falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
Browse Source

x86: cabac: remove hardcoded struct offsets from inline asm 

Signed-off-by: Mans Rullgard <mans@mansr.com>
tags/n0.8
Mans Rullgard 14 years ago
parent
34ee43fc0f
commit
6b712acc0e
2 changed files with 41 additions and 40 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +21
    -26
      libavcodec/cabac.h
  2. +20
    -14
      libavcodec/x86/h264_i386.h

+ 21
- 26
libavcodec/cabac.h View File

@@ -27,6 +27,8 @@
#ifndef AVCODEC_CABAC_H
#define AVCODEC_CABAC_H

#include <stddef.h>

#include "put_bits.h"

//#undef NDEBUG
@@ -307,17 +309,6 @@ static inline void renorm_cabac_decoder_once(CABACContext *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;

@@ -347,7 +338,7 @@ static av_always_inline int get_cabac_inline(CABACContext *c, uint8_t * const st
#endif /* HAVE_FAST_CMOV */


#define BRANCHLESS_GET_CABAC(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
#define BRANCHLESS_GET_CABAC(ret, cabac, statep, low, lowword, range, tmp, tmpbyte, byte) \
"movzbl "statep" , "ret" \n\t"\
"mov "range" , "tmp" \n\t"\
"and $0xC0 , "range" \n\t"\
@@ -361,13 +352,13 @@ static av_always_inline int get_cabac_inline(CABACContext *c, uint8_t * const st
"shl %%cl , "low" \n\t"\
"test "lowword" , "lowword" \n\t"\
" jnz 1f \n\t"\
"mov "BYTE"("cabac"), %%"REG_c" \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"\
"mov %%"REG_c" , "byte "("cabac") \n\t"\
"lea -1("low") , %%ecx \n\t"\
"xor "low" , %%ecx \n\t"\
"shr $15 , %%ecx \n\t"\
@@ -379,14 +370,16 @@ static av_always_inline int get_cabac_inline(CABACContext *c, uint8_t * const st
"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"
"movl %a3(%2), %%esi \n\t"
"movl %a4(%2), %%ebx \n\t"
BRANCHLESS_GET_CABAC("%0", "%2", "(%1)", "%%ebx", "%%bx", "%%esi", "%%edx", "%%dl", "%a5")
"movl %%esi, %a3(%2) \n\t"
"movl %%ebx, %a4(%2) \n\t"

:"=&a"(bit)
:"r"(state), "r"(c)
:"r"(state), "r"(c),
"i"(offsetof(CABACContext, range)), "i"(offsetof(CABACContext, low)),
"i"(offsetof(CABACContext, bytestream))
: "%"REG_c, "%ebx", "%edx", "%esi", "memory"
);
bit&=1;
@@ -442,8 +435,8 @@ static int av_unused get_cabac_bypass(CABACContext *c){
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"
"movl %a2(%1), %%ebx \n\t"
"movl %a3(%1), %%eax \n\t"
"shl $17, %%ebx \n\t"
"add %%eax, %%eax \n\t"
"sub %%ebx, %%eax \n\t"
@@ -454,19 +447,21 @@ static av_always_inline int get_cabac_bypass_sign(CABACContext *c, int val){
"sub %%edx, %%ecx \n\t"
"test %%ax, %%ax \n\t"
" jnz 1f \n\t"
"mov "BYTE "(%1), %%"REG_b" \n\t"
"mov %a4(%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"
"mov %%"REG_b", %a4(%1) \n\t"
"1: \n\t"
"movl %%eax, "LOW "(%1) \n\t"
"movl %%eax, %a3(%1) \n\t"

:"+c"(val)
:"r"(c)
:"r"(c),
"i"(offsetof(CABACContext, range)), "i"(offsetof(CABACContext, low)),
"i"(offsetof(CABACContext, bytestream))
: "%eax", "%"REG_b, "%edx", "memory"
);
return val;


+ 20
- 14
libavcodec/x86/h264_i386.h View File

@@ -29,6 +29,8 @@
#ifndef AVCODEC_X86_H264_I386_H
#define AVCODEC_X86_H264_I386_H

#include <stddef.h>

#include "libavcodec/cabac.h"

