[flang][hlfir] Add hlfir.count intrinsic

Adds a new HLFIR operation for the COUNT intrinsic according to
the design set out in flang/docs/HighLevel.md. This patch includes all
the necessary changes to create a new HLFIR operation and lower it into
the fir runtime call.

Author was @jacob-crawley. Minor adjustments by @tblah

Differential Revision: https://reviews.llvm.org/D152521

8 files changed, 459 insertions, 3 deletions

diff --git a/flang/include/flang/Optimizer/HLFIR/HLFIROps.td b/flang/include/flang/Optimizer/HLFIR/HLFIROps.td
index 2dd85a2..beb50a4 100644
--- a/flang/include/flang/Optimizer/HLFIR/HLFIROps.td
+++ b/flang/include/flang/Optimizer/HLFIR/HLFIROps.td
@@ -359,6 +359,27 @@ def hlfir_AnyOp : hlfir_Op<"any", []> {
   let hasVerifier = 1;
 }
 
+def hlfir_CountOp : hlfir_Op<"count", [AttrSizedOperandSegments]> {
+  let summary = "COUNT transformational intrinsic";
+  let description = [{
+    Takes a logical and counts the number of true values.
+  }];
+
+  let arguments = (ins
+    AnyFortranLogicalArrayObject:$mask,
+    Optional<AnyIntegerType>:$dim,
+    Optional<AnyIntegerType>:$kind
+  );
+
+  let results = (outs AnyFortranValue);
+
+  let assemblyFormat = [{
+    $mask  (`dim` $dim^)? (`kind` $kind^)?  attr-dict `:` functional-type(operands, results)
+  }];
+
+  let hasVerifier = 1;
+}
+
 
 def hlfir_ProductOp : hlfir_Op<"product", [AttrSizedOperandSegments,
     DeclareOpInterfaceMethods<ArithFastMathInterface>]> {
@@ -430,7 +451,7 @@ def hlfir_SumOp : hlfir_Op<"sum", [AttrSizedOperandSegments,
   let hasVerifier = 1;
 }
 
-def hlifr_DotProductOp : hlfir_Op<"dot_product",
+def hlfir_DotProductOp : hlfir_Op<"dot_product",
     [DeclareOpInterfaceMethods<ArithFastMathInterface>]> {
   let summary = "DOT_PRODUCT transformational intrinsic";
   let description = [{
diff --git a/flang/lib/Lower/ConvertCall.cpp b/flang/lib/Lower/ConvertCall.cpp
index 604291c..4f4505c 100644
--- a/flang/lib/Lower/ConvertCall.cpp
+++ b/flang/lib/Lower/ConvertCall.cpp
@@ -1497,6 +1497,18 @@ genHLFIRIntrinsicRefCore(PreparedActualArguments &loweredActuals,
 
     return {hlfir::EntityWithAttributes{dotProductOp.getResult()}};
   }
+  if (intrinsicName == "count") {
+    llvm::SmallVector<mlir::Value> operands = getOperandVector(loweredActuals);
+    mlir::Value array = operands[0];
+    mlir::Value dim = operands[1];
+    if (dim)
+      dim = hlfir::loadTrivialScalar(loc, builder, hlfir::Entity{dim});
+    mlir::Value kind = operands[2];
+    mlir::Type resultTy = computeResultType(array, *callContext.resultType);
+    hlfir::CountOp countOp =
+        builder.create<hlfir::CountOp>(loc, resultTy, array, dim, kind);
+    return {hlfir::EntityWithAttributes{countOp.getResult()}};
+  }
 
   // TODO add hlfir operations for other transformational intrinsics here
 
diff --git a/flang/lib/Optimizer/HLFIR/IR/HLFIROps.cpp b/flang/lib/Optimizer/HLFIR/IR/HLFIROps.cpp
index 7a00bfa..21a44c0 100644
--- a/flang/lib/Optimizer/HLFIR/IR/HLFIROps.cpp
+++ b/flang/lib/Optimizer/HLFIR/IR/HLFIROps.cpp
@@ -510,6 +510,43 @@ mlir::LogicalResult hlfir::AnyOp::verify() {
 }
 
 //===----------------------------------------------------------------------===//
+// CountOp
+//===----------------------------------------------------------------------===//
+
+mlir::LogicalResult hlfir::CountOp::verify() {
+  mlir::Operation *op = getOperation();
+
+  auto results = op->getResultTypes();
+  assert(results.size() == 1);
+  mlir::Value mask = getMask();
+  mlir::Value dim = getDim();
+
+  fir::SequenceType maskTy =
+      hlfir::getFortranElementOrSequenceType(mask.getType())
+          .cast<fir::SequenceType>();
+  llvm::ArrayRef<int64_t> maskShape = maskTy.getShape();
+
+  mlir::Type resultType = results[0];
+  if (auto resultExpr = mlir::dyn_cast_or_null<hlfir::ExprType>(resultType)) {
+    if (maskShape.size() > 1 && dim != nullptr) {
+      if (!resultExpr.isArray())
+        return emitOpError("result must be an array");
+
+      llvm::ArrayRef<int64_t> resultShape = resultExpr.getShape();
+      // Result has rank n-1
+      if (resultShape.size() != (maskShape.size() - 1))
+        return emitOpError("result rank must be one less than MASK");
+    } else {
+      return emitOpError("result must be of numerical scalar type");
+    }
+  } else if (!hlfir::isFortranScalarNumericalType(resultType)) {
+    return emitOpError("result must be of numerical scalar type");
+  }
+
+  return mlir::success();
+}
+
+//===----------------------------------------------------------------------===//
 // ConcatOp
 //===----------------------------------------------------------------------===//
 
