Philip Rebohle
26afaba410
[dxbc] Implement MSAD4 instruction
...
Apparently we're required to support this for Direct3D 11.2+.
There are currently no known games that require this instruction.
2019-09-16 12:42:14 +02:00
Philip Rebohle
72a356fe01
[d3d11] Implement GetResourceTiling
...
When called on a regular resource, native D3D11 will zero
out the return values. We should match this behaviour.
2019-09-16 12:42:14 +02:00
Philip Rebohle
7ba7178d14
[d3d11] Implement ID3D11Device2
...
Adds some extra validation to resource creation so that
apps cannot accidentally create tiled resources.
2019-09-16 12:42:14 +02:00
Philip Rebohle
c1a7243811
[d3d11] Implement ID3D11DeviceContext2
...
We don't support tiled resources nor markers for now.
Marker support could be added through debug layers.
2019-09-16 12:42:14 +02:00
Philip Rebohle
6e28ab956d
[d3d11] Implement D3D11.2 feature queries
2019-09-16 12:42:14 +02:00
Philip Rebohle
d7476f3fb8
[d3d11] Include D3D11.2 headers
2019-09-16 12:42:14 +02:00
Philip Rebohle
f2d461ab5b
[d3d11] Prevent log spam from repeated Get/SetResourceMinLOD calls
2019-09-16 12:41:44 +02:00
Philip Rebohle
3e17ccfaaa
[d3d11] Implement missing D3D11_FEATURE_D3D9_OPTIONS feature query
2019-09-16 02:21:46 +02:00
Philip Rebohle
c852b6de34
[d3d11] Optimize viewport updates if there is only one single viewport
2019-09-13 14:38:51 +02:00
Philip Rebohle
160b684d5a
[d3d11] Add range check for scissors and make failing it unlikely
2019-09-13 11:49:00 +02:00
Philip Rebohle
e2808309a3
[metæ] Release 1.3.4
2019-09-08 21:03:46 +02:00
Philip Rebohle
f57c021ab5
[util] Enable d3d11.allowMapFlagNoWait for Control
2019-09-08 10:35:57 +02:00
Philip Rebohle
b8888ffe6a
[d3d11] Don't create linearly tiled compressed images
...
Quantum Break uses this for texture uploads, which causes unnecessary
GPU synchronization and queue submissions when doing texture uploads.
2019-09-07 20:16:16 +02:00
orbea
c57e63f7d2
Work around build failure with winegcc/clang. ( #1184 )
...
Fixes #1182 .
2019-09-05 18:01:33 +02:00
Philip Rebohle
b0552751ad
[d3d11] Fix incorrect AddRef -> Release
...
Spotted by @jp7677
2019-08-29 21:24:47 +02:00
Philip Rebohle
2776ef43a3
[meta] Release 1.3.3
2019-08-29 20:56:47 +02:00
Philip Rebohle
30d19cd0f2
[d3d11] Rename some hazard tracking methods for clarity
2019-08-29 19:10:53 +02:00
Philip Rebohle
6be124e2cd
[d3d11] Check for conflicts withing RTV and UAV lists upon binding them
2019-08-29 19:10:53 +02:00
Philip Rebohle
e07ef1ec40
[d3d11] Resolve pipeline hazards when binding render targets
2019-08-29 19:10:53 +02:00
Philip Rebohle
a36f056572
[d3d11] Resolve pipeline hazards when binding compute shader UAVs
2019-08-29 19:10:53 +02:00
Philip Rebohle
c8c781c88b
[d3d11] Add check whether a viewed resource has a given bind flag
2019-08-29 19:10:50 +02:00
Philip Rebohle
afc8e4c29d
[d3d11] Filter redundant OMSetRenderTargets calls
...
Since rebinding render targets is a rather expensive
operation, we should avoid doing so whenever possible.
Affects Resident Evil 2 and Devil May Cry 5.
2019-08-26 23:59:08 +02:00
Philip Rebohle
4789790087
[d3d11] Remove outdated comment
...
We actually handle this properly now.
2019-08-26 23:43:47 +02:00
Philip Rebohle
5756e5c921
[d3d11] Check for shader resource view hazards
...
Fixes incorrect behaviour in games that try to use a currently bound
UAV or render target as a shader resource at the same time.
Fixes visual artifacts in Shining Resonance Refrain on AMD hardware.
2019-08-26 23:43:47 +02:00
Philip Rebohle
137589d755
[d3d11] Add state to track potentially hazardous bound SRVs
2019-08-26 23:29:01 +02:00
Philip Rebohle
4064dd3737
[d3d11] Add bound compute shader UAV mask
...
Will be used for efficient hazard tracking.
2019-08-26 23:29:01 +02:00
Philip Rebohle
8208cedfa9
[d3d11] Add common view info struct to all view types
...
Will be used for hazard detection.
2019-08-26 23:29:01 +02:00
Philip Rebohle
5ae5053a2a
[d3d11] Remove templated SetUnorderedAccessViews
...
Not needed because the CS and OM paths are separate anyway.
2019-08-26 23:29:01 +02:00
Philip Rebohle
08e3500beb
[d3d11] Don't use .at()
2019-08-26 23:29:01 +02:00
Philip Rebohle
473025a93b
[dxvk] Add function to find set bits in binding mask
2019-08-26 23:29:01 +02:00
Philip Rebohle
d2d19b0dec
[dxvk] Add function to set bind mask bit to a given value
2019-08-26 23:29:00 +02:00
Jacek Caban
1981140257
[dxvk] Explicitly include unordered_map in dxvk_renderpass.h
...
When using libstdc++, it's somehow implicitly included, but it's not the case for libc++.
2019-08-26 20:07:58 +02:00
Joshua Ashton
8e54477c2a
[util] Use R string literals for app compat regexes
2019-08-24 21:32:51 +02:00
Philip Rebohle
54ca5900e1
[util] Use regular expressions on full exe path to match app profiles
...
This allows us to detect applications with non-unique executable names
or with variable executable names more reliably, and also allows us to
merge profiles for games that have multiple known exe names.
The matching is also no longer case-sensitive.
2019-08-18 17:49:22 +02:00
Philip Rebohle
c934333a5c
[util] Add method to retrieve full exe path
2019-08-17 11:50:39 +02:00
Philip Rebohle
d38607c9be
[dxvk] Implement depth-stencil resolve
2019-08-13 15:16:09 +02:00
Philip Rebohle
e54dfab471
[dxvk] Support depth-stencil resolve using VK_KHR_depth_stencil_resolve
2019-08-13 15:16:09 +02:00
Philip Rebohle
7e95493fba
[dxvk] Support shader-based depth-stencil resolve
...
Requires VK_EXT_shader_stencil_export for stencil resolves.
2019-08-13 15:16:00 +02:00
Philip Rebohle
a516ca5b85
[dxvk] Add meta shaders for depth-stencil resolve
2019-08-13 14:37:50 +02:00
Philip Rebohle
acf0001fd1
[dxvk] Make device properties accessible from the device
2019-08-13 14:37:50 +02:00
Philip Rebohle
2f15cb7602
[dxvk] Enable VK_KHR_depth_stencil_resolve if available
2019-08-13 12:53:31 +02:00
Philip Rebohle
d127c08487
[dxvk] Enable VK_KHR_create_renderpass2 if available
2019-08-13 12:52:54 +02:00
Philip Rebohle
1999ca9dd4
[util] Set d3d11.dcSingleUseMode=False for SteamVR perf test
...
*Actually* fixes #1160 .
2019-08-10 11:21:47 +02:00
Philip Rebohle
4cce07ccd9
[d3d11] Update resources mapped on deferred contexts in Map()
...
Apparently it is legal to use mapped buffers on deferred contexts
in D3D11, so we have to execute our update code immediately.
Fixes #1160 .
2019-08-10 11:11:44 +02:00
Joshua Ashton
d579f07238
[util] Fix compiling with clang/clang-cl
2019-08-07 06:08:54 +02:00
Philip Rebohle
590834660e
Revert "[dxvk] Use vkCmdUpdateBuffer to clear tiny buffers"
...
This reverts commit 311661e404Fixes #1155 .
2019-08-07 05:41:33 +02:00
Philip Rebohle
bd40b05720
[d3d11] Recreate swap chain immediately after synchronous present fails
...
Fully restores pre-1.3 behaviour in case async present is disabled.
Apparently this is necessary to avoid hangs on Nvidia for some reason.
