src/multiarray/lowlevel_strided_loops.c.src File Reference

#include "Python.h"
#include "structmember.h"
#include <numpy/arrayobject.h>
#include <numpy/npy_cpu.h>
#include <numpy/halffloat.h>
#include "lowlevel_strided_loops.h"

Defines
#define	PY_SSIZE_T_CLEAN
#define	NPY_NO_DEPRECATED_API   NPY_API_VERSION
#define	_MULTIARRAYMODULE
#define	_ALIGN(type)   offsetof(struct {char c; type v;}, v)
#define	NPY_USE_UNALIGNED_ACCESS   0
#define	_NPY_NOP1(x)   (x)
#define	_NPY_NOP2(x)   (x)
#define	_NPY_NOP4(x)   (x)
#define	_NPY_NOP8(x)   (x)
#define	_NPY_SWAP2(x)   npy_bswap2(x)
#define	_NPY_SWAP4(x)   npy_bswap4(x)
#define	_NPY_SWAP_PAIR4(x)
#define	_NPY_SWAP8(x)   npy_bswap8(x)
#define	_NPY_SWAP_PAIR8(x)
#define	_NPY_SWAP_INPLACE2(x)   npy_bswap2_unaligned(x)
#define	_NPY_SWAP_INPLACE4(x)   npy_bswap4_unaligned(x)
#define	_NPY_SWAP_INPLACE8(x)   npy_bswap8_unaligned(x)
#define	_NPY_SWAP_INPLACE16(x)
Functions
static void	_strided_to_strided (char *dst, npy_intp dst_stride, char *src, npy_intp src_stride, npy_intp N, npy_intp src_itemsize, NpyAuxData *NPY_UNUSED(data))
static void	_swap_strided_to_strided (char *dst, npy_intp dst_stride, char *src, npy_intp src_stride, npy_intp N, npy_intp src_itemsize, NpyAuxData *NPY_UNUSED(data))
static void	_swap_pair_strided_to_strided (char *dst, npy_intp dst_stride, char *src, npy_intp src_stride, npy_intp N, npy_intp src_itemsize, NpyAuxData *NPY_UNUSED(data))
static void	_contig_to_contig (char *dst, npy_intp NPY_UNUSED(dst_stride), char *src, npy_intp NPY_UNUSED(src_stride), npy_intp N, npy_intp src_itemsize, NpyAuxData *NPY_UNUSED(data))
NPY_NO_EXPORT PyArray_StridedUnaryOp *	PyArray_GetStridedCopyFn (int aligned, npy_intp src_stride, npy_intp dst_stride, npy_intp itemsize)
NPY_NO_EXPORT PyArray_StridedUnaryOp *	function (int aligned, npy_intp src_stride, npy_intp dst_stride, npy_intp itemsize)
NPY_NO_EXPORT PyArray_StridedUnaryOp *	PyArray_GetStridedNumericCastFn (int aligned, npy_intp src_stride, npy_intp dst_stride, int src_type_num, int dst_type_num)
NPY_NO_EXPORT npy_intp	PyArray_TransferNDimToStrided (npy_intp ndim, char *dst, npy_intp dst_stride, char *src, npy_intp *src_strides, npy_intp src_strides_inc, npy_intp *coords, npy_intp coords_inc, npy_intp *shape, npy_intp shape_inc, npy_intp count, npy_intp src_itemsize, PyArray_StridedUnaryOp *stransfer, NpyAuxData *data)
NPY_NO_EXPORT npy_intp	PyArray_TransferStridedToNDim (npy_intp ndim, char *dst, npy_intp *dst_strides, npy_intp dst_strides_inc, char *src, npy_intp src_stride, npy_intp *coords, npy_intp coords_inc, npy_intp *shape, npy_intp shape_inc, npy_intp count, npy_intp src_itemsize, PyArray_StridedUnaryOp *stransfer, NpyAuxData *data)
NPY_NO_EXPORT npy_intp	PyArray_TransferMaskedStridedToNDim (npy_intp ndim, char *dst, npy_intp *dst_strides, npy_intp dst_strides_inc, char *src, npy_intp src_stride, npy_uint8 *mask, npy_intp mask_stride, npy_intp *coords, npy_intp coords_inc, npy_intp *shape, npy_intp shape_inc, npy_intp count, npy_intp src_itemsize, PyArray_MaskedStridedUnaryOp *stransfer, NpyAuxData *data)
NPY_NO_EXPORT int mapiter_trivial_	name (PyArrayObject *self, PyArrayObject *ind, PyArrayObject *result)
NPY_NO_EXPORT int mapiter_	name (PyArrayMapIterObject *mit)

