[mlir][Linalg] NFC - Cleanup explicitly instantiated paterns 2/n - Loops.cpp

[nicolasvasilache]

Summary

This revision belongs to a series of patches that reduce reliance of Linalg transformations on templated rewrite and conversion patterns.
Instead, this uses a MatchAnyTag pattern for the vast majority of cases and dispatches internally.

[pifon2a]

Nice, it became much cleaner with llvm::TypeSwitch.

[mravishankar]

Nice cleanup! Concerned about drop in coverage though.

mlir/lib/Dialect/Linalg/Transforms/Loops.cpp

313–315

Nit: Crux of this method and the one below are the same. Can we just have a emitScalarPoolingMinMaxOpImplementation and combine the logic (the explicit instantiation would be just calling this impl function). But maybe not be worth it, so a suggestion that can be ignored.

455–460

TIL TypeSwitch! This looks great! Thanks!

mlir/test/Dialect/Linalg/loops.mlir

2 ↗

(On Diff #297269)

Does this need to be split. Having the entire command line in one line looks fine to me.

3 ↗

(On Diff #297269)

Does R_UN work? Will this not stop the test from executing?

6 ↗

(On Diff #297269)

Is this actually running the test?

81 ↗

(On Diff #297269)

This seems to be dropping coverage a lot. For example, the pooling ops, conv ops, etc. arent being tests. Not sure whats happening here?

[nicolasvasilache]

Revert spurious test file changes.

[mravishankar]

Thanks!

[nicolasvasilache]

Factor out common code.

[nicolasvasilache]

Rebase

[nicolasvasilache]

This was landed as c303d9b394427e93aa772d543426715b24f98fd1

mlir/lib/Dialect/Linalg/Transforms/Loops.cpp

 #include "mlir/Dialect/StandardOps/EDSC/Intrinsics.h"
 #include "mlir/IR/AffineExpr.h"
 #include "mlir/IR/AffineMap.h"
 #include "mlir/IR/BlockAndValueMapping.h"
 #include "mlir/Support/LLVM.h"
 #include "mlir/Transforms/DialectConversion.h"
 #include "mlir/Transforms/FoldUtils.h"
 
+#include "llvm/ADT/TypeSwitch.h"
+
 using namespace mlir;
 using namespace mlir::edsc;
 using namespace mlir::edsc::intrinsics;
 using namespace mlir::linalg;
 
 using edsc::op::operator+;
 
 static SmallVector<Value, 8> makeCanonicalAffineApplies(OpBuilder &b,

 
 template <typename IndexedValueType, typename OpType>
 static void inlineRegionAndEmitStore(OpType op, ArrayRef<Value> indexedValues,
                                      ArrayRef<SmallVector<Value, 8>> indexing,
                                      ArrayRef<Value> outputBuffers) {
   assert(op.getOperation()->getNumRegions() == 1 &&
          "Expected single region op");
   auto &b = ScopedContext::getBuilderRef();
-  auto &block = op.region().front();
+  auto &block = op.getOperation()->getRegion(0).front();
   BlockAndValueMapping map;
   map.map(block.getArguments(), indexedValues);
   for (auto &op : block.without_terminator()) {
     assert(op.getNumRegions() == 0 && "expected a non-nested region");
     auto *newOp = b.clone(op, map);
     map.map(op.getResults(), newOp->getResults());
   }
 
static InputAndOutputIndices getInputAndOutputIndices(ArrayRef<Value> allIvs,
   auto mapsRange = op.indexing_maps().template getAsRange<AffineMapAttr>();
   auto maps = llvm::to_vector<8>(
       llvm::map_range(mapsRange, [](AffineMapAttr a) { return a.getValue(); }));
   return InputAndOutputIndices{
       makeCanonicalAffineApplies(b, loc, maps[0], allIvs),
       makeCanonicalAffineApplies(b, loc, maps[2], allIvs)};
 }
 
-namespace {
-
 /// Emits the MLIR for the scalar part of the generic op by:
 ///   1. Emitting load ops for each input and output view in order. This is
 ///      achieved by applying the appropriate input or output map to the
 ///      enclosing induction variables.
 ///   2. Emitting a call to `op.fun()` that takes as arguments the scalars
 ///      from point 1. above.
 ///   3. Emitting store ops to store the results of 2. to the output
 ///      views.

