Invariant start/end intrinsics overloaded for address space

Summary:
The llvm.invariant.start and llvm.invariant.end intrinsics currently
support specifying invariant memory objects only in the default address
space.

With this change, these intrinsics are overloaded for any adddress space
for memory objects
and we can use these llvm invariant intrinsics in non-default address
spaces.

Example: llvm.invariant.start.p1i8(i64 4, i8 addrspace(1)* %ptr)

This overloaded intrinsic is needed for representing final or invariant
memory in managed languages.

Reviewers: apilipenko, reames

Subscribers: llvm-commits
llvm-svn: 276447

llvm/docs/LangRef.rst

@@ -11842,10 +11842,11 @@ object following this intrinsic may be removed as dead.
 
 Syntax:
 """""""
+This is an overloaded intrinsic. The memory object can belong to any address space.
 
 ::
 
-      declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
+      declare {}* @llvm.invariant.start.p0i8(i64 <size>, i8* nocapture <ptr>)
 
 Overview:
 """""""""
@@ -11872,10 +11873,11 @@ unchanging.
 
 Syntax:
 """""""
+This is an overloaded intrinsic. The memory object can belong to any address space.
 
 ::
 
-      declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
+      declare void @llvm.invariant.end.p0i8({}* <start>, i64 <size>, i8* nocapture <ptr>)
 
 Overview:
 """""""""

llvm/include/llvm/IR/Intrinsics.td

@@ -554,11 +554,11 @@ def int_lifetime_end    : Intrinsic<[],
                                     [llvm_i64_ty, llvm_ptr_ty],
                                     [IntrArgMemOnly, NoCapture<1>]>;
 def int_invariant_start : Intrinsic<[llvm_descriptor_ty],
-                                    [llvm_i64_ty, llvm_ptr_ty],
+                                    [llvm_i64_ty, llvm_anyptr_ty],
                                     [IntrArgMemOnly, NoCapture<1>]>;
 def int_invariant_end   : Intrinsic<[],
                                     [llvm_descriptor_ty, llvm_i64_ty,
-                                     llvm_ptr_ty],
+                                     llvm_anyptr_ty],
                                     [IntrArgMemOnly, NoCapture<2>]>;
 
 def int_invariant_group_barrier : Intrinsic<[llvm_ptr_ty], 

llvm/lib/IR/AutoUpgrade.cpp

@@ -148,7 +148,31 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
     }
     break;
   }
-
+  case 'i': {
+    if (Name.startswith("invariant.start")) {
+      auto Args = F->getFunctionType()->params();
+      Type* ObjectPtr[1] = {Args[1]};
+      if (F->getName() !=
+          Intrinsic::getName(Intrinsic::invariant_start, ObjectPtr)) {
+        F->setName(Name + ".old");
+        NewFn = Intrinsic::getDeclaration(
+            F->getParent(), Intrinsic::invariant_start, ObjectPtr);
+        return true;
+      }
+    }
+    if (Name.startswith("invariant.end")) {
+      auto Args = F->getFunctionType()->params();
+      Type* ObjectPtr[1] = {Args[2]};
+      if (F->getName() !=
+          Intrinsic::getName(Intrinsic::invariant_end, ObjectPtr)) {
+        F->setName(Name + ".old");
+        NewFn = Intrinsic::getDeclaration(F->getParent(),
+                                          Intrinsic::invariant_end, ObjectPtr);
+        return true;
+      }
+    }
+    break;
+  }
   case 'm': {
     if (Name.startswith("masked.load.")) {
       Type *Tys[] = { F->getReturnType(), F->arg_begin()->getType() };
@@ -1339,6 +1363,8 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
     return;
   }
 
+  case Intrinsic::invariant_start:
+  case Intrinsic::invariant_end:
   case Intrinsic::masked_load:
   case Intrinsic::masked_store: {
     SmallVector<Value *, 4> Args(CI->arg_operands().begin(),

llvm/test/Assembler/auto_upgrade_intrinsics.ll

@@ -76,6 +76,21 @@ define void @tests.masked.store(<2 x double>* %ptr, <2 x i1> %mask, <2 x double>
   ret void
 }
 
