return s;
}
+/// UseMacro - Examine the prototype string to determine if the intrinsic
+/// should be defined as a preprocessor macro instead of an inline function.
+static bool UseMacro(const std::string &proto) {
+ // If this builtin takes an immediate argument, we need to #define it rather
+ // than use a standard declaration, so that SemaChecking can range check
+ // the immediate passed by the user.
+ if (proto.find('i') != std::string::npos)
+ return true;
+
+ // Pointer arguments need to use macros to avoid hiding aligned attributes
+ // from the pointer type.
+ if (proto.find('p') != std::string::npos ||
+ proto.find('c') != std::string::npos)
+ return true;
+
+ return false;
+}
+
+/// MacroArgUsedDirectly - Return true if argument i for an intrinsic that is
+/// defined as a macro should be accessed directly instead of being first
+/// assigned to a local temporary.
+static bool MacroArgUsedDirectly(const std::string &proto, unsigned i) {
+ return (proto[i] == 'i' || proto[i] == 'p' || proto[i] == 'c');
+}
+
// Generate the string "(argtype a, argtype b, ...)"
static std::string GenArgs(const std::string &proto, StringRef typestr) {
- bool define = proto.find('i') != std::string::npos;
+ bool define = UseMacro(proto);
char arg = 'a';
std::string s;
for (unsigned i = 1, e = proto.size(); i != e; ++i, ++arg) {
if (define) {
- // Immediate macro arguments are used directly instead of being assigned
+ // Some macro arguments are used directly instead of being assigned
// to local temporaries; prepend an underscore prefix to make their
// names consistent with the local temporaries.
- if (proto[i] == 'i')
+ if (MacroArgUsedDirectly(proto, i))
s += "__";
} else {
s += TypeString(proto[i], typestr) + " __";
static std::string GenMacroLocals(const std::string &proto, StringRef typestr) {
char arg = 'a';
std::string s;
+ bool generatedLocal = false;
for (unsigned i = 1, e = proto.size(); i != e; ++i, ++arg) {
// Do not create a temporary for an immediate argument.
// That would defeat the whole point of using a macro!
- if (proto[i] == 'i') continue;
+ if (proto[i] == 'i')
+ continue;
+ generatedLocal = true;
+
+ // For other (non-immediate) arguments that are used directly, a local
+ // temporary is still needed to get the correct type checking, even though
+ // that temporary is not used for anything.
+ if (MacroArgUsedDirectly(proto, i)) {
+ s += TypeString(proto[i], typestr) + " __";
+ s.push_back(arg);
+ s += "_ = (__";
+ s.push_back(arg);
+ s += "); (void)__";
+ s.push_back(arg);
+ s += "_; ";
+ continue;
+ }
s += TypeString(proto[i], typestr) + " __";
s.push_back(arg);
s += "); ";
}
- s += "\\\n ";
+ if (generatedLocal)
+ s += "\\\n ";
return s;
}
StringRef typestr) {
bool quad;
unsigned nElts = GetNumElements(typestr, quad);
-
- // If this builtin takes an immediate argument, we need to #define it rather
- // than use a standard declaration, so that SemaChecking can range check
- // the immediate passed by the user.
- bool define = proto.find('i') != std::string::npos;
+ bool define = UseMacro(proto);
std::string ts = TypeString(proto[0], typestr);
std::string s;
- if (op == OpHi || op == OpLo) {
- s = "union { " + ts + " r; double d; } u; u.d = ";
- } else if (!define) {
+ if (!define) {
s = "return ";
}
case OpMul:
s += "__a * __b;";
break;
- case OpMullN:
- s += Extend(typestr, "__a") + " * " +
- Extend(typestr, Duplicate(nElts << (int)quad, typestr, "__b")) + ";";
- break;
case OpMullLane:
- s += Extend(typestr, "__a") + " * " +
- Extend(typestr, SplatLane(nElts, "__b", "__c")) + ";";
- break;
- case OpMull:
- s += Extend(typestr, "__a") + " * " + Extend(typestr, "__b") + ";";
+ s += MangleName("vmull", typestr, ClassS) + "(__a, " +
+ SplatLane(nElts, "__b", "__c") + ");";
break;
case OpMlaN:
s += "__a + (__b * " + Duplicate(nElts, typestr, "__c") + ");";
s += "__a + (__b * __c);";
break;
case OpMlalN:
- s += "__a + (" + Extend(typestr, "__b") + " * " +
- Extend(typestr, Duplicate(nElts, typestr, "__c")) + ");";
+ s += "__a + " + MangleName("vmull", typestr, ClassS) + "(__b, " +
+ Duplicate(nElts, typestr, "__c") + ");";
break;
case OpMlalLane:
- s += "__a + (" + Extend(typestr, "__b") + " * " +
- Extend(typestr, SplatLane(nElts, "__c", "__d")) + ");";
+ s += "__a + " + MangleName("vmull", typestr, ClassS) + "(__b, " +
+ SplatLane(nElts, "__c", "__d") + ");";
break;
case OpMlal:
- s += "__a + (" + Extend(typestr, "__b") + " * " +
- Extend(typestr, "__c") + ");";
+ s += "__a + " + MangleName("vmull", typestr, ClassS) + "(__b, __c);";
break;
case OpMlsN:
s += "__a - (__b * " + Duplicate(nElts, typestr, "__c") + ");";
s += "__a - (__b * __c);";
break;
case OpMlslN:
- s += "__a - (" + Extend(typestr, "__b") + " * " +
- Extend(typestr, Duplicate(nElts, typestr, "__c")) + ");";
+ s += "__a - " + MangleName("vmull", typestr, ClassS) + "(__b, " +
+ Duplicate(nElts, typestr, "__c") + ");";
break;
case OpMlslLane:
- s += "__a - (" + Extend(typestr, "__b") + " * " +
- Extend(typestr, SplatLane(nElts, "__c", "__d")) + ");";
+ s += "__a - " + MangleName("vmull", typestr, ClassS) + "(__b, " +
+ SplatLane(nElts, "__c", "__d") + ");";
break;
case OpMlsl:
- s += "__a - (" + Extend(typestr, "__b") + " * " +
- Extend(typestr, "__c") + ");";
+ s += "__a - " + MangleName("vmull", typestr, ClassS) + "(__b, __c);";
break;
case OpQDMullLane:
s += MangleName("vqdmull", typestr, ClassS) + "(__a, " +
s += ", (int64x1_t)__b, 0, 1);";
break;
case OpHi:
- s += "(((float64x2_t)__a)[1]);";
+ s += "(" + ts +
+ ")__builtin_shufflevector((int64x2_t)__a, (int64x2_t)__a, 1);";
break;
case OpLo:
- s += "(((float64x2_t)__a)[0]);";
+ s += "(" + ts +
+ ")__builtin_shufflevector((int64x2_t)__a, (int64x2_t)__a, 0);";
break;
case OpDup:
s += Duplicate(nElts, typestr, "__a") + ";";
throw "unknown OpKind!";
break;
}
- if (op == OpHi || op == OpLo) {
- if (!define)
- s += " return";
- s += " u.r;";
- }
return s;
}
// sret-like argument.
bool sret = (proto[0] >= '2' && proto[0] <= '4');
- // If this builtin takes an immediate argument, we need to #define it rather
- // than use a standard declaration, so that SemaChecking can range check
- // the immediate passed by the user.
- bool define = proto.find('i') != std::string::npos;
+ bool define = UseMacro(proto);
// Check if the prototype has a scalar operand with the type of the vector
// elements. If not, bitcasting the args will take care of arg checking.
StringRef outTypeStr, StringRef inTypeStr,
OpKind kind, ClassKind classKind) {
assert(!proto.empty() && "");
- bool define = proto.find('i') != std::string::npos;
+ bool define = UseMacro(proto);
std::string s;
// static always inline + return type
OS << " " << TypeString('d', TDTypeVec[i]) << ";\n";
}
OS << "\n";
- OS << "typedef __attribute__((__vector_size__(8))) "
- "double float64x1_t;\n";
- OS << "typedef __attribute__((__vector_size__(16))) "
- "double float64x2_t;\n";
- OS << "\n";
// Emit struct typedefs.
for (unsigned vi = 2; vi != 5; ++vi) {
std::vector<Record*> RV = Records.getAllDerivedDefinitions("Inst");
- // Emit vmovl and vabd intrinsics first so they can be used by other
+ // Emit vmovl, vmull and vabd intrinsics first so they can be used by other
// intrinsics. (Some of the saturating multiply instructions are also
// used to implement the corresponding "_lane" variants, but tablegen
// sorts the records into alphabetical order so that the "_lane" variants
// come after the intrinsics they use.)
emitIntrinsic(OS, Records.getDef("VMOVL"));
+ emitIntrinsic(OS, Records.getDef("VMULL"));
emitIntrinsic(OS, Records.getDef("VABD"));
for (unsigned i = 0, e = RV.size(); i != e; ++i) {
Record *R = RV[i];
- if (R->getName() != "VMOVL" && R->getName() != "VABD")
+ if (R->getName() != "VMOVL" &&
+ R->getName() != "VMULL" &&
+ R->getName() != "VABD")
emitIntrinsic(OS, R);
}
for (unsigned ti = 0, te = TypeVec.size(); ti != te; ++ti) {
std::string namestr, shiftstr, rangestr;
- // Builtins which are overloaded by type will need to have their upper
- // bound computed at Sema time based on the type constant.
- if (Proto.find('s') == std::string::npos) {
+ if (R->getValueAsBit("isVCVT_N")) {
+ // VCVT between floating- and fixed-point values takes an immediate
+ // in the range 1 to 32.
+ ck = ClassB;
+ rangestr = "l = 1; u = 31"; // upper bound = l + u
+ } else if (Proto.find('s') == std::string::npos) {
+ // Builtins which are overloaded by type will need to have their upper
+ // bound computed at Sema time based on the type constant.
ck = ClassB;
if (R->getValueAsBit("isShift")) {
shiftstr = ", true";
void NeonEmitter::runTests(raw_ostream &OS) {
OS <<
"// RUN: %clang_cc1 -triple thumbv7-apple-darwin \\\n"
- "// RUN: -target-cpu cortex-a8 -ffreestanding -S -o - %s | FileCheck %s\n"
+ "// RUN: -target-cpu cortex-a9 -ffreestanding -S -o - %s | FileCheck %s\n"
"\n"
"#include <arm_neon.h>\n"
"\n";
projects / oota-llvm.git / blobdiff