Clean up type flags for overloaded Neon builtins. No functional change.

This patch just adds a simple NeonTypeFlags class to replace the various
hardcoded constants that had been used until now.  Unfortunately I couldn't
figure out a good way to avoid duplicating that class between clang and
TableGen, but since it's small and rarely changes, that's not so bad.

llvm-svn: 144054
This commit is contained in:
Bob Wilson 2011-11-08 01:16:11 +00:00
parent 1654bb22ca
commit 98bc98caa8
5 changed files with 150 additions and 65 deletions

View File

@ -56,6 +56,46 @@ namespace clang {
};
}
/// NeonTypeFlags - Flags to identify the types for overloaded Neon
/// builtins. These must be kept in sync with the flags in
/// utils/TableGen/NeonEmitter.h.
class NeonTypeFlags {
enum {
EltTypeMask = 0xf,
UnsignedFlag = 0x10,
QuadFlag = 0x20
};
uint32_t Flags;
public:
enum EltType {
Int8,
Int16,
Int32,
Int64,
Poly8,
Poly16,
Float16,
Float32
};
NeonTypeFlags(unsigned F) : Flags(F) {}
NeonTypeFlags(EltType ET, bool IsUnsigned, bool IsQuad) : Flags(ET) {
if (IsUnsigned)
Flags |= UnsignedFlag;
if (IsQuad)
Flags |= QuadFlag;
}
EltType getEltType() const { return (EltType)(Flags & EltTypeMask); }
bool isPoly() const {
EltType ET = getEltType();
return ET == Poly8 || ET == Poly16;
}
bool isUnsigned() const { return (Flags & UnsignedFlag) != 0; }
bool isQuad() const { return (Flags & QuadFlag) != 0; }
};
} // end namespace clang.
#endif

View File

@ -1186,17 +1186,23 @@ Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID,
}
}
static llvm::VectorType *GetNeonType(LLVMContext &C, unsigned type, bool q) {
switch (type) {
default: break;
case 0:
case 5: return llvm::VectorType::get(llvm::Type::getInt8Ty(C), 8 << (int)q);
case 6:
case 7:
case 1: return llvm::VectorType::get(llvm::Type::getInt16Ty(C),4 << (int)q);
case 2: return llvm::VectorType::get(llvm::Type::getInt32Ty(C),2 << (int)q);
case 3: return llvm::VectorType::get(llvm::Type::getInt64Ty(C),1 << (int)q);
case 4: return llvm::VectorType::get(llvm::Type::getFloatTy(C),2 << (int)q);
static llvm::VectorType *GetNeonType(LLVMContext &C, NeonTypeFlags Type) {
int IsQuad = Type.isQuad();
switch (Type.getEltType()) {
default: break;
case NeonTypeFlags::Int8:
case NeonTypeFlags::Poly8:
return llvm::VectorType::get(llvm::Type::getInt8Ty(C), 8 << IsQuad);
case NeonTypeFlags::Int16:
case NeonTypeFlags::Poly16:
case NeonTypeFlags::Float16:
return llvm::VectorType::get(llvm::Type::getInt16Ty(C), 4 << IsQuad);
case NeonTypeFlags::Int32:
return llvm::VectorType::get(llvm::Type::getInt32Ty(C), 2 << IsQuad);
case NeonTypeFlags::Int64:
return llvm::VectorType::get(llvm::Type::getInt64Ty(C), 1 << IsQuad);
case NeonTypeFlags::Float32:
return llvm::VectorType::get(llvm::Type::getFloatTy(C), 2 << IsQuad);
};
return 0;
}
@ -1363,14 +1369,12 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
}
// Determine the type of this overloaded NEON intrinsic.
unsigned type = Result.getZExtValue();
bool usgn = type & 0x08;
bool quad = type & 0x10;
bool poly = (type & 0x7) == 5 || (type & 0x7) == 6;
(void)poly; // Only used in assert()s.
NeonTypeFlags Type(Result.getZExtValue());
bool usgn = Type.isUnsigned();
bool quad = Type.isQuad();
bool rightShift = false;
llvm::VectorType *VTy = GetNeonType(getLLVMContext(), type & 0x7, quad);
llvm::VectorType *VTy = GetNeonType(getLLVMContext(), Type);
llvm::Type *Ty = VTy;
if (!Ty)
return 0;
@ -1429,34 +1433,43 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
return EmitNeonCall(F, Ops, "vcnt");
}
case ARM::BI__builtin_neon_vcvt_f16_v: {
assert((type & 0x7) == 7 && !quad && "unexpected vcvt_f16_v builtin");
assert(Type.getEltType() == NeonTypeFlags::Float16 && !quad &&
"unexpected vcvt_f16_v builtin");
Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vcvtfp2hf);
return EmitNeonCall(F, Ops, "vcvt");
}
case ARM::BI__builtin_neon_vcvt_f32_f16: {
assert((type & 0x7) == 7 && !quad && "unexpected vcvt_f32_f16 builtin");
assert(Type.getEltType() == NeonTypeFlags::Float16 && !quad &&
"unexpected vcvt_f32_f16 builtin");
Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vcvthf2fp);
return EmitNeonCall(F, Ops, "vcvt");
}
case ARM::BI__builtin_neon_vcvt_f32_v:
case ARM::BI__builtin_neon_vcvtq_f32_v: {
case ARM::BI__builtin_neon_vcvtq_f32_v:
Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
Ty = GetNeonType(getLLVMContext(), 4, quad);
Ty = GetNeonType(getLLVMContext(),
NeonTypeFlags(NeonTypeFlags::Float32, false, quad));
return usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt")
: Builder.CreateSIToFP(Ops[0], Ty, "vcvt");
}
case ARM::BI__builtin_neon_vcvt_s32_v:
case ARM::BI__builtin_neon_vcvt_u32_v:
case ARM::BI__builtin_neon_vcvtq_s32_v:
case ARM::BI__builtin_neon_vcvtq_u32_v: {
Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(getLLVMContext(), 4, quad));
llvm::Type *FloatTy =
GetNeonType(getLLVMContext(),
NeonTypeFlags(NeonTypeFlags::Float32, false, quad));
Ops[0] = Builder.CreateBitCast(Ops[0], FloatTy);
return usgn ? Builder.CreateFPToUI(Ops[0], Ty, "vcvt")
: Builder.CreateFPToSI(Ops[0], Ty, "vcvt");
}
case ARM::BI__builtin_neon_vcvt_n_f32_v:
case ARM::BI__builtin_neon_vcvtq_n_f32_v: {
llvm::Type *Tys[2] = { GetNeonType(getLLVMContext(), 4, quad), Ty };
Int = usgn ? Intrinsic::arm_neon_vcvtfxu2fp : Intrinsic::arm_neon_vcvtfxs2fp;
llvm::Type *FloatTy =
GetNeonType(getLLVMContext(),
NeonTypeFlags(NeonTypeFlags::Float32, false, quad));
llvm::Type *Tys[2] = { FloatTy, Ty };
Int = usgn ? Intrinsic::arm_neon_vcvtfxu2fp
: Intrinsic::arm_neon_vcvtfxs2fp;
Function *F = CGM.getIntrinsic(Int, Tys);
return EmitNeonCall(F, Ops, "vcvt_n");
}
@ -1464,8 +1477,12 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
case ARM::BI__builtin_neon_vcvt_n_u32_v:
case ARM::BI__builtin_neon_vcvtq_n_s32_v:
case ARM::BI__builtin_neon_vcvtq_n_u32_v: {
llvm::Type *Tys[2] = { Ty, GetNeonType(getLLVMContext(), 4, quad) };
Int = usgn ? Intrinsic::arm_neon_vcvtfp2fxu : Intrinsic::arm_neon_vcvtfp2fxs;
llvm::Type *FloatTy =
GetNeonType(getLLVMContext(),
NeonTypeFlags(NeonTypeFlags::Float32, false, quad));
llvm::Type *Tys[2] = { Ty, FloatTy };
Int = usgn ? Intrinsic::arm_neon_vcvtfp2fxu
: Intrinsic::arm_neon_vcvtfp2fxs;
Function *F = CGM.getIntrinsic(Int, Tys);
return EmitNeonCall(F, Ops, "vcvt_n");
}
@ -1656,12 +1673,12 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
}
case ARM::BI__builtin_neon_vmul_v:
case ARM::BI__builtin_neon_vmulq_v:
assert(poly && "vmul builtin only supported for polynomial types");
assert(Type.isPoly() && "vmul builtin only supported for polynomial types");
return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vmulp, Ty),
Ops, "vmul");
case ARM::BI__builtin_neon_vmull_v:
Int = usgn ? Intrinsic::arm_neon_vmullu : Intrinsic::arm_neon_vmulls;
Int = poly ? (unsigned)Intrinsic::arm_neon_vmullp : Int;
Int = Type.isPoly() ? (unsigned)Intrinsic::arm_neon_vmullp : Int;
return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull");
case ARM::BI__builtin_neon_vpadal_v:
case ARM::BI__builtin_neon_vpadalq_v: {

View File

@ -245,27 +245,25 @@ Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
// Get the valid immediate range for the specified NEON type code.
static unsigned RFT(unsigned t, bool shift = false) {
bool quad = t & 0x10;
switch (t & 0x7) {
case 0: // i8
return shift ? 7 : (8 << (int)quad) - 1;
case 1: // i16
return shift ? 15 : (4 << (int)quad) - 1;
case 2: // i32
return shift ? 31 : (2 << (int)quad) - 1;
case 3: // i64
return shift ? 63 : (1 << (int)quad) - 1;
case 4: // f32
assert(!shift && "cannot shift float types!");
return (2 << (int)quad) - 1;
case 5: // poly8
return shift ? 7 : (8 << (int)quad) - 1;
case 6: // poly16
return shift ? 15 : (4 << (int)quad) - 1;
case 7: // float16
assert(!shift && "cannot shift float types!");
return (4 << (int)quad) - 1;
NeonTypeFlags Type(t);
int IsQuad = Type.isQuad();
switch (Type.getEltType()) {
case NeonTypeFlags::Int8:
case NeonTypeFlags::Poly8:
return shift ? 7 : (8 << IsQuad) - 1;
case NeonTypeFlags::Int16:
case NeonTypeFlags::Poly16:
return shift ? 15 : (4 << IsQuad) - 1;
case NeonTypeFlags::Int32:
return shift ? 31 : (2 << IsQuad) - 1;
case NeonTypeFlags::Int64:
return shift ? 63 : (1 << IsQuad) - 1;
case NeonTypeFlags::Float16:
assert(!shift && "cannot shift float types!");
return (4 << IsQuad) - 1;
case NeonTypeFlags::Float32:
assert(!shift && "cannot shift float types!");
return (2 << IsQuad) - 1;
}
return 0;
}
@ -288,8 +286,8 @@ bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
return true;
TV = Result.getLimitedValue(32);
if ((TV > 31) || (mask & (1 << TV)) == 0)
TV = Result.getLimitedValue(64);
if ((TV > 63) || (mask & (1 << TV)) == 0)
return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
<< TheCall->getArg(ArgNo)->getSourceRange();
}

View File

@ -833,7 +833,6 @@ static std::string GenOpString(OpKind op, const std::string &proto,
static unsigned GetNeonEnum(const std::string &proto, StringRef typestr) {
unsigned mod = proto[0];
unsigned ret = 0;
if (mod == 'v' || mod == 'f')
mod = proto[1];
@ -851,35 +850,32 @@ static unsigned GetNeonEnum(const std::string &proto, StringRef typestr) {
// Based on the modifying character, change the type and width if necessary.
type = ModType(mod, type, quad, poly, usgn, scal, cnst, pntr);
if (usgn)
ret |= 0x08;
if (quad && proto[1] != 'g')
ret |= 0x10;
NeonTypeFlags::EltType ET;
switch (type) {
case 'c':
ret |= poly ? 5 : 0;
ET = poly ? NeonTypeFlags::Poly8 : NeonTypeFlags::Int8;
break;
case 's':
ret |= poly ? 6 : 1;
ET = poly ? NeonTypeFlags::Poly16 : NeonTypeFlags::Int16;
break;
case 'i':
ret |= 2;
ET = NeonTypeFlags::Int32;
break;
case 'l':
ret |= 3;
ET = NeonTypeFlags::Int64;
break;
case 'h':
ret |= 7;
ET = NeonTypeFlags::Float16;
break;
case 'f':
ret |= 4;
ET = NeonTypeFlags::Float32;
break;
default:
throw "unhandled type!";
break;
}
return ret;
NeonTypeFlags Flags(ET, usgn, quad && proto[1] != 'g');
return Flags.getFlags();
}
// Generate the definition for this intrinsic, e.g. __builtin_neon_cls(a)

View File

@ -86,6 +86,40 @@ enum ClassKind {
ClassB // bitcast arguments with enum argument to specify type
};
/// NeonTypeFlags - Flags to identify the types for overloaded Neon
/// builtins. These must be kept in sync with the flags in
/// include/clang/Basic/TargetBuiltins.h.
class NeonTypeFlags {
enum {
EltTypeMask = 0xf,
UnsignedFlag = 0x10,
QuadFlag = 0x20
};
uint32_t Flags;
public:
enum EltType {
Int8,
Int16,
Int32,
Int64,
Poly8,
Poly16,
Float16,
Float32
};
NeonTypeFlags(unsigned F) : Flags(F) {}
NeonTypeFlags(EltType ET, bool IsUnsigned, bool IsQuad) : Flags(ET) {
if (IsUnsigned)
Flags |= UnsignedFlag;
if (IsQuad)
Flags |= QuadFlag;
}
uint32_t getFlags() const { return Flags; }
};
namespace llvm {
class NeonEmitter : public TableGenBackend {