 ///          store %14#0, %arg1[%i, %j, %k] :
 ///            memref<?x?x?Xf32, stride_specification>
 ///          store %14#1, %arg2[%i, %k, %j] :
 ///            memref<?x?x?Xf32, stride_specification>
 ///       }
 ///      }
 ///    }
 /// ```
-// TODO: need a LinalgStructuredOpInterface.
-template <typename IndexedValueType, typename LinalgStructuredOpType>
-void emitScalarImplementation(ArrayRef<Value> allIvs,
-                              LinalgStructuredOpType linalgOp) {
+template <typename IndexedValueType>
+static void emitScalarImplementation(ArrayRef<Value> allIvs,
+                                     LinalgOp linalgOp) {
   assert(linalgOp.hasBufferSemantics() &&
          "expected linalg op with buffer semantics");
   auto &b = ScopedContext::getBuilderRef();
   auto loc = ScopedContext::getLocation();
   unsigned nInputs = linalgOp.getNumInputs();
   unsigned nOutputs = linalgOp.getNumOutputs();
   SmallVector<Value, 4> indexedValues;
   indexedValues.reserve(nInputs + nOutputs);
 
   auto attr = linalgOp.template getAttrOfType<IntegerAttr>("symbol_source");
   auto allIvsPlusDims = SmallVector<Value, 4>(allIvs.begin(), allIvs.end());
   if (attr) {
-    auto operand = linalgOp.getOperand(attr.getInt());
+    auto operand = linalgOp.getOperation()->getOperand(attr.getInt());
     auto shapedType = operand.getType().template cast<ShapedType>();
     allIvsPlusDims.reserve(allIvs.size() + shapedType.getRank());
     for (unsigned idx = 0, e = shapedType.getRank(); idx < e; ++idx)
       allIvsPlusDims.push_back(b.create<DimOp>(loc, operand, idx));
   }
 
   // TODO: Avoid the loads if the corresponding argument of the
   // region has no uses.
indexing.push_back(makeCanonicalAffineApplies(
         b, loc, linalgOp.getOutputIndexingMap(i), allIvsPlusDims));
     outputBuffers.push_back(linalgOp.getOutputBuffer(i));
   }
   inlineRegionAndEmitStore<IndexedValueType>(linalgOp, indexedValues, indexing,
                                              outputBuffers);
 }
 
 template <typename IndexedValueType>
-void emitScalarImplementation(ArrayRef<Value> allIvs, CopyOp copyOp) {
+static void emitScalarImplementation(ArrayRef<Value> allIvs, CopyOp copyOp) {
   assert(copyOp.hasBufferSemantics() &&
          "expected linalg op with buffer semantics");
   auto nPar = copyOp.getNumParallelLoops();
   assert(nPar == allIvs.size());
   auto inputIvs =
       permuteIvs(allIvs.take_front(nPar), copyOp.inputPermutation());
   auto outputIvs =
       permuteIvs(allIvs.take_front(nPar), copyOp.outputPermutation());
   SmallVector<Value, 8> iivs(inputIvs.begin(), inputIvs.end());
   SmallVector<Value, 8> oivs(outputIvs.begin(), outputIvs.end());
   IndexedValueType O(copyOp.getOutputBuffer(0)), I(copyOp.getInput(0));
   // Emit the proper scalar assignment, whether we are dealing with a 0-D or
   // an n-D loop nest; with or without permutations.
   // clang-format off
     nPar > 0 ? O(oivs) = I(iivs) :
                O() = I();
   // clang-format on
 }
 
