There's no good reason for this to be public.
Move it into the writers, and expose a 'disable_robustness' option to
turn it off. This can be expanded to hold more fine-grain control in the
future.
Change-Id: I6ea6e54a27b2ae0fbcba5fdf45539063045cc15a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/122203
Commit-Queue: Ben Clayton <bclayton@google.com>
Kokoro: Ben Clayton <bclayton@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Auto-Submit: Ben Clayton <bclayton@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
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>
This CL implements f16 in uniform and storage address space, allowing
using f16 types in uniform and storage buffers on all backends. Tint
uint tests and Dawn E2E tests are added to validate the f16 types work
as expected.
Bug: tint:1473, tint:1502
Change-Id: I15e3de1033d3727f2ea33f4657f682c5f13c2153
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/106320
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
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>
Attempting to paper over all the MSL standard library holes for packed_vector in the MSL writer added complexity to the writer, produced messy output, and didn't actually catch all the cases where casts were needed.
Add a new PackedVec3 transform that applies the packed_vector -> vec casts in a smarter, more precise way.
Fixed: tint:1534
Change-Id: I73ce7e5a62fbc9cb04e1093133070f5fb8965dce
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/107340
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: Ben Clayton <bclayton@chromium.org>
This CL removes the parsing of a `-` in front of numerics when lexed.
This will cause the number to become a UnaryOperator negation then the
number instead of a negative number.
Bug: tint:1679, tint:1141, tint:1570
Change-Id: I217c0ffcbe5be934c8d56bd83141b47ade83bc60
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/106463
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Similar to the handling of packed values in the arithmetic operators
the shift operators need to cast to the unpacked type before doing the
as_type casts.
Bug: tint:1542
Change-Id: I4289c45ab0a067ce122f61675fe5e251a83b6f8b
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/105720
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
In 104681 the vectors were cast to themselves to fixup an issue
with `packed_int`. That CL used an `as_type` which does a bit cast.
A `packed_int` can not be bitcast to an `int`. This CL changes to
a type cast, so instead of `as_type<int3>()` it does `int3()`.
Bug: tint:1677
Change-Id: I72218c06853e4e5ae1a0d34e2fc3e1ca597de993
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/104682
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Auto-Submit: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: Ben Clayton <bclayton@google.com>
This CL updates the internals to use AddressSpace instead of the old
StorageClass name.
Bug: tint:1404
Change-Id: Iecc208e839453437f4d630f65e0152206a52db7e
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/104420
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Auto-Submit: Dan Sinclair <dsinclair@chromium.org>
Currently in the MSL backend we cast int values to uint in order to get
the correct WGSL behaviour for over/under flow. This fails in the case
of host shareable buffers as they use `packed` types which need to get
cast to the non-packed version first.
Bug: tint:1677
Change-Id: I57b70abaa8ca614472a26d63f19c1aef2bd64668
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/103986
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Shuffle the transform orders to ensure that these are embedded in a structure before running the Std140 transform.
Add more end-to-end tests for these.
As pointed out in tint:1673, arrays of matrices are not correctly decomposed by the Std140 transform.
This will be addressed by a later change.
Bug: tint:1673
Change-Id: I47c93e458ff48578922d576819792e8ed3a5723c
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/102541
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Zhaoming Jiang <zhaoming.jiang@intel.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: David Neto <dneto@google.com>
Since GLSL ES does not support the offset= attribute, struct members
with explicit @align or @size attributes require adding explicit
padding members. This in turn requires rewriting any constructor
calls to initialize the new padding to zero, handled in the same
transform.
Note that this is currently overly-verbose, and will add padding where
GLSL doesn't technically need it (e.g., padding a vec3 out to 16 bytes).
Bug: tint:1415
Change-Id: Ia9ba513066a0e84f4c43247fcbbe02f5fadd6630
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/101720
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Stephen White <senorblanco@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Modify the AddSpirvBlockAttribute transform to fix top-level structure
access of uniform, storage and push-constant buffers for use in the
GLSL backend. The small change to the transform makes the transform
wrap host-sharable buffers, if they're also used as a
non-host-sharable structure. Also rename the transform to
AddBlockAttrbibute in order to reflect its wider applicability.
Change-Id: Ib2bf4ebf6bce72790791dbae9387032be765e4b9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/101061
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Stephen White <senorblanco@chromium.org>
This transform breaks up matNx2<f32> matrices used in uniform buffers
into column vectors, which fixes std140 layout rules.
Used by the SPIR-V and GLSL backends.
Re-enable tests that were disabled for these cases.
Bug: tint:1632
Change-Id: I596d016582b4189a0b413d762b3e7eabd3504b22
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/100907
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: Ben Clayton <bclayton@chromium.org>
This CL add missing type conversions for f16, especially for SPIRV
backend which require special handling. A transform,
VectorizeMatrixConversions, are also added for SPIRV to replace a matrix
conversion to a matrix construction with converted column vectors.
Unittests for the transform and SPIRV writer, and end-to-end tests for
all conversion rules are added.
