slang.git/source/slang/core.meta.slang, branch master

Immutable access qualifier for pointers and use `__ldg` on cuda. (#8710)

2025-10-16T03:59:47+00:00

This PR implements `Access.Immutable` to allow pointers to immutable data. The new type `ImmutablePtr` is defined as an alias of `Ptr`. By forming a immutable pointer, the programmer is conveying to the compiler that the data at the pointer address will never change during the execution of the current program. Therefore loads from immutable pointers can be deduplicated by the compiler, and will translate to `__ldg` when generating code for CUDA. The SPIRV backend is not changed in this PR, since the current SPIRV spec makes it very difficult to specify loads from immutable address without generating tons of wrappers and boilerplate type declarations. We would like to see the spec evolved a bit to around its support of `NonWritable` physical storage pointers or immutable loads before we attempt to express such immutability in SPIRV. For now we simply emit ordinary pointers and loads when generating spirv. --------- Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

Clean up Slang IR representation of undefined values (#8708)

2025-10-15T01:00:47+00:00

Prior to this change, the Slang IR used a single opcode (`kIROp_Undefined`) to encode all cases of undefined values. The particular motivation for this change was a need to distinguish those undefined values that represent a load from an uninitialized memory location versus other sorts of undefined values. If transforming a variable into SSA form results in `undefined` values in cases where the a `load` was executed without a prior `store`, that represents an error on the programmer's part, and should be diagnosed. However, other cases of undefined values can arise during program transformation and optimization, and should not typically result in diagnostics being emitted. While it was not the original motivation for this change, it is also worth noting that the LLVM project has transitioned from initially using only a single `undef` instruction to having a more nuanced model, and the same factors that motivated their shift also apply to the Slang IR. Counter-intuitively, the semantics of undefined values actually need to be carefully defined. Concretely, this change splits the pre-existing `undefined` opcode into two sub-cases: - `kIROp_LoadFromUninitializedMemory`, to represent the case of loading from a memory location (such as a local variable) that has not been initialized. - `kIROp_Poison`, corresponding to the LLVM `poison` value. Our poison instruction is intended to have semantics comparable to LLVM's equivalent. Conceptually, any operation that is invoked with a poison value as input will (with a few exceptions) produce a poison value as output. One can think of the behavior of `poison` as similar to how not-a-number values propagate in floating-point computations: by default they "infect" the result of any computation they are involved in. This semantic choice helps to ensure that many optimizations end up being correct in the presence of undefined values, even if they did not specifically account for them. The `kIROp_LoadFromUninitializedMemory` case is comparable to the combination of `freeze` and `undef` in LLVM. An LLVM `undef` value has semantics that allow *each* use of that value to be replaced with a *different* arbitrary value; these semantics cause many optimizations to only be correct in the absence of undefined values. An LLVM `freeze` instruction can take an undefined value as input, and produces a single value that is still arbitrary, but must be consistent across all uses. The latter semantics are what we want, since a given `load` from an uninitialized memory location will yield an arbitrary-but-fixed value. Note that we intentionally do not have a direct analogue to LLVM's `undef` instruction, because of the way that `undef` causes so many complications when trying to write optimizations. We also do not add a `kIROp_Freeze` instruction in this change, but that is simply because we currently have no need for it. Existing code that was creating `IRUndefined` values has been updated to create either `IRPoison` or `IRLoadFromUninitializedMemory` values, as appropriate to the use case. Code that was checking for the `kIROp_Undefined` opcode has been updated to either check for both of the new opcodes (in the case of `switch` statements), or to use `as` to perform a dynamic cast to the common base type of the two new instructions. Note that this change does not alter the way that instructions representing undefined values are typically emitted as ordinary instructions in the block that produces an undefined value. While emitting `IRLoadFromUninitializedMemory` as an ordinary instruction is exactly what we want, the `IRPoison` case would actually be better represented in Slang IR as a "hoistable" instruction, so that there would only be a singular `poison` value of each type. Changing `IRPoison` to be hoistable would be a good follow-up change, but might run into more challenges depending on what assumptions (if any) the codebase is making about where undefined values get emitted. --------- Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

Addition of `Load`/`Store` coherent operations (#8395)

