pixman: Branch 'master' - 2 commits

[Siarhei Siamashka]

 pixman/pixman-combine32.c |   58 ++++++++++++++++++++++++++++++++++++++++------
 test/utils-prng.c         |   12 +++++++++
 2 files changed, 63 insertions(+), 7 deletions(-)

New commits:
commit fdab3c1b6cd9c5e197ec3f6bc0a03da32880e317
Author: Siarhei Siamashka <siarhei.siamashka at gmail.com>
Date:   Sat Dec 8 15:16:51 2012 +0200

    test: Workaround unaligned MOVDQA bug (http://gcc.gnu.org/PR55614)
    
    Just use SSE2 intrinsics to do unaligned memory accesses as
    a workaround for this gcc bug related to vector extensions.

diff --git a/test/utils-prng.c b/test/utils-prng.c
index 7c2dd6a..967b898 100644
--- a/test/utils-prng.c
+++ b/test/utils-prng.c
@@ -27,6 +27,10 @@
 #include "utils.h"
 #include "utils-prng.h"
 
+#if defined(GCC_VECTOR_EXTENSIONS_SUPPORTED) && defined(__SSE2__)
+#include <xmmintrin.h>
+#endif
+
 void smallprng_srand_r (smallprng_t *x, uint32_t seed)
 {
     uint32_t i;
@@ -77,6 +81,14 @@ store_rand_128_data (void *addr, prng_rand_128_data_t *d, int aligned)
         *(uint8x16 *)addr = d->vb;
         return;
     }
+    else
+    {
+#ifdef __SSE2__
+        /* workaround for http://gcc.gnu.org/PR55614 */
+        _mm_storeu_si128 (addr, _mm_loadu_si128 ((__m128i *)d));
+        return;
+#endif
+    }
 #endif
     /* we could try something better for unaligned writes (packed attribute),
      * but GCC is not very reliable: http://gcc.gnu.org/PR55454 */
commit 2bc59006d7fe91abf68a2061ad86c06e1b2964ab
Author: Siarhei Siamashka <siarhei.siamashka at gmail.com>
Date:   Fri Nov 30 12:00:47 2012 +0200

    Improve performance of combine_over_u
    
    The generic C over_u combiner can be a lot faster with the
    addition of special shortcuts for 0xFF and 0x00 alpha/mask
    values. This is already implemented in C and SSE2 fast paths.
    
    Profiling the run of cairo-perf-trace benchmarks with PIXMAN_DISABLE
    environment variable set to "fast mmx sse2" on Intel Core i7:
    
    === before ===
    
    37.32%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] combine_over_u
    21.37%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] bits_image_fetch_bilinear_no_repeat_8888
    13.51%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] bits_image_fetch_bilinear_affine_none_a8r8g8b8
     2.96%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] radial_compute_color
     2.74%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] fetch_scanline_a8
     2.71%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] fetch_scanline_x8r8g8b8
     2.17%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] _pixman_gradient_walker_pixel
     1.86%  cairo-perf-trac  libcairo.so.2.11200.0 [.] _cairo_tor_scan_converter_generate
     1.57%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] bits_image_fetch_bilinear_affine_pad_a8r8g8b8
     0.97%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] combine_in_reverse_u
     0.96%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] combine_over_ca
    
    === after ===
    
    28.79%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] bits_image_fetch_bilinear_no_repeat_8888
    18.44%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] bits_image_fetch_bilinear_affine_none_a8r8g8b8
    15.54%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] combine_over_u
     3.94%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] radial_compute_color
     3.69%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] fetch_scanline_a8
     3.69%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] fetch_scanline_x8r8g8b8
     2.94%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] _pixman_gradient_walker_pixel
     2.52%  cairo-perf-trac  libcairo.so.2.11200.0 [.] _cairo_tor_scan_converter_generate
     2.08%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] bits_image_fetch_bilinear_affine_pad_a8r8g8b8
     1.31%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] combine_in_reverse_u
     1.29%  cairo-perf-trac  libpixman-1.so.0.29.1 [.] combine_over_ca

diff --git a/pixman/pixman-combine32.c b/pixman/pixman-combine32.c
index 54cc877..3ac7576 100644
--- a/pixman/pixman-combine32.c
+++ b/pixman/pixman-combine32.c
@@ -196,14 +196,58 @@ combine_over_u (pixman_implementation_t *imp,
 {
     int i;
 
-    for (i = 0; i < width; ++i)
+    if (!mask)
     {
-	uint32_t s = combine_mask (src, mask, i);
-	uint32_t d = *(dest + i);
-	uint32_t ia = ALPHA_8 (~s);
-
-	UN8x4_MUL_UN8_ADD_UN8x4 (d, ia, s);
-	*(dest + i) = d;
+	for (i = 0; i < width; ++i)
+	{
+	    uint32_t s = *(src + i);
+	    uint32_t a = ALPHA_8 (s);
+	    if (a == 0xFF)
+	    {
+		*(dest + i) = s;
+	    }
+	    else if (s)
+	    {
+		uint32_t d = *(dest + i);
+		uint32_t ia = a ^ 0xFF;
+		UN8x4_MUL_UN8_ADD_UN8x4 (d, ia, s);
+		*(dest + i) = d;
+	    }
+	}
+    }
+    else
+    {
+	for (i = 0; i < width; ++i)
+	{
+	    uint32_t m = ALPHA_8 (*(mask + i));
+	    if (m == 0xFF)
+	    {
+		uint32_t s = *(src + i);
+		uint32_t a = ALPHA_8 (s);
+		if (a == 0xFF)
+		{
+		    *(dest + i) = s;
+		}
+		else if (s)
+		{
+		    uint32_t d = *(dest + i);
+		    uint32_t ia = a ^ 0xFF;
+		    UN8x4_MUL_UN8_ADD_UN8x4 (d, ia, s);
+		    *(dest + i) = d;
+		}
+	    }
+	    else if (m)
+	    {
+		uint32_t s = *(src + i);
+		if (s)
+		{
+		    uint32_t d = *(dest + i);
+		    UN8x4_MUL_UN8 (s, m);
+		    UN8x4_MUL_UN8_ADD_UN8x4 (d, ALPHA_8 (~s), s);
+		    *(dest + i) = d;
+		}
+	    }
+	}
     }
 }