 template <typename IndexedValueType>
-void emitScalarImplementation(ArrayRef<Value> allIvs, FillOp fillOp) {
+static void emitScalarImplementation(ArrayRef<Value> allIvs, FillOp fillOp) {
   assert(fillOp.hasBufferSemantics() &&
          "expected linalg op with buffer semantics");
   auto nPar = fillOp.getNumParallelLoops();
   assert(nPar == allIvs.size());
   auto ivs = SmallVector<Value, 4>(allIvs.begin(), allIvs.begin() + nPar);
   IndexedValueType O(fillOp.getOutputBuffer(0));
   // Emit the proper scalar assignment, whether we are dealing with a 0-D or
   // an n-D loop nest; with or without permutations.
   nPar > 0 ? O(ivs) = fillOp.value() : O() = fillOp.value();
 }
 
 template <typename IndexedValueType>
-Value getConvOpInput(ConvOp convOp, StdIndexedValue im,
+static Value getConvOpInput(ConvOp convOp, StdIndexedValue im,
                             MutableArrayRef<Value> imIdx) {
   // TODO: add a level of indirection to linalg.generic.
   if (!convOp.padding())
     return im(imIdx);
 
   auto *context = ScopedContext::getContext();
   Value zeroIndex = std_constant_index(0);
   SmallVector<Value, 8> conds;
   SmallVector<Value, 8> clampedImIdx;
if (hasPadding(convOp)) {
     Value paddedInput = getConvOpInput<IndexedValueType>(convOp, I, imIdx);
     O(oIdx) += F(fIdx) * paddedInput;
   } else {
     IndexedValueType I(convOp.input());
     O(oIdx) += F(fIdx) * I(imIdx);
   }
 }
 
-template <typename IndexedValueType>
-void emitScalarImplementation(ArrayRef<Value> allIvs, PoolingMaxOp op) {
+template <typename IndexedValueType, typename OpType>
+static void emitPoolingMinMaxScalarImplementation(ArrayRef<Value> allIvs,
+                                                  OpType op) {
   InputAndOutputIndices indices = getInputAndOutputIndices(allIvs, op);
   // Emit scalar form.
   IndexedValueType output(op.output());
   IndexedValueType input(op.input());
   Value lhs = output(indices.outputs);
   Value rhs = input(indices.inputs);
   using edsc::op::sgt;
-  Value maxValue = std_select(sgt(lhs, rhs), lhs, rhs);
-  output(indices.outputs) = maxValue;
+  using edsc::op::slt;
+  Value value = std::is_same<OpType, PoolingMinOp>()
+                    ? std_select(slt(lhs, rhs), lhs, rhs)
+                    : std_select(sgt(lhs, rhs), lhs, rhs);
+  output(indices.outputs) = value;
 }
 
 template <typename IndexedValueType>
-void emitScalarImplementation(ArrayRef<Value> allIvs, PoolingMinOp op) {
-  InputAndOutputIndices indices = getInputAndOutputIndices(allIvs, op);
-  // Emit scalar form.
-  IndexedValueType output(op.output());
-  IndexedValueType input(op.input());
-  Value lhs = output(indices.outputs);
-  Value rhs = input(indices.inputs);
-  using edsc::op::slt;
-  Value minValue = std_select(slt(lhs, rhs), lhs, rhs);
-  output(indices.outputs) = minValue;
+static void emitScalarImplementation(ArrayRef<Value> allIvs, PoolingMaxOp op) {
+  emitPoolingMinMaxScalarImplementation<IndexedValueType, PoolingMaxOp>(allIvs,
+                                                                        op);
 }
+
 template <typename IndexedValueType>
-void emitScalarImplementation(ArrayRef<Value> allIvs, PoolingSumOp op) {
+static void emitScalarImplementation(ArrayRef<Value> allIvs, PoolingMinOp op) {
+  emitPoolingMinMaxScalarImplementation<IndexedValueType, PoolingMinOp>(allIvs,
+                                                                        op);
+}
+
+template <typename IndexedValueType>
+static void emitScalarImplementation(ArrayRef<Value> allIvs, PoolingSumOp op) {
   auto indices = getInputAndOutputIndices(allIvs, op);
   IndexedValueType input(op.input()), output(op.output());
 
