Expose "noduplicate" attribute as a property for intrinsics.

The "noduplicate" function attribute exists to prevent certain optimizations
from duplicating calls to the function. This is important on platforms where
certain function call duplications are unsafe (for example execution barriers
for CUDA and OpenCL).

This patch makes it possible to specify intrinsics as "noduplicate" and
translates that to the appropriate function attribute.

llvm-svn: 204200

7 files changed, 109 insertions, 6 deletions

diff --git a/llvm/include/llvm/IR/Intrinsics.td b/llvm/include/llvm/IR/Intrinsics.td
index 755c9c2b..94c7b50 100644
--- a/llvm/include/llvm/IR/Intrinsics.td
+++ b/llvm/include/llvm/IR/Intrinsics.td
@@ -69,6 +69,10 @@ class ReadNone<int argNo> : IntrinsicProperty {
 
 def IntrNoReturn : IntrinsicProperty;
 
+// IntrNoduplicate - Calls to this intrinsic cannot be duplicated.
+// Parallels the noduplicate attribute on LLVM IR functions.
+def IntrNoDuplicate : IntrinsicProperty;
+
 //===----------------------------------------------------------------------===//
 // Types used by intrinsics.
 //===----------------------------------------------------------------------===//
diff --git a/llvm/include/llvm/IR/IntrinsicsNVVM.td b/llvm/include/llvm/IR/IntrinsicsNVVM.td
index a372c22..7f72ce8b 100644
--- a/llvm/include/llvm/IR/IntrinsicsNVVM.td
+++ b/llvm/include/llvm/IR/IntrinsicsNVVM.td
@@ -730,15 +730,15 @@ def llvm_anyi64ptr_ty     : LLVMAnyPointerType<llvm_i64_ty>;     // (space)i64*
 
 // Bar.Sync
   def int_cuda_syncthreads : GCCBuiltin<"__syncthreads">,
-      Intrinsic<[], [], []>;
+      Intrinsic<[], [], [IntrNoDuplicate]>;
   def int_nvvm_barrier0 : GCCBuiltin<"__nvvm_bar0">,
-      Intrinsic<[], [], []>;
+      Intrinsic<[], [], [IntrNoDuplicate]>;
   def int_nvvm_barrier0_popc : GCCBuiltin<"__nvvm_bar0_popc">,
-      Intrinsic<[llvm_i32_ty], [llvm_i32_ty], []>;
+      Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoDuplicate]>;
   def int_nvvm_barrier0_and : GCCBuiltin<"__nvvm_bar0_and">,
-      Intrinsic<[llvm_i32_ty], [llvm_i32_ty], []>;
+      Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoDuplicate]>;
   def int_nvvm_barrier0_or : GCCBuiltin<"__nvvm_bar0_or">,
-      Intrinsic<[llvm_i32_ty], [llvm_i32_ty], []>;
+      Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoDuplicate]>;
 
   // Membar
   def int_nvvm_membar_cta : GCCBuiltin<"__nvvm_membar_cta">,
diff --git a/llvm/test/CodeGen/NVPTX/noduplicate-syncthreads.ll b/llvm/test/CodeGen/NVPTX/noduplicate-syncthreads.ll
new file mode 100644
index 0000000..64745fc
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/noduplicate-syncthreads.ll
@@ -0,0 +1,74 @@
+; RUN: opt < %s -O3 -S | FileCheck %s
+
+; Make sure the call to syncthreads is not duplicate here by the LLVM
+; optimizations, because it has the noduplicate attribute set.
+
+; CHECK: call void @llvm.cuda.syncthreads
+; CHECK-NOT: call void @llvm.cuda.syncthreads
+
+; Function Attrs: nounwind
+define void @foo(float* %output) #1 {
+entry:
+  %output.addr = alloca float*, align 8
+  store float* %output, float** %output.addr, align 8
+  %0 = load float** %output.addr, align 8
+  %arrayidx = getelementptr inbounds float* %0, i64 0
+  %1 = load float* %arrayidx, align 4
+  %conv = fpext float %1 to double
+  %cmp = fcmp olt double %conv, 1.000000e+01
+  br i1 %cmp, label %if.then, label %if.else
+
+if.then:                                          ; preds = %entry
+  %2 = load float** %output.addr, align 8
+  %3 = load float* %2, align 4
+  %conv1 = fpext float %3 to double
+  %add = fadd double %conv1, 1.000000e+00
+  %conv2 = fptrunc double %add to float
+  store float %conv2, float* %2, align 4
+  br label %if.end
+
+if.else:                                          ; preds = %entry
+  %4 = load float** %output.addr, align 8
+  %5 = load float* %4, align 4
+  %conv3 = fpext float %5 to double
+  %add4 = fadd double %conv3, 2.000000e+00
+  %conv5 = fptrunc double %add4 to float
+  store float %conv5, float* %4, align 4
+  br label %if.end
+
+if.end:                                           ; preds = %if.else, %if.then
+  call void @llvm.cuda.syncthreads()
+  %6 = load float** %output.addr, align 8
+  %arrayidx6 = getelementptr inbounds float* %6, i64 0
+  %7 = load float* %arrayidx6, align 4
+  %conv7 = fpext float %7 to double
+  %cmp8 = fcmp olt double %conv7, 1.000000e+01
+  br i1 %cmp8, label %if.then9, label %if.else13
+
+if.then9:                                         ; preds = %if.end
+  %8 = load float** %output.addr, align 8
+  %9 = load float* %8, align 4
+  %conv10 = fpext float %9 to double
+  %add11 = fadd double %conv10, 3.000000e+00
+  %conv12 = fptrunc double %add11 to float
+  store float %conv12, float* %8, align 4
+  br label %if.end17
+
+if.else13:                                        ; preds = %if.end
+  %10 = load float** %output.addr, align 8
+  %11 = load float* %10, align 4
+  %conv14 = fpext float %11 to double
+  %add15 = fadd double %conv14, 4.000000e+00
+  %conv16 = fptrunc double %add15 to float
+  store float %conv16, float* %10, align 4
+  br label %if.end17
+
+if.end17:                                         ; preds = %if.else13, %if.then9
+  ret void
+}
+
+; Function Attrs: noduplicate nounwind
+declare void @llvm.cuda.syncthreads() #2
+
+!0 = metadata !{void (float*)* @foo, metadata !"kernel", i32 1}
+!1 = metadata !{null, metadata !"align", i32 8}
diff --git a/llvm/test/Feature/intrinsic-noduplicate.ll b/llvm/test/Feature/intrinsic-noduplicate.ll
new file mode 100644
index 0000000..9a2b0ab
--- /dev/null
+++ b/llvm/test/Feature/intrinsic-noduplicate.ll
@@ -0,0 +1,9 @@
+; RUN: llvm-as < %s | llvm-dis | FileCheck %s
+
+; Make sure LLVM knows about the noduplicate attribute on the
+; llvm.cuda.syncthreads intrinsic.
+
+declare void @llvm.cuda.syncthreads()
+
+; CHECK: declare void @llvm.cuda.syncthreads() #[[ATTRNUM:[0-9]+]]
+; CHECK: attributes #[[ATTRNUM]] = { noduplicate nounwind }
diff --git a/llvm/utils/TableGen/CodeGenIntrinsics.h b/llvm/utils/TableGen/CodeGenIntrinsics.h
index edbb18b..06daa97 100644
--- a/llvm/utils/TableGen/CodeGenIntrinsics.h
+++ b/llvm/utils/TableGen/CodeGenIntrinsics.h
@@ -73,6 +73,9 @@ namespace llvm {
     /// canThrow - True if the intrinsic can throw.
     bool canThrow;
 
