[flang-commits] [flang] [Flang] Add Maxloc to fir simplify intrinsics pass (PR #75463)

David Green via flang-commits flang-commits at lists.llvm.org
Thu Dec 14 04:09:29 PST 2023


https://github.com/davemgreen updated https://github.com/llvm/llvm-project/pull/75463

>From a0d17ce16c94f78696bd3e1f0fd2a61c8f8b15ce Mon Sep 17 00:00:00 2001
From: David Green <david.green at arm.com>
Date: Thu, 14 Dec 2023 11:00:06 +0000
Subject: [PATCH 1/2] [Flang] Add Maxloc to fir simplify intrinsics pass

This takes the code from D144103 and extends it to maxloc, to allow the
simplifyMinMaxlocReduction method to work with both min and max intrinsics by
switching condition and limit/initial value.
---
 .../Transforms/SimplifyIntrinsics.cpp         |  79 +++---
 flang/test/Transforms/simplifyintrinsics.fir  | 247 ++++++++++++++++++
 2 files changed, 292 insertions(+), 34 deletions(-)

diff --git a/flang/lib/Optimizer/Transforms/SimplifyIntrinsics.cpp b/flang/lib/Optimizer/Transforms/SimplifyIntrinsics.cpp
index 8ecf7fb44f15d0..c19d7bd2d7200f 100644
--- a/flang/lib/Optimizer/Transforms/SimplifyIntrinsics.cpp
+++ b/flang/lib/Optimizer/Transforms/SimplifyIntrinsics.cpp
@@ -99,8 +99,8 @@ class SimplifyIntrinsicsPass
   void simplifyLogicalDim1Reduction(fir::CallOp call,
                                     const fir::KindMapping &kindMap,
                                     GenReductionBodyTy genBodyFunc);
-  void simplifyMinlocReduction(fir::CallOp call,
-                               const fir::KindMapping &kindMap);
+  void simplifyMinMaxlocReduction(fir::CallOp call,
+                                  const fir::KindMapping &kindMap, bool isMax);
   void simplifyReductionBody(fir::CallOp call, const fir::KindMapping &kindMap,
                              GenReductionBodyTy genBodyFunc,
                              fir::FirOpBuilder &builder,
@@ -357,12 +357,11 @@ using MinlocBodyOpGeneratorTy = llvm::function_ref<mlir::Value(
     fir::FirOpBuilder &, mlir::Location, const mlir::Type &, mlir::Value,
     mlir::Value, llvm::SmallVector<mlir::Value, Fortran::common::maxRank> &)>;
 
-static void
-genMinlocReductionLoop(fir::FirOpBuilder &builder, mlir::func::FuncOp &funcOp,
-                       InitValGeneratorTy initVal,
-                       MinlocBodyOpGeneratorTy genBody, unsigned rank,
-                       mlir::Type elementType, mlir::Location loc, bool hasMask,
-                       mlir::Type maskElemType, mlir::Value resultArr) {
+static void genMinMaxlocReductionLoop(
+    fir::FirOpBuilder &builder, mlir::func::FuncOp &funcOp,
+    InitValGeneratorTy initVal, MinlocBodyOpGeneratorTy genBody, unsigned rank,
+    mlir::Type elementType, mlir::Location loc, bool hasMask,
+    mlir::Type maskElemType, mlir::Value resultArr) {
 
   mlir::IndexType idxTy = builder.getIndexType();
 
@@ -751,20 +750,23 @@ static mlir::FunctionType genRuntimeMinlocType(fir::FirOpBuilder &builder,
                                  {boxRefType, boxType, boxType}, {});
 }
 
-static void genRuntimeMinlocBody(fir::FirOpBuilder &builder,
-                                 mlir::func::FuncOp &funcOp, unsigned rank,
-                                 int maskRank, mlir::Type elementType,
-                                 mlir::Type maskElemType,
-                                 mlir::Type resultElemTy) {
-  auto init = [](fir::FirOpBuilder builder, mlir::Location loc,
-                 mlir::Type elementType) {
+static void genRuntimeMinMaxlocBody(fir::FirOpBuilder &builder,
+                                    mlir::func::FuncOp &funcOp, bool isMax,
+                                    unsigned rank, int maskRank,
+                                    mlir::Type elementType,
+                                    mlir::Type maskElemType,
+                                    mlir::Type resultElemTy) {
+  auto init = [isMax](fir::FirOpBuilder builder, mlir::Location loc,
+                      mlir::Type elementType) {
     if (auto ty = elementType.dyn_cast<mlir::FloatType>()) {
       const llvm::fltSemantics &sem = ty.getFloatSemantics();
       return builder.createRealConstant(
-          loc, elementType, llvm::APFloat::getLargest(sem, /*Negative=*/false));
+          loc, elementType, llvm::APFloat::getLargest(sem, /*Negative=*/isMax));
     }
     unsigned bits = elementType.getIntOrFloatBitWidth();
-    int64_t maxInt = llvm::APInt::getSignedMaxValue(bits).getSExtValue();
+    int64_t maxInt = (isMax ? llvm::APInt::getSignedMinValue(bits)
+                            : llvm::APInt::getSignedMaxValue(bits))
+                         .getSExtValue();
     return builder.createIntegerConstant(loc, elementType, maxInt);
   };
 