   // Emit scalar form.
   output(indices.outputs) += input(indices.inputs);
 }
+
 /// Emits the MLIR for the scalar part of the indexed generic op by:
 ///   1. Emitting load ops for each input and output view in order. This is
 ///      achieved by applying the appropriate input or output map to the
 ///      enclosing induction variables.
 ///   2. Emitting a call to `op.fun()` that takes as arguments the induction
 ///      variables and the scalars from point 1. above.
 ///   3. Emitting store ops to store the results of 2. to the output views.
 ///
for (unsigned i = 0; i < nOutputs; ++i) {
     indexing.push_back(makeCanonicalAffineApplies(
         b, loc, indexedGenericOp.getOutputIndexingMap(i), allIvs));
     outputBuffers.push_back(indexedGenericOp.getOutputBuffer(i));
   }
   inlineRegionAndEmitStore<IndexedValueType>(indexedGenericOp, indexedValues,
                                              indexing, outputBuffers);
 }
 
-template <typename LoopTy, typename ConcreteOpTy>
-Optional<LinalgLoops> linalgOpToLoopsImpl(Operation *op, OpBuilder &builder) {
+template <typename LoopTy>
+static Optional<LinalgLoops> linalgOpToLoopsImpl(Operation *op,
+                                                 OpBuilder &builder) {
   using IndexedValueTy = typename GenerateLoopNest<LoopTy>::IndexedValueTy;
 
   ScopedContext scope(builder, op->getLoc());
 
   // The flattened loopToOperandRangesMaps is expected to be an invertible
   // permutation map (which is asserted in the inverse calculation).
-  auto linalgOp = cast<ConcreteOpTy>(op);
+  auto linalgOp = cast<LinalgOp>(op);
   assert(linalgOp.hasBufferSemantics() &&
          "expected linalg op with buffer semantics");
   auto mapsRange =
       linalgOp.indexing_maps().template getAsRange<AffineMapAttr>();
   auto maps = llvm::to_vector<8>(
       llvm::map_range(mapsRange, [](AffineMapAttr a) { return a.getValue(); }));
   SmallVector<Value, 8> sizes = getShape(builder, linalgOp);
   AffineMap map = concatAffineMaps(maps);
   auto loopRanges = emitLoopRanges(scope.getBuilderRef(), scope.getLocation(),
                                    map, getShape(builder, linalgOp));
   SmallVector<Value, 4> allIvs;
   GenerateLoopNest<LoopTy>::doit(
       loopRanges, /*iterInitArgs*/ {}, linalgOp.iterator_types().getValue(),
       [&](ValueRange ivs, ValueRange iterArgs) -> scf::ValueVector {
         assert(iterArgs.empty() && "unexpected iterArgs");
         allIvs.append(ivs.begin(), ivs.end());
-        emitScalarImplementation<IndexedValueTy>(allIvs, linalgOp);
+        llvm::TypeSwitch<Operation *>(op)
+            .Case<CopyOp, FillOp, ConvOp, PoolingMaxOp, PoolingMinOp,
+                  PoolingSumOp, IndexedGenericOp, LinalgOp>([&](auto op) {
+              emitScalarImplementation<IndexedValueTy>(allIvs, op);
+            })
+            .Default([&](Operation *op) { assert(false && "unexpected op"); });
         return scf::ValueVector{};
       });
   // Number of loop ops might be different from the number of ivs since some
   // loops like affine.parallel and scf.parallel have multiple ivs.
   llvm::SetVector<Operation *> loopSet;
   for (Value iv : allIvs) {
     if (!iv)
       return {};
     // The induction variable is a block argument of the entry block of the
     // loop operation.
     BlockArgument ivVal = iv.dyn_cast<BlockArgument>();
     if (!ivVal)
       return {};
     loopSet.insert(ivVal.getOwner()->getParentOp());
   }
   LinalgLoops loops(loopSet.begin(), loopSet.end());
   return loops;
 }
 
-template <typename LoopType, typename ConcreteOp>
+namespace {
+template <typename LoopType>
 class LinalgRewritePattern : public RewritePattern {
 public:
-  explicit LinalgRewritePattern(MLIRContext *context)
-      : RewritePattern(ConcreteOp::getOperationName(), 1, context) {}
+  LinalgRewritePattern() : RewritePattern(/*benefit=*/1, MatchAnyOpTypeTag()) {}
 
