[mlir][Linalg] Refine how broadcast dims are treated (llvm#99015)

This PR fixes how broadcast dims (identified as "zero" results in
permutation maps) corresponding to a reduction iterator are vectorised
in the case of generic Ops. Here's an example:

```mlir
  #map = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
  #map1 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, 0)>

  func.func @generic_with_reduction_and_broadcast(%arg0: tensor<1x12x197x197xf32>) -> (tensor<1x12x197x1xf32>) {
    %0 = tensor.empty() : tensor<1x12x197x1xf32>

    %1 = linalg.generic {indexing_maps = [#map, #map1],
                        iterator_types = ["parallel", "parallel", "parallel", "reduction"]}
      ins(%arg0 : tensor<1x12x197x197xf32>)
      outs(%0 : tensor<1x12x197x1xf32>) {

    ^bb0(%in: f32, %out: f32):
      %818 = arith.addf %in, %out : f32
      linalg.yield %818 : f32
    } -> tensor<1x12x197x1xf32>
    return %1 : tensor<1x12x197x1xf32>
  }
```

This is a perfectly valid Generic Op, but currently triggers two issues
in the vectoriser. The root cause is this map:

```mlir
  #map1 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, 0)>
```

This map triggers an assert in `reindexIndexingMap` -  this hook
incorrectly assumes that every result in the input map is a `dim`
expression and that there are no constants. That's not the case in this
example. `reindexIndexingMap` is extended to allow maps like the one
above. For now, only constant "zero" results are allowed. This can be
extended in the future once a good motivating example is available.

Separately, the permutation map highlighted above "breaks" mask
calculation (ATM masks are always computed, even in the presence of
static shapes). When applying the following permutation:
```mlir
  (d0, d1, d2, d3) -> (d0, d1, d2, 0)
```

to these canonical shapes (corresponding to the example above):
```
  (1, 12, 197, 197)
```
we end up with the following error:
```bash
error: vector types must have positive constant sizes but got 1, 12, 197, 0
```

The error makes sense and indicates that we should update the
permutation map above to:
```
  (d0, d1, d2, d3) -> (d0, d1, d2)
```

This would correctly give the following vector type:
```
  vector<1x12x197xi1>
```

Fixes llvm#97247

mlir/include/mlir/IR/AffineMap.h

@@ -354,6 +354,24 @@ class AffineMap {
   /// returns the resulting values. `this` must be symbol-less.
   SmallVector<int64_t, 4> compose(ArrayRef<int64_t> values) const;
 
+  /// Returns the number of "zero" results (constant values == 0) in this map.
+  ///
+  /// Example:
+  ///   * For `(d0, d1) -> (d0, d1, 0)` returns 1
+  ///   * For `(d0, d1, d2) -> (d0, d1)` returns 0
+  ///   * For `(d0, d1, d2) -> (d0, 0, d1, 0, d2)` returns 2
+  size_t getNumOfZeroResults() const;
+
+  /// Returns the AffineMap resulting from removing "zero" results (constant
+  /// values == 0) from this map.
+  ///
+  /// Example:
+  ///   * For `(d0, d1) -> (d0, d1, 0)` returns `(d0, d1) -> (d0, d1)`
+  ///   * For `(d0, d1, d2) -> (d0, d1)` returns `(d0, d1, d2) -> (d0, d1)`
+  ///   * For `(d0, d1, d2) -> (d0, 0, d1, 0, d2)` returns
+  ///     `(d0, d1, d2) -> (d0, d1, d2)`
+  AffineMap dropZeroResults();
+
   /// Returns true if the AffineMap represents a subset (i.e. a projection) of a
   /// symbol-less permutation map. `allowZeroInResults` allows projected
   /// permutation maps with constant zero result expressions.

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

@@ -224,10 +224,10 @@ struct VectorizationState {
   /// Masks an operation with the canonical vector mask if the operation needs
   /// masking. Returns the masked operation or the original operation if masking
   /// is not needed. If provided, the canonical mask for this operation is
-  /// permuted using `maybeMaskingMap`.
+  /// permuted using `maybeIndexingMap`.
   Operation *
   maskOperation(RewriterBase &rewriter, Operation *opToMask, LinalgOp linalgOp,
-                std::optional<AffineMap> maybeMaskingMap = std::nullopt);
+                std::optional<AffineMap> maybeIndexingMap = std::nullopt);
 
 private:
   /// Initializes the iteration space static sizes using the Linalg op
@@ -422,16 +422,28 @@ Value VectorizationState::getOrCreateMaskFor(
   return mask;
 }
 
