vectorize: trivial handling for F-order arrays

This extends the trivial handling to support trivial handling for
Fortran-order arrays (i.e. column major): if inputs aren't all
C-contiguous, but *are* all F-contiguous, the resulting array will be
F-contiguous and we can do trivial processing.

For anything else (e.g. C-contiguous, or inputs requiring non-trivial
processing), the result is in (numpy-default) C-contiguous layout.

include/pybind11/numpy.h

@@ -1052,11 +1052,14 @@ private:
     std::array<common_iter, N> m_common_iterator;
 };
 
-// Populates the shape and number of dimensions for the set of buffers.  Returns true if the
-// broadcast is "trivial"--that is, has each buffer being either a singleton or a full-size,
-// C-contiguous storage buffer.
+enum class broadcast_trivial { non_trivial, c_trivial, f_trivial };
+
+// Populates the shape and number of dimensions for the set of buffers.  Returns a broadcast_trivial
+// enum value indicating whether the broadcast is "trivial"--that is, has each buffer being either a
+// singleton or a full-size, C-contiguous (`c_trivial`) or Fortran-contiguous (`f_trivial`) storage
+// buffer; returns `non_trivial` otherwise.
 template <size_t N>
-bool broadcast(const std::array<buffer_info, N> &buffers, size_t &ndim, std::vector<size_t> &shape) {
+broadcast_trivial broadcast(const std::array<buffer_info, N> &buffers, size_t &ndim, std::vector<size_t> &shape) {
     ndim = std::accumulate(buffers.begin(), buffers.end(), size_t(0), [](size_t res, const buffer_info& buf) {
         return std::max(res, buf.ndim);
     });
@@ -1064,14 +1067,12 @@ bool broadcast(const std::array<buffer_info, N> &buffers, size_t &ndim, std::vec
     shape.clear();
     shape.resize(ndim, 1);
 
-    bool trivial_broadcast = true;
+    // Figure out the output size, and make sure all input arrays conform (i.e. are either size 1 or
+    // the full size).
     for (size_t i = 0; i < N; ++i) {
-        trivial_broadcast = trivial_broadcast && (buffers[i].size == 1 || buffers[i].ndim == ndim);
-        size_t expect_stride = buffers[i].itemsize;
         auto res_iter = shape.rbegin();
-        auto stride_iter = buffers[i].strides.rbegin();
-        auto shape_iter = buffers[i].shape.rbegin();
-        while (shape_iter != buffers[i].shape.rend()) {
+        auto end = buffers[i].shape.rend();
+        for (auto shape_iter = buffers[i].shape.rbegin(); shape_iter != end; ++shape_iter, ++res_iter) {
             const auto &dim_size_in = *shape_iter;
             auto &dim_size_out = *res_iter;
 
@@ -1080,21 +1081,54 @@ bool broadcast(const std::array<buffer_info, N> &buffers, size_t &ndim, std::vec
                 dim_size_out = dim_size_in;
             else if (dim_size_in != 1 && dim_size_in != dim_size_out)
                 pybind11_fail("pybind11::vectorize: incompatible size/dimension of inputs!");
-
-            if (trivial_broadcast && buffers[i].size > 1) {
-                if (dim_size_in == dim_size_out && expect_stride == *stride_iter) {
-                    expect_stride *= dim_size_in;
-                    ++stride_iter;
-                } else {
-                    trivial_broadcast = false;
-                }
-            }
-
-            ++shape_iter;
-            ++res_iter;
         }
     }
-    return trivial_broadcast;
+
+    bool trivial_broadcast_c = true;
+    bool trivial_broadcast_f = true;
+    for (size_t i = 0; i < N && (trivial_broadcast_c || trivial_broadcast_f); ++i) {
+        if (buffers[i].size == 1)
+            continue;
+
+        // Require the same number of dimensions:
+        if (buffers[i].ndim != ndim)
+            return broadcast_trivial::non_trivial;
+
+        // Require all dimensions be full-size:
+        if (!std::equal(buffers[i].shape.cbegin(), buffers[i].shape.cend(), shape.cbegin()))
+            return broadcast_trivial::non_trivial;
+
+        // Check for C contiguity (but only if previous inputs were also C contiguous)
+        if (trivial_broadcast_c) {
+            size_t expect_stride = buffers[i].itemsize;
+            auto end = buffers[i].shape.crend();
+            for (auto shape_iter = buffers[i].shape.crbegin(), stride_iter = buffers[i].strides.crbegin();
+                    trivial_broadcast_c && shape_iter != end; ++shape_iter, ++stride_iter) {
+                if (expect_stride == *stride_iter)
+                    expect_stride *= *shape_iter;
+                else
+                    trivial_broadcast_c = false;
+            }
+        }
+
+        // Check for Fortran contiguity (if previous inputs were also F contiguous)
+        if (trivial_broadcast_f) {
+            size_t expect_stride = buffers[i].itemsize;
+            auto end = buffers[i].shape.cend();
+            for (auto shape_iter = buffers[i].shape.cbegin(), stride_iter = buffers[i].strides.cbegin();
+                    trivial_broadcast_f && shape_iter != end; ++shape_iter, ++stride_iter) {
+                if (expect_stride == *stride_iter)
+                    expect_stride *= *shape_iter;
+                else
+                    trivial_broadcast_f = false;
+            }
+        }
+    }
+
+    return
+        trivial_broadcast_c ? broadcast_trivial::c_trivial :
+        trivial_broadcast_f ? broadcast_trivial::f_trivial :
+        broadcast_trivial::non_trivial;
 }
 
