| Age | Commit message (Collapse) | Author |
|---|
|
* Address issues with GLSL style global in/out vars (#6669)
Asserts and segfaults were observed trying to compile a simple
vertex shader like:
````
in int2 inPos;
[shader("vertex")]
main(uniform int2 test1, int2 test2, out float4 pos: SV_Position)
void main()
{
// Bogus use of all input vars to prevent optimizing out.
pos = float4(inPos.x, test1.x, test2.y, 0);
}
````
Further investigation found that while replacing "uniform int2 test1"
with "int2 test1" allowed for successful compilation, the resulting
output shader would have overlapping location qualifiers. For example,
compiling the above with "int2 test1" to glsl might give:
````
...
layout(location = 0) in ivec2 test1_0;
layout(location = 1) in ivec2 test2_0;
layout(location = 0) in ivec2 translatedGlobalParams_inPos_0;
...
````
This was because Slang does not actually support mixing GLSL style global
in/out vars and entry point params. However, this is never checked for
or noted in documentation. Slang source also assumes input shaders do not
mix these and these assumptions ultimately led to the observed asserts
and seg faults when using uniform entry point params.
This change makes updates to throw an error when the compiler detects that
it is trying to translate global in/out variables into entry point params
when an entry point already contains parameters, allowing for compilation
to fail gracefully.
Certain tests have been updated to avoid mixing GLSL style global in/out
vars and entry point params. This was mostly for tests that were using
functions like WaveGetLaneIndex which use global in vars for certain
platforms (see __builtinWaveLaneIndex).
* Address issues with GLSL style global in/out vars - updates 1 (#6669)
Update addresses review feedback to support mixing GLSL-flavored global
in/out vars and entrypoint parameters when either all global in/out vars
or all entry point params have a system value binding semantic.
* Address issues with GLSL style global in/out vars - updates 2 (#6669)
This update attempts to actually allow mixing GLSL style global in
vars and entry point vars.
Change attempts to recalculate offsets when adding the global input
vars into the recreated entry point params layout.
Additional updates were made to:
-resolve further issues uncovered with entry point uniform params.
-Address improper use of SV_DispatchThreadID in wave-get-lane-index.slang
for metal. "thread_position_in_grid" is not supported for signed integer
scalars or vectors.
-Fix a spirv casting conflict due to the implementation of
gl_PrimitiveID.get conflicting with PrimitiveIndex().
-Add a call to remove a global var in replaceUsesOfGlobalVar(). The global
var is already replaced in this function and keeping it around can prevent
it from being cleaned up by DCE if it still has decorations.
* format code
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
|
|
* Disable Link-Time-Optimization by default
LTO was requested for the release package a while ago.
When we added it, LTO was enabled by default although it was needed only for
the release packages.
Later we found that the Release build cannot be incrementally recompiled
when LTO is enabled. It sometimes works fine, but it required full recompilation
when it doesn't work. We added a new CMake option, `SLANG_ENABLE_RELEASE_LTO`,
to disable it for developers. But many Slang developers don't know the
option exists.
I was going to update the document, CONTRIBUTING.md, but I thought it
will be better to change the default behavior.
* Fix a compiler warning treated as an error on linux
A padding variable was uninitialized, which is fine, but the compiler
was complaining about it.
* Fix other gcc error for uninitialized variable
* Fix more compile warning treated as error
* Fix compiler warning from gcc 11
It appears that this is a valid warning that the `delete this` is done
on an offset 8 when the class uses multiple inheritance.
The compiler warning is following:
```
In file included from /home/runner/work/slang/slang/source/core/slang-memory-file-system.h:5,
from /home/runner/work/slang/slang/tools/slang-unit-test/unit-test-module-ptr.cpp:3:
In destructor ‘virtual Slang::ComBaseObject::~ComBaseObject()’,
inlined from ‘uint32_t Slang::ComBaseObject::_releaseImpl()’ at /home/runner/work/slang/slang/source/core/slang-com-object.h:49:16,
inlined from ‘virtual uint32_t Slang::MemoryFileSystem::release()’ at /home/runner/work/slang/slang/source/core/slang-memory-file-system.h:34:5:
/home/runner/work/slang/slang/source/core/slang-com-object.h:33:31: error: ‘void operator delete(void*, std::size_t)’ called on pointer ‘<unknown>’ with nonzero offset 8 [-Werror=free-nonheap-object]
33 | virtual ~ComBaseObject() {}
| ^
In destructor ‘virtual Slang::ComBaseObject::~ComBaseObject()’,
inlined from ‘uint32_t Slang::ComBaseObject::_releaseImpl()’ at /home/runner/work/slang/slang/source/core/slang-com-object.h:49:16,
inlined from ‘virtual uint32_t Slang::MemoryFileSystem::release()’ at /home/runner/work/slang/slang/source/core/slang-memory-file-system.h:34:5,
inlined from ‘Slang::ComPtr<T>::~ComPtr() [with T = ISlangMutableFileSystem]’ at /home/runner/work/slang/slang/include/slang-com-ptr.h:113:34,
inlined from ‘void _modulePtr_impl(UnitTestContext*)’ at /home/runner/work/slang/slang/tools/slang-unit-test/unit-test-module-ptr.cpp:92:1:
/home/runner/work/slang/slang/source/core/slang-com-object.h:33:31: error: ‘void operator delete(void*, std::size_t)’ called on pointer ‘<unknown>’ with nonzero offset 8 [-Werror=free-nonheap-object]
33 | virtual ~ComBaseObject() {}
| ^
/home/runner/work/slang/slang/tools/slang-unit-test/unit-test-module-ptr.cpp: In function ‘void _modulePtr_impl(UnitTestContext*)’:
/home/runner/work/slang/slang/tools/slang-unit-test/unit-test-module-ptr.cpp:36:69: note: returned from ‘void* operator new(std::size_t)’
36 | ComPtr<ISlangMutableFileSystem>(new Slang::MemoryFileSystem());
| ^
```
The problem is on the fact that `ComBaseObject` is not the first in the
multiple inheritance:
```
class MemoryFileSystem : public ISlangMutableFileSystem, public ComBaseObject
{
public:
// ISlangUnknown
SLANG_COM_BASE_IUNKNOWN_ALL
```
It should be:
```
class MemoryFileSystem : public ComBaseObject, public ISlangMutableFileSystem
```
The chain of ComObject release is little complicated and it is easy to
make a mistake. Here is summary with details,
1. `release()` is declared as a pure-virtual in ISlangUnknown, which is
one of the base classes of `ISlangMutableFileSystem`.
```
struct ISlangUnknown
{
virtual SLANG_NO_THROW uint32_t SLANG_MCALL release() = 0;
```
2. `release()` is implemented with the macro
`SLANG_COM_BASE_IUNKNOWN_RELEASE`.
```
SLANG_NO_THROW uint32_t SLANG_MCALL release() SLANG_OVERRIDE \
{ \
return _releaseImpl(); \
}
inline uint32_t ComBaseObject::_releaseImpl()
{
// Check there is a ref count to avoid underflow
SLANG_ASSERT(m_refCount != 0);
const uint32_t count = --m_refCount;
if (count == 0)
{
delete this;
}
return count;
}
```
3. The instance of `MemoryFileSystem` is handled by ComPtr. And
`ComPtr::~ComPtr()` calls the `release()`.
```
ComPtr<ISlangMutableFileSystem> memoryFileSystem =
ComPtr<ISlangMutableFileSystem>(new Slang::MemoryFileSystem());
SLANG_FORCE_INLINE ~ComPtr()
{
if (m_ptr)
((Ptr)m_ptr)->release();
}
```
4. When `delete this` is called, because ComBaseObject is not the first
in the multiple inheritance, `this` is 8 byte off from the actual
instance address.
A fix for this is to change the order of the inheritance and make
ComBaseObject to be the first in the order.
|
|
Close #7315.
We have couple mis-definition in capability.
sm_50 shouldn't require cuda compute_9_0, drop it to compute_6_0
unpack should only require compute_6_0
subgroup_ballot will require sm_60
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
WGSL doesn't support isnan and isinf, because it assumes that it always uses fast-math and fast-math doesnt' handle NaN as defined in IEEE standard.
The initial implementation used a clever workaround but it stopped working from some point.
This PR implemented isnan and isinf with a bitwise operation, which can be expensive.
But that seems to be an only option at the moment.
|
|
* Fix#6993 - Emit Diagnostic Warning and Fix SIGSEGV
* Update external/slang-rhi submodule
* Add checks for valid stage names for paq in SemanticsVisitor check
* format code
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
Co-authored-by: Ellie Hermaszewska <ellieh@nvidia.com>
|
|
* Fix missing debug info in the included slang file
Issue:
https://github.com/shader-slang/slang/issues/7271
Debug info including DebugFunction, DebugLocation, and DebugValue
are missing in IR for "#included" Slang shader file.
The included shader file was not added to TranslationUnit's source
file list, therefore mapSourceFileToDebugSourceInst.add() was not
called for the source in generateIRForTranslationUnit(), and later
mapSourceFileToDebugSourceInst.tryGetValue() could not get value
for the source to add DebugLocationDecoration, which led to missing
DebugFunction, DebugLocation and other debug info for the included
file in IR.
Adding the include file in TranslationUnit's source file list fixes
the issue.
* Add source file using PreprocessorHandler
Call _addSourceFile from FrontEndPreprocessorHandler::handleFileDependency.
* Just use FrontEndPreprocessorHandler
* Make _addSourceFile public
* format code
* Distingush the included source file
* Add m_includedFileSet to avoid adding dup file
HashSet<SourceFile*> m_includedFileSet;
---------
Co-authored-by: ArielG-NV <159081215+ArielG-NV@users.noreply.github.com>
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
|
|
* Add legalization for 0-sized arrays.
* Allow 0-sized arrays in the front-end.
* More tests.
* Add `Conditional<T, hasValue>` type to core module.
* Update toc.
* Fix wording.
* Update test.
|
|
* Make interface types non c-style.
* Make Optional<T> work with autodiff and existential types.
* Fix.
* patch behind slang 2026.
* Fix warnings.
* cleanup.
* Fix tests.
* Fix.
* Fix com interface lowering.
* Add comment to test.
* regenerate command line reference
* Add test for passing `none` to autodiff function.
* Fix recording of `getDynamicObjectRTTIBytes`.
* Fix nested Optional types.
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
|
|
When we return a raw point to a module, we should decrement the
reference count. The module is owned by its session so it should be
valid as long as the session is valid.
|
|
|
|
|
|
|
|
|
|
|
|
* Add check for the variable requirement
This change adds the capability check for the variables requirement.
With this check, the shader
```
[require(cpp_cuda_glsl_hlsl_metal_spirv)]
Buffer<float> InputTyped;
[require(cpp_cuda_glsl_hlsl_metal_spirv)]
RWBuffer<float> OutputTyped;
```
will issue error if targeting to WSGL
e.g. `.\build\Debug\bin\slangc .\tests\wgsl_no_buffer.slang -o
wgsl_no_buffer.txt -target wgsl -entry Main -stage compute`
.\tests\wgsl_no_buffer.slang(2): error 36108: 'InputTyped' has dependencies that are not compatible on the required target 'wgsl'.
Buffer<float> InputTyped;
^~~~~~~~~~
.\tests\wgsl_no_buffer.slang(4): error 36108: 'OutputTyped' has dependencies that are not compatible on the required target 'wgsl'.
RWBuffer<float> OutputTyped;
^~~~~~~~~~~
Fixes #6304
* Add var capability tests
* Do capability checks for global var only
* Add inferredCapabilityRequirements to var capability check
* Add requirement to the intrinsic types Buffer/RWBuffer
* format code
* Update capabliity test
* use DefaultDataLayout as default data layout
* Use visitMemberExpr to check the capabilities
* Update the cap tests to match the error messages
* update test to use the ScalarDataLayout for hlsl target
* Update tests check condition to use error number only
* Add default push_constant data layout type
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
|
|
(#7283)
* Ensure we do not have an initExpr on a var inside an InterfaceDecl
Ensure we do not have an initExpr on a var inside an InterfaceDecl. If we do, send an error.
Ensure the language server does not segfault with this error as per the issue.
* format code
* split tests
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
|
|
* Fix lua header file path
Add two missed files in #7167
* Fix lua header file path
Add two missed files in #7167
* Leave lua/ in the path to avoid name conflict
* Remove xxx from path of SLANG_OVERRIDE_xxx_PATH
Change SLANG_OVERRIDE_xxx_PATH from path-to-parent-folder/xxx
to path-to-parent-folder and add "xxx/" back to "#include",
which helps to avoid the potential name conflict of external tools.
* format code
---------
Co-authored-by: Yong He <yonghe@outlook.com>
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
|
|
The files `slang-fossil.{h,cpp}` define a new serialization format that is designed to support data being memory-mapped in and then traversed as-is.
The `docs/design/serialization.md` document was updated with details on this new format.
The `slang-serialize-fossil.{h,cpp}` files define implementations of the recently introduced `ISerializerImpl` interface for reading/writing this new binary format.
The overall structure of these implementations is heavily based on the existing RIFF implementation from `slang-serialize-riff.{h,cpp}`.
Switching the AST serialization over to use this format required almost no changes to `slang-serialize-ast.cpp`.
The new format is more space-efficient than the RIFF-based format in memory (by factor of over 2x), but is actually *worse* than the RIFF-based format in terms of how it affects the size of `slang.dll`, because the new format is seemingly less amenable to LZ4 compression.
A few pieces of utility code were added or moved as part of this work:
* The `core/slang-internally-linked-list.*` implementation is just a type that was used as part of `core/slang-riff.*`, but that wasn't really RIFF-specific.
* The `core/slang-blob-builder.*` files implement a low-level utility for building a binary format in memory out of "chunks". The overall structure of this type is based on the RIFF-specific builder implementation, but has been generalized so that it should apply to other kinds of binary serialization.
* The `core/slang-relative-ptr.h` file implements a simple relative pointer type, which is currently only used by the `slang-fossil.h` format.
If there are concerns about adopting the new format immediately for the AST, this change could be modified to introduce all the new code, but leave the AST serialization using the previous RIFF-based format.
|
|
* Enable LSS hit object test
Enabled LSS SER tests now that PR #7211, which added SER support to OptiX,
has been merged.
Ran: ./build/Debug/bin/slangc.exe tests/cuda/lss-test.slang -target ptx
-Xnvrtc -I"C:/ProgramData/NVIDIA Corporation/OptiX SDK 9.0.0/include"
and confirmed that the HitObject intrinsic is called.
eg:
call (%f15, %f16, %f17, %f18, %f19, %f20, %f21, %f22),
_optix_hitobject_get_linear_curve_vertex_data, ();
* format code
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
|
|
* Fix SPIRV specialization constant with floating-point operations
* Improve test
* WIP
* Restrict `OpSpecConstantOp` allowed operations based on SPIRV specifications
* Fix typo on floating type check
* Emit error on float to int spec cosnt int val casts
|
|
With this PR, MapElement works for the following signatures:
- CoopMat<...>::MapElement(functype(...));
- CoopMat<...>::MapElement(capturing-lambda);
- CoopMat<...>::MapElement(not-capturing-lambda);
- Tuple<CoopMat<...>,...>::MapElement(functype(...));
- Tuple<CoopMat<...>,...>::MapElement(capturing-lambda);
- Tuple<CoopMat<...>,...>::MapElement(not-capturing-lambda);
|
|
* Language version + tuple syntax.
