[flang[OpenACC] Lower wait directive

[clementval]

Summary

This patch adds lowering for the !$acc wait directive
from the PFT to OpenACC dialect.

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

[PeteSteinfeld]

I get errors when I try to build this patch. Here's the first error:

[5791/5857] Building CXX object tools/flang/lib/Lower/CMakeFiles/obj.FortranLower.dir/OpenACC.cpp.o
FAILED: tools/flang/lib/Lower/CMakeFiles/obj.FortranLower.dir/OpenACC.cpp.o
/usr/bin/c++ -DFLANG_LITTLE_ENDIAN=1 -DGTEST_HAS_RTTI=0 -D_DEBUG -D_GNU_SOURCE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -I/local/home/psteinfeld/main/wait/build/tools/flang/lib/Lower -I/local/home/psteinfeld/main/wait/flang/lib/Lower -I/local/home/psteinfeld/main/wait/flang/include -I/local/home/psteinfeld/main/wait/build/tools/flang/include -I/local/home/psteinfeld/main/wait/build/include -I/local/home/psteinfeld/main/wait/llvm/include -isystem /local/home/psteinfeld/main/wait/llvm/../mlir/include -isystem /local/home/psteinfeld/main/wait/build/tools/mlir/include -isystem /local/home/psteinfeld/main/wait/build/tools/clang/include -isystem /local/home/psteinfeld/main/wait/llvm/../clang/include -fPIC -fno-semantic-interposition -fvisibility-inlines-hidden -Werror=date-time -Wall -Wextra -Wno-unused-parameter -Wwrite-strings -Wcast-qual -Wno-missing-field-initializers -pedantic -Wno-long-long -Wimplicit-fallthrough -Wno-maybe-uninitialized -Wno-class-memaccess -Wno-noexcept-type -Wdelete-non-virtual-dtor -Wno-comment -Wmisleading-indentation -fdiagnostics-color -ffunction-sections -fdata-sections -Werror -Wno-deprecated-copy -fno-strict-aliasing -fno-semantic-interposition -O3 -DNDEBUG  -fno-exceptions -fno-rtti -UNDEBUG -std=c++17 -MD -MT tools/flang/lib/Lower/CMakeFiles/obj.FortranLower.dir/OpenACC.cpp.o -MF tools/flang/lib/Lower/CMakeFiles/obj.FortranLower.dir/OpenACC.cpp.o.d -o tools/flang/lib/Lower/CMakeFiles/obj.FortranLower.dir/OpenACC.cpp.o -c /local/home/psteinfeld/main/wait/flang/lib/Lower/OpenACC.cpp
/local/home/psteinfeld/main/wait/flang/lib/Lower/OpenACC.cpp: In function ‘void genACC(Fortran::lower::AbstractConverter&, Fortran::lower::pft::Evaluation&, const Fortran::parser::OpenACCWaitConstruct&)’:
/local/home/psteinfeld/main/wait/flang/lib/Lower/OpenACC.cpp:972:7: error: ‘genIfClause’ was not declared in this scope
       genIfClause(converter, ifClause, ifCond, stmtCtx);
       ^~~~~~~~~~~
/local/home/psteinfeld/main/wait/flang/lib/Lower/OpenACC.cpp:972:7: note: suggested alternative: ‘ifClause’
       genIfClause(converter, ifClause, ifCond, stmtCtx);
       ^~~~~~~~~~~
       ifClause
/local/home/psteinfeld/main/wait/flang/lib/Lower/OpenACC.cpp:975:7: error: ‘genAsyncClause’ was not declared in this scope
       genAsyncClause(converter, asyncClause, async, addAsyncAttr, stmtCtx);
       ^~~~~~~~~~~~~~
/local/home/psteinfeld/main/wait/flang/lib/Lower/OpenACC.cpp:975:7: note: suggested alternative: ‘asyncClause’
       genAsyncClause(converter, asyncClause, async, addAsyncAttr, stmtCtx);
       ^~~~~~~~~~~~~~
       asyncClause
At global scope:
cc1plus: error: unrecognized command line option ‘-Wno-deprecated-copy’ [-Werror]
cc1plus: all warnings being treated as errors

[clementval]

Rebase

[PeteSteinfeld]

All builds and tests correctly and looks good.

flang/lib/Lower/OpenACC.cpp

static void genACC(Fortran::lower::AbstractConverter &converter,
                    const Fortran::parser::OpenACCWaitConstruct &waitConstruct) {
 
   const auto &waitArgument =
       std::get<std::optional<Fortran::parser::AccWaitArgument>>(
           waitConstruct.t);
   const auto &accClauseList =
       std::get<Fortran::parser::AccClauseList>(waitConstruct.t);
 
-  mlir::Value ifCond, waitDevnum, async;
-  SmallVector<mlir::Value, 2> waitOperands;
+  mlir::Value ifCond, asyncOperand, waitDevnum, async;
+  SmallVector<mlir::Value> waitOperands;
 
   // Async clause have optional values but can be present with
   // no value as well. When there is no value, the op has an attribute to
   // represent the clause.
   bool addAsyncAttr = false;
 