-/// Masks an operation with the canonical vector mask if the operation needs
-/// masking. Returns the masked operation or the original operation if masking
-/// is not needed. If provided, the canonical mask for this operation is
-/// permuted using `maybeMaskingMap`.
 Operation *
 VectorizationState::maskOperation(RewriterBase &rewriter, Operation *opToMask,
                                   LinalgOp linalgOp,
-                                  std::optional<AffineMap> maybeMaskingMap) {
+                                  std::optional<AffineMap> maybeIndexingMap) {
   LDBG("Trying to mask: " << *opToMask << "\n");
 
+  std::optional<AffineMap> maybeMaskingMap = std::nullopt;
+  // The Operand indexing map may contain "zero" results, e.g.:
+  //    (d0, d1, d2, d3) -> (d0, d1, d2, 0)
+  // When applied to canonical vector shapes like these:
+  //    (1, 16, 16, 4)
+  // we would get:
+  //    (1, 16, 16, 0)
+  // Instead, we should extract the following map permutation map for masking:
+  //    (d0, d1, d2, d3) -> (d0, d1, d2)
+  // This way, the corresponding vector/mask type will be:
+  //    vector<1x16x16xty>
+  // rather than:
+  //    vector<1x16x16x0xty>
+  if (maybeIndexingMap)
+    maybeMaskingMap = maybeIndexingMap->dropZeroResults();
+
   // Create or retrieve mask for this operation.
   Value mask =
       getOrCreateMaskFor(rewriter, opToMask, linalgOp, maybeMaskingMap);
@@ -476,7 +488,8 @@ static AffineMap reindexIndexingMap(AffineMap map) {
   assert(map.isProjectedPermutation(/*allowZeroInResults=*/true) &&
          "expected projected permutation");
   auto res = compressUnusedDims(map);
-  assert(res.getNumDims() == res.getNumResults() &&
+  assert(res.getNumDims() ==
+             (res.getNumResults() - res.getNumOfZeroResults()) &&
          "expected reindexed map with same number of dims and results");
   return res;
 }
@@ -1349,16 +1362,6 @@ vectorizeAsLinalgGeneric(RewriterBase &rewriter, VectorizationState &state,
     // permutation map and masking map.
     AffineMap indexingMap = linalgOp.getMatchingIndexingMap(opOperand);
 
-    // Remove zeros from indexing map to use it as masking map.
-    SmallVector<int64_t> zeroPos;
-    auto results = indexingMap.getResults();
-    for (const auto &result : llvm::enumerate(results)) {
-      if (isa<AffineConstantExpr>(result.value())) {
-        zeroPos.push_back(result.index());
-      }
-    }
-    AffineMap maskingMap = indexingMap.dropResults(zeroPos);
-
     AffineMap readMap;
     VectorType readType;
     Type elemType = getElementTypeOrSelf(opOperand->get());
@@ -1388,7 +1391,7 @@ vectorizeAsLinalgGeneric(RewriterBase &rewriter, VectorizationState &state,
     Operation *read = rewriter.create<vector::TransferReadOp>(
         loc, readType, opOperand->get(), indices, readMap,
         ArrayRef<bool>(inBounds));
-    read = state.maskOperation(rewriter, read, linalgOp, maskingMap);
+    read = state.maskOperation(rewriter, read, linalgOp, indexingMap);
     Value readValue = read->getResult(0);
 
     // 3.b. If masked, set in-bounds to true. Masking guarantees that the access

mlir/lib/IR/AffineMap.cpp

@@ -592,6 +592,29 @@ SmallVector<int64_t, 4> AffineMap::compose(ArrayRef<int64_t> values) const {
   return res;
 }
 
+size_t AffineMap::getNumOfZeroResults() const {
+  size_t res = 0;
+  for (auto expr : getResults()) {
+    auto constExpr = dyn_cast<AffineConstantExpr>(expr);
+    if (constExpr && constExpr.getValue() == 0)
+      res++;
+  }
+
+  return res;
+}
+
+AffineMap AffineMap::dropZeroResults() {
+  auto exprs = llvm::to_vector(getResults());
+  SmallVector<AffineExpr> newExprs;
+
+  for (auto expr : getResults()) {
+    auto constExpr = dyn_cast<AffineConstantExpr>(expr);
+    if (!constExpr || constExpr.getValue() != 0)
+      newExprs.push_back(expr);
+  }
+  return AffineMap::get(getNumDims(), getNumSymbols(), newExprs, getContext());
+}
+
 bool AffineMap::isProjectedPermutation(bool allowZeroInResults) const {
   if (getNumSymbols() > 0)
     return false;