diff --git a/flang/lib/Optimizer/HLFIR/Transforms/LowerHLFIRIntrinsics.cpp b/flang/lib/Optimizer/HLFIR/Transforms/LowerHLFIRIntrinsics.cpp
index b88b30e..d3c604e 100644
--- a/flang/lib/Optimizer/HLFIR/Transforms/LowerHLFIRIntrinsics.cpp
+++ b/flang/lib/Optimizer/HLFIR/Transforms/LowerHLFIRIntrinsics.cpp
@@ -262,6 +262,39 @@ using AnyOpConversion = HlfirReductionIntrinsicConversion<hlfir::AnyOp>;
 
 using AllOpConversion = HlfirReductionIntrinsicConversion<hlfir::AllOp>;
 
+struct CountOpConversion : public HlfirIntrinsicConversion<hlfir::CountOp> {
+  using HlfirIntrinsicConversion<hlfir::CountOp>::HlfirIntrinsicConversion;
+
+  mlir::LogicalResult
+  matchAndRewrite(hlfir::CountOp count,
+                  mlir::PatternRewriter &rewriter) const override {
+    fir::KindMapping kindMapping{rewriter.getContext()};
+    fir::FirOpBuilder builder{rewriter, kindMapping};
+    const mlir::Location &loc = count->getLoc();
+
+    mlir::Type i32 = builder.getI32Type();
+    mlir::Type logicalType = fir::LogicalType::get(
+        builder.getContext(), builder.getKindMap().defaultLogicalKind());
+
+    llvm::SmallVector<IntrinsicArgument, 3> inArgs;
+    inArgs.push_back({count.getMask(), logicalType});
+    inArgs.push_back({count.getDim(), i32});
+    inArgs.push_back({count.getKind(), i32});
+
+    auto *argLowering = fir::getIntrinsicArgumentLowering("count");
+    llvm::SmallVector<fir::ExtendedValue, 3> args =
+        lowerArguments(count, inArgs, rewriter, argLowering);
+
+    mlir::Type scalarResultType = hlfir::getFortranElementType(count.getType());
+
+    auto [resultExv, mustBeFreed] =
+        fir::genIntrinsicCall(builder, loc, "count", scalarResultType, args);
+
+    processReturnValue(count, resultExv, mustBeFreed, builder, rewriter);
+    return mlir::success();
+  }
+};
+
 struct MatmulOpConversion : public HlfirIntrinsicConversion<hlfir::MatmulOp> {
   using HlfirIntrinsicConversion<hlfir::MatmulOp>::HlfirIntrinsicConversion;
 
