[Mlir-commits] [mlir] [MLIR][Linalg] Fix insert_slice fusion with rank reduction (PR #130961)

Thomas Preud'homme llvmlistbot at llvm.org
Wed May 21 01:21:49 PDT 2025


https://github.com/RoboTux updated https://github.com/llvm/llvm-project/pull/130961

>From a370cd2d44b2715470c49dfb8b013d12dcff9826 Mon Sep 17 00:00:00 2001
From: Thomas Preud'homme <thomas.preudhomme at arm.com>
Date: Wed, 12 Mar 2025 13:22:14 +0000
Subject: [PATCH 1/8] [MLIR][Linalg] Fix insert_slice fusion with rank
 reduction

Insert_slice fusion with a linalg producer does not account for
possible rank-reduction in the insert_slice return type. When that
happens, a tosa.cast gets generated due to the type mismatch which is
invalid for tensor with different rank. This later trips other pass.
---
 mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp | 36 ++++++++++-
 .../Dialect/Linalg/tile-and-fuse-tensors.mlir | 63 +++++++++++++++++++
 2 files changed, 97 insertions(+), 2 deletions(-)

diff --git a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
index 223d728b0b27d..81b204df5a0aa 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
@@ -18,6 +18,7 @@
 #include "mlir/Dialect/Linalg/Utils/Utils.h"
 #include "mlir/Dialect/MemRef/IR/MemRef.h"
 #include "mlir/Dialect/Tensor/IR/Tensor.h"
+#include "mlir/Dialect/Tensor/Utils/Utils.h"
 #include "mlir/IR/AffineExpr.h"
 #include "mlir/IR/AffineMap.h"
 #include "mlir/IR/Dominance.h"
@@ -26,6 +27,7 @@
 #include "mlir/Transforms/RegionUtils.h"
 #include "llvm/ADT/MapVector.h"
 #include "llvm/ADT/ScopeExit.h"
+#include "llvm/ADT/SmallBitVector.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 
@@ -235,6 +237,31 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpOperand &consumerOpOperand) {
   return fuseProducerOfTensor(b, producerOpResult, consumerOpOperand);
 }
 
+/// Create tensor.collapse_shape to drop dimensions in `dropDims` in tensor
+/// `from`.
+tensor::CollapseShapeOp collapseTo(OpBuilder &b, Location loc, Value from,
+                                   const llvm::SmallBitVector &dropDims) {
+  auto fromType = cast<ShapedType>(from.getType());
+  assert(fromType.getRank() == dropDims.size());
+  SmallVector<ReassociationIndices, 2> reassocIdxsVec;
+  ReassociationIndices reassocIdxs;
+
+  bool foundKeptDim = false;
+  for (int dim = 0; dim < fromType.getRank(); dim++) {
+    if (!dropDims.test(dim)) {
+      if (foundKeptDim) {
+        reassocIdxsVec.push_back(reassocIdxs);
+        reassocIdxs.clear();
+      }
+      foundKeptDim = true;
+    }
+    reassocIdxs.push_back(dim);
+  }
+  if (!reassocIdxs.empty())
+    reassocIdxsVec.push_back(reassocIdxs);
+  return b.create<tensor::CollapseShapeOp>(loc, from, reassocIdxsVec);
+}
+
 FailureOr<FusionInfo>
 mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpResult producerOpResult,
                                    OpOperand &consumerOpOperand) {
@@ -255,6 +282,7 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpResult producerOpResult,
                << "\nNot fusable, not an extract_slice op: " << inputTensor);
     return failure();
   }
+  llvm::SmallBitVector droppedDims = sliceOp.getDroppedDims();
 
   // If producer is already in the same block as consumer, we are done.
   if (consumerOpOperand.get().getParentBlock() ==
@@ -272,12 +300,16 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpResult producerOpResult,
            consumerOpOperand);
 
   // Replace use.
+  Value def = fusedProducer->getResult(producerOpResult.getResultNumber());
+  Type consumerType = consumerOpOperand.get().getType();
+  // Rank-reduction occured as part of the extract_slice.
+  if (cast<ShapedType>(consumerType).getRank() !=
+      cast<ShapedType>(def.getType()).getRank())
+    def = collapseTo(b, fusedProducer.getLoc(), def, droppedDims);
   // Canonicalizations are not guaranteed to have happened before constructing
   // `fusedProducer`. In the tensor case this can result in temporary type
   // mismatches. Insert a `tensor.cast` op to propagate the transformation
   // invariant that types are compatible.
-  Value def = fusedProducer->getResult(producerOpResult.getResultNumber());
-  Type consumerType = consumerOpOperand.get().getType();
   if (consumerType != def.getType())
     def = b.create<tensor::CastOp>(fusedProducer.getLoc(), consumerType, def);
   consumerOpOperand.set(def);
diff --git a/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir b/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
index 0f27a92c119cf..b4fbdfacde899 100644
--- a/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
+++ b/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
@@ -318,3 +318,66 @@ func.func @pad_generic_static(%small_input: tensor<58x1xf32>, %large_input: tens
   }
   return %for0 : tensor<64x128xf32>
 }
+
+// -----
+
+func.func @rank_reduced_extract_slice(%arg0: tensor<6x6x1x1x1x1xf32>, %arg1: tensor<6x6x1x1xf32>, %arg2: tensor<4x6xf32>) -> tensor<4x6xf32> {
+  %c0 = arith.constant 0 : index
+  %c2 = arith.constant 2 : index
+  %c6 = arith.constant 6 : index
+  %cst = arith.constant 0.0 : f32
+  %init1 = tensor.empty() : tensor<6x6x1x1x1x1xf32>
+  %fill1 = linalg.fill ins(%cst : f32) outs(%init1 : tensor<6x6x1x1x1x1xf32>) -> tensor<6x6x1x1x1x1xf32>
+  %0 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3, d4, d6)>, affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d6, d5)>, affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3, d4, d5)>], iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel", "parallel", "reduction"]} ins(%arg0, %arg1 : tensor<6x6x1x1x1x1xf32>, tensor<6x6x1x1xf32>) outs(%fill1 : tensor<6x6x1x1x1x1xf32>) {
+  ^bb0(%in: f32, %in_1: f32, %out: f32):
+    %10 = arith.mulf %in, %in_1 : f32
+    %11 = arith.addf %out, %10 : f32
+    linalg.yield %11 : f32
+  } -> tensor<6x6x1x1x1x1xf32>
+  %init2 = tensor.empty() : tensor<4x6xf32>
+  %1 = scf.for %arg4 = %c0 to %c6 step %c2 iter_args(%arg3 = %init2) -> (tensor<4x6xf32>) {
+    %2 = tensor.extract_slice %0[0, %arg4, 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1] : tensor<6x6x1x1x1x1xf32> to tensor<6x2xf32>
+    %init3 = tensor.empty() : tensor<4x2xf32>
+    %fill3 = linalg.fill ins(%cst : f32) outs(%init3 : tensor<4x2xf32>) -> tensor<4x2xf32>
+    %3 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2) -> (d0, d2)>, affine_map<(d0, d1, d2) -> (d2, d1)>, affine_map<(d0, d1, d2) -> (d0, d1)>], iterator_types = ["parallel", "parallel", "reduction"]} ins(%arg2, %2 : tensor<4x6xf32>, tensor<6x2xf32>) outs(%fill3 : tensor<4x2xf32>) {
+    ^bb0(%in: f32, %in_1: f32, %out: f32):
+      %20 = arith.mulf %in, %in_1 : f32
+      %21 = arith.addf %out, %20 : f32
+      linalg.yield %21 : f32
+    } -> tensor<4x2xf32>
+    %4 = tensor.insert_slice %3 into %arg3[0, %arg4] [4, 2] [1, 1]  : tensor<4x2xf32> into tensor<4x6xf32>
+    scf.yield %4 : tensor<4x6xf32>
+  }
+  return %1 : tensor<4x6xf32>
+}
+
+//       CHECK: func @rank_reduced_extract_slice(
+//  CHECK-SAME: %[[ARG0:[0-9a-z]*]]: tensor<6x6x1x1x1x1xf32>
+//  CHECK-SAME: %[[ARG1:[0-9a-z]*]]: tensor<6x6x1x1xf32>
+//  CHECK-SAME: %[[ARG2:[0-9a-z]*]]: tensor<4x6xf32>
+
+//   CHECK-DAG: %[[C0:.*]] = arith.constant 0 : index
+//   CHECK-DAG: %[[C2:.*]] = arith.constant 2 : index
+//   CHECK-DAG: %[[C6:.*]] = arith.constant 6 : index
+//       CHECK: %[[EMPTY_PROD:.*]] = tensor.empty() : tensor<6x6x1x1x1x1xf32>
+//       CHECK: %[[FILL_PROD:.*]] = linalg.fill ins({{%.*}} : f32)
+//  CHECK-SAME:     outs(%[[EMPTY_PROD]] : tensor<6x6x1x1x1x1xf32>) -> tensor<6x6x1x1x1x1xf32>
+//       CHECK: %[[EMPTY_FOR:.*]] = tensor.empty() : tensor<4x6xf32>
+//       CHECK: %[[EMPTY_CONS:.*]] = tensor.empty() : tensor<4x2xf32>
+//       CHECK: %[[FILL_CONS:.*]] = linalg.fill ins({{%.*}} : f32)
+//  CHECK-SAME:     outs(%[[EMPTY_CONS]] : tensor<4x2xf32>) -> tensor<4x2xf32>
+//       CHECK: %[[FOR:.*]] = scf.for %[[I:[0-9a-z]*]] = %[[C0]] to %[[C6]] step %[[C2]] iter_args(%[[ARG_ITER:.*]] = %[[EMPTY_FOR]])
+//   CHECK-DAG:   %[[ARG0_SLICE:.*]] = tensor.extract_slice %[[ARG0]][0, %[[I]], 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1]  : tensor<6x6x1x1x1x1xf32> to tensor<6x2x1x1x1x1xf32>
+//   CHECK-DAG:   %[[ARG1_SLICE:.*]] = tensor.extract_slice %[[ARG1]][0, %[[I]], 0, 0] [6, 2, 1, 1] [1, 1, 1, 1]  : tensor<6x6x1x1xf32> to tensor<6x2x1x1xf32>
+//   CHECK-DAG:   %[[FILL_PROD_SLICE:.*]] = tensor.extract_slice %[[FILL_PROD]][0, %[[I]], 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1] : tensor<6x6x1x1x1x1xf32> to tensor<6x2x1x1x1x1xf32>
+
+//       CHECK:    %[[MMUL_PROD:.*]] = linalg.generic
+//  CHECK-SAME:        ins(%[[ARG0_SLICE]], %[[ARG1_SLICE]] : tensor<6x2x1x1x1x1xf32>, tensor<6x2x1x1xf32>)
+//  CHECK-SAME:        outs(%[[FILL_PROD_SLICE]] : tensor<6x2x1x1x1x1xf32>)
+//       CHECK:    %[[PROD_COLLAPSE:.*]] = tensor.collapse_shape %[[MMUL_PROD]] {{\[\[0\], \[1, 2, 3, 4, 5\]\]}} : tensor<6x2x1x1x1x1xf32> into tensor<6x2xf32>
+//       CHECK:    %[[MMUL_CONS:.*]] = linalg.generic
+//  CHECK-SAME:        ins(%[[ARG2]], %[[PROD_COLLAPSE]] : tensor<4x6xf32>, tensor<6x2xf32>)
+//  CHECK-SAME:        outs(%[[FILL_CONS]] : tensor<4x2xf32>)
+//       CHECK:   %[[CONS_SLICE:.*]] = tensor.insert_slice %[[MMUL_CONS]] into %[[ARG_ITER]][0, %[[I]]] [4, 2] [1, 1] : tensor<4x2xf32> into tensor<4x6xf32>
+//       CHECK:   scf.yield %[[CONS_SLICE]] : tensor<4x6xf32>
+//       CHECK: return %[[FOR]] : tensor<4x6xf32>

