| 2143d69b |
/*
* Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
* |
| d814a839 |
* This file is part of FFmpeg. |
| 2143d69b |
* |
| d814a839 |
* FFmpeg is free software; you can redistribute it and/or |
| 2143d69b |
* 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.
* |
| d814a839 |
* FFmpeg is distributed in the hope that it will be useful, |
| 2143d69b |
* 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 |
| d814a839 |
* License along with FFmpeg; if not, write to the Free Software |
| 2143d69b |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#ifndef AVCODEC_X86_CABAC_H
#define AVCODEC_X86_CABAC_H
#include "libavcodec/cabac.h"
#include "libavutil/attributes.h" |
| 3741aa37 |
#include "libavutil/macros.h" |
| c318626c |
#include "libavutil/x86/asm.h" |
| 2143d69b |
#include "config.h"
|
| 32de2805 |
#if (defined(__i386) && defined(__clang__) && (__clang_major__<2 || (__clang_major__==2 && __clang_minor__<10)))\ |
| d1e6e5c8 |
|| ( !defined(__clang__) && defined(__llvm__) && __GNUC__==4 && __GNUC_MINOR__==2 && __GNUC_PATCHLEVEL__<=1)\
|| (defined(__INTEL_COMPILER) && defined(_MSC_VER)) |
| 32de2805 |
# define BROKEN_COMPILER 1
#else
# define BROKEN_COMPILER 0
#endif
|
| 8123e090 |
#if HAVE_INLINE_ASM
|
| fa6fa216 |
#ifndef UNCHECKED_BITSTREAM_READER
#define UNCHECKED_BITSTREAM_READER !CONFIG_SAFE_BITSTREAM_READER
#endif
#if UNCHECKED_BITSTREAM_READER
#define END_CHECK(end) ""
#else
#define END_CHECK(end) \
"cmp "end" , %%"REG_c" \n\t"\
"jge 1f \n\t"
#endif
|
| 82c71913 |
#ifdef BROKEN_RELOCATIONS
#define TABLES_ARG , "r"(tables)
#if HAVE_FAST_CMOV
#define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \
"cmp "low" , "tmp" \n\t"\
"cmova %%ecx , "range" \n\t"\
"sbb %%rcx , %%rcx \n\t"\
"and %%ecx , "tmp" \n\t"\
"xor %%rcx , "retq" \n\t"\
"sub "tmp" , "low" \n\t"
#else /* HAVE_FAST_CMOV */
#define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \ |
| 41ed7ab4 |
/* P4 Prescott has crappy cmov,sbb,64-bit shift so avoid them */ \ |
| 82c71913 |
"sub "low" , "tmp" \n\t"\
"sar $31 , "tmp" \n\t"\
"sub %%ecx , "range" \n\t"\
"and "tmp" , "range" \n\t"\
"add %%ecx , "range" \n\t"\
"shl $17 , %%ecx \n\t"\
"and "tmp" , %%ecx \n\t"\
"sub %%ecx , "low" \n\t"\
"xor "tmp" , "ret" \n\t"\
"movslq "ret" , "retq" \n\t"
#endif /* HAVE_FAST_CMOV */
#define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, norm_off, lps_off, mlps_off, tables) \
"movzbl "statep" , "ret" \n\t"\
"mov "range" , "tmp" \n\t"\
"and $0xC0 , "range" \n\t"\
"lea ("ret", "range", 2), %%ecx \n\t"\
"movzbl "lps_off"("tables", %%rcx), "range" \n\t"\
"sub "range" , "tmp" \n\t"\
"mov "tmp" , %%ecx \n\t"\
"shl $17 , "tmp" \n\t"\
BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \
"movzbl "norm_off"("tables", "rangeq"), %%ecx \n\t"\
"shl %%cl , "range" \n\t"\
"movzbl "mlps_off"+128("tables", "retq"), "tmp" \n\t"\
"shl %%cl , "low" \n\t"\
"mov "tmpbyte" , "statep" \n\t"\
"test "lowword" , "lowword" \n\t"\
"jnz 2f \n\t"\
"mov "byte" , %%"REG_c" \n\t"\ |
| fa6fa216 |
END_CHECK(end)\ |
| 82c71913 |
