switch float math udfs from std to libm

Author: leoyvens

datafusion/functions/Cargo.toml

@@ -86,6 +86,7 @@ regex = { workspace = true, optional = true }
 sha2 = { version = "^0.10.1", optional = true }
 unicode-segmentation = { version = "^1.7.1", optional = true }
 uuid = { version = "1.7", features = ["v4"], optional = true }
+libm = "0.2.11"
 
 [dev-dependencies]
 arrow = { workspace = true, features = ["test_utils"] }

datafusion/functions/src/macros.rs

@@ -157,6 +157,9 @@ macro_rules! make_math_unary_udf {
                 ColumnarValue, Documentation, ScalarUDFImpl, Signature, Volatility,
             };
 
+            #[allow(unused_imports)]
+            use $crate::macros::StdFloat as _;
+
             #[derive(Debug)]
             pub struct $UDF {
                 signature: Signature,
@@ -218,12 +221,16 @@ macro_rules! make_math_unary_udf {
                         DataType::Float64 => Arc::new(
                             args[0]
                                 .as_primitive::<Float64Type>()
-                                .unary::<_, Float64Type>(|x: f64| f64::$UNARY_FUNC(x)),
+                                .unary::<_, Float64Type>(|x: f64| {
+                                    libm::Libm::<f64>::$UNARY_FUNC(x)
+                                }),
                         ) as ArrayRef,
                         DataType::Float32 => Arc::new(
                             args[0]
                                 .as_primitive::<Float32Type>()
-                                .unary::<_, Float32Type>(|x: f32| f32::$UNARY_FUNC(x)),
+                                .unary::<_, Float32Type>(|x: f32| {
+                                    libm::Libm::<f32>::$UNARY_FUNC(x)
+                                }),
                         ) as ArrayRef,
                         other => {
                             return exec_err!(
@@ -333,7 +340,7 @@ macro_rules! make_math_binary_udf {
                             let result = arrow::compute::binary::<_, _, _, Float64Type>(
                                 y,
                                 x,
-                                |y, x| f64::$BINARY_FUNC(y, x),
+                                |y, x| libm::Libm::<f64>::$BINARY_FUNC(y, x),
                             )?;
                             Arc::new(result) as _
                         }
@@ -343,7 +350,7 @@ macro_rules! make_math_binary_udf {
                             let result = arrow::compute::binary::<_, _, _, Float32Type>(
                                 y,
                                 x,
-                                |y, x| f32::$BINARY_FUNC(y, x),
+                                |y, x| libm::Libm::<f32>::$BINARY_FUNC(y, x),
                             )?;
                             Arc::new(result) as _
                         }
@@ -365,3 +372,30 @@ macro_rules! make_math_binary_udf {
         }
     };
 }
+
+/// Adds methods that exist in std but are missing from libm,
+/// because they are not platform intrinsics but just conveniences.
+pub trait StdFloat<T> {
+    fn to_radians(x: T) -> T;
+    fn to_degrees(x: T) -> T;
+}
+
+impl StdFloat<f64> for libm::Libm<f64> {
+    fn to_radians(x: f64) -> f64 {
+        x.to_radians()
+    }
+
+    fn to_degrees(x: f64) -> f64 {
+        x.to_degrees()
+    }
+}
+
+impl StdFloat<f32> for libm::Libm<f32> {
+    fn to_radians(x: f32) -> f32 {
+        x.to_radians()
+    }
+
+    fn to_degrees(x: f32) -> f32 {
+        x.to_degrees()
+    }
+}

datafusion/functions/src/math/mod.rs

@@ -177,7 +177,7 @@ make_math_unary_udf!(
     LnFunc,
     LN,
     ln,
-    ln,
+    log,
     super::ln_order,
     super::bounds::unbounded_bounds,
     super::get_ln_doc

datafusion/sqllogictest/test_files/scalar.slt

@@ -93,7 +93,7 @@ select abs(a), abs(b), abs(c) from signed_integers;
 query RRR rowsort
 select acos(0), acos(0.5), acos(1);
 ----
-1.5707963267948966 1.0471975511965976 0
+1.5707963267948966 1.0471975511965979 0
 
 # acos scalar nulls
 query R rowsort
@@ -140,7 +140,7 @@ NaN 9.90349 NaN
 query RRR rowsort
 select asin(0), asin(0.5), asin(1);
 ----
-0 0.5235987755982988 1.5707963267948966
+0 0.5235987755982989 1.5707963267948966
 
 # asin scalar nulls
 query R rowsort
@@ -362,7 +362,7 @@ select round(cos(a), 5), round(cos(b), 5), round(cos(c), 5) from signed_integers
 query RRR rowsort
 select cosh(1), cosh(2), cosh(3);
 ----
-1.5430806348152437 3.7621956910836314 10.067661995777765
+1.543080634815244 3.7621956910836314 10.067661995777765
 
 # cosh scalar nulls
 query R rowsort
@@ -385,7 +385,7 @@ select round(cosh(a), 5), round(cosh(b), 5), round(cosh(c), 5) from small_floats
 query RRR rowsort
 select exp(0), exp(1), exp(2);
 ----
-1 2.718281828459045 7.38905609893065
+1 2.7182818284590455 7.38905609893065
 
 # exp scalar nulls
 query R rowsort
@@ -454,7 +454,7 @@ select floor(a), floor(b), floor(c) from signed_integers;
 query RRR rowsort
 select ln(1), ln(exp(1)), ln(3);
 ----
-0 1 1.0986122886681098
+0 1 1.0986122886681096
 
 # ln scalar nulls
 query R rowsort
@@ -872,7 +872,7 @@ select round(sin(a), 5), round(sin(b), 5), round(sin(c), 5) from small_floats;
 query RRR rowsort
 select sinh(1), sinh(2), sinh(3);
 ----
-1.1752011936438014 3.6268604078470186 10.017874927409903
+1.1752011936438014 3.626860407847019 10.017874927409903
 
 # sinh scalar nulls
 query R rowsort
@@ -937,7 +937,7 @@ NaN
 query RRR rowsort
 select tan(0), tan(pi() / 6), tan(pi() / 4);
 ----
-0 0.5773502691896256 0.9999999999999999
+0 0.5773502691896257 0.9999999999999999
 
 # tan scalar nulls
 query R rowsort