Geoffrey Martin-Noble [Sat, 1 Jun 2019 00:18:59 +0000 (17:18 -0700)]
Consistently use int64_t for shape-related values in shaped types
We want to support 64-bit shapes (even when the compiler is on a 32-bit architecture). Using int64_t consistently allows us to sidestep the bugginess of unsigned arithmetic.
Still unsigned: kind, memory space, and bit width. The first two are basically enums. We could have a discussion about the last one, but it's basically just a very large enum as well and we're not doing any math on it, I think.
--
PiperOrigin-RevId:
250985791
Geoffrey Martin-Noble [Fri, 31 May 2019 23:41:21 +0000 (16:41 -0700)]
Static cast size_t -> int64_t instead of vice versa for equals comparisons
These were just introduced by a previous CL moving MemRef getRank to return int64_t. size_t could be smaller than 64 bits and in equals comparisons, signed vs unsigned doesn't matter. In these cases, we know right now that the particular int64_t is not larger than max size_t (because it currently comes directly from a size() call), the alternative cast plus equals comparison is always safe, so we might as well do it that way and no longer require reasoning deeper into the callstack.
We are already assuming that size() calls fit into int64_t in a number of other cases like the aforementioned getRank() (since exabytes of RAM are rare). If we want to avoid this assumption we will have to come up with a principled way to do it throughout.
--
PiperOrigin-RevId:
250980297
River Riddle [Fri, 31 May 2019 23:03:26 +0000 (16:03 -0700)]
Add support for providing an output stream to the SourceMgrDiagnosticHandlers.
--
PiperOrigin-RevId:
250974331
Amit Sabne [Fri, 31 May 2019 20:56:47 +0000 (13:56 -0700)]
Loop invariant code motion - remove reliance on getForwardSlice. Add more tests.
--
PiperOrigin-RevId:
250950703
River Riddle [Fri, 31 May 2019 20:48:43 +0000 (13:48 -0700)]
NFC: Cleanup method definitions within Parser and add header blocks to improve readability.
--
PiperOrigin-RevId:
250949195
Mehdi Amini [Fri, 31 May 2019 20:46:23 +0000 (13:46 -0700)]
Fix CMake builds: sources not used in every targets must be marked with LLVM_OPTIONAL_SOURCES
--
PiperOrigin-RevId:
250948796
Geoffrey Martin-Noble [Fri, 31 May 2019 20:28:37 +0000 (13:28 -0700)]
Get rid of separate getRank() on MemRef now that it subclasses ShapedType
We are moving towards int64_t for these shape/dimension -related values to avoid buginess with unsigned arithmetic
--
PiperOrigin-RevId:
250945322
Geoffrey Martin-Noble [Fri, 31 May 2019 20:28:19 +0000 (13:28 -0700)]
Some cleanup of ShapedType now that MemRef subclasses it.
Extract common methods into ShapedType.
Simplify methods.
Remove some extraneous asserts.
Replace sentinel value with a helper method to check the same.
--
PiperOrigin-RevId:
250945261
Geoffrey Martin-Noble [Fri, 31 May 2019 20:28:01 +0000 (13:28 -0700)]
Make MemRefType subclass ShapedType
MemRefs have the same notion of shape, rank, and fixed element type. This allows us to reuse utilities based on shape for memref.
All dyn_cast and isa calls for ShapedType have been checked and either modified to explicitly check for vector or tensor, or confirmed to not depend on the result being a vector or tensor.
Discussion in https://groups.google.com/a/tensorflow.org/forum/#!topic/mlir/cHLoyfGu8y8
--
PiperOrigin-RevId:
250945184
Geoffrey Martin-Noble [Fri, 31 May 2019 20:01:45 +0000 (13:01 -0700)]
Don't use ShapedType to indicate vector or tensor type
MemRefType may soon subclass ShapedType. ShapedType only guarantees that something has a shape (possibly dynamic), rank (or explicitly unranked), and fixed element type.
--
PiperOrigin-RevId:
250940537
River Riddle [Fri, 31 May 2019 19:53:22 +0000 (12:53 -0700)]
Fix an msan warning for use of an uninitialized variable.
