Make RID_Owner lock-free for fetching.

This PR makes RID_Owner lock free for fetching values, this should give a very
significant peformance boost where used.

Some considerations:

* A maximum number of elements to alocate must be given (by default 256k).
* Access to the RID structure is still safe given they are independent from addition/removals.
* RID access was never really thread-safe in the sense that the contents of the data are not protected anyway. Each server needs to implement locking as it sees fit.
This commit is contained in:
Juan Linietsky 2023-12-19 17:34:48 +01:00 committed by Dario
parent 0a9d8f04c1
commit 08947d366f
1 changed files with 68 additions and 71 deletions

View File

@ -32,7 +32,7 @@
#define RID_OWNER_H
#include "core/os/memory.h"
#include "core/os/spin_lock.h"
#include "core/os/mutex.h"
#include "core/string/print_string.h"
#include "core/templates/hash_set.h"
#include "core/templates/list.h"
@ -69,42 +69,54 @@ public:
template <typename T, bool THREAD_SAFE = false>
class RID_Alloc : public RID_AllocBase {
T **chunks = nullptr;
struct Chunk {
T data;
uint32_t validator;
};
Chunk **chunks = nullptr;
uint32_t **free_list_chunks = nullptr;
uint32_t **validator_chunks = nullptr;
uint32_t elements_in_chunk;
uint32_t max_alloc = 0;
uint32_t alloc_count = 0;
uint32_t chunk_limit = 0;
const char *description = nullptr;
mutable SpinLock spin_lock;
mutable Mutex mutex;
_FORCE_INLINE_ RID _allocate_rid() {
if constexpr (THREAD_SAFE) {
spin_lock.lock();
mutex.lock();
}
if (alloc_count == max_alloc) {
//allocate a new chunk
uint32_t chunk_count = alloc_count == 0 ? 0 : (max_alloc / elements_in_chunk);
if (THREAD_SAFE && chunk_count == chunk_limit) {
mutex.unlock();
if (description != nullptr) {
ERR_FAIL_V_MSG(RID(), vformat("Element limit for RID of type '%s' reached.", String(description)));
} else {
ERR_FAIL_V_MSG(RID(), "Element limit reached.");
}
}
//grow chunks
chunks = (T **)memrealloc(chunks, sizeof(T *) * (chunk_count + 1));
chunks[chunk_count] = (T *)memalloc(sizeof(T) * elements_in_chunk); //but don't initialize
//grow validators
validator_chunks = (uint32_t **)memrealloc(validator_chunks, sizeof(uint32_t *) * (chunk_count + 1));
validator_chunks[chunk_count] = (uint32_t *)memalloc(sizeof(uint32_t) * elements_in_chunk);
if constexpr (!THREAD_SAFE) {
chunks = (Chunk **)memrealloc(chunks, sizeof(Chunk *) * (chunk_count + 1));
}
chunks[chunk_count] = (Chunk *)memalloc(sizeof(Chunk) * elements_in_chunk); //but don't initialize
//grow free lists
if constexpr (!THREAD_SAFE) {
free_list_chunks = (uint32_t **)memrealloc(free_list_chunks, sizeof(uint32_t *) * (chunk_count + 1));
}
free_list_chunks[chunk_count] = (uint32_t *)memalloc(sizeof(uint32_t) * elements_in_chunk);
//initialize
for (uint32_t i = 0; i < elements_in_chunk; i++) {
// Don't initialize chunk.
validator_chunks[chunk_count][i] = 0xFFFFFFFF;
chunks[chunk_count][i].validator = 0xFFFFFFFF;
free_list_chunks[chunk_count][i] = alloc_count + i;
}
@ -122,14 +134,13 @@ class RID_Alloc : public RID_AllocBase {
id <<= 32;
id |= free_index;
validator_chunks[free_chunk][free_element] = validator;
validator_chunks[free_chunk][free_element] |= 0x80000000; //mark uninitialized bit
chunks[free_chunk][free_element].validator = validator;
chunks[free_chunk][free_element].validator |= 0x80000000; //mark uninitialized bit
alloc_count++;
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
mutex.unlock();
}
return _make_from_id(id);
@ -156,16 +167,10 @@ public:
if (p_rid == RID()) {
