[MLIR][OpenMP] Lowering support for Order clause in SIMD directive (#96866)

Author: harishch4

mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp

@@ -1460,6 +1460,18 @@ convertOmpParallel(omp::ParallelOp opInst, llvm::IRBuilderBase &builder,
   return bodyGenStatus;
 }
 
+/// Convert Order attribute to llvm::omp::OrderKind.
+static llvm::omp::OrderKind
+convertOrderKind(std::optional<omp::ClauseOrderKind> o) {
+  if (!o)
+    return llvm::omp::OrderKind::OMP_ORDER_unknown;
+  switch (*o) {
+  case omp::ClauseOrderKind::Concurrent:
+    return llvm::omp::OrderKind::OMP_ORDER_concurrent;
+  }
+  llvm_unreachable("Unknown ClauseOrderKind kind");
+}
+
 /// Converts an OpenMP simd loop into LLVM IR using OpenMPIRBuilder.
 static LogicalResult
 convertOmpSimd(Operation &opInst, llvm::IRBuilderBase &builder,
@@ -1539,11 +1551,12 @@ convertOmpSimd(Operation &opInst, llvm::IRBuilderBase &builder,
     safelen = builder.getInt64(safelenVar.value());
 
   llvm::MapVector<llvm::Value *, llvm::Value *> alignedVars;
-  ompBuilder->applySimd(
-      loopInfo, alignedVars,
-      simdOp.getIfExpr() ? moduleTranslation.lookupValue(simdOp.getIfExpr())
-                         : nullptr,
-      llvm::omp::OrderKind::OMP_ORDER_unknown, simdlen, safelen);
+  llvm::omp::OrderKind order = convertOrderKind(simdOp.getOrderVal());
+  ompBuilder->applySimd(loopInfo, alignedVars,
+                        simdOp.getIfExpr()
+                            ? moduleTranslation.lookupValue(simdOp.getIfExpr())
+                            : nullptr,
+                        order, simdlen, safelen);
 
   builder.restoreIP(afterIP);
   return success();

mlir/test/Target/LLVMIR/openmp-llvm.mlir

@@ -826,6 +826,26 @@ llvm.func @simd_if(%arg0: !llvm.ptr {fir.bindc_name = "n"}, %arg1: !llvm.ptr {fi
 
 // -----
 
+// CHECK-LABEL: @simd_order
+llvm.func @simd_order() {
+  %0 = llvm.mlir.constant(10 : i64) : i64
+  %1 = llvm.mlir.constant(1 : i64) : i64
+  %2 = llvm.alloca %1 x i64 : (i64) -> !llvm.ptr
+  omp.simd order(concurrent) safelen(2) {
+    omp.loop_nest (%arg0) : i64 = (%1) to (%0) inclusive step (%1) {
+      llvm.store %arg0, %2 : i64, !llvm.ptr
+      omp.yield
+    }
+  }
+  llvm.return
+}
+// If clause order(concurrent) is specified then the memory instructions
+// are marked parallel even if 'safelen' is finite.
+// CHECK: llvm.loop.parallel_accesses
+// CHECK-NEXT: llvm.loop.vectorize.enable
+// CHECK-NEXT: llvm.loop.vectorize.width{{.*}}i64 2
+// -----
+
 llvm.func @body(i64)
 
 llvm.func @test_omp_wsloop_ordered(%lb : i64, %ub : i64, %step : i64) -> () {