[mlir][linalg] Make Linalg vectorizer lower affine.apply (take 2)

[awarzynski]

Summary

As discussed in [1], it is possible that the input to the Linalg
vectorizer contains affine.apply ops. Such operations are not
vectarizable at the moment, but this can be fixed by simply converting
them to arithmetic operations. This is basically what this patch
introduces.

The IR change enabled in this patch could be part of a larger set of
"linalgOp pre-processing" transformations that happens right before
vectorization starts but after we know we can vectorize the op. I am
leaving this as a TODO.

[1] https://github.com/iree-org/iree/issues/10876

Co-authored-by: Thomas Raoux <[e-mail]>

[awarzynski]

ATM this is simply https://reviews.llvm.org/D142371 + https://reviews.llvm.org/D143243 with a couple extra affine.apply Ops in the test.

[awarzynski]

Add another small test case

[dcaballe]

LGTM! Just a comment about the tests

mlir/test/Dialect/Linalg/vectorization.mlir

305

Perhaps it would make sense to create simpler test so that it's simpler to understand the generated output (talking about myself here :))

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

 //===- Vectorization.cpp - Implementation of linalg Vectorization ---------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file implements the linalg dialect Vectorization transformations.
 //
 //===----------------------------------------------------------------------===//
+#include "mlir/Dialect/Affine/Utils.h"
 
 #include "mlir/Analysis/SliceAnalysis.h"
 #include "mlir/Dialect/Affine/IR/AffineOps.h"
 #include "mlir/Dialect/Arith/IR/Arith.h"
 #include "mlir/Dialect/Func/IR/FuncOps.h"
 #include "mlir/Dialect/Linalg/IR/Linalg.h"
 #include "mlir/Dialect/Linalg/Transforms/Transforms.h"
 #include "mlir/Dialect/Linalg/Utils/Utils.h"
mlir::linalg::vectorizeLinalgOpPrecondition(LinalgOp linalgOp,
   }
   if (failed(reductionPreconditions(linalgOp))) {
     LDBG("precondition failed: reduction preconditions\n");
     return failure();
   }
   return success();
 }
 
+/// Converts affine.apply Ops to arithmetic operations.
+static void convertAffineApply(RewriterBase &rewriter, LinalgOp linalgOp) {
+  OpBuilder::InsertionGuard g(rewriter);
+  auto toReplace = linalgOp.getBlock()->getOps<AffineApplyOp>();
+
+  for (auto op : make_early_inc_range(toReplace)) {
+    rewriter.setInsertionPoint(op);
+    auto expanded = expandAffineExpr(
+        rewriter, op->getLoc(), op.getAffineMap().getResult(0),
+        op.getOperands().take_front(op.getAffineMap().getNumDims()),
+        op.getOperands().take_back(op.getAffineMap().getNumSymbols()));
+    rewriter.replaceOp(op, expanded);
+  }
+}
+
 /// Emit a suitable vector form for a Linalg op. If provided, `inputVectorSizes`
 /// are used to vectorize this operation. `inputVectorSizes` must match the rank
 /// of the iteration space of the operation and the input vector sizes must be
 /// greater than or equal to their counterpart iteration space sizes, if static.
 /// `inputVectorShapes` also allows the vectorization of operations with dynamic
 /// shapes.
 LogicalResult mlir::linalg::vectorize(RewriterBase &rewriter, LinalgOp linalgOp,
                                       ArrayRef<int64_t> inputVectorSizes,
LogicalResult mlir::linalg::vectorize(RewriterBase &rewriter, LinalgOp linalgOp,
   FailureOr<Operation *> convOr = vectorizeConvolution(rewriter, linalgOp);
   if (succeeded(convOr)) {
     llvm::append_range(results, (*convOr)->getResults());
   } else {
     if (failed(vectorizeLinalgOpPrecondition(linalgOp, inputVectorSizes,
                                              vectorizeNDExtract)))
       return failure();
     LDBG("Vectorize generic by broadcasting to the canonical vector shape\n");
+
+    // Pre-process before proceeding.
+    convertAffineApply(rewriter, linalgOp);
+
     // TODO: 'vectorize' takes in a 'RewriterBase' which is up-casted to
     // 'OpBuilder' when it is passed over to some methods like
     // 'vectorizeAsLinalgGeneric'. This is highly problematic: if we erase an op
     // within these methods, the actual rewriter won't be notified and we will
     // end up with read-after-free issues!
     if (failed(vectorizeAsLinalgGeneric(rewriter, state, linalgOp, results)))
       return failure();
   }

