Nicolas Vasilache [Tue, 25 Jun 2019 19:02:56 +0000 (12:02 -0700)]
Internal change
PiperOrigin-RevId:
255018853
Andy Davis [Tue, 25 Jun 2019 17:29:53 +0000 (10:29 -0700)]
Support printing SSA ids in affine.load/store which do not have special names.
PiperOrigin-RevId:
254997746
River Riddle [Tue, 25 Jun 2019 16:42:52 +0000 (09:42 -0700)]
Update the OperationFolder to find a valid insertion point when materializing constants.
The OperationFolder currently just inserts into the entry block of a Function, but regions may be isolated above, i.e. explicit capture only, and blindly inserting constants may break the invariants of these regions.
PiperOrigin-RevId:
254987796
Alex Zinenko [Tue, 25 Jun 2019 16:04:13 +0000 (09:04 -0700)]
GPUtoNVVM: adjust integer bitwidth when lowering special register ops
GPU dialect operations (launch and launch_func) use `index` type for thread and
block index values inside the kernel, for compatibility with affine loops.
NVVM dialect operations, following the NVVM intrinsics, use `!llvm.i32` type,
which does not necessarily have the same bit width as the lowered `index` type.
Optionally sign-extend (indices are signed) or truncate the result of the NVVM
dialect operation to the bit width of the lowered `index` type before passing
it to other operations. This behavior is consistent with `std.index_cast`. We
cannot use the latter since we are targeting LLVM dialect types directly,
rather than standard integer types.
PiperOrigin-RevId:
254980868
Stephan Herhut [Tue, 25 Jun 2019 13:44:58 +0000 (06:44 -0700)]
Add gpu::GPUDialect::isKernel helper.
Also some mild cleanup of the kernel to cubin conversion pass.
PiperOrigin-RevId:
254959303
Nicolas Vasilache [Tue, 25 Jun 2019 13:35:28 +0000 (06:35 -0700)]
Add missing dependence to MLIRLLVMIR
PiperOrigin-RevId:
254957980
Alex Zinenko [Tue, 25 Jun 2019 13:30:32 +0000 (06:30 -0700)]
NVVM target: emit nvvm.annotations for kernel functions
PTX backend in LLVM expects additional module-level metadata
`!nvvm.annotations` that lists functions that can be used as GPU kernels.
Generate this metadata based on the `gpu.kernel` attribute attached to
functions. This attribute is added automatically by the kernel outlining pass
in the GPU dialect lowering flow.
PiperOrigin-RevId:
254957345
Alex Zinenko [Tue, 25 Jun 2019 11:21:10 +0000 (04:21 -0700)]
gpu.launch_func: add accessors to grid configuration operands
Add accessor functions that return `gpu::KernelDim3` containing the respective
operands for grid and block size accepted by `gpu.launch_func`. Use the same
signature as for `gpu.launch`.
PiperOrigin-RevId:
254942674
Mehdi Amini [Tue, 25 Jun 2019 00:51:44 +0000 (17:51 -0700)]
Print both type in diagnostic emitted on mismatch between def/use of SSA Values (NFC)
The error would look like:
path/filename.mlir:32:23: error: use of value '%28' expects different type than prior uses: ''i32'' vs ''!_tf.control''
PiperOrigin-RevId:
254874859
River Riddle [Tue, 25 Jun 2019 00:36:05 +0000 (17:36 -0700)]
NFC: Move the ArgConverter methods out-of-line to improve readability.
PiperOrigin-RevId:
254872695
Nicolas Vasilache [Mon, 24 Jun 2019 22:23:26 +0000 (15:23 -0700)]
Fix OSS build
PiperOrigin-RevId:
254847773
Nicolas Vasilache [Mon, 24 Jun 2019 21:35:21 +0000 (14:35 -0700)]
Split test-specific passes out of mlir-opt
Instead put their impl in test/lib and link them into mlir-test-opt
PiperOrigin-RevId:
254837439
River Riddle [Mon, 24 Jun 2019 19:03:44 +0000 (12:03 -0700)]
Cache several common LLVMTypes in the LLVMDialect.
