Stella Laurenzo [Wed, 3 Apr 2019 19:40:30 +0000 (12:40 -0700)]
Refactor Quantization.md to separate TFLite native quantization scheme from the more experimental, generalized scheme.
PiperOrigin-RevId:
241785572
Nicolas Vasilache [Wed, 3 Apr 2019 19:33:01 +0000 (12:33 -0700)]
Linalg portion of the tutorial - part 3-2
This CL adds support for lowering tensor contractions to loops declaratively.
This is done thanks to two properties of the such operations:
1. the definition of an AffineMap getLoopsToOperandRangesMap for each op which maps iteration space dimensions to ranges of the view operands, in their order of occurrence;
2. the definition of a scalar implementation for each op which creates the computation inside the loops given enclosing parallel and reduction loops,
All the other properties are derived in a generic fashion from these 2 properties and a few analyses.
A lowerToLoops transformation is added as well as a test that exercises it.
--
PiperOrigin-RevId:
241783992
Lei Zhang [Wed, 3 Apr 2019 19:29:14 +0000 (12:29 -0700)]
[TableGen] Allow additional result patterns not directly used for replacement
This CL looses the requirement that all result patterns in a rewrite rule must
replace a result of the root op in the source pattern. Now only the last N
result pattern-generated ops are used to replace a N-result source op.
This allows to generate additional ops to aid building up final ops used to
replace the source op.
--
PiperOrigin-RevId:
241783192
Stella Laurenzo [Wed, 3 Apr 2019 18:16:32 +0000 (11:16 -0700)]
Initial release of the Quantization dialect
Includes a draft of documentation for the quantization setup.
Given how many comments such docs have garnered in the past, I've biased towards a lightly edited first-draft so that people can argue about terminology, approach and structure without having spent too much time on it.
Note that the sections under "Uniform quantization" were cribbed nearly verbatim from internal documentation that Daniel wrote.
PiperOrigin-RevId:
241768668
Lei Zhang [Wed, 3 Apr 2019 17:48:07 +0000 (10:48 -0700)]
[TableGen] Fix convertFromStorage for OptionalAttr
OptionalAttr is just wrapping around the actual attribute; so it should just use
the actual attribute's `convertFromStorage` to read the value and wrap it around
with `Optional<>` to return. Previously it was mandating how the actual attribute
reads the value with `{0}.getValue()`.
--
PiperOrigin-RevId:
241762355
Alex Zinenko [Wed, 3 Apr 2019 17:38:47 +0000 (10:38 -0700)]
Tutorial Linalg1: implement conversion to the LLVM Dialect
Implement conversion from the Linalg dialect to the LLVM dialect using a simple
set of DialectOpConverters and by plugging them into the dialect conversion
infrastructure. View and Range Linalg types are converted into descriptors
that store the dynamic values in an LLVM aggregate type, similarly to memrefs.
Slice operations create new descriptors based on the original descriptors and
thus remove the constraint on ViewTypes not being acceptable as function
arguments.
--
PiperOrigin-RevId:
241760189
Mehdi Amini [Wed, 3 Apr 2019 16:40:08 +0000 (09:40 -0700)]
Add mention of the TensorFlow Code of Conduct in the Readme
PiperOrigin-RevId:
241748009
Mehdi Amini [Wed, 3 Apr 2019 16:29:52 +0000 (09:29 -0700)]
Add missing source file in CMakeLists.txt
--
PiperOrigin-RevId:
241746122
Lei Zhang [Wed, 3 Apr 2019 16:29:25 +0000 (09:29 -0700)]
[TableGen] Remove asserts for attributes in aggregate builders
Attributes can have default values or be optional. Checking the validity of
attributes in aggregate builder should consider that. And to be accurate,
we should check all required attributes are indeed provided in the list.
This is actually duplicating the work done by verifier. Checking the validity
of attributes should be the responsiblity of verifiers. This CL removes
the assertion for attributes in aggregate builders for the above reason.
(Assertions for operands/results are still kept since they are trivial.)
Also added more tests for aggregate builders.
--
PiperOrigin-RevId:
241746059
Mehdi Amini [Wed, 3 Apr 2019 16:20:26 +0000 (09:20 -0700)]
Fix CMake build
Some files were not built anymore internally but still referenced
from CMake. Delete them and unreference them in the CMake files.
