[Mlir-commits] [llvm] [mlir] [MLIR][OpenMP] Lowering nontemporal clause to LLVM IR for SIMD directive (PR #118751)

Kaviya Rajendiran llvmlistbot at llvm.org
Tue Mar 4 22:51:05 PST 2025


https://github.com/kaviya2510 updated https://github.com/llvm/llvm-project/pull/118751

>From a7a485940e3582e5375a39599a8676d2fabb3388 Mon Sep 17 00:00:00 2001
From: Kaviya Rajendiran <kaviyara2000 at gmail.com>
Date: Fri, 6 Dec 2024 14:49:39 +0530
Subject: [PATCH 1/2] [MLIR][OpenMP] Lowering nontemporal clause to LLVM IR for
 SIMD directive

---
 .../llvm/Frontend/OpenMP/OMPIRBuilder.h       |  2 +-
 llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp     | 85 +++++++++++++++-
 .../OpenMP/OpenMPToLLVMIRTranslation.cpp      | 15 +--
 .../Target/LLVMIR/openmp-nontemporal.mlir     | 96 +++++++++++++++++++
 mlir/test/Target/LLVMIR/openmp-todo.mlir      | 13 ---
 5 files changed, 190 insertions(+), 21 deletions(-)
 create mode 100644 mlir/test/Target/LLVMIR/openmp-nontemporal.mlir

diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
index 80b4aa2bd2855..fc726ec6cf4b4 100644
--- a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
+++ b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
@@ -1245,7 +1245,7 @@ class OpenMPIRBuilder {
   void applySimd(CanonicalLoopInfo *Loop,
                  MapVector<Value *, Value *> AlignedVars, Value *IfCond,
                  omp::OrderKind Order, ConstantInt *Simdlen,
-                 ConstantInt *Safelen);
+                 ConstantInt *Safelen, ArrayRef<Value *> NontempralVars = {});
 
   /// Generator for '#omp flush'
   ///
diff --git a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
index e34e93442ff85..bff201a1377c8 100644
--- a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
+++ b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
@@ -5385,10 +5385,86 @@ OpenMPIRBuilder::getOpenMPDefaultSimdAlign(const Triple &TargetTriple,
   return 0;
 }
 