mlir/test/Dialect/Linalg/vectorization-with-patterns.mlir

@@ -1964,3 +1964,43 @@ module attributes {transform.with_named_sequence} {
 //       CHECK:     %[[VAL_8:.*]] = vector.transpose %[[VAL_7]], [1, 0] : vector<1x4xf32> to vector<4x1xf32>
 //       CHECK:     vector.transfer_write %[[VAL_8]], %{{.*}} {in_bounds = [true, true]} : vector<4x1xf32>, tensor<4x1xf32>
 //       CHECK:     vector.transfer_write %[[VAL_7]], %{{.*}} {in_bounds = [true, true]} : vector<1x4xf32>, tensor<1x4xf32>
+
+// -----
+
+// Extracted from: https://github.com/llvm/llvm-project/issues/97247
+
+#map = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
+#map1 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, 0)>
+
+func.func @generic_with_reduction_and_broadcast(%arg0: tensor<1x12x197x197xf32>) -> (tensor<1x12x197x1xf32>) {
+  %0 = tensor.empty() : tensor<1x12x197x1xf32>
+  %1 = linalg.generic {indexing_maps = [#map, #map1], iterator_types = ["parallel", "parallel", "parallel", "reduction"]} ins(%arg0 : tensor<1x12x197x197xf32>) outs(%0 : tensor<1x12x197x1xf32>) {
+  ^bb0(%in: f32, %out: f32):
+    %818 = arith.addf %in, %out : f32
+    linalg.yield %818 : f32
+  } -> tensor<1x12x197x1xf32>
+  return %1 : tensor<1x12x197x1xf32>
+}
+module attributes {transform.with_named_sequence} {
+  transform.named_sequence @__transform_main(%arg0: !transform.any_op {transform.readonly}) {
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg0 : (!transform.any_op) -> !transform.any_op
+    %1 = transform.get_parent_op %0 {isolated_from_above} : (!transform.any_op) -> !transform.any_op
+    %2 = transform.structured.vectorize_children_and_apply_patterns %1 : (!transform.any_op) -> !transform.any_op
+    transform.yield
+  }
+}
+
+// CHECK: #[[$ATTR_32:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2)>
+
+// CHECK-LABEL:   func.func @generic_with_reduction_and_broadcast(
+// CHECK-SAME:                                                    %[[VAL_0:.*]]: tensor<1x12x197x197xf32>) -> tensor<1x12x197x1xf32> {
+// CHECK:           %[[VAL_1:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK:           %[[VAL_2:.*]] = arith.constant 0 : index
+// CHECK:           %[[VAL_3:.*]] = tensor.empty() : tensor<1x12x197x1xf32>
+// CHECK:           %[[VAL_4:.*]] = vector.transfer_read %[[VAL_0]]{{\[}}%[[VAL_2]], %[[VAL_2]], %[[VAL_2]], %[[VAL_2]]], %[[VAL_1]] {in_bounds = [true, true, true, true]} : tensor<1x12x197x197xf32>, vector<1x12x197x197xf32>
+// CHECK:           %[[VAL_5:.*]] = vector.transfer_read %[[VAL_3]]{{\[}}%[[VAL_2]], %[[VAL_2]], %[[VAL_2]], %[[VAL_2]]], %[[VAL_1]] {in_bounds = [true, true, true], permutation_map = #[[$ATTR_32]]} : tensor<1x12x197x1xf32>, vector<1x12x197xf32>
+// CHECK:           %[[VAL_6:.*]] = vector.multi_reduction <add>, %[[VAL_4]], %[[VAL_5]] [3] : vector<1x12x197x197xf32> to vector<1x12x197xf32>
+// CHECK:           %[[VAL_7:.*]] = vector.broadcast %[[VAL_6]] : vector<1x12x197xf32> to vector<1x1x12x197xf32>
+// CHECK:           %[[VAL_8:.*]] = vector.transpose %[[VAL_7]], [1, 2, 3, 0] : vector<1x1x12x197xf32> to vector<1x12x197x1xf32>
+// CHECK:           %[[VAL_9:.*]] = vector.transfer_write %[[VAL_8]], %[[VAL_3]]{{\[}}%[[VAL_2]], %[[VAL_2]], %[[VAL_2]], %[[VAL_2]]] {in_bounds = [true, true, true, true]} : vector<1x12x197x1xf32>, tensor<1x12x197x1xf32>
+// CHECK:           return %[[VAL_9]] : tensor<1x12x197x1xf32>

