Skip to content

add new MKL vm functions #81

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 5 commits into from
Jul 12, 2025
Merged

add new MKL vm functions #81

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
6c7c143
add new vm functions
May 9, 2025
495a300
Merge branch 'main' into update-ufuncs
vtavana Jun 20, 2025
e0a666e
address comments
Jun 26, 2025
55ce77c
Merge branch 'main' into update-ufuncs
vtavana Jul 11, 2025
d0ad565
Merge branch 'main' into update-ufuncs
vtavana Jul 12, 2025
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
5 changes: 5 additions & 0 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,11 @@ All notable changes to this project will be documented in this file.
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

## [dev] (MM/DD/YYYY)

### Added
* Added mkl implementation for floating point data-types of `exp2`, `log2`, `fabs`, `copysign`, `nextafter`, `fmax`, `fmin` and `remainder` functions [gh-81](https://github.com/IntelPython/mkl_umath/pull/81)

## [0.2.0] (06/DD/2025)
This release updates `mkl_umath` to be aligned with both numpy-1.26.x and numpy-2.x.x.

Expand Down
208 changes: 144 additions & 64 deletions mkl_umath/src/mkl_umath_loops.c.src
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -126,7 +126,7 @@
} while(0)

/* for pointers p1, and p2 pointing at contiguous arrays n-elements of size s, are arrays disjoint or same
* when these conditions are not met VML functions may product incorrect output
* when these conditions are not met VML functions may produce incorrect output
*/
#define DISJOINT_OR_SAME(p1, p2, n, s) (((p1) == (p2)) || ((p2) + (n)*(s) < (p1)) || ((p1) + (n)*(s) < (p2)) )

Expand Down Expand Up @@ -219,6 +219,7 @@ divmod@c@(@type@ a, @type@ b, @type@ *modulus)
** FLOAT LOOPS **
*****************************************************************************
*/
/* TODO: Use MKL for pow, arctan2, fmod, hypot, i0 */

/**begin repeat
* Float types
Expand Down Expand Up @@ -334,22 +335,26 @@ mkl_umath_@TYPE@_exp(char **args, const npy_intp *dimensions, const npy_intp *st
* #scalarf = exp2f, exp2#
*/

/* TODO: Use VML */
void
mkl_umath_@TYPE@_exp2(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
const int contig = IS_UNARY_CONT(@type@, @type@);
const int disjoint_or_same = DISJOINT_OR_SAME(args[0], args[1], dimensions[0], sizeof(@type@));
const int can_vectorize = contig && disjoint_or_same;

UNARY_LOOP_DISPATCH(
@type@, @type@
,
can_vectorize
,
const @type@ in1 = *(@type@ *)ip1;
*(@type@ *)op1 = @scalarf@(in1);
)
if(can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD) {
CHUNKED_VML_CALL2(v@c@Exp2, dimensions[0], @type@, args[0], args[1]);
/* v@c@Exp2(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
} else {
UNARY_LOOP_DISPATCH(
@type@, @type@
,
can_vectorize
,
const @type@ in1 = *(@type@ *)ip1;
*(@type@ *)op1 = @scalarf@(in1);
)
}
}

/**end repeat**/
Expand Down Expand Up @@ -460,22 +465,27 @@ mkl_umath_@TYPE@_log(char **args, const npy_intp *dimensions, const npy_intp *st
* #scalarf = log2f, log2#
*/

/* TODO: Use VML */
void
mkl_umath_@TYPE@_log2(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
const int contig = IS_UNARY_CONT(@type@, @type@);
const int disjoint_or_same = DISJOINT_OR_SAME(args[0], args[1], dimensions[0], sizeof(@type@));
const int can_vectorize = contig && disjoint_or_same;

UNARY_LOOP_DISPATCH(
@type@, @type@
,
can_vectorize
,
const @type@ in1 = *(@type@ *)ip1;
*(@type@ *)op1 = @scalarf@(in1);
)
if (can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD)
{
CHUNKED_VML_CALL2(v@c@Log2, dimensions[0], @type@, args[0], args[1]);
/* v@c@Log2(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
} else {
UNARY_LOOP_DISPATCH(
@type@, @type@
,
can_vectorize
,
const @type@ in1 = *(@type@ *)ip1;
*(@type@ *)op1 = @scalarf@(in1);
)
}
}