+static void appendNontemporalVars(BasicBlock *Block,
+                                  SmallVectorImpl<Value *> &NontemporalVars) {
+  for (Instruction &I : *Block) {
+    if (const CallInst *CI = dyn_cast<CallInst>(&I)) {
+      if (CI->getIntrinsicID() == Intrinsic::memcpy) {
+        llvm::Value *DestPtr = CI->getArgOperand(0);
+        llvm::Value *SrcPtr = CI->getArgOperand(1);
+        for (const llvm::Value *Var : NontemporalVars) {
+          if (Var == SrcPtr) {
+            NontemporalVars.push_back(DestPtr);
+            break;
+          }
+        }
+      }
+    }
+  }
+}
+
+/** Attach nontemporal metadata to the load/store instructions of nontemporal
+ *  variables of \p Block
+ * Nontemporal variables may be a scalar, fixed size or allocatable
+ * or pointer array
+ *
+ * Example scenarios for nontemporal variables:
+ * Case 1: Scalar variable
+ *    If the nontemporal variable is a scalar, it is allocated on stack.Load and
+ * store instructions directly access the alloca pointer of the scalar
+ * variable for fetching information about scalar variable or  writing
+ * into the scalar variable. Mark those load and store instructions as
+ * non-temporal.
+ *
+ * Case 2: Fixed Size array
+ *    If the nontemporal variable is a fixed-size array, it is allocated
+ * as a contiguous block of memory. It uses one GEP instruction, to compute the
+ * address of each individual array elements and perform load or store
+ * operation on it. Mark those load and store instructions as non-temporal.
+ *
+ * Case 3: Allocatable array
+ *     For an allocatable array, which might involve runtime type descriptor,
+ * needs to navigate through descriptors using two or more GEP and load
+ * instructions to compute the address of each individual element in an array.
+ * Mark those load or store which access the individual array elements as
+ * non-temporal.
+ */
+static void addNonTemporalMetadata(BasicBlock *Block, MDNode *Nontemporal,
+                                   SmallVectorImpl<Value *> &NontemporalVars) {
+  appendNontemporalVars(Block, NontemporalVars);
+  for (Instruction &I : *Block) {
+    llvm::Value *mem_ptr = nullptr;
+    bool MetadataFlag = true;
+    if (llvm::LoadInst *li = dyn_cast<llvm::LoadInst>(&I)) {
+      if (!(li->getType()->isPointerTy()))
+        mem_ptr = li->getPointerOperand();
+    } else if (llvm::StoreInst *si = dyn_cast<llvm::StoreInst>(&I))
+      mem_ptr = si->getPointerOperand();
+    if (mem_ptr) {
+      while (mem_ptr && !(isa<llvm::AllocaInst>(mem_ptr))) {
+        if (llvm::GetElementPtrInst *gep =
+                dyn_cast<llvm::GetElementPtrInst>(mem_ptr)) {
+          llvm::Type *sourceType = gep->getSourceElementType();
+          if (sourceType->isStructTy() && gep->getNumIndices() >= 2 &&
+              !(gep->hasAllZeroIndices())) {
+            MetadataFlag = false;
+            break;
+          }
+          mem_ptr = gep->getPointerOperand();
+        } else if (llvm::LoadInst *li = dyn_cast<llvm::LoadInst>(mem_ptr))
+          mem_ptr = li->getPointerOperand();
+      }
+      if (MetadataFlag && is_contained(NontemporalVars, mem_ptr))
+        I.setMetadata(LLVMContext::MD_nontemporal, Nontemporal);
+    }
+  }
+}
+
 void OpenMPIRBuilder::applySimd(CanonicalLoopInfo *CanonicalLoop,
                                 MapVector<Value *, Value *> AlignedVars,
                                 Value *IfCond, OrderKind Order,
-                                ConstantInt *Simdlen, ConstantInt *Safelen) {
+                                ConstantInt *Simdlen, ConstantInt *Safelen,
+                                ArrayRef<Value *> NontemporalVarsIn) {
   LLVMContext &Ctx = Builder.getContext();
 
   Function *F = CanonicalLoop->getFunction();
@@ -5486,6 +5562,13 @@ void OpenMPIRBuilder::applySimd(CanonicalLoopInfo *CanonicalLoop,
   }
 
   addLoopMetadata(CanonicalLoop, LoopMDList);
+  SmallVector<Value *> NontemporalVars{NontemporalVarsIn};
+  // Set nontemporal metadata to load and stores of nontemporal values
+  if (NontemporalVars.size()) {
+    MDNode *NontemporalNode = MDNode::getDistinct(Ctx, {});
+    for (BasicBlock *BB : Reachable)
+      addNonTemporalMetadata(BB, NontemporalNode, NontemporalVars);
+  }
 }
 
 /// Create the TargetMachine object to query the backend for optimization
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index 32c7c501d03c3..1c9690a1c7b68 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -189,10 +189,6 @@ static LogicalResult checkImplementationStatus(Operation &op) {
     if (!op.getLinearVars().empty() || !op.getLinearStepVars().empty())
       result = todo("linear");
   };
-  auto checkNontemporal = [&todo](auto op, LogicalResult &result) {
-    if (!op.getNontemporalVars().empty())
-      result = todo("nontemporal");
-  };
   auto checkNowait = [&todo](auto op, LogicalResult &result) {
     if (op.getNowait())
       result = todo("nowait");
@@ -300,7 +296,6 @@ static LogicalResult checkImplementationStatus(Operation &op) {
       })
       .Case([&](omp::SimdOp op) {
         checkLinear(op, result);
-        checkNontemporal(op, result);
         checkReduction(op, result);
       })
       .Case<omp::AtomicReadOp, omp::AtomicWriteOp, omp::AtomicUpdateOp,
@@ -2527,6 +2522,14 @@ convertOmpSimd(Operation &opInst, llvm::IRBuilderBase &builder,
 
