[flang] Lower character related intrinsic

[clementval]

Summary

This patch adds lowering for the following character related intrinsics:

• len
• len_trim
• lge, lgt, lle and llt

This patch is part of the upstreaming effort from fir-dev branch.

[PeteSteinfeld]

All builds and tests correctly and looks good.

flang/lib/Lower/IntrinsicCall.cpp

 #include "flang/Lower/Runtime.h"
 #include "flang/Lower/StatementContext.h"
 #include "flang/Lower/SymbolMap.h"
 #include "flang/Lower/Todo.h"
 #include "flang/Optimizer/Builder/Character.h"
 #include "flang/Optimizer/Builder/Complex.h"
 #include "flang/Optimizer/Builder/FIRBuilder.h"
 #include "flang/Optimizer/Builder/MutableBox.h"
+#include "flang/Optimizer/Builder/Runtime/Character.h"
 #include "flang/Optimizer/Builder/Runtime/Inquiry.h"
 #include "flang/Optimizer/Builder/Runtime/RTBuilder.h"
 #include "flang/Optimizer/Builder/Runtime/Reduction.h"
 #include "flang/Optimizer/Dialect/FIROpsSupport.h"
 #include "flang/Optimizer/Support/FatalError.h"
 #include "mlir/Dialect/LLVMIR/LLVMDialect.h"
 #include "llvm/Support/CommandLine.h"
 
fir::ExtendedValue genAllocated(mlir::Type,
                                   llvm::ArrayRef<fir::ExtendedValue>);
   fir::ExtendedValue genAny(mlir::Type, llvm::ArrayRef<fir::ExtendedValue>);
   fir::ExtendedValue genAssociated(mlir::Type,
                                    llvm::ArrayRef<fir::ExtendedValue>);
   fir::ExtendedValue genChar(mlir::Type, llvm::ArrayRef<fir::ExtendedValue>);
   mlir::Value genDim(mlir::Type, llvm::ArrayRef<mlir::Value>);
   fir::ExtendedValue genDotProduct(mlir::Type,
                                    llvm::ArrayRef<fir::ExtendedValue>);
+  template <mlir::arith::CmpIPredicate pred>
+  fir::ExtendedValue genCharacterCompare(mlir::Type,
+                                         llvm::ArrayRef<fir::ExtendedValue>);
   template <Extremum, ExtremumBehavior>
   mlir::Value genExtremum(mlir::Type, llvm::ArrayRef<mlir::Value>);
   /// Lowering for the IAND intrinsic. The IAND intrinsic expects two arguments
   /// in the llvm::ArrayRef.
   mlir::Value genIand(mlir::Type, llvm::ArrayRef<mlir::Value>);
   mlir::Value genIbits(mlir::Type, llvm::ArrayRef<mlir::Value>);
   fir::ExtendedValue genLbound(mlir::Type, llvm::ArrayRef<fir::ExtendedValue>);
   fir::ExtendedValue genNull(mlir::Type, llvm::ArrayRef<fir::ExtendedValue>);
+  fir::ExtendedValue genLen(mlir::Type, llvm::ArrayRef<fir::ExtendedValue>);
+  fir::ExtendedValue genLenTrim(mlir::Type, llvm::ArrayRef<fir::ExtendedValue>);
   fir::ExtendedValue genSize(mlir::Type, llvm::ArrayRef<fir::ExtendedValue>);
   fir::ExtendedValue genSum(mlir::Type, llvm::ArrayRef<fir::ExtendedValue>);
   fir::ExtendedValue genUbound(mlir::Type, llvm::ArrayRef<fir::ExtendedValue>);
 
   /// Define the different FIR generators that can be mapped to intrinsic to
   /// generate the related code.
   using ElementalGenerator = decltype(&IntrinsicLibrary::genAbs);
   using ExtendedGenerator = decltype(&IntrinsicLibrary::genSum);

