[vlc-commits] deinterlace/x86: update yadif asm to libav's external asm

Janne Grunau git at videolan.org
Tue Feb 5 11:23:57 CET 2019


vlc | branch: master | Janne Grunau <janne-vlc at jannau.net> | Sun Aug 14 12:59:47 2016 +0200| [97b29c73693b4093263cdae3b9b9e2b60f696bdc] | committer: Jean-Baptiste Kempf

deinterlace/x86: update yadif asm to libav's external asm

Signed-off-by: Jean-Baptiste Kempf <jb at videolan.org>

> http://git.videolan.org/gitweb.cgi/vlc.git/?a=commit;h=97b29c73693b4093263cdae3b9b9e2b60f696bdc
---

 modules/video_filter/Makefile.am                  |   5 +-
 modules/video_filter/deinterlace/algo_yadif.c     |  19 +-
 modules/video_filter/deinterlace/yadif.h          |  68 +-----
 modules/video_filter/deinterlace/yadif_template.h | 282 ----------------------
 modules/video_filter/deinterlace/yadif_x86.asm    | 257 ++++++++++++++++++++
 5 files changed, 276 insertions(+), 355 deletions(-)

diff --git a/modules/video_filter/Makefile.am b/modules/video_filter/Makefile.am
index d3db0c0d0a..51ee92aff1 100644
--- a/modules/video_filter/Makefile.am
+++ b/modules/video_filter/Makefile.am
@@ -131,11 +131,14 @@ libdeinterlace_plugin_la_SOURCES = \
 	video_filter/deinterlace/algo_basic.c video_filter/deinterlace/algo_basic.h \
 	video_filter/deinterlace/algo_x.c video_filter/deinterlace/algo_x.h \
 	video_filter/deinterlace/algo_yadif.c video_filter/deinterlace/algo_yadif.h \
-	video_filter/deinterlace/yadif.h video_filter/deinterlace/yadif_template.h \
+	video_filter/deinterlace/yadif.h \
 	video_filter/deinterlace/algo_phosphor.c video_filter/deinterlace/algo_phosphor.h \
 	video_filter/deinterlace/algo_ivtc.c video_filter/deinterlace/algo_ivtc.h
 # inline ASM doesn't build with -O0
 libdeinterlace_plugin_la_CFLAGS = $(AM_CFLAGS) -O2
+if HAVE_X86ASM
+libdeinterlace_plugin_la_SOURCES += video_filter/deinterlace/yadif_x86.asm
+endif
 if HAVE_NEON
 libdeinterlace_plugin_la_SOURCES += video_filter/deinterlace/merge_arm.S
 libdeinterlace_plugin_la_CFLAGS += -DCAN_COMPILE_ARM
diff --git a/modules/video_filter/deinterlace/algo_yadif.c b/modules/video_filter/deinterlace/algo_yadif.c
index 900111f4cf..6fdecf53a5 100644
--- a/modules/video_filter/deinterlace/algo_yadif.c
+++ b/modules/video_filter/deinterlace/algo_yadif.c
@@ -112,23 +112,18 @@ int RenderYadif( filter_t *p_filter, picture_t *p_dst, picture_t *p_src,
         void (*filter)(uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next,
                        int w, int prefs, int mrefs, int parity, int mode);
 
-/* android clang build for x86 fails as not enough registers are available */
-#if !defined(__ANDROID__)
-# if defined(HAVE_YADIF_SSSE3)
+#if defined(HAVE_X86ASM)
         if( vlc_CPU_SSSE3() )
-            filter = yadif_filter_line_ssse3;
+            filter = vlcpriv_yadif_filter_line_ssse3;
         else
-# endif
-# if defined(HAVE_YADIF_SSE2)
         if( vlc_CPU_SSE2() )
-            filter = yadif_filter_line_sse2;
+            filter = vlcpriv_yadif_filter_line_sse2;
         else
-# endif
-# if defined(HAVE_YADIF_MMX)