@@ -405,14 +438,14 @@ public:
     patterns.insert<MatmulOpConversion, MatmulTransposeOpConversion,
                     AllOpConversion, AnyOpConversion, SumOpConversion,
                     ProductOpConversion, TransposeOpConversion,
-                    DotProductOpConversion>(context);
+                    CountOpConversion, DotProductOpConversion>(context);
     mlir::ConversionTarget target(*context);
     target.addLegalDialect<mlir::BuiltinDialect, mlir::arith::ArithDialect,
                            mlir::func::FuncDialect, fir::FIROpsDialect,
                            hlfir::hlfirDialect>();
     target.addIllegalOp<hlfir::MatmulOp, hlfir::MatmulTransposeOp, hlfir::SumOp,
                         hlfir::ProductOp, hlfir::TransposeOp, hlfir::AnyOp,
-                        hlfir::AllOp, hlfir::DotProductOp>();
+                        hlfir::AllOp, hlfir::DotProductOp, hlfir::CountOp>();
     target.markUnknownOpDynamicallyLegal(
         [](mlir::Operation *) { return true; });
     if (mlir::failed(
diff --git a/flang/test/HLFIR/count-lowering.fir b/flang/test/HLFIR/count-lowering.fir
new file mode 100644
index 0000000..0d9cc34
--- /dev/null
+++ b/flang/test/HLFIR/count-lowering.fir
@@ -0,0 +1,164 @@
+// Test hlfir.count operation lowering to fir runtime call
+// RUN: fir-opt %s -lower-hlfir-intrinsics | FileCheck %s
+
+func.func @_QPcount1(%arg0: !fir.box<!fir.array<?x!fir.logical<4>>> {fir.bindc_name = "a"}, %arg1: !fir.ref<i32> {fir.bindc_name = "s"}) {
+  %0:2 = hlfir.declare %arg0 {uniq_name = "_QFcount1Ea"} : (!fir.box<!fir.array<?x!fir.logical<4>>>) -> (!fir.box<!fir.array<?x!fir.logical<4>>>, !fir.box<!fir.array<?x!fir.logical<4>>>)
+  %1:2 = hlfir.declare %arg1 {uniq_name = "_QFcount1Es"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+  %2 = hlfir.count %0#0 {fastmath = #arith.fastmath<contract>} : (!fir.box<!fir.array<?x!fir.logical<4>>>) -> i32
+  hlfir.assign %2 to %1#0 : i32, !fir.ref<i32>
+  return
+}
+// CHECK-LABEL: func.func @_QPcount1(
+// CHECK:           %[[ARG0:.*]]: !fir.box<!fir.array<?x!fir.logical<4>>>
+// CHECK:           %[[ARG1:.*]]: !fir.ref<i32>
+// CHECK-DAG:     %[[MASK:.*]]:2 = hlfir.declare %[[ARG0]]
+// CHECK-DAG:     %[[RES:.*]]:2 = hlfir.declare %[[ARG1]]
+// CHECK-DAG:     %[[MASK_ARG:.*]] = fir.convert %[[MASK]]#1 : (!fir.box<!fir.array<?x!fir.logical<4>>>) -> !fir.box<none>
+// CHECK:         %[[RET_ARG:.*]] = fir.call @_FortranACount(%[[MASK_ARG]], %[[LOC_STR:.*]], %[[LOC_N:.*]], %[[C1:.*]]) : (!fir.box<none>, !fir.ref<i8>, i32, i32) -> i64
+// CHECK-NEXT:    %[[RET:.*]] = fir.convert %[[RET_ARG]] : (i64) -> i32
+// CHECK-NEXT:    hlfir.assign %[[RET]] to %[[RES]]#0 : i32, !fir.ref<i32>
+// CHECK-NEXT:    return
+// CHECK-NEXT:  }
+
+func.func @_QPcount2(%arg0: !fir.box<!fir.array<?x?x!fir.logical<4>>> {fir.bindc_name = "a"}, %arg1: !fir.box<!fir.array<?xi32>> {fir.bindc_name = "s"}, %arg2: !fir.ref<i32> {fir.bindc_name = "d"}) {
+  %0:2 = hlfir.declare %arg0 {uniq_name = "_QFcount2Ea"} : (!fir.box<!fir.array<?x?x!fir.logical<4>>>) -> (!fir.box<!fir.array<?x?x!fir.logical<4>>>, !fir.box<!fir.array<?x?x!fir.logical<4>>>)
+  %1:2 = hlfir.declare %arg2 {uniq_name = "_QFcount2Ed"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+  %2:2 = hlfir.declare %arg1 {uniq_name = "_QFcount2Es"} : (!fir.box<!fir.array<?xi32>>) -> (!fir.box<!fir.array<?xi32>>, !fir.box<!fir.array<?xi32>>)