   llvm::MapVector<llvm::Value *, llvm::Value *> alignedVars;
   llvm::omp::OrderKind order = convertOrderKind(simdOp.getOrder());
+
+  llvm::SmallVector<llvm::Value *> nontemporalVars;
+  mlir::OperandRange nontemporals = simdOp.getNontemporalVars();
+  for (mlir::Value nontemporal : nontemporals) {
+    llvm::Value *nt = moduleTranslation.lookupValue(nontemporal);
+    nontemporalVars.push_back(nt);
+  }
+
   llvm::BasicBlock *sourceBlock = builder.GetInsertBlock();
   std::optional<ArrayAttr> alignmentValues = simdOp.getAlignments();
   mlir::OperandRange operands = simdOp.getAlignedVars();
@@ -2558,7 +2561,7 @@ convertOmpSimd(Operation &opInst, llvm::IRBuilderBase &builder,
                         simdOp.getIfExpr()
                             ? moduleTranslation.lookupValue(simdOp.getIfExpr())
                             : nullptr,
-                        order, simdlen, safelen);
+                        order, simdlen, safelen, nontemporalVars);
 
   return cleanupPrivateVars(builder, moduleTranslation, simdOp.getLoc(),
                             llvmPrivateVars, privateDecls);
diff --git a/mlir/test/Target/LLVMIR/openmp-nontemporal.mlir b/mlir/test/Target/LLVMIR/openmp-nontemporal.mlir
new file mode 100644
index 0000000000000..f8cee94be4ff7
--- /dev/null
+++ b/mlir/test/Target/LLVMIR/openmp-nontemporal.mlir
@@ -0,0 +1,96 @@
+// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s
+
+// -----
+// CHECK-LABEL: @simd_nontemporal
+llvm.func @simd_nontemporal() {
+  %0 = llvm.mlir.constant(10 : i64) : i64
+  %1 = llvm.mlir.constant(1 : i64) : i64
+  %2 = llvm.alloca %1 x i64 : (i64) -> !llvm.ptr
+  %3 = llvm.alloca %1 x i64 : (i64) -> !llvm.ptr
+  //CHECK: %[[A_ADDR:.*]] = alloca i64, i64 1, align 8
+  //CHECK: %[[B_ADDR:.*]] = alloca i64, i64 1, align 8
+  //CHECK: %[[B:.*]] = load i64, ptr %[[B_ADDR]], align 4, !nontemporal !1, !llvm.access.group !2
+  //CHECK: store i64 %[[B]], ptr %[[A_ADDR]], align 4, !nontemporal !1, !llvm.access.group !2
+  omp.simd nontemporal(%2, %3 : !llvm.ptr, !llvm.ptr) {
+    omp.loop_nest (%arg0) : i64 = (%1) to (%0) inclusive step (%1) {
+      %4 = llvm.load %3 : !llvm.ptr -> i64
+      llvm.store %4, %2 : i64, !llvm.ptr
+      omp.yield
+    }
+  }
+  llvm.return
+}
+
+// -----
+
+//CHECK-LABEL:  define void @_QPtest(ptr %0, ptr %1) {
+llvm.func @_QPtest(%arg0: !llvm.ptr {fir.bindc_name = "n"}, %arg1: !llvm.ptr {fir.bindc_name = "a"}) {
+    %0 = llvm.mlir.constant(1 : i32) : i32
+    %1 = llvm.alloca %0 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> {alignment = 8 : i64} : (i32) -> !llvm.ptr
+    %2 = llvm.alloca %0 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> {alignment = 8 : i64} : (i32) -> !llvm.ptr
+    %3 = llvm.mlir.constant(1 : i64) : i64
+    %4 = llvm.alloca %3 x i32 {bindc_name = "i", pinned} : (i64) -> !llvm.ptr
+    %6 = llvm.load %arg0 : !llvm.ptr -> i32
+    //CHECK:  %[[A_VAL1:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, align 8
+    //CHECK:  %[[A_VAL2:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, align 8
+    omp.simd nontemporal(%arg1 : !llvm.ptr) {
+      omp.loop_nest (%arg2) : i32 = (%0) to (%6) inclusive step (%0) {
+        llvm.store %arg2, %4 : i32, !llvm.ptr
+        //CHECK:  call void @llvm.memcpy.p0.p0.i32(ptr %[[A_VAL2]], ptr %1, i32 48, i1 false)
+        %7 = llvm.mlir.constant(48 : i32) : i32
+        "llvm.intr.memcpy"(%2, %arg1, %7) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i32) -> ()
+        %8 = llvm.load %4 : !llvm.ptr -> i32
+        %9 = llvm.sext %8 : i32 to i64