return nullptr;
}
if constexpr (THREAD_SAFE) {
spin_lock.lock();
}
uint64_t id = p_rid.get_id();
uint32_t idx = uint32_t(id & 0xFFFFFFFF);
if (unlikely(idx >= max_alloc)) {
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
}
return nullptr;
}
@ -174,38 +179,26 @@ public:
uint32_t validator = uint32_t(id >> 32);
Chunk &c = chunks[idx_chunk][idx_element];
if (unlikely(p_initialize)) {
if (unlikely(!(validator_chunks[idx_chunk][idx_element] & 0x80000000))) {
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
}
if (unlikely(!(c.validator & 0x80000000))) {
ERR_FAIL_V_MSG(nullptr, "Initializing already initialized RID");
}
if (unlikely((validator_chunks[idx_chunk][idx_element] & 0x7FFFFFFF) != validator)) {
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
}
if (unlikely((c.validator & 0x7FFFFFFF) != validator)) {
ERR_FAIL_V_MSG(nullptr, "Attempting to initialize the wrong RID");
}
validator_chunks[idx_chunk][idx_element] &= 0x7FFFFFFF; //initialized
c.validator &= 0x7FFFFFFF; //initialized
} else if (unlikely(validator_chunks[idx_chunk][idx_element] != validator)) {
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
}
if ((validator_chunks[idx_chunk][idx_element] & 0x80000000) && validator_chunks[idx_chunk][idx_element] != 0xFFFFFFFF) {
} else if (unlikely(c.validator != validator)) {
if ((c.validator & 0x80000000) && c.validator != 0xFFFFFFFF) {
ERR_FAIL_V_MSG(nullptr, "Attempting to use an uninitialized RID");
}
return nullptr;
}
T *ptr = &chunks[idx_chunk][idx_element];
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
}
T *ptr = &c.data;
return ptr;
}
@ -222,14 +215,14 @@ public:
_FORCE_INLINE_ bool owns(const RID &p_rid) const {
if constexpr (THREAD_SAFE) {
spin_lock.lock();
mutex.lock();
}
uint64_t id = p_rid.get_id();
uint32_t idx = uint32_t(id & 0xFFFFFFFF);
if (unlikely(idx >= max_alloc)) {
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
mutex.unlock();
}
return false;
}
@ -239,10 +232,10 @@ public:
uint32_t validator = uint32_t(id >> 32);
bool owned = (validator != 0x7FFFFFFF) && (validator_chunks[idx_chunk][idx_element] & 0x7FFFFFFF) == validator;
bool owned = (validator != 0x7FFFFFFF) && (chunks[idx_chunk][idx_element].validator & 0x7FFFFFFF) == validator;
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
mutex.unlock();
}
return owned;
@ -250,14 +243,14 @@ public:
_FORCE_INLINE_ void free(const RID &p_rid) {
if constexpr (THREAD_SAFE) {
spin_lock.lock();
mutex.lock();
}
uint64_t id = p_rid.get_id();
uint32_t idx = uint32_t(id & 0xFFFFFFFF);
if (unlikely(idx >= max_alloc)) {
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
mutex.unlock();
}
ERR_FAIL();
}
@ -266,26 +259,26 @@ public:
uint32_t idx_element = idx % elements_in_chunk;
uint32_t validator = uint32_t(id >> 32);
if (unlikely(validator_chunks[idx_chunk][idx_element] & 0x80000000)) {
if (unlikely(chunks[idx_chunk][idx_element].validator & 0x80000000)) {
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
mutex.unlock();
}
ERR_FAIL_MSG("Attempted to free an uninitialized or invalid RID.");
} else if (unlikely(validator_chunks[idx_chunk][idx_element] != validator)) {
ERR_FAIL_MSG("Attempted to free an uninitialized or invalid RID");
} else if (unlikely(chunks[idx_chunk][idx_element].validator != validator)) {
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
mutex.unlock();
}
ERR_FAIL();
}
chunks[idx_chunk][idx_element].~T();
validator_chunks[idx_chunk][idx_element] = 0xFFFFFFFF; // go invalid
chunks[idx_chunk][idx_element].data.~T();
chunks[idx_chunk][idx_element].validator = 0xFFFFFFFF; // go invalid
alloc_count--;
free_list_chunks[alloc_count / elements_in_chunk][alloc_count % elements_in_chunk] = idx;
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
mutex.unlock();
}
}
@ -294,34 +287,35 @@ public:
}
void get_owned_list(List<RID> *p_owned) const {
if constexpr (THREAD_SAFE) {
spin_lock.lock();
mutex.lock();
}
for (size_t i = 0; i < max_alloc; i++) {
uint64_t validator = validator_chunks[i / elements_in_chunk][i % elements_in_chunk];
uint64_t validator = chunks[i / elements_in_chunk][i % elements_in_chunk].validator;
if (validator != 0xFFFFFFFF) {
p_owned->push_back(_make_from_id((validator << 32) | i));
}
}
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
mutex.unlock();
}
}
//used for fast iteration in the elements or RIDs
void fill_owned_buffer(RID *p_rid_buffer) const {
if constexpr (THREAD_SAFE) {
spin_lock.lock();
mutex.lock();
}
uint32_t idx = 0;
for (size_t i = 0; i < max_alloc; i++) {
uint64_t validator = validator_chunks[i / elements_in_chunk][i % elements_in_chunk];
uint64_t validator = chunks[i / elements_in_chunk][i % elements_in_chunk].validator;
if (validator != 0xFFFFFFFF) {
p_rid_buffer[idx] = _make_from_id((validator << 32) | i);
idx++;
}
}
if constexpr (THREAD_SAFE) {
spin_lock.unlock();
mutex.unlock();
}
}
@ -329,8 +323,13 @@ public:
description = p_descrption;
}
RID_Alloc(uint32_t p_target_chunk_byte_size = 65536) {
RID_Alloc(uint32_t p_target_chunk_byte_size = 65536, uint32_t p_maximum_number_of_elements = 262144) {
elements_in_chunk = sizeof(T) > p_target_chunk_byte_size ? 1 : (p_target_chunk_byte_size / sizeof(T));
if constexpr (THREAD_SAFE) {
chunk_limit = (p_maximum_number_of_elements / elements_in_chunk) + 1;
chunks = (Chunk **)memalloc(sizeof(Chunk *) * chunk_limit);
free_list_chunks = (uint32_t **)memalloc(sizeof(uint32_t *) * chunk_limit);
}
}
~RID_Alloc() {
@ -339,12 +338,12 @@ public:
alloc_count, description ? description : typeid(T).name()));
for (size_t i = 0; i < max_alloc; i++) {
uint64_t validator = validator_chunks[i / elements_in_chunk][i % elements_in_chunk];
uint64_t validator = chunks[i / elements_in_chunk][i % elements_in_chunk].validator;
if (validator & 0x80000000) {
continue; //uninitialized
}
if (validator != 0xFFFFFFFF) {
chunks[i / elements_in_chunk][i % elements_in_chunk].~T();
chunks[i / elements_in_chunk][i % elements_in_chunk].data.~T();
}
}
}
@ -352,14 +351,12 @@ public:
uint32_t chunk_count = max_alloc / elements_in_chunk;
for (uint32_t i = 0; i < chunk_count; i++) {
memfree(chunks[i]);
memfree(validator_chunks[i]);
memfree(free_list_chunks[i]);
}
if (chunks) {
memfree(chunks);
memfree(free_list_chunks);
memfree(validator_chunks);
}
}
};
@ -419,8 +416,8 @@ public:
alloc.set_description(p_descrption);
}
RID_PtrOwner(uint32_t p_target_chunk_byte_size = 65536) :
alloc(p_target_chunk_byte_size) {}
RID_PtrOwner(uint32_t p_target_chunk_byte_size = 65536, uint32_t p_maximum_number_of_elements = 262144) :
alloc(p_target_chunk_byte_size, p_maximum_number_of_elements) {}
};
template <typename T, bool THREAD_SAFE = false>
@ -473,8 +470,8 @@ public:
void set_description(const char *p_descrption) {
alloc.set_description(p_descrption);
}
RID_Owner(uint32_t p_target_chunk_byte_size = 65536) :
alloc(p_target_chunk_byte_size) {}
RID_Owner(uint32_t p_target_chunk_byte_size = 65536, uint32_t p_maximum_number_of_elements = 262144) :
alloc(p_target_chunk_byte_size, p_maximum_number_of_elements) {}
};
#endif // RID_OWNER_H