   LogicalResult matchAndRewrite(Operation *op,
                                 PatternRewriter &rewriter) const override {
-    if (!linalgOpToLoopsImpl<LoopType, ConcreteOp>(op, rewriter))
+    if (!isa<LinalgOp>(op))
+      return failure();
+    if (!linalgOpToLoopsImpl<LoopType>(op, rewriter))
       return failure();
     rewriter.eraseOp(op);
     return success();
   }
 };
 
-template <typename LoopType, typename ConcreteOp>
-void insertOnePattern(OwningRewritePatternList &patterns, MLIRContext *ctx) {
-  patterns.insert<LinalgRewritePattern<LoopType, ConcreteOp>>(ctx);
-}
+struct FoldAffineOp;
+} // namespace
 
-template <typename LoopType, typename... Args>
-void insertPatterns(OwningRewritePatternList &patterns, MLIRContext *ctx) {
-  (void)std::initializer_list<int>{
-      0, (insertOnePattern<LoopType, Args>(patterns, ctx), 0)...};
+template <typename LoopType>
+static void lowerLinalgToLoopsImpl(FuncOp funcOp, MLIRContext *context) {
+  OwningRewritePatternList patterns;
+  patterns.insert<LinalgRewritePattern<LoopType>>();
+  DimOp::getCanonicalizationPatterns(patterns, context);
+  AffineApplyOp::getCanonicalizationPatterns(patterns, context);
+  patterns.insert<FoldAffineOp>(context);
+  // Just apply the patterns greedily.
+  applyPatternsAndFoldGreedily(funcOp, patterns);
 }
 
+namespace {
 /// Local folding pattern for AffineApplyOp that we can apply greedily.
 /// This replaces AffineApplyOp by the proper value in cases where the
 /// associated map is trivial.
 /// A trivial map here is defined as a map with a single result and either:
 ///   1. Zero operand + returns a single AffineConstantExpr
 ///   2. One operand + returns a single AffineDimExpr
 ///   3. One operand + returns a single AffineSymbolExpr
 //
LogicalResult matchAndRewrite(Operation *op,
     }
     if (expr.dyn_cast<AffineDimExpr>() || expr.dyn_cast<AffineSymbolExpr>()) {
       rewriter.replaceOp(op, op->getOperand(0));
       return success();
     }
     return failure();
   }
 };
-} // namespace
 
-template <typename LoopType>
-static void lowerLinalgToLoopsImpl(FuncOp funcOp, MLIRContext *context) {
-  OwningRewritePatternList patterns;
-  // Canonicalization and folding patterns applied greedily allow cleaning up
-  // the emitted IR on the fly.
-  // TODO: fold view and subview ops?
-  insertPatterns<LoopType,
-#define GET_OP_LIST
-#include "mlir/Dialect/Linalg/IR/LinalgStructuredOps.cpp.inc"
-                 >(patterns, context);
-
-  DimOp::getCanonicalizationPatterns(patterns, context);
-  AffineApplyOp::getCanonicalizationPatterns(patterns, context);
-  patterns.insert<FoldAffineOp>(context);
-  // Just apply the patterns greedily.
-  applyPatternsAndFoldGreedily(funcOp, patterns);
-}
-
-namespace {
 struct LowerToAffineLoops
     : public LinalgLowerToAffineLoopsBase<LowerToAffineLoops> {
   void runOnFunction() override {
     lowerLinalgToLoopsImpl<AffineForOp>(getFunction(), &getContext());
   }
 };
+
 struct LowerToLoops : public LinalgLowerToLoopsBase<LowerToLoops> {
   void runOnFunction() override {
     lowerLinalgToLoopsImpl<scf::ForOp>(getFunction(), &getContext());
   }
 };
+
 struct LowerToParallelLoops
     : public LinalgLowerToParallelLoopsBase<LowerToParallelLoops> {
   void runOnFunction() override {
     lowerLinalgToLoopsImpl<scf::ParallelOp>(getFunction(), &getContext());
   }
 };
 } // namespace
 
 std::unique_ptr<OperationPass<FuncOp>> mlir::createConvertLinalgToLoopsPass() {
   return std::make_unique<LowerToLoops>();
 }
 
 std::unique_ptr<OperationPass<FuncOp>>
 mlir::createConvertLinalgToParallelLoopsPass() {
   return std::make_unique<LowerToParallelLoops>();
 }
 
 std::unique_ptr<OperationPass<FuncOp>>
 mlir::createConvertLinalgToAffineLoopsPass() {
   return std::make_unique<LowerToAffineLoops>();
 }
 
