Skip to content

Make the thunderx2 zdot kernel compatible with LLVM21 in Windows on Arm #5575

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

Open
martin-frbg wants to merge 3 commits into
base: develop
Choose a base branch
from
+30 −17
Open

Make the thunderx2 zdot kernel compatible with LLVM21 in Windows on Arm #5575

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
cfa28bc
Support compilation with LLVM for Windows on Arm
martin-frbg Dec 19, 2025
ac2c663
remove special handling of C/ZDOT for LLVM on WoA
martin-frbg Dec 19, 2025
d39b777
Make .align conditional on not being on WoA and strip CRLF endings
martin-frbg Dec 24, 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
10 changes: 0 additions & 10 deletions kernel/arm64/KERNEL.NEOVERSEN1
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,18 +102,8 @@ ZNRM2KERNEL = znrm2.S

DDOTKERNEL = dot.c
SDOTKERNEL = dot.c
ifeq ($(OSNAME), WINNT)
ifeq ($(C_COMPILER), CLANG)
CDOTKERNEL = zdot.S
ZDOTKERNEL = zdot.S
else
CDOTKERNEL = zdot_thunderx2t99.c
ZDOTKERNEL = zdot_thunderx2t99.c
endif
else
CDOTKERNEL = zdot_thunderx2t99.c
ZDOTKERNEL = zdot_thunderx2t99.c
endif
DSDOTKERNEL = dot.S

DGEMM_BETA = dgemm_beta.S
Expand Down
37 changes: 30 additions & 7 deletions kernel/arm64/zdot_thunderx2t99.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#endif

#include "common.h"

#ifdef _MSC_VER
#include <complex.h>
#endif
#include <arm_neon.h>

#define N "x0" /* vector length */
Expand Down Expand Up @@ -197,14 +199,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#if defined(SMP)
extern int blas_level1_thread_with_return_value(int mode, BLASLONG m, BLASLONG n,
BLASLONG k, void *alpha, void *a, BLASLONG lda, void *b, BLASLONG ldb,
void *c, BLASLONG ldc, int (*function)(), int nthreads);
void *c, BLASLONG ldc, int (*function)(void), int nthreads);
#endif

static void zdot_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, OPENBLAS_COMPLEX_FLOAT *result)
{
FLOAT dotr = 0.0, doti = 0.0;
FLOAT dotr = 0.0, doti = 0.0;

#ifdef _MSC_VER
CREAL(*result) = 0.0;
CIMAG(*result) = 0.0;
#else
OPENBLAS_COMPLEX_FLOAT cf = OPENBLAS_MAKE_COMPLEX_FLOAT(0.0, 0.0);
*result = cf;
*result = cf;
#endif

if ( n < 0 ) return;

Expand Down Expand Up @@ -235,8 +243,9 @@ static void zdot_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLON
" asr "J", "N", #"N_DIV_SHIFT" \n"
" cmp "J", xzr \n"
" beq 3f //dot_kernel_F1 \n"

#ifndef _MSC_VER
" .align 5 \n"
#endif
"2: //dot_kernel_F: \n"
" "KERNEL_F" \n"
" subs "J", "J", #1 \n"
Expand Down Expand Up @@ -297,10 +306,14 @@ static void zdot_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLON
"v23", "v24", "v25", "v26", "v27", "v28", "v29",
"v30", "v31"
);

#ifdef _MSC_VER
CREAL(*result) = dotr;
CIMAG(*result) = doti;
#else
cf=OPENBLAS_MAKE_COMPLEX_FLOAT(dotr, doti);
*result = cf;
return;
#endif
return;
}

#if defined(SMP)
Expand All @@ -320,7 +333,13 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
int nthreads;
FLOAT dummy_alpha;
#endif
#ifdef _MSC_VER
OPENBLAS_COMPLEX_FLOAT zdot;
CREAL(zdot) = 0.0;
CIMAG(zdot) = 0.0;
#else
OPENBLAS_COMPLEX_FLOAT zdot = OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0);
#endif

#if defined(SMP)
if (inc_x == 0 || inc_y == 0 || n <= 10000)
Expand All @@ -347,7 +366,11 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA

ptr = (OPENBLAS_COMPLEX_FLOAT *)result;
for (i = 0; i < nthreads; i++) {
#ifdef _MSC_VER
CREAL(zdot)+= CREAL(*ptr);CIMAG(zdot)+=CIMAG(*ptr);
#else
zdot = OPENBLAS_MAKE_COMPLEX_FLOAT (CREAL(zdot) + CREAL(*ptr), CIMAG(zdot) + CIMAG(*ptr));
#endif
ptr = (void *)(((char *)ptr) + sizeof(double) * 2);
}
}
Expand Down