Search Results

Search found 3677 results on 148 pages for 'concurrent vector'.

Page 64/148 | < Previous Page | 60 61 62 63 64 65 66 67 68 69 70 71  | Next Page >

• Creating an object in the loop

  - by Jacob

  std::vector<double> C(4); for(int i = 0; i < 1000;++i) for(int j = 0; j < 2000; ++j) { C[0] = 1.0; C[1] = 1.0; C[2] = 1.0; C[3] = 1.0; } is much faster than for(int i = 0; i < 1000;++i) for(int j = 0; j < 2000; ++j) { std::vector<double> C(4); C[0] = 1.0; C[1] = 1.0; C[2] = 1.0; C[3] = 1.0; } I realize this happens because std::vector is repeatedly being created and instantiated in the loop, but I was under the impression this would be optimized away. Is it completely wrong to keep variables local in a loop whenever possible? I was under the (perhaps false) impression that this would provide optimization opportunities for the compiler. EDIT: I use VC++2005 (release mode) with full optimization (/Ox)

  Read the article

• Link compatibility between C++ and D

  - by Caspin

  D easily interfaces with C. D just as easily interfaces with C++, but (and it's a big but) the C++ needs to be extremely trivial. The code cannot use: namespaces templates multiple inheritance mix virtual with non-virtual methods more? I completely understand the inheritance restriction. The rest however, feel like artificial limitations. Now I don't want to be able to use std::vector<T> directly, but I would really like to be able to link with std::vector<int> as an externed template. The C++ interfacing page has this particularly depressing comment. D templates have little in common with C++ templates, and it is very unlikely that any sort of reasonable method could be found to express C++ templates in a link-compatible way with D. This means that the C++ STL, and C++ Boost, likely will never be accessible from D. Admittedly I'll probably never need std::vector while coding in D, but I'd love to use QT or boost. So what's the deal. Why is it so hard to express non-trivial C++ classes in D? Would it not be worth it to add some special annotations or something to express at least namespaces?

  Read the article

• Creating a new object destroys an older object with different name in C++

  - by Mikael

  First question here! So, I am having some problems with pointers in Visual C++ 2008. I'm writing a program which will control six cameras and do some processing on them so to clean things up I have created a Camera Manager class. This class handles all operations which will be carried out on all the cameras. Below this is a Camera class which interacts with each individual camera driver and does some basic image processing. Now, the idea is that when the manager is initialised it creates two cameras and adds them to a vector so that I can access them later. The catch here is that when I create the second camera (camera2) the first camera's destructor is called for some reason, which then disconnects the camera. Normally I'd assume that the problem is somewhere in the Camera class, but in this case everything works perfectly as long as I don't create the camera2 object. What's gone wrong? CameraManager.h: #include "stdafx.h" #include <vector> #include "Camera.h" class CameraManager{ std::vector<Camera> cameras; public: CameraManager(); ~CameraManager(); void CaptureAll(); void ShowAll(); }; CameraManager.cpp: #include "stdafx.h" #include "CameraManager.h" CameraManager::CameraManager() { printf("Camera Manager: Initializing\n"); [...] Camera *camera1 = new Camera(NodeInfo,1, -44,0,0); cameras.push_back(*camera1); // Adding the following two lines causes camera1's destructor to be called. Why? Camera *camera2 = new Camera(NodeInfo,0, 44,0,0); cameras.push_back(*camera2); printf("Camera Manager: Ready\n"); }

  Read the article

• Getting functions of inherited functions to be called

  - by wrongusername

  Let's say I have a base class Animal from which a class Cow inherits, and a Barn class containing an Animal vector, and let's say the Animal class has a virtual function scream(), which Cow overrides. With the following code: Animal.h #ifndef _ANIMAL_H #define _ANIMAL_H #include <iostream> using namespace std; class Animal { public: Animal() {}; virtual void scream() {cout << "aaaAAAAAAAAAAGHHHHHHHHHH!!! ahhh..." << endl;} }; #endif /* _ANIMAL_H */ Cow.h #ifndef _COW_H #define _COW_H #include "Animal.h" class Cow: public Animal { public: Cow() {} void scream() {cout << "MOOooooOOOOOOOO!!!" << endl;} }; #endif /* _COW_H */ Barn.h #ifndef _BARN_H #define _BARN_H #include "Animal.h" #include <vector> class Barn { std::vector<Animal> animals; public: Barn() {} void insertAnimal(Animal animal) {animals.push_back(animal);} void tortureAnimals() { for(int a = 0; a < animals.size(); a++) animals[a].scream(); } }; #endif /* _BARN_H */ and finally main.cpp #include <stdlib.h> #include "Barn.h" #include "Cow.h" #include "Chicken.h" /* * */ int main(int argc, char** argv) { Barn barn; barn.insertAnimal(Cow()); barn.tortureAnimals(); return (EXIT_SUCCESS); } I get this output: aaaAAAAAAAAAAGHHHHHHHHHH!!! ahhh... How should I code this to get MOOooooOOOOOOOO!!! (and whatever other classes inheriting Animal wants scream() to be) instead?

  Read the article

• C++ trouble with pointers to objects

  - by Zibd

  I have a class with a vector of pointers to objects. I've introduced some elements on this vector, and on my main file I've managed to print them and add others with no problems. Now I'm trying to remove an element from that vector and check to see if it's not NULL but it is not working. I'm filling it with on class Test: Other *a = new Other(1,1); Other *b = new Other(2,2); Other *c = new Other(3,3); v->push_back(a); v->push_back(b); v->push_back(c); And on my main file I have: Test t; (...) Other *pointer = t.vect->at(0); delete t.vect->at(0); t.vect->erase(t.vect->begin()); if (pointer == NULL) { cout << "Nothing here.."; } // Never enters here..