 
 /// Structure describing what needs to be done to lower intrinsic "name".
 struct IntrinsicHandler {
   const char *name;
   IntrinsicLibrary::Generator generator;
   // The following may be omitted in the table below.
   Fortran::lower::IntrinsicArgumentLoweringRules argLoweringRules = {};
   bool isElemental = true;
+  /// Code heavy intrinsic can be outlined to make FIR
+  /// more readable.
+  bool outline = false;
 };
 
 constexpr auto asValue = Fortran::lower::LowerIntrinsicArgAs::Value;
 constexpr auto asAddr = Fortran::lower::LowerIntrinsicArgAs::Addr;
 constexpr auto asBox = Fortran::lower::LowerIntrinsicArgAs::Box;
 constexpr auto asInquired = Fortran::lower::LowerIntrinsicArgAs::Inquired;
 using I = IntrinsicLibrary;
 
static constexpr IntrinsicHandler handlers[]{
     {"char", &I::genChar},
     {"dim", &I::genDim},
     {"dot_product",
      &I::genDotProduct,
      {{{"vector_a", asBox}, {"vector_b", asBox}}},
      /*isElemental=*/false},
     {"iand", &I::genIand},
     {"ibits", &I::genIbits},
+    {"len",
+     &I::genLen,
+     {{{"string", asInquired}, {"kind", asValue}}},
+     /*isElemental=*/false},
+    {"len_trim", &I::genLenTrim},
+    {"lge", &I::genCharacterCompare<mlir::arith::CmpIPredicate::sge>},
+    {"lgt", &I::genCharacterCompare<mlir::arith::CmpIPredicate::sgt>},
+    {"lle", &I::genCharacterCompare<mlir::arith::CmpIPredicate::sle>},
+    {"llt", &I::genCharacterCompare<mlir::arith::CmpIPredicate::slt>},
     {"min", &I::genExtremum<Extremum::Min, ExtremumBehavior::MinMaxss>},
     {"null", &I::genNull, {{{"mold", asInquired}}}, /*isElemental=*/false},
     {"sum",
      &I::genSum,
      {{{"array", asBox},
        {"dim", asValue},
        {"mask", asBox, handleDynamicOptional}}},
      /*isElemental=*/false},
     {"ubound",
      &I::genUbound,
      {{{"array", asBox}, {"dim", asValue}, {"kind", asValue}}},
      /*isElemental=*/false},
 };
 
 static const IntrinsicHandler *findIntrinsicHandler(llvm::StringRef name) {
   auto compare = [](const IntrinsicHandler &handler, llvm::StringRef name) {
     return name.compare(handler.name) > 0;
   };
   auto result =
       std::lower_bound(std::begin(handlers), std::end(handlers), name, compare);
   return result != std::end(handlers) && result->name == name ? result
                                                               : nullptr;
 }
 
+/// To make fir output more readable for debug, one can outline all intrinsic
+/// implementation in wrappers (overrides the IntrinsicHandler::outline flag).
+static llvm::cl::opt<bool> outlineAllIntrinsics(
+    "outline-intrinsics",
+    llvm::cl::desc(
+        "Lower all intrinsic procedure implementation in their own functions"),
+    llvm::cl::init(false));
+
 //===----------------------------------------------------------------------===//
 // Math runtime description and matching utility
 //===----------------------------------------------------------------------===//
 
 /// Command line option to modify math runtime version used to implement
 /// intrinsics.
 enum MathRuntimeVersion { fastVersion, llvmOnly };
 llvm::cl::opt<MathRuntimeVersion> mathRuntimeVersion(
invokeHandler(IntrinsicLibrary::ExtendedGenerator generator,
   return std::invoke(generator, lib, *resultType, args);
 }
 
 fir::ExtendedValue
 IntrinsicLibrary::genIntrinsicCall(llvm::StringRef name,
                                    llvm::Optional<mlir::Type> resultType,
                                    llvm::ArrayRef<fir::ExtendedValue> args) {
   if (const IntrinsicHandler *handler = findIntrinsicHandler(name)) {
-    bool outline = false;
+    bool outline = handler->outline || outlineAllIntrinsics;
     return std::visit(
         [&](auto &generator) -> fir::ExtendedValue {
           return invokeHandler(generator, *handler, resultType, args, outline,
                                *this);
         },
         handler->generator);
   }
 
mlir::Value IntrinsicLibrary::genIbits(mlir::Type resultType,
   auto mask = builder.create<mlir::arith::ShRUIOp>(loc, ones, shiftCount);
   auto res1 = builder.create<mlir::arith::ShRSIOp>(loc, args[0], pos);
   auto res2 = builder.create<mlir::arith::AndIOp>(loc, res1, mask);
   auto lenIsZero = builder.create<mlir::arith::CmpIOp>(
       loc, mlir::arith::CmpIPredicate::eq, len, zero);
   return builder.create<mlir::arith::SelectOp>(loc, lenIsZero, zero, res2);
 }
 
