Merge remote-tracking branch 'qatar/master'

* qatar/master: (28 commits)
  Replace usages of av_get_bits_per_sample_fmt() with av_get_bytes_per_sample().
  x86: cabac: fix register constraints for 32-bit mode
  cabac: move x86 asm to libavcodec/x86/cabac.h
  x86: h264: cast pointers to intptr_t rather than int
  x86: h264: remove hardcoded edi in decode_significance_8x8_x86()
  x86: h264: remove hardcoded esi in decode_significance[_8x8]_x86()
  x86: h264: remove hardcoded edx in decode_significance[_8x8]_x86()
  x86: h264: remove hardcoded eax in decode_significance[_8x8]_x86()
  x86: cabac: change 'a' constraint to 'r' in get_cabac_inline()
  x86: cabac: remove hardcoded esi in get_cabac_inline()
  x86: cabac: remove hardcoded edx in get_cabac_inline()
  x86: cabac: remove unused macro parameter
  x86: cabac: remove hardcoded ebx in inline asm
  x86: cabac: remove hardcoded struct offsets from inline asm
  cabac: remove inline asm under #if 0
  cabac: remove BRANCHLESS_CABAC_DECODER switch
  cabac: remove #if 0 cascade under never-set #ifdef ARCH_X86_DISABLED
  document libswscale bump
  error_resilience: skip last-MV predictor step if MVs are not available.
  error_resilience: actually add counter when adding a MV predictor.
  ...

Conflicts:
	Changelog
	libavcodec/error_resilience.c
	libavfilter/defaults.c
	libavfilter/vf_drawtext.c
	libswscale/swscale.h
	tests/ref/vsynth1/error
	tests/ref/vsynth2/error

Merged-by: Michael Niedermayer <michaelni@gmx.at>

1 files changed, 26 insertions, 335 deletions

diff --git a/libavcodec/cabac.h b/libavcodec/cabac.h
index 0af77e426f..1b2d53f3d5 100644
--- a/libavcodec/cabac.h
+++ b/libavcodec/cabac.h
@@ -27,16 +27,15 @@
 #ifndef AVCODEC_CABAC_H
 #define AVCODEC_CABAC_H
 
+#include <stddef.h>
+
 #include "put_bits.h"
 
 //#undef NDEBUG
 #include <assert.h>
-#include "libavutil/x86_cpu.h"
 
 #define CABAC_BITS 16
 #define CABAC_MASK ((1<<CABAC_BITS)-1)
-#define BRANCHLESS_CABAC_DECODER 1
-//#define ARCH_X86_DISABLED 1
 
 typedef struct CABACContext{
     int low;
@@ -57,6 +56,9 @@ extern uint8_t ff_h264_mps_state[2*64];     ///< transIdxMPS
 extern uint8_t ff_h264_lps_state[2*64];     ///< transIdxLPS
 extern const uint8_t ff_h264_norm_shift[512];
 
+#if ARCH_X86
+#   include "x86/cabac.h"
+#endif
 
 void ff_init_cabac_encoder(CABACContext *c, uint8_t *buf, int buf_size);
 void ff_init_cabac_decoder(CABACContext *c, const uint8_t *buf, int buf_size);
@@ -270,26 +272,6 @@ static void refill(CABACContext *c){
     c->bytestream+= CABAC_BITS/8;
 }
 
-#if ! ( ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS) )
-static void refill2(CABACContext *c){
-    int i, x;
-
-    x= c->low ^ (c->low-1);
-    i= 7 - ff_h264_norm_shift[x>>(CABAC_BITS-1)];
-
-    x= -CABAC_MASK;
-
-#if CABAC_BITS == 16
-        x+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);
-#else
-        x+= c->bytestream[0]<<1;
-#endif
-
-    c->low += x<<i;
-    c->bytestream+= CABAC_BITS/8;
-}
-#endif
-
 static inline void renorm_cabac_decoder(CABACContext *c){
     while(c->range < 0x100){
         c->range+= c->range;
@@ -300,267 +282,38 @@ static inline void renorm_cabac_decoder(CABACContext *c){
 }
 