>From ce067327de33ee88397573ead73291764f15c627 Mon Sep 17 00:00:00 2001
From: Thomas Preud'homme <thomas.preudhomme at arm.com>
Date: Tue, 25 Mar 2025 22:49:26 +0000
Subject: [PATCH 2/8] Add more comments and simplify test

---
 mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp | 12 ++-
 .../Dialect/Linalg/tile-and-fuse-tensors.mlir | 88 ++++++++-----------
 2 files changed, 45 insertions(+), 55 deletions(-)

diff --git a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
index 81b204df5a0aa..d18d6f7ff8dd8 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
@@ -239,14 +239,20 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpOperand &consumerOpOperand) {
 
 /// Create tensor.collapse_shape to drop dimensions in `dropDims` in tensor
 /// `from`.
-tensor::CollapseShapeOp collapseTo(OpBuilder &b, Location loc, Value from,
-                                   const llvm::SmallBitVector &dropDims) {
+static tensor::CollapseShapeOp collapseTo(OpBuilder &b, Location loc, Value from,
+                                          const llvm::SmallBitVector &dropDims) {
   auto fromType = cast<ShapedType>(from.getType());
-  assert(fromType.getRank() == dropDims.size());
+  assert(fromType.getRank() == dropDims.size() && "dropDims dimension does not match from tensor rank");
+  // Computed reassociation map for the corresponding tensor.collapse_shape.
   SmallVector<ReassociationIndices, 2> reassocIdxsVec;
+  // Current reassociation indices to add dropped dimension to.
   ReassociationIndices reassocIdxs;
 
   bool foundKeptDim = false;
+  // Dropped dimensions might be at the beginning or end of the shape so
+  // combine all contiguous dimensions before and after a given non dropped
+  // dimension in reassocIdxs until another non dropped dimension is found.
+  // When that happens, add the reassociation indices to the map.
   for (int dim = 0; dim < fromType.getRank(); dim++) {
     if (!dropDims.test(dim)) {
       if (foundKeptDim) {
diff --git a/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir b/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
index b4fbdfacde899..46b70a9c0edba 100644
--- a/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
+++ b/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
@@ -321,63 +321,47 @@ func.func @pad_generic_static(%small_input: tensor<58x1xf32>, %large_input: tens
 
 // -----
 
-func.func @rank_reduced_extract_slice(%arg0: tensor<6x6x1x1x1x1xf32>, %arg1: tensor<6x6x1x1xf32>, %arg2: tensor<4x6xf32>) -> tensor<4x6xf32> {
-  %c0 = arith.constant 0 : index
-  %c2 = arith.constant 2 : index
-  %c6 = arith.constant 6 : index
+func.func @rank_reduced_extract_slice(%cond : i1) -> tensor<6x2xf32> {
   %cst = arith.constant 0.0 : f32
-  %init1 = tensor.empty() : tensor<6x6x1x1x1x1xf32>
-  %fill1 = linalg.fill ins(%cst : f32) outs(%init1 : tensor<6x6x1x1x1x1xf32>) -> tensor<6x6x1x1x1x1xf32>
-  %0 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3, d4, d6)>, affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d6, d5)>, affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2, d3, d4, d5)>], iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel", "parallel", "reduction"]} ins(%arg0, %arg1 : tensor<6x6x1x1x1x1xf32>, tensor<6x6x1x1xf32>) outs(%fill1 : tensor<6x6x1x1x1x1xf32>) {
-  ^bb0(%in: f32, %in_1: f32, %out: f32):
-    %10 = arith.mulf %in, %in_1 : f32
-    %11 = arith.addf %out, %10 : f32
-    linalg.yield %11 : f32
+  %cst1 = arith.constant 1.0 : f32
+
+  %empty1 = tensor.empty() : tensor<6x6x1x1x1x1xf32>
+  %init1 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d2, d3, d4, d5)>], iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel", "parallel"]} outs(%empty1 : tensor<6x6x1x1x1x1xf32>) {
+  ^bb0(%out: f32):
+    linalg.yield %cst : f32
   } -> tensor<6x6x1x1x1x1xf32>
-  %init2 = tensor.empty() : tensor<4x6xf32>
-  %1 = scf.for %arg4 = %c0 to %c6 step %c2 iter_args(%arg3 = %init2) -> (tensor<4x6xf32>) {
-    %2 = tensor.extract_slice %0[0, %arg4, 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1] : tensor<6x6x1x1x1x1xf32> to tensor<6x2xf32>
-    %init3 = tensor.empty() : tensor<4x2xf32>
-    %fill3 = linalg.fill ins(%cst : f32) outs(%init3 : tensor<4x2xf32>) -> tensor<4x2xf32>
-    %3 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2) -> (d0, d2)>, affine_map<(d0, d1, d2) -> (d2, d1)>, affine_map<(d0, d1, d2) -> (d0, d1)>], iterator_types = ["parallel", "parallel", "reduction"]} ins(%arg2, %2 : tensor<4x6xf32>, tensor<6x2xf32>) outs(%fill3 : tensor<4x2xf32>) {
-    ^bb0(%in: f32, %in_1: f32, %out: f32):
-      %20 = arith.mulf %in, %in_1 : f32
-      %21 = arith.addf %out, %20 : f32
-      linalg.yield %21 : f32
-    } -> tensor<4x2xf32>
-    %4 = tensor.insert_slice %3 into %arg3[0, %arg4] [4, 2] [1, 1]  : tensor<4x2xf32> into tensor<4x6xf32>
-    scf.yield %4 : tensor<4x6xf32>
+
+  %if = scf.if %cond -> tensor<6x2xf32> {
+    %extract0 = tensor.extract_slice %init1[0, 0, 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1] : tensor<6x6x1x1x1x1xf32> to tensor<6x2xf32>
+
+    %init2 = tensor.empty() : tensor<6x2xf32>
+    %add1 = linalg.generic {indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>, affine_map<(d0, d1) -> (d0, d1)>], iterator_types = ["parallel", "parallel"]} ins(%extract0 : tensor<6x2xf32>) outs(%init2 : tensor<6x2xf32>) {
+      ^bb0(%in: f32, %out: f32):
+        %add = arith.addf %in, %cst1 : f32
+        linalg.yield %add : f32
+    } -> tensor<6x2xf32>
+    scf.yield %add1 : tensor<6x2xf32>
+  } else {
+    %extract2 = tensor.extract_slice %init1[0, 2, 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1] : tensor<6x6x1x1x1x1xf32> to tensor<6x2xf32>
+    scf.yield %extract2 : tensor<6x2xf32>
   }
-  return %1 : tensor<4x6xf32>
+
+  return %if : tensor<6x2xf32>
 }
 
 //       CHECK: func @rank_reduced_extract_slice(
-//  CHECK-SAME: %[[ARG0:[0-9a-z]*]]: tensor<6x6x1x1x1x1xf32>
-//  CHECK-SAME: %[[ARG1:[0-9a-z]*]]: tensor<6x6x1x1xf32>
-//  CHECK-SAME: %[[ARG2:[0-9a-z]*]]: tensor<4x6xf32>
+//  CHECK-SAME: %[[COND:[0-9a-z]*]]: i1
 
