| Age | Commit message (Collapse) | Author |
|---|
|
differentiable args (#4901)
* Explicitly detach derivative when forming a non-differentiable struct out of differentiable args
This fixes an issue where initializer lists get optimized out and lose information about non-differentiability.
There are 2 places where this could have been fixed:
1. When coercing initializer-list exprs, we can check for non-differentiable aggregate types and use a detach derivative on all the args.
2. Add an extra case in the peephole optimization step that adds detach-derivative when simplifying a make-struct of a non-differentiable type.
Even though solution 2 is more elegant, this PR goes with solution 1 simply to avoid having to use a differentiable-type-conformance-context that is used in the auto-diff IR passes to check for differentiability.
* Change test name + add expected vals
|
|
* Fix Varying Variable Location Assignments With Hull Shaders
Fixes: #4913
Fixes: #4540
Changes:
1. Added `kIROp_ControlBarrier` to HLSL/GLSL emitting.
2. Added a method to track 'used' and 'unused' varyings for when legalizing GLSL. This allows us to assign correct offsets to automatically added varyings
* Added a `ZeroLSB` check to UIntSet for this purpose
* add missing return
* code comment adjustment
* cleanup
* comment and HLSL controlBarrier mistake
* assume space for glsl/spriv varying is irrelevant
|
|
Fixes: #4899
Fixes invalid test results since `{...}` was differentiating the constructor of `myvector` when it should not (see #4877). This change modifies the test so it is correct so other PRs may be merged if indirectly/directly fixing the old use-case for this test.
|
|
* Implement `-fvk-use-dx-layout`
Fixes: #4126
Changes:
* Added fvk-use-dx-layout
* Modified `HLSLConstantBufferLayoutRulesImpl` for correctness (ex: Array is always 16 byte aligned)
* Added kFXCShaderResourceLayoutRulesFamilyImpl and kFXCConstantBufferLayoutRulesFamilyImpl to handle fvk-use-dx-layout
* Added `ConstantBufferLayoutRules` to manage constant buffer rules
* Added `alignCompositeElementOfNonAggregate`/`alignCompositeElementOfAggregate` to handle forced alignment of composites for ConstantBuffers
* `StructuredBuffer` rules are mostly equal to `scalar` layout, not much was needed to be changed to support this behavior.
* seperate legacy constant buffer and how Slang does constant-buffer normally
* undo an addition
* remove accidental test
* Address review and fix
Address review and remove GLSL support since GLSL requires a seperate legalization (need to linearlize structs like with `legalizeMetalIR` to assign explicit offsets)
* comments
* remove aggregate and non-aggregate logic
We don't need this distinction for the logic
---------
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
* Fix the slang-test bug
Since we reorganize the build directory, now the libraries are
located at different directory with executables in non-Windows
platform, we have to change the code on how to find the dll directory.
* Integrate the record/replay test into slang-unit-test
We create a unit-test-record-replay.cpp to run the converted slang
examples in child process as our tests for the record-replay layer.
* Disable the test on Apple
Due to the limitation of current examples, we temporarily disable them
on apples.
Change the ci to make this test only be run on the gpu-equipped runners,
for other runners we add a white-list file
"expected-failure-record-replay-tests.txt".
* Remove 'hello-world' example from unit test
"hello-world" doesn't use gfx abstract library, instead it uses vk directly, it's
not a preferable way. So we will drop this test, instead, we will use cpu-hello-world
example.
|
|
|
|
* Fix a false failure when using the latest SPIR-V
---------
Co-authored-by: Ellie Hermaszewska <github@sub.monoid.al>
Co-authored-by: ArielG-NV <159081215+ArielG-NV@users.noreply.github.com>
Co-authored-by: Ellie Hermaszewska <ellieh@nvidia.com>
|
|
* Support dependent generic constraints.
* Fix warning.
* Update comment.
* Fix.
* Add a test case to verify fix of #3804.
* Address review.
|
|
* Fix `tests\autodiff\reverse-while-loop-3.slang` test
Fixes: #4876
Fix `tests\autodiff\reverse-while-loop-3.slang` test to have correct listed output values
* set default to 1 (different val)
---------
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
* Tuple swizzling and element access.
* Update proposal status.
