yading@10: /* yading@10: * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder yading@10: * Copyright (c) 2003 Michael Niedermayer yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: /** yading@10: * @file yading@10: * H.264 / AVC / MPEG4 part10 codec. yading@10: * non-MMX i386-specific optimizations for H.264 yading@10: * @author Michael Niedermayer yading@10: */ yading@10: yading@10: #ifndef AVCODEC_X86_H264_I386_H yading@10: #define AVCODEC_X86_H264_I386_H yading@10: yading@10: #include yading@10: yading@10: #include "libavcodec/cabac.h" yading@10: #include "cabac.h" yading@10: yading@10: #if HAVE_INLINE_ASM yading@10: yading@10: //FIXME use some macros to avoid duplicating get_cabac (cannot be done yet yading@10: //as that would make optimization work hard) yading@10: #if HAVE_7REGS yading@10: #define decode_significance decode_significance_x86 yading@10: static int decode_significance_x86(CABACContext *c, int max_coeff, yading@10: uint8_t *significant_coeff_ctx_base, yading@10: int *index, x86_reg last_off){ yading@10: void *end= significant_coeff_ctx_base + max_coeff - 1; yading@10: int minusstart= -(intptr_t)significant_coeff_ctx_base; yading@10: int minusindex= 4-(intptr_t)index; yading@10: int bit; yading@10: x86_reg coeff_count; yading@10: yading@10: #ifdef BROKEN_RELOCATIONS yading@10: void *tables; yading@10: yading@10: __asm__ volatile( yading@10: "lea "MANGLE(ff_h264_cabac_tables)", %0 \n\t" yading@10: : "=&r"(tables) yading@10: ); yading@10: #endif yading@10: yading@10: __asm__ volatile( yading@10: "3: \n\t" yading@10: yading@10: BRANCHLESS_GET_CABAC("%4", "%q4", "(%1)", "%3", "%w3", yading@10: "%5", "%q5", "%k0", "%b0", yading@10: "%c11(%6)", "%c12(%6)", yading@10: AV_STRINGIFY(H264_NORM_SHIFT_OFFSET), yading@10: AV_STRINGIFY(H264_LPS_RANGE_OFFSET), yading@10: AV_STRINGIFY(H264_MLPS_STATE_OFFSET), yading@10: "%13") yading@10: yading@10: "test $1, %4 \n\t" yading@10: " jz 4f \n\t" yading@10: "add %10, %1 \n\t" yading@10: yading@10: BRANCHLESS_GET_CABAC("%4", "%q4", "(%1)", "%3", "%w3", yading@10: "%5", "%q5", "%k0", "%b0", yading@10: "%c11(%6)", "%c12(%6)", yading@10: AV_STRINGIFY(H264_NORM_SHIFT_OFFSET), yading@10: AV_STRINGIFY(H264_LPS_RANGE_OFFSET), yading@10: AV_STRINGIFY(H264_MLPS_STATE_OFFSET), yading@10: "%13") yading@10: yading@10: "sub %10, %1 \n\t" yading@10: "mov %2, %0 \n\t" yading@10: "movl %7, %%ecx \n\t" yading@10: "add %1, %%"REG_c" \n\t" yading@10: "movl %%ecx, (%0) \n\t" yading@10: yading@10: "test $1, %4 \n\t" yading@10: " jnz 5f \n\t" yading@10: yading@10: "add"OPSIZE" $4, %2 \n\t" yading@10: yading@10: "4: \n\t" yading@10: "add $1, %1 \n\t" yading@10: "cmp %8, %1 \n\t" yading@10: " jb 3b \n\t" yading@10: "mov %2, %0 \n\t" yading@10: "movl %7, %%ecx \n\t" yading@10: "add %1, %%"REG_c" \n\t" yading@10: "movl %%ecx, (%0) \n\t" yading@10: "5: \n\t" yading@10: "add %9, %k0 \n\t" yading@10: "shr $2, %k0 \n\t" yading@10: : "=&q"(coeff_count), "+r"(significant_coeff_ctx_base), "+m"(index), yading@10: "+&r"(c->low), "=&r"(bit), "+&r"(c->range) yading@10: : "r"(c), "m"(minusstart), "m"(end), "m"(minusindex), "m"(last_off), yading@10: "i"(offsetof(CABACContext, bytestream)), yading@10: "i"(offsetof(CABACContext, bytestream_end)) yading@10: TABLES_ARG yading@10: : "%"REG_c, "memory" yading@10: ); yading@10: return coeff_count; yading@10: } yading@10: yading@10: #define decode_significance_8x8 decode_significance_8x8_x86 yading@10: static int decode_significance_8x8_x86(CABACContext *c, yading@10: uint8_t *significant_coeff_ctx_base, yading@10: int *index, uint8_t *last_coeff_ctx_base, const uint8_t *sig_off){ yading@10: int minusindex= 4-(intptr_t)index; yading@10: int bit; yading@10: x86_reg coeff_count; yading@10: x86_reg last=0; yading@10: x86_reg state; yading@10: yading@10: #ifdef BROKEN_RELOCATIONS yading@10: void *tables; yading@10: yading@10: __asm__ volatile( yading@10: "lea "MANGLE(ff_h264_cabac_tables)", %0 \n\t" yading@10: : "=&r"(tables) yading@10: ); yading@10: #endif yading@10: yading@10: __asm__ volatile( yading@10: "mov %1, %6 \n\t" yading@10: "3: \n\t" yading@10: yading@10: "mov %10, %0 \n\t" yading@10: "movzbl (%0, %6), %k6 \n\t" yading@10: "add %9, %6 \n\t" yading@10: yading@10: BRANCHLESS_GET_CABAC("%4", "%q4", "(%6)", "%3", "%w3", yading@10: "%5", "%q5", "%k0", "%b0", yading@10: "%c12(%7)", "%c13(%7)", yading@10: AV_STRINGIFY(H264_NORM_SHIFT_OFFSET), yading@10: AV_STRINGIFY(H264_LPS_RANGE_OFFSET), yading@10: AV_STRINGIFY(H264_MLPS_STATE_OFFSET), yading@10: "%15") yading@10: yading@10: "mov %1, %k6 \n\t" yading@10: "test $1, %4 \n\t" yading@10: " jz 4f \n\t" yading@10: yading@10: #ifdef BROKEN_RELOCATIONS yading@10: "movzbl %c14(%15, %q6), %k6\n\t" yading@10: #else yading@10: "movzbl "MANGLE(ff_h264_cabac_tables)"+%c14(%k6), %k6\n\t" yading@10: #endif yading@10: "add %11, %6 \n\t" yading@10: yading@10: BRANCHLESS_GET_CABAC("%4", "%q4", "(%6)", "%3", "%w3", yading@10: "%5", "%q5", "%k0", "%b0", yading@10: "%c12(%7)", "%c13(%7)", yading@10: AV_STRINGIFY(H264_NORM_SHIFT_OFFSET), yading@10: AV_STRINGIFY(H264_LPS_RANGE_OFFSET), yading@10: AV_STRINGIFY(H264_MLPS_STATE_OFFSET), yading@10: "%15") yading@10: yading@10: "mov %2, %0 \n\t" yading@10: "mov %1, %k6 \n\t" yading@10: "movl %k6, (%0) \n\t" yading@10: yading@10: "test $1, %4 \n\t" yading@10: " jnz 5f \n\t" yading@10: yading@10: "add"OPSIZE" $4, %2 \n\t" yading@10: yading@10: "4: \n\t" yading@10: "addl $1, %k6 \n\t" yading@10: "mov %k6, %1 \n\t" yading@10: "cmpl $63, %k6 \n\t" yading@10: " jb 3b \n\t" yading@10: "mov %2, %0 \n\t" yading@10: "movl %k6, (%0) \n\t" yading@10: "5: \n\t" yading@10: "addl %8, %k0 \n\t" yading@10: "shr $2, %k0 \n\t" yading@10: : "=&q"(coeff_count), "+m"(last), "+m"(index), "+&r"(c->low), yading@10: "=&r"(bit), "+&r"(c->range), "=&r"(state) yading@10: : "r"(c), "m"(minusindex), "m"(significant_coeff_ctx_base), yading@10: "m"(sig_off), "m"(last_coeff_ctx_base), yading@10: "i"(offsetof(CABACContext, bytestream)), yading@10: "i"(offsetof(CABACContext, bytestream_end)), yading@10: "i"(H264_LAST_COEFF_FLAG_OFFSET_8x8_OFFSET) TABLES_ARG yading@10: : "%"REG_c, "memory" yading@10: ); yading@10: return coeff_count; yading@10: } yading@10: #endif /* HAVE_7REGS && !defined(BROKEN_RELOCATIONS) */ yading@10: yading@10: #endif /* HAVE_INLINE_ASM */ yading@10: #endif /* AVCODEC_X86_H264_I386_H */