blob: 1d42edd8a94a866407f3f22268ece7e0586fd36d [file] [log] [blame]
//===- SveEmitter.cpp - Generate arm_sve.h for use with clang -*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This tablegen backend is responsible for emitting arm_sve.h, which includes
// a declaration and definition of each function specified by the ARM C/C++
// Language Extensions (ACLE).
//
// For details, visit:
// https://developer.arm.com/architectures/system-architectures/software-standards/acle
//
// Each SVE instruction is implemented in terms of 1 or more functions which
// are suffixed with the element type of the input vectors. Functions may be
// implemented in terms of generic vector operations such as +, *, -, etc. or
// by calling a __builtin_-prefixed function which will be handled by clang's
// CodeGen library.
//
// See also the documentation in include/clang/Basic/arm_sve.td.
//
//===----------------------------------------------------------------------===//
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/TableGen/Record.h"
#include "llvm/TableGen/Error.h"
#include <string>
#include <sstream>
#include <set>
#include <cctype>
#include <tuple>
using namespace llvm;
enum ClassKind {
ClassNone,
ClassS, // signed/unsigned, e.g., "_s8", "_u8" suffix
ClassG, // Overloaded name without type suffix
};
using TypeSpec = std::string;
namespace {
class ImmCheck {
unsigned Arg;
unsigned Kind;
unsigned ElementSizeInBits;
public:
ImmCheck(unsigned Arg, unsigned Kind, unsigned ElementSizeInBits = 0)
: Arg(Arg), Kind(Kind), ElementSizeInBits(ElementSizeInBits) {}
ImmCheck(const ImmCheck &Other) = default;
~ImmCheck() = default;
unsigned getArg() const { return Arg; }
unsigned getKind() const { return Kind; }
unsigned getElementSizeInBits() const { return ElementSizeInBits; }
};
class SVEType {
TypeSpec TS;
bool Float, Signed, Immediate, Void, Constant, Pointer, BFloat;
bool DefaultType, IsScalable, Predicate, PredicatePattern, PrefetchOp;
unsigned Bitwidth, ElementBitwidth, NumVectors;
public:
SVEType() : SVEType(TypeSpec(), 'v') {}
SVEType(TypeSpec TS, char CharMod)
: TS(TS), Float(false), Signed(true), Immediate(false), Void(false),
Constant(false), Pointer(false), BFloat(false), DefaultType(false),
IsScalable(true), Predicate(false), PredicatePattern(false),
PrefetchOp(false), Bitwidth(128), ElementBitwidth(~0U), NumVectors(1) {
if (!TS.empty())
applyTypespec();
applyModifier(CharMod);
}
bool isPointer() const { return Pointer; }
bool isVoidPointer() const { return Pointer && Void; }
bool isSigned() const { return Signed; }
bool isImmediate() const { return Immediate; }
bool isScalar() const { return NumVectors == 0; }
bool isVector() const { return NumVectors > 0; }
bool isScalableVector() const { return isVector() && IsScalable; }
bool isChar() const { return ElementBitwidth == 8; }
bool isVoid() const { return Void & !Pointer; }
bool isDefault() const { return DefaultType; }
bool isFloat() const { return Float && !BFloat; }
bool isBFloat() const { return BFloat && !Float; }
bool isFloatingPoint() const { return Float || BFloat; }
bool isInteger() const { return !isFloatingPoint() && !Predicate; }
bool isScalarPredicate() const {
return !isFloatingPoint() && Predicate && NumVectors == 0;
}
bool isPredicateVector() const { return Predicate; }
bool isPredicatePattern() const { return PredicatePattern; }
bool isPrefetchOp() const { return PrefetchOp; }
bool isConstant() const { return Constant; }
unsigned getElementSizeInBits() const { return ElementBitwidth; }
unsigned getNumVectors() const { return NumVectors; }
unsigned getNumElements() const {
assert(ElementBitwidth != ~0U);
return Bitwidth / ElementBitwidth;
}
unsigned getSizeInBits() const {
return Bitwidth;
}
/// Return the string representation of a type, which is an encoded
/// string for passing to the BUILTIN() macro in Builtins.def.
std::string builtin_str() const;
/// Return the C/C++ string representation of a type for use in the
/// arm_sve.h header file.
std::string str() const;
private:
/// Creates the type based on the typespec string in TS.
void applyTypespec();
/// Applies a prototype modifier to the type.
void applyModifier(char Mod);
};
class SVEEmitter;
/// The main grunt class. This represents an instantiation of an intrinsic with
/// a particular typespec and prototype.
class Intrinsic {
/// The unmangled name.
std::string Name;
/// The name of the corresponding LLVM IR intrinsic.
std::string LLVMName;
/// Intrinsic prototype.
std::string Proto;
/// The base type spec for this intrinsic.
TypeSpec BaseTypeSpec;
/// The base class kind. Most intrinsics use ClassS, which has full type
/// info for integers (_s32/_u32), or ClassG which is used for overloaded
/// intrinsics.
ClassKind Class;
/// The architectural #ifdef guard.
std::string Guard;
// The merge suffix such as _m, _x or _z.
std::string MergeSuffix;
/// The types of return value [0] and parameters [1..].
std::vector<SVEType> Types;
/// The "base type", which is VarType('d', BaseTypeSpec).
