In C++ argument evaluation order is undefined. MSVC and Clang evaluate these in different orders, leading to hilarity when writing tests that expect a deterministic ordering.
Pull out all the argument expressions to create() in the clone functions so a cloned program is deterministic in its ordering between compilers.
Change-Id: I8e2de31398960c480ce7ee1dfaac4f67652d2dbc
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/41544
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: dan sinclair <dsinclair@chromium.org>
Auto-Submit: Ben Clayton <bclayton@google.com>
Change 41302 correctly fixed up Module::Clone(), but this wasn't actually called by the CloneContext, as Module::Clone() returns a new Module, where as the CloneContext needs to clone into an existing Module.
Refactor the code so that this duplicated logic is moved into a single Module::Copy() method.
Fixed: 1177275
Change-Id: Ia8c45ef05e03b2891b5785ee6f425dd01cb989c6
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/41542
Reviewed-by: James Price <jrprice@google.com>
Reviewed-by: dan sinclair <dsinclair@chromium.org>
Commit-Queue: dan sinclair <dsinclair@chromium.org>
And add tests for IntrinsicTable.
Drop all the type unwrapping - be precise:
* Display the actual argument types in the signature mismatch message
* Only dereference pointer arguments if the parameter does not expect a pointer
Correctly match access control on storage types
Note that I was mistaken in tint:486 - the TypeDeterminer is resolving identifiers to variables correctly as pointer types. The confustion here was probably due to all the UnwrapAll() calls, which have now all gone.
Fixed: tint:486
Change-Id: I239eabd1fedfc082566c4af616ccfc58786cae25
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/41280
Commit-Queue: Ben Clayton <bclayton@google.com>
Auto-Submit: Ben Clayton <bclayton@google.com>
Reviewed-by: dan sinclair <dsinclair@chromium.org>
Instead of emitting all global variables and then functions, emit
global declarations in the order they were added to the AST.
This fixes issues where the reording might generate an invalid WGSL
program from a valid input (e.g. when declaring a global variable with
the same name as a variable inside a function that precedes it).
This also unifies the implementation of Generate() and
GenerateEntryPoint(), to avoid implementing the same logic twice.
Change-Id: I60a4e5ed4a054562cdcc3d028f8d577434a6d713
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/41303
Reviewed-by: dan sinclair <dsinclair@chromium.org>
Commit-Queue: dan sinclair <dsinclair@chromium.org>
Instead of validating all global variables and then functions,
validate global declarations in the order they were added to the AST.
This fixes false-positive "redeclared identifier" errors when a global
variable is declared after a function that declares a variable of the
same name, and false-negative "identifier not declared" errors when a
global variable is declared after a function that tries to use it.
Change-Id: Ibf5e5265bc2f8ca892096f0420757b70e1984525
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/41302
Commit-Queue: James Price <jrprice@google.com>
Reviewed-by: dan sinclair <dsinclair@chromium.org>
Adds a vector<CastableBase*> to ast::Module which stores the list of
global variables, functions, and types, in the order that they were
declared.
This will be used to fix validation and backend issues around name
uniqueness.
Change-Id: I14491f6ebc0fc7341bd3fb3b3f408faa234a91f7
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/41301
Commit-Queue: James Price <jrprice@google.com>
Reviewed-by: dan sinclair <dsinclair@chromium.org>
While traversing and resolving the AST, TypeOf() was called to look up
the semantic node for the given AST expression in order to fetch the
resolved type. However, for CallExpression semantic nodes are
constructed at the end of the AST traversal, and GetType() for these
would unexpectedly return nullptr, causing a crash.
To fix, have TypeDeterminer maintain an internal map of ast::Expression
to resolved type. Always populate this internal map whenever SetType() is
called. At the end of the AST traversal, have CreateSemanticNodes()
construct the semantic nodes for any ast::Expression nodes that do not
already have a semantic node assigned.
With this, GetType() will always return the type set with SetType().
Fixes tint -> dawn autoroller.
