diff --git a/libswscale/ops_memcpy.c b/libswscale/ops_memcpy.c
index 52ba64b0c3..735006ab18 100644
--- a/libswscale/ops_memcpy.c
+++ b/libswscale/ops_memcpy.c
@@ -49,9 +49,11 @@ static void process(const SwsOpExec *exec, const void *priv,
         uint8_t *out = exec->out[i];
         const int idx = p->index[i];
         const int bytes = x_end * exec->block_size_out[i];
-        const int use_loop = exec->out_stride[i] > bytes + SWS_MAX_PADDING;
+        const int use_loop = exec->out_stride[i] < 0 ||
+                             exec->out_stride[i] > bytes + SWS_MAX_PADDING;
+        const size_t span = (size_t) (lines - 1) * exec->out_stride[i] + bytes;
         if (idx < 0 && !use_loop) {
-            memset(out, p->clear_value[i], exec->out_stride[i] * lines);
+            memset(out, p->clear_value[i], span);
         } else if (idx < 0) {
             for (int y = y_start; y < y_end; y++) {
                 memset(out, p->clear_value[i], bytes);
@@ -61,7 +63,7 @@ static void process(const SwsOpExec *exec, const void *priv,
             av_assert1(exec->out_stride[i] == exec->in_stride[idx]);
             continue; /* plane was already ref'd */
         } else if (exec->out_stride[i] == exec->in_stride[idx] && !use_loop) {
-            memcpy(out, exec->in[idx], exec->out_stride[i] * lines);
+            memcpy(out, exec->in[idx], span);
         } else {
             const uint8_t *in = exec->in[idx];
             for (int y = y_start; y < y_end; y++) {