--
PiperOrigin-RevId:
241744718
Mehdi Amini [Wed, 3 Apr 2019 15:28:29 +0000 (08:28 -0700)]
Update Readme to mention that the TensorFlow specific will be upstreamed later
--
PiperOrigin-RevId:
241735845
Alex Zinenko [Wed, 3 Apr 2019 13:11:03 +0000 (06:11 -0700)]
Drop MLIREmitter-based version of the EDSC
This version has been deprecated and can now be removed completely since the
last remaining user (Python bindings) migrated to declarative builders.
Several functions in lib/EDSC/Types.cpp construct core IR objects for the C
bindings. Move these functions into lib/EDSC/CoreAPIs.cpp until we decide
where they should live.
This completes the migration from the delayed-construction EDSC to Declarative
Builders.
--
PiperOrigin-RevId:
241716729
Alex Zinenko [Wed, 3 Apr 2019 12:43:44 +0000 (05:43 -0700)]
Python bindings: drop MLIREmitter and related functionality
This completes the transition of Python bindings to use the declarative
builders infrastructure instead of the now-deprecated EDSC emitter
infrastructure. The relevant unit tests have been replicated using the new
functionality and the remaining end-to-end compilation tests have been updated
accordingly. The latter show an improvement in brevity and readability.
--
PiperOrigin-RevId:
241713489
Alex Zinenko [Wed, 3 Apr 2019 12:39:02 +0000 (05:39 -0700)]
Python bindings: support __floordiv__ for index types
The original reimplementation of EDSC as declarative builders and the
subsequent rework of Python bindings forbade to use the (true) division
operator for values of the index types without providing an alternative. Index
types only support floor and ceil division through affine maps. Expose this to
Python bindings through a `__floordiv__` function on `ValueHandle`s.
--
PiperOrigin-RevId:
241713093
Mehdi Amini [Wed, 3 Apr 2019 08:47:13 +0000 (01:47 -0700)]
Remove section of the README copied from the Rationale document, it does not fit here.
--
PiperOrigin-RevId:
241687723
Nicolas Vasilache [Wed, 3 Apr 2019 07:46:40 +0000 (00:46 -0700)]
Fix test that fails on non-determinism in LowerVectorTransfers
This CL fixes the non-determinism across compilers in an edsc::select expression used in LowerVectorTransfers. This is achieved by factoring the expression out of the function call to ensure a deterministic order of evaluation.
Since the expression is now factored out, fewer IR is generated and the test is updated accordingly.
--
PiperOrigin-RevId:
241679962
Mehdi Amini [Wed, 3 Apr 2019 01:14:56 +0000 (18:14 -0700)]
Fix path for the examples in Toy tutorial Ch1
PiperOrigin-RevId:
241643625
Mehdi Amini [Wed, 3 Apr 2019 01:07:26 +0000 (18:07 -0700)]
Add link to the Toy tutorial in the Readme
--
PiperOrigin-RevId:
241642739
River Riddle [Tue, 2 Apr 2019 22:50:51 +0000 (15:50 -0700)]
Add fake-use for variable in LLVMDialect.cpp to fix opt build.
--
PiperOrigin-RevId:
241620442
Mehdi Amini [Tue, 2 Apr 2019 22:41:22 +0000 (15:41 -0700)]
Fix minor typo (triggered a GCC warning, NFC)
--
PiperOrigin-RevId:
241618796
Alex Zinenko [Tue, 2 Apr 2019 22:33:54 +0000 (15:33 -0700)]
Introduce custom format for the LLVM IR Dialect
Historically, the LLVM IR dialect has been using the generic form of MLIR
operation syntax. It is verbose and often redundant. Introduce the custom
printing and parsing for all existing operations in the LLVM IR dialect.
Update the relevant documentation and tests.
--
PiperOrigin-RevId:
241617393
Mehdi Amini [Tue, 2 Apr 2019 22:15:20 +0000 (15:15 -0700)]
Add -DLLVM_BUILD_EXAMPLES to the mlir build instruction: this enables the tutorial examples
Seems like a better default for getting started.
--
PiperOrigin-RevId:
241614043
Mehdi Amini [Tue, 2 Apr 2019 22:13:26 +0000 (15:13 -0700)]
Fix Toy tutorial gcc 7 build (NFC)
--
PiperOrigin-RevId:
241613710
Nicolas Vasilache [Tue, 2 Apr 2019 21:35:09 +0000 (14:35 -0700)]
Linalg portion of the tutorial - part 3
This CL starts the third part of the Linalg tutorial by adding support for ops to declare how they lower themselves to other ops.