Fixed: tint:488
Change-Id: I2830c496d9b2e4807ec01ed69aeafb3912f4a890
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/40606
Reviewed-by: dan sinclair <dsinclair@chromium.org>
Commit-Queue: Ben Clayton <bclayton@google.com>
Auto-Submit: Ben Clayton <bclayton@google.com>
Use the new semantic::Intrinsic::Parameters() information to determine whether a parameter is a pointer.
Don't generate a load if the parameter expects a pointer.
Fixed: tint:361
Change-Id: I1420a6b0e22d52f67a5e52151fb073ac33df5bd5
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/40508
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: dan sinclair <dsinclair@chromium.org>
Provides a centeralized table for all intrinsic overloads.
IntrinsicTable::Lookup() takes the intrinsic type and list of arguments, returning either the matched overload, or a sensible error message.
The validator has expectations that the TypeDeterminer resolves the return type of an intrinsic call, even when the signature doesn't match. To handle this, create semantic::Intrinsic nodes even when the overload fails to match. A significant portion of the Validator's logic for handling intrinsics can be removed (future change).
There are a number of benefits to migrating the TypeDeterminer and Validator over to the IntrinsicTable:
* There's far less intrininsic-bespoke code to maintain (no more duplicate `kIntrinsicData` tables in TypeDeterminer and Validator).
* Adding or adjusting an intrinsic overload involves adding or adjusting a single Register() line.
* Error messages give helpful suggestions for related overloads when given incorrect arguments.
* Error messages are consistent for all intrinsics.
* Error messages are far more understandable than those produced by the TypeDeterminer.
* Further improvements on the error messages produced by the IntrinsicTable will benefit _all_ the intrinsics and their overloads.
* The IntrinsicTable generates correct parameter information, including whether parameters are pointers or not.
* The IntrinsicTable will help with implementing autocomplete for a language server
Change-Id: I4bfa88533396b0b372aef41a62fe47b738531aed
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/40504
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: dan sinclair <dsinclair@chromium.org>
semantic::Intrinsic derives from semantic::CallTarget, which can be obtained from the Target() accessor on the CallExpression.
Flesh out semantic::Parameter to contain a `Usage` - extra metadata for the parameter.
The information in `Intrinsic` is enough to remove the `semantic::IntrinsicCall` and `semantic::TextureIntrinsicCall` types.
Change-Id: Ida9c193674ad8605d8f12f6a1d27f38c7d008434
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/40503
Reviewed-by: dan sinclair <dsinclair@chromium.org>
Commit-Queue: Ben Clayton <bclayton@google.com>
It previously said:
"a struct containing a runtime-sized array must be in the 'storage' storage class"
This had be looking at the variable's storage class, when the error was actually trying to tell me I needed `[[block]]` on the struct declaration.
Change-Id: I7a23a0c0c35508bdac20c808d2635592638dfa77
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/40602
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: Sarah Mashayekhi <sarahmashay@google.com>
textureLoad():
* Does not operate on cube textures
* Does not have a `level` parameter for storage textures
* Requires an `array_index` argument for arrayed texture types
textureSample():
* Only supports f32 data types for sampled textures
Bug: tint:449
Change-Id: I30b3a0c32245d5e6c4b1f3aeef112a67ffb1d055
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/40506
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: Ryan Harrison <rharrison@chromium.org>
Reviewed-by: dan sinclair <dsinclair@chromium.org>
For consistency with the other .cc file names in this directory.
The reason we need the sem_ prefix is because all the tint lib code is compiled as a single target, and .GN cannot cope with .cc files with the same file name, even if they're in different directories.
This could be fixed by building each directory as a separate target, but we have circular dependencies that currently make this impossible.
Change-Id: I2f1379ff11dd863f81662449f73d67832f0a8a4e
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/40501
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: dan sinclair <dsinclair@chromium.org>
CallTarget holds parameter information. This is simple to extract from an ast::Function.
CallTarget will also be used for intrinsics, which can be overloaded. CallTarget will hold the resolved overload parameter signature.
Bug: tint:361
Change-Id: I4dadc4a99293f12ede9e9cbd9132ba5f9b9830ed
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/40284
Reviewed-by: dan sinclair <dsinclair@chromium.org>
Commit-Queue: Ben Clayton <bclayton@google.com>
