Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

ENH: Provide a hook for gufuncs to process core dimensions. #26908

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 8 commits into from
Jul 22, 2024
Merged

ENH: Provide a hook for gufuncs to process core dimensions. #26908

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
a3f61e1
ENH: Provide a hook for gufuncs to process core dimensions.
WarrenWeckesser Jul 10, 2024
7ab2434
TST: Add a test of conv1d_full with both input core dims 0
WarrenWeckesser Jul 15, 2024
aedce12
TST: Remove the test gufunc wl1_pdist
WarrenWeckesser Jul 15, 2024
a8dcfa8
DOC: Add comment about the PyUFunc_ProcessCoreDimsFunc type.
WarrenWeckesser Jul 15, 2024
ee89a1f
DOC: Copy-edit PyUFunc_ProcessCoreDimsFunc comment.
WarrenWeckesser Jul 15, 2024
e71f7ae
DOC: Work on the reference doc for the process_core_dim_sizes function.
WarrenWeckesser Jul 16, 2024
30418e0
DOC: Copy-edit, fix a few typos.
WarrenWeckesser Jul 16, 2024
0586b38
MAINT: Update 2.1 C-API version and use it
seberg Jul 22, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 7 additions & 0 deletions 7 numpy/_core/include/numpy/ufuncobject.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,6 +65,9 @@ typedef int (PyUFunc_TypeResolutionFunc)(
PyObject *type_tup,
PyArray_Descr **out_dtypes);

typedef int (PyUFunc_ProcessCoreDimsFunc)(
mhvk marked this conversation as resolved.
Show resolved Hide resolved
struct _tagPyUFuncObject *ufunc,
npy_intp *core_dim_sizes);

typedef struct _tagPyUFuncObject {
PyObject_HEAD
Expand Down Expand Up @@ -191,6 +194,10 @@ typedef struct _tagPyUFuncObject {
/* A PyListObject of `(tuple of DTypes, ArrayMethod/Promoter)` */
PyObject *_loops;
#endif
#if NPY_FEATURE_VERSION >= NPY_2_1_API_VERSION
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just noticed, but just noting to not forget: NPY_2_1_API_VERSION isn't defined yet I think. But don't worry about it, I can just do it (maybe even later today) since evertying else should be good.

/* User defined function to process core dimensions of a gufunc. */
PyUFunc_ProcessCoreDimsFunc *process_core_dims_func;
#endif
} PyUFuncObject;

#include "arrayobject.h"
Expand Down
252 changes: 252 additions & 0 deletions 252 numpy/_core/src/umath/_umath_tests.c.src
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -761,6 +761,205 @@ add_INT32_negative_indexed(PyObject *module, PyObject *dict) {
return 0;
}

// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
// Define the gufunc 'conv1d_full'
// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

#define MIN(a, b) (((a) < (b)) ? (a) : (b))
#define MAX(a, b) (((a) < (b)) ? (b) : (a))

int conv1d_full_process_core_dims(PyUFuncObject *ufunc,
npy_intp *core_dim_sizes)
{
//
// core_dim_sizes will hold the core dimensions [m, n, p].
// p will be -1 if the caller did not provide the out argument.
//
npy_intp m = core_dim_sizes[0];
npy_intp n = core_dim_sizes[1];
npy_intp p = core_dim_sizes[2];
npy_intp required_p = m + n - 1;

if (p == -1) {
core_dim_sizes[2] = required_p;
return 0;
}
if (p != required_p) {
PyErr_Format(PyExc_ValueError,
"conv1d_full: the core dimension p of the out parameter "
"does not equal m + n - 1, where m and n are the core "
"dimensions of the inputs x and y; got m=%zd and n=%zd so "
"p must be %zd, but got p=%zd.",
m, n, required_p, p);
return -1;
}
return 0;
}

static void
conv1d_full_double_loop(char **args,
npy_intp const *dimensions,
npy_intp const *steps,
void *NPY_UNUSED(func))
{
// Input and output arrays
char *p_x = args[0];
char *p_y = args[1];
char *p_out = args[2];
// Number of loops of pdist calculations to execute.
npy_intp nloops = dimensions[0];
// Core dimensions
npy_intp m = dimensions[1];
npy_intp n = dimensions[2];
npy_intp p = dimensions[3]; // Must be m + n - 1.
// Core strides
npy_intp x_stride = steps[0];
npy_intp y_stride = steps[1];
npy_intp out_stride = steps[2];
// Inner strides
npy_intp x_inner_stride = steps[3];
npy_intp y_inner_stride = steps[4];
npy_intp out_inner_stride = steps[5];

