Skip to content

Fix complex scalar pointers in strided batched GEMM #6086

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
Arthur031221 wants to merge 2 commits into
base: develop
Choose a base branch
from
+68 −4
Open

Fix complex scalar pointers in strided batched GEMM #6086

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b974855
Fix complex scalars in strided batched GEMM
Arthur031221 Oct 1, 2026
e862d77
Resolve small-kernel addresses in strided batched GEMM
Arthur031221 Oct 3, 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
3 changes: 3 additions & 0 deletions CONTRIBUTORS.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -289,3 +289,6 @@ hheei <hheei@users.noreply.github.com>

* Hugo Meiland <hugo@meiland.nl>
* [2026-08-09] Add Cortex-A72 DGEMM 6x8 microkernel and blocking

* Arthur031221 <levi74108520963@gmail.com>
* [2026-10-02] Fix complex scaling in strided batched GEMM.
13 changes: 9 additions & 4 deletions interface/gemm_batch_strided.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -366,18 +366,18 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE transa, enum CBLAS_TRANS
#if !defined(COMPLEX)
if(beta == 0.0){
group_mode=mode | BLAS_SMALL_B0_OPT;
group_small_matrix_opt_routine=(void *)(gemm_small_kernel_b0[(group_transb<<2)|group_transa]);
group_small_matrix_opt_routine=SMALL_KERNEL_ADDR(gemm_small_kernel_b0, ((group_transb<<2)|group_transa));
}else{
group_mode=mode | BLAS_SMALL_OPT;
group_small_matrix_opt_routine=(void *)(gemm_small_kernel[(group_transb<<2)|group_transa]);
group_small_matrix_opt_routine=SMALL_KERNEL_ADDR(gemm_small_kernel, ((group_transb<<2)|group_transa));
}
#else
if(beta[0] == 0.0 && beta[1] == 0.0){
group_mode=mode | BLAS_SMALL_B0_OPT;
group_small_matrix_opt_routine=(void *)(zgemm_small_kernel_b0[(group_transb<<2)|group_transa]);
group_small_matrix_opt_routine=SMALL_KERNEL_ADDR(zgemm_small_kernel_b0, ((group_transb<<2)|group_transa));
}else{
group_mode=mode | BLAS_SMALL_OPT;
group_small_matrix_opt_routine=(void *)(zgemm_small_kernel[(group_transb<<2)|group_transa]);
group_small_matrix_opt_routine=SMALL_KERNEL_ADDR(zgemm_small_kernel, ((group_transb<<2)|group_transa));
}

#endif
Expand All @@ -396,8 +396,13 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE transa, enum CBLAS_TRANS
args_array[i].lda=group_lda;
args_array[i].ldb=group_ldb;
args_array[i].ldc=group_ldc;
#ifdef COMPLEX
args_array[i].alpha=alpha;
args_array[i].beta=beta;
#else
args_array[i].alpha=&alpha;
args_array[i].beta=&beta;
#endif

#if defined(CBLAS)
if (order == CblasColMajor) {
Expand Down
28 changes: 28 additions & 0 deletions utest/test_extensions/test_cgemm.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -271,4 +271,32 @@ CTEST(cgemm, transa_conjnotransb)

ASSERT_DBL_NEAR_TOL(0.0f, norm, SINGLE_EPS);
}

#ifndef NO_CBLAS
CTEST(cgemm, strided_batch_complex_scalars)
{
float a[] = {1.0f, 2.0f};
float b[] = {3.0f, 4.0f};
float c[] = {5.0f, 6.0f};
float alpha[] = {2.0f, 3.0f};
float beta[] = {4.0f, -1.0f};

cblas_cgemm_batch_strided(CblasColMajor, CblasNoTrans, CblasNoTrans,
1, 1, 1, alpha, a, 1, 2, b, 1, 2,
beta, c, 1, 2, 1);

ASSERT_DBL_NEAR_TOL(-14.0, c[0], SINGLE_EPS);
ASSERT_DBL_NEAR_TOL(24.0, c[1], SINGLE_EPS);

beta[0] = beta[1] = 0.0;
c[0] = 5.0;
c[1] = 6.0;
cblas_cgemm_batch_strided(CblasColMajor, CblasNoTrans, CblasNoTrans,
1, 1, 1, alpha, a, 1, 2, b, 1, 2,
beta, c, 1, 2, 1);

ASSERT_DBL_NEAR_TOL(-40.0, c[0], SINGLE_EPS);
ASSERT_DBL_NEAR_TOL(5.0, c[1], SINGLE_EPS);
}
#endif
#endif
28 changes: 28 additions & 0 deletions utest/test_extensions/test_zgemm.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -271,4 +271,32 @@ CTEST(zgemm, transa_conjnotransb)

ASSERT_DBL_NEAR_TOL(0.0, norm, DOUBLE_EPS);
}

#ifndef NO_CBLAS
CTEST(zgemm, strided_batch_complex_scalars)
{
double a[] = {1.0, 2.0};
double b[] = {3.0, 4.0};
double c[] = {5.0, 6.0};
double alpha[] = {2.0, 3.0};
double beta[] = {4.0, -1.0};

cblas_zgemm_batch_strided(CblasColMajor, CblasNoTrans, CblasNoTrans,
1, 1, 1, alpha, a, 1, 2, b, 1, 2,
beta, c, 1, 2, 1);

ASSERT_DBL_NEAR_TOL(-14.0, c[0], DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(24.0, c[1], DOUBLE_EPS);

beta[0] = beta[1] = 0.0;
c[0] = 5.0;
c[1] = 6.0;
cblas_zgemm_batch_strided(CblasColMajor, CblasNoTrans, CblasNoTrans,
1, 1, 1, alpha, a, 1, 2, b, 1, 2,
beta, c, 1, 2, 1);

ASSERT_DBL_NEAR_TOL(-40.0, c[0], DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(5.0, c[1], DOUBLE_EPS);
}
#endif
#endif
Loading