--
PiperOrigin-RevId:
250939301
River Riddle [Fri, 31 May 2019 16:27:52 +0000 (09:27 -0700)]
Remove the newline from the mlir-opt 'split-input-file' flag marker. This fixes support for DOS style new lines(/r/n).
--
PiperOrigin-RevId:
250899420
River Riddle [Fri, 31 May 2019 16:24:48 +0000 (09:24 -0700)]
Abstract the internal storage of the NamedAttributeList into a new attribute, DictionaryAttr. This attribute maintains a sorted list of NamedAttributes. This will allow for operations/functions to maintain sub dictionaries of attributes.
The syntax is the same as top level attribute dictionaries:
{sub_dictionary: {fn: @someFn, boolAttr: true}}
--
PiperOrigin-RevId:
250898950
Jacques Pienaar [Fri, 31 May 2019 01:42:16 +0000 (18:42 -0700)]
Disable named attribute in ODG for ArgOrResultElementTypeIs as was taking address of r-value.
--
PiperOrigin-RevId:
250805965
Lei Zhang [Thu, 30 May 2019 23:50:16 +0000 (16:50 -0700)]
[ODS] Support region names and constraints
Similar to arguments and results, now we require region definition in ops to
be specified as a DAG expression with the 'region' operator. This way we can
specify the constraints for each region and optionally give the region a name.
Two kinds of region constraints are added, one allowing any region, and the
other requires a certain number of blocks.
--
PiperOrigin-RevId:
250790211
Geoffrey Martin-Noble [Thu, 30 May 2019 23:02:49 +0000 (16:02 -0700)]
Replace checks against numDynamicDims with hasStaticShape
--
PiperOrigin-RevId:
250782165
River Riddle [Thu, 30 May 2019 21:58:28 +0000 (14:58 -0700)]
Replace a usage of std::vector with SmallVector to allow constructing with non-constant iterators on MSVC.
--
PiperOrigin-RevId:
250769027
Geoffrey Martin-Noble [Thu, 30 May 2019 18:01:17 +0000 (11:01 -0700)]
Make getRank abort for unranked type
This better matches the other methods in ShapedType which only make sense for ranked types. There's now an explicit hasRank for checking the rank. Actual call sites rarely used the "-1" sentinel to combine checking for rankedness and checking that rank is a certain value. And in most cases they should actually be checking for rankedness at a higher level using type predicates. Using an explicit method is clearer than a sentinel anyway.
--
PiperOrigin-RevId:
250720853
Jacques Pienaar [Thu, 30 May 2019 17:57:23 +0000 (10:57 -0700)]
Allow argument and result names replacement in predicates.
This allow specifying $x to refer to an operand's named argument (operand or attribute) or result. Skip variadic operands/results for now pending autogenerated discussion of their accessors.
This adds a new predicate, following feedback on the naming but does not remove the old one. Post feedback I'll do that, potentially in follow up.
--
PiperOrigin-RevId:
250720003
Jacques Pienaar [Thu, 30 May 2019 17:02:57 +0000 (10:02 -0700)]
Avoid unspecified files error by explicitly including for now.
--
PiperOrigin-RevId:
250708336
Jacques Pienaar [Thu, 30 May 2019 16:42:17 +0000 (09:42 -0700)]
Add keywords in target_link_libraries post add_llvm_executable.
--
PiperOrigin-RevId:
250704528
River Riddle [Thu, 30 May 2019 15:23:49 +0000 (08:23 -0700)]
Replace usages of 'add_executable' with 'add_llvm_executable'.
--
PiperOrigin-RevId:
250691487
Jacques Pienaar [Thu, 30 May 2019 04:15:43 +0000 (21:15 -0700)]
Fix 5 ClangTidy - Readability findings.
* the 'empty' method should be used to check for emptiness instead of 'size'
* using decl 'CapturableHandle' is unused
* redundant get() call on smart pointer
* using decl 'apply' is unused
* using decl 'ScopeGuard' is unused
--
PiperOrigin-RevId:
250623863
River Riddle [Thu, 30 May 2019 03:31:12 +0000 (20:31 -0700)]
NFC: Inline toString as operations can be streamed directly into raw_ostream.
