[mlir][TilingInterface] Handle multi operand consumer fusion. (llvm#145193)

For consumer fusion cases of this form

```
%0:2 = scf.forall .. shared_outs(%arg0 = ..., %arg0 = ...) {

  tensor.parallel_insert_slice ... into %arg0
  tensor.parallel_insert_slice ... into %arg1
}
%1 = linalg.generic ... ins(%0#0, %0#1)
```

the current consumer fusion that handles one slice at a time cannot fuse
the consumer into the loop, since fusing along one slice will create and
SSA violation on the other use from the `scf.forall`. The solution is to
allow consumer fusion to allow considering multiple slices at once. This
PR changes the `TilingInterface` methods related to consumer fusion,
i.e.

- `getTiledImplementationFromOperandTile`
- `getIterationDomainFromOperandTile`

to allow fusion while considering multiple operands. It is upto the
`TilingInterface` implementation to return an error if a list of tiles
of the operands cannot result in a consistent implementation of the
tiled operation.

The Linalg operation implementation of `TilingInterface` has been
modified to account for these changes and allow cases where operand
tiles that can result in a consistent tiling implementation are handled.

---------

Signed-off-by: MaheshRavishankar <mahesh.ravishankar@gmail.com>

Author: MaheshRavishankar

mlir/include/mlir/Dialect/SCF/Transforms/TileUsingInterface.h

@@ -313,19 +313,23 @@ tileConsumerAndFuseProducersUsingSCF(RewriterBase &rewriter,
                                      TilingInterface consumer,
                                      const SCFTileAndFuseOptions &options);
 
-/// Fuse the consumer of the source of `candidateSliceOp` by computing the
-/// required slice of the consumer in-place.  Note that the method
-/// replaces the uses of `candidateSliceOp` with the tiled and fused consumer
-/// value but does not delete the slice operation.
+/// Fuse the consumer `candidateSlices` by computing the required slice of the
+/// consumer in-place. All the entries of `candidateSlices` are expected to map
+/// to the same consumer. The method returns an error if the consumer cannot be
+/// tiled in a manner that is consistent for all the passed slices. Note that
+/// the method replaces the uses of `candidateSlices` with the tiled and fused
+/// consumer value but does not delete the slice operations.
 struct SCFFuseConsumerOfSliceResult {
-  OpOperand *origConsumerOperand; // Original untiled consumer's operand.
-  OpOperand
-      *tiledAndFusedConsumerOperand; // Tiled and fused consumer's operand.
+  // Original untiled consumer operands.
+  SmallVector<OpOperand *> origConsumerOperands;
+  // Tiled and fused consumer operands.
+  SmallVector<OpOperand *> tiledAndFusedConsumerOperands;
   SmallVector<Operation *> tiledOps;
 };
 FailureOr<scf::SCFFuseConsumerOfSliceResult>
-tileAndFuseConsumerOfSlice(RewriterBase &rewriter, Operation *candidateSliceOp,
-                           MutableArrayRef<LoopLikeOpInterface> loops);
+tileAndFuseConsumerOfSlices(RewriterBase &rewriter,
+                            ArrayRef<Operation *> candidateSlices,
+                            MutableArrayRef<LoopLikeOpInterface> loops);
 
 /// Method to lower an `op` that implements the `TilingInterface` to
 /// loops/scalars.

mlir/include/mlir/Dialect/Tensor/Transforms/Transforms.h

@@ -31,12 +31,16 @@ namespace tensor {
 FailureOr<TilingResult> replaceExtractSliceWithTiledProducer(
     OpBuilder &builder, tensor::ExtractSliceOp sliceOp, OpResult producerOp);
 
-/// Method to swap an `tensor.insert_slice` with its consumer when the
-/// consumer implements the `TilingInterface`.
+/// Method to swap `tensor.insert_slice`s with their consumers when the
+/// consumer implements the `TilingInterface`. The size of `sliceOps` and
+/// `consumerOperands` is expected to be the same. Every entry in
+/// `consumerOperands` represents a use of the the corresponding
+/// entry in `sliceOps` in the consumer. All entries of `consumerOperands` is
+/// expected to be uses in the same consumer.
 FailureOr<TilingResult>
-replaceInsertSliceWithTiledConsumer(OpBuilder &builder,
-                                    OffsetSizeAndStrideOpInterface sliceOp,
-                                    OpOperand &consumerOp);
+replaceInsertSlicesWithTiledConsumer(OpBuilder &builder,
+                                     ArrayRef<tensor::InsertSliceOp> sliceOps,
+                                     ArrayRef<OpOperand *> consumerOperands);
 
 //===----------------------------------------------------------------------===//
 // Populate functions.

mlir/include/mlir/IR/OpDefinition.h

@@ -272,7 +272,7 @@ class OpFoldResult : public PointerUnion<Attribute, Value> {
   using PointerUnion<Attribute, Value>::PointerUnion;
 
 public:
-  void dump() const { llvm::errs() << *this << "\n"; }
+  LLVM_DUMP_METHOD void dump() const { llvm::errs() << *this << "\n"; }
 
   MLIRContext *getContext() const {
     PointerUnion pu = *this;

mlir/include/mlir/Interfaces/TilingInterface.td

@@ -202,28 +202,28 @@ def TilingInterface : OpInterface<"TilingInterface"> {
       InterfaceMethod<
         /*desc=*/[{
           Method to generate the tiled implementation of an operation that uses
-          exactly a tile of the given operand.
+          the exact tiles of the given operands.
 