LLVM is not thread-safe which means that several of the 'get' methods for LLVMType must be double locked to ensure thread-safety. This cl adds static caching, i.e. no lookups or locking, for several simple LLVM types(i1, half, void, etc.). It also cleans up the implementation of the double locking that is required for some types. In the future we could add a form of dynamic caching to only need to lock one mutex in the best case, but that requires analysis on the memory overhead/vs time lost to taking two locks.
PiperOrigin-RevId:
254806747
Mahesh Ravishankar [Mon, 24 Jun 2019 17:59:05 +0000 (10:59 -0700)]
Add SPIR-V Load/Store operations. Currently this only support memory
operands being None, Volatile, Aligned and Nontemporal
PiperOrigin-RevId:
254792353
River Riddle [Mon, 24 Jun 2019 16:22:07 +0000 (09:22 -0700)]
NFC: Simplify Operation::getContext to use the context within the location.
PiperOrigin-RevId:
254771979
Nicolas Vasilache [Mon, 24 Jun 2019 16:12:47 +0000 (09:12 -0700)]
Cleanup test following bad merge
PiperOrigin-RevId:
254770395
Nicolas Vasilache [Mon, 24 Jun 2019 16:00:06 +0000 (09:00 -0700)]
Use linalg.view_slice in tiling and fusion
This CL makes use of view_slice in tiling and fusion.
Using a higher level IR element greatly simplifies the IR produced during tiling and fusion.
Lowering to LLVM is updated to first translate view_slice into a sequence of dim, range and cmpi.
This level will also be useful when lowering to affine.
PiperOrigin-RevId:
254767814
Jacques Pienaar [Mon, 24 Jun 2019 15:56:57 +0000 (08:56 -0700)]
Remove leftover change from splitting mlir-opt change.
PiperOrigin-RevId:
254767366
Jacques Pienaar [Mon, 24 Jun 2019 15:41:52 +0000 (08:41 -0700)]
Split out mlir-opt main into separate file.
Enable reusing the real mlir-opt main from unit tests and in case where
additional initialization needs to happen before main is invoked (e.g., when
using different command line flag libraries).
PiperOrigin-RevId:
254764575
Nicolas Vasilache [Mon, 24 Jun 2019 14:59:59 +0000 (07:59 -0700)]
Add higher-level linalg.view_slice operation.
This will be useful to simplify the IR emitted during transformations as well as lowering to affine.
PiperOrigin-RevId:
254757641
Andy Davis [Mon, 24 Jun 2019 14:31:52 +0000 (07:31 -0700)]
Add parsing/printing for new affine.load and affine.store operations.
The new operations affine.load and affine.store will take composed affine maps by construction.
These operations will eventually replace load and store operations currently used in affine regions and operated on by affine transformation and analysis passes.
PiperOrigin-RevId:
254754048
River Riddle [Mon, 24 Jun 2019 08:27:22 +0000 (01:27 -0700)]
Update CSE to respect nested regions that are isolated from above. This cl also removes the unused 'NthRegionIsIsolatedFromAbove' trait as it was replaced with a more general 'IsIsolatedFromAbove'.
PiperOrigin-RevId:
254709704
Jacques Pienaar [Sat, 22 Jun 2019 22:06:17 +0000 (15:06 -0700)]
Change string to std::string.
PiperOrigin-RevId:
254580672
River Riddle [Sat, 22 Jun 2019 21:39:16 +0000 (14:39 -0700)]
Allow converting Diagnostic to LogicalResult::Failure. This matches the behavior of InFlightDiagnostic and enables notes to be converted to failure.
PiperOrigin-RevId:
254579098
Jacques Pienaar [Sat, 22 Jun 2019 21:29:23 +0000 (14:29 -0700)]
Linalg1LLVMConversion should depend on MLIRLLVMIR to ensure ops generated.
PiperOrigin-RevId:
254578530
River Riddle [Sat, 22 Jun 2019 18:48:43 +0000 (11:48 -0700)]
Add a new dialect hook 'materializeConstant' to create a constant operation that materializes an attribute value with the given type. This effectively adds support for dialect specific constant values that have different invariants than std.constant. 'OperationFolder' is updated to use this new hook, or attempt to default to std.constant when legal.
