Last week I was debugging a Windows hook function, and I discovered a side effect of installing a hook:

This article is a follow-up on a previous article: 'Manipulating LabVIEW arrays in C++' in which

Nice series of posts from TechEd Barcelona: http://msmvps.com/blogs/vandooren/archive/2006/11/07/Tech_2D00_Ed-developers-Barcelona_3A00_-Monday.aspx

Nice series of posts from TechEd Barcelona (including reviews of some VC++ sessions): http://msmvps.com/blogs/vandooren/archive/2006/11/07/Tech_2D00_Ed-developers-Barcelona_3A00_-Monday.aspx

Nice series of posts from TechEd Barcelona (including reviews of some VC++ sessions): http://msmvps.com/blogs/vandooren/archive/2006/11/07/Tech_2D00_Ed-developers-Barcelona_3A00_-Monday.aspx

Hello!

Your solution is usefull also to solve an old problem here in my lab.  I'm new in C#. I have a problem with your sourcecode.  My compiler (Borland Dev.Studio) can't load the:

using System.Collections.Generic;

Any ideas?

 regards, Damjan

KRIZMANCIC at TASC dot INFM dot it

VC2005 inserts appropriate memory barriers for volatile accesses when you use a CPU that requires them. The code should be safe for all CPUs with VC2005. This is also true for Intel C++ (although volatile's behaviour is controllable by a compiler switch there, so care is required).

Another possible enhancement would be to drop the use of volatile, and instead insert the appropriate memory barrier operations directly (the old "volatile is neither necessary nor sufficient in general" adage). This is more explicit, and can be more efficient on some architectures, since you are explicitly specifying the exact barriers you require, which gives the optimizer more freedom.

Another enhancement you could add would be blocking read and write methods, though that's trickier, since you need a fast-pathed event object to avoid kernel calls where possible.

I am getting the error...

VI version is later than the current LabVIEW version.  An error occurred loading VI 'CleanupCallback.vi'.  LabVIEW load error code 9: VI version (8.2) is newer than the LabVIEW version (8.2b23).

Strange....its basically the same version.  Just off by a little b23.  You would think a newer version would be somewhat backwards compatable. Guess not.  So, how can I find out what is in "get callback function pointer" and cleanup callback?  Great article..can't wait to impliement it.

Updated Labview and it worked flawless.  This is awesome.

Thanks, Joe

Hello Joe,

How did you update labview? I mean 8.2 is the latest version. I am getting the same error you did (for another vi though).

George

(mail: gpouikli at ee.duth.gr)

YOU NEED TO DESIGN MFC FOR USB

Hi Joe

I have the same problem that you had.

Can you tell me how can I update LabVIEW

Sorry, dude.  Firewire never really caught on in the industry, which is why the option is not in Vista.  Are you still using floppy disks?  :)

Hi

Thanks very much for this link, but I have a question when the callback function is a VI code and the caller is a dll function. Could you please read this post and give me a reply to"paulbin@126.com"?

Thanks very much again.

Sorry for lose link.

http://forums.ni.com/ni/board/message?board.id=170&message.id=237441

luogo fine, sapete..

Glad to hear you liked the PowerShell session,

also nice to see you noticed the divider example :

>Everybody seems to find it perfectly obvious that 44/7 equates to 6.2857

As PowerShell is a Shell, the engine has a powerfull type converter and does a lot of parsing to make interactive use easier,and to behave as an admin would expect.

to force PowerShell into a more developer like behavour cast it explicitly :

PoSH> [int](44/7)

6

I'm sure you will enjoy the book.

Greetings /\/\o\/\/

i don't get it. " Each of the pointers (indices really) is updated only by one thread.". Why? can you please explain a bit more?

Scommettevo che avete speso molto tempo lavorare al luogo. Ha valso la pena di fare;)

Scommettevo che avete speso molto tempo lavorare al luogo. Ha valso la pena di fare;)

Interesting comments.. :D

Interesting comments.. :D

Lavoro eccellente! ..ringraziamenti per le informazioni..realmente lo apprezzo: D

I think this is not guaranteed to work if there is more than one thread on either end, if I understand correctly?  