+        %10 = llvm.getelementptr %2[0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)>
+        %11 = llvm.load %10 : !llvm.ptr -> !llvm.ptr
+        %12 = llvm.mlir.constant(0 : index) : i64
+        %13 = llvm.getelementptr %2[0, 7, %12, 0] : (!llvm.ptr, i64) -> !llvm.ptr, !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)>
+        %14 = llvm.load %13 : !llvm.ptr -> i64
+        %15 = llvm.getelementptr %2[0, 7, %12, 1] : (!llvm.ptr, i64) -> !llvm.ptr, !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)>
+        %16 = llvm.load %15 : !llvm.ptr -> i64
+        %17 = llvm.getelementptr %2[0, 7, %12, 2] : (!llvm.ptr, i64) -> !llvm.ptr, !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)>
+        %18 = llvm.load %17 : !llvm.ptr -> i64
+        %19 = llvm.mlir.constant(0 : i64) : i64
+        %20 = llvm.sub %9, %14 overflow<nsw> : i64
+        %21 = llvm.mul %20, %3 overflow<nsw> : i64
+        %22 = llvm.mul %21, %3 overflow<nsw> : i64
+        %23 = llvm.add %22,%19 overflow<nsw> : i64
+        %24 = llvm.mul %3, %16 overflow<nsw> : i64
+        //CHECK:  %[[VAL1:.*]] = getelementptr float, ptr {{.*}}, i64 %{{.*}}
+        //CHECK:  %[[LOAD_A:.*]] = load float, ptr %[[VAL1]], align 4, !nontemporal 
+        //CHECK:  %[[RES:.*]] = fadd contract float %[[LOAD_A]], 2.000000e+01
+        %25 = llvm.getelementptr %11[%23] : (!llvm.ptr, i64) -> !llvm.ptr, f32
+        %26 = llvm.load %25 : !llvm.ptr -> f32
+        %27 = llvm.mlir.constant(2.000000e+01 : f32) : f32
+        %28 = llvm.fadd %26, %27 {fastmathFlags = #llvm.fastmath<contract>} : f32
+        //CHECK:  call void @llvm.memcpy.p0.p0.i32(ptr %[[A_VAL1]], ptr %1, i32 48, i1 false)
+        %29 = llvm.mlir.constant(48 : i32) : i32
+        "llvm.intr.memcpy"(%1, %arg1, %29) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i32) -> ()
+        %30 = llvm.load %4 : !llvm.ptr -> i32
+        %31 = llvm.sext %30 : i32 to i64
+        %32 = llvm.getelementptr %1[0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)>
+        %33 = llvm.load %32 : !llvm.ptr -> !llvm.ptr
+        %34 = llvm.mlir.constant(0 : index) : i64
+        %35 = llvm.getelementptr %1[0, 7, %34, 0] : (!llvm.ptr, i64) -> !llvm.ptr, !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)>
+        %36 = llvm.load %35 : !llvm.ptr -> i64
+        %37 = llvm.getelementptr %1[0, 7, %34, 1] : (!llvm.ptr, i64) -> !llvm.ptr, !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)>
+        %38 = llvm.load %37 : !llvm.ptr -> i64
+        %39 = llvm.getelementptr %1[0, 7, %34, 2] : (!llvm.ptr, i64) -> !llvm.ptr, !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)>
+        %40 = llvm.load %39 : !llvm.ptr -> i64
+        %41 = llvm.sub %31, %36 overflow<nsw> : i64
+        %42 = llvm.mul %41, %3 overflow<nsw> : i64
+        %43 = llvm.mul %42, %3 overflow<nsw> : i64
+        %44 = llvm.add %43,%19 overflow<nsw> : i64
+        %45 = llvm.mul %3, %38 overflow<nsw> : i64
+        //CHECK:  %[[VAL2:.*]] = getelementptr float, ptr %{{.*}}, i64 %{{.*}}
+        //CHECK:  store float %[[RES]], ptr %[[VAL2]], align 4, !nontemporal 
+        %46 = llvm.getelementptr %33[%44] : (!llvm.ptr, i64) -> !llvm.ptr, f32
+        llvm.store %28, %46 : f32, !llvm.ptr
+        omp.yield
+      }
+    }
+    llvm.return
+  }
+// -----
+ 
+
diff --git a/mlir/test/Target/LLVMIR/openmp-todo.mlir b/mlir/test/Target/LLVMIR/openmp-todo.mlir
index f907bb3f94a2a..a6ed73b158bdb 100644
--- a/mlir/test/Target/LLVMIR/openmp-todo.mlir
+++ b/mlir/test/Target/LLVMIR/openmp-todo.mlir
@@ -184,19 +184,6 @@ llvm.func @simd_linear(%lb : i32, %ub : i32, %step : i32, %x : !llvm.ptr) {
 
