#define CABAC_BITS 16
#define CABAC_MASK ((1<<CABAC_BITS)-1)
+#define BRANCHLESS_CABAD 1
typedef struct CABACContext{
int low;
#ifdef STRICT_LIMITS
int symCount;
#endif
- uint8_t lps_range[2*66][4]; ///< rangeTabLPS
- uint8_t lps_state[2*65]; ///< transIdxLPS
- uint8_t mps_state[2*65]; ///< transIdxMPS
+ uint8_t lps_range[2*65][4]; ///< rangeTabLPS
+ uint8_t lps_state[2*64]; ///< transIdxLPS
+ uint8_t mps_state[2*64]; ///< transIdxMPS
const uint8_t *bytestream_start;
const uint8_t *bytestream;
const uint8_t *bytestream_end;
extern const uint8_t ff_h264_lps_range[64][4];
extern const uint8_t ff_h264_mps_state[64];
extern const uint8_t ff_h264_lps_state[64];
-extern const uint8_t ff_h264_norm_shift[256];
+extern const uint8_t ff_h264_norm_shift[128];
void ff_init_cabac_encoder(CABACContext *c, uint8_t *buf, int buf_size);
}
static void refill(CABACContext *c){
- if(c->bytestream <= c->bytestream_end)
#if CABAC_BITS == 16
c->low+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);
#else
c->bytestream+= CABAC_BITS/8;
}
-#if 1 /* all use commented */
static void refill2(CABACContext *c){
int i, x;
x= c->low ^ (c->low-1);
- i= 9 - ff_h264_norm_shift[x>>(CABAC_BITS+1)];
+ i= 7 - ff_h264_norm_shift[x>>(CABAC_BITS+1)];
x= -CABAC_MASK;
- if(c->bytestream <= c->bytestream_end)
#if CABAC_BITS == 16
x+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);
#else
c->low += x<<i;
c->bytestream+= CABAC_BITS/8;
}
-#endif
static inline void renorm_cabac_decoder(CABACContext *c){
while(c->range < (0x200 << CABAC_BITS)){
static int get_cabac(CABACContext *c, uint8_t * const state){
//FIXME gcc generates duplicate load/stores for c->low and c->range
-//START_TIMER
+#ifdef ARCH_X86
+ int bit;
+
+#define LOW "0"
+#define RANGE "4"
+#define LPS_RANGE "12"
+#define LPS_STATE "12+2*65*4"
+#define MPS_STATE "12+2*65*4+2*64"
+#define BYTESTART "12+2*65*4+4*64"
+#define BYTE "16+2*65*4+4*64"
+#define BYTEEND "20+2*65*4+4*64"
+#ifndef BRANCHLESS_CABAD
+ asm volatile(
+ "movzbl (%1), %%eax \n\t"
+ "movl "RANGE "(%2), %%ebx \n\t"
+ "movl "RANGE "(%2), %%edx \n\t"
+ "shrl $23, %%ebx \n\t"
+ "leal "LPS_RANGE"(%2, %%eax, 4), %%esi \n\t"
+ "movzbl (%%ebx, %%esi), %%esi \n\t"
+ "shll $17, %%esi \n\t"
+ "movl "LOW "(%2), %%ebx \n\t"
+//eax:state ebx:low, edx:range, esi:RangeLPS
+ "subl %%esi, %%edx \n\t"
+ "cmpl %%edx, %%ebx \n\t"
+ " ja 1f \n\t"
+ "cmp $0x2000000, %%edx \n\t" //FIXME avoidable
+ "setb %%cl \n\t"
+ "shl %%cl, %%edx \n\t"
+ "shl %%cl, %%ebx \n\t"
+ "movzbl "MPS_STATE"(%2, %%eax), %%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"
+ "movl "BYTE "(%2), %%esi \n\t"
+ "subl $0xFFFF, %%ebx \n\t"
+ "movzwl (%%esi), %%ecx \n\t"
+ "bswap %%ecx \n\t"
+ "shrl $15, %%ecx \n\t"
+ "addl $2, %%esi \n\t"
+ "addl %%ecx, %%ebx \n\t"
+ "movl %%esi, "BYTE "(%2) \n\t"
+ "jmp 2f \n\t"
+ "1: \n\t"
+//eax:state ebx:low, edx:range, esi:RangeLPS
+ "subl %%edx, %%ebx \n\t"
+ "movl %%esi, %%edx \n\t"
+ "shr $19, %%esi \n\t"
+ "movzbl " MANGLE(ff_h264_norm_shift) "(%%esi), %%ecx \n\t"
+ "shll %%cl, %%ebx \n\t"
+ "shll %%cl, %%edx \n\t"
+ "movzbl "LPS_STATE"(%2, %%eax), %%ecx \n\t"
+ "movb %%cl, (%1) \n\t"
+ "addl $1, %%eax \n\t"
+ "test %%bx, %%bx \n\t"
+ " jnz 2f \n\t"
+
+ "movl "BYTE "(%2), %%ecx \n\t"
+ "movzwl (%%ecx), %%esi \n\t"
+ "bswap %%esi \n\t"
+ "shrl $15, %%esi \n\t"
+ "subl $0xFFFF, %%esi \n\t"
+ "addl $2, %%ecx \n\t"