@@ -797,18 +799,24 @@ static void genRuntimeMinlocBody(fir::FirOpBuilder &builder,
   }
 
   auto genBodyOp =
-      [&rank, &resultArr](
-          fir::FirOpBuilder builder, mlir::Location loc, mlir::Type elementType,
-          mlir::Value elem1, mlir::Value elem2,
-          llvm::SmallVector<mlir::Value, Fortran::common::maxRank> indices)
+      [&rank, &resultArr,
+       isMax](fir::FirOpBuilder builder, mlir::Location loc,
+              mlir::Type elementType, mlir::Value elem1, mlir::Value elem2,
+              llvm::SmallVector<mlir::Value, Fortran::common::maxRank> indices)
       -> mlir::Value {
     mlir::Value cmp;
     if (elementType.isa<mlir::FloatType>()) {
       cmp = builder.create<mlir::arith::CmpFOp>(
-          loc, mlir::arith::CmpFPredicate::OLT, elem1, elem2);
+          loc,
+          isMax ? mlir::arith::CmpFPredicate::OGT
+                : mlir::arith::CmpFPredicate::OLT,
+          elem1, elem2);
     } else if (elementType.isa<mlir::IntegerType>()) {
       cmp = builder.create<mlir::arith::CmpIOp>(
-          loc, mlir::arith::CmpIPredicate::slt, elem1, elem2);
+          loc,
+          isMax ? mlir::arith::CmpIPredicate::sgt
+                : mlir::arith::CmpIPredicate::slt,
+          elem1, elem2);
     } else {
       llvm_unreachable("unsupported type");
     }
@@ -875,9 +883,8 @@ static void genRuntimeMinlocBody(fir::FirOpBuilder &builder,
   // bit of a hack - maskRank is set to -1 for absent mask arg, so don't
   // generate high level mask or element by element mask.
   bool hasMask = maskRank > 0;
-
-  genMinlocReductionLoop(builder, funcOp, init, genBodyOp, rank, elementType,
-                         loc, hasMask, maskElemType, resultArr);
+  genMinMaxlocReductionLoop(builder, funcOp, init, genBodyOp, rank, elementType,
+                            loc, hasMask, maskElemType, resultArr);
 }
 
 /// Generate function type for the simplified version of RTNAME(DotProduct)
@@ -1150,8 +1157,8 @@ void SimplifyIntrinsicsPass::simplifyLogicalDim1Reduction(
                         intElementType);
 }
 
-void SimplifyIntrinsicsPass::simplifyMinlocReduction(
-    fir::CallOp call, const fir::KindMapping &kindMap) {
+void SimplifyIntrinsicsPass::simplifyMinMaxlocReduction(
+    fir::CallOp call, const fir::KindMapping &kindMap, bool isMax) {
 
   mlir::Operation::operand_range args = call.getArgs();
 
@@ -1217,11 +1224,11 @@ void SimplifyIntrinsicsPass::simplifyMinlocReduction(
   auto typeGenerator = [rank](fir::FirOpBuilder &builder) {
     return genRuntimeMinlocType(builder, rank);
   };
-  auto bodyGenerator = [rank, maskRank, inputType, logicalElemType,
-                        outType](fir::FirOpBuilder &builder,
-                                 mlir::func::FuncOp &funcOp) {
-    genRuntimeMinlocBody(builder, funcOp, rank, maskRank, inputType,
-                         logicalElemType, outType);
+  auto bodyGenerator = [rank, maskRank, inputType, logicalElemType, outType,
+                        isMax](fir::FirOpBuilder &builder,
+                               mlir::func::FuncOp &funcOp) {
+    genRuntimeMinMaxlocBody(builder, funcOp, isMax, rank, maskRank, inputType,
+                            logicalElemType, outType);
   };
 