+// LEN
+// Note that this is only used for an unrestricted intrinsic LEN call.
+// Other uses of LEN are rewritten as descriptor inquiries by the front-end.
+fir::ExtendedValue
+IntrinsicLibrary::genLen(mlir::Type resultType,
+                         llvm::ArrayRef<fir::ExtendedValue> args) {
+  // Optional KIND argument reflected in result type and otherwise ignored.
+  assert(args.size() == 1 || args.size() == 2);
+  mlir::Value len = fir::factory::readCharLen(builder, loc, args[0]);
+  return builder.createConvert(loc, resultType, len);
+}
+
+// LEN_TRIM
+fir::ExtendedValue
+IntrinsicLibrary::genLenTrim(mlir::Type resultType,
+                             llvm::ArrayRef<fir::ExtendedValue> args) {
+  // Optional KIND argument reflected in result type and otherwise ignored.
+  assert(args.size() == 1 || args.size() == 2);
+  const fir::CharBoxValue *charBox = args[0].getCharBox();
+  if (!charBox)
+    TODO(loc, "character array len_trim");
+  auto len =
+      fir::factory::CharacterExprHelper(builder, loc).createLenTrim(*charBox);
+  return builder.createConvert(loc, resultType, len);
+}
+
+// LGE, LGT, LLE, LLT
+template <mlir::arith::CmpIPredicate pred>
+fir::ExtendedValue
+IntrinsicLibrary::genCharacterCompare(mlir::Type type,
+                                      llvm::ArrayRef<fir::ExtendedValue> args) {
+  assert(args.size() == 2);
+  return fir::runtime::genCharCompare(
+      builder, loc, pred, fir::getBase(args[0]), fir::getLen(args[0]),
+      fir::getBase(args[1]), fir::getLen(args[1]));
+}
+
 // Compare two FIR values and return boolean result as i1.
 template <Extremum extremum, ExtremumBehavior behavior>
 static mlir::Value createExtremumCompare(mlir::Location loc,
                                          fir::FirOpBuilder &builder,
                                          mlir::Value left, mlir::Value right) {
   static constexpr mlir::arith::CmpIPredicate integerPredicate =
       extremum == Extremum::Max ? mlir::arith::CmpIPredicate::sgt
                                 : mlir::arith::CmpIPredicate::slt;

flang/lib/Optimizer/Builder/FIRBuilder.cpp

return box.match(
       },
       [&](const fir::BoxValue &x) -> mlir::Value {
         assert(x.isCharacter());
         if (!x.getExplicitParameters().empty())
           return x.getExplicitParameters()[0];
         return fir::factory::CharacterExprHelper{builder, loc}
             .readLengthFromBox(x.getAddr());
       },
-      [&](const fir::MutableBoxValue &) -> mlir::Value {
-        // MutableBoxValue must be read into another category to work with them
-        // outside of allocation/assignment contexts.
-        fir::emitFatalError(loc, "readCharLen on MutableBoxValue");
+      [&](const fir::MutableBoxValue &x) -> mlir::Value {
+        return readCharLen(builder, loc,
+                           fir::factory::genMutableBoxRead(builder, loc, x));
       },
       [&](const auto &) -> mlir::Value {
         fir::emitFatalError(
             loc, "Character length inquiry on a non-character entity");
       });
 }
 