 static inline void renorm_cabac_decoder_once(CABACContext *c){
-#ifdef ARCH_X86_DISABLED
-    int temp;
-#if 0
-    //P3:683    athlon:475
-    __asm__(
-        "lea -0x100(%0), %2         \n\t"
-        "shr $31, %2                \n\t"  //FIXME 31->63 for x86-64
-        "shl %%cl, %0               \n\t"
-        "shl %%cl, %1               \n\t"
-        : "+r"(c->range), "+r"(c->low), "+c"(temp)
-    );
-#elif 0
-    //P3:680    athlon:474
-    __asm__(
-        "cmp $0x100, %0             \n\t"
-        "setb %%cl                  \n\t"  //FIXME 31->63 for x86-64
-        "shl %%cl, %0               \n\t"
-        "shl %%cl, %1               \n\t"
-        : "+r"(c->range), "+r"(c->low), "+c"(temp)
-    );
-#elif 1
-    int temp2;
-    //P3:665    athlon:517
-    __asm__(
-        "lea -0x100(%0), %%eax      \n\t"
-        "cltd                       \n\t"
-        "mov %0, %%eax              \n\t"
-        "and %%edx, %0              \n\t"
-        "and %1, %%edx              \n\t"
-        "add %%eax, %0              \n\t"
-        "add %%edx, %1              \n\t"
-        : "+r"(c->range), "+r"(c->low), "+a"(temp), "+d"(temp2)
-    );
-#elif 0
-    int temp2;
-    //P3:673    athlon:509
-    __asm__(
-        "cmp $0x100, %0             \n\t"
-        "sbb %%edx, %%edx           \n\t"
-        "mov %0, %%eax              \n\t"
-        "and %%edx, %0              \n\t"
-        "and %1, %%edx              \n\t"
-        "add %%eax, %0              \n\t"
-        "add %%edx, %1              \n\t"
-        : "+r"(c->range), "+r"(c->low), "+a"(temp), "+d"(temp2)
-    );
-#else
-    int temp2;
-    //P3:677    athlon:511
-    __asm__(
-        "cmp $0x100, %0             \n\t"
-        "lea (%0, %0), %%eax        \n\t"
-        "lea (%1, %1), %%edx        \n\t"
-        "cmovb %%eax, %0            \n\t"
-        "cmovb %%edx, %1            \n\t"
-        : "+r"(c->range), "+r"(c->low), "+a"(temp), "+d"(temp2)
-    );
-#endif
-#else
-    //P3:675    athlon:476
     int shift= (uint32_t)(c->range - 0x100)>>31;
     c->range<<= shift;
     c->low  <<= shift;
-#endif
     if(!(c->low & CABAC_MASK))
         refill(c);
 }
 
-static av_always_inline int get_cabac_inline(CABACContext *c, uint8_t * const state){
-    //FIXME gcc generates duplicate load/stores for c->low and c->range
-#define LOW          "0"
-#define RANGE        "4"
-#if ARCH_X86_64
-#define BYTESTART   "16"
-#define BYTE        "24"
-#define BYTEEND     "32"
-#else
-#define BYTESTART   "12"
-#define BYTE        "16"
-#define BYTEEND     "20"
-#endif
-#if ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS)
-    int bit;
-
-#ifndef BRANCHLESS_CABAC_DECODER
-    __asm__ volatile(
-        "movzbl (%1), %0                        \n\t"
-        "movl "RANGE    "(%2), %%ebx            \n\t"
-        "movl "RANGE    "(%2), %%edx            \n\t"
-        "andl $0xC0, %%ebx                      \n\t"
-        "movzbl "MANGLE(ff_h264_lps_range)"(%0, %%ebx, 2), %%esi\n\t"
-        "movl "LOW      "(%2), %%ebx            \n\t"
-//eax:state ebx:low, edx:range, esi:RangeLPS
-        "subl %%esi, %%edx                      \n\t"
-        "movl %%edx, %%ecx                      \n\t"
-        "shll $17, %%ecx                        \n\t"
-        "cmpl %%ecx, %%ebx                      \n\t"
-        " ja 1f                                 \n\t"
+#ifndef get_cabac_inline
+static void refill2(CABACContext *c){
+    int i, x;
 