   mlir::func::FuncOp newFunc =
@@ -1367,7 +1374,11 @@ void SimplifyIntrinsicsPass::runOnOperation() {
           return;
         }
         if (funcName.starts_with(RTNAME_STRING(Minloc))) {
-          simplifyMinlocReduction(call, kindMap);
+          simplifyMinMaxlocReduction(call, kindMap, false);
+          return;
+        }
+        if (funcName.starts_with(RTNAME_STRING(Maxloc))) {
+          simplifyMinMaxlocReduction(call, kindMap, true);
           return;
         }
       }
diff --git a/flang/test/Transforms/simplifyintrinsics.fir b/flang/test/Transforms/simplifyintrinsics.fir
index 39483a9cc18fe8..7dc758a522cd85 100644
--- a/flang/test/Transforms/simplifyintrinsics.fir
+++ b/flang/test/Transforms/simplifyintrinsics.fir
@@ -2348,3 +2348,250 @@ func.func @_QPtestminloc_doesntwork1d_unknownmask(%arg0: !fir.ref<!fir.array<10x
 // CHECK-NOT:         fir.call @_FortranAMinlocInteger4x1_i32_contract_simplified({{.*}}) fastmath<contract> : (!fir.ref<!fir.box<none>>, !fir.box<none>, !fir.box<none>) -> ()
 // CHECK:             fir.call @_FortranAMinlocInteger4({{.*}}) fastmath<contract> : (!fir.ref<!fir.box<none>>, !fir.box<none>, i32, !fir.ref<i8>, i32, !fir.box<none>, i1) -> none
 // CHECK-NOT:         fir.call @_FortranAMinlocInteger4x1_i32_contract_simplified({{.*}}) fastmath<contract> : (!fir.ref<!fir.box<none>>, !fir.box<none>, !fir.box<none>) -> ()
+
+// -----
+// Check Maxloc simplifies similarly to minloc
+func.func @_QPtestmaxloc_works1d(%arg0: !fir.ref<!fir.array<10xi32>> {fir.bindc_name = "a"}, %arg1: !fir.ref<!fir.array<10x!fir.logical<4>>> {fir.bindc_name = "b"}) -> !fir.array<1xi32> {
+  %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xi32>>>
+  %c10 = arith.constant 10 : index
+  %c10_0 = arith.constant 10 : index
+  %c1 = arith.constant 1 : index
+  %1 = fir.alloca !fir.array<1xi32> {bindc_name = "testmaxloc_works1d", uniq_name = "_QFtestmaxloc_works1dEtestmaxloc_works1d"}
+  %2 = fir.shape %c1 : (index) -> !fir.shape<1>
+  %3 = fir.array_load %1(%2) : (!fir.ref<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.array<1xi32>
+  %4 = fir.shape %c10 : (index) -> !fir.shape<1>
+  %5 = fir.embox %arg0(%4) : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<10xi32>>
+  %6 = fir.shape %c10_0 : (index) -> !fir.shape<1>
+  %7 = fir.embox %arg1(%6) : (!fir.ref<!fir.array<10x!fir.logical<4>>>, !fir.shape<1>) -> !fir.box<!fir.array<10x!fir.logical<4>>>
+  %c4 = arith.constant 4 : index
+  %false = arith.constant false
+  %8 = fir.zero_bits !fir.heap<!fir.array<?xi32>>
+  %c0 = arith.constant 0 : index
+  %9 = fir.shape %c0 : (index) -> !fir.shape<1>
+  %10 = fir.embox %8(%9) : (!fir.heap<!fir.array<?xi32>>, !fir.shape<1>) -> !fir.box<!fir.heap<!fir.array<?xi32>>>
+  fir.store %10 to %0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+  %11 = fir.address_of(@_QQclXea5bcf7f706678e1796661f8916f3379) : !fir.ref<!fir.char<1,55>>
+  %c5_i32 = arith.constant 5 : i32
+  %12 = fir.convert %0 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> !fir.ref<!fir.box<none>>
+  %13 = fir.convert %5 : (!fir.box<!fir.array<10xi32>>) -> !fir.box<none>
+  %14 = fir.convert %c4 : (index) -> i32
+  %15 = fir.convert %11 : (!fir.ref<!fir.char<1,55>>) -> !fir.ref<i8>
+  %16 = fir.convert %7 : (!fir.box<!fir.array<10x!fir.logical<4>>>) -> !fir.box<none>
+  %17 = fir.call @_FortranAMaxlocInteger4(%12, %13, %14, %15, %c5_i32, %16, %false) fastmath<contract> : (!fir.ref<!fir.box<none>>, !fir.box<none>, i32, !fir.ref<i8>, i32, !fir.box<none>, i1) -> none
+  %18 = fir.load %0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+  %c0_1 = arith.constant 0 : index
+  %19:3 = fir.box_dims %18, %c0_1 : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
+  %20 = fir.box_addr %18 : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.heap<!fir.array<?xi32>>
+  %21 = fir.shape_shift %19#0, %19#1 : (index, index) -> !fir.shapeshift<1>
+  %22 = fir.array_load %20(%21) : (!fir.heap<!fir.array<?xi32>>, !fir.shapeshift<1>) -> !fir.array<?xi32>
+  %c1_2 = arith.constant 1 : index
+  %c0_3 = arith.constant 0 : index
+  %23 = arith.subi %c1, %c1_2 : index
+  %24 = fir.do_loop %arg2 = %c0_3 to %23 step %c1_2 unordered iter_args(%arg3 = %3) -> (!fir.array<1xi32>) {
