Skip to content

Resolve compilation errors with ICX compiler from 2026.0 release #155

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
antonwolfy merged 3 commits into from
Jan 30, 2026
Merged

Resolve compilation errors with ICX compiler from 2026.0 release #155

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b0a5f3c
Fix compilation errors with icx compiler from 2026.0 release
antonwolfy Jan 30, 2026
0d523c4
Add PR to the changelog
antonwolfy Jan 30, 2026
726c611
Rename mkl_op_type to mkl_ftype (aka MKL floating type)
antonwolfy Jan 30, 2026
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
2 changes: 2 additions & 0 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,8 @@ 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] - YYYY-MM-DD
### Fixed
* Build with ICX compiler from 2026.0 release [gh-155](https://github.com/IntelPython/mkl_umath/pull/155)

### Removed
* Dropped support for Python 3.9 [gh-125](https://github.com/IntelPython/mkl_umath/pull/125)
Expand Down
73 changes: 43 additions & 30 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 @@ -73,22 +73,23 @@

#define MKL_INT_MAX ((npy_intp) ((~((MKL_UINT) 0)) >> 1))

#define CHUNKED_VML_CALL2(vml_func, n, type, in1, op1) \
#define CHUNKED_VML_CALL2(vml_func, n, type, mkl_type, mkl_ftype, in1, op1) \
do { \
npy_intp _n_ = (n); \
const npy_intp _chunk_size = MKL_INT_MAX; \
type *in1p = (type *) (in1); \
type *op1p = (type *) (op1); \
while (_n_ > 0) { \
npy_intp _current_chunk = (_n_ > _chunk_size) ? _chunk_size : _n_; \
vml_func((MKL_INT) _current_chunk, in1p, op1p); \
vml_func((MKL_INT) _current_chunk, (mkl_type *)(void *) in1p, \
(mkl_ftype *)(void *) op1p); \
_n_ -= _current_chunk; \
in1p += _current_chunk; \
op1p += _current_chunk; \
} \
} while (0)

#define CHUNKED_VML_CALL3(vml_func, n, type, in1, in2, op1) \
#define CHUNKED_VML_CALL3(vml_func, n, type, mkl_type, in1, in2, op1) \
do { \
npy_intp _n_ = (n); \
const npy_intp _chunk_size = MKL_INT_MAX; \
Expand All @@ -97,7 +98,8 @@
type *op1p = (type *) (op1); \
while (_n_ > 0) { \
npy_intp _current_chunk = (_n_ > _chunk_size) ? _chunk_size : _n_; \
vml_func((MKL_INT) _current_chunk, in1p, in2p, op1p); \
vml_func((MKL_INT) _current_chunk, (mkl_type *)(void *) in1p, \
(mkl_type *)(void *) in2p, (mkl_type *)(void *) op1p); \
_n_ -= _current_chunk; \
in1p += _current_chunk; \
in2p += _current_chunk; \
Expand Down Expand Up @@ -323,7 +325,7 @@ mkl_umath_@TYPE@_@kind@(char **args, const npy_intp *dimensions, const npy_intp

if (IS_BINARY_CONT(@type@, @type@)) {
if (dimensions[0] > VML_ASM_THRESHOLD && disjoint_or_same1 && disjoint_or_same2) {
CHUNKED_VML_CALL3(v@s@@VML@, dimensions[0], @type@, args[0], args[1], args[2]);
CHUNKED_VML_CALL3(v@s@@VML@, dimensions[0], @type@, @type@, args[0], args[1], args[2]);
/* v@s@@VML@(dimensions[0], (@type@*) args[0], (@type@*) args[1], (@type@*) args[2]); */
}
else {
Expand Down Expand Up @@ -482,7 +484,7 @@ mkl_umath_@TYPE@_multiply(char **args, const npy_intp *dimensions, const npy_int

if (IS_BINARY_CONT(@type@, @type@)) {
if (dimensions[0] > VML_ASM_THRESHOLD && disjoint_or_same1 && disjoint_or_same2) {
CHUNKED_VML_CALL3(v@s@Mul, dimensions[0], @type@, args[0], args[1], args[2]);
CHUNKED_VML_CALL3(v@s@Mul, dimensions[0], @type@, @type@, args[0], args[1], args[2]);
/* v@s@Mul(dimensions[0], (@type@*) args[0], (@type@*) args[1], (@type@*) args[2]); */
}
else {
Expand Down Expand Up @@ -633,7 +635,7 @@ mkl_umath_@TYPE@_divide(char **args, const npy_intp *dimensions, const npy_intp