+
+declare {}* @llvm.invariant.start(i64, i8* nocapture) nounwind readonly
+declare void @llvm.invariant.end({}*, i64, i8* nocapture) nounwind
+
+define void @tests.invariant.start.end() {
+  ; CHECK-LABEL: @tests.invariant.start.end(
+  %a = alloca i8
+  %i = call {}* @llvm.invariant.start(i64 1, i8* %a)
+  ; CHECK: call {}* @llvm.invariant.start.p0i8
+  store i8 0, i8* %a
+  call void @llvm.invariant.end({}* %i, i64 1, i8* %a)
+  ; CHECK: call void @llvm.invariant.end.p0i8
+  ret void
+}
+
 @__stack_chk_guard = external global i8*
 declare void @llvm.stackprotectorcheck(i8**)
 

llvm/test/CodeGen/AMDGPU/promote-alloca-invariant-markers.ll

@@ -1,7 +1,7 @@
 ; RUN: llc -march=amdgcn -mattr=+promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
 
-declare {}* @llvm.invariant.start(i64, i8* nocapture) #0
+declare {}* @llvm.invariant.start.p0i8(i64, i8* nocapture) #0
-declare void @llvm.invariant.end({}*, i64, i8* nocapture) #0
+declare void @llvm.invariant.end.p0i8({}*, i64, i8* nocapture) #0
 declare i8* @llvm.invariant.group.barrier(i8*) #1
 
 ; GCN-LABEL: {{^}}use_invariant_promotable_lds:
@@ -14,8 +14,8 @@ bb:
   %tmp2 = getelementptr inbounds i32, i32 addrspace(1)* %arg, i64 1
   %tmp3 = load i32, i32 addrspace(1)* %tmp2
   store i32 %tmp3, i32* %tmp
-  %tmp4 = call {}* @llvm.invariant.start(i64 4, i8* %tmp1) #0
+  %tmp4 = call {}* @llvm.invariant.start.p0i8(i64 4, i8* %tmp1) #0
-  call void @llvm.invariant.end({}* %tmp4, i64 4, i8* %tmp1) #0
+  call void @llvm.invariant.end.p0i8({}* %tmp4, i64 4, i8* %tmp1) #0
   %tmp5 = call i8* @llvm.invariant.group.barrier(i8* %tmp1) #1
   ret void
 }

llvm/test/Feature/memorymarkers.ll

@@ -6,8 +6,8 @@ declare void @_Z3barRKi(i32*)
 
 declare void @llvm.lifetime.start(i64, i8* nocapture) nounwind
 declare void @llvm.lifetime.end(i64, i8* nocapture) nounwind
-declare {}* @llvm.invariant.start(i64, i8* nocapture) readonly nounwind
+declare {}* @llvm.invariant.start.p0i8(i64, i8* nocapture) readonly nounwind
-declare void @llvm.invariant.end({}*, i64, i8* nocapture) nounwind
+declare void @llvm.invariant.end.p0i8({}*, i64, i8* nocapture) nounwind
 
 define i32 @_Z4foo2v() nounwind {
 entry:
@@ -24,12 +24,12 @@ entry:
   store i32 5, i32* %1, align 4
 
   ;; Constructor has finished here.
-  %inv = call {}* @llvm.invariant.start(i64 8, i8* %y)
+  %inv = call {}* @llvm.invariant.start.p0i8(i64 8, i8* %y)
   call void @_Z3barRKi(i32* %0) nounwind
   %2 = load i32, i32* %0, align 8
 
   ;; Destructor is run here.
-  call void @llvm.invariant.end({}* %inv, i64 8, i8* %y)
+  call void @llvm.invariant.end.p0i8({}* %inv, i64 8, i8* %y)
   ;; Destructor is done here.
   call void @llvm.lifetime.end(i64 8, i8* %y)
   ret i32 %2

llvm/test/Transforms/GlobalOpt/invariant-nodatalayout.ll

@@ -1,14 +1,14 @@
 ; RUN: opt -globalopt -S -o - < %s | FileCheck %s
 ; The check here is that it doesn't crash.
 
