From 7a28de08f384702aa6b6fb7482a410754972d622 Mon Sep 17 00:00:00 2001
From: Justin Holewinski <jholewinski@nvidia.com>
Date: Fri, 27 Jun 2014 18:36:11 +0000
Subject: [NVPTX] Add reflect intrinsic (better than matching by function name)

Also clean up some of the logic in NVVMReflect.cpp while we're messing around in there.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211948 91177308-0d34-0410-b5e6-96231b3b80d8
---
 include/llvm/IR/IntrinsicsNVVM.td  |  3 ++
 lib/Target/NVPTX/NVVMReflect.cpp   | 69 ++++++++++++++++++++++++++------------
 test/CodeGen/NVPTX/nvvm-reflect.ll | 14 ++++++++
 3 files changed, 64 insertions(+), 22 deletions(-)

diff --git a/include/llvm/IR/IntrinsicsNVVM.td b/include/llvm/IR/IntrinsicsNVVM.td
index a53d6eaf46..6baf01875d 100644
--- a/include/llvm/IR/IntrinsicsNVVM.td
+++ b/include/llvm/IR/IntrinsicsNVVM.td
@@ -888,6 +888,9 @@ def int_nvvm_compiler_error :
 def int_nvvm_compiler_warn :
     Intrinsic<[], [llvm_anyptr_ty], [], "llvm.nvvm.compiler.warn">;
 
+def int_nvvm_reflect :
+  Intrinsic<[llvm_i32_ty], [llvm_anyptr_ty], [IntrNoMem], "llvm.nvvm.reflect">;
+
 // isspacep.{const, global, local, shared}
 def int_nvvm_isspacep_const
   : Intrinsic<[llvm_i1_ty], [llvm_ptr_ty], [IntrNoMem],
diff --git a/lib/Target/NVPTX/NVVMReflect.cpp b/lib/Target/NVPTX/NVVMReflect.cpp
index cb8bd72602..a8d6b95ae4 100644
--- a/lib/Target/NVPTX/NVVMReflect.cpp
+++ b/lib/Target/NVPTX/NVVMReflect.cpp
@@ -22,6 +22,7 @@
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/Type.h"
 #include "llvm/Pass.h"
@@ -47,17 +48,16 @@ class NVVMReflect : public ModulePass {
 private:
   StringMap<int> VarMap;
   typedef DenseMap<std::string, int>::iterator VarMapIter;
-  Function *ReflectFunction;
 
 public:
   static char ID;
-  NVVMReflect() : ModulePass(ID), ReflectFunction(nullptr) {
+  NVVMReflect() : ModulePass(ID) {
     initializeNVVMReflectPass(*PassRegistry::getPassRegistry());
     VarMap.clear();
   }
 
   NVVMReflect(const StringMap<int> &Mapping)
-  : ModulePass(ID), ReflectFunction(nullptr) {
+  : ModulePass(ID) {
     initializeNVVMReflectPass(*PassRegistry::getPassRegistry());
     for (StringMap<int>::const_iterator I = Mapping.begin(), E = Mapping.end();
          I != E; ++I) {
@@ -70,6 +70,8 @@ public:
   }
   bool runOnModule(Module &) override;
 
+private:
+  bool handleFunction(Function *ReflectFunction);
   void setVarMap();
 };
 }
@@ -120,19 +122,7 @@ void NVVMReflect::setVarMap() {
   }
 }
 