  Read the article

• SSE (SIMD extensions) support in gcc

  - by goldenmean

  Hi, I see a code as below: include "stdio.h" #define VECTOR_SIZE 4 typedef float v4sf __attribute__ ((vector_size(sizeof(float)*VECTOR_SIZE))); // vector of four single floats typedef union f4vector { v4sf v; float f[VECTOR_SIZE]; } f4vector; void print_vector (f4vector *v) { printf("%f,%f,%f,%f\n", v->f[0], v->f[1], v->f[2], v->f[3]); } int main() { union f4vector a, b, c; a.v = (v4sf){1.2, 2.3, 3.4, 4.5}; b.v = (v4sf){5., 6., 7., 8.}; c.v = a.v + b.v; print_vector(&a); print_vector(&b); print_vector(&c); } This code builds fine and works expectedly using gcc (it's inbuild SSE / MMX extensions and vector data types. this code is doing a SIMD vector addition using 4 single floats. I want to understand in detail what does each keyword/function call on this typedef line means and does: typedef float v4sf __attribute__ ((vector_size(sizeof(float)*VECTOR_SIZE))); What is the vector_size() function return; What is the __attribute__ keyword for Here is the float data type being type defined to vfsf type? I understand the rest part. thanks, -AD

  Read the article

• "end()" iterator for back inserters?

  - by Thanatos

  For iterators such as those returned from std::back_inserter(), is there something that can be used as an "end" iterator? This seems a little nonsensical at first, but I have an API which is: template<typename InputIterator, typename OutputIterator> void foo( InputIterator input_begin, InputIterator input_end, OutputIterator output_begin, OutputIterator output_end ); foo performs some operation on the input sequence, generating an output sequence. (Who's length is known to foo but may or may not be equal to the input sequence's length.) The taking of the output_end parameter is the odd part: std::copy doesn't do this, for example, and assumes you're not going to pass it garbage. foo does it to provide range checking: if you pass a range too small, it throws an exception, in the name of defensive programming. (Instead of potentially overwriting random bits in memory.) Now, say I want to pass foo a back inserter, specifically one from a std::vector which has no limit outside of memory constraints. I still need a "end" iterator - in this case, something that will never compare equal. (Or, if I had a std::vector but with a restriction on length, perhaps it might sometimes compare equal?) How do I go about doing this? I do have the ability to change foo's API - is it better to not check the range, and instead provide an alternate means to get the required output range? (Which would be needed anyways for raw arrays, but not required for back inserters into a vector.) This would seem less robust, but I'm struggling to make the "robust" (above) work.

  Read the article

• Java constructor using generic types

  - by Beer Me

  I'm having a hard time wrapping my head around Java generic types. Here's a simple piece of code that in my mind should work, but I'm obviously doing something wrong. Eclipse reports this error in BreweryList.java: The method breweryMethod() is undefined for the type <T> The idea is to fill a Vector with instances of objects that are a subclass of the Brewery class, so the invocation would be something like: BreweryList breweryList = new BreweryList(BrewerySubClass.class, list); BreweryList.java package com.beerme.test; import java.util.Vector; public class BreweryList<T extends Brewery> extends Vector<T> { public BreweryList(Class<T> c, Object[] j) { super(); for (int i = 0; i < j.length; i++) { T item = c.newInstance(); // breweryMethod() is an instance method // of Brewery, of which <T> is a subclass (right?) c.breweryMethod(); // "The method breweryMethod() is undefined // for the type <T>" } } } Brewery.java package com.beerme.test; public class Brewery { public Brewery() { super(); } protected void breweryMethod() { } } BrewerySubClass.java package com.beerme.test; public class BrewerySubClass extends Brewery { public BrewerySubClass() { super(); } public void brewerySubClassMethod() { } } I'm sure this is a complete-generics-noob question, but I'm stuck. Thanks for any tips!

  Read the article

• Matrix multiplication using pairs

  - by sc_ray

  Hi, I am looking into alternate ways to do a Matrix Multiplication. Instead of storing my matrix as a two-dimensional array, I am using a vector such as vector<pair<pair<int,int >,int > > to store my matrix. The pair within my pair (pair) stores my indices (i,j) and the other int stores the value for the given (i,j) pair. I thought I might have some luck implementing my sparse array this way. The problem is when I try to multiply this matrix with itself. If this was a 2-d array implementation, I would have multiplied the matrix as follows: for(i=0; i<row1; i++) { for(j=0; j<col1; j++) { C[i][j] = 0; for(k=0; k<col2; k++) C[i][j] += A[i][j] * A[j][k]; } } Can somebody point out a way to achieve the same result using my vector of 'pair of pairs'? Thanks

  Read the article

• Java constructor using generic types

  - by user37903

  I'm having a hard time wrapping my head around Java generic types. Here's a simple piece of code that in my mind should work, but I'm obviously doing something wrong. Eclipse reports this error in BreweryList.java: The method initBreweryFromObject() is undefined for the type <T> The idea is to fill a Vector with instances of objects that are a subclass of the Brewery class, so the invocation would be something like: BreweryList breweryList = new BreweryList(BrewerySubClass.class, list); BreweryList.java package com.beerme.test; import java.util.Vector; public class BreweryList<T extends Brewery> extends Vector<T> { public BreweryList(Class<T> c, Object[] j) { super(); for (int i = 0; i < j.length; i++) { T item = c.newInstance(); // initBreweryFromObject() is an instance method // of Brewery, of which <T> is a subclass (right?) c.initBreweryFromObject(); // "The method initBreweryFromObject() is undefined // for the type <T>" } } } Brewery.java package com.beerme.test; public class Brewery { public Brewery() { super(); } protected void breweryMethod() { } } BrewerySubClass.java package com.beerme.test; public class BrewerySubClass extends Brewery { public BrewerySubClass() { super(); } public void androidMethod() { } } I'm sure this is a complete-generics-noob question, but I'm stuck. Thanks for any tips!