 // -----
 
-llvm.func @simd_nontemporal(%lb : i32, %ub : i32, %step : i32, %x : !llvm.ptr) {
-  // expected-error at below {{not yet implemented: Unhandled clause nontemporal in omp.simd operation}}
-  // expected-error at below {{LLVM Translation failed for operation: omp.simd}}
-  omp.simd nontemporal(%x : !llvm.ptr) {
-    omp.loop_nest (%iv) : i32 = (%lb) to (%ub) step (%step) {
-      omp.yield
-    }
-  }
-  llvm.return
-}
-
-// -----
-
 omp.declare_reduction @add_f32 : f32
 init {
 ^bb0(%arg: f32):

>From 53cc22910ede9382ecb2258f390515a530edc140 Mon Sep 17 00:00:00 2001
From: Kaviya Rajendiran <kaviyara2000 at gmail.com>
Date: Wed, 5 Mar 2025 12:20:45 +0530
Subject: [PATCH 2/2] [MLIR][OpenMP] Created callback function for adding the
 metadata of nontemporal and handled the translation in
 OpenMPToLLVMIRTranslation.cpp

---
 .../llvm/Frontend/OpenMP/OMPIRBuilder.h       | 16 +++-
 llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp     | 91 ++-----------------
 .../OpenMP/OpenMPToLLVMIRTranslation.cpp      | 91 +++++++++++++++++--
 3 files changed, 104 insertions(+), 94 deletions(-)

diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
index fc726ec6cf4b4..60c64e369ca54 100644
--- a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
+++ b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
@@ -1229,6 +1229,9 @@ class OpenMPIRBuilder {
   void unrollLoopPartial(DebugLoc DL, CanonicalLoopInfo *Loop, int32_t Factor,
                          CanonicalLoopInfo **UnrolledCLI);
 
+  using NonTemporalBodyGenCallbackTy =
+      function_ref<void(llvm::BasicBlock *BB, MDNode *NontemporalNode)>;
+
   /// Add metadata to simd-ize a loop. If IfCond is not nullptr, the loop
   /// is cloned. The metadata which prevents vectorization is added to
   /// to the cloned loop. The cloned loop is executed when ifCond is evaluated
@@ -1242,10 +1245,15 @@ class OpenMPIRBuilder {
   /// \param Order       The enum to map order clause.
   /// \param Simdlen     The Simdlen length to apply to the simd loop.
   /// \param Safelen     The Safelen length to apply to the simd loop.
-  void applySimd(CanonicalLoopInfo *Loop,
-                 MapVector<Value *, Value *> AlignedVars, Value *IfCond,
-                 omp::OrderKind Order, ConstantInt *Simdlen,
-                 ConstantInt *Safelen, ArrayRef<Value *> NontempralVars = {});
+  /// \param NontemporalCBFunc    Call back function for nontemporal.
+  /// \param NontemporalVars      Array of nontemporal vars.
+  void applySimd(
+      CanonicalLoopInfo *Loop, MapVector<Value *, Value *> AlignedVars,
+      Value *IfCond, omp::OrderKind Order, ConstantInt *Simdlen,
+      ConstantInt *Safelen,
+      NonTemporalBodyGenCallbackTy NontemporalCBFunc = [](BasicBlock *,
+                                                          MDNode *) {},
+      ArrayRef<Value *> NontempralVars = {});
 