-//   CHECK-DAG: %[[C0:.*]] = arith.constant 0 : index
-//   CHECK-DAG: %[[C2:.*]] = arith.constant 2 : index
-//   CHECK-DAG: %[[C6:.*]] = arith.constant 6 : index
 //       CHECK: %[[EMPTY_PROD:.*]] = tensor.empty() : tensor<6x6x1x1x1x1xf32>
-//       CHECK: %[[FILL_PROD:.*]] = linalg.fill ins({{%.*}} : f32)
-//  CHECK-SAME:     outs(%[[EMPTY_PROD]] : tensor<6x6x1x1x1x1xf32>) -> tensor<6x6x1x1x1x1xf32>
-//       CHECK: %[[EMPTY_FOR:.*]] = tensor.empty() : tensor<4x6xf32>
-//       CHECK: %[[EMPTY_CONS:.*]] = tensor.empty() : tensor<4x2xf32>
-//       CHECK: %[[FILL_CONS:.*]] = linalg.fill ins({{%.*}} : f32)
-//  CHECK-SAME:     outs(%[[EMPTY_CONS]] : tensor<4x2xf32>) -> tensor<4x2xf32>
-//       CHECK: %[[FOR:.*]] = scf.for %[[I:[0-9a-z]*]] = %[[C0]] to %[[C6]] step %[[C2]] iter_args(%[[ARG_ITER:.*]] = %[[EMPTY_FOR]])
-//   CHECK-DAG:   %[[ARG0_SLICE:.*]] = tensor.extract_slice %[[ARG0]][0, %[[I]], 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1]  : tensor<6x6x1x1x1x1xf32> to tensor<6x2x1x1x1x1xf32>
-//   CHECK-DAG:   %[[ARG1_SLICE:.*]] = tensor.extract_slice %[[ARG1]][0, %[[I]], 0, 0] [6, 2, 1, 1] [1, 1, 1, 1]  : tensor<6x6x1x1xf32> to tensor<6x2x1x1xf32>
-//   CHECK-DAG:   %[[FILL_PROD_SLICE:.*]] = tensor.extract_slice %[[FILL_PROD]][0, %[[I]], 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1] : tensor<6x6x1x1x1x1xf32> to tensor<6x2x1x1x1x1xf32>
-
-//       CHECK:    %[[MMUL_PROD:.*]] = linalg.generic
-//  CHECK-SAME:        ins(%[[ARG0_SLICE]], %[[ARG1_SLICE]] : tensor<6x2x1x1x1x1xf32>, tensor<6x2x1x1xf32>)
-//  CHECK-SAME:        outs(%[[FILL_PROD_SLICE]] : tensor<6x2x1x1x1x1xf32>)
-//       CHECK:    %[[PROD_COLLAPSE:.*]] = tensor.collapse_shape %[[MMUL_PROD]] {{\[\[0\], \[1, 2, 3, 4, 5\]\]}} : tensor<6x2x1x1x1x1xf32> into tensor<6x2xf32>
-//       CHECK:    %[[MMUL_CONS:.*]] = linalg.generic
-//  CHECK-SAME:        ins(%[[ARG2]], %[[PROD_COLLAPSE]] : tensor<4x6xf32>, tensor<6x2xf32>)
-//  CHECK-SAME:        outs(%[[FILL_CONS]] : tensor<4x2xf32>)
-//       CHECK:   %[[CONS_SLICE:.*]] = tensor.insert_slice %[[MMUL_CONS]] into %[[ARG_ITER]][0, %[[I]]] [4, 2] [1, 1] : tensor<4x2xf32> into tensor<4x6xf32>
-//       CHECK:   scf.yield %[[CONS_SLICE]] : tensor<4x6xf32>
-//       CHECK: return %[[FOR]] : tensor<4x6xf32>
+//       CHECK: %[[FILL_PROD:.*]] = linalg.generic
+//  CHECK-SAME:     outs(%[[EMPTY_PROD]] : tensor<6x6x1x1x1x1xf32>)
+
+//       CHECK: %[[EMPTY_CONS:.*]] = tensor.empty() : tensor<6x2xf32>
+//       CHECK: %[[EXTRACT_SLICE_CONS:.*]] = tensor.extract_slice %[[EMPTY_PROD]][0, 0, 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1] : tensor<6x6x1x1x1x1xf32> to tensor<6x2x1x1x1x1xf32>
+
+//       CHECK: %[[FILL_CONS:.*]] = linalg.generic
+//  CHECK-SAME:     outs(%[[EXTRACT_SLICE_CONS]] : tensor<6x2x1x1x1x1xf32>)
+//       CHECK: %[[CONS_COLLAPSE:.*]] = tensor.collapse_shape %[[FILL_CONS]] {{\[\[0\], \[1, 2, 3, 4, 5\]\]}} : tensor<6x2x1x1x1x1xf32> into tensor<6x2xf32>
+//       CHECK: %[[ADD1_CONS:.*]] = linalg.generic
+//  CHECK-SAME:     ins(%[[CONS_COLLAPSE]] : tensor<6x2xf32>)
+//  CHECK-SAME:     outs(%[[EMPTY_CONS]] : tensor<6x2xf32>)

>From b51029d13fb67e2860fdf3832f6b66ea1b30544f Mon Sep 17 00:00:00 2001
From: Thomas Preud'homme <thomas.preudhomme at arm.com>
Date: Wed, 26 Mar 2025 00:01:55 +0000
Subject: [PATCH 3/8] Fix clang-format

---
 mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
index d18d6f7ff8dd8..bcb21263ee68f 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
@@ -239,10 +239,12 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpOperand &consumerOpOperand) {
 
 /// Create tensor.collapse_shape to drop dimensions in `dropDims` in tensor
 /// `from`.
-static tensor::CollapseShapeOp collapseTo(OpBuilder &b, Location loc, Value from,
-                                          const llvm::SmallBitVector &dropDims) {
+static tensor::CollapseShapeOp
+collapseTo(OpBuilder &b, Location loc, Value from,
+           const llvm::SmallBitVector &dropDims) {
   auto fromType = cast<ShapedType>(from.getType());
-  assert(fromType.getRank() == dropDims.size() && "dropDims dimension does not match from tensor rank");
+  assert(fromType.getRank() == dropDims.size() &&
+         "dropDims dimension does not match from tensor rank");
   // Computed reassociation map for the corresponding tensor.collapse_shape.
   SmallVector<ReassociationIndices, 2> reassocIdxsVec;
   // Current reassociation indices to add dropped dimension to.

>From b19b6490e71181c7e956223b3e4f33330fd0fde7 Mon Sep 17 00:00:00 2001
From: Thomas Preud'homme <thomas.preudhomme at arm.com>
Date: Wed, 7 May 2025 10:56:37 +0100
Subject: [PATCH 4/8] Address comments

- rename collapseTo to better reflect its usage
- assert it only collapse unit dimensions
- rename ReassociationIndices-using variables to reassocGroup and
  reassocMaps, the same terminology used in tensor.collapse_shape
  documentation
- use more representative test with comments to better explain what the
  patch does
---
 mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp |  34 +++---
 .../Dialect/Linalg/tile-and-fuse-tensors.mlir | 107 +++++++++++-------
 2 files changed, 87 insertions(+), 54 deletions(-)

diff --git a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
index 3655c43940b06..c52f347135a9e 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
@@ -236,37 +236,39 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpOperand &consumerOpOperand) {
   return fuseProducerOfTensor(b, producerOpResult, consumerOpOperand);
 }
 
-/// Create tensor.collapse_shape to drop dimensions in `dropDims` in tensor
+/// Create tensor.collapse_shape to drop unit dimensions in `dropDims` in tensor
 /// `from`.
 static tensor::CollapseShapeOp
-collapseTo(OpBuilder &b, Location loc, Value from,
-           const llvm::SmallBitVector &dropDims) {
+dropGivenUnitDims(OpBuilder &b, Location loc, Value from,
+                  const llvm::SmallBitVector &dropDims) {
   auto fromType = cast<ShapedType>(from.getType());
-  assert(fromType.getRank() == dropDims.size() &&
+  assert(fromType.getRank() == static_cast<int64_t>(dropDims.size()) &&
          "dropDims dimension does not match from tensor rank");
   // Computed reassociation map for the corresponding tensor.collapse_shape.
-  SmallVector<ReassociationIndices, 2> reassocIdxsVec;
-  // Current reassociation indices to add dropped dimension to.
-  ReassociationIndices reassocIdxs;
+  SmallVector<ReassociationIndices, 2> reassocMaps;
+  // Current reassociation group to add dropped dimension to.
+  ReassociationIndices reassocGroup;
 
   bool foundKeptDim = false;
   // Dropped dimensions might be at the beginning or end of the shape so
   // combine all contiguous dimensions before and after a given non dropped
-  // dimension in reassocIdxs until another non dropped dimension is found.
+  // dimension in reassocGroup until another non dropped dimension is found.
   // When that happens, add the reassociation indices to the map.
   for (int dim = 0; dim < fromType.getRank(); dim++) {
-    if (!dropDims.test(dim)) {
+    if (dropDims.test(dim))
+      assert(fromType.getShape()[dim] == 1 && "Dropping non unit dimension");
+    else {
       if (foundKeptDim) {
-        reassocIdxsVec.push_back(reassocIdxs);
-        reassocIdxs.clear();
+        reassocMaps.push_back(reassocGroup);
+        reassocGroup.clear();
       }
       foundKeptDim = true;
     }
-    reassocIdxs.push_back(dim);
+    reassocGroup.push_back(dim);
   }
-  if (!reassocIdxs.empty())
-    reassocIdxsVec.push_back(reassocIdxs);
-  return b.create<tensor::CollapseShapeOp>(loc, from, reassocIdxsVec);
+  if (!reassocGroup.empty())
+    reassocMaps.push_back(reassocGroup);
+  return b.create<tensor::CollapseShapeOp>(loc, from, reassocMaps);
 }
 
