This adds SPIR-V assembly and WGSL tests derived from VK-GL-CTS commit
571256871c2e2f03995373e1e4a02958d8cd8cf5. The following procedure was
followed:
- Those .amber files in VK-GL-CTS wholly owned by Google were
identified
- All GLSL and SPIR-V shaders were extracted from the Amber files and
converted into SPIR-V binaries
- The compact-ids pass of spirv-opt was applied to each binary
- Duplicate binaries were removed
- spirv-opt -O was used to obtain an optimized version of each remaining
binary, with duplicates discarded
- Binaries that failed validation using spirv-val with target
environment SPIR-V 1.3 were discarded
- Those binaries that tint could not successfully convert into WGSL were
put aside for further investigation
- SPIR-V assembly versions of the remaining binaries are included in
this CL
- test-runner with -generate-expected and -generate-skip was used to
generate expected .spvasm, .msl, .hlsl and .wgsl outputs for these
SPIR-V assembly tests
- Each successfully-generated .expected.wgsl is included in this CL
again, as a WGLSL test
- test-runner with -generate-expected and -generate-skip was used again,
to generate expected outputs for these WGSL tests
Change-Id: Ibe9baf2729cf97e0b633db9a426f53362a5de540
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58842
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
For loop initializers and continuing statements do not have a BlockStatement as their parent.
Handle removal of these statements with a new Transform::RemoveStatement() helper
Fixed: tint:990
Change-Id: I24e7b18dcf71d3ef0a4d3ee68b9f68518e0eb5e8
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58063
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: James Price <jrprice@google.com>
CloneContext::Replace(T* what, T* with) is bug-prone, as complex transforms may want to clone `what` multiple times, or not at all. In both cases, this will likely result in an ICE as either the replacement will be reachable multiple times, or not at all.
The CTS test: webgpu:shader,execution,robust_access:linear_memory:storageClass="storage";storageMode="read_write";access="read";atomic=true;baseType="i32"
Was triggering this brokenness with DecomposeMemoryAccess's use of CloneContext::Replace(T*, T*).
Switch the usage of CloneContext::Replace(T*, T*) to the new function form.
As std::function is copyable, it cannot hold a captured std::unique_ptr.
This prevented the Replace() lambdas from capturing the necessary `BufferAccess` data, as this held a `std::unique_ptr<Offset>`.
To fix this, use a `BlockAllocator` for Offsets, and use raw pointers instead.
Because the function passed to Replace() is called just before the node is cloned, insertion of new functions will occur just before the currently evaluated module-scope entity.
This allows us to remove the "insert_after" arguments to LoadFunc(), StoreFunc(), and AtomicFunc().
We can also kill the icky InsertGlobal() and TypeDeclOf() helpers.
Bug: tint:993
Change-Id: I60972bc13a2fa819a163ee2671f61e82d0e68d2a
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58222
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: James Price <jrprice@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
This uses FXC compilation failure mitigation for _any_ vector index assignment that has a non-constant index. FXC can still fall over if the loop calls a function that performs the dynamic index.
Use some vector swizzle logic to avoid branches in the helper.
Fixed: tint:980
Change-Id: I2a759d88a7d884bc61b4631cf57feb4acc8178de
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/57882
Auto-Submit: Ben Clayton <bclayton@google.com>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
This change introduces sem::CompoundStatement, a new base class for
statements that can hold other statements.
sem::BlockStatements now derives from sem::CompoundStatement, and
this change introduces the following new CompoundStatements:
* `sem::IfStatement`
* `sem::ElseStatement`
* `sem::ForLoopStatement`
* `sem::LoopStatement`
* `sem::SwitchStatement`.
These new CompoundStatements are now inserted into the semantic
tree as now documented in `docs/compound_statements.md`.
The `sem::BlockStatement::FindFirstParent()` methods have been
moved down to `sem::Statement`.
The `Resolver::BlockScope()` method has been replaced with
`Resolver::Scope()` which now maintains the `current_statement_`,
`current_compound_statement_ ` and `current_block_`. This
simplifies statement nesting.
The most significant change in behavior is that statements now
always have a parent, so calling Block() on the initializer or
continuing of a for-loop statement will now return the
BlockStatement that holds the for-loop. Before this would
return nullptr.
Fixed: tint:979
Change-Id: I90e38fd719da2a281ed9210e975ab96171cb6842
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/57707
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: James Price <jrprice@google.com>
Use the new semantic constant value information to significantly reduce the complex indexing logic emitted for UBO accesses.
This will dramatically reduce the number of `for` loops that are decayed to `while` loops.
Change-Id: I1b0adb5edde2b4ed39c6beafc2e28106b86e0edd
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/57701
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: James Price <jrprice@google.com>
Move the bulk of the constant evaulation logic out of transform::FoldConstants and into Resolver and sem::Expression.
transform::FoldConstants now replace TypeConstructor nodes that have a constant value on the expression.
This is ground work to:
* Cleaning up the HLSL uniform buffer indexing, which is `/` and `%` arithmatic heavy
* Prepares us to handle `constexpr` when it lands in the spec
* Provide a centralized place to do constant evaluation, instead of the
having similar logic scattered around the codebase.
Change-Id: I3e2f542be692046a8d243b62a82556db519953e7
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/57426
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Reviewed-by: James Price <jrprice@google.com>
By generating a helper function for these, we can keep the atomic expression pre-statement-free. This can help prevent for-loops from being transformed into while loops, which can upset FXC.
We can't do the same for workgroup storage atomics, as the InterlockedXXX() methods have the workgroup-storage expression as the first argument, and I'm not aware of any way to make a user-declared parameter be `groupshared`.
Change-Id: I8669127a58dc9cae95ce316523029064b5c9b5fa
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/57462
Commit-Queue: James Price <jrprice@google.com>
Auto-Submit: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: James Price <jrprice@google.com>
And remove the u32 overload of frexp (it's not in the spec).
Brings the number of failing tint end to end tests for MSL down to 19/1098.
The WG still haven't found consensus on reworking these two intrinsics.
It's very likely that their signature will change so that they return a structure instead of returning a value and outputing another as a pointer.
Until the WG makes a decision, let's implement these according to the current spec.
Some overloads are still failing due to MSL missing overloads of the pointer parameter being in the `threadgroup` address space.
I'm holding off fixing these until we know what's happening with these intrinsics.
See also:
https://github.com/gpuweb/gpuweb/issues/1480https://github.com/gpuweb/gpuweb/issues/1846
Change-Id: Ib6764e6659d840db41bc65fed2b8b283d1056c3d
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/57421
Commit-Queue: Ben Clayton <bclayton@chromium.org>
Commit-Queue: Ben Clayton <bclayton@google.com>
Kokoro: Ben Clayton <bclayton@google.com>
Reviewed-by: James Price <jrprice@google.com>
