How to avoid divergent branch in simple if statements in Cuda - c

I am wondering, how can anyone avoid branching in kernels when the threads have to compare and store values either from local, shared or global variables. For example the following code checks a shared variable and sets a bool flag to true accordingly
if ( shared_variable < local_value ){
shared_bool_var = true;
}
__syncthreads();
The problem here is that all threads access the same variable and all will overwrite to true.
So i would use a threadId.x check to only let one thread access that variable but this would cause branch divergence.
if ( threadIdx.x == 0 && shared_variable < local_value ){
shared_bool_var = true;
}
__syncthreads();
The question here is what should I prefer to do? In both cases it seems safe since the syncthread will protect from Hazards (read before write etc). My preference is the second solution but usually the code is not that simple.
In the aforementioned case, is it safe to allow all threads to access 1 shared memory location or this would cause a bank conflict or serialization of memory access?
Thanks

One important thing to note: semantically and functionally speaking, both code stanzas are not equivalent:
// set var to true if ANY thread in the block verifies the predicate
if (shared_variable < local_value) {
shared_bool_var = true;
}
// set var to true if THE FIRST thread in the block verifies the predicate
if (threadIdx.x == 0 && shared_variable < local_value) {
shared_bool_var = true;
}
But back to your question:
In the aforementioned case, is it safe to allow all threads to access 1 shared memory location or this would cause a bank conflict or serialization of memory access?
After verification in the CUDA programming guide, it seems there is some kind of write-collapsing mechanism that prevents serialization of write-accesses to the same address: instead, only one thread writes its value (but which thread is undefined).
CC 1.x:
If a non-atomic instruction executed by a warp writes to the same location in shared memory for more than one of the threads of the warp, only one thread per half-warp performs a write and which thread performs the final write is undefined.
CC 2.x and above:
A shared memory request for a warp does not generate a bank conflict between two threads that access any address within the same 32-bit word (even though the two addresses fall in the same bank): In that case, [...] for write accesses, each address is written by only one of the threads (which thread performs the write is undefined).
Additionally:
So i would use a threadId.x check to only let one thread access that variable but this would cause branch divergence.
This isn't "more divergent" than the first code. The first stanza exhibits divergence whenever a whole warp doesn't evaluate the predicate identically. The second stanza exhibits divergence only in the first warp of every block. In both cases, none of these branches have an impact on performance: there is no else body and the if body is a single instruction.

Related

pointer shared between two threads without mutex [duplicate]