 FailureOr<FusionInfo>
@@ -312,7 +314,7 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpResult producerOpResult,
   // Rank-reduction occured as part of the extract_slice.
   if (cast<ShapedType>(consumerType).getRank() !=
       cast<ShapedType>(def.getType()).getRank())
-    def = collapseTo(b, fusedProducer.getLoc(), def, droppedDims);
+    def = dropGivenUnitDims(b, fusedProducer.getLoc(), def, droppedDims);
   // Canonicalizations are not guaranteed to have happened before constructing
   // `fusedProducer`. In the tensor case this can result in temporary type
   // mismatches. Insert a `tensor.cast` op to propagate the transformation
diff --git a/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir b/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
index 46b70a9c0edba..693a2bb29f76e 100644
--- a/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
+++ b/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
@@ -321,47 +321,78 @@ func.func @pad_generic_static(%small_input: tensor<58x1xf32>, %large_input: tens
 
 // -----
 
-func.func @rank_reduced_extract_slice(%cond : i1) -> tensor<6x2xf32> {
+#map0 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d3)>
+#map1 = affine_map<(d0, d1, d2, d3) -> (d0, d3, d2)>
+#map2 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2)>
+#map3 = affine_map<(d0, d1, d2) -> (d0, d2)>
+#map4 = affine_map<(d0, d1, d2) -> (d2, d1)>
+#map5 = affine_map<(d0, d1, d2) -> (d0, d1)>
+func.func @rank_reduced_extract_slice(%arg0: tensor<1x6x5xf32>, %arg1: tensor<1x5x6xf32>, %arg2: tensor<4x6xf32>) -> tensor<4x6xf32> {
+  %c0 = arith.constant 0 : index
+  %c2 = arith.constant 2 : index
+  %c6 = arith.constant 6 : index
   %cst = arith.constant 0.0 : f32
-  %cst1 = arith.constant 1.0 : f32
-
-  %empty1 = tensor.empty() : tensor<6x6x1x1x1x1xf32>
-  %init1 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d2, d3, d4, d5)>], iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel", "parallel"]} outs(%empty1 : tensor<6x6x1x1x1x1xf32>) {
-  ^bb0(%out: f32):
-    linalg.yield %cst : f32
-  } -> tensor<6x6x1x1x1x1xf32>
-
-  %if = scf.if %cond -> tensor<6x2xf32> {
-    %extract0 = tensor.extract_slice %init1[0, 0, 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1] : tensor<6x6x1x1x1x1xf32> to tensor<6x2xf32>
-
-    %init2 = tensor.empty() : tensor<6x2xf32>
-    %add1 = linalg.generic {indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>, affine_map<(d0, d1) -> (d0, d1)>], iterator_types = ["parallel", "parallel"]} ins(%extract0 : tensor<6x2xf32>) outs(%init2 : tensor<6x2xf32>) {
-      ^bb0(%in: f32, %out: f32):
-        %add = arith.addf %in, %cst1 : f32
-        linalg.yield %add : f32
-    } -> tensor<6x2xf32>
-    scf.yield %add1 : tensor<6x2xf32>
-  } else {
-    %extract2 = tensor.extract_slice %init1[0, 2, 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1] : tensor<6x6x1x1x1x1xf32> to tensor<6x2xf32>
-    scf.yield %extract2 : tensor<6x2xf32>
+  %init1 = tensor.empty() : tensor<1x6x6xf32>
+  %fill1 = linalg.fill ins(%cst : f32) outs(%init1 : tensor<1x6x6xf32>) -> tensor<1x6x6xf32>
+  %0 = linalg.generic
+    {indexing_maps = [#map0, #map1, #map2], iterator_types = ["parallel", "parallel", "parallel", "reduction"]}
+    ins(%arg0, %arg1 : tensor<1x6x5xf32>, tensor<1x5x6xf32>) outs(%fill1 : tensor<1x6x6xf32>) {
+  ^bb0(%in: f32, %in_1: f32, %out: f32):
+    %10 = arith.mulf %in, %in_1 : f32
+    %11 = arith.addf %out, %10 : f32
+    linalg.yield %11 : f32
+  } -> tensor<1x6x6xf32>
+  %init2 = tensor.empty() : tensor<4x6xf32>
+  %1 = scf.for %arg4 = %c0 to %c6 step %c2 iter_args(%arg3 = %init2) -> (tensor<4x6xf32>) {
+    %2 = tensor.extract_slice %0[0, 0, %arg4] [1, 6, 2] [1, 1, 1] : tensor<1x6x6xf32> to tensor<6x2xf32>
+    %init3 = tensor.empty() : tensor<4x2xf32>
+    %fill3 = linalg.fill ins(%cst : f32) outs(%init3 : tensor<4x2xf32>) -> tensor<4x2xf32>
+    %3 = linalg.generic
+     {indexing_maps = [#map3, #map4, #map5], iterator_types = ["parallel", "parallel", "reduction"]}
+     ins(%arg2, %2 : tensor<4x6xf32>, tensor<6x2xf32>) outs(%fill3 : tensor<4x2xf32>) {
+    ^bb0(%in: f32, %in_1: f32, %out: f32):
+      %20 = arith.mulf %in, %in_1 : f32
+      %21 = arith.addf %out, %20 : f32
+      linalg.yield %21 : f32
+    } -> tensor<4x2xf32>
+    %4 = tensor.insert_slice %3 into %arg3[0, %arg4] [4, 2] [1, 1]  : tensor<4x2xf32> into tensor<4x6xf32>
+    scf.yield %4 : tensor<4x6xf32>
   }
-
-  return %if : tensor<6x2xf32>
+  return %1 : tensor<4x6xf32>
 }
 
 //       CHECK: func @rank_reduced_extract_slice(
-//  CHECK-SAME: %[[COND:[0-9a-z]*]]: i1
-
-//       CHECK: %[[EMPTY_PROD:.*]] = tensor.empty() : tensor<6x6x1x1x1x1xf32>
-//       CHECK: %[[FILL_PROD:.*]] = linalg.generic
-//  CHECK-SAME:     outs(%[[EMPTY_PROD]] : tensor<6x6x1x1x1x1xf32>)
+//  CHECK-SAME: %[[ARG0:[0-9a-z]*]]: tensor<1x6x5xf32>
+//  CHECK-SAME: %[[ARG1:[0-9a-z]*]]: tensor<1x5x6xf32>
+//  CHECK-SAME: %[[ARG2:[0-9a-z]*]]: tensor<4x6xf32>
 
-//       CHECK: %[[EMPTY_CONS:.*]] = tensor.empty() : tensor<6x2xf32>
-//       CHECK: %[[EXTRACT_SLICE_CONS:.*]] = tensor.extract_slice %[[EMPTY_PROD]][0, 0, 0, 0, 0, 0] [6, 2, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1] : tensor<6x6x1x1x1x1xf32> to tensor<6x2x1x1x1x1xf32>
-
-//       CHECK: %[[FILL_CONS:.*]] = linalg.generic
-//  CHECK-SAME:     outs(%[[EXTRACT_SLICE_CONS]] : tensor<6x2x1x1x1x1xf32>)
-//       CHECK: %[[CONS_COLLAPSE:.*]] = tensor.collapse_shape %[[FILL_CONS]] {{\[\[0\], \[1, 2, 3, 4, 5\]\]}} : tensor<6x2x1x1x1x1xf32> into tensor<6x2xf32>
-//       CHECK: %[[ADD1_CONS:.*]] = linalg.generic
-//  CHECK-SAME:     ins(%[[CONS_COLLAPSE]] : tensor<6x2xf32>)
-//  CHECK-SAME:     outs(%[[EMPTY_CONS]] : tensor<6x2xf32>)
+//   CHECK-DAG: %[[C0:.*]] = arith.constant 0 : index
+//   CHECK-DAG: %[[C2:.*]] = arith.constant 2 : index
+//   CHECK-DAG: %[[C6:.*]] = arith.constant 6 : index
+//   CHECK-DAG: %[[CST:.*]] = arith.constant 0.000000e+00 : f32
+//       CHECK: %[[EMPTY_PROD:.*]] = tensor.empty() : tensor<1x6x6xf32>
+//  CHECK-NEXT: %[[FILL_PROD:.*]] = linalg.fill ins(%[[CST]] : f32) outs(%[[EMPTY_PROD]] : tensor<1x6x6xf32>) -> tensor<1x6x6xf32>
+//  CHECK-NEXT: %[[EMPTY_FOR:.*]] = tensor.empty() : tensor<4x6xf32>
+//  CHECK-NEXT: %[[EMPTY_CONS:.*]] = tensor.empty() : tensor<4x2xf32>
+//  CHECK-NEXT: %[[FILL_CONS:.*]] = linalg.fill ins(%[[CST]] : f32)
+
+//  For loop right after tensor alloc & fill, no linalg.generic.
+//  CHECK-NEXT: %[[FOR:.*]] = scf.for %[[I:[0-9a-z]*]] = %[[C0]] to %[[C6]] step %[[C2]] iter_args(%[[ARG_ITER:.*]] = %[[EMPTY_FOR]])
+
+//  Producer linalg.generic now inside the loop, with tiled args sliced before
+//  it.
+//   CHECK-DAG:   %[[ARG1_SLICE:.*]] = tensor.extract_slice %[[ARG1]][0, 0, %[[I]]] [1, 5, 2] [1, 1, 1]  : tensor<1x5x6xf32> to tensor<1x5x2xf32>
+//   CHECK-DAG:   %[[PROD_SLICE:.*]] = tensor.extract_slice %[[FILL_PROD]][0, 0, %[[I]]] [1, 6, 2] [1, 1, 1]  : tensor<1x6x6xf32> to tensor<1x6x2xf32>
+//       CHECK:    %[[MMUL_PROD:.*]] = linalg.generic
+//  CHECK-SAME:        ins(%[[ARG0]], %[[ARG1_SLICE]] : tensor<1x6x5xf32>, tensor<1x5x2xf32>)
+//  CHECK-SAME:        outs(%[[PROD_SLICE]] : tensor<1x6x2xf32>)
+//
+//  Consumer uses a rank-reduced version of producer result so a collapse_shape
+//  is generated.
+//       CHECK:    %[[PROD_COLLAPSE:.*]] = tensor.collapse_shape %[[MMUL_PROD]] {{\[\[0, 1\], \[2\]\]}} : tensor<1x6x2xf32> into tensor<6x2xf32>
+//       CHECK:    %[[MMUL_CONS:.*]] = linalg.generic
+//  CHECK-SAME:        ins(%[[ARG2]], %[[PROD_COLLAPSE]] : tensor<4x6xf32>, tensor<6x2xf32>)
+//  CHECK-SAME:        outs(%[[FILL_CONS]] : tensor<4x2xf32>)
+//       CHECK:   %[[CONS_SLICE:.*]] = tensor.insert_slice %[[MMUL_CONS]] into %[[ARG_ITER]][0, %[[I]]] [4, 2] [1, 1] : tensor<4x2xf32> into tensor<4x6xf32>
+//       CHECK:   scf.yield %[[CONS_SLICE]] : tensor<4x6xf32>
+//       CHECK: return %[[FOR]] : tensor<4x6xf32>

