2be0d6842e
[dxvk] Add option to limit memory chunk size
2023-03-01 11:53:18 +01:00
1c2df54bdf
[dxvk] Improve logging in case of memory errors
2023-01-14 03:39:55 +01:00
2922b780c1
[dxvk] Properly handle bufferImageGranularity for images
...
Fixes validation errors about dedicated allocations being
bigger than the image's memory requirement on Nvidia.
2023-01-14 03:39:55 +01:00
8052347a22
[dxvk] Keep more empty system memory chunks alive
2023-01-10 03:42:37 +01:00
0b9f785bc5
[dxvk] Remove broken memory budget logic for UMA systems
...
All heaps on an UMA systems are by definition device-local, so this
just leads to unnecessary issues.
2022-11-01 12:48:03 +01:00
f1f8d45fcd
[dxvk] Rework allocation logic for large resources
...
This may reduce internal fragmentation with very large resources.
We previously changed behaviour to not do this in order to reduce memory
pressure in the average case, however by trying to suballocate from existing
chunks and falling back to a dedicated allocation on failure, rather than
allocating a new chunk, we can mostly avoid that situation.
2022-09-24 17:06:00 +02:00
15588004b4
[dxvk] Bump memory chunk size to 256 MiB.
2022-09-24 17:06:00 +02:00
51e0a56243
[dxvk] Use new pseudo-features where applicable
2022-09-04 18:28:27 +02:00
f9db4921e0
[dxvk] Implement sparse memory allocator
2022-08-26 05:53:03 +02:00
a2a21cb4d3
[dxvk] Clean up internal memory allocation API
2022-08-25 18:23:02 +02:00
67f937111d
[dxvk] Remove shrinkNvidiaHVVHeap workaround
...
No longer needed.
2022-08-10 16:39:58 +02:00
a76b5693f3
[dxvk] Clean up remaining object creation code
2022-07-18 14:52:12 +02:00
81b89cf31d
[dxvk] Simplify Nvidia HVV workaround
...
We no longer support 465 series drivers, so the check was obsolete.
2022-07-14 21:21:34 +02:00
131af0d677
[dxvk] Replace VK_KHR_driver_properties with core feature
2022-07-14 21:21:34 +02:00
425fce9200
[dxvk] Introduce transient memory flag for staging buffers
...
Potentially reduces fragmentation by putting short-lived staging buffers
and sysmem resources created by the application into different memory pools.
2022-02-13 02:08:20 +01:00
9d4be00fa7
[dxvk] Allow large sysmem allocations on 64-bit platforms again
...
Since we frequently discard staging buffers now, having larger chunks
is actually beneficial again.
2022-02-13 02:08:20 +01:00
a268de51e3
[dxvk] Don't shrink HVV heap if it's the main video memory heap
...
May fix issues on RBAR-enabled Nvidia systems.
2022-01-16 02:47:51 +01:00
db2e32ede6
[dxvk] Be smarter about which chunks to free
...
Freeing all empty chunks immediately may cause issues if an app constantly
allocates and frees a small number of resources that don't fit into any
existing chunk, so try to keep one around. Aggressively free everything
under memory pressure if necessary.
2022-01-13 18:39:31 +01:00
15ab07ab94
[dxvk] Replace allocation priority with allocation flags
2022-01-13 18:39:31 +01:00
d34bbdb58e
[dxvk] Free empty memory chunks
2022-01-13 18:22:44 +01:00
778aee0372
[dxvk] Rework Nvidia HVV option
2021-11-02 00:52:25 +01:00
aa4b7c9f92
[dxvk] Rework 32-bit check
2021-10-01 14:18:42 +02:00
cd8a2bcfcd
[dxvk] Use custom sync primitives
2021-06-29 00:51:20 +02:00
1c2edabbcb
[dxvk] Disable NVIDIA HVV bug workaround on fixed drivers
...
Reviewed-by: Arthur Huillet <ahuillet@nvidia.com>
2021-03-05 11:21:50 +01:00
b25d6ba615
[dxvk] Add option to disable workaround for NVIDIA HVV bug
...
Adds a new dxvk.halveNvidiaHVVHeap option.
Reviewed-by: Arthur Huillet <ahuillet@nvidia.com>
2021-03-05 11:21:50 +01:00
16a51f3c03
[dxvk] Only use half of the DEVICE_LOCAL | HOST_VISIBLE heap on Nvidia
...
Seems to help with random crashes in FFXIV and potentially
other games on 450 series drivers.
2020-08-22 11:51:56 +02:00
e435e071e0
[dxvk] Introduce memory heap budget
...
Allows more fine-grained control of memory allocations for specific
heaps. For now, target 80% for device-local heaps on UMA devices.
2020-08-22 11:50:37 +02:00
48777c8fcf
[dxvk] Remove optional memory property flags one by one
...
