Philip Rebohle
|
d729567e7a
|
[dxvk] Refactor AccessOp
Allows us to do some more fun stuff with it going forward.
|
2025-03-07 12:43:31 +01:00 |
|
Philip Rebohle
|
a08579e555
|
[dxvk] Fix clear after late resolve
|
2025-03-06 09:54:34 +01:00 |
|
Philip Rebohle
|
45ec01a0a1
|
[d3d11] Add TRANSFER_SRC usage to icb
Needed for defrag.
|
2025-03-06 09:54:34 +01:00 |
|
Philip Rebohle
|
542e0d2ab0
|
[dxvk,d3d9,d3d11] Decouple ID3DUserDefinedAnnotation from internal markers
Some games nope out when we expose debug markers, so add a debug mode
that doesn't while still providing internal markers and debug names.
|
2025-03-06 09:54:34 +01:00 |
|
Philip Rebohle
|
91b48dd31d
|
[d3d11] Make user defined annotation more robust
|
2025-03-06 09:54:34 +01:00 |
|
GranMinigun
|
d8eb4d0d66
|
[util] Spoof vendor ID for CivCity: Rome
|
2025-03-06 01:25:30 +01:00 |
|
Philip Rebohle
|
f161d9bc7b
|
[dxvk] Fix synchronization on swapchain tear-down
|
2025-03-05 23:16:28 +01:00 |
|
Philip Rebohle
|
634f38b38d
|
[dxvk] Enable sparse binding features if supported
|
2025-03-05 23:15:47 +01:00 |
|
Philip Rebohle
|
ff8378be19
|
[dxvk] Fix uploadImage barriers
|
2025-03-05 23:15:47 +01:00 |
|
Philip Rebohle
|
ad75fb17cd
|
[dxvk] Fix uploadBuffer barriers
|
2025-03-05 23:15:47 +01:00 |
|
Philip Rebohle
|
76244812fb
|
[dxvk] Move compressed image initialization to transfer queue
This is relatively common.
|
2025-03-05 23:15:47 +01:00 |
|
Philip Rebohle
|
9c0dec9f58
|
[dxvk] Move buffer initialization to transfer queue
|
2025-03-05 23:15:47 +01:00 |
|
Philip Rebohle
|
a4a5bf5d63
|
[dxvk] Add helpers for transfer queue release barriers
|
2025-03-05 23:15:47 +01:00 |
|
Philip Rebohle
|
49f0968f57
|
[dxvk] Use sparse residency for zero buffer
Saves a small amount of memory since reading from an unbound
buffer will return zero anyway.
|
2025-03-05 23:15:47 +01:00 |
|
Philip Rebohle
|
ec0deb73da
|
[dxvk] Fix missing zero buffer tracking
Scary.
|
2025-03-05 23:15:47 +01:00 |
|
WinterSnowfall
|
fed51e6c92
|
[dxso] Fix uninitialized m_maxDefinedConstant variable
|
2025-03-05 15:02:11 +01:00 |
|
Philip Rebohle
|
196258111c
|
[dxvk] Be more conservative with CS data alignment
Fixes #4734. Apparently, weak alignment hints are an error rather
than getting ignored.
|
2025-03-05 13:12:23 +01:00 |
|
Philip Rebohle
|
d04fe1cdc0
|
[meta] Add some missing documentation to dxvk.conf
|
2025-03-04 21:49:39 +01:00 |
|
Philip Rebohle
|
3cf453160c
|
[d3d11] Ensure required image usage flags for video blits
Fixes some validation errors.
|
2025-03-04 21:05:59 +01:00 |
|
Philip Rebohle
|
9769df9dd8
|
[d3d11] Implement ID3D11On12Device1
|
2025-03-02 19:21:39 +01:00 |
|
Philip Rebohle
|
106032fa65
|
[dxvk] Use granular image access tracking in copyImageHw
|
2025-03-02 14:01:09 +01:00 |
|
Philip Rebohle
|
d2b53b76de
|
[dxvk] Use granular image access tracking in copyBufferToImageHw
|
2025-03-02 14:01:09 +01:00 |
|
Philip Rebohle
|
02dc403074
|
[dxvk] Use granular image access tracking in copyImageToBufferCs
|
2025-03-02 14:01:09 +01:00 |
|
Philip Rebohle
|
420d083677
|
[dxvk] Use granular image access tracking in copyImageToBufferHw
|
2025-03-02 14:01:09 +01:00 |
|
Philip Rebohle
|
cd29f0feaa
|
[dxvk] Add barrier helpers for granular image access tracking
|
2025-03-02 14:01:09 +01:00 |
|
Philip Rebohle
|
baaf72a373
|
[dxvk] Add helpers to compute more granular image address ranges
|
2025-03-02 14:01:09 +01:00 |
|
Philip Rebohle
|
1f0ad760e1
|
[dxvk] Refactor address range for barrier tracking
Allows us to bump the size and offset to 64 bits without increasing the
total size of the node.