>From 49deedf97d448f7bb71d47da0b0a1bbe7b53db96 Mon Sep 17 00:00:00 2001
From: Thomas Preud'homme <thomas.preudhomme at arm.com>
Date: Mon, 19 May 2025 23:05:52 +0100
Subject: [PATCH 5/8] Clean up code

dropGivenUnitDims():
- move assert out of loop
- rework algorithm to make grouping more explicit and avoid complex
  nested ifs
- fix occured typo

Test: remove all tensor.empty and linalg.fill
---
 mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp | 42 +++++++++----------
 .../Dialect/Linalg/tile-and-fuse-tensors.mlir | 37 +++++++---------
 2 files changed, 36 insertions(+), 43 deletions(-)

diff --git a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
index c52f347135a9e..d69c85984aa4e 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
@@ -242,32 +242,30 @@ static tensor::CollapseShapeOp
 dropGivenUnitDims(OpBuilder &b, Location loc, Value from,
                   const llvm::SmallBitVector &dropDims) {
   auto fromType = cast<ShapedType>(from.getType());
-  assert(fromType.getRank() == static_cast<int64_t>(dropDims.size()) &&
+  int64_t rank = fromType.getRank();
+  assert(rank == static_cast<int64_t>(dropDims.size()) &&
          "dropDims dimension does not match from tensor rank");
+  assert(llvm::all_of(
+             dropDims.set_bits(),
+             [&](unsigned dim) { return fromType.getShape()[dim] == 1; }) &&
+         "Dropping non unit dimension");
   // Computed reassociation map for the corresponding tensor.collapse_shape.
   SmallVector<ReassociationIndices, 2> reassocMaps;
   // Current reassociation group to add dropped dimension to.
-  ReassociationIndices reassocGroup;
-
-  bool foundKeptDim = false;
-  // Dropped dimensions might be at the beginning or end of the shape so
-  // combine all contiguous dimensions before and after a given non dropped
-  // dimension in reassocGroup until another non dropped dimension is found.
-  // When that happens, add the reassociation indices to the map.
-  for (int dim = 0; dim < fromType.getRank(); dim++) {
-    if (dropDims.test(dim))
-      assert(fromType.getShape()[dim] == 1 && "Dropping non unit dimension");
-    else {
-      if (foundKeptDim) {
-        reassocMaps.push_back(reassocGroup);
-        reassocGroup.clear();
-      }
-      foundKeptDim = true;
-    }
-    reassocGroup.push_back(dim);
+
+  int64_t nextDimToGroup = 0;
+  llvm::SmallBitVector keptDims(dropDims);
+  keptDims.flip();
+  int64_t lastSetBit = keptDims.find_last();
+  for(int64_t setBit : keptDims.set_bits()) {
+    // Group consecutive dropped dimension with the next non-dropped dimension.
+    // If this is the last set dimension, also group all subsequent dropped
+    // dimension, if any.
+    int64_t upTo = setBit == lastSetBit ? rank - 1 : setBit;
+    auto seq = llvm::seq_inclusive(nextDimToGroup, upTo);
+    reassocMaps.emplace_back(llvm::make_range(seq.begin(), seq.end()));
+    nextDimToGroup = setBit + 1;
   }
-  if (!reassocGroup.empty())
-    reassocMaps.push_back(reassocGroup);
   return b.create<tensor::CollapseShapeOp>(loc, from, reassocMaps);
 }
 
@@ -311,7 +309,7 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpResult producerOpResult,
   // Replace use.
   Value def = fusedProducer->getResult(producerOpResult.getResultNumber());
   Type consumerType = consumerOpOperand.get().getType();
-  // Rank-reduction occured as part of the extract_slice.
+  // Rank-reduction occurred as part of the extract_slice.
   if (cast<ShapedType>(consumerType).getRank() !=
       cast<ShapedType>(def.getType()).getRank())
     def = dropGivenUnitDims(b, fusedProducer.getLoc(), def, droppedDims);
diff --git a/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir b/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
index 693a2bb29f76e..9340e70b4d507 100644
--- a/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
+++ b/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
@@ -327,35 +327,32 @@ func.func @pad_generic_static(%small_input: tensor<58x1xf32>, %large_input: tens
 #map3 = affine_map<(d0, d1, d2) -> (d0, d2)>
 #map4 = affine_map<(d0, d1, d2) -> (d2, d1)>
 #map5 = affine_map<(d0, d1, d2) -> (d0, d1)>
-func.func @rank_reduced_extract_slice(%arg0: tensor<1x6x5xf32>, %arg1: tensor<1x5x6xf32>, %arg2: tensor<4x6xf32>) -> tensor<4x6xf32> {
+func.func @rank_reduced_extract_slice(
+    %arg0: tensor<1x6x5xf32>, %arg1: tensor<1x5x6xf32>, %arg2: tensor<4x6xf32>,
+    %arg3: tensor<1x6x6xf32>, %arg4: tensor<4x6xf32>, %arg5: tensor<4x2xf32>
+) -> tensor<4x6xf32> {
   %c0 = arith.constant 0 : index
   %c2 = arith.constant 2 : index
   %c6 = arith.constant 6 : index
-  %cst = arith.constant 0.0 : f32
-  %init1 = tensor.empty() : tensor<1x6x6xf32>
-  %fill1 = linalg.fill ins(%cst : f32) outs(%init1 : tensor<1x6x6xf32>) -> tensor<1x6x6xf32>
   %0 = linalg.generic
     {indexing_maps = [#map0, #map1, #map2], iterator_types = ["parallel", "parallel", "parallel", "reduction"]}
-    ins(%arg0, %arg1 : tensor<1x6x5xf32>, tensor<1x5x6xf32>) outs(%fill1 : tensor<1x6x6xf32>) {
+    ins(%arg0, %arg1 : tensor<1x6x5xf32>, tensor<1x5x6xf32>) outs(%arg3 : tensor<1x6x6xf32>) {
   ^bb0(%in: f32, %in_1: f32, %out: f32):
     %10 = arith.mulf %in, %in_1 : f32
     %11 = arith.addf %out, %10 : f32
     linalg.yield %11 : f32
   } -> tensor<1x6x6xf32>
-  %init2 = tensor.empty() : tensor<4x6xf32>
-  %1 = scf.for %arg4 = %c0 to %c6 step %c2 iter_args(%arg3 = %init2) -> (tensor<4x6xf32>) {
-    %2 = tensor.extract_slice %0[0, 0, %arg4] [1, 6, 2] [1, 1, 1] : tensor<1x6x6xf32> to tensor<6x2xf32>
-    %init3 = tensor.empty() : tensor<4x2xf32>
-    %fill3 = linalg.fill ins(%cst : f32) outs(%init3 : tensor<4x2xf32>) -> tensor<4x2xf32>
+  %1 = scf.for %arg7 = %c0 to %c6 step %c2 iter_args(%arg6 = %arg4) -> (tensor<4x6xf32>) {
+    %2 = tensor.extract_slice %0[0, 0, %arg7] [1, 6, 2] [1, 1, 1] : tensor<1x6x6xf32> to tensor<6x2xf32>
     %3 = linalg.generic
      {indexing_maps = [#map3, #map4, #map5], iterator_types = ["parallel", "parallel", "reduction"]}
-     ins(%arg2, %2 : tensor<4x6xf32>, tensor<6x2xf32>) outs(%fill3 : tensor<4x2xf32>) {
+     ins(%arg2, %2 : tensor<4x6xf32>, tensor<6x2xf32>) outs(%arg5 : tensor<4x2xf32>) {
     ^bb0(%in: f32, %in_1: f32, %out: f32):
       %20 = arith.mulf %in, %in_1 : f32
       %21 = arith.addf %out, %20 : f32
       linalg.yield %21 : f32
     } -> tensor<4x2xf32>
-    %4 = tensor.insert_slice %3 into %arg3[0, %arg4] [4, 2] [1, 1]  : tensor<4x2xf32> into tensor<4x6xf32>
+    %4 = tensor.insert_slice %3 into %arg6[0, %arg7] [4, 2] [1, 1]  : tensor<4x2xf32> into tensor<4x6xf32>
     scf.yield %4 : tensor<4x6xf32>
   }
   return %1 : tensor<4x6xf32>
@@ -365,24 +362,22 @@ func.func @rank_reduced_extract_slice(%arg0: tensor<1x6x5xf32>, %arg1: tensor<1x
 //  CHECK-SAME: %[[ARG0:[0-9a-z]*]]: tensor<1x6x5xf32>
 //  CHECK-SAME: %[[ARG1:[0-9a-z]*]]: tensor<1x5x6xf32>
 //  CHECK-SAME: %[[ARG2:[0-9a-z]*]]: tensor<4x6xf32>
+//  CHECK-SAME: %[[ARG3:[0-9a-z]*]]: tensor<1x6x6xf32>
+//  CHECK-SAME: %[[ARG4:[0-9a-z]*]]: tensor<4x6xf32>
+//  CHECK-SAME: %[[ARG5:[0-9a-z]*]]: tensor<4x2xf32>
 