---

Define Documentation

#define _ALIGN

(

type

)

offsetof(struct {char c; type v;}, v)

used for some alignment checks

#define _MULTIARRAYMODULE

#define _NPY_NOP1

(

x

)

(x)

#define _NPY_NOP2

(

x

)

(x)

#define _NPY_NOP4

(

x

)

(x)

#define _NPY_NOP8

(

x

)

(x)

#define _NPY_SWAP2

(

x

)

npy_bswap2(x)

#define _NPY_SWAP4

(

x

)

npy_bswap4(x)

#define _NPY_SWAP8

(

x

)

npy_bswap8(x)

#define _NPY_SWAP_INPLACE16

(

x

)

Value:

{ \
        char a = (x)[0]; (x)[0] = (x)[15]; (x)[15] = a; \
        a = (x)[1]; (x)[1] = (x)[14]; (x)[14] = a; \
        a = (x)[2]; (x)[2] = (x)[13]; (x)[13] = a; \
        a = (x)[3]; (x)[3] = (x)[12]; (x)[12] = a; \
        a = (x)[4]; (x)[4] = (x)[11]; (x)[11] = a; \
        a = (x)[5]; (x)[5] = (x)[10]; (x)[10] = a; \
        a = (x)[6]; (x)[6] = (x)[9]; (x)[9] = a; \
        a = (x)[7]; (x)[7] = (x)[8]; (x)[8] = a; \
        }

#define _NPY_SWAP_INPLACE2

(

x

)

npy_bswap2_unaligned(x)

#define _NPY_SWAP_INPLACE4

(

x

)

npy_bswap4_unaligned(x)

#define _NPY_SWAP_INPLACE8

(

x

)

npy_bswap8_unaligned(x)

#define _NPY_SWAP_PAIR4

(

x

)

Value:

(((((npy_uint32)x)&0xffu) << 8) | \
                       ((((npy_uint32)x)&0xff00u) >> 8) | \
                       ((((npy_uint32)x)&0xff0000u) << 8) | \
                       ((((npy_uint32)x)&0xff000000u) >> 8))

#define _NPY_SWAP_PAIR8

(

x

)

Value:

(((((npy_uint64)x)&0xffULL) << 24) | \
                       ((((npy_uint64)x)&0xff00ULL) << 8) | \
                       ((((npy_uint64)x)&0xff0000ULL) >> 8) | \
                       ((((npy_uint64)x)&0xff000000ULL) >> 24) | \
                       ((((npy_uint64)x)&0xff00000000ULL) << 24) | \
                       ((((npy_uint64)x)&0xff0000000000ULL) << 8) | \
                       ((((npy_uint64)x)&0xff000000000000ULL) >> 8) | \
                       ((((npy_uint64)x)&0xff00000000000000ULL) >> 24))

#define NPY_NO_DEPRECATED_API   NPY_API_VERSION

#define NPY_USE_UNALIGNED_ACCESS   0

Disable harmless compiler warning "4116: unnamed type definition in parentheses" which is caused by the _ALIGN macro.

x86 platform works with unaligned access but the compiler is allowed to assume all data is aligned to its size by the C standard. This means it can vectorize instructions peeling only by the size of the type, if the data is not aligned to this size one ends up with data not correctly aligned for SSE instructions (16 byte). So this flag can only be enabled if autovectorization is disabled.

#define PY_SSIZE_T_CLEAN

---

Function Documentation

static void _contig_to_contig	(	char *	dst,
		npy_intp	NPY_UNUSEDdst_stride,
		char *	src,
		npy_intp	NPY_UNUSEDsrc_stride,
		npy_intp	N,
		npy_intp	src_itemsize,
		NpyAuxData *	NPY_UNUSEDdata
	)		[static]

Referenced by _swap_pair_strided_to_strided().

static void _strided_to_strided	(	char *	dst,
		npy_intp	dst_stride,
		char *	src,
		npy_intp	src_stride,
		npy_intp	N,
		npy_intp	src_itemsize,
		NpyAuxData *	NPY_UNUSEDdata
	)		[static]

******* STRIDED COPYING/SWAPPING SPECIALIZED FUNCTIONS ********

begin repeat

#elsize = 1, 2, 4, 8, 16# #elsize_half = 0, 1, 2, 4, 8# #type = npy_uint8, npy_uint16, npy_uint32, npy_uint64, npy_uint128#

begin repeat1

oper = strided_to_strided, strided_to_contig,

contig_to_strided, contig_to_contig#

System Message: WARNING/2 (<string>, line 4) Definition list ends without a blank line; unexpected unindent.

