door closed by defaul and vcg:: namespave

This commit is contained in:
Federico Ponchio 2012-05-29 12:14:27 +00:00
parent 75ebfa9d5a
commit 0d4a84f012
4 changed files with 204 additions and 191 deletions

View File

@ -21,245 +21,253 @@ using namespace std;
/** Cache virtual base class. You are required to implement the pure virtual functions get, drop and size.
*/
namespace vcg {
template <typename Token> class Transfer;
template <typename Token>
class Cache: public Provider<Token> {
public:
///true if this is the last cache (the one we use the data from)
bool final;
//if true the cache will exit at the first opportunity
bool quit;
///keeps track of changes (if 1 then something was loaded or dropped
mt::atomicInt new_data;
///callback for new_data
void (*callback)(void *data);
///true if this is the last cache (the one we use the data from)
bool final;
//if true the cache will exit at the first opportunity
bool quit;
///keeps track of changes (if 1 then something was loaded or dropped
mt::atomicInt new_data;
///callback for new_data
void (*callback)(void *data);
///data is fetched from here
Provider<Token> *input;
///data is fetched from here
Provider<Token> *input;
///threads running over cache...
std::vector<Transfer<Token> *> transfers;
///threads running over cache...
std::vector<Transfer<Token> *> transfers;
protected:
///max space available
uint64_t s_max;
///current space used
uint64_t s_curr;
///max space available
uint64_t s_max;
///current space used
uint64_t s_curr;
public:
Cache(uint64_t _capacity = INT_MAX):
final(false), quit(false), new_data(false), input(NULL), s_max(_capacity), s_curr(0) {}
virtual ~Cache() {}
Cache(uint64_t _capacity = INT_MAX):
final(false), quit(false), new_data(false), input(NULL), s_max(_capacity), s_curr(0) {}
virtual ~Cache() {}
void setInputCache(Provider<Token> *p) { input = p; }
uint64_t capacity() { return s_max; }
uint64_t size() { return s_curr; }
void setCapacity(uint64_t c) { s_max = c; }
void setInputCache(Provider<Token> *p) { input = p; }
uint64_t capacity() { return s_max; }
uint64_t size() { return s_curr; }
void setCapacity(uint64_t c) { s_max = c; }
///return true if the cache is waiting for priority to change
bool newData() {
bool r = new_data.testAndSetOrdered(1, 0); //if changed is 1, r is true
return r;
}
///empty the cache. Make sure no resource is locked before calling this.
/// Require pause or stop before. Ensure there no locked item
void flush() {
//std::vector<Token *> tokens;
{
for(int i = 0; i < this->heap.size(); i++) {
Token *token = &(this->heap[i]);
//tokens.push_back(token);
s_curr -= drop(token);
assert(!(token->count >= Token::LOCKED));
if(final)
token->count.testAndSetOrdered(Token::READY, Token::CACHE);
input->heap.push(token);
}
this->heap.clear();
///return true if the cache is waiting for priority to change
bool newData() {
bool r = new_data.testAndSetOrdered(1, 0); //if changed is 1, r is true
return r;
}
if(!s_curr == 0) {
std::cerr << "Cache size after flush is not ZERO!\n";
s_curr = 0;
}
}
///empty the cache. Make sure no resource is locked before calling this.
/// Require pause or stop before. Ensure there no locked item
template <class FUNCTOR> void flush(FUNCTOR functor) {
std::vector<Token *> tokens;
{
int count = 0;
mt::mutexlocker locker(&(this->heap_lock));
for(int k = 0; k < this->heap.size(); k++) {
Token *token = &this->heap[k];
if(functor(token)) { //drop it
tokens.push_back(token);
s_curr -= drop(token);
assert(token->count < Token::LOCKED);
if(final)
token->count.testAndSetOrdered(Token::READY, Token::CACHE);
} else
this->heap.at(count++) = token;
}
this->heap.resize(count);
this->heap_dirty = true;
///empty the cache. Make sure no resource is locked before calling this.
/// Require pause or stop before. Ensure there no locked item
void flush() {
//std::vector<Token *> tokens;
{
for(int i = 0; i < this->heap.size(); i++) {
Token *token = &(this->heap[i]);
//tokens.push_back(token);
s_curr -= drop(token);
assert(!(token->count >= Token::LOCKED));
if(final)
token->count.testAndSetOrdered(Token::READY, Token::CACHE);
input->heap.push(token);
}
this->heap.clear();
}
if(!s_curr == 0) {
std::cerr << "Cache size after flush is not ZERO!\n";
s_curr = 0;
}
}
{
mt::mutexlocker locker(&(input->heap_lock));
for(unsigned int i = 0; i < tokens.size(); i++) {
input->heap.push(tokens[i]);
}
}
}
virtual void abort() {}
///empty the cache. Make sure no resource is locked before calling this.
/// Require pause or stop before. Ensure there no locked item
template <class FUNCTOR> void flush(FUNCTOR functor) {
std::vector<Token *> tokens;
{
int count = 0;
mt::mutexlocker locker(&(this->heap_lock));
for(int k = 0; k < this->heap.size(); k++) {
Token *token = &this->heap[k];
if(functor(token)) { //drop it
tokens.push_back(token);
s_curr -= drop(token);
assert(token->count < Token::LOCKED);
if(final)
token->count.testAndSetOrdered(Token::READY, Token::CACHE);
} else
this->heap.at(count++) = token;
}
this->heap.resize(count);
this->heap_dirty = true;
}
{
mt::mutexlocker locker(&(input->heap_lock));
for(unsigned int i = 0; i < tokens.size(); i++) {
input->heap.push(tokens[i]);
}
}
}
virtual void abort() {}
protected:
///return the space used in the cache by the loaded resource
virtual int size(Token *token) = 0;
///returns amount of space used in cache -1 for failed transfer
virtual int get(Token *token) = 0;
///return amount removed
virtual int drop(Token *token) = 0;
///make sure the get function do not access token after abort is returned.
///return the space used in the cache by the loaded resource
virtual int size(Token *token) = 0;
///returns amount of space used in cache -1 for failed transfer
virtual int get(Token *token) = 0;
///return amount removed
virtual int drop(Token *token) = 0;
///make sure the get function do not access token after abort is returned.
///called in as first thing in run()
virtual void begin() {}
///called in as last thing in run()
virtual void end() {}
///called in as first thing in run()
virtual void begin() {}
virtual void middle() {}
///called in as last thing in run()
virtual void end() {}
///[should be protected]
void run() {
assert(input);
/* basic operation of the cache:
///[should be protected]
void run() {
assert(input);
/* basic operation of the cache:
1) transfer first element of input_cache if
cache has room OR first element in input as higher priority of last element