mlir/lib/Dialect/Linalg/Utils/Utils.cpp

bool allIndexingsAreProjectedPermutation(LinalgOp op) {
   });
 }
 
 bool hasOnlyScalarElementwiseOp(Region &r) {
   if (!llvm::hasSingleElement(r))
     return false;
   for (Operation &op : r.front()) {
     if (!(isa<arith::ConstantOp, func::ConstantOp, tensor::ExtractOp,
-              linalg::YieldOp, linalg::IndexOp>(op) ||
+              linalg::YieldOp, linalg::IndexOp, AffineApplyOp>(op) ||
           OpTrait::hasElementwiseMappableTraits(&op)) ||
         llvm::any_of(op.getResultTypes(),
                      [](Type type) { return !type.isIntOrIndexOrFloat(); }))
       return false;
   }
   return true;
 }
 

mlir/test/Dialect/Linalg/vectorization.mlir

 ^bb1(%arg1: !pdl.operation):
   %0 = transform.structured.match ops{["linalg.generic"]} in %arg1 : (!pdl.operation) -> !pdl.operation
   %1 = get_closest_isolated_parent %0 : (!pdl.operation) -> !pdl.operation
   %2 = transform.structured.vectorize %1
 }
 
 // -----
 
+#map0 = affine_map<(d0) -> (d0)>
+
+func.func @vectorize_affine_apply(%arg0: tensor<5xf32>, %arg3: index) -> tensor<5xi32> {
+  %0 = tensor.empty() : tensor<5xi32>
+  %1 = linalg.generic {indexing_maps = [#map0, #map0],
+                       iterator_types = ["parallel"]}
+    ins(%arg0 : tensor<5xf32>)
+    outs(%0 : tensor<5xi32>) {
+  ^bb0(%arg1: f32, %arg2: i32):
+    %2 = linalg.index 0 : index
+    %11 = affine.apply affine_map<() -> (123)>()
+    %12 = affine.apply affine_map<(d0, d1) -> (d0 + d1)>(%2, %11)
+    %13 = affine.apply affine_map<(d0)[s0] -> (d0 + s0)>(%12)[%arg3]
+    %14 = affine.apply affine_map<(d0) -> (d0 + 1)>(%13)
+    %3 = arith.index_cast %14 : index to i32
+    linalg.yield %3 : i32
+  } -> tensor<5xi32>
+  return %1 : tensor<5xi32>
+}
+
+// CHECK-LABEL:  func.func @vectorize_affine_apply
+// CHECK-SAME: %arg0: tensor<5xf32>
+// CHECK-SAME: %[[ARG1:.*]]: index
+// CHECK:   %[[CST:.*]] = arith.constant dense<[123, 124, 125, 126, 127]> : vector<5xindex>
+// CHECK:   %[[CST_0:.*]] = arith.constant dense<1> : vector<5xindex>
+// CHECK:   %[[C0:.*]] = arith.constant 0 : index
+// CHECK:   %[[EMPTY:.*]] = tensor.empty() : tensor<5xi32>
+// CHECK:   %[[BCAST:.*]] = vector.broadcast %[[ARG1]] : index to vector<5xindex>
+// CHECK:   %[[SUM1:.*]] = arith.addi %[[BCAST]], %[[CST]] : vector<5xindex>
+// CHECK:   %[[SUM2:.*]] = arith.addi %[[SUM1]], %[[CST_0]] : vector<5xindex>
+// CHECK:   %[[CAST:.*]] = arith.index_cast %[[SUM2]] : vector<5xindex> to vector<5xi32>
+// CHECK:   vector.transfer_write %[[CAST]], %[[EMPTY]][%[[C0:.*]]] {in_bounds = [true]} : vector<5xi32>, tensor<5xi32>
+
+transform.sequence failures(propagate) {
+ ^bb1(%arg1: !pdl.operation):
+   %0 = transform.structured.match ops{["linalg.generic"]} in %arg1 : (!pdl.operation) -> !pdl.operation
+   %1 = get_closest_isolated_parent %0 : (!pdl.operation) -> !pdl.operation
+   %2 = transform.structured.vectorize %1 { vectorize_nd_extract }
+}
+
+// -----
+
 // CHECK-LABEL: func @test_vectorize_fill
 func.func @test_vectorize_fill(%A : memref<8x16xf32>, %arg0 : f32) {
   //       CHECK: %[[V:.*]] = vector.broadcast {{.*}} : f32 to vector<8x16xf32>
   //       CHECK: vector.transfer_write %[[V]], {{.*}} : vector<8x16xf32>, memref<8x16xf32>
   linalg.fill ins(%arg0 : f32) outs(%A : memref<8x16xf32>)
   return
 }