-// TODO: gradually remove this layer as more ops become "named".
-template <typename LoopTy>
-static Optional<LinalgLoops> linalgOpToLoopsImplSwitch(Operation *op,
-                                                       OpBuilder &builder) {
-  assert(isa<LinalgOp>(op) && "LinalgOp expected");
-  if (isa<CopyOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, CopyOp>(op, builder);
-  if (isa<FillOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, FillOp>(op, builder);
-  if (isa<ConvOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, ConvOp>(op, builder);
-  if (isa<PoolingMaxOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, PoolingMaxOp>(op, builder);
-  if (isa<PoolingMinOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, PoolingMinOp>(op, builder);
-  if (isa<PoolingSumOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, PoolingSumOp>(op, builder);
-  if (isa<IndexedGenericOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, IndexedGenericOp>(op, builder);
-
-  // TODO: Cases below are generic and need a LinalgStructuredOpInterface.
-  if (isa<GenericOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, GenericOp>(op, builder);
-  if (isa<MatmulOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, MatmulOp>(op, builder);
-  if (isa<MatvecOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, MatvecOp>(op, builder);
-  if (isa<VecmatOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, VecmatOp>(op, builder);
-  if (isa<DotOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, DotOp>(op, builder);
-  if (isa<BatchMatmulOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, BatchMatmulOp>(op, builder);
-  if (isa<ConvWOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, ConvWOp>(op, builder);
-  if (isa<ConvNWCOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, ConvNWCOp>(op, builder);
-  if (isa<ConvNCWOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, ConvNCWOp>(op, builder);
-  if (isa<ConvHWOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, ConvHWOp>(op, builder);
-  if (isa<ConvNHWCOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, ConvNHWCOp>(op, builder);
-  if (isa<ConvNCHWOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, ConvNCHWOp>(op, builder);
-  if (isa<ConvDHWOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, ConvDHWOp>(op, builder);
-  if (isa<ConvNDHWCOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, ConvNDHWCOp>(op, builder);
-  if (isa<ConvNCDHWOp>(op))
-    return linalgOpToLoopsImpl<LoopTy, ConvNCDHWOp>(op, builder);
-  llvm_unreachable("Unexpected op in linalgOpToLoopsImpl");
-}
-
 SmallVector<Range, 4> mlir::linalg::emitLoopRanges(OpBuilder &b, Location loc,
                                                    AffineMap map,
                                                    ValueRange viewSizes) {
   unsigned numDims = map.getNumDims(), numRes = map.getNumResults();
   unsigned numSym = map.getNumSymbols();
   assert(viewSizes.size() == numRes + numSym &&
          "viewSizes must contain sizes of all views and values for symbols");
   SmallVector<Range, 4> res(numDims);
SmallVector<Range, 4> mlir::linalg::emitLoopRanges(OpBuilder &b, Location loc,
   }
   return res;
 }
 
 /// Emits a loop nest with the proper body for `op`.
 template <typename LoopTy>
 Optional<LinalgLoops> mlir::linalg::linalgLowerOpToLoops(OpBuilder &builder,
                                                          Operation *op) {
-  return linalgOpToLoopsImplSwitch<LoopTy>(op, builder);
+  return linalgOpToLoopsImpl<LoopTy>(op, builder);
 }
 
 template Optional<LinalgLoops>
 mlir::linalg::linalgLowerOpToLoops<AffineForOp>(OpBuilder &builder,
                                                 Operation *op);
 template Optional<LinalgLoops>
 mlir::linalg::linalgLowerOpToLoops<scf::ForOp>(OpBuilder &builder,
                                                Operation *op);