avfilter: x86: Use more precise compile template names
[ffmpeg.git] / libavfilter / x86 / yadif_template.c
index 5788a08..02b0c9f 100644 (file)
@@ -1,24 +1,24 @@
 /*
  * Copyright (C) 2006 Michael Niedermayer <michaelni@gmx.at>
  *
- * This file is part of FFmpeg.
+ * This file is part of Libav.
  *
- * FFmpeg is free software; you can redistribute it and/or modify
+ * Libav 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,
+ * 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 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.,
+ * with Libav; if not, write to the Free Software Foundation, Inc.,
  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  */
 
-#ifdef COMPILE_TEMPLATE_SSE
+#ifdef COMPILE_TEMPLATE_SSE2
 #define MM "%%xmm"
 #define MOV  "movq"
 #define MOVQ "movdqa"
@@ -62,7 +62,7 @@
             MOVQ"      "MM"2, "MM"5 \n\t"\
             "pxor      "MM"3, "MM"4 \n\t"\
             "pavgb     "MM"3, "MM"5 \n\t"\
-            "pand     "MANGLE(ff_pb_1)", "MM"4 \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 */\
@@ -92,7 +92,7 @@
 
 #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(ff_pw_1)", "MM"6 \n\t"\
+            "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"\
             "por       "MM"5, "MM"3 \n\t"\
             MOVQ"      "MM"3, "MM"1 \n\t"
 
-void RENAME(ff_yadif_filter_line)(uint8_t *dst,
-                                  uint8_t *prev, uint8_t *cur, uint8_t *next,
-                                  int w, int refs, int parity, int mode)
+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)
 {
-    DECLARE_ALIGNED(16, uint8_t, tmp0[16]);
-    DECLARE_ALIGNED(16, uint8_t, tmp1[16]);
-    DECLARE_ALIGNED(16, uint8_t, tmp2[16]);
-    DECLARE_ALIGNED(16, uint8_t, tmp3[16]);
+    DECLARE_ALIGNED(16, uint8_t, tmp)[16*4];
     int x;
 
 #define FILTER\
@@ -124,9 +121,9 @@ void RENAME(ff_yadif_filter_line)(uint8_t *dst,
             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, %[tmp0] \n\t" /* c */\
-            MOVQ"      "MM"3, %[tmp1] \n\t" /* d */\
-            MOVQ"      "MM"1, %[tmp2] \n\t" /* e */\
+            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] */\
@@ -148,7 +145,7 @@ void RENAME(ff_yadif_filter_line)(uint8_t *dst,
             "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, %[tmp3] \n\t" /* diff */\
+            MOVQ"      "MM"2, 48(%[tmp]) \n\t" /* diff */\
 \
             "paddw     "MM"0, "MM"1 \n\t"\
             "paddw     "MM"0, "MM"0 \n\t"\
@@ -167,7 +164,7 @@ void RENAME(ff_yadif_filter_line)(uint8_t *dst,
             "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(ff_pw_1)", "MM"0 \n\t" /* spatial_score */\
+            "psubw    "MANGLE(pw_1)", "MM"0 \n\t" /* spatial_score */\
 \
             CHECK(-2,0)\
             CHECK1\
@@ -179,7 +176,7 @@ void RENAME(ff_yadif_filter_line)(uint8_t *dst,
             CHECK2\
 \
             /* if(p->mode<2) ... */\
-            MOVQ"    %[tmp3], "MM"6 \n\t" /* diff */\
+            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] */\
@@ -190,9 +187,9 @@ void RENAME(ff_yadif_filter_line)(uint8_t *dst,
             "paddw     "MM"5, "MM"3 \n\t"\
             "psrlw     $1,    "MM"2 \n\t" /* b */\
             "psrlw     $1,    "MM"3 \n\t" /* f */\
-            MOVQ"    %[tmp0], "MM"4 \n\t" /* c */\
-            MOVQ"    %[tmp1], "MM"5 \n\t" /* d */\
-            MOVQ"    %[tmp2], "MM"7 \n\t" /* e */\
+            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"\
@@ -211,7 +208,7 @@ void RENAME(ff_yadif_filter_line)(uint8_t *dst,
             "pmaxsw    "MM"4, "MM"6 \n\t" /* diff= MAX3(diff, min, -max); */\
             "1: \n\t"\
 \
-            MOVQ"    %[tmp1], "MM"2 \n\t" /* d */\
+            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 */\
@@ -219,16 +216,13 @@ void RENAME(ff_yadif_filter_line)(uint8_t *dst,
             "pminsw    "MM"3, "MM"1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
             "packuswb  "MM"1, "MM"1 \n\t"\
 \
-            :[tmp0]"=m"(tmp0),\
-             [tmp1]"=m"(tmp1),\
-             [tmp2]"=m"(tmp2),\
-             [tmp3]"=m"(tmp3)\
-            :[prev] "r"(prev),\
+            ::[prev] "r"(prev),\
              [cur]  "r"(cur),\
              [next] "r"(next),\
-             [prefs]"r"((x86_reg)refs),\
-             [mrefs]"r"((x86_reg)-refs),\
-             [mode] "g"(mode)\
+             [prefs]"r"((x86_reg)prefs),\
+             [mrefs]"r"((x86_reg)mrefs),\
+             [mode] "g"(mode),\
+             [tmp]  "r"(tmp)\
         );\
         __asm__ volatile(MOV" "MM"1, %0" :"=m"(*dst));\
         dst += STEP;\
@@ -265,4 +259,3 @@ void RENAME(ff_yadif_filter_line)(uint8_t *dst,
 #undef CHECK1
 #undef CHECK2
 #undef FILTER
-