2019-08-06 09:11:04 +02:00
Philip Rebohle
9fdd7bf789
[meta] Release 1.3.2
2019-08-05 20:38:22 +02:00
Philip Rebohle
bc0537df37
[d3d11] Remove d3d11.asyncPresent option
2019-08-04 21:49:55 +02:00
Philip Rebohle
a558f82b5f
[dxvk] Implement asynchronous presentation option in the backend
2019-08-04 21:49:55 +02:00
Philip Rebohle
77fde83479
[dxvk] Move some code to where it makes more sense
2019-08-04 21:37:26 +02:00
Philip Rebohle
6b2f4f93cc
[d3d11] Fix incorrect tracking of mapped image subresources
...
When returnig DXGI_ERROR_WAS_STILL_DRAWING, we should not mark
the subresource as mapped, and instead ignore subsequent calls
to Unmap.
2019-08-03 19:11:19 +02:00
Philip Rebohle
140a2c1017
Revert "[d3d11] Return error when mapping an already mapped image"
...
This reverts commit 1cc531eaf8Fixes #1148 , #1149 .
2019-08-03 19:08:22 +02:00
Philip Rebohle
02d92210ad
[dxvk] Avoid redundant descriptor set updates when binding buffers (v2)
...
We need to check not just the buffer object but also the length of the
bound buffer range, since this information will be written into the
descriptor and cannot be changed via dynamic offsets.
Fixes: f501ebc
2019-08-01 21:30:27 +02:00
Philip Rebohle
af15d85baa
[dxvk] Add config option to enable the HUD
2019-08-01 12:37:55 +02:00
Philip Rebohle
4fd41f8550
[dxvk] Don't allow common objects to be reference-counted
2019-07-30 20:06:59 +02:00
Philip Rebohle
65c1e58bd9
[dxvk] Use DxvkObjects to create and pass around common objects
2019-07-30 20:00:38 +02:00
Philip Rebohle
35679d2ba7
[dxvk] Add class that bundles objects shared between device and context
...
Uses lazy initialization for various meta objects.
2019-07-30 20:00:38 +02:00
Philip Rebohle
cfd7033e40
[dxvk] Pass DxvkDevice to DxvkRenderPassPool constructor
2019-07-30 20:00:38 +02:00
Philip Rebohle
e12103121f
[dxvk] Pass DxvkDevice to DxvkEventPool constructor
2019-07-30 20:00:38 +02:00
Philip Rebohle
40028a5b8c
[dxvk] Pass DxvkDevice to DxvkMetaClearObjects constructor
2019-07-30 20:00:38 +02:00
Philip Rebohle
6a58c432ec
[dxvk] Pass DxvkDevice to DxvkMetaPackObjects constructor
2019-07-30 20:00:38 +02:00
Philip Rebohle
9a2da555c0
[util] Add helper for lazy initialization
...
a
2019-07-30 20:00:35 +02:00
Philip Rebohle
23379b6b9c
[dxvk] Check render pass before pipeline state during pipeline lookup
...
Checking one pointer is a bit cheaper than comparing a 1600 byte struct.
2019-07-30 16:19:16 +02:00
Philip Rebohle
a7666aad82
[dxvk] Refactor the way render passes to pipeline compiler methods
2019-07-30 13:17:56 +02:00
Philip Rebohle
13bc3df92f
[dxvk] Refactor render pass objects to not use reference counting
...
Like pipeline objects, we keep these around anyway so there's no
reason to add ref count overhead. Also use a hash map to perform
the lookup.
2019-07-30 13:13:02 +02:00
Philip Rebohle
d01b6baf38
[dxvk] Introduce compilePipeline method to Dxvk*Pipeline classes
...
Decouples the act of synchronously retrieving a handle
from asynchronously compiling the pipeline.
2019-07-30 12:14:52 +02:00
Philip Rebohle
3dc33c64a9
[dxvk] Introduce DxvkComputePipelineInstance
...
Same as the graphics pipeline equivalent.
2019-07-30 11:07:07 +02:00
Philip Rebohle
20b0cbdfb6
[dxvk] Rename compilePipeline -> createPipeline
...
More in line with Vulkan naming.
2019-07-30 11:06:25 +02:00
Philip Rebohle
6ab074c95b
[dxbc] Only use atomic append/consume optimization in compute shaders
...
- For fragment shaders, this isn't safe since ballots include helper invocations
- For vertex shaders, if drivers don't support subgroup operations in those
stages, we don't want it to affect the performance of compute shadres.
2019-07-26 14:07:39 +02:00
Philip Rebohle
dfe2922136
[dxvk] Fix partial clears for mismatched framebuffer attachment sizes
...
Fixes an issue in Borderlands: The Pre-Sequel, which binds a 512x512
color attachment and a full-screen depth buffer at the same time and
then attempts to clear the depth buffer.