 //   CHECK-DAG: %[[C0:.*]] = arith.constant 0 : index
 //   CHECK-DAG: %[[C2:.*]] = arith.constant 2 : index
 //   CHECK-DAG: %[[C6:.*]] = arith.constant 6 : index
-//   CHECK-DAG: %[[CST:.*]] = arith.constant 0.000000e+00 : f32
-//       CHECK: %[[EMPTY_PROD:.*]] = tensor.empty() : tensor<1x6x6xf32>
-//  CHECK-NEXT: %[[FILL_PROD:.*]] = linalg.fill ins(%[[CST]] : f32) outs(%[[EMPTY_PROD]] : tensor<1x6x6xf32>) -> tensor<1x6x6xf32>
-//  CHECK-NEXT: %[[EMPTY_FOR:.*]] = tensor.empty() : tensor<4x6xf32>
-//  CHECK-NEXT: %[[EMPTY_CONS:.*]] = tensor.empty() : tensor<4x2xf32>
-//  CHECK-NEXT: %[[FILL_CONS:.*]] = linalg.fill ins(%[[CST]] : f32)
 
 //  For loop right after tensor alloc & fill, no linalg.generic.
-//  CHECK-NEXT: %[[FOR:.*]] = scf.for %[[I:[0-9a-z]*]] = %[[C0]] to %[[C6]] step %[[C2]] iter_args(%[[ARG_ITER:.*]] = %[[EMPTY_FOR]])
+//   CHECK-NOT: linalg.generic
+//  CHECK-NEXT: %[[FOR:.*]] = scf.for %[[I:[0-9a-z]*]] = %[[C0]] to %[[C6]] step %[[C2]] iter_args(%[[ARG_ITER:.*]] = %[[ARG4]])
 
 //  Producer linalg.generic now inside the loop, with tiled args sliced before
 //  it.
 //   CHECK-DAG:   %[[ARG1_SLICE:.*]] = tensor.extract_slice %[[ARG1]][0, 0, %[[I]]] [1, 5, 2] [1, 1, 1]  : tensor<1x5x6xf32> to tensor<1x5x2xf32>
-//   CHECK-DAG:   %[[PROD_SLICE:.*]] = tensor.extract_slice %[[FILL_PROD]][0, 0, %[[I]]] [1, 6, 2] [1, 1, 1]  : tensor<1x6x6xf32> to tensor<1x6x2xf32>
+//   CHECK-DAG:   %[[PROD_SLICE:.*]] = tensor.extract_slice %[[ARG3]][0, 0, %[[I]]] [1, 6, 2] [1, 1, 1]  : tensor<1x6x6xf32> to tensor<1x6x2xf32>
 //       CHECK:    %[[MMUL_PROD:.*]] = linalg.generic
 //  CHECK-SAME:        ins(%[[ARG0]], %[[ARG1_SLICE]] : tensor<1x6x5xf32>, tensor<1x5x2xf32>)
 //  CHECK-SAME:        outs(%[[PROD_SLICE]] : tensor<1x6x2xf32>)
@@ -392,7 +387,7 @@ func.func @rank_reduced_extract_slice(%arg0: tensor<1x6x5xf32>, %arg1: tensor<1x
 //       CHECK:    %[[PROD_COLLAPSE:.*]] = tensor.collapse_shape %[[MMUL_PROD]] {{\[\[0, 1\], \[2\]\]}} : tensor<1x6x2xf32> into tensor<6x2xf32>
 //       CHECK:    %[[MMUL_CONS:.*]] = linalg.generic
 //  CHECK-SAME:        ins(%[[ARG2]], %[[PROD_COLLAPSE]] : tensor<4x6xf32>, tensor<6x2xf32>)
-//  CHECK-SAME:        outs(%[[FILL_CONS]] : tensor<4x2xf32>)
+//  CHECK-SAME:        outs(%[[ARG5]] : tensor<4x2xf32>)
 //       CHECK:   %[[CONS_SLICE:.*]] = tensor.insert_slice %[[MMUL_CONS]] into %[[ARG_ITER]][0, %[[I]]] [4, 2] [1, 1] : tensor<4x2xf32> into tensor<4x6xf32>
 //       CHECK:   scf.yield %[[CONS_SLICE]] : tensor<4x6xf32>
 //       CHECK: return %[[FOR]] : tensor<4x6xf32>

>From cf20e80c175bbfd6990aa3d58f0d972e796f7510 Mon Sep 17 00:00:00 2001
From: Thomas Preud'homme <thomas.preudhomme at arm.com>
Date: Tue, 20 May 2025 11:33:20 +0100
Subject: [PATCH 6/8] Fix codestyle

---
 mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
index d69c85984aa4e..f983fb5e40fa7 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
@@ -257,7 +257,7 @@ dropGivenUnitDims(OpBuilder &b, Location loc, Value from,
   llvm::SmallBitVector keptDims(dropDims);
   keptDims.flip();
   int64_t lastSetBit = keptDims.find_last();
-  for(int64_t setBit : keptDims.set_bits()) {
+  for (int64_t setBit : keptDims.set_bits()) {
     // Group consecutive dropped dimension with the next non-dropped dimension.
     // If this is the last set dimension, also group all subsequent dropped
     // dimension, if any.

>From 42d8959bc2594be9218c1e709320e048cde8cc84 Mon Sep 17 00:00:00 2001
From: Thomas Preud'homme <thomas.preudhomme at arm.com>
Date: Tue, 20 May 2025 13:26:57 +0100
Subject: [PATCH 7/8] Move dropGivenUnitDims to Tensor Utils

---
 .../include/mlir/Dialect/Tensor/Utils/Utils.h |  5 +++
 mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp | 36 ++-----------------
 mlir/lib/Dialect/Tensor/Utils/Utils.cpp       | 33 +++++++++++++++++
 3 files changed, 40 insertions(+), 34 deletions(-)

diff --git a/mlir/include/mlir/Dialect/Tensor/Utils/Utils.h b/mlir/include/mlir/Dialect/Tensor/Utils/Utils.h
index 22ca8a99dd7db..6c2a55f67db87 100644
--- a/mlir/include/mlir/Dialect/Tensor/Utils/Utils.h
+++ b/mlir/include/mlir/Dialect/Tensor/Utils/Utils.h
@@ -43,6 +43,11 @@ FailureOr<RankedTensorType>
 computeTransposedType(RankedTensorType rankedTensorType,
                       ArrayRef<int64_t> transposeVector);
 
+/// Create tensor.collapse_shape to drop unit dimensions in `dropDims` in tensor
+/// `from`.
+CollapseShapeOp dropGivenUnitDims(OpBuilder &b, Location loc, Value from,
+                                  const llvm::SmallBitVector &dropDims);
+
 /// A tensor.insert_slice is a cast-like operation if it merely rank-extends the
 /// source tensor or inserts the source tensor into a destination tensor with
 /// the same shape.
diff --git a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
index f983fb5e40fa7..e3673e2a385c0 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
@@ -236,39 +236,6 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpOperand &consumerOpOperand) {
   return fuseProducerOfTensor(b, producerOpResult, consumerOpOperand);
 }
 
