diff --git a/av1/encoder/tune_vmaf.c b/av1/encoder/tune_vmaf.c
index 847571f..fdb7c77 100644
--- a/av1/encoder/tune_vmaf.c
+++ b/av1/encoder/tune_vmaf.c
@@ -247,6 +247,8 @@
 
 // 8-tap Gaussian convolution filter with sigma = 1.0, sums to 128,
 // all co-efficients must be even.
+// The array is of size 9 to allow passing gauss_filter + 1 to
+// _mm_loadu_si128() in prepare_coeffs_6t().
 DECLARE_ALIGNED(16, static const int16_t, gauss_filter[9]) = { 0,  8, 30, 52,
                                                                30, 8, 0,  0 };
 static AOM_INLINE void gaussian_blur(const int bit_depth,