/**end repeat**/
Expand Down Expand Up @@ -957,14 +967,20 @@ mkl_umath_@TYPE@_fabs(char **args, const npy_intp *dimensions, const npy_intp *s
const int disjoint_or_same = DISJOINT_OR_SAME(args[0], args[1], dimensions[0], sizeof(@type@));
const int can_vectorize = contig && disjoint_or_same;

UNARY_LOOP_DISPATCH(
@type@, @type@
,
can_vectorize
,
const @type@ in1 = *(@type@ *)ip1;
*(@type@ *)op1 = @scalarf@(in1);
)
if( can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD)
{
CHUNKED_VML_CALL2(v@c@Abs, dimensions[0], @type@, args[0], args[1]);
/* v@c@Abs(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
} else {
UNARY_LOOP_DISPATCH(
@type@, @type@
,
can_vectorize
,
const @type@ in1 = *(@type@ *)ip1;
*(@type@ *)op1 = @scalarf@(in1);
)
}
}

/**end repeat**/
Expand Down Expand Up @@ -1230,7 +1246,6 @@ pairwise_sum_@TYPE@(char *a, npy_intp n, npy_intp stride)
* #type = npy_float, npy_double#
* #TYPE = FLOAT, DOUBLE#
* #c = f, #
* #C = F, #
* #s = s, d#
* #SUPPORTED_BY_VML = 1, 1#
*/
Expand Down Expand Up @@ -1959,20 +1974,46 @@ mkl_umath_@TYPE@_spacing(char **args, const npy_intp *dimensions, const npy_intp
void
mkl_umath_@TYPE@_copysign(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
BINARY_LOOP {
const @type@ in1 = *(@type@ *)ip1;
const @type@ in2 = *(@type@ *)ip2;
*((@type@ *)op1)= copysign@c@(in1, in2);
#if @SUPPORTED_BY_VML@
const int contig = IS_BINARY_CONT(@type@, @type@);
const int disjoint_or_same1 = DISJOINT_OR_SAME(args[0], args[2], dimensions[0], sizeof(@type@));
const int disjoint_or_same2 = DISJOINT_OR_SAME(args[1], args[2], dimensions[0], sizeof(@type@));
const int can_vectorize = contig && disjoint_or_same1 && disjoint_or_same2;

if(can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD) {
CHUNKED_VML_CALL3(v@s@CopySign, dimensions[0], @type@, args[0], args[1], args[2]);
/* v@s@CopySign(dimensions[0], (@type@*) args[0], (@type@*) args[1], (@type@*) args[2]); */
} else
#endif
{
BINARY_LOOP {
const @type@ in1 = *(@type@ *)ip1;
const @type@ in2 = *(@type@ *)ip2;
*((@type@ *)op1)= copysign@c@(in1, in2);
}
}
}

void
mkl_umath_@TYPE@_nextafter(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
BINARY_LOOP {
const @type@ in1 = *(@type@ *)ip1;
const @type@ in2 = *(@type@ *)ip2;
*((@type@ *)op1)= nextafter@c@(in1, in2);
#if @SUPPORTED_BY_VML@
const int contig = IS_BINARY_CONT(@type@, @type@);
const int disjoint_or_same1 = DISJOINT_OR_SAME(args[0], args[2], dimensions[0], sizeof(@type@));
const int disjoint_or_same2 = DISJOINT_OR_SAME(args[1], args[2], dimensions[0], sizeof(@type@));
const int can_vectorize = contig && disjoint_or_same1 && disjoint_or_same2;

if(can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD) {
CHUNKED_VML_CALL3(v@s@NextAfter, dimensions[0], @type@, args[0], args[1], args[2]);
/* v@s@NextAfter(dimensions[0], (@type@*) args[0], (@type@*) args[1], (@type@*) args[2]); */
} else
#endif
{
BINARY_LOOP {
const @type@ in1 = *(@type@ *)ip1;
const @type@ in2 = *(@type@ *)ip2;
*((@type@ *)op1)= nextafter@c@(in1, in2);
}
}
}

