These sources cannot build without the IR being explicitly enabled, as
they depend on code in headers that would be `#ifdef`'d out.
Change-Id: Id542f41679c3e6e6f210ca632e85ca796e61a746
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133402
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Auto-Submit: James Price <jrprice@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: James Price <jrprice@google.com>
Commit-Queue: Austin Eng <enga@chromium.org>
Kokoro: James Price <jrprice@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Also add `_ir` to the names of test files.This makes it less likely to
confuse IR code with non-IR code.
Bug: tint:1906
Change-Id: I2db85440c815c2834d7bab1899afa0fafc89ce6a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133222
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
When converting an AST expression to IR, check for a sem::Load node
and emit a load instruction if present.
Update conversion of compound assignment and increment/decrement to
load from the LHS.
Convert load instructions to inline variable references when going
back to the AST.
Bug: tint:1718
Change-Id: Ib2b850efb304a71eff95aadac825f015623b6eb3
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133220
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: James Price <jrprice@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Change the type of a `var` to a pointer. Fold away address-of and
indirection.
Fixed: tint:1912
Change-Id: Ib1f07538c63df9349c5b6171062b6f79750c1439
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133400
Commit-Queue: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Adds support for both `If` and `Block` flow nodes as branch
targets. Also support a nullptr branch target by emitting
OpUnreachable.
Bug: tint:1906
Change-Id: I1adea83ce6c7c85c6a2e2dae9327499cb7f850bd
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132861
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: James Price <jrprice@google.com>
Use it in the SPIR-V writer.
Bug: tint:1718, tint:1906
Change-Id: If8f29300712c457a02ddc9eb2fd76e0b49ee8fea
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132682
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Enable the transform manager to run a pipeline that mixes AST and IR
transforms, automatically converting the current program as necessary.
Bug: tint:1718
Change-Id: I8df76db61edd94e0b1d7c2aaabc18b394db3d8de
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132502
Commit-Queue: James Price <jrprice@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: James Price <jrprice@google.com>
The RHS of a compound assignment statement may need to be
materialized. This was showing up when converting things like `i += 1`
to IR, as abstract types were creeping into the IR.
Change-Id: Idf9b1523d1751e26c28a795af07769ca85a65f14
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133221
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: James Price <jrprice@google.com>
This CL adds phony assignment to the IR. The assignment part is ignored
and the RHS of the expression is generated. This creates a result value
which is then never used.
Bug: tint:1918
Change-Id: Ic87fdcb387cb4d9783c4dbbe26ebc76f67a3cdd9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133260
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
The Instructions are now inheriting from Value so they can use the value
block allocator. Remove the instruction allocator.
Bug: tint:1718
Change-Id: Ia8f3b8b106d03883b3c72ebead392ae972a86c4b
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133164
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: dan sinclair <dsinclair@google.com>
This CL adds the ability to add parameters to blocks and emits the
params as needed in the disassembly.
The ShortCircuit is updated to use block arguments instead of creating
a return value.
Bug: tint:1909
Change-Id: I92afa6cf8ff4e01bfa3de46e76c26c465f0d6062
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133200
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Reviewed-by: Ben Clayton <bclayton@google.com>
The address space and access mode both exist on the type of the Var,
storing the information into Var is redundant. This CL removes them from
var in favour of the type fields.
Bug: tint:1718
Change-Id: I0198a8794d6359834592562251b2788e8a0347ca
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133163
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
And implement functions with return values.
Bug: tint:1902
Change-Id: Id4015aa83bf75de2a0f3dfdbfe19f728c05226c8
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133142
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: Ben Clayton <bclayton@google.com>
Kokoro: Ben Clayton <bclayton@google.com>
Returns with values requires fleshing out of functions, which comes next
Bug: tint:1902
Change-Id: I5f956805441b99038f2d48758d1bd767ea4e1a1d
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133141
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Kokoro: Ben Clayton <bclayton@google.com>
Traverse the block nodes, and if statements.
Bug: tint:1902
Change-Id: Ie5533acdc65378bfea91b46a62090c4d3216b303
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133100
Commit-Queue: Ben Clayton <bclayton@google.com>
Kokoro: Ben Clayton <bclayton@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
This CL adds increment and decrement statement support into the IR.