SVEType BaseType;
uint64_t Flags;
SmallVector<ImmCheck, 2> ImmChecks;
public:
Intrinsic(StringRef Name, StringRef Proto, uint64_t MergeTy,
StringRef MergeSuffix, uint64_t MemoryElementTy, StringRef LLVMName,
uint64_t Flags, ArrayRef<ImmCheck> ImmChecks, TypeSpec BT,
ClassKind Class, SVEEmitter &Emitter, StringRef Guard);
~Intrinsic()=default;
std::string getName() const { return Name; }
std::string getLLVMName() const { return LLVMName; }
std::string getProto() const { return Proto; }
TypeSpec getBaseTypeSpec() const { return BaseTypeSpec; }
SVEType getBaseType() const { return BaseType; }
StringRef getGuard() const { return Guard; }
ClassKind getClassKind() const { return Class; }
SVEType getReturnType() const { return Types[0]; }
ArrayRef<SVEType> getTypes() const { return Types; }
SVEType getParamType(unsigned I) const { return Types[I + 1]; }
unsigned getNumParams() const { return Proto.size() - 1; }
uint64_t getFlags() const { return Flags; }
bool isFlagSet(uint64_t Flag) const { return Flags & Flag;}
ArrayRef<ImmCheck> getImmChecks() const { return ImmChecks; }
/// Return the type string for a BUILTIN() macro in Builtins.def.
std::string getBuiltinTypeStr();
/// Return the name, mangled with type information. The name is mangled for
/// ClassS, so will add type suffixes such as _u32/_s32.
std::string getMangledName() const { return mangleName(ClassS); }
/// Returns true if the intrinsic is overloaded, in that it should also generate
/// a short form without the type-specifiers, e.g. 'svld1(..)' instead of
/// 'svld1_u32(..)'.
static bool isOverloadedIntrinsic(StringRef Name) {
auto BrOpen = Name.find("[");
auto BrClose = Name.find(']');
return BrOpen != std::string::npos && BrClose != std::string::npos;
}
/// Return true if the intrinsic takes a splat operand.
bool hasSplat() const {
// These prototype modifiers are described in arm_sve.td.
return Proto.find_first_of("ajfrKLR@") != std::string::npos;
}
/// Return the parameter index of the splat operand.
unsigned getSplatIdx() const {
// These prototype modifiers are described in arm_sve.td.
auto Idx = Proto.find_first_of("ajfrKLR@");
assert(Idx != std::string::npos && Idx > 0 &&
"Prototype has no splat operand");
return Idx - 1;
}
/// Emits the intrinsic declaration to the ostream.
void emitIntrinsic(raw_ostream &OS) const;
private:
std::string getMergeSuffix() const { return MergeSuffix; }
std::string mangleName(ClassKind LocalCK) const;
std::string replaceTemplatedArgs(std::string Name, TypeSpec TS,
std::string Proto) const;
};
class SVEEmitter {
private:
// The reinterpret builtins are generated separately because they
// need the cross product of all types (121 functions in total),
// which is inconvenient to specify in the arm_sve.td file or
// generate in CGBuiltin.cpp.
struct ReinterpretTypeInfo {
const char *Suffix;
const char *Type;
const char *BuiltinType;
};
SmallVector<ReinterpretTypeInfo, 12> Reinterprets = {
{"s8", "svint8_t", "q16Sc"}, {"s16", "svint16_t", "q8Ss"},
{"s32", "svint32_t", "q4Si"}, {"s64", "svint64_t", "q2SWi"},
{"u8", "svuint8_t", "q16Uc"}, {"u16", "svuint16_t", "q8Us"},
{"u32", "svuint32_t", "q4Ui"}, {"u64", "svuint64_t", "q2UWi"},
{"f16", "svfloat16_t", "q8h"}, {"bf16", "svbfloat16_t", "q8y"},
{"f32", "svfloat32_t", "q4f"}, {"f64", "svfloat64_t", "q2d"}};
RecordKeeper &Records;
llvm::StringMap<uint64_t> EltTypes;
llvm::StringMap<uint64_t> MemEltTypes;
llvm::StringMap<uint64_t> FlagTypes;
llvm::StringMap<uint64_t> MergeTypes;
llvm::StringMap<uint64_t> ImmCheckTypes;
public:
SVEEmitter(RecordKeeper &R) : Records(R) {
for (auto *RV : Records.getAllDerivedDefinitions("EltType"))
EltTypes[RV->getNameInitAsString()] = RV->getValueAsInt("Value");
for (auto *RV : Records.getAllDerivedDefinitions("MemEltType"))
MemEltTypes[RV->getNameInitAsString()] = RV->getValueAsInt("Value");
for (auto *RV : Records.getAllDerivedDefinitions("FlagType"))
FlagTypes[RV->getNameInitAsString()] = RV->getValueAsInt("Value");
for (auto *RV : Records.getAllDerivedDefinitions("MergeType"))
MergeTypes[RV->getNameInitAsString()] = RV->getValueAsInt("Value");
for (auto *RV : Records.getAllDerivedDefinitions("ImmCheckType"))
ImmCheckTypes[RV->getNameInitAsString()] = RV->getValueAsInt("Value");
}
/// Returns the enum value for the immcheck type
unsigned getEnumValueForImmCheck(StringRef C) const {
auto It = ImmCheckTypes.find(C);
if (It != ImmCheckTypes.end())
return It->getValue();
llvm_unreachable("Unsupported imm check");
}
/// Returns the enum value for the flag type
uint64_t getEnumValueForFlag(StringRef C) const {
auto Res = FlagTypes.find(C);
if (Res != FlagTypes.end())
return Res->getValue();
llvm_unreachable("Unsupported flag");
}
// Returns the SVETypeFlags for a given value and mask.
uint64_t encodeFlag(uint64_t V, StringRef MaskName) const {
auto It = FlagTypes.find(MaskName);
if (It != FlagTypes.end()) {
uint64_t Mask = It->getValue();
unsigned Shift = llvm::countTrailingZeros(Mask);
return (V << Shift) & Mask;
}
llvm_unreachable("Unsupported flag");
}
// Returns the SVETypeFlags for the given element type.
uint64_t encodeEltType(StringRef EltName) {
auto It = EltTypes.find(EltName);
if (It != EltTypes.end())
return encodeFlag(It->getValue(), "EltTypeMask");
llvm_unreachable("Unsupported EltType");
}
// Returns the SVETypeFlags for the given memory element type.
uint64_t encodeMemoryElementType(uint64_t MT) {
return encodeFlag(MT, "MemEltTypeMask");
}
// Returns the SVETypeFlags for the given merge type.
uint64_t encodeMergeType(uint64_t MT) {
return encodeFlag(MT, "MergeTypeMask");
}
// Returns the SVETypeFlags for the given splat operand.
unsigned encodeSplatOperand(unsigned SplatIdx) {
assert(SplatIdx < 7 && "SplatIdx out of encodable range");
return encodeFlag(SplatIdx + 1, "SplatOperandMask");
}
// Returns the SVETypeFlags value for the given SVEType.