-#if 1
-        //athlon:4067 P3:4110
-        "lea -0x100(%%edx), %%ecx               \n\t"
-        "shr $31, %%ecx                         \n\t"
-        "shl %%cl, %%edx                        \n\t"
-        "shl %%cl, %%ebx                        \n\t"
+    x= c->low ^ (c->low-1);
+    i= 7 - ff_h264_norm_shift[x>>(CABAC_BITS-1)];
+
+    x= -CABAC_MASK;
+
+#if CABAC_BITS == 16
+        x+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);
 #else
-        //athlon:4057 P3:4130
-        "cmp $0x100, %%edx                      \n\t" //FIXME avoidable
-        "setb %%cl                              \n\t"
-        "shl %%cl, %%edx                        \n\t"
-        "shl %%cl, %%ebx                        \n\t"
+        x+= c->bytestream[0]<<1;
 #endif
-        "movzbl "MANGLE(ff_h264_mps_state)"(%0), %%ecx   \n\t"
-        "movb %%cl, (%1)                        \n\t"
-//eax:state ebx:low, edx:range, esi:RangeLPS
-        "test %%bx, %%bx                        \n\t"
-        " jnz 2f                                \n\t"
-        "mov  "BYTE     "(%2), %%"REG_S"        \n\t"
-        "subl $0xFFFF, %%ebx                    \n\t"
-        "movzwl (%%"REG_S"), %%ecx              \n\t"
-        "bswap %%ecx                            \n\t"
-        "shrl $15, %%ecx                        \n\t"
-        "add  $2, %%"REG_S"                     \n\t"
-        "addl %%ecx, %%ebx                      \n\t"
-        "mov  %%"REG_S", "BYTE    "(%2)         \n\t"
-        "jmp 2f                                 \n\t"
-        "1:                                     \n\t"
-//eax:state ebx:low, edx:range, esi:RangeLPS
-        "subl %%ecx, %%ebx                      \n\t"
-        "movl %%esi, %%edx                      \n\t"
-        "movzbl " MANGLE(ff_h264_norm_shift) "(%%esi), %%ecx   \n\t"
-        "shll %%cl, %%ebx                       \n\t"
-        "shll %%cl, %%edx                       \n\t"
-        "movzbl "MANGLE(ff_h264_lps_state)"(%0), %%ecx   \n\t"
-        "movb %%cl, (%1)                        \n\t"
-        "add  $1, %0                            \n\t"
-        "test %%bx, %%bx                        \n\t"
-        " jnz 2f                                \n\t"
-
-        "mov  "BYTE     "(%2), %%"REG_c"        \n\t"
-        "movzwl (%%"REG_c"), %%esi              \n\t"
-        "bswap %%esi                            \n\t"
-        "shrl $15, %%esi                        \n\t"
-        "subl $0xFFFF, %%esi                    \n\t"
-        "add  $2, %%"REG_c"                     \n\t"
-        "mov  %%"REG_c", "BYTE    "(%2)         \n\t"
-
-        "leal -1(%%ebx), %%ecx                  \n\t"
-        "xorl %%ebx, %%ecx                      \n\t"
-        "shrl $15, %%ecx                        \n\t"
-        "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx   \n\t"
-        "neg %%ecx                              \n\t"
-        "add $7, %%ecx                          \n\t"
-
-        "shll %%cl , %%esi                      \n\t"
-        "addl %%esi, %%ebx                      \n\t"
-        "2:                                     \n\t"
-        "movl %%edx, "RANGE    "(%2)            \n\t"
-        "movl %%ebx, "LOW      "(%2)            \n\t"
-        :"=&a"(bit) //FIXME this is fragile gcc either runs out of registers or miscompiles it (for example if "+a"(bit) or "+m"(*state) is used
-        :"r"(state), "r"(c)
-        : "%"REG_c, "%ebx", "%edx", "%"REG_S, "memory"
-    );
-    bit&=1;
-#else /* BRANCHLESS_CABAC_DECODER */
-
-
-#if HAVE_FAST_CMOV
-#define BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
-        "mov    "tmp"       , %%ecx                                     \n\t"\
-        "shl    $17         , "tmp"                                     \n\t"\
-        "cmp    "low"       , "tmp"                                     \n\t"\
-        "cmova  %%ecx       , "range"                                   \n\t"\
-        "sbb    %%ecx       , %%ecx                                     \n\t"\
-        "and    %%ecx       , "tmp"                                     \n\t"\
-        "sub    "tmp"       , "low"                                     \n\t"\
-        "xor    %%ecx       , "ret"                                     \n\t"
-#else /* HAVE_FAST_CMOV */
