Add from_ieee754_32 Presto scalar function

facebookincubator · Mar 28, 2024 · 0d9494e · 0d9494e
1 parent 93b2544
commit 0d9494e
Show file tree

Hide file tree

Showing 4 changed files with 86 additions and 4 deletions.
diff --git a/velox/docs/functions/presto/binary.rst b/velox/docs/functions/presto/binary.rst
@@ -26,6 +26,15 @@ Binary Functions
 
     Decodes binary data from the hex encoded ``string``.
 
+.. function:: from_ieee754_32(binary) -> real
+
+    Decodes the 32-bit big-endian ``binary`` in IEEE 754 single-precision floating-point format.
+    Throws a VeloxException if input size is shorter / longer than 32 bits.
+
+.. function:: from_ieee754_64(binary) -> double
+
+    Decodes the 64-bit big-endian ``binary`` in IEEE 754 double-precision floating-point format.
+
 .. function:: hmac_md5(binary, key) -> varbinary
 
     Computes the HMAC with md5 of ``binary`` with the given ``key``.
@@ -98,10 +107,6 @@ Binary Functions
 
     Encodes ``double`` in a 64-bit big-endian binary according to IEEE 754 double-precision floating-point format.
 
-.. function:: from_ieee754_64(binary) -> double
-
-    Decodes the 64-bit big-endian ``binary`` in IEEE 754 double-precision floating-point format.
-
 .. function:: xxhash64(binary) -> varbinary
 
     Computes the xxhash64 hash of ``binary``.
diff --git a/velox/functions/prestosql/BinaryFunctions.h b/velox/functions/prestosql/BinaryFunctions.h
@@ -424,4 +424,21 @@ struct ToIEEE754Bits32 {
   }
 };
 
+template <typename T>
+struct FromIEEE754Bits32 {
+  VELOX_DEFINE_FUNCTION_TYPES(T);
+
+  FOLLY_ALWAYS_INLINE void call(
+      out_type<float>& result,
+      const arg_type<Varbinary>& input) {
+    static constexpr auto kTypeLength = sizeof(int32_t);
+    VELOX_USER_CHECK_EQ(
+        input.size(),
+        kTypeLength,
+        "Input floating-point value must be exactly 4 bytes long");
+    memcpy(&result, input.data(), kTypeLength);
+    result = folly::Endian::big(result);
+  }
+};
+
 } // namespace facebook::velox::functions
diff --git a/velox/functions/prestosql/registration/BinaryFunctionsRegistration.cpp b/velox/functions/prestosql/registration/BinaryFunctionsRegistration.cpp
@@ -66,6 +66,8 @@ void registerSimpleFunctions(const std::string& prefix) {
       {prefix + "from_ieee754_64"});
   registerFunction<ToIEEE754Bits32, Varbinary, float>(
       {prefix + "to_ieee754_32"});
+  registerFunction<FromIEEE754Bits32, float, Varbinary>(
+      {prefix + "from_ieee754_32"});
 }
 } // namespace
 

diff --git a/velox/functions/prestosql/tests/BinaryFunctionsTest.cpp b/velox/functions/prestosql/tests/BinaryFunctionsTest.cpp
@@ -18,6 +18,7 @@
 #include <array>
 #include <limits>
 #include "velox/common/base/VeloxException.h"
+#include "velox/common/base/tests/GTestUtils.h"
 #include "velox/expression/Expr.h"
 #include "velox/functions/prestosql/tests/utils/FunctionBaseTest.h"
 
@@ -697,4 +698,61 @@ TEST_F(BinaryFunctionsTest, toIEEE754Bits32) {
       hexToDec("FF7FFFFF"),
       toIEEE754Bits32(std::numeric_limits<float>::lowest()));
 }
+
+TEST_F(BinaryFunctionsTest, fromIEEE754Bits32) {
+  const auto fromIEEE754Bits32 = [&](const std::optional<std::string>& arg) {
+    return evaluateOnce<float, std::string>(
+        "from_ieee754_32(c0)", {arg}, {VARBINARY()});
+  };
+
+  const auto toIEEE754Bits32 = [&](std::optional<float> arg) {
+    return evaluateOnce<std::string, float>("to_ieee754_32(c0)", arg);
+  };
+
+  EXPECT_EQ(std::nullopt, fromIEEE754Bits32(std::nullopt));
+  EXPECT_EQ(1.0f, fromIEEE754Bits32(hexToDec("3F800000")));
+  EXPECT_EQ(3.14f, fromIEEE754Bits32(hexToDec("4048F5C3")));
+  EXPECT_EQ(3.4028235E38f, fromIEEE754Bits32(hexToDec("7f7fffff")));
+  EXPECT_EQ(-3.4028235E38f, fromIEEE754Bits32(hexToDec("ff7fffff")));
+  EXPECT_EQ(1.4E-45f, fromIEEE754Bits32(hexToDec("00000001")));
+  EXPECT_EQ(-1.4E-45f, fromIEEE754Bits32(hexToDec("80000001")));
+  EXPECT_EQ(
+      std::numeric_limits<float>::infinity(),
+      fromIEEE754Bits32(hexToDec("7f800000")));
+  EXPECT_EQ(
+      -std::numeric_limits<float>::infinity(),
+      fromIEEE754Bits32(hexToDec("ff800000")));
+  EXPECT_THROW(fromIEEE754Bits32("YQ"), VeloxUserError);
+  EXPECT_EQ(3.4028235E38f, fromIEEE754Bits32(toIEEE754Bits32(3.4028235E38f)));
+  EXPECT_EQ(-3.4028235E38f, fromIEEE754Bits32(toIEEE754Bits32(-3.4028235E38f)));
+  EXPECT_EQ(1.4E-45f, fromIEEE754Bits32(toIEEE754Bits32(1.4E-45f)));
+  EXPECT_EQ(-1.4E-45f, fromIEEE754Bits32(toIEEE754Bits32(-1.4E-45f)));
+  EXPECT_EQ(
+      std::numeric_limits<float>::infinity(),
+      fromIEEE754Bits32(
+          toIEEE754Bits32(std::numeric_limits<float>::infinity())));
+  EXPECT_EQ(
+      -std::numeric_limits<float>::infinity(),
+      fromIEEE754Bits32(
+          toIEEE754Bits32(-std::numeric_limits<float>::infinity())));
+  EXPECT_EQ(
+      std::numeric_limits<float>::max(),
+      fromIEEE754Bits32(toIEEE754Bits32(std::numeric_limits<float>::max())));
+  EXPECT_EQ(
+      std::numeric_limits<float>::min(),
+      fromIEEE754Bits32(toIEEE754Bits32(std::numeric_limits<float>::min())));
+  EXPECT_TRUE(
+      std::isnan(fromIEEE754Bits32(
+                     toIEEE754Bits32(std::numeric_limits<float>::quiet_NaN()))
+                     .value()));
+  EXPECT_TRUE(std::isnan(
+      fromIEEE754Bits32(
+          toIEEE754Bits32(std::numeric_limits<float>::signaling_NaN()))
+          .value()));
+  EXPECT_TRUE(
+      std::isnan(fromIEEE754Bits32(toIEEE754Bits32(std::nan("nan"))).value()));
+  VELOX_ASSERT_THROW(
+      fromIEEE754Bits32(hexToDec("0000000000000001")),
+      "Input floating-point value must be exactly 4 bytes long");
+}
 } // namespace