Add speculatable function attribute
This attribute tells the optimizer that the function may be speculated.
Patch by Tom Stellard
llvm-svn: 301680
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index b0a3158..bf4973c 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -1535,6 +1535,17 @@
``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
diff --git a/llvm/include/llvm/Bitcode/LLVMBitCodes.h b/llvm/include/llvm/Bitcode/LLVMBitCodes.h
index 03eac80..8ee1e4b 100644
--- a/llvm/include/llvm/Bitcode/LLVMBitCodes.h
+++ b/llvm/include/llvm/Bitcode/LLVMBitCodes.h
@@ -545,7 +545,8 @@
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 {
diff --git a/llvm/include/llvm/IR/Attributes.td b/llvm/include/llvm/IR/Attributes.td
index 7b63638..75867a6 100644
--- a/llvm/include/llvm/IR/Attributes.td
+++ b/llvm/include/llvm/IR/Attributes.td
@@ -137,6 +137,9 @@
/// +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">;
diff --git a/llvm/include/llvm/IR/Function.h b/llvm/include/llvm/IR/Function.h
index f48d436..4ba34d9 100644
--- a/llvm/include/llvm/IR/Function.h
+++ b/llvm/include/llvm/IR/Function.h
@@ -418,6 +418,14 @@
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 {
diff --git a/llvm/include/llvm/IR/Intrinsics.td b/llvm/include/llvm/IR/Intrinsics.td
index 309b214..cf60f79 100644
--- a/llvm/include/llvm/IR/Intrinsics.td
+++ b/llvm/include/llvm/IR/Intrinsics.td
@@ -98,6 +98,9 @@
// 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.
//===----------------------------------------------------------------------===//
diff --git a/llvm/lib/AsmParser/LLLexer.cpp b/llvm/lib/AsmParser/LLLexer.cpp
index 49a8ce4..85152ff 100644
--- a/llvm/lib/AsmParser/LLLexer.cpp
+++ b/llvm/lib/AsmParser/LLLexer.cpp
@@ -648,6 +648,7 @@
KEYWORD(returned);
KEYWORD(returns_twice);
KEYWORD(signext);
+ KEYWORD(speculatable);
KEYWORD(sret);
KEYWORD(ssp);
KEYWORD(sspreq);
diff --git a/llvm/lib/AsmParser/LLParser.cpp b/llvm/lib/AsmParser/LLParser.cpp
index a6a9b32..f422787 100644
--- a/llvm/lib/AsmParser/LLParser.cpp
+++ b/llvm/lib/AsmParser/LLParser.cpp
@@ -1095,6 +1095,7 @@
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:
diff --git a/llvm/lib/AsmParser/LLToken.h b/llvm/lib/AsmParser/LLToken.h
index 33f8e63..9f9f4f5 100644
--- a/llvm/lib/AsmParser/LLToken.h
+++ b/llvm/lib/AsmParser/LLToken.h
@@ -198,6 +198,7 @@
kw_returned,
kw_returns_twice,
kw_signext,
+ kw_speculatable,
kw_ssp,
kw_sspreq,
kw_sspstrong,
diff --git a/llvm/lib/Bitcode/Reader/BitcodeReader.cpp b/llvm/lib/Bitcode/Reader/BitcodeReader.cpp
index 6b29f6d..782738f 100644
--- a/llvm/lib/Bitcode/Reader/BitcodeReader.cpp
+++ b/llvm/lib/Bitcode/Reader/BitcodeReader.cpp
@@ -1119,6 +1119,7 @@
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 @@
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:
diff --git a/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp b/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp
index eb8b432..54839b4 100644
--- a/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp
+++ b/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp
@@ -688,6 +688,8 @@
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:
diff --git a/llvm/lib/IR/Attributes.cpp b/llvm/lib/IR/Attributes.cpp
index 6c4d0f0..f0db6b4 100644
--- a/llvm/lib/IR/Attributes.cpp
+++ b/llvm/lib/IR/Attributes.cpp
@@ -315,6 +315,8 @@
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))
diff --git a/llvm/lib/IR/Verifier.cpp b/llvm/lib/IR/Verifier.cpp
index 5f60774..b65f3ef 100644
--- a/llvm/lib/IR/Verifier.cpp
+++ b/llvm/lib/IR/Verifier.cpp
@@ -1203,9 +1203,9 @@
void Verifier::visitModuleIdents(const Module &M) {
const NamedMDNode *Idents = M.getNamedMetadata("llvm.ident");
- if (!Idents)
+ if (!Idents)
return;
-
+
// llvm.ident takes a list of metadata entry. Each entry has only one string.
// Scan each llvm.ident entry and make sure that this requirement is met.
for (const MDNode *N : Idents->operands()) {
@@ -1215,7 +1215,7 @@
("invalid value for llvm.ident metadata entry operand"
"(the operand should be a string)"),
N->getOperand(0));
- }
+ }
}
void Verifier::visitModuleFlags(const Module &M) {
@@ -1352,6 +1352,7 @@
case Attribute::InaccessibleMemOnly:
case Attribute::InaccessibleMemOrArgMemOnly:
case Attribute::AllocSize:
+ case Attribute::Speculatable:
return true;
default:
break;
@@ -1837,7 +1838,7 @@
Assert(ExpectedNumArgs <= (int)CS.arg_size(),
"gc.statepoint too few arguments according to length fields", &CI);
- // Check that the only uses of this gc.statepoint are gc.result or
+ // Check that the only uses of this gc.statepoint are gc.result or
// gc.relocate calls which are tied to this statepoint and thus part
// of the same statepoint sequence
for (const User *U : CI.users()) {
@@ -2610,6 +2611,15 @@
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);
@@ -3908,7 +3918,7 @@
// If the intrinsic takes MDNode arguments, verify that they are either global
// or are local to *this* function.