uint64_t encodeTypeFlags(const SVEType &T);
/// Emit arm_sve.h.
void createHeader(raw_ostream &o);
/// Emit all the __builtin prototypes and code needed by Sema.
void createBuiltins(raw_ostream &o);
/// Emit all the information needed to map builtin -> LLVM IR intrinsic.
void createCodeGenMap(raw_ostream &o);
/// Emit all the range checks for the immediates.
void createRangeChecks(raw_ostream &o);
/// Create the SVETypeFlags used in CGBuiltins
void createTypeFlags(raw_ostream &o);
/// Create intrinsic and add it to \p Out
void createIntrinsic(Record *R, SmallVectorImpl<std::unique_ptr<Intrinsic>> &Out);
};
} // end anonymous namespace
//===----------------------------------------------------------------------===//
// Type implementation
//===----------------------------------------------------------------------===//
std::string SVEType::builtin_str() const {
std::string S;
if (isVoid())
return "v";
if (isVoidPointer())
S += "v";
else if (!isFloatingPoint())
switch (ElementBitwidth) {
case 1: S += "b"; break;
case 8: S += "c"; break;
case 16: S += "s"; break;
case 32: S += "i"; break;
case 64: S += "Wi"; break;
case 128: S += "LLLi"; break;
default: llvm_unreachable("Unhandled case!");
}
else if (isFloat())
switch (ElementBitwidth) {
case 16: S += "h"; break;
case 32: S += "f"; break;
case 64: S += "d"; break;
default: llvm_unreachable("Unhandled case!");
}
else if (isBFloat()) {
assert(ElementBitwidth == 16 && "Not a valid BFloat.");
S += "y";
}
if (!isFloatingPoint()) {
if ((isChar() || isPointer()) && !isVoidPointer()) {
// Make chars and typed pointers explicitly signed.
if (Signed)
S = "S" + S;
else if (!Signed)
S = "U" + S;
} else if (!isVoidPointer() && !Signed) {
S = "U" + S;
}
}
// Constant indices are "int", but have the "constant expression" modifier.
if (isImmediate()) {
assert(!isFloat() && "fp immediates are not supported");
S = "I" + S;
}
if (isScalar()) {
if (Constant) S += "C";
if (Pointer) S += "*";
return S;
}
assert(isScalableVector() && "Unsupported type");
return "q" + utostr(getNumElements() * NumVectors) + S;
}
std::string SVEType::str() const {
if (isPredicatePattern())
return "sv_pattern";
if (isPrefetchOp())
return "sv_prfop";
std::string S;
if (Void)
S += "void";
else {
if (isScalableVector())
S += "sv";
if (!Signed && !isFloatingPoint())
S += "u";
if (Float)
S += "float";
else if (isScalarPredicate() || isPredicateVector())
S += "bool";
else if (isBFloat())
S += "bfloat";
else
S += "int";
if (!isScalarPredicate() && !isPredicateVector())
S += utostr(ElementBitwidth);
if (!isScalableVector() && isVector())
S += "x" + utostr(getNumElements());
if (NumVectors > 1)
S += "x" + utostr(NumVectors);
if (!isScalarPredicate())
S += "_t";
}
if (Constant)
S += " const";
if (Pointer)
S += " *";
return S;
}
void SVEType::applyTypespec() {
for (char I : TS) {
switch (I) {
case 'P':
Predicate = true;
break;
case 'U':
Signed = false;
break;
case 'c':
ElementBitwidth = 8;
break;
case 's':
ElementBitwidth = 16;
break;
case 'i':
ElementBitwidth = 32;
break;
case 'l':
ElementBitwidth = 64;
break;
case 'h':
Float = true;
ElementBitwidth = 16;
break;
case 'f':
Float = true;
ElementBitwidth = 32;
break;
case 'd':
Float = true;
ElementBitwidth = 64;
break;
case 'b':
BFloat = true;
Float = false;
ElementBitwidth = 16;
break;
default:
llvm_unreachable("Unhandled type code!");
}
}
assert(ElementBitwidth != ~0U && "Bad element bitwidth!");
}
void SVEType::applyModifier(char Mod) {
switch (Mod) {
case '2':
NumVectors = 2;
break;
case '3':
NumVectors = 3;
break;
case '4':
NumVectors = 4;
break;
case 'v':
Void = true;
break;
case 'd':
DefaultType = true;
break;
case 'c':
Constant = true;
LLVM_FALLTHROUGH;
case 'p':
Pointer = true;
Bitwidth = ElementBitwidth;
NumVectors = 0;
break;
case 'e':
Signed = false;
ElementBitwidth /= 2;
break;
case 'h':
ElementBitwidth /= 2;
break;
case 'q':
ElementBitwidth /= 4;
break;
case 'b':
Signed = false;
Float = false;
BFloat = false;
ElementBitwidth /= 4;
break;
case 'o':
ElementBitwidth *= 4;
break;
case 'P':
Signed = true;
Float = false;
BFloat = false;
Predicate = true;
Bitwidth = 16;
ElementBitwidth = 1;
break;
case 's':
case 'a':
Bitwidth = ElementBitwidth;
NumVectors = 0;
break;
case 'R':
ElementBitwidth /= 2;
NumVectors = 0;
break;
case 'r':
ElementBitwidth /= 4;
NumVectors = 0;
break;
case '@':
Signed = false;
Float = false;
BFloat = false;
ElementBitwidth /= 4;
NumVectors = 0;
break;
case 'K':
Signed = true;
Float = false;
BFloat = false;
Bitwidth = ElementBitwidth;
NumVectors = 0;
break;
case 'L':
Signed = false;
Float = false;
BFloat = false;
Bitwidth = ElementBitwidth;
NumVectors = 0;
break;
case 'u':
Predicate = false;
Signed = false;
Float = false;
BFloat = false;
break;
case 'x':
Predicate = false;
Signed = true;
Float = false;
BFloat = false;
break;
case 'i':
Predicate = false;
Float = false;
BFloat = false;
ElementBitwidth = Bitwidth = 64;
NumVectors = 0;
Signed = false;
Immediate = true;
break;
case 'I':
Predicate = false;
Float = false;
BFloat = false;
ElementBitwidth = Bitwidth = 32;
NumVectors = 0;
Signed = true;
Immediate = true;
PredicatePattern = true;
break;
case 'J':
Predicate = false;
Float = false;
BFloat = false;
ElementBitwidth = Bitwidth = 32;
NumVectors = 0;
Signed = true;
Immediate = true;
PrefetchOp = true;
break;
