Stephan Herhut [Wed, 28 Aug 2019 08:50:28 +0000 (01:50 -0700)]
Port mlir-cuda-runner to use dialect conversion framework.
Instead of lowering the program in two steps (Standard->LLVM followed
by GPU->NVVM), leading to invalid IR inbetween, the runner now uses
one pattern based rewrite step to go directly from Standard+GPU to
LLVM+NVVM.
PiperOrigin-RevId:
265861934
Uday Bondhugula [Wed, 28 Aug 2019 00:56:25 +0000 (17:56 -0700)]
Refactor / improve replaceAllMemRefUsesWith
Refactor replaceAllMemRefUsesWith to split it into two methods: the new
method does the replacement on a single op, and is used by the existing
one.
- make the methods return LogicalResult instead of bool
- Earlier, when replacement failed (due to non-deferencing uses of the
memref), the set of ops that had already been processed would have
been replaced leaving the IR in an inconsistent state. Now, a
pass is made over all ops to first check for non-deferencing
uses, and then replacement is performed. No test cases were affected
because all clients of this method were first checking for
non-deferencing uses before calling this method (for other reasons).
This isn't true for a use case in another upcoming PR (scalar
replacement); clients can now bail out with consistent IR on failure
of replaceAllMemRefUsesWith. Add test case.
- multiple deferencing uses of the same memref in a single op is
possible (we have no such use cases/scenarios), and this has always
remained unsupported. Add an assertion for this.
- minor fix to another test pipeline-data-transfer case.
Signed-off-by: Uday Bondhugula <uday@polymagelabs.com>
Closes tensorflow/mlir#87
PiperOrigin-RevId:
265808183
River Riddle [Tue, 27 Aug 2019 19:43:55 +0000 (12:43 -0700)]
Update Ch.2 of the Toy tutorial.
The code and documentation for this chapter of the tutorial have been updated to follow the new flow. The toy 'array' type has been replaced by usages of the MLIR tensor type. The code has also been cleaned up and modernized.
Closes tensorflow/mlir#101
PiperOrigin-RevId:
265744086
MLIR Team [Tue, 27 Aug 2019 17:55:47 +0000 (10:55 -0700)]
Add 3 additional intrinsic ops to NVVM dialect, in preparation to implement block-wide reduce.
PiperOrigin-RevId:
265720077
Lei Zhang [Tue, 27 Aug 2019 17:50:58 +0000 (10:50 -0700)]
[spirv] Fix the entry block to start with OpLabel
Each basic block in SPIR-V must start with an OpLabel instruction.
We don't support control flow yet, so this CL just makes sure that
the entry block follows this rule and is valid.
PiperOrigin-RevId:
265718841
Mahesh Ravishankar [Tue, 27 Aug 2019 17:49:53 +0000 (10:49 -0700)]
Enhance GPU To SPIR-V conversion to support builtins and load/store ops.
To support a conversion of a simple load-compute-store kernel from GPU
dialect to SPIR-V dialect, the conversion of operations like
"gpu.block_dim", "gpu.thread_id" which allow threads to get the launch
conversion is needed. In SPIR-V these are specified as global
variables with builin attributes. This CL adds support to specify
builtin variables in SPIR-V conversion framework. This is used to
convert the relevant operations from GPU dialect to SPIR-V dialect.
Also add support for conversion of load/store operation in Standard
dialect to SPIR-V dialect.
To simplify the conversion add a method to build a spv.AccessChain
operation that automatically determines the return type based on the
base pointer type and the indices provided.
PiperOrigin-RevId:
265718525
Denis Khalikov [Tue, 27 Aug 2019 17:41:07 +0000 (10:41 -0700)]
[spirv] Add Block decoration for spv.struct.
Add Block decoration for top-level spv.struct.
Closes tensorflow/mlir#102
PiperOrigin-RevId:
265716241
River Riddle [Tue, 27 Aug 2019 00:34:06 +0000 (17:34 -0700)]
NFC: Remove the explicit context from Operation::create and OperationState.
The context can easily be recovered from the Location in these situations.
PiperOrigin-RevId:
265578574
Eric Schweitz [Mon, 26 Aug 2019 21:18:47 +0000 (14:18 -0700)]
Add FPToSI/FPExt/FPTrunc cast ops to the LLVM dialect.
Closes tensorflow/mlir#99
PiperOrigin-RevId:
265538731
River Riddle [Mon, 26 Aug 2019 20:53:22 +0000 (13:53 -0700)]
NFC: Remove unnecessary context parameters from several Location getters.
The context can be recovered by other means in these methods and doesn't need to be passed explicitly.
PiperOrigin-RevId:
265532956
MLIR Team [Mon, 26 Aug 2019 17:27:38 +0000 (10:27 -0700)]
Update documentation for custom rewrite specs.
PiperOrigin-RevId:
265485862
Andy Ly [Mon, 26 Aug 2019 16:44:09 +0000 (09:44 -0700)]
Support folding of ops with inner ops in GreedyPatternRewriteDriver.
This fixes a bug when folding ops with inner ops and inner ops are still being visited.
PiperOrigin-RevId:
265475780
Alina Sbirlea [Sat, 24 Aug 2019 04:00:40 +0000 (21:00 -0700)]
NFC: Add doc for id-punct
PiperOrigin-RevId:
265190168
River Riddle [Sat, 24 Aug 2019 01:11:29 +0000 (18:11 -0700)]
NFC: Update Ch.1 of the Toy tutorial.
