[flang][OpenMP] Enable lastprivate on simd

[NimishMishra]

This PR enables the lowering of lastprivate when defined on simd construct

[llvmbot]

@llvm/pr-subscribers-flang-openmp

@llvm/pr-subscribers-flang-fir-hlfir

Author: None (NimishMishra)

Changes

This PR enables the lowering of lastprivate when defined on simd construct

---

Full diff: https://github.com/llvm/llvm-project/pull/93786.diff

2 Files Affected:

• (modified) flang/lib/Lower/OpenMP/DataSharingProcessor.cpp (+3-3)
• (modified) flang/test/Lower/OpenMP/simd.f90 (+27)

diff --git a/flang/lib/Lower/OpenMP/DataSharingProcessor.cpp b/flang/lib/Lower/OpenMP/DataSharingProcessor.cpp
index 557a9685024c5..ab5f2b8ac2429 100644
--- a/flang/lib/Lower/OpenMP/DataSharingProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/DataSharingProcessor.cpp
@@ -199,12 +199,12 @@ void DataSharingProcessor::insertLastPrivateCompare(mlir::Operation *op) {
   for (const omp::Clause &clause : clauses) {
     if (clause.id != llvm::omp::OMPC_lastprivate)
       continue;
-    // TODO: Add lastprivate support for simd construct
-    if (mlir::isa<mlir::omp::WsloopOp>(op)) {
+    if (mlir::isa<mlir::omp::WsloopOp>(op) or
+        mlir::isa<mlir::omp::SimdOp>(op)) {
       // Update the original variable just before exiting the worksharing
       // loop. Conversion as follows:
       //
-      // omp.wsloop {             omp.wsloop {
+      // omp.wsloop / omp.simd {    omp.wsloop / omp.simd {
       //   omp.loop_nest {          omp.loop_nest {
       //     ...                      ...
       //     store          ===>      store
diff --git a/flang/test/Lower/OpenMP/simd.f90 b/flang/test/Lower/OpenMP/simd.f90
index 223b248b79348..2f9720d776490 100644
--- a/flang/test/Lower/OpenMP/simd.f90
+++ b/flang/test/Lower/OpenMP/simd.f90
@@ -182,3 +182,30 @@ subroutine simd_with_collapse_clause(n)
   end do
   !$OMP END SIMD
 end subroutine
+
+!CHECK: func.func @_QPlastprivate_with_simd() {
+subroutine lastprivate_with_simd
+
+!CHECK: %[[VAR_SUM:.*]] = fir.alloca f32 {bindc_name = "sum", uniq_name = "_QFlastprivate_with_simdEsum"}
+!CHECK: %[[VAR_SUM_DECLARE:.*]]:2 = hlfir.declare %[[VAR_SUM]] {{.*}}
+!CHECK: %[[VAR_SUM_PINNED:.*]] = fir.alloca f32 {bindc_name = "sum", pinned, uniq_name = "_QFlastprivate_with_simdEsum"}
+!CHECK: %[[VAR_SUM_PINNED_DECLARE:.*]]:2 = hlfir.declare %[[VAR_SUM_PINNED]] {{.*}}
+  implicit none
+  integer :: i
+  real :: sum
+
+!CHECK: omp.simd {
+!CHECK: omp.loop_nest (%[[ARG:.*]]) : i32 = ({{.*}} to ({{.*}}) inclusive step ({{.*}}) {
+!CHECK: %[[ADD_RESULT:.*]] = arith.addi {{.*}}
+!CHECK: %[[ADD_RESULT_CONVERT:.*]] = fir.convert %[[ADD_RESULT]] : (i32) -> f32
+!CHECK: hlfir.assign %[[ADD_RESULT_CONVERT]] to %[[VAR_SUM_PINNED_DECLARE]]#0 : f32, !fir.ref<f32> 
+  !$omp simd lastprivate(sum)
+  do i = 1, 100
+     sum = i + 1
+  end do
+!CHECK: %[[SELECT_RESULT:.*]] = arith.select {{.*}}, {{.*}}, {{.*}} : i1
+!CHECK: fir.if %[[SELECT_RESULT]] {
+!CHECK: %[[LOADED_SUM:.*]] = fir.load %[[VAR_SUM_PINNED_DECLARE]]#0 : !fir.ref<f32>
+!CHECK: hlfir.assign %[[LOADED_SUM]] to %[[VAR_SUM_DECLARE]]#0 temporary_lhs : f32, !fir.ref<f32>
+!CHECK: }
+end subroutine

[tblah]

LGTM thanks

[luporl]

Suggested change

	if (mlir::isa<mlir::omp::WsloopOp>(op) or
	if (mlir::isa<mlir::omp::WsloopOp>(op) ||

[ergawy]

Just 2 nits. LG, thanks!

[ergawy]

nit: to separate this from the previous blocks of checks.

Suggested change

	!CHECK: func.func @_QPlastprivate_with_simd() {
	!CHECK-LABEL: func.func @_QPlastprivate_with_simd() {

[ergawy]

nit: can you add the omp ops terminators so that we have balances braces and also it is more visible how the structure of the omp region looks like in relation to the lastprivate operations.