+    %26 = fir.array_fetch %22, %arg2 : (!fir.array<?xi32>, index) -> i32
+    %27 = fir.array_update %arg3, %26, %arg2 : (!fir.array<1xi32>, i32, index) -> !fir.array<1xi32>
+    fir.result %27 : !fir.array<1xi32>
+  }
+  fir.array_merge_store %3, %24 to %1 : !fir.array<1xi32>, !fir.array<1xi32>, !fir.ref<!fir.array<1xi32>>
+  fir.freemem %20 : !fir.heap<!fir.array<?xi32>>
+  %25 = fir.load %1 : !fir.ref<!fir.array<1xi32>>
+  return %25 : !fir.array<1xi32>
+}
+
+// CHECK-LABEL:   func.func @_QPtestmaxloc_works1d(
+// CHECK-SAME:                                     %[[INARR:.*]]: !fir.ref<!fir.array<10xi32>> {fir.bindc_name = "a"},
+// CHECK-SAME:                                     %[[MASK:.*]]: !fir.ref<!fir.array<10x!fir.logical<4>>> {fir.bindc_name = "b"}) -> !fir.array<1xi32> {
+// CHECK:           %[[OUTARR:.*]] = fir.alloca !fir.box<!fir.heap<!fir.array<?xi32>>>
+// CHECK:           %[[SIZE10_0:.*]] = arith.constant 10 : index
+// CHECK:           %[[SIZE10_1:.*]] = arith.constant 10 : index
+// CHECK:           %[[INARR_SHAPE:.*]] = fir.shape %[[SIZE10_0]] : (index) -> !fir.shape<1>
+// CHECK:           %[[BOX_INARR:.*]] = fir.embox %[[INARR]](%[[INARR_SHAPE]]) : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<10xi32>>
+// CHECK:           %[[MASK_SHAPE:.*]] = fir.shape %[[SIZE10_1]] : (index) -> !fir.shape<1>
+// CHECK:           %[[BOX_MASK:.*]] = fir.embox %[[MASK]](%[[MASK_SHAPE]]) : (!fir.ref<!fir.array<10x!fir.logical<4>>>, !fir.shape<1>) -> !fir.box<!fir.array<10x!fir.logical<4>>>
+// CHECK:           %[[REF_BOX_OUTARR_NONE:.*]] = fir.convert %[[OUTARR]] : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> !fir.ref<!fir.box<none>>
+// CHECK:           %[[BOX_INARR_NONE:.*]] = fir.convert %[[BOX_INARR]] : (!fir.box<!fir.array<10xi32>>) -> !fir.box<none>
+// CHECK:           %[[BOX_MASK_NONE:.*]] = fir.convert %[[BOX_MASK]] : (!fir.box<!fir.array<10x!fir.logical<4>>>) -> !fir.box<none>
+// CHECK:           fir.call @_FortranAMaxlocInteger4x1_Logical4x1_i32_contract_simplified(%[[REF_BOX_OUTARR_NONE]], %[[BOX_INARR_NONE]], %[[BOX_MASK_NONE]]) fastmath<contract> : (!fir.ref<!fir.box<none>>, !fir.box<none>, !fir.box<none>) -> ()
+
+// CHECK-LABEL:   func.func private @_FortranAMaxlocInteger4x1_Logical4x1_i32_contract_simplified(
+// CHECK-SAME:             %[[REF_BOX_OUTARR_NONE:.*]]: !fir.ref<!fir.box<none>>,
+// CHECK-SAME:             %[[BOX_INARR_NONE:.*]]: !fir.box<none>,
+// CHECK-SAME:             %[[BOX_MASK_NONE:.*]]: !fir.box<none>) attributes {llvm.linkage = #llvm.linkage<linkonce_odr>} {
+// CHECK:           %[[FLAG_ALLOC:.*]] = fir.alloca i32
+// CHECK:           %[[INIT_OUT_IDX:.*]] = arith.constant 0 : i32
+// CHECK:           %[[OUTARR_SIZE:.*]] = arith.constant 1 : index
+// CHECK:           %[[OUTARR:.*]] = fir.allocmem !fir.array<1xi32>
+// CHECK:           %[[OUTARR_SHAPE:.*]] = fir.shape %[[OUTARR_SIZE]] : (index) -> !fir.shape<1>
+// CHECK:           %[[BOX_OUTARR:.*]] = fir.embox %[[OUTARR]](%[[OUTARR_SHAPE]]) : (!fir.heap<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.box<!fir.heap<!fir.array<1xi32>>>
+// CHECK:           %[[BOX_MASK:.*]] = fir.convert %[[BOX_MASK_NONE]] : (!fir.box<none>) -> !fir.box<!fir.array<?x!fir.logical<4>>>
+// CHECK:           %[[OUTARR_IDX0:.*]] = arith.constant 0 : index
+// CHECK:           %[[OUTARR_ITEM0:.*]] = fir.coordinate_of %[[BOX_OUTARR]], %[[OUTARR_IDX0]] : (!fir.box<!fir.heap<!fir.array<1xi32>>>, index) -> !fir.ref<i32>
+// CHECK:           fir.store %[[INIT_OUT_IDX]] to %[[OUTARR_ITEM0]] : !fir.ref<i32>
+// CHECK:           %[[CINDEX_0:.*]] = arith.constant 0 : index
+// CHECK:           %[[BOX_INARR:.*]] = fir.convert %[[BOX_INARR_NONE]] : (!fir.box<none>) -> !fir.box<!fir.array<?xi32>>
+// CHECK:           %[[FLAG_SET:.*]] = arith.constant 1 : i32
+// CHECK:           %[[FLAG_EMPTY:.*]] = arith.constant 0 : i32
+// CHECK:           fir.store %[[FLAG_EMPTY]] to %[[FLAG_ALLOC]] : !fir.ref<i32>
+// CHECK:           %[[MAX:.*]] = arith.constant -2147483648 : i32
+// CHECK:           %[[CINDEX_1:.*]] = arith.constant 1 : index
+// CHECK:           %[[DIM_INDEX0:.*]] = arith.constant 0 : index