--
PiperOrigin-RevId:
250619765
Geoffrey Martin-Noble [Thu, 30 May 2019 00:15:51 +0000 (17:15 -0700)]
Use hasRank instead of comparing rank to -1
--
PiperOrigin-RevId:
250598252
River Riddle [Wed, 29 May 2019 23:46:17 +0000 (16:46 -0700)]
Move CondBranchOp to the ODG framework.
--
PiperOrigin-RevId:
250593367
Geoffrey Martin-Noble [Wed, 29 May 2019 23:07:17 +0000 (16:07 -0700)]
Exclude all ShapedType subclasses other than TensorType subclasses from having non-scalar elements.
The current logic assumes that ShapedType indicates a vector or tensor, which will not be true soon when MemRef subclasses ShapedType
--
PiperOrigin-RevId:
250586364
Geoffrey Martin-Noble [Wed, 29 May 2019 23:06:49 +0000 (16:06 -0700)]
Check for Vector or Tensor Type rather than ShapedType when matching constant splat
Only vectors or tensors can be constant splats (or element attrs). This is in preparation for making MemRef subclass ShapedType
--
PiperOrigin-RevId:
250586281
Geoffrey Martin-Noble [Wed, 29 May 2019 22:51:17 +0000 (15:51 -0700)]
Avoid dyn_cast to ShapedType
ShapedType just indicates shape/rank/element semantics. It's generally not useful for other type checking. This check already checks for vector or tensor type, so we can use a direct cast if we check those first.
Related to making MemRefType a subclass of ShapedType
--
PiperOrigin-RevId:
250583231
Geoffrey Martin-Noble [Wed, 29 May 2019 22:34:50 +0000 (15:34 -0700)]
Make it clear that ElementsAttr is only for static shaped vectors or tensors.
This is in preparation for making MemRef a subclass of ShapedType, but also UnrankedTensor should already be excluded.
--
PiperOrigin-RevId:
250580197
MLIR Team [Wed, 29 May 2019 21:56:41 +0000 (14:56 -0700)]
Remove "size" property of affine maps.
--
PiperOrigin-RevId:
250572818
Andy Davis [Wed, 29 May 2019 21:02:14 +0000 (14:02 -0700)]
LoopFusionUtils CL 2/n: Factor out and generalize slice union computation.
*) Factors slice union computation out of LoopFusion into Analysis/Utils (where other iteration slice utilities exist).
*) Generalizes slice union computation to take the union of slices computed on all loads/stores pairs between source and destination loop nests.
*) Fixes a bug in FlatAffineConstraints::addSliceBounds where redundant constraints were added.
*) Takes care of a TODO to expose FlatAffineConstraints::mergeAndAlignIds as a public method.
--
PiperOrigin-RevId:
250561529
Geoffrey Martin-Noble [Wed, 29 May 2019 20:58:10 +0000 (13:58 -0700)]
Change "may never" to "must never"
"may never" indicates that it's allowed to never hold a value. I believe the intent here is that it is not allowed to ever hold a value.
--
PiperOrigin-RevId:
250560623
River Riddle [Wed, 29 May 2019 20:47:24 +0000 (13:47 -0700)]
Add whole_archive_link support for MSVC.
This issue was discovered by @loic-joly-sonarsource with a proposed fix
in https://github.com/tensorflow/mlir/pull/22.
--
PiperOrigin-RevId:
250558504
Geoffrey Martin-Noble [Wed, 29 May 2019 20:46:41 +0000 (13:46 -0700)]
Change elements literal parsing to not rely on shaped type being a vector or tensor.
This is in preparation for making MemRef a ShapedType. In general, a shaped type should be anything with shape, rank, and element type properties, so use sites shouldn't assume more than that.
I also pulled the trailing comma parsing out the parseElementsLiteralType (new name) method. It seems weird to have the method parse the type + a trailing comma, even if all call sites currently need that. It's surprising behavior without looking at the implementation.