2019-07-24 23:46:02 +02:00
Philip Rebohle
9c5102e257
[dxvk] Don't duplicate geometry shader system value outputs
...
Fixes #1121 . The basic idea here is that all built-ins that can be
written by the GS will be consumed as built-ins by the FS anyway,
so we do not need to keep the o# variable around.
2019-07-24 18:18:20 +02:00
Philip Rebohle
eaa41eb76c
[dxvk] Don't use reference counting for pipeline objects
...
Again not necessary since these objects are persistent.
Eliminates refcount overhead of pipeline lookups entirely.
2019-07-23 13:15:06 +02:00
Philip Rebohle
8cd13cc5bd
[dxvk] Use shader key structs directly for pipeline lookups
...
Removes some overhead and unnecessary ref count changes on shaders.
2019-07-23 13:00:04 +02:00
Philip Rebohle
8d4996bcda
[dxvk] Use shader key structs to store shaders in DxvkPipeline objects
...
Mostly a code cleanup to make constructing these objects a bit easier.
2019-07-23 12:48:11 +02:00
Philip Rebohle
604e89b97a
[dxvk] Rename Dxvk*PipelineKey -> Dxvk*PipelineShaders
2019-07-23 12:41:09 +02:00
Philip Rebohle
70294aac44
[dxvk] Disable resource tracking for DxvkPipeline objects
...
This isn't necessary at all since these objects are persistent.
2019-07-23 12:34:48 +02:00
Philip Rebohle
00cf2a20a3
[d3d11] Don't allow the creation of buffers with a size of zero
2019-07-21 20:47:42 +02:00
Philip Rebohle
f5cec978c8
[meta] Release 1.3.1
2019-07-20 20:33:19 +02:00
Philip Rebohle
c89bec5abd
[dxvk] Don't log submission errors on presentation
...
Makes no sense and only leads to confusion.
2019-07-20 20:25:18 +02:00
Philip Rebohle
3f4c9a3bb5
[hud] Add GPU load monitor
2019-07-18 23:23:36 +02:00
Philip Rebohle
5bb20cceb6
[dxvk] Add GPU idle time to stat counters
2019-07-18 23:23:12 +02:00
Philip Rebohle
3d86ecd94d
[dxvk] Estimate GPU idle time based on cleanup thread activity
...
We'll assume that GPU idle time == time spent waiting for new
command lists to be added to the queue of the cleanup thread.
This isn't entirely accurate, especially if CPU load is very
high, but should be good enough.
2019-07-18 22:54:11 +02:00
Philip Rebohle
02d917c680
[dxvk] Fix meta copy operation for mipmapped images
2019-07-18 19:52:24 +02:00
Philip Rebohle
fb9ea958a1
[dxvk] Fix meta copy operation for 1D images
2019-07-18 19:50:57 +02:00
Philip Rebohle
2905ba82d2
[dxvk] Remove old meta vertex and geometry shaders
2019-07-18 18:59:11 +02:00
Philip Rebohle
8889d6402e
[dxvk] Use new fullscreen shaders for mip gen operations
2019-07-18 18:57:18 +02:00
Philip Rebohle
e91efb6dc2
[dxvk] Use new fullscreen shaders for meta resolve operations
2019-07-18 18:57:17 +02:00
Philip Rebohle
792f15680a
[dxvk] Use new fullscreen shaders for meta copy operations
2019-07-18 18:57:17 +02:00
Philip Rebohle
07408bcdcc
[dxvk] Add new vertex and geometry shaders for fullscreen passes
...
These new shaders are aimed to be used by all meta operations
and will work without geometry shaders on supported hardware.
2019-07-18 18:57:15 +02:00
Philip Rebohle
92d6f26130
[d3d11] Always enable depth-stencil attachment usage for depth-stencil images
...
Allows us to use framebuffer copies even if the game itself does not
intend to render to the image.
Improves performance in Final Fantasy XIV on RADV.
2019-07-18 17:26:02 +02:00
Philip Rebohle
8a9cee903b
[dxvk] Use render pass copy for depth-stencil images if beneficial
2019-07-18 17:26:02 +02:00
Philip Rebohle
a08f9d0897
[dxvk] Implement device- and driver-specific performance hints
...
These are meant to be read by the DxvkContext in order to choose
a fast path for certasin operations.
2019-07-18 17:25:56 +02:00
Philip Rebohle
e611dff45e
[dxvk] Support depth-stencil meta-copy operations
2019-07-18 17:25:56 +02:00
Philip Rebohle
0dd8cba199
[dxvk] Support depth-stencil formats for meta copy objects
2019-07-18 17:25:50 +02:00
Philip Rebohle
89516e2da2
[dxvk] Add meta copy shaders for depth-stencil formats
2019-07-18 17:25:48 +02:00
Philip Rebohle
677e33b9c9
[dxvk] Enable VK_EXT_shader_stencil_export if available
2019-07-18 17:25:48 +02:00
Philip Rebohle
f0fb25c082
[dxvk] Log more device info on device creation
...
This should tell us whether extension features are actually used.
2019-07-18 13:37:14 +02:00
Philip Rebohle
f3943934a7
[util] Improve multi-line logging
2019-07-18 13:34:52 +02:00
Philip Rebohle
f16ba4794b
[d3d11] Use unlikely() for some query code
...
This is a somewhat hot path in some games, so why not.
2019-07-17 20:35:00 +02:00
Philip Rebohle
c4b56b9d8d
[d3d11] Use private temporary references for queries
...
Like the previous commit, just with queries.
2019-07-17 20:16:19 +02:00
Philip Rebohle
7225674088
[d3d11] Use private temporary references for state objects
...
We really shouldn't be altering the application-visible ref
count when sending these objects to the CS thread.
2019-07-17 20:01:57 +02:00
Philip Rebohle
21a2ce045f
[util] Add convenience method to return public/private references
2019-07-17 19:59:50 +02:00
Philip Rebohle
af8e1a3d47
[d3d11] Simplify ApplyPrimitiveTopology code
2019-07-17 15:54:59 +02:00
Philip Rebohle
3f30fbd098
[d3d11] Simplify BindIndexBuffer code
2019-07-17 14:41:00 +02:00
Philip Rebohle
63fe899bdc
[dxvk] Don't check if bindings have changed in the backend
...
The state tracker should perform these checks before sending commands
off to the backend anyway, so checking again in the backend is redundant.
2019-07-17 14:26:55 +02:00
Philip Rebohle
b6c395c013
[dxvk] Don't track command count in CS chunks
...
We weren't using this at all, and it's not necessary
to check whether the chunk is empty either.
2019-07-17 12:52:25 +02:00
Philip Rebohle
11b7fc8914
[d3d11] Catch invalid ClearUnorderedAccessViewFloat calls
...
This method cannot be called on integer UAVs.
2019-07-17 11:47:42 +02:00
Philip Rebohle
b20ceec727
[d3d11] Handle integer formats in ClearRenderTargetView correctly
...
We're supposed to apply the same color conversion as in ClearView.
2019-07-17 11:47:42 +02:00
Philip Rebohle
c6ea115ca3
[dxvk] Add method to retrieve buffer view format info
2019-07-17 11:47:42 +02:00
Philip Rebohle
c29314de5a
[dxvk] Change DxvkImageView::formatInfo to return view format info
...
Makes more sense and won't break any existing code using it.
2019-07-17 11:33:39 +02:00
Philip Rebohle
7895272806
[dxvk] Use correct command buffer for buffer updates
...
Fixes a regression in SpellForce 3.
2019-07-16 23:47:33 +02:00
Philip Rebohle
493b55b073
[dxvk] Compute memory chunk size per memory type rather than per heap
2019-07-16 09:59:44 +02:00
Philip Rebohle
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
Philip Rebohle
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
Philip Rebohle
ad8fdcac07
[dxvk] Don't mark queries as stalling if DONOTFLUSH is set
2019-07-16 01:19:02 +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
d8e31f221f
[dxvk] Fix offset in DxvkBuffer::subSlice
2019-07-15 04:06:12 +02:00
Philip Rebohle
f88658d88c
[dxvk] Template BindPoint parameter in some DxvkContext methods
...
Again, there's no reason not to do this since the argument is always
constant, and we can skip some work in the compute case.
2019-07-14 20:09:35 +02:00
Philip Rebohle
def05f007c
[dxvk] Template Index parameter of commitGraphicsState
...
This is always constant anyway, so avoid some unnecessary code.
2019-07-14 19:57:50 +02:00
Philip Rebohle
8dfdda7a39
[dxvk] Get rid of validateCompute/GraphicsState
...
After all this time we haven't found a single situation where
we need more validation, so scrap it. Checking whether there
is an active render pass was redundant anyway.
