How to compare and increment an atomic variable - multithreading

I want to know is it possible to perform comparison and increment of an atomic variable using a single atomic operation. This is what I have written so far(snippet code of a thread)
std::atomic<int> counter; //global variable
if(counter<25)
{
counter++;
}
else
{
//send serial/socket data
}
I know that I am doing wrong since atomic variable counter is accessed two times(once for getting the data and other for increment). But this may cause a problem if another thread performs some update operation on 'counter' after getting the variable value and before increment. So I want to know is it possible to do both these operations in a single shot. Also I don't want to use mutex.

In if(counter<25) counter++; there is a race condition between reading the counter and updating it (i.e. an atomic load followed by atomic load-modify-store).
It needs a compare-exchange loop to make sure the value read has not changed since. If it has changed, the operation needs to be retried.
Something like the following:
std::atomic<int> counter;
auto value = counter.load(std::memory_order_relaxed);
while(value < 25) {
if(counter.compare_exchange_weak(value, value + 1, std::memory_order_release, std::memory_order_relaxed))
break; // Succeeded incrementing counter.
// compare_exchange_weak failed because counter has changed.
// compare_exchange_weak reloaded value with the new value of counter.
// Retry.
}
if(!(value < 25))
// Failed to increment because counter is not less than 25.

closer code might be "something like":
const int GUARD = 25;
auto value = counter.load()
if (value < GUARD)
{
auto expectValue = value;
auto newValue = expectValue + 1;
for(;;)
{
if (counter.cmpxchg(expect,new))
{
break; // we did it!
}
newValue = expectValue + 1;
if (expectValue >= GUARD)
{
break; // someone else did it!
}
// someone else incremented, now we increment again
// cmpxchg updates expected with the new value
}
if (newValue <= GUARD)
{
// do the work you only want to do 25 times.
}
}

Related

How to recreate swapchain after vkAcquireNextImageKHR is VK_SUBOPTIMAL_KHR?