+// CHECK:           %[[DIMS:.*]]:3 = fir.box_dims %[[BOX_INARR]], %[[DIM_INDEX0]] : (!fir.box<!fir.array<?xi32>>, index) -> (index, index, index)
+// CHECK:           %[[EXTENT:.*]] = arith.subi %[[DIMS]]#1, %[[CINDEX_1]] : index
+// CHECK:           %[[DOLOOP:.*]] = fir.do_loop %[[ITER:.*]] = %[[CINDEX_0]] to %[[EXTENT]] step %[[CINDEX_1]] iter_args(%[[MIN:.*]] = %[[MAX]]) -> (i32) {
+// CHECK:             %[[MASK_ITEM:.*]] = fir.coordinate_of %[[BOX_MASK]], %[[ITER]] : (!fir.box<!fir.array<?x!fir.logical<4>>>, index) -> !fir.ref<!fir.logical<4>>
+// CHECK:             %[[MASK_ITEMVAL:.*]] = fir.load %[[MASK_ITEM]] : !fir.ref<!fir.logical<4>>
+// CHECK:             %[[MASK_IF_ITEM:.*]] = fir.convert %[[MASK_ITEMVAL]] : (!fir.logical<4>) -> i1
+// CHECK:             %[[IF_MASK:.*]] = fir.if %[[MASK_IF_ITEM]] -> (i32) {
+// CHECK:               %[[FLAG_SET2:.*]] = arith.constant 1 : i32
+// CHECK:               fir.store %[[FLAG_SET2]] to %[[FLAG_ALLOC]] : !fir.ref<i32>
+// CHECK:               %[[INARR_ITEM:.*]] = fir.coordinate_of %[[BOX_INARR]], %[[ITER]] : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
+// CHECK:               %[[INARR_ITEMVAL:.*]] = fir.load %[[INARR_ITEM]] : !fir.ref<i32>
+// CHECK:               %[[NEW_MIN:.*]] = arith.cmpi sgt, %[[INARR_ITEMVAL]], %[[MIN]] : i32
+// CHECK:               %[[IF_NEW_MIN:.*]] = fir.if %[[NEW_MIN]] -> (i32) {
+// CHECK:                 %[[ONE:.*]] = arith.constant 1 : i32
+// CHECK:                 %[[OUTARR_IDX:.*]] = arith.constant 0 : index
+// CHECK:                 %[[OUTARR_ITEM:.*]] = fir.coordinate_of %[[BOX_OUTARR]], %[[OUTARR_IDX]] : (!fir.box<!fir.heap<!fir.array<1xi32>>>, index) -> !fir.ref<i32>
+// CHECK:                 %[[ITER_I32:.*]] = fir.convert %[[ITER]] : (index) -> i32
+// CHECK:                 %[[FORTRAN_IDX:.*]] = arith.addi %[[ITER_I32]], %[[ONE]] : i32
+// CHECK:                 fir.store %[[FORTRAN_IDX]] to %[[OUTARR_ITEM]] : !fir.ref<i32>
+// CHECK:                 fir.result %[[INARR_ITEMVAL]] : i32
+// CHECK:               } else {
+// CHECK:                 fir.result %[[MIN]] : i32
+// CHECK:               }
+// CHECK:               fir.result %[[IF_NEW_MIN:.*]] : i32
+// CHECK:             } else {
+// CHECK:               fir.result %[[MIN]] : i32
+// CHECK:             }
+// CHECK:             fir.result %[[IF_MASK:.*]] : i32
+// CHECK:           }
+// CHECK:           %[[FLAG_VAL:.*]] = fir.load %[[FLAG_ALLOC]] : !fir.ref<i32>
+// CHECK:           %[[FLAG_WAS_SET:.*]] = arith.cmpi eq, %[[FLAG_VAL]], %[[FLAG_SET]] : i32
+// CHECK:           fir.if %[[FLAG_WAS_SET]] {
+// CHECK:             %[[TEST_MAX:.*]] = arith.constant -2147483648 : i32
+// CHECK:             %[[INIT_NOT_CHANGED:.*]] = arith.cmpi eq, %[[TEST_MAX]], %[[DO_LOOP:.*]] : i32
+// CHECK:             fir.if %[[INIT_NOT_CHANGED]] {
+// CHECK:               %[[FLAG_OUTARR_IDX:.*]] = arith.constant 0 : index
+// CHECK:               %[[FLAG_OUTARR_ITEM:.*]] = fir.coordinate_of %[[BOX_OUTARR]], %[[FLAG_OUTARR_IDX]] : (!fir.box<!fir.heap<!fir.array<1xi32>>>, index) -> !fir.ref<i32>
+// CHECK:               fir.store %[[FLAG_SET]] to %[[FLAG_OUTARR_ITEM]] : !fir.ref<i32>
+// CHECK:             }
+// CHECK:           }
+// CHECK:           %[[REF_BOX_OUTARR:.*]] = fir.convert %[[REF_BOX_OUTARR_NONE]] : (!fir.ref<!fir.box<none>>) -> !fir.ref<!fir.box<!fir.heap<!fir.array<1xi32>>>>
+// CHECK:           fir.store %[[BOX_OUTARR]] to %[[REF_BOX_OUTARR]] : !fir.ref<!fir.box<!fir.heap<!fir.array<1xi32>>>>
+// CHECK:           return
+// CHECK:         }
+
+// -----
+// Check Maxloc simplifies correctly for 1D case with scalar mask and f64 input
+
+func.func @_QPtestmaxloc_works1d_scalarmask_f64(%arg0: !fir.ref<!fir.array<10xf64>> {fir.bindc_name = "a"}, %arg1: !fir.ref<!fir.logical<4>> {fir.bindc_name = "b"}) -> !fir.array<1xi32> {
+  %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xi32>>>
+  %c10 = arith.constant 10 : index
+  %c1 = arith.constant 1 : index
+  %1 = fir.alloca !fir.array<1xi32> {bindc_name = "testmaxloc_works1d_scalarmask_f64", uniq_name = "_QFtestmaxloc_works1d_scalarmask_f64Etestminloc_works1d_scalarmask_f64"}
+  %2 = fir.shape %c1 : (index) -> !fir.shape<1>
