Add support for __nvvm_reflect changes in libdevice in CUDA-7.0

Summary:
CUDA 7.0's libdevice uses slightly different IR to call __nvvm_reflect
and that triggers an assertion in nvvm_reflect optimization pass. This
change allows nvvm_reflect pass to deal with both old and new ways to
pass an argument to __nvvm_reflect.

Test Plan: ninja check-all

Reviewers: eliben, echristo

Subscribers: jholewinski, llvm-commits

Differential Revision: http://reviews.llvm.org/D8399

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@232732 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/NVPTX/NVVMReflect.cpp b/lib/Target/NVPTX/NVVMReflect.cpp
index a8d6b95ae4ec6619d1ee63659ba939f0729abe8a..12d99336ebd658401f5aa21eddd21dd58a1c55aa 100644 (file)
--- a/lib/Target/NVPTX/NVVMReflect.cpp
+++ b/lib/Target/NVPTX/NVVMReflect.cpp
@@ -137,6 +137,26 @@ bool NVVMReflect::handleFunction(Function *ReflectFunction) {
   // ConstantArray can be found successfully, see if it can be
   // found in VarMap. If so, replace the uses of CallInst with the
   // value found in VarMap. If not, replace the use  with value 0.
+
+  // IR for __nvvm_reflect calls differs between CUDA versions:
+  // CUDA 6.5 and earlier uses this sequence:
+  //    %ptr = tail call i8* @llvm.nvvm.ptr.constant.to.gen.p0i8.p4i8
+  //        (i8 addrspace(4)* getelementptr inbounds
+  //           ([8 x i8], [8 x i8] addrspace(4)* @str, i32 0, i32 0))
+  //    %reflect = tail call i32 @__nvvm_reflect(i8* %ptr)
+  //
+  // Value returned by Sym->getOperand(0) is a Constant with a
+  // ConstantDataSequential operand which can be converted to string and used
+  // for lookup.
+  //
+  // CUDA 7.0 does it slightly differently:
+  //   %reflect = call i32 @__nvvm_reflect(i8* addrspacecast
+  //        (i8 addrspace(1)* getelementptr inbounds
+  //           ([8 x i8], [8 x i8] addrspace(1)* @str, i32 0, i32 0) to i8*))
+  //
+  // In this case, we get a Constant with a GlobalVariable operand and we need
+  // to dig deeper to find its initializer with the string we'll use for lookup.
+
   for (User *U : ReflectFunction->users()) {
     assert(isa<CallInst>(U) && "Only a call instruction can use _reflect");
     CallInst *Reflect = cast<CallInst>(U);
@@ -158,16 +178,23 @@ bool NVVMReflect::handleFunction(Function *ReflectFunction) {
     const Value *Sym = GEP->getOperand(0);
     assert(isa<Constant>(Sym) && "Format of _reflect function not recognized");
 
-    const Constant *SymStr = cast<Constant>(Sym);
+    const Value *Operand = cast<Constant>(Sym)->getOperand(0);
+    if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(Operand)) {
+      // For CUDA-7.0 style __nvvm_reflect calls we need to find operand's
+      // initializer.
+      assert(GV->hasInitializer() &&
+             "Format of _reflect function not recognized");
+      const Constant *Initializer = GV->getInitializer();
+      Operand = Initializer;
+    }
 
-    assert(isa<ConstantDataSequential>(SymStr->getOperand(0)) &&
+    assert(isa<ConstantDataSequential>(Operand) &&
            "Format of _reflect function not recognized");
-
-    assert(cast<ConstantDataSequential>(SymStr->getOperand(0))->isCString() &&
+    assert(cast<ConstantDataSequential>(Operand)->isCString() &&
            "Format of _reflect function not recognized");
 
     std::string ReflectArg =
-        cast<ConstantDataSequential>(SymStr->getOperand(0))->getAsString();
+        cast<ConstantDataSequential>(Operand)->getAsString();
 
     ReflectArg = ReflectArg.substr(0, ReflectArg.size() - 1);
     DEBUG(dbgs() << "Arg of _reflect : " << ReflectArg << "\n");

diff --git a/test/CodeGen/NVPTX/nvvm-reflect.ll b/test/CodeGen/NVPTX/nvvm-reflect.ll
index a203626f4e73dc8359fb140ad193e12b14356aec..8c75dfc30a56f9d250d0bfb8009f093ab5ed64fc 100644 (file)
--- a/test/CodeGen/NVPTX/nvvm-reflect.ll
+++ b/test/CodeGen/NVPTX/nvvm-reflect.ll
@@ -46,3 +46,34 @@ define i32 @intrinsic() {
   %reflect = tail call i32 @llvm.nvvm.reflect.p0i8(i8* %ptr)
   ret i32 %reflect
 }
+
+; CUDA-7.0 passes __nvvm_reflect argument slightly differently.
+; Verify that it works, too
+
+@"$str" = private addrspace(1) constant [8 x i8] c"USE_MUL\00"
+
+define float @bar(float %a, float %b) {
+; USE_MUL_0: define float @bar
+; USE_MUL_0-NOT: call i32 @__nvvm_reflect
+; USE_MUL_1: define float @bar
+; USE_MUL_1-NOT: call i32 @__nvvm_reflect
+  %reflect = call i32 @__nvvm_reflect(i8* addrspacecast (i8 addrspace(1)* getelementptr inbounds ([8 x i8], [8 x i8] addrspace(1)* @"$str", i32 0, i32 0) to i8*))
+  %cmp = icmp ne i32 %reflect, 0
+  br i1 %cmp, label %use_mul, label %use_add
+
+use_mul:
+; USE_MUL_1: fmul float %a, %b
+; USE_MUL_0-NOT: fadd float %a, %b
+  %ret1 = fmul float %a, %b
+  br label %exit
+
+use_add:
+; USE_MUL_0: fadd float %a, %b
+; USE_MUL_1-NOT: fmul float %a, %b
+  %ret2 = fadd float %a, %b
+  br label %exit
+
+exit:
+  %ret = phi float [%ret1, %use_mul], [%ret2, %use_add]
+  ret float %ret
+}