And add basic support for builtins returning structures.
Bug tint:1581
Change-Id: I67f987339b9a344e1915c69c9991803f0665305d
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/111242
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
With this CL, binary ops add, subtract, multiply, and divide of concrete
values will now produce an error if the result is inf/NaN, as it was
doing with abstract values. This also affects the cross builtin, which
is written in terms of subtract and multiply.
Bug: tint:1581
Bug: tint:1747
Change-Id: Ib1d0d8deddc82c67ab53729a6011937636fcc1a5
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/110163
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
This is the same as was done for const eval builtin tests, allowing the
error cases to be defined in the same place as we define the positive
cases.
Also got rid fo the 'overflow' flag, which was used to skip abstract
cases in the unit test. Instead, I modified the tests to only add
overflow cases if not abstract.
This change will make it easier to update tests when we make Inf/NaN
failures for concrete float operations.
Bug: tint:1581
Bug: tint:1747
Change-Id: I7e5d8f9b24ca486aaa03a3b1bd07ccedb09411c9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/110043
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Also:
Print unrepresentable numbers with higher precision - otherwise values can round, and diagnostics can be very confusing.
Improve diagnostic distinction between `( )` `[ ]` interval ranges.
Change-Id: I9269fbf1738f0bce5f2ddb5a387687543fd5d0bb
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/108700
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Rules for shift left of concrete values are now split between signed and
unsigned. Shifting unsigned values no longer fails with "sign change"
errors. Furthermore, shifting unsigned values must only shift out 0s.
See https://github.com/gpuweb/gpuweb/pull/3539.
Bug: tint:1701
Bug: tint:1717
Change-Id: Iba2799f4b02cdc77cc58a6c7c104aaa408f0f0f9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/106381
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
googletest doesn't print test parameters properly when they are a tuple
of variants.
Change-Id: I070697bb0118282dd4841df23c936e0171934628
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/106380
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
In WGSL, we can shift left abstracts by >= 64, as long as the result is
representable in the data type we choose for it. When shifting 0, we can
shift by any positive u32 value (result is always 0), but in C++, it's
UB to shift by more than the bit width of the data type, so we need to
handle this. This bug was caught by ClusterFuzz.
Bug: chromium:1372963
Change-Id: I638ca190b93538908ca6472f3735627ea8531c5a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/106266
Reviewed-by: David Neto <dneto@google.com>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
The reason for slow compile times is because the very large variants of
builder::Value<T>s combined with the many std::visits over these
variants result in many combinatorial instantiations of the visit
callbacks.
To address this, I added a polymorphic base class ValueBase to Value<T>,
and replaced most of the std::visit-based compile time code with runtime
virtual calls. For the two heaviest users of std::visit over the large
variants, compiles times dropped more than half (clang-10, debug):
const_eval_binary_op_test.cc: 19.079s to 7.736s
const_eval_unary_op_test.cc: 10.021s to 4.789s
Bug: tint:1711
Change-Id: Iba05e6ae1004ef0814250e2a8ea50aa2b26b85f2
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/105782
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
The single file was taking too long to compile, and was becoming
difficult to work in.
Bug: tint:1711
Change-Id: Ibaaa5dd72aef02cdffe80156848d010ff84c9553
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/105740
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