PiperOrigin-RevId:
254570153
River Riddle [Sat, 22 Jun 2019 18:08:52 +0000 (11:08 -0700)]
NFC: Remove the 'context' parameter from OperationState.
Now that Locations are Attributes they contain a direct reference to the MLIRContext, i.e. the context can be directly accessed from the given location instead of being explicitly passed in.
PiperOrigin-RevId:
254568329
Jacques Pienaar [Sat, 22 Jun 2019 15:02:00 +0000 (08:02 -0700)]
Add missing build dep.
PiperOrigin-RevId:
254557652
Feng Liu [Sat, 22 Jun 2019 04:43:49 +0000 (21:43 -0700)]
Add int8 predicates to the tf lite ops
PiperOrigin-RevId:
254519710
River Riddle [Sat, 22 Jun 2019 03:20:27 +0000 (20:20 -0700)]
Define a ModuleOp that represents a Module as an Operation.
The ModuleOp contains a single region that must contain a single block. This block must be terminated by a new pseudo operation 'module_terminator'. The syntax for this operations is as follows:
`module` (`attributes` attr-dict)? region
Example:
module {
...
}
module attributes { ... } {
...
}
PiperOrigin-RevId:
254513752
Nicolas Vasilache [Sat, 22 Jun 2019 01:55:02 +0000 (18:55 -0700)]
Remove dead function
PiperOrigin-RevId:
254507641
Nicolas Vasilache [Sat, 22 Jun 2019 01:31:52 +0000 (18:31 -0700)]
Add a Linalg convolution op.
This CL adds a conv op that corresponds to the TF description along with its lowering to loops (https://www.tensorflow.org/api_docs/python/tf/nn/convolution).
The dimension of the convolution is inferred from the rank of the views. The other logical
dimensions correspond to the TF description.
The computation of tiled views need to be updated to work for the input tensor. This is left for a future CL.
PiperOrigin-RevId:
254505644
River Riddle [Sat, 22 Jun 2019 01:27:49 +0000 (18:27 -0700)]
Refactor the location classes to be attributes instead of separate IR classes.
This will allow for locations to be used in the same contexts as attributes. Given that attributes are nullable types, the 'Location' class now represents a non-nullable wrapper around a 'LocationAttr'. This preserves the desired semantics we have for non-optional locations.
PiperOrigin-RevId:
254505278
River Riddle [Sat, 22 Jun 2019 00:00:01 +0000 (17:00 -0700)]
Simplify PredecessorIterator by using mapped_iterator.
PiperOrigin-RevId:
254495164
Lei Zhang [Fri, 21 Jun 2019 21:51:58 +0000 (14:51 -0700)]
[spirv] Basic serializer and deserializer
This CL adds the basic SPIR-V serializer and deserializer for converting
SPIR-V module into the binary format and back. Right now only an empty
module with addressing model and memory model is supported; (de)serialize
other components will be added gradually with subsequent CLs.
The purpose of this library is to enable importing SPIR-V binary modules
to run transformations on them and exporting SPIR-V modules to be consumed
by execution environments. The focus is transformations, which inevitably
means changes to the binary module; so it is not designed to be a general
tool for investigating the SPIR-V binary module and does not guarantee
roundtrip equivalence (at least for now).
PiperOrigin-RevId:
254473019
River Riddle [Fri, 21 Jun 2019 20:45:21 +0000 (13:45 -0700)]
NFC: Drop the unnecessary dependence on TransformsUtils from Parser. This is a historical dependency that is no longer needed.
PiperOrigin-RevId:
254460518
River Riddle [Fri, 21 Jun 2019 20:43:38 +0000 (13:43 -0700)]
Define the 'reference' type for the remaining value-typed iterators. This allows for them to be used with llvm::enumerate and other various iterator utilities.
PiperOrigin-RevId:
254460201
River Riddle [Fri, 21 Jun 2019 17:26:57 +0000 (10:26 -0700)]
Define the 'reference' type for the OperandType and ResultType iterators so that they can be used with llvm::enumerate.