+  %3 = fir.load %1#0 : !fir.ref<i32>
+  %4 = hlfir.count %0#0 dim %3 {fastmath = #arith.fastmath<contract>} : (!fir.box<!fir.array<?x?x!fir.logical<4>>>, i32) -> !hlfir.expr<?xi32>
+  hlfir.assign %4 to %2#0 : !hlfir.expr<?xi32>, !fir.box<!fir.array<?xi32>>
+  hlfir.destroy %4 : !hlfir.expr<?xi32>
+  return
+}
+// CHECK-LABEL: func.func @_QPcount2(
+// CHECK:           %[[ARG0:.*]]: !fir.box<!fir.array<?x?x!fir.logical<4>>>
+// CHECK:           %[[ARG1:.*]]: !fir.box<!fir.array<?xi32>
+// CHECK:           %[[ARG2:.*]]: !fir.ref<i32>
+// CHECK-DAG:     %[[MASK:.*]]:2 = hlfir.declare %[[ARG0]]
+// CHECK-DAG:     %[[DIM_VAR:.*]]:2 = hlfir.declare %[[ARG2]]
+// CHECK-DAG:     %[[RES:.*]]:2 = hlfir.declare %[[ARG1]]
+
+// CHECK-DAG:     %[[RET_BOX:.*]] = fir.alloca !fir.box<!fir.heap<!fir.array<?xi32>>>
+// CHECK-DAG:     %[[RET_ADDR:.*]] = fir.zero_bits !fir.heap<!fir.array<?xi32>>
+// CHECK-DAG:     %[[C0:.*]] = arith.constant 0 : index
+// CHECK-DAG:     %[[RET_SHAPE:.*]] = fir.shape %[[C0]] : (index) -> !fir.shape<1>
+// CHECK-DAG:     %[[RET_EMBOX:.*]] = fir.embox %[[RET_ADDR]](%[[RET_SHAPE]])
+// CHECK-DAG:     fir.store %[[RET_EMBOX]] to %[[RET_BOX]]
+
+// CHECK-DAG:     %[[DIM:.*]] = fir.load %[[DIM_VAR]]#0 : !fir.ref<i32>
+// CHECK-DAG:     %[[RET_ARG:.*]] = fir.convert %[[RET_BOX]]
+// CHECK-DAG:     %[[MASK_ARG:.*]] = fir.convert %[[MASK]]#1
+
+// CHECK:         %[[NONE:.*]] = fir.call @_FortranACountDim(%[[RET_ARG]], %[[MASK_ARG]], %[[DIM]], %[[LOC_STR:.*]], %[[LOC_N:.*]]) : (!fir.ref<!fir.box<none>>, !fir.box<none>, i32, i32, !fir.ref<i8>, i32) -> none
+// CHECK:         %[[RET:.*]] = fir.load %[[RET_BOX]]
+// CHECK:         %[[BOX_DIMS:.*]]:3 = fir.box_dims %[[RET]]
+// CHECK-NEXT:    %[[ADDR:.*]] = fir.box_addr %[[RET]]
+// CHECK-NEXT:    %[[SHIFT:.*]] = fir.shape_shift %[[BOX_DIMS]]#0, %[[BOX_DIMS]]#1
+// CHECK-NEXT:    %[[TMP:.*]]:2 = hlfir.declare %[[ADDR]](%[[SHIFT]]) {uniq_name = ".tmp.intrinsic_result"}
+// CHECK:         %[[TRUE:.*]] = arith.constant true
+// CHECK:         %[[EXPR:.*]] = hlfir.as_expr %[[TMP]]#0 move %[[TRUE]] : (!fir.box<!fir.array<?xi32>>, i1) -> !hlfir.expr<?xi32>
+// CHECK:         hlfir.assign %[[EXPR]] to %[[RES]]#0
+// CHECK:         hlfir.destroy %[[EXPR]]
+// CHECK-NEXT:    return
+// CHECK-NEXT:  }
+
+func.func @_QPcount3(%arg0: !fir.ref<!fir.array<2xi32>> {fir.bindc_name = "s"}) {
+  %0 = fir.address_of(@_QFcount3Ea) : !fir.ref<!fir.array<2x2x!fir.logical<4>>>
+  %c2 = arith.constant 2 : index
+  %c2_0 = arith.constant 2 : index
+  %1 = fir.shape %c2, %c2_0 : (index, index) -> !fir.shape<2>
+  %2:2 = hlfir.declare %0(%1) {uniq_name = "_QFcount3Ea"} : (!fir.ref<!fir.array<2x2x!fir.logical<4>>>, !fir.shape<2>) -> (!fir.ref<!fir.array<2x2x!fir.logical<4>>>, !fir.ref<!fir.array<2x2x!fir.logical<4>>>)
+  %c2_1 = arith.constant 2 : index
+  %3 = fir.shape %c2_1 : (index) -> !fir.shape<1>
+  %4:2 = hlfir.declare %arg0(%3) {uniq_name = "_QFcount3Es"} : (!fir.ref<!fir.array<2xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<2xi32>>, !fir.ref<!fir.array<2xi32>>)
+  %c1_i32 = arith.constant 1 : i32
+  %5 = hlfir.count %2#0 dim %c1_i32 {fastmath = #arith.fastmath<contract>} : (!fir.ref<!fir.array<2x2x!fir.logical<4>>>, i32) -> !hlfir.expr<2xi32>