+  %3 = fir.array_load %1(%2) : (!fir.ref<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.array<1xi32>
+  %4 = fir.shape %c10 : (index) -> !fir.shape<1>
+  %5 = fir.embox %arg0(%4) : (!fir.ref<!fir.array<10xf64>>, !fir.shape<1>) -> !fir.box<!fir.array<10xf64>>
+  %6 = fir.embox %arg1 : (!fir.ref<!fir.logical<4>>) -> !fir.box<!fir.logical<4>>
+  %c4 = arith.constant 4 : index
+  %false = arith.constant false
+  %7 = fir.zero_bits !fir.heap<!fir.array<?xi32>>
+  %c0 = arith.constant 0 : index
+  %8 = fir.shape %c0 : (index) -> !fir.shape<1>
+  %9 = fir.embox %7(%8) : (!fir.heap<!fir.array<?xi32>>, !fir.shape<1>) -> !fir.box<!fir.heap<!fir.array<?xi32>>>
+  fir.store %9 to %0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+  %10 = fir.address_of(@_QQclX66951c28c5b8bab5cdb25c1ac762b978) : !fir.ref<!fir.char<1,65>>
+  %c6_i32 = arith.constant 6 : i32
+  %11 = fir.convert %0 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> !fir.ref<!fir.box<none>>
+  %12 = fir.convert %5 : (!fir.box<!fir.array<10xf64>>) -> !fir.box<none>
+  %13 = fir.convert %c4 : (index) -> i32
+  %14 = fir.convert %10 : (!fir.ref<!fir.char<1,65>>) -> !fir.ref<i8>
+  %15 = fir.convert %6 : (!fir.box<!fir.logical<4>>) -> !fir.box<none>
+  %16 = fir.call @_FortranAMaxlocReal8(%11, %12, %13, %14, %c6_i32, %15, %false) fastmath<contract> : (!fir.ref<!fir.box<none>>, !fir.box<none>, i32, !fir.ref<i8>, i32, !fir.box<none>, i1) -> none
+  %17 = fir.load %0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+  %c0_0 = arith.constant 0 : index
+  %18:3 = fir.box_dims %17, %c0_0 : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
+  %19 = fir.box_addr %17 : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.heap<!fir.array<?xi32>>
+  %20 = fir.shape_shift %18#0, %18#1 : (index, index) -> !fir.shapeshift<1>
+  %21 = fir.array_load %19(%20) : (!fir.heap<!fir.array<?xi32>>, !fir.shapeshift<1>) -> !fir.array<?xi32>
+  %c1_1 = arith.constant 1 : index
+  %c0_2 = arith.constant 0 : index
+  %22 = arith.subi %c1, %c1_1 : index
+  %23 = fir.do_loop %arg2 = %c0_2 to %22 step %c1_1 unordered iter_args(%arg3 = %3) -> (!fir.array<1xi32>) {
+    %25 = fir.array_fetch %21, %arg2 : (!fir.array<?xi32>, index) -> i32
+    %26 = fir.array_update %arg3, %25, %arg2 : (!fir.array<1xi32>, i32, index) -> !fir.array<1xi32>
+    fir.result %26 : !fir.array<1xi32>
+  }
+  fir.array_merge_store %3, %23 to %1 : !fir.array<1xi32>, !fir.array<1xi32>, !fir.ref<!fir.array<1xi32>>
+  fir.freemem %19 : !fir.heap<!fir.array<?xi32>>
+  %24 = fir.load %1 : !fir.ref<!fir.array<1xi32>>
+  return %24 : !fir.array<1xi32>
+}
+
+// CHECK-LABEL:   func.func @_QPtestmaxloc_works1d_scalarmask_f64(
+// CHECK-SAME:                                     %[[INARR:.*]]: !fir.ref<!fir.array<10xf64>> {fir.bindc_name = "a"},
+// CHECK-SAME:                                     %[[MASK:.*]]: !fir.ref<!fir.logical<4>> {fir.bindc_name = "b"}) -> !fir.array<1xi32> {
+// CHECK:           fir.call @_FortranAMaxlocReal8x1_Logical4x0_i32_contract_simplified({{.*}}, {{.*}}, {{.*}}) fastmath<contract> : (!fir.ref<!fir.box<none>>, !fir.box<none>, !fir.box<none>) -> ()
+
+// CHECK-LABEL:   func.func private @_FortranAMaxlocReal8x1_Logical4x0_i32_contract_simplified(
+// CHECK-SAME:             %[[REF_BOX_OUTARR_NONE:.*]]: !fir.ref<!fir.box<none>>,
+// CHECK-SAME:             %[[BOX_INARR_NONE:.*]]: !fir.box<none>,
+// CHECK-SAME:             %[[BOX_MASK_NONE:.*]]: !fir.box<none>) attributes {llvm.linkage = #llvm.linkage<linkonce_odr>} {
+// CHECK:           %[[FLAG_ALLOC:.*]] = fir.alloca i32
+// CHECK:           %[[INIT_OUT_IDX:.*]] = arith.constant 0 : i32
+// CHECK:           %[[OUTARR_SIZE:.*]] = arith.constant 1 : index
+// CHECK:           %[[OUTARR:.*]] = fir.allocmem !fir.array<1xi32>
+// CHECK:           %[[OUTARR_SHAPE:.*]] = fir.shape %[[OUTARR_SIZE]] : (index) -> !fir.shape<1>
+// CHECK:           %[[BOX_OUTARR:.*]] = fir.embox %[[OUTARR]](%[[OUTARR_SHAPE]]) : (!fir.heap<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.box<!fir.heap<!fir.array<1xi32>>>
+// CHECK:           %[[OUTARR_IDX0:.*]] = arith.constant 0 : index
+// CHECK:           %[[OUTARR_ITEM0:.*]] = fir.coordinate_of %[[BOX_OUTARR]], %[[OUTARR_IDX0]] : (!fir.box<!fir.heap<!fir.array<1xi32>>>, index) -> !fir.ref<i32>