Is there a problem with multiple threads using the same integer memory location between pthreads in a C program without any synchronization utilities?
To simplify the issue,
Only one thread will write to the integer
Multiple threads will read the integer
This pseudo-C illustrates what I am thinking
void thread_main(int *a) {
//wait for something to finish
//dereference 'a', make decision based on its value
}
int value = 0;
for (int i=0; i<10; i++)
pthread_create(NULL,NULL,thread_main,&value);
}
// do something
value = 1;
I assume it is safe, since an integer occupies one processor word, and reading/writing to a word should be the most atomic of operations, right?
Your pseudo-code is NOT safe.
Although accessing a word-sized integer is indeed atomic, meaning that you'll never see an intermediate value, but either "before write" or "after write", this isn't enough for your outlined algorithm.
You are relying on the relative order of the write to a and making some other change that wakes the thread. This is not an atomic operation and is not guaranteed on modern processors.
You need some sort of memory fence to prevent write reordering. Otherwise it's not guaranteed that other threads EVER see the new value.
Unlike java where you explicitly start a thread, posix threads start executing immediatelly.
So there is no guarantee that the value you set to 1 in main function (assuming that is what you refer in your pseudocode) will be executed before or after the threads try to access it.
So while it is safe to read the integer concurrently, you need to do some synchronization if you need to write to the value in order to be used by the threads.
Otherwise there is no guarantee what is the value they will read (in order to act depending on the value as you note).
You should not be making assumptions on multithreading e.g.that there is some processing in each thread befor accessing the value etc.
There are no guarantees
I wouldn't count on it. The compiler may emit code that assumes it knows what the value of 'value' is at any given time in a CPU register without re-loading it from memory.
EDIT:
Ben is correct (and I'm an idiot for saying he wasn't) that there is the possibility that the cpu will re-order the instructions and execute them down multiple pipelines at the same time. This means that the value=1 could possibly get set before the pipeline performing "the work" finished. In my defense (not a full idiot?) I have never seen this happen in real life and we do have an extensive thread library and we do run exhaustive long term tests and this pattern is used throughout. I would have seen it if it were happening, but none of our tests ever crash or produce the wrong answer. But... Ben is correct, the possibility exists. It is probably happening all the time in our code, but the re-ordering is not setting flags early enough that the consumers of the data protected by the flags can use the data before its finished. I will be changing our code to include barriers, because there is no guarantee that this will continue to work in the wild. I believe the correct solution is similar to this:
Threads that read the value:
...
if (value)
{
__sync_synchronize(); // don't pipeline any of the work until after checking value
DoSomething();
}
...
The thread that sets the value:
...
DoStuff()
__sync_synchronize(); // Don't pipeline "setting value" until after finishing stuff
value = 1; // Stuff Done
...
That being said, I found this to be a simple explanation of barriers.
COMPILER BARRIER
Memory barriers affect the CPU. Compiler barriers affect the compiler. Volatile will not keep the compiler from re-ordering code. Here for more info.
I believe you can use this code to keep gcc from rearranging the code during compile time:
#define COMPILER_BARRIER() __asm__ __volatile__ ("" ::: "memory")
So maybe this is what should really be done?
#define GENERAL_BARRIER() do { COMPILER_BARRIER(); __sync_synchronize(); } while(0)
Threads that read the value:
...
if (value)
{
GENERAL_BARRIER(); // don't pipeline any of the work until after checking value
DoSomething();
}
...
The thread that sets the value:
...
DoStuff()
GENERAL_BARRIER(); // Don't pipeline "setting value" until after finishing stuff
value = 1; // Stuff Done
...
Using GENERAL_BARRIER() keeps gcc from re-ordering the code and also keeps the cpu from re-ordering the code. Now, I wonder if gcc wont re-order code over its memory barrier builtin, __sync_synchronize(), which would make the use of COMPILER_BARRIER redundant.
X86
As Ben points out, different architectures have different rules regarding how they rearrange code in the execution pipelines. Intel seems to be fairly conservative. So the barriers might not be required nearly as much on Intel. Not a good reason to avoid the barriers though, since that could change.
ORIGINAL POST:
We do this all the time. its perfectly safe (not for all situations, but a lot). Our application runs on 1000's of servers in a huge farm with 16 instances per server and we don't have race conditions. You are correct to wonder why people use mutexes to protect already atomic operations. In many situations the lock is a waste of time. Reading and writing to 32 bit integers on most architectures is atomic. Don't try that with 32 bit bit-fields though!
Processor write re-ordering is not going to affect one thread reading a global value set by another thread. In fact, the result using locks is the same as the result not without locks. If you win the race and check the value before its changed ... well that's the same as winning the race to lock the value so no-one else can change it while you read it. Functionally the same.
The volatile keyword tells the compiler not to store a value in a register, but to keep referring to the original memory location. this should have no effect unless you are optimizing code. We have found that the compiler is pretty smart about this and have not run into a situation yet where volatile changed anything. The compiler seems to be pretty good at coming up with candidates for register optimization. I suspect that the const keyword might encourage register optimization on a variable.
The compiler might re-order code in a function if it knows the end result will not be different. I have not seen the compiler do this with global variables, because the compiler has no idea how changing the order of a global variable will affect code outside of the immediate function.
If a function is acting up, you can control the optimization level at the function level using __attrribute__.
Now, that said, if you use that flag as a gateway to allow only one thread of a group to perform some work, that wont work. Example: Thread A and Thread B both could read the flag. Thread A gets scheduled out. Thread B sets the flag to 1 and starts working. Thread A wakes up and sets the flag to 1 and starts working. Ooops! To avoid locks and still do something like that you need to look into atomic operations, specifically gcc atomic builtins like __sync_bool_compare_and_swap(value, old, new). This allows you to set value = new if value is currently old. In the previous example, if value = 1, only one thread (A or B) could execute __sync_bool_compare_and_swap(&value, 1, 2) and change value from 1 to 2. The losing thread would fail. __sync_bool_compare_and_swap returns the success of the operation.
Deep down, there is a "lock" when you use the atomic builtins, but it is a hardware instruction and very fast when compared to using mutexes.
That said, use mutexes when you have to change a lot of values at the same time. atomic operations (as of todayu) only work when all the data that has to change atomicly can fit into a contiguous 8,16,32,64 or 128 bits.
Assume the first thing you're doing in thread func in sleeping for a second. So value after that will be definetly 1.
In any instant you should at least declare the shared variable volatile. However you should in all cases prefer some other form of thread IPC or synchronisation; in this case it looks like a condition variable is what you actually need.
Hm, I guess it is secure, but why don't you just declare a function that returns the value to the other threads, as they will only read it?
Because the simple idea of passing pointers to separate threads is already a security fail, in my humble opinion. What I'm telling you is: why to give a (modifiable, public accessible) integer address when you only need the value?