This vulkan tutorial discusses swapchain recreation:
You could also decide to [recreate the swapchain] that if the swap chain is suboptimal, but I've chosen to proceed anyway in that case because we've already acquired an image.
My question is: how would one recreate the swapchain and not proceed in this case of VK_SUBOPTIMAL_KHR?
To see what I mean, let's look at the tutorial's render function:
void drawFrame() {
vkWaitForFences(device, 1, &inFlightFences[currentFrame], VK_TRUE, UINT64_MAX);
uint32_t imageIndex;
VkResult result = vkAcquireNextImageKHR(device, swapChain, UINT64_MAX, imageAvailableSemaphores[currentFrame], VK_NULL_HANDLE, &imageIndex);
if (result == VK_ERROR_OUT_OF_DATE_KHR) {
recreateSwapChain();
return;
/* else if (result == VK_SUBOPTIMAL_KHR) { createSwapchain(); ??? } */
} else if (result != VK_SUCCESS && result != VK_SUBOPTIMAL_KHR) {
throw std::runtime_error("failed to acquire swap chain image!");
}
if (imagesInFlight[imageIndex] != VK_NULL_HANDLE) {
vkWaitForFences(device, 1, &imagesInFlight[imageIndex], VK_TRUE, UINT64_MAX);
}
imagesInFlight[imageIndex] = inFlightFences[currentFrame];
VkSubmitInfo submitInfo{};
submitInfo.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
VkSemaphore waitSemaphores[] = {imageAvailableSemaphores[currentFrame]};
VkPipelineStageFlags waitStages[] = {VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT};
submitInfo.waitSemaphoreCount = 1;
submitInfo.pWaitSemaphores = waitSemaphores;
submitInfo.pWaitDstStageMask = waitStages;
submitInfo.commandBufferCount = 1;
submitInfo.pCommandBuffers = &commandBuffers[imageIndex];
VkSemaphore signalSemaphores[] = {renderFinishedSemaphores[currentFrame]};
submitInfo.signalSemaphoreCount = 1;
submitInfo.pSignalSemaphores = signalSemaphores;
vkResetFences(device, 1, &inFlightFences[currentFrame]);
if (vkQueueSubmit(graphicsQueue, 1, &submitInfo, inFlightFences[currentFrame]) != VK_SUCCESS) {
throw std::runtime_error("failed to submit draw command buffer!");
}
VkPresentInfoKHR presentInfo{};
presentInfo.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR;
presentInfo.waitSemaphoreCount = 1;
presentInfo.pWaitSemaphores = signalSemaphores;
VkSwapchainKHR swapChains[] = {swapChain};
presentInfo.swapchainCount = 1;
presentInfo.pSwapchains = swapChains;
presentInfo.pImageIndices = &imageIndex;
result = vkQueuePresentKHR(presentQueue, &presentInfo);
if (result == VK_ERROR_OUT_OF_DATE_KHR || result == VK_SUBOPTIMAL_KHR || framebufferResized) {
framebufferResized = false;
recreateSwapChain();
} else if (result != VK_SUCCESS) {
throw std::runtime_error("failed to present swap chain image!");
}
currentFrame = (currentFrame + 1) % MAX_FRAMES_IN_FLIGHT;
}
The trouble is as follows:
vkAcquireImageKHR succeeds, signaling the semaphore and returning a valid, suboptimal image
Recreate the swapchain
We can't present the image from 1 with the swapchain from 2 due to VUID-VkPresentInfoKHR-pImageIndices-01430. We need to call vkAcquireImageKHR again to get a new image.
When we call vkAcquireImageKHR again, the semaphore is in the signaled state which is not allowed (VUID-vkAcquireNextImageKHR-semaphore-01286), we need to 'unsignal' it.
Is the best solution here to destroy and recreate the semaphore?
Ad 3: you can use the old images (and swapchain) if you properly use the oldSwapchain parameter when creating the new swapchain. Which is what I assume the tutorial suggests.
Anyway. What I do is that I paranoidly sanitize that toxic semaphore like this:
// cleanup dangerous semaphore with signal pending from vkAcquireNextImageKHR (tie it to a specific queue)
// https://github.com/KhronosGroup/Vulkan-Docs/issues/1059
void cleanupUnsafeSemaphore( VkQueue queue, VkSemaphore semaphore ){
const VkPipelineStageFlags psw = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT;
VkSubmitInfo submit_info = {};
submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
submit_info.waitSemaphoreCount = 1;
submit_info.pWaitSemaphores = &semaphore;
submit_info.pWaitDstStageMask;
vkQueueSubmit( queue, 1, &submit_info, VK_NULL_HANDLE );
}
After that the semaphore can be properly catched with a fence or vkQueueWaitIdle, and then destroyed or reused.
I just destroy them, because the new semaphore count might differ, and I don't really consider swapchain recreation a hotspot (and also I just use vkDeviceWaitIdle in such case).

Barrier thread synchronisation in pairs

I want to implement a function that blocks incoming threads until 2 threads of the same colour meet. Each thread is modelled by a unique id and colour. The function should return the id of the other thread it is matched to. I tried to use the idea for rendezvous and barrier to solve the question, which worked when there are only 2 threads at any one time, but when there are more than 2 threads, the id returned is incorrect. I am trying to solve for 1 colour first as the other 2 colors would have similar logic, but I am not sure what I need to change?
int pack_thread(int colour, int id) {
if (colour == 1) {
sem_wait(&d); map[0][count[0]] = id; count[0]++;
if (count[0] == 1) {
sem_post(&d);
sem_post(&a1); sem_wait(&a2); sem_post(&barr);
return map[0][1];
} else if (count[0] == 2) {
sem_post(&d);
sem_wait(&a1); sem_post(&a2);
return map[0][0];
} else {
sem_post(&d);
sem_wait(&barr);
map[0][1] = id;
count[0] = 0;
sem_post(&a2);sem_wait(&a1);
return map[0][0];
}
} else {
return 0;
}

Why is the following a memory leak? [duplicate]

I've got code that looks like this:
for (std::list<item*>::iterator i=items.begin();i!=items.end();i++)
{
bool isActive = (*i)->update();
//if (!isActive)
// items.remove(*i);
//else
other_code_involving(*i);
}
items.remove_if(CheckItemNotActive);
I'd like remove inactive items immediately after update them, inorder to avoid walking the list again. But if I add the commented-out lines, I get an error when I get to i++: "List iterator not incrementable". I tried some alternates which didn't increment in the for statement, but I couldn't get anything to work.
What's the best way to remove items as you are walking a std::list?
You have to increment the iterator first (with i++) and then remove the previous element (e.g., by using the returned value from i++). You can change the code to a while loop like so:
std::list<item*>::iterator i = items.begin();
while (i != items.end())
{
bool isActive = (*i)->update();
if (!isActive)
{
items.erase(i++); // alternatively, i = items.erase(i);
}
else
{
other_code_involving(*i);
++i;
}
}
You want to do:
i= items.erase(i);
That will correctly update the iterator to point to the location after the iterator you removed.
You need to do the combination of Kristo's answer and MSN's:
// Note: Using the pre-increment operator is preferred for iterators because
// there can be a performance gain.
//
// Note: As long as you are iterating from beginning to end, without inserting
// along the way you can safely save end once; otherwise get it at the
// top of each loop.
std::list< item * >::iterator iter = items.begin();
std::list< item * >::iterator end = items.end();
while (iter != end)
{
item * pItem = *iter;
if (pItem->update() == true)
{
other_code_involving(pItem);
++iter;
}
else
{
// BTW, who is deleting pItem, a.k.a. (*iter)?
iter = items.erase(iter);
}
}
Of course, the most efficient and SuperCool® STL savy thing would be something like this:
// This implementation of update executes other_code_involving(Item *) if
// this instance needs updating.
//
// This method returns true if this still needs future updates.
//
bool Item::update(void)
{
if (m_needsUpdates == true)
{
m_needsUpdates = other_code_involving(this);
}
return (m_needsUpdates);
}
// This call does everything the previous loop did!!! (Including the fact
// that it isn't deleting the items that are erased!)
items.remove_if(std::not1(std::mem_fun(&Item::update)));
I have sumup it, here is the three method with example:
1. using while loop
list<int> lst{4, 1, 2, 3, 5};
auto it = lst.begin();
while (it != lst.end()){
if((*it % 2) == 1){
it = lst.erase(it);// erase and go to next
} else{
++it; // go to next
}
}
for(auto it:lst)cout<<it<<" ";
cout<<endl; //4 2
2. using remove_if member funtion in list:
list<int> lst{4, 1, 2, 3, 5};
lst.remove_if([](int a){return a % 2 == 1;});
for(auto it:lst)cout<<it<<" ";
cout<<endl; //4 2
3. using std::remove_if funtion combining with erase member function:
list<int> lst{4, 1, 2, 3, 5};
lst.erase(std::remove_if(lst.begin(), lst.end(), [](int a){
return a % 2 == 1;
}), lst.end());
for(auto it:lst)cout<<it<<" ";
cout<<endl; //4 2
4. using for loop , should note update the iterator:
list<int> lst{4, 1, 2, 3, 5};
for(auto it = lst.begin(); it != lst.end();++it){
if ((*it % 2) == 1){
it = lst.erase(it); erase and go to next(erase will return the next iterator)
--it; // as it will be add again in for, so we go back one step
}
}
for(auto it:lst)cout<<it<<" ";
cout<<endl; //4 2
Use std::remove_if algorithm.
Edit:
Work with collections should be like:
prepare collection.
process collection.
Life will be easier if you won't mix this steps.
std::remove_if. or list::remove_if ( if you know that you work with list and not with the TCollection )
std::for_each
The alternative for loop version to Kristo's answer.
You lose some efficiency, you go backwards and then forward again when deleting but in exchange for the extra iterator increment you can have the iterator declared in the loop scope and the code looking a bit cleaner. What to choose depends on priorities of the moment.
The answer was totally out of time, I know...
typedef std::list<item*>::iterator item_iterator;
for(item_iterator i = items.begin(); i != items.end(); ++i)
{
bool isActive = (*i)->update();
if (!isActive)
{
items.erase(i--);
}
else
{
other_code_involving(*i);
}
}
Here's an example using a for loop that iterates the list and increments or revalidates the iterator in the event of an item being removed during traversal of the list.
for(auto i = items.begin(); i != items.end();)
{
if(bool isActive = (*i)->update())
{
other_code_involving(*i);
++i;
}
else
{
i = items.erase(i);
}
}
items.remove_if(CheckItemNotActive);
Removal invalidates only the iterators that point to the elements that are removed.
So in this case after removing *i , i is invalidated and you cannot do increment on it.
What you can do is first save the iterator of element that is to be removed , then increment the iterator and then remove the saved one.
If you think of the std::list like a queue, then you can dequeue and enqueue all the items that you want to keep, but only dequeue (and not enqueue) the item you want to remove. Here's an example where I want to remove 5 from a list containing the numbers 1-10...
std::list<int> myList;
int size = myList.size(); // The size needs to be saved to iterate through the whole thing
for (int i = 0; i < size; ++i)
{
int val = myList.back()
myList.pop_back() // dequeue
if (val != 5)
{
myList.push_front(val) // enqueue if not 5
}
}
myList will now only have numbers 1-4 and 6-10.
Iterating backwards avoids the effect of erasing an element on the remaining elements to be traversed:
typedef list<item*> list_t;
for ( list_t::iterator it = items.end() ; it != items.begin() ; ) {
--it;
bool remove = <determine whether to remove>
if ( remove ) {
items.erase( it );
}
}
PS: see this, e.g., regarding backward iteration.
PS2: I did not thoroughly tested if it handles well erasing elements at the ends.
You can write
std::list<item*>::iterator i = items.begin();
while (i != items.end())
{
bool isActive = (*i)->update();
if (!isActive) {
i = items.erase(i);
} else {
other_code_involving(*i);
i++;
}
}
You can write equivalent code with std::list::remove_if, which is less verbose and more explicit
items.remove_if([] (item*i) {
bool isActive = (*i)->update();
if (!isActive)
return true;
other_code_involving(*i);
return false;
});
The std::vector::erase std::remove_if idiom should be used when items is a vector instead of a list to keep compexity at O(n) - or in case you write generic code and items might be a container with no effective way to erase single items (like a vector)
items.erase(std::remove_if(begin(items), end(items), [] (item*i) {
bool isActive = (*i)->update();
if (!isActive)
return true;
other_code_involving(*i);
return false;
}));
do while loop, it's flexable and fast and easy to read and write.
auto textRegion = m_pdfTextRegions.begin();
while(textRegion != m_pdfTextRegions.end())
{
if ((*textRegion)->glyphs.empty())
{
m_pdfTextRegions.erase(textRegion);
textRegion = m_pdfTextRegions.begin();
}
else
textRegion++;
}
I'd like to share my method. This method also allows the insertion of the element to the back of the list during iteration
#include <iostream>
#include <list>
int main(int argc, char **argv) {
std::list<int> d;
for (int i = 0; i < 12; ++i) {
d.push_back(i);
}
auto it = d.begin();
int nelem = d.size(); // number of current elements
for (int ielem = 0; ielem < nelem; ++ielem) {
auto &i = *it;
if (i % 2 == 0) {
it = d.erase(it);
} else {
if (i % 3 == 0) {
d.push_back(3*i);
}
++it;
}
}
for (auto i : d) {
std::cout << i << ", ";
}
std::cout << std::endl;
// result should be: 1, 3, 5, 7, 9, 11, 9, 27,
return 0;
}
I think you have a bug there, I code this way:
for (std::list<CAudioChannel *>::iterator itAudioChannel = audioChannels.begin();
itAudioChannel != audioChannels.end(); )
{
CAudioChannel *audioChannel = *itAudioChannel;
std::list<CAudioChannel *>::iterator itCurrentAudioChannel = itAudioChannel;
itAudioChannel++;
if (audioChannel->destroyMe)
{
audioChannels.erase(itCurrentAudioChannel);
delete audioChannel;
continue;
}
audioChannel->Mix(outBuffer, numSamples);
}

In thread, when loop too much times, Invoke waste a lot of time

for (int i = 0; i < 100,000; i++)
{
threadEvent.Invoke(i, new EventArgs());// tell processbar value
}
threadEvent += new EventHandler(method_threadEvent);
void method_threadEvent(object sender, EventArgs e)
{
int nowValue = Convert.ToInt32(sender);
nowValueDelegate now = new nowValueDelegate(setNow);
this.Invoke(now, nowValue);
}
private void setNow(int nowValue)
{
this.progressBar1.Value = nowValue;
}
private delegate void nowValueDelegate(int nowValue);
in the loop i do nothing, but it also waste alot of time !
why threadEvent.Invoke spend so much time ?
Invoking is an expensive operation, because it has to cross thread boundaries.
It's best to reduce the amount of invokes, by for instance only updating the progress bar for each percentage of work you do, rather than for each iteration of the loop. That way, only 100 updates need to be processed, rather than one for each iteration.
First thing you need to do is to calculate or estimate the current progress.
For a typical loop
for (int i = 0; i < someValue; ++i)
{
... // Work here
}
A good estimate of progress is (i / someValue) * 100, which gives the percentage of the loop that has been completed. To update the progress to the UI thread only when the next percentage has been reached you could do something in the line of:
int percentCompleted = 0;
threadEvent.Invoke(percentCompleted, new EventArgs()); // Initial progressbar value
for (int i = 0; i < someValue; ++i)
{
int newlyCompleted = (i / someValue) * 100;
if (newlyCompleted > percentCompleted)
threadEvent.Invoke(percentCompleted, new EventArgs());
percentCompleted = newlyCompleted;
... // Work here
}
Now finally, you could use BeginInvoke instead of Invoke to make sure the worker thread doesn't wait for the threadEvent to complete (PostMessage behaviour). This works well here because there is no return value from threadEvent that you need.