The DevDays in Argentina are the same... VC++ simply doesn't exist, VB and C# only...mmm... this doesn't smell well...

Regards

I just want to know is it possible to

Create a thread safe producer consumer CircularQueue in C++ without using locks ...

As I already mentioned before, it's a small world after all. (Hearing that annoying Disney tune yet

Congratulations on new arrival in your family, happiness and health to Nele ;-)

Very interesting reading.

Thanks for posting this for those of us who can't be there.

Ken

MVP [ASP.NET]

What is the corresponding C# "out" for C++/CLI?

C# you can have foo (out int [] a)  or foo (ref int [] a)

I know how to do the second one in C++/CLI like

foo( array<int>^% a)

Can you tell the first one in C++/CLI?

Great article that clarify the intricacies of copy contructor and assignement operator

Great article that clarifies the intricacies of copy contructor and assignement operator

> Kumar said:

>

> I just want to know is it possible to

> Create a thread safe producer consumer

> CircularQueue in C++ without using locks ...

Yes, but the code above won't generally work unless.

1. The m_Read and m_Write updates are atomic

2. PushElement: Has a memory barrier around the data copy and m_Write update

3. PopElement: Has a memory barrier around the m_Write access in the if() clause and the data read.

I'm not sure about what Biff said, (#2,3)

Why do we need a memory barrier around data copy and m_write for instance?

There won't be any reordering for those in a single thread.

Thus we don't need to worry about those here(in 2 threads) either.

One of the issues involved in running a pharmaceutical process network is that QA has to give its blessing

I got a question about the multithreaded case.

>OK this is royally late, but to G: this will work in a >multithreaded situation without any problems.

     bool PopElement(T &Element)

     {

           if(m_Read == m_Write)

                 return false;

**** what if some other thread changed the value of m_Read at this point?

           int nextElement = (m_Read + 1) % Size;

           Element = m_Data[m_Read];

           m_Read = nextElement;

           return true;

     }

Pingback from  Domaining - Information on Domains and Domaining &raquo; Active Directory disjointed namespace problems

Have fun crashing just by looking at those two lines you should obviously see that this cannot be threadsafe:

           int nextElement = (m_Write + 1) % Size;

EDIT: Yes this is threadsafe. Size does not change over the lifetime of the object, and m_Write is only written to by the writer

           if(nextElement != m_Read)

Unless your CPU has an atomic +1 % Size compare instruction

It will do something like

1. Reg = 1

2. Reg = Reg + m_Write

3. Reg = Reg % Size

4. Compare Reg, m_Read

If another thread now updates m_Write during step 3 and 4 you have a serious Problem.

EDIT: Since the queue has only 1 writer, there is no problem. With multiple writers your point is valid, but this queue is meant for 1 writer hread and 1 reader thread.

Same goes for the position counter updates when you want multiple writers or readers you'd need to use Interlocked functions here or a Mutex.

EDIT: True, but again this class is only meant for single reader and single writer scenarios

The minimal needs to get a single reader/ single writer queue  threadsafe are 2 Semaphores to handle the full/empty checks.

EDIT: No, it is not. 32 bit reads and writes are atomic. The writer checks before each update to see if there is at least one free element. It does not matter if the reader reads more or not, because that will only make more items available.

Similarly, If the reader checks if there is at least 1 element in the queue, the write pointer may be updated or not it doesn't matter. Only 1 element will be read or written at a time, so if the write pointer changes, that doesn't matter.

For adding multiple multiple readers and writers you've to make the m_Read/m_Write increments atomic. For example use InterlockedIncrement or place a Mutex around the lines.

This problem has extensivly studied btw:

en.wikipedia.org/.../Producers-consumers_problem

EDIT: I just read the article, and the only reason semaphores are needed is because the reader and writer share a wakeup mechanism. The buffering itself does not need semaphores or protection in the single reader single writer scenario.

My class does not have or need a wakeup mechanism, so they don't need semaphores either.