2) make room until eliminating an element would leave space. */
begin();
while(!this->quit) {
input->check_queue.enter(); //wait for cache below to load something or priorities to change
if(this->quit) break;
begin();
while(!this->quit) {
input->check_queue.enter(true); //wait for cache below to load something or priorities to change
if(this->quit) break;
if(unload() || load()) {
new_data.testAndSetOrdered(0, 1); //if not changed, set as changed
input->check_queue.open(); //we signal ourselves to check again
}
input->check_queue.leave();
middle();
if(unload() || load()) {
new_data.testAndSetOrdered(0, 1); //if not changed, set as changed
input->check_queue.open(); //we signal ourselves to check again
}
input->check_queue.leave();
}
this->quit = false; //in case someone wants to restart;
end();
}
this->quit = false; //in case someone wants to restart;
end();
}
/** Checks wether we need to make room in the cache because of:
/** Checks wether we need to make room in the cache because of:
size() - sizeof(lowest priority item) > capacity()
**/
bool unload() {
Token *remove = NULL;
//make room int the cache checking that:
//1 we need to make room (capacity < current)
if(size() > capacity()) {
bool unload() {
Token *remove = NULL;
//make room int the cache checking that:
//1 we need to make room (capacity < current)
if(size() > capacity()) {
mt::mutexlocker locker(&(this->heap_lock));
mt::mutexlocker locker(&(this->heap_lock));
//2 we have some element not in the upper caches (heap.size() > 0
if(this->heap.size()) {
Token &last = this->heap.min();
int itemsize = size(&last);
//2 we have some element not in the upper caches (heap.size() > 0
if(this->heap.size()) {
Token &last = this->heap.min();
int itemsize = size(&last);
//3 after removing the item, we are still full (avoids bouncing items)
if(size() - itemsize > capacity()) {
//3 after removing the item, we are still full (avoids bouncing items)
if(size() - itemsize > capacity()) {
//4 item to remove is not locked. (only in last cache. you can't lock object otherwise)
if(!final) { //not final we can drop when we want
remove = this->heap.popMin();
} else {
last.count.testAndSetOrdered(Token::READY, Token::CACHE);
if(last.count <= Token::CACHE) { //was not locked and now can't be locked, remove it.
remove = this->heap.popMin();
} else { //last item is locked need to reorder stack
remove = this->heap.popMin();
this->heap.push(remove);
return true;
//4 item to remove is not locked. (only in last cache. you can't lock object otherwise)
if(!final) { //not final we can drop when we want
remove = this->heap.popMin();
} else {
last.count.testAndSetOrdered(Token::READY, Token::CACHE);
if(last.count <= Token::CACHE) { //was not locked and now can't be locked, remove it.
remove = this->heap.popMin();
} else { //last item is locked need to reorder stack
remove = this->heap.popMin();
this->heap.push(remove);
cout << "Reordering stack something (what?)\n";
return true;
}
}
}
}
}
}
}
if(remove) {
{
mt::mutexlocker input_locker(&(input->heap_lock));
int size = drop(remove);
assert(size >= 0);
s_curr -= size;
input->heap.push(remove);
}
cout << "Removing something (what?)\n";
return true;
}
return false;
}
if(remove) {
{
mt::mutexlocker input_locker(&(input->heap_lock));
int size = drop(remove);
assert(size >= 0);
s_curr -= size;
input->heap.push(remove);
}
return true;
}
return false;
}
///should be protected
bool load() {
Token *insert = NULL;
Token *last = NULL; //we want to lock only one heap at once to avoid deadlocks.
///should be protected
bool load() {
Token *insert = NULL;
Token *last = NULL; //we want to lock only one heap at once to avoid deadlocks.
/* check wether we have room (curr < capacity) or heap is empty.
/* check wether we have room (curr < capacity) or heap is empty.
empty heap is bad: we cannot drop anything to make room, and cache above has nothing to get.
this should not happen if we set correct cache sizes, but if it happens.... */
{
mt::mutexlocker locker(&(this->heap_lock));
this->rebuild();
if(size() > capacity() && this->heap.size() > 0) {
last = &(this->heap.min()); //no room, set last so we might check for a swap.
}
}
{
mt::mutexlocker input_locker(&(input->heap_lock));
input->rebuild(); //if dirty rebuild
if(input->heap.size()) { //we need something in input to tranfer.
Token &first = input->heap.max();
if(first.count > Token::REMOVE &&
(!last || first.priority > last->priority)) { //if !last we already decided we want a transfer., otherwise check for a swap
insert = input->heap.popMax(); //remove item from heap, while we transfer it.
}
}
}
if(insert) { //we want to fetch something
int size = get(insert);
if(size >= 0) { //success
s_curr += size;
{
mt::mutexlocker locker(&(this->heap_lock));
if(final)
insert->count.ref(); //now lock is 0 and can be locked
this->heap.push(insert);
mt::mutexlocker locker(&(this->heap_lock));
this->rebuild();
if(size() > capacity() && this->heap.size() > 0) {
last = &(this->heap.min()); //no room, set last so we might check for a swap.
}
}
this->check_queue.open(); //we should signal the parent cache that we have a new item
return true;
} else { //failed transfer put it back, we will keep trying to transfer it...
mt::mutexlocker input_locker(&(input->heap_lock));
input->heap.push(insert);
{
mt::mutexlocker input_locker(&(input->heap_lock));
input->rebuild(); //if dirty rebuild
if(input->heap.size()) { //we need something in input to tranfer.
Token &first = input->heap.max();
if(first.count > Token::REMOVE &&
(!last || first.priority > last->priority)) { //if !last we already decided we want a transfer., otherwise check for a swap
insert = input->heap.popMax(); //remove item from heap, while we transfer it.
}
}
}
if(insert) { //we want to fetch something
int size = get(insert);
if(size >= 0) { //success
s_curr += size;
{
mt::mutexlocker locker(&(this->heap_lock));
if(final)
insert->count.ref(); //now lock is 0 and can be locked
this->heap.push(insert);
}
this->check_queue.open(); //we should signal the parent cache that we have a new item
return true;
} else { //failed transfer put it back, we will keep trying to transfer it...
mt::mutexlocker input_locker(&(input->heap_lock));
input->heap.push(insert);
return false;
}
}
return false;
}
}
return false;
}
};
} //namespace
/* TODO use the following class to allow multiple cache transfers at the same time */
/*