Change the use of 'array' to 'tensor' to reflect the new flow that the tutorial will follow. Also tidy up some of the documentation, code comments, and fix a few out-dated links.
PiperOrigin-RevId:
265174676
Nicolas Vasilache [Sat, 24 Aug 2019 00:44:55 +0000 (17:44 -0700)]
Lower linalg.copy to LLVM dialect in the presence of transposes.
Add an extra RewritePattern that does not convert types to rewrite a CopyOp that has non-identity permutations into a sequence of TransposeOp followed by a CopyOp without such permutations.
This RewitePattern is made to fail in the non-permutation case so that the conversion pattern can kick in to lower to LLVM.
This is an instance of A->A->B lowering where A->A is done by a RewritePattern in case_1 and A->B is done by a ConversionPatternRewriter when not(case_1).
PiperOrigin-RevId:
265171380
Nicolas Vasilache [Sat, 24 Aug 2019 00:28:51 +0000 (17:28 -0700)]
Lower linalg.transpose to LLVM dialect
Add a conversion pattern that transforms a linalg.transpose op into:
1. A function entry `alloca` operation to allocate a ViewDescriptor.
2. A load of the ViewDescriptor from the pointer allocated in 1.
3. Updates to the ViewDescriptor to introduce the data ptr, offset, size
and stride. Size and stride are permutations of the original values.
4. A store of the resulting ViewDescriptor to the alloca'ed pointer.
The linalg.transpose op is replaced by the alloca'ed pointer.
PiperOrigin-RevId:
265169112
Nicolas Vasilache [Fri, 23 Aug 2019 21:47:46 +0000 (14:47 -0700)]
Add a linalg.transpose op
A linalg.transpose op is a pure metadata operation that takes a view + permutation map and produces
another view of the same underlying data, with a different reindexing. This is a
pure metadata operation that does not touch the underlying data.
Example:
```
%t = linalg.transpose %v (i, j) -> (j, i) : !linalg.view<?x?xf32>
```
PiperOrigin-RevId:
265139429
River Riddle [Fri, 23 Aug 2019 18:28:19 +0000 (11:28 -0700)]
NFC: Add a note to 'applyPatternsGreedily' that it also performs folding/dce.
Fixes tensorflow/mlir#72
PiperOrigin-RevId:
265097597
Nicolas Vasilache [Fri, 23 Aug 2019 18:08:59 +0000 (11:08 -0700)]
Add lowering of linalg.copy to an external C++ library and a test.
This CL extends support for lowering of linalg to external C++ libraries with CopyOp. Currently this can only work when the permutation maps in the copies are identity. Future support for permutations will be added later.
PiperOrigin-RevId:
265093025
Nicolas Vasilache [Fri, 23 Aug 2019 18:08:33 +0000 (11:08 -0700)]
Update Linalg slice and subview documentation - NFC
PiperOrigin-RevId:
265092922
Lei Zhang [Fri, 23 Aug 2019 18:07:13 +0000 (11:07 -0700)]
[spirv] NFC: move SPIR-V control flow ops to a separate file
This CL is also purely moving code around for better file organization.
PiperOrigin-RevId:
265092566
Chris Lattner [Fri, 23 Aug 2019 17:35:24 +0000 (10:35 -0700)]
Introduce the ability for "isolated from above" ops to introduce shadowing
names for the basic block arguments in their body.
PiperOrigin-RevId:
265084627
MLIR Team [Fri, 23 Aug 2019 17:34:04 +0000 (10:34 -0700)]
NFC: Update in-code documentation. Make the two grammar definitions of static-dimension-list consistent.
PiperOrigin-RevId:
265084348
Lei Zhang [Fri, 23 Aug 2019 17:26:13 +0000 (10:26 -0700)]
[spirv] NFC: move arithmetic and logical ops to separate files
This is purely moving code around for better file organization.
PiperOrigin-RevId:
265082517
Nicolas Vasilache [Fri, 23 Aug 2019 17:15:29 +0000 (10:15 -0700)]
Fix BufferAllocOp builder.
One of the BufferAllocOp builders was improperly specified which triggered infinite recursion. This CL fixes it.
PiperOrigin-RevId:
265080371
River Riddle [Fri, 23 Aug 2019 17:08:42 +0000 (10:08 -0700)]
NFC: Move the operation, region, and block sections to after the dialect section.
Operations/Regions/Blocks represent the core IR building blocks and should be introduced before types and attributes.
PiperOrigin-RevId:
265079103
MLIR Team [Fri, 23 Aug 2019 02:05:07 +0000 (19:05 -0700)]
Add I32ElementsAttr to OpBase
PiperOrigin-RevId:
264969142
River Riddle [Fri, 23 Aug 2019 01:58:51 +0000 (18:58 -0700)]
Add iterator support to ElementsAttr and SparseElementsAttr.
This will allow iterating the values of a non-opaque ElementsAttr, with all of the types currently supported by DenseElementsAttr. This should help reduce the amount of specialization on DenseElementsAttr.
PiperOrigin-RevId:
264968151
River Riddle [Fri, 23 Aug 2019 00:51:06 +0000 (17:51 -0700)]
NFC: Cleanup the Attribute section in the LangRef.