  Read the article

• Optimizing a "set in a string list" to a "set as a matrix" operation

  - by Eric Fournier

  I have a set of strings which contain space-separated elements. I want to build a matrix which will tell me which elements were part of which strings. For example: "" "A B C" "D" "B D" Should give something like: A B C D 1 2 1 1 1 3 1 4 1 1 Now I've got a solution, but it runs slow as molasse, and I've run out of ideas on how to make it faster: reverseIn <- function(vector, value) { return(value %in% vector) } buildCategoryMatrix <- function(valueVector) { allClasses <- c() for(classVec in unique(valueVector)) { allClasses <- unique(c(allClasses, strsplit(classVec, " ", fixed=TRUE)[[1]])) } resMatrix <- matrix(ncol=0, nrow=length(valueVector)) splitValues <- strsplit(valueVector, " ", fixed=TRUE) for(cat in allClasses) { if(cat=="") { catIsPart <- (valueVector == "") } else { catIsPart <- sapply(splitValues, reverseIn, cat) } resMatrix <- cbind(resMatrix, catIsPart) } colnames(resMatrix) <- allClasses return(resMatrix) } Profiling the function gives me this: $by.self self.time self.pct total.time total.pct "match" 31.20 34.74 31.24 34.79 "FUN" 30.26 33.70 74.30 82.74 "lapply" 13.56 15.10 87.86 97.84 "%in%" 12.92 14.39 44.10 49.11 So my actual questions would be: - Where are the 33% spent in "FUN" coming from? - Would there be any way to speed up the %in% call? I tried turning the strings into factors prior to going into the loop so that I'd be matching numbers instead of strings, but that actually makes R crash. I've also tried going for partial matrix assignment (IE, resMatrix[i,x] <- 1) where i is the number of the string and x is the vector of factors. No dice there either, as it seems to keep on running infinitely.

  Read the article

• Another boost error

  - by user1676605

  On this code I get the enourmous error static void ParseTheCommandLine(int argc, char *argv[]) { int count; int seqNumber; namespace po = boost::program_options; std::string appName = boost::filesystem::basename(argv[0]); po::options_description desc("Generic options"); desc.add_options() ("version,v", "print version string") ("help", "produce help message") ("sequence-number", po::value<int>(&seqNumber)->default_value(0), "sequence number") ("pem-file", po::value< vector<string> >(), "pem file") ; po::positional_options_description p; p.add("pem-file", -1); po::variables_map vm; po::store(po::command_line_parser(argc, argv). options(desc).positional(p).run(), vm); po::notify(vm); if (vm.count("pem file")) { cout << "Pem files are: " << vm["pem-file"].as< vector<string> >() << "\n"; } cout << "Sequence number is " << seqNumber << "\n"; exit(1); ../../../FIXMarketDataCommandLineParameters/FIXMarketDataCommandLineParameters.hpp|98|error: no match for ‘operator<<’ in ‘std::operator<< [with _Traits = std::char_traits](((std::basic_ostream &)(& std::cout)), ((const char*)"Pem files are: ")) << ((const boost::program_options::variable_value*)vm.boost::program_options::variables_map::operator[](((const std::string&)(& std::basic_string, std::allocator (((const char*)"pem-file"), ((const std::allocator&)((const std::allocator*)(& std::allocator()))))))))-boost::program_options::variable_value::as with T = std::vector, std::allocator , std::allocator, std::allocator ’|

  Read the article

• sending address of a variable declared on the stack?

  - by kobac

  I have a doubt concerning declaring variables, their scope, and if their address could be sent to other functions even if they are declared on the stack? class A{ AA a; void f1(){ B b; aa.f2(&b); } }; class AA{ B* mb; f2(B* b){ mb = b; //... } }; Afterwards, I use my AA::mb pointer in the code. So things I would like to know are following. When the program exits A::f1() function, b variable since declared as a local variable and placed on the stack, can't be used anymore afterwards. What happens with the validity of the AA::mb pointer? It contains the address of the local variable which could not be available anymore, so the pointer isn't valid anymore? If B class is a std::<vector>, and AA::mb is not a pointer anymore to that vector, but a vector collection itself for example. I would like to avoid copying all of it's contents in AA::f2() to a member AA::mb in line mb = b. Which solution would you recommend since I can't assign a pointer to it, because it'll be destroyed when the program exits AA::f2()

  Read the article

• optimization math computation (multiplication and summing)

  - by wiso

  Suppose you want to compute the sum of the square of the differences of items: $\sum_{i=1}^{N-1} (x_i - x_{i+1})^2$, the simplest code (the input is std::vector<double> xs, the ouput sum2) is: double sum2 = 0.; double prev = xs[0]; for (vector::const_iterator i = xs.begin() + 1; i != xs.end(); ++i) { sum2 += (prev - (*i)) * (prev - (*i)); // only 1 - with compiler optimization prev = (*i); } I hope that the compiler do the optimization in the comment above. If N is the length of xs you have N-1 multiplications and 2N-3 sums (sums means + or -). Now suppose you know this variable: sum = $x_1^2 + x_N^2 + 2 sum_{i=2}^{N-1} x_i^2$ Expanding the binomial square: $sum_i^{N-1} (x_i-x_{i+1})^2 = sum - 2\sum_{i=1}^{N-1} x_i x_{i+1}$ so the code becomes: double sum2 = 0.; double prev = xs[0]; for (vector::const_iterator i = xs.begin() + 1; i != xs.end(); ++i) { sum2 += (*i) * prev; prev = (*i); } sum2 = -sum2 * 2. + sum; Here I have N multiplications and N-1 additions. In my case N is about 100. Well, compiling with g++ -O2 I got no speed up (I try calling the inlined function 2M times), why?

  Read the article

• Could this C cast to avoid a signed/unsigned comparison make any sense?

  - by sharptooth

  I'm reviewing a C++ project and see effectively the following: std::vector<SomeType> objects; //then later int size = (int)objects.size(); for( int i = 0; i < size; ++i ) { process( objects[i] ); } Here's what I see. std::vector::size() returns size_t that can be of some size not related to the size of int. Even if sizeof(int) == sizeof(size_t) int is signed and can't hold all possible values of size_t. So the code above could only process the lower part of a very long vector and contains a bug. That said I'm curious of why the author might have written this? My only guess is that first he omitted the (int) cast and the compiler emitted something like Visual C++ C4018 warning: warning C4018: '<' : signed/unsigned mismatch so the author though that the best way to avoid the compiler warning would be to simply cast the size_t to int thus making the compiler shut up. Is there any other possible sane reason for that C cast?