"add"OPSIZE" $2 , "byte" \n\t"\ |
| fa6fa216 |
"1: \n\t"\ |
| 82c71913 |
"movzwl (%%"REG_c") , "tmp" \n\t"\
"lea -1("low") , %%ecx \n\t"\
"xor "low" , %%ecx \n\t"\
"shr $15 , %%ecx \n\t"\
"bswap "tmp" \n\t"\
"shr $15 , "tmp" \n\t"\
"movzbl "norm_off"("tables", %%rcx), %%ecx \n\t"\
"sub $0xFFFF , "tmp" \n\t"\
"neg %%ecx \n\t"\
"add $7 , %%ecx \n\t"\
"shl %%cl , "tmp" \n\t"\
"add "tmp" , "low" \n\t"\
"2: \n\t"
#else /* BROKEN_RELOCATIONS */ |
| 1898c2f4 |
#define TABLES_ARG NAMED_CONSTRAINTS_ARRAY_ADD(ff_h264_cabac_tables) |
| 82c71913 |
#define RIP_ARG
|
| 2143d69b |
#if HAVE_FAST_CMOV |
| 5520df6a |
#define BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)\ |
| 2143d69b |
"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"\ |
| 6c325765 |
"xor %%ecx , "ret" \n\t"\
"sub "tmp" , "low" \n\t" |
| 2143d69b |
#else /* HAVE_FAST_CMOV */ |
| 5520df6a |
#define BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)\ |
| 2143d69b |
"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 */
|
| 82c71913 |
#define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, norm_off, lps_off, mlps_off, tables) \ |
| 2143d69b |
"movzbl "statep" , "ret" \n\t"\
"mov "range" , "tmp" \n\t"\
"and $0xC0 , "range" \n\t"\ |
| 7f668cd2 |
"movzbl "MANGLE(ff_h264_cabac_tables)"+"lps_off"("ret", "range", 2), "range" \n\t"\ |
| 2143d69b |
"sub "range" , "tmp" \n\t"\ |
| 5520df6a |
BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp) \ |
| 7f668cd2 |
"movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"("range"), %%ecx \n\t"\ |
| 2143d69b |
"shl %%cl , "range" \n\t"\ |
| 7f668cd2 |
"movzbl "MANGLE(ff_h264_cabac_tables)"+"mlps_off"+128("ret"), "tmp" \n\t"\ |
| 2143d69b |
"shl %%cl , "low" \n\t"\ |
| 6c325765 |
"mov "tmpbyte" , "statep" \n\t"\ |
| 2143d69b |
"test "lowword" , "lowword" \n\t"\ |
| a9401981 |
" jnz 2f \n\t"\ |
| 2138a89e |
"mov "byte" , %%"REG_c" \n\t"\ |
| fa6fa216 |
END_CHECK(end)\ |
| 2138a89e |
"add"OPSIZE" $2 , "byte" \n\t"\ |
| fa6fa216 |
"1: \n\t"\ |
| 2143d69b |
"movzwl (%%"REG_c") , "tmp" \n\t"\
"lea -1("low") , %%ecx \n\t"\
"xor "low" , %%ecx \n\t"\
"shr $15 , %%ecx \n\t"\ |
| 6c325765 |
"bswap "tmp" \n\t"\
"shr $15 , "tmp" \n\t"\ |
| 7f668cd2 |
"movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"(%%ecx), %%ecx \n\t"\ |
| 6c325765 |
"sub $0xFFFF , "tmp" \n\t"\ |
| 2143d69b |
"neg %%ecx \n\t"\
"add $7 , %%ecx \n\t"\
"shl %%cl , "tmp" \n\t"\
"add "tmp" , "low" \n\t"\ |
| a9401981 |
"2: \n\t" |
| 2143d69b |
|
| 82c71913 |
#endif /* BROKEN_RELOCATIONS */ |
| f247f4cf |
|
| 32de2805 |
#if HAVE_7REGS && !BROKEN_COMPILER |
| 2143d69b |
#define get_cabac_inline get_cabac_inline_x86
static av_always_inline int get_cabac_inline_x86(CABACContext *c,
uint8_t *const state)
{ |
| 2138a89e |
int bit, tmp; |
| 82c71913 |
#ifdef BROKEN_RELOCATIONS
void *tables;
__asm__ volatile(
"lea "MANGLE(ff_h264_cabac_tables)", %0 \n\t"
: "=&r"(tables) |
| 1898c2f4 |
: NAMED_CONSTRAINTS_ARRAY(ff_h264_cabac_tables) |
| 82c71913 |
);
#endif |
| 2143d69b |
__asm__ volatile( |
| 82c71913 |
