pixman: Branch 'master' - 2 commits

[Søren Sandmann Pedersen]

 pixman/pixman-sse2.c |   51 +++++++++++++++++++++++++++++++++------------------
 1 file changed, 33 insertions(+), 18 deletions(-)

New commits:
commit 3f5d6f90b753175a888f36a93d1e79fdc80d95de
Author: Søren Sandmann Pedersen <sandmann at daimi.au.dk>
Date:   Wed Sep 17 09:50:57 2008 -0400

    Don't include stdio.h

diff --git a/pixman/pixman-sse2.c b/pixman/pixman-sse2.c
index 0c671ed..0f36436 100644
--- a/pixman/pixman-sse2.c
+++ b/pixman/pixman-sse2.c
@@ -1,4 +1,3 @@
-#include <stdio.h>
 /*
  * Copyright © 2008 Rodrigo Kumpera
  * Copyright © 2008 André Tupinambá
commit eba402092082bf48072671e04e224589af872acd
Author: Søren Sandmann Pedersen <sandmann at daimi.au.dk>
Date:   Sun Sep 14 14:58:00 2008 -0400

    [sse2] Fix rounding bug in conversion from 565 to 8888
    
    When converting from 565 to 8888, replicate the topmost bits instead
    of appending zeros.

diff --git a/pixman/pixman-sse2.c b/pixman/pixman-sse2.c
index cc08189..0c671ed 100644
--- a/pixman/pixman-sse2.c
+++ b/pixman/pixman-sse2.c
@@ -1,3 +1,4 @@
+#include <stdio.h>
 /*
  * Copyright © 2008 Rodrigo Kumpera
  * Copyright © 2008 André Tupinambá
@@ -73,6 +74,9 @@ static __m128i MaskRed;
 static __m128i MaskGreen;
 static __m128i MaskBlue;
 
+static __m128i Mask565FixRB;
+static __m128i Mask565FixG;
+
 /* -------------------------------------------------------------------------------------------------
  * SSE2 Inlines
  */
@@ -89,26 +93,37 @@ unpack_128_2x128 (__m128i data, __m128i* dataLo, __m128i* dataHi)
     *dataHi = _mm_unpackhi_epi8 (data, _mm_setzero_si128 ());
 }
 
+static inline __m128i
+unpack565to8888 (__m128i lo)
+{
+    __m128i r, g, b, rb, t;
+    
+    r = _mm_and_si128 (_mm_slli_epi32 (lo, 8), MaskRed);
+    g = _mm_and_si128 (_mm_slli_epi32 (lo, 5), MaskGreen);
+    b = _mm_and_si128 (_mm_slli_epi32 (lo, 3), MaskBlue);
+
+    rb = _mm_or_si128 (r, b);
+    t  = _mm_and_si128 (rb, Mask565FixRB);
+    t  = _mm_srli_epi32 (t, 5);
+    rb = _mm_or_si128 (rb, t);
+
+    t  = _mm_and_si128 (g, Mask565FixG);
+    t  = _mm_srli_epi32 (t, 6);
+    g  = _mm_or_si128 (g, t);
+    
+    return _mm_or_si128 (rb, g);
+}
+
 static inline void
 unpack565_128_4x128 (__m128i data, __m128i* data0, __m128i* data1, __m128i* data2, __m128i* data3)
 {
     __m128i lo, hi;
-    __m128i r, g, b;
 
     lo = _mm_unpacklo_epi16 (data, _mm_setzero_si128 ());
     hi = _mm_unpackhi_epi16 (data, _mm_setzero_si128 ());
 
-    r = _mm_and_si128 (_mm_slli_epi32 (lo, 8), MaskRed);
-    g = _mm_and_si128 (_mm_slli_epi32 (lo, 5), MaskGreen);
-    b = _mm_and_si128 (_mm_slli_epi32 (lo, 3), MaskBlue);
-
-    lo = _mm_or_si128 (_mm_or_si128 (r, g), b);
-
-    r = _mm_and_si128 (_mm_slli_epi32 (hi, 8), MaskRed);
-    g = _mm_and_si128 (_mm_slli_epi32 (hi, 5), MaskGreen);
-    b = _mm_and_si128 (_mm_slli_epi32 (hi, 3), MaskBlue);
-
-    hi = _mm_or_si128 (_mm_or_si128 (r, g), b);
+    lo = unpack565to8888 (lo);
+    hi = unpack565to8888 (hi);
 
     unpack_128_2x128 (lo, data0, data1);
     unpack_128_2x128 (hi, data2, data3);
@@ -2297,7 +2312,8 @@ fbComposeSetupSSE2(void)
         MaskRed   = createMask_2x32_128 (0x00f80000, 0x00f80000);
         MaskGreen = createMask_2x32_128 (0x0000fc00, 0x0000fc00);
         MaskBlue  = createMask_2x32_128 (0x000000f8, 0x000000f8);
-
+	Mask565FixRB = createMask_2x32_128 (0x00e000e0, 0x00e000e0);
+	Mask565FixG = createMask_2x32_128  (0x0000c000, 0x0000c000);
         Mask0080 = createMask_16_128 (0x0080);
         Mask00ff = createMask_16_128 (0x00ff);
         Mask0101 = createMask_16_128 (0x0101);
@@ -2484,6 +2500,7 @@ fbCompositeSolid_nx0565sse2 (pixman_op_t op,
         while (w && (unsigned long)dst & 15)
         {
             d = *dst;
+
             *dst++ = pack565_32_16 (pack_1x64_32 (over_1x64 (_mm_movepi64_pi64 (xmmSrc),
                                                              _mm_movepi64_pi64 (xmmAlpha),
                                                              expand565_16_1x64 (d))));
@@ -2498,15 +2515,14 @@ fbCompositeSolid_nx0565sse2 (pixman_op_t op,
             /* fill cache line with next memory */
             cachePrefetchNext ((__m128i*)dst);
 
-            xmmDst = load128Aligned ((__m128i*)dst);
-
-            unpack565_128_4x128 (xmmDst, &xmmDst0, &xmmDst1, &xmmDst2, &xmmDst3);
-
+	    xmmDst = load128Aligned ((__m128i*)dst);
+	    
+	    unpack565_128_4x128 (xmmDst, &xmmDst0, &xmmDst1, &xmmDst2, &xmmDst3);
+	    
             over_2x128 (&xmmSrc, &xmmSrc, &xmmAlpha, &xmmAlpha, &xmmDst0, &xmmDst1);
             over_2x128 (&xmmSrc, &xmmSrc, &xmmAlpha, &xmmAlpha, &xmmDst2, &xmmDst3);
 
             xmmDst = pack565_4x128_128 (&xmmDst0, &xmmDst1, &xmmDst2, &xmmDst3);
-
             save128Aligned ((__m128i*)dst, xmmDst);
 
             dst += 8;