Tweak pdf14_preserve_backdrop (for speed).

First the current code clears the backdrop, then we copy into it.

In many cases we overwrite exactly the area we just cleared. Spot
this, and avoid it.

We *could* optimise this further in cases where we aren't overwriting
exactly the same region. Wait to see if this is justified.

1 files changed, 14 insertions, 1 deletions

diff --git a/base/gxblend1.c b/base/gxblend1.c
index e30d1fdcc..23b9ae638 100644
--- a/base/gxblend1.c
+++ b/base/gxblend1.c
@@ -123,6 +123,11 @@ copy_plane_part(byte *des_ptr, int des_rowstride, byte *src_ptr, int src_rowstri
 {
     int y;
 
+    if (width == des_rowstride && width == src_rowstride) {
+        width *= height;
+        height = 1;
+    }
+
     for (y = 0; y < height; ++y) {
         memcpy(des_ptr, src_ptr, width);
         des_ptr += des_rowstride;
@@ -267,7 +272,15 @@ pdf14_preserve_backdrop(pdf14_buf *buf, pdf14_buf *tos, bool knockout_buff)
            has a region outside the existing tos group.  Need to check if this
            is getting clipped in which case we need to fix the allocation of
            the buffer to be smaller */
-        memset(buf_plane, 0, n_planes * buf->planestride);
+        if (x0 > buf->rect.p.x || x1 < buf->rect.q.x ||
+            y0 > buf->rect.p.y || y1 < buf->rect.q.y) {
+            /* FIXME: There is potential for more optimisation here,
+             * but I don't know how often we hit this case. */
+            memset(buf_plane, 0, n_planes * buf->planestride);
+        } else if (n_planes > tos->n_chan) {
+            /* We *could* get away with not blanking any tag plane too... */
+            memset(buf->data + tos->n_chan * buf->planestride, 0, (n_planes - tos->n_chan) * buf->planestride);
+        }
         buf_plane += x0 - buf->rect.p.x +
                     (y0 - buf->rect.p.y) * buf->rowstride;
         tos_plane += x0 - tos->rect.p.x +