+  hlfir.assign %5 to %4#0 : !hlfir.expr<2xi32>, !fir.ref<!fir.array<2xi32>>
+  hlfir.destroy %5 : !hlfir.expr<2xi32>
+  return
+}
+// CHECK-LABEL:  func.func @_QPcount3(
+// CHECK:           %[[ARG0:.*]]: !fir.ref<!fir.array<2xi32>>
+// CHECK-DAG:     %[[RET_BOX:.*]] = fir.alloca !fir.box<!fir.heap<!fir.array<?xi32>>>
+// CHECK-DAG:     %[[RET_ADDR:.*]] = fir.zero_bits !fir.heap<!fir.array<?xi32>>
+// CHECK-DAG:     %[[C0:.*]] = arith.constant 0 : index
+// CHECK-DAG:     %[[RET_SHAPE:.*]] = fir.shape %[[C0]] : (index) -> !fir.shape<1>
+// CHECK-DAG:     %[[RET_EMBOX:.*]] = fir.embox %[[RET_ADDR]](%[[RET_SHAPE]])
+// CHECK-DAG:     fir.store %[[RET_EMBOX]] to %[[RET_BOX]]
+// CHECK-DAG:     %[[RES:.*]]:2 = hlfir.declare %[[ARG0]](%[[RES_SHAPE:.*]])
+
+// CHECK-DAG:     %[[MASK_ADDR:.*]] = fir.address_of
+// CHECK-DAG:     %[[MASK_VAR:.*]]:2 = hlfir.declare %[[MASK_ADDR]](%[[MASK_SHAPE:.*]])
+// CHECK-DAG:     %[[MASK_BOX:.*]] = fir.embox %[[MASK_VAR]]#1(%[[MASK_SHAPE:.*]])
+
+// CHECK-DAG:     %[[DIM:.*]] = arith.constant 1 : i32
+
+// CHECK-DAG:     %[[RET_ARG:.*]] = fir.convert %[[RET_BOX]]
+// CHECK-DAG:     %[[MASK_ARG:.*]] = fir.convert %[[MASK_BOX]] : (!fir.box<!fir.array<2x2x!fir.logical<4>>>) -> !fir.box<none>
+// CHECK:         %[[NONE:.*]] = fir.call @_FortranACountDim(%[[RET_ARG]], %[[MASK_ARG]], %[[DIM]], %[[LOC_STR:.*]], %[[LOC_N:.*]])
+// CHECK:         %[[RET:.*]] = fir.load %[[RET_BOX]]
+// CHECK:         %[[BOX_DIMS:.*]]:3 = fir.box_dims %[[RET]]
+// CHECK-NEXT:    %[[ADDR:.*]] = fir.box_addr %[[RET]]
+// CHECK-NEXT:    %[[SHIFT:.*]] = fir.shape_shift %[[BOX_DIMS]]#0, %[[BOX_DIMS]]#1
+// CHECK-NEXT:    %[[TMP:.*]]:2 = hlfir.declare %[[ADDR]](%[[SHIFT]]) {uniq_name = ".tmp.intrinsic_result"}
+// CHECK:         %[[TRUE:.*]] = arith.constant true
+// CHECK:         %[[EXPR:.*]] = hlfir.as_expr %[[TMP]]#0 move %[[TRUE]] : (!fir.box<!fir.array<?xi32>>, i1) -> !hlfir.expr<?xi32>
+// CHECK:         hlfir.assign %[[EXPR]] to %[[RES]]
+// CHECK:         hlfir.destroy %[[EXPR]]
+// CHECK-NEXT:    return
+// CHECK-NEXT:  }
+
+func.func @_QPcount4(%arg0: !fir.box<!fir.array<?x?x!fir.logical<4>>> {fir.bindc_name = "a"}, %arg1: !fir.box<!fir.array<?xi32>> {fir.bindc_name = "s"}, %arg2: !fir.ref<i32> {fir.bindc_name = "d"}) {
+  %0:2 = hlfir.declare %arg0 {uniq_name = "_QFcount4Ea"} : (!fir.box<!fir.array<?x?x!fir.logical<4>>>) -> (!fir.box<!fir.array<?x?x!fir.logical<4>>>, !fir.box<!fir.array<?x?x!fir.logical<4>>>)
+  %1:2 = hlfir.declare %arg2 {uniq_name = "_QFcount4Ed"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+  %2:2 = hlfir.declare %arg1 {uniq_name = "_QFcount4Es"} : (!fir.box<!fir.array<?xi32>>) -> (!fir.box<!fir.array<?xi32>>, !fir.box<!fir.array<?xi32>>)
+  %c8_i32 = arith.constant 8 : i32
+  %3 = fir.load %1#0 : !fir.ref<i32>
+  %4 = hlfir.count %0#0 dim %3 kind %c8_i32 {fastmath = #arith.fastmath<contract>} : (!fir.box<!fir.array<?x?x!fir.logical<4>>>, i32, i32) -> !hlfir.expr<?xi64>
+  %5 = hlfir.shape_of %4 : (!hlfir.expr<?xi64>) -> !fir.shape<1>
+  %6 = hlfir.elemental %5 : (!fir.shape<1>) -> !hlfir.expr<?xi32> {
+  ^bb0(%arg3: index):
+    %7 = hlfir.apply %4, %arg3 : (!hlfir.expr<?xi64>, index) -> i64
+    %8 = fir.convert %7 : (i64) -> i32
+    hlfir.yield_element %8 : i32
+  }
+  hlfir.assign %6 to %2#0 : !hlfir.expr<?xi32>, !fir.box<!fir.array<?xi32>>
