diff --git a/decoder/ihevcd_iquant_itrans_recon_ctb.c b/decoder/ihevcd_iquant_itrans_recon_ctb.c
index a558644..88ace44 100644
--- a/decoder/ihevcd_iquant_itrans_recon_ctb.c
+++ b/decoder/ihevcd_iquant_itrans_recon_ctb.c
@@ -567,9 +567,11 @@
     /* Intra 32x32 Y                                                         */
     /* Inter 32x32 Y                                                         */
     /*************************************************************************/
-    WORD32 scaling_mat_offset[] =
+    /* Only first 20 entries are used. Array is extended to avoid out of bound
+       reads. Skip CUs (64x64) read this table, but don't really use the value */
+    static const WORD32 scaling_mat_offset[] =
       { 0, 16, 32, 48, 64, 80, 96, 160, 224, 288, 352, 416, 480, 736, 992,
-        1248, 1504, 1760, 2016, 3040 };
+        1248, 1504, 1760, 2016, 3040, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
 
     PROFILE_DISABLE_IQ_IT_RECON_INTRA_PRED();
 
diff --git a/decoder/ihevcd_sao.c b/decoder/ihevcd_sao.c
index 2702317..dc852c6 100644
--- a/decoder/ihevcd_sao.c
+++ b/decoder/ihevcd_sao.c
@@ -568,10 +568,11 @@
     UWORD8  *pu1_sao_src_top_left_luma_bot_left;
     UWORD8 *au1_sao_src_top_left_chroma_bot_left;
     UWORD8 *pu1_sao_src_top_left_chroma_bot_left;
-
-    WORD8 ai1_offset_y[5];
-    WORD8 ai1_offset_cb[5];
-    WORD8 ai1_offset_cr[5];
+    /* Only 5 values are used, but arrays are large
+     enough so that SIMD functions can read 64 bits at a time */
+    WORD8 ai1_offset_y[8];
+    WORD8 ai1_offset_cb[8];
+    WORD8 ai1_offset_cr[8];
     WORD32  chroma_yuv420sp_vu = ps_sao_ctxt->is_chroma_yuv420sp_vu;
 
     PROFILE_DISABLE_SAO();