PiperOrigin-RevId:
254422623
Krzysztof Drewniak [Fri, 21 Jun 2019 17:23:18 +0000 (10:23 -0700)]
Clarify documentation for MLIR to indicate that the build instructions run the test suite.
PiperOrigin-RevId:
254421795
River Riddle [Fri, 21 Jun 2019 16:29:46 +0000 (09:29 -0700)]
Add support for 1->N type mappings in the dialect conversion infrastructure. To support these mappings a hook must be overridden on the type converter: 'materializeConversion' :to generate a cast operation from the new types to the old type. This operation is automatically erased if all uses are removed, otherwise it remains in the IR for the user to handle.
PiperOrigin-RevId:
254411383
River Riddle [Fri, 21 Jun 2019 16:20:42 +0000 (09:20 -0700)]
Cache instances of several common attributes(e.g. BoolAttr, UnitAttr) and types(I1/I16/I32/etc.) when creating the MLIRContext. This allows for these symbols to be accessed without the need to perform any lookups/locking.
PiperOrigin-RevId:
254410080
Nicolas Vasilache [Fri, 21 Jun 2019 12:34:58 +0000 (05:34 -0700)]
Support for 0-D case in Linalg ops
This CL adds support for O-D ops in Linalg ops by:
1. making the CopyOp maps optional instead of default valued
2. allowing certain map operations to accept and return empty maps
3. making linalg::LowerToLoops aware of these changes
4. providing a proper 0-D impl for CopyOp and FillOp
5. adding the relevant tests
PiperOrigin-RevId:
254381908
River Riddle [Fri, 21 Jun 2019 10:26:39 +0000 (03:26 -0700)]
Add an overload to 'PatternRewriter::inlineRegionBefore' that accepts a parent region for the insertion position. This allows for inlining the given region into the end of another region.
PiperOrigin-RevId:
254367375
Mehdi Amini [Fri, 21 Jun 2019 07:44:11 +0000 (00:44 -0700)]
Add missing override on overridden virtual method (NFC)
PiperOrigin-RevId:
254349019
Nicolas Vasilache [Thu, 20 Jun 2019 22:10:35 +0000 (15:10 -0700)]
Uniformize usage of OpBuilder& (NFC)
Historically the pointer-based version of builders was used.
This CL uniformizes to OpBuilder &
PiperOrigin-RevId:
254280885
Mahesh Ravishankar [Thu, 20 Jun 2019 20:19:41 +0000 (13:19 -0700)]
Fix a warning in enum value ecoding for SPIRV image types. Instead of
using the enum class in the packed struct, use unsigned value for
encoding and convert it lazily
PiperOrigin-RevId:
254258341
Mehdi Amini [Thu, 20 Jun 2019 15:58:24 +0000 (08:58 -0700)]
TensorFlow operands can be broadcasted with each other and with the result, relax the verifier.
PiperOrigin-RevId:
254206814
Lei Zhang [Thu, 20 Jun 2019 12:59:19 +0000 (05:59 -0700)]
Add definition for OperandElementTypeIterator and ResultElementTypeIterator
These are useful utility iterators helping use to get the element types of
operands/results of shaped types.
Also defined ranges for these iterators.
PiperOrigin-RevId:
254180888
River Riddle [Thu, 20 Jun 2019 06:39:06 +0000 (23:39 -0700)]
Rename ConversionTarget::isLegal to isDynamicallyLegal to better represent what the function is actually checking.
PiperOrigin-RevId:
254141073
Mahesh Ravishankar [Wed, 19 Jun 2019 23:09:57 +0000 (16:09 -0700)]
Add SPIRV Image Type according to the spec described here :
https://www.khronos.org/registry/spir-v/specs/1.0/SPIRV.html#OpTypeImage.
Add new enums to describe Image dimensionality, Image Depth, Arrayed
information, Sampling, Sampler User information, and Image format.
Doesn's support the Optional Access qualifier at this stage
Fix Enum generator for tblgen to add "_" at the beginning if the enum
starts with a number.