 mlir::Value fir::factory::readExtent(fir::FirOpBuilder &builder,

flang/test/Lower/Intrinsics/len.f90

+! RUN: bbc -emit-fir %s -o - | FileCheck %s
+
+! CHECK-LABEL: len_test
+subroutine len_test(i, c)
+    integer :: i
+    character(*) :: c
+    ! CHECK: %[[c:.*]]:2 = fir.unboxchar %arg1
+    ! CHECK: %[[xx:.*]] = fir.convert %[[c]]#1 : (index) -> i32
+    ! CHECK: fir.store %[[xx]] to %arg0
+    i = len(c)
+  end subroutine
+  
+  ! CHECK-LABEL: len_test_array
+  ! CHECK-SAME: %[[arg0:.*]]: !fir.ref<i32> {fir.bindc_name = "i"}, %[[arg1:.*]]: !fir.boxchar<1> {fir.bindc_name = "c"}
+  subroutine len_test_array(i, c)
+    integer :: i
+    character(*) :: c(100)
+    ! CHECK: %[[c:.*]]:2 = fir.unboxchar %[[arg1]]
+    ! CHECK: %[[xx:.*]] = fir.convert %[[c]]#1 : (index) -> i32
+    ! CHECK: fir.store %[[xx]] to %[[arg0]]
+    i = len(c)
+  end subroutine
+  
+  ! CHECK-LABEL: func @_QPlen_test_assumed_shape_array(
+  ! CHECK-SAME:  %[[VAL_0:.*]]: !fir.ref<i32> {fir.bindc_name = "i"},
+  ! CHECK-SAME:  %[[VAL_1:.*]]: !fir.box<!fir.array<?x!fir.char<1,?>>> {fir.bindc_name = "c"}) {
+  subroutine len_test_assumed_shape_array(i, c)
+    integer :: i
+    character(*) :: c(:)
+  ! CHECK:  %[[VAL_2:.*]] = fir.box_elesize %[[VAL_1]] : (!fir.box<!fir.array<?x!fir.char<1,?>>>) -> index
+  ! CHECK:  %[[VAL_3:.*]] = fir.convert %[[VAL_2]] : (index) -> i32
+  ! CHECK:  fir.store %[[VAL_3]] to %[[VAL_0]] : !fir.ref<i32>
+    i = len(c)
+  end subroutine
+  
+  ! CHECK-LABEL: func @_QPlen_test_array_alloc(
+  ! CHECK-SAME:  %[[VAL_0:.*]]: !fir.ref<i32> {fir.bindc_name = "i"},
+  ! CHECK-SAME:  %[[VAL_1:.*]]: !fir.ref<!fir.box<!fir.heap<!fir.array<?x!fir.char<1,?>>>>> {fir.bindc_name = "c"}) {
+  subroutine len_test_array_alloc(i, c)
+    integer :: i
+    character(:), allocatable :: c(:)
+  ! CHECK:  %[[VAL_2:.*]] = fir.load %[[VAL_1]] : !fir.ref<!fir.box<!fir.heap<!fir.array<?x!fir.char<1,?>>>>>
+  ! CHECK:  %[[VAL_3:.*]] = fir.box_elesize %[[VAL_2]] : (!fir.box<!fir.heap<!fir.array<?x!fir.char<1,?>>>>) -> index
+  ! CHECK:  %[[VAL_4:.*]] = fir.convert %[[VAL_3]] : (index) -> i32
+  ! CHECK:  fir.store %[[VAL_4]] to %[[VAL_0]] : !fir.ref<i32>
+    i = len(c)
+  end subroutine
+  
+  ! CHECK-LABEL: func @_QPlen_test_array_local_alloc(
+  ! CHECK-SAME:  %[[VAL_0:.*]]: !fir.ref<i32> {fir.bindc_name = "i"})
+  subroutine len_test_array_local_alloc(i)
+    integer :: i
+    character(:), allocatable :: c(:)
+  ! CHECK:  %[[VAL_5:.*]] = fir.alloca index {uniq_name = "_QFlen_test_array_local_allocEc.len"}
+  ! CHECK:  %[[VAL_7:.*]] = arith.constant 10 : i32
+  ! CHECK:  %[[VAL_10:.*]] = fir.convert %[[VAL_7]] : (i32) -> index
+  ! CHECK:  fir.store %[[VAL_10]] to %[[VAL_5]] : !fir.ref<index>
+    allocate(character(10):: c(100))
+  ! CHECK:  %[[VAL_13:.*]] = fir.load %[[VAL_5]] : !fir.ref<index>
+  ! CHECK:  %[[VAL_14:.*]] = fir.convert %[[VAL_13]] : (index) -> i32
+  ! CHECK:  fir.store %[[VAL_14]] to %[[VAL_0]] : !fir.ref<i32>
+    i = len(c)
+  end subroutine
+  
+  ! CHECK-LABEL: func @_QPlen_test_alloc_explicit_len(
+  ! CHECK-SAME:  %[[VAL_0:.*]]: !fir.ref<i32> {fir.bindc_name = "i"},
+  ! CHECK-SAME:  %[[VAL_1:.*]]: !fir.ref<i32> {fir.bindc_name = "n"},
+  ! CHECK-SAME:  %[[VAL_2:.*]]: !fir.ref<!fir.box<!fir.heap<!fir.array<?x!fir.char<1,?>>>>> {fir.bindc_name = "c"}) {
+  subroutine len_test_alloc_explicit_len(i, n, c)
+    integer :: i
+    integer :: n
+    character(n), allocatable :: c(:)
+  ! CHECK:  %[[VAL_3:.*]] = fir.load %[[VAL_1]] : !fir.ref<i32>
+  ! CHECK:  fir.store %[[VAL_3]] to %[[VAL_0]] : !fir.ref<i32>
+    i = len(c)
+  end subroutine

