x86: AVX2 predict_16x16_p

[x264] / common / osdep.h

diff --git a/common/osdep.h b/common/osdep.h
index 6249dbe0c53085167a4556ce85c4a8136abbfc90..e4b22f0cf2908d197fd045bc38670f4a231a5186 100644 (file)
--- a/common/osdep.h
+++ b/common/osdep.h
@@ -1,7 +1,7 @@
 /*****************************************************************************
  * osdep.h: platform-specific code
  *****************************************************************************
- * Copyright (C) 2007-2011 x264 project
+ * Copyright (C) 2007-2013 x264 project
  *
  * Authors: Loren Merritt <lorenm@u.washington.edu>
  *          Laurent Aimar <fenrir@via.ecp.fr>
@@ -31,15 +31,10 @@
 #define _FILE_OFFSET_BITS 64
 #include <stdio.h>
 #include <sys/stat.h>
+#include <inttypes.h>
 
 #include "config.h"
 
-#if HAVE_STDINT_H
-#include <stdint.h>
-#else
-#include <inttypes.h>
-#endif
-
 #if !HAVE_LOG2F
 #define log2f(x) (logf(x)/0.693147180559945f)
 #define log2(x) (log(x)/0.693147180559945)
@@ -84,6 +79,7 @@
 #else
 #define DECLARE_ALIGNED( var, n ) var __attribute__((aligned(n)))
 #endif
+#define ALIGNED_32( var ) DECLARE_ALIGNED( var, 32 )
 #define ALIGNED_16( var ) DECLARE_ALIGNED( var, 16 )
 #define ALIGNED_8( var )  DECLARE_ALIGNED( var, 8 )
 #define ALIGNED_4( var )  DECLARE_ALIGNED( var, 4 )
@@ -94,24 +90,41 @@
 // - armcc can't either, but is nice enough to actually tell you so
 // - Apple gcc only maintains 4 byte alignment
 // - llvm can align the stack, but only in svn and (unrelated) it exposes bugs in all released GNU binutils...
+
+#define ALIGNED_ARRAY_EMU( mask, type, name, sub1, ... )\
+    uint8_t name##_u [sizeof(type sub1 __VA_ARGS__) + mask]; \
+    type (*name) __VA_ARGS__ = (void*)((intptr_t)(name##_u+mask) & ~mask)
+
 #if ARCH_ARM && SYS_MACOSX
-#define ALIGNED_ARRAY_8( type, name, sub1, ... )\
-    uint8_t name##_u [sizeof(type sub1 __VA_ARGS__) + 7]; \
-    type (*name) __VA_ARGS__ = (void*)((intptr_t)(name##_u+7) & ~7)
+#define ALIGNED_ARRAY_8( ... ) ALIGNED_ARRAY_EMU( 7, __VA_ARGS__ )
 #else
 #define ALIGNED_ARRAY_8( type, name, sub1, ... )\
     ALIGNED_8( type name sub1 __VA_ARGS__ )
 #endif
 
 #if ARCH_ARM
-#define ALIGNED_ARRAY_16( type, name, sub1, ... )\
-    uint8_t name##_u [sizeof(type sub1 __VA_ARGS__) + 15];\
-    type (*name) __VA_ARGS__ = (void*)((intptr_t)(name##_u+15) & ~15)
+#define ALIGNED_ARRAY_16( ... ) ALIGNED_ARRAY_EMU( 15, __VA_ARGS__ )
 #else
 #define ALIGNED_ARRAY_16( type, name, sub1, ... )\
     ALIGNED_16( type name sub1 __VA_ARGS__ )
 #endif
 
+#define EXPAND(x) x
+
+#define ALIGNED_ARRAY_32( ... ) EXPAND( ALIGNED_ARRAY_EMU( 31, __VA_ARGS__ ) )
+#define ALIGNED_ARRAY_64( ... ) EXPAND( ALIGNED_ARRAY_EMU( 63, __VA_ARGS__ ) )
+
+/* For AVX2 */
+#if ARCH_X86 || ARCH_X86_64
+#define NATIVE_ALIGN 32
+#define ALIGNED_N ALIGNED_32
+#define ALIGNED_ARRAY_N ALIGNED_ARRAY_32
+#else
+#define NATIVE_ALIGN 16
+#define ALIGNED_N ALIGNED_16
+#define ALIGNED_ARRAY_N ALIGNED_ARRAY_16
+#endif
+
 #define UNINIT(x) x=x
 
 #if defined(__GNUC__) && (__GNUC__ > 3 || __GNUC__ == 3 && __GNUC_MINOR__ > 0)
@@ -148,7 +161,7 @@ static inline int x264_pthread_create( x264_pthread_t *t, void *a, void *(*f)(vo
      return 0;
 }
 #define x264_pthread_join(t,s)       { long tmp; \
-                                       wait_for_thread(t,(s)?(long*)(*(s)):&tmp); }
+                                       wait_for_thread(t,(s)?(long*)(s):&tmp); }
 
 #elif HAVE_POSIXTHREAD
 #include <pthread.h>
@@ -203,6 +216,25 @@ int x264_threading_init( void );
 #define x264_threading_init() 0
 #endif
 
+static ALWAYS_INLINE int x264_pthread_fetch_and_add( int *val, int add, x264_pthread_mutex_t *mutex )
+{
+#if HAVE_THREAD
+#if defined(__GNUC__) && (__GNUC__ > 4 || __GNUC__ == 4 && __GNUC_MINOR__ > 0) && ARCH_X86
+    return __sync_fetch_and_add( val, add );
+#else
+    x264_pthread_mutex_lock( mutex );
+    int res = *val;
+    *val += add;
+    x264_pthread_mutex_unlock( mutex );
+    return res;
+#endif
+#else
+    int res = *val;
+    *val += add;
+    return res;
+#endif
+}
+
 #define WORD_SIZE sizeof(void*)
 
 #define asm __asm__
@@ -253,6 +285,13 @@ static ALWAYS_INLINE uint16_t endian_fix16( uint16_t x )
 }
 #endif
 
+/* For values with 4 bits or less. */
+static int ALWAYS_INLINE x264_ctz_4bit( uint32_t x )
+{
+    static uint8_t lut[16] = {4,0,1,0,2,0,1,0,3,0,1,0,2,0,1,0};
+    return lut[x];
+}
+
 #if defined(__GNUC__) && (__GNUC__ > 3 || __GNUC__ == 3 && __GNUC_MINOR__ > 3)
 #define x264_clz(x) __builtin_clz(x)
 #define x264_ctz(x) __builtin_ctz(x)