--
PiperOrigin-RevId:
250558363
River Riddle [Wed, 29 May 2019 20:35:44 +0000 (13:35 -0700)]
Disable the mlir-cpu-runner tests on win32. llvm orc does not support the COFF rtld.
--
PiperOrigin-RevId:
250556116
River Riddle [Wed, 29 May 2019 20:33:22 +0000 (13:33 -0700)]
Replace the Function reference methods from the OpAsmParser/OpAsmPrinter with usages of FunctionAttr.
--
PiperOrigin-RevId:
250555680
River Riddle [Wed, 29 May 2019 20:23:47 +0000 (13:23 -0700)]
Add an example of accessing the impl storage to the 'DefiningAttributesAndTypes' document.
--
PiperOrigin-RevId:
250553898
River Riddle [Wed, 29 May 2019 20:00:14 +0000 (13:00 -0700)]
Remove the Toy/* file path from the location check lines. This assumes a specific file path format that breaks on MSVC.
--
PiperOrigin-RevId:
250549223
River Riddle [Wed, 29 May 2019 18:59:26 +0000 (11:59 -0700)]
NFC: Switch std::vector to SmallVector. This fixes a compiler error on MSVC as the StringMap key dereference method is non-const.
--
PiperOrigin-RevId:
250538085
Ben Vanik [Wed, 29 May 2019 18:40:05 +0000 (11:40 -0700)]
[TableGen] Making printer support $cppClass substitution (similar to parser).
--
PiperOrigin-RevId:
250534216
Lei Zhang [Wed, 29 May 2019 17:47:16 +0000 (10:47 -0700)]
[spirv] Add ModuleOp
This op defines a SPIR-V module using a MLIR region. The region contains
one block. Module-level operations, including functions definitions,
are all placed in this block.
This CL extracts common definitions from SPIRVOps.td into SPIRVBase.td.
The new op is placed in SPIRVStructureOps.td.
--
PiperOrigin-RevId:
250522320
Rasmus Munk Larsen [Wed, 29 May 2019 16:59:21 +0000 (09:59 -0700)]
Get rid of redundant verifier, which is implied by the type traits.
--
PiperOrigin-RevId:
250511694
Rasmus Munk Larsen [Wed, 29 May 2019 16:22:30 +0000 (09:22 -0700)]
Add a rank op to MLIR. Example:
%1 = rank %0 : index
--
PiperOrigin-RevId:
250505411
Nicolas Vasilache [Wed, 29 May 2019 16:12:47 +0000 (09:12 -0700)]
Add lowering linalg.for to LLVM IR
This CL adds lowering of linalg.for to LLVM IR and adds an IR test.
This also replaces the usage of affine.for with linalg.for and enables the LLVM IR path in the integration test.
--
PiperOrigin-RevId:
250503798
Alex Zinenko [Wed, 29 May 2019 12:47:15 +0000 (05:47 -0700)]
EDSC: use llvm::function_ref instead of std::function
Region body constructors in EDSC now take a callback to the function that fills
in the body. This callback is called immediately and not stored, so it is
sufficient to pass a reference to it and avoid a potentially expensive copy.
--
PiperOrigin-RevId:
250473793
Nicolas Vasilache [Tue, 28 May 2019 22:05:51 +0000 (15:05 -0700)]
Add a linalg.for operation to support non-affine loop constructs
The affine.for operation has restrictions that make it suitable for dependence analysis. The Linalg dialect aims at being more general.
This CL introduces linalg.for, and its associated terminator, along with a simple roundtripping test.
A `linalg.for` only takes one value of index type for lower bound, upper bound and step.
Example usage:
```
linalg.for %iv = %lb to %ub step %step {
... // body
}
```
--
PiperOrigin-RevId:
250369722
Lei Zhang [Tue, 28 May 2019 19:14:26 +0000 (12:14 -0700)]
Remove unused IsSingleResult template
--
PiperOrigin-RevId:
250335025
River Riddle [Tue, 28 May 2019 17:23:06 +0000 (10:23 -0700)]
NFC: Replace Function::args_iterator with Block::args_iterator.
--
PiperOrigin-RevId:
250311059
Jacques Pienaar [Tue, 28 May 2019 17:21:25 +0000 (10:21 -0700)]
Fix mismatched-tags warning.
