Commit Graph

262 Commits

Author SHA1 Message Date
Adam Sawicki
ae3970387e Added internal class VmaDefragmentationAlgorithm_Fast::FeeSpaceDatabase.
Defragmentation algorithm is pretty much finished now!
2018-12-06 18:06:08 +01:00
Adam Sawicki
2af57d7f47 Fixed bug in VmaBlockMetadata_Generic::IsBufferImageGranularityConflictPossible. Another minor fix. 2018-12-06 15:35:05 +01:00
Adam Sawicki
0eeed59309 Merge branch 'master' into allocation_defragmentation_strategies
# Conflicts:
#	docs/html/vk__mem__alloc_8h_source.html
2018-12-06 14:52:11 +01:00
Adam Sawicki
e8b01485ce Small change not to #include <vulkan/vulkan.h> if it was already included (possible from a different path) before including VMA. Issue #50 Thanks @baldurk ! 2018-12-06 14:47:59 +01:00
Adam Sawicki
647cf24b7f VmaDefragmentationAlgorithm_Fast: Added support for memmove() of overlapping memory regions when defragmenting on CPU. 2018-11-23 17:58:00 +01:00
Adam Sawicki
9a4f508f88 Added VmaDefragmentationAlgorithm_Fast - fast path for defragmentation, a completely new algorithm. 2018-11-23 17:26:05 +01:00
Adam Sawicki
da85ec36bf Made VmaDefragmentationAlgorithm an abstract class and specific implementation as new class VmaDefragmentationAlgorithm_Generic, so user can easily plug his own defragmentation algorithms. 2018-11-22 17:32:44 +01:00
Adam Sawicki
fb00cc9ea8 Removed VMA_DEFRAGMENTATION_* flags. VmaDefragmentationInfo2::flags is now reserved for future use and should be 0. Research various options and chosen some parameters as default. 2018-11-22 17:10:07 +01:00
Adam Sawicki
434ac86781 Minor improvement in vmaDefragmentationBegin. 2018-11-22 16:20:49 +01:00
Adam Sawicki
52076ebf26 Added VmaDefragmentationInfo2::poolCount, pPools. Added test for it - TestDefragmentationWholePool. Removed VmaDefragmentationStats::allocationsLost. Optimized defragmentation algorithm. 2018-11-22 16:14:50 +01:00
Adam Sawicki
3b392258e2 Documentation: Added mention of VK_AMD_memory_overallocation_behavior extension. 2018-11-20 11:33:07 +01:00
Adam Sawicki
8ec2ab69d6 VmaReplay: Added parameter --DefragmentatationFlags <Flags>. 2018-11-16 17:22:20 +01:00
Adam Sawicki
5249980132 Implemeneted VMA_DEFRAGMENTATION_FAST_ALGORITHM_BIT, VMA_DEFRAGMENTATION_OPTIMAL_ALGORITHM_BIT. Not tested yet.
Added VMA_ALLOCATION_INTERNAL_STRATEGY_MIN_OFFSET.
2018-11-16 16:15:18 +01:00
Adam Sawicki
9948ca0ceb Removed VMA_DEFRAGMENTATION_CAN_MAKE_LOST_BIT. Not going to implement it for now... 2018-11-16 14:46:04 +01:00
Adam Sawicki
48a12aa835 Merge branch 'master' into allocation_defragmentation_strategies
# Conflicts:
#	docs/html/vk__mem__alloc_8h_source.html
#	src/Tests.cpp
#	src/vk_mem_alloc.h
2018-11-16 14:36:57 +01:00
Adam Sawicki
99f2d3543c Minor update in documentation. 2018-11-16 13:48:11 +01:00
Adam Sawicki
7f97202add Updated documentation of memory mapping with known bug in MoltenVK, based on #47. Thanks @DiegoAce ! 2018-11-16 13:43:34 +01:00
Adam Sawicki
b0c363693f Added function vmaResizeAllocation.
Added tests: function TestResize. Bumped CSV recording file format version to 1.4.
2018-11-13 16:17:38 +01:00
Adam Sawicki
1146b45c63 Fixed aligned_alloc for Android API < 16. Issue #49. Thanks @zhaijialong ! 2018-11-09 11:21:03 +01:00
Adam Sawicki
7a6e442b66 Fixes in documentation: "Features not supported". 2018-11-08 18:46:04 +01:00
Adam Sawicki
370ab184f8 Changed functions to validate and return VK_ERROR_VALIDATION_FAILED_EXT when trying to allocate memory of size 0, create buffer with size 0, or image with one of the dimensions 0.
That's because vkCreateBuffer returns VK_SUCCESS for buffer with size = 0, so VMA then proceeded to allocation of size 0, which is a critical error, checked by an assert, but apparently some users don't enable asserts in VMA.

