[Flang][OpenMP] Lowering nontemporal clause to MLIR for SIMD directive

[harishch4]

Currently, Flang throws a "not yet implemented: Unhandled clause NONTEMPORAL in SIMD construct" error when encountering nontemporal clause. This patch adds support for this clause in SIMD construct.

[llvmbot]

@llvm/pr-subscribers-flang-openmp

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

Author: None (harishch4)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/108339.diff

4 Files Affected:

• (modified) flang/lib/Lower/OpenMP/ClauseProcessor.cpp (+12)
• (modified) flang/lib/Lower/OpenMP/ClauseProcessor.h (+1)
• (modified) flang/lib/Lower/OpenMP/OpenMP.cpp (+2-2)
• (modified) flang/test/Lower/OpenMP/simd.f90 (+18)

diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index f336d213cc8620..fa8a4309700f45 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -1009,6 +1009,18 @@ bool ClauseProcessor::processMap(
   return clauseFound;
 }
 
+bool ClauseProcessor::processNontemporal(
+    mlir::omp::NontemporalClauseOps &result) const {
+  return findRepeatableClause<omp::clause::Nontemporal>(
+      [&](const omp::clause::Nontemporal &clause, const parser::CharBlock &) {
+        for (const Object &object : clause.v) {
+          semantics::Symbol *sym = object.sym();
+          mlir::Value symVal = converter.getSymbolAddress(*sym);
+          result.nontemporalVars.push_back(symVal);
+        }
+      });
+}
+
 bool ClauseProcessor::processReduction(
     mlir::Location currentLocation, mlir::omp::ReductionClauseOps &result,
     llvm::SmallVectorImpl<mlir::Type> *outReductionTypes,
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.h b/flang/lib/Lower/OpenMP/ClauseProcessor.h
index 8d02d368f4ee04..be1d8a66ddfccd 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.h
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.h
@@ -121,6 +121,7 @@ class ClauseProcessor {
       llvm::SmallVectorImpl<const semantics::Symbol *> *mapSyms = nullptr,
       llvm::SmallVectorImpl<mlir::Location> *mapSymLocs = nullptr,
       llvm::SmallVectorImpl<mlir::Type> *mapSymTypes = nullptr) const;
+  bool processNontemporal(mlir::omp::NontemporalClauseOps &result) const;
   bool processReduction(
       mlir::Location currentLocation, mlir::omp::ReductionClauseOps &result,
       llvm::SmallVectorImpl<mlir::Type> *reductionTypes = nullptr,
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 233aacb40354d4..99114dc38a249c 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -1123,13 +1123,13 @@ static void genSimdClauses(lower::AbstractConverter &converter,
   ClauseProcessor cp(converter, semaCtx, clauses);
   cp.processAligned(clauseOps);
   cp.processIf(llvm::omp::Directive::OMPD_simd, clauseOps);
+  cp.processNontemporal(clauseOps);
   cp.processOrder(clauseOps);
   cp.processReduction(loc, clauseOps);
   cp.processSafelen(clauseOps);
   cp.processSimdlen(clauseOps);
 
-  cp.processTODO<clause::Linear, clause::Nontemporal>(
-      loc, llvm::omp::Directive::OMPD_simd);
+  cp.processTODO<clause::Linear>(loc, llvm::omp::Directive::OMPD_simd);
 }
 
 static void genSingleClauses(lower::AbstractConverter &converter,
diff --git a/flang/test/Lower/OpenMP/simd.f90 b/flang/test/Lower/OpenMP/simd.f90
index 2127451878849b..bdc6a1eef01565 100644
--- a/flang/test/Lower/OpenMP/simd.f90
+++ b/flang/test/Lower/OpenMP/simd.f90
@@ -223,3 +223,21 @@ subroutine simdloop_aligned_allocatable()
     A(i) = i
   end do
 end subroutine
+
+!CHECK-LABEL: func @_QPsimd_with_nontemporal_clause
+subroutine simd_with_nontemporal_clause(n)
+  !CHECK: %[[A_DECL:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "_QFsimd_with_nontemporal_clauseEa"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+  !CHECK: %[[C_DECL:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "_QFsimd_with_nontemporal_clauseEc"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+  integer :: i, n
+  integer :: A, B, C
+  !CHECK: %[[LB:.*]] = arith.constant 1 : i32
+  !CHECK: %[[UB:.*]] = fir.load %{{.*}}#0 : !fir.ref<i32>
+  !CHECK: %[[STEP:.*]] = arith.constant 1 : i32
+  !CHECK: omp.simd nontemporal(%[[A_DECL]]#1, %[[C_DECL]]#1 : !fir.ref<i32>, !fir.ref<i32>) {
+  !CHECK-NEXT: omp.loop_nest (%[[I:.*]]) : i32 = (%[[LB]]) to (%[[UB]]) inclusive step (%[[STEP]]) {
+  !$OMP SIMD NONTEMPORAL(A, C)
+  do i = 1, n
+    C = A + B
+  end do
+  !$OMP END SIMD
+end subroutine

[skatrak]

This LGTM, thank you!

[NimishMishra]

I am okay with the change, but it would be good to add a description for this PR before merging. LGTM otherwise.