--
PiperOrigin-RevId:
250310618
River Riddle [Tue, 28 May 2019 15:57:38 +0000 (08:57 -0700)]
Tidy up a few parser functions in the ModuleParser. This also adds a missing error for attribute aliases that contain '.' characters.
--
PiperOrigin-RevId:
250291646
Lei Zhang [Tue, 28 May 2019 15:03:46 +0000 (08:03 -0700)]
[ODS] Support numRegions in Op definition
--
PiperOrigin-RevId:
250282024
Alex Zinenko [Tue, 28 May 2019 14:41:17 +0000 (07:41 -0700)]
Do not assume Blocks belong to Functions
Fix Block::splitBlock and Block::eraseFromFunction that erronously assume
blocks belong to functions. They now belong to regions. When splitting, new
blocks should be created in the same region as the existing block. When
erasing a block, it should be removed from the region rather than from the
function body that transitively contains the region.
Also rename Block::eraseFromFunction to Block::erase for consistency with other
IR containers.
--
PiperOrigin-RevId:
250278272
Stephan Herhut [Tue, 28 May 2019 11:45:50 +0000 (04:45 -0700)]
Fix translation of NVVM special registers to intrinsics.
The original implementation did not map the return value of the intrinsics
call to the result value of the special register op. Uses of the result
value hence hit a nullpointer.
--
PiperOrigin-RevId:
250255436
Alex Zinenko [Tue, 28 May 2019 07:49:23 +0000 (00:49 -0700)]
Decouple affine->standard lowering from the pass
The lowering from the Affine dialect to the Standard dialect was originally
implemented as a standalone pass. However, it may be used by other passes
willing to lower away some of the affine constructs as a part of their
operation. Decouple the transformation functions from the pass infrastructure
and expose the entry point for the lowering.
Also update the lowering functions to use `LogicalResult` instead of bool for
return values.
--
PiperOrigin-RevId:
250229198
Jacques Pienaar [Tue, 28 May 2019 04:18:19 +0000 (21:18 -0700)]
Fix correspondence between trait names in ODS and C++ class names.
Make the correspondence between the ODS and C++ side clearer.
--
PiperOrigin-RevId:
250211194
Jacques Pienaar [Tue, 28 May 2019 03:04:56 +0000 (20:04 -0700)]
Change pattern test to use TestDialect instead.
Verify pattern specification, added benefit, named pattern and location recording using TestDialect. Naming is verified via explicitly adding named pattern to TestPatternDriver pass. Refactoring test to verify the desired functionality rather than generated code.
--
PiperOrigin-RevId:
250205618
Jacques Pienaar [Tue, 28 May 2019 00:13:55 +0000 (17:13 -0700)]
Use SourceMgrDiagnosticHandler in mlir-translate for translations from MLIR input.
Report errors using the file and line location using SourceMgr's diagnostic reporting. Reduce some horizontal white spacing too.
--
PiperOrigin-RevId:
250193646
River Riddle [Mon, 27 May 2019 16:05:17 +0000 (09:05 -0700)]
Move the 'is_detected' utility out of StorageUniquer and into STLExtras to allow reuse throughout the codebase.
--
PiperOrigin-RevId:
250160508
Jacques Pienaar [Mon, 27 May 2019 07:51:27 +0000 (00:51 -0700)]
Avoid unused variable warning.
--
PiperOrigin-RevId:
250109802
River Riddle [Mon, 27 May 2019 02:49:57 +0000 (19:49 -0700)]
NFC: Add a missing include for std::isalnum and std::digit.
--
PiperOrigin-RevId:
250085298
River Riddle [Sun, 26 May 2019 20:01:39 +0000 (13:01 -0700)]
NFC: Cleanup a switch statement that only has a default case.
--
PiperOrigin-RevId:
250062529
River Riddle [Sun, 26 May 2019 15:28:28 +0000 (08:28 -0700)]
Add a 'getDialectNamespace' utility to a few dialects.
