Add speculatable function attribute

This attribute tells the optimizer that the function may be speculated.

Patch by Tom Stellard

llvm-svn: 301680

docs/LangRef.rst

@@ -1535,6 +1535,17 @@ example:
 ``sanitize_thread``
     This attribute indicates that ThreadSanitizer checks
     (dynamic thread safety analysis) are enabled for this function.
+``speculatable``
+    This function attribute indicates that the function does not have any
+    effects besides calculating its result and does not have undefined behavior.
+    Note that ``speculatable`` is not enough to conclude that along any
+    particular exection path the number of calls to this function will not be
+    externally observable. This attribute is only valid on functions
+    and declarations, not on individual call sites. If a function is
+    incorrectly marked as speculatable and really does exhibit
+    undefined behavior, the undefined behavior may be observed even
+    if the call site is dead code.
+
 ``ssp``
     This attribute indicates that the function should emit a stack
     smashing protector. It is in the form of a "canary" --- a random value

include/llvm/Bitcode/LLVMBitCodes.h

@@ -545,7 +545,8 @@ enum AttributeKindCodes {
   ATTR_KIND_INACCESSIBLEMEM_ONLY = 49,
   ATTR_KIND_INACCESSIBLEMEM_OR_ARGMEMONLY = 50,
   ATTR_KIND_ALLOC_SIZE = 51,
-  ATTR_KIND_WRITEONLY = 52
+  ATTR_KIND_WRITEONLY = 52,
+  ATTR_KIND_SPECULATABLE = 53
 };
 
 enum ComdatSelectionKindCodes {

include/llvm/IR/Attributes.td

@@ -137,6 +137,9 @@ def SExt : EnumAttr<"signext">;
 /// +1 bias 0 means unaligned (different from alignstack=(1)).
 def StackAlignment : EnumAttr<"alignstack">;
 
+/// Function can be speculated.
+def Speculatable : EnumAttr<"speculatable">;
+
 /// Stack protection.
 def StackProtect : EnumAttr<"ssp">;
 

include/llvm/IR/Function.h

@@ -418,6 +418,14 @@ public:
     removeFnAttr(Attribute::Convergent);
   }
 
+  /// @brief Determine if the call has sideeffects.
+  bool isSpeculatable() const {
+    return hasFnAttribute(Attribute::Speculatable);
+  }
+  void setSpeculatable() {
+    addFnAttr(Attribute::Speculatable);
+  }
+
   /// Determine if the function is known not to recurse, directly or
   /// indirectly.
   bool doesNotRecurse() const {

include/llvm/IR/Intrinsics.td

@@ -98,6 +98,9 @@ def IntrNoDuplicate : IntrinsicProperty;
 // Parallels the convergent attribute on LLVM IR functions.
 def IntrConvergent : IntrinsicProperty;
 
+// This property indicates that the intrinsic is safe to speculate.
+def IntrSpeculatable : IntrinsicProperty;
+
 //===----------------------------------------------------------------------===//
 // Types used by intrinsics.
 //===----------------------------------------------------------------------===//

lib/AsmParser/LLLexer.cpp

@@ -648,6 +648,7 @@ lltok::Kind LLLexer::LexIdentifier() {
   KEYWORD(returned);
   KEYWORD(returns_twice);
   KEYWORD(signext);
+  KEYWORD(speculatable);
   KEYWORD(sret);
   KEYWORD(ssp);
   KEYWORD(sspreq);

lib/AsmParser/LLParser.cpp

@@ -1095,6 +1095,7 @@ bool LLParser::ParseFnAttributeValuePairs(AttrBuilder &B,
     case lltok::kw_readonly: B.addAttribute(Attribute::ReadOnly); break;
     case lltok::kw_returns_twice:
       B.addAttribute(Attribute::ReturnsTwice); break;
+    case lltok::kw_speculatable: B.addAttribute(Attribute::Speculatable); break;
     case lltok::kw_ssp: B.addAttribute(Attribute::StackProtect); break;
     case lltok::kw_sspreq: B.addAttribute(Attribute::StackProtectReq); break;
     case lltok::kw_sspstrong:

lib/AsmParser/LLToken.h

@@ -198,6 +198,7 @@ enum Kind {
   kw_returned,
   kw_returns_twice,
   kw_signext,
+  kw_speculatable,
   kw_ssp,
   kw_sspreq,
   kw_sspstrong,

lib/Bitcode/Reader/BitcodeReader.cpp

@@ -1119,6 +1119,7 @@ static uint64_t getRawAttributeMask(Attribute::AttrKind Val) {
   case Attribute::SwiftSelf:       return 1ULL << 51;
   case Attribute::SwiftError:      return 1ULL << 52;
   case Attribute::WriteOnly:       return 1ULL << 53;
+  case Attribute::Speculatable:    return 1ULL << 54;
   case Attribute::Dereferenceable:
     llvm_unreachable("dereferenceable attribute not supported in raw format");
     break;
@@ -1315,6 +1316,8 @@ static Attribute::AttrKind getAttrFromCode(uint64_t Code) {
     return Attribute::ReturnsTwice;
   case bitc::ATTR_KIND_S_EXT:
     return Attribute::SExt;
+  case bitc::ATTR_KIND_SPECULATABLE:
+    return Attribute::Speculatable;
   case bitc::ATTR_KIND_STACK_ALIGNMENT:
     return Attribute::StackAlignment;
   case bitc::ATTR_KIND_STACK_PROTECT:

lib/Bitcode/Writer/BitcodeWriter.cpp

@@ -688,6 +688,8 @@ static uint64_t getAttrKindEncoding(Attribute::AttrKind Kind) {
     return bitc::ATTR_KIND_RETURNS_TWICE;
   case Attribute::SExt:
     return bitc::ATTR_KIND_S_EXT;
+  case Attribute::Speculatable:
+    return bitc::ATTR_KIND_SPECULATABLE;
   case Attribute::StackAlignment:
     return bitc::ATTR_KIND_STACK_ALIGNMENT;
   case Attribute::StackProtect:

lib/IR/Attributes.cpp

@@ -315,6 +315,8 @@ std::string Attribute::getAsString(bool InAttrGrp) const {
     return "returns_twice";
   if (hasAttribute(Attribute::SExt))
     return "signext";
+  if (hasAttribute(Attribute::Speculatable))
+    return "speculatable";
   if (hasAttribute(Attribute::StackProtect))
     return "ssp";
   if (hasAttribute(Attribute::StackProtectReq))

lib/IR/Verifier.cpp

@@ -1352,6 +1352,7 @@ static bool isFuncOnlyAttr(Attribute::AttrKind Kind) {
   case Attribute::InaccessibleMemOnly:
   case Attribute::InaccessibleMemOrArgMemOnly:
   case Attribute::AllocSize:
+  case Attribute::Speculatable:
     return true;
   default:
     break;
@@ -2610,6 +2611,15 @@ void Verifier::verifyCallSite(CallSite CS) {
   Assert(verifyAttributeCount(Attrs, CS.arg_size()),
          "Attribute after last parameter!", I);
 
+  if (Attrs.hasAttribute(AttributeList::FunctionIndex, Attribute::Speculatable)) {
+    // Don't allow speculatable on call sites, unless the underlying function
+    // declaration is also speculatable.
+    Function *Callee
+      = dyn_cast<Function>(CS.getCalledValue()->stripPointerCasts());
+    Assert(Callee && Callee->isSpeculatable(),
+           "speculatable attribute may not apply to call sites", I);
+  }
+
   // Verify call attributes.
   verifyFunctionAttrs(FTy, Attrs, I);
 

test/Bitcode/attributes.ll

@@ -204,7 +204,7 @@ define void @f34()
 ; CHECK: define void @f34()
 {
         call void @nobuiltin() nobuiltin
-; CHECK: call void @nobuiltin() #33
+; CHECK: call void @nobuiltin() #34
         ret void;
 }
 
@@ -334,6 +334,11 @@ define void @f56() writeonly
   ret void
 }
 
+; CHECK: define void @f57() #33
+define void @f57() speculatable {
+  ret void
+}
+
 ; CHECK: attributes #0 = { noreturn }
 ; CHECK: attributes #1 = { nounwind }
 ; CHECK: attributes #2 = { readnone }
@@ -367,4 +372,5 @@ define void @f56() writeonly
 ; CHECK: attributes #30 = { allocsize(0) }
 ; CHECK: attributes #31 = { allocsize(0,1) }
 ; CHECK: attributes #32 = { writeonly }
-; CHECK: attributes #33 = { nobuiltin }
+; CHECK: attributes #33 = { speculatable }
+; CHECK: attributes #34 = { nobuiltin }

test/Bitcode/compatibility.ll

@@ -1246,7 +1246,7 @@ exit:
   ; CHECK: select <2 x i1> <i1 true, i1 false>, <2 x i8> <i8 2, i8 3>, <2 x i8> <i8 3, i8 2>
 
   call void @f.nobuiltin() builtin
-  ; CHECK: call void @f.nobuiltin() #41
+  ; CHECK: call void @f.nobuiltin() #42
 
   call fastcc noalias i32* @f.noalias() noinline
   ; CHECK: call fastcc noalias i32* @f.noalias() #12
@@ -1613,6 +1613,9 @@ normal:
 declare void @f.writeonly() writeonly
 ; CHECK: declare void @f.writeonly() #40
 
+declare void @f.speculatable() speculatable
+; CHECK: declare void @f.speculatable() #41
+
 ;; Constant Expressions
 
 define i8** @constexpr() {
@@ -1661,7 +1664,8 @@ define i8** @constexpr() {
 ; CHECK: attributes #38 = { nounwind readonly }
 ; CHECK: attributes #39 = { inaccessiblemem_or_argmemonly nounwind }
 ; CHECK: attributes #40 = { writeonly }
-; CHECK: attributes #41 = { builtin }
+; CHECK: attributes #41 = { speculatable }
+; CHECK: attributes #42 = { builtin }
 