View File

@ -5,6 +5,7 @@
/** Allows to insert tokens, update priorities and generally control the cache.
*/
namespace vcg {
template <class Token>
class Controller {
@ -66,7 +67,6 @@ class Controller {
///ensure that added tokens are processed and existing ones have their priority updated.
///potential bug! update is done on the heaps, if something is in transit...
void updatePriorities() {
if(tokens.size()) {
mt::mutexlocker l(&provider.heap_lock);
for(unsigned int i = 0; i < tokens.size(); i++)
@ -140,6 +140,7 @@ class Controller {
void resume() {
assert(!stopped);
assert(paused);
cout << "Resume" << endl;
//unlock and open all doors
for(unsigned int i = 0; i < caches.size(); i++) {
@ -177,5 +178,5 @@ class Controller {
}
};
} //namespace
#endif // CONTROLLER_H

View File

@ -17,6 +17,8 @@
You should never interact with this class.
*/
namespace vcg {
template <typename Token>
class Provider: public mt::thread {
public:
@ -76,5 +78,5 @@ class Provider: public mt::thread {
}
};
} //namespace
#endif

View File

@ -15,6 +15,7 @@
The Priority template argument can simply be a floating point number
or something more complex, (frame and error in pixel); the only
requirement is the existence of a < comparison operator */
namespace vcg {
template <typename Priority>
class Token {
@ -83,4 +84,5 @@ class Token {
}
};
} //namespace
#endif // GCACHE_H