Merge pull request #326 from PyO3/extraction-error

Avoid the overhead of creating a PyErr for downcasting.

Author: adamreichold

benches/array.rs

@@ -30,6 +30,27 @@ fn extract_failure(bencher: &mut Bencher) {
     });
 }
 
+#[bench]
+fn downcast_success(bencher: &mut Bencher) {
+    Python::with_gil(|py| {
+        let any: &PyAny = PyArray2::<f64>::zeros(py, (10, 10), false);
+
+        bencher.iter(|| {
+            black_box(any).downcast::<PyArray2<f64>>().unwrap();
+        });
+    });
+}
+
+#[bench]
+fn downcast_failure(bencher: &mut Bencher) {
+    Python::with_gil(|py| {
+        let any: &PyAny = PyArray2::<i32>::zeros(py, (10, 10), false);
+
+        bencher.iter(|| {
+            black_box(any).downcast::<PyArray2<f64>>().unwrap_err();
+        });
+    });
+}
 struct Iter(Range<usize>);
 
 impl Iterator for Iter {

src/array.rs

@@ -26,7 +26,7 @@ use crate::cold;
 use crate::convert::{ArrayExt, IntoPyArray, NpyIndex, ToNpyDims, ToPyArray};
 use crate::dtype::{Element, PyArrayDescr};
 use crate::error::{
-    BorrowError, DimensionalityError, FromVecError, NotContiguousError, TypeError,
+    BorrowError, DimensionalityError, FromVecError, IgnoreError, NotContiguousError, TypeError,
     DIMENSIONALITY_MISMATCH_ERR, MAX_DIMENSIONALITY_ERR,
 };
 use crate::npyffi::{self, npy_intp, NPY_ORDER, PY_ARRAY_API};
@@ -131,7 +131,7 @@ unsafe impl<T: Element, D: Dimension> PyTypeInfo for PyArray<T, D> {
     }
 
     fn is_type_of(ob: &PyAny) -> bool {
-        <&Self>::extract(ob).is_ok()
+        Self::extract::<IgnoreError>(ob).is_ok()
     }
 }
 
@@ -145,30 +145,7 @@ impl<T, D> IntoPy<PyObject> for PyArray<T, D> {
 
 impl<'py, T: Element, D: Dimension> FromPyObject<'py> for &'py PyArray<T, D> {
     fn extract(ob: &'py PyAny) -> PyResult<Self> {
-        // Check if the object is an array.
-        let array = unsafe {
-            if npyffi::PyArray_Check(ob.py(), ob.as_ptr()) == 0 {
-                return Err(PyDowncastError::new(ob, PyArray::<T, D>::NAME).into());
-            }
-            &*(ob as *const PyAny as *const PyArray<T, D>)
-        };
-
-        // Check if the dimensionality matches `D`.
-        let src_ndim = array.ndim();
-        if let Some(dst_ndim) = D::NDIM {
-            if src_ndim != dst_ndim {
-                return Err(DimensionalityError::new(src_ndim, dst_ndim).into());
-            }
-        }
-
-        // Check if the element type matches `T`.
-        let src_dtype = array.dtype();
-        let dst_dtype = T::get_dtype(ob.py());
-        if !src_dtype.is_equiv_to(dst_dtype) {
-            return Err(TypeError::new(src_dtype, dst_dtype).into());
-        }
-
-        Ok(array)
+        PyArray::extract(ob)
     }
 }
 
@@ -390,6 +367,36 @@ impl<T, D> PyArray<T, D> {
 }
 
 impl<T: Element, D: Dimension> PyArray<T, D> {
+    fn extract<'py, E>(ob: &'py PyAny) -> Result<&'py Self, E>
+    where
+        E: From<PyDowncastError<'py>> + From<DimensionalityError> + From<TypeError<'py>>,
+    {
+        // Check if the object is an array.
+        let array = unsafe {
+            if npyffi::PyArray_Check(ob.py(), ob.as_ptr()) == 0 {
+                return Err(PyDowncastError::new(ob, Self::NAME).into());
+            }
+            &*(ob as *const PyAny as *const Self)
+        };
+
+        // Check if the dimensionality matches `D`.
+        let src_ndim = array.ndim();
+        if let Some(dst_ndim) = D::NDIM {
+            if src_ndim != dst_ndim {
+                return Err(DimensionalityError::new(src_ndim, dst_ndim).into());
+            }
+        }
+
+        // Check if the element type matches `T`.
+        let src_dtype = array.dtype();
+        let dst_dtype = T::get_dtype(ob.py());
+        if !src_dtype.is_equiv_to(dst_dtype) {
+            return Err(TypeError::new(src_dtype, dst_dtype).into());
+        }
+
+        Ok(array)
+    }
+
     /// Same as [`shape`][Self::shape], but returns `D` insead of `&[usize]`.
     #[inline(always)]
     pub fn dims(&self) -> D {

src/dtype.rs

@@ -117,9 +117,12 @@ impl PyArrayDescr {
 
     /// Returns true if two type descriptors are equivalent.
     pub fn is_equiv_to(&self, other: &Self) -> bool {
+        let self_ptr = self.as_dtype_ptr();
+        let other_ptr = other.as_dtype_ptr();
+
         unsafe {
-            PY_ARRAY_API.PyArray_EquivTypes(self.py(), self.as_dtype_ptr(), other.as_dtype_ptr())
-                != 0
+            self_ptr == other_ptr
+                || PY_ARRAY_API.PyArray_EquivTypes(self.py(), self_ptr, other_ptr) != 0
         }
     }
 
@@ -413,7 +416,7 @@ fn npy_int_type_lookup<T, T0, T1, T2>(npy_types: [NPY_TYPES; 3]) -> NPY_TYPES {
 
 fn npy_int_type<T: Bounded + Zero + Sized + PartialEq>() -> NPY_TYPES {
     let is_unsigned = T::min_value() == T::zero();
-    let bit_width = size_of::<T>() << 3;
+    let bit_width = 8 * size_of::<T>();
 
     match (is_unsigned, bit_width) {
         (false, 8) => NPY_TYPES::NPY_BYTE,
@@ -449,6 +452,7 @@ macro_rules! impl_element_scalar {
         $(#[$meta])*
         unsafe impl Element for $ty {
             const IS_COPY: bool = true;
+
             fn get_dtype(py: Python) -> &PyArrayDescr {
                 PyArrayDescr::from_npy_type(py, $npy_type)
             }

src/error.rs

@@ -162,3 +162,18 @@ impl fmt::Display for BorrowError {
 }
 
 impl_pyerr!(BorrowError);
+
+/// An internal type used to ignore certain error conditions
+///
+/// This is beneficial when those errors will never reach a public API anyway
+/// but dropping them will improve performance.
+pub(crate) struct IgnoreError;
+
+impl<E> From<E> for IgnoreError
+where
+    PyErr: From<E>,
+{
+    fn from(_err: E) -> Self {
+        Self
+    }
+}