struct Comp {bool operator () (const Data* d1, const Data* d2) const { return d1->i < d2->i; }};
using MySet = std::multiset<const Data*, Comp>;
int main() {
// Add some thousand pointers. Will be sorted according to struct member
MySet mySet{ dpd.begin(), dpd.end() };
// Extract a range of data. integer values between 42 and 52
const Data* p42 = dpd[21];
const Data* p52 = dpd[26];
// Show result
for (auto iptr = mySet.lower_bound(p42); iptr != mySet.upper_bound(p52); ++iptr)
std::cout << (*iptr)->i << '\n';
// Insert a new element
Data d1{ 47 };
mySet.insert(&d1);
// Show again
std::cout << "\n\n";
for (auto iptr = mySet.lower_bound(p42); iptr != mySet.upper_bound(p52); ++iptr)
std::cout << (*iptr)->i << '\n';
}