We don't always want to remove HOST_CACHED if no DEVICE_LOCAL
type exists, so iterate over the bits one by one.
2020-08-13 18:47:52 +02:00
ca4c03284f
[dxvk] Don't oversubscribe memory heaps on UMA devices
...
Otherwise, we seem to suffer a major performance penalty
on setups with insufficient dedicated system memory.
2020-01-28 16:31:34 +00:00
60145f0167
[dxvk] Don't enable VK_KHR_dedicated_allocation, VK_KHR_get_memory_requirements2
...
Core in Vulkan 1.1.
2020-01-17 17:46:59 +01:00
7a18cb1227
[dxvk] Remove unused memory stat counters
2019-12-13 12:08:36 +01:00
493b55b073
[dxvk] Compute memory chunk size per memory type rather than per heap
2019-07-16 09:59:44 +02:00
6936da17d9
[dxvk] Try harder to allocate memory from a given memory type
...
Before failing, see if we can't allocate a smaller chunk size.
2019-07-16 09:59:44 +02:00
18aada29ef
[dxvk] Don't put large resources into their own memory allocations
...
This approach currently has two issues:
- We might fail to allocate the resource on the desired memory type
even if there is a chunk available that it would tif in
- With 128MB chunks, this no longer seems to be beneficial anyway
2019-07-16 09:24:55 +02:00
e116ff49e6
[dxvk] Rename memoryAvailable -> memoryBudget
2019-07-13 19:04:33 +02:00
3b1376b2fe
[dxvk] Increase memory chunk size to 128 MiB
...
Nvidia drivers apparently don't like smaller chunks very much.
May increase overall memory consumption and fragmentation.
2019-07-04 16:11:13 +02:00
3b128179ab
[dxvk] Print VK_EXT_memory_budget stats on allocation failure
...
Based on PR #1112 , but using a cleaner method to check
support for the VK_EXT_memory_budget extension.
Suggested-by: Liam Middlebrook <lmiddlebrook@nvidia.com>
2019-07-03 11:11:12 +02:00
cb806a5d56
[dxvk] Free memory if vkMapMemory fails
...
Otherwise, we leak the allocated memory chunk and make the
problem even worse than it already is.
2019-07-02 01:26:54 +02:00
138dde6c3d
[dxvk] Prefer VRAM allocation over dedicated sysmem allocation
...
May in some cases improve performance when under memory pressure:
If a dedicated allocation is preferred, but the alloaction fails,
try to allocate memory from an already allocated chunk instead
of falling back to system memory right away.
2019-06-20 10:34:54 +02:00
0a35ae6fca
[dxvk] Log memory utilization per heap on memory allocation failures
2019-06-19 14:47:09 +02:00
f272071d8d
[dxvk] Don't enforce HOST_CACHED flag when allocating memory
...
The better fix would be to support non-coherent memory properly,
but this will have to do for now. Fixes #947 .
2019-03-14 16:47:17 +01:00
c360a196fb
[dxvk] Restore dedicated allocation behaviour
...
Whoops.
2019-01-30 15:49:06 +01:00
b76f470c5f
[dxvk] Remove dxvk.allowMemoryOvercommit option
...
The default behaviour is now to allow overallocation by default,
which makes sense especially when VK_EXT_memory_priority is used.
Does not seem to affect Nvidia drivers.
2019-01-30 13:52:43 +01:00
dfa8524e73
[dxvk] Wire up memory priority to device memory allocations
2019-01-30 13:52:43 +01:00
03b488772d
[dxvk] Store device instead of adapter in memory allocator
2019-01-30 13:52:43 +01:00
134613375d
[dxvk] Introduce concept of priorities to device-local memory allocations
...
The priority defines the importance of any given allocation being
resident in device-local memory rather than system memory, with
values between 0 (low) and 1 (high).
Suballocations will only work on chunks with the requested priority.
This ensures that high-priority allocations are reserved solely for
high-priority resources, and that high-priority resources will be
allocated from a high-priority memory chunk.
2019-01-30 13:52:43 +01:00
1af96347e1
[dxvk] Forward memory allocation statistics to the adapter
2018-11-15 17:08:05 +01:00
eac86fab15
[dxvk] Do not map allocations that don't need a HOST_VISIBLE memory type
...
Should reduce the amount of virtual memory used on systems without
dedicated VRAM. Refs #640 .
2018-09-17 09:08:00 +02:00
b2c4855490
[dxvk] Use global user config for backend options
2018-08-07 16:59:49 +02:00
43cbe42ea8
[dxvk] Pick memory chunk size dynamically
...
Reduces the number of Vulkan memory allocations, and
can lead to lower CPU overhead in certain situations.
Improves Shadow Warrior 2 by ~2% on RADV.
2018-07-09 19:18:39 +02:00
Philip Rebohle
Liam Middlebrook