--
PiperOrigin-RevId:
250049416
Lei Zhang [Sun, 26 May 2019 12:43:20 +0000 (05:43 -0700)]
[spirv] Start the SPIR-V dialect
This CL sets up the basic structure for a SPIR-V dialect: operation
definition specification, dialect registration, testing, etc.
A single op, FMul, is defined and tested to showcase.
The SPIR-V dialect aims to be a simple proxy for the SPIR-V binary format
to enable straightforward and lightweight conversion from/to the binary
format. Ops in this dialect should stay as the same semantic level and
try to be a mechanical mapping to the corresponding SPIR-V instructions;
but they can deviate representationally to allow using MLIR mechanisms.
--
PiperOrigin-RevId:
250040830
River Riddle [Sun, 26 May 2019 00:22:27 +0000 (17:22 -0700)]
Add a templated wrapper around RewritePattern that allows for defining match/rewrite methods with an instance of the source op instead of a raw Operation*.
--
PiperOrigin-RevId:
250003405
Jacques Pienaar [Sat, 25 May 2019 18:03:51 +0000 (11:03 -0700)]
Use fused location for rewritten ops in generated rewrites.
This does tracks the location by recording all the ops in the source pattern and using the fused location for the transformed op. Track the locations via the rewrite state which is a bit heavy weight, in follow up to change to matchAndRewrite this will be addressed (and need for extra array go away).
--
PiperOrigin-RevId:
249986555
Jacques Pienaar [Sat, 25 May 2019 18:03:25 +0000 (11:03 -0700)]
Make scope explicit to avoid misleading-indentation warnings.
--
PiperOrigin-RevId:
249986537
Jacques Pienaar [Sat, 25 May 2019 17:55:20 +0000 (10:55 -0700)]
Address some build warnings.
--
PiperOrigin-RevId:
249986120
River Riddle [Sat, 25 May 2019 02:35:23 +0000 (19:35 -0700)]
Add support to RewritePattern for specifying the potential operations that can be generated during a rewrite. This will enable analyses to start understanding the possible effects of applying a rewrite pattern.
--
PiperOrigin-RevId:
249936309
Jacques Pienaar [Sat, 25 May 2019 02:31:09 +0000 (19:31 -0700)]
Fix incorrect result type inference for nested tuples & remove hacky const case.
Builders can be further refined but wanted to address bug where the result type was used beyond the first depth.
--
PiperOrigin-RevId:
249936004
River Riddle [Sat, 25 May 2019 02:22:00 +0000 (19:22 -0700)]
Update the type printer for Diagnostic to automatically wrap the type with ''.
--
PiperOrigin-RevId:
249935489
River Riddle [Sat, 25 May 2019 01:49:45 +0000 (18:49 -0700)]
Update the m_Constant matcher to enable matching derived attribute types.
--
PiperOrigin-RevId:
249933184
River Riddle [Sat, 25 May 2019 01:17:50 +0000 (18:17 -0700)]
Rename DialectConversion to TypeConverter and split out pattern construction. This simplifies building the conversion pattern list from multiple sources.
--
PiperOrigin-RevId:
249930583
River Riddle [Sat, 25 May 2019 01:01:38 +0000 (18:01 -0700)]
Move the definitions of LoadOp and StoreOp to the ODG framework.
--
PiperOrigin-RevId:
249928980
River Riddle [Sat, 25 May 2019 01:01:20 +0000 (18:01 -0700)]
Move the definitions of CmpIOp, CmpFOp, and SelectOp to the ODG framework.
--
PiperOrigin-RevId:
249928953
Geoffrey Martin-Noble [Sat, 25 May 2019 00:46:58 +0000 (17:46 -0700)]
Replace checks for rank -1 with direct calls to hasRank
Also removed a redundant check for rank after already checking for static shape (which implies rank)
--
PiperOrigin-RevId:
249927636
Geoffrey Martin-Noble [Sat, 25 May 2019 00:44:21 +0000 (17:44 -0700)]
Limit the number of places where shaped type has to explicitly reference its base classes.
Introduces a hasRank() method to make checking for rank a bit easier.
