Place the `#[track_caller]` attribute on the `check` and
`validation_error` functions in the Naga `wgsl_errors` test module, so
that panics or assertion failures will point to the actual test,
not to the code in the utility function.
Makes the dual source implementation in wgpu WebGPU spec compliant.
Furthermore, makes the dual source blending extension available when targeting WebGPU.
This allows abstract-typed expressions to be used for some or all of
the switch selector and case selectors. If these are all not
convertible to the same concrete scalar integer type we return an
error. If all the selector expressions are abstract then they are
concretized to i32.
The note previously provided by the relevant error message, suggesting
adding or removing the `u` suffix from case values, has been
removed. While useful for simple literal values, it was comically
incorrect for more complex case expressions. The error message should
still be useful enough to allow the user to easily identify the
problem.
When the user provides values for a module's overrides, rather than
replacing override-sized array types with ordinary array types (which
could require adjusting type handles throughout the module), instead
edit all overrides to have initializers that are fully-evaluated
constant expressions. Then, change all backends to handle
override-sized arrays by retrieving their overrides' values.
For arrays whose sizes are override expressions, not simple references
to a specific override's value, let front ends built array types that
refer to anonymous overrides whose initializers are the necessary
expression.
This means that all arrays whose sizes are override expressions are
references to some `Override`. Remove `naga::PendingArraySize`, and
let `ArraySize::Pending` hold a `Handle<Override>` in all cases.
Expand `tests/gpu-tests/shader/array_size_overrides.rs` to include the
test case that motivated this approach.
Simplify `naga::back::spv::LocalType` to have only one variant for
representing pointers, and use the SPIR-V type as the base type, so
that `LocalType` values are equal when the SPIR-V pointer types they
map to would be equal as well.
This avoids emitting multiple `OpTypePointer` instructions, which
SPIR-V would interpret as distinct types, for equivalent Naga IR
pointer types.
Change the `get_pointer_type_id` utility function accordingly, and
adjust its callers.
Change `LocalType::from_inner` into `Writer::localtype_from_inner`, so
that it can call `Writer` methods to build base types. Adjust users.
Instead allow the const to be converted each time it is const
evaluated as part of another expression. This allows an abstract const
to be used as a different type depending on the context.
As a result, abstract types may now find their way in to the IR, which
we don't want. This occurs because the compact pass treats named
expressions as used, mostly so that our snapshot tests are more
useful, and therefore does not remove them. To prevent this, we avoid
adding abstract-typed local consts to the named expressions list. This
will have no functional effect on any shaders produced by the
backends, but some unused local const declarations will no longer be
present.
Naga currently trips up on shaders containing a bitshift where the
left operand is an AbstractInt and the right operand is *not* a const
expression. This is due to the fact that resulting type of
`AbstractInt << u32` is abstract, and we are unable to evaluate it
away during const evaluation. This results in abstract-typed
expressions in our IR, which causes various issues.
For other binary operations, the type of the left and right operands
would be related, therefore the presence of a concrete right operand
would tell us which type to convert the left operand to. But for shift
operations the right operand's scalar type is always a u32, and does
not influence the left operand's type.
Luckily the WGSL spec[1] tells us how to avoid this problem:
6.1.3. Overload Resolution
2. Eliminate any candidate where one of its subexpressions resolves
to an abstract type after feasible automatic conversions, but
another of the candidate’s subexpressions is not a
const-expression.
This, for example, specifies we eliminate the `AbstractInt << u32:
AbstractInt` overload candidate, leaving `i32 << u32: i32` as the
remaining candidate with the lowest conversion rank. The equivalent
also applies for right-bitshifts, and for `vecN<AbstractInt>`
operands.
[1] https://www.w3.org/TR/WGSL/#overload-resolution-section
When an entry point's return type is anonymous, have
`naga::proc::Namer` assign it a name based on its shader stage.
Remove bespoke logic for this from the WGSL backend.
Fixes#7263.