if (IS_BINARY_CONT(@type@, @type@)) {
if (dimensions[0] > VML_D_THRESHOLD && disjoint_or_same1 && disjoint_or_same2) {
CHUNKED_VML_CALL3(v@s@Div, dimensions[0], @type@, args[0], args[1], args[2]);
CHUNKED_VML_CALL3(v@s@Div, dimensions[0], @type@, @type@, args[0], args[1], args[2]);
/* v@s@Div(dimensions[0], (@type@*) args[0], (@type@*) args[1], (@type@*) args[2]); */
}
else {
Expand Down Expand Up @@ -784,7 +786,7 @@ mkl_umath_@TYPE@_@kind@(char **args, const npy_intp *dimensions, const npy_intp
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]);
CHUNKED_VML_CALL3(v@s@@VML@, dimensions[0], @type@, @type@, args[0], args[1], args[2]);
/* v@s@@VML@(dimensions[0], (@type@*) args[0], (@type@*) args[1], (@type@*) args[2]); */
}
else {
Expand Down Expand Up @@ -822,7 +824,7 @@ mkl_umath_@TYPE@_@kind@(char **args, const npy_intp *dimensions, const npy_intp
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]);
CHUNKED_VML_CALL3(v@s@@VML@, dimensions[0], @type@, @type@, args[0], args[1], args[2]);
/* v@s@@VML@(dimensions[0], (@type@*) args[0], (@type@*) args[1], (@type@*) args[2]); */
}
else {
Expand All @@ -849,7 +851,7 @@ mkl_umath_@TYPE@_@kind@(char **args, const npy_intp *dimensions, const npy_intp

if (can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD)
{
CHUNKED_VML_CALL2(v@s@@VML@, dimensions[0], @type@, args[0], args[1]);
CHUNKED_VML_CALL2(v@s@@VML@, dimensions[0], @type@, @type@, @type@, args[0], args[1]);
/* v@s@@VML@(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
}
else {
Expand Down Expand Up @@ -880,7 +882,7 @@ mkl_umath_@TYPE@_@kind@(char **args, const npy_intp *dimensions, const npy_intp
if (can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD)
{
ignore_fpstatus = 1;
CHUNKED_VML_CALL2(v@s@@VML@, dimensions[0], @type@, args[0], args[1]);
CHUNKED_VML_CALL2(v@s@@VML@, dimensions[0], @type@, @type@, @type@, args[0], args[1]);
/* v@s@Exp(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
}
else {
Expand Down Expand Up @@ -909,7 +911,7 @@ mkl_umath_@TYPE@_absolute(char **args, const npy_intp *dimensions, const npy_int
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]);
CHUNKED_VML_CALL2(v@s@Abs, dimensions[0], @type@, @type@, @type@, args[0], args[1]);
/* v@s@Abs(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
}
else {
Expand All @@ -931,7 +933,7 @@ mkl_umath_@TYPE@_reciprocal(char **args, const npy_intp *dimensions, const npy_i
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]);
CHUNKED_VML_CALL2(v@s@Inv, dimensions[0], @type@, @type@, @type@, args[0], args[1]);
/* v@s@Inv(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
}
else {
Expand All @@ -950,7 +952,7 @@ mkl_umath_@TYPE@_square(char **args, const npy_intp *dimensions, const npy_intp
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]);
CHUNKED_VML_CALL2(v@s@Sqr, dimensions[0], @type@, @type@, @type@, args[0], args[1]);
/* v@s@Sqr(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
}
else {
Expand Down Expand Up @@ -1187,7 +1189,9 @@ mkl_umath_@TYPE@_ldexp_long(char **args, const npy_intp *dimensions, const npy_i
* complex types
* #TYPE = CFLOAT, CDOUBLE#
* #ftype = npy_float, npy_double#
* #type = npy_cfloat, npy_cdouble#
* #type = npy_cfloat, npy_cdouble#
* #mkl_type = MKL_Complex8, MKL_Complex16#
* #mkl_ftype = float, double#
* #c = f, #
* #C = F, #
* #s = c, z#
Expand Down Expand Up @@ -1281,14 +1285,14 @@ pairwise_sum_@TYPE@(@ftype@ *rr, @ftype@ * ri, char * a, npy_intp n, npy_intp st
void
mkl_umath_@TYPE@_@kind@(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
const int contig = IS_BINARY_CONT(@type@, @type@);
const int contig = IS_BINARY_CONT(@mkl_type@, @mkl_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_ASM_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]); */
CHUNKED_VML_CALL3(v@s@@VML@, dimensions[0], @type@, @mkl_type@, args[0], args[1], args[2]);
/* v@s@@VML@(dimensions[0], (@mkl_type@*) args[0], (@mkl_type@*) args[1], (@mkl_type@*) args[2]); */
}
else {
if (IS_BINARY_REDUCE && @PW@) {
Expand Down Expand Up @@ -1319,14 +1323,14 @@ mkl_umath_@TYPE@_@kind@(char **args, const npy_intp *dimensions, const npy_intp
void
mkl_umath_@TYPE@_multiply(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
const int contig = IS_BINARY_CONT(@type@, @type@);
const int contig = IS_BINARY_CONT(@mkl_type@, @mkl_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_ASM_THRESHOLD) {
CHUNKED_VML_CALL3(v@s@Mul, dimensions[0], @type@, args[0], args[1], args[2]);
/* v@s@Mul(dimensions[0], (@type@*) args[0], (@type@*) args[1], (@type@*) args[2]); */
CHUNKED_VML_CALL3(v@s@Mul, dimensions[0], @type@, @mkl_type@, args[0], args[1], args[2]);
/* v@s@Mul(dimensions[0], (@mkl_type@*) args[0], (@mkl_type@*) args[1], (@mkl_type@*) args[2]); */
}
else {
BINARY_LOOP {
Expand All @@ -1343,14 +1347,14 @@ mkl_umath_@TYPE@_multiply(char **args, const npy_intp *dimensions, const npy_int
void
mkl_umath_@TYPE@_divide(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
const int contig = IS_BINARY_CONT(@type@, @type@);
const int contig = IS_BINARY_CONT(@mkl_type@, @mkl_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_D_THRESHOLD) {
CHUNKED_VML_CALL3(v@s@Div, dimensions[0], @type@, args[0], args[1], args[2]);
/* v@s@Div(dimensions[0], (@type@*) args[0], (@type@*) args[1], (@type@*) args[2]); */
CHUNKED_VML_CALL3(v@s@Div, dimensions[0], @type@, @mkl_type@, args[0], args[1], args[2]);
/* v@s@Div(dimensions[0], (@mkl_type@*) args[0], (@mkl_type@*) args[1], (@mkl_type@*) args[2]); */
}
else {
BINARY_LOOP {
Expand Down Expand Up @@ -1493,13 +1497,13 @@ mkl_umath_@TYPE@_reciprocal(char **args, const npy_intp *dimensions, const npy_i