+  hlfir.destroy %6 : !hlfir.expr<?xi32>
+  hlfir.destroy %4 : !hlfir.expr<?xi64>
+  return
+}
+// CHECK-LABEL: func.func @_QPcount4(
+// CHECK:           %[[ARG0:.*]]: !fir.box<!fir.array<?x?x!fir.logical<4>>>
+// CHECK:           %[[ARG1:.*]]: !fir.box<!fir.array<?xi32>
+// CHECK:           %[[ARG2:.*]]: !fir.ref<i32>
+// CHECK-DAG:     %[[MASK:.*]]:2 = hlfir.declare %[[ARG0]]
+// CHECK-DAG:     %[[DIM_VAR:.*]]:2 = hlfir.declare %[[ARG2]]
+// CHECK-DAG:     %[[RES:.*]]:2 = hlfir.declare %[[ARG1]]
+
+// CHECK-DAG:     %[[RET_BOX:.*]] = fir.alloca !fir.box<!fir.heap<!fir.array<?xi64>>>
+// CHECK-DAG:     %[[RET_ADDR:.*]] = fir.zero_bits !fir.heap<!fir.array<?xi64>>
+// CHECK-DAG:     %[[C0:.*]] = arith.constant 0 : index
+// CHECK-DAG:     %[[RET_SHAPE:.*]] = fir.shape %[[C0]] : (index) -> !fir.shape<1>
+// CHECK-DAG:     %[[RET_EMBOX:.*]] = fir.embox %[[RET_ADDR]](%[[RET_SHAPE]])
+// CHECK-DAG:     fir.store %[[RET_EMBOX]] to %[[RET_BOX]]
+
+// CHECK-DAG:     %[[DIM:.*]] = fir.load %[[DIM_VAR]]#0 : !fir.ref<i32>
+// CHECK-DAG:     %[[KIND:.*]] = arith.constant 8 : i32
+// CHECK-DAG:     %[[RET_ARG:.*]] = fir.convert %[[RET_BOX]]
+// CHECK-DAG:     %[[MASK_ARG:.*]] = fir.convert %[[MASK]]#1
+
+// CHECK:         %[[NONE:.*]] = fir.call @_FortranACountDim(%[[RET_ARG]], %[[MASK_ARG]], %[[DIM]], %[[KIND]], %[[LOC_STR:.*]], %[[LOC_N:.*]]) : (!fir.ref<!fir.box<none>>, !fir.box<none>, i32, i32, !fir.ref<i8>, i32) -> none
+// CHECK:         %[[RET:.*]] = fir.load %[[RET_BOX]]
+// CHECK:         %[[BOX_DIMS:.*]]:3 = fir.box_dims %[[RET]]
+// CHECK-NEXT:    %[[ADDR:.*]] = fir.box_addr %[[RET]]
+// CHECK-NEXT:    %[[SHIFT:.*]] = fir.shape_shift %[[BOX_DIMS]]#0, %[[BOX_DIMS]]#1
+// CHECK-NEXT:    %[[TMP:.*]]:2 = hlfir.declare %[[ADDR]](%[[SHIFT]]) {uniq_name = ".tmp.intrinsic_result"}
+// CHECK:         %[[TRUE:.*]] = arith.constant true
+// CHECK:         %[[EXPR:.*]] = hlfir.as_expr %[[TMP]]#0 move %[[TRUE]] : (!fir.box<!fir.array<?xi64>>, i1) -> !hlfir.expr<?xi64>
+// CHECK-NEXT:    %[[OUT_SHAPE:.*]] = hlfir.shape_of %[[EXPR]]
+// CHECK-NEXT:    %[[OUT:.*]] = hlfir.elemental %[[OUT_SHAPE]] : (!fir.shape<1>) -> !hlfir.expr<?xi32>
+// CHECK-DAG:       hlfir.assign %[[OUT]] to %[[RES]]#0
+// CHECK-NEXT:      hlfir.destroy %[[OUT]] : !hlfir.expr<?xi32>
+// CHECK-NEXT:      hlfir.destroy %[[EXPR]] : !hlfir.expr<?xi64>
+// CHECK-NEXT:    return
+// CHECK-NEXT:  }
diff --git a/flang/test/HLFIR/count.fir b/flang/test/HLFIR/count.fir
new file mode 100644
index 0000000..c25f9b9
--- /dev/null
+++ b/flang/test/HLFIR/count.fir
@@ -0,0 +1,83 @@
+// Test hlfir.count operation parse, verify (no errors), and unparse
+
+// RUN: fir-opt %s | fir-opt | FileCheck %s
+
+// mask is an expression of known shape
+func.func @count0(%arg0: !hlfir.expr<2x!fir.logical<4>>) {
+  %count = hlfir.count %arg0 : (!hlfir.expr<2x!fir.logical<4>>) -> i32
+  return
+}
+// CHECK:      func.func @count0(%[[ARRAY:.*]]: !hlfir.expr<2x!fir.logical<4>>) {
+// CHECK-NEXT:   %[[COUNT:.*]] = hlfir.count %[[ARRAY]] : (!hlfir.expr<2x!fir.logical<4>>) -> i32
+// CHECK-NEXT:   return
+// CHECK-NEXT: }
+
+// mask is an expression of assumed shape
+func.func @count1(%arg0: !hlfir.expr<?x!fir.logical<4>>) {
+  %count = hlfir.count %arg0 : (!hlfir.expr<?x!fir.logical<4>>) -> i32
+  return
+}
+// CHECK:      func.func @count1(%[[ARRAY:.*]]: !hlfir.expr<?x!fir.logical<4>>) {