Expand Down Expand Up @@ -2009,39 +2050,65 @@ mkl_umath_@TYPE@_@kind@(char **args, const npy_intp *dimensions, const npy_intp

/**begin repeat1
* #kind = fmax, fmin#
* #VML = Fmax, Fmin#
* #OP = >=, <=#
**/
void
mkl_umath_@TYPE@_@kind@(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
/* */
if (IS_BINARY_REDUCE) {
BINARY_REDUCE_LOOP(@type@) {
const @type@ in2 = *(@type@ *)ip2;
/* Order of operations important for MSVC 2015 */
io1 = (io1 @OP@ in2 || isnan(in2)) ? io1 : in2;
#if @SUPPORTED_BY_VML@
const int contig = IS_BINARY_CONT(@type@, @type@);
const int disjoint_or_same1 = DISJOINT_OR_SAME(args[0], args[2], dimensions[0], sizeof(@type@));
const int disjoint_or_same2 = DISJOINT_OR_SAME(args[1], args[2], dimensions[0], sizeof(@type@));
const int can_vectorize = contig && disjoint_or_same1 && disjoint_or_same2;

if(can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD) {
CHUNKED_VML_CALL3(v@s@@VML@, dimensions[0], @type@, args[0], args[1], args[2]);
/* v@s@@VML@(dimensions[0], (@type@*) args[0], (@type@*) args[1], (@type@*) args[2]); */
} else
#endif
{
if (IS_BINARY_REDUCE) {
BINARY_REDUCE_LOOP(@type@) {
const @type@ in2 = *(@type@ *)ip2;
/* Order of operations important for MSVC 2015 */
io1 = (io1 @OP@ in2 || isnan(in2)) ? io1 : in2;
}
*((@type@ *)iop1) = io1;
}
*((@type@ *)iop1) = io1;
}
else {
BINARY_LOOP {
const @type@ in1 = *(@type@ *)ip1;
const @type@ in2 = *(@type@ *)ip2;
/* Order of operations important for MSVC 2015 */
*((@type@ *)op1) = (in1 @OP@ in2 || isnan(in2)) ? in1 : in2;
else {
BINARY_LOOP {
const @type@ in1 = *(@type@ *)ip1;
const @type@ in2 = *(@type@ *)ip2;
/* Order of operations important for MSVC 2015 */
*((@type@ *)op1) = (in1 @OP@ in2 || isnan(in2)) ? in1 : in2;
}
}
feclearexcept(FE_ALL_EXCEPT); /* clear floatstatus */
}
feclearexcept(FE_ALL_EXCEPT); /* clear floatstatus */
}
/**end repeat1**/

void
mkl_umath_@TYPE@_remainder(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
BINARY_LOOP {
const @type@ in1 = *(@type@ *)ip1;
const @type@ in2 = *(@type@ *)ip2;
divmod@c@(in1, in2, (@type@ *)op1);
#if @SUPPORTED_BY_VML@
const int contig = IS_BINARY_CONT(@type@, @type@);
const int disjoint_or_same1 = DISJOINT_OR_SAME(args[0], args[2], dimensions[0], sizeof(@type@));
const int disjoint_or_same2 = DISJOINT_OR_SAME(args[1], args[2], dimensions[0], sizeof(@type@));
const int can_vectorize = contig && disjoint_or_same1 && disjoint_or_same2;

if(can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD) {
CHUNKED_VML_CALL3(v@s@Remainder, dimensions[0], @type@, args[0], args[1], args[2]);
/* v@s@Remainder(dimensions[0], (@type@*) args[0], (@type@*) args[1], (@type@*) args[2]); */
} else
#endif
{
BINARY_LOOP {
const @type@ in1 = *(@type@ *)ip1;
const @type@ in2 = *(@type@ *)ip2;
divmod@c@(in1, in2, (@type@ *)op1);
}
}
}