   /// Generator for '#omp flush'
   ///
diff --git a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
index bff201a1377c8..5973c0d63aa5d 100644
--- a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
+++ b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
@@ -5385,86 +5385,11 @@ OpenMPIRBuilder::getOpenMPDefaultSimdAlign(const Triple &TargetTriple,
   return 0;
 }
 
-static void appendNontemporalVars(BasicBlock *Block,
-                                  SmallVectorImpl<Value *> &NontemporalVars) {
-  for (Instruction &I : *Block) {
-    if (const CallInst *CI = dyn_cast<CallInst>(&I)) {
-      if (CI->getIntrinsicID() == Intrinsic::memcpy) {
-        llvm::Value *DestPtr = CI->getArgOperand(0);
-        llvm::Value *SrcPtr = CI->getArgOperand(1);
-        for (const llvm::Value *Var : NontemporalVars) {
-          if (Var == SrcPtr) {
-            NontemporalVars.push_back(DestPtr);
-            break;
-          }
-        }
-      }
-    }
-  }
-}
-
-/** Attach nontemporal metadata to the load/store instructions of nontemporal
- *  variables of \p Block
- * Nontemporal variables may be a scalar, fixed size or allocatable
- * or pointer array
- *
- * Example scenarios for nontemporal variables:
- * Case 1: Scalar variable
- *    If the nontemporal variable is a scalar, it is allocated on stack.Load and
- * store instructions directly access the alloca pointer of the scalar
- * variable for fetching information about scalar variable or  writing
- * into the scalar variable. Mark those load and store instructions as
- * non-temporal.
- *
- * Case 2: Fixed Size array
- *    If the nontemporal variable is a fixed-size array, it is allocated
- * as a contiguous block of memory. It uses one GEP instruction, to compute the
- * address of each individual array elements and perform load or store
- * operation on it. Mark those load and store instructions as non-temporal.
- *
- * Case 3: Allocatable array
- *     For an allocatable array, which might involve runtime type descriptor,
- * needs to navigate through descriptors using two or more GEP and load
- * instructions to compute the address of each individual element in an array.
- * Mark those load or store which access the individual array elements as
- * non-temporal.
- */
-static void addNonTemporalMetadata(BasicBlock *Block, MDNode *Nontemporal,
-                                   SmallVectorImpl<Value *> &NontemporalVars) {
-  appendNontemporalVars(Block, NontemporalVars);
-  for (Instruction &I : *Block) {
-    llvm::Value *mem_ptr = nullptr;
-    bool MetadataFlag = true;
-    if (llvm::LoadInst *li = dyn_cast<llvm::LoadInst>(&I)) {
-      if (!(li->getType()->isPointerTy()))
-        mem_ptr = li->getPointerOperand();
-    } else if (llvm::StoreInst *si = dyn_cast<llvm::StoreInst>(&I))
-      mem_ptr = si->getPointerOperand();
-    if (mem_ptr) {
-      while (mem_ptr && !(isa<llvm::AllocaInst>(mem_ptr))) {
-        if (llvm::GetElementPtrInst *gep =
-                dyn_cast<llvm::GetElementPtrInst>(mem_ptr)) {
-          llvm::Type *sourceType = gep->getSourceElementType();
-          if (sourceType->isStructTy() && gep->getNumIndices() >= 2 &&
-              !(gep->hasAllZeroIndices())) {
-            MetadataFlag = false;
-            break;
-          }
-          mem_ptr = gep->getPointerOperand();
-        } else if (llvm::LoadInst *li = dyn_cast<llvm::LoadInst>(mem_ptr))
-          mem_ptr = li->getPointerOperand();
-      }
-      if (MetadataFlag && is_contained(NontemporalVars, mem_ptr))
-        I.setMetadata(LLVMContext::MD_nontemporal, Nontemporal);
-    }
-  }
-}
-
-void OpenMPIRBuilder::applySimd(CanonicalLoopInfo *CanonicalLoop,
-                                MapVector<Value *, Value *> AlignedVars,
-                                Value *IfCond, OrderKind Order,
-                                ConstantInt *Simdlen, ConstantInt *Safelen,
-                                ArrayRef<Value *> NontemporalVarsIn) {
+void OpenMPIRBuilder::applySimd(
+    CanonicalLoopInfo *CanonicalLoop, MapVector<Value *, Value *> AlignedVars,
+    Value *IfCond, OrderKind Order, ConstantInt *Simdlen, ConstantInt *Safelen,
+    OpenMPIRBuilder::NonTemporalBodyGenCallbackTy NontemporalCBFunc,
+    ArrayRef<Value *> NontemporalVarsIn) {
   LLVMContext &Ctx = Builder.getContext();
 