Bug: tint:1718
Change-Id: Ieba2ab7a4c9232dc9d76a8605637cd5ea46b0e08
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133162
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: James Price <jrprice@google.com>
This will allow for both AST transforms and IR transforms to be used
in the same transform manager.
The transform manager is no longer a Tranform subclass.
Bug: tint:1718
Change-Id: I59bd87806090fa365ce4b575710c5ffcfc657bd8
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132501
Commit-Queue: James Price <jrprice@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
This is the first commit to implementing an IR -> AST conversion path.
Not much currently implemented, but enough to start building upon.
Started implementing AST -> IR -> AST roundtrip tests.
Bug: tint:1902
Change-Id: I17fa9f3bc609b48ecb916ab47ae61bfa36821b44
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/131180
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
This CL updates the disassembly output to make it clearer when
instructions are flow node traversals, and when they're actual
instructions in a block.
Bug: tint:1718
Change-Id: I95e103a5672b5579081b73e9633bf07b31e51187
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133161
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
The scope stack for a loop is slightly different from other constructs
as the block for the body and the block for the continuing need to share
the scope. This CL fixes the loop conversion to special case the scope
stack for the loop body so the continuing block can see the variables.
Bug: tint:1718
Change-Id: I2e4898b36f1541b48a4e349955833b155332947a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133160
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
This is unneeded, as we can now determine if a function is an entry
point by checking its pipeline stage.
Bug: tint:1718
Change-Id: Icb09b39b7a63cf9a9a36ccad5e144eceec1dfdeb
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/131521
Commit-Queue: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
This CL adds the pipeline stage and workgroup_size as optional
parameters when creating a function in the IR.
Bug: tint:1718
Change-Id: Iae65dcb9557a644a17ec67fc5269d0c2db3f8aba
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133001
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Reviewed-by: Ben Clayton <bclayton@google.com>
This CL updates the SPIRV IR generator tests to use the `Branch` method
of the builder.
Bug: tint:1718
Change-Id: I16961382b797dc90277529f706ee240be282d623
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/133000
Reviewed-by: James Price <jrprice@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Kokoro: James Price <jrprice@google.com>
Black-box the BuilderImpl by putting the BuilderImpl into the
from_program.cc file.
Why:
* It means there's only a single definition of all the methods that need
to be maintained, instead of pointlessly splitting code between two
files (.cc / .h).
* It removes all the implementation details from the header.
* It removes a whole bunch of transitive includes, slowing compiles.
* It prevents the temptation for future #includes to
private-implementation details.
* It reduces the amount of symbols declared outside of anonymous
namespaces, reducing symbol polution and the amount of work on the
linker.
Bug: tint:1718
Change-Id: I82838089f784ab003dae4ef06545bba1ca2401cc
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132321
Commit-Queue: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
For the `if`, `switch` and `function` flow nodes this Cl makes a few
required fields part of the constructor. This cuts down boilerplate when
creating new nodes.
Bug: tint:1718
Change-Id: I739bcefc2ed36b0203b57974b50bb2b79f6e1684
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132980
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Reviewed-by: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
This CL moves the module from being owned by the builder, to be passed
into the builder and stored as a reference. This allows the transforms
to create builders based on the same module as needed.
Bug: tint:1718
Change-Id: I4863f368582242626c2a2c6996fd62f053314bb1
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132862
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Dan Sinclair <dsinclair@chromium.org>
Reviewed-by: James Price <jrprice@google.com>
Remove the direct use of BuilderImpl from TestHelperBase to cut down the amount
of internal state management required by the tests, and removing confusing
conflation between the BuilderImpl and Builder.
This change removes the following methods from TestHelperBase:
* CreateBuilder()
* InjectFlowBlock()
* CreateEmptyBuilder()
* FlowNodeForAstNode()
Tests now just use FromProgram() function for testing AST -> IR.
The downside to the black-box testing is that the per-method granularity of the
unit testing increases to whole FromProgram() granularity. However, my personal
opinion is that this is more than offset by the lack of state leakage from the
implementation to the tests.