 template <typename Func, typename Return, typename... Args>
@@ -1116,32 +1150,42 @@ struct vectorize_helper {
         /* Determine dimensions parameters of output array */
         size_t ndim = 0;
         std::vector<size_t> shape(0);
-        bool trivial_broadcast = broadcast(buffers, ndim, shape);
+        auto trivial = broadcast(buffers, ndim, shape);
 
         size_t size = 1;
         std::vector<size_t> strides(ndim);
         if (ndim > 0) {
-            strides[ndim-1] = sizeof(Return);
-            for (size_t i = ndim - 1; i > 0; --i) {
-                strides[i - 1] = strides[i] * shape[i];
-                size *= shape[i];
+            if (trivial == broadcast_trivial::f_trivial) {
+                strides[0] = sizeof(Return);
+                for (size_t i = 1; i < ndim; ++i) {
+                    strides[i] = strides[i - 1] * shape[i - 1];
+                    size *= shape[i - 1];
+                }
+                size *= shape[ndim - 1];
+            }
+            else {
+                strides[ndim-1] = sizeof(Return);
+                for (size_t i = ndim - 1; i > 0; --i) {
+                    strides[i - 1] = strides[i] * shape[i];
+                    size *= shape[i];
+                }
+                size *= shape[0];
             }
-            size *= shape[0];
         }
 
         if (size == 1)
             return cast(f(*reinterpret_cast<Args *>(buffers[Index].ptr)...));
 
-        array_t<Return, array::c_style> result(shape, strides);
+        array_t<Return> result(shape, strides);
         auto buf = result.request();
         auto output = (Return *) buf.ptr;
 
-        if (trivial_broadcast) {
-            /* Call the function */
+        /* Call the function */
+        if (trivial == broadcast_trivial::non_trivial) {
+            apply_broadcast<Index...>(buffers, buf, index);
+        } else {
             for (size_t i = 0; i < size; ++i)
                 output[i] = f((reinterpret_cast<Args *>(buffers[Index].ptr)[buffers[Index].size == 1 ? 0 : i])...);
-        } else {
-            apply_broadcast<Index...>(buffers, buf, index);
         }
 
         return result;

tests/test_numpy_vectorize.cpp

@@ -41,6 +41,10 @@ test_initializer numpy_vectorize([](py::module &m) {
 
 
     // Internal optimization test for whether the input is trivially broadcastable:
+    py::enum_<py::detail::broadcast_trivial>(m, "trivial")
+        .value("f_trivial", py::detail::broadcast_trivial::f_trivial)
+        .value("c_trivial", py::detail::broadcast_trivial::c_trivial)
+        .value("non_trivial", py::detail::broadcast_trivial::non_trivial);
     m.def("vectorized_is_trivial", [](
                 py::array_t<int, py::array::forcecast> arg1,
                 py::array_t<float, py::array::forcecast> arg2,

tests/test_numpy_vectorize.py

@@ -24,6 +24,20 @@ def test_vectorize(capture):
             my_func(x:int=1, y:float=2, z:float=3)
             my_func(x:int=3, y:float=4, z:float=3)
         """
+        with capture:
+            a = np.array([[1, 2], [3, 4]], order='F')
+            b = np.array([[10, 20], [30, 40]], order='F')
+            c = 3
+            result = f(a, b, c)
+            assert np.allclose(result, a * b * c)
+            assert result.flags.f_contiguous
+        # All inputs are F order and full or singletons, so we the result is in col-major order:
+        assert capture == """
+            my_func(x:int=1, y:float=10, z:float=3)
+            my_func(x:int=3, y:float=30, z:float=3)
+            my_func(x:int=2, y:float=20, z:float=3)
+            my_func(x:int=4, y:float=40, z:float=3)
+        """
         with capture:
             a, b, c = np.array([[1, 3, 5], [7, 9, 11]]), np.array([[2, 4, 6], [8, 10, 12]]), 3
             assert np.allclose(f(a, b, c), a * b * c)
@@ -105,29 +119,43 @@ def test_docs(doc):
 
 
 def test_trivial_broadcasting():
-    from pybind11_tests import vectorized_is_trivial
+    from pybind11_tests import vectorized_is_trivial, trivial, vectorized_func
 
