Accept any type in the intrinsics definition, and then manually
validate that there are no atomics in the type. Add manual E2E tests
for composite types.
Use the BuiltinPolyfill transform to implement it for all backends.
Update the uniformity analysis with special-case tags for the builtin.
Fixed: tint:1780
Change-Id: I95786dff4df70a0b16ed1c53b853b5d0ec6bc501
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/114862
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: James Price <jrprice@google.com>
Kokoro: James Price <jrprice@google.com>
Start using Tint's ClampFragDepth transform in the Vulkan backend when
needed in order to correctly clamp @builtin(frag_depth) on Vulkan. Do
this by always reserving 8 bytes of push constant space to contain the
f32 min and max values from the last viewport command.
Reenables relevant CTS tests that were suppressed on Vulkan.
Bug: dawn:1125, dawn:1576, dawn:1616
Change-Id: I38f4f6c3c51c99b5e591a780fea9859537529534
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/105642
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Loko Kung <lokokung@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
The toggle D3D12ForceClearCopyableDepthStencilTextureOnCreation can be
disabled when the toggle D3D12Allocate2DTexturewithCopyDstAsCommittedResource
is enabled because with D3D12Allocate2DTexturewithCopyDstAsCommittedResource
all the depth stencil textures (can only be 2D textures) with CopyDst usage
have already been created with CreateCommittedResource() instead of
CreatePlacedResource(), thus the driver issue about creating depth stencil
texture on a dirty heap with CreatePlacedResource() won't be triggered.
Bug: dawn:1487, chromium:1237175
Change-Id: I872d4d95e6e05e1bcf9489b31a72e61f957de3e5
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/116129
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
For logical binary expressions that can be short-circuited, if the rhs
tree contained a mix of constant and runtime expressions, we would
erroneously mark the node as runtime, although some of its children were
resolved as kNotEvaluated. This would then fail during backend
generation.
This is a fork of 115820, addressing review comments, as amaiorano is OOO this week.
Bug: chromium:1403752
Bug: tint:1581
Change-Id: I18682c7fe1db092d280390881ff86b3c0db23e9b
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/116020
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Auto-Submit: Ben Clayton <bclayton@google.com>
This CL updates ir:Value to track the instructions which use a given
value. The instructions add their usage upon construction. This
necessitates making the values non-const in a lot of places as they get
changed by the instruction.
The `result` value is moved up to the base instruction class as it
should exist in all instructions.
Bug: tint:1718
Change-Id: Id7ab6e43d48caea502756d274dd6be2e1e4240f1
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/116141
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
This Cl adds a couple cleanups requested in 114202 as a followup.
Templates updated to have the EnableIf in the `template` block. The code
for `create` of a Splat or Composite is moved to a helper method.
Bug: tint:1718
Change-Id: Ib302d78633c6102cfbe17d63f0a4841ecf147472
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/116100
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Before this change, the variadic function template version of
WrapInFunction would be selected when passing a utils::VectorRef<const
ast::Statement*>, even though an overload exists for that type. The
reason is that during type deduction, the compiler will select templates
over non-templates in its overload set. The only way around this was to
avoid type-deduction by explicitly casting the argument to
utils::VectorRef<const ast::Statement*>.
This CL adds a CanWrapInStatement metafunction that evaluates to true if
the arg type is one that could be passed to
ProgramBuilder::WrapInStatement. This is used to SFINAE in the variadic
args version of WrapInFunction.
Change-Id: I8aa3d69e2ce7324fd60b1b2a5906a51d51b549a3
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/115502
Reviewed-by: David Neto <dneto@google.com>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
This change is a necessary to support workgroupUniformLoad in a
following patch. Otherwise, there is no change to the set of shaders
that are accepted or rejected by the analysis.
We now distinguish between uniformity requirements on the contents of
a pointer parameter versus the pointer value itself when generating
tags for function parameters.
Whilst processing an expression, if we see a sem::Load node we pass a
flag down through child expressions to indicate that we will be
loading from the result. When processing an identifier expression, we
can then select between adding an edge to the source of the
pointer/reference versus the contents of the root identifier that it
corresponds to.
Since the pointers passed to atomic builtins can be uniform, we
special-case them to capture the fact that their return value is
always considered non-uniform.
The arrayLength builtin no longer needs special-casing.
Added many tests to cover various cases that are now captured
differently in the graph. There are two cases that are disabled as
they require variable pointers to trigger the uniformity violation.
Bug: tint:1780
Change-Id: I03edb65f22a6ffb0e7daf8b2f590f5de898e6262
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/114861
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: James Price <jrprice@google.com>
Reviewed-by: David Neto <dneto@google.com>
Variables declared in for-loop initializers were not being tracked
properly across iterations as a check was wrongly determining them to
be declared inside the loop body.
Also fixes an issue where variables declared in for-loop initializers
were still considered to be in scope after loop exit.
Change-Id: I2ce3a519be45c8daba31bf00e8b2614f0bd6a2de
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/114364
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Auto-Submit: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: James Price <jrprice@google.com>
When useing Bitcast to or from a class type, gcc warns even if the type
is trivially copyable. Fixed this by static_asserting that both types
are trivially copyable, and casting the pointers to std::byte*.
Change-Id: Ibb420f2dcdd35cfb187d74983fa8ab9b50d10c85
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/115180
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
Reviewed-by: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