void
mkl_umath_@TYPE@_conjugate(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func)) {
const int contig = IS_UNARY_CONT(@type@, @type@);
const int contig = IS_UNARY_CONT(@mkl_type@, @mkl_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@Conj, dimensions[0], @type@, args[0], args[1]);
/* v@s@Conj(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
CHUNKED_VML_CALL2(v@s@Conj, dimensions[0], @type@, @mkl_type@, @mkl_type@, args[0], args[1]);
/* v@s@Conj(dimensions[0], (@mkl_type@*) args[0], (@mkl_type@*) args[1]); */
}
else {
UNARY_LOOP {
Expand All @@ -1511,18 +1515,27 @@ mkl_umath_@TYPE@_conjugate(char **args, const npy_intp *dimensions, const npy_in
}
}

/**begin repeat
* complex types
* #TYPE = CFLOAT, CDOUBLE#
* #ftype = npy_float, npy_double#
* #type = npy_cfloat, npy_cdouble#
* #mkl_ftype = float, double#
* #mkl_type = MKL_Complex8, MKL_Complex16#

*/
void
mkl_umath_@TYPE@_absolute(char **args, const npy_intp *dimensions, const npy_intp *steps, void *NPY_UNUSED(func))
{
const int contig = IS_UNARY_CONT(@type@, @ftype@);
const int contig = IS_UNARY_CONT(@mkl_type@, @mkl_ftype@);
const int disjoint_or_same = DISJOINT_OR_SAME_TWO_DTYPES(args[0], args[1], dimensions[0], sizeof(@type@), sizeof(@ftype@));
const int can_vectorize = contig && disjoint_or_same;
int ignore_fpstatus = 0;

if (can_vectorize && dimensions[0] > VML_TRANSCEDENTAL_THRESHOLD) {
ignore_fpstatus = 1;
CHUNKED_VML_CALL2(v@s@Abs, dimensions[0], @type@, args[0], args[1]);
/* v@s@Abs(dimensions[0], (@type@*) args[0], (@type@*) args[1]); */
CHUNKED_VML_CALL2(v@s@Abs, dimensions[0], @type@, @mkl_type@, @mkl_ftype@, args[0], args[1]);
/* v@s@Abs(dimensions[0], (@mkl_type@*) args[0], (@mkl_ftype@*) args[1]); */
}
else {
UNARY_LOOP {
Expand Down
Loading