II. How to improve the efficiency of the use of memory
I have seen two of the following:
1. Design of the 2-level configurator in STL. I have implemented a simple vector, efficiency and STL Vector in accordance with the thinking of STL. The difference is far;). STL in memory allocation (alloc) The block has great effort (of course, not just memory usage). It is great to improve efficiency. Oh, I want to see more details, I will go to the teacher's << STL source analysis >> Chapter 2
2. SLAB algorithm, (The following is some of the information I collected)
3. The following is a description of the SLAB algorithm:
4. Assign objects: IF (free object position in the cache area corresponding to the object) {
5. Get this object without initialization}
6. Else {
7. Assign memory
8. Initialization object}
9. Release the object: Simply use the corresponding position to be idle without in the cache, without having to analyze.
10. If the system resources are insufficient: some unused object spaces are obtained from the cache, they analyze these objects, and release the space occupied by these objects to the system.
The following is a schematic diagram of the data structure of SLAB
11.
Cache_cache
Cache1 SLAB1 SLAB2 SLAB3 SLAB4
Cache2
C_firstp c_freep c_lastp
Cache3
Cache4
Cache5 s_freep
It is very simple to understand the code of the SLAB algorithm or the use of your own program. As long as you add 2 macro. If you don't want to download Myicq code, you can go here.
http://yj000885.sosoo.net/3ddoc/testslab.htm
It is to be explained. In general, it is not necessary to define the management method of memory. Generally, a large project, which is relatively high in performance (such as game engine). Generally, the way you define your own definition.
There is also an introduction to << Windows Advanced Programming >>. I think it can also improve performance to some extent.
1. This is a platform-related approach. Suitable for Win32 platforms.
We know, when calling new, the object is allocated on the heap. If there is a heap of each object, then create an object, destroy an object, then you need to assign a heap again, then It is more destroyed. The overhead is relatively large.
You can use the following ideas to reduce overhead. Create a heap for each class, all instances of this class are allocated on that heap. This can reduce overhead. Look at the following code:
Class testclass
{
Private:
Static hheap s_heap;
STATIC uint s_unumallosinheap;
PUBLIC:
// Reserve Operator New, Delete
Void * Operator New (size_t size)
{
IF (S_HHEAP == NULL)
{
_Try {
S_HHEAP = HeapCreate (/ * Stack of initial size * /, 0, 0);
}
_EXCEPT (Exception_Handler)
{
Return NULL;
}
............ .....
Void * p = (void *) HeapAlloc (S_HHEAP, 0, size); / / olive the processing exception code
IF (p) s_unumallocsinheap ;
Return P;
}
Void Operator Delete (Void * P)
{
IF (HEAPFREE (S_HHEAP, 0, P)
S_unumallocsinheap -;
IF (0 == s_unumallocsinheap) destroyed
S_HHEAP = NULL;
}
}
HHEAP TESTCLASS :: S_HHEAP = NULL;
UINT TESTCLASS :: S_unumallocsinheap = 0;
You can save overhead when there are multiple objects of multiple objects.
I have written so much. I am a bit sleepy. There are still many problems in my code. Please ask everyone to advice. Thank you!