* Add a section on dialect attribute values and attribute aliases
* Move FloatAttr into its alphabetically correct place
* Add a "Standard Attribute Values" section
PiperOrigin-RevId:
264959306
River Riddle [Thu, 22 Aug 2019 23:43:06 +0000 (16:43 -0700)]
NFC: Cleanup the type system section of the LangRef.
* Alphabetize the type definitions
* Make 'Dialect specific types' a type-system subsection
* Merge Builtin types and Standard types
PiperOrigin-RevId:
264947721
Lei Zhang [Thu, 22 Aug 2019 23:01:13 +0000 (16:01 -0700)]
[spirv] Add support for extension (de)serialization
Only a few important KHR extensions are registered to the
SPIR-V dialect for now.
PiperOrigin-RevId:
264939428
River Riddle [Thu, 22 Aug 2019 22:53:41 +0000 (15:53 -0700)]
NFC: Rework and cleanup the High-Level structure and Dialect sections.
Both sections are out-of-date and need to be updated. The dialect section is particularly bad in that it never actually mentions what a 'Dialect' is.
PiperOrigin-RevId:
264937905
River Riddle [Thu, 22 Aug 2019 20:20:28 +0000 (13:20 -0700)]
NFC: Remove mentions of the TensorFlow dialect from the langref.
PiperOrigin-RevId:
264904489
Nicolas Vasilache [Thu, 22 Aug 2019 19:46:30 +0000 (12:46 -0700)]
Avoid overflow when lowering linalg.slice
linalg.subview used to lower to a slice with a bounded range resulting in correct bounded accesses. However linalg.slice could still index out of bounds. This CL moves the bounding to linalg.slice.
LLVM select and cmp ops gain a more idiomatic builder.
PiperOrigin-RevId:
264897125
River Riddle [Thu, 22 Aug 2019 18:31:01 +0000 (11:31 -0700)]
NFC: Avoid reconstructing the OpInterface methods.
PiperOrigin-RevId:
264881293
Lei Zhang [Thu, 22 Aug 2019 18:15:05 +0000 (11:15 -0700)]
[spirv] Add support for capability (de)serialization
This CL pulls in capabilities defined in the spec and adds
support for (de)serialize capabilities of a spv.module.
PiperOrigin-RevId:
264877413
Logan Chien [Thu, 22 Aug 2019 17:36:01 +0000 (10:36 -0700)]
Add Positive{I32,I64}Attr and HasAnyRankOfPred
This commit adds `PositiveI32Attr` and `PositiveI64Attr` to match positive
integers but not zero nor negative integers. This commit also adds
`HasAnyRankOfPred` to match tensors with the specified ranks.
PiperOrigin-RevId:
264867046
Jacques Pienaar [Thu, 22 Aug 2019 16:45:04 +0000 (09:45 -0700)]
Split out parsing location into separate functions per instance
Split out method into specialized instances + add an early exit. Should be NFC, but simplifies reading the logic slightly IMHO.
PiperOrigin-RevId:
264855529
Nicolas Vasilache [Thu, 22 Aug 2019 16:38:16 +0000 (09:38 -0700)]
Let LLVMOpLowering specify a PatternBenefit - NFC
Currently the benefit is always set to 1 which limits the ability to do A->B->C lowering
PiperOrigin-RevId:
264854146
River Riddle [Thu, 22 Aug 2019 14:15:09 +0000 (07:15 -0700)]
NFC: Fix path of LinalgLibraryOpInterfaces inc files.
PiperOrigin-RevId:
264827908
River Riddle [Thu, 22 Aug 2019 03:57:23 +0000 (20:57 -0700)]
Add support for generating operation interfaces from the ODS framework.
Operation interfaces generally require a bit of boilerplate code to connect all of the pieces together. This cl introduces mechanisms in the ODS to allow for generating operation interfaces via the 'OpInterface' class.
Providing a definition of the `OpInterface` class will auto-generate the c++
classes for the interface. An `OpInterface` includes a name, for the c++ class,
along with a list of interface methods. There are two types of methods that can be used with an interface, `InterfaceMethod` and `StaticInterfaceMethod`. They are both comprised of the same core components, with the distinction that `StaticInterfaceMethod` models a static method on the derived operation.
An `InterfaceMethod` is comprised of the following components:
* ReturnType
- A string corresponding to the c++ return type of the method.
* MethodName
- A string corresponding to the desired name of the method.
* Arguments
- A dag of strings that correspond to a c++ type and variable name
respectively.
* MethodBody (Optional)
- An optional explicit implementation of the interface method.
def MyInterface : OpInterface<"MyInterface"> {
let methods = [
// A simple non-static method with no inputs.
InterfaceMethod<"unsigned", "foo">,
// A new non-static method accepting an input argument.
InterfaceMethod<"Value *", "bar", (ins "unsigned":$i)>,
// Query a static property of the derived operation.
StaticInterfaceMethod<"unsigned", "fooStatic">,
// Provide the definition of a static interface method.
// Note: `ConcreteOp` corresponds to the derived operation typename.
StaticInterfaceMethod<"Operation *", "create",
(ins "OpBuilder &":$builder, "Location":$loc), [{
return builder.create<ConcreteOp>(loc);
}]>,
// Provide a definition of the non-static method.
// Note: `op` corresponds to the derived operation variable.
InterfaceMethod<"unsigned", "getNumInputsAndOutputs", (ins), [{
return op.getNumInputs() + op.getNumOutputs();
}]>,
];
PiperOrigin-RevId:
264754898
River Riddle [Thu, 22 Aug 2019 02:03:13 +0000 (19:03 -0700)]
Avoid assigning to an unchecked Error.
Fixes tensorflow/mlir#97
PiperOrigin-RevId:
264743395
Lei Zhang [Thu, 22 Aug 2019 01:53:32 +0000 (18:53 -0700)]
Point to spv.AccessChain when reporting spv.AccessChain errors
PiperOrigin-RevId:
264742130
Nicolas Vasilache [Thu, 22 Aug 2019 01:36:15 +0000 (18:36 -0700)]
Remove dead getLLVMLibraryCallImplDefinition in Linalg's LowerToLLVMDialect.cpp - NFC
PiperOrigin-RevId:
264740014
Nicolas Vasilache [Thu, 22 Aug 2019 01:15:39 +0000 (18:15 -0700)]
Reduce reliance on custom grown Jit implementation - NFC
This CL makes use of the standard LLVM LLJIT and removes the need for a custom JIT implementation within MLIR.
To achieve this, one needs to clone (i.e. serde) the produced llvm::Module into a new LLVMContext. This is currently necessary because the llvm::LLVMContext is owned by the LLVMDialect, somewhat deep in the call hierarchy.
In the future we should remove the reliance of serding the llvm::Module by allowing the injection of an LLVMContext from the top-level. Unfortunately this will require deeper API changes and impact multiple places. It is therefore left for future work.
PiperOrigin-RevId:
264737459
Lei Zhang [Thu, 22 Aug 2019 01:04:56 +0000 (18:04 -0700)]
Remove the wrapping function in SPIR-V (de)serialization
Previously Module and Function are builtinn constructs in MLIR.
Due to the structural requirements we must wrap the SPIR-V
module inside a Function inside a Module. Now the requirement
is lifted and we can remove the wrapping function! :)
PiperOrigin-RevId:
264736051
MLIR Team [Thu, 22 Aug 2019 00:51:27 +0000 (17:51 -0700)]
NFC: Update in-code documentation for type.
PiperOrigin-RevId:
264734014
Chintan Kaur [Thu, 22 Aug 2019 00:45:06 +0000 (17:45 -0700)]
Fix minor typos in TestingGuide and OpDefinitions.
PiperOrigin-RevId:
264733092
MLIR Team [Wed, 21 Aug 2019 23:50:55 +0000 (16:50 -0700)]
NFC: Update in-code documentation for function-type.
PiperOrigin-RevId:
264723462
River Riddle [Wed, 21 Aug 2019 23:50:30 +0000 (16:50 -0700)]
Add a hook to the OpAsmDialectInterface to allow providing a special name for the operation result.
This generalizes the current special handling for constant operations(they get named 'cst'/'true'/'false'/etc.)
PiperOrigin-RevId:
264723379
MLIR Team [Wed, 21 Aug 2019 21:28:10 +0000 (14:28 -0700)]
[TableGen] Add a `StaticShapeMemRefOf` trait.
The trait specifies that the `MemRefOf` has to have a static shape.
PiperOrigin-RevId:
264692758
River Riddle [Wed, 21 Aug 2019 20:00:30 +0000 (13:00 -0700)]
Automated rollback of commit
b9dc2e481818315f2f0d87455349f497f6118a4c
PiperOrigin-RevId:
264672975
River Riddle [Wed, 21 Aug 2019 19:16:23 +0000 (12:16 -0700)]
NFC: Make the ModuleState field in the ModulePrinter optional.
The ModuleState is only used for printing aliases, which is only done when printing the top-level module.
PiperOrigin-RevId:
264664138
River Riddle [Wed, 21 Aug 2019 17:23:14 +0000 (10:23 -0700)]
Add iterator support to ElementsAttr and SparseElementsAttr.
This will allow iterating the values of a non-opaque ElementsAttr, with all of the types currently supported by DenseElementsAttr. This should help reduce the amount of specialization on DenseElementsAttr.
PiperOrigin-RevId:
264637293
River Riddle [Wed, 21 Aug 2019 16:41:37 +0000 (09:41 -0700)]
Move the parser extensions for aliases currently on Dialect to a new OpAsmDialectInterface.
This will allow for adding more hooks for controlling parser behavior without bloating Dialect in the common case. This cl also adds iteration support to the DialectInterfaceCollection.
PiperOrigin-RevId:
264627846
Lei Zhang [Wed, 21 Aug 2019 15:17:19 +0000 (08:17 -0700)]
[spirv] Support i1 as bool type
PiperOrigin-RevId:
264612014
Lei Zhang [Wed, 21 Aug 2019 12:35:07 +0000 (05:35 -0700)]
Support variadic ops in declarative rewrite rules
This CL extends declarative rewrite rules to support matching and
generating ops with variadic operands/results. For this, the
generated `matchAndRewrite()` method for each pattern now are
changed to
* Use "range" types for the local variables used to store captured
values (`operand_range` for operands, `ArrayRef<Value *>` for
values, *Op for results). This allows us to have a unified way
of handling both single values and value ranges.