Added tests: function TestInvalidAllocations.
2018-11-08 16:31:00 +01:00
Adam Sawicki
dec65ac4eb Small addition to documentation. 2018-11-06 12:06:06 +01:00
Adam Sawicki
fa251ad7c8 Locked right mutexes during defragmentation. GPU memory defragmentation is now finished! 2018-10-18 15:22:40 +02:00
Adam Sawicki
76c5bcabfd Further refactoring of defragmentation, preparing for defragmentation of GPU memory. 2018-10-18 14:14:28 +02:00
Adam Sawicki
29b04041f7 Further refactoring of defragmentation classes. 2018-10-18 13:11:00 +02:00
Adam Sawicki
fc9db8dda7 Minor addition to documentation. #46 2018-10-18 12:03:50 +02:00
Adam Sawicki
2dcfcf8b63 More refactoring in preparation for GPU memory defragmentation. Created classes: VmaBlockDefragmentationContext, VmaBlockVectorDefragmentationContext. Class VmaDefragmentationContext_T is no longer empty. 2018-10-17 17:26:39 +02:00
Adam Sawicki
a9f030d7ba Further refactoring of defragmentation. 2018-10-17 15:20:36 +02:00
Adam Sawicki
a114419b23 Refactored defragmentation code. Removed class VmaDefragmentator. Renamed struct to VmaDefragmentationMove.
Minor fixes in documentation and comments.
2018-10-16 15:30:55 +02:00
Adam Sawicki
ef6cc40b59 Defragmentation: added support for memory that is HOST_VISIBLE but not HOST_COHERENT - calling Invalidate and Flush internally.
Fixed a bug in VmaDefragmentator::Defragment.
2018-10-16 14:20:47 +02:00
Adam Sawicki
fa87ae34a6 Small addition to documentation. 2018-10-15 18:15:11 +02:00
Adam Sawicki
6826f2d768 Small refactoring: Removed function VmaBlockVector::DestroyDefragmentator. 2018-10-10 16:49:36 +02:00
Adam Sawicki
f863a1dbd0 Used VMA_RW_MUTEX in 2 other places for optimization. 2018-10-10 16:43:44 +02:00
Adam Sawicki
de9181b2ba Added configuration macro VMA_USE_STL_SHARED_MUTEX.
Added internal class VmaRWMutex, macro VMA_RW_MUTEX, classes VmaMutexLockRead, VmaMutexLockWrite. Used it in VmaAllocator_T::m_PoolsMutex.
2018-10-10 16:26:26 +02:00
Adam Sawicki
b0d4afbeed Refactored defragmentation. Added class VmaDefragmentationAlgorithm, extracted from VmaDefragmentator. 2018-10-09 15:35:46 +02:00
Adam Sawicki
2aad902a66 Minor update in VmaDefragmentationInfo2. 2018-10-09 13:43:34 +02:00
Adam Sawicki
ad0989bfb4 Fixed bug in VmaAllocator_T::Defragment. 2018-10-09 13:26:33 +02:00
Adam Sawicki
012a4ac697 Designed and documented new interface for defragmentation.
Added: VmaDefragmentationContext, VmaDefragmentationFlagBits, VmaDefragmentationFlags, VmaDefragmentationInfo2, vmaDefragmentationBegin(), vmaDefragmentationEnd(). Extended VmaDefragmentationStats by allocationsLost. Deprecated VmaDefragmentationInfo2, vmaDefragment().
2018-10-09 13:25:01 +02:00
Adam Sawicki
7ce9630bbf Improved documentation chapters: Defragmentation, vmaDefragment(), Features not supported. 2018-09-28 14:51:09 +02:00
Adam Sawicki
6aa6255741 Fixing macro VMA_VALIDATE for Linux compilation. 2018-09-24 15:16:47 +02:00
Adam Sawicki
02246f3eb8 CHANGING BRANCHING STRATEGY! No more "development" branch, now in-progress development will happen on "master".
Bumped version number to "development".
2018-09-24 15:10:06 +02:00
Adam Sawicki
3951aa5bc3 Merge branch 'allocation_defragmentation_strategies' into development
# Conflicts:
#	docs/html/search/all_f.js
#	docs/html/vk__mem__alloc_8h.html
#	docs/html/vk__mem__alloc_8h_source.html
#	src/vk_mem_alloc.h
2018-09-21 16:48:42 +02:00
Adam Sawicki
c6432d1d45 Buddy allocation algorithm finished and documented! 2018-09-21 16:44:16 +02:00
Adam Sawicki
0591535eb1 VmaBlockMetadata_Buddy: Fixed allocation of Node objects to use provided CPU allocation callbacks. 2018-09-21 15:23:32 +02:00
Adam Sawicki
1e8cf94558 VmaBlockMetadata_Buddy: Added (simple way of) respecting bufferImageGranularity.
Minor fixes in documentation.
2018-09-21 15:10:04 +02:00
Adam Sawicki
9933c5cadf VmaBlockMetadata_Buddy: Added respecting of allocation alignment. 2018-09-21 14:57:24 +02:00
Adam Sawicki
a01d4587df VmaBlockMetadata_Buddy: Introduced m_LevelCount to limit number of levels in use by particular memory block, considering new constant MIN_NODE_SIZE. 2018-09-21 14:22:35 +02:00
Adam Sawicki
d6e6d6bdf2 VmaBlockMetadata_Buddy: Introduced concept of m_UsableSize to always use powers of two even when memory block size is not. 2018-09-21 14:07:02 +02:00
Adam Sawicki
a79d2746f1 Added dummy implementation of VmaBlockMetadata_Buddy methods related to lost allocations. Lost allocations are not supported in buddy algorithm for now. 2018-09-21 13:26:12 +02:00
Adam Sawicki
14d9e1a938 Minor optimization: Skip call to vkUnmapMemory before vkFreeMemory for dedicated allocations, which is allowed by the spec. 2018-09-17 17:06:42 +02:00