2025-10-10T17:09:24+00:00

Fixes: https://github.com/shader-slang/slang/issues/7634 Duplicate of PR https://github.com/shader-slang/slang/pull/8052 Primary Changes: * Added `storeCoherent` and `loadCoherent` for coherent load/store via pointers. This is backed by `IRMemoryScopeAttr` which is an `IRAttr` attached to `IRLoad` and `IRStore` * Logic in `source\slang\slang-emit-spirv.cpp` for load/store emitting has been reworked to be less messy and more maintainable * Add to `hlsl.meta.slang` coop vector and coop matrix coherent load/store operations Secondary Changes: * Added a missing load/store test for coop matrix: `tests\cooperative-matrix\load-store-pointer.slang` --------- Co-authored-by: ArielG-NV Co-authored-by: ArielG-NV <159081215+ArielG-NV@users.noreply.github.com> Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com> Co-authored-by: Nathan V. Morrical

Rename some symbols related to pointers types (#8592)

2025-10-03T04:48:11+00:00

Note that while this change touched a large numer of files, there are no changes to functionality being made here. The only things being done are renaming various symbols and, in a few cases, updating or adding comments for consistency with the new names. The core of the naming changes are: * Most things named to refer to `OutType` (e.g., `IROutType`, `IRBuilder::getOutType()`, etc.) have been consistently renamed to refer to `OutParamType`, to emphasize that the relevant AST/IR node types are only intended for use to represent `out` parameters. * The same change as described above for `OutType` is also made for `RefType`, which becomes `RefParamType` in most cases. One mess that this exposes is the way that the `ExplicitRef` type in the core module currently lowers to `IRRefParamType`. This change sticks to the rule of not making functional changes, so that mess is left as-is for now. * Names referring to `InOutType` have been changed to instead refer to `BorrowInOutType`. The intention with this naming change is to emphasize that the Slang rules for `inout` are semantically those of a borrow (or at least our interpretation of what a borrow means). * Names referring to `ConstRefType` have been changed to instead refer to `BorrowInType`. This change starts work on clarifying that the existing `__constref` modifier was never intended to be a read-only analogue of `__ref`, and instead is the input-only analogue of `inout`. * The `ParameterDirection` enum type has been changed to `ParamPassingMode`, to reflect the fact that the concept of "direction" fails to capture what is actually being encoded, particularly once we have modes beyond simple `in`/`out`/`inout`. While this change does not alter behavior in any case (the user-exposed Slang language is unchanged), it is intended to set up subsequence changes that will work to make the handling of these types in the compiler more nuanced and correct. Breaking this part of the change out separately is primarily motivated by a desire to minimize the effort for reviewers. --------- Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

Rewriting the lower-buffer-element-type pass to avoid unnecessary packing/unpacking. (#8526)

2025-09-30T00:45:08+00:00

Part of the effort to improve the performance of generated SPIRV code. The existing lower-buffer-element-type pass works by loading the entire buffer element content from memory, and translate it to logical type stored in a local variable at the earliest reference of a buffer handle. This means that is can generate inefficient code that reads more than necessary. Consider this example: ``` struct BigStruct { bool values[1024]; } ConstantBuffer cb; void test(BigStruct v) { if (v.values[0]) { printf("ok"); } } [numthreads(1,1,1)] void computeMain() { test(cb); } ``` In IR, the `computeMain` function before lower-buffer-element-type pass is something like following: ``` func test: %v = param : BigStruct %barr = fieldExtract(%v, "values") %element = elementExtract(%barr, 0) ... // uses %element func computeMain: %v = load(cb) call %test %v ``` The existing lower-buffer-element-type pass will rewrite the bool array in `BigStruct` into `int` array so it is legal in SPIRV. However, it does so by inserting the translation on the first `load` of the constant buffer: ``` struct BigStruct_std430 { int values[1024]; } var cb : ConstantBuffer; func computeMain: %tmpVar : var call %unpackStorage(%tmpVar, cb) %v : BigStruct = load %tmpVar call %test %v ``` This means that the entire array will be loaded and translated to int, before calling `test`, which only uses one element. It turns out that the downstream compiler isn't always able to optimize out this inefficient translation/copy. This PR completely rewrites the way buffer-element-type lowering is handled to avoid producing this inefficient code. It works in two parts: first we turn on the `transformParamsToConstRef` pass for SPIRV target as well, so we will translate the `test` function to take the `v` parameter as `constref`. The second part is a redesigned buffer-element-type pass that defers the storage-type to logical-type translation until a value is actually used by a `load` instruction. In this example, after `transformParamsToConstRef`, the IR is: ``` func test: %v = param : ConstRef %barr = fieldAddr(%v, "values") %elementPtr = elementAddr(%barr, 0) %element = load(%elementPtr) ... // uses %element func computeMain: call %test %cb ``` The new `buffer-element-type-lowering` pass will take this IR, and insert translation at latest possible time across the entire call graph, and translate the IR into: ``` func test: %v = param : ConstRef %barr = fieldAddr(%v, "values") %elementPtr : ptr = elementAddr(%barr, 0) %element_int = load(%elementPtr) %element = cast(%element_int) : %bool ... // uses %element func computeMain: call %test %cb ``` In this new IR, there is no longer a load and conversion of the entire array. See new comment in `slang-ir-lower-buffer-element-type.cpp` for more details of how the pass works. This PR also address many other issues surfaced by turning on `transformParamsToConstRef` pass on SPIRV backend. --------- Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

