Inline asm for VP56 arith coder

This is a lot more reliable to get cmov rather than trying to trick gcc into
generating it, useful since it's 2% faster overall.

Patch by Eli Friedman <eli.friedman at gmail>

Originally committed as revision 24471 to svn://svn.ffmpeg.org/ffmpeg/trunk

2 files changed, 61 insertions, 5 deletions

diff --git a/libavcodec/vp56.h b/libavcodec/vp56.h
index ad07a49e9b..1eacdc2158 100644
--- a/libavcodec/vp56.h
+++ b/libavcodec/vp56.h
@@ -208,23 +208,25 @@ static av_always_inline unsigned int vp56_rac_renorm(VP56RangeCoder *c)
     return code_word;
 }
 
+#if ARCH_X86
+#include "x86/vp56_arith.h"
+#endif
+
+#ifndef vp56_rac_get_prob
+#define vp56_rac_get_prob vp56_rac_get_prob
 static inline int vp56_rac_get_prob(VP56RangeCoder *c, uint8_t prob)
 {
-    /* Don't put c->high in a local variable; if we do that, gcc gets
-     * the stupids and turns the code below into a branch again. */
     unsigned int code_word = vp56_rac_renorm(c);
     unsigned int low = 1 + (((c->high - 1) * prob) >> 8);
     unsigned int low_shift = low << 8;
     int bit = code_word >= low_shift;
 
-    /* Incantation to convince GCC to turn these into conditional moves
-     * instead of branches -- faster, as this branch is basically
-     * unpredictable. */
     c->high = bit ? c->high - low : low;
     c->code_word = bit ? code_word - low_shift : code_word;
 
     return bit;
 }
+#endif
 
 // branchy variant, to be used where there's a branch based on the bit decoded
 static av_always_inline int vp56_rac_get_prob_branchy(VP56RangeCoder *c, int prob)
diff --git a/libavcodec/x86/vp56_arith.h b/libavcodec/x86/vp56_arith.h
new file mode 100644
index 0000000000..d05dcb6ed5
--- /dev/null
+++ b/libavcodec/x86/vp56_arith.h
@@ -0,0 +1,54 @@
+/**
+ * VP5 and VP6 compatible video decoder (arith decoder)
+ *
+ * Copyright (C) 2006  Aurelien Jacobs <aurel@gnuage.org>
+ * Copyright (C) 2010  Eli Friedman
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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.
+ *
+ * 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser 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
+ */
+
+#ifndef AVCODEC_X86_VP56_ARITH_H
+#define AVCODEC_X86_VP56_ARITH_H
+
+#if HAVE_FAST_CMOV
+#define vp56_rac_get_prob vp56_rac_get_prob
+static inline int vp56_rac_get_prob(VP56RangeCoder *c, uint8_t prob)
+{
+    unsigned int code_word = vp56_rac_renorm(c);
+    unsigned int high = c->high;
+    unsigned int low = 1 + (((high - 1) * prob) >> 8);
+    unsigned int low_shift = low << 8;
+    int bit = 0;
+
+    __asm__(
+        "subl  %4, %1      \n\t"
+        "subl  %3, %2      \n\t"
+        "leal (%2, %3), %3 \n\t"
+        "setae %b0         \n\t"
+        "cmovb %4, %1      \n\t"
+        "cmovb %3, %2      \n\t"
+        : "+q"(bit), "+r"(high), "+r"(code_word), "+r"(low_shift)
+        : "r"(low)
+    );
+
+    c->high      = high;
+    c->code_word = code_word;
+    return bit;
+}
+#endif
+
+#endif /* AVCODEC_X86_VP56_ARITH_H */