|
2025-03-02 14:01:09 +01:00 |
|
Philip Rebohle
|
fe58b393d4
|
[util] Add helper for 2D and 3D morton codes
|
2025-03-02 14:01:09 +01:00 |
|
Philip Rebohle
|
c21f4e0190
|
[util] Add 48-bit integer type
|
2025-03-02 14:01:09 +01:00 |
|
Philip Rebohle
|
416f9c5a4a
|
[d3d11] Embed UpdateBuffer data in CS chunk
Tiny optimization that gets rid of a copy and also lets us use chunk
memory more efficiently.
|
2025-03-02 13:29:48 +01:00 |
|
Philip Rebohle
|
5b68884fd9
|
[d3d11] Fix confusing debug color for GPU synchronization
Purple was already used for various other things.
|
2025-03-01 16:37:39 +01:00 |
|
Robin Kertels
|
8f84085370
|
[dxvk] Add CS thread load to the HUD
|
2025-03-01 14:23:22 +01:00 |
|
Philip Rebohle
|
b35e69b467
|
[d3d11] Skip empty draws
Watch Dogs 2 likes to do this with graphics UAVs bound, which
introduces quite a bit of unnecessary GPU synchronization.
|
2025-03-01 11:30:38 +01:00 |
|
Philip Rebohle
|
11dc0e7ce8
|
[util] Enable compute UAV barriers for Watch Dogs 2
|
2025-02-28 23:57:26 +01:00 |
|
Philip Rebohle
|
6b5d595b3e
|
[d3d11] Add option to force compute shader UAV synchronization
|
2025-02-28 23:57:26 +01:00 |
|
Philip Rebohle
|
396a4e0235
|
[dxbc] Add option to implicitly synchronize UAV accesses
|
2025-02-28 23:57:26 +01:00 |
|
Philip Rebohle
|
c04410ca00
|
[dxbc] Increase maximum size of embedded icbs to 256 bytes
Gives drivers more info, while still avoiding duplication of large ICBs
when compiling the same shader into a large number of pipelines.
|
2025-02-28 12:07:49 +01:00 |
|
Philip Rebohle
|
fcbdff3b72
|
[d3d9] Set correct input topology for SWVP emulation
Otherwise we'll run the input patch pass, even if it doesn't do anything.
|
2025-02-26 20:58:47 +01:00 |
|
Philip Rebohle
|
0359a3521d
|
[dxvk] Patch geometry shader if input topology does not match pipeline
|
2025-02-26 20:53:43 +01:00 |
|
Philip Rebohle
|
b41253e18d
|
[dxvk] Add pass to patch GS input topology if necessary
|
2025-02-26 20:53:43 +01:00 |
|
Philip Rebohle
|
96a260e94e
|
[dxbc] Write back GS input topology to shader info
|
2025-02-26 20:53:43 +01:00 |
|
Philip Rebohle
|
2cb9d40f88
|
[dxvk] Add input topology field to shader info
|
2025-02-26 20:53:43 +01:00 |
|
Philip Rebohle
|
2cd305e606
|
[dxvk] Enable uniformBufferStandardLayout feature
Core and required in Vulkan 1.2, but keep it optional anyway for now
since we're not really doing anything critical with it.
|
2025-02-26 20:52:25 +01:00 |
|
Philip Rebohle
|
ec5bd04378
|
[dxvk] Remove uniform data from shaders
No longer used
|
2025-02-26 20:52:25 +01:00 |
|
Philip Rebohle
|
43f360d20c
|
[d3d11] Move immediate constant buffers to non-mapped VRAM
This has the advantage that icbs can be defragmented and don't
take up space in HVV chunks that would potentially go unused.
|
2025-02-26 20:52:25 +01:00 |
|
Philip Rebohle
|
6f59124e9a
|
[dxbc] Retrieve icb data directly from shader module
|
2025-02-26 20:52:25 +01:00 |
|
Philip Rebohle
|
d6d13edb7a
|
[dxbc] Use smallest aligned vector type for buffer-backed ubo
|
2025-02-26 20:52:25 +01:00 |
|
Philip Rebohle
|
31192b6d3f
|
[dxbc] Rework embedded immediate constant buffers
Considerably reduces the size of immediate constant buffer arrays when
not all vector components are used. Also adds bound-checking.
|
2025-02-26 20:52:25 +01:00 |
|
WinterSnowfall
|
31a4679960
|
[dxvk] Don't skip CPU devices when a device filter is set
|
2025-02-25 21:39:22 +01:00 |
|
Philip Rebohle
|
13bd234cea
|
[dxvk] Actually fix multidraw workaround
Otherwise we'll end up using the feature anyway.
|
2025-02-25 14:18:05 +01:00 |
|