From 4cf5bfc72a724fb653c48338b93e91dccea238af Mon Sep 17 00:00:00 2001
From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
Date: Wed, 18 Nov 2009 06:08:48 +0200
Subject: [PATCH 4/5] Support of overlapping src/dst for pixman_blt_neon

---
 pixman/pixman-arm-neon.c |   62 +++++++++++++++++++++++++++++++++++++--------
 1 files changed, 51 insertions(+), 11 deletions(-)

diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index 6808b36..7feee1d 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -168,26 +168,66 @@ pixman_blt_neon (uint32_t *src_bits,
                  int       width,
                  int       height)
 {
-    if (src_bpp != dst_bpp)
+    uint8_t *   src_bytes;
+    uint8_t *   dst_bytes;
+    int         bpp;
+
+    if (src_bpp != dst_bpp || src_bpp & 7)
 	return FALSE;
 
+    bpp = src_bpp >> 3;
+    width *= bpp;
+    src_stride *= 4;
+    dst_stride *= 4;
+    src_bytes = (uint8_t *)src_bits + src_y * src_stride + src_x * bpp;
+    dst_bytes = (uint8_t *)dst_bits + dst_y * dst_stride + dst_x * bpp;
+
+    if (src_bpp != 16 && src_bpp != 32)
+    {
+	pixman_blt_helper (src_bytes, dst_bytes, src_stride, dst_stride,
+	                   width, height);
+	return TRUE;
+    }
+
+    if (src_bytes < dst_bytes && 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;
+
+	if (src_bytes + width > dst_bytes)
+	{
+	    /* TODO: reverse scanline copy using NEON */
+	    while (--height >= 0)
+	    {
+		memmove (dst_bytes, src_bytes, width);
+		dst_bytes += dst_stride;
+		src_bytes += src_stride;
+	    }
+	    return TRUE;
+	}
+    }
+
     switch (src_bpp)
     {
     case 16:
 	pixman_composite_src_0565_0565_asm_neon (
-		width, height,
-		(uint16_t *)(((char *) dst_bits) +
-		dst_y * dst_stride * 4 + dst_x * 2), dst_stride * 2,
-		(uint16_t *)(((char *) src_bits) +
-		src_y * src_stride * 4 + src_x * 2), src_stride * 2);
+		width >> 1,
+		height,
+		(uint16_t *) dst_bytes,
+		dst_stride >> 1,
+		(uint16_t *) src_bytes,
+		src_stride >> 1);
 	return TRUE;
     case 32:
 	pixman_composite_src_8888_8888_asm_neon (
-		width, height,
-		(uint32_t *)(((char *) dst_bits) +
-		dst_y * dst_stride * 4 + dst_x * 4), dst_stride,
-		(uint32_t *)(((char *) src_bits) +
-		src_y * src_stride * 4 + src_x * 4), src_stride);
+		width >> 2,
+		height,
+		(uint32_t *) dst_bytes,
+		dst_stride >> 2,
+		(uint32_t *) src_bytes,
+		src_stride >> 2);
 	return TRUE;
     default:
 	return FALSE;
-- 
1.6.6.1