This is partially to make it easier to make MemRef subclass ShapedType
--
PiperOrigin-RevId:
249927442
Jacques Pienaar [Fri, 24 May 2019 23:17:52 +0000 (16:17 -0700)]
Use TestDialect to test traits instead of unittest.
--
PiperOrigin-RevId:
249916947
Geoffrey Martin-Noble [Fri, 24 May 2019 22:31:53 +0000 (15:31 -0700)]
Add a type-constrained nested tuple type.
This is useful for dialects that use tuples but only support a subset of types.
--
PiperOrigin-RevId:
249910133
Lei Zhang [Fri, 24 May 2019 21:39:05 +0000 (14:39 -0700)]
Add TestLoopFusion.cpp to CMakeLists.txt
--
PiperOrigin-RevId:
249901490
River Riddle [Fri, 24 May 2019 20:28:55 +0000 (13:28 -0700)]
Add operand type iterators to Operation and cleanup usages of operand->getType. This also simplifies some lingering usages of result->getType.
--
PiperOrigin-RevId:
249889174
Andy Davis [Fri, 24 May 2019 20:17:44 +0000 (13:17 -0700)]
Add LoopFusionUtils.cpp to CMakeLists.
--
PiperOrigin-RevId:
249887371
Andy Davis [Fri, 24 May 2019 17:54:22 +0000 (10:54 -0700)]
Affine Loop Fusion Utility Module (1/n).
*) Adds LoopFusionUtils which will expose a set of loop fusion utilities (e.g. dependence checks, fusion cost/storage reduction, loop fusion transformation) for use by loop fusion algorithms. Support for checking block-level fusion-preventing dependences is added in this CL (additional loop fusion utilities will be added in subsequent CLs).
*) Adds TestLoopFusion test pass for testing LoopFusionUtils at a fine granularity.
*) Adds unit test for testing dependence check for block-level fusion-preventing dependences.
--
PiperOrigin-RevId:
249861071
River Riddle [Fri, 24 May 2019 17:36:16 +0000 (10:36 -0700)]
NFC: Rename DialectConversionPattern to ConversionPattern.
--
PiperOrigin-RevId:
249857277
Alex Zinenko [Fri, 24 May 2019 14:00:29 +0000 (07:00 -0700)]
Detemplatize convertRegion in DialectConversion
Originally, FunctionConverter::convertRegion in the DialectConversion framework
was implemented as a function template because it was creating a new region in
the parent object, which could have been an op or a function. Since
DialectConversion now operates in place, new region is no longer created so
there is no need for convertRegion to be aware of the parent, only of the error
reporting location.
--
PiperOrigin-RevId:
249826392
Stephan Herhut [Fri, 24 May 2019 11:49:56 +0000 (04:49 -0700)]
Add support for llvm.constant with StringAttr as value.
These are translated to an llvm::ConstantDataArray on translation to llvm IR
proper.
--
PiperOrigin-RevId:
249813111
River Riddle [Fri, 24 May 2019 00:01:16 +0000 (17:01 -0700)]
Move explicit calls to Module::getNamedFunction outside of the operations that contain FunctionAttr. It is up to the users of operations to query the module for a specific function referenced by a FunctionAttr.
--
PiperOrigin-RevId:
249743109
River Riddle [Thu, 23 May 2019 23:16:34 +0000 (16:16 -0700)]
Introduce a new common diagnostic handler ScopedDiagnosticHandler to simplify saving and restoring the currently registered handler.
--
PiperOrigin-RevId:
249735912
Jacques Pienaar [Thu, 23 May 2019 23:08:52 +0000 (16:08 -0700)]
Add pattern file location to generated code to trace origin of pattern.
--
PiperOrigin-RevId:
249734666
Geoffrey Martin-Noble [Thu, 23 May 2019 22:43:11 +0000 (15:43 -0700)]
Make TupleOf take a list of allowed types.
This better matches other container types. Seems better to do that, even though tuples are a little different, since they don't have a single element type.
Also fixed its description to mention the element type.
--
PiperOrigin-RevId:
249730341
River Riddle [Thu, 23 May 2019 22:11:19 +0000 (15:11 -0700)]
Add a new TestDialect directory in tests/. This directory defines a fake 'TestDialect' that allows for the use of FileCheck to test things that aren't currently used anywhere else in tree. As a first order, this should simplify the tests used for tablegen components revolving around operation constraints/patterns.