-#define BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
-        "mov    "tmp"       , %%ecx                                     \n\t"\
-        "shl    $17         , "tmp"                                     \n\t"\
-        "sub    "low"       , "tmp"                                     \n\t"\
-        "sar    $31         , "tmp"                                     \n\t" /*lps_mask*/\
-        "sub    %%ecx       , "range"                                   \n\t" /*RangeLPS - range*/\
-        "and    "tmp"       , "range"                                   \n\t" /*(RangeLPS - range)&lps_mask*/\
-        "add    %%ecx       , "range"                                   \n\t" /*new range*/\
-        "shl    $17         , %%ecx                                     \n\t"\
-        "and    "tmp"       , %%ecx                                     \n\t"\
-        "sub    %%ecx       , "low"                                     \n\t"\
-        "xor    "tmp"       , "ret"                                     \n\t"
-#endif /* HAVE_FAST_CMOV */
-
-
-#define BRANCHLESS_GET_CABAC(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
-        "movzbl "statep"    , "ret"                                     \n\t"\
-        "mov    "range"     , "tmp"                                     \n\t"\
-        "and    $0xC0       , "range"                                   \n\t"\
-        "movzbl "MANGLE(ff_h264_lps_range)"("ret", "range", 2), "range" \n\t"\
-        "sub    "range"     , "tmp"                                     \n\t"\
-        BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
-        "movzbl " MANGLE(ff_h264_norm_shift) "("range"), %%ecx          \n\t"\
-        "shl    %%cl        , "range"                                   \n\t"\
-        "movzbl "MANGLE(ff_h264_mlps_state)"+128("ret"), "tmp"          \n\t"\
-        "mov    "tmpbyte"   , "statep"                                  \n\t"\
-        "shl    %%cl        , "low"                                     \n\t"\
-        "test   "lowword"   , "lowword"                                 \n\t"\
-        " jnz   1f                                                      \n\t"\
-        "mov "BYTE"("cabac"), %%"REG_c"                                 \n\t"\
-        "movzwl (%%"REG_c")     , "tmp"                                 \n\t"\
-        "bswap  "tmp"                                                   \n\t"\
-        "shr    $15         , "tmp"                                     \n\t"\
-        "sub    $0xFFFF     , "tmp"                                     \n\t"\
-        "add    $2          , %%"REG_c"                                 \n\t"\
-        "mov    %%"REG_c"   , "BYTE    "("cabac")                       \n\t"\
-        "lea    -1("low")   , %%ecx                                     \n\t"\
-        "xor    "low"       , %%ecx                                     \n\t"\
-        "shr    $15         , %%ecx                                     \n\t"\
-        "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx            \n\t"\
-        "neg    %%ecx                                                   \n\t"\
-        "add    $7          , %%ecx                                     \n\t"\
-        "shl    %%cl        , "tmp"                                     \n\t"\
-        "add    "tmp"       , "low"                                     \n\t"\
-        "1:                                                             \n\t"
-
-    __asm__ volatile(
-        "movl "RANGE    "(%2), %%esi            \n\t"
-        "movl "LOW      "(%2), %%ebx            \n\t"
-        BRANCHLESS_GET_CABAC("%0", "%2", "(%1)", "%%ebx", "%%bx", "%%esi", "%%edx", "%%dl")
-        "movl %%esi, "RANGE    "(%2)            \n\t"
-        "movl %%ebx, "LOW      "(%2)            \n\t"
-
-        :"=&a"(bit)
-        :"r"(state), "r"(c)
-        : "%"REG_c, "%ebx", "%edx", "%esi", "memory"
-    );
-    bit&=1;
-#endif /* BRANCHLESS_CABAC_DECODER */
-#else /* ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS) */
+
+    c->low += x<<i;
+    c->bytestream+= CABAC_BITS/8;
+}
+
+static av_always_inline int get_cabac_inline(CABACContext *c, uint8_t * const state){
     int s = *state;
     int RangeLPS= ff_h264_lps_range[2*(c->range&0xC0) + s];
-    int bit, lps_mask av_unused;
+    int bit, lps_mask;
 