+// CHECK-NEXT:   %[[COUNT:.*]] = hlfir.count %[[ARRAY]] : (!hlfir.expr<?x!fir.logical<4>>) -> i32
+// CHECK-NEXT:   return
+// CHECK-NEXT: }
+
+// mask is a boxed array
+func.func @count2(%arg0: !fir.box<!fir.array<2x!fir.logical<4>>>) {
+  %count = hlfir.count %arg0 : (!fir.box<!fir.array<2x!fir.logical<4>>>) -> i32
+  return
+}
+// CHECK:      func.func @count2(%[[ARRAY:.*]]: !fir.box<!fir.array<2x!fir.logical<4>>>) {
+// CHECK-NEXT:   %[[COUNT:.*]] = hlfir.count %[[ARRAY]] : (!fir.box<!fir.array<2x!fir.logical<4>>>) -> i32
+// CHECK-NEXT:   return
+// CHECK-NEXT: }
+
+// mask is an assumed shape boxed array
+func.func @count3(%arg0: !fir.box<!fir.array<?x!fir.logical<4>>>) {
+  %count = hlfir.count %arg0 : (!fir.box<!fir.array<?x!fir.logical<4>>>) -> i32
+  return
+}
+// CHECK:      func.func @count3(%[[ARRAY:.*]]: !fir.box<!fir.array<?x!fir.logical<4>>>) {
+// CHECK-NEXT:   %[[COUNT:.*]] = hlfir.count %[[ARRAY]] : (!fir.box<!fir.array<?x!fir.logical<4>>>) -> i32
+// CHECK-NEXT:   return
+// CHECK-NEXT: }
+
+// mask is a 2-dimensional array
+func.func @count4(%arg0: !fir.box<!fir.array<?x?x!fir.logical<4>>>){
+  %count = hlfir.count %arg0 : (!fir.box<!fir.array<?x?x!fir.logical<4>>>) -> i32
+  return
+}
+// CHECK:      func.func @count4(%[[ARRAY:.*]]: !fir.box<!fir.array<?x?x!fir.logical<4>>>) {
+// CHECK-NEXT:   %[[COUNT:.*]] = hlfir.count %[[ARRAY]] : (!fir.box<!fir.array<?x?x!fir.logical<4>>>) -> i32
+// CHECK-NEXT:   return
+// CHECK-NEXT: }
+
+// mask and dim argument
+func.func @count5(%arg0: !fir.box<!fir.array<?x!fir.logical<4>>>, %arg1: i32) {
+  %count = hlfir.count %arg0 dim %arg1 : (!fir.box<!fir.array<?x!fir.logical<4>>>, i32) -> i32
+  return
+}
+// CHECK:      func.func @count5(%[[ARRAY:.*]]: !fir.box<!fir.array<?x!fir.logical<4>>>, %[[DIM:.*]]: i32) {
+// CHECK-NEXT:   %[[COUNT:.*]] = hlfir.count %[[ARRAY]] dim %[[DIM]] : (!fir.box<!fir.array<?x!fir.logical<4>>>, i32) -> i32
+// CHECK-NEXT:   return
+// CHECK-NEXT: }
+
+// mask is a 2 dimensional array with dim
+func.func @count6(%arg0: !fir.box<!fir.array<?x?x!fir.logical<4>>>, %arg1: i32) {
+  %count = hlfir.count %arg0 dim %arg1 : (!fir.box<!fir.array<?x?x!fir.logical<4>>>, i32) -> i32
+  return
+}
+// CHECK:      func.func @count6(%[[ARRAY:.*]]: !fir.box<!fir.array<?x?x!fir.logical<4>>>, %[[DIM:.*]]: i32) {
+// CHECK-NEXT:   %[[ANY:.*]] = hlfir.count %[[ARRAY]] dim %[[DIM]] : (!fir.box<!fir.array<?x?x!fir.logical<4>>>, i32) -> i32
+// CHECK-NEXT:   return
+// CHECK-NEXT: }
+
+// mask is of a different logical type
+func.func @count7(%arg0: !fir.box<!fir.array<?x?x!fir.logical<8>>>, %arg1: i32) {
+  %count = hlfir.count %arg0 dim %arg1 : (!fir.box<!fir.array<?x?x!fir.logical<8>>>, i32) -> i32
+  return
+}
+// CHECK:      func.func @count7(%[[ARRAY:.*]]: !fir.box<!fir.array<?x?x!fir.logical<8>>>, %[[DIM:.*]]: i32) {
+// CHECK-NEXT:   %[[ANY:.*]] = hlfir.count %[[ARRAY]] dim %[[DIM]] : (!fir.box<!fir.array<?x?x!fir.logical<8>>>, i32) -> i32
+// CHECK-NEXT:   return
+// CHECK-NEXT: }
\ No newline at end of file
diff --git a/flang/test/HLFIR/invalid.fir b/flang/test/HLFIR/invalid.fir
index 01bccdf..6db1b79 100644
--- a/flang/test/HLFIR/invalid.fir
+++ b/flang/test/HLFIR/invalid.fir
@@ -369,6 +369,30 @@ func.func @bad_all6(%arg0: !hlfir.expr<?x!fir.logical<4>>) {
 }
 