PiperOrigin-RevId:
254091423
Lei Zhang [Wed, 19 Jun 2019 21:32:07 +0000 (14:32 -0700)]
Print proper message saying variadic ops are not supported in RewriterGen
Support for ops with variadic operands/results will come later; but right now
a proper message helps to avoid deciphering confusing error messages later in
the compilation stage.
PiperOrigin-RevId:
254071820
Lei Zhang [Wed, 19 Jun 2019 21:28:19 +0000 (14:28 -0700)]
Add missing MLIRStandardOps dependency for MLIRGPU
PiperOrigin-RevId:
254070969
River Riddle [Wed, 19 Jun 2019 20:58:31 +0000 (13:58 -0700)]
Refactor the TypeConverter to support more robust type conversions:
* Support for 1->0 type mappings, i.e. when the argument is being removed.
* Reordering types when converting a type signature.
* Adding new inputs when converting a type signature.
This cl also lays down the initial foundation for supporting 1->N type mappings, but full support will come in a followup.
Moving forward, function signature changes will be driven by populating a SignatureConversion instance. This class contains all of the necessary information for adding/removing/remapping function signatures; e.g. addInputs, addResults, remapInputs, etc.
PiperOrigin-RevId:
254064665
River Riddle [Wed, 19 Jun 2019 18:55:27 +0000 (11:55 -0700)]
Extract the function symbol table functionality, i.e. mapping and name uniquing, out of Module and into a new class SymbolTable. As modules become operations it is necessary to extract out this functionality that cannot be represented with a generic operation.
PiperOrigin-RevId:
254041734
Geoffrey Martin-Noble [Wed, 19 Jun 2019 18:30:16 +0000 (11:30 -0700)]
Remove unnecessary -verify-diagnostics
These were likely added in error because of confusion about the flag when it was just called "-verify". The extra flag doesn't cause much harm, but it does make mlir-opt do more work and clutter the RUN line
PiperOrigin-RevId:
254037016
Geoffrey Martin-Noble [Wed, 19 Jun 2019 18:21:41 +0000 (11:21 -0700)]
Rename -verify mlir-opt flag to -verify-expected-diagnostics
This name has caused some confusion because it suggests that it's running op verification (and that this verification isn't getting run by default).
PiperOrigin-RevId:
254035268
Lei Zhang [Wed, 19 Jun 2019 18:13:36 +0000 (11:13 -0700)]
Export symbols in cpu runner cblas library
By default MSVC does not export any symbol and does not create a companion
.lib for a .dll. This will cause problems when trying to link against the
library.
PiperOrigin-RevId:
254033454
Stephan Herhut [Wed, 19 Jun 2019 13:22:36 +0000 (06:22 -0700)]
Add a pass that translates GPU.launch_func into a series of runtime calls.
This does not map the calls to the CUDA libary directly but uses a slim wrapper
ABI on top that has more convenient types for code generation and is stable. Such
ABI is expected to be provided by the actual runner.
PiperOrigin-RevId:
253983833
River Riddle [Wed, 19 Jun 2019 01:26:26 +0000 (18:26 -0700)]
Simplify usages of SplatElementsAttr now that it inherits from DenseElementsAttr.
PiperOrigin-RevId:
253910543
River Riddle [Tue, 18 Jun 2019 23:41:00 +0000 (16:41 -0700)]
NFC: Reorder the attribute classes alphabetically to improve readability.
PiperOrigin-RevId:
253894445
River Riddle [Tue, 18 Jun 2019 20:56:54 +0000 (13:56 -0700)]
NFC: Append 'Location' to the end of each the location kinds. This is in preparation for making the location classes attributes instead of separate IR types.
PiperOrigin-RevId:
253860058
River Riddle [Tue, 18 Jun 2019 20:35:02 +0000 (13:35 -0700)]
Replace usages of 'UniquedFilename' with 'Identifier' and remove it. Identifier already contains all of the necessary functionality/verification, so having a separate class for filenames is unnecessary.
