2bdcc3dd2f
Ensure pixel history is consistent about unknown depth or colour data
2025-10-27 11:50:35 +00:00
9c001c3f59
Fix D3D12 pixel history not properly picking up discarded fragments
2025-10-27 11:50:35 +00:00
f459440d03
Fix issues with mismatching float/uint bindings in D3D11 pixel history
2025-10-27 11:50:35 +00:00
eb3136808d
Fix handling of 3D attachments in pixel history
2025-10-27 11:50:35 +00:00
f12b2573ad
Fix calculation of subresource for mip/slice D3D11 pixel history
2025-10-27 11:50:18 +00:00
bc1caaffee
Ensure all UAVs are unbound for D3D11 pixel history pixel copy
...
* This is necessary so the source texture can be bound as an SRV - we already
unbind all RTVs.
2025-10-27 11:50:18 +00:00
30a2b704e2
Correctly handle some fragments discarding when doing stencil counting
...
* If some fragments discard then our stencil counting goes off-by-one. Use
primitive IDs to render primitives in isolation and determine which ones
discarded so we can adjust stencil counting as needed.
2025-10-27 11:50:18 +00:00
74ecbd235a
Explicitly fetch pre/postmod values per-fragment with stencil counting
...
* This is more precise than propagating pre/post mod
2025-10-27 11:50:18 +00:00
4d96b492ec
Determine which events had some fragments discard in GL pixel history
2025-10-27 11:50:18 +00:00
c17557eb5f
Always use full precision format for shader output in GL pixel history
2025-10-27 11:50:18 +00:00
e17a10e329
Fetch clean values for per-pixel pre/post mod in GL pixel history
...
* We can't assume that premod == postmod in all cases. For colour history on a
colour target it will be, but depth values can change indepednently. Instead
we explicitly fetch pre/post mod before and after each event instead of just
postmod.
2025-10-27 11:50:18 +00:00
dceced1496
Add a clean FBO to copy from for target image
...
* The target image won't always be bound to the (or any) current framebuffer, if
e.g. it's written via image storage. Ensuring we still have a framebuffer to
copy from keeps most code the same.
2025-10-27 11:50:18 +00:00
15fb73a83d
Disable depth bounds test as needed when we want depth testing off
2025-10-27 11:50:18 +00:00
802a8aa845
Note whether GL pixel history target is depth or colour
2025-10-27 11:50:17 +00:00
0be1edd78b
Add separate pre/post mod reading in GL pixel history
...
* We will want to be able to batch these independently and this approach is
easier with fixed slots than having a slot per type in the same framebuffer.
2025-10-27 11:50:17 +00:00
426f9724c0
Mark unbound pixel shader in GL pixel history
2025-10-27 11:50:17 +00:00
a4ad0f0d2b
Ensure GL pixel history finds all possible modifying events
...
* All tests must be disabled as much as possible, including depth clipping and
depth bounds, as well as binding a fixed shader for potential shader discards.
2025-10-27 11:50:17 +00:00
724bce2d20
Don't ignore based on bound framebuffer for clear or direct writes
2025-10-27 11:50:17 +00:00
d768527918
Check for discarding at a per-pixel level for test failures on GL
2025-10-27 11:50:17 +00:00
d2a524ee5b
Check for depth clipping and depth bounds failures on GL pixel history
2025-10-27 11:50:17 +00:00
1c970a7868
Add global helper for naming D3D12 objects with normal strings
2025-10-27 11:50:17 +00:00
384b952625
Use uint/sint type for per-fragment postmod and shaderOut
...
* We also use a UAV clear for integer targets as needed
2025-10-27 11:50:17 +00:00
0cbaefe557
Match target subresource in use for D3D12 pixel history work
2025-10-27 11:50:17 +00:00
6f98e46da2
Add helper for quantised-checking of depth test failures on fragments
2025-10-27 11:50:16 +00:00
50976369db
Depth clears are considered to have depth in D3D12 pixel history
2025-10-27 11:50:16 +00:00
cc05571135
For vulkan pixel history on MSAA images, always rebind compute state
...
* The copying of a pixel happens via compute for MSAA targets so the compute
state becomes dirty in CopyPixelForPixelHistory - we rebind all state to
ensure it is properly re-set. This is not needed if non-MSAA.