Parsing currently fails for shaders that attempt to dynamically index
an abstract-typed array (or vector, etc), like so:
var x = array(1, 2, 3)[i];
This is caused by attempting to concretize the Expression::Access
expression, which the ConstantEvaluator fails to do so due to the
presence of a non-constant expression.
To solve this, this patch concretizes the base type *prior* to
indexing it (for non-constant indices), meaning the constant evaluator
never sees any non-constant expressions. This matches the WGSL
specification:
When an abstract array value e is indexed by an expression that is
not a const-expression, then the array is concretized before the
index is applied.
(Similar applies for both vectors and matrices, too.)
This may be somewhat non-optimal in that if there are multiple
accesses of the same abstract expression, we will produce duplicated
concretized versions of that expression. This seems unlikely to be a
major issue in practice, and we can always improve this if and when we
encounter a real issue caused by it.
When spilling arrays and matrices so that we can access them with an
index computed at runtime, if we need to spill the same expression again,
don't wipe out our record of the temporary variable we used the first
time. Just re-use it.
Fixes#7048.
If it is undefined behaviour for loops to be infinite, then, when
encountering an infinite loop, downstream compilers are able to make
certain optimizations that may be unsafe. For example, omitting bounds
checks. To prevent this, we must ensure that any loops emitted by our
backends are provably bounded. We already do this for both the MSL and
HLSL backends. This patch makes us do so for SPIRV as well.
The construct used is the same as for HLSL and MSL backends: use a
vec2<u32> to emulate a 64-bit counter, which is incremented every
iteration and breaks after 2^64 iterations.
While the implementation is fairly verbose for the SPIRV backend, the
logic is simple enough. The one point of note is that SPIRV requires
`OpVariable` instructions with a `Function` storage class to be
located at the start of the first block of the function. We therefore
remember the IDs generated for each loop counter variable in a
function whilst generating the function body's code. The instructions
to declare these variables are then emitted in `Function::to_words()`
prior to emitting the function's body.
As this may negatively impact shader performance, this workaround can
be disabled using the same mechanism as for other backends: eg calling
Device::create_shader_module_trusted() and setting the
ShaderRuntimeChecks::force_loop_bounding flag to false.
Instead allow the const to be converted and each time it is const
evaluated as part of another expression. This allows an abstract const
to be used as a different type depending on the context.
A consequence of this is that abstract types may now find their way to
the validation stage, which we don't want. We therefore additionally
now ensure that the compact pass removes global constants of abstract
types. This will have no *functional* effect on shaders generated by
the backends, as the expressions belonging to the abstract consts in
the IR will not actually be used, as any usage in the input shader
will have been const-evaluated away. Certain unused const declarations
will now be removed, however, as can be seen by the effect on the
snapshot outputs.
Integer division or modulo is undefined behaviour in SPIR-V when the
divisor is zero, or when the dividend is the most negative number
representable by the result type and the divisor is negative one.
This patch makes us avoid this undefined behaviour and instead ensures
we adhere to the WGSL spec in these cases: for divisions the
expression evaluates to the value of the dividend, and for modulos the
expression evaluates to zero.
Similarily to how we handle these cases for the MSL and HLSL backends,
prior to emitting each function we emit code for any helper functions
required by that function's expressions. In this case that is helper
functions for integer division and modulo. Then, when emitting the
actual function's body, if we encounter an expression which needs
wrapped we instead emit a function call to the helper.
Emit helper functions for MathFunction::Abs and UnaryOperator::Negate
with a signed integer scalar or vector operand. And for
BinaryOperator::Divide and BinaryOperator::Modulo with signed or
unsigned integer scalar or vector operands.
Abs and Negate are written to avoid signed integer overflow when the
operand equals INT_MIN. This is achieved by bitcasting the value to
unsigned, using the negation operator, then bitcasting the result back
to signed. As HLSL's bitcast functions asint() and asuint() only work
for 32-bit types, we only use this workaround in such cases.
Division and Modulo avoid undefined behaviour for INT_MIN / -1 and
divide-by-zero by using 1 for the divisor instead. Additionally we
avoid undefined behaviour when using the modulo operator on operands
of mixed signedness by using the equation from the WGSL spec, using
division, subtraction and multiplication, rather than HLSL's modulus
operator.