Tests are added that demonstrate matmul lowering to a loop over matvec and matvec lowering to a loop over dot.
This is part of a list of CLs that add new Transforms and Analyses to Linalg3: it iseasier to integrate in small chunks.
As part of working with the TensorContractionBase template class and in an effort to add pieces incrementally without copying code, it is easiest to define operations ahead of time in Linalg2/TensorOps.h and gradually implement them as needed. This CL performs the necessary refactoring for this to happen.
--
PiperOrigin-RevId:
241605869
River Riddle [Tue, 2 Apr 2019 21:02:32 +0000 (14:02 -0700)]
Update the dialect attribute verification hooks to return LogicalResult instead of bool. This updates Function::emitError to return LogicalResult as well.
--
PiperOrigin-RevId:
241598892
Mehdi Amini [Tue, 2 Apr 2019 20:32:53 +0000 (13:32 -0700)]
Provide default `affineMapComposition` and memory space on `MemRefType::get()` (NFC)
This matches the builder API.
--
PiperOrigin-RevId:
241592869
Mehdi Amini [Tue, 2 Apr 2019 20:11:20 +0000 (13:11 -0700)]
Chapter 2 of the Toy tutorial
This introduces a basic MLIRGen through straight AST traversal,
without dialect registration at this point.
--
PiperOrigin-RevId:
241588354
River Riddle [Tue, 2 Apr 2019 20:09:34 +0000 (13:09 -0700)]
Rewrite the verify hooks on operations to use LogicalResult instead of bool. This also changes the return of Operation::emitError/emitOpError to LogicalResult as well.
--
PiperOrigin-RevId:
241588075
Nicolas Vasilache [Tue, 2 Apr 2019 17:32:18 +0000 (10:32 -0700)]
Linalg portion of the tutorial - part 2
The second part of the Linalg tutorial introduces:
1. the TensorContractionBase type from which all tensor contractions derive;
2. a basic set of operations DotOp, MatvecOp and MatmulOp;
3. a helper function `createFullyComposedView` that walks the producers of a SliceOp up until the root ViewOp and returns a single ViewOp;
4. programmatic examples to test MLIR construction involving these types.
This CL also refactors file organization so that:
1. clients only need to include Ops.h and Types.h while keeping independent small files separate for the purpose of the tutorial;
2. each step of the tutorial has its own linalgxxx include directory and each include explicitly states in which part of the tutorial a particular concept was introduced.
Lastly the following cleanups are applied:
1. ValueOrSliceOp is removed in favor of simpler helper function.
2. methods that walk back the chain of ops are removed from the core ops and added to a separate Analysis.
3. various additional cleanups.
--
PiperOrigin-RevId:
241555769
River Riddle [Tue, 2 Apr 2019 17:24:11 +0000 (10:24 -0700)]
Update the Function and Module verifiers to return LogicalResult instead of bool.
--
PiperOrigin-RevId:
241553930
Mehdi Amini [Tue, 2 Apr 2019 17:02:07 +0000 (10:02 -0700)]
Initial version for chapter 1 of the Toy tutorial
--
PiperOrigin-RevId:
241549247
Andy Davis [Tue, 2 Apr 2019 13:37:40 +0000 (06:37 -0700)]
Enable producer-consumer fusion for liveout memrefs if consumer read region matches producer write region.
--
PiperOrigin-RevId:
241517207
River Riddle [Tue, 2 Apr 2019 10:48:45 +0000 (03:48 -0700)]
Update the pass ir-printing test to not rely on rtti type pretty printing.
--
PiperOrigin-RevId:
241498090
River Riddle [Tue, 2 Apr 2019 03:43:13 +0000 (20:43 -0700)]
Remove MLPatternLoweringPass and rewrite LowerVectorTransfers to use RewritePattern instead.
--
PiperOrigin-RevId:
241455472
Lei Zhang [Tue, 2 Apr 2019 00:40:58 +0000 (17:40 -0700)]
[TableGen] Add Confined, IntMinValue, and ArrayMinCount for attribute constraints
This CL introduces Confined as a general mechanism to compose complex attribute
constraints out of more primitive ones. It's particularly useful for automatically
generating op definitions from some external source, where we can have random
combinations of primitive constraints and it would be impractical to define a case
for each of such combination.
Two primitive attribute constraints, IntMinValue and ArrayMinCount, are added to be
used together with Confined.