#src_contig = 0, 0, 1 ,1# #dst_contig = 0, 1, 0 ,1#

begin repeat2

#swap = _NPY_NOP, _NPY_NOP, _NPY_SWAP_INPLACE, _NPY_SWAP,

_NPY_SWAP_INPLACE, _NPY_SWAP_PAIR#

System Message: WARNING/2 (<string>, line 4) Definition list ends without a blank line; unexpected unindent.

#prefix = , _aligned, _swap, _aligned_swap, _swap_pair, _aligned_swap_pair# #is_aligned = 0, 1, 0, 1, 0, 1# #minelsize = 1, 1, 2, 2, 4, 4# #is_swap = 0, 0, 1, 1, 2, 2#

<

&#64;elsize&#64; >= &#64;minelsize&#64;

end repeat2*

end repeat1*

end repeat*

static void _swap_pair_strided_to_strided	(	char *	dst,
		npy_intp	dst_stride,
		char *	src,
		npy_intp	src_stride,
		npy_intp	N,
		npy_intp	src_itemsize,
		NpyAuxData *	NPY_UNUSEDdata
	)		[static]

general in-place swap

general in-place swap

References _contig_to_contig().

static void _swap_strided_to_strided	(	char *	dst,
		npy_intp	dst_stride,
		char *	src,
		npy_intp	src_stride,
		npy_intp	N,
		npy_intp	src_itemsize,
		NpyAuxData *	NPY_UNUSEDdata
	)		[static]

general in-place swap

NPY_NO_EXPORT PyArray_StridedUnaryOp* function	(	int	aligned,
		npy_intp	src_stride,
		npy_intp	dst_stride,
		npy_intp	itemsize
	)

PyArray_GetStridedCopySwapFn and PyArray_GetStridedCopySwapPairFn are nearly identical, so can do a repeat for them.

begin repeat

function = PyArray_GetStridedCopySwapFn, PyArray_GetStridedCopySwapPairFn# #tag = , _pair# #not_pair = 1, 0#

Skip the "unaligned" versions on CPUs which support unaligned memory accesses.

<

!NPY_USE_UNALIGNED_ACCESS

contiguous dst

constant src

begin repeat1

#elsize = 2, 4, 8, 16#

end repeat1*

contiguous src

begin repeat1

#elsize = 2, 4, 8, 16#

end repeat1*

general src

begin repeat1

#elsize = 2, 4, 8, 16#

end repeat1*

general dst

constant src

begin repeat1

#elsize = 2, 4, 8, 16#

end repeat1*

contiguous src

begin repeat1

#elsize = 2, 4, 8, 16#

end repeat1*

begin repeat1

#elsize = 2, 4, 8, 16#

end repeat1*

contiguous dst

contiguous src

begin repeat1

#elsize = 2, 4, 8, 16#

end repeat1*

general src

begin repeat1

#elsize = 2, 4, 8, 16#

end repeat1*

general dst

contiguous src

begin repeat1

#elsize = 2, 4, 8, 16#

end repeat1*

general src

begin repeat1

#elsize = 2, 4, 8, 16#

end repeat1*

NPY_NO_EXPORT int mapiter_trivial_ name	(	PyArrayObject *	self,
		PyArrayObject *	ind,
		PyArrayObject *	result
	)

System Message: ERROR/3 (<string>, line 1) Document or section may not begin with a transition.

---

System Message: ERROR/3 (<string>, line 1) Document may not end with a transition.

************ MapIter (Advanced indexing) Get/Set ***************

System Message: ERROR/3 (<string>, line 1) Document or section may not begin with a transition.

---

System Message: ERROR/3 (<string>, line 1) Document may not end with a transition.

begin repeat

name = set, get# #isget = 0, 1#

Advanded indexing iteration of arrays when there is a single indexing array which has the same memory order as the value array and both can be trivally iterated (single stride, aligned, no casting necessary).

Check the indices beforehand

Reset the ind_ptr and itersize, due to broadcasting it is always the size of ind.

Optimization for aligned types that do not need the api

begin repeat1

#elsize = 1, 2, 4, 8, 0# #copytype = npy_uint8, npy_uint16, npy_uint32, npy_uint64, 0#

<

!&#64;isget&#64;

end repeat1*

NPY_NO_EXPORT int mapiter_ name

(

PyArrayMapIterObject *

mit

)

General advanced indexing iteration.

Cached mit info

Constant information

Fill constant information

Alignment information (swapping is never needed, since we buffer), could also check extra_op is buffered, but it should rarely matter.

Item by item copy situation, the operand is buffered so use copyswap.

We have only one iterator handling everything

****** Optimized inner loops without subspace ********

begin repeat1

#one_iter = 1, 0# #numiter = 1, numiter#

Optimization for aligned types that do not need the api

begin repeat2

#elsize = 1, 2, 4, 8, 0# #copytype = npy_uint8, npy_uint16, npy_uint32, npy_uint64, 0#

Outer iteration (safe because mit->size != 0)

When the API is needed the casting might fail TODO: (only if buffering is enabled).

advance indexing arrays

advance extra operand

end repeat2*

end repeat1*

************* Nested Iteration Situation **************

<

has three states

Use strided transfer functions for the inner loop

Get a dtype transfer function, since there are no buffers, this is safe.

subspace is trivially iterable. manipulate pointers to avoid expensive resetting

begin repeat1

#one_iter = 1, 0# #numiter = 1, numiter#

Outer iteration (safe because mit->size != 0)

Resetting is slow, so try to avoid resetting if subspace iteration is trivial. Watch out: reset_offsets are kept outside of the loop, assuming the subspaces of different external iterations share the same structure.

slower resetting: first iteration or non-trivial subspace

(can't really fail, since no buffering necessary)

reset_offsets are nonzero for negative strides.

use the faster adjustment further on

faster resetting if the subspace iteration is trival. reset_offsets are zero for positive strides, for negative strides this shifts the pointer to the last item.

When the API is needed the casting might fail TODO: Could only check if casting is unsafe, or even just

System Message: ERROR/3 (<string>, line 3) Unexpected indentation.

<blockquote> not at all...</blockquote>

end repeat1*

NPY_NO_EXPORT PyArray_StridedUnaryOp* PyArray_GetStridedCopyFn	(	int	aligned,
		npy_intp	src_stride,
		npy_intp	dst_stride,
		npy_intp	itemsize
	)

Gives back a function pointer to a specialized function for copying strided memory. Returns NULL if there is a problem with the inputs.

aligned:

Should be 1 if the src and dst pointers are always aligned, 0 otherwise.

src_stride:

Should be the src stride if it will always be the same, NPY_MAX_INTP otherwise.

dst_stride:

Should be the dst stride if it will always be the same, NPY_MAX_INTP otherwise.

itemsize:

Should be the item size if it will always be the same, 0 otherwise.

Skip the "unaligned" versions on CPUs which support unaligned memory accesses.

<

!NPY_USE_UNALIGNED_ACCESS

contiguous dst

constant src

begin repeat

#elsize = 1, 2, 4, 8, 16#

end repeat*

contiguous src

general src

begin repeat

#elsize = 1, 2, 4, 8, 16#

end repeat*

general dst

constant src

begin repeat

#elsize = 1, 2, 4, 8, 16#

end repeat*

contiguous src

begin repeat

#elsize = 1, 2, 4, 8, 16#

end repeat*

begin repeat

#elsize = 1, 2, 4, 8, 16#

end repeat*

contiguous dst

contiguous src

general src

begin repeat

#elsize = 2, 4, 8, 16#

end repeat*

general dst

contiguous src

begin repeat

#elsize = 2, 4, 8, 16#

end repeat*

general src

begin repeat

#elsize = 2, 4, 8, 16#

end repeat*

Referenced by _strided_to_strided_truncate_copy().

NPY_NO_EXPORT PyArray_StridedUnaryOp* PyArray_GetStridedNumericCastFn	(	int	aligned,
		npy_intp	src_stride,
		npy_intp	dst_stride,
		int	src_type_num,
		int	dst_type_num
	)

end repeat*

******* STRIDED CASTING SPECIALIZED FUNCTIONS ********

begin repeat <blockquote>

#NAME1 = BOOL,

UBYTE, USHORT, UINT, ULONG, ULONGLONG, BYTE, SHORT, INT, LONG, LONGLONG, HALF, FLOAT, DOUBLE, LONGDOUBLE, CFLOAT, CDOUBLE, CLONGDOUBLE#

#name1 = bool,

ubyte, ushort, uint, ulong, ulonglong, byte, short, int, long, longlong, half, float, double, longdouble, cfloat, cdouble, clongdouble#

#type1 = npy_bool,

npy_ubyte, npy_ushort, npy_uint, npy_ulong, npy_ulonglong, npy_byte, npy_short, npy_int, npy_long, npy_longlong, npy_half, npy_float, npy_double, npy_longdouble, npy_cfloat, npy_cdouble, npy_clongdouble#

#rtype1 = npy_bool,

npy_ubyte, npy_ushort, npy_uint, npy_ulong, npy_ulonglong, npy_byte, npy_short, npy_int, npy_long, npy_longlong, npy_half, npy_float, npy_double, npy_longdouble, npy_float, npy_double, npy_longdouble#

System Message: WARNING/2 (<string>, line 23) Definition list ends without a blank line; unexpected unindent.

#is_bool1 = 1, 0*17# #is_half1 = 0*11, 1, 0*6# #is_float1 = 0*12, 1, 0, 0, 1, 0, 0# #is_double1 = 0*13, 1, 0, 0, 1, 0# #is_complex1 = 0*15, 1*3# </blockquote>

begin repeat1 <blockquote>

#NAME2 = BOOL,

UBYTE, USHORT, UINT, ULONG, ULONGLONG, BYTE, SHORT, INT, LONG, LONGLONG, HALF, FLOAT, DOUBLE, LONGDOUBLE, CFLOAT, CDOUBLE, CLONGDOUBLE#

#name2 = bool,

ubyte, ushort, uint, ulong, ulonglong, byte, short, int, long, longlong, half, float, double, longdouble, cfloat, cdouble, clongdouble#

#type2 = npy_bool,

npy_ubyte, npy_ushort, npy_uint, npy_ulong, npy_ulonglong, npy_byte, npy_short, npy_int, npy_long, npy_longlong, npy_half, npy_float, npy_double, npy_longdouble, npy_cfloat, npy_cdouble, npy_clongdouble#

#rtype2 = npy_bool,

npy_ubyte, npy_ushort, npy_uint, npy_ulong, npy_ulonglong, npy_byte, npy_short, npy_int, npy_long, npy_longlong, npy_half, npy_float, npy_double, npy_longdouble, npy_float, npy_double, npy_longdouble#

System Message: WARNING/2 (<string>, line 23) Definition list ends without a blank line; unexpected unindent.

#is_bool2 = 1, 0*17# #is_half2 = 0*11, 1, 0*6# #is_float2 = 0*12, 1, 0, 0, 1, 0, 0# #is_double2 = 0*13, 1, 0, 0, 1, 0# #is_complex2 = 0*15, 1*3# </blockquote>

begin repeat2

#prefix = _aligned,,_aligned_contig,_contig# #aligned = 1,0,1,0# #contig = 0,0,1,1#

end repeat2*

end repeat1*

end repeat*

begin repeat <blockquote>

#NAME1 = BOOL,

UBYTE, USHORT, UINT, ULONG, ULONGLONG, BYTE, SHORT, INT, LONG, LONGLONG, HALF, FLOAT, DOUBLE, LONGDOUBLE, CFLOAT, CDOUBLE, CLONGDOUBLE#

#name1 = bool,

ubyte, ushort, uint, ulong, ulonglong, byte, short, int, long, longlong, half, float, double, longdouble, cfloat, cdouble, clongdouble#

#type1 = npy_bool,

npy_ubyte, npy_ushort, npy_uint, npy_ulong, npy_ulonglong, npy_byte, npy_short, npy_int, npy_long, npy_longlong, npy_half, npy_float, npy_double, npy_longdouble, npy_cfloat, npy_cdouble, npy_clongdouble#

</blockquote>

printf("test fn d - second dn", NPY_&#64;NAME1&#64;, dst_type_num);

begin repeat1 <blockquote>

#NAME2 = BOOL,

UBYTE, USHORT, UINT, ULONG, ULONGLONG, BYTE, SHORT, INT, LONG, LONGLONG, HALF, FLOAT, DOUBLE, LONGDOUBLE, CFLOAT, CDOUBLE, CLONGDOUBLE#

#name2 = bool,

ubyte, ushort, uint, ulong, ulonglong, byte, short, int, long, longlong, half, float, double, longdouble, cfloat, cdouble, clongdouble#

#type2 = npy_bool,

npy_ubyte, npy_ushort, npy_uint, npy_ulong, npy_ulonglong, npy_byte, npy_short, npy_int, npy_long, npy_longlong, npy_half, npy_float, npy_double, npy_longdouble, npy_cfloat, npy_cdouble, npy_clongdouble#

</blockquote>

printf("ret fn d dn", NPY_&#64;NAME1&#64;, NPY_&#64;NAME2&#64;);

end repeat1*

printf("switched test fn d - second dn", NPY_&#64;NAME1&#64;, dst_type_num);

end repeat*

NPY_NO_EXPORT npy_intp PyArray_TransferMaskedStridedToNDim	(	npy_intp	ndim,
		char *	dst,
		npy_intp *	dst_strides,
		npy_intp	dst_strides_inc,
		char *	src,
		npy_intp	src_stride,
		npy_uint8 *	mask,
		npy_intp	mask_stride,
		npy_intp *	coords,
		npy_intp	coords_inc,
		npy_intp *	shape,
		npy_intp	shape_inc,
		npy_intp	count,
		npy_intp	src_itemsize,
		PyArray_MaskedStridedUnaryOp *	stransfer,
		NpyAuxData *	data
	)

See documentation of arguments in lowlevel_strided_loops.h

Finish off dimension 0

If it's 1-dimensional, there's no more to copy

Adjust the src and dst pointers

Finish off dimension 1

If it's 2-dimensional, there's no more to copy

General-case loop for everything else

Iteration structure for dimensions 2 and up

Copy the coordinates and shape

Adjust the dst pointer from the dimension 0 and 1 loop

Increment to the next coordinate

If the last dimension rolled over, we're done

A loop for dimensions 0 and 1

References check_and_adjust_index(), NPY_AUXDATA_FREE, NPY_BEGIN_THREADS, NPY_BEGIN_THREADS_DEF, NPY_END_THREADS, and NpyIter_ResetBasePointers().

NPY_NO_EXPORT npy_intp PyArray_TransferNDimToStrided	(	npy_intp	ndim,
		char *	dst,
		npy_intp	dst_stride,
		char *	src,
		npy_intp *	src_strides,
		npy_intp	src_strides_inc,
		npy_intp *	coords,
		npy_intp	coords_inc,
		npy_intp *	shape,
		npy_intp	shape_inc,
		npy_intp	count,
		npy_intp	src_itemsize,
		PyArray_StridedUnaryOp *	stransfer,
		NpyAuxData *	data
	)

************ PRIMITIVE FLAT TO/FROM NDIM FUNCTIONS *************

See documentation of arguments in lowlevel_strided_loops.h

Finish off dimension 0

If it's 1-dimensional, there's no more to copy

Adjust the src and dst pointers

Finish off dimension 1

If it's 2-dimensional, there's no more to copy

General-case loop for everything else

Iteration structure for dimensions 2 and up

Copy the coordinates and shape

Adjust the src pointer from the dimension 0 and 1 loop

Increment to the next coordinate

If the last dimension rolled over, we're done

A loop for dimensions 0 and 1

NPY_NO_EXPORT npy_intp PyArray_TransferStridedToNDim	(	npy_intp	ndim,
		char *	dst,
		npy_intp *	dst_strides,
		npy_intp	dst_strides_inc,
		char *	src,
		npy_intp	src_stride,
		npy_intp *	coords,
		npy_intp	coords_inc,
		npy_intp *	shape,
		npy_intp	shape_inc,
		npy_intp	count,
		npy_intp	src_itemsize,
		PyArray_StridedUnaryOp *	stransfer,
		NpyAuxData *	data
	)

See documentation of arguments in lowlevel_strided_loops.h

Finish off dimension 0

If it's 1-dimensional, there's no more to copy

Adjust the src and dst pointers

Finish off dimension 1

If it's 2-dimensional, there's no more to copy

General-case loop for everything else

Iteration structure for dimensions 2 and up

Copy the coordinates and shape

Adjust the dst pointer from the dimension 0 and 1 loop

Increment to the next coordinate

If the last dimension rolled over, we're done

A loop for dimensions 0 and 1

Referenced by npyiter_allocate_buffers().