* Fix compile error.
* regenerate documentation Table of Contents
* Fix.
* regenerate command line reference
* Fix.
* Fix.
* Fix more test failures.
* revert empty line change,
* Retrigger CI
* #version->#lang
* Update source/core/slang-type-text-util.cpp
Co-authored-by: ArielG-NV <159081215+ArielG-NV@users.noreply.github.com>
* Remove comments.
* Fix parsing logic.
* Fix parser.
* Fix parser.
* update test comment
* Update options.
* regenerate documentation Table of Contents
* regenerate command line reference
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
Co-authored-by: ArielG-NV <159081215+ArielG-NV@users.noreply.github.com>
|
|
* change default descriptor binding to be VkMutable
|
|
Fixes #6987
|
|
* WiP: LSS intrinsics: initial commit
* format code
* Fix CI failures
* Address review comment
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
|
|
The `SLANG_UNREFLECTED` macro has been completely meaningless since we switched away from the old AST serialization/reflection approach, so the lingering uses of it in the code as pointless at best and misleading/confusing at worst.
|
|
to (#6326) (#7194)
* Fix operator precedence in OptiX ray payload pointer casting
Added extra parentheses around the cast to ensure proper operator precedence when
dereferencing the OptiX ray payload pointer. This fixes the issue where the compiler
was treating the expression as (RayPayload_0 *)getOptiXRayPayloadPtr()->color_0 instead of
((RayPayload_0 *)getOptiXRayPayloadPtr())->color_0.
Error:
nvrtc 12.9: tests/cuda/optix-cluster.slang(17): error : expression
must have pointer-to-class type but it has type "void *"
nvrtc 12.9: note : (RayPayload_0
*)getOptiXRayPayloadPtr()->color_0 = color_1;
nvrtc 12.9: note : ^
Tested using:
./build/Debug/bin/slangc -target ptx -Xnvrtc
-I"/home/haaggarwal/NVIDIA-OptiX-SDK-9.0.0-linux64-x86_64/include"
-DSLANG_CUDA_ENABLE_OPTIX -entry closestHitShaderA
./tests/cuda/optix-cluster.slang
* Fix Check
|
|
* Implement shader execution reordering support for OptiX
Added OptiX backend support for Shader Execution Reordering (SER) features as outlined in issue #6647. This implementation:
1. Added CUDA target support for HitObject API
2. Implemented core SER functionality (TraceRay, MakeHit/Miss, Invoke)
3. Added OptiX-specific hit object handling functions
4. Added test case for OptiX SER functionality
* format code
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
|
|
|
|
* Properly implement WaveMask* variants of WaveMultiPrefix* intrinsics
* More partitioned intrinsics
* More partitioned intrinsics and cleaned up non-prefixed WaveMask* implementations
* Refactor HLSL WaveMultiPrefix* implementations
* fix cap atoms
* Clean up implementation
* Add GLSL intrinsics and cleanup
* Add tests
* Fix affected capability test
* Update and fix tests
* Move expected.txt file
* Refactor WaveMask* to call WaveMulti*
* Refactor SPIRV/GLSL preamble code
* Enable emit-via-glsl tests
* remove wave_multi_prefix capability in favor of subgroup_partitioned
* Update docs
* Update cap atoms doc
|
|
* List all source files in debug source file list
The source file which does not participate in the line table is
missing from the debug source file list. Always copy IRDebugSource
instruction in linkIR() to fix the issue.
* Update the code to address review
* Add [[fallthrough]]
|
|
* Implement throw statement
It already existed in the IR, so only parsing, checking and lowering was
missing.
* Initial catch implementation
Likely very broken.
* Error out when catch() isn't last in scope
* Prevent accessing variables from scope preceding catch
As those may actually not be available at that point.