Though not explicitly specified one way or the other, we have been
informed by DirectX engineers that signed integer overflow may be
undefined behaviour in some cases. To avoid this, we therefore bitcast
signed operands to unsigned prior to performing addition, subtraction,
or multiplication, then bitcast the result back to signed. As signed
types are represented as two's complement, this gives the correct
result whilst avoid any potential undefined behaviour.
Unfortunately HLSL's bitcast functions asint() and asuint() only work
for the 32-bit int and uint types. We therefore only apply this
workaround for 32-bit signed arithmetic. Support for other bit widths
could be added in the future, but extra care must be taken when
converting from unsigned to signed to avoid undefined or
implemented-defined behaviour.
There is no literal suffix in HLSL for the integer type. We can,
however, wrap integer literals in a constructor style cast. This
avoids ambiguity when passing literals to overloaded functions, which
we'll make use of in the subsequent patch in this PR.
Adds infrastructure to the MSL backend for emitting helper functions,
based upon the existing HLSL backend equivalent. Emit helper functions
for MathFunction::Abs and UnaryOperator::Negate with a signed integer
scalar or vector operand. And for BinaryOperator::Divide and
BinaryOperator::Modulo with signed or unsigned integer scalar or
vector operands.
Abs and Negate are written to avoid signed integer overflow when the
operand equals INT_MIN. This is achieved by bitcasting the value to
unsigned and using the negation operator, then bitcasting the result
back to signed.
Division and Modulo avoid undefined bevaviour for INT_MIN / -1 and
divide-by-zero by using 1 for the divisor instead. Additionally we
avoid undefined behaviour when using the modulo operator when either
or both operands are negative by using the equation from the WGSL
spec, using division, subtraction and multiplication, rather than
MSL's modulus operator.
Lastly, as the usage of the wrapper function for unary integer
negation makes the negation_avoids_prefix_decrement() testcase less
interesting, we extend it to additionally test negating floats.
Attempting to convert a non-abstract type will always fail, which can
result in unrelated errors being misreported as type conversion
errors. For example, in #7035 we made it so that abstract types can be
used as return values. This changed the final testcase in the
invalid_functions() test to fail due to failing to convert a u32 to an
atomic<u32>, rather than with a NonConstructibleReturnType error.
This patch makes it so that we do not attempt to convert non-abstract
types in the first place. This avoids the AutoConversion error for
that testcase, meaning the NonConstructibleReturnType is once again
reported.
Various callsites of try_automatic_conversions() have been updated to
ensure they still return an InitializationTypeMismatch error if the
types are mismatched, even if try_automatic_conversions() succeeds (eg
when conversion was not attempted due to the type being concrete). To
reduce code duplication these callsites were all adapted to use the
existing type_and_init() helper function.
Lastly, a couple of tests that expected to result in a AutoConversion
error have been adjusted to ensure this still occurs, by ensuring the
type used is abstract.
When converting the underlying scalar type of an array during const
evaluation, we currently use the resulting base type's size as the array
stride for the resulting type. For certain types, this may not match the
required alignment and will therefore result in a validation error.
For example, `array<vec3<f32>, N>` should have a stride of 16. But if
declared with an abstract initializer, eg `array(vec3(0.0))` we will
incorrectly determine the stride to be 12.
To solve this, we use the proc::Layouter struct to determine the
required array stride during const evaluation. To avoid repeating
layouting work, we reuse the Lowerer's layouter, passing it through the
various *Contexts through to the ConstantEvaluator.
When lowering a return statement, call expression_for_abstract()
rather than expression() to avoid concretizing the return value. Then,
if the function has a return type, call try_automatic_conversions() to
attempt to convert our return value to the correct type.
This has the unfortunate side effect that some errors that would have
been caught by the validator are instead encountered as conversion
errors by the parser. This may result in a slightly less descriptive
error message in some cases. (See the change to the invalid_functions()
test, for example.)