2019-07-14 19:34:56 +02:00
Philip Rebohle
a93dd74f71
[dxvk] Don't use derivative pipelines
...
No driver appears to be taking advantage of this, so why bother.
2019-07-14 13:58:00 +02:00
Philip Rebohle
03c6df56c1
[d3d11] Initialize subresourceLayers in the inner loop
...
Not doing so causes Dirt Rally to fail uploading some textures
properly for some extremely weird reason.
2019-07-13 23:39:58 +02:00
Philip Rebohle
7cb385facd
[meta] Release 1.3
2019-07-13 19:15:30 +02:00
Philip Rebohle
e116ff49e6
[dxvk] Rename memoryAvailable -> memoryBudget
2019-07-13 19:04:33 +02:00
Joshua Ashton
fac3ae9f83
[spirv] Implement opCross
2019-07-12 23:32:09 +02:00
Joshua Ashton
f280386aec
[spirv] Implement opFMix
2019-07-12 23:32:09 +02:00
Philip Rebohle
8644d75722
[dxbc] Use OpDemoteToHelperInvocationEXT for discards if supported
2019-07-11 19:33:17 +02:00
Philip Rebohle
e901e1269d
[spirv] Add support for OpDemoteToHelperInvocationEXT
2019-07-11 19:33:17 +02:00
Philip Rebohle
7b34b8515b
[d3d11] Enable shaderDemoteToHelperInvocation feature
2019-07-11 19:33:17 +02:00
Philip Rebohle
b2a53a2413
[dxvk] Enable VK_EXT_shader_demote_to_helper_invocation if available
2019-07-11 19:33:17 +02:00
Liam Middlebrook
2e83aae472
[dxbc] Print shader signatures at debug loglevel
2019-07-11 12:40:44 +02:00
Liam Middlebrook
ff4fa29304
[dxbc] Add string output function for DxbcRegMask
2019-07-11 12:40:44 +02:00
Liam Middlebrook
2164a44887
[dxbc] Add ostream operator for DxbcScalarType
2019-07-11 12:40:44 +02:00
Philip Rebohle
06d4e06d8a
[dxbc] Enable early discard on RADV/ACO
...
Still can't enable on LLVM due to GPU hangs. The way we detect it
may have to change in the future, but for now, this should do.
2019-07-08 16:30:56 +02:00
Robin Kertels
47f7333c18
[d3d11] Fix RSGetViewports and RSGetScissorRects behaviour
...
Fixes #1116 .
2019-07-08 13:14:42 +02:00
Philip Rebohle
24e1969dc4
[dxvk] Remove obsolete DxvkEvent class
2019-07-08 00:16:03 +02:00
Philip Rebohle
15072afa1f
[d3d11] Use new signals as frame synchronization events
2019-07-08 00:16:03 +02:00
Philip Rebohle
f0acc40e50
[dxvk] Add support for new signals to the backend
2019-07-08 00:16:03 +02:00
Philip Rebohle
4e8122eda9
[dxvk] Add new signal class
2019-07-08 00:16:00 +02:00
Philip Rebohle
c631953ab6
[d3d11] Don't immediately synchronize after present
...
This will actually enable asynchronous presentation.
Improves performance in Quake Champions.
2019-07-05 21:20:09 +02:00
Philip Rebohle
77db8158c8
[dxvk] Add option to toggle asynchronous presentation
2019-07-05 21:20:09 +02:00
Philip Rebohle
ed5c43a14d
[dxvk] Implement asynchronous presentation
...
Off-loads the vkQueuePresentKHR call to the queue submission thread
to avoid synchronization with that thread on a present call.
2019-07-05 15:11:59 +02:00
Philip Rebohle
0900f5d1bc
[dxvk] Fix typo
2019-07-05 15:09:15 +02:00
Philip Rebohle
1cc531eaf8
[d3d11] Return error when mapping an already mapped image
...
Final Fantasy XIV does this. Should avoid some unnecessary work.
2019-07-05 10:25:06 +02:00
Philip Rebohle
2f64f5b4e7
[dxvk] Check whether CS thread is busy before synchronizing with it
...
Reduces unnecessary locking overhead, which may be relevant if this
function gets called frequently by GetData or WaitForResource.
2019-07-04 21:37:17 +02:00
Philip Rebohle
f6dbf5bbf0
[dxvk] Increase staging buffer size to 32 MiB
...
Since the chunk size was also doubled to 128 MiB and we want
staging buffers to get their own allocations.