* Cleanup.
* Fix merrge error.
* Address review.
|
|
* Variadic Generics Part 2: IR lowering and specialization.
* Update design doc status.
* Update design doc.
* Resolve review comments.
|
|
|
|
|
|
Previously, the warnings were:
```
environment.slang(22): warning 41022: inout parameter 'seed' is never written to
public float3 environment_sample(StructuredBuffer <Environment_sample_data> sample_buffer, inout int seed)
^~~~~~~~~~~~~~~~~~
hit.slang(5): warning 41022: inout parameter 'seed' is never written to
float3 sample_lights(inout uint seed)
^~~~~~~~~~~~~
```
With this commit they should not be emitted.
|
|
* Refactoring benchmark CI pipeline
* Combine setup and run steps
|
|
* Do not zero-initialize groupshared and rayquery variables
Fixes: #4824
`-zero-initialize` option will explicitly not:
1. Set any groupshared values to defaults
2. Set any rayQuery object to a default state (currently invalid code generation)
* grammer
* disallow groupshared initializers
disallow groupshared initializers & adjust tests accordingly
* remove disallowed groupshared-init expression
* do not default init if non-copyable
---------
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
* Fix the issue that NonUniformResourceIndex is ignored
Fix the issue that after `specializeFunctionCalls`,
`NonUniformResourceIndex` is ignored in the generated specialized
function.
The reason is that if the function has a non-uniform resource parameter,
we will legalize it by replacing the resource parameter with a index,
and indexing of the resource will be moved inside the specialized function.
e.g.
```
void func(ResourceType resource) { ... }
func(resource[NonUniformResourceIndex(0)])
```
will be specialized into
```
void func(int index) { resource[index]; }
func(0);
```
In this case, inside the function, we will loose the information about
whether the resource is a non-uniform. So we add the handling for this
corner case by adding insert a `NonUniformResourceIndex` into the
specialized function:
```
void func(int index) {
int nonUniformIdx = NonUniformResourceIndex(index);
resource[nonUniformIdx];
}
```
* Fix the issue that arguments mismatch after specilization callsite
specializeCall() call could cause arguments mismatch with the parameters
of the specialized function.
For example, if the function parameter contains a resource type
```
void func(ResourceType res) { ... }
int index = ...
func(resources[index]);
```
This will be specialized into
```
void func(int index) { resources[index] }
int index = ...
func(index);
```
However, if we have more than 1 call sites, and the other call site
doesn't use `int` as the index, e.g.
```
uint index = ...
func(resources[index]);
```
this call site will be specialized into
```
uint index = ...
func(index);
```
this will be invalid, because the argument doesn't match the parameter.
so we just add the data type of the new arguments into the function key such that
For the uniformity info, we add a new attribute "IROp_NonUniformAttr",
so we will form a IRAttributedType that encodes both uniformity and data
type, and use it as the key of call info. So if there is call site using the different
data type for the resource index, we will specialize a new function for this.
* Handle the intCast and uintCast operation
Since after intCast/uintCast of nonuniformIndex, it's still a
nonuniformIndex. So we will handle this case as well.
Also, add a new test to cover this.
|
|
Adds a new Github CI action for benchmarking the slangc compiler on the MDL shaders. For now, the results are only dumped to the output of the CI, which can be later viewed through raw logs. The next step is to use github-action-benchmark to push these results into a page which will show the benchmark results over time as commits are pushed.
|
|
Fixes #4818
Emit memory qualifier once for GLSL targets.
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
* Support unicode identifier names.
* Fix.
* Fix language server.
* Fix build errors.
* Fix.
* Fix offset translation in language server.
|
|
Addresses #4698 as one approach to diagnose the potential problem.
Emit warnings when a user marks a parameter as `inout` but never writes to it in the function. A new intrinsic function `unmodified(out T)` has been added to explicitly indicate that an `inout` variable will not be modified in the function.
This is only one way to address the specific validation error in #4698. In general it seems that DXC does some more extensive checks on actual struct fields (as opposed to observing arbitrary struct writes), so that will be the next step.
|
|
* Fix auto-diff synthesized method naming conventions
* Update tests; remove unused var
|
|
Closes #4746.
