mmx: Add nearest over_8888_n_8888
authorMatt Turner <mattst88@gmail.com>
Wed, 2 Jan 2013 05:18:09 +0000 (21:18 -0800)
committerMatt Turner <mattst88@gmail.com>
Fri, 5 Sep 2014 07:22:04 +0000 (00:22 -0700)
lowlevel-blt-bench -n, over_8888_n_8888, 15 iterations on Loongson 2f:

           Before          After
          Mean StdDev     Mean StdDev   Change
    L1     9.7   0.01     19.2   0.02   +98.2%
    L2     9.6   0.11     19.2   0.16   +99.5%
    M      7.3   0.02     12.5   0.01   +72.0%
    HT     6.6   0.01     13.4   0.02  +103.2%
    VT     6.4   0.01     12.6   0.03   +96.1%
    R      6.3   0.01     11.2   0.01   +76.5%
    RT     4.4   0.01      8.1   0.03   +82.6%

pixman/pixman-mmx.c

index f9a92ce09555f5f79650b5d23cf7f9ee99799731..63f4cdf82f2bf12400d0fa311c79142c9b0f51ab 100644 (file)
@@ -3555,6 +3555,59 @@ mmx_composite_over_reverse_n_8888 (pixman_implementation_t *imp,
     _mm_empty ();
 }
 
+static force_inline void
+scaled_nearest_scanline_mmx_8888_n_8888_OVER (const uint32_t * mask,
+                                             uint32_t *       dst,
+                                             const uint32_t * src,
+                                             int32_t          w,
+                                             pixman_fixed_t   vx,
+                                             pixman_fixed_t   unit_x,
+                                             pixman_fixed_t   src_width_fixed,
+                                             pixman_bool_t    zero_src)
+{
+    __m64 mm_mask;
+
+    if (zero_src || (*mask >> 24) == 0)
+       return;
+
+    mm_mask = expand_alpha (load8888 (mask));
+
+    while (w)
+    {
+       uint32_t s = *(src + pixman_fixed_to_int (vx));
+       vx += unit_x;
+       while (vx >= 0)
+           vx -= src_width_fixed;
+
+       if (s)
+       {
+           __m64 ms = load8888 (&s);
+           __m64 alpha = expand_alpha (ms);
+           __m64 dest  = load8888 (dst);
+
+           store8888 (dst, (in_over (ms, alpha, mm_mask, dest)));
+       }
+
+       dst++;
+       w--;
+    }
+
+    _mm_empty ();
+}
+
+FAST_NEAREST_MAINLOOP_COMMON (mmx_8888_n_8888_cover_OVER,
+                             scaled_nearest_scanline_mmx_8888_n_8888_OVER,
+                             uint32_t, uint32_t, uint32_t, COVER, TRUE, TRUE)
+FAST_NEAREST_MAINLOOP_COMMON (mmx_8888_n_8888_pad_OVER,
+                             scaled_nearest_scanline_mmx_8888_n_8888_OVER,
+                             uint32_t, uint32_t, uint32_t, PAD, TRUE, TRUE)
+FAST_NEAREST_MAINLOOP_COMMON (mmx_8888_n_8888_none_OVER,
+                             scaled_nearest_scanline_mmx_8888_n_8888_OVER,
+                             uint32_t, uint32_t, uint32_t, NONE, TRUE, TRUE)
+FAST_NEAREST_MAINLOOP_COMMON (mmx_8888_n_8888_normal_OVER,
+                             scaled_nearest_scanline_mmx_8888_n_8888_OVER,
+                             uint32_t, uint32_t, uint32_t, NORMAL, TRUE, TRUE)
+
 #define BSHIFT ((1 << BILINEAR_INTERPOLATION_BITS))
 #define BMSK (BSHIFT - 1)
 
@@ -3995,6 +4048,15 @@ static const pixman_fast_path_t mmx_fast_paths[] =
     PIXMAN_STD_FAST_PATH    (IN,   a8,       null,     a8,       mmx_composite_in_8_8              ),
     PIXMAN_STD_FAST_PATH    (IN,   solid,    a8,       a8,       mmx_composite_in_n_8_8            ),
 
+    SIMPLE_NEAREST_SOLID_MASK_FAST_PATH (OVER, a8r8g8b8, a8r8g8b8, mmx_8888_n_8888                 ),
+    SIMPLE_NEAREST_SOLID_MASK_FAST_PATH (OVER, a8b8g8r8, a8b8g8r8, mmx_8888_n_8888                 ),
+    SIMPLE_NEAREST_SOLID_MASK_FAST_PATH (OVER, a8r8g8b8, x8r8g8b8, mmx_8888_n_8888                 ),
+    SIMPLE_NEAREST_SOLID_MASK_FAST_PATH (OVER, a8b8g8r8, x8b8g8r8, mmx_8888_n_8888                 ),
+    SIMPLE_NEAREST_SOLID_MASK_FAST_PATH_NORMAL (OVER, a8r8g8b8, a8r8g8b8, mmx_8888_n_8888          ),
+    SIMPLE_NEAREST_SOLID_MASK_FAST_PATH_NORMAL (OVER, a8b8g8r8, a8b8g8r8, mmx_8888_n_8888          ),
+    SIMPLE_NEAREST_SOLID_MASK_FAST_PATH_NORMAL (OVER, a8r8g8b8, x8r8g8b8, mmx_8888_n_8888          ),
+    SIMPLE_NEAREST_SOLID_MASK_FAST_PATH_NORMAL (OVER, a8b8g8r8, x8b8g8r8, mmx_8888_n_8888          ),
+
     SIMPLE_BILINEAR_FAST_PATH (SRC, a8r8g8b8,          a8r8g8b8, mmx_8888_8888                     ),
     SIMPLE_BILINEAR_FAST_PATH (SRC, a8r8g8b8,          x8r8g8b8, mmx_8888_8888                     ),
     SIMPLE_BILINEAR_FAST_PATH (SRC, x8r8g8b8,          x8r8g8b8, mmx_8888_8888                     ),