flang/test/Lower/Intrinsics/len_trim.f90

+! RUN: bbc -emit-fir %s -o - | FileCheck %s
+
+! CHECK-LABEL: len_trim_test
+integer function len_trim_test(c)
+character(*) :: c
+ltrim = len_trim(c)
+! CHECK-DAG: %[[c0:.*]] = arith.constant 0 : index
+! CHECK-DAG: %[[c1:.*]] = arith.constant 1 : index
+! CHECK-DAG: %[[cm1:.*]] = arith.constant -1 : index
+! CHECK-DAG: %[[lastChar:.*]] = arith.subi {{.*}}, %[[c1]]
+! CHECK: %[[iterateResult:.*]]:2 = fir.iterate_while (%[[index:.*]] = %[[lastChar]] to %[[c0]] step %[[cm1]]) and ({{.*}}) iter_args({{.*}}) {
+  ! CHECK: %[[addr:.*]] = fir.coordinate_of {{.*}}, %[[index]]
+  ! CHECK: %[[codeAddr:.*]] = fir.convert %[[addr]]
+  ! CHECK: %[[code:.*]] = fir.load %[[codeAddr]]
+  ! CHECK: %[[bool:.*]] = arith.cmpi eq
+  ! CHECK: fir.result %[[bool]], %[[index]]
+! CHECK: }
+! CHECK: %[[len:.*]] = arith.addi %[[iterateResult]]#1, %[[c1]]
+! CHECK: select %[[iterateResult]]#0, %[[c0]], %[[len]]
+end function

flang/test/Lower/Intrinsics/lge_lgt_lle_llt.f90

+! RUN: bbc -emit-fir %s -o - | FileCheck %s
+
+subroutine lge_test
+    character*3 :: c1(3)
+    character*7 :: c2(3)
+    ! c1(1) = 'a'; c1(2) = 'B'; c1(3) = 'c';
+    ! c2(1) = 'A'; c2(2) = 'b'; c2(3) = 'c';
+    ! CHECK: BeginExternalListOutput
+    ! CHECK: fir.do_loop
+    ! CHECK: CharacterCompareScalar1
+    ! CHECK: OutputDescriptor
+    ! CHECK: EndIoStatement
+    print*, lge(c1, c2)
+    ! CHECK: BeginExternalListOutput
+    ! CHECK: fir.do_loop
+    ! CHECK: CharacterCompareScalar1
+    ! CHECK: OutputDescriptor
+    ! CHECK: EndIoStatement
+    print*, lgt(c1, c2)
+    ! CHECK: BeginExternalListOutput
+    ! CHECK: fir.do_loop
+    ! CHECK: CharacterCompareScalar1
+    ! CHECK: OutputDescriptor
+    ! CHECK: EndIoStatement
+    print*, lle(c1, c2)
+    ! CHECK: BeginExternalListOutput
+    ! CHECK: fir.do_loop
+    ! CHECK: CharacterCompareScalar1
+    ! CHECK: OutputDescriptor
+    ! CHECK: EndIoStatement
+    print*, llt(c1, c2)
+  end
+  
 No newline at end of file