case 'k':
Predicate = false;
Signed = true;
Float = false;
BFloat = false;
ElementBitwidth = Bitwidth = 32;
NumVectors = 0;
break;
case 'l':
Predicate = false;
Signed = true;
Float = false;
BFloat = false;
ElementBitwidth = Bitwidth = 64;
NumVectors = 0;
break;
case 'm':
Predicate = false;
Signed = false;
Float = false;
BFloat = false;
ElementBitwidth = Bitwidth = 32;
NumVectors = 0;
break;
case 'n':
Predicate = false;
Signed = false;
Float = false;
BFloat = false;
ElementBitwidth = Bitwidth = 64;
NumVectors = 0;
break;
case 'w':
ElementBitwidth = 64;
break;
case 'j':
ElementBitwidth = Bitwidth = 64;
NumVectors = 0;
break;
case 'f':
Signed = false;
ElementBitwidth = Bitwidth = 64;
NumVectors = 0;
break;
case 'g':
Signed = false;
Float = false;
BFloat = false;
ElementBitwidth = 64;
break;
case 't':
Signed = true;
Float = false;
BFloat = false;
ElementBitwidth = 32;
break;
case 'z':
Signed = false;
Float = false;
BFloat = false;
ElementBitwidth = 32;
break;
case 'O':
Predicate = false;
Float = true;
ElementBitwidth = 16;
break;
case 'M':
Predicate = false;
Float = true;
BFloat = false;
ElementBitwidth = 32;
break;
case 'N':
Predicate = false;
Float = true;
ElementBitwidth = 64;
break;
case 'Q':
Constant = true;
Pointer = true;
Void = true;
NumVectors = 0;
break;
case 'S':
Constant = true;
Pointer = true;
ElementBitwidth = Bitwidth = 8;
NumVectors = 0;
Signed = true;
break;
case 'W':
Constant = true;
Pointer = true;
ElementBitwidth = Bitwidth = 8;
NumVectors = 0;
Signed = false;
break;
case 'T':
Constant = true;
Pointer = true;
ElementBitwidth = Bitwidth = 16;
NumVectors = 0;
Signed = true;
break;
case 'X':
Constant = true;
Pointer = true;
ElementBitwidth = Bitwidth = 16;
NumVectors = 0;
Signed = false;
break;
case 'Y':
Constant = true;
Pointer = true;
ElementBitwidth = Bitwidth = 32;
NumVectors = 0;
Signed = false;
break;
case 'U':
Constant = true;
Pointer = true;
ElementBitwidth = Bitwidth = 32;
NumVectors = 0;
Signed = true;
break;
case 'A':
Pointer = true;
ElementBitwidth = Bitwidth = 8;
NumVectors = 0;
Signed = true;
break;
case 'B':
Pointer = true;
ElementBitwidth = Bitwidth = 16;
NumVectors = 0;
Signed = true;
break;
case 'C':
Pointer = true;
ElementBitwidth = Bitwidth = 32;
NumVectors = 0;
Signed = true;
break;
case 'D':
Pointer = true;
ElementBitwidth = Bitwidth = 64;
NumVectors = 0;
Signed = true;
break;
case 'E':
Pointer = true;
ElementBitwidth = Bitwidth = 8;
NumVectors = 0;
Signed = false;
break;
case 'F':
Pointer = true;
ElementBitwidth = Bitwidth = 16;
NumVectors = 0;
Signed = false;
break;
case 'G':
Pointer = true;
ElementBitwidth = Bitwidth = 32;
NumVectors = 0;
Signed = false;
break;
default:
llvm_unreachable("Unhandled character!");
}
}
//===----------------------------------------------------------------------===//
// Intrinsic implementation
//===----------------------------------------------------------------------===//
Intrinsic::Intrinsic(StringRef Name, StringRef Proto, uint64_t MergeTy,
StringRef MergeSuffix, uint64_t MemoryElementTy,
StringRef LLVMName, uint64_t Flags,
ArrayRef<ImmCheck> Checks, TypeSpec BT, ClassKind Class,
SVEEmitter &Emitter, StringRef Guard)
: Name(Name.str()), LLVMName(LLVMName), Proto(Proto.str()),
BaseTypeSpec(BT), Class(Class), Guard(Guard.str()),
MergeSuffix(MergeSuffix.str()), BaseType(BT, 'd'), Flags(Flags),
ImmChecks(Checks.begin(), Checks.end()) {
// Types[0] is the return value.
for (unsigned I = 0; I < Proto.size(); ++I) {
SVEType T(BaseTypeSpec, Proto[I]);
Types.push_back(T);
// Add range checks for immediates
if (I > 0) {
if (T.isPredicatePattern())
ImmChecks.emplace_back(
I - 1, Emitter.getEnumValueForImmCheck("ImmCheck0_31"));
else if (T.isPrefetchOp())
ImmChecks.emplace_back(
I - 1, Emitter.getEnumValueForImmCheck("ImmCheck0_13"));
}
}
// Set flags based on properties
this->Flags |= Emitter.encodeTypeFlags(BaseType);
this->Flags |= Emitter.encodeMemoryElementType(MemoryElementTy);
this->Flags |= Emitter.encodeMergeType(MergeTy);
if (hasSplat())
this->Flags |= Emitter.encodeSplatOperand(getSplatIdx());
}
std::string Intrinsic::getBuiltinTypeStr() {
std::string S = getReturnType().builtin_str();
for (unsigned I = 0; I < getNumParams(); ++I)
S += getParamType(I).builtin_str();
return S;
}
std::string Intrinsic::replaceTemplatedArgs(std::string Name, TypeSpec TS,
std::string Proto) const {
std::string Ret = Name;
while (Ret.find('{') != std::string::npos) {
size_t Pos = Ret.find('{');
size_t End = Ret.find('}');
unsigned NumChars = End - Pos + 1;
assert(NumChars == 3 && "Unexpected template argument");
SVEType T;
char C = Ret[Pos+1];
switch(C) {
default:
llvm_unreachable("Unknown predication specifier");
case 'd':
T = SVEType(TS, 'd');
break;
case '0':
case '1':
case '2':
case '3':
T = SVEType(TS, Proto[C - '0']);
break;
}
// Replace templated arg with the right suffix (e.g. u32)
std::string TypeCode;
if (T.isInteger())
TypeCode = T.isSigned() ? 's' : 'u';
else if (T.isPredicateVector())
TypeCode = 'b';
else if (T.isBFloat())
TypeCode = "bf";
else
TypeCode = 'f';
Ret.replace(Pos, NumChars, TypeCode + utostr(T.getElementSizeInBits()));
}
return Ret;
}
std::string Intrinsic::mangleName(ClassKind LocalCK) const {
std::string S = getName();
if (LocalCK == ClassG) {
// Remove the square brackets and everything in between.
while (S.find("[") != std::string::npos) {
auto Start = S.find("[");
auto End = S.find(']');
S.erase(Start, (End-Start)+1);
}
} else {
// Remove the square brackets.
while (S.find("[") != std::string::npos) {
auto BrPos = S.find('[');
if (BrPos != std::string::npos)
S.erase(BrPos, 1);
BrPos = S.find(']');
if (BrPos != std::string::npos)
S.erase(BrPos, 1);
}
}
// Replace all {d} like expressions with e.g. 'u32'
return replaceTemplatedArgs(S, getBaseTypeSpec(), getProto()) +
getMergeSuffix();
}
void Intrinsic::emitIntrinsic(raw_ostream &OS) const {
// Use the preprocessor to
if (getClassKind() != ClassG || getProto().size() <= 1) {
OS << "#define " << mangleName(getClassKind())
<< "(...) __builtin_sve_" << mangleName(ClassS)
<< "(__VA_ARGS__)\n";
} else {
std::string FullName = mangleName(ClassS);
std::string ProtoName = mangleName(ClassG);
OS << "__aio __attribute__((__clang_arm_builtin_alias("
<< "__builtin_sve_" << FullName << ")))\n";
OS << getTypes()[0].str() << " " << ProtoName << "(";
for (unsigned I = 0; I < getTypes().size() - 1; ++I) {
if (I != 0)
OS << ", ";
OS << getTypes()[I + 1].str();
}
OS << ");\n";
}
}
//===----------------------------------------------------------------------===//
// SVEEmitter implementation
//===----------------------------------------------------------------------===//
uint64_t SVEEmitter::encodeTypeFlags(const SVEType &T) {
if (T.isFloat()) {
switch (T.getElementSizeInBits()) {
case 16:
return encodeEltType("EltTyFloat16");
case 32:
return encodeEltType("EltTyFloat32");
case 64:
return encodeEltType("EltTyFloat64");
default:
llvm_unreachable("Unhandled float element bitwidth!");
}
}
if (T.isBFloat()) {
assert(T.getElementSizeInBits() == 16 && "Not a valid BFloat.");
return encodeEltType("EltTyBFloat16");
}
if (T.isPredicateVector()) {
switch (T.getElementSizeInBits()) {
case 8:
return encodeEltType("EltTyBool8");
case 16:
return encodeEltType("EltTyBool16");
case 32:
return encodeEltType("EltTyBool32");
case 64:
return encodeEltType("EltTyBool64");
default:
llvm_unreachable("Unhandled predicate element bitwidth!");
}
}
switch (T.getElementSizeInBits()) {
case 8:
return encodeEltType("EltTyInt8");
case 16:
return encodeEltType("EltTyInt16");
case 32:
return encodeEltType("EltTyInt32");
case 64:
return encodeEltType("EltTyInt64");
default:
llvm_unreachable("Unhandled integer element bitwidth!");
}
}
void SVEEmitter::createIntrinsic(
Record *R, SmallVectorImpl<std::unique_ptr<Intrinsic>> &Out) {
StringRef Name = R->getValueAsString("Name");
StringRef Proto = R->getValueAsString("Prototype");
StringRef Types = R->getValueAsString("Types");
StringRef Guard = R->getValueAsString("ArchGuard");
StringRef LLVMName = R->getValueAsString("LLVMIntrinsic");
uint64_t Merge = R->getValueAsInt("Merge");
StringRef MergeSuffix = R->getValueAsString("MergeSuffix");
uint64_t MemEltType = R->getValueAsInt("MemEltType");
std::vector<Record*> FlagsList = R->getValueAsListOfDefs("Flags");
std::vector<Record*> ImmCheckList = R->getValueAsListOfDefs("ImmChecks");
int64_t Flags = 0;
for (auto FlagRec : FlagsList)
Flags |= FlagRec->getValueAsInt("Value");
// Create a dummy TypeSpec for non-overloaded builtins.
if (Types.empty()) {
assert((Flags & getEnumValueForFlag("IsOverloadNone")) &&
"Expect TypeSpec for overloaded builtin!");
Types = "i";
}
// Extract type specs from string
SmallVector<TypeSpec, 8> TypeSpecs;
TypeSpec Acc;
for (char I : Types) {
Acc.push_back(I);
if (islower(I)) {
TypeSpecs.push_back(TypeSpec(Acc));
Acc.clear();
}
}
// Remove duplicate type specs.
llvm::sort(TypeSpecs);
TypeSpecs.erase(std::unique(TypeSpecs.begin(), TypeSpecs.end()),
TypeSpecs.end());
// Create an Intrinsic for each type spec.
for (auto TS : TypeSpecs) {
// Collate a list of range/option checks for the immediates.
SmallVector<ImmCheck, 2> ImmChecks;
for (auto *R : ImmCheckList) {
int64_t Arg = R->getValueAsInt("Arg");
int64_t EltSizeArg = R->getValueAsInt("EltSizeArg");
int64_t Kind = R->getValueAsDef("Kind")->getValueAsInt("Value");
assert(Arg >= 0 && Kind >= 0 && "Arg and Kind must be nonnegative");
unsigned ElementSizeInBits = 0;
if (EltSizeArg >= 0)
ElementSizeInBits =
SVEType(TS, Proto[EltSizeArg + /* offset by return arg */ 1])
.getElementSizeInBits();
ImmChecks.push_back(ImmCheck(Arg, Kind, ElementSizeInBits));
}
Out.push_back(std::make_unique<Intrinsic>(
Name, Proto, Merge, MergeSuffix, MemEltType, LLVMName, Flags, ImmChecks,
TS, ClassS, *this, Guard));
// Also generate the short-form (e.g. svadd_m) for the given type-spec.
if (Intrinsic::isOverloadedIntrinsic(Name))
Out.push_back(std::make_unique<Intrinsic>(
Name, Proto, Merge, MergeSuffix, MemEltType, LLVMName, Flags,
ImmChecks, TS, ClassG, *this, Guard));
}
}
void SVEEmitter::createHeader(raw_ostream &OS) {
OS << "/*===---- arm_sve.h - ARM SVE intrinsics "
"-----------------------------------===\n"
" *\n"
" *\n"
" * Part of the LLVM Project, under the Apache License v2.0 with LLVM "
"Exceptions.\n"
" * See https://llvm.org/LICENSE.txt for license information.\n"
" * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception\n"
" *\n"
" *===-----------------------------------------------------------------"
"------===\n"
" */\n\n";
OS << "#ifndef __ARM_SVE_H\n";
OS << "#define __ARM_SVE_H\n\n";
OS << "#if !defined(__ARM_FEATURE_SVE)\n";
OS << "#error \"SVE support not enabled\"\n";
OS << "#else\n\n";
OS << "#if !defined(__LITTLE_ENDIAN__)\n";
OS << "#error \"Big endian is currently not supported for arm_sve.h\"\n";
OS << "#endif\n";
OS << "#include <stdint.h>\n\n";
OS << "#ifdef __cplusplus\n";
OS << "extern \"C\" {\n";
OS << "#else\n";
OS << "#include <stdbool.h>\n";
OS << "#endif\n\n";
OS << "typedef __fp16 float16_t;\n";
OS << "typedef float float32_t;\n";
OS << "typedef double float64_t;\n";
OS << "typedef __SVInt8_t svint8_t;\n";
OS << "typedef __SVInt16_t svint16_t;\n";
OS << "typedef __SVInt32_t svint32_t;\n";
OS << "typedef __SVInt64_t svint64_t;\n";
OS << "typedef __SVUint8_t svuint8_t;\n";
OS << "typedef __SVUint16_t svuint16_t;\n";
OS << "typedef __SVUint32_t svuint32_t;\n";
OS << "typedef __SVUint64_t svuint64_t;\n";
OS << "typedef __SVFloat16_t svfloat16_t;\n\n";
OS << "#if defined(__ARM_FEATURE_SVE_BF16) && "
"!defined(__ARM_FEATURE_BF16_SCALAR_ARITHMETIC)\n";
OS << "#error \"__ARM_FEATURE_BF16_SCALAR_ARITHMETIC must be defined when "
"__ARM_FEATURE_SVE_BF16 is defined\"\n";
OS << "#endif\n\n";
OS << "#if defined(__ARM_FEATURE_SVE_BF16)\n";
OS << "typedef __SVBFloat16_t svbfloat16_t;\n";
OS << "#endif\n\n";
OS << "#if defined(__ARM_FEATURE_BF16_SCALAR_ARITHMETIC)\n";
OS << "#include <arm_bf16.h>\n";
OS << "typedef __bf16 bfloat16_t;\n";
OS << "#endif\n\n";
OS << "typedef __SVFloat32_t svfloat32_t;\n";
OS << "typedef __SVFloat64_t svfloat64_t;\n";
OS << "typedef __clang_svint8x2_t svint8x2_t;\n";
OS << "typedef __clang_svint16x2_t svint16x2_t;\n";
OS << "typedef __clang_svint32x2_t svint32x2_t;\n";
OS << "typedef __clang_svint64x2_t svint64x2_t;\n";
OS << "typedef __clang_svuint8x2_t svuint8x2_t;\n";
OS << "typedef __clang_svuint16x2_t svuint16x2_t;\n";
OS << "typedef __clang_svuint32x2_t svuint32x2_t;\n";
OS << "typedef __clang_svuint64x2_t svuint64x2_t;\n";
OS << "typedef __clang_svfloat16x2_t svfloat16x2_t;\n";
OS << "typedef __clang_svfloat32x2_t svfloat32x2_t;\n";
OS << "typedef __clang_svfloat64x2_t svfloat64x2_t;\n";
OS << "typedef __clang_svint8x3_t svint8x3_t;\n";
OS << "typedef __clang_svint16x3_t svint16x3_t;\n";
OS << "typedef __clang_svint32x3_t svint32x3_t;\n";
OS << "typedef __clang_svint64x3_t svint64x3_t;\n";
OS << "typedef __clang_svuint8x3_t svuint8x3_t;\n";
OS << "typedef __clang_svuint16x3_t svuint16x3_t;\n";
OS << "typedef __clang_svuint32x3_t svuint32x3_t;\n";
OS << "typedef __clang_svuint64x3_t svuint64x3_t;\n";
OS << "typedef __clang_svfloat16x3_t svfloat16x3_t;\n";
OS << "typedef __clang_svfloat32x3_t svfloat32x3_t;\n";
OS << "typedef __clang_svfloat64x3_t svfloat64x3_t;\n";
OS << "typedef __clang_svint8x4_t svint8x4_t;\n";
OS << "typedef __clang_svint16x4_t svint16x4_t;\n";
OS << "typedef __clang_svint32x4_t svint32x4_t;\n";
OS << "typedef __clang_svint64x4_t svint64x4_t;\n";
OS << "typedef __clang_svuint8x4_t svuint8x4_t;\n";
OS << "typedef __clang_svuint16x4_t svuint16x4_t;\n";
OS << "typedef __clang_svuint32x4_t svuint32x4_t;\n";
OS << "typedef __clang_svuint64x4_t svuint64x4_t;\n";
OS << "typedef __clang_svfloat16x4_t svfloat16x4_t;\n";
OS << "typedef __clang_svfloat32x4_t svfloat32x4_t;\n";
OS << "typedef __clang_svfloat64x4_t svfloat64x4_t;\n";
OS << "typedef __SVBool_t svbool_t;\n\n";
OS << "#ifdef __ARM_FEATURE_SVE_BF16\n";
OS << "typedef __clang_svbfloat16x2_t svbfloat16x2_t;\n";
OS << "typedef __clang_svbfloat16x3_t svbfloat16x3_t;\n";
OS << "typedef __clang_svbfloat16x4_t svbfloat16x4_t;\n";
OS << "#endif\n";
OS << "typedef enum\n";
OS << "{\n";
OS << " SV_POW2 = 0,\n";
OS << " SV_VL1 = 1,\n";
OS << " SV_VL2 = 2,\n";
OS << " SV_VL3 = 3,\n";
OS << " SV_VL4 = 4,\n";
OS << " SV_VL5 = 5,\n";
OS << " SV_VL6 = 6,\n";
OS << " SV_VL7 = 7,\n";
OS << " SV_VL8 = 8,\n";
OS << " SV_VL16 = 9,\n";
OS << " SV_VL32 = 10,\n";
OS << " SV_VL64 = 11,\n";
OS << " SV_VL128 = 12,\n";
OS << " SV_VL256 = 13,\n";
OS << " SV_MUL4 = 29,\n";
OS << " SV_MUL3 = 30,\n";
OS << " SV_ALL = 31\n";
OS << "} sv_pattern;\n\n";
OS << "typedef enum\n";
OS << "{\n";
OS << " SV_PLDL1KEEP = 0,\n";
OS << " SV_PLDL1STRM = 1,\n";
OS << " SV_PLDL2KEEP = 2,\n";
OS << " SV_PLDL2STRM = 3,\n";
OS << " SV_PLDL3KEEP = 4,\n";
OS << " SV_PLDL3STRM = 5,\n";
OS << " SV_PSTL1KEEP = 8,\n";
OS << " SV_PSTL1STRM = 9,\n";
OS << " SV_PSTL2KEEP = 10,\n";
OS << " SV_PSTL2STRM = 11,\n";
OS << " SV_PSTL3KEEP = 12,\n";
OS << " SV_PSTL3STRM = 13\n";
OS << "} sv_prfop;\n\n";
OS << "/* Function attributes */\n";
OS << "#define __aio static inline __attribute__((__always_inline__, "
"__nodebug__, __overloadable__))\n\n";
// Add reinterpret functions.
for (auto ShortForm : { false, true } )
for (const ReinterpretTypeInfo &From : Reinterprets)
for (const ReinterpretTypeInfo &To : Reinterprets) {
const bool IsBFloat = StringRef(From.Suffix).equals("bf16") ||
StringRef(To.Suffix).equals("bf16");
if (IsBFloat)
OS << "#if defined(__ARM_FEATURE_SVE_BF16)\n";
if (ShortForm) {
OS << "__aio " << From.Type << " svreinterpret_" << From.Suffix;
OS << "(" << To.Type << " op) {\n";
OS << " return __builtin_sve_reinterpret_" << From.Suffix << "_"
<< To.Suffix << "(op);\n";
OS << "}\n\n";
} else
OS << "#define svreinterpret_" << From.Suffix << "_" << To.Suffix
<< "(...) __builtin_sve_reinterpret_" << From.Suffix << "_"
<< To.Suffix << "(__VA_ARGS__)\n";
if (IsBFloat)
OS << "#endif /* #if defined(__ARM_FEATURE_SVE_BF16) */\n";
}
SmallVector<std::unique_ptr<Intrinsic>, 128> Defs;
std::vector<Record *> RV = Records.getAllDerivedDefinitions("Inst");
for (auto *R : RV)
createIntrinsic(R, Defs);
// Sort intrinsics in header file by following order/priority:
// - Architectural guard (i.e. does it require SVE2 or SVE2_AES)
// - Class (is intrinsic overloaded or not)
// - Intrinsic name
std::stable_sort(
Defs.begin(), Defs.end(), [](const std::unique_ptr<Intrinsic> &A,
const std::unique_ptr<Intrinsic> &B) {
auto ToTuple = [](const std::unique_ptr<Intrinsic> &I) {
return std::make_tuple(I->getGuard(), (unsigned)I->getClassKind(), I->getName());
};
return ToTuple(A) < ToTuple(B);
});
StringRef InGuard = "";
for (auto &I : Defs) {
// Emit #endif/#if pair if needed.
if (I->getGuard() != InGuard) {
if (!InGuard.empty())
OS << "#endif //" << InGuard << "\n";
InGuard = I->getGuard();
if (!InGuard.empty())
OS << "\n#if " << InGuard << "\n";
}
// Actually emit the intrinsic declaration.
I->emitIntrinsic(OS);
}
if (!InGuard.empty())
OS << "#endif //" << InGuard << "\n";
OS << "#if defined(__ARM_FEATURE_SVE_BF16)\n";
OS << "#define svcvtnt_bf16_x svcvtnt_bf16_m\n";
OS << "#define svcvtnt_bf16_f32_x svcvtnt_bf16_f32_m\n";
OS << "#endif /*__ARM_FEATURE_SVE_BF16 */\n\n";
OS << "#if defined(__ARM_FEATURE_SVE2)\n";
OS << "#define svcvtnt_f16_x svcvtnt_f16_m\n";
OS << "#define svcvtnt_f16_f32_x svcvtnt_f16_f32_m\n";
OS << "#define svcvtnt_f32_x svcvtnt_f32_m\n";
OS << "#define svcvtnt_f32_f64_x svcvtnt_f32_f64_m\n\n";
OS << "#define svcvtxnt_f32_x svcvtxnt_f32_m\n";
OS << "#define svcvtxnt_f32_f64_x svcvtxnt_f32_f64_m\n\n";
OS << "#endif /*__ARM_FEATURE_SVE2 */\n\n";
OS << "#ifdef __cplusplus\n";
OS << "} // extern \"C\"\n";
OS << "#endif\n\n";
OS << "#endif /*__ARM_FEATURE_SVE */\n\n";
OS << "#endif /* __ARM_SVE_H */\n";
}
void SVEEmitter::createBuiltins(raw_ostream &OS) {
std::vector<Record *> RV = Records.getAllDerivedDefinitions("Inst");
SmallVector<std::unique_ptr<Intrinsic>, 128> Defs;
for (auto *R : RV)
createIntrinsic(R, Defs);
// The mappings must be sorted based on BuiltinID.