-    assert vectorized_is_trivial(1, 2, 3)
-    assert vectorized_is_trivial(np.array(1), np.array(2), 3)
-    assert vectorized_is_trivial(np.array([1, 3]), np.array([2, 4]), 3)
-    assert vectorized_is_trivial(
+    assert vectorized_is_trivial(1, 2, 3) == trivial.c_trivial
+    assert vectorized_is_trivial(np.array(1), np.array(2), 3) == trivial.c_trivial
+    assert vectorized_is_trivial(np.array([1, 3]), np.array([2, 4]), 3) == trivial.c_trivial
+    assert trivial.c_trivial == vectorized_is_trivial(
         np.array([[1, 3, 5], [7, 9, 11]]), np.array([[2, 4, 6], [8, 10, 12]]), 3)
-    assert not vectorized_is_trivial(
-        np.array([[1, 2, 3], [4, 5, 6]]), np.array([2, 3, 4]), 2)
-    assert not vectorized_is_trivial(
-        np.array([[1, 2, 3], [4, 5, 6]]), np.array([[2], [3]]), 2)
+    assert vectorized_is_trivial(
+        np.array([[1, 2, 3], [4, 5, 6]]), np.array([2, 3, 4]), 2) == trivial.non_trivial
+    assert vectorized_is_trivial(
+        np.array([[1, 2, 3], [4, 5, 6]]), np.array([[2], [3]]), 2) == trivial.non_trivial
     z1 = np.array([[1, 2, 3, 4], [5, 6, 7, 8]], dtype='int32')
     z2 = np.array(z1, dtype='float32')
     z3 = np.array(z1, dtype='float64')
-    assert vectorized_is_trivial(z1, z2, z3)
-    assert not vectorized_is_trivial(z1[::2, ::2], 1, 1)
-    assert vectorized_is_trivial(1, 1, z1[::2, ::2])
-    assert not vectorized_is_trivial(1, 1, z3[::2, ::2])
-    assert vectorized_is_trivial(z1, 1, z3[1::4, 1::4])
+    assert vectorized_is_trivial(z1, z2, z3) == trivial.c_trivial
+    assert vectorized_is_trivial(1, z2, z3) == trivial.c_trivial
+    assert vectorized_is_trivial(z1, 1, z3) == trivial.c_trivial
+    assert vectorized_is_trivial(z1, z2, 1) == trivial.c_trivial
+    assert vectorized_is_trivial(z1[::2, ::2], 1, 1) == trivial.non_trivial
+    assert vectorized_is_trivial(1, 1, z1[::2, ::2]) == trivial.c_trivial
+    assert vectorized_is_trivial(1, 1, z3[::2, ::2]) == trivial.non_trivial
+    assert vectorized_is_trivial(z1, 1, z3[1::4, 1::4]) == trivial.c_trivial
 
     y1 = np.array(z1, order='F')
     y2 = np.array(y1)
     y3 = np.array(y1)
-    assert not vectorized_is_trivial(y1, y2, y3)
-    assert not vectorized_is_trivial(y1, z2, z3)
-    assert not vectorized_is_trivial(y1, 1, 1)
+    assert vectorized_is_trivial(y1, y2, y3) == trivial.f_trivial
+    assert vectorized_is_trivial(y1, 1, 1) == trivial.f_trivial
+    assert vectorized_is_trivial(1, y2, 1) == trivial.f_trivial
+    assert vectorized_is_trivial(1, 1, y3) == trivial.f_trivial
+    assert vectorized_is_trivial(y1, z2, 1) == trivial.non_trivial
+    assert vectorized_is_trivial(z1[1::4, 1::4], y2, 1) == trivial.f_trivial
+    assert vectorized_is_trivial(y1[1::4, 1::4], z2, 1) == trivial.c_trivial
+
+    assert vectorized_func(z1, z2, z3).flags.c_contiguous
+    assert vectorized_func(y1, y2, y3).flags.f_contiguous
+    assert vectorized_func(z1, 1, 1).flags.c_contiguous
+    assert vectorized_func(1, y2, 1).flags.f_contiguous
+    assert vectorized_func(z1[1::4, 1::4], y2, 1).flags.f_contiguous
+    assert vectorized_func(y1[1::4, 1::4], z2, 1).flags.c_contiguous