  Read the article

• Python - creating a list with 2 characteristics bug

  - by user2733911

  The goal is to create a list of 99 elements. All elements must be 1s or 0s. The first element must be a 1. There must be 7 1s in total. import random import math import time # constants determined through testing generation_constant = 0.96 def generate_candidate(): coin_vector = [] coin_vector.append(1) for i in range(0, 99): random_value = random.random() if (random_value > generation_constant): coin_vector.append(1) else: coin_vector.append(0) return coin_vector def validate_candidate(vector): vector_sum = sum(vector) sum_test = False if (vector_sum == 7): sum_test = True first_slot = vector[0] first_test = False if (first_slot == 1): first_test = True return (sum_test and first_test) vector1 = generate_candidate() while (validate_candidate(vector1) == False): vector1 = generate_candidate() print vector1, sum(vector1), validate_candidate(vector1) Most of the time, the output is correct, saying something like [1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0] 7 True but sometimes, the output is: [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] 2 False What exactly am I doing wrong?

  Read the article

• C#/.NET Little Wonders: Interlocked CompareExchange()

  - by James Michael Hare

  Once again, in this series of posts I look at the parts of the .NET Framework that may seem trivial, but can help improve your code by making it easier to write and maintain. The index of all my past little wonders posts can be found here. Two posts ago, I discussed the Interlocked Add(), Increment(), and Decrement() methods (here) for adding and subtracting values in a thread-safe, lightweight manner.  Then, last post I talked about the Interlocked Read() and Exchange() methods (here) for safely and efficiently reading and setting 32 or 64 bit values (or references).  This week, we’ll round out the discussion by talking about the Interlocked CompareExchange() method and how it can be put to use to exchange a value if the current value is what you expected it to be. Dirty reads can lead to bad results Many of the uses of Interlocked that we’ve explored so far have centered around either reading, setting, or adding values.  But what happens if you want to do something more complex such as setting a value based on the previous value in some manner? Perhaps you were creating an application that reads a current balance, applies a deposit, and then saves the new modified balance, where of course you’d want that to happen atomically.  If you read the balance, then go to save the new balance and between that time the previous balance has already changed, you’ll have an issue!  Think about it, if we read the current balance as $400, and we are applying a new deposit of $50.75, but meanwhile someone else deposits $200 and sets the total to $600, but then we write a total of $450.75 we’ve lost $200! Now, certainly for int and long values we can use Interlocked.Add() to handles these cases, and it works well for that.  But what if we want to work with doubles, for example?  Let’s say we wanted to add the numbers from 0 to 99,999 in parallel.  We could do this by spawning several parallel tasks to continuously add to a total: 1: double total = 0; 2:  3: Parallel.For(0, 10000, next => 4: { 5: total += next; 6: }); Were this run on one thread using a standard for loop, we’d expect an answer of 4,999,950,000 (the sum of all numbers from 0 to 99,999).  But when we run this in parallel as written above, we’ll likely get something far off.  The result of one of my runs, for example, was 1,281,880,740.  That is way off!  If this were banking software we’d be in big trouble with our clients.  So what happened?  The += operator is not atomic, it will read in the current value, add the result, then store it back into the total.  At any point in all of this another thread could read a “dirty” current total and accidentally “skip” our add.   So, to clean this up, we could use a lock to guarantee concurrency: 1: double total = 0.0; 2: object locker = new object(); 3:  4: Parallel.For(0, count, next => 5: { 6: lock (locker) 7: { 8: total += next; 9: } 10: }); Which will give us the correct result of 4,999,950,000.  One thing to note is that locking can be heavy, especially if the operation being locked over is trivial, or the life of the lock is a high percentage of the work being performed concurrently.  In the case above, the lock consumes pretty much all of the time of each parallel task – and the task being locked on is relatively trivial. Now, let me put in a disclaimer here before we go further: For most uses, lock is more than sufficient for your needs, and is often the simplest solution!    So, if lock is sufficient for most needs, why would we ever consider another solution?  The problem with locking is that it can suspend execution of your thread while it waits for the signal that the lock is free.  Moreover, if the operation being locked over is trivial, the lock can add a very high level of overhead.  This is why things like Interlocked.Increment() perform so well, instead of locking just to perform an increment, we perform the increment with an atomic, lockless method. As with all things performance related, it’s important to profile before jumping to the conclusion that you should optimize everything in your path.  If your profiling shows that locking is causing a high level of waiting in your application, then it’s time to consider lighter alternatives such as Interlocked. CompareExchange() – Exchange existing value if equal some value So let’s look at how we could use CompareExchange() to solve our problem above.  The general syntax of CompareExchange() is: T CompareExchange<T>(ref T location, T newValue, T expectedValue) If the value in location == expectedValue, then newValue is exchanged.  Either way, the value in location (before exchange) is returned. Actually, CompareExchange() is not one method, but a family of overloaded methods that can take int, long, float, double, pointers, or references.  It cannot take other value types (that is, can’t CompareExchange() two DateTime instances directly).  Also keep in mind that the version that takes any reference type (the generic overload) only checks for reference equality, it does not call any overridden Equals(). So how does this help us?  Well, we can grab the current total, and exchange the new value if total hasn’t changed.  This would look like this: 1: // grab the snapshot 2: double current = total; 3:  4: // if the total hasn’t changed since I grabbed the snapshot, then 5: // set it to the new total 6: Interlocked.CompareExchange(ref total, current + next, current); So what the code above says is: if the amount in total (1st arg) is the same as the amount in current (3rd arg), then set total to current + next (2nd arg).  This check and exchange pair is atomic (and thus thread-safe). This works if total is the same as our snapshot in current, but the problem, is what happens if they aren’t the same?  Well, we know that in either case we will get the previous value of total (before the exchange), back as a result.  