Fix power function Inf handling (#11295)

Summary:
Pull Request resolved: #11295

Fix bug introduced in D64145275 that returned NaN whenever the expononent was infinity.

Reviewed By: spershin

Differential Revision: D64573431

fbshipit-source-id: 7346b79732e5850579e5f048d0c85c6772a8cd42

velox/functions/prestosql/Arithmetic.h

@@ -229,9 +229,8 @@ struct PowerFunction {
   template <typename TInput>
   FOLLY_ALWAYS_INLINE void
   call(double& result, const TInput& a, const TInput& b) {
-    result = (std::isnan(a) && b != 0) || std::isnan(b) || std::isinf(b)
-        ? std::numeric_limits<double>::quiet_NaN()
-        : pow(a, b);
+    result =
+        std::isnan(b) ? std::numeric_limits<double>::quiet_NaN() : pow(a, b);
   }
 };
 

velox/functions/prestosql/tests/ArithmeticTest.cpp

@@ -292,9 +292,18 @@ TEST_F(ArithmeticTest, powerNan) {
 }
 
 TEST_F(ArithmeticTest, powerInf) {
-  std::vector<double> baseDouble = {1, kInf, kInf, kInf};
-  std::vector<double> exponentDouble = {kInf, 1, kNan, 0};
-  std::vector<double> expectedDouble = {kNan, kInf, kNan, 1};
+  // base | exp | result
+  //----------------------
+  //   1    Inf     1
+  //   1   -Inf     1
+  //   2    Inf    Inf
+  //   2   -Inf     0
+  //  Inf    1     Inf
+  //  Inf    0      1
+  //  Inf   NaN    NaN
+  std::vector<double> baseDouble = {1, 1, 2, 2, kInf, kInf, kInf};
+  std::vector<double> exponentDouble = {kInf, -kInf, kInf, -kInf, 1, 0, kNan};
+  std::vector<double> expectedDouble = {1, 1, kInf, 0, kInf, 1, kNan};
 
   // Check using function name and alias.
   assertExpression<double>(