Race condition and mutex

I have 2 questions regarding to threads, one is about race condition and the other is about mutex.
So the first question :
I've read about race condition in wikipedia page :
http://en.wikipedia.org/wiki/Race_condition
And in the example of race condition between 2 threads this is shown :
http://i60.tinypic.com/2vrtuz4.png[
Now so far I believed that threads works parallel to each other, but judging from this picture it's seems that I interpreted on how actions done by the computer wrong.
From this picture only 1 action is done at a time, and although the threads gets switched from time to time and the other thread gets to do some actions this is still 1 action at a time done by the computer. Is it really like this ? There's no "real" parallel computing, just 1 action done at a time in a very fast rate which gives the illusion of parallel computing ?
This leads me to my second question about mutex.
I've read that if threads read/write to the same memory we need some sort of synchronization mechanism. I've read the normal data types won't do and we need a mutex.
Let's take for example the following code :
#include <stdio.h>
#include <stdbool.h>
#include <windows.h>
#include <process.h>
bool lock = false;
void increment(void*);
void decrement(void*);
int main()
{
int n = 5;
HANDLE hIncrement = (HANDLE)_beginthread(increment, 0, (void*)&n);
HANDLE hDecrement = (HANDLE)_beginthread(decrement, 0, (void*)&n);
WaitForSingleObject(hIncrement, 1000 * 500);
WaitForSingleObject(hDecrement, 1000 * 500);
return 0;
}
void increment(void *p)
{
int *n = p;
for(int i = 0; i < 10; i++)
{
while (lock)
{
}
lock = true;
(*n)++;
lock = false;
}
}
void decrement(void *p)
{
int *n = p;
for(int i = 0; i < 10; i++)
{
while (lock)
{
}
lock = true;
(*n)--;
lock = false;
}
}
Now in my example here, I use bool lock as my synchronization mechanism to avoid a race condition between the 2 threads over the memory space pointed by pointer n.
Now what I did here won't obviously work because although I avoided a race condition over the memory space pointed by pointer n between the 2 threads a new race condition over bool lock variable may occur.
Let's consider the following sequence of events (A = increment thread, B = decrement thread) :
A gets out of the while loop since lock is false
A gets to set lock to true
B waits in the while loop because lock is set to true
A increment the value pointed by n
A sets lock to false
A gets to the while loop
A gets out of the while loop since lock is false
B gets out of the while loop since lock is false
A sets lock to true
B sets lock to true
and from here we get an unexpected behavior of 2 un-synchronized threads because the bool lock is not race condition proof.
Ok, so far this is my understanding and the solution to our problem above we need a mutex.
I'm fine with that, a data type that will magically be condition race proof.
I just don't understand how with mutex type it won't happen where as with every other type it will and here lies my problem, I want to understand why mutex and how this is happening.
About your first question: Whether or not there are actually several different threads running at once, or whether it is just implemented as as fast switching, is a matter of your hardware. Typical PCs these days have several cores (often with more than one thread each), so you have to assume that things actually DO happen at the same time.
But even if you have only a single-core system, things are not quite so easy. This is because the compiler is usually allowed to re-order instructions in order to optimize code. It can also e.g. choose to cache a variable in a CPU register instead of loading it from memory every time you access it, and it also doesn't have to write it back to memory every time you write to that variable. The compiler is allowed to do that as long as the result is the same AS IF it had run your original code in its original order - as long as nobody else is looking closely at what's actually going on, such as a different thread.
And once you actually do have different cores, consider that they all have their own CPU registers and even their own cache. Even if a thread on one core wrote to a certain variable, as long as that core doesn't write its cache back to the shared memory a different core won't see that change.
In short, you have to be very careful in making any assumptions about what happens when two threads access variables at the same time, especially in C/C++. The interactions can be so surprising that I'd say, to stay on the safe side, you should make sure that there are no race conditions in your code, e.g. by always using mutexes for accessing memory that is shared between threads.
Which is where we can neatly segway into the second question: What's so special about mutexes, and how can they work if all basic data types are not threadsafe?
The thing about mutexes is that they are implemented with a lot of knowledge about the system for which they are being used (hardware and operating system), and with either the direct help or a deep knowledge of the compiler itself.
The C language does not give you direct access to all the capabilities of your hardware and operating system, because platforms can be very different from each other. Instead, C focuses on providing a level of abstraction that allows you to compile the same code for many different platforms. The different "basic" data types are just something that the C standard came up with as a set of data types which can in some way be supported on almost any platform - but the actual hardware that your program will be compiled for is usually not limited to those types and operations.
In other word, not everything that you can do with your PC can be expressed in terms of C's ints, bytes, assignments, arithmetic operators and so on. For example, PCs often calculate with 80-bit floating point types which are usually not mapped directly to a C floating point type at all. More to the point of our topic, there are also CPU instructions that influence how multiple CPU cores will work together. Additionally, if you know the CPU, you often know a few things about the behaviour of the basic types that the C standard doesn't guarantee (for example, whether loads and stores to 32-bit integers are atomic). With that extra knowledge, it can become possible to implement mutexes for that particular platform, and it will often require code that is e.g. written directly in assembly language, because the necessary features are not available in plain C.