   Function *F = CanonicalLoop->getFunction();
@@ -5562,12 +5487,12 @@ void OpenMPIRBuilder::applySimd(CanonicalLoopInfo *CanonicalLoop,
   }
 
   addLoopMetadata(CanonicalLoop, LoopMDList);
-  SmallVector<Value *> NontemporalVars{NontemporalVarsIn};
+
   // Set nontemporal metadata to load and stores of nontemporal values
-  if (NontemporalVars.size()) {
+  if (NontemporalVarsIn.size()) {
     MDNode *NontemporalNode = MDNode::getDistinct(Ctx, {});
     for (BasicBlock *BB : Reachable)
-      addNonTemporalMetadata(BB, NontemporalNode, NontemporalVars);
+      NontemporalCBFunc(BB, NontemporalNode);
   }
 }
 
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index 1c9690a1c7b68..ef313388df8e0 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -2462,6 +2462,25 @@ convertOrderKind(std::optional<omp::ClauseOrderKind> o) {
   llvm_unreachable("Unknown ClauseOrderKind kind");
 }
 
+static void
+appendNontemporalVars(llvm::BasicBlock *Block,
+                      SmallVectorImpl<llvm::Value *> &NontemporalVars) {
+  for (llvm::Instruction &I : *Block) {
+    if (const llvm::CallInst *CI = dyn_cast<llvm::CallInst>(&I)) {
+      if (CI->getIntrinsicID() == llvm::Intrinsic::memcpy) {
+        llvm::Value *DestPtr = CI->getArgOperand(0);
+        llvm::Value *SrcPtr = CI->getArgOperand(1);
+        for (const llvm::Value *Var : NontemporalVars) {
+          if (Var == SrcPtr) {
+            NontemporalVars.push_back(DestPtr);
+            break;
+          }
+        }
+      }
+    }
+  }
+}
+
 /// Converts an OpenMP simd loop into LLVM IR using OpenMPIRBuilder.
 static LogicalResult
 convertOmpSimd(Operation &opInst, llvm::IRBuilderBase &builder,
@@ -2523,13 +2542,71 @@ convertOmpSimd(Operation &opInst, llvm::IRBuilderBase &builder,
   llvm::MapVector<llvm::Value *, llvm::Value *> alignedVars;
   llvm::omp::OrderKind order = convertOrderKind(simdOp.getOrder());
 
-  llvm::SmallVector<llvm::Value *> nontemporalVars;
+  llvm::SmallVector<llvm::Value *> nontemporalOrigVars;
   mlir::OperandRange nontemporals = simdOp.getNontemporalVars();
   for (mlir::Value nontemporal : nontemporals) {
     llvm::Value *nt = moduleTranslation.lookupValue(nontemporal);
-    nontemporalVars.push_back(nt);
+    nontemporalOrigVars.push_back(nt);
   }
 