Thus, we can test this against our snapshot to see if it was the value we expected: 1: // if the value returned is != current, then our snapshot must be out of date 2: // which means we didn't (and shouldn't) apply current + next 3: if (Interlocked.CompareExchange(ref total, current + next, current) != current) 4: { 5: // ooops, total was not equal to our snapshot in current, what should we do??? 6: } So what do we do if we fail?  That’s up to you and the problem you are trying to solve.  It’s possible you would decide to abort the whole transaction, or perhaps do a lightweight spin and try again.  Let’s try that: 1: double current = total; 2:  3: // make first attempt... 4: if (Interlocked.CompareExchange(ref total, current + i, current) != current) 5: { 6: // if we fail, go into a spin wait, spin, and try again until succeed 7: var spinner = new SpinWait(); 8:  9: do 10: { 11: spinner.SpinOnce(); 12: current = total; 13: } 14: while (Interlocked.CompareExchange(ref total, current + i, current) != current); 15: } 16:  This is not trivial code, but it illustrates a possible use of CompareExchange().  What we are doing is first checking to see if we succeed on the first try, and if so great!  If not, we create a SpinWait and then repeat the process of SpinOnce(), grab a fresh snapshot, and repeat until CompareExchnage() succeeds.  You may wonder why not a simple do-while here, and the reason it’s more efficient to only create the SpinWait until we absolutely know we need one, for optimal efficiency. Though not as simple (or maintainable) as a simple lock, this will perform better in many situations.  Comparing an unlocked (and wrong) version, a version using lock, and the Interlocked of the code, we get the following average times for multiple iterations of adding the sum of 100,000 numbers: 1: Unlocked money average time: 2.1 ms 2: Locked money average time: 5.1 ms 3: Interlocked money average time: 3 ms So the Interlocked.CompareExchange(), while heavier to code, came in lighter than the lock, offering a good compromise of safety and performance when we need to reduce contention. CompareExchange() - it’s not just for adding stuff… So that was one simple use of CompareExchange() in the context of adding double values -- which meant we couldn’t have used the simpler Interlocked.Add() -- but it has other uses as well. If you think about it, this really works anytime you want to create something new based on a current value without using a full lock.  For example, you could use it to create a simple lazy instantiation implementation.  In this case, we want to set the lazy instance only if the previous value was null: 1: public static class Lazy<T> where T : class, new() 2: { 3: private static T _instance; 4:  5: public static T Instance 6: { 7: get 8: { 9: // if current is null, we need to create new instance 10: if (_instance == null) 11: { 12: // attempt create, it will only set if previous was null 13: Interlocked.CompareExchange(ref _instance, new T(), (T)null); 14: } 15:  16: return _instance; 17: } 18: } 19: } So, if _instance == null, this will create a new T() and attempt to exchange it with _instance.  If _instance is not null, then it does nothing and we discard the new T() we created. This is a way to create lazy instances of a type where we are more concerned about locking overhead than creating an accidental duplicate which is not used.  In fact, the BCL implementation of Lazy<T> offers a similar thread-safety choice for Publication thread safety, where it will not guarantee only one instance was created, but it will guarantee that all readers get the same instance.  Another possible use would be in concurrent collections.  Let’s say, for example, that you are creating your own brand new super stack that uses a linked list paradigm and is “lock free”.  We could use Interlocked.CompareExchange() to be able to do a lockless Push() which could be more efficient in multi-threaded applications where several threads are pushing and popping on the stack concurrently. Yes, there are already concurrent collections in the BCL (in .NET 4.0 as part of the TPL), but it’s a fun exercise!  So let’s assume we have a node like this: 1: public sealed class Node<T> 2: { 3: // the data for this node 4: public T Data { get; set; } 5:  6: // the link to the next instance 7: internal Node<T> Next { get; set; } 8: } Then, perhaps, our stack’s Push() operation might look something like: 1: public sealed class SuperStack<T> 2: { 3: private volatile T _head; 4:  5: public void Push(T value) 6: { 7: var newNode = new Node<int> { Data = value, Next = _head }; 8:  9: if (Interlocked.CompareExchange(ref _head, newNode, newNode.Next) != newNode.Next) 10: { 11: var spinner = new SpinWait(); 12:  13: do 14: { 15: spinner.SpinOnce(); 16: newNode.Next = _head; 17: } 18: while (Interlocked.CompareExchange(ref _head, newNode, newNode.Next) != newNode.Next); 19: } 20: } 21:  22: // ... 23: } Notice a similar paradigm here as with adding our doubles before.  What we are doing is creating the new Node with the data to push, and with a Next value being the original node referenced by _head.  This will create our stack behavior (LIFO – Last In, First Out).  Now, we have to set _head to now refer to the newNode, but we must first make sure it hasn’t changed! So we check to see if _head has the same value we saved in our snapshot as newNode.Next, and if so, we set _head to newNode.  This is all done atomically, and the result is _head’s original value, as long as the original value was what we assumed it was with newNode.Next, then we are good and we set it without a lock!  If not, we SpinWait and try again. Once again, this is much lighter than locking in highly parallelized code with lots of contention.  If I compare the method above with a similar class using lock, I get the following results for pushing 100,000 items: 1: Locked SuperStack average time: 6 ms 2: Interlocked SuperStack average time: 4.5 ms So, once again, we can get more efficient than a lock, though there is the cost of added code complexity.  Fortunately for you, most of the concurrent collection you’d ever need are already created for you in the System.Collections.Concurrent (here) namespace – for more information, see my Little Wonders – The Concurent Collections Part 1 (here), Part 2 (here), and Part 3 (here). Summary We’ve seen before how the Interlocked class can be used to safely and efficiently add, increment, decrement, read, and exchange values in a multi-threaded environment.  In addition to these, Interlocked CompareExchange() can be used to perform more complex logic without the need of a lock when lock contention is a concern. The added efficiency, though, comes at the cost of more complex code.  As such, the standard lock is often sufficient for most thread-safety needs.  But if profiling indicates you spend a lot of time waiting for locks, or if you just need a lock for something simple such as an increment, decrement, read, exchange, etc., then consider using the Interlocked class’s methods to reduce wait. Technorati Tags: C#,CSharp,.NET,Little Wonders,Interlocked,CompareExchange,threading,concurrency