memory barrier in linux kernel's ext2 function ext2_statfs()

Could anyone explain why linux kernel's ext2 function
int ext2_statfs (struct dentry * dentry, struct kstatfs * buf)
issues smp_rmb() andsmp_wmb() in
else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
case ?
This was added in the upstream kernel commit 2235219b7721b8e74de6841e79240936561a2b63 which omits unnecessary calculation of .statfs, but couldn't understand why memory barriers had been added.
Since the function starts by doing a spin_lock (and unlocks at the end),
spin_lock(&sbi->s_lock);
it is protected against concurrent accesses. So why the need of smp_wmb and smp_rmb...
As for the write operation,
sbi->s_overhead_last = overhead;
smp_wmb();
sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
it appears that the smp_wmb would prevent any compiler or processor optimization (since there is no dependency between the two writes in terms of memory accesses) that would change the order of the two assignments. Since there is no concurrent accesses in ext2_statfs, the problem could happen in another function that may have a similar behavior and check if there is a block count change and, if not, use the overhead from sbi->s_overhead_last - which would be wrong if another thread would be, at this very time, in another function, in the middle of the two write operations in the wrong order.
This is probably a precaution as any similar write access to the sbi properties would spin_lock it (with the same spinlock) beforehand. Also it enhances the fact that the two operations must be done in this order.
As for the read access the need is less obvious and would require to analyze further the dependencies between the sbi and es blocks.

What prevents a race condition when checking a semaphore value?

