Generic C implementation of pixman_blt with overlapping support

This was ported from meta-oe's patch [1]:
Uses memcpy/memmove functions to copy pixels, can handle the
case when both source and destination areas are in the same
image (this is useful for scrolling).

It is assumed that copying direction is only important when
using the same image for both source and destination (and
src_stride == dst_stride). Copying direction is undefined
for the images with different source and destination stride
which happen to be in the overlapped areas (but this is an
unrealistic case anyway).

[1] http://cgit.openembedded.org/meta-openembedded/tree/meta-oe/recipes-graphics/xorg-lib/pixman-0.26.2/0008-Generic-C-implementation-of-pixman_blt-with-overlapp.patch

Upstream-Status: Pending
Signed-off-by: Andreas Müller <schnitzeltony@googlemail.com>
Signed-off-by: Constantin Musca <constantinx.musca@intel.com>

Index: pixman-0.28.0/pixman/pixman-general.c
===================================================================
--- pixman-0.28.0.orig/pixman/pixman-general.c
+++ pixman-0.28.0/pixman/pixman-general.c
@@ -214,6 +214,41 @@ static const pixman_fast_path_t general_
     { PIXMAN_OP_NONE }
 };
 
+static pixman_bool_t
+general_blt (pixman_implementation_t *imp,
+             uint32_t *               src_bits,
+             uint32_t *               dst_bits,
+             int                      src_stride,
+             int                      dst_stride,
+             int                      src_bpp,
+             int                      dst_bpp,
+             int                      src_x,
+             int                      src_y,
+             int                      dest_x,
+             int                      dest_y,
+             int                      width,
+             int                      height)
+{
+    uint8_t *dst_bytes = (uint8_t *)dst_bits;
+    uint8_t *src_bytes = (uint8_t *)src_bits;
+    int bpp;
+
+    if (src_bpp != dst_bpp || src_bpp & 7)
+        return FALSE;
+
+    bpp = src_bpp >> 3;
+    width *= bpp;
+    src_stride *= 4;
+    dst_stride *= 4;
+    pixman_blt_helper (src_bytes + src_y * src_stride + src_x * bpp,
+                       dst_bytes + dest_y * dst_stride + dest_x * bpp,
+                       src_stride,
+                       dst_stride,
+                       width,
+                       height);
+    return TRUE;
+}
+
 pixman_implementation_t *
 _pixman_implementation_create_general (void)
 {
@@ -222,6 +257,7 @@ _pixman_implementation_create_general (v
     _pixman_setup_combiner_functions_32 (imp);
     _pixman_setup_combiner_functions_float (imp);
 
+    imp->blt = general_blt;
     imp->src_iter_init = general_src_iter_init;
     imp->dest_iter_init = general_dest_iter_init;
 
Index: pixman-0.28.0/pixman/pixman-private.h
===================================================================
--- pixman-0.28.0.orig/pixman/pixman-private.h
+++ pixman-0.28.0/pixman/pixman-private.h
@@ -1109,6 +1109,45 @@ void pixman_timer_register (pixman_timer
 
 #endif /* PIXMAN_TIMERS */
 
+/* a helper function, can blit 8-bit images with src/dst overlapping support */
+static inline void
+pixman_blt_helper (uint8_t *src_bytes,
+                   uint8_t *dst_bytes,
+                   int      src_stride,
+                   int      dst_stride,
+                   int      width,
+                   int      height)
+{
+    /*
+     * The second part of this check is not strictly needed, but it prevents
+     * unnecessary upside-down processing of areas which belong to different
+     * images. Upside-down processing can be slower with fixed-distance-ahead
+     * prefetch and perceived as having more tearing.
+     */
+    if (src_bytes < dst_bytes + width &&
+        src_bytes + src_stride * height > dst_bytes) {
+        src_bytes += src_stride * height - src_stride;
+        dst_bytes += dst_stride * height - dst_stride;
+        dst_stride = -dst_stride;
+        src_stride = -src_stride;
+
+        /* Horizontal scrolling to the left needs memmove */
+        if (src_bytes + width > dst_bytes) {
+            while (--height >= 0) {
+                memmove (dst_bytes, src_bytes, width);
+                dst_bytes += dst_stride;
+                src_bytes += src_stride;
+            }
+            return;
+        }
+    }
+    while (--height >= 0) {
+        memcpy (dst_bytes, src_bytes, width);
+        dst_bytes += dst_stride;
+        src_bytes += src_stride;
+    }
+}
+
 #endif /* __ASSEMBLER__ */
 
 #endif /* PIXMAN_PRIVATE_H */