  Read the article

• Does the .NET Framework need to be reoptimized after upgrading to a new CPU microarchitecture?

  - by Louis

  I believe that the .NET Framework will optimize certain binaries targeting features specific to the machine it's installed on. After changing the CPU from an Intel Nehalem to a Haswell chip, should the optimization be run again manually? If so, what is the process for that? Between generations here are some notable additions: Westmere: AES instruction set Sandy Bridge: Advanced Vector Extensions Ivy Bridge: RdRand (hardware random number generator), F16C (16-bit Floating-point conversion instructions) Haswell: Haswell New Instructions (includes Advanced Vector Extensions 2 (AVX2), gather, BMI1, BMI2, ABM and FMA3 support) So my, albeit naive, thought process was that the optimizations could take advantage of these in general cases. For example, perhaps calls to the Random library could utilize the hardware-RNG on Ivy Bridge and later models.

  Read the article

• How do you traverse and store XML in Blackberry Java app?

  - by Greg

  I'm having a problem accessing the contents of an XML document. My goal is this: Take an XML source and parse it into a fair equivalent of an associative array, then store it as a persistable object. the xml is pretty simple: <root> <element> <category_id>1</category_id> <name>Cars</name> </element> <element> <category_id>2</category_id> <name>Boats</name> </element> </root> Basic java class below. I'm pretty much just calling save(xml) after http response above. Yes, the xml is properly formatted. import java.io.IOException; import java.util.Hashtable; import org.w3c.dom.Document; import org.w3c.dom.Node; import org.w3c.dom.NodeList; import java.util.Vector; import net.rim.device.api.system.PersistentObject; import net.rim.device.api.system.PersistentStore; import net.rim.device.api.xml.parsers.DocumentBuilder; import net.rim.device.api.xml.parsers.DocumentBuilderFactory; public class database{ private static PersistentObject storeVenue; static final long key = 0x2ba5f8081f7ef332L; public Hashtable hashtable; public Vector venue_list; String _node,_element; public database() { storeVenue = PersistentStore.getPersistentObject(key); } public void save(Document xml) { venue_list = new Vector(); storeVenue.setContents(venue_list); Hashtable categories = new Hashtable(); try{ DocumentBuilderFactory docBuilderFactory = DocumentBuilderFactory. newInstance(); DocumentBuilder docBuilder = docBuilderFactory.newDocumentBuilder(); docBuilder.isValidating(); xml.getDocumentElement ().normalize (); NodeList list=xml.getElementsByTagName("*"); _node=new String(); _element = new String(); for (int i=0;i<list.getLength();i++){ Node value=list.item(i).getChildNodes().item(0); _node=list.item(i).getNodeName(); _element=value.getNodeValue(); categories.put(_element, _node); } } catch (Exception e){ System.out.println(e.toString()); } venue_list.addElement(categories); storeVenue.commit(); } The code above is the work in progress, and is most likely heavily flawed. However, I have been at this for days now. I can never seem to get all child nodes, or the name / value pair. When I print out the vector as a string, I usually end up with results like this: [{ = root, = element}] and that's it. No "category_id", no "name" Ideally, I would end up with something like [{1 = cars, 2 = boats}] Any help is appreciated. Thanks

  Read the article

• Identifier is undefined

  - by hawk

  I wrote the following code in C++ using VS2012 Express. void ac_search( uint num_patterns, uint pattern_length, const char *patterns, uint num_records, uint record_length, const char *records, int *matches, Node* trie) { // Irrelevant code omitted. } vector<int> ac_benchmark_search( uint num_patterns, uint pattern_length, const char *patterns, uint num_records, uint record_length, const char *records, double &time) { // Prepare the container for the results vector<int> matches(num_records * num_patterns); Trie T; Node* trie = T.addWord(records, num_records, record_length); // error line ac_search(num_patterns, pattern_length, patterns, num_records, record_length, records, matches.data(), trie); // Irrelevant code omitted. return matches; } I get the error identifier "ac_search" is undefined at the function invoking line. I am a bit confused here. because the function ac_search is declared as a global (not inside any container). Why can't I call it at this place? Am I missing something? Update I tried ignore irrelevant code and then included it gradually and found that everything is fine until I include the outer loop of ac_search I get the aforementioned error. here is updated code of the function ac_search: void ac_cpu_string_search(uint num_patterns, uint pattern_length, const char *patterns, uint num_records, uint record_length, const char *records, int *matches, Node* trie) { // Loop over all records //for (uint record_number = 0; record_number < num_records; ++record_number) //{ // // Loop over all patterns for (uint pattern_number = 0; pattern_number < num_patterns; ++pattern_number) { // Execute string search const char *ptr_record = &records[record_number * record_length]; const char *ptr_match = std::strstr(ptr_record, &patterns[pattern_number * pattern_length]); // If pattern was found, then calculate offset, otherwise result is -1 if (ptr_match) { matches[record_number * num_patterns + pattern_number] = static_cast<int>(std::distance(ptr_record, ptr_match)); } else { matches[record_number * num_patterns + pattern_number] = -1; } // } //} } Update 2 I think the error has something to do with the function addWord which belongs to the class Trie. When I commented out this function, I did not get the error anymore. Node* Trie::addWord(const char *records, uint num_records, uint record_length) { // Loop over all records for (uint record_number = 0; record_number < num_records; ++record_number) { const char *ptr_record = &records[record_number * record_length]; string s = ptr_record; Node* current = root; if ( s.length() == 0 ) { current->setWordMarker(); // an empty word return; } for ( int i = 0; i < s.length(); i++ ) { Node* child = current->findChild(s[i]); if ( child != NULL ) { current = child; } else { Node* tmp = new Node(); tmp->setContent(s[i]); current->appendChild(tmp); current = tmp; } if ( i == s.length() - 1 ) current->setWordMarker(); } return current; } void ac_search( uint num_patterns, uint pattern_length, const char *patterns, uint num_records, uint record_length, const char *records, int *matches, Node* trie) { // Irrelevant code omitted. } vector<int> ac_benchmark_search( uint num_patterns, uint pattern_length, const char *patterns, uint num_records, uint record_length, const char *records, double &time) { // Prepare the container for the results vector<int> matches(num_records * num_patterns); Trie T; Node* trie = T.addWord(records, num_records, record_length); // error line ac_search(num_patterns, pattern_length, patterns, num_records, record_length, records, matches.data(), trie); // Irrelevant code omitted. return matches; }

