C++ How to add two vectors asynchronously - c++11

Good morning everyone.
I am new to C++ 11 multithreading theme and trying to write down a code that add two vectors of equal sizes in asynchronous way.
This means, if I have two vectors:
vector<int> fisrt = {1, 2, 3};
vector<int> second = {3, 2, 1};
first += second; // first = {4, 4, 4}
I wrote Paginator class which splits vector on "pages" with appropriate page_size.
My idea of vectors addition in asynchronous way is next: split vectors on pages with choosen page_size, and add pages of first and second vectors in asynchronous way.
Paginator class implementation
template<class Iter>
class IterRange {
public:
explicit IterRange(Iter first, Iter last) : first_(first), last_(last) {}
Iter begin() { return first_; }
const Iter begin() const { return first_; }
Iter end() { return last_; }
const Iter end() const { return last_; }
private:
Iter first_;
Iter last_;
};
template<class Iter>
IterRange<Iter> MakeIterRange(Iter first, Iter last) {
return IterRange<Iter> {first, last };
}
template<class Iter>
class Paginator {
public:
Paginator(Iter first, Iter last, size_t page_size) : page_size_(page_size) {
size_t pages_count = static_cast<size_t> (floor((double)distance(first, last) / page_size_));
pages_.reserve(pages_count);
size_t page_id = 0u;
Iter begin_page = first;
for (page_id, begin_page; page_id < pages_count; ++page_id, begin_page += page_size_) {
pages_.push_back(MakeIterRange( begin_page, begin_page + page_size ));
}
// If some elements less than page_size_ is left
if (begin_page != last) {
pages_.push_back(MakeIterRange(begin_page, begin_page + distance(begin_page, last)));
}
}
auto begin() { return pages_.begin(); }
auto begin() const { return pages_.begin(); }
auto end() { return pages_.end(); }
auto end() const { return pages_.end(); }
private:
size_t page_size_;
vector<IterRange<Iter>> pages_;
};
template<class Iter>
Paginator<Iter> MakePaginator(Iter first, Iter last, size_t page_size) {
return{ first, last, page_size };
}
template<class Container> // And the same for non constant Container
auto Paginate(const Container & c, size_t page_size) {
return MakePaginator(begin(c), end(c), page_size);
}
This Paginate procedure is used in operator+= in my Matrix class.
Matrix class fields are:
Matrix sizes along Ox and Oy direction respectively: size_t nx_, size_t ny_.
Vector of (nx_ * ny_) size, which stores all elements in matrix: vector body_.
Operator += for Matrix
template
inline Matrix & Matrix::operator+=(const Matrix & other) {
size_t threads_numb = thread::hardware_concurrency();
size_t page_size = static_cast<size_t> (ceil((double)body_.size() / threads_numb));
vector<future<void>> futures;
auto page_1 = page::Paginate(body_, page_size);
auto page_2 = page::Paginate(other.body_, page_size);
auto it_2 = page_2.begin();
for (auto it = page_1.begin(); it != page_1.end(); ++it, ++it_2) {
futures.push_back(
async([it, it_2] { transform(it->begin(), it->end(), it_2->begin(), it->begin(), plus<T>()); })
);
}
return *this;
}
But as a result I get iterator out of range error! How could I fix this?
P.S. Sorry for bad representation of first string in operator +=. Could not fix this problem :(