PiperOrigin-RevId:
253855505
Lei Zhang [Tue, 18 Jun 2019 18:15:55 +0000 (11:15 -0700)]
[spirv] Add spv.Variable
This is a direct modelling of SPIR-V's OpVariable. The custom assembly format
parsers/prints descriptor in a nicer way if presents. There are other common
decorations that can appear on variables like builtin, which can be supported
later.
This CL additionally deduplicates the parser/printer/verifier declaration
in op definitions by adding defaults to SPV_Op base.
by adding
PiperOrigin-RevId:
253828254
Mehdi Amini [Tue, 18 Jun 2019 16:38:16 +0000 (09:38 -0700)]
Add a setAttrList() method on mlir::Operation
This is an efficient method to copy attributes from one operation to
another.
PiperOrigin-RevId:
253806004
Alex Zinenko [Tue, 18 Jun 2019 16:12:32 +0000 (09:12 -0700)]
Fix GPUToNVVM naming: NNVM should have been NVVM
Rename `createLowerGpuOpsToNNVMOpsPass` to `createLowerGpuOpsToNVVMOpsPass`.
PiperOrigin-RevId:
253801577
Andy Davis [Tue, 18 Jun 2019 15:52:09 +0000 (08:52 -0700)]
Factor fusion compute cost calculation out of LoopFusion and into LoopFusionUtils (NFC).
PiperOrigin-RevId:
253797886
Jacques Pienaar [Tue, 18 Jun 2019 14:50:40 +0000 (07:50 -0700)]
Add missing cmake dependency from Linalg to MLIRStandardToLLVM.
PiperOrigin-RevId:
253788170
Alex Zinenko [Tue, 18 Jun 2019 14:19:38 +0000 (07:19 -0700)]
Put createGpuKernelOutliningPass in MLIR namespace
This function was declared in the `mlir` namespace but defined in the global
namespace, leading to linking errors when used.
PiperOrigin-RevId:
253784410
Alex Zinenko [Tue, 18 Jun 2019 12:01:30 +0000 (05:01 -0700)]
Make examples/Linalg3 depend on the new standard to LLVM conversion library.
PiperOrigin-RevId:
253767820
Stephan Herhut [Tue, 18 Jun 2019 12:00:13 +0000 (05:00 -0700)]
Use llvm::StringSwitch in lowering of GPU ops to NVVM ops.
PiperOrigin-RevId:
253767688
Stephan Herhut [Tue, 18 Jun 2019 10:26:31 +0000 (03:26 -0700)]
Add a pass that translates a CUDA kernel function (tagged with nvvm.kernel) to
a CUBIN blob for execution on CUDA GPUs.
This is a first in a series of patches to build a simple CUDA runner to allow
experimenting with MLIR code on GPUs.
PiperOrigin-RevId:
253758915
Stephan Herhut [Tue, 18 Jun 2019 09:24:16 +0000 (02:24 -0700)]
Add mlir::Function::eraseBody helper to remove the body of a function.
This does not add any new functionality but gives clearing the blocks of the
body region a more obvious name.
PiperOrigin-RevId:
253751382
River Riddle [Tue, 18 Jun 2019 02:46:31 +0000 (19:46 -0700)]
Fix the detection of boolean splat values in DenseElementsAttr for arrays with <=15 values.
PiperOrigin-RevId:
253711372
Mehdi Amini [Tue, 18 Jun 2019 00:24:51 +0000 (17:24 -0700)]
Add a TableGen Type predicate to check that an operand type can be broadcasted to a particular output type
PiperOrigin-RevId:
253694397
Mehdi Amini [Mon, 17 Jun 2019 22:31:53 +0000 (15:31 -0700)]
Refactor generic op printing: extract a public printFunctionalType() on OpAsmPrinter (NFC)
PiperOrigin-RevId:
253674584
Lei Zhang [Mon, 17 Jun 2019 21:47:22 +0000 (14:47 -0700)]
[spirv] Add spv.constant
This CL defines a single `spv.constant` op to represent various
flavors of SPIR-V constant instructions.