--
PiperOrigin-RevId:
241435955
Nicolas Vasilache [Mon, 1 Apr 2019 22:15:09 +0000 (15:15 -0700)]
Linalg portion of the tutorial - part 1
The first part of the Linalg tutorial introduces:
1. the RangeType and ViewType;
2. operations on those, namely RangeOp, ViewOp and SliceOp;
3. programmatic examples to test MLIR construction involving these types, ops and affine.for loops (with a mock custom op called "some_consumer").
--
PiperOrigin-RevId:
241409949
Mehdi Amini [Mon, 1 Apr 2019 18:28:27 +0000 (11:28 -0700)]
Use `this->` to access inherited member from dependent type (Fix GCC7 build, NFC)
--
PiperOrigin-RevId:
241364505
Mehdi Amini [Mon, 1 Apr 2019 18:19:02 +0000 (11:19 -0700)]
Minor fixes to README.md (NFC)
In particular fix the links to work on GitHub, and do not include RTTI to build.
--
PiperOrigin-RevId:
241362244
Mehdi Amini [Mon, 1 Apr 2019 17:42:34 +0000 (10:42 -0700)]
Add a getLoc() method on mlir::Value that returns the loc of the defining Operations if any, otherwise an unknown location
--
PiperOrigin-RevId:
241354085
Mehdi Amini [Mon, 1 Apr 2019 17:36:50 +0000 (10:36 -0700)]
Add a getContext() convenient accessor on mlir::Value (NFC)
--
PiperOrigin-RevId:
241352789
Mehdi Amini [Mon, 1 Apr 2019 17:17:45 +0000 (10:17 -0700)]
Fix MacOS build
This is making up for some differences in standard library and linker flags.
It also get rid of the requirement to build with RTTI.
--
PiperOrigin-RevId:
241348845
River Riddle [Mon, 1 Apr 2019 17:03:18 +0000 (10:03 -0700)]
Rewrite ResultTypeIterator to be a mapped_iterator instead of an IndexedAccessorIterator. IndexedAccessorIterator makes some assumptions about the element type that do not hold for the result type, i.e. pointer elements. A mapped_iterator also better models the behavior of a ResultTypeIterator.
--
PiperOrigin-RevId:
241345949
Feng Liu [Mon, 1 Apr 2019 17:01:47 +0000 (10:01 -0700)]
Support 0-d tensor type attributes
This CL fixes the parser and printer to support the 0-d tensor type attributes.
--
PiperOrigin-RevId:
241345329
Feng Liu [Mon, 1 Apr 2019 16:39:59 +0000 (09:39 -0700)]
Minor fixes on the typo/naming/style in the Pattern.cpp file
--
PiperOrigin-RevId:
241341334
Lei Zhang [Mon, 1 Apr 2019 15:58:53 +0000 (08:58 -0700)]
[TableGen] Add EnumAttrCase and EnumAttr
This CL adds EnumAttr as a general mechanism for modelling enum attributes. Right now
it is using StringAttr under the hood since MLIR does not have native support for enum
attributes.
--
PiperOrigin-RevId:
241334043
River Riddle [Mon, 1 Apr 2019 06:30:22 +0000 (23:30 -0700)]
Add a flag to Dialect that allows for dialects to enable support for unregistered operations. This flag is off by default and can be toggled via the 'allowUnknownOperations(...)' method. This means that moving forward an error will be emitted for unknown operations if the dialect does not explicitly allow it.
Example:
func @unknown_std_op() {
%0 = "std.foo_bar_op"() : () -> index
return
}
Will result in:
error: unregistered operation 'std.foo_bar_op' found in dialect ('std') that does not allow unknown operations
--
PiperOrigin-RevId:
241266009
River Riddle [Mon, 1 Apr 2019 05:08:12 +0000 (22:08 -0700)]
Update variable in PassTiming to refer to system_clock instead of high_resolution_clock.
--
PiperOrigin-RevId:
241260071
River Riddle [Sat, 30 Mar 2019 05:30:54 +0000 (22:30 -0700)]
Rename the 'namePrefix' field in the Dialect class to 'name' and tidy some comments to make it clear that 'name' refers to the dialect namespace.
--
PiperOrigin-RevId:
241103116
Chris Lattner [Sat, 30 Mar 2019 05:23:34 +0000 (22:23 -0700)]
Implement basic IR support for a builtin complex<> type. As with tuples, we
have no standard ops for working with these yet, this is simply enough to
represent and round trip them in the printer and parser.