+    /// isNoDuplicate - True if the intrinsic is marked as noduplicate.
+    bool isNoDuplicate;
+
     /// isNoReturn - True if the intrinsic is no-return.
     bool isNoReturn;
 
diff --git a/llvm/utils/TableGen/CodeGenTarget.cpp b/llvm/utils/TableGen/CodeGenTarget.cpp
index 1f47675..884af4c 100644
--- a/llvm/utils/TableGen/CodeGenTarget.cpp
+++ b/llvm/utils/TableGen/CodeGenTarget.cpp
@@ -446,6 +446,7 @@ CodeGenIntrinsic::CodeGenIntrinsic(Record *R) {
   isCommutative = false;
   canThrow = false;
   isNoReturn = false;
+  isNoDuplicate = false;
 
   if (DefName.size() <= 4 ||
       std::string(DefName.begin(), DefName.begin() + 4) != "int_")
@@ -570,6 +571,8 @@ CodeGenIntrinsic::CodeGenIntrinsic(Record *R) {
       isCommutative = true;
     else if (Property->getName() == "Throws")
       canThrow = true;
+    else if (Property->getName() == "IntrNoDuplicate")
+      isNoDuplicate = true;
     else if (Property->getName() == "IntrNoReturn")
       isNoReturn = true;
     else if (Property->isSubClassOf("NoCapture")) {
diff --git a/llvm/utils/TableGen/IntrinsicEmitter.cpp b/llvm/utils/TableGen/IntrinsicEmitter.cpp
index cf6934c..1b28128 100644
--- a/llvm/utils/TableGen/IntrinsicEmitter.cpp
+++ b/llvm/utils/TableGen/IntrinsicEmitter.cpp
@@ -502,6 +502,9 @@ struct AttributeComparator {
     if (L->canThrow != R->canThrow)
       return R->canThrow;
 
+    if (L->isNoDuplicate != R->isNoDuplicate)
+      return R->isNoDuplicate;
+
     if (L->isNoReturn != R->isNoReturn)
       return R->isNoReturn;
 
@@ -616,7 +619,8 @@ EmitAttributes(const std::vector<CodeGenIntrinsic> &Ints, raw_ostream &OS) {
 
     ModRefKind modRef = getModRefKind(intrinsic);
 
-    if (!intrinsic.canThrow || modRef || intrinsic.isNoReturn) {
+    if (!intrinsic.canThrow || modRef || intrinsic.isNoReturn ||
+        intrinsic.isNoDuplicate) {
       OS << "      const Attribute::AttrKind Atts[] = {";
       bool addComma = false;
       if (!intrinsic.canThrow) {
@@ -629,6 +633,12 @@ EmitAttributes(const std::vector<CodeGenIntrinsic> &Ints, raw_ostream &OS) {
         OS << "Attribute::NoReturn";
         addComma = true;
       }
+      if (intrinsic.isNoDuplicate) {
+        if (addComma)
+          OS << ",";
+        OS << "Attribute::NoDuplicate";
+        addComma = true;
+      }
 
       switch (modRef) {
       case MRK_none: break;