From ed1a11ed52bbd1f15bb9b0416d69b7924bee3191 Mon Sep 17 00:00:00 2001 From: Diego Biurrun Date: Thu, 24 Oct 2013 02:24:05 +0200 Subject: [PATCH] gradfun: x86: Factor out common code for some gradfun_filter_line() variants --- libavfilter/x86/vf_gradfun_init.c | 29 +++++++++++++---------------- 1 file changed, 13 insertions(+), 16 deletions(-) diff --git a/libavfilter/x86/vf_gradfun_init.c b/libavfilter/x86/vf_gradfun_init.c index d45fb9906c..3f23bf6799 100644 --- a/libavfilter/x86/vf_gradfun_init.c +++ b/libavfilter/x86/vf_gradfun_init.c @@ -42,13 +42,13 @@ void ff_gradfun_blur_line_movdqu_sse2(intptr_t x, uint16_t *buf, uint8_t *src1, uint8_t *src2); #if HAVE_YASM -static void gradfun_filter_line_mmxext(uint8_t *dst, uint8_t *src, uint16_t *dc, - int width, int thresh, - const uint16_t *dithers) +static void gradfun_filter_line(uint8_t *dst, uint8_t *src, uint16_t *dc, + int width, int thresh, const uint16_t *dithers, + int alignment) { intptr_t x; - if (width & 3) { - x = width & ~3; + if (width & alignment) { + x = width & ~alignment; ff_gradfun_filter_line_c(dst + x, src + x, dc + x / 2, width - x, thresh, dithers); width = x; @@ -58,21 +58,18 @@ static void gradfun_filter_line_mmxext(uint8_t *dst, uint8_t *src, uint16_t *dc, thresh, dithers); } +static void gradfun_filter_line_mmxext(uint8_t *dst, uint8_t *src, uint16_t *dc, + int width, int thresh, + const uint16_t *dithers) +{ + gradfun_filter_line(dst, src, dc, width, thresh, dithers, 3); +} + static void gradfun_filter_line_ssse3(uint8_t *dst, uint8_t *src, uint16_t *dc, int width, int thresh, const uint16_t *dithers) { - intptr_t x; - if (width & 7) { - // could be 10% faster if I somehow eliminated this - x = width & ~7; - ff_gradfun_filter_line_c(dst + x, src + x, dc + x / 2, - width - x, thresh, dithers); - width = x; - } - x = -width; - ff_gradfun_filter_line_ssse3(x, dst + width, src + width, dc + width / 2, - thresh, dithers); + gradfun_filter_line(dst, src, dc, width, thresh, dithers, 7); } static void gradfun_blur_line_sse2(uint16_t *dc, uint16_t *buf, uint16_t *buf1,