For safe wakeup you need a semaphore indeed (which is why my production code has them for exactly this feature) but the queueing mechanism itself needs no protection. Which you could have figured out for yourself, had you read the article more closely

To proof invadility of your code you could do this btw:

...

int nextElement = (m_Read + 1) % Size;

Sleep(10 + rand() % 10);

Element = m_Data[m_Read];

...

EDIT: Yes I know, but ONLY if there are multiple readers! The m_Read is only updated in PopElement, and only 1 thread is ever calling that Method, so no matter how long you sleep in there, it will never go wrong.

Now make 1 writer thread that writes in 1 2 3 4 5 6 etc

And make 10 reader threads you'll for sure get duplicate numbers and some missing numbers in the output.

Yes indeed. That's why it's described as a class for a single reader and single writer. Which I also explain under the heading 'Why this is safe'

Thanks a lot!

Had this problem too, though you was the first to find a solution.

:)

Many Regards,

Jens- Birger Schlie

You need to change your reply to G. G was already asking what if there are multiple readers or multiple writers (and you obviously misunderstood his question). That's why both Allen and noobo critiqued your code (and you never clarified in your original article that it's a single-producer single-consumer que).

You can use IMEX=1 extended property when defining connection string for OLE DB driver and get all cells for all columns as text.

That's true. But there was a time when VMWare actually worked on FreeBSD and I used to rely on it quite heaviliy (I think it was up to VMWare 2.0). I do not remember if it just worked or if it was indeed a platform officially supported by VMWare. But given this fact, I consider it even more unfortunate that it does not run on BSD...

I used to have that dilbert strip, but can't find it anymore. do you have the date from it?

I'm new in .net, I have the similar problem, I need your help.

The following is the prototype of the callback function:

VOID(_stdcall *VideoFilter)(VOID *pcontext, BYTE *pData,ULONG dataLength)

Can you give me the corresponding .NET code?

Thank you in advance.

would u please tell me the diifference b/w c++ and c++/cli?

I am absolutely agreee with you. I work in company wich uses this syntax and we`re tring to migrate C# but our CEO does not want to do this.  

That's why there is the VBA option "option compare database" that Access uses by default for all VBA code. I think case insensitiveness is also the default behavior for SQL server and most other DBMS -- so I think Access is not really to blame here... But that does not change the fact that Access can be a pain to work with ;)

Most MS SQL Server installation are also 'case insensitive', like Jet. So, based from your comment, changing Jet to MS SQL Server everywere, yield the same observable effects, so,... MS SQL Server ain't a database?

Na, sounds more you are just too fast to blame other of your (wrong) assumptions about code.

If the queue size is one, the writer will not be able to use it.

Hi,

i am trying to implement a hardware in labview for my master thesis (sports science). as i am no programmer i am having not enough knowledge to get the things working right.

the problem i have got is simmilar to the one you solved.

i need to suply a function with a pointer to a callback function defined as

typedef BOOL (_stdcall *CHANNEL_EVENT_FUNC)(UCHAR ucANTChannel, UCHAR ucEvent);

it is used in

_declspec(dllexport) void ANT_AssignChannelEventFunction(UCHAR ucANTChannel, // Initialize Channel pointers

                                   CHANNEL_EVENT_FUNC pfChannelEvent,

                                   UCHAR *pucRxBuffer);

used in the c file as following

__declspec (dllexport) void ANT_AssignChannelEventFunction(UCHAR ucLink, CHANNEL_EVENT_FUNC pfLinkEvent, UCHAR *pucRxBuffer)

{

  sLink[ucLink].pfLinkEvent = pfLinkEvent;

  sLink[ucLink].pucRxBuffer = pucRxBuffer;

}

if you could help me by solving the problem i would be very thankfull or if you could make things work. i would pay for it too.

thank you a lot

martin

(email martin.boecskoer(at)aon.at)

Start slow and taper off...

I found this list of article on Raymond's blog . Raymond's blog is one of the more interesting

References are shorthand for pointers.

There's no other way to get it in a programmer's head what references really are. Sure, technically, they are slightly different, and it's possible that a new compiler technology may one day come along in which references and pointers are separated - but it's still going to behave in the ways that you've outlined.

Of course, as an ex-Fortran programmer (everything is passed by reference, even constants!), it's not too hard for me to deal with.

It seems that next IEEE standard might have something called "64-bit decimal floating-point" which will address the original problems with floating points and decimal numbers.

This paper has more info on subject

ieeexplore.ieee.org/.../login.jsp

We need to rid ourselves of mechanical hard drives before we'll see instant-on computers.  However, this is something that does need to happen, in my opinion.  If you've had any experience with the end-user side of computers, you'll realize its not the logging in part that is the delay.  Sure it adds a little extra bit of boot time, but the speed of a computer degrades over time, no matter how effective you are at managing performance through diagnostics and tune-ups.

I'd gladly welcome Instant-on.

TY.PC in my room has no net, hooked up a external HD to my friends,tried installing,but that didnt even work.Thanks for this info.

Fing BS  they dont include a install disc.BS that you need a net connection for a mp3 player.

Very useful code, thanks. I was had exactly the same requirement when I came across it.

One note: the actual FIFO depth (the effective buffering achievable) is Size - 1, not Size.

Thanks, however, my computer my computer says that my computer does not enough storage at step 5. However, I doubt that it needs that much GB. Got any ideas?

This implementation does not work for one reason, and is not useful for much more than a thought experiment for another reason. First, it doesn't work because of the following race condition in PushElement:

1            int nextElement = (m_Write + 1) % Size;

2            if(nextElement != m_Read)

3           {

4                  m_Data[m_Write] = Element;

5                  m_Write = nextElement;

6                  return true;

7            }

If any producer thread gets to line 2 while any other producer thread is executing 2-5, both will have the same m_Write index, and an item will be lost in the queue.

This is a fairly obvious issue, and also highly likely given how much time other threads have to race in and screw things up. This is a major problem.

However, that aside, the second major problem is more related to the intended functionality. This queue provides no way for a reader thread to idle on an empty queue. Therefore, readers are doomed to either poll the queue in tight loops when it is empty (not ideal if your CPUs are busy doing other things) or they must insert arbitrarily long pauses when the queue is empty (not ideal if latency matters). This problem makes this implementation not useful in most real scenarios, in addition to not being thread-safe.

JC

PS, a similar and equally obvious race condition exists in pop:

1            int nextElement = (m_Read + 1) % Size;

2            Element = m_Data[m_Read];

3            m_Read = nextElement;

It should be easily identifiable. If any consumer thread gets to 2 while any other consumer thread is executing 1-3 then both consumer threads use the same value of m_Read, thus returning the same item twice.

This implementation is far from thread-safe, and not for any subtle reasons having to do with memory barriers, optimizations, etc. It's just plain broken. Locking around the lines I mentioned to make the operations atomic would provide the necessary safety.

A second option on Windows is to modify the read/write indices using InterlockedIncrement(), which does atomic increment and returns of an integer (search for InterlockedIncrement on msdn.microsoft.com, there are many other Interlocked* functions as well), instead of attempting to do it over a few non-atomic statements and crossing your fingers that another thread doesn't run at the same time.

JC

Thanks for simplicity. And do you known how can I do this in SmartDevice (Native) projects?

I see one PPC program has been written in mixed mode. But it may be throught COM ?

Now, I can't access to guy with that very interesting project.

I work in mobile device programming sector. But some reason for to leave in C++ (big-big-big old C/C++ code). Then -> Native C++ (Smart Device projects). But I known, that big part of OS (Win CE, especially GUI-elements) has been written in .NET.

IMHO: "native way" for using that elements - using Managed C++. But "Managed C++ in SmartDevice projects" - it is not easy...

How can I do it? (very bad question, IMHO, but - "no way for me":)

Hi,

I am not sure if this question is in the same vein as what you have blogged on "Preventing a DLL from being unloaded by the app that uses it ". But essentially, i have implemented a hook similar to the code snipped you provided on "The windows hook hack" article. The problem is when I run(debug) my application that uses this dll, I am unable to recompile my dll after I have closed the application that was using this dll. The error message received was "error PRJ0008 : Could not delete file 'c:\...\mydll.dll'.

Somehow the dll is still being used and it cannot be deleted.

Can you please help? Thanks.

One of the things that everyone has to do sooner or later is parsing command line arguments into program

Hi vanDooren,

Thanks. Currently i call UnhookWindowsHookEx in dllmain when DLL_PROCESS_DETACH is detected. I suppose thats not the right way?

Thanks,

John

Hi Bruno,

Thank you so much. It definitely helped. :)

My original intention was to have my dll create a hook so that I could retrieve WM_GETMESSAGE (DBT_DEVICEARRIVAL and DBT_DEVICEREMOVECOMPLETE to be precise) so that the dll would know when my generic usb HID device was attached/removed. Suffice to say my dll never got that far. I know this is probably off topic but would you kindly point me to the right way of doing that as well?

Thanks,

John :)

Hello All,

Please can anyone help me to learn the device driver writing in VC# or VC++ and i am new comer in this field.

I know VC#.net and have little knowledge about VC++

anyone help me asap.

Thanks

Anand

Thanks loads  - this is exactly what i needed - short & sweet ! keep it up!

This worked!! For the last year I have been trying to figure out some way to use the software since I did the 2.0 update (which then took away my ability to sync my zune with my computer because of its lack of internet... little side note they didn't tell anyone) and had been unable to figure anything out that didn't involve hacking into parts of my computer that I would rather not do. I wasn't sure if your suggestion was going to work, but it did! I did get an error message after following your steps, saying I was missing a particular .dll file, but all I had to do to fix that was download and install the microsoft netscape 2.0 to correct that issue and then everything worked the way it was meant to. Thanks so much!!

my girlfriend got the Zune and it's horrible.

she could have gotten something better for the fraction

of the price.

Last week I had a very peculiar problem with the Terminal Server licensing. I couldn't connect to

i need to pass the below structure from labVIEW to C DLL Can any one help me out .....is thr an option to pass without writting a wrapper?PL do help .....

typedef struct

{

unsigned long NumOfParams;

SCONFIG *ConfigPtr;

} SCONFIG_LIST

typedef struct

{

unsigned long Parameter;

unsigned long Value;

} SCONFIG

pl do help me....

i need to pass the below structure from labVIEW to C DLL Can you help me out in this....is there an option to pass without writting a wrapper?or could how could you pass these kinda structures in labVIEW.

typedef struct

{

unsigned long NumOfParams;

SCONFIG *ConfigPtr;

} SCONFIG_LIST

typedef struct

{

unsigned long Parameter;

unsigned long Value;

} SCONFIG

kindly help me

Thanks for the concise explanation, great work.

I know this is a VC++ forum, but do you know if the same code will work in Java?

Pingback from  Kerberos Query | keyongtech

Hi I followed the link, I was looking for engineering blogs. Thanks for the share.

I am currently working on some projects where I have to program a DCOM server. There are several reasons

Pingback from  aufgeben und nach C#? | hilpers

>>> There are several reasons why it has to be DCOM and C++,  instead of e.g. .NET remoting and C#.

I'd be interested in knowing these reasons.

THX

Thank you for the detailed explanation. That does make sense.

I'm looking forward to read the other articles in this series.

Pingback from  Addressing Physical memory in XP using C++ | keyongtech

Pingback from  how do static import libraries actually work? | keyongtech

Pingback from  Semi-newbie question on Visual C++ vs. Visual C# | keyongtech

Another way the code could be unsafe, that you don't mention in your article, is if a thread is interrupted during the update of the read or write pointer. Sure, you claim in another comment that "32 bit reads and writes are atomic" but, of course, that statement does not generally hold true.

hehe I did that but luckily on a subset of users :)

Int is also defined to be >= 16 bits, so they are seldom atomic on 8-bit platforms. And the answer is yes, C++ is used there too... :)

There is no guarantee what-so-ever that int is always atomic. Even on 32-bit CPUs the memory subsystem may be limited to 16-bit, requiring two bus cycles for int access, possibly breaking atomicity. Unaligned accesses may also be non-atomic.