  Read the article

• STL find performs bettern than hand-crafter loop

  - by dusha

  Hello all, I have some question. Given the following C++ code fragment: #include <boost/progress.hpp> #include <vector> #include <algorithm> #include <numeric> #include <iostream> struct incrementor { incrementor() : curr_() {} unsigned int operator()() { return curr_++; } private: unsigned int curr_; }; template<class Vec> char const* value_found(Vec const& v, typename Vec::const_iterator i) { return i==v.end() ? "no" : "yes"; } template<class Vec> typename Vec::const_iterator find1(Vec const& v, typename Vec::value_type val) { return find(v.begin(), v.end(), val); } template<class Vec> typename Vec::const_iterator find2(Vec const& v, typename Vec::value_type val) { for(typename Vec::const_iterator i=v.begin(), end=v.end(); i<end; ++i) if(*i==val) return i; return v.end(); } int main() { using namespace std; typedef vector<unsigned int>::const_iterator iter; vector<unsigned int> vec; vec.reserve(10000000); boost::progress_timer pt; generate_n(back_inserter(vec), vec.capacity(), incrementor()); //added this line, to avoid any doubts, that compiler is able to // guess the data is sorted random_shuffle(vec.begin(), vec.end()); cout << "value generation required: " << pt.elapsed() << endl; double d; pt.restart(); iter found=find1(vec, vec.capacity()); d=pt.elapsed(); cout << "first search required: " << d << endl; cout << "first search found value: " << value_found(vec, found)<< endl; pt.restart(); found=find2(vec, vec.capacity()); d=pt.elapsed(); cout << "second search required: " << d << endl; cout << "second search found value: " << value_found(vec, found)<< endl; return 0; } On my machine (Intel i7, Windows Vista) STL find (call via find1) runs about 10 times faster than the hand-crafted loop (call via find2). I first thought that Visual C++ performs some kind of vectorization (may be I am mistaken here), but as far as I can see assembly does not look the way it uses vectorization. Why is STL loop faster? Hand-crafted loop is identical to the loop from the STL-find body. I was asked to post program's output. Without shuffle: value generation required: 0.078 first search required: 0.008 first search found value: no second search required: 0.098 second search found value: no With shuffle (caching effects): value generation required: 1.454 first search required: 0.009 first search found value: no second search required: 0.044 second search found value: no Many thanks, dusha. P.S. I return the iterator and write out the result (found or not), because I would like to prevent compiler optimization, that it thinks the loop is not required at all. The searched value is obviously not in the vector.

  Read the article

• Can't get Jacobi algorithm to work in Objective-C

  - by Chris Long

  Hi, For some reason, I can't get this program to work. I've had other CS majors look at it and they can't figure it out either. This program performs the Jacobi algorithm (you can see step-by-step instructions and a MATLAB implementation here). BTW, it's different from the Wikipedia article of the same name. Since NSArray is one-dimensional, I added a method that makes it act like a two-dimensional C array. After running the Jacobi algorithm many times, the diagonal entries in the NSArray (i[0][0], i[1][1], etc.) are supposed to get bigger and the others approach 0. For some reason though, they all increase exponentially. For instance, i[2][4] should equal 0.0000009, not 9999999, while i[2][2] should be big. Thanks in advance, Chris NSArray+Matrix.m @implementation NSArray (Matrix) @dynamic offValue, transposed; - (double)offValue { double sum = 0.0; for ( MatrixItem *item in self ) if ( item.nonDiagonal ) sum += pow( item.value, 2.0 ); return sum; } - (NSMutableArray *)transposed { NSMutableArray *transpose = [[[NSMutableArray alloc] init] autorelease]; int i, j; for ( i = 0; i < 5; i++ ) { for ( j = 0; j < 5; j++ ) { [transpose addObject:[self objectAtRow:j andColumn:i]]; } } return transpose; } - (id)objectAtRow:(NSUInteger)row andColumn:(NSUInteger)column { NSUInteger index = 5 * row + column; return [self objectAtIndex:index]; } - (NSMutableArray *)multiplyWithMatrix:(NSArray *)array { NSMutableArray *result = [[NSMutableArray alloc] init]; int i = 0, j = 0, k = 0; double value; for ( i = 0; i < 5; i++ ) { value = 0.0; for ( j = 0; j < 5; j++ ) { for ( k = 0; k < 5; k++ ) { MatrixItem *firstItem = [self objectAtRow:i andColumn:k]; MatrixItem *secondItem = [array objectAtRow:k andColumn:j]; value += firstItem.value * secondItem.value; } MatrixItem *item = [[MatrixItem alloc] initWithValue:value]; item.row = i; item.column = j; [result addObject:item]; } } return result; } @end Jacobi_AlgorithmAppDelegate.m // ... - (void)jacobiAlgorithmWithEntry:(MatrixItem *)entry { MatrixItem *b11 = [matrix objectAtRow:entry.row andColumn:entry.row]; MatrixItem *b22 = [matrix objectAtRow:entry.column andColumn:entry.column]; double muPlus = ( b22.value + b11.value ) / 2.0; muPlus += sqrt( pow((b22.value - b11.value), 2.0) + 4.0 * pow(entry.value, 2.0) ); Vector *u1 = [[[Vector alloc] initWithX:(-1.0 * entry.value) andY:(b11.value - muPlus)] autorelease]; [u1 normalize]; Vector *u2 = [[[Vector alloc] initWithX:-u1.y andY:u1.x] autorelease]; NSMutableArray *g = [[[NSMutableArray alloc] init] autorelease]; for ( int i = 0; i <= 24; i++ ) { MatrixItem *item = [[[MatrixItem alloc] init] autorelease]; if ( i == 6*entry.row ) item.value = u1.x; else if ( i == 6*entry.column ) item.value = u2.y; else if ( i == ( 5*entry.row + entry.column ) || i == ( 5*entry.column + entry.row ) ) item.value = u1.y; else if ( i % 6 == 0 ) item.value = 1.0; else item.value = 0.0; [g addObject:item]; } NSMutableArray *firstResult = [[g.transposed multiplyWithMatrix:matrix] autorelease]; matrix = [firstResult multiplyWithMatrix:g]; } // ...

