[flang][openacc] Generate loop nest as column major

[clementval]

Summary

Address comment from D153455

[jeanPerier]

Thanks, LGTM

flang/lib/Lower/OpenACC.cpp

if (auto refTy = mlir::dyn_cast<fir::ReferenceType>(ty)) {
     if (auto seqTy = mlir::dyn_cast<fir::SequenceType>(refTy.getEleTy())) {
       if (seqTy.hasDynamicExtents())
         TODO(loc, "OpenACC reduction on array with dynamic extents");
       mlir::Type idxTy = builder.getIndexType();
       mlir::Type refTy = fir::ReferenceType::get(seqTy.getEleTy());
 
       llvm::SmallVector<fir::DoLoopOp> loops;
       llvm::SmallVector<mlir::Value> ivs;
-      for (auto ext : seqTy.getShape()) {
+      for (auto ext : llvm::reverse(seqTy.getShape())) {
         auto lb = builder.create<mlir::arith::ConstantOp>(
             loc, idxTy, builder.getIntegerAttr(idxTy, 0));
         auto ub = builder.create<mlir::arith::ConstantOp>(
             loc, idxTy, builder.getIntegerAttr(idxTy, ext - 1));
         auto step = builder.create<mlir::arith::ConstantOp>(
             loc, idxTy, builder.getIntegerAttr(idxTy, 1));
         auto loop = builder.create<fir::DoLoopOp>(loc, lb, ub, step,
                                                   /*unordered=*/false);

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

 
 ! CHECK-LABEL: acc.reduction.recipe @reduction_add_ref_100x10x2xi32 : !fir.ref<!fir.array<100x10x2xi32>> reduction_operator <add> init {
 ! CHECK: ^bb0(%{{.*}}: !fir.ref<!fir.array<100x10x2xi32>>):
 ! CHECK:   %[[CST:.*]] = arith.constant dense<0> : vector<100x10x2xi32>
 ! CHECK:   acc.yield %[[CST]] : vector<100x10x2xi32>
 ! CHECK: } combiner {
 ! CHECK: ^bb0(%[[ARG0:.*]]: !fir.ref<!fir.array<100x10x2xi32>>, %[[ARG1:.*]]: !fir.ref<!fir.array<100x10x2xi32>>):
 ! CHECK:   %[[LB0:.*]] = arith.constant 0 : index
-! CHECK:   %[[UB0:.*]] = arith.constant 99 : index
+! CHECK:   %[[UB0:.*]] = arith.constant 1 : index
 ! CHECK:   %[[STEP0:.*]] = arith.constant 1 : index
 ! CHECK:   fir.do_loop %[[IV0:.*]] = %[[LB0]] to %[[UB0]] step %[[STEP0]] {
 ! CHECK:     %[[LB1:.*]] = arith.constant 0 : index
 ! CHECK:     %[[UB1:.*]] = arith.constant 9 : index
 ! CHECK:     %[[STEP1:.*]] = arith.constant 1 : index
 ! CHECK:     fir.do_loop %[[IV1:.*]] = %[[LB1]] to %[[UB1]] step %[[STEP1]] {
 ! CHECK:       %[[LB2:.*]] = arith.constant 0 : index
-! CHECK:       %[[UB2:.*]] = arith.constant 1 : index
+! CHECK:       %[[UB2:.*]] = arith.constant 99 : index
 ! CHECK:       %[[STEP2:.*]] = arith.constant 1 : index
 ! CHECK:       fir.do_loop %[[IV2:.*]] = %[[LB2]] to %[[UB2]] step %[[STEP2]] {
 ! CHECK:         %[[COORD1:.*]] = fir.coordinate_of %[[ARG0]], %[[IV0]], %[[IV1]], %[[IV2]] : (!fir.ref<!fir.array<100x10x2xi32>>, index, index, index) -> !fir.ref<i32>
 ! CHECK:         %[[COORD2:.*]] = fir.coordinate_of %[[ARG1]], %[[IV0]], %[[IV1]], %[[IV2]] : (!fir.ref<!fir.array<100x10x2xi32>>, index, index, index) -> !fir.ref<i32>
 ! CHECK:         %[[LOAD1:.*]] = fir.load %[[COORD1]] : !fir.ref<i32>
 ! CHECK:         %[[LOAD2:.*]] = fir.load %[[COORD2]] : !fir.ref<i32>
 ! CHECK:         %[[COMBINED:.*]] = arith.addi %[[LOAD1]], %[[LOAD2]] : i32
 ! CHECK:         fir.store %[[COMBINED]] to %[[COORD1]] : !fir.ref<i32>
 ! CHECK:       }
 ! CHECK:     }
 ! CHECK:   }
 ! CHECK:   acc.yield %[[ARG0]] : !fir.ref<!fir.array<100x10x2xi32>>
 ! CHECK: }
 
 ! CHECK-LABEL: acc.reduction.recipe @reduction_add_ref_100x10xi32 : !fir.ref<!fir.array<100x10xi32>> reduction_operator <add> init {
 ! CHECK: ^bb0(%{{.*}}: !fir.ref<!fir.array<100x10xi32>>):
 ! CHECK: %[[CST:.*]] = arith.constant dense<0> : vector<100x10xi32>
 ! CHECK: acc.yield %[[CST]] : vector<100x10xi32>
 ! CHECK: } combiner {
 ! CHECK: ^bb0(%[[ARG0:.*]]: !fir.ref<!fir.array<100x10xi32>>, %[[ARG1:.*]]: !fir.ref<!fir.array<100x10xi32>>):
 ! CHECK:   %[[LB0:.*]] = arith.constant 0 : index
-! CHECK:   %[[UB0:.*]] = arith.constant 99 : index
+! CHECK:   %[[UB0:.*]] = arith.constant 9 : index
 ! CHECK:   %[[STEP0:.*]] = arith.constant 1 : index
 ! CHECK:   fir.do_loop %[[IV0:.*]] = %[[LB0]] to %[[UB0]] step %[[STEP0]] {
 ! CHECK:     %[[LB1:.*]] = arith.constant 0 : index
-! CHECK:     %[[UB1:.*]] = arith.constant 9 : index
+! CHECK:     %[[UB1:.*]] = arith.constant 99 : index
 ! CHECK:     %[[STEP1:.*]] = arith.constant 1 : index
 ! CHECK:     fir.do_loop %[[IV1:.*]] = %[[LB1]] to %[[UB1]] step %[[STEP1]] {
 ! CHECK:       %[[COORD1:.*]] = fir.coordinate_of %[[ARG0]], %[[IV0]], %[[IV1]] : (!fir.ref<!fir.array<100x10xi32>>, index, index) -> !fir.ref<i32>
 ! CHECK:       %[[COORD2:.*]] = fir.coordinate_of %[[ARG1]], %[[IV0]], %[[IV1]] : (!fir.ref<!fir.array<100x10xi32>>, index, index) -> !fir.ref<i32>
 ! CHECK:       %[[LOAD1]] = fir.load %[[COORD1]] : !fir.ref<i32>
 ! CHECK:       %[[LOAD2]] = fir.load %[[COORD2]] : !fir.ref<i32>
 ! CHECK:       %[[COMBINED:.*]] = arith.addi %[[LOAD1]], %[[LOAD2]] : i32
 ! CHECK:       fir.store %[[COMBINED]] to %[[COORD1]] : !fir.ref<i32>