MLIR][LLVMIR] Adding scan lowering to llvm on the mlir side

Author: anchuraj

flang/lib/Lower/OpenMP/OpenMP.cpp

@@ -2326,12 +2326,50 @@ genParallelOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
 
 static mlir::omp::ScanOp
 genScanOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
-          semantics::SemanticsContext &semaCtx, mlir::Location loc,
-          const ConstructQueue &queue, ConstructQueue::const_iterator item) {
+          semantics::SemanticsContext &semaCtx, lower::pft::Evaluation &eval,
+          mlir::Location loc, const ConstructQueue &queue,
+          ConstructQueue::const_iterator item) {
   mlir::omp::ScanOperands clauseOps;
   genScanClauses(converter, semaCtx, item->clauses, loc, clauseOps);
-  return mlir::omp::ScanOp::create(converter.getFirOpBuilder(),
-                                   converter.getCurrentLocation(), clauseOps);
+  mlir::omp::ScanOp scanOp = mlir::omp::ScanOp::create(
+      converter.getFirOpBuilder(), converter.getCurrentLocation(), clauseOps);
+  
+  /// Scan redution is not implemented with nested workshare loops, linear clause, tiling
+  mlir::omp::LoopNestOp loopNestOp =
+      scanOp->getParentOfType<mlir::omp::LoopNestOp>();
+  mlir::omp::WsloopOp wsLoopOp =
+      scanOp->getParentOfType<mlir::omp::WsloopOp>();
+  bool isNested = (loopNestOp.getNumLoops() > 1) || (wsLoopOp && (wsLoopOp->getParentOfType<mlir::omp::WsloopOp>()));
+  if(isNested)
+    TODO(loc, "Scan directive inside nested workshare loops");
+  if(wsLoopOp && !wsLoopOp.getLinearVars().empty())
+    TODO(loc, "Scan directive with linear clause");
+  if(loopNestOp.getTileSizes())
+    TODO(loc, "Scan directive with loop tiling");
+  
+  // All loop indices should be loaded after the scan construct as otherwise,
+  // it would result in using the index variable across scan directive.
+  // (`Intra-iteration dependences from a statement in the structured
+  // block sequence that precede a scan directive to a statement in the
+  // structured block sequence that follows a scan directive must not exist,
+  // except for dependences for the list items specified in an inclusive or
+  // exclusive clause.`).
+  // TODO: Nested loops are not handled.
+  mlir::Region &region = loopNestOp->getRegion(0);
+  mlir::Value indexVal = fir::getBase(region.getArgument(0));
+  lower::pft::Evaluation *doConstructEval = eval.parentConstruct;
+  fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+  lower::pft::Evaluation *doLoop = &doConstructEval->getFirstNestedEvaluation();
+  auto *doStmt = doLoop->getIf<parser::NonLabelDoStmt>();
+  assert(doStmt && "Expected do loop to be in the nested evaluation");
+  const auto &loopControl =
+      std::get<std::optional<parser::LoopControl>>(doStmt->t);
+  const parser::LoopControl::Bounds *bounds =
+      std::get_if<parser::LoopControl::Bounds>(&loopControl->u);
+  mlir::Operation *storeOp =
+      setLoopVar(converter, loc, indexVal, bounds->name.thing.symbol);
+  firOpBuilder.setInsertionPointAfter(storeOp);
+  return scanOp;
 }
 
 static mlir::omp::SectionsOp
@@ -3416,7 +3454,7 @@ static void genOMPDispatch(lower::AbstractConverter &converter,
                                   loc, queue, item);
     break;
   case llvm::omp::Directive::OMPD_scan:
-    newOp = genScanOp(converter, symTable, semaCtx, loc, queue, item);
+    newOp = genScanOp(converter, symTable, semaCtx, eval, loc, queue, item);
     break;
   case llvm::omp::Directive::OMPD_section:
     llvm_unreachable("genOMPDispatch: OMPD_section");

flang/test/Lower/OpenMP/Todo/nested-wsloop-scan.f90

@@ -0,0 +1,34 @@
+! Tests scan reduction behavior when used in nested workshare loops
+
+! RUN: %not_todo_cmd %flang_fc1 -emit-hlfir -fopenmp -o - %s 2>&1 | FileCheck %s
+
+program nested_scan_example
+  implicit none
+  integer, parameter :: n = 4, m = 5
+  integer :: a(n, m), b(n, m)
+  integer :: i, j
+  integer :: row_sum, col_sum
+
+  do i = 1, n
+     do j = 1, m
+        a(i, j) = i + j
+     end do
+  end do
+
+  !$omp parallel do reduction(inscan, +: row_sum) private(col_sum, j)
+  do i = 1, n
+     row_sum = row_sum + i
+     !$omp scan inclusive(row_sum)
+
+     col_sum = 0
+     !$omp parallel do reduction(inscan, +: col_sum)
+     do j = 1, m
+        col_sum = col_sum + a(i, j)
+        !CHECK: not yet implemented: Scan directive inside nested workshare loops
+        !$omp scan inclusive(col_sum)
+        b(i, j) = col_sum + row_sum 
+     end do
+     !$omp end parallel do
+  end do
+  !$omp end parallel do
+end program nested_scan_example

flang/test/Lower/OpenMP/Todo/wsloop-scan-collapse.f90

@@ -0,0 +1,29 @@
+! Tests scan reduction behavior when used in nested workshare loops
+
+! RUN: %not_todo_cmd %flang_fc1 -emit-hlfir -fopenmp -o - %s 2>&1 | FileCheck %s
+
+program nested_loop_example
+  implicit none
+  integer :: i, j, x
+  integer, parameter :: N = 100, M = 200
+  real :: A(N, M), B(N, M)
+  x = 0
+
+  do i = 1, N
+    do j = 1, M
+      A(i, j) = i * j
+    end do
+  end do
+  
+  !$omp parallel do collapse(2) reduction(inscan, +:x)
+  do i = 1, N
+    do j = 1, M
+      x = x + A(i,j)
+      !CHECK: not yet implemented: Scan directive inside nested workshare loops
+      !$omp scan inclusive(x)
+      B(i, j) = x
+    end do
+  end do
+  !$omp end parallel do
+
+end program nested_loop_example