PiperOrigin-RevId:
253665716
Lei Zhang [Mon, 17 Jun 2019 20:29:06 +0000 (13:29 -0700)]
[spirv] Define common types using op definition spec
This CL also tightens spv.FMul to only accept 16/32/64-bit floats.
PiperOrigin-RevId:
253649352
Stephan Herhut [Mon, 17 Jun 2019 19:47:09 +0000 (12:47 -0700)]
Verification for gpu.launch_func should rely on function type and not on the
actual body of a kernel function. This enables using gpu.launch_func with
external kernel declarations.
PiperOrigin-RevId:
253639770
Lei Zhang [Mon, 17 Jun 2019 18:50:16 +0000 (11:50 -0700)]
Also consider attributes when getting context for Operation
This CL also updates to use containing region as a fallback way to find
context since functions will eventually become ops with regions.
PiperOrigin-RevId:
253627322
Lei Zhang [Mon, 17 Jun 2019 18:42:37 +0000 (11:42 -0700)]
[spirv] Add doc regarding the SPIR-V dialect
* Add basic design philosophy
* Add type syntax to both the doc and type parser comments
PiperOrigin-RevId:
253625782
Alex Zinenko [Mon, 17 Jun 2019 18:35:05 +0000 (11:35 -0700)]
Introduce std.index_cast and its lowering+translation to LLVM
Index types integers of platform-specific bit width. They are used to index
memrefs and as loop induction variables, however they could not be obtained
from an integer until now, making it virtually impossible to express indirect
accesses (given that memrefs of indices are not allowed) or data-dependent
loops. Introduce `std.index_cast` to transform indices into integers and vice
versa. The semantics of this cast is to sign-extend when casting to a wider
integer, and to truncate when casting to a narrower integer. It belongs to
StandardOps because both types it operates on are standard types, and because
its results are likely to be used in std.load and std.store.
Introduce llvm.sext, llvm.zext and llvm.trunc operations to the LLVM dialect.
Provide the conversion of `std.index_cast` to llvm.sext or llvm.trunc,
depending on the actual bitwidth of `index` known during the conversion.
PiperOrigin-RevId:
253624100
Alex Zinenko [Mon, 17 Jun 2019 18:30:16 +0000 (11:30 -0700)]
Factor Region::getUsedValuesDefinedAbove into Transforms/RegionUtils
Arguably, this function is only useful for transformations and should not
pollute the main IR. Also make sure it accepts a the resulting container
by-reference instead of returning it.
PiperOrigin-RevId:
253622981
Andy Davis [Mon, 17 Jun 2019 16:59:35 +0000 (09:59 -0700)]
LoopFusion: adds support for computing forward computation slices, which will enable fusion of consumer loop nests into their producers in subsequent CLs.
PiperOrigin-RevId:
253601994
Stephan Herhut [Mon, 17 Jun 2019 10:56:20 +0000 (03:56 -0700)]
Add lowering pass from GPU dialect operations to LLVM/NVVM intrinsics.
PiperOrigin-RevId:
253551452
Chris Lattner [Sun, 16 Jun 2019 03:19:36 +0000 (20:19 -0700)]
Move the Region type out to its own .h/.cpp file instead of putting it into
Block.h/cpp.
This doesn't change much but makes it easier to find.
PiperOrigin-RevId:
253423041
River Riddle [Sat, 15 Jun 2019 02:54:56 +0000 (19:54 -0700)]
Use braces when building an 'llvm_return' via EDSC intrinsics. This fixes a build error on MSVC where it is unable to properly handled template type aliases.
PiperOrigin-RevId:
253339204
Jing Pu [Fri, 14 Jun 2019 23:06:49 +0000 (16:06 -0700)]
Remove dead code.
PiperOrigin-RevId:
253314416
Alex Zinenko [Fri, 14 Jun 2019 22:54:21 +0000 (15:54 -0700)]
Start moving conversions to {lib,include/mlir}/Conversion
Conversions from dialect A to dialect B depend on both A and B. Therefore, it
is reasonable for them to live in a separate library that depends on both
DialectA and DialectB library, and does not forces dependees of DialectA or
DialectB to also link in the conversion. Create the directory layout for the
conversions and move the Standard to LLVM dialect conversion as the first
example.