for (npy_intp loop = 0; loop < nloops; ++loop, p_x += x_stride,
p_y += y_stride,
p_out += out_stride) {
// Basic implementation of 1d convolution
for (npy_intp k = 0; k < p; ++k) {
double sum = 0.0;
for (npy_intp i = MAX(0, k - n + 1); i < MIN(m, k + 1); ++i) {
double x_i = *(double *)(p_x + i*x_inner_stride);
double y_k_minus_i = *(double *)(p_y + (k - i)*y_inner_stride);
sum += x_i * y_k_minus_i;
}
*(double *)(p_out + k*out_inner_stride) = sum;
}
}
}

static PyUFuncGenericFunction conv1d_full_functions[] = {
(PyUFuncGenericFunction) &conv1d_full_double_loop
};
static void *const conv1d_full_data[] = {NULL};
static const char conv1d_full_typecodes[] = {NPY_DOUBLE, NPY_DOUBLE, NPY_DOUBLE};


// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
// Define the gufunc 'wl1_pdist' (weighted l1 pairwise distances)
//
// 'wl1_pdist' has shape signature (m,n?),(n?)->(p), where p must
// be m*(m-1)/2. The second input parameter provides weights for
// the components.
// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

int wl1_pdist_process_core_dims(PyUFuncObject *ufunc,
npy_intp *core_dim_sizes)
{
npy_intp m_even, m_odd;
npy_intp required_p;

//
// core_dim_sizes will hold the core dimensions [m, n, p].
// p will be -1 if the caller did not provide the out argument.
//
npy_intp m = core_dim_sizes[0];
npy_intp p = core_dim_sizes[2];
if (m == 0) {
PyErr_SetString(PyExc_ValueError,
"pdist: the core dimension m of the input parameter "
"x must be at least 1; got 0.");
return -1;
}

// Identify the even and odd factors of the expression m*(m - 1).
npy_intp r = m % 2;
m_even = m - r;
m_odd = m - (1 - r);

if (m_even/2 > NPY_MAX_INTP/m_odd) {
PyErr_Format(PyExc_ValueError,
"wl1_pdist: the core dimension m=%zd of the input parameter "
"x is too big; the calculation of the output size "
"p = m*(m - 1)/2 results in integer overflow.", m);
return -1;
}

required_p = m_odd*(m_even/2);

if (p == -1) {
core_dim_sizes[2] = required_p;
return 0;
}
if (p != required_p) {
PyErr_Format(PyExc_ValueError,
"wl1_pdist: the core dimension p of the out parameter "
"does not equal m*(m - 1)/2, where m is the first core "
"dimension of the input x; got m=%zd, so p must be %zd, "
"but got p=%zd).",
m, required_p, p);
return -1;
}
return 0;
}

static void
wl1_pdist_double_loop(char **args,
npy_intp const *dimensions,
npy_intp const *steps,
void *NPY_UNUSED(func))
{
// Input and output arrays
char *p_x = args[0];
char *p_w = args[1];
char *p_out = args[2];
// Number of loops of pdist calculations to execute.
npy_intp nloops = dimensions[0];
// Core dimensions
npy_intp m = dimensions[1];
npy_intp n = dimensions[2];
// npy_intp p = dimensions[3]; // Must be m*(m-1)/2; unused here.
// Core strides
npy_intp x_stride = steps[0];
npy_intp w_stride = steps[1];
npy_intp out_stride = steps[2];
// x array strides
npy_intp x_row_stride = steps[3];
npy_intp x_col_stride = steps[4];
// w array strides
npy_intp w_inner_stride = steps[5];
// out array strides
npy_intp out_inner_stride = steps[6];

for (npy_intp loop = 0; loop < nloops; ++loop, p_x += x_stride,
p_w += w_stride,
p_out += out_stride) {
npy_intp k_out = 0;
for (npy_intp i = 0; i < m - 1; ++i) {
char *p_rowi = p_x + i*x_row_stride;
for (npy_intp j = i + 1; j < m; ++j) {
char *p_rowj = p_x + j*x_row_stride;
double sum = 0.0;
for (npy_intp k = 0; k < n; ++k) {
double x1 = *(double *)(p_rowi + k*x_col_stride);
double x2 = *(double *)(p_rowj + k*x_col_stride);
double w = *(double *)(p_w + k*w_inner_stride);
sum += w*npy_fabs(x1 - x2);
}
*(double *)(p_out + k_out*out_inner_stride) = sum;
++k_out;
}
}
}
}