Split overloaded uses of RefType in front-end (#8427)

2025-09-23T01:20:13+00:00

Overview ======== This change is the start of an attempt to address how the Slang compiler codebase has ended up conflating two similar, but semantically distinct, concepts: * The long-standing notion of `ref` parameters (only allowed for use in the builtin modules), which are encoded using a wrapper `Type` in the AST as part of the representation of the parameters of a `FuncType`. * A recently-introduced notion of explicit reference types that mirror the built-in `Ptr` type, with a relationship comparable to that between pointer and reference types in C++. The change splits the `Ref` type in the core module into two distinct types, with one for each of the two use cases. Similarly, the `RefType` class in the compiler's AST is split into two distinct classes, to represent the two cases. Background ========== The `Ref` type in the core module (hidden and not intended for users to ever see or use) was originally introduced to encode the `ref` parameter-passing mode, comparable to the hidden `Out` and `InOut` types used to encode `out` and `inout` parameter-passing modes. The `Ref` type in the core module was encoded as a instance of the `RefType` class in the Slang AST (similar to how `Out` mapped to an `OutType`). These AST classes were *only* intended to be used by the compiler front-end as part of its encoding of function types. The `FuncType` class needed a way to distinguish an `inout int` parameter from a plain (implicitly `in`) `int` parameter, so these wrapper like `RefType` and `OutType` were introduced to encode both the parameter type (`T`) and the parameter-passing mode in a form that could be passed around as a `Type`. Notably, the `Ref` type (and `Out`, etc.) were *not* intended to be type names that ever get uttered in Slang code (not even in the builtin modules), and the vast majority of the compiler code was not supposed to ever encounter them. They were an implementation detail of `FuncType`, and nothing else. (In hindsight it may have been a mistake to use a nominal type declared in the core module to implement these wrappers; it might have been a good idea to use an entirely separate class of `Type` for this case...) Recent changes to the builtin modules introduced functions that wanted to *return* a reference (so that the parameter-passing-mode modifiers like `ref` could not trivially be used), and as part of those changes the appealingly-named `Ref` type in the core module was re-used for this new case. Builtin operations were declared with an explicit `Ref` return type, and parts of the compiler front-end that had previously been blissfully unaware of the AST's `RefType` (and `InOutType`, etc.) had to start accounting for the possibility that an explicit `Ref` would show up. Related changes also introduced a comparable conflation of the (unfortunately-named) `constref` parameter-passing modifier and builtin operations that wanted to return an explicit reference that is read-only. Both use cases were mapped to the core-module `ConstRef` type, which appeared in the AST as an instance of the `ConstRefType` class. The overlapping use of `ConstRef`` is actually significantly more troublesome than the `Ref` case because, despite what its name implies, `constref` was not really supposed to be the read-only analogue of `ref`, but rather it is closer to the "immutable value borrow" analogue to `inout`'s "mutable value borrow." The semantics of a "value borrow" vs. a "memory reference" in Slang have not been very carefully codified, and the conflation around `ConstRef` has contributed to things becoming increasingly muddy in the compiler back-end. Main Changes ============ Core Module ----------- The `Ref` type has been replaced with two distinct types, with one for each use case: * `RefParam` is intended for use when encoding a `ref` parameter in a function type * `ExplicitRef` is intended for use when an operation in a builtin module wants to return a reference The other types used to represent parameter-passing modes (e.g., `InOut`) were renamed to better indicate that their role in defining parameter types (e.g., `InOutParam`). The `ExplicitRef` type was given additional generic parameters for the allowed access and the address space, akin to what `Ptr` now supports. The pointer dereference operator (prefix `*`) in the core module should now properly propagate the access and address space of the pointer over to the reference that gets returned. The two distinct use cases of `ConstRef` were not split in the way as `Ref`, instead the case for the `constref` parameter-passing mode uses `ConstParamRef`, while cases that previously used `ConstRef` to represent a read-only explicit reference instead now use `ExplicitRef`. Prior to this change there were two subscripts declared on pointers: one in the `Ptr` type itself, and another in an `extension` for pointers with `Access.ReadWrite`. The comments on the code seemed to indicate that the catch-all subscript used to only have a `get` accessor, while the `ref` was only available on read-write pointers, but it seems that subsequent changes converted the default subscript to support `ref`. This change eliminates the subscript added via `extension`, since it is redundant. AST and Front-End ================= Similar to the changes in the core module, the AST `RefType` class was split into: * `RefParamType` for the case of encoding `ref` parameters * `ExplicitRefType` for the case where the user meant an