- for (Value *V : CS.args())
+ for (Value *V : CS.args())
if (auto *MD = dyn_cast<MetadataAsValue>(V))
visitMetadataAsValue(*MD, CS.getCaller());
@@ -3981,7 +3991,7 @@
auto IsValidAlignment = [&](uint64_t Alignment) {
return isPowerOf2_64(Alignment) && ElementSizeVal.ule(Alignment);
};
-
+
uint64_t DstAlignment = CS.getParamAlignment(1),
SrcAlignment = CS.getParamAlignment(2);
@@ -4220,7 +4230,7 @@
}
case Intrinsic::masked_load: {
Assert(CS.getType()->isVectorTy(), "masked_load: must return a vector", CS);
-
+
Value *Ptr = CS.getArgOperand(0);
//Value *Alignment = CS.getArgOperand(1);
Value *Mask = CS.getArgOperand(2);
@@ -4230,12 +4240,12 @@
// DataTy is the overloaded type
Type *DataTy = cast<PointerType>(Ptr->getType())->getElementType();
- Assert(DataTy == CS.getType(),
+ Assert(DataTy == CS.getType(),
"masked_load: return must match pointer type", CS);
Assert(PassThru->getType() == DataTy,
"masked_load: pass through and data type must match", CS);
Assert(Mask->getType()->getVectorNumElements() ==
- DataTy->getVectorNumElements(),
+ DataTy->getVectorNumElements(),
"masked_load: vector mask must be same length as data", CS);
break;
}
@@ -4249,10 +4259,10 @@
// DataTy is the overloaded type
Type *DataTy = cast<PointerType>(Ptr->getType())->getElementType();
- Assert(DataTy == Val->getType(),
+ Assert(DataTy == Val->getType(),
"masked_store: storee must match pointer type", CS);
Assert(Mask->getType()->getVectorNumElements() ==
- DataTy->getVectorNumElements(),
+ DataTy->getVectorNumElements(),
"masked_store: vector mask must be same length as data", CS);
break;
}
diff --git a/llvm/test/Bitcode/attributes.ll b/llvm/test/Bitcode/attributes.ll
index 9fdf54b..18aa12c 100644
--- a/llvm/test/Bitcode/attributes.ll
+++ b/llvm/test/Bitcode/attributes.ll
@@ -204,7 +204,7 @@
; CHECK: define void @f34()
{
call void @nobuiltin() nobuiltin
-; CHECK: call void @nobuiltin() #33
+; CHECK: call void @nobuiltin() #34
ret void;
}
@@ -334,6 +334,11 @@
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 @@
; 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 }
diff --git a/llvm/test/Bitcode/compatibility.ll b/llvm/test/Bitcode/compatibility.ll
index b1f52bb..bc77e05 100644
--- a/llvm/test/Bitcode/compatibility.ll
+++ b/llvm/test/Bitcode/compatibility.ll
@@ -3,7 +3,7 @@
; Please update this file when making any IR changes. Information on the
; release process for this file is available here:
;
-; http://llvm.org/docs/DeveloperPolicy.html#ir-backwards-compatibility
+; http://llvm.org/docs/DeveloperPolicy.html#ir-backwards-compatibility
; RUN: llvm-as < %s | llvm-dis | llvm-as | llvm-dis | FileCheck %s
; RUN-PR24755: verify-uselistorder < %s
@@ -1246,7 +1246,7 @@
; 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 @@
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 @@
; 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
diff --git a/llvm/test/Verifier/speculatable-callsite-invalid.ll b/llvm/test/Verifier/speculatable-callsite-invalid.ll
new file mode 100644
index 0000000..f9a1adf
--- /dev/null
+++ b/llvm/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 }
diff --git a/llvm/test/Verifier/speculatable-callsite.ll b/llvm/test/Verifier/speculatable-callsite.ll
new file mode 100644
index 0000000..fafed83
--- /dev/null
+++ b/llvm/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 }
diff --git a/llvm/utils/TableGen/CodeGenIntrinsics.h b/llvm/utils/TableGen/CodeGenIntrinsics.h
index 6df0e6a..d126f27 100644
--- a/llvm/utils/TableGen/CodeGenIntrinsics.h
+++ b/llvm/utils/TableGen/CodeGenIntrinsics.h
@@ -123,6 +123,9 @@
/// 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;
diff --git a/llvm/utils/TableGen/CodeGenTarget.cpp b/llvm/utils/TableGen/CodeGenTarget.cpp
index 03c58ac..aad9005 100644
--- a/llvm/utils/TableGen/CodeGenTarget.cpp
+++ b/llvm/utils/TableGen/CodeGenTarget.cpp
@@ -515,6 +515,7 @@
isNoReturn = false;
isNoDuplicate = false;
isConvergent = false;
+ isSpeculatable = false;
if (DefName.size() <= 4 ||
std::string(DefName.begin(), DefName.begin() + 4) != "int_")
@@ -653,6 +654,8 @@
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));
diff --git a/llvm/utils/TableGen/IntrinsicEmitter.cpp b/llvm/utils/TableGen/IntrinsicEmitter.cpp
index e979b94..72f2dce 100644
--- a/llvm/utils/TableGen/IntrinsicEmitter.cpp
+++ b/llvm/utils/TableGen/IntrinsicEmitter.cpp
@@ -476,6 +476,9 @@
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 @@
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 @@
OS << "Attribute::Convergent";
addComma = true;
}
+ if (intrinsic.isSpeculatable) {
+ if (addComma)
+ OS << ",";
+ OS << "Attribute::Speculatable";
+ addComma = true;
+ }
switch (intrinsic.ModRef) {
case CodeGenIntrinsic::NoMem: