prepare for header-only version

pocketfft.cc

@@ -30,7 +30,7 @@
 
 using namespace std;
 
-namespace {
+namespace pocketfft_private {
 
 template<typename T> struct arr
   {
@@ -458,6 +458,10 @@ template<bool bwd, typename T> NOINLINE void pass3 (size_t ido, size_t l1,
       }
   }
 
+#undef PARTSTEP3b
+#undef PARTSTEP3a
+#undef PREP3
+
 template<bool bwd, typename T> NOINLINE void pass4 (size_t ido, size_t l1,
   const T * restrict cc, T * restrict ch, const cmplx<T0> * restrict wa)
   {
@@ -562,6 +566,10 @@ template<bool bwd, typename T> NOINLINE void pass5 (size_t ido, size_t l1,
       }
   }
 
+#undef PARTSTEP5b
+#undef PARTSTEP5a
+#undef PREP5
+
 #define PREP7(idx) \
         T t1 = CC(idx,0,k), t2, t3, t4, t5, t6, t7; \
         PMC (t2,t7,CC(idx,1,k),CC(idx,6,k)); \
@@ -627,6 +635,11 @@ template<bool bwd, typename T> NOINLINE void pass7(size_t ido, size_t l1,
       }
   }
 
+#undef PARTSTEP7
+#undef PARTSTEP7a0
+#undef PARTSTEP7a
+#undef PREP7
+
 #define PREP11(idx) \
         T t1 = CC(idx,0,k), t2, t3, t4, t5, t6, t7, t8, t9, t10, t11; \
         PMC (t2,t11,CC(idx,1,k),CC(idx,10,k)); \
@@ -703,6 +716,11 @@ template<bool bwd, typename T> NOINLINE void pass11 (size_t ido, size_t l1,
       }
   }
 
+#undef PARTSTEP11
+#undef PARTSTEP11a0
+#undef PARTSTEP11a
+#undef PREP11
+
 #define CX(a,b,c) cc[(a)+ido*((b)+l1*(c))]
 #define CX2(a,b) cc[(a)+idl1*(b)]
 #define CH2(a,b) ch[(a)+idl1*(b)]
@@ -857,7 +875,6 @@ template<bool bwd, typename T> NOINLINE void pass_all(T c[], T0 fact)
 #undef WA
 #undef CC
 #undef CH
-#undef PMC
 
   public:
     template<typename T> NOINLINE void forward(T c[], T0 fct)
@@ -1434,8 +1451,8 @@ template<typename T>NOINLINE void radb5(size_t ido, size_t l1,
       }
   }
 
-#undef CC
 #undef CH
+#undef CC
 #define CC(a,b,c) cc[(a)+ido*((b)+cdim*(c))]
 #define CH(a,b,c) ch[(a)+ido*((b)+l1*(c))]
 #define C1(a,b,c) cc[(a)+ido*((b)+l1*(c))]
@@ -1572,10 +1589,11 @@ template<typename T> NOINLINE void radbg(size_t ido, size_t ip, size_t l1,
 #undef C2
 #undef CH2
 
-#undef CC
 #undef CH
-#undef PM
+#undef CC
+
 #undef MULPM
+#undef PM
 #undef WA
 
 template<typename T> void copy_and_norm(T *c, T *p1, size_t n, T0 fct)
@@ -2362,6 +2380,8 @@ template<typename T> NOINLINE void pocketfft_general_r(
     }
   }
 
+#undef HAVE_VECSUPPORT
+
 //
 // Python interface
 //
@@ -2598,7 +2618,7 @@ Parameters
 a : numpy.ndarray (np.complex64 or np.complex128)
     The input data
 axes : list of integers
-    The axes on which the FFT is carried out.
+    The axes along which the FFT is carried out.
     If not set, all axes will be transformed.
 fct : float
     Normalization factor
@@ -2609,7 +2629,7 @@ inplace : bool
 Returns
 -------
 np.ndarray (same shape and data type as a)
-    the transformed data
+    The transformed data.
 )DELIM";
 
 const char *ifftn_DS = R"DELIM(Performs a backward complex FFT.
@@ -2619,7 +2639,7 @@ Parameters
 a : numpy.ndarray (np.complex64 or np.complex128)
     The input data
 axes : list of integers
-    The axes on which the FFT is carried out.
+    The axes along which the FFT is carried out.
     If not set, all axes will be transformed.
 fct : float
     Normalization factor
@@ -2640,7 +2660,7 @@ Parameters
 a : numpy.ndarray (np.float32 or np.float64)
     The input data
 axes : list of integers
-    The axes on which the FFT is carried out.
+    The axes along which the FFT is carried out.
     If not set, all axes will be transformed in ascending order.
 fct : float
     Normalization factor
@@ -2653,6 +2673,28 @@ np.ndarray (np.complex64 or np.complex128)
     was n on input, it is n//2+1 on output.
 )DELIM";
 
+const char *rfft_scipy_DS = R"DELIM(Performs a forward real-valued FFT.
+
+Parameters
+----------
+a : numpy.ndarray (np.float32 or np.float64)
+    The input data
+axis : int
+    The axis along which the FFT is carried out.
+fct : float
+    Normalization factor
+inplace : bool
+    if False, returns the result in a new array and leaves the input unchanged.
+    if True, stores the result in the input array and returns a handle to it.
+
+Returns
+-------
+np.ndarray (np.float32 or np.float64)
+    The transformed data. The shape is identical to that of the input array.
+    Along the transformed axis, values are arranged in
+    FFTPACK half-complex order, i.e. `a[0].re, a[1].re, a[1].im, a[2].re ...`.
+)DELIM";
+
 const char *irfftn_DS = R"DELIM(Performs a backward real-valued FFT.
 
 Parameters
@@ -2660,7 +2702,7 @@ Parameters
 a : numpy.ndarray (np.complex64 or np.complex128)
     The input data
 axes : list of integers
-    The axes on which the FFT is carried out.
+    The axes along which the FFT is carried out.
     If not set, all axes will be transformed in ascending order.
 lastsize : the output size of the last axis to be transformed.
     If the corresponding input axis has size n, this can be 2*n-2 or 2*n-1.
@@ -2675,6 +2717,27 @@ np.ndarray (np.float32 or np.float64)
     entries.
 )DELIM";
 
+const char *irfft_scipy_DS = R"DELIM(Performs a backward real-valued FFT.
+
+Parameters
+----------
+a : numpy.ndarray (np.float32 or np.float64)
+    The input data. Along the transformed axis, values are expected in
+    FFTPACK half-complex order, i.e. `a[0].re, a[1].re, a[1].im, a[2].re ...`.
+axis : int
+    The axis along which the FFT is carried out.
+fct : float
+    Normalization factor
+inplace : bool
+    if False, returns the result in a new array and leaves the input unchanged.
+    if True, stores the result in the input array and returns a handle to it.
+
+Returns
+-------
+np.ndarray (np.float32 or np.float64)
+    The transformed data. The shape is identical to that of the input array.
+)DELIM";
+
 const char *hartley_DS = R"DELIM(Performs a Hartley transform.
 For every requested axis, a 1D forward Fourier transform is carried out,
 and the sum of real and imaginary parts of the result is stored in the output
@@ -2685,7 +2748,7 @@ Parameters
 a : numpy.ndarray (np.float32 or np.float64)
     The input data
 axes : list of integers
-    The axes on which the transform is carried out.
+    The axes along which the transform is carried out.
     If not set, all axes will be transformed.
 fct : float
     Normalization factor
@@ -2704,6 +2767,7 @@ np.ndarray (same shape and data type as a)
 PYBIND11_MODULE(pypocketfft, m)
   {
   using namespace pybind11::literals;
+  using namespace pocketfft_private;
 
   m.doc() = pypocketfft_DS;
   m.def("fftn",&fftn, fftn_DS, "a"_a, "axes"_a=py::none(), "fct"_a=1.,
@@ -2711,12 +2775,12 @@ PYBIND11_MODULE(pypocketfft, m)
   m.def("ifftn",&ifftn, ifftn_DS, "a"_a, "axes"_a=py::none(), "fct"_a=1.,
     "inplace"_a=false);
   m.def("rfftn",&rfftn, rfftn_DS, "a"_a, "axes"_a=py::none(), "fct"_a=1.);
-  m.def("rfft_scipy",&rfft_scipy, "a"_a, "axis"_a, "fct"_a=1.,
-    "inplace"_a=false);
-  m.def("irfft_scipy",&irfft_scipy, "a"_a, "axis"_a, "fct"_a=1.,
+  m.def("rfft_scipy",&rfft_scipy, rfft_scipy_DS, "a"_a, "axis"_a, "fct"_a=1.,
     "inplace"_a=false);
   m.def("irfftn",&irfftn, irfftn_DS, "a"_a, "axes"_a=py::none(), "lastsize"_a=0,
     "fct"_a=1.);
+  m.def("irfft_scipy",&irfft_scipy, irfft_scipy_DS, "a"_a, "axis"_a, "fct"_a=1.,
+    "inplace"_a=false);
   m.def("hartley",&hartley, hartley_DS, "a"_a, "axes"_a=py::none(), "fct"_a=1.,
     "inplace"_a=false);
   m.def("hartley2",&hartley2, "a"_a, "axes"_a=py::none(), "fct"_a=1.,