052961ea08
Added DXBCContainer::GetGlobalShaderFlags()
2024-12-11 14:36:09 +00:00
daa0392170
Extend D3D12 GetShaderBlob() to allow passing arbitrary compile options
...
Pass via ShaderCompileFlag with name "@compile_option" and value is the compile option to add
Added helper to return D3D12 Shader Cache m_CompileFlags
2024-12-11 14:34:40 +00:00
cfa392e279
DXIL Debugger handle Half types in GatherPSInputDataForInitialValues()
2024-12-11 14:25:50 +00:00
8679f8a9ba
DXIL Debugger handle Half types in GetInterpolationModeForInputParam()
2024-12-11 14:25:10 +00:00
b6dca8fb47
DXIL Debugger do not flush denorm for Half types
2024-12-11 14:24:39 +00:00
6fddc34077
DXIL Debugger support for DXOp::IMul, DXOp::UMul, DXOp::UDiv
...
Prevent divide by zero for integer divides. Set GeneratedNanOrInf for divide by zero.
2024-12-11 06:55:59 +00:00
a40d729bbe
Use mad() intrinsic in D3D12_Shader_Debug_Zoo SM6+ tests
2024-12-10 17:36:30 +00:00
ae7159fc63
DXIL Debugger support for DXOp::IMad, DXOp::UMad
2024-12-10 17:36:26 +00:00
3ada6022e1
DXIL Debugger support for Operation::CompareExchange
2024-12-10 16:39:35 +00:00
4f7dd54d9e
Fix file encoding error in nvapi header
2024-12-10 15:16:19 +00:00
eab775181a
Always pass CAPTURE_REPLAY RT pipeline create flags
...
This is required for RT pipelines when capture/replay handles are used,
both during capture and replay.
2024-12-10 15:16:08 +00:00
6a761c2196
Enable RT pipeline capture/replay feature alongside RT pipelines
...
Required for using capture/replay shader handles.
2024-12-10 15:16:08 +00:00
06047f10dd
DXIL Debugger support for DXOp::AtomicCompareExchange
2024-12-10 13:50:40 +00:00
8138a632f2
DXIL Debugger support for DXOp::AtomicBinOp
2024-12-10 13:49:02 +00:00
8de06cf324
D3D12 Shader Debug Zoo : add a test for UAV Interlocked*
...
RWTexture<int>
InterlockedAdd
InterlockedAnd
InterlockedOr
InterlockedXor
InterlockedMin
InterlockedMax
InterlockedCompareExchange
InterlockedCompareStore
InterlockedExchange
2024-12-10 13:46:47 +00:00
8f75a1658f
DXIL Debugger treat global variables as always in scope
2024-12-10 11:38:26 +00:00
9e0e25a576
DXIL Debugger save previous block, Phi Variables for Operation::Switch
2024-12-10 06:21:35 +00:00
24faca1ae9
DXIL Debugger ignore global variable zeroinitializer
...
The backing memory is initialised to zero on creation
2024-12-10 06:21:35 +00:00
9d32418e33
Fix GL_Buffer_Updates to not rely on exact thumbnail
2024-12-10 00:06:39 +00:00
8cabf9eaf0
Use dedicated lists and allocators for overlay to reset them
2024-12-09 16:19:20 +00:00
aa30ac2c10
Don't overlap storage with multiple inline UBO updates in one template
...
* Previously we were writing the inline UBO data to the same byte storage
meaning the last update's data would write over all updates.
2024-12-09 16:19:20 +00:00
0e406b0e4d
Fix some issues with descriptor display with inline UBOs
2024-12-09 16:19:20 +00:00
c4caa7395c
Keep GPUBuffer fully unwrapped to avoid problems during capture
2024-12-09 16:19:20 +00:00
ca5c0e8a3e
DXIL ControlFlow optimisations
...
Lazy evaluate the connections map
2024-12-09 15:31:11 +00:00
89c5724477
DXIL Debugger support for TextureStore
2024-12-09 15:31:11 +00:00
23c7e1cfea
D3D12 Shader Debug Zoo : add a test for Texture2D Store
2024-12-09 14:35:49 +00:00
0fcecf0a15
D3D12_Shader_Debug_Zoo only do 16bit tests if runtime supports it
...