Bug tint:1718
Change-Id: Iba2560e0fbcbd3dfb936694e50997d716f09fbd8
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132960
Kokoro: Ben Clayton <bclayton@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Commit-Queue: Ben Clayton <bclayton@google.com>
Now that AST transforms live in the AST namespace, these prefixes are
no longer necessary.
Change-Id: I658746ac04220075653ec57d6dc998947232d8bc
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132425
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: James Price <jrprice@google.com>
All AST transforms live inside this namespace. There will be a
corresponding ir::transform namespace for IR transforms.
Change-Id: I543972451c08e7df5632b835257999a8af3701d1
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132424
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: James Price <jrprice@google.com>
This prepares for the addition of IR transforms. Subsequent patches
will introduce an ast::transform namespace and refactor the Transform
class into a common base class for the two types of transform.
Bug: tint:1718
Change-Id: I0088d8356341e861df68df1f4f755ee82757dd2d
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132423
Commit-Queue: James Price <jrprice@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
There is no `builtin` subdirectory of `ast`, and this additional
`builtin` namespaces causes awkward conflicts in a subsequent patch
that moves AST transforms under the `ast` namespace.
Change-Id: I54b11f37b686875f1dab7deb9781e978a639d6e9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132422
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Make sure we return `false` when something goes wrong during
codegen. Also, add some more TINT_ICE calls in places where we know
stuff is unimplemented. This makes it easier to track current progress
against Tint's E2E tests.
Bug: tint:1906
Change-Id: Ic5885201d7b4f286d8f282df5a2074a017a98477
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132421
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: James Price <jrprice@google.com>
ir::Value has a virtual Type() method. Implement this on the subclasses
that actually have a type.
Removed pointless memory usage on instructions that have no value.
Also un-getter some fields to match the other IR code.
Change-Id: Ibf545a582d5cbb820410c9aaec0312692e803373
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/131745
Reviewed-by: James Price <jrprice@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Missing utils:: namespace on TrimSuffix().
Change-Id: I184bee79dc2c807730c1be0a532b290441d5f4ae
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132323
Commit-Queue: Ben Clayton <bclayton@google.com>
Kokoro: Ben Clayton <bclayton@google.com>
Reviewed-by: Ryan Harrison <rharrison@chromium.org>
Reviewed-by: James Price <jrprice@google.com>
Auto-Submit: Ben Clayton <bclayton@google.com>
Register the name of 'var' and 'let' variables when building the IR.
Use the names of these variables when printing the disassembly.
Change-Id: I56ee24a29333f1bc8f97459bc1cfca5c3a59e79d
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/131741
Auto-Submit: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: James Price <jrprice@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Adds a `Value()` method to the writer which gets the ID for a value,
which could either be a constant or an instruction result.
Bug: tint:1906
Change-Id: I57f5e0cfea1e3b8322702090714bac49d003ba75
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132285
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
All instructions are unimplemented, and the only supported branch
target is the function terminator.
No new tests, as this only handles returning from an empty function.
Bug: tint:1906
Change-Id: I05cf5d252d96d37757c3ac1f84b79d0586c36b97
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132283
Commit-Queue: James Price <jrprice@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Use a hashmap to de-duplicate constants.
Bug: tint:1906
Change-Id: Ic2654e1819d8c3c0199ce3f723f20877495ed0f0
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/132282
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: James Price <jrprice@google.com>
There's a few reasons for this change:
* Not all values have identifiers, and carrying redundant fields is
inefficent.
* Not all IDs will be integers - much like LLVM IR and SPIR-V, we will
likely want to disassemble with textual identifiers, so a uint32_t
is not ideal, and a std::string is even more bloat for each value.
* Transforms don't use identifiers, but instead raw pointers. We don't
want to encourage using IDs as they're simply a less-efficient way to
refer to values.
* This makes values consistent with types and flow-control blocks, as
they will both have their disassembly ID generated by the
disassembler.
The next step will be to add a hashmap to the module so that
pre-declared value names can be stored out-of-band.
Bug: tint:1718
Change-Id: I80beafc165f2bde54cc44a91015776926ca952b2
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/131740
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: James Price <jrprice@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