     c->range -= RangeLPS;
-#ifndef BRANCHLESS_CABAC_DECODER
-    if(c->low < (c->range<<(CABAC_BITS+1))){
-        bit= s&1;
-        *state= ff_h264_mps_state[s];
-        renorm_cabac_decoder_once(c);
-    }else{
-        bit= ff_h264_norm_shift[RangeLPS];
-        c->low -= (c->range<<(CABAC_BITS+1));
-        *state= ff_h264_lps_state[s];
-        c->range = RangeLPS<<bit;
-        c->low <<= bit;
-        bit= (s&1)^1;
-
-        if(!(c->low & CABAC_MASK)){
-            refill2(c);
-        }
-    }
-#else /* BRANCHLESS_CABAC_DECODER */
     lps_mask= ((c->range<<(CABAC_BITS+1)) - c->low)>>31;
 
     c->low -= (c->range<<(CABAC_BITS+1)) & lps_mask;
@@ -575,10 +328,9 @@ static av_always_inline int get_cabac_inline(CABACContext *c, uint8_t * const st
     c->low  <<= lps_mask;
     if(!(c->low & CABAC_MASK))
         refill2(c);
-#endif /* BRANCHLESS_CABAC_DECODER */
-#endif /* ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS) */
     return bit;
 }
+#endif
 
 static int av_noinline av_unused get_cabac_noinline(CABACContext *c, uint8_t * const state){
     return get_cabac_inline(c,state);
@@ -589,36 +341,6 @@ static int av_unused get_cabac(CABACContext *c, uint8_t * const state){
 }
 
 static int av_unused get_cabac_bypass(CABACContext *c){
-#if 0 //not faster
-    int bit;
-    __asm__ volatile(
-        "movl "RANGE    "(%1), %%ebx            \n\t"
-        "movl "LOW      "(%1), %%eax            \n\t"
-        "shl $17, %%ebx                         \n\t"
-        "add %%eax, %%eax                       \n\t"
-        "sub %%ebx, %%eax                       \n\t"
-        "cltd                                   \n\t"
-        "and %%edx, %%ebx                       \n\t"
-        "add %%ebx, %%eax                       \n\t"
-        "test %%ax, %%ax                        \n\t"
-        " jnz 1f                                \n\t"
-        "movl "BYTE     "(%1), %%"REG_b"        \n\t"
-        "subl $0xFFFF, %%eax                    \n\t"
-        "movzwl (%%"REG_b"), %%ecx              \n\t"
-        "bswap %%ecx                            \n\t"
-        "shrl $15, %%ecx                        \n\t"
-        "addl $2, %%"REG_b"                     \n\t"
-        "addl %%ecx, %%eax                      \n\t"
-        "movl %%"REG_b", "BYTE     "(%1)        \n\t"
-        "1:                                     \n\t"
-        "movl %%eax, "LOW      "(%1)            \n\t"
-
-        :"=&d"(bit)
-        :"r"(c)
-        : "%eax", "%"REG_b, "%ecx", "memory"
-    );
-    return bit+1;
-#else
     int range;
     c->low += c->low;
 
@@ -632,42 +354,11 @@ static int av_unused get_cabac_bypass(CABACContext *c){
         c->low -= range;
         return 1;
     }
-#endif
 }
 
 
+#ifndef get_cabac_bypass_sign
 static av_always_inline int get_cabac_bypass_sign(CABACContext *c, int val){
-#if ARCH_X86 && HAVE_EBX_AVAILABLE
-    __asm__ volatile(
-        "movl "RANGE    "(%1), %%ebx            \n\t"
-        "movl "LOW      "(%1), %%eax            \n\t"
-        "shl $17, %%ebx                         \n\t"
-        "add %%eax, %%eax                       \n\t"
-        "sub %%ebx, %%eax                       \n\t"
-        "cltd                                   \n\t"
-        "and %%edx, %%ebx                       \n\t"
-        "add %%ebx, %%eax                       \n\t"
-        "xor %%edx, %%ecx                       \n\t"
-        "sub %%edx, %%ecx                       \n\t"
-        "test %%ax, %%ax                        \n\t"
-        " jnz 1f                                \n\t"
-        "mov  "BYTE     "(%1), %%"REG_b"        \n\t"
-        "subl $0xFFFF, %%eax                    \n\t"
-        "movzwl (%%"REG_b"), %%edx              \n\t"
-        "bswap %%edx                            \n\t"
-        "shrl $15, %%edx                        \n\t"
-        "add  $2, %%"REG_b"                     \n\t"
-        "addl %%edx, %%eax                      \n\t"
-        "mov  %%"REG_b", "BYTE     "(%1)        \n\t"
-        "1:                                     \n\t"
-        "movl %%eax, "LOW      "(%1)            \n\t"
-
-        :"+c"(val)
-        :"r"(c)
-        : "%eax", "%"REG_b, "%edx", "memory"
-    );
-    return val;
-#else
     int range, mask;
     c->low += c->low;
 
@@ -680,8 +371,8 @@ static av_always_inline int get_cabac_bypass_sign(CABACContext *c, int val){
     range &= mask;
     c->low += range;
     return (val^mask)-mask;
-#endif
 }
+#endif
 
 /**
  *