+// CHECK:           fir.store %[[INIT_OUT_IDX]] to %[[OUTARR_ITEM0]] : !fir.ref<i32>
+// CHECK:           %[[BOX_MASK:.*]] = fir.convert %[[BOX_MASK_NONE]] : (!fir.box<none>) -> !fir.box<!fir.array<1xi1>>
+// CHECK:           %[[MASK_IDX0:.*]] = arith.constant 0 : index
+// CHECK:           %[[MASK_ITEM:.*]] = fir.coordinate_of %[[BOX_MASK]], %[[MASK_IDX0]] : (!fir.box<!fir.array<1xi1>>, index) -> !fir.ref<i1>
+// CHECK:           %[[MASK:.*]] = fir.load %[[MASK_ITEM]] : !fir.ref<i1>
+// CHECK:           %[[INIT_RES:.*]] = fir.if %[[MASK]] -> (f64) {
+// CHECK:             %[[C_INDEX0:.*]] = arith.constant 0 : index
+// CHECK:             %[[BOX_INARR:.*]] = fir.convert %[[BOX_INARR_NONE]] : (!fir.box<none>) -> !fir.box<!fir.array<?xf64>>
+// CHECK:             %[[FLAG_SET:.*]] = arith.constant 1 : i32
+// CHECK:             %[[FLAG_EMPTY:.*]] = arith.constant 0 : i32
+// CHECK:             fir.store %[[FLAG_EMPTY]] to %[[FLAG_ALLOC]] : !fir.ref<i32>
+// CHECK:             %[[MAX:.*]] = arith.constant -1.7976931348623157E+308 : f64
+// CHECK:             %[[C_INDEX1:.*]] = arith.constant 1 : index
+// CHECK:             %[[DIM_INDEX:.*]] = arith.constant 0 : index
+// CHECK:             %[[DIMS:.*]]:3 = fir.box_dims %[[BOX_INARR]], %[[DIM_INDEX]] : (!fir.box<!fir.array<?xf64>>, index) -> (index, index, index)
+// CHECK:             %[[EXTENT:.*]] = arith.subi %[[DIMS]]#1, %[[C_INDEX1]] : index
+// CHECK:             %[[DOLOOP:.*]] = fir.do_loop %[[ITER:.*]] = %[[C_INDEX0]] to %[[EXTENT]] step %[[C_INDEX1]] iter_args(%[[MIN:.*]] = %[[MAX]]) -> (f64) {
+// CHECK:               %[[INARR_ITEM:.*]] = fir.coordinate_of %[[BOX_INARR]], %[[ITER]] : (!fir.box<!fir.array<?xf64>>, index) -> !fir.ref<f64>
+// CHECK:               %[[INARR_ITEMVAL:.*]] = fir.load %[[INARR_ITEM]] : !fir.ref<f64>
+// CHECK:               %[[NEW_MIN:.*]] = arith.cmpf ogt, %[[INARR_ITEMVAL]], %[[MIN]] fastmath<{{.*}}> : f64
+// CHECK:               %[[IF_NEW_MIN:.*]] = fir.if %[[NEW_MIN]] -> (f64) {
+// CHECK:                 %[[ONE:.*]] = arith.constant 1 : i32
+// CHECK:                 %[[OUTARR_IDX:.*]] = arith.constant 0 : index
+// CHECK:                 %[[OUTARR_ITEM:.*]] = fir.coordinate_of %[[BOX_OUTARR]], %[[OUTARR_IDX]] : (!fir.box<!fir.heap<!fir.array<1xi32>>>, index) -> !fir.ref<i32>
+// CHECK:                 %[[ITER_I32:.*]] = fir.convert %[[ITER]] : (index) -> i32
+// CHECK:                 %[[FORTRAN_IDX:.*]] = arith.addi %[[ITER_I32]], %[[ONE]] : i32
+// CHECK:                 fir.store %[[FORTRAN_IDX]] to %[[OUTARR_ITEM]] : !fir.ref<i32>
+// CHECK:                 fir.result %[[INARR_ITEMVAL]] : f64
+// CHECK:               } else {
+// CHECK:                 fir.result %[[MIN]] : f64
+// CHECK:               }
+// CHECK:               fir.result %[[IF_NEW_MIN:.*]] : f64
+// CHECK:             }
+// CHECK:           }
+// CHECK:           %[[FLAG_CHECK:.*]] = arith.constant 1 : i32
+// CHECK:           %[[FLAG_VAL:.*]] = fir.load %[[FLAG_ALLOC]] : !fir.ref<i32>
+// CHECK:           %[[FLAG_WAS_SET:.*]] = arith.cmpi eq, %[[FLAG_VAL]], %[[FLAG_CHECK]] : i32
+// CHECK:           fir.if %[[FLAG_WAS_SET]] {
+// CHECK:             %[[TEST_MAX:.*]] = arith.constant -1.7976931348623157E+308 : f64
+// CHECK:             %[[INIT_NOT_CHANGED:.*]] = arith.cmpf oeq, %[[TEST_MAX]], %[[INIT_RES:.*]] fastmath<{{.*}}> : f64
+// CHECK:             fir.if %[[INIT_NOT_CHANGED]] {
+// CHECK:               %[[FLAG_OUTARR_IDX:.*]] = arith.constant 0 : index
+// CHECK:               %[[FLAG_OUTARR_ITEM:.*]] = fir.coordinate_of %[[BOX_OUTARR]], %[[FLAG_OUTARR_IDX]] : (!fir.box<!fir.heap<!fir.array<1xi32>>>, index) -> !fir.ref<i32>
+// CHECK:               fir.store %[[FLAG_CHECK]] to %[[FLAG_OUTARR_ITEM]] : !fir.ref<i32>
+// CHECK:             }
+// CHECK:           }
+// CHECK:           %[[REF_BOX_OUTARR:.*]] = fir.convert %[[VAL_0]] : (!fir.ref<!fir.box<none>>) -> !fir.ref<!fir.box<!fir.heap<!fir.array<1xi32>>>>
+// CHECK:           fir.store %[[BOX_OUTARR]] to %[[REF_BOX_OUTARR]] : !fir.ref<!fir.box<!fir.heap<!fir.array<1xi32>>>>
+// CHECK:           return
+// CHECK:         }