This commit adds a support for "NumThreads" and "Shader" attribute keyword, which is in CamelCasing starting with an upper case letter.
The attribute keywords in HLSL are case-insensitive. As an example, one of D3D documents says,
"The attribute name "Shader" is case insensitive."
https://microsoft.github.io/DirectX-Specs/d3d/WorkGraphs.html
Slang, however, doesn't support the case-insensitivity. They should be all lower-case or CamelCasing starting with an upper case.
|
|
* Add embedded precompiled binary IR ops
Add IR operations to embed precompiled DXIL or SPIR-V blobs
into IR. Adds a BlobLit literal that is mostly identical to
StringLit except for its inability to be displayed, e.g.
in dumped IR. In the future, the blob might be dumped as
hexadecimal, but for now it is summarized as "<binary blob>".
* EmbeddedDXIL and SPIR-V options
The options, '-embed-dxil' and '-embed-spirv' in slangc, will
cause a target dxil or spirv to be compiled and stored in the
translation unit IR when written to a slang-module. Subsequent
changes actually implement the options.
* Per-translation unit DXIL precompilation
When -embed-dxil is specified, perform a precompilation to DXIL of
each TU, linked only with stdlib. Embed the resulting DXIL for
the TU in a IR op. Being part of IR, the precompiled DXIL can be
serialized to disk in a slang-module.
Upon loading slang-modules, the new IR op will be searched for and
the precompiled DXIL blob is saved with the loaded Module. During
linking, if all the Modules have precompiled blobs they will be
sent to the downstream compile commands as libraries instead of
source, skipping the downstream compilation, using DXC only for
linking.
Fixes Issue #4580
* Remove placeholder embedded SPIRV support
Code was added only to sketch out how other precompiled bins
will be supported.
* Remove the rest of the SPIRV placeholder support
* Fix warnings, test error on non-windows
* Remove lib_6_6 hack, add dxil_lib capability
* Allocate blob value from irmodule memarena
* Add null check after memarena allocation
* Restore the request->e2erequest code path for generatewholeprogram
* Update capability handling, move EmbedDXIL enum to end to preserve abi
* Remove lib_6_6 hack
* Move ICompileRequest functions to end
|
|
|
|
|
|
* Allow impliocit 'uniform' entrypoint parameters.
* Fix.
* Fix.
* Fix.
* Fix.
|
|
* Allow generic type deduction from ParameterBlock arguments.
* Fix test.
* Update expected failure list.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
* Proper warning generation for target switches and intrinsic asm
* Relaxing terminators
* Fix compiler warnings
* Rectified target switch reachability check
* Simplify target switch reachability check
* Refactoring variable names
* Using getBlocks
* Moving ad hoc special case to diagnostics source
* Using the LINE directive for testing
* Simplifying reliance on target switches
* Skipping IR generation for empty target switches
---------
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
* Fix IR lowering for generic interface types.
* Fix.
* Fix.
|
|
|
|
|
|
(#4751)
|
|
|
|
* Fix SPIRV emit for small-integer texture types.
* Disable -emit-spirv-via-glsl test.
|
|
|
|
* Disable warnings for input global variables
* Update comment to reflect actual check
Co-authored-by: ArielG-NV <159081215+ArielG-NV@users.noreply.github.com>
* Update comments in uninitialized-globals.slang
* Update uninitialized-globals.slang
* Refactoring test variable
* Typo in test
---------
Co-authored-by: ArielG-NV <159081215+ArielG-NV@users.noreply.github.com>
|
|
Fixes bug #3180
This test verifies the check for illegal swizzle on vector types, whose
element is not a basic Type (int, float).
The check captures the failure within swizzle access on a nested vector
subscript element.
vector<vector<int, 2>, 2> a
int b = a[0].x; // illegal
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
* Add `_Internal`/`External` atom validation and use enforcement.
Fixes: #4676
Changes:
* Added `validateInternalAtomExternalAtomPair` to the capability generator to ensure all `_Internal` atoms have a corresponding `External` atom.
* Validation of 'RequireCapabilityAttribute' warns if a user uses an '_Internal' atom.
* Added 'External' atoms to atoms with an already existing '_Internal' atom.
* Printing an atom removes '_'.