Related

Runtime polymorphism c++11 and operator overloading

Let's say I am trying to implement some math vector class.
As vector interface will be used in multiple places: array based vector, matrices return columns and rows as vector interface objects and etc.
I would like to overload +,- operators for my vectors. Each operator should return new constructed object of some vector implementation class.
But as you know operator overloading should return a value or a reference. I can not return a value, as I need runtime polymorphism, so I am left with references. But to have a reference that does not die after the function call object should be created in the heap.
So how should I manage the situation?
P.S. I could create a shared_ptr and return a reference to containing value, but it does not look like a good practice.
typedef unsigned int vector_idx_t;
template <class T, vector_idx_t size>
class vector {
public:
virtual ~vector();
virtual T& operator[](const vector_idx_t idx) = 0;
virtual vector<T, size>& operator+ (const T& a) const = 0;
virtual vector<T, size>& operator- (const T& a) const = 0;
virtual vector<T, size>& operator* (const T& a) const = 0;
virtual vector<T, size>& operator/ (const T& a) const = 0;
virtual vector<T, size>& operator+ (const vector<T, size>& vec2) const = 0;
virtual vector<T, size>& operator- (const vector<T, size>& vec2) const = 0;
};
template <class T, vector_idx_t size>
class array_vector: public vector<T, size> {
private:
std::array<T, size> m_elements;
public:
array_vector();
array_vector(std::array<T, size> elements);
array_vector(const vector<T, size>& vec2);
array_vector(std::initializer_list<T> elems);
virtual ~array_vector();
virtual T& operator[](const vector_idx_t idx) {
return m_elements[idx];
}
virtual vector<T, size>& operator+ (const T& a) const {
std::array<T, size> e;
for (vector_idx_t i = 0; i < size; ++i) {
e[i] = m_elements[i] + a;
}
auto v = std::make_shared<array_vector<T, size>>(elems);
return *v;
}
};
I suggest a slight modification to your design for accommodating the polymorphic nature of the implementation.
Don't make vector polymorphic.
Use a Data class to contain the implementation specific details of vector.
Make Data polymorphic.
That will allow you to return vectors by value or by reference, as appropriate to an interface.
Polymorphism by subtype is not the answer to all problems. I understand what are you trying to do but I don't exactly understand why a polymorphic by template solution is not enough and you need to have virtual operators (which don't mix well at all with polymorphism by subtype).
You want to be able to define operations on mixed types of vectors so that you can compute results between real containers and proxy to containers.
This first of all should require that you have a basic final type that you need, a proxy to a matrix column is not a real container but rather a view of a container, so adding two of them should return a real container (eg. a container backed by an actual std::array?).
A similar design could be managed by something like
template<typename ContainerType, typename ElementType>
class vector_of : public ContainerType
{
public:
vector_of(const ContainerType& container) : ContainerType(container) { }
vector_of<ContainerType, ElementType> operator+(const ElementType& a) const
{
vector_of<ContainerType, ElementType> copy = vector_of<ContainerType,ElementType>(*this);
std::for_each(copy.begin(), copy.end(), [&a](ElementType& element) { element += a; });
}
template<typename T>
vector_of<ContainerType, ElementType> operator+(const vector_of<T, ElementType>& a) const
{
vector_of<ContainerType, ElementType> copy(*this);
auto it = copy.begin();
auto it2 = a.begin();
while (it != copy.end() && it2 != a.end())
{
*it += *it2;
++it;
++it2;
}
return copy;
}
};
The trick here is that operator+ is a template method which accepts a generic container of ElementType elements. The code assumes that these kind of containers provide a begin and end methods which return an iterator (which is a smart choice in any case because it works well with STL).
With you can do things like:
class MatrixRowProxy
{
private:
int* data;
size_t length;
public:
MatrixRowProxy(int* data, size_t length) : data(data), length(length) { }
int* begin() const { return data; }
int* end() const { return data + length; }
};
vector_of<std::array<int, 5>, int> base = vector_of<std::array<int, 5>, int>({ 1, 2, 3, 4, 5 });
vector_of<std::vector<int>, int> element = vector_of<std::vector<int>, int>({ 2, 3, 4, 5, 6 });
int* data = new int[5] { 10, 20, 30, 40, 50};
vector_of<MatrixRowProxy, int> proxy = vector_of<MatrixRowProxy, int>(MatrixRowProxy(data, 5));
auto result = base + element + proxy;
for (const auto& t : result)
std::cout << t << std::endl;
So you can add heterogeneous kinds of vectors without the need of any virtual method.
Of course these methods require to create a new resulting object in the methods. This is done by copying this into a new vector_of<ContainerType, ElementType>. Nothing prevents you from adding a third template argument like VectorFactory which takes care of this so that you could use vectors which are only wrappers also on LHS of such operators.