Check using Native16BitShaderOpsSupported
2024-12-09 12:50:43 +00:00
acc29c9291
Use unwrapped memory map/unmap functions for GPUBuffer
...
* If we use the wrapped functions this can deadlock on the CapTransitionLock
when using GPUBuffers during capture.
2024-12-06 16:25:08 +00:00
bc5bd00205
Cap oversized soft-memory allocations at 512MB
...
* This allows a larger soft-memory limit without allocating too much memory in
one block, but still allows low soft-memory limits to cap the size of memory
blocks.
2024-12-06 16:25:08 +00:00
9703cf23e3
Don't use BeginQuery/EndQuery to initialise AS queries
2024-12-06 15:59:22 +00:00
1e977fe889
Use configurable overestimate to account for incorrect maxVertex
...
* We do this already on D3D12, this applies the same change to Vulkan -
maxVertex can't really be trusted but we hope it is approximately right or the
VBs are small, we don't want to copy the whole VB since it may be massive.
2024-12-06 15:59:22 +00:00
ab6381319f
Skip options5 entirely
...
* No longer modifying this one.
2024-12-06 13:38:49 +00:00
8de3242825
Enable all features in options5
2024-12-06 13:38:48 +00:00
c85835ce4b
Enable DXIL Debugging
...
Remove config option D3D12_Experimental_EnableDXILShaderDebugging
2024-12-06 11:13:29 +00:00
f7cc72c9da
DXIL Debugger change SSA lifetime to use block and instruction index
...
Use control flow to determine the ordering of ExectionPoint's (block + instruction index)
Can't use pure instruction index because programs do not have to have linear flow and monotonically increasing instruction index
Can assign to an SSA ID at an instruction index larger than when the SSA ID is accessed, as long as the assignment occurs before the access in control flow (block) ordering
SSA ID assignments inside loop blocks are considered to be live until the next uniform block after the loop block the SSA ID assignment occurs in
2024-12-06 11:13:28 +00:00
933772c3ec
DXIL ControlFlow Updates added IsForwardConnection()
...
Cache 2D bool array of IsBlock B in any path ahead of Block A
2024-12-06 11:13:28 +00:00
2e6368e6c2
DXIL ControlFlow optimisations, new APIs
...
Use rdcarray instead of unordered_map, unordered_set for hot containers
Made struct public and added new APIs
rdcarray<uint32_t> GetUniformBlocks();
rdcarray<uint32_t> GetLoopBlocks();
uint32_t GetNextUniformBlock(uint32_t from);
2024-12-06 11:13:28 +00:00
4102b66ec0
DXIL Debugger Added m_Assigned container
...
To track/assert if an SSA ID is used before it has been assigned
2024-12-06 11:13:28 +00:00
93bba6375d
DXIL disassembly custom decode asint(), asfloat()
...
Before:
int _1374 = (int)(_1364); // bitcast
float _1377 = (float)(_1376); // bitcast
After:
int _1374 = asint(_1364); // bitcast
float _1377 = asfloat(_1376); // bitcast
2024-12-06 11:13:28 +00:00
87547d6074
x86 compile fix
2024-12-06 11:04:19 +00:00
70c6dd32a2
Implement auditing of indirect RT dispatches
2024-12-05 13:31:35 +00:00
bdd479a5d4
Use RT indirect estimate directly
2024-12-05 13:31:23 +00:00
3366b25f0d
Add appropriate barriers for indirect RT arguments during patching
2024-12-05 13:31:23 +00:00
dcf9ef00dc
Fix missing flag to indicate that indirect count buffer is unused
2024-12-05 13:31:23 +00:00
e2f67eb9f6
Ignore a couple of DXBC chunks without processing or warning
2024-12-05 13:31:23 +00:00
a26e599c02
Rebase addresses to current form in indirect ray dispatches
2024-12-05 13:31:23 +00:00
9d698acbf4
Reset lod bias on GL when rendering textures internally
2024-12-05 13:31:23 +00:00
4bc41a6296
Fix nv aftermath include order
2024-12-05 13:31:23 +00:00
9bc648cf3b
Fix fetching DRED contexts
...
* Though these are broken as they are not properly processing BeginEvent strings
which are not packed/encoded
2024-12-05 13:31:23 +00:00
72fffe4cf2
Run patching for indirect RT executes while auditing, but don't execute
2024-12-05 13:31:23 +00:00
Jake Turner