--
PiperOrigin-RevId:
249724328
Geoffrey Martin-Noble [Thu, 23 May 2019 21:41:32 +0000 (14:41 -0700)]
Rename Float/Integer to AnyFloat/AnyInteger
This emphasizes that it is potentially less constrained than you might desire (especially dialects will frequently not support all bit widths), and better matches the other type names, especially the container types.
--
PiperOrigin-RevId:
249718409
Nicolas Vasilache [Thu, 23 May 2019 20:00:16 +0000 (13:00 -0700)]
Fix Linalg tiling for the partial tile case.
This CL prepares for mixing lowering of tiled linalg operations to loops with load and store operations. In particular it is necessary to capture partial tile information in views. This CL makes slice ops during Linalg tiling properly stop at partial tile boundaries by implementing `min` with a `cmpi` and `select` over values of index type.
To be consistent with lowering to loops, the implementation of tiling also drops specifics of accessing values via ranges and instead uses ranges of the form
`[0, dim(view), 1]` for creating view slices. This simplifies the code for the implementation of tiling and utils.
This also allows removing restrictions around needing a View or SliceOp defined in the current function context (as well as all it RangeOps). The restriction removal is tested by making the dot test operate directly on views.
The above is still subject to folding of the linalg.dim operation left for a future CL.
At this time, mixing tiling and lowering to loops all the way to execution is not yet functional because affine.for does not allow arbitrarily defined values of index type as its operands.
The previously introduced linalg.range_intersection was not sufficient to capture the necessary information and still required dealing with max quantities.
A followup CL will remove linalg.range_intersection.
--
PiperOrigin-RevId:
249698823
River Riddle [Thu, 23 May 2019 19:51:13 +0000 (12:51 -0700)]
Remove a stale declaration in FunctionAttr.
--
PiperOrigin-RevId:
249697345
Rob Suderman [Thu, 23 May 2019 18:22:13 +0000 (11:22 -0700)]
Added the ability to run a mapping function across the values of an elements
attr. This supports both the SplatElementsAttr and DenseElementsAttr for both
float and integer inputs / outputs.
--
PiperOrigin-RevId:
249681056
River Riddle [Thu, 23 May 2019 18:11:18 +0000 (11:11 -0700)]
Apply operation rewrites before updating arguments.
--
PiperOrigin-RevId:
249678839
River Riddle [Thu, 23 May 2019 16:23:33 +0000 (09:23 -0700)]
Decouple running a conversion from the DialectConversion class. The DialectConversion class is only necessary for type signature changes(block arguments or function arguments). This isn't always desired when performing a dialect conversion. This allows for those conversions without this need to run per function instead of per module.
--
PiperOrigin-RevId:
249657549
Nicolas Vasilache [Thu, 23 May 2019 14:00:18 +0000 (07:00 -0700)]
Fix Linalg lowering to loops
This CL makes lowering to loops always be a:
```
%D = linalg.dim %view, constant : !linalg.view<...>
affine.for %ix = %c0 to %D {
...
}
```
This form composes correctly with tiling and is also the proper way to emit loops from views that across function boundaries.
The previous version that would extract the range_min/max/step was composing incorrectly with tiling (i.e. would shift by range_min both in the loop bounds and in the slice) and would not work across function boundaries.
The relevant tests are updated and a new test `dot_view`---which lowers to loops from views passed as function parameters---is added.
When additional context is available, the linalg.dim operations should be folded away but this is left for a future CL.
--
PiperOrigin-RevId:
249634712
Thomas Joerg [Thu, 23 May 2019 09:16:18 +0000 (02:16 -0700)]
Automated rollback of changelist
247713812.
PiperOrigin-RevId:
249605627
Stephan Herhut [Thu, 23 May 2019 09:03:14 +0000 (02:03 -0700)]
Add LLVM::IntToPtrOp and LLVM::PtrToIntOp to LLVM dialect.
--
PiperOrigin-RevId:
249604199