* Add IError and use it in Result type lowering
* Add diagnostic tests
* Allow caught throws in non-throw functions
* Fix catch propagating between functions & SPIR-V merge issue
* Add test for non-trivial error types
* Fix MSVC build
* Fix invalid value type from Result lowering
* Also lower error handling in templates
* Lower result types only after specialization
* Attempt to disambiguate error enums by witness table
* Revert matching by witness, types should be distinct too
* Don't assert valueField when getting Result's error value
It may not exist if the function returns void, but getting the error
value is still legitimate.
* Update tests for new error numbers & get rid of expected.txt
* Change catch lowering to resemble breaking a loop
... To make SPIR-V happy.
* Fix dead catch blocks and invalid cached dominator tree
* More SPIR-V adjustment
* Lower catch as two nested loops
* Add defer interaction test and revert broken defer changes
* Fix enum type when throwing literals
* Cleanup and bikeshedding
* Document error handling mechanism
* Fix table of contents
* Use boolean tag in Result<T, E>
* Use anyValue storage for Result<T,E>
* Remove IError
* Fix formatting
* Eradicate success values from docs and tests
* Use parseModernParamDecl for catch parameter
* Implement do-catch syntax
* Implement catch-all
* Fix formatting
* Fix marshalling native calls that throw
---------
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
* Add pointer/T* variants for coop vec load/store
* fix stride decoration and improved test
* fix compile warnings
* Improve test
* Use `coopVecLoad` function in test
|
|
* Add default constructor for Ptr type
* Make pointers c-style type, remove __init() constructor
|
|
* Implement default initializer list for C-Style type member
Close #6189.
Previsouly, for the C-Style member in a struct, if it doesn't have any initialize
expression, when we synthesize the ctor, we will not associate the
default value for the parameter corresponding to that member.
This bring some trouble that existing slang users has to add '= {}' to
every struct fields in order to make all the parameters in the synthesized ctor having
a default value, so people can still use `Struct a = {}` to create a
struct.
To make this use case convenience, we will automatically associated a
'= {}' as the default value for this case.
This PR also add support for empty initializing link-time sized vector/matrix by "= {}".
In addition, this PR also fix a bug in auto diff where we should not report error when proccessing
transpose on an empty struct.
|
|
(#7213)
* Make sizeof(generic) work as compile-time constant
* format code
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
|
|
fixes: [#7143](https://github.com/shader-slang/slang/issues/7143)
fixes: [#7146](https://github.com/shader-slang/slang/issues/7146)
Goal of PR:
* This PR is part of the larger #7115 refactor to how dynamic dispatch works.
* The first step is to add the `-std <std-revision>` flag.
* The second step is to provide basic `dyn` keyword support in AST. This does not include `varDecl` support since most of these interactions require `some` keyword support.
Future PR(s) goal:
* Support `some` keyword in AST. With this we will also implement all varDecl interactions between `dyn` and `some`.
* Add IR support for `some` and `dyn`.
Breakdown of PR:
* most of the logic is in `validateDyn.*`. This was done so that in the future when we implement more features we will have an easy time removing/adding restrictions to `dyn` interfaces.
Breaking changes:
* As per spec (https://github.com/shader-slang/spec/pull/14/files), any type conforming to a `dyn` interface errors if member list contains one of the following: opaque type, non copyable type, or unsized type.
* Due to the breaking change, the test `tests\compute\dynamic-dispatch-bindless-texture.slang` is incorrect. This has been fixed.
|
|
* Fix incorrect way of type checking inside of CoopVec Multiply
Related to an issue #7201.
The operator `is` is silently failing when the right-hand-side operand is an interface type.
This PR uses concrete types to check the type.
|
|
* Add inverse hyperbolic derivatives
* Add test
|
|
This change takes the new approach to serialization that was used for the AST and generalizes it in a few ways:
* The new approach is no longer tangled up with the RIFF format.
The serialization system supports multiple different implementations of the underlying format.
The existing RIFF format is now supported as one back-end, but support for others will follow in subsequent changes.
* The new approach is no longer deeply specialized to AST serialization.