* Create local variables for each operand for op creation. If the
operand is variadic, then a `SmallVector<Value*>` will be created
to collect all values for that operand; otherwise a `Value*` will
be created.
* Use a collective result type builder. All result types are
specified via a single parameter to the builder.
We can use one result pattern to replace multiple results of the
matched root op. When that happens, it will require specifying
types for multiple results. Add a new collective-type builder.
PiperOrigin-RevId:
264588559
Lei Zhang [Wed, 21 Aug 2019 11:45:23 +0000 (04:45 -0700)]
Materialize spv.constants at use sites
In SPIR-V binary format, constants are placed at the module level
and referenced by instructions inside functions using their result
<id>s. To model this natively (using SSA values for result <id>s),
it means we need to have implicit capturing functions. We will
lose the ability to have function passes if going down that path.
Instead, this CL changes to materialize constants at their use
sites in deserialization. It's cheap to copy constants in MLIR
given that attributes is uniqued to MLIRContext. By localizing
constants into functions, we can preserve isolated functions.
PiperOrigin-RevId:
264582532
River Riddle [Wed, 21 Aug 2019 02:58:35 +0000 (19:58 -0700)]
NFC: Keep the dialect list in the context sorted by namespace.
Most dialects are initialized statically, which does not have a guaranteed initialization order. By keeping the dialect list sorted, we can guarantee a deterministic iteration order of dialects.
PiperOrigin-RevId:
264522875
River Riddle [Wed, 21 Aug 2019 01:49:08 +0000 (18:49 -0700)]
NFC: Use a DenseSet instead of a DenseMap for DialectInterfaceCollection.
The interfaces are looked up by dialect, which can always be retrieved from an interface instance.
PiperOrigin-RevId:
264516023
River Riddle [Wed, 21 Aug 2019 01:41:38 +0000 (18:41 -0700)]
NFC: Move the LangRef documentation on StandardOps to a new document.
The LangRef should contain documentation about the core system, and standard ops is a dialect just like any other. This will also simplify the transition when StandardOps is eventually split apart.
PiperOrigin-RevId:
264514988
River Riddle [Tue, 20 Aug 2019 22:36:08 +0000 (15:36 -0700)]
NFC: Move AffineOps dialect to the Dialect sub-directory.
PiperOrigin-RevId:
264482571
Lei Zhang [Tue, 20 Aug 2019 20:33:41 +0000 (13:33 -0700)]
Add spv.specConstant and spv._reference_of
Similar to global variables, specialization constants also live
in the module scope and can be referenced by instructions in
functions in native SPIR-V. A direct modelling would be to allow
functions in the SPIR-V dialect to implicit capture, but it means
we are losing the ability to write passes for Functions. While
in SPIR-V normally we want to process the module as a whole,
it's not common to see multiple functions get used so we'd like
to leave the door open for those cases. Therefore, similar to
global variables, we introduce spv.specConstant to model three
SPIR-V instructions: OpSpecConstantTrue, OpSpecConstantFalse,
and OpSpecConstant. They do not return SSA value results;
instead they have symbols and can only be referenced by the
symbols. To use it in a function, we need to have another op
spv._reference_of to turn the symbol into an SSA value. This
breaks the tie and makes functions still explicit capture.
Previously specialization constants were handled similarly as
normal constants. That is incorrect given that specialization
constant actually acts more like variable (without need to
load and store). E.g., they cannot be de-duplicated like normal
constants.
This CL also refines various documents and comments.
PiperOrigin-RevId:
264455172
Denis Khalikov [Tue, 20 Aug 2019 18:02:57 +0000 (11:02 -0700)]
[spirv] Support (de)serialization of spv.struct
Support (de)serialization of spv.struct with offset decorations.
Closes tensorflow/mlir#94
PiperOrigin-RevId:
264421427
Diego Caballero [Tue, 20 Aug 2019 17:43:45 +0000 (10:43 -0700)]
Fix build of affine load/store with empty map
tensorflow/mlir#58 fixed and exercised
verification of load/store ops using empty affine maps. Unfortunately,
it didn't exercise the creation of them. This PR addresses that aspect.
It removes the assumption of AffineMap having at least one result and
stores a pointer to MLIRContext as member of AffineMap.
* Add empty map support to affine.store + test
* Move MLIRContext to AffineMapStorage
Closes tensorflow/mlir#74
PiperOrigin-RevId:
264416260
Andy Davis [Tue, 20 Aug 2019 17:38:27 +0000 (10:38 -0700)]
Update MLIR code examples in Passes.md doc to use new affine.load/store dma_start/wait operations.
PiperOrigin-RevId:
264415037
Zhang [Tue, 20 Aug 2019 15:53:02 +0000 (08:53 -0700)]
Update Ch-2.md
--
406f1e8211f8f5017f44f46af750dec061e707a2 by Zhang <5205699+Naville@users.noreply.github.com>:
Update Ch-2.md
Closes tensorflow/mlir#93
PiperOrigin-RevId:
264392995
Alex Zinenko [Tue, 20 Aug 2019 14:51:32 +0000 (07:51 -0700)]
ConvertLaunchFuncToCudaCalls: use LLVM dialect globals
This conversion has been using a stack-allocated array of i8 to store the
null-terminated kernel name in order to pass it to the CUDA wrappers expecting
a C string because the LLVM dialect was missing support for globals. Now that
the suport is introduced, use a global instead.
Refactor global string construction from GenerateCubinAccessors into a common
utility function living in the LLVM namespace.
PiperOrigin-RevId:
264382489
Alex Zinenko [Tue, 20 Aug 2019 14:45:47 +0000 (07:45 -0700)]
JitRunner: support entry functions returning void
JitRunner can use as entry points functions that produce either a single
'!llvm.f32' value or a list of memrefs. Memref support is legacy and was
introduced before MLIR could lower memref allocation and deallocation to
malloc/free calls so as to allocate the memory externally, and is likely to be
dropped in the future since it unconditionally runs affine+standard-to-llvm
lowering on the module instead of accepting the LLVM dialect. CUDA runner
relies on memref-based flow in the runner without actually returning anything.
Introduce a runner flow to use functions that return void as entry points.
PiperOrigin-RevId:
264381686
Alex Zinenko [Tue, 20 Aug 2019 13:38:28 +0000 (06:38 -0700)]
LLVM dialect: prefix operations that correspond to intrinsics with "intr."
LLVM intrinsics have an open name space and their names can potentially overlap
with names of LLVM instructions (LLVM intrinsics are functions, not
instructions). In MLIR, LLVM intrinsics are modeled as operations, so it needs
to make sure their names cannot clash with the instructions. Use the "intr."
prefix for intrinsics in the LLVM dialect.
PiperOrigin-RevId:
264372173
Nicolas Vasilache [Tue, 20 Aug 2019 08:59:58 +0000 (01:59 -0700)]
Add support for LLVM lowering of binary ops on n-D vector types
This CL allows binary operations on n-D vector types to be lowered to LLVMIR by performing an (n-1)-D extractvalue, 1-D vector operation and an (n-1)-D insertvalue.
PiperOrigin-RevId:
264339118
Uday Bondhugula [Tue, 20 Aug 2019 08:52:39 +0000 (01:52 -0700)]
Fix AffineExpr::simplifyAdd bug
- fix missing check while simplifying an expression with floordiv to a
mod
- fixes issue tensorflow/mlir#82
Signed-off-by: Uday Bondhugula <uday@polymagelabs.com>
Closes tensorflow/mlir#84
PiperOrigin-RevId:
264338353
Chintan Kaur [Tue, 20 Aug 2019 00:29:22 +0000 (17:29 -0700)]
Fix minor typos and add missing syntax in the documentation.
PiperOrigin-RevId:
264281501
Nicolas Vasilache [Tue, 20 Aug 2019 00:11:12 +0000 (17:11 -0700)]
Move Linalg and VectorOps dialects to the Dialect subdir - NFC
PiperOrigin-RevId:
264277760
River Riddle [Tue, 20 Aug 2019 00:01:13 +0000 (17:01 -0700)]
Add a DialectConversion document detailing the conversion infrastructure.
This is an important piece of the infrastructure that is missing proper high level documentation on usage.
PiperOrigin-RevId:
264275482
Rob Suderman [Mon, 19 Aug 2019 22:56:55 +0000 (15:56 -0700)]
Add DictionaryAttr to OpBase.td
PiperOrigin-RevId:
264262369
River Riddle [Mon, 19 Aug 2019 22:26:43 +0000 (15:26 -0700)]
Allow isolated regions to form isolated SSA name scopes in the printer.
This will allow for naming values the same as existing SSA values for regions attached to operations that are isolated from above. This fits in with how the system already allows separate name scopes for sibling regions. This name shadowing can be enabled in the custom parser of operations by setting the 'enableNameShadowing' flag to true when calling 'parseRegion'.
%arg = constant 10 : i32
foo.op {
%arg = constant 10 : i32
}
PiperOrigin-RevId:
264255999
Nicolas Vasilache [Mon, 19 Aug 2019 21:36:49 +0000 (14:36 -0700)]
Add alignment support to linalg.buffer_alloc
This CL adds an integer attribute to linalg.buffer_alloc and lowering to LLVM.
The alignment is constrained to be a positive power of 2.
Lowering to LLVM produces the pattern:
```
%[[alloc:.*]] = llvm.call @malloc(%[[s]]) : (!llvm.i64) -> !llvm<"i8*">
%[[cast:.*]] = llvm.bitcast %[[alloc]] : !llvm<"i8*"> to !llvm.i64
%[[rem:.*]] = llvm.urem %[[cast]], %[[c16]] : !llvm.i64
%[[drem:.*]] = llvm.sub %[[c16]], %[[rem]] : !llvm.i64
%[[off:.*]] = llvm.urem %[[drem]], %[[c16]] : !llvm.i64
llvm.getelementptr %{{.*}}[%[[off]]] : (!llvm<"i8*">, !llvm.i64) -> !llvm<"i8*">
```
where `ptr` is aligned on `align` by computing the address
`ptr + (align - ptr % align) % align`.
To allow dealloc op to still be able to free memory, additional information is needed in
the buffer type. The buffer type is thus extended with an extra i8* for the base allocation address.
PiperOrigin-RevId:
264244455
River Riddle [Mon, 19 Aug 2019 19:43:46 +0000 (12:43 -0700)]
Add support for Operation interfaces.
Operation interfaces, as the name suggests, are those registered at the
Operation level. These interfaces provide an opaque view into derived
operations, by providing a virtual interface that must be implemented. As an
example, the Linalg dialect implements an interface LinalgOp that provides
general queries about some of the dialects library operations. These queries may
provide things like: the number of parallel loops, the number of inputs and
outputs, etc.
Operation interfaces are defined by overriding the CRTP base class OpInterface.
This class takes as a template parameter, a `Traits` class that defines a
Concept and a Model class. These classes provide an implementation of
concept-based polymorphism, where the Concept defines a set of virtual methods
that are overridden by the Model that is templated on the concrete operation
type. It is important to note that these classes should be pure in that they
contain no non-static data members.
PiperOrigin-RevId:
264218741
River Riddle [Mon, 19 Aug 2019 19:12:50 +0000 (12:12 -0700)]
NFC: Don't assume that all operation traits are within the 'OpTrait::' namespace.
This places an unnecessary restriction that all traits are within this namespace.
PiperOrigin-RevId:
264212000
Mahesh Ravishankar [Mon, 19 Aug 2019 18:38:53 +0000 (11:38 -0700)]
Fix parsing/printing of spv.globalVariable and spv._address_of
Change the prining/parsing of spv.globalVariable to print the type of
the variable after the ':' to be consistent with MLIR convention.
The spv._address_of should print the variable type after the ':'. It was
mistakenly printing the address of the return value. Add a (missing)
test that should have caught that.
Also move spv.globalVariable and spv._address_of tests to
structure-ops.mlir.
PiperOrigin-RevId:
264204686
River Riddle [Mon, 19 Aug 2019 18:00:47 +0000 (11:00 -0700)]
NFC: Move LLVMIR, SDBM, and StandardOps to the Dialect/ directory.
PiperOrigin-RevId:
264193915
Lei Zhang [Mon, 19 Aug 2019 17:57:43 +0000 (10:57 -0700)]
[spirv] Add spv.ReturnValue
This CL adds the spv.ReturnValue op and its tests. Also adds a
InFunctionScope trait to make sure that the op stays inside
a function. To be consistent, ModuleOnly trait is changed to
InModuleScope.
PiperOrigin-RevId:
264193081
Nicolas Vasilache [Mon, 19 Aug 2019 17:21:15 +0000 (10:21 -0700)]
Refactor linalg lowering to LLVM
The linalg.view type used to be lowered to a struct containing a data pointer, offset, sizes/strides information. This was problematic when passing to external functions due to ABI, struct padding and alignment issues.
The linalg.view type is now lowered to LLVMIR as a *pointer* to a struct containing the data pointer, offset and sizes/strides. This simplifies the interfacing with external library functions and makes it trivial to add new functions without creating a shim that would go from a value type struct to a pointer type.
The consequences are that:
1. lowering explicitly uses llvm.alloca in lieu of llvm.undef and performs the proper llvm.load/llvm.store where relevant.
2. the shim creation function `getLLVMLibraryCallDefinition` disappears.
3. views are passed by pointer, scalars are passed by value. In the future, other structs will be passed by pointer (on a per-need basis).
PiperOrigin-RevId:
264183671
Nicolas Vasilache [Mon, 19 Aug 2019 01:54:50 +0000 (18:54 -0700)]
Add alignment support for llvm.alloca
Extend the LLVM dialect AllocaOp with an alignment attribute.
PiperOrigin-RevId:
264068306
Jacques Pienaar [Sun, 18 Aug 2019 18:32:26 +0000 (11:32 -0700)]
InitLLVM already initializes PrettyStackTraceProgram
Remove extra PrettyStackTraceProgram and use InitLLVM consistently.
PiperOrigin-RevId:
264041205
Jacques Pienaar [Sat, 17 Aug 2019 18:05:35 +0000 (11:05 -0700)]
Change from llvm::make_unique to std::make_unique
Switch to C++14 standard method as llvm::make_unique has been removed (
https://reviews.llvm.org/D66259). Also mark some targets as c++14 to ease next
integrates.
PiperOrigin-RevId:
263953918
River Riddle [Sat, 17 Aug 2019 17:22:19 +0000 (10:22 -0700)]
NFC: Add header blocks to improve readability.
PiperOrigin-RevId:
263951251
Mahesh Ravishankar [Sat, 17 Aug 2019 17:19:48 +0000 (10:19 -0700)]
Add spirv::GlobalVariableOp that allows module level definition of variables
FuncOps in MLIR use explicit capture. So global variables defined in
module scope need to have a symbol name and this should be used to
refer to the variable within the function. This deviates from SPIR-V
spec, which assigns an SSA value to variables at all scopes that can
be used to refer to the variable, which requires SPIR-V functions to
allow implicit capture. To handle this add a new op,
spirv::GlobalVariableOp that can be used to define module scope
variables.
Since instructions need an SSA value, an new spirv::AddressOfOp is
added to convert a symbol reference to an SSA value for use with other
instructions.
This also means the spirv::EntryPointOp instruction needs to change to
allow initializers to be specified using symbol reference instead of
SSA value
The current spirv::VariableOp which returns an SSA value (as defined
by SPIR-V spec) can still be used to define function-scope variables.
PiperOrigin-RevId:
263951109
River Riddle [Sat, 17 Aug 2019 02:21:50 +0000 (19:21 -0700)]
NFC: Modernize and cleanup standard ops.
