[mlir][vector]add extractInsertFoldConstantOp fold function and apply it to extractOp and insertOp.

mlir/lib/Dialect/Vector/IR/VectorOps.cpp

@@ -1989,6 +1989,45 @@ static Value foldScalarExtractFromFromElements(ExtractOp extractOp) {
   return fromElementsOp.getElements()[flatIndex];
 }
 
+// If the dynamic indices of `extractOp` or `insertOp` are result of
+// `constantOp`, then fold it.
+template <typename OpType, typename AdaptorType>
+static Value extractInsertFoldConstantOp(OpType op, AdaptorType adaptor,
+                                         SmallVectorImpl<Value> &operands) {
+  std::vector<int64_t> staticPosition = op.getStaticPosition().vec();
+  OperandRange dynamicPosition = op.getDynamicPosition();
+  ArrayRef<Attribute> dynamicPositionAttr = adaptor.getDynamicPosition();
+
+  // If the dynamic operands is empty, it is returned directly.
+  if (!dynamicPosition.size())
+    return {};
+
+  // `index` is used to iterate over the `dynamicPosition`.
+  unsigned index = 0;
+
+  // `opChange` is a flag. If it is true, it means to update `op` in place.
+  bool opChange = false;
+  for (unsigned i = 0, e = staticPosition.size(); i < e; ++i) {
+    if (!ShapedType::isDynamic(staticPosition[i]))
+      continue;
+    Attribute positionAttr = dynamicPositionAttr[index];
+    Value position = dynamicPosition[index++];
+    if (auto attr = mlir::dyn_cast_if_present<IntegerAttr>(positionAttr)) {
+      staticPosition[i] = attr.getInt();
+      opChange = true;
+      continue;
+    }
+    operands.push_back(position);
+  }
+
+  if (opChange) {
+    op.setStaticPosition(staticPosition);
+    op.getOperation()->setOperands(operands);
+    return op.getResult();
+  }
+  return {};
+}
+
 /// Fold an insert or extract operation into an poison value when a poison index
 /// is found at any dimension of the static position.
 static Attribute foldPoisonIndexInsertExtractOp(MLIRContext *context,
@@ -2035,6 +2074,9 @@ OpFoldResult ExtractOp::fold(FoldAdaptor adaptor) {
     return val;
   if (auto val = foldScalarExtractFromFromElements(*this))
     return val;
+  SmallVector<Value> operands = {getVector()};
+  if (auto val = extractInsertFoldConstantOp(*this, adaptor, operands))
+    return val;
   return OpFoldResult();
 }
 
@@ -3094,6 +3136,9 @@ OpFoldResult vector::InsertOp::fold(FoldAdaptor adaptor) {
   // (type mismatch).
   if (getNumIndices() == 0 && getSourceType() == getType())
     return getSource();
+  SmallVector<Value> operands = {getSource(), getDest()};
+  if (auto val = extractInsertFoldConstantOp(*this, adaptor, operands))
+    return val;
   if (auto res = foldPoisonIndexInsertExtractOp(
           getContext(), adaptor.getStaticPosition(), kPoisonIndex))
     return res;

mlir/test/Conversion/VectorToLLVM/vector-to-llvm-interface.mlir

@@ -530,6 +530,25 @@ func.func @extract_scalar_from_vec_0d_index(%arg0: vector<index>) -> index {
 
 // -----
 
+func.func @extract_scalar_from_vec_2d_f32_dynamic_idxs_compile_time_const(%arg : vector<32x1xi32>) -> i32 {
+  %0 = arith.constant 0 : index
+  %1 = vector.extract %arg[%0, %0] : i32 from vector<32x1xi32>
+  return %1 : i32
+}
+
+// At compile time, since the indices of extractOp are constants,
+// they will be collapsed and folded away; therefore, the lowering works.
+
+// CHECK-LABEL: @extract_scalar_from_vec_2d_f32_dynamic_idxs_compile_time_const
+//  CHECK-SAME:   %[[ARG:.*]]: vector<32x1xi32>) -> i32 {
+//       CHECK:   %[[CAST:.*]] = builtin.unrealized_conversion_cast %[[ARG]] : vector<32x1xi32> to !llvm.array<32 x vector<1xi32>>
+//       CHECK:   %[[VEC_0:.*]] = llvm.extractvalue %[[CAST]][0] : !llvm.array<32 x vector<1xi32>>
+//       CHECK:   %[[C0:.*]] = llvm.mlir.constant(0 : i64) : i64
+//       CHECK:   %[[RES:.*]] = llvm.extractelement %[[VEC_0]]{{\[}}%[[C0]] : i64] : vector<1xi32>
+//       CHECK:   return %[[RES]] : i32
+
+// -----
+
 //===----------------------------------------------------------------------===//
 // vector.insertelement
 //===----------------------------------------------------------------------===//
@@ -781,6 +800,29 @@ func.func @insert_scalar_into_vec_2d_f32_dynamic_idx_scalable(%arg0: vector<1x[1
 