Ok, on x86 with MSVC++ it may be a valid assumption, I don't know. But the vast majority of computer systems in the world does not belong to that specific category, hence the 'not generally true'. For sure, the C++ standard doesn't give any such guarantees, and code like this that depend on implementation specific details should come with warning labels like the one you already have for hardware reordering.

C++0x will address the concurrency issues, probably because this type of synchronization just cannot be done portably and safely in plain C++. (From what I understand, C++0x will also make your method undefined behavior?)

What you need to do is a test case and prove it.. once you do ( and I doubt you will write one that will not fail every 7 hours or less)..

I agree they didn't read on single writer, single reader..

But now, you have to ask yourself, where is the proof? You might be just lucky and you are.

People already pointed out to you, you MUST have barriers for it to work. You didn't listen there at all.

Also, volatile is not the way to do it, and volatile on array is just nonsense.

Hi, so I wonder, where would we see 8bits nowadays? I can imagine some embedded applications or mobile phones or ...? And what is the solution for those? Just locks?

I was thinking you could actually use this pattern and still get to the multiple readers/writers by simply implementing joins and splits:

Multiple writers:

Writer thread A writes to QueueA

Writer thread B writes to QueueB

Join thread J reads QueueA and QueueB and writes to QueueX

Reader thread X reads QueueX

Multiple readers:

Writer thread X writes to QueueX

Split thread S reads QueueX and writes to QueueA and QueueB

Reader thread A reads QueueA

Reader thread B reads QueueB

You keep the 1 to 1 model....

I have a laptop and a desktop, but I could only get my laptop to install the software at a coffee shop. I can't wait to see if this works, because it was very short-sighted and annoying of microsoft to make their software this way.

Thx very helpful....

Though i dont know exaclty the real problem, but since the kdc error keep appearing in event viewer, so i did like you wrote here...

this line:

LPOLESTR strings[] = {L"One", L"Two", L"Three", NULL};

should be replaced by

LPOLESTR strings[] = {L"One", L"Two", L"Three"};

the C/C++ guarantee that one past the last element of an array will be a valid value, you will not be able to dereference it, but you will be able to compare to it

so adding a NULL  to your array is useless.

apart from that, nice introductory article.

I worked at Matrox 20 years ago and we had a similar producer/consumer fifo for interface between a graphic card  and the PC. There is one assumption (for intel platform) that you do not mention: the integer must be aligned on a 32 bit boundary. if you do not do it then the read or write will not be atomic and you will have problems between the producer and the consumer thread.

if you force the indexes to be in memory locations not divisible by 4, something like:

#pragma pack(1)

struct fifoInfo

{

char something;

int readIndex;

int writeIndex;

volatile T m_Data[Size]

};

the read & write index will not be updated atomicly. unless the latest intel architecture has change tremendously.

I had this same event happen.

I was running Server 2008 for 120 days and when the terminal license server expired, it was offline for a day or so. I reset the actual server and now I have been running Terminal License Free for about 3 months.

I have purchased 20 extra license just in case it flips out on one day and tries connecting to the terminal licensing server, but for now, its alll good.

Strange.

you can doubt as much as you can, read TC++PL 3rd edition, chapter 5, section 3 state:

"Taking a pointer to the element one beyond the end of an array is guaranteed to work."

see also 2.7.2

I've often seen this error in C++ samples from microsoft. which probably explains why this error is so frequent.

Of course dereferencing that pointer one past the last element is a big no-no.

Thanks!  This helped me solve a problem i was having.

Hi vanDooren,

Your description here is valuable to me since I'm having trouble to unload the dll that contains some Detours code. And Your solution 3) on Dec. 17 exactly matches my requirement.

However, I don't know how to obtain the function pointer in my DLL module which unhooks the system calls.  

Existing sample codes I found leverage GetProcAddress() to obtain a well-known function pointer (which is assumed to be the same across different processes) in Kernel32.dll (like LoadLibrary). But the function in my DLL is not shared across process. Even it's shared, different process may load it onto different memory address.

Are you aware of any solution to this?

Thank you very much!

Frank

Hi vanDooren,

Thanks.

