[IRGen] Emit llvm fneg operation for Builtin.fneg_*

rdar://150722907

The original handling was a workaround before LLVM added the fneg operation. Now that it has it, we should use it.

Author: drexin

lib/IRGen/GenBuiltin.cpp

@@ -588,9 +588,7 @@ void irgen::emitBuiltinCall(IRGenFunction &IGF, const BuiltinInfo &Builtin,
   }
 
   if (Builtin.ID == BuiltinValueKind::FNeg) {
-    llvm::Value *rhs = args.claimNext();
-    llvm::Value *lhs = llvm::ConstantFP::get(rhs->getType(), "-0.0");
-    llvm::Value *v = IGF.Builder.CreateFSub(lhs, rhs);
+    llvm::Value *v = IGF.Builder.CreateFNeg(args.claimNext());
     return out.add(v);
   }
   if (Builtin.ID == BuiltinValueKind::AssumeTrue) {

test/IRGen/builtins.swift

@@ -329,9 +329,9 @@ func fneg_test(_ half: Builtin.FPIEEE16,
                double: Builtin.FPIEEE64)
   -> (Builtin.FPIEEE16, Builtin.FPIEEE32, Builtin.FPIEEE64)
 {
-  // CHECK: fsub half 0xH8000, {{%.*}}
-  // CHECK: fsub float -0.000000e+00, {{%.*}}
-  // CHECK: fsub double -0.000000e+00, {{%.*}}
+  // CHECK: fneg half
+  // CHECK: fneg float
+  // CHECK: fneg double
   return (Builtin.fneg_FPIEEE16(half),
           Builtin.fneg_FPIEEE32(single),
           Builtin.fneg_FPIEEE64(double))