//FIXME use some macros to avoid duplicating get_cabac (cannot be done yet
@@ -42,20 +44,20 @@ static int decode_significance_x86(CABACContext *c, int max_coeff,
int minusindex= 4-(int)index;
int coeff_count;
__asm__ volatile(
"movl "RANGE "(%3), %%esi \n\t"
"movl "LOW "(%3), %%ebx \n\t"
"movl %a8(%3), %%esi \n\t"
"movl %a9(%3), %%ebx \n\t"

"2: \n\t"

BRANCHLESS_GET_CABAC("%%edx", "%3", "(%1)", "%%ebx",
"%%bx", "%%esi", "%%eax", "%%al")
"%%bx", "%%esi", "%%eax", "%%al", "%a10")

"test $1, %%edx \n\t"
" jz 3f \n\t"
"add %7, %1 \n\t"

BRANCHLESS_GET_CABAC("%%edx", "%3", "(%1)", "%%ebx",
"%%bx", "%%esi", "%%eax", "%%al")
"%%bx", "%%esi", "%%eax", "%%al", "%a10")

"sub %7, %1 \n\t"
"mov %2, %%"REG_a" \n\t"
@@ -81,10 +83,12 @@ static int decode_significance_x86(CABACContext *c, int max_coeff,
"add %6, %%eax \n\t"
"shr $2, %%eax \n\t"

"movl %%esi, "RANGE "(%3) \n\t"
"movl %%ebx, "LOW "(%3) \n\t"
"movl %%esi, %a8(%3) \n\t"
"movl %%ebx, %a9(%3) \n\t"
:"=&a"(coeff_count), "+r"(significant_coeff_ctx_base), "+m"(index)
:"r"(c), "m"(minusstart), "m"(end), "m"(minusindex), "m"(last_off)
:"r"(c), "m"(minusstart), "m"(end), "m"(minusindex), "m"(last_off),
"i"(offsetof(CABACContext, range)), "i"(offsetof(CABACContext, low)),
"i"(offsetof(CABACContext, bytestream))
: "%"REG_c, "%ebx", "%edx", "%esi", "memory"
);
return coeff_count;
@@ -97,8 +101,8 @@ static int decode_significance_8x8_x86(CABACContext *c,
int coeff_count;
x86_reg last=0;
__asm__ volatile(
"movl "RANGE "(%3), %%esi \n\t"
"movl "LOW "(%3), %%ebx \n\t"
"movl %a8(%3), %%esi \n\t"
"movl %a9(%3), %%ebx \n\t"

"mov %1, %%"REG_D" \n\t"
"2: \n\t"
@@ -108,7 +112,7 @@ static int decode_significance_8x8_x86(CABACContext *c,
"add %5, %%"REG_D" \n\t"

BRANCHLESS_GET_CABAC("%%edx", "%3", "(%%"REG_D")", "%%ebx",
"%%bx", "%%esi", "%%eax", "%%al")
"%%bx", "%%esi", "%%eax", "%%al", "%a10")

"mov %1, %%edi \n\t"
"test $1, %%edx \n\t"
@@ -119,7 +123,7 @@ static int decode_significance_8x8_x86(CABACContext *c,
"add %7, %%"REG_D" \n\t"

BRANCHLESS_GET_CABAC("%%edx", "%3", "(%%"REG_D")", "%%ebx",
"%%bx", "%%esi", "%%eax", "%%al")
"%%bx", "%%esi", "%%eax", "%%al", "%a10")

"mov %2, %%"REG_a" \n\t"
"mov %1, %%edi \n\t"
@@ -142,10 +146,12 @@ static int decode_significance_8x8_x86(CABACContext *c,
"addl %4, %%eax \n\t"
"shr $2, %%eax \n\t"

"movl %%esi, "RANGE "(%3) \n\t"
"movl %%ebx, "LOW "(%3) \n\t"
"movl %%esi, %a8(%3) \n\t"
"movl %%ebx, %a9(%3) \n\t"
:"=&a"(coeff_count),"+m"(last), "+m"(index)
:"r"(c), "m"(minusindex), "m"(significant_coeff_ctx_base), "m"(sig_off), "m"(last_off)
:"r"(c), "m"(minusindex), "m"(significant_coeff_ctx_base), "m"(sig_off), "m"(last_off),
"i"(offsetof(CABACContext, range)), "i"(offsetof(CABACContext, low)),
"i"(offsetof(CABACContext, bytestream))
: "%"REG_c, "%ebx", "%edx", "%esi", "%"REG_D, "memory"
);
return coeff_count;


Write Preview
Loading…
Cancel
Save