mlir/test/Dialect/Linalg/vectorization.mlir

@@ -147,6 +147,51 @@ module attributes {transform.with_named_sequence} {
 
 // -----
 
+#map = affine_map<(d0, d1) -> (d0, d1)>
+#map1 = affine_map<(d0, d1) -> (d0, 0)>
+
+func.func @dynamic_generic_with_reduction_and_broadcast(%arg0: tensor<?x?xf32>, %init: tensor<?x?xf32>) -> (tensor<?x?xf32>) {
+  %0 = linalg.generic { indexing_maps = [#map, #map1],
+                        iterator_types = ["parallel", "reduction"]}
+    ins(%arg0 : tensor<?x?xf32>)
+    outs(%init : tensor<?x?xf32>) {
+  ^bb0(%in: f32, %out: f32):
+    %1 = arith.addf %in, %out : f32
+    linalg.yield %1 : f32
+  } -> tensor<?x?xf32>
+  return %0 : tensor<?x?xf32>
+}
+// CHECK: #[[$MAP:.+]] = affine_map<(d0, d1) -> (d0)>
+
+// CHECK-LABEL:   func.func @dynamic_generic_with_reduction_and_broadcast(
+// CHECK-SAME:      %[[VAL_0:.*]]: tensor<?x?xf32>,
+// CHECK-SAME:      %[[VAL_1:.*]]: tensor<?x?xf32>) -> tensor<?x?xf32> {
+// CHECK:           %[[VAL_2:.*]] = arith.constant 0 : index
+// CHECK:           %[[VAL_3:.*]] = tensor.dim %[[VAL_0]], %[[VAL_2]] : tensor<?x?xf32>
+// CHECK:           %[[VAL_4:.*]] = arith.constant 1 : index
+// CHECK:           %[[VAL_5:.*]] = tensor.dim %[[VAL_0]], %[[VAL_4]] : tensor<?x?xf32>
+// CHECK:           %[[VAL_6:.*]] = arith.constant 0 : index
+// CHECK:           %[[VAL_7:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK:           %[[VAL_8:.*]] = vector.create_mask %[[VAL_3]], %[[VAL_5]] : vector<4x4xi1>
+// CHECK:           %[[VAL_9:.*]] = vector.mask %[[VAL_8]] { vector.transfer_read %[[VAL_0]]{{\[}}%[[VAL_6]], %[[VAL_6]]], %[[VAL_7]] {in_bounds = [true, true]} : tensor<?x?xf32>, vector<4x4xf32> } : vector<4x4xi1> -> vector<4x4xf32>
+// CHECK:           %[[VAL_10:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK:           %[[VAL_11:.*]] = vector.create_mask %[[VAL_3]] : vector<4xi1>
+// CHECK:           %[[VAL_12:.*]] = vector.mask %[[VAL_11]] { vector.transfer_read %[[VAL_1]]{{\[}}%[[VAL_6]], %[[VAL_6]]], %[[VAL_10]] {in_bounds = [true], permutation_map = #[[$MAP]]} : tensor<?x?xf32>, vector<4xf32> } : vector<4xi1> -> vector<4xf32>
+// CHECK:           %[[VAL_13:.*]] = vector.mask %[[VAL_8]] { vector.multi_reduction <add>, %[[VAL_9]], %[[VAL_12]] [1] : vector<4x4xf32> to vector<4xf32> } : vector<4x4xi1> -> vector<4xf32>
+// CHECK:           %[[VAL_14:.*]] = arith.constant 0 : index
+// CHECK:           %[[VAL_15:.*]] = vector.mask %[[VAL_11]] { vector.transfer_write %[[VAL_13]], %[[VAL_1]]{{\[}}%[[VAL_14]], %[[VAL_14]]] {in_bounds = [true], permutation_map = #[[$MAP]]} : vector<4xf32>, tensor<?x?xf32> } : vector<4xi1> -> tensor<?x?xf32>
+// CHECK:           return %[[VAL_15]] : tensor<?x?xf32>
+
+module attributes {transform.with_named_sequence} {
+  transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
+    %0 = transform.structured.match ops{["linalg.generic"]} in %arg1 : (!transform.any_op) -> !transform.any_op
+    transform.structured.vectorize %0 vector_sizes [4, 4] : !transform.any_op
+    transform.yield
+  }
+}
+
+// -----
+
 func.func @vectorize_dynamic_2d_transpose(%arg0: tensor<?x?xf32>,
                                           %arg1: tensor<?x?xf32>,
                                           %arg2: tensor<?x?xf32>) -> tensor<?x?xf32> {