2019-07-04 20:39:27 +02:00
Philip Rebohle
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
Philip Rebohle
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
Philip Rebohle
ac9610f377
[dxvk] Add VK_EXT_memory_budget as a passive extension
2019-07-03 11:02:13 +02:00
Philip Rebohle
f21c02caef
[dxvk] Add support for passive device extensions
...
These extensions only affect physical device functionality and
do not have to be enabled during device creation.
2019-07-03 10:59:48 +02:00
Philip Rebohle
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
Joshua Ashton
1dadba3cce
[dxvk] Give null values to empty DxvkGraphicsPipelineInstances
...
Silences an MSVC warning.
2019-07-01 02:46:54 +02:00
Joshua Ashton
c381e8a29e
[util] Add correct type suffixes to Sha1Hash dword function to silence compiler warning
2019-07-01 02:46:54 +02:00
Philip Rebohle
67122d9246
[d3d11] Enable initial image uploads over SDMA
2019-06-29 01:53:43 +02:00
Philip Rebohle
9d902418c0
[d3d11] Enable initial buffer uploads over SDMA
2019-06-29 01:53:43 +02:00
Philip Rebohle
d5c0a8c842
[dxvk] Upload HUD font texture via SDMA
2019-06-29 01:53:43 +02:00
Philip Rebohle
c7c90830c4
[dxvk] Implement image uploads on SDMA
2019-06-29 01:53:32 +02:00
Philip Rebohle
eb71c62b33
[dxvk] Implement buffer uploads on SDMA
2019-06-29 01:53:32 +02:00
Philip Rebohle
0a40e2c868
[dxvk] Support queue ownership transfer barriers
2019-06-29 01:53:32 +02:00
Philip Rebohle
f85b9088d4
[dxvk] Explicitly specify the command buffer for copy commands
2019-06-29 01:53:32 +02:00
Philip Rebohle
be24ca6099
[dxvk] Rename and add SDMA barrier sets
2019-06-29 01:53:32 +02:00
Philip Rebohle
4f3dcf2bc8
[dxvk] Add SDMA command buffer
...
This new command buffer will be submitted to the transfer queue,
if available, otherwise it will be the first buffer submitted
to the graphics queue.
2019-06-29 01:53:32 +02:00
Philip Rebohle
4e0de6bc20
[dxvk] Refactor queue submission
2019-06-29 01:53:32 +02:00
Philip Rebohle
545cd52020
[dxvk] Create transfer queue command pool if available
2019-06-29 01:53:32 +02:00
Philip Rebohle
d8163c4446
[dxvk] Don't pass queue handles to DxvkCommandList
...
Instead, pull them from the device as needed. This coupling would
only make sense if we required one command list per queue family.
2019-06-29 01:53:32 +02:00
Philip Rebohle
d2d11bf995
[dxvk] Add option to enable or disable the transfer queue
2019-06-29 01:53:32 +02:00
Philip Rebohle
191bba660b
[dxvk] Provide way to access queue properties from device
2019-06-29 01:53:32 +02:00
Philip Rebohle
1c39765b86
[dxvk] Rework queue family selection
...
This allows us to support multiple queues more easily.
2019-06-29 01:53:32 +02:00
Philip Rebohle
110fc8f833
[dxvk] Properly clear used command buffer flags
...
Fixes an issue where we would always submit the init buffer, even
if it is empty.
2019-06-29 01:53:29 +02:00
Philip Rebohle
770ec2c4db
[dxvk] Get rid of array in present vertex shader
2019-06-29 01:34:16 +02:00
Philip Rebohle
42e61020e4
[d3d11] Remove unused SetRenderTargets method
2019-06-27 15:54:42 +02:00
Philip Rebohle
2148619f3c
[d3d11] Spill render pass when restoring context state as needed
2019-06-27 15:54:42 +02:00
Philip Rebohle
a704e6d27e
[d3d11] Fix UAV binding in OMSetRenderTargets{,AndUnorderedAccessViews}
...
Unlike for compute shaders, we're supposed to replace all UAV bindings
when binding render targets. We also should spill the render pass when
disabling UAV rendering to avoid read-after-write hazards.
Fixes a potential synchronization bug encountered in Devil May Cry 5.