Recursive indexing for tensor in C++11

I have a tensor classes of rank N which wrap data stored in an array. For example, a rank-3 tensor would have dimensions (d0,d1,d2) and a unique element would be accessed with the multi-index (i0,i1,i2) from the underlying array of length d0*d1*d2. If d0=d1=d2=10, i0=1, i1=2, i2=3, then element 123 of the array would be accessed.
I've implemented a recursively defined class which computes single array index from the multi-index as follows:
template<size_t N>
class TensorIndex : TensorIndex<N-1> {
private:
size_t d;
public:
template<typename...Ds>
TensorIndex( size_t d0, Ds...ds ) : TensorIndex<N-1>( ds... ), d(d0) {}
template<typename...Is>
size_t index( size_t i0, Is...is ) {
return i0+d*TensorIndex<N-1>::index(is...);
}
};
template<>
struct TensorIndex<1> {
TensorIndex( size_t ) {}
size_t index( size_t i ) { return i; }
};
Which reverses the desired order.
TensorIndex<3> g(10,10,10);
std::cout << g.index(1,2,3) << std::endl;
outputs 321. What would be a simple way to reverse the order of the arguments for the constructor and index functions?
Edit:
I tried implementing using the suggested approach of reversing the variadic arguments, but this was suboptimal as it required reversing the arguments for both index and the constructor and the necessary helper functions for these two cases would appear slightly different. The initializer list answer looks more straightforward.
No need of recursion nor to reverse, you can use initializer-list to call an evaluation function that accumulates index from left to right. The function object called in the initalizer-list should have a non-void return type :
#include <cstddef>
#include <iostream>
using namespace std;
template<size_t N>
class TensorIndex {
public:
template<typename... Args>
TensorIndex(Args... args) : dims{static_cast<size_t>(args)...}
{
static_assert(sizeof...(Args) == N,
"incorrect number of arguments for TensorIndex constructor");
}
template<typename... Args>
size_t index(Args... args) {
static_assert(sizeof...(Args) == N,
"incorrect number of arguments for TensorIndex::index()");
IndexEval eval{dims};
Pass pass{eval(args)...}; // evaluate from left to right : initializer-list
return eval.get_res();
}
private:
const size_t dims[N];
class IndexEval {
size_t k = 0;
size_t res = 0;
const size_t* dims;
public:
IndexEval(const size_t* dims) : dims{dims} {}
size_t operator()(size_t i) {
return res = res * dims[k++] + i;
}
size_t get_res() const { return res; }
};
struct Pass {
template<typename... Args> Pass(Args...) {}
};
};
int main()
{
TensorIndex<3> g(10, 10, 10);
cout << g.index(1, 2, 3) << endl;
}

Vector proxy const and non const operation

I wrote a matrix template which is based on a std::vector. I am facing a problem when I want to proceed to an mathematical operation between two matrices: one which is a const reference and the other one which is not const and not a reference.
I am passing through a proxy to retrieve data stored in the std::vector:
template<typename T>
class Mat
{
public:
typedef T value_type;
typedef T& reference;
typedef const T& const_reference;
typedef typename std::vector<T>::size_type size_type;
private:
class Proxy
{
public:
Proxy(Mat& o, value_type i) : o(o), i(i)
{}
operator value_type()
{ return o.m[i]; }
reference operator=(const_reference other)
{ return o.m[i] = other; }
reference operator+=(const_reference other)
{ return o.m[i] += other; }
value_type operator*(const_reference other)
{ return o.m[i] * other; }
private:
Mat& o;
size_type i;
};
// Matrix methods and operator overloads.
Proxy operator[](size_type i)
{ return Proxy(*this, i); }
const Proxy operator[](size_type i) const
{ return Proxy(*this, i); }
Here is a sample from an extract of a test class which triggered an error:
void Test::buildMemoryMatrix(const Mat<int>& state)
{
Mat<int> t = state.getTransposed();
for(Mat<int>::size_type i = 0; i < state.rows(); ++i)
{
for(Mat<int>::size_type j = 0; j < t.cols(); ++j)
{
for(Mat<int>::size_type k = 0; k < state.cols(); ++k)
{
if(i == j)
{ continue; }
memory[i * memory.cols() + j] += state[i * state.cols() + k] * t[k * t.cols() + j];
}
}
}
}
Some precisions about the Mat<int> methods and Test class members:
The memory is a member of Test. It is represented as a Mat<int> object;
Mat::getTransposed() returns a transposed of a matrix;
Mat::rows() returns the number of rows of a matrix;
Mat::cols() returns the number of columns of a matrix
The following line give me some trouble:
memory[i * memory.cols() + j] += state[i * state.cols() + k] * t[k * t.cols() + j];
passing ‘const Mat::Proxy’ as ‘this’ argument discards qualifiers
[-fpermissive]
How to solve this problem?
Thanks for your answers.

How to force reorganization of a priority_queue?