How to compile a program in Befunge-93?

How do I save and compile a program in Befunge-93?
What file extension do I need to save the file as (like fileName.what)?
Then how do I compile it?
Befunge is typically an interpreted language. In fact it was designed to be difficult to compile because the programs are self-modifying (but that hasn't stopped people from trying to write a compiler).
You should save the file however you like (a .bf extension is typically used) and run the interpreter (bef), passing the filename as a command line argument.
You can see the usage instructions in the source code for the interpreter.
As it is an interpreted language, it usually won't be compiled. A C# Befunge 93 interpreter below, that will read code as an array[] of strings.
string befunge93(string[] p) {
for (int i=0;i<p.Length;i++)
p[i]=p[i].Replace("\"","ª"); // REMOVE QUOTATIONS TO MAKE THINGS EASIER
int l=0; int c=0; // current line and colum
var stack=new Stack<int>(new int[20000]); // initialize stack with many zeroes.
string res=""; // output limited to 1000 chars
int limit=100000; // max commands limited to 10^5
bool smode=false; // string mode
char direction='>';
//-------- MAIN LOOP ---------------
while (res.Length < 1000 && limit--> 0 )
{
var ch=p[l][c];
if (ch=='ª') // ", TOGGLE STRING MODE
smode = !smode;
else if (smode) // STRING MODE => PUSH CHAR
stack.Push((int)ch);
else if (ch==',') // OUTPUT CHAR FROM STACK
res+=(char)stack.Pop();
else if (new Regex(#"[><^v]").IsMatch(""+ch)) // CHANGE DIRECTION
direction = ch;
else if (new Regex(#"\d").IsMatch(""+ch)) // CHANGE DIRECTION
stack.Push(ch-'0');
else if (ch=='*') // MULTIPLICATION
stack.Push(stack.Pop()*stack.Pop());
else if (ch=='+') // SUM
stack.Push(stack.Pop()+stack.Pop());
else if (ch=='`') // GREATER THEN
stack.Push(stack.Pop()>=stack.Pop()?0:1);
else if (ch=='!') // NOT
stack.Push(stack.Pop()==0?1:0);
else if (ch=='.') // OUTPUT NUMBER
res+=$"{stack.Pop()} ";
else if (ch=='#') // JUMP NEXT COMMAND
move();
else if (ch=='$') // DISCARD ITEM FROM STACK
stack.Pop();
else if (ch=='\\') // SWAP
{ var a=stack.Pop(); var b=stack.Pop(); stack.Push(a); stack.Push(b); }
else if (ch=='%') // modulo
{ var a=stack.Pop(); var b=stack.Pop(); stack.Push(b%a); }
else if (ch=='-') // MINUS
{ var a=stack.Pop(); var b=stack.Pop(); stack.Push(b-a); }
else if (ch=='/') // DIVISION
{ var a=stack.Pop(); var b=stack.Pop(); stack.Push(b/a); }
else if (ch==':') // DUPLICATE
stack.Push(stack.First());
else if (ch=='_' || ch=='|') // CONDITIONALS: MOVE IF STACK ZERO, LEFT OTHERWISE
{
var last = stack.Pop();
if (ch=='_') direction=last==0?'>':'<'; // right if stack was zero. Left otherwise
else direction=last==0?'v':'^'; // ch=='|'
}
else if (ch==' '); // SPACE - DO NOTHING.
else return res;
move();
}
return res;
void move() // move cursor
{
switch(direction)
{
case '>':
if (++c==p[0].Length) c=0; break;
case '<':
if (--c==-1) c=p[0].Length-1; break;
case 'v':
if (++l==p.Length) l=0; break;
case '^':
if (--l==-1) l=p.Length-1; break;
default: break;
}
}
}

Resources