-declare {}* @llvm.invariant.start(i64 %size, i8* nocapture %ptr)
+declare {}* @llvm.invariant.start.p0i8(i64 %size, i8* nocapture %ptr)
 
 @object1 = global { i32, i32 } zeroinitializer
 ; CHECK: @object1 = global { i32, i32 } zeroinitializer
 
 define void @ctor1() {
   %ptr = bitcast {i32, i32}* @object1 to i8*
-  call {}* @llvm.invariant.start(i64 4, i8* %ptr)
+  call {}* @llvm.invariant.start.p0i8(i64 4, i8* %ptr)
   ret void
 }
 

llvm/test/Transforms/GlobalOpt/invariant.ll

@@ -3,10 +3,10 @@
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
-declare {}* @llvm.invariant.start(i64 %size, i8* nocapture %ptr)
+declare {}* @llvm.invariant.start.p0i8(i64 %size, i8* nocapture %ptr)
 
 define void @test1(i8* %ptr) {
-  call {}* @llvm.invariant.start(i64 4, i8* %ptr)
+  call {}* @llvm.invariant.start.p0i8(i64 4, i8* %ptr)
   ret void
 }
 
@@ -25,7 +25,7 @@ define void @ctor1() {
 define void @ctor2() {
   store i32 -1, i32* @object2
   %A = bitcast i32* @object2 to i8*
-  %B = call {}* @llvm.invariant.start(i64 4, i8* %A)
+  %B = call {}* @llvm.invariant.start.p0i8(i64 4, i8* %A)
   %C = bitcast {}* %B to i8*
   ret void
 }
@@ -36,7 +36,7 @@ define void @ctor2() {
 define void @ctor3() {
   store i32 -1, i32* @object3
   %A = bitcast i32* @object3 to i8*
-  call {}* @llvm.invariant.start(i64 3, i8* %A)
+  call {}* @llvm.invariant.start.p0i8(i64 3, i8* %A)
   ret void
 }
 
@@ -46,7 +46,7 @@ define void @ctor3() {
 define void @ctor4() {
   store i32 -1, i32* @object4
   %A = bitcast i32* @object4 to i8*
-  call {}* @llvm.invariant.start(i64 -1, i8* %A)
+  call {}* @llvm.invariant.start.p0i8(i64 -1, i8* %A)
   ret void
 }
 

llvm/test/Transforms/InstCombine/invariant.ll

@@ -2,15 +2,27 @@
 ; RUN: opt < %s -instcombine -S | FileCheck %s
 
 declare void @g(i8*)
+declare void @g_addr1(i8 addrspace(1)*)
 
-declare {}* @llvm.invariant.start(i64, i8* nocapture) nounwind readonly
+declare {}* @llvm.invariant.start.p0i8(i64, i8* nocapture) nounwind readonly
+declare {}* @llvm.invariant.start.p1i8(i64, i8 addrspace(1)* nocapture) nounwind readonly
 
 define i8 @f() {
   %a = alloca i8                                  ; <i8*> [#uses=4]
   store i8 0, i8* %a
-  %i = call {}* @llvm.invariant.start(i64 1, i8* %a) ; <{}*> [#uses=0]
+  %i = call {}* @llvm.invariant.start.p0i8(i64 1, i8* %a) ; <{}*> [#uses=0]
-  ; CHECK: call {}* @llvm.invariant.start
+  ; CHECK: call {}* @llvm.invariant.start.p0i8
   call void @g(i8* %a)
   %r = load i8, i8* %a                                ; <i8> [#uses=1]
   ret i8 %r
 }
+
+; make sure llvm.invariant.call in non-default addrspace are also not eliminated.
+define i8 @f_addrspace1(i8 addrspace(1)* %a) {
+  store i8 0, i8 addrspace(1)* %a
+  %i = call {}* @llvm.invariant.start.p1i8(i64 1, i8 addrspace(1)* %a) ; <{}*> [#uses=0]
+  ; CHECK: call {}* @llvm.invariant.start.p1i8
+  call void @g_addr1(i8 addrspace(1)* %a)
+  %r = load i8, i8 addrspace(1)* %a                                ; <i8> [#uses=1]
+  ret i8 %r
+}