Bug: tint:1473, tint:1502, chromium:1356215
Change-Id: Iaff125e5dd295d35c4ab74757eb56b642802a51a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/100483
Reviewed-by: David Neto <dneto@google.com>
Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
This CL changes the MSL emission for struct initializers to emit the
struct name first.
`const a = {.f=float3(1)}` becomes `const a = Normals{.f=float3(1)}`.
This fixes an issues where the initialization happens inside an array
which the downstream compiler rejected without the explicit struct
naming.
Bug: tint:1641
Change-Id: I948b9ca94f4b89eac6d5bbbaa615b3d71d50c737
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/98760
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Reviewed-by: Ben Clayton <bclayton@google.com>
This patch add DXC compile flag "-enable-16bit-types" and change profile
to SM6.2 when validating generated HLSL using DXC if f16 extension is
enabled in the WGSL program.
The patch add Tint end-to-end test cases for expressions using f16 type,
including constructor, binary operator, splat, zero-init, and others.
Testcases that use f16 types in uniform or storage buffer are SKIPped,
because such usage is not implemented yet.
Bug: tint:1473, tint:1502
Change-Id: I481ab3d12cbb822f11ef85ba807bca3f9770089b
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/96252
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@chromium.org>
Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
Change tint's `--fxc` flag to take the path of the FXC compiler DLL.
Have tint attempt to validate with both FXC and DXC if `--validate` is
passed.
Fix the 'dirsWithNoPassExpectations' logic which looks like it got
broken with the tint -> dawn merge. It also incorrectly applied
filepath.FromSlash() on windows.
Change-Id: I0f46aa5c21bc48a2abc48402c41f846aff4a8633
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/96800
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Commit-Queue: Ben Clayton <bclayton@chromium.org>
These were never part of the spec, and they were not correctly
implemented for all backends.
Fixed: tint:1597
Change-Id: If1a23f1619c61c53baae277f1cf37aee4460ab7b
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/95952
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Enable the parsing of 'const'.
Warn on use of module-scope 'let', and automatically replace with 'const'.
Fixed: tint:1580
Change-Id: I214aabca80686dc6b60ae21a7a57fbfb4898ea83
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/93786
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Module-scope 'let' is getting replaced by 'const'. For all backends,
'const' will be inlined into the place of usage, making most of these
tests produce no output (if replaced with 'const'). Instead switch to
emitting with 'var'.
Bug: tint:1580
Change-Id: Ied5ddf9cdb7fbd3cef8e7b0c6f4983748aaa3d07
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/94688
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Ben Clayton <bclayton@chromium.org>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
If a module-scope private variable is only referenced within a single
function, promote it to a function scope declaration instead of
passing it as a parameter. This reduces the number of a function
parameters that are needed in some cases.
Bug: tint:1509
Change-Id: I8951f6216bc7e4cf5abfda314bea1e9ed3ded560
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/94002
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Use the constant evaluated value instead of manually traversing
variables to find the literal value. This is a small step towards
supporting 'const' values for array sizes.
Also make our OOB-byte related error diagnostics consistent.
Bug: tint:1580
Change-Id: Idf9eb22cdbf69d750218c554e9f826c30458c6b9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/94600
Reviewed-by: David Neto <dneto@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Once 'const' is introduced, let will no longer resolve to a
creation-time constant value.
Add temporary code into each of the writers to prevent constants that
originate from a 'let' from being inlined. This will reduce the amount
of noise in later CLs.
Bug: tint:1580
Change-Id: Id3493a43ac09fe9f042ff2d517d04b2ae854d43e
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/94541
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: David Neto <dneto@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Ben Clayton <bclayton@chromium.org>
This patch fix the issue of identity matrix constructors in HLSL. This
patch also fix dawn e2e tests for identity matrix constructors and zero
matrix constructors.
Bug: tint:1596, tint:1545
Change-Id: I6c41eb299c1d5f89cf18720611f450abae26d3f4
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/94042
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
And remove the WrapArraysInStructs transform.
Wrapping arrays in structures becomes troublesome for `const` arrays, as
currently WGSL does not allow `const` structures.
MSL 2.0+ has a builtin array<> helper, but we're targetting MSL 1.2, so
we have to emit our own. Fortunately, it can be done with a few lines of
templated code.
This produces significantly cleaner output.
Change-Id: Ifc92ef21e09befa252a07c856c4b5afdc51cc2e4
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/94540
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Ben Clayton <bclayton@chromium.org>
Reviewed-by: David Neto <dneto@google.com>
https://github.com/gpuweb/gpuweb/issues/3056 proposes that creation-time
constant expressions should support arrays. We have verbal agreement to
update to spec to support const arrays.
Bug: tint:1580
Change-Id: If460f729384d700a1c2149c5b7dbe8613a960184
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/94330
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: David Neto <dneto@google.com>
Aside from more SPIR-V using OpConstantNull, this is a no-op change.
By refactoring these methods, they're easier to read, and contain less
lambda-heavy code.
Change-Id: I89c26b2b9f1cd0785d86fb3293f7cfda550bef2e
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/94331
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: Ben Clayton <bclayton@google.com>
Ben Clayton
James Price
Antonio Maiorano
Zhaoming Jiang
dan sinclair
Stephen White
Loko Kung