PiperOrigin-RevId:
253312252
River Riddle [Fri, 14 Jun 2019 18:56:41 +0000 (11:56 -0700)]
Disallow using NOperands/NResults when N < 2. We have special traits for the case of 0/1 that we explicitly check for throughout the codebase. This also fixes weird build failures in MSVC where it doesn't properly handle template type aliases.
PiperOrigin-RevId:
253269936
Nicolas Vasilache [Fri, 14 Jun 2019 17:18:08 +0000 (10:18 -0700)]
Add Linalg CopyOp
This CL adds a generic CopyOp to Linalg and its lowering to loops.
The CopyOp supports input and output permutation maps.
When combined with tiling and allocating a new local buffer, this should provide basic support for implementing simple memory transfers with coalescing.
At the moment, lowering copies to a library call is not supported.
PiperOrigin-RevId:
253250497
River Riddle [Fri, 14 Jun 2019 15:41:32 +0000 (08:41 -0700)]
Add an overloaded 'get' method to DenseElementsAttr that accepts an initializer_list.
PiperOrigin-RevId:
253234385
Alex Zinenko [Fri, 14 Jun 2019 14:28:41 +0000 (07:28 -0700)]
Explicitly construct ArrayRef in AttributeTest.cpp
Some compilers find initializer list constructors from boolean literals
ambiguous between ArrayRef<bool> and ArrayRef<Attribute>. Call the
ArrayRef<bool> constructor explicitly to disambiguate.
PiperOrigin-RevId:
253224859
Alex Zinenko [Fri, 14 Jun 2019 08:56:19 +0000 (01:56 -0700)]
Convert a nest affine loops to a GPU kernel
This converts entire loops into threads/blocks. No check on the size of the
block or grid, or on the validity of parallelization is performed, it is under
the responsibility of the caller to strip-mine the loops and to perform the
dependence analysis before calling the conversion.
PiperOrigin-RevId:
253189268
River Riddle [Fri, 14 Jun 2019 00:24:33 +0000 (17:24 -0700)]
Refactor SplatElementsAttr to inherit from DenseElementsAttr as opposed to being a separate Attribute type. DenseElementsAttr provides a better internal representation for splat values as well as better API for accessing elements.
PiperOrigin-RevId:
253138287
River Riddle [Thu, 13 Jun 2019 23:09:56 +0000 (16:09 -0700)]
NFC: Fix a narrowing conversion from size_t to int64_t when constructing a VectorType.
PiperOrigin-RevId:
253125435
River Riddle [Thu, 13 Jun 2019 23:02:36 +0000 (16:02 -0700)]
NFC: Fix a warning for casting away const qualifiers.
PiperOrigin-RevId:
253124057
Geoffrey Martin-Noble [Thu, 13 Jun 2019 21:52:29 +0000 (14:52 -0700)]
Add ability to verify type matching between operands/results
This extends and generalizes the functionality for checking that element types match
PiperOrigin-RevId:
253110512
Mahesh Ravishankar [Thu, 13 Jun 2019 20:47:08 +0000 (13:47 -0700)]
Add a definition of the library function to use when Linalg ops are
lowered to LLVM, instead of expecting one to exist in the Module
PiperOrigin-RevId:
253097382
River Riddle [Thu, 13 Jun 2019 20:22:32 +0000 (13:22 -0700)]
Add several utility 'getValues<T>' functions to DenseElementsAttr that return ranges as opposed to filling a SmallVector. This is much more efficient for the general case and allows for avoiding constructing APInt/APFloat/Attribute when possible.
PiperOrigin-RevId:
253092550
Alex Zinenko [Thu, 13 Jun 2019 14:08:46 +0000 (07:08 -0700)]
Disallow non-index operands and results in affine.apply
`affine.apply` is supposed to operate on values of index types in context of
affine loops. It is possible to programmatically constuct an `affine.apply`
that takes values of other types as operands or returns them, but it would not
be parseable. Disallow such cases in the verifier.
PiperOrigin-RevId:
253021704