I have a priority_queue that contains a vector with some objects.
std::priority_queue<std::shared_ptr<Foo>, std::vector<std::shared_ptr<Foo>>, foo_less> foo_queue;
It has a foo_queue function that will order the priority_queue.
Now, from outside of the priority_queue, I want to change some object value that must influenciate the ordering of the priority_queue.
My question is:
How can I set some kind of "refresh" that will trigger the priority_queue to run the foo_queue() in order to keep it orderes all the time?
Make your own priority queue using the standard heap algorithms and a
vector. When you want to change a key, find and remove that value from
the underlying vector and call make_heap on the vector. Alter the key and then
push it back onto the heap. So the cost is a linear search of the vector
to find the value and a call to make_heap (which I think is also linear).
#include <iostream>
#include <vector>
#include <algorithm>
template <class T, class Container = std::vector<T>,
class Compare = std::less<T> >
class my_priority_queue {
protected:
Container c;
Compare comp;
public:
explicit my_priority_queue(const Container& c_ = Container(),
const Compare& comp_ = Compare())
: c(c_), comp(comp_)
{
std::make_heap(c.begin(), c.end(), comp);
}
bool empty() const { return c.empty(); }
std::size_t size() const { return c.size(); }
const T& top() const { return c.front(); }
void push(const T& x)
{
c.push_back(x);
std::push_heap(c.begin(), c.end(), comp);
}
void pop()
{
std::pop_heap(c.begin(), c.end(), comp);
c.pop_back();
}
void remove(const T& x)
{
auto it = std::find(c.begin(), c.end(), x);
if (it != c.end()) {
c.erase(it);
std::make_heap(c.begin(), c.end(), comp);
}
}
};
class Foo {
int x_;
public:
Foo(int x) : x_(x) {}
bool operator<(const Foo& f) const { return x_ < f.x_; }
bool operator==(const Foo& f) const { return x_ == f.x_; }
int get() const { return x_; }
void set(int x) { x_ = x; }
};
int main() {
my_priority_queue<Foo> q;
for (auto x: {7, 1, 9, 5}) q.push(Foo(x));
while (!q.empty()) {
std::cout << q.top().get() << '\n';
q.pop();
}
std::cout << '\n';
for (auto x: {7, 1, 9, 5}) q.push(Foo(x));
Foo x = Foo(5);
q.remove(x);
x.set(8);
q.push(x);
while (!q.empty()) {
std::cout << q.top().get() << '\n';
q.pop();
}
}

Is there a way to find the C++ mangled name to use in GetProcAddress?

