In order to preserve padding properly for MSL, we need to use its
packed_vec type for all vec3 types in storage buffers, not just struct
members. This commit includes a complete rewrite of the PackedVec3
transform to achieve this. The key details are:
* An internal `__packed_vec3<>` type was added, which corresponds to a
`type::Vector` with an additional flag to indicate that it will be
emitted as packed vector.
* The `PackedVec3` transform replaces all vec3 types used in
host-shareable address spaces with the internal `__packed_vec3`
type. This includes vec3 types that appear as the store type of a
pointer.
* When used as an array element, these `__packed_vec3` types are
wrapped in a struct that contains a single `__packed_vec3`
member. This allows us to add an `@align()` attribute that ensures
that `array<vec3<T>>` still has the correct array element stride.
* When the `vec3<T>` appears as a struct member in the input program,
we apply the `@align()` to that member to ensure that we do not
change its offset.
* Matrix types with three rows that are used in memory are replaced
with an array of columns, where each column uses a `__packed_vec3`
inside an aligned wrapper structure as above.
* Accesses to host-shareable memory that involve any of these types
invoke a "pack" or "unpack" helper function to convert them to the
equivalent type that uses `__packed_vec3` or a regular `vec3` as
required.
* The `chromium_internal_relaxed_uniform_layout` extension is used to
avoid issues where modifying a type in the uniform address space
triggers stricter layout validation rules.
Bug: tint:1571
Fixed: tint:1837
Change-Id: Idaf2da2f5bcb2be00c85ec657edfb614186476bb
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/121200
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Change the DecomposeMemoryAccess to behave more like the DirectVariableAccess transform, in that it'll inline the access of buffer variable into the load / store helper functions, instead of passing the array down.
This avoids large array copies observed with FXC, which can have *severe* performance costs.
Fixed: tint:1819
Change-Id: I52eb3f908813f72ab9da446743e24a2637158309
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/121460
Kokoro: Kokoro <noreply+kokoro@google.com>
Auto-Submit: Ben Clayton <bclayton@google.com>
Reviewed-by: James Price <jrprice@google.com>
Commit-Queue: James Price <jrprice@google.com>
When calling the generated helper, the column and row arguments were
swapped.
Improved the unit tests to actually show this, rather than passing in a
single value for both column and row.
Bug: tint:1824
Bug: tint:1333
Change-Id: I32a92dec5e594dabd9d8d2b08474c0d6f3645520
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/118420
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
This fixes edge-cases, like the condition expression being a type-cast,
which DXC apparently sees as a variable re-declaration. Example:
fn foo(x : f32) {
switch (i32(x)) {
default {
}
}
}
was emitted as HLSL:
void foo(float x) {
int(x);
do {
} while (false);
}
The `int(x)` is seen as a re-declaration of `x` by DXC.
We fix this by only emitted the condition expression if it has
side-effects (which currently means it contains a call expression).
Bug: tint:1820
Change-Id: I7e4320fa09ea2d634c9e324cb0b752b0ee7dcde9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/118161
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: James Price <jrprice@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
The HLSL `sign` method returns an `int` result (scalar or vector). The
WGSL `sign` expects the result to be the same type as the argument. This
CL injects a cast to the correct type after the `sign` call in the HLSL
generated source.
Bug: tint:1795
Change-Id: I51fed24b5b8b752b6b27fdfb5dd47eb803902793
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/116692
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
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>
Overriding the alignment to 1 would cause nested structures to be
incorrectly laid out. The fix: Don't override the alignment.
All struct layout validation works on the sem offsets, so none of this
has to change.
Bug: tint:1776
Change-Id: Ic01d45fb2790cd823ed9a55e336860ebdc351aea
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/112603
Kokoro: Ben Clayton <bclayton@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: Ben Clayton <bclayton@google.com>
Update CreateASTTypeFor() to handle a potential edge-case described in tint:1764.
We haven't seen this issue happen in production, nor can I find a way to trigger this with the tint executable, but try to handle this before we encounter a nasty bug.
Fixed: tint:1764
Change-Id: I496932955a6fdcbe26eacef8dcd04988f92545a1
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/111040
Auto-Submit: Ben Clayton <bclayton@google.com>
Reviewed-by: James Price <jrprice@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
There's a reason the overload of `ctx.Replace()` that takes a pointer to the replacement is deprecated - it doesn't play well when used as part of another replacement.
Switch to using the callback overload of Replace() to fix bad transform output.
Bug: tint:1386647
Change-Id: I94292eeb65d24d7b2446b16b8b4ad13bdd27965a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/111000
Auto-Submit: Ben Clayton <bclayton@google.com>
Commit-Queue: James Price <jrprice@google.com>
Reviewed-by: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Passing a dereferenced value from Hashmap::Find() directly into Hashmap::Add() is a potential cause of UAF, as the insertion may reallocate the map, invalidating the input reference.
I'll try to think of ways to make this foot-gun harder to do, but this CL fixes the immediate bug found by fuzzers.
Bug: chromium:1383755
Change-Id: I4f8b2fcb0745b008a47ef9947c330afb9ac4e78f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/110020
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: James Price <jrprice@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Discard statements no longer affect the behavior or uniformity
analysis. Update the resolver, validator, and several tests to reflect
this.
Some E2E tests were removed as they had loops that are now considered
to be infinite.
Use the DemoteToHelper transform to emulate the correct semantics on
platforms where discard is (or may) terminate the invocation in a
manner that would affect derivative operations.
We no longer need the UnwindDiscardFunctions transform for HLSL, which
already implements the correct semantics. However, we still run the
DemoteToHelper transform for the HLSL backend due to issues with FXC's
handling of discard statements (see crbug.com/tint/1118).
Fixed: tint:1723
Change-Id: Ib49ff187919ae81c4af8675e1b66acd57e2ff7d2
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/109003
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: James Price <jrprice@google.com>
Replace the ShouldRun() method with Apply() which will do the
transformation if it needs to be done, otherwise returns
'SkipTransform'.
This reduces a bunch of duplicated scanning between the old ShouldRun()
and Transform().
This change also adjusts code style to make the transforms more
consistent.
Change-Id: I9a6b10cb8b4ed62676b12ef30fb7764d363386c6
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/107681
Reviewed-by: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
This CL make transform AddBlockAttribute always try to wrap types used
by buffer variables into a struct, in order to generate valid GLSL code
for assigning one buffer struct variable to another buffer struct
variable.
Fixed: tint:1735
Change-Id: I009d8a9ca7ecea1dc0ad6164275c964a18acb33f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/108023
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