-/// Create tensor.collapse_shape to drop unit dimensions in `dropDims` in tensor
-/// `from`.
-static tensor::CollapseShapeOp
-dropGivenUnitDims(OpBuilder &b, Location loc, Value from,
-                  const llvm::SmallBitVector &dropDims) {
-  auto fromType = cast<ShapedType>(from.getType());
-  int64_t rank = fromType.getRank();
-  assert(rank == static_cast<int64_t>(dropDims.size()) &&
-         "dropDims dimension does not match from tensor rank");
-  assert(llvm::all_of(
-             dropDims.set_bits(),
-             [&](unsigned dim) { return fromType.getShape()[dim] == 1; }) &&
-         "Dropping non unit dimension");
-  // Computed reassociation map for the corresponding tensor.collapse_shape.
-  SmallVector<ReassociationIndices, 2> reassocMaps;
-  // Current reassociation group to add dropped dimension to.
-
-  int64_t nextDimToGroup = 0;
-  llvm::SmallBitVector keptDims(dropDims);
-  keptDims.flip();
-  int64_t lastSetBit = keptDims.find_last();
-  for (int64_t setBit : keptDims.set_bits()) {
-    // Group consecutive dropped dimension with the next non-dropped dimension.
-    // If this is the last set dimension, also group all subsequent dropped
-    // dimension, if any.
-    int64_t upTo = setBit == lastSetBit ? rank - 1 : setBit;
-    auto seq = llvm::seq_inclusive(nextDimToGroup, upTo);
-    reassocMaps.emplace_back(llvm::make_range(seq.begin(), seq.end()));
-    nextDimToGroup = setBit + 1;
-  }
-  return b.create<tensor::CollapseShapeOp>(loc, from, reassocMaps);
-}
-
 FailureOr<FusionInfo>
 mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpResult producerOpResult,
                                    OpOperand &consumerOpOperand) {
@@ -312,7 +279,8 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpResult producerOpResult,
   // Rank-reduction occurred as part of the extract_slice.
   if (cast<ShapedType>(consumerType).getRank() !=
       cast<ShapedType>(def.getType()).getRank())
-    def = dropGivenUnitDims(b, fusedProducer.getLoc(), def, droppedDims);
+    def =
+        tensor::dropGivenUnitDims(b, fusedProducer.getLoc(), def, droppedDims);
   // Canonicalizations are not guaranteed to have happened before constructing
   // `fusedProducer`. In the tensor case this can result in temporary type
   // mismatches. Insert a `tensor.cast` op to propagate the transformation
diff --git a/mlir/lib/Dialect/Tensor/Utils/Utils.cpp b/mlir/lib/Dialect/Tensor/Utils/Utils.cpp
index c3d56759a896a..53a219dff48c5 100644
--- a/mlir/lib/Dialect/Tensor/Utils/Utils.cpp
+++ b/mlir/lib/Dialect/Tensor/Utils/Utils.cpp
@@ -94,6 +94,39 @@ mlir::tensor::computeTransposedType(RankedTensorType rankedTensorType,
   return transposedTensorType;
 }
 
+/// Create tensor.collapse_shape to drop unit dimensions in `dropDims` in tensor
+/// `from`.
+CollapseShapeOp
+mlir::tensor::dropGivenUnitDims(OpBuilder &b, Location loc, Value from,
+                                const llvm::SmallBitVector &dropDims) {
+  auto fromType = cast<ShapedType>(from.getType());
+  int64_t rank = fromType.getRank();
+  assert(rank == static_cast<int64_t>(dropDims.size()) &&
+         "dropDims dimension does not match from tensor rank");
+  assert(llvm::all_of(
+             dropDims.set_bits(),
+             [&](unsigned dim) { return fromType.getShape()[dim] == 1; }) &&
+         "Dropping non unit dimension");
+  // Computed reassociation map for the corresponding tensor.collapse_shape.
+  SmallVector<ReassociationIndices, 2> reassocMaps;
+  // Current reassociation group to add dropped dimension to.
+
+  int64_t nextDimToGroup = 0;
+  llvm::SmallBitVector keptDims(dropDims);
+  keptDims.flip();
+  int64_t lastSetBit = keptDims.find_last();
+  for (int64_t setBit : keptDims.set_bits()) {
+    // Group consecutive dropped dimension with the next non-dropped dimension.
+    // If this is the last set dimension, also group all subsequent dropped
+    // dimension, if any.
+    int64_t upTo = setBit == lastSetBit ? rank - 1 : setBit;
+    auto seq = llvm::seq_inclusive(nextDimToGroup, upTo);
+    reassocMaps.emplace_back(llvm::make_range(seq.begin(), seq.end()));
+    nextDimToGroup = setBit + 1;
+  }
+  return b.create<tensor::CollapseShapeOp>(loc, from, reassocMaps);
+}
+
 bool mlir::tensor::isCastLikeInsertSliceOp(InsertSliceOp op) {
   llvm::SmallBitVector droppedDims = op.getDroppedDims();
   int64_t srcDim = 0;

>From 6fc320ddd5696da921939eae2bea403c37c8568b Mon Sep 17 00:00:00 2001
From: Thomas Preud'homme <thomas.preudhomme at arm.com>
Date: Tue, 20 May 2025 22:43:32 +0100
Subject: [PATCH 8/8] Address review comments

Utils:
- drop comments on implementation
- rename from into src

Fusion:
- restrict live range of droppedDims
- clarify comment for rank-reduction check

Test:
- Use more descriptive SSA and FileCheck variables
- Emphasize the rank-reducing extract_slice in the input IR as the key
  aspect of the test.
---
 .../include/mlir/Dialect/Tensor/Utils/Utils.h |  4 +-
 mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp |  8 +--
 mlir/lib/Dialect/Tensor/Utils/Utils.cpp       | 14 +++--
 .../Dialect/Linalg/tile-and-fuse-tensors.mlir | 51 ++++++++++---------
 4 files changed, 41 insertions(+), 36 deletions(-)

diff --git a/mlir/include/mlir/Dialect/Tensor/Utils/Utils.h b/mlir/include/mlir/Dialect/Tensor/Utils/Utils.h
index 6c2a55f67db87..1a4733df3f187 100644
--- a/mlir/include/mlir/Dialect/Tensor/Utils/Utils.h
+++ b/mlir/include/mlir/Dialect/Tensor/Utils/Utils.h
@@ -44,8 +44,8 @@ computeTransposedType(RankedTensorType rankedTensorType,
                       ArrayRef<int64_t> transposeVector);
 
 /// Create tensor.collapse_shape to drop unit dimensions in `dropDims` in tensor
-/// `from`.
-CollapseShapeOp dropGivenUnitDims(OpBuilder &b, Location loc, Value from,
+/// `src`.
+CollapseShapeOp dropGivenUnitDims(OpBuilder &b, Location loc, Value src,
                                   const llvm::SmallBitVector &dropDims);
 
 /// A tensor.insert_slice is a cast-like operation if it merely rank-extends the
diff --git a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
index e3673e2a385c0..4fc8a17554435 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
@@ -256,7 +256,6 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpResult producerOpResult,
                << "\nNot fusable, not an extract_slice op: " << inputTensor);
     return failure();
   }
-  llvm::SmallBitVector droppedDims = sliceOp.getDroppedDims();
 
   // If producer is already in the same block as consumer, we are done.
   if (consumerOpOperand.get().getParentBlock() ==
@@ -276,11 +275,14 @@ mlir::linalg::fuseProducerOfTensor(OpBuilder &b, OpResult producerOpResult,
   // Replace use.
   Value def = fusedProducer->getResult(producerOpResult.getResultNumber());
   Type consumerType = consumerOpOperand.get().getType();
-  // Rank-reduction occurred as part of the extract_slice.
+  // Check if rank-reduction occurred as part of the extract_slice. If yes,
+  // collapse the dropped dimensions.
   if (cast<ShapedType>(consumerType).getRank() !=
-      cast<ShapedType>(def.getType()).getRank())
+      cast<ShapedType>(def.getType()).getRank()) {
+    llvm::SmallBitVector droppedDims = sliceOp.getDroppedDims();
     def =
         tensor::dropGivenUnitDims(b, fusedProducer.getLoc(), def, droppedDims);
+  }
   // Canonicalizations are not guaranteed to have happened before constructing
   // `fusedProducer`. In the tensor case this can result in temporary type
   // mismatches. Insert a `tensor.cast` op to propagate the transformation
diff --git a/mlir/lib/Dialect/Tensor/Utils/Utils.cpp b/mlir/lib/Dialect/Tensor/Utils/Utils.cpp
index 53a219dff48c5..11ae0108594dd 100644
--- a/mlir/lib/Dialect/Tensor/Utils/Utils.cpp
+++ b/mlir/lib/Dialect/Tensor/Utils/Utils.cpp
@@ -94,18 +94,16 @@ mlir::tensor::computeTransposedType(RankedTensorType rankedTensorType,
   return transposedTensorType;
 }
 
-/// Create tensor.collapse_shape to drop unit dimensions in `dropDims` in tensor
-/// `from`.
 CollapseShapeOp
-mlir::tensor::dropGivenUnitDims(OpBuilder &b, Location loc, Value from,
+mlir::tensor::dropGivenUnitDims(OpBuilder &b, Location loc, Value src,
                                 const llvm::SmallBitVector &dropDims) {
-  auto fromType = cast<ShapedType>(from.getType());
-  int64_t rank = fromType.getRank();
+  auto srcType = cast<ShapedType>(src.getType());
+  int64_t rank = srcType.getRank();
   assert(rank == static_cast<int64_t>(dropDims.size()) &&
-         "dropDims dimension does not match from tensor rank");
+         "dropDims dimension does not match src tensor rank");
   assert(llvm::all_of(
              dropDims.set_bits(),
-             [&](unsigned dim) { return fromType.getShape()[dim] == 1; }) &&
+             [&](unsigned dim) { return srcType.getShape()[dim] == 1; }) &&
          "Dropping non unit dimension");
   // Computed reassociation map for the corresponding tensor.collapse_shape.
   SmallVector<ReassociationIndices, 2> reassocMaps;
@@ -124,7 +122,7 @@ mlir::tensor::dropGivenUnitDims(OpBuilder &b, Location loc, Value from,
     reassocMaps.emplace_back(llvm::make_range(seq.begin(), seq.end()));
     nextDimToGroup = setBit + 1;
   }
-  return b.create<tensor::CollapseShapeOp>(loc, from, reassocMaps);
+  return b.create<tensor::CollapseShapeOp>(loc, src, reassocMaps);
 }
 
