[APFloat] Prevent construction of APFloat with Semantics and FP value

Constructor invocations such as `APFloat(APFloat::IEEEdouble(), 0.0)`
may seem like they accept a FP (floating point) value, but the overload
they reach is actually the `integerPart` one, not a `float` or `double`
overload (which only exists when `fltSemantics` isn't passed).

This may lead to possible loss of data, by the conversion from `float`
or `double` to `integerPart`.

To prevent future mistakes, a new constructor overload, which accepts
any FP value and marked with `delete`, to prevent its usage.

Fixes PR34095.

Differential Revision: https://reviews.llvm.org/D70425

Author: ekatz

llvm/include/llvm/ADT/APFloat.h

@@ -851,6 +851,9 @@ class APFloat : public APFloatBase {
   APFloat(const fltSemantics &Semantics) : U(Semantics) {}
   APFloat(const fltSemantics &Semantics, StringRef S);
   APFloat(const fltSemantics &Semantics, integerPart I) : U(Semantics, I) {}
+  template <typename T, typename = typename std::enable_if<
+                            std::is_floating_point<T>::value>::type>
+  APFloat(const fltSemantics &Semantics, T V) = delete;
   // TODO: Remove this constructor. This isn't faster than the first one.
   APFloat(const fltSemantics &Semantics, uninitializedTag)
       : U(Semantics, uninitialized) {}

llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp

@@ -3386,7 +3386,7 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
 
     if (const ConstantFP *C = dyn_cast<ConstantFP>(Src)) {
       const APFloat &ArgVal = C->getValueAPF();
-      APFloat Val(ArgVal.getSemantics(), 1.0);
+      APFloat Val(ArgVal.getSemantics(), 1);
       APFloat::opStatus Status = Val.divide(ArgVal,
                                             APFloat::rmNearestTiesToEven);
       // Only do this if it was exact and therefore not dependent on the

llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp

@@ -1735,7 +1735,7 @@ Value *LibCallSimplifier::optimizePow(CallInst *Pow, IRBuilder<> &B) {
     // TODO: This whole transformation should be backend specific (e.g. some
     //       backends might prefer libcalls or the limit for the exponent might
     //       be different) and it should also consider optimizing for size.
-    APFloat LimF(ExpoF->getSemantics(), 33.0),
+    APFloat LimF(ExpoF->getSemantics(), 33),
             ExpoA(abs(*ExpoF));
     if (ExpoA.compare(LimF) == APFloat::cmpLessThan) {
       // This transformation applies to integer or integer+0.5 exponents only.

llvm/unittests/ADT/APFloatTest.cpp

@@ -520,9 +520,9 @@ TEST(APFloatTest, FMA) {
 
   // Test x87 extended precision case from http://llvm.org/PR20728.
   {
-    APFloat M1(APFloat::x87DoubleExtended(), 1.0);
-    APFloat M2(APFloat::x87DoubleExtended(), 1.0);
-    APFloat A(APFloat::x87DoubleExtended(), 3.0);
+    APFloat M1(APFloat::x87DoubleExtended(), 1);
+    APFloat M2(APFloat::x87DoubleExtended(), 1);
+    APFloat A(APFloat::x87DoubleExtended(), 3);
 
     bool losesInfo = false;
     M1.fusedMultiplyAdd(M1, A, APFloat::rmNearestTiesToEven);
@@ -600,9 +600,9 @@ TEST(APFloatTest, Denormal) {
   {
     const char *MinNormalStr = "1.17549435082228750797e-38";
     EXPECT_FALSE(APFloat(APFloat::IEEEsingle(), MinNormalStr).isDenormal());
-    EXPECT_FALSE(APFloat(APFloat::IEEEsingle(), 0.0).isDenormal());
+    EXPECT_FALSE(APFloat(APFloat::IEEEsingle(), 0).isDenormal());
 
-    APFloat Val2(APFloat::IEEEsingle(), 2.0e0);
+    APFloat Val2(APFloat::IEEEsingle(), 2);
     APFloat T(APFloat::IEEEsingle(), MinNormalStr);
     T.divide(Val2, rdmd);
     EXPECT_TRUE(T.isDenormal());
@@ -612,9 +612,9 @@ TEST(APFloatTest, Denormal) {
   {
     const char *MinNormalStr = "2.22507385850720138309e-308";
     EXPECT_FALSE(APFloat(APFloat::IEEEdouble(), MinNormalStr).isDenormal());
-    EXPECT_FALSE(APFloat(APFloat::IEEEdouble(), 0.0).isDenormal());
+    EXPECT_FALSE(APFloat(APFloat::IEEEdouble(), 0).isDenormal());
 
-    APFloat Val2(APFloat::IEEEdouble(), 2.0e0);
+    APFloat Val2(APFloat::IEEEdouble(), 2);
     APFloat T(APFloat::IEEEdouble(), MinNormalStr);
     T.divide(Val2, rdmd);
     EXPECT_TRUE(T.isDenormal());
@@ -624,9 +624,9 @@ TEST(APFloatTest, Denormal) {
   {
     const char *MinNormalStr = "3.36210314311209350626e-4932";
     EXPECT_FALSE(APFloat(APFloat::x87DoubleExtended(), MinNormalStr).isDenormal());
-    EXPECT_FALSE(APFloat(APFloat::x87DoubleExtended(), 0.0).isDenormal());
+    EXPECT_FALSE(APFloat(APFloat::x87DoubleExtended(), 0).isDenormal());
 
-    APFloat Val2(APFloat::x87DoubleExtended(), 2.0e0);
+    APFloat Val2(APFloat::x87DoubleExtended(), 2);
     APFloat T(APFloat::x87DoubleExtended(), MinNormalStr);
     T.divide(Val2, rdmd);
     EXPECT_TRUE(T.isDenormal());
@@ -636,9 +636,9 @@ TEST(APFloatTest, Denormal) {
   {
     const char *MinNormalStr = "3.36210314311209350626267781732175260e-4932";
     EXPECT_FALSE(APFloat(APFloat::IEEEquad(), MinNormalStr).isDenormal());
-    EXPECT_FALSE(APFloat(APFloat::IEEEquad(), 0.0).isDenormal());
+    EXPECT_FALSE(APFloat(APFloat::IEEEquad(), 0).isDenormal());
 
-    APFloat Val2(APFloat::IEEEquad(), 2.0e0);
+    APFloat Val2(APFloat::IEEEquad(), 2);
     APFloat T(APFloat::IEEEquad(), MinNormalStr);
     T.divide(Val2, rdmd);
     EXPECT_TRUE(T.isDenormal());
@@ -1153,8 +1153,8 @@ TEST(APFloatTest, makeNaN) {
 #ifdef GTEST_HAS_DEATH_TEST
 #ifndef NDEBUG
 TEST(APFloatTest, SemanticsDeath) {
-  EXPECT_DEATH(APFloat(APFloat::IEEEsingle(), 0.0f).convertToDouble(), "Float semantics are not IEEEdouble");
-  EXPECT_DEATH(APFloat(APFloat::IEEEdouble(), 0.0 ).convertToFloat(),  "Float semantics are not IEEEsingle");
+  EXPECT_DEATH(APFloat(APFloat::IEEEsingle(), 0).convertToDouble(), "Float semantics are not IEEEdouble");
+  EXPECT_DEATH(APFloat(APFloat::IEEEdouble(), 0).convertToFloat(),  "Float semantics are not IEEEsingle");
 }
 
 TEST(APFloatTest, StringDecimalDeath) {