+ "movl %%ecx, "BYTE "(%2) \n\t"
+
+ "leal -1(%%ebx), %%ecx \n\t"
+ "xorl %%ebx, %%ecx \n\t"
+ "shrl $17, %%ecx \n\t"
+ "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx \n\t"
+ "neg %%cl \n\t"
+ "add $7, %%cl \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"
+ "andl $1, %%eax \n\t"
+
+ :"=&a"(bit) //FIXME this is fragile gcc either runs out of registers or misscompiles it (for example if "+a"(bit) or "+m"(*state) is used
+ :"r"(state), "r"(c)
+ : "%ecx", "%ebx", "%edx", "%esi"
+ );
+#else
+ asm volatile(
+ "movzbl (%1), %%eax \n\t"
+ "movl "RANGE "(%2), %%ebx \n\t"
+ "movl "RANGE "(%2), %%edx \n\t"
+ "shrl $23, %%ebx \n\t"
+ "leal "LPS_RANGE"(%2, %%eax, 4), %%esi \n\t"
+ "movzbl (%%ebx, %%esi), %%esi \n\t"
+ "shll $17, %%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"
+ "subl %%ebx, %%edx \n\t"
+ "sarl $31, %%edx \n\t" //lps_mask
+ "subl %%ecx, %%esi \n\t" //RangeLPS - range
+ "andl %%edx, %%esi \n\t" //(RangeLPS - range)&lps_mask
+ "addl %%ecx, %%esi \n\t" //new range
+ "andl %%edx, %%ecx \n\t"
+ "subl %%ecx, %%ebx \n\t"
+
+//eax:state ebx:low edx:mask esi:range
+ "xorl %%edx, %%eax \n\t"
+ "movzbl "MPS_STATE"(%2, %%eax), %%ecx \n\t"
+ "movb %%cl, (%1) \n\t"
+
+ "movl %%esi, %%edx \n\t"
+//eax:bit ebx:low edx:range esi:range
+
+ "shr $19, %%esi \n\t"
+ "movzbl " MANGLE(ff_h264_norm_shift) "(%%esi), %%ecx \n\t"
+ "shll %%cl, %%ebx \n\t"
+ "shll %%cl, %%edx \n\t"
+ "test %%bx, %%bx \n\t"
+ " jnz 1f \n\t"
+
+ "movl "BYTE "(%2), %%ecx \n\t"
+ "movzwl (%%ecx), %%esi \n\t"
+ "bswap %%esi \n\t"
+ "shrl $15, %%esi \n\t"
+ "subl $0xFFFF, %%esi \n\t"
+ "addl $2, %%ecx \n\t"
+ "movl %%ecx, "BYTE "(%2) \n\t"
+
+ "leal -1(%%ebx), %%ecx \n\t"
+ "xorl %%ebx, %%ecx \n\t"
+ "shrl $17, %%ecx \n\t"
+ "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx \n\t"
+ "neg %%cl \n\t"
+ "add $7, %%cl \n\t"
+
+ "shll %%cl , %%esi \n\t"
+ "addl %%esi, %%ebx \n\t"
+ "1: \n\t"
+ "movl %%edx, "RANGE "(%2) \n\t"
+ "movl %%ebx, "LOW "(%2) \n\t"
+ "andl $1, %%eax \n\t"
+ :"=&a"(bit)
+ :"r"(state), "r"(c)
+ : "%ecx", "%ebx", "%edx", "%esi"
+ );
+#endif
+#else
int s = *state;
int RangeLPS= c->lps_range[s][c->range>>(CABAC_BITS+7)]<<(CABAC_BITS+1);
int bit, lps_mask attribute_unused;
c->range -= RangeLPS;
-#if 1
+#ifndef BRANCHLESS_CABAD
if(c->low < c->range){
bit= s&1;
-#ifdef ARCH_X86
- //P3:627
-asm(
- "addb $2, %b0 \n\t"
- " js 1f \n\t"
- "movb %b0, %1 \n\t"
- "1: \n\t"
- : "+q"(s), "=m"(*state)
-);
-#else
- *state= c->mps_state[s]; //P3:655
-/* if(s<126) //P3:657
- *state= s+2;*/
- s+=2; //P3:631
- if(s<128)
- *state= s;
-#endif
+ *state= c->mps_state[s];
renorm_cabac_decoder_once(c);
}else{
- bit= ff_h264_norm_shift[RangeLPS>>17];
+ bit= ff_h264_norm_shift[RangeLPS>>19];
c->low -= c->range;
*state= c->lps_state[s];
-// c->range = RangeLPS;
-// renorm_cabac_decoder(c);
c->range = RangeLPS<<bit;
c->low <<= bit;
bit= (s&1)^1;
c->low -= c->range & lps_mask;
c->range += (RangeLPS - c->range) & lps_mask;
- bit= (s^lps_mask)&1;
- *state= c->mps_state[s - (128&lps_mask)];
+ s^=lps_mask;
+ *state= c->mps_state[s];
+ bit= s&1;
- lps_mask= ff_h264_norm_shift[c->range>>(CABAC_BITS+2)];
+ lps_mask= ff_h264_norm_shift[c->range>>(CABAC_BITS+3)];
c->range<<= lps_mask;
c->low <<= lps_mask;
if(!(c->low & CABAC_MASK))
refill2(c);
#endif
-//STOP_TIMER("get_cabac")
+#endif
return bit;
}
projects / ffmpeg / blobdiff