explicit reference type All the other classes that represent wrappers for encoding parameter-passing modes (e.g., `OutType`) were similarly renamed (e.g., `OutParamType`). The `ConstRefType` class was simply renamed to `ConstRefParamType`, because any use cases of `ConstRefType` that intended an explicit reference type will now use `ExplicitRefType` with `Acccess.Read`. For convenience, this change includes type aliases to map the old names for these types over to the new ones (e.g., `using OutType = OutParamType`) so that the change doesn't need to affect quite so many lines of code. The `RefType` and `ConstRefType` names are intentionally left undefined, since it woudl be unsafe to assume that existing use sites should default to either of the two possible interpretations. All use cases of `RefType` and `ConstRefType` (and their former shared base class `RefTypeBase`) were audited and updated to refer to either `RefParamType`/`ConstRefParamType` or `ExplicitRefType`, as appropriate (based on whether the context of the code indicated it was working with parameter-passing mode wrapper types, or explicit reference types). In many (many) cases comments were added to the code that was updated (and some unrelated code that needed to be audited along the way) to note cases where there appears to be something fishy going on in the compiler and/or there are obvious opportunities for next-step improvement. The `QualType` constructor used to infer l-value-ness when passed a `RefType` or `ConstRefType`; that code was introduced to support explicit reference types. The code was updated to consult the access argument of an `ExplicitRefType` to try and determine the right l-value-ness to use. There is some ambiguity about what should be done in the case where the value of the generic argument representing the access cannot be statically determined; a better solution may be needed. Many other cases in the front-end that were working with `RefType` and `ConstRefType` for explicit references also need to figure out l-value-ness, and these were changed to rely on the logic already added to `QualType` so that it wouldn't have to be duplicated. It isn't clear if this structure is the best way to tackle the problem, but it seems to at least be an upgrade over the more strictly ad-hoc logic that was in place before. Future Work =========== IR-Level Work ------------- The most obvious next step to take is that the split that was made in the compiler front-end needs to be properly plumbed through all of the back-end. There appears to be a lot of code in the back end of the compiler that has made the same conflation of `ref` parameters and explicit reference types that the front-end did. In practice, any uses of `ExplicitRef` in the front-end should desugar into plain pointer-based code in the IR. Clean Up Parameter-Passing Modes -------------------------------- The code that handles different parameter-passing modes (`ParameterDirection`s) and their wrapper types is somewhat scattered and messy (as found while auditing use cases of `RefType`). A cleanup pass is warranted to ensure that most code only needs to think about `ParameterDirection`s. There should ideally be only a single operation in the front-end that handles determining the `ParameterDirection` of a parameter based on its modifiers. Similarly, there should be one operation to wrap a value type based on a parameter direction, and one operation to derive a `ParameterDirection` from the wrapper type. Ideally, the accessors for `FuncType` should not provide unrestricted access to the potentially-wrapped parameter types, and should instead return some kind of `ParamInfo` struct that encodes both a `ParameterDirection` and the unwrapped `Type` of the parameter. Clean Up `QualType` ------------------- A significant piece of future work that appears required is to drastically clean up and improve the way that `QualType`s are represente and handled in the front-end. There are currently various distinct `bool` flags in `QualType` (some with very unclear meaning) and differnet parts of the codebase consult/modify only subsets of them; a clear enumeration of the "value categories" (to use the C++ terminology) that Slang supports could be quite helpful. Naively, a `QualType` should at least encode the basic information that a `Ptr` type encodes: * A value type * Allowed access (read-only, read-write, etc.) * Address space The main additional thing that a `QualType` needs is a way to distinguish cases where an expression evaluates to: * A reference to a memory location, where all the information from a `Ptr` is relevant * A simple value, such that the access and address space are irrelevant * A reference to an abstract storage location (a `property`, `subscript`, or an implicit conversion that needs to support being an l-value), in which case address space is irrelevant and the "allowed access" basically amounts to a listing of the accessors the storage location supports Eliminate Explicit Reference Types ---------------------------------- Finally, twe should eventually eliminate the `ExplicitRef` type from the core module (and all of the supporting code from the front-end), since the feature is not a good fit for the Slang language. We should find some other way to decorate operations in the builtin module that need to returns a reference rather than a value (note how `ref` accessors already avoided exposing explicit reference types, by design). --------- Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