--
PiperOrigin-RevId:
241102728
Jacques Pienaar [Sat, 30 Mar 2019 05:10:12 +0000 (22:10 -0700)]
Add build files and update README.
* Add initial version of build files;
* Update README with instructions to download and build MLIR from github;
--
PiperOrigin-RevId:
241102092
River Riddle [Fri, 29 Mar 2019 21:18:08 +0000 (14:18 -0700)]
Assert that registered dialects have unique names. This creates a guarantee that the namespace of a dialect can be used a unique key.
--
PiperOrigin-RevId:
241049578
River Riddle [Fri, 29 Mar 2019 21:06:51 +0000 (14:06 -0700)]
Remove the MLIRContext parameter from Dialect::parseType. Dialects already have access to the context via Dialect::getContext.
--
PiperOrigin-RevId:
241047077
River Riddle [Fri, 29 Mar 2019 20:59:28 +0000 (13:59 -0700)]
Update TypeBase::verifyConstructionInvariants to use a LogicalResult return instead of bool.
--
PiperOrigin-RevId:
241045568
River Riddle [Fri, 29 Mar 2019 20:15:06 +0000 (13:15 -0700)]
Replace usages of instruction with operation in the g3 documents.
PiperOrigin-RevId:
241037784
Nicolas Vasilache [Fri, 29 Mar 2019 18:48:20 +0000 (11:48 -0700)]
Cleanup SuperVectorization dialect printing and parsing.
On the read side,
```
%3 = vector_transfer_read %arg0, %i2, %i1, %i0 {permutation_map: (d0, d1, d2)->(d2, d0)} : (memref<?x?x?xf32>, index, index, index) -> vector<32x256xf32>
```
becomes:
```
%3 = vector_transfer_read %arg0[%i2, %i1, %i0] {permutation_map: (d0, d1, d2)->(d2, d0)} : memref<?x?x?xf32>, vector<32x256xf32>
```
On the write side,
```
vector_transfer_write %0, %arg0, %c3, %c3 {permutation_map: (d0, d1)->(d0)} : vector<128xf32>, memref<?x?xf32>, index, index
```
becomes
```
vector_transfer_write %0, %arg0[%c3, %c3] {permutation_map: (d0, d1)->(d0)} : vector<128xf32>, memref<?x?xf32>
```
Documentation will be cleaned up in a followup commit that also extracts a proper .md from the top of the file comments.
PiperOrigin-RevId:
241021879
Feng Liu [Fri, 29 Mar 2019 16:54:11 +0000 (09:54 -0700)]
remove the const quantifier before temp variable
PiperOrigin-RevId:
240997262
Nicolas Vasilache [Fri, 29 Mar 2019 16:47:30 +0000 (09:47 -0700)]
Make createMaterializeVectorsPass take a vectorSize parameter - NFC
This CL allows the programmatic control of the target hardware vector size when creating a MaterializeVectorsPass.
This is useful for registering passes for the tutorial.
PiperOrigin-RevId:
240996136
Feng Liu [Fri, 29 Mar 2019 16:36:09 +0000 (09:36 -0700)]
[TableGen] Support benefit score in pattern definition.
A integer number can be specified in the pattern definition and used as the
adjustment to the default benefit score in the generated rewrite pattern C++
definition.
PiperOrigin-RevId:
240994192
Nicolas Vasilache [Fri, 29 Mar 2019 16:34:06 +0000 (09:34 -0700)]
Refactor vectorization patterns
This CL removes the reliance of the vectorize pass on the specification of a `fastestVaryingDim` parameter. This parameter is a restriction meant to more easily target a particular loop/memref combination for vectorization and is mainly used for testing.
This also had the side-effect of restricting vectorization patterns to only the ones in which all memrefs were contiguous along the same loop dimension. This simple restriction prevented matmul to vectorize in 2-D.
this CL removes the restriction and adds the matmul test which vectorizes in 2-D along the parallel loops. Support for reduction loops is left for future work.
PiperOrigin-RevId:
240993827
River Riddle [Fri, 29 Mar 2019 16:18:45 +0000 (09:18 -0700)]
Slight rewording of TupleType rationale.
PiperOrigin-RevId:
240991400
River Riddle [Fri, 29 Mar 2019 15:59:20 +0000 (08:59 -0700)]
Update the QuickstartRewrites document to include information about the new 'matchAndRewrite' functionality in RewritePatterns.
PiperOrigin-RevId:
240987764
River Riddle [Fri, 29 Mar 2019 15:52:06 +0000 (08:52 -0700)]
[PassManager] Add a utility class, PrettyStackTraceParallelDiagnosticEntry, to emit any queued up diagnostics in the event of a crash when multi-threading.