BRANCHLESS_GET_CABAC("%0", "%q0", "(%4)", "%1", "%w1",
"%2", "%q2", "%3", "%b3", |
| 8ec0204e |
"%c6(%5)", "%c7(%5)",
AV_STRINGIFY(H264_NORM_SHIFT_OFFSET),
AV_STRINGIFY(H264_LPS_RANGE_OFFSET),
AV_STRINGIFY(H264_MLPS_STATE_OFFSET),
"%8") |
| 66bdc585 |
: "=&r"(bit), "=&r"(c->low), "=&r"(c->range), "=&q"(tmp) |
| 951014e5 |
: "r"(state), "r"(c), |
| a9401981 |
"i"(offsetof(CABACContext, bytestream)), |
| 8ec0204e |
"i"(offsetof(CABACContext, bytestream_end))
TABLES_ARG |
| 66bdc585 |
,"1"(c->low), "2"(c->range) |
| 2143d69b |
: "%"REG_c, "memory"
);
return bit & 1;
} |
| d1e6e5c8 |
#endif /* HAVE_7REGS && !BROKEN_COMPILER */ |
| 2143d69b |
|
| 41efb8d9 |
#if !BROKEN_COMPILER |
| 2143d69b |
#define get_cabac_bypass_sign get_cabac_bypass_sign_x86
static av_always_inline int get_cabac_bypass_sign_x86(CABACContext *c, int val)
{
x86_reg tmp;
__asm__ volatile( |
| 8ec0204e |
"movl %c6(%2), %k1 \n\t"
"movl %c3(%2), %%eax \n\t" |
| 95bfa4ea |
"shl $17, %k1 \n\t"
"add %%eax, %%eax \n\t"
"sub %k1, %%eax \n\t" |
| 0f2588d7 |
"cdq \n\t" |
| 95bfa4ea |
"and %%edx, %k1 \n\t"
"add %k1, %%eax \n\t"
"xor %%edx, %%ecx \n\t"
"sub %%edx, %%ecx \n\t"
"test %%ax, %%ax \n\t"
"jnz 1f \n\t" |
| 8ec0204e |
"mov %c4(%2), %1 \n\t" |
| 95bfa4ea |
"subl $0xFFFF, %%eax \n\t"
"movzwl (%1), %%edx \n\t"
"bswap %%edx \n\t"
"shrl $15, %%edx \n\t" |
| fa6fa216 |
#if UNCHECKED_BITSTREAM_READER |
| d40ff29c |
"add $2, %1 \n\t" |
| 95bfa4ea |
"addl %%edx, %%eax \n\t" |
| 7427d1ca |
"mov %1, %c4(%2) \n\t" |
| fa6fa216 |
#else
"addl %%edx, %%eax \n\t"
"cmp %c5(%2), %1 \n\t"
"jge 1f \n\t"
"add"OPSIZE" $2, %c4(%2) \n\t"
#endif |
| 95bfa4ea |
"1: \n\t" |
| 8ec0204e |
"movl %%eax, %c3(%2) \n\t" |
| 2143d69b |
|
| db025929 |
: "+c"(val), "=&r"(tmp)
: "r"(c),
"i"(offsetof(CABACContext, low)),
"i"(offsetof(CABACContext, bytestream)), |
| a0bdcb01 |
"i"(offsetof(CABACContext, bytestream_end)), |
| db025929 |
"i"(offsetof(CABACContext, range))
: "%eax", "%edx", "memory" |
| 2143d69b |
);
return val;
}
|
| d222f6e3 |
#define get_cabac_bypass get_cabac_bypass_x86
static av_always_inline int get_cabac_bypass_x86(CABACContext *c)
{
x86_reg tmp;
int res;
__asm__ volatile(
"movl %c6(%2), %k1 \n\t"
"movl %c3(%2), %%eax \n\t"
"shl $17, %k1 \n\t"
"add %%eax, %%eax \n\t"
"sub %k1, %%eax \n\t" |
| 0f2588d7 |
"cdq \n\t" |
| d222f6e3 |
"and %%edx, %k1 \n\t"
"add %k1, %%eax \n\t"
"inc %%edx \n\t"
"test %%ax, %%ax \n\t"
"jnz 1f \n\t"
"mov %c4(%2), %1 \n\t"
"subl $0xFFFF, %%eax \n\t"
"movzwl (%1), %%ecx \n\t"
"bswap %%ecx \n\t"
"shrl $15, %%ecx \n\t"
"addl %%ecx, %%eax \n\t"
"cmp %c5(%2), %1 \n\t"
"jge 1f \n\t"
"add"OPSIZE" $2, %c4(%2) \n\t"
"1: \n\t"
"movl %%eax, %c3(%2) \n\t"
: "=&d"(res), "=&r"(tmp)
: "r"(c),
"i"(offsetof(CABACContext, low)),
"i"(offsetof(CABACContext, bytestream)),
"i"(offsetof(CABACContext, bytestream_end)),
"i"(offsetof(CABACContext, range))
: "%eax", "%ecx", "memory"
);
return res;
} |
| 66537c7e |
#endif /* !BROKEN_COMPILER */ |
| d222f6e3 |
|
| 8123e090 |
#endif /* HAVE_INLINE_ASM */ |
| 2143d69b |
#endif /* AVCODEC_X86_CABAC_H */ |