[Pixman] [PATCH] sse2: faster bilinear scaling (use _mm_loadl_epi64)
Matt Turner
mattst88 at gmail.com
Mon Jun 25 07:49:31 PDT 2012
On Mon, Jun 25, 2012 at 12:50 AM, Siarhei Siamashka
<siarhei.siamashka at gmail.com> wrote:
> Using _mm_loadl_epi64() to load two pixels at once (pairs of top
> and bottom pixels) is faster than loading each pixel separately
> and combining them with _mm_set_epi32().
>
> === cairo-perf-trace ===
>
> before: image firefox-fishtank 66.912 66.931 0.13% 3/3
> after: image firefox-fishtank 57.584 58.349 0.74% 3/3
>
> === lowlevel-blt-bench ===
>
> before: src_8888_8888 = L1: 181.10 L2: 179.14 M:178.08 ( 11.02%) HT:153.22 VT:133.45 R:142.24 RT: 95.32
> after: src_8888_8888 = L1: 228.68 L2: 225.75 M:223.98 ( 14.23%) HT:185.32 VT:155.06 R:162.73 RT:102.52
> ---
> pixman/pixman-sse2.c | 15 +++++++--------
> 1 files changed, 7 insertions(+), 8 deletions(-)
>
> diff --git a/pixman/pixman-sse2.c b/pixman/pixman-sse2.c
> index 0604254..ef82a18 100644
> --- a/pixman/pixman-sse2.c
> +++ b/pixman/pixman-sse2.c
> @@ -5377,17 +5377,16 @@ FAST_NEAREST_MAINLOOP_COMMON (sse2_8888_n_8888_none_OVER,
> #define BILINEAR_INTERPOLATE_ONE_PIXEL(pix) \
> do { \
> __m128i xmm_wh, xmm_lo, xmm_hi, a; \
> - /* fetch 2x2 pixel block into sse2 register */ \
> - uint32_t tl = src_top [pixman_fixed_to_int (vx)]; \
> - uint32_t tr = src_top [pixman_fixed_to_int (vx) + 1]; \
> - uint32_t bl = src_bottom [pixman_fixed_to_int (vx)]; \
> - uint32_t br = src_bottom [pixman_fixed_to_int (vx) + 1]; \
> - a = _mm_set_epi32 (tr, tl, br, bl); \
> + /* fetch 2x2 pixel block into sse2 registers */ \
> + __m128i tltr = _mm_loadl_epi64 ( \
> + (__m128i *)&src_top[pixman_fixed_to_int (vx)]); \
> + __m128i blbr = _mm_loadl_epi64 ( \
> + (__m128i *)&src_bottom[pixman_fixed_to_int (vx)]); \
> vx += unit_x; \
> /* vertical interpolation */ \
> - a = _mm_add_epi16 (_mm_mullo_epi16 (_mm_unpackhi_epi8 (a, xmm_zero), \
> + a = _mm_add_epi16 (_mm_mullo_epi16 (_mm_unpacklo_epi8 (tltr, xmm_zero), \
> xmm_wt), \
> - _mm_mullo_epi16 (_mm_unpacklo_epi8 (a, xmm_zero), \
> + _mm_mullo_epi16 (_mm_unpacklo_epi8 (blbr, xmm_zero), \
> xmm_wb)); \
> /* calculate horizontal weights */ \
> xmm_wh = _mm_add_epi16 (xmm_addc, \
> --
> 1.7.3.4
Great. I'm glad that suggestion actually made a difference.
More information about the Pixman
mailing list