 bool mlir::tensor::isCastLikeInsertSliceOp(InsertSliceOp op) {
diff --git a/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir b/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
index 9340e70b4d507..fd755a208b2c9 100644
--- a/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
+++ b/mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir
@@ -328,43 +328,48 @@ func.func @pad_generic_static(%small_input: tensor<58x1xf32>, %large_input: tens
 #map4 = affine_map<(d0, d1, d2) -> (d2, d1)>
 #map5 = affine_map<(d0, d1, d2) -> (d0, d1)>
 func.func @rank_reduced_extract_slice(
-    %arg0: tensor<1x6x5xf32>, %arg1: tensor<1x5x6xf32>, %arg2: tensor<4x6xf32>,
-    %arg3: tensor<1x6x6xf32>, %arg4: tensor<4x6xf32>, %arg5: tensor<4x2xf32>
+    %prod_in: tensor<1x6x5xf32>, %prod_weight: tensor<1x5x6xf32>,
+    %cons_in: tensor<4x6xf32>, %prod_init: tensor<1x6x6xf32>,
+    %for_iv_init: tensor<4x6xf32>, %cons_init: tensor<4x2xf32>
 ) -> tensor<4x6xf32> {
   %c0 = arith.constant 0 : index
   %c2 = arith.constant 2 : index
   %c6 = arith.constant 6 : index
-  %0 = linalg.generic
+  %mmul_prod = linalg.generic
     {indexing_maps = [#map0, #map1, #map2], iterator_types = ["parallel", "parallel", "parallel", "reduction"]}
-    ins(%arg0, %arg1 : tensor<1x6x5xf32>, tensor<1x5x6xf32>) outs(%arg3 : tensor<1x6x6xf32>) {
+    ins(%prod_in, %prod_weight : tensor<1x6x5xf32>, tensor<1x5x6xf32>) outs(%prod_init : tensor<1x6x6xf32>) {
   ^bb0(%in: f32, %in_1: f32, %out: f32):
     %10 = arith.mulf %in, %in_1 : f32
     %11 = arith.addf %out, %10 : f32
     linalg.yield %11 : f32
   } -> tensor<1x6x6xf32>
-  %1 = scf.for %arg7 = %c0 to %c6 step %c2 iter_args(%arg6 = %arg4) -> (tensor<4x6xf32>) {
-    %2 = tensor.extract_slice %0[0, 0, %arg7] [1, 6, 2] [1, 1, 1] : tensor<1x6x6xf32> to tensor<6x2xf32>
-    %3 = linalg.generic
+  %for = scf.for %arg7 = %c0 to %c6 step %c2 iter_args(%arg6 = %for_iv_init) -> (tensor<4x6xf32>) {
+
+    // Extract slice with rank-reduced result type. When fused in the loop
+    // with sliced operands, the producer linalg must have its now sliced
+    // result be rank-reduced as well to match consumer's use type.
+    %prod_slice = tensor.extract_slice %mmul_prod[0, 0, %arg7] [1, 6, 2] [1, 1, 1] : tensor<1x6x6xf32> to tensor<6x2xf32>
+    %mmul_cons = linalg.generic
      {indexing_maps = [#map3, #map4, #map5], iterator_types = ["parallel", "parallel", "reduction"]}
-     ins(%arg2, %2 : tensor<4x6xf32>, tensor<6x2xf32>) outs(%arg5 : tensor<4x2xf32>) {
+     ins(%cons_in, %prod_slice : tensor<4x6xf32>, tensor<6x2xf32>) outs(%cons_init : tensor<4x2xf32>) {
     ^bb0(%in: f32, %in_1: f32, %out: f32):
       %20 = arith.mulf %in, %in_1 : f32
       %21 = arith.addf %out, %20 : f32
       linalg.yield %21 : f32
     } -> tensor<4x2xf32>
-    %4 = tensor.insert_slice %3 into %arg6[0, %arg7] [4, 2] [1, 1]  : tensor<4x2xf32> into tensor<4x6xf32>
+    %4 = tensor.insert_slice %mmul_cons into %arg6[0, %arg7] [4, 2] [1, 1]  : tensor<4x2xf32> into tensor<4x6xf32>
     scf.yield %4 : tensor<4x6xf32>
   }
-  return %1 : tensor<4x6xf32>
+  return %for : tensor<4x6xf32>
 }
 
 //       CHECK: func @rank_reduced_extract_slice(
-//  CHECK-SAME: %[[ARG0:[0-9a-z]*]]: tensor<1x6x5xf32>
-//  CHECK-SAME: %[[ARG1:[0-9a-z]*]]: tensor<1x5x6xf32>
-//  CHECK-SAME: %[[ARG2:[0-9a-z]*]]: tensor<4x6xf32>
-//  CHECK-SAME: %[[ARG3:[0-9a-z]*]]: tensor<1x6x6xf32>
-//  CHECK-SAME: %[[ARG4:[0-9a-z]*]]: tensor<4x6xf32>
-//  CHECK-SAME: %[[ARG5:[0-9a-z]*]]: tensor<4x2xf32>
+//  CHECK-SAME: %[[PROD_IN:[0-9a-z]*]]: tensor<1x6x5xf32>
+//  CHECK-SAME: %[[PROD_WEIGHT:[0-9a-z]*]]: tensor<1x5x6xf32>
+//  CHECK-SAME: %[[CONS_IN:[0-9a-z]*]]: tensor<4x6xf32>
+//  CHECK-SAME: %[[PROD_INIT:[0-9a-z]*]]: tensor<1x6x6xf32>
+//  CHECK-SAME: %[[FOR_IV_INIT:[0-9a-z]*]]: tensor<4x6xf32>
+//  CHECK-SAME: %[[CONS_INIT:[0-9a-z]*]]: tensor<4x2xf32>
 
 //   CHECK-DAG: %[[C0:.*]] = arith.constant 0 : index
 //   CHECK-DAG: %[[C2:.*]] = arith.constant 2 : index
@@ -372,22 +377,22 @@ func.func @rank_reduced_extract_slice(
 
 //  For loop right after tensor alloc & fill, no linalg.generic.
 //   CHECK-NOT: linalg.generic
-//  CHECK-NEXT: %[[FOR:.*]] = scf.for %[[I:[0-9a-z]*]] = %[[C0]] to %[[C6]] step %[[C2]] iter_args(%[[ARG_ITER:.*]] = %[[ARG4]])
+//  CHECK-NEXT: %[[FOR:.*]] = scf.for %[[I:[0-9a-z]*]] = %[[C0]] to %[[C6]] step %[[C2]] iter_args(%[[ARG_ITER:.*]] = %[[FOR_IV_INIT]])
 
 //  Producer linalg.generic now inside the loop, with tiled args sliced before
 //  it.
-//   CHECK-DAG:   %[[ARG1_SLICE:.*]] = tensor.extract_slice %[[ARG1]][0, 0, %[[I]]] [1, 5, 2] [1, 1, 1]  : tensor<1x5x6xf32> to tensor<1x5x2xf32>
-//   CHECK-DAG:   %[[PROD_SLICE:.*]] = tensor.extract_slice %[[ARG3]][0, 0, %[[I]]] [1, 6, 2] [1, 1, 1]  : tensor<1x6x6xf32> to tensor<1x6x2xf32>
+//   CHECK-DAG:   %[[PROD_WEIGHT_SLICE:.*]] = tensor.extract_slice %[[PROD_WEIGHT]][0, 0, %[[I]]] [1, 5, 2] [1, 1, 1]  : tensor<1x5x6xf32> to tensor<1x5x2xf32>
+//   CHECK-DAG:   %[[PROD_INIT_SLICE:.*]] = tensor.extract_slice %[[PROD_INIT]][0, 0, %[[I]]] [1, 6, 2] [1, 1, 1]  : tensor<1x6x6xf32> to tensor<1x6x2xf32>
 //       CHECK:    %[[MMUL_PROD:.*]] = linalg.generic
-//  CHECK-SAME:        ins(%[[ARG0]], %[[ARG1_SLICE]] : tensor<1x6x5xf32>, tensor<1x5x2xf32>)
-//  CHECK-SAME:        outs(%[[PROD_SLICE]] : tensor<1x6x2xf32>)
+//  CHECK-SAME:        ins(%[[PROD_IN]], %[[PROD_WEIGHT_SLICE]] : tensor<1x6x5xf32>, tensor<1x5x2xf32>)
+//  CHECK-SAME:        outs(%[[PROD_INIT_SLICE]] : tensor<1x6x2xf32>)
 //
 //  Consumer uses a rank-reduced version of producer result so a collapse_shape
 //  is generated.
 //       CHECK:    %[[PROD_COLLAPSE:.*]] = tensor.collapse_shape %[[MMUL_PROD]] {{\[\[0, 1\], \[2\]\]}} : tensor<1x6x2xf32> into tensor<6x2xf32>
 //       CHECK:    %[[MMUL_CONS:.*]] = linalg.generic
-//  CHECK-SAME:        ins(%[[ARG2]], %[[PROD_COLLAPSE]] : tensor<4x6xf32>, tensor<6x2xf32>)
-//  CHECK-SAME:        outs(%[[ARG5]] : tensor<4x2xf32>)
+//  CHECK-SAME:        ins(%[[CONS_IN]], %[[PROD_COLLAPSE]] : tensor<4x6xf32>, tensor<6x2xf32>)
+//  CHECK-SAME:        outs(%[[CONS_INIT]] : tensor<4x2xf32>)
 //       CHECK:   %[[CONS_SLICE:.*]] = tensor.insert_slice %[[MMUL_CONS]] into %[[ARG_ITER]][0, %[[I]]] [4, 2] [1, 1] : tensor<4x2xf32> into tensor<4x6xf32>
 //       CHECK:   scf.yield %[[CONS_SLICE]] : tensor<4x6xf32>
 //       CHECK: return %[[FOR]] : tensor<4x6xf32>



More information about the Mlir-commits mailing list