diff options
author | Matt Turner <mattst88@gmail.com> | 2013-01-01 21:18:09 -0800 |
---|---|---|
committer | Matt Turner <mattst88@gmail.com> | 2014-03-13 19:45:51 -0700 |
commit | 14c379391cf8045f507dd6d61f8fe9a5ec27cf56 (patch) | |
tree | 1157e9fca4a1974fce091ef61bbe80b6170ad14c | |
parent | 82d094654a46bd97d47f1f132a01ae0a74b986f3 (diff) |
mmx: Add nearest over_8888_n_8888
Unscaled:
over_8888_n_8888 = L1: 395.65 L2: 395.19 M:372.80 ( 19.59%) HT:257.44 VT:194.30 R:164.31 RT: 75.20 ( 882Kops/s)
Before:
over_8888_n_8888 = L1: 107.65 L2: 106.34 M:103.18 ( 5.44%) HT: 49.62 VT: 45.55 R: 43.48 RT: 20.38 ( 242Kops/s)
After:
over_8888_n_8888 = L1: 224.50 L2: 223.42 M:215.18 ( 11.30%) HT:172.70 VT:148.65 R:127.62 RT: 64.50 ( 615Kops/s)
-rw-r--r-- | pixman/pixman-mmx.c | 62 |
1 files changed, 62 insertions, 0 deletions
diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c index f9a92ce..63f4cdf 100644 --- a/pixman/pixman-mmx.c +++ b/pixman/pixman-mmx.c @@ -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 ), |