PiperOrigin-RevId:
263891926
River Riddle [Sat, 17 Aug 2019 00:59:03 +0000 (17:59 -0700)]
NFC: Refactor the PassInstrumentation framework to operate on Operation instead of llvm::Any.
Now that functions and modules are operations, Operation makes more sense as the opaque object to refer to both.
PiperOrigin-RevId:
263883913
River Riddle [Fri, 16 Aug 2019 21:45:37 +0000 (14:45 -0700)]
NFC: Move the Type::is* predicates to StandardTypes.cpp
These methods are currently defined 'inline' in StandardTypes.h, but this may create linker errors if StandardTypes.h isn't included at the use site.
PiperOrigin-RevId:
263850328
MLIR Team [Fri, 16 Aug 2019 18:00:31 +0000 (11:00 -0700)]
Fix minor typos in the documentation
PiperOrigin-RevId:
263805025
Denis Khalikov [Fri, 16 Aug 2019 17:17:47 +0000 (10:17 -0700)]
[spirv] Extend spv.array with Layoutinfo
Extend spv.array with Layoutinfo to support (de)serialization.
Closes tensorflow/mlir#80
PiperOrigin-RevId:
263795304
River Riddle [Fri, 16 Aug 2019 17:16:09 +0000 (10:16 -0700)]
Refactor DialectConversion to convert the signatures of blocks when they are moved.
Often we want to ensure that block arguments are converted before operations that use them. This refactors the current implementation to be cleaner/less frequent by triggering conversion when a set of blocks are moved/inlined; or when legalization is successful.
PiperOrigin-RevId:
263795005
Jacques Pienaar [Fri, 16 Aug 2019 15:34:37 +0000 (08:34 -0700)]
C++14 is now default enabled in LLVM, remove obsolete CMake flag.
PiperOrigin-RevId:
263776602
Jacques Pienaar [Fri, 16 Aug 2019 15:08:11 +0000 (08:08 -0700)]
Remove C++11 requirement set in cmakelists
C++14 is now the required.
PiperOrigin-RevId:
263772579
Nicolas Vasilache [Fri, 16 Aug 2019 10:52:56 +0000 (03:52 -0700)]
Extend vector.outerproduct with an optional 3rd argument
This CL adds an optional third argument to the vector.outerproduct instruction.
When such a third argument is specified, it is added to the result of the outerproduct and is lowered to FMA intrinsic when the lowering supports it.
In the future, we can add an attribute on the `vector.outerproduct` instruction to modify the operations for which to emit code (e.g. "+/*", "max/+", "min/+", "log/exp" ...).
This CL additionally performs minor cleanups in the vector lowering and adds tests to improve coverage.
This has been independently verified to result in proper fma instructions for haswell as follows.
Input:
```
func @outerproduct_add(%arg0: vector<17xf32>, %arg1: vector<8xf32>, %arg2: vector<17x8xf32>) -> vector<17x8xf32> {
%2 = vector.outerproduct %arg0, %arg1, %arg2 : vector<17xf32>, vector<8xf32>
return %2 : vector<17x8xf32>
}
}
```
Command:
```
mlir-opt vector-to-llvm.mlir -vector-lower-to-llvm-dialect --disable-pass-threading | mlir-opt -lower-to-cfg -lower-to-llvm | mlir-translate --mlir-to-llvmir | opt -O3 | llc -O3 -march=x86-64 -mcpu=haswell -mattr=fma,avx2
```
Output:
```
outerproduct_add: # @outerproduct_add
# %bb.0:
...
vmovaps 112(%rbp), %ymm8
vbroadcastss %xmm0, %ymm0
...
vbroadcastss 64(%rbp), %ymm15
vfmadd213ps 144(%rbp), %ymm8, %ymm0 # ymm0 = (ymm8 * ymm0) + mem
...
vfmadd213ps 400(%rbp), %ymm8, %ymm9 # ymm9 = (ymm8 * ymm9) + mem
...
```
PiperOrigin-RevId:
263743359
Mahesh Ravishankar [Thu, 15 Aug 2019 17:54:22 +0000 (10:54 -0700)]
Simplify the classes that support SPIR-V conversion.
Modify the Type converters to have a SPIRVBasicTypeConverter which
only handles conversion from standard types to SPIRV types. Rename
SPIRVEntryFnConverter to SPIRVTypeConverter. This contains the
SPIRVBasicTypeConverter within it.
Remove SPIRVFnLowering class and have separate utility methods to
lower a function as entry function or a non-entry function. The
current setup could end with diamond inheritence that is not very
friendly to use. For example, you could define the following Op
conversion methods that lower from a dialect "Foo" which resuls in
diamond inheritance.
template<typename OpTy>
class FooDialect : public SPIRVOpLowering<OpTy> {...};
class FooFnLowering : public FooDialect, SPIRVFnLowering {...};
PiperOrigin-RevId:
263597101
Mahesh Ravishankar [Thu, 15 Aug 2019 17:52:24 +0000 (10:52 -0700)]
Add BuiltIn EnumAttr to SPIR-V dialect
Generate the EnumAttr to represent BuiltIns in SPIR-V dialect. The
builtIn can be specified as a StringAttr with value being the
name of the builtin. Extend Decoration (de)serialization to handle
BuiltIns.
Also fix an error in the SPIR-V dialect generator script.
PiperOrigin-RevId:
263596624