-        if( vlc_CPU_MMX() )
-            filter = yadif_filter_line_mmx;
+#if defined(__i386__)
+        if( vlc_CPU_MMXEXT() )
+            filter = vlcpriv_yadif_filter_line_mmxext;
         else
-# endif
+#endif
 #endif
             filter = yadif_filter_line_c;
 
diff --git a/modules/video_filter/deinterlace/yadif.h b/modules/video_filter/deinterlace/yadif.h
index af16443d08..bb07d9c1b5 100644
--- a/modules/video_filter/deinterlace/yadif.h
+++ b/modules/video_filter/deinterlace/yadif.h
@@ -22,66 +22,6 @@
 #   include "config.h"
 #endif
 
-#include <stdalign.h>
-
-#if defined(__GNUC__)
-# define ATTR_USED __attribute__((used))
-#else
-# define ATTR_USED
-#endif
-
-typedef intptr_t x86_reg;
-typedef struct { uint64_t a, b; } xmm_reg;
-
-const ATTR_USED alignas (16) xmm_reg pb_1 = {
-    0x0101010101010101ULL, 0x0101010101010101ULL
-};
-const ATTR_USED alignas (16) xmm_reg pw_1 = {
-    0x0001000100010001ULL, 0x0001000100010001ULL
-};
-
-#ifdef CAN_COMPILE_SSSE3
-#if defined(__SSE__) || defined(__GNUC__) || defined(__clang__)
-// ================ SSSE3 =================
-#define HAVE_YADIF_SSSE3
-#define COMPILE_TEMPLATE_SSE 1
-#define COMPILE_TEMPLATE_SSSE3 1
-#define VLC_TARGET VLC_SSE
-#define RENAME(a) a ## _ssse3
-#include "yadif_template.h"
-#undef COMPILE_TEMPLATE_SSE
-#undef COMPILE_TEMPLATE_SSSE3
-#undef VLC_TARGET
-#undef RENAME
-#endif
-#endif
-
-#ifdef CAN_COMPILE_SSE2
-#if defined(__SSE__) || defined(__GNUC__) || defined(__clang__)
-// ================= SSE2 =================
-#define HAVE_YADIF_SSE2
-#define COMPILE_TEMPLATE_SSE 1
-#define VLC_TARGET VLC_SSE
-#define RENAME(a) a ## _sse2
-#include "yadif_template.h"
-#undef COMPILE_TEMPLATE_SSE
-#undef VLC_TARGET
-#undef RENAME
-#endif
-#endif
-
-#ifdef CAN_COMPILE_MMX
-#if defined(__MMX__) || defined(__GNUC__) || defined(__clang__)
-// ================ MMX =================
-#define HAVE_YADIF_MMX
-#define VLC_TARGET VLC_MMX
-#define RENAME(a) a ## _mmx
-#include "yadif_template.h"
-#undef VLC_TARGET
-#undef RENAME
-#endif
-#endif
-
 #define FFABS abs
 
 #define CHECK(j)\
@@ -151,3 +91,11 @@ static void yadif_filter_line_c_16bit(uint8_t *dst8, uint8_t *prev8, uint8_t *cu
     prefs /= 2;
     FILTER
 }
+
+#if defined(__i386__) || defined(__x86_64__)
+void vlcpriv_yadif_filter_line_ssse3(uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int prefs, int mrefs, int parity, int mode);
+void vlcpriv_yadif_filter_line_sse2(uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int prefs, int mrefs, int parity, int mode);
+#endif
+#if defined(__i386__)
+void vlcpriv_yadif_filter_line_mmxext(uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int prefs, int mrefs, int parity, int mode);
+#endif
diff --git a/modules/video_filter/deinterlace/yadif_template.h b/modules/video_filter/deinterlace/yadif_template.h
deleted file mode 100644
index df4897db35..0000000000
--- a/modules/video_filter/deinterlace/yadif_template.h
+++ /dev/null
@@ -1,282 +0,0 @@
-/*
- * Copyright (C) 2006 Michael Niedermayer <michaelni at gmx.at>
- *
- * This file is part of FFmpeg.