PiperOrigin-RevId:
240986566
MLIR Team [Fri, 29 Mar 2019 15:06:25 +0000 (08:06 -0700)]
Enable input-reuse fusion to search function arguments for fusion candidates (takes care of a TODO, enables another tutorial test case).
PiperOrigin-RevId:
240979894
River Riddle [Fri, 29 Mar 2019 02:53:02 +0000 (19:53 -0700)]
Change the vectorizer test pass to output via diagnostics instead of llvm::outs. This allows for the output to be deterministic when multi-threading is enabled.
PiperOrigin-RevId:
240905858
MLIR Team [Fri, 29 Mar 2019 02:49:31 +0000 (19:49 -0700)]
Support for type constraints across operand and results
--
PiperOrigin-RevId:
240905555
Tatiana Shpeisman [Fri, 29 Mar 2019 01:28:14 +0000 (18:28 -0700)]
Fixed a few instances of inconsistent grammar.
PiperOrigin-RevId:
240896336
Jacques Pienaar [Fri, 29 Mar 2019 01:00:57 +0000 (18:00 -0700)]
Address some errors from g++
These fail with:
could not convert ‘module’ from ‘llvm::orc::ThreadSafeModule’ to
‘llvm::Expected<llvm::orc::ThreadSafeModule>’
PiperOrigin-RevId:
240892583
Jacques Pienaar [Fri, 29 Mar 2019 00:37:29 +0000 (17:37 -0700)]
Add README file for MLIR.
PiperOrigin-RevId:
240889350
River Riddle [Fri, 29 Mar 2019 00:36:06 +0000 (17:36 -0700)]
Remove the LowerEDSCTestPass.
Most of the tests have been ported to be unit-tests and this pass is problematic in the way it depends on TableGen-generated files. This pass is also non-deterministic during multi-threading and a blocker to turning it on by default.
PiperOrigin-RevId:
240889154
River Riddle [Thu, 28 Mar 2019 23:18:52 +0000 (16:18 -0700)]
Tidy up a few comments and error messages related to parsing multi-result operations.
PiperOrigin-RevId:
240876306
Jacques Pienaar [Thu, 28 Mar 2019 23:13:32 +0000 (16:13 -0700)]
Remove extra qualification
PiperOrigin-RevId:
240875432
Alex Zinenko [Thu, 28 Mar 2019 23:10:07 +0000 (16:10 -0700)]
LLVM IR Dialect: separate the conversion tool from the conversion pass
Originally, the conversion to the LLVM IR dialect had been implemented as pass.
The common conversion infrastructure was factored into DialectConversion from
which the conversion pass inherited. The conversion being a pass is
undesirable for callers that only need the conversion done, for example as a
part of sequence of conversions or outside the pass manager infrastructure.
Split the LLVM IR Dialect conversion into the conversion proper and the
conversion pass, where the latter contains the former instead of inheriting.
NFC.
PiperOrigin-RevId:
240874740
Alex Zinenko [Thu, 28 Mar 2019 22:58:53 +0000 (15:58 -0700)]
Dialect Conversion: convert regions of operations when cloning them
Dialect conversion currently clones the operations that did not match any
pattern. This includes cloning any regions that belong to these operations.
Instead, apply conversion recursively to the nested regions.
Note that if an operation matched one of the conversion patterns, it is up to
the pattern rewriter to fill in the regions of the converted operation. This
may require calling back to the converter and is left for future work.
PiperOrigin-RevId:
240872410
Nicolas Vasilache [Thu, 28 Mar 2019 22:12:18 +0000 (15:12 -0700)]
NFC - Handle IndexedValue corner case
Implicit conversion don't play nicely in expressions such as:
`C() = A(i) * B(i)`.
Make `C()` return an IndexedValue instead of casting to ValueHandle.
This prevents double capture errors and is useful for the tutorial.
PiperOrigin-RevId:
240863223
River Riddle [Thu, 28 Mar 2019 21:58:52 +0000 (14:58 -0700)]
Change the muli-return syntax for operations. The name of the operation result now contains the number of results that it refers to if the number of results is greater than 1.
Example:
%call:2 = call @multi_return() : () -> (f32, i32)
use(%calltensorflow/mlir#0, %calltensorflow/mlir#1)
This cl also adds parser support for uniquely named result values. This means that a test writer can now write something like:
%foo, %bar = call @multi_return() : () -> (f32, i32)
use(%foo, %bar)
Note: The printer will still print the collapsed form.