* Fixed some incorrect which were checking for the incorrect warning/error (capability4.slang, capability5.slang, capability6.slang).
* switch capability name to use `UnownedStringSlice` instead of `const char*`
switch capability name to use `UnownedStringSlice` instead of `const char*`, this includes using functions like `.startsWith`.
* grammer
---------
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
* Overhaul IR lowering of pointer types.
* Propagate address space in IRBuilder.
* Fixup.
* Fix.
* Fix.
* Change how Ptr type is printed to text.
* Fix.
|
|
* Check extensions before function parameters.
Fix decl ref formation for synthesized differentiable requirements that are inside an extension.
* Fix clang errors.
* More clang fix.
* Fix warnings.
* Fix build error.
* Fix.
* Fix typo.
|
|
* Allow implicitly casting enum types to bool.
* Fix.
|
|
Closes #4395
This commit allows Slang to use 1-dimensional matrix when targetting
HLSL. The 1-dimensional matrix is supported by DXC natively.
GLSL/Vulkan doesn't support the 1-dimensional matrix natively. It is
not trivial for Slang to convert all of matrix functions to vector or
scalar at the emitting step. We can implement this later if there are
needs. This commit disallows the use of 1-dimensional matrix for
targetting GLSL/Vulkan by the capability system; in other words, the
new 1-dimentional functions have "[require(hlsl)]".
|
|
* Add ResourceArray intrinsic type
* Move aliased parameter generation to GLSL legalization
* Add DynamicResourceEntry type for proxying layout of GenericResourceArray
* Reimplement as DynamicResource
* Add reflection test
* Don't reuse alias cache between different parameters
* Add dynamic cast extensions for buffer types
* Minor format fix
* Fix VarDecl diagnostics after finding non-appliable initializer candidates
---------
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
|
|
* Allow only specific spv storage classes for binding decoration
In
https://registry.khronos.org/vulkan/specs/1.3/html/chap37.html#VUID-StandaloneSpirv-DescriptorSet-06491
it states that
If a variable is decorated by DescriptorSet or Binding, the Storage
class must be UniformConstant, Uniform and StorageBuffer.
So apply this rule to our emit-spirv logic.
* Add a unit test
* Address few comments
|
|
Fixes: #4704
Change the type we assign when getting the member variable of a legalized `ConstantBuffer<T[...]>`.
|
|
fixes: #4700
Changes:
* If a uniform object (which uses uniform locations) has explicit bindings we will warn to use `ConstantBuffer<T>` instead. We check for a warning specifically when we know an object uses uniform layouts because objects may not use a uniform-layout register even if tagged with `uniform`. A good example of this is `uniform ConstantBuffer<T>`.
|
|
Fixes: #4675
Fixes: #4683
Fixes: #4443
Fixes: #4585
Fixes: #4172
Made the following changes:
1. All capability diagnostic printing logic tries to simplify before printing. This means that we do not print atoms which imply another atom.
2. Do not print the `_` prefix part of atom names since it is misleading users on what they should use to solve a capability issue encountered. (`_Internal` `External` atom changes are not in this PR)
3. Bundle together printing of all sets which contain exactly the same atoms (excluding abstract atoms). This allows printing the following `vertex/fragment/hull/domain/... + glsl` instead of `vertex + glsl | fragment + glsl | hull + glsl | domain + glsl | ....`
4. Rework how entry-point errors are reported to users (example at bottom of PR comment)
5. Rework how atom-provenance data is collected to be leaner and more useful so we can rework the errors. There are 2 notable changes here:
* We no longer store a list which describes where the first of an `CapabilityAtom` comes from. This heavily simplifies AST logic for the capability system. AST parsing of capabilities is much faster. The trade-off is faster AST parsing and correct AST node data for slower diagnostics if an error is found
* atom-provenance data now stores a reference to an atom's use-site to provide information on **where** and **what** is wrong with user code versus only sharing **what** and not where.
|
|
* Fix for invalid swizzle causing crash
Fixes #4689
If swizzle code is provided 5+ element swizzle the checkSwizzleExpr code will do an out of bounds array access and crash.
* switch test to check for to ensure no crash
* cleanup swizzle errors to only emit once
---------
Co-authored-by: Yong He <yonghe@outlook.com>
|