 ;; Metadata
 

test/Verifier/speculatable-callsite-invalid.ll

@@ -0,0 +1,24 @@
+; RUN: not llvm-as %s -o /dev/null 2>&1 | FileCheck %s
+
+; Make sure that speculatable is not allowed on a call site if the
+; declaration is not also speculatable.
+
+declare i32 @not_speculatable()
+
+; CHECK: speculatable attribute may not apply to call sites
+; CHECK-NEXT: %ret = call i32 @not_speculatable() #0
+define i32 @call_not_speculatable() {
+  %ret = call i32 @not_speculatable() #0
+  ret i32 %ret
+}
+
+@gv = internal unnamed_addr constant i32 0
+
+; CHECK: speculatable attribute may not apply to call sites
+; CHECK-NEXT: %ret = call float bitcast (i32* @gv to float ()*)() #0
+define float @call_bitcast_speculatable() {
+  %ret = call float bitcast (i32* @gv to float()*)() #0
+  ret float %ret
+}
+
+attributes #0 = { speculatable }

test/Verifier/speculatable-callsite.ll

@@ -0,0 +1,20 @@
+; RUN: llvm-as %s -o /dev/null
+
+; Make sure speculatable is accepted on a call site if the declaration
+; is also speculatable.
+
+declare i32 @speculatable() #0
+
+; Make sure this the attribute is accepted on the call site if the
+; declaration matches.
+define i32 @call_speculatable() {
+  %ret = call i32 @speculatable() #0
+  ret i32 %ret
+}
+
+define float @call_bitcast_speculatable() {
+  %ret = call float bitcast (i32()* @speculatable to float()*)() #0
+  ret float %ret
+}
+
+attributes #0 = { speculatable }

utils/TableGen/CodeGenIntrinsics.h

@@ -123,6 +123,9 @@ struct CodeGenIntrinsic {
   /// True if the intrinsic is marked as convergent.
   bool isConvergent;
 
+  // True if the intrinsic is marked as speculatable.
+  bool isSpeculatable;
+
   enum ArgAttribute { NoCapture, Returned, ReadOnly, WriteOnly, ReadNone };
   std::vector<std::pair<unsigned, ArgAttribute>> ArgumentAttributes;
 

utils/TableGen/CodeGenTarget.cpp

@@ -515,6 +515,7 @@ CodeGenIntrinsic::CodeGenIntrinsic(Record *R) {
   isNoReturn = false;
   isNoDuplicate = false;
   isConvergent = false;
+  isSpeculatable = false;
 
   if (DefName.size() <= 4 ||
       std::string(DefName.begin(), DefName.begin() + 4) != "int_")
@@ -653,6 +654,8 @@ CodeGenIntrinsic::CodeGenIntrinsic(Record *R) {
       isConvergent = true;
     else if (Property->getName() == "IntrNoReturn")
       isNoReturn = true;
+    else if (Property->getName() == "IntrSpeculatable")
+      isSpeculatable = true;
     else if (Property->isSubClassOf("NoCapture")) {
       unsigned ArgNo = Property->getValueAsInt("ArgNo");
       ArgumentAttributes.push_back(std::make_pair(ArgNo, NoCapture));

utils/TableGen/IntrinsicEmitter.cpp

@@ -476,6 +476,9 @@ struct AttributeComparator {
     if (L->isConvergent != R->isConvergent)
       return R->isConvergent;
 
+    if (L->isSpeculatable != R->isSpeculatable)
+      return R->isSpeculatable;
+
     // Try to order by readonly/readnone attribute.
     CodeGenIntrinsic::ModRefBehavior LK = L->ModRef;
     CodeGenIntrinsic::ModRefBehavior RK = R->ModRef;
@@ -600,7 +603,7 @@ void IntrinsicEmitter::EmitAttributes(const CodeGenIntrinsicTable &Ints,
     if (!intrinsic.canThrow ||
         intrinsic.ModRef != CodeGenIntrinsic::ReadWriteMem ||
         intrinsic.isNoReturn || intrinsic.isNoDuplicate ||
-        intrinsic.isConvergent) {
+        intrinsic.isConvergent || intrinsic.isSpeculatable) {
       OS << "      const Attribute::AttrKind Atts[] = {";
       bool addComma = false;
       if (!intrinsic.canThrow) {
@@ -625,6 +628,12 @@ void IntrinsicEmitter::EmitAttributes(const CodeGenIntrinsicTable &Ints,
         OS << "Attribute::Convergent";
         addComma = true;
       }
+      if (intrinsic.isSpeculatable) {
+        if (addComma)
+          OS << ",";
+        OS << "Attribute::Speculatable";
+        addComma = true;
+      }
 
       switch (intrinsic.ModRef) {
       case CodeGenIntrinsic::NoMem: