APFloat: allow 64-bit of payload

[jfb]

Summary

The APFloat and Constant APIs taking an APInt allow arbitrary payloads, and that's great. There's a convenience API which takes an unsigned, and that's silly because it then directly creates a 64-bit APInt. Just change it to 64-bits directly.

[jfb]

Testing-wise, this isn't tested in APFloatTest.cpp, so I'm keeping the status-quo.
The only place where it's used in an odd way is for the floating-literals.s test (which tries to match the GNU as behavior). The test makes me uneasy because it's using ~0 to create a .single NaN... and I have no idea what's the right behavior for .double... This is better fixed in a follow-up.

[rjmccall]

This truncates the payload to fit into the significand field. Presumably that includes setting the top bit of the significand to make this a NaN and not an infinity. Does it also always form a qNaN, or can you create an sNaN this way?

[jfb]

In D55460#1324136, @rjmccall wrote:

This truncates the payload to fit into the significand field. Presumably that includes setting the top bit of the significand to make this a NaN and not an infinity. Does it also always form a qNaN, or can you create an sNaN this way?

Someone wrote a test for this! http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/ADT/APFloatTest.cpp?r1=97364&r2=97363&pathrev=97364

[rjmccall]

In D55460#1324143, @jfb wrote:

In D55460#1324136, @rjmccall wrote:

This truncates the payload to fit into the significand field. Presumably that includes setting the top bit of the significand to make this a NaN and not an infinity. Does it also always form a qNaN, or can you create an sNaN this way?

Someone wrote a test for this! http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/ADT/APFloatTest.cpp?r1=97364&r2=97363&pathrev=97364

Heh, blast from the past. I guess I could also have looked at the right implementation to see that it always made a qNaN.

Anyway, yes, this seems reasonable.

[rjmccall]

Oh, actually, please embellish the unit test to include some >32-bit payloads when building double-typed NaNs.

[jfb]

• Add tests

[jfb]

In D55460#1324156, @rjmccall wrote:

Oh, actually, please embellish the unit test to include some >32-bit payloads when building double-typed NaNs.

Done, I added some for 32-bit as well, where the original weren't touching top bits either.

[jfb]

• Also add ConstantFP NaN getters which match the APFloat ones (with getQNaN / getSNaN and APInt parameters).

[rjmccall]

LGTM.

include/llvm/ADT/APFloat.h

static APFloat getInf(const fltSemantics &Sem, bool Negative = false) {
     APFloat Val(Sem, uninitialized);
     Val.makeInf(Negative);
     return Val;
   }
 
   /// Factory for NaN values.
   ///
   /// \param Negative - True iff the NaN generated should be negative.
-  /// \param type - The unspecified fill bits for creating the NaN, 0 by
+  /// \param payload - The unspecified fill bits for creating the NaN, 0 by
   /// default.  The value is truncated as necessary.
   static APFloat getNaN(const fltSemantics &Sem, bool Negative = false,
-                        unsigned type = 0) {
-    if (type) {
-      APInt fill(64, type);
-      return getQNaN(Sem, Negative, &fill);
+                        uint64_t payload = 0) {
+    if (payload) {
+      APInt intPayload(64, payload);
+      return getQNaN(Sem, Negative, &intPayload);
     } else {
       return getQNaN(Sem, Negative, nullptr);
     }
   }
 
   /// Factory for QNaN values.
   static APFloat getQNaN(const fltSemantics &Sem, bool Negative = false,
                          const APInt *payload = nullptr) {

include/llvm/IR/Constants.h

public:
   static Constant *get(Type* Ty, double V);
 
   /// If Ty is a vector type, return a Constant with a splat of the given
   /// value. Otherwise return a ConstantFP for the given value.
   static Constant *get(Type *Ty, const APFloat &V);
 
   static Constant *get(Type* Ty, StringRef Str);
   static ConstantFP *get(LLVMContext &Context, const APFloat &V);
-  static Constant *getNaN(Type *Ty, bool Negative = false, unsigned type = 0);
+  static Constant *getNaN(Type *Ty, bool Negative = false, uint64_t Payload = 0);
   static Constant *getNegativeZero(Type *Ty);
   static Constant *getInfinity(Type *Ty, bool Negative = false);
 
   /// Return true if Ty is big enough to represent V.
   static bool isValueValidForType(Type *Ty, const APFloat &V);
   inline const APFloat &getValueAPF() const { return Val; }
 
   /// Return true if the value is positive or negative zero.

lib/IR/Constants.cpp

Constant *ConstantFP::get(Type *Ty, StringRef Str) {
 
   // For vectors, broadcast the value.
   if (VectorType *VTy = dyn_cast<VectorType>(Ty))
     return ConstantVector::getSplat(VTy->getNumElements(), C);
 
   return C;
 }
 
-Constant *ConstantFP::getNaN(Type *Ty, bool Negative, unsigned Type) {
+Constant *ConstantFP::getNaN(Type *Ty, bool Negative, uint64_t Payload) {
   const fltSemantics &Semantics = *TypeToFloatSemantics(Ty->getScalarType());
-  APFloat NaN = APFloat::getNaN(Semantics, Negative, Type);
+  APFloat NaN = APFloat::getNaN(Semantics, Negative, Payload);
   Constant *C = get(Ty->getContext(), NaN);
 
   if (VectorType *VTy = dyn_cast<VectorType>(Ty))
     return ConstantVector::getSplat(VTy->getNumElements(), C);
 
   return C;
 }