PiperOrigin-RevId:
240860058
MLIR Team [Thu, 28 Mar 2019 21:54:49 +0000 (14:54 -0700)]
Remove overly conservative check in LoopFusion pass (enables fusion in tutorial example).
PiperOrigin-RevId:
240859227
River Riddle [Thu, 28 Mar 2019 18:40:17 +0000 (11:40 -0700)]
[PassManager] Define a ParallelDiagnosticHandler to ensure that diagnostics are still produced in a deterministic order when multi-threading.
PiperOrigin-RevId:
240817922
River Riddle [Thu, 28 Mar 2019 18:25:19 +0000 (11:25 -0700)]
Rename InstOperand to OpOperand.
PiperOrigin-RevId:
240814651
Dimitrios Vytiniotis [Thu, 28 Mar 2019 16:20:13 +0000 (09:20 -0700)]
Remove global LLVM CLI variables from library code
Plus move parsing code into the MLIR CPU runner binary.
PiperOrigin-RevId:
240786709
River Riddle [Thu, 28 Mar 2019 15:24:38 +0000 (08:24 -0700)]
Replace remaining usages of the Instruction class with Operation.
PiperOrigin-RevId:
240777521
Nicolas Vasilache [Thu, 28 Mar 2019 01:04:19 +0000 (18:04 -0700)]
Cleanup vectorize_1d.mlir test - NFC
This CL splits a large monolithic test function into smaller ones that are each CHECK-LABEL'd
PiperOrigin-RevId:
240684979
Nicolas Vasilache [Thu, 28 Mar 2019 00:50:34 +0000 (17:50 -0700)]
Make vectorization aware of loop semantics
Now that we have a dependence analysis, we can check that loops are indeed parallel and make vectorization correct.
PiperOrigin-RevId:
240682727
River Riddle [Thu, 28 Mar 2019 00:43:08 +0000 (17:43 -0700)]
Update the multi-threaded pass timing to not assume that total time will be different from user time.
PiperOrigin-RevId:
240681618
Jacques Pienaar [Thu, 28 Mar 2019 00:33:40 +0000 (17:33 -0700)]
Move test of trait using dialect ops, to dialects of ops.
PiperOrigin-RevId:
240680010
MLIR Team [Wed, 27 Mar 2019 23:16:15 +0000 (16:16 -0700)]
Small edit for clarity. ("Zero dimensions" reads to me as "rank of zero.")
PiperOrigin-RevId:
240664300
Jacques Pienaar [Wed, 27 Mar 2019 21:52:34 +0000 (14:52 -0700)]
Move TF dialect test to dialect.
PiperOrigin-RevId:
240646586
Nicolas Vasilache [Wed, 27 Mar 2019 21:12:01 +0000 (14:12 -0700)]
Give the Vectorize pass a virtualVectorSize argument.
This CL allows vectorization to be called and configured in other ways than just via command line arguments.
This allows triggering vectorization programmatically.
PiperOrigin-RevId:
240638208
Lei Zhang [Wed, 27 Mar 2019 21:09:19 +0000 (14:09 -0700)]
[TableGen] Change names for Builder* and OperationState* parameters to avoid collision
The `Builder*` parameter is unused in both generated build() methods so that we can
leave it unnamed. Changed stand-alone parameter build() to take `_tblgen_state` instead
of `result` to allow `result` to avoid having name collisions with op operand,
attribute, or result.
PiperOrigin-RevId:
240637700
River Riddle [Wed, 27 Mar 2019 21:02:26 +0000 (14:02 -0700)]
Add support for multi-threaded pass timing.
When multi-threading is enabled in the pass manager the meaning of the display
slightly changes. First, a new timing column is added, `User Time`, that
displays the total time spent across all threads. Secondly, the `Wall Time`
column displays the longest individual time spent amongst all of the threads.
This means that the `Wall Time` column will continue to give an indicator on the
perceived time, or clock time, whereas the `User Time` will display the total
cpu time.
Example:
$ mlir-opt foo.mlir -experimental-mt-pm -cse -canonicalize -convert-to-llvmir -pass-timing
===-------------------------------------------------------------------------===
... Pass execution timing report ...