It turns out I have to inject the binary executable code to the remote thread and have that code to call the GetProcAddress("MyExportedFunc") to obtain the function pointer.

A weird thing is: when I WriteProcessMemory to copy the &LocalThread to RemoteAddress, the copied binary isn't exactly the same as the original code. There are 4bit error in the 5th byte

I verify this by reading the runtime memory address in the processes, respectively. And I also memcpy the binary code to a local buffer and the same error happens.

I guess this may be due to Windows Vista has some protection on reading data from the code segment.

I work around this by declaring a const char[] to maintain the machine code I want to inject and now everything sees ok.

Thank you for your help and hope my experience can be of any use.

Frank

By the way, I believe your idea of injecting second dll to the remote process and having that dll unhook the first one would definitely work.

And it seems easier. I try my current method is just out of curiosity.

I am a little late to this party but how in the God's name do I remove this compiler option from the set of compiler switches?  I have migrated my project to VS 2008 and I get that dreaded D9035 warning which says that the /clr:oldSyntax option will be deprecated in a future release.  Ok I get that but how do I remove it?

So good article!I have kown lot of  detals about COM,Thanks very much! I think I will read more of your blog articles in the feature !

What happens if the Task is already running when the Scheduler calls it again.  Is there a configuration setting that allows you to stop the current or delay the new task until the previous one finishes, etc.?

I had the same issue as you and used a program called IsUsedBy mstsoftware.com to determine what program is using the Task.  It showed me I had 17 calls to the program that had been abandoned.  A reboot would have fixed mine as well but I would not have known.

Wusthof are excellent choice for kitchen knives, there are many types of Wusthof knife. If you got the right type to cut meat, then Wusthof will be better choice than The Kai. Yes its true there are saying that japan knives are sharper than any other knives in the world, but nowadays choice of sharp knives also available from different brand.

Hi Bruno,

thank you a lot for this useful article.

I'd like to ask you for an expert advice :-)  since I'm not sure about 1 thing. I was considering why elements in the queue need to be volatile.

Proper instruction ordering is still ensured even if queue is not volatile: since m_Read and m_Write are volatile, and they have proper Acquire/Release semantics - the compiler will not reorder memory instructions and so flag is always set after the queue has really been updated.

So it appears to me that the only important reason why the queue is volatile is when there are multiple processors, each one having its own cache. That way reads/writes to the queue are immediately reflected in the shared memory, am I right ?

I'm asking because type T in my case is a class which is not under my control, and compiler insists on having operator= for that class defined as volatile.

And one more question: If I'm right with the above presumption, then even if I do use locks to synchronize access to the queue, it still needs to be volatile to work in multi-processor enviroments. Is this correct... ?

I thank you in advance for your reply,

S. Angelovic

Hi,

To have atomic manipulation on basic types like integers (readIndex and writeIndex ) you can use (only under linux with the pthread library) the __sync_fetch_and_add() function to increment it for example.

I guess you have to align your basic type.

Theses atomic functions are made to be used with basic types, not structures.

Nicolas.

I bought a zune and found that microsoft made it so that the zune software can't be run on a linux-based operating system with wine. this annoyed me, but i prepared a desktop computer with XP installed so that i can get the zune software going on it. come to find out, i needed an internet connection. i was pissed. thanks for the tutorial though. helped me out a lot!

From what I read your "right way" shouldn't be done this way. The MSDN says calling LoadLibrary() in a DLLMain() should not be done which you do. I don't even understand why this doesn't get a lock-up, you call LoadLibrary() on a DLL that hasn't finished loading. From my understanding this should go into an infinite loop. I didn't read the referenced article on the NT Loader though. I assume it tags loaded-but-not-intialized DLLs?

Parallel programming is used specifically to serve working software developers, not just computer scientists. It is a complete, highly accessible pattern language that will help any experienced developer "think parallel"-and start writing effective parallel code almost immediately. Instead of formal theory, it deliver proven solutions to the challenges faced by parallel programmers, and pragmatic guidance for using today's parallel APIs in the real world.

I am definitely bookmarking this page and sharing it with my friends.

:)