-  auto &firOpBuilder = converter.getFirOpBuilder();
-  auto currentLocation = converter.getCurrentLocation();
+  fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+  mlir::Location currentLocation = converter.getCurrentLocation();
   Fortran::lower::StatementContext stmtCtx;
 
   if (waitArgument) { // wait has a value.
     const Fortran::parser::AccWaitArgument &waitArg = *waitArgument;
     const std::list<Fortran::parser::ScalarIntExpr> &waitList =
         std::get<std::list<Fortran::parser::ScalarIntExpr>>(waitArg.t);
     for (const Fortran::parser::ScalarIntExpr &value : waitList) {
       mlir::Value v = fir::getBase(
           converter.genExprValue(*Fortran::semantics::GetExpr(value), stmtCtx));
       waitOperands.push_back(v);
     }
 
     const std::optional<Fortran::parser::ScalarIntExpr> &waitDevnumValue =
         std::get<std::optional<Fortran::parser::ScalarIntExpr>>(waitArg.t);
     if (waitDevnumValue)
       waitDevnum = fir::getBase(converter.genExprValue(
           *Fortran::semantics::GetExpr(*waitDevnumValue), stmtCtx));
   }
 
   // Lower clauses values mapped to operands.
   // Keep track of each group of operands separatly as clauses can appear
   // more than once.
-  for (const auto &clause : accClauseList.v) {
+  for (const Fortran::parser::AccClause &clause : accClauseList.v) {
     if (const auto *ifClause =
             std::get_if<Fortran::parser::AccClause::If>(&clause.u)) {
-      mlir::Value cond = fir::getBase(converter.genExprValue(
-          *Fortran::semantics::GetExpr(ifClause->v), stmtCtx));
-      ifCond = firOpBuilder.createConvert(currentLocation,
-                                          firOpBuilder.getI1Type(), cond);
+      genIfClause(converter, ifClause, ifCond, stmtCtx);
     } else if (const auto *asyncClause =
                    std::get_if<Fortran::parser::AccClause::Async>(&clause.u)) {
-      const auto &asyncClauseValue = asyncClause->v;
-      if (asyncClauseValue) { // async has a value.
-        async = fir::getBase(converter.genExprValue(
-            *Fortran::semantics::GetExpr(*asyncClauseValue), stmtCtx));
-      } else {
-        addAsyncAttr = true;
-      }
+      genAsyncClause(converter, asyncClause, async, addAsyncAttr, stmtCtx);
     }
   }
 
   // Prepare the operand segement size attribute and the operands value range.
-  SmallVector<mlir::Value, 8> operands;
-  SmallVector<int32_t, 4> operandSegments;
+  SmallVector<mlir::Value> operands;
+  SmallVector<int32_t> operandSegments;
   addOperands(operands, operandSegments, waitOperands);
   addOperand(operands, operandSegments, async);
   addOperand(operands, operandSegments, waitDevnum);
   addOperand(operands, operandSegments, ifCond);
 
-  auto waitOp = createSimpleOp<mlir::acc::WaitOp>(firOpBuilder, currentLocation,
-                                                  operands, operandSegments);
+  mlir::acc::WaitOp waitOp = createSimpleOp<mlir::acc::WaitOp>(
+      firOpBuilder, currentLocation, operands, operandSegments);
 
   if (addAsyncAttr)
     waitOp.asyncAttr(firOpBuilder.getUnitAttr());
 }
 
 void Fortran::lower::genOpenACCConstruct(
     Fortran::lower::AbstractConverter &converter,
     Fortran::lower::pft::Evaluation &eval,

flang/test/Lower/OpenACC/acc-wait.f90

+! This test checks lowering of OpenACC wait directive.
+
+! RUN: bbc -fopenacc -emit-fir %s -o - | FileCheck %s
+
+subroutine acc_update
+  integer :: async = 1
+  logical :: ifCondition = .TRUE.
+
+  !$acc wait
+!CHECK: acc.wait{{$}}
+
+  !$acc wait if(.true.)
+!CHECK: [[IF1:%.*]] = arith.constant true
+!CHECK: acc.wait if([[IF1]]){{$}}
+
+  !$acc wait if(ifCondition)
+!CHECK: [[IFCOND:%.*]] = fir.load %{{.*}} : !fir.ref<!fir.logical<4>>
+!CHECK: [[IF2:%.*]] = fir.convert [[IFCOND]] : (!fir.logical<4>) -> i1
+!CHECK: acc.wait if([[IF2]]){{$}}
+
+  !$acc wait(1, 2)
+!CHECK: [[WAIT1:%.*]] = arith.constant 1 : i32
+!CHECK: [[WAIT2:%.*]] = arith.constant 2 : i32
+!CHECK: acc.wait([[WAIT1]], [[WAIT2]] : i32, i32){{$}}
+
+  !$acc wait(1) async
+!CHECK: [[WAIT3:%.*]] = arith.constant 1 : i32
+!CHECK: acc.wait([[WAIT3]] : i32) attributes {async}
+
+  !$acc wait(1) async(async)
+!CHECK: [[WAIT3:%.*]] = arith.constant 1 : i32
+!CHECK: [[ASYNC1:%.*]] = fir.load %{{.*}} : !fir.ref<i32>
+!CHECK: acc.wait([[WAIT3]] : i32) async([[ASYNC1]] : i32){{$}}
+
+  !$acc wait(devnum: 3: queues: 1, 2)
+!CHECK: [[WAIT1:%.*]] = arith.constant 1 : i32
+!CHECK: [[WAIT2:%.*]] = arith.constant 2 : i32
+!CHECK: [[DEVNUM:%.*]] = arith.constant 3 : i32
+!CHECK: acc.wait([[WAIT1]], [[WAIT2]] : i32, i32) wait_devnum([[DEVNUM]] : i32){{$}}
+
+end subroutine acc_update