===-------------------------------------------------------------------------===
Total Execution Time: 0.0078 seconds
---User Time--- ---Wall Time--- --- Name ---
0.0175 ( 88.3%) 0.0055 ( 70.4%) Function Pipeline
0.0018 ( 9.3%) 0.0006 ( 8.1%) CSE
0.0013 ( 6.3%) 0.0004 ( 5.8%) (A) DominanceInfo
0.0017 ( 8.7%) 0.0006 ( 7.1%) FunctionVerifier
0.0128 ( 64.6%) 0.0039 ( 50.5%) Canonicalizer
0.0011 ( 5.7%) 0.0004 ( 4.7%) FunctionVerifier
0.0004 ( 2.1%) 0.0004 ( 5.2%) ModuleVerifier
0.0010 ( 5.3%) 0.0010 ( 13.4%) LLVMLowering
0.0009 ( 4.3%) 0.0009 ( 11.0%) ModuleVerifier
0.0198 (100.0%) 0.0078 (100.0%) Total
PiperOrigin-RevId:
240636269
River Riddle [Wed, 27 Mar 2019 21:02:02 +0000 (14:02 -0700)]
Replace usages of Instruction with Operation in the Transforms/ directory.
PiperOrigin-RevId:
240636130
Mehdi Amini [Wed, 27 Mar 2019 20:57:02 +0000 (13:57 -0700)]
Simplify API uses of `getContext()` (NFC)
The Pass base class is providing a convenience getContext() accessor.
PiperOrigin-RevId:
240634961
Mehdi Amini [Wed, 27 Mar 2019 20:38:48 +0000 (13:38 -0700)]
Allow to mutate the type of MLIR Value in-place
This avoid trashing memory by cloning and replaceAllUseswith when
performing type inference.
PiperOrigin-RevId:
240631137
Jacques Pienaar [Wed, 27 Mar 2019 20:24:05 +0000 (13:24 -0700)]
Fix include path in test pass.
PiperOrigin-RevId:
240628260
Jacques Pienaar [Wed, 27 Mar 2019 19:36:57 +0000 (12:36 -0700)]
Initialize std::atomic directly.
Avoids error in OSS build:
error: copying variable of type 'std::atomic<unsigned int>' invokes deleted constructor
PiperOrigin-RevId:
240618765
Mehdi Amini [Wed, 27 Mar 2019 19:20:51 +0000 (12:20 -0700)]
Add a method to swap the type of a function in-place
This is motivated by the need to translate function across dialect which
requires morphing their type, as well as the Toy tutorial part on interprocedural
shape inference.
The alternative is cloning the function, but it is heavy and it seems like an
arbitrary restriction to forbid morphing the function type.
PiperOrigin-RevId:
240615755
Jacques Pienaar [Wed, 27 Mar 2019 18:36:36 +0000 (11:36 -0700)]
Use dereference instead of implicit conversion for IndexedValue to Value*.
Avoids ambiguous constructor error on some compilers.
PiperOrigin-RevId:
240606838
Jacques Pienaar [Wed, 27 Mar 2019 17:38:23 +0000 (10:38 -0700)]
Add missing numeric header for std::accumulate.
PiperOrigin-RevId:
240593135
Alex Zinenko [Wed, 27 Mar 2019 16:39:31 +0000 (09:39 -0700)]
LLVM IR Conversion: support zero-dimensional memrefs
The spec allows zero-dimensional memrefs to exist and treats them essentially
as single-element buffers. Unlike single-dimensional memrefs of static shape
<1xTy>, zero-dimensional memrefs do not require indices to access the only
element they store. Add support of zero-dimensional memrefs to the LLVM IR
conversion. In particular, such memrefs are converted into bare pointers, and
accesses to them are converted to bare loads and stores, without the overhead
of `getelementptr %buffer, 0`.
PiperOrigin-RevId:
240579456
Alex Zinenko [Wed, 27 Mar 2019 16:16:27 +0000 (09:16 -0700)]
LLVM IR Conversion: keep LLVM dialect types as is during conversion
When converting to the LLVM IR Dialect, it is possible for the input IR to
contain LLVM IR Dialect operation and/or types, for example, some functions may
have been coverted to the LLVM IR Dialect already, or may have been created
using this dialect directly. Make sure that type conversion keeps LLVM IR
Dialect types unmodified and does not error out. Operations are already kept
as is.
PiperOrigin-RevId:
240574972
River Riddle [Wed, 27 Mar 2019 15:55:17 +0000 (08:55 -0700)]
Replace usages of Instruction with Operation in the /Analysis directory.
PiperOrigin-RevId:
240569775