 // -----
+func.func @bad_count1(%arg0: !hlfir.expr<?x?x!fir.logical<4>>, %arg1: i32) {
+  // expected-error@+1 {{'hlfir.count' op result must be an array}}
+  %0 = hlfir.count %arg0 dim %arg1 : (!hlfir.expr<?x?x!fir.logical<4>>, i32) -> !hlfir.expr<i32>
+}
+
+// -----
+func.func @bad_count2(%arg0: !hlfir.expr<?x?x!fir.logical<4>>, %arg1: i32){
+  // expected-error@+1 {{'hlfir.count' op result rank must be one less than MASK}}
+  %0 = hlfir.count %arg0 dim %arg1 : (!hlfir.expr<?x?x!fir.logical<4>>, i32) -> !hlfir.expr<?x?x!fir.logical<4>>
+}
+
+// -----
+func.func @bad_count3(%arg0: !hlfir.expr<?x!fir.logical<4>>, %arg1: i32) {
+  // expected-error@+1 {{'hlfir.count' op result must be of numerical scalar type}}
+  %0 = hlfir.count %arg0 dim %arg1 : (!hlfir.expr<?x!fir.logical<4>>, i32) -> !hlfir.expr<i32>
+}
+
+// -----
+func.func @bad_count4(%arg0: !hlfir.expr<?x!fir.logical<4>>, %arg1: i32) {
+  // expected-error@+1 {{'hlfir.count' op result must be of numerical scalar type}}
+  %0 = hlfir.count %arg0 dim %arg1 : (!hlfir.expr<?x!fir.logical<4>>, i32) -> !fir.logical<4>
+}
+
+// -----
 func.func @bad_product1(%arg0: !hlfir.expr<?xi32>, %arg1: i32, %arg2: !fir.box<!fir.logical<4>>) {
   // expected-error@+1 {{'hlfir.product' op result must have the same element type as ARRAY argument}}
   %0 = hlfir.product %arg0 dim %arg1 mask %arg2 : (!hlfir.expr<?xi32>, i32, !fir.box<!fir.logical<4>>) -> f32
diff --git a/flang/test/Lower/HLFIR/count.f90 b/flang/test/Lower/HLFIR/count.f90
new file mode 100644
index 0000000..25c7484
--- /dev/null
+++ b/flang/test/Lower/HLFIR/count.f90
@@ -0,0 +1,82 @@
+! Test lowering of COUNT intrinsic to HLFIR
+! RUN: bbc -emit-hlfir -o - %s 2>&1 | FileCheck %s
+
+! simple 1 argument COUNT
+subroutine count1(a, s)
+  logical :: a(:)
+  integer :: s
+  s = COUNT(a)
+end subroutine
+! CHECK-LABEL: func.func @_QPcount1(
+! CHECK:           %[[ARG0:.*]]: !fir.box<!fir.array<?x!fir.logical<4>>>
+! CHECK:           %[[ARG1:.*]]: !fir.ref<i32>
+! CHECK-DAG:     %[[MASK:.*]]:2 = hlfir.declare %[[ARG0]]
+! CHECK-DAG:     %[[OUT:.*]]:2 = hlfir.declare %[[ARG1]]
+! CHECK-NEXT:    %[[EXPR:.*]] = hlfir.count %[[MASK]]#0 : (!fir.box<!fir.array<?x!fir.logical<4>>>) -> i32
+! CHECK-NEXT:    hlfir.assign %[[EXPR]] to %[[OUT]]#0 : i32, !fir.ref<i32>
+! CHECK-NEXT:    return
+! CHECK-NEXT:  }
+
+! count with by-ref DIM argument
+subroutine count2(a, s, d)
+  logical :: a(:,:)
+  integer :: s(:), d
+  s = COUNT(a, d)
+end subroutine
+! CHECK-LABEL: func.func @_QPcount2(
+! CHECK:           %[[ARG0:.*]]: !fir.box<!fir.array<?x?x!fir.logical<4>>>
+! CHECK:           %[[ARG1:.*]]: !fir.box<!fir.array<?xi32>>
+! CHECK:           %[[ARG2:.*]]: !fir.ref<i32>
+! CHECK-DAG:     %[[MASK:.*]]:2 = hlfir.declare %[[ARG0]]
+! CHECK-DAG:     %[[DIM_REF:.*]]:2 = hlfir.declare %[[ARG2]]
+! CHECK-DAG:     %[[OUT:.*]]:2 = hlfir.declare %[[ARG1]]
+! CHECK-DAG:     %[[DIM:.*]] = fir.load %[[DIM_REF]]#0 : !fir.ref<i32>
+! CHECK-DAG:     %[[EXPR:.*]] = hlfir.count %[[MASK]]#0 dim %[[DIM]] : (!fir.box<!fir.array<?x?x!fir.logical<4>>>, i32) -> !hlfir.expr<?xi32>
+! CHECK-NEXT:    hlfir.assign %[[EXPR]] to %[[OUT]]#0 : !hlfir.expr<?xi32>, !fir.box<!fir.array<?xi32>>
+! CHECK-NEXT:    hlfir.destroy %[[EXPR]] : !hlfir.expr<?xi32>
+! CHECK-NEXT:    return
+! CHECK-NEXT:  }
+
+! count with DIM argument by-val, mask isn't boxed
+subroutine count3(s)
+  integer :: s(2)
+  logical :: a(2,2) = reshape((/.true.,.false.,.true.,.false./), shape(a))
+  s = COUNT(a, 1)
+end subroutine
+! CHECK-LABEL: func.func @_QPcount3(
+! CHECK:           %[[ARG0:.*]]: !fir.ref<!fir.array<2xi32>>
+! CHECK-DAG:     %[[ADDR:.*]] = fir.address_of{{.*}} : !fir.ref<!fir.array<2x2x!fir.logical<4>>>
+! CHECK-DAG:     %[[MASK_SHAPE:.*]] = fir.shape {{.*}} -> !fir.shape<2>
+! CHECK-DAG:     %[[MASK:.*]]:2 = hlfir.declare %[[ADDR]](%[[MASK_SHAPE]])
+! CHECK-DAG:     %[[OUT_SHAPE:.*]] = fir.shape {{.*}} -> !fir.shape<1>
+! CHECK-DAG:     %[[OUT:.*]]:2 = hlfir.declare %[[ARG0]](%[[OUT_SHAPE]])
+! CHECK-DAG:     %[[C1:.*]] = arith.constant 1 : i32
+! CHECK-DAG:     %[[EXPR:.*]] = hlfir.count %[[MASK]]#0 dim %[[C1]] : (!fir.ref<!fir.array<2x2x!fir.logical<4>>>, i32) -> !hlfir.expr<2xi32>
+! CHECK-DAG:     hlfir.assign %[[EXPR]] to %[[OUT]]#0 : !hlfir.expr<2xi32>, !fir.ref<!fir.array<2xi32>>
+! CHECK-NEXT:    hlfir.destroy %[[EXPR]] : !hlfir.expr<2xi32>
+! CHECK-NEXT:    return
+! CHECK-NEXT:  }
+
+! count with dim and kind arguments
+subroutine count4(a, s, d)
+  logical :: a(:,:)
+  integer :: s(:), d
+  s = COUNT(a, d, 8)
+end subroutine
+! CHECK-LABEL: func.func @_QPcount4(
+! CHECK:           %[[ARG0:.*]]: !fir.box<!fir.array<?x?x!fir.logical<4>>>
+! CHECK:           %[[ARG1:.*]]: !fir.box<!fir.array<?xi32>>
+! CHECK:           %[[ARG2:.*]]: !fir.ref<i32>
+! CHECK-DAG:     %[[MASK:.*]]:2 = hlfir.declare %[[ARG0]]
+! CHECK-DAG:     %[[DIM_REF:.*]]:2 = hlfir.declare %[[ARG2]]
+! CHECK-DAG:     %[[OUT:.*]]:2 = hlfir.declare %[[ARG1]]
+! CHECK-DAG:     %[[C8:.*]] = arith.constant 8 : i32
+! CHECK-DAG:     %[[DIM:.*]] = fir.load %[[DIM_REF]]#0 : !fir.ref<i32>
+! CHECK-DAG:     %[[EXPR:.*]] = hlfir.count %[[MASK]]#0 dim %[[DIM]] kind %[[C8]] : (!fir.box<!fir.array<?x?x!fir.logical<4>>>, i32, i32) -> !hlfir.expr<?xi64>
+! CHECK-DAG:     %[[RES_SHAPE:.*]] = hlfir.shape_of %[[EXPR]]
+! CHECK-DAG:     %[[RES:.*]] = hlfir.elemental %[[RES_SHAPE]] : (!fir.shape<1>) -> !hlfir.expr<?xi32>
+! CHECK-DAG:     hlfir.assign %[[RES]] to %[[OUT]]#0
+! CHECK-NEXT:    hlfir.destroy %[[RES]] : !hlfir.expr<?xi32>
+! CHECK-NEXT:    hlfir.destroy %[[EXPR]] : !hlfir.expr<?xi64>
+! CHECK-NEXT:    return
+! CHECK-NEXT:  }