I'm studying multithreading and trying to understand the concept of semaphores and mutual exclusion. Most of the examples I find online use some sort of library (e.g. pthread) to implement the semaphore or mutex, but I'm more interested in the implementation of a simple semaphore that establishes a critical section -- no more than one thread accessing a particular region of memory.
For this task, I believe I would need a mutex (a.k.a. a binary semaphore if I understand the terminology correctly). I can see how the semaphore would prevent a race condition by "locking" the section of code to a single thread, but what prevents a race condition from occurring at the semaphore itself?
I imagine a binary semaphore to hold an int value to keep track of the lock:
Semaphore
---------
int lock = 1;
unsigned P(void){
if(lock > 0){
lock--;
return 0; /* success */
}
return 1; /* fail */
}
void V(void){
lock++;
}
Suppose two threads call the P function at the same time, they both reach the if(lock > 0) check at the same time and evaluate the condition as true -- this creates a race condition where both threads are granted access to the same region of memory at the same time.
So what prevents this race condition from occurring in real world implementations of semaphores?
Locking and relasing semaphores and/or mutexes happen as atomic operations, this means the CPU cannot be withdrawn from the current process. This ensures, that as soon as a mutex-lock is started (it consists of either a single or a few CPU-instruction (microcode)), the process keeps the CPU until the locking/releasing is done.
There are also different ways to implement threading, which can either be a direct support by CPU (kernel-space) or through a library (such as pthreads) in user-space.
From OSDev.org
An atomic operation is an operation that will always be executed without any other process being able to read or change state that is read or changed during the operation. It is effectively executed as a single step, and is an important quality in a number of algorithms that deal with multiple indepent processes, both in synchronization and algorithms that update shared data without requiring synchronization.
Here is a nice article on atomicity, too (although in Delphi).
The most common (although definitely not the only) way to implement most locking primitives are compare-and-set instructions. An normal move instruction would just set the value of a memory location to whatever value you ask it to while a compare-and-set instruction does "atomically set this memory location to value X only if the value of the memory location is Y, then set some flag if the operation succeeded or not". The keyword "atomic" is that the CPU can in hardware make sure that nothing else can interfere with that operation.
Using a compare-and-swap instruction your example P could be implemented as:
int oldlock;
retry:
oldlock = lock;
if (oldlock > 0) {
if (compare_and_swap(&lock, oldlock, oldlock - 1))
goto retry;
return 0;
}
return 1;
Of course reality is much more complex than that, but compare-and-set is easy to understand and explain and has the nice property that it can implement (almost?) all other locking primitives.
Here's a wikipedia article.
The difference between a semaphore (or a mutex) and a "normal" variable isn't that big. Those libraries which offer you this functionality just make sure that the semaphore is only accessed through atomic operations. There are multiple ways to achieve that:
Special assembly instructions which guarantee atomic access, e.g.: TSL or XCHG.
Turning off the scheduler's interrupts before the variable gets accessed and afterwards turning them back on again. So the scheduler can't remove your process from the CPU. But you have to be aware that this only works on single CPU systems.
Using language specific features like Java's synchronise keyword.
An example of how to use the TSL instruction:
enter_region: ; A "jump to" tag; function entry point.
tsl reg, flag ; Test and Set Lock; flag is the
; shared variable; it is copied
; into the register reg and flag
; then atomically set to 1.
cmp reg, #0 ; Was flag zero on entry_region?
jnz enter_region ; Jump to enter_region if
; reg is non-zero; i.e.,
; flag was non-zero on entry.
ret ; Exit; i.e., flag was zero on
; entry. If we get here, tsl
; will have set it non-zero; thus,
; we have claimed the resource
; associated with flag.
leave_region:
move flag, #0 ; store 0 in flag
ret ; return to caller
By the way, as you already and correctly pointed out, a mutex is just a special kind of a semaphore, only allowing FALSE (represented by 0 in C) and TRUE (represented by 1 or any other value != 0) for it's internal int variable. Thus making it a so called binary semaphore.

Can an integer be shared between threads safely?