Expand All @@ -2059,9 +2126,11 @@ void
mkl_umath_@TYPE@_square(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(data))
{
#if @SUPPORTED_BY_VML@
if(IS_UNARY_CONT(@type@, @type@) &&
dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD &&
DISJOINT_OR_SAME(args[0], args[1], dimensions[0], sizeof(@type@)) ) {
const int contig = IS_UNARY_CONT(@type@, @type@);
const int disjoint_or_same = DISJOINT_OR_SAME(args[0], args[1], dimensions[0], sizeof(@type@));
const int can_vectorize = contig && disjoint_or_same;

if(can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD) {
CHUNKED_VML_CALL2(v@s@Sqr, dimensions[0], @type@, args[0], args[1]);
/* v@s@Sqr(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
} else
Expand All @@ -2078,9 +2147,11 @@ void
mkl_umath_@TYPE@_reciprocal(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(data))
{
#if @SUPPORTED_BY_VML@
if(IS_UNARY_CONT(@type@, @type@) &&
dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD &&
DISJOINT_OR_SAME(args[0], args[1], dimensions[0], sizeof(@type@)) ) {
const int contig = IS_UNARY_CONT(@type@, @type@);
const int disjoint_or_same = DISJOINT_OR_SAME(args[0], args[1], dimensions[0], sizeof(@type@));
const int can_vectorize = contig && disjoint_or_same;

if(can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD) {
CHUNKED_VML_CALL2(v@s@Inv, dimensions[0], @type@, args[0], args[1]);
/* v@s@Inv(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
} else
Expand Down Expand Up @@ -2114,9 +2185,11 @@ void
mkl_umath_@TYPE@_absolute(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
#if @SUPPORTED_BY_VML@
if(IS_UNARY_CONT(@type@, @type@) &&
dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD &&
DISJOINT_OR_SAME(args[0], args[1], dimensions[0], sizeof(@type@)) ) {
const int contig = IS_UNARY_CONT(@type@, @type@);
const int disjoint_or_same = DISJOINT_OR_SAME(args[0], args[1], dimensions[0], sizeof(@type@));
const int can_vectorize = contig && disjoint_or_same;

if(can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD) {
CHUNKED_VML_CALL2(v@s@Abs, dimensions[0], @type@, args[0], args[1]);
/* v@s@Abs(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
} else
Expand Down Expand Up @@ -2162,6 +2235,7 @@ mkl_umath_@TYPE@_sign(char **args, const npy_intp *dimensions, const npy_intp *s
}
}

/* TODO: USE MKL */
void
mkl_umath_@TYPE@_modf(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
Expand Down Expand Up @@ -2261,6 +2335,7 @@ mkl_umath_@TYPE@_ldexp_long(char **args, const npy_intp *dimensions, const npy_i
** COMPLEX LOOPS **
*****************************************************************************
*/
/* TODO: USE MKL for pow, exp, ln, log10, sqrt, trigonometric functions and hyperbolic functions */

#define CGE(xr,xi,yr,yi) ((xr > yr && !isnan(xi) && !isnan(yi)) \
|| (xr == yr && xi >= yi))
Expand Down Expand Up @@ -2363,6 +2438,7 @@ pairwise_sum_@TYPE@(@ftype@ *rr, @ftype@ * ri, char * a, npy_intp n,
}
}

/* TODO: USE MKL */
/**begin repeat1
* arithmetic
* #kind = add, subtract#
Expand Down Expand Up @@ -2396,6 +2472,7 @@ mkl_umath_@TYPE@_@kind@(char **args, const npy_intp *dimensions, const npy_intp
}
/**end repeat1**/

/* TODO: USE MKL */
void
mkl_umath_@TYPE@_multiply(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
Expand All @@ -2409,6 +2486,7 @@ mkl_umath_@TYPE@_multiply(char **args, const npy_intp *dimensions, const npy_int
}
}

/* TODO: USE MKL */
void
mkl_umath_@TYPE@_divide(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
Expand Down Expand Up @@ -2557,6 +2635,7 @@ mkl_umath_@TYPE@__ones_like(char **args, const npy_intp *dimensions, const npy_i
}
}

/* TODO: USE MKL */
void
mkl_umath_@TYPE@_conjugate(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func)) {
UNARY_LOOP {
Expand All @@ -2567,6 +2646,7 @@ mkl_umath_@TYPE@_conjugate(char **args, const npy_intp *dimensions, const npy_in
}
}

/* TODO: USE MKL */
void
mkl_umath_@TYPE@_absolute(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
Expand Down
Loading
Loading