Added __magic_enum (#8436)

2025-09-17T12:46:27+00:00

Fixes #8406 (and #8410). `AddressSpace`, `MemoryScope` and `AccessQualifier` are no longer `BaseType`. I added a new `__magic_enum` (very similar to `__magic_type`) syntax to be able to easily create values or these enums from the compiler. (I don't know if it was the right way to do it, but it works and the changes are small enough?). I had a weird bug: `tests/language-feature/capability/address-of.slang` was failing in `IRBuilder::_findOrEmitConstant(IRConstant& keyInst)`. When needing a new `u64(0)`, it did not find it in the `ConstantMap` first, but then failed to add it right after because it already existed in the map! But this was triggered by `IRPtrType* IRBuilder::getPtrType(IROp op, IRType* valueType, AccessQualifier accessQualifier, AddressSpace addressSpace)`, which is a strange coincidence... but I could not find the issue in what I did. I ended up bumping unordered_dense, and it solved the issue (so there was a bug in there).

[CBP] Pointer frontend changes + groupshared pointer support (#7848)

2025-08-29T22:52:34+00:00

Resolves #7628 Resolves: #8197 Primary Goals: 1. Add `Access` to pointer 2. AddressSpace::GroupShared support for pointers (SPIR-V) 3. Add `__getAddress()` to replace `&` * `&` is not updated to `require(cpu)` since slangpy uses `&`. This means we must: (1) merge PR; (2) replace `&` with `__getAddress()`; (3) add `require(cpu)` to `&` Changes: * Added to `Ptr` the `Access` generic argument & logic (for `Access::Read`). * Moved the generic argument `AddressSpace` from `Ptr` to the end of the type. * Added pointer casting support between any `Ptr` as long as the `AddressSpace` is the same * Disallow globallycoherent T* and coherent T* * Disallow const T*, T const*, and const T* * Fixed .natvis display of `ConstantValue` `ValOperandNode` * Support generic resolution of type-casted integers * Added `VariablePointer` emitting for spirv + other minor logic needed for groupshared pointers Breaking Changes: * Anyone using the `AddressSpace` of `Ptr` will now have to account for the `Access` argument * we disallow various syntax paired with `Ptr` and `T*` --------- Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

Fix mesh shader OutputIndices subscript error by adding missing ref accessor (#7929)

2025-08-22T20:48:20+00:00

Fixes the Slang compiler internal error "subscript had no getter" when reading from mesh shader output index arrays (e.g., `triangles[0].x`). ## Problem The `OutputIndices` struct was missing a `ref` accessor in its `__subscript` implementation, causing the compiler to fail when trying to materialize subscript expressions as r-values. ## Solution Added the missing `ref` accessor to `OutputIndices.__subscript` using the `kIROp_MeshOutputRef` intrinsic operation, matching the pattern used in `OutputVertices` and `OutputPrimitives`. ## Files Changed - `source/slang/core.meta.slang` - Added missing `ref` accessor - `tests/bugs/gh-7925.slang` - Test case to reproduce and verify the fix Fixes #7925 Generated with [Claude Code](https://claude.ai/code) --------- Co-authored-by: Claude Co-authored-by: Lujin Wang Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: slangbot Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

Add matrix select intrinsic (#7566)

2025-07-31T22:12:21+00:00

* Add matrix select intrinsic * Fix hlsl test * Restrict matrix select to HLSL * Better test for HLSL side * Select route for GLSL/SPIRV * Exclude matrices from select legalization * Exclude CUDA from select test * Inline and move * format code --------- Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>