+  /** Call back function to attach nontemporal metadata to the load/store
+   * instructions of nontemporal variables of Block.
+   * Nontemporal variables may be a scalar, fixed size or allocatable
+   * or pointer array
+   *
+   * Example scenarios for nontemporal variables:
+   * Case 1: Scalar variable
+   *    If the nontemporal variable is a scalar, it is allocated on stack.Load
+   * and store instructions directly access the alloca pointer of the scalar
+   * variable for fetching information about scalar variable or  writing
+   * into the scalar variable. Mark those load and store instructions as
+   * non-temporal.
+   *
+   * Case 2: Fixed Size array
+   *    If the nontemporal variable is a fixed-size array, it is allocated
+   * as a contiguous block of memory. It uses one GEP instruction, to compute
+   * the address of each individual array elements and perform load or store
+   * operation on it. Mark those load and store instructions as non-temporal.
+   *
+   * Case 3: Allocatable array
+   *     For an allocatable array, which might involve runtime type descriptor,
+   * needs to navigate through descriptors using two or more GEP and load
+   * instructions to compute the address of each individual element in an array.
+   * Mark those load or store which access the individual array elements as
+   * non-temporal.
+   */
+  auto addNonTemporalMetadataCB = [&](llvm::BasicBlock *Block,
+                                      llvm::MDNode *Nontemporal) {
+    SmallVector<llvm::Value *> NontemporalVars{nontemporalOrigVars};
+    appendNontemporalVars(Block, NontemporalVars);
+    for (llvm::Instruction &I : *Block) {
+      llvm::Value *mem_ptr = nullptr;
+      bool MetadataFlag = true;
+      if (llvm::LoadInst *li = dyn_cast<llvm::LoadInst>(&I)) {
+        if (!(li->getType()->isPointerTy()))
+          mem_ptr = li->getPointerOperand();
+      } else if (llvm::StoreInst *si = dyn_cast<llvm::StoreInst>(&I))
+        mem_ptr = si->getPointerOperand();
+      if (mem_ptr) {
+        while (mem_ptr && !(isa<llvm::AllocaInst>(mem_ptr))) {
+          if (llvm::GetElementPtrInst *gep =
+                  dyn_cast<llvm::GetElementPtrInst>(mem_ptr)) {
+            llvm::Type *sourceType = gep->getSourceElementType();
+            if (sourceType->isStructTy() && gep->getNumIndices() >= 2 &&
+                !(gep->hasAllZeroIndices())) {
+              MetadataFlag = false;
+              break;
+            }
+            mem_ptr = gep->getPointerOperand();
+          } else if (llvm::LoadInst *li = dyn_cast<llvm::LoadInst>(mem_ptr))
+            mem_ptr = li->getPointerOperand();
+        }
+        if (MetadataFlag && is_contained(NontemporalVars, mem_ptr))
+          I.setMetadata(llvm::LLVMContext::MD_nontemporal, Nontemporal);
+      }
+    }
+  };
+
   llvm::BasicBlock *sourceBlock = builder.GetInsertBlock();
   std::optional<ArrayAttr> alignmentValues = simdOp.getAlignments();
   mlir::OperandRange operands = simdOp.getAlignedVars();
@@ -2557,11 +2634,11 @@ convertOmpSimd(Operation &opInst, llvm::IRBuilderBase &builder,
 
   builder.SetInsertPoint(*regionBlock, (*regionBlock)->begin());
   llvm::CanonicalLoopInfo *loopInfo = findCurrentLoopInfo(moduleTranslation);
-  ompBuilder->applySimd(loopInfo, alignedVars,
-                        simdOp.getIfExpr()
-                            ? moduleTranslation.lookupValue(simdOp.getIfExpr())
-                            : nullptr,
-                        order, simdlen, safelen, nontemporalVars);
+  ompBuilder->applySimd(
+      loopInfo, alignedVars,
+      simdOp.getIfExpr() ? moduleTranslation.lookupValue(simdOp.getIfExpr())
+                         : nullptr,
+      order, simdlen, safelen, addNonTemporalMetadataCB, nontemporalOrigVars);
 
   return cleanupPrivateVars(builder, moduleTranslation, simdOp.getLoc(),
                             llvmPrivateVars, privateDecls);



More information about the Mlir-commits mailing list