static PyUFuncGenericFunction wl1_pdist_functions[] = {
(PyUFuncGenericFunction) &wl1_pdist_double_loop
};
static void *const wl1_pdist_data[] = {NULL};
static const char wl1_pdist_typecodes[] = {NPY_DOUBLE, NPY_DOUBLE, NPY_DOUBLE};


static PyMethodDef UMath_TestsMethods[] = {
{"test_signature", UMath_Tests_test_signature, METH_VARARGS,
"Test signature parsing of ufunc. \n"
Expand Down Expand Up @@ -830,6 +1029,59 @@ PyMODINIT_FUNC PyInit__umath_tests(void) {
return NULL;
}

// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
// Define the gufunc 'conv1d_full'
// Shape signature is (m),(n)->(p) where p must be m + n - 1.
// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

PyUFuncObject *gufunc = (PyUFuncObject *) PyUFunc_FromFuncAndDataAndSignature(
conv1d_full_functions,
conv1d_full_data,
conv1d_full_typecodes,
1, 2, 1, PyUFunc_None, "conv1d_full",
"convolution of x and y ('full' mode)",
0, "(m),(n)->(p)");
if (gufunc == NULL) {
Py_DECREF(m);
return NULL;
}
gufunc->process_core_dims_func = &conv1d_full_process_core_dims;

int status = PyModule_AddObject(m, "conv1d_full", (PyObject *) gufunc);
if (status == -1) {
Py_DECREF(gufunc);
Py_DECREF(m);
return NULL;
}

// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
// Define the gufunc 'wl1_pdist'
// Shape signature is (m,n?),(n?)->(p), where p must be m*(m-1)/2.
// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

gufunc = (PyUFuncObject *) PyUFunc_FromFuncAndDataAndSignature(
wl1_pdist_functions,
wl1_pdist_data,
wl1_pdist_typecodes,
1, 2, 1, PyUFunc_None, "wl1_pdist",
"pairwise l1 distance (city block distance) "
"of rows in x",
0, "(m,n?),(n?)->(p)");
if (gufunc == NULL) {
Py_DECREF(m);
return NULL;
}
gufunc->process_core_dims_func = &wl1_pdist_process_core_dims;

status = PyModule_AddObject(m, "wl1_pdist", (PyObject *) gufunc);
if (status == -1) {
Py_DECREF(gufunc);
Py_DECREF(m);
return NULL;
}

// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

#if Py_GIL_DISABLED
// signal this module supports running with the GIL disabled
PyUnstable_Module_SetGIL(m, Py_MOD_GIL_NOT_USED);
Expand Down
9 changes: 9 additions & 0 deletions 9 numpy/_core/src/umath/ufunc_object.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1590,6 +1590,13 @@ _get_coredim_sizes(PyUFuncObject *ufunc, PyArrayObject **op,
}
}

if (ufunc->process_core_dims_func != NULL) {
int status = ufunc->process_core_dims_func(ufunc, core_dim_sizes);
if (status != 0) {
return -1;
}
}

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So one thing we do not do here is sanity check the output. I.e. if the core_dim_size is already set but the user mutates it, we could be in trouble.

To be clear, I think that is OK, but it should be documented.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It is easy enough to pass a copy to process_core_dims_func and verify on return that nothing was changed that shouldn't be changed. I'll add that.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think it matters, just occured to me that we should document that you may only validate (raise an error) or fill in -1.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I wavered a bit about whether or not to test, but in the end, making a mistake in the processing here is no different from making a mistake in the actual function that gets called - one can get a core dump regardless. So, agreed with @seberg that a check after the call is not needed.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

OK, I won't add the check!

/*
* Make sure no core dimension is unspecified.
*/
Expand Down Expand Up @@ -4689,6 +4696,8 @@ PyUFunc_FromFuncAndDataAndSignatureAndIdentity(PyUFuncGenericFunction *func, voi
/* Type resolution and inner loop selection functions */
ufunc->type_resolver = &PyUFunc_DefaultTypeResolver;

ufunc->process_core_dims_func = NULL;

ufunc->op_flags = NULL;
ufunc->_loops = NULL;
if (nin + nout != 0) {
Expand Down
Loading
Loading
Morty Proxy This is a proxified and sanitized view of the page, visit original site.