  Read the article

• c++ protected pointer member to the same class and access privileges

  - by aajmakin

  Hi, Example code is included at the bottom of the message. I'm puzzled about the protected access specifier in a class. I have define a class node which has a protected string member name string name; and a vector of node pointers vector args; Before I thought that a member function of node could not do args[0]-name but a program that does just this does compile and run. However, now I would like to inherit this class and access the name field in one of the args array pointers from this derived class args[0]-name but this does not compile. When I compile the example code below with the commented sections uncommented, the compiler reports: Compiler output: g++ test.cc -o test test.cc: In member function 'void foo::newnode::print_args2()': test.cc:22: error: 'std::string foo::node::name' is protected test.cc:61: error: within this context Compilation exited abnormally with code 1 at Thu Jun 17 12:40:12 Questions: Why can I access the name field of the node pointers in args in class node, because this is what I would excpect from a similarly defined private field in Java. How can I access those fields from the derived class. Example code: #include <iostream> #include <vector> using namespace std; namespace foo { class node; typedef std::vector<node*> nodes; class node { public: node (string _name); void print_args (); void add_node (node* a); protected: nodes args; string name; }; } foo::node::node (string _name) : args(0) { name = _name; } void foo::node::add_node (node* a) { args.push_back(a); } void foo::node::print_args () { for (int i = 0; i < args.size(); i++) { cout << "node " << i << ": " << args[i]->name << endl; } } // namespace foo // { // class newnode : public node // { // public: // newnode (string _name) : node(_name) {} // void print_args2 (); // protected: // }; // } // void foo::newnode::print_args2 () // { // for (int i = 0; i < args.size(); i++) // { // cout << "node " << i << ": " << args[i]->name << endl; // } // } int main (int argc, char** argv) { foo::node a ("a"); foo::node b ("b"); foo::node c ("c"); a.add_node (&b); a.add_node (&c); a.print_args (); // foo::newnode newa ("newa"); // foo::newnode newb ("newb"); // foo::newnode newc ("newc"); // newa.add_node (&newb); // newa.add_node (&newc); // newa.print_args2 (); return 0; }

  Read the article

• C++ Problem resolution - is it the best way to simulate a "tuple"?

  - by fbin

  Hi everyone! I've got the following problem: "Write a template function vectorMAXMIN() that will accept a vector and a number indicating the size of the vector and will return the max and the min values of the vector"... So i think in it... Create a class vector to avoid the "size" passing value and control the insertions and can get from this the max and min values... ( dunno if it's a good idea ) The problem is "how to return a tuple?" When i read the problem, i thought in a tuple to return "max, min values" is it correct? The code: #include <iostream> template < typename T > class _tuple { public: T _Max; T _Min; }; template < typename T > class _vector { public: _vector( int cnt = 0); ~_vector(); _tuple< T > get_tuple( void ); void insert( const T ); private: T *ptr; int cnt; int MAX; }; template < typename T > _vector< T >::_vector( int N ) { ptr = new T [N] ; MAX = N; cnt = 0; } template < typename T > _tuple<T> _vector< T >::get_tuple( void ) { _tuple< T > _mytuple; _mytuple._Max = ptr[0]; _mytuple._Min = ptr[0]; for( int i = 1; i < cnt; i++) { if( _mytuple._Max > ptr[i] ) _mytuple._Max = ptr[i]; if( _mytuple._Min < ptr[i] ) _mytuple._Min = ptr[i]; } return _mytuple; } template < typename T > void _vector< T >::insert( const T element) { if( cnt == MAX ) std::cerr << "Error: Out of range!" << std::endl; else { ptr[cnt] = element; cnt++; } } template < typename T > _vector< T >::~_vector() { delete [] ptr; } int main() { _vector< int > v; _tuple < int > t; v.insert(2); v.insert(1); v.insert(5); v.insert(0); v.insert(4); t = v.get_tuple(); std::cout << "MAX:" << t._Max; std::cout << " MIN:" << t._Min; return 0; }

  Read the article

• Referencing variables in a structure / C++

  - by user1628622

  Below, I provided a minimal example of code I created. I managed to get this code working, but I'm not sure if the practice being employed is sound. In essence, what I am trying to do is have the 'Parameter' class reference select elements in the 'States' class, so variables in States can be changed via Parameters. Questions I have: is the approach taken OK? If not, is there a better way to achieve what I am aiming for? Example code: struct VAR_TYPE{ public: bool is_fixed; // If is_fixed = true, then variable is a parameter double value; // Numerical value std::string name; // Description of variable (to identify it by name) }; struct NODE{ public: VAR_TYPE X, Y, Z; /* VAR_TYPE is a structure of primitive types */ }; class States{ private: std::vector <NODE_ptr> node; // shared ptr to struct NODE std::vector <PROP_DICTIONARY_ptr> property; // CAN NOT be part of Parameter std::vector <ELEMENT_ptr> element; // CAN NOT be part of Parameter public: /* ect */ void set_X_reference ( Parameter &T , int i ) { T.push_var( &node[i]->X ); } void set_Y_reference ( Parameter &T , int i ) { T.push_var( &node[i]->Y ); } void set_Z_reference ( Parameter &T , int i ) { T.push_var( &node[i]->Z ); } bool get_node_bool_X( int i ) { return node[i]->X.is_fixed; } // repeat for Y and Z }; class Parameter{ private: std::vector <VAR_TYPE*> var; public: /* ect */ }; int main(){ States S; Parameter P; /* Here I initialize and set S, and do other stuff */ // Now I assign components in States to Parameters for(int n=0 ; n<S.size_of_nodes() ; n++ ){ if ( S.get_node_bool_X(n)==true ){ S.set_X_reference ( P , n ); }; // repeat if statement for Y and Z }; /* Now P points selected to data in S, and I can * modify the contents of S through P */ return 0; }; Update The reason this issue cropped up is I am working with Fortran legacy code. To sum up this Fotran code - it's a numerical simulation of a flight vehicle. This code has a fairly rigid procedural framework one must work within, which comes with a pre-defined list of allowable Fortran types. The Fortran glue code can create an instance of a C++ object (in actuality, a reference from the perspective of Fortran), but is not aware what is contained in it (other means are used to extract C++ data into Fortran). The problem that I encountered is when a C++ module is dynamically linked to the Fortran glue code, C++ objects have to be initialized each instance the C++ code is called. This happens by virtue of how the Fortran template is defined. To avoid this cycle of re-initializing objects, I plan to use 'State' as a container class. The Fortran code allows a 'State' object, which has an arbitrary definition; but I plan to use it to harness all relevant information about the model. The idea is to use the Parameters class (which is exposed and updated by the Fortran code) to update variables in States.

  Read the article

< Previous Page | 60 61 62 63 64 65 66 67 68 69 70 71  | Next Page >