>From 874fb39277c981daf251f39f8f6d6ee112bbbe20 Mon Sep 17 00:00:00 2001
From: David Green <david.green at arm.com>
Date: Thu, 14 Dec 2023 12:09:15 +0000
Subject: [PATCH 2/2] Test fixup

---
 flang/test/Transforms/simplifyintrinsics.fir | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/flang/test/Transforms/simplifyintrinsics.fir b/flang/test/Transforms/simplifyintrinsics.fir
index 7dc758a522cd85..0bd6ac7c436ff2 100644
--- a/flang/test/Transforms/simplifyintrinsics.fir
+++ b/flang/test/Transforms/simplifyintrinsics.fir
@@ -2423,7 +2423,6 @@ func.func @_QPtestmaxloc_works1d(%arg0: !fir.ref<!fir.array<10xi32>> {fir.bindc_
 // CHECK:           %[[OUTARR:.*]] = fir.allocmem !fir.array<1xi32>
 // CHECK:           %[[OUTARR_SHAPE:.*]] = fir.shape %[[OUTARR_SIZE]] : (index) -> !fir.shape<1>
 // CHECK:           %[[BOX_OUTARR:.*]] = fir.embox %[[OUTARR]](%[[OUTARR_SHAPE]]) : (!fir.heap<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.box<!fir.heap<!fir.array<1xi32>>>
-// CHECK:           %[[BOX_MASK:.*]] = fir.convert %[[BOX_MASK_NONE]] : (!fir.box<none>) -> !fir.box<!fir.array<?x!fir.logical<4>>>
 // CHECK:           %[[OUTARR_IDX0:.*]] = arith.constant 0 : index
 // CHECK:           %[[OUTARR_ITEM0:.*]] = fir.coordinate_of %[[BOX_OUTARR]], %[[OUTARR_IDX0]] : (!fir.box<!fir.heap<!fir.array<1xi32>>>, index) -> !fir.ref<i32>
 // CHECK:           fir.store %[[INIT_OUT_IDX]] to %[[OUTARR_ITEM0]] : !fir.ref<i32>
@@ -2432,6 +2431,7 @@ func.func @_QPtestmaxloc_works1d(%arg0: !fir.ref<!fir.array<10xi32>> {fir.bindc_
 // CHECK:           %[[FLAG_SET:.*]] = arith.constant 1 : i32
 // CHECK:           %[[FLAG_EMPTY:.*]] = arith.constant 0 : i32
 // CHECK:           fir.store %[[FLAG_EMPTY]] to %[[FLAG_ALLOC]] : !fir.ref<i32>
+// CHECK:           %[[BOX_MASK:.*]] = fir.convert %[[BOX_MASK_NONE]] : (!fir.box<none>) -> !fir.box<!fir.array<?x!fir.logical<4>>>
 // CHECK:           %[[MAX:.*]] = arith.constant -2147483648 : i32
 // CHECK:           %[[CINDEX_1:.*]] = arith.constant 1 : index
 // CHECK:           %[[DIM_INDEX0:.*]] = arith.constant 0 : index
@@ -2442,8 +2442,7 @@ func.func @_QPtestmaxloc_works1d(%arg0: !fir.ref<!fir.array<10xi32>> {fir.bindc_
 // CHECK:             %[[MASK_ITEMVAL:.*]] = fir.load %[[MASK_ITEM]] : !fir.ref<!fir.logical<4>>
 // CHECK:             %[[MASK_IF_ITEM:.*]] = fir.convert %[[MASK_ITEMVAL]] : (!fir.logical<4>) -> i1
 // CHECK:             %[[IF_MASK:.*]] = fir.if %[[MASK_IF_ITEM]] -> (i32) {
-// CHECK:               %[[FLAG_SET2:.*]] = arith.constant 1 : i32
-// CHECK:               fir.store %[[FLAG_SET2]] to %[[FLAG_ALLOC]] : !fir.ref<i32>
+// CHECK:               fir.store %[[FLAG_SET]] to %[[FLAG_ALLOC]] : !fir.ref<i32>
 // CHECK:               %[[INARR_ITEM:.*]] = fir.coordinate_of %[[BOX_INARR]], %[[ITER]] : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
 // CHECK:               %[[INARR_ITEMVAL:.*]] = fir.load %[[INARR_ITEM]] : !fir.ref<i32>
 // CHECK:               %[[NEW_MIN:.*]] = arith.cmpi sgt, %[[INARR_ITEMVAL]], %[[MIN]] : i32



More information about the flang-commits mailing list