 // -----
 
+func.func @insert_scalar_from_vec_2d_f32_dynamic_idxs_compile_time_const(%arg : vector<4x1xi32>) -> vector<4x1xi32> {
+  %0 = arith.constant 0 : index
+  %1 = arith.constant 1 : i32
+  %res = vector.insert %1, %arg[%0, %0] : i32 into vector<4x1xi32>
+  return %res : vector<4x1xi32>
+}
+
+// At compile time, since the indices of insertOp are constants,
+// they will be collapsed and folded away; therefore, the lowering works.
+
+// CHECK-LABEL: @insert_scalar_from_vec_2d_f32_dynamic_idxs_compile_time_const
+//  CHECK-SAME:   %[[ARG:.*]]: vector<4x1xi32>) -> vector<4x1xi32> {
+//       CHECK:   %[[CAST:.*]] = builtin.unrealized_conversion_cast %[[ARG]] : vector<4x1xi32> to !llvm.array<4 x vector<1xi32>>
+//       CHECK:   %[[C1:.*]] = arith.constant 1 : i32
+//       CHECK:   %[[VEC_0:.*]] = llvm.extractvalue %[[CAST]][0] : !llvm.array<4 x vector<1xi32>>
+//       CHECK:   %[[C0:.*]] = llvm.mlir.constant(0 : i64) : i64
+//       CHECK:   %[[VEC_1:.*]] = llvm.insertelement %[[C1]], %[[VEC_0]]{{\[}}%[[C0]] : i64] : vector<1xi32>
+//       CHECK:   %[[VEC_2:.*]] = llvm.insertvalue %[[VEC_1]], %[[CAST]][0] : !llvm.array<4 x vector<1xi32>>
+//       CHECK:   %[[RES:.*]] = builtin.unrealized_conversion_cast %[[VEC_2]] : !llvm.array<4 x vector<1xi32>> to vector<4x1xi32>
+//       CHECK:   return %[[RES]] : vector<4x1xi32>
+
+// -----
+
 //===----------------------------------------------------------------------===//
 // vector.type_cast
 //

mlir/test/Dialect/Vector/canonicalize.mlir

@@ -3171,3 +3171,29 @@ func.func @contiguous_scatter_step(%base: memref<?xf32>,
     memref<?xf32>, vector<16xindex>, vector<16xi1>, vector<16xf32>
   return
 }
+
+// -----
+
+// CHECK-LABEL: @fold_extract_constant_indices
+//   CHECK-SAME:   %[[ARG:.*]]: vector<32x1xi32>) -> i32 {
+//        CHECK:   %[[RES:.*]] = vector.extract %[[ARG]][0, 0] : i32 from vector<32x1xi32>
+//        CHECK:   return %[[RES]] : i32
+func.func @fold_extract_constant_indices(%arg : vector<32x1xi32>) -> i32 {
+  %0 = arith.constant 0 : index
+  %1 = vector.extract %arg[%0, %0] : i32 from vector<32x1xi32>
+  return %1 : i32
+}
+
+// -----
+
+// CHECK-LABEL: @fold_insert_constant_indices
+//  CHECK-SAME:   %[[ARG:.*]]: vector<4x1xi32>) -> vector<4x1xi32> {
+//       CHECK:   %[[VAL:.*]] = arith.constant 1 : i32
+//       CHECK:   %[[RES:.*]] = vector.insert %[[VAL]], %[[ARG]] [0, 0] : i32 into vector<4x1xi32>
+//       CHECK:   return %[[RES]] : vector<4x1xi32>
+func.func @fold_insert_constant_indices(%arg : vector<4x1xi32>) -> vector<4x1xi32> {
+  %0 = arith.constant 0 : index
+  %1 = arith.constant 1 : i32
+  %res = vector.insert %1, %arg[%0, %0] : i32 into vector<4x1xi32>
+  return %res : vector<4x1xi32>
+}

mlir/test/Dialect/Vector/vector-warp-distribute.mlir

@@ -778,12 +778,11 @@ func.func @warp_constant(%laneid: index) -> (vector<1xf32>) {
 
 // CHECK-PROP-LABEL: func.func @vector_extract_1d(
 //   CHECK-PROP-DAG:   %[[C5_I32:.*]] = arith.constant 5 : i32
-//   CHECK-PROP-DAG:   %[[C1:.*]] = arith.constant 1 : index
 //       CHECK-PROP:   %[[R:.*]] = gpu.warp_execute_on_lane_0(%{{.*}})[32] -> (vector<2xf32>) {
 //       CHECK-PROP:     %[[V:.*]] = "some_def"() : () -> vector<64xf32>
 //       CHECK-PROP:     gpu.yield %[[V]] : vector<64xf32>
 //       CHECK-PROP:   }
-//       CHECK-PROP:   %[[E:.*]] = vector.extract %[[R]][%[[C1]]] : f32 from vector<2xf32>
+//       CHECK-PROP:   %[[E:.*]] = vector.extract %[[R]][1] : f32 from vector<2xf32>
 //       CHECK-PROP:   %[[SHUFFLED:.*]], %{{.*}} = gpu.shuffle  idx %[[E]], %[[C5_I32]]
 //       CHECK-PROP:   return %[[SHUFFLED]] : f32
 func.func @vector_extract_1d(%laneid: index) -> (f32) {