The common "solution" to use GetProcAddress with C++ is "extern "C", but that breaks overloading. Name mangling allows multiple functions to co-exist, as long as their signature differs. But is there a way to find these mangled names for GetProcAddress?
The VC++ compiler knows its own name mangling scheme, so why not use that? Inside template<typename T> T GetProcAddress(HMODULE h, const char* name), the macro __FUNCDNAME__ contains the mangled name of GetProcAddress. That includes the T part. So, inside GetProcAddress<void(*)(int), we have a substring with the mangled name of void(*)(int). From that, we can trivially derive the mangled name of void foo(int);
This code relies on the VC++ macro __FUNCDNAME__. For MinGW you'd need to base this on __PRETTY_FUNCTION__ instead.
FARPROC GetProcAddress_CppImpl(HMODULE h, const char* name, std::string const& Signature)
{
// The signature of T appears twice in the signature of T GetProcAddress<T>(HMODULE, const char*)
size_t len = Signature.find("##YA");
std::string templateParam = Signature.substr(0, len);
std::string returnType = Signature.substr(len+4);
returnType.resize(templateParam.size()); // Strip off our own arguments (HMODULE and const char*)
assert(templateParam == returnType);
// templateParam and returnType are _pointers_ to functions (P6), so adjust to function type (Y)
std::string funName = "?" + std::string(name) + "##Y" + templateParam.substr(2);
return ::GetProcAddress(h, funName.c_str());
}
template <typename T>
T GetProcAddress(HMODULE h, const char* name)
{
// Get our own signature. We use `const char* name` to keep it simple.
std::string Signature = __FUNCDNAME__ + 18; // Ignore prefix "??$GetProcAddress#"
return reinterpret_cast<T>(GetProcAddress_CppImpl(h, name, Signature));
}
// Showing the result
struct Dummy { };
__declspec(dllexport) void foo( const char* s)
{
std::cout << s;
}
__declspec(dllexport) void foo( int i, Dummy )
{
std::cout << "Overloaded foo(), got " << i << std::endl;
}
__declspec(dllexport) void foo( std::string const& s )
{
std::cout << "Overloaded foo(), got " << s << std::endl;
}
__declspec(dllexport) int foo( std::map<std::string, double> volatile& )
{
std::cout << "Overloaded foo(), complex type\n";
return 42;
}
int main()
{
HMODULE h = GetModuleHandleW(0);
foo("Hello, ");
auto pFoo1 = GetProcAddress<void (*)( const char*)>(h, "foo");
// This templated version of GetProcAddress is typesafe: You can't pass
// a float to pFoo1. That is a compile-time error.
pFoo1(" world\n");
auto pFoo2 = GetProcAddress<void (*)( int, Dummy )>(h, "foo");
pFoo2(42, Dummy()); // Again, typesafe.
auto pFoo3 = GetProcAddress<void (*)( std::string const& )>(h, "foo");
pFoo3("std::string overload\n");
auto pFoo4 = GetProcAddress<int (*)( std::map<std::string, double> volatile& )>(h, "foo");
// pFoo4 != NULL, this overload exists.
auto pFoo5 = GetProcAddress<void (*)( float )>(h, "foo");
// pFoo5==NULL - no such overload.
}
Use dumpbin /exports 'file.dll' to get the decorated / undecorated name of all the symbols.
It's impossible to do it just by using GetProcAddress. However, one way to do it would be to enumerate all the exported functions for that particular module, and do a pattern matching to find all the mangled names.
More specifically, refer to this answer here. The only change you will need to make would be to pass in TRUE for MappedAsImage parameter and the return value of GetModuleHandle for Base parameter to ImageDirectoryEntryToData function call.
void EnumerateExportedFunctions(HMODULE hModule, vector<string>& slListOfDllFunctions)
{
DWORD *dNameRVAs(0);
_IMAGE_EXPORT_DIRECTORY *ImageExportDirectory;
unsigned long cDirSize;
_LOADED_IMAGE LoadedImage;
string sName;
slListOfDllFunctions.clear();
ImageExportDirectory = (_IMAGE_EXPORT_DIRECTORY*)
ImageDirectoryEntryToData(hModule,
TRUE, IMAGE_DIRECTORY_ENTRY_EXPORT, &cDirSize);
if (ImageExportDirectory != NULL)
{
dNameRVAs = (DWORD *)ImageRvaToVa(LoadedImage.FileHeader,
LoadedImage.MappedAddress,
ImageExportDirectory->AddressOfNames, NULL);
for(size_t i = 0; i < ImageExportDirectory->NumberOfNames; i++)
{
sName = (char *)ImageRvaToVa(LoadedImage.FileHeader,
LoadedImage.MappedAddress,
dNameRVAs[i], NULL);
slListOfDllFunctions.push_back(sName);
}
}
}
I can't quite fathom why you'd ever want/need a constexpr version of MSalters' solution, but here it is, complete with namespace mangling. Use as
using F = int(double*);
constexpr auto f = mangled::name<F>([]{ return "foo::bar::frobnicate"; });
constexpr const char* cstr = f.data();
where F is the function signature and foo::bar::frobnicate is the (possibly qualified) name of the function.
#include<string_view>
#include<array>
namespace mangled
{
namespace detail
{
template<typename F>
inline constexpr std::string_view suffix()
{
auto str = std::string_view(__FUNCDNAME__);
return str.substr(14, str.size() - 87);
}
template<typename L>
struct constexpr_string
{
constexpr constexpr_string(L) {}
static constexpr std::string_view data = L{}();
};
template<typename Name>
inline constexpr int qualifiers()
{
int i = -2, count = -1;
while(i != std::string_view::npos)
{
i = Name::data.find("::", i + 2);
count++;
}
return count;
}
template<typename Name>
inline constexpr auto split()
{
std::array<std::string_view, qualifiers<Name>() + 1> arr = {};
int prev = -2;
for(int i = arr.size() - 1; i > 0; i--)
{
int cur = Name::data.find("::", prev + 2);
arr[i] = Name::data.substr(prev + 2, cur - prev - 2);
prev = cur;
}
arr[0] = Name::data.substr(prev + 2);
return arr;
}
template<typename F, typename Name>
struct name_builder
{
static constexpr auto suf = detail::suffix<F>();
static constexpr auto toks = split<Name>();
static constexpr auto len = Name::data.size() + suf.size() - toks.size() + 6;
static constexpr auto str = [] {
std::array<char, len> arr = {};
arr[0] = '?';
int i = 1;
for(int t = 0; t < toks.size(); t++)
{
if(t > 0)
arr[i++] = '#';
for(auto c : toks[t])
arr[i++] = c;
}
arr[i++] = '#';
arr[i++] = '#';
arr[i++] = 'Y';
for(auto c : suf)
arr[i++] = c;
return arr;
}();
};
}
template<typename F, typename LambdaString>
inline constexpr std::string_view name(LambdaString)
{
using Cs = detail::constexpr_string<LambdaString>;
using N = detail::name_builder<F, Cs>;
return {N::str.data(), N::len};
}
}
GodBolt

Resources