           This method is required to allow operations to be "tiled and fused"
-          with an (already tiled) producer. Given a tile of the producer, this
-          method generates the tile of the consumer that uses exactly this
-          produced tile. In some sense it is the "reverse" of
+          with an (already tiled) producer. Given tiles of the producer, this
+          method generates the tile of the consumer that uses exactly these
+          produced tiles. In some sense it is the "reverse" of
           `generateResultTileValue`.
-          - `operandNumber` is the result of the producer used by the consumer.
-          - `offsets` is the offset of the slice of the producer result used by
-            the tiled implementation of the consumer.
-          - `sizes` is the size of the slice of the producer result used by the
+          - `operandNumbers` is the list of operands whose tiles are "producers".
+          - `allOffsets` is the offset of the slice of the producer used by the
+            tiled implementation of the consumer.
+          - `allSizes` is the size of the slice of the producer used by the
             consumer.
-          If it is illegal to fuse with a producer along the given operand for
+          If it is illegal to fuse with a producer along the given operand tiles for
           an operation, the implementation should return a failure.
         }],
         /*retType=*/"::mlir::FailureOr<::mlir::TilingResult>",
-        /*methodName=*/"getTiledImplementationFromOperandTile",
+        /*methodName=*/"getTiledImplementationFromOperandTiles",
         /*args=*/(ins
           "::mlir::OpBuilder &":$b,
-          "unsigned":$operandNumber,
-          "::mlir::ArrayRef<::mlir::OpFoldResult>":$offsets,
-          "::mlir::ArrayRef<::mlir::OpFoldResult>":$sizes),
+          "::mlir::ArrayRef<unsigned>":$operandNumbers,
+          "::mlir::ArrayRef<::mlir::SmallVector<::mlir::OpFoldResult>>":$allOffsets,
+          "::mlir::ArrayRef<::mlir::SmallVector<::mlir::OpFoldResult>>":$allSizes),
         /*methodBody=*/"",
         /*defaultImplementation=*/[{
           return failure();
@@ -235,16 +235,17 @@ def TilingInterface : OpInterface<"TilingInterface"> {
           tile of the operand.
 
           This method is required to allow operations to be "tiled and fused"
-          with an (already tiled) producer. Given a tile of an operand,
-          returns the tile of the iteration space that uses this tile.
-          - `operandNumber` is the result of the producer used by the consumer.
-          - `offsets` is the offset of the slice of the producer result used by
+          with an (already tiled) producer. Given tiles of operands,
+          returns the tile of the iteration space that uses these tiles.
+          - `operandNumbers` is the list of operands whose tiles are "produced"
+            by the producer(s).
+          - `allOffsets` is the offset of the slice of the producers used by
             the tiled implementation of the consumer.
-          - `sizes` is the size of the slice of the producer result used by the
+          - `allSizes` is the size of the slice of the producers used by the
             consumer.
-          If it is illegal to fuse with a producer along the given operand for
-          an operation, or if this mapping cannot be computed, the
-          implementation should return a failure.
+          If it is illegal to fuse with the producer slices for an operation,
+          or if this mapping cannot be computed, the implementation should
+          return a failure.
 
           Note that unlike the "tile consumer and fuse producer" case, the
           "tile producer and fuse consumer" requires an additional method to get
@@ -285,17 +286,17 @@ def TilingInterface : OpInterface<"TilingInterface"> {
           transformation. It does not provide guarantees on whether such a
           transformation is profitable.
 
-          For most cases `getTiledImplementationFromOperandTile` could be a
-          implemented using `getIterationDomainTileFromOperandTile` +
+          For most cases `getTiledImplementationFromOperandTiles` could be a
+          implemented using `getIterationDomainTileFromOperandTiles` +
           `getTiledImplementation` methods.
         }],
         /*retType=*/"::llvm::LogicalResult",
-        /*methodName=*/"getIterationDomainTileFromOperandTile",
+        /*methodName=*/"getIterationDomainTileFromOperandTiles",
         /*args=*/(ins
           "::mlir::OpBuilder &":$b,
-          "unsigned":$operandNumber,
-          "::mlir::ArrayRef<::mlir::OpFoldResult> ":$offsets,
-          "::mlir::ArrayRef<::mlir::OpFoldResult> ":$sizes,
+          "::mlir::ArrayRef<unsigned>":$operandNumbers,
+          "::mlir::ArrayRef<::mlir::SmallVector<::mlir::OpFoldResult>> ":$allOffsets,
+          "::mlir::ArrayRef<::mlir::SmallVector<::mlir::OpFoldResult>> ":$allSizes,
           "::mlir::SmallVectorImpl<::mlir::OpFoldResult> &":$iterDomainOffsets,
           "::mlir::SmallVectorImpl<::mlir::OpFoldResult> &":$iterDomainSizes),
         /*methodBody=*/"",