2019-06-27 15:54:42 +02:00
Joshua Ashton
194db57a4d
[spirv] Expose MatrixStride decoration for members
2019-06-27 10:49:49 +02:00
Joshua Ashton
80808d743b
[spirv] Add generic memberDecorate
2019-06-27 10:49:49 +02:00
Joshua Ashton
1b8c54eb84
[spirv] Expose opMatrixTimesMatrix, opMatrixTimesVector and opVectorTimesMatrix
2019-06-27 10:49:49 +02:00
Philip Rebohle
afe2b487a6
[dxvk] Avoid redundant vertex and index buffer tracking
...
Same optimization as for regular resources. Mostly reduces load
on the cleanup thread.
2019-06-27 01:51:15 +02:00
Philip Rebohle
c3ebf4658c
[dxvk] Reorder command list reset operations by importance
2019-06-27 00:02:43 +02:00
Philip Rebohle
7491c06389
[meta] Release 1.2.3
2019-06-26 17:33:57 +02:00
Philip Rebohle
5728d10587
[d3d11] Copy initial texture data to mapped buffer, if available
...
Otherwise, if an application maps the image right after creating it,
we might end up reading garbage data or overriding the image data.
2019-06-25 14:22:47 +02:00
Philip Rebohle
4d4db6c683
[dxvk] Use void pointers for packImageData
...
Makes this function a bit less annoying to use.
2019-06-25 14:17:03 +02:00
Philip Rebohle
3293bd21d1
[dxvk] Avoid updating binding mask on hot path in updateShaderResources
...
Assuming that everything is bound by default allows us to save a few
CPU cycles per descriptor to update as long as that assumption holds
true, while adding only a small constant cost for the update operation.
2019-06-23 23:43:25 +02:00
Philip Rebohle
b7769759f2
[dxvk] Implement comparison and setting multiple bits for DxvkBindingSet
2019-06-23 23:41:56 +02:00
Philip Rebohle
5e3336d79b
[dxvk] Clean up updateShaderResources a bit
2019-06-23 23:21:47 +02:00
Philip Rebohle
81e7a8d1f6
[dxvk] Avoid redundant resource tracking
...
Only tracks a shader resource if the resource bound to the slot
has actually changed, or if we began recording a new command
buffer. Reduces the number of atomic operations in some games.
2019-06-23 18:55:49 +02:00
Philip Rebohle
924ca9eaec
[dxvk] Add more generic version of DxvkBindingMask
2019-06-23 18:45:34 +02:00
Philip Rebohle
f58242c58d
[dxvk] Fix poor API design around defining resource slots
2019-06-23 15:49:29 +02:00
Andrew Eikum
326eb0bf80
[dxgi] Don't fail leaving fullscreen if window is already destroyed
...
Unreal Engine 4 games destroy the window, then call
SetFullscreenMode(FALSE). If this call fails, they pop up an error
dialog. Wine tests show that this call should succeed.
2019-06-21 17:27:34 +02:00
Philip Rebohle
62372f5d47
[dxvk] Release staging buffer memory for HUD context
...
We really don't want to waste 32MB of system RAM because of
one single texture update all the time.
2019-06-21 16:43:22 +02:00
Philip Rebohle
34cdba1df5
[dxvk] Add method to explicitly release staging buffer memory
2019-06-21 16:43:22 +02:00
Philip Rebohle
1c9bc235d0
[dxvk] Remove old staging buffer code
2019-06-21 16:43:22 +02:00
Philip Rebohle
8c1a56af89
[dxvk] Remove old staging buffer allocator from DxvkCommandList
2019-06-21 16:43:22 +02:00
Philip Rebohle
24c282378e
[dxvk] Use new staging buffer allocator in DxvkContext
2019-06-21 16:43:22 +02:00
Philip Rebohle
970deb452e
[dxvk] Implement new staging buffer allocator
2019-06-21 16:43:19 +02:00
Philip Rebohle
a41bd8c4a0
[d3d11] Further optimize constant buffer binding
...
Saves a few CPU cycles on the more common SetConstantBuffers method,
compared to SetConstantBuffers1.
2019-06-20 22:49:19 +02:00
Philip Rebohle
3141467c37
[d3d11] Template all functions that take ShaderStage as an argument
...
Basically what D9VK does. Doesn't seem to affect performance, but why not.
2019-06-20 22:09:02 +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
Joshua Ashton
a4a4f5a822
[util] Correct return value of SetThreadDescription
2019-06-19 19:43:36 +02:00
Philip Rebohle
0a35ae6fca
[dxvk] Log memory utilization per heap on memory allocation failures
2019-06-19 14:47:09 +02:00