The old code had things like serialization for `List`s and `Dictionary`s, but it was embedded inside the `AST{Encoding|Decoding}Context`, and thus couldn't be leveraged for other serialization tasks.
This change factors out a completely AST-independent `Serializer` implementation, with an `ASTSerializer` layered on top of it to provide the additional context needed.
* There is less duplication of code between reading and writing of serialized data.
The old code had both the `ASTEncodingContext` and `ASTDecodingContext`, with serialization logic for most types being implemented in both, but with the constraint that those implementations needed to be kept in sync to avoid serialization-related runtime failures.
A key property of the revamped approach is that a single `serialize()` method for a type implements both the reading and writing directions of serialization.
|
|
semantics (#7150)
* Map SV_VertexID to `gl_VertexIndex - gl_BaseVertex`, provide SV_Vulkan* SV semantics
* Fix docs
* Regenerate toc
* Fix affected pointer-2 test
* Add tests
---------
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
The user can explicitly use Vulkan memory model, or it will be
automatically used when cooperative-matrix is used.
When vulkan memory model is used, two keywords, "Coherent" and
"Volatile", are not allowed.
There are many differences regarding atomic and texture but
this PR has changes limited to support `globallycoherent`
keyword. When variables with `globallycoherent` is used with `OpLoad`, it
will use additional options, `MakePointerAvailable|NonPrivatePointer`,
that will provide the same effect. For `OpStore`, it will use
`MakePointerVisible|NonPrivatePointer`.
|
|
* Add noreturn attribute to IgnoreHit
* Revert "Add noreturn attribute to IgnoreHit"
This reverts commit 3cf2354dada71b9a8713b08f3a2e261de4aabfa4.
* Fix: Preserve inout param modifications with OptiX IgnoreHit()
Issue #6326 identified that in OptiX, when using IgnoreHit() (which
maps to the "noreturn" optixIgnoreIntersection() intrinsic), any
modifications made to 'inout' parameters within the shader would be
lost. This was due to IgnoreHit() preventing the execution of the
copy-back operation from the temporary variable (used to implement
'inout' semantics) to the original parameter.
This commit introduces a new IR pass, 'undoParameterCopy', specifically
for CUDA/OptiX targets to address this. The pass operates as follows:
1. Identifies temporary IR variables created for 'inout' parameters,
which are now decorated with 'TempCallArgVarDecoration'.
2. Maps these temporary variables back to their original parameter
storage (e.g., the OptiX payload pointer).
3. Replaces all uses of the temporary variable directly with the
original parameter pointer.
4. Removes the temporary variable declaration and its initializing store
(which copied from the original parameter to the temporary).
By transforming the IR to operate directly on the original parameter
storage before any potential call to IgnoreHit(), this fix ensures
that all modifications are preserved, correctly resolving issue #6326.
The pass is integrated into the compilation flow for relevant targets.
* Refactor(IR): Optimize GetOptiXRayPayloadPtr for better DCE/CSE
To allow for more effective dead code elimination (DCE) and
common subexpression elimination (CSE) of `getOptiXRayPayloadPtr`
instructions, this commit:
- Marks `kIROp_GetOptiXRayPayloadPtr` as side-effect-free within
`IRInst::mightHaveSideEffects` (in `slang-ir.cpp`).
- Flags `GetOptiXRayPayloadPtr` as `HOISTABLE` in its definition
within `slang-ir-inst-defs.h`.
This addresses scenarios where multiple, potentially redundant,
calls to `getOptiXRayPayloadPtr` might appear in the IR,
allowing optimizers to produce cleaner and potentially more
efficient code for OptiX targets. This change supports efforts
to refine IR handling for ray-tracing shader stages.