llvm::sort(Defs, [](const std::unique_ptr<Intrinsic> &A,
const std::unique_ptr<Intrinsic> &B) {
return A->getMangledName() < B->getMangledName();
});
OS << "#ifdef GET_SVE_BUILTINS\n";
for (auto &Def : Defs) {
// Only create BUILTINs for non-overloaded intrinsics, as overloaded
// declarations only live in the header file.
if (Def->getClassKind() != ClassG)
OS << "BUILTIN(__builtin_sve_" << Def->getMangledName() << ", \""
<< Def->getBuiltinTypeStr() << "\", \"n\")\n";
}
// Add reinterpret builtins
for (const ReinterpretTypeInfo &From : Reinterprets)
for (const ReinterpretTypeInfo &To : Reinterprets)
OS << "BUILTIN(__builtin_sve_reinterpret_" << From.Suffix << "_"
<< To.Suffix << +", \"" << From.BuiltinType << To.BuiltinType
<< "\", \"n\")\n";
OS << "#endif\n\n";
}
void SVEEmitter::createCodeGenMap(raw_ostream &OS) {
std::vector<Record *> RV = Records.getAllDerivedDefinitions("Inst");
SmallVector<std::unique_ptr<Intrinsic>, 128> Defs;
for (auto *R : RV)
createIntrinsic(R, Defs);
// The mappings must be sorted based on BuiltinID.
llvm::sort(Defs, [](const std::unique_ptr<Intrinsic> &A,
const std::unique_ptr<Intrinsic> &B) {
return A->getMangledName() < B->getMangledName();
});
OS << "#ifdef GET_SVE_LLVM_INTRINSIC_MAP\n";
for (auto &Def : Defs) {
// Builtins only exist for non-overloaded intrinsics, overloaded
// declarations only live in the header file.
if (Def->getClassKind() == ClassG)
continue;
uint64_t Flags = Def->getFlags();
auto FlagString = std::to_string(Flags);
std::string LLVMName = Def->getLLVMName();
std::string Builtin = Def->getMangledName();
if (!LLVMName.empty())
OS << "SVEMAP1(" << Builtin << ", " << LLVMName << ", " << FlagString
<< "),\n";
else
OS << "SVEMAP2(" << Builtin << ", " << FlagString << "),\n";
}
OS << "#endif\n\n";
}
void SVEEmitter::createRangeChecks(raw_ostream &OS) {
std::vector<Record *> RV = Records.getAllDerivedDefinitions("Inst");
SmallVector<std::unique_ptr<Intrinsic>, 128> Defs;
for (auto *R : RV)
createIntrinsic(R, Defs);
// The mappings must be sorted based on BuiltinID.
llvm::sort(Defs, [](const std::unique_ptr<Intrinsic> &A,
const std::unique_ptr<Intrinsic> &B) {
return A->getMangledName() < B->getMangledName();
});
OS << "#ifdef GET_SVE_IMMEDIATE_CHECK\n";
// Ensure these are only emitted once.
std::set<std::string> Emitted;
for (auto &Def : Defs) {
if (Emitted.find(Def->getMangledName()) != Emitted.end() ||
Def->getImmChecks().empty())
continue;
OS << "case SVE::BI__builtin_sve_" << Def->getMangledName() << ":\n";
for (auto &Check : Def->getImmChecks())
OS << "ImmChecks.push_back(std::make_tuple(" << Check.getArg() << ", "
<< Check.getKind() << ", " << Check.getElementSizeInBits() << "));\n";
OS << " break;\n";
Emitted.insert(Def->getMangledName());
}
OS << "#endif\n\n";
}
/// Create the SVETypeFlags used in CGBuiltins
void SVEEmitter::createTypeFlags(raw_ostream &OS) {
OS << "#ifdef LLVM_GET_SVE_TYPEFLAGS\n";
for (auto &KV : FlagTypes)
OS << "const uint64_t " << KV.getKey() << " = " << KV.getValue() << ";\n";
OS << "#endif\n\n";
OS << "#ifdef LLVM_GET_SVE_ELTTYPES\n";
for (auto &KV : EltTypes)
OS << " " << KV.getKey() << " = " << KV.getValue() << ",\n";
OS << "#endif\n\n";
OS << "#ifdef LLVM_GET_SVE_MEMELTTYPES\n";
for (auto &KV : MemEltTypes)
OS << " " << KV.getKey() << " = " << KV.getValue() << ",\n";
OS << "#endif\n\n";
OS << "#ifdef LLVM_GET_SVE_MERGETYPES\n";
for (auto &KV : MergeTypes)
OS << " " << KV.getKey() << " = " << KV.getValue() << ",\n";
OS << "#endif\n\n";
OS << "#ifdef LLVM_GET_SVE_IMMCHECKTYPES\n";
for (auto &KV : ImmCheckTypes)
OS << " " << KV.getKey() << " = " << KV.getValue() << ",\n";
OS << "#endif\n\n";
}
namespace clang {
void EmitSveHeader(RecordKeeper &Records, raw_ostream &OS) {
SVEEmitter(Records).createHeader(OS);
}
void EmitSveBuiltins(RecordKeeper &Records, raw_ostream &OS) {
SVEEmitter(Records).createBuiltins(OS);
}
void EmitSveBuiltinCG(RecordKeeper &Records, raw_ostream &OS) {
SVEEmitter(Records).createCodeGenMap(OS);
}
void EmitSveRangeChecks(RecordKeeper &Records, raw_ostream &OS) {
SVEEmitter(Records).createRangeChecks(OS);
}
void EmitSveTypeFlags(RecordKeeper &Records, raw_ostream &OS) {
SVEEmitter(Records).createTypeFlags(OS);
}
} // End namespace clang