diff --git a/tests/unit/test_fmha_hd64_smem_p.cu b/tests/unit/test_fmha_hd64_smem_p.cu
index aca89f4b..c19882bc 100644
--- a/tests/unit/test_fmha_hd64_smem_p.cu
+++ b/tests/unit/test_fmha_hd64_smem_p.cu
@@ -247,8 +247,7 @@ int main() {
     float cos_sim=0,na=0,nb=0;
     for (int d=0;d<HD;d++) { float a=bf16_to_f32_host(h_o[d]),b=h_o_scalar[d]; if(fabsf(b)>1e-4f) { cos_sim+=a*b; na+=a*a; nb+=b*b; } }
     cos_sim /= (sqrtf(na)*sqrtf(nb)+1e-10f);
-    printf("Filtered cosine (|ref|>1e-4): %.8f
-", cos_sim);
+    printf("Filtered cosine: %.8f\n", cos_sim);
     printf("Max rel err: %.8f | cosine: %.8f\n", rel_err, cos_sim);
     printf("Test %s\n", cos_sim > 0.999f ? "PASSED" : "FAILED");