* Remove debugging code
* Refactor UndoParameterCopyVisitor for improved performance
- Optimized IR traversal by combining multiple passes into a single scan
- Removed unnecessary dictionary, immediately replace uses when a temp var is found
- Reduced duplicate code paths by checking for both temp vars and redundant stores in one loop
- Better handling of the 'changed' flag to ensure DCE only runs when needed
- Results in fewer instruction traversals and improved efficiency for large functions
* Add Test
|
|
(#7010)
* Fix structured buffer get dimensions
* Further fixes and added tests
* Remove unnecessary include
* Fix test issues
* attempt to fix wgpu crash
* test remove half usage in test
* attempt to fix WGPU test issue
* Another attempt to fix WGSL test - make test similar to the existing GetDimensions test
---------
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
Fix correct bindings for bindless resource model [spirv and glsl]
fixes: #6952
Problem:
* Currently all bindless objects are placed in the same set (fine) and same binding (incorrect behavior for vulkan). This is incorrect since as per [spec](https://registry.khronos.org/vulkan/specs/latest/man/html/VkDescriptorType.html), only 1 resource type may be written to each index inside a set (these rules are loosened with VK_EXT_mutable_descriptor_type)
* This means currently generated bindings do not work in practice if we (for example) use `Sampler2D.Handle` and `Texture1D.Handle` in a shader since we would place 2 incompatible objects in the same binding-index and set.
Solution:
* `__getDynamicResourceHeap` was modified to allow bindings to chosen dynamically for a descriptor
* use `IOpaqueDescriptor` to check compile-time information of resource types so that we can identify different resources
* Using this information of `IOpaqueDescriptor`, we modify `defaultGetDescriptorFromHandle` to provide a binding model (1 resource per binding-index) which produces legal spirv/glsl.
* To support `VK_EXT_mutable_descriptor_type` the function `defaultGetDescriptorFromHandle` has a set of options (`BindlessDescriptorOptions`) for a user to pick-from to support their binding model. Capabilities are not used here for flexibility purposes (specifically old shaders mixed with modern vulkan extensions).
Other changes:
* Added `TexelBuffer` DescriptorKind to aid in generating correct bindings
* format code
* Add to docs bindless changes, make AccelerationStructure use its handle directly, adjust tests accordingly
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
|
|
* Enforce rule that `export`/`extern` (non cpp) must be `const`
fixes: #5570
Problem:
1. we allow non-const-link-time-var to be linked to a const-link-time-var.
2. problem is that: module use site has const var, so, we emit OpStore %Ptr %Const in IR, this is expected, this is good. We fail because we in reality have a OpStore %Ptr %Var (fails since we need a OpLoad in-between) in IR since the module with our link-time-variable-value is a regular variable.
3. We loose the float_litteral talked about inside the github issue since, we technically don't use our variable "VAL" (we never OpLoad from it), so spirv-opt removes the float_litteral, this is a byproduct of the actual issue.
Solution:
* `export`/`extern` variables must always be `const`. This excludes `__extern_cpp` since `cpp` does not exhibit this issue and works differently.
* format code
* changel logic and tests to only ensure `static const` with `export`/`extern`
* changing the rules: only reqirement is that if we have const we must have static
* remove a spirrious change made
* fix merge
---------
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
* Fix HLSL ByteAddressBuffer Load* parameter integer type
* Fix tests
* Fix load with alignment function signature clash
* Fix LoadAligned tests
|
|
|
|
Fixes #7127
|
|
Fixes issue #6898
The -emit-spirv-via-glsl slang-test option has been broken for
some amount of time. Tests that were using it were operating as
if using -emit-spirv-directly, leading to many duplicated tests.
After fixing the test option, there were an number of errors that
appeared as a result.
This change fixes the broken test option and the resulting test
errors. Some of the test errors revealed some legitimate issues,
such as:
-The GLSL bitCount instrinsic only supports 32-bit integers and
requires emulation for other bit widths.
-Emitting GLSL 8-bit and 16-bit glsl integer types did not emit
the proper extension requirements
-Emitting GLSL and casting for 16-bit integers was missing a
closing parenthesis.
-Missing profile for GL_EXT_shader_explicit_arithmetic_types
-Missing toType cases for UInt8/Int8 for the kIROp_BitCast case
in tryEmitInstExprImpl.
|