- *
- * FFmpeg is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
- *
- * FFmpeg is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License along
- * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
- * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
- */
-
-/* For some reason clang doens't like that %%rip macro */
-
-#if defined(__x86_64__) && !defined(__APPLE__)
-#define MANGLE(a) "" #a "(%%rip)"
-#define MANGLEVARIABLES [mode] "g"(mode),
-#else
-#define MANGLE(a) "" "%["#a"]"
-#define MANGLEVARIABLES [pw_1] "m"(pw_1),\
-             [pb_1] "m"(pb_1),\
-             [mode] "g"(mode),
-#endif
-
-#ifdef COMPILE_TEMPLATE_SSE
-#define REGMM "xmm"
-#define MM "%%"REGMM
-#define MOV  "movq"
-#define MOVQ "movdqa"
-#define MOVQU "movdqu"
-#define STEP 8
-#define LOAD(mem,dst) \
-            MOV"       "mem", "dst" \n\t"\
-            "punpcklbw "MM"7, "dst" \n\t"
-#define PSRL1(reg) "psrldq $1, "reg" \n\t"
-#define PSRL2(reg) "psrldq $2, "reg" \n\t"
-#define PSHUF(src,dst) "movdqa "dst", "src" \n\t"\
-                       "psrldq $2, "src"     \n\t"
-#else
-#define REGMM "mm"
-#define MM "%%"REGMM
-#define MOV  "movd"
-#define MOVQ "movq"
-#define MOVQU "movq"
-#define STEP 4
-#define LOAD(mem,dst) \
-            MOV"       "mem", "dst" \n\t"\
-            "punpcklbw "MM"7, "dst" \n\t"
-#define PSRL1(reg) "psrlq $8, "reg" \n\t"
-#define PSRL2(reg) "psrlq $16, "reg" \n\t"
-#define PSHUF(src,dst) "pshufw $9, "dst", "src" \n\t"
-#endif
-
-#ifdef COMPILE_TEMPLATE_SSSE3
-#define PABS(tmp,dst) \
-            "pabsw     "dst", "dst" \n\t"
-#else
-#define PABS(tmp,dst) \
-            "pxor     "tmp", "tmp" \n\t"\
-            "psubw    "dst", "tmp" \n\t"\
-            "pmaxsw   "tmp", "dst" \n\t"
-#endif
-
-
-#define CHECK(pj,mj) \
-            MOVQU" "#pj"(%[cur],%[mrefs]), "MM"2 \n\t" /* cur[x-refs-1+j] */\
-            MOVQU" "#mj"(%[cur],%[prefs]), "MM"3 \n\t" /* cur[x+refs-1-j] */\
-            MOVQ"      "MM"2, "MM"4 \n\t"\
-            MOVQ"      "MM"2, "MM"5 \n\t"\
-            "pxor      "MM"3, "MM"4 \n\t"\
-            "pavgb     "MM"3, "MM"5 \n\t"\
-            "pand      "MANGLE(pb_1)", "MM"4 \n\t"\
-            "psubusb   "MM"4, "MM"5 \n\t"\
-            PSRL1(MM"5")                 \
-            "punpcklbw "MM"7, "MM"5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
-            MOVQ"      "MM"2, "MM"4 \n\t"\
-            "psubusb   "MM"3, "MM"2 \n\t"\
-            "psubusb   "MM"4, "MM"3 \n\t"\
-            "pmaxub    "MM"3, "MM"2 \n\t"\
-            MOVQ"      "MM"2, "MM"3 \n\t"\
-            MOVQ"      "MM"2, "MM"4 \n\t" /* ABS(cur[x-refs-1+j] - cur[x+refs-1-j]) */\
-            PSRL1(MM"3")                  /* ABS(cur[x-refs  +j] - cur[x+refs  -j]) */\
-            PSRL2(MM"4")                  /* ABS(cur[x-refs+1+j] - cur[x+refs+1-j]) */\
-            "punpcklbw "MM"7, "MM"2 \n\t"\
-            "punpcklbw "MM"7, "MM"3 \n\t"\
-            "punpcklbw "MM"7, "MM"4 \n\t"\
-            "paddw     "MM"3, "MM"2 \n\t"\
-            "paddw     "MM"4, "MM"2 \n\t" /* score */
-
-#define CHECK1 \
-            MOVQ"      "MM"0, "MM"3 \n\t"\
-            "pcmpgtw   "MM"2, "MM"3 \n\t" /* if(score < spatial_score) */\
-            "pminsw    "MM"2, "MM"0 \n\t" /* spatial_score= score; */\
-            MOVQ"      "MM"3, "MM"6 \n\t"\
-            "pand      "MM"3, "MM"5 \n\t"\
-            "pandn     "MM"1, "MM"3 \n\t"\
-            "por       "MM"5, "MM"3 \n\t"\
-            MOVQ"      "MM"3, "MM"1 \n\t" /* spatial_pred= (cur[x-refs+j] + cur[x+refs-j])>>1; */
-
-#define CHECK2 /* pretend not to have checked dir=2 if dir=1 was bad.\
-                  hurts both quality and speed, but matches the C version. */\
-            "paddw    "MANGLE(pw_1)", "MM"6 \n\t"\
-            "psllw     $14,   "MM"6 \n\t"\
-            "paddsw    "MM"6, "MM"2 \n\t"\
-            MOVQ"      "MM"0, "MM"3 \n\t"\
-            "pcmpgtw   "MM"2, "MM"3 \n\t"\
-            "pminsw    "MM"2, "MM"0 \n\t"\
-            "pand      "MM"3, "MM"5 \n\t"\
-            "pandn     "MM"1, "MM"3 \n\t"\
-            "por       "MM"5, "MM"3 \n\t"\
-            MOVQ"      "MM"3, "MM"1 \n\t"
-
-#if defined(__MINGW32__) && defined(_WIN32) && !defined(_WIN64)
-__attribute__((__force_align_arg_pointer__))
-#endif
-VLC_TARGET static void RENAME(yadif_filter_line)(uint8_t *dst,
-                              uint8_t *prev, uint8_t *cur, uint8_t *next,
-                              int w, int prefs, int mrefs, int parity, int mode)
-{
-    uint8_t tmpU[5*16];
-    uint8_t *tmp= (uint8_t*)(((uintptr_t)(tmpU+15)) & ~15);
-    int x;
-
-#define FILTER\
-    for(x=0; x<w; x+=STEP){\
-        __asm__ volatile(\
-            "pxor      "MM"7, "MM"7 \n\t"\
-            LOAD("(%[cur],%[mrefs])", MM"0") /* c = cur[x-refs] */\
-            LOAD("(%[cur],%[prefs])", MM"1") /* e = cur[x+refs] */\
-            LOAD("(%["prev2"])", MM"2") /* prev2[x] */\
-            LOAD("(%["next2"])", MM"3") /* next2[x] */\
-            MOVQ"      "MM"3, "MM"4 \n\t"\
-            "paddw     "MM"2, "MM"3 \n\t"\
-            "psraw     $1,    "MM"3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
-            MOVQ"      "MM"0,   (%[tmp]) \n\t" /* c */\
-            MOVQ"      "MM"3, 16(%[tmp]) \n\t" /* d */\
-            MOVQ"      "MM"1, 32(%[tmp]) \n\t" /* e */\
-            "psubw     "MM"4, "MM"2 \n\t"\
-            PABS(      MM"4", MM"2") /* temporal_diff0 */\
-            LOAD("(%[prev],%[mrefs])", MM"3") /* prev[x-refs] */\
-            LOAD("(%[prev],%[prefs])", MM"4") /* prev[x+refs] */\
-            "psubw     "MM"0, "MM"3 \n\t"\
-            "psubw     "MM"1, "MM"4 \n\t"\
-            PABS(      MM"5", MM"3")\
-            PABS(      MM"5", MM"4")\
-            "paddw     "MM"4, "MM"3 \n\t" /* temporal_diff1 */\
-            "psrlw     $1,    "MM"2 \n\t"\
-            "psrlw     $1,    "MM"3 \n\t"\
-            "pmaxsw    "MM"3, "MM"2 \n\t"\
-            LOAD("(%[next],%[mrefs])", MM"3") /* next[x-refs] */\
-            LOAD("(%[next],%[prefs])", MM"4") /* next[x+refs] */\
-            "psubw     "MM"0, "MM"3 \n\t"\
-            "psubw     "MM"1, "MM"4 \n\t"\
-            PABS(      MM"5", MM"3")\
-            PABS(      MM"5", MM"4")\
-            "paddw     "MM"4, "MM"3 \n\t" /* temporal_diff2 */\
-            "psrlw     $1,    "MM"3 \n\t"\
-            "pmaxsw    "MM"3, "MM"2 \n\t"\
-            MOVQ"      "MM"2, 48(%[tmp]) \n\t" /* diff */\
-\
-            "paddw     "MM"0, "MM"1 \n\t"\
-            "paddw     "MM"0, "MM"0 \n\t"\
-            "psubw     "MM"1, "MM"0 \n\t"\
-            "psrlw     $1,    "MM"1 \n\t" /* spatial_pred */\
-            PABS(      MM"2", MM"0")      /* ABS(c-e) */\
-\
-            MOVQU" -1(%[cur],%[mrefs]), "MM"2 \n\t" /* cur[x-refs-1] */\
-            MOVQU" -1(%[cur],%[prefs]), "MM"3 \n\t" /* cur[x+refs-1] */\
-            MOVQ"      "MM"2, "MM"4 \n\t"\
-            "psubusb   "MM"3, "MM"2 \n\t"\
-            "psubusb   "MM"4, "MM"3 \n\t"\
-            "pmaxub    "MM"3, "MM"2 \n\t"\
-            PSHUF(MM"3", MM"2") \
-            "punpcklbw "MM"7, "MM"2 \n\t" /* ABS(cur[x-refs-1] - cur[x+refs-1]) */\
-            "punpcklbw "MM"7, "MM"3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
-            "paddw     "MM"2, "MM"0 \n\t"\
-            "paddw     "MM"3, "MM"0 \n\t"\
-            "psubw    "MANGLE(pw_1)", "MM"0 \n\t" /* spatial_score */\
-\
-            CHECK(-2,0)\
-            CHECK1\
-            CHECK(-3,1)\
-            CHECK2\
-            CHECK(0,-2)\
-            CHECK1\
-            CHECK(1,-3)\
-            CHECK2\
-\
-            /* if(p->mode<2) ... */\
-            MOVQ"    48(%[tmp]), "MM"6 \n\t" /* diff */\
-            "cmpl      $2, %[mode] \n\t"\
-            "jge       1f \n\t"\
-            LOAD("(%["prev2"],%[mrefs],2)", MM"2") /* prev2[x-2*refs] */\
-            LOAD("(%["next2"],%[mrefs],2)", MM"4") /* next2[x-2*refs] */\
-            LOAD("(%["prev2"],%[prefs],2)", MM"3") /* prev2[x+2*refs] */\
-            LOAD("(%["next2"],%[prefs],2)", MM"5") /* next2[x+2*refs] */\
-            "paddw     "MM"4, "MM"2 \n\t"\
-            "paddw     "MM"5, "MM"3 \n\t"\
-            "psrlw     $1,    "MM"2 \n\t" /* b */\
-            "psrlw     $1,    "MM"3 \n\t" /* f */\
-            MOVQ"   (%[tmp]), "MM"4 \n\t" /* c */\
-            MOVQ" 16(%[tmp]), "MM"5 \n\t" /* d */\
-            MOVQ" 32(%[tmp]), "MM"7 \n\t" /* e */\
-            "psubw     "MM"4, "MM"2 \n\t" /* b-c */\
-            "psubw     "MM"7, "MM"3 \n\t" /* f-e */\
-            MOVQ"      "MM"5, "MM"0 \n\t"\
-            "psubw     "MM"4, "MM"5 \n\t" /* d-c */\
-            "psubw     "MM"7, "MM"0 \n\t" /* d-e */\
-            MOVQ"      "MM"2, "MM"4 \n\t"\
-            "pminsw    "MM"3, "MM"2 \n\t"\
-            "pmaxsw    "MM"4, "MM"3 \n\t"\
-            "pmaxsw    "MM"5, "MM"2 \n\t"\
-            "pminsw    "MM"5, "MM"3 \n\t"\
-            "pmaxsw    "MM"0, "MM"2 \n\t" /* max */\
-            "pminsw    "MM"0, "MM"3 \n\t" /* min */\
-            "pxor      "MM"4, "MM"4 \n\t"\
-            "pmaxsw    "MM"3, "MM"6 \n\t"\
-            "psubw     "MM"2, "MM"4 \n\t" /* -max */\
-            "pmaxsw    "MM"4, "MM"6 \n\t" /* diff= MAX3(diff, min, -max); */\
-            "1: \n\t"\
-\
-            MOVQ" 16(%[tmp]), "MM"2 \n\t" /* d */\
-            MOVQ"      "MM"2, "MM"3 \n\t"\
-            "psubw     "MM"6, "MM"2 \n\t" /* d-diff */\
-            "paddw     "MM"6, "MM"3 \n\t" /* d+diff */\
-            "pmaxsw    "MM"2, "MM"1 \n\t"\
-            "pminsw    "MM"3, "MM"1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
-            "packuswb  "MM"1, "MM"1 \n\t"\
-\
-            ::[prev] "r"(prev),\
-             [cur]  "r"(cur),\
-             [next] "r"(next),\
-             [prefs]"r"((x86_reg)prefs),\
-             [mrefs]"r"((x86_reg)mrefs),\
-             MANGLEVARIABLES\
-             [tmp]  "r"(tmp)\
-        );\
-        __asm__ volatile(MOV" "MM"1, %0" :"=m"(*dst));\
-        dst += STEP;\
-        prev+= STEP;\
-        cur += STEP;\
-        next+= STEP;\
-    }
-
-    if (parity) {
-#define prev2 "prev"
-#define next2 "cur"
-        FILTER
-#undef prev2
-#undef next2
-    } else {
-#define prev2 "cur"
-#define next2 "next"
-        FILTER
-#undef prev2
-#undef next2
-    }
-}
-#undef STEP
-#undef REGMM
-#undef MM
-#undef MOV
-#undef MOVQ
-#undef MOVQU
-#undef PSHUF
-#undef PSRL1
-#undef PSRL2
-#undef LOAD
-#undef PABS
-#undef CHECK
-#undef CHECK1
-#undef CHECK2
-#undef FILTER
-
diff --git a/modules/video_filter/deinterlace/yadif_x86.asm b/modules/video_filter/deinterlace/yadif_x86.asm
new file mode 100644
index 0000000000..fc1e5624c5
--- /dev/null
+++ b/modules/video_filter/deinterlace/yadif_x86.asm
@@ -0,0 +1,257 @@
+;*****************************************************************************
+;* x86-optimized functions for yadif filter
+;*
+;* Copyright (C) 2006 Michael Niedermayer <michaelni at gmx.at>
+;* Copyright (c) 2013 Daniel Kang <daniel.d.kang at gmail.com>
+;*
+;* This file is part of Libav.
+;*
+;* Libav is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* Libav is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with Libav; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "x86util.asm"
+
+SECTION_RODATA
+
+pb_1: times 16 db 1
+pw_1: times  8 dw 1
+
+SECTION .text
+
+%macro CHECK 2
+    movu      m2, [curq+t1+%1]
+    movu      m3, [curq+t0+%2]
+    mova      m4, m2
+    mova      m5, m2
+    pxor      m4, m3
+    pavgb     m5, m3
+    pand      m4, [pb_1]
+    psubusb   m5, m4
+%if mmsize == 16
+    psrldq    m5, 1
+%else
+    psrlq     m5, 8
+%endif
+    punpcklbw m5, m7
+    mova      m4, m2
+    psubusb   m2, m3
+    psubusb   m3, m4
+    pmaxub    m2, m3
+    mova      m3, m2
+    mova      m4, m2
+%if mmsize == 16
+    psrldq    m3, 1
+    psrldq    m4, 2
+%else
+    psrlq     m3, 8
+    psrlq     m4, 16
+%endif
+    punpcklbw m2, m7
+    punpcklbw m3, m7
+    punpcklbw m4, m7
+    paddw     m2, m3
+    paddw     m2, m4
+%endmacro
+
+%macro CHECK1 0
+    mova    m3, m0
+    pcmpgtw m3, m2
+    pminsw  m0, m2
+    mova    m6, m3
+    pand    m5, m3
+    pandn   m3, m1
+    por     m3, m5
+    mova    m1, m3
+%endmacro
+
+%macro CHECK2 0
+    paddw   m6, [pw_1]
+    psllw   m6, 14
+    paddsw  m2, m6
+    mova    m3, m0
+    pcmpgtw m3, m2
+    pminsw  m0, m2
+    pand    m5, m3
+    pandn   m3, m1
+    por     m3, m5
+    mova    m1, m3
+%endmacro
+
+%macro LOAD 2
+    movh      m%1, %2
+    punpcklbw m%1, m7
+%endmacro
+
+%macro FILTER 3
+.loop%1:
+    pxor         m7, m7
+    LOAD          0, [curq+t1]
+    LOAD          1, [curq+t0]
+    LOAD          2, [%2]
+    LOAD          3, [%3]
+    mova         m4, m3
+    paddw        m3, m2
+    psraw        m3, 1
+    mova   [rsp+ 0], m0
+    mova   [rsp+16], m3
+    mova   [rsp+32], m1
+    psubw        m2, m4
+    ABS1         m2, m4
+    LOAD          3, [prevq+t1]
+    LOAD          4, [prevq+t0]
+    psubw        m3, m0
+    psubw        m4, m1
+    ABS1         m3, m5
+    ABS1         m4, m5
+    paddw        m3, m4
+    psrlw        m2, 1
+    psrlw        m3, 1
+    pmaxsw       m2, m3
+    LOAD          3, [nextq+t1]
+    LOAD          4, [nextq+t0]
+    psubw        m3, m0
+    psubw        m4, m1
+    ABS1         m3, m5
+    ABS1         m4, m5
+    paddw        m3, m4
+    psrlw        m3, 1
+    pmaxsw       m2, m3
+    mova   [rsp+48], m2
+
+    paddw        m1, m0
+    paddw        m0, m0
+    psubw        m0, m1
+    psrlw        m1, 1
+    ABS1         m0, m2
+
+    movu         m2, [curq+t1-1]
+    movu         m3, [curq+t0-1]
+    mova         m4, m2
+    psubusb      m2, m3
+    psubusb      m3, m4
+    pmaxub       m2, m3
+%if mmsize == 16
+    mova         m3, m2
+    psrldq       m3, 2
+%else
+    pshufw       m3, m2, q0021
+%endif
+    punpcklbw    m2, m7
+    punpcklbw    m3, m7
+    paddw        m0, m2
+    paddw        m0, m3
+    psubw        m0, [pw_1]
+
+    CHECK -2, 0
+    CHECK1
+    CHECK -3, 1
+    CHECK2
+    CHECK 0, -2
+    CHECK1
+    CHECK 1, -3
+    CHECK2
+
+    mova         m6, [rsp+48]
+    cmp   DWORD r8m, 2
+    jge .end%1
+    LOAD          2, [%2+t1*2]
+    LOAD          4, [%3+t1*2]
+    LOAD          3, [%2+t0*2]
+    LOAD          5, [%3+t0*2]
+    paddw        m2, m4
+    paddw        m3, m5
+    psrlw        m2, 1
+    psrlw        m3, 1
+    mova         m4, [rsp+ 0]
+    mova         m5, [rsp+16]
+    mova         m7, [rsp+32]
+    psubw        m2, m4
+    psubw        m3, m7
+    mova         m0, m5
+    psubw        m5, m4
+    psubw        m0, m7
+    mova         m4, m2
+    pminsw       m2, m3
+    pmaxsw       m3, m4
+    pmaxsw       m2, m5
+    pminsw       m3, m5
+    pmaxsw       m2, m0
+    pminsw       m3, m0
+    pxor         m4, m4
+    pmaxsw       m6, m3
+    psubw        m4, m2
+    pmaxsw       m6, m4
+
+.end%1:
+    mova         m2, [rsp+16]
+    mova         m3, m2
+    psubw        m2, m6
+    paddw        m3, m6
+    pmaxsw       m1, m2
+    pminsw       m1, m3
+    packuswb     m1, m1
+
+    movh     [dstq], m1
+    add        dstq, mmsize/2
+    add       prevq, mmsize/2
+    add        curq, mmsize/2
+    add       nextq, mmsize/2
+    sub   DWORD r4m, mmsize/2
+    jg .loop%1
+%endmacro
+
+%macro YADIF 0
+%if ARCH_X86_32
+cglobal yadif_filter_line, 4, 6, 8, 80, dst, prev, cur, next, w, prefs, \
+                                        mrefs, parity, mode
+%else
+cglobal yadif_filter_line, 4, 7, 8, 80, dst, prev, cur, next, w, prefs, \
+                                        mrefs, parity, mode
+%endif
+    cmp      DWORD wm, 0
+    jle .ret
+%if ARCH_X86_32
+    mov            r4, r5mp
+    mov            r5, r6mp
+    DECLARE_REG_TMP 4,5
+%else
+    movsxd         r5, DWORD r5m
+    movsxd         r6, DWORD r6m
+    DECLARE_REG_TMP 5,6
+%endif
+
+    cmp DWORD paritym, 0
+    je .parity0
+    FILTER 1, prevq, curq
+    jmp .ret
+
+.parity0:
+    FILTER 0, curq, nextq
+
+.ret:
+%if mmsize == 8
+    emms
+%endif
+    RET
+%endmacro
+
+INIT_XMM ssse3
+YADIF
+INIT_XMM sse2
+YADIF
+%if ARCH_X86_32
+INIT_MMX mmxext
+YADIF
+%endif



More information about the vlc-commits mailing list