2025-10-27 11:50:16 +00:00
7ef6d5d3e0
If there's no depth attachment, disable depth testing during history
...
* When we add a depth attachment we want to ensure this doesn't "activate" a
dormant depth test/bounds state which was inactive before due to the lack of
an attachment, so turn them off explicitly in our replacement pipelines.
2025-10-27 11:48:07 +00:00
cf42a3ea5c
The depth bounds test doesn't run if there is no depth target bound
2025-10-27 11:45:36 +00:00
f626625490
Query correct attachment for mip/slice in GL clears
2025-10-27 11:45:35 +00:00
1521278fd5
Only avoid duplicate failures when exceptions are raised
2025-10-27 11:45:35 +00:00
929548f478
Add dummy reads to pipes for linux for some reason
2025-10-27 11:45:35 +00:00
969c299448
Silence silly validation warning
2025-10-27 11:45:35 +00:00
fccf3c0408
DXIL Shader Debugger fixes for GPU batching of SampleGather operations
2025-10-27 10:11:17 +00:00
80307c944c
Fetch workgroup inputs for DXIL shader debugger
...
Used when debugging a subgroup with workgroup operations get the entire workgroup.
Get the subgroup (and workgroup) layout of threads from the GPU instead of assuming tightly wrapped.
Ensure the lane input data is organised to match the simulation expectations
2025-10-25 16:09:19 +01:00
a704828303
DXIL Debugger Load operations copy from backing memory
...
Don't use the cached value, fixes problems with GSM on a POD type (not array)
Atomic operations always load GSM from global backing memory
2025-10-25 15:57:00 +01:00
f9788d1289
DXIL Debugger ignore duplicate change if baseMemoryId == ptrId
2025-10-25 15:26:03 +01:00
17e97ca1b0
DXIL Debugger support for runtime Bitcast on a pointer source
2025-10-25 15:25:27 +01:00
5b47b73f4d
Change DXIL UAV out of bounds or null write to be a warning not error
2025-10-25 15:25:13 +01:00
cb83fac970
DXIL debugger support for GPU batching of Math and SampleGather ops
2025-10-25 15:12:39 +01:00
6bb8070fd2
Add a lock around Atomic memory operations in DXIL debugger
2025-10-25 15:00:07 +01:00
02f08bb051
DXOp::WaveIsFirstLane considers helper lanes as active
2025-10-25 14:46:38 +01:00
7b146ec97d
Convert DXIL Debugger to run simulation multi-threaded
...
Enabled by default : config option "D3D12_DXIL_Debug_EnableShaderDebugMT"
By default run jobs in a hot spin loop in a single JobQueue job
Each JobQueue Job runs SimulationJobHelper which is a while loop looking for threads to simulate
Developer option "D3D12_Hack_ShaderDebugUsesJobSystemJobs" to run simulation steps in individual job system jobs.
2025-10-25 14:33:12 +01:00
0e20fbf9fb
Added caching to DXIL APIWrapper and Debugger
...
Added TypedUAV Load/Store to APIWrapper
2025-10-25 14:17:40 +01:00
f72c1a3769
DXIL debugger support for queued GPU operations (run on Device Thread)
2025-10-25 14:04:19 +01:00
516dec8e17
DXIL debugger support for queue and run simulation on Device Thread
2025-10-25 13:49:28 +01:00
0243927d43
DXIL Debugger can simulate multiple instructions per step
2025-10-25 13:36:33 +01:00
fe8e63f7ec
Reworked DXIL debugger simulation in preparation for multithreading
...
Rearranged to isolate the simulation of a single lane into a helper function: InternalStepThread(uint32_t lane)
2025-10-25 13:24:43 +01:00
c4fa11b6e6
Assert for DXIL the subgroup is converged for subgroup ops
2025-10-25 13:12:20 +01:00
34a8af762b
Cache DXIL SSA ID Result names
...
The disassembly will populate the cache and the debugger will reuse the cache
2025-10-25 13:00:24 +01:00
99b3137b41
Add asserts for DXIL ThreadState APIs to be called on the device thread
2025-10-25 12:47:25 +01:00
baldurk