-bool NVVMReflect::runOnModule(Module &M) {
-  if (!NVVMReflectEnabled)
-    return false;
-
-  setVarMap();
-
-  ReflectFunction = M.getFunction(NVVM_REFLECT_FUNCTION);
-
-  // If reflect function is not used, then there will be
-  // no entry in the module.
-  if (!ReflectFunction)
-    return false;
-
+bool NVVMReflect::handleFunction(Function *ReflectFunction) {
   // Validate _reflect function
   assert(ReflectFunction->isDeclaration() &&
          "_reflect function should not have a body");
@@ -155,13 +145,15 @@ bool NVVMReflect::runOnModule(Module &M) {
            "Only one operand expect for _reflect function");
     // In cuda, we will have an extra constant-to-generic conversion of
     // the string.
-    const Value *conv = Reflect->getArgOperand(0);
-    assert(isa<CallInst>(conv) && "Expected a const-to-gen conversion");
-    const CallInst *ConvCall = cast<CallInst>(conv);
-    const Value *str = ConvCall->getArgOperand(0);
-    assert(isa<ConstantExpr>(str) &&
+    const Value *Str = Reflect->getArgOperand(0);
+    if (isa<CallInst>(Str)) {
+      // CUDA path
+      const CallInst *ConvCall = cast<CallInst>(Str);
+      Str = ConvCall->getArgOperand(0);
+    }
+    assert(isa<ConstantExpr>(Str) &&
            "Format of _reflect function not recognized");
-    const ConstantExpr *GEP = cast<ConstantExpr>(str);
+    const ConstantExpr *GEP = cast<ConstantExpr>(Str);
 
     const Value *Sym = GEP->getOperand(0);
     assert(isa<Constant>(Sym) && "Format of _reflect function not recognized");
@@ -195,3 +187,36 @@ bool NVVMReflect::runOnModule(Module &M) {
     ToRemove[i]->eraseFromParent();
   return true;
 }
+
+bool NVVMReflect::runOnModule(Module &M) {
+  if (!NVVMReflectEnabled)
+    return false;
+
+  setVarMap();
+
+
+  bool Res = false;
+  std::string Name;
+  Type *Tys[1];
+  Type *I8Ty = Type::getInt8Ty(M.getContext());
+  Function *ReflectFunction;
+
+  // Check for standard overloaded versions of llvm.nvvm.reflect
+
+  for (unsigned i = 0; i != 5; ++i) {
+    Tys[0] = PointerType::get(I8Ty, i);
+    Name = Intrinsic::getName(Intrinsic::nvvm_reflect, Tys);
+    ReflectFunction = M.getFunction(Name);
+    if(ReflectFunction != 0) {
+      Res |= handleFunction(ReflectFunction);
+    }
+  }
+
+  ReflectFunction = M.getFunction(NVVM_REFLECT_FUNCTION);
+  // If reflect function is not used, then there will be
+  // no entry in the module.
+  if (ReflectFunction != 0)
+    Res |= handleFunction(ReflectFunction);
+
+  return Res;
+}
diff --git a/test/CodeGen/NVPTX/nvvm-reflect.ll b/test/CodeGen/NVPTX/nvvm-reflect.ll
index 45438847b8..21e9c69e65 100644
--- a/test/CodeGen/NVPTX/nvvm-reflect.ll
+++ b/test/CodeGen/NVPTX/nvvm-reflect.ll
@@ -32,3 +32,17 @@ exit:
   %ret = phi float [%ret1, %use_mul], [%ret2, %use_add]
   ret float %ret
 }
+
+declare i32 @llvm.nvvm.reflect.p0i8(i8*)
+
+; USE_MUL_0: define i32 @intrinsic
+; USE_MUL_1: define i32 @intrinsic
+define i32 @intrinsic() {
+; USE_MUL_0-NOT: call i32 @llvm.nvvm.reflect
+; USE_MUL_0: ret i32 0
+; USE_MUL_1-NOT: call i32 @llvm.nvvm.reflect
+; USE_MUL_1: ret i32 1
+  %ptr = tail call i8* @llvm.nvvm.ptr.constant.to.gen.p0i8.p4i8(i8 addrspace(4)* getelementptr inbounds ([8 x i8] addrspace(4)* @str, i32 0, i32 0))
+  %reflect = tail call i32 @llvm.nvvm.reflect.p0i8(i8* %ptr)
+  ret i32 %reflect
+}
-- 
cgit v1.2.3