Is there a problem with multiple threads using the same integer memory location between pthreads in a C program without any synchronization utilities?
To simplify the issue,
Only one thread will write to the integer
Multiple threads will read the integer
This pseudo-C illustrates what I am thinking
void thread_main(int *a) {
//wait for something to finish
//dereference 'a', make decision based on its value
}
int value = 0;
for (int i=0; i<10; i++)
pthread_create(NULL,NULL,thread_main,&value);
}
// do something
value = 1;
I assume it is safe, since an integer occupies one processor word, and reading/writing to a word should be the most atomic of operations, right?
Your pseudo-code is NOT safe.
Although accessing a word-sized integer is indeed atomic, meaning that you'll never see an intermediate value, but either "before write" or "after write", this isn't enough for your outlined algorithm.
You are relying on the relative order of the write to a and making some other change that wakes the thread. This is not an atomic operation and is not guaranteed on modern processors.
You need some sort of memory fence to prevent write reordering. Otherwise it's not guaranteed that other threads EVER see the new value.
Unlike java where you explicitly start a thread, posix threads start executing immediatelly.
So there is no guarantee that the value you set to 1 in main function (assuming that is what you refer in your pseudocode) will be executed before or after the threads try to access it.
So while it is safe to read the integer concurrently, you need to do some synchronization if you need to write to the value in order to be used by the threads.
Otherwise there is no guarantee what is the value they will read (in order to act depending on the value as you note).
You should not be making assumptions on multithreading e.g.that there is some processing in each thread befor accessing the value etc.
There are no guarantees
I wouldn't count on it. The compiler may emit code that assumes it knows what the value of 'value' is at any given time in a CPU register without re-loading it from memory.
EDIT:
Ben is correct (and I'm an idiot for saying he wasn't) that there is the possibility that the cpu will re-order the instructions and execute them down multiple pipelines at the same time. This means that the value=1 could possibly get set before the pipeline performing "the work" finished. In my defense (not a full idiot?) I have never seen this happen in real life and we do have an extensive thread library and we do run exhaustive long term tests and this pattern is used throughout. I would have seen it if it were happening, but none of our tests ever crash or produce the wrong answer. But... Ben is correct, the possibility exists. It is probably happening all the time in our code, but the re-ordering is not setting flags early enough that the consumers of the data protected by the flags can use the data before its finished. I will be changing our code to include barriers, because there is no guarantee that this will continue to work in the wild. I believe the correct solution is similar to this:
Threads that read the value:
...
if (value)
{
__sync_synchronize(); // don't pipeline any of the work until after checking value
DoSomething();
}
...
The thread that sets the value:
...
DoStuff()
__sync_synchronize(); // Don't pipeline "setting value" until after finishing stuff
value = 1; // Stuff Done
...
That being said, I found this to be a simple explanation of barriers.
COMPILER BARRIER
Memory barriers affect the CPU. Compiler barriers affect the compiler. Volatile will not keep the compiler from re-ordering code. Here for more info.
I believe you can use this code to keep gcc from rearranging the code during compile time:
#define COMPILER_BARRIER() __asm__ __volatile__ ("" ::: "memory")
So maybe this is what should really be done?
#define GENERAL_BARRIER() do { COMPILER_BARRIER(); __sync_synchronize(); } while(0)
Threads that read the value:
...
if (value)
{
GENERAL_BARRIER(); // don't pipeline any of the work until after checking value
DoSomething();
}
...
The thread that sets the value:
...
DoStuff()
GENERAL_BARRIER(); // Don't pipeline "setting value" until after finishing stuff
value = 1; // Stuff Done
...
Using GENERAL_BARRIER() keeps gcc from re-ordering the code and also keeps the cpu from re-ordering the code. Now, I wonder if gcc wont re-order code over its memory barrier builtin, __sync_synchronize(), which would make the use of COMPILER_BARRIER redundant.
X86
As Ben points out, different architectures have different rules regarding how they rearrange code in the execution pipelines. Intel seems to be fairly conservative. So the barriers might not be required nearly as much on Intel. Not a good reason to avoid the barriers though, since that could change.
ORIGINAL POST:
We do this all the time. its perfectly safe (not for all situations, but a lot). Our application runs on 1000's of servers in a huge farm with 16 instances per server and we don't have race conditions. You are correct to wonder why people use mutexes to protect already atomic operations. In many situations the lock is a waste of time. Reading and writing to 32 bit integers on most architectures is atomic. Don't try that with 32 bit bit-fields though!
Processor write re-ordering is not going to affect one thread reading a global value set by another thread. In fact, the result using locks is the same as the result not without locks. If you win the race and check the value before its changed ... well that's the same as winning the race to lock the value so no-one else can change it while you read it. Functionally the same.
The volatile keyword tells the compiler not to store a value in a register, but to keep referring to the original memory location. this should have no effect unless you are optimizing code. We have found that the compiler is pretty smart about this and have not run into a situation yet where volatile changed anything. The compiler seems to be pretty good at coming up with candidates for register optimization. I suspect that the const keyword might encourage register optimization on a variable.
The compiler might re-order code in a function if it knows the end result will not be different. I have not seen the compiler do this with global variables, because the compiler has no idea how changing the order of a global variable will affect code outside of the immediate function.
If a function is acting up, you can control the optimization level at the function level using __attrribute__.
Now, that said, if you use that flag as a gateway to allow only one thread of a group to perform some work, that wont work. Example: Thread A and Thread B both could read the flag. Thread A gets scheduled out. Thread B sets the flag to 1 and starts working. Thread A wakes up and sets the flag to 1 and starts working. Ooops! To avoid locks and still do something like that you need to look into atomic operations, specifically gcc atomic builtins like __sync_bool_compare_and_swap(value, old, new). This allows you to set value = new if value is currently old. In the previous example, if value = 1, only one thread (A or B) could execute __sync_bool_compare_and_swap(&value, 1, 2) and change value from 1 to 2. The losing thread would fail. __sync_bool_compare_and_swap returns the success of the operation.
Deep down, there is a "lock" when you use the atomic builtins, but it is a hardware instruction and very fast when compared to using mutexes.
That said, use mutexes when you have to change a lot of values at the same time. atomic operations (as of todayu) only work when all the data that has to change atomicly can fit into a contiguous 8,16,32,64 or 128 bits.
Assume the first thing you're doing in thread func in sleeping for a second. So value after that will be definetly 1.
In any instant you should at least declare the shared variable volatile. However you should in all cases prefer some other form of thread IPC or synchronisation; in this case it looks like a condition variable is what you actually need.
Hm, I guess it is secure, but why don't you just declare a function that returns the value to the other threads, as they will only read it?
Because the simple idea of passing pointers to separate threads is already a security fail, in my humble opinion. What I'm telling you is: why to give a (modifiable, public accessible) integer address when you only need the value?

Resources