From 94b115ebdc0bc437e8c26c481d564c691f51c30e Mon Sep 17 00:00:00 2001 From: Martin Kroeker Date: Wed, 14 Oct 2026 14:19:28 +0200 Subject: [PATCH] Use explicit zeroing of y for beta=0 in the non-unit stride path as well --- kernel/arm64/sbgemv_n_neon.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff --git a/kernel/arm64/sbgemv_n_neon.c b/kernel/arm64/sbgemv_n_neon.c index ff730407fd..eb1063ba21 100644 --- a/kernel/arm64/sbgemv_n_neon.c +++ b/kernel/arm64/sbgemv_n_neon.c @@ -494,10 +494,17 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT alpha, bfloat16 *a, BLASLONG lda, } BLASLONG iy = 0; - for (i = 0; i < m; i++) { + if (beta == 0) { + for (i = 0; i < m; i++) { + y[iy] = 0.; + iy += incy; + } + } else { + for (i = 0; i < m; i++) { y[iy] *= beta; iy += incy; } + } for (j = 0; j < n; j++) { x0 = alpha * vcvtah_f32_bf16(*x_ptr);