Philip Rebohle
c547039bef
[dxvk] Decide memory priority based on access flags rather than usage
...
Buffers used as shader resources are storage buffers as well,
and should not take priority over other read-only resources.
2020-11-27 12:20:55 +01:00
Philip Rebohle
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
Philip Rebohle
1211bb5e5f
[dxvk] Sort buffer slices before returning them to the buffer
...
Buffer slices often get shuffled over time due to timing and thread
synchronization, which makes it less and less likely for the dynamic
uniform buffer binding optimization to be effective. Sorting the
slices beforehand addresses the issue and may help CPU performance.
2019-11-23 00:42:54 +01:00
Philip Rebohle
cee7db1c57
[dxvk] Always align texel buffers to at least 16 bytes
...
Since we don't know the view format in advance, and some
drivers require single-texel alignment.
2019-11-18 13:45:35 +01:00
Philip Rebohle
3457e312b0
[dxvk] Always align index buffer slices to 256 Bytes
...
To be on the safe side.
2019-11-14 14:32:01 +01:00
Philip Rebohle
6948d18f5f
[dxvk] Always align vertex buffer slices to 256 bytes
...
Works around an unknown geometry rendering isssue in Warhammer Chaosbane.
Should probably be investigated at some point.
2019-11-10 22:37:11 +01:00
Philip Rebohle
d1e9e1392d
[dxvk] Lazily allocate slices of small buffers
...
Reduces memory overhead in case games create small static
buffers. This is somewhat common for index buffers used
to render fullscreen quads or triangles.
2019-10-26 15:24:55 +02:00
Philip Rebohle
e868f829b5
[dxvk] Pad buffers more tightly
...
This will allow for smaller buffers to be laid out more efficiently.
2019-10-26 15:24:54 +02:00
Philip Rebohle
bd5630439e
[dxvk] Inline freeBufferSlice implementation
...
May save a cycle or two in invalidateBuffer.
2019-10-11 13:06:55 +02:00
Philip Rebohle
acf6c27e76
[dxvk] Limit maximum size of multi-slice buffers
...
Reduces memory fragmentation caused by frequently renamed buffers.
2019-07-15 04:14:23 +02:00
Philip Rebohle
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
Philip Rebohle
01a7e06ad7
[dxvk] Inline DxvkBuffer::allocSlice and DxvkBuffer::freeSlice
...
Seems to take some load off both the main rendering thread and the
cleanup thread in SotTR.
2019-02-04 10:01:29 +01:00
Philip Rebohle
e7bf1647dd
[dxvk] Use high memory priority for GPU-writable resources
2019-01-30 13:52:43 +01:00
Philip Rebohle
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
Philip Rebohle
9dbe352c51
[dxvk] Fix incorrect buffer view range for views with non-zero offset
2019-01-18 21:26:48 +01:00
Philip Rebohle
8b5db80fbd
[dxvk] Reimplement DxvkBuffer
...
Avoids the DxvkPhysicalBuffer indirection and keeps all buffer
objects alive until the DxvkBuffer itself gets destroyed.
2019-01-09 18:04:46 +01:00
Philip Rebohle
cc61e38b9c
[dxvk] Reimplement DxvkBufferView
...
Instead of recreating the buffer view every single time the
underlying buffer gets invalidated, this keeps all buffer
views around until the object itself gets destroyed.
2019-01-09 18:02:00 +01:00
Philip Rebohle
504de3994b
[dxvk] Inline DxvkBufferView::updateView method
...
Most of the time the condition is false, so we should help
the compiler where we can to optimize for the common case.
2018-11-09 11:40:44 +01:00
Philip Rebohle
387f41ede5
[dxvk] Make DxvkBuffer destructor explicit
...
Prevents compiler from inlining massive amounts of code where it's
not needed. This is more consistent with the other classes as well.
2018-09-29 20:19:07 +02:00
Philip Rebohle
19f74a89cd
[dxvk] Optimize buffer renaming
...
Small reduction in CPU overhead by using spinlocks instead of
a full mutex for the two buffer free lists.
2018-07-19 09:44:58 +02:00
Philip Rebohle
32cd85dc11
[dxvk] Discard slices from old physical buffers
...
Improves effectiveness of an optimization that allows
using dynamic buffer offsets for descriptor sets.
2018-06-22 00:27:52 +02:00
Philip Rebohle
f762811af0
[dxvk] Use multiple free lists for physical buffer slices
...
Reduces lock contention and slightly improves performance in games
that rely heavily on the buffer renaming mechanism if the lock
protecting the original free list was contested.
2018-03-19 14:53:59 +01:00
Philip Rebohle
0eaa3fea3b
[dxvk] Implement thread-safe buffer renaming
...
This is required for resource mapping on deferred contexts.
May also fix a potential synchronization issue where a buffer
could be mapped multiple times before the CS thread would mark
the physical buffer as used, which would result in invalid data.
2018-03-19 02:18:44 +01:00
Philip Rebohle
360bf3ee4c
[dxvk] Recreate buffer views if necessary
...
Fixes broken animations and artifacts in The Witcher 3.
2018-03-07 13:54:28 +01:00
Philip Rebohle
cea86472bf
[dxvk] Fixed incorrect buffer create info
2018-01-29 11:44:02 +01:00
Philip Rebohle
d74be35e8c
[dxvk] Refactored physical buffer
...
Class is now much more generic, which should make things
easier for the upcomping deferred context implementation
2018-01-29 00:01:00 +01:00
Philip Rebohle
e4d49aeb66
[dxvk] Fixed reported size of staging buffers
...
Fixes staging slice allocation in Tomb Raider (2013),
and potentially other games which allocate large staging
buffers.
2018-01-24 15:44:40 +01:00
Philip Rebohle
a0acbeec72
[dxvk] Optimized buffer renaming
...
Helps applications which frequently update buffers
through either Map()/Unmap() or UpdateSubresource.
2018-01-19 00:20:05 +01:00
Philip Rebohle
6e6c290e01
[d3d11] Use DxvkPhysicalBufferSlice for resource copies and buffer views
2018-01-18 18:33:13 +01:00
Philip Rebohle
f41a7c7c87
[dxvk] Use DxvkPhysicalBufferSlice for buffer memory barriers
2018-01-18 17:32:34 +01:00
Philip Rebohle
a87ae8aba4
[dxvk] Added DxvkPhysicalBuffer to back virtual buffers
...
This is the first step to optimizing buffer updates for applications
that frequently invalidate buffers. The goal is to reduce the number
of buffer allocations per frame and reduce the cost of invalidation.
2018-01-18 15:52:57 +01:00
Philip Rebohle
f68655feff
[dxvk] Separated buffer renaming from allocation
2017-12-20 02:58:36 +01:00
Philip Rebohle
d3b2174180
[dxvk] Implemented buffer renaming
2017-12-16 13:21:11 +01:00
Philip Rebohle
b5d068366d
[dxvk] Implemented shader resource binding
2017-12-03 20:23:26 +01:00
Philip Rebohle
f924931d2d
[general] Removed TRACE calls for now
2017-11-26 14:01:41 +01:00
Philip Rebohle
47bdf9408b
[dxvk] Added buffer view class
2017-11-21 19:50:57 +01:00
Philip Rebohle
2c3674190a
[dxvk] Implemented proper resource binding
2017-10-15 19:23:10 +02:00
Philip Rebohle
44d9bd9000
[dxvk] Added buffer creation
2017-10-15 14:36:41 +02:00
Philip Rebohle
00e63d71a9
Initial commit
2017-10-10 23:32:13 +02:00
