;***************************************************************************** ;* MMX/SSE2/AVX-optimized 10-bit H.264 qpel code ;***************************************************************************** ;* Copyright (C) 2011 x264 project ;* ;* Authors: Daniel Kang ;* ;* This file is part of Libav. ;* ;* Libav is free software; you can redistribute it and/or ;* 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. ;* ;* Libav is distributed in the hope that it will be useful, ;* 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 ;* License along with Libav; if not, write to the Free Software ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA ;****************************************************************************** %include "libavutil/x86/x86inc.asm" %include "libavutil/x86/x86util.asm" SECTION_RODATA 32 cextern pw_16 cextern pw_1 cextern pb_0 pw_pixel_max: times 8 dw ((1 << 10)-1) pad10: times 8 dw 10*1023 pad20: times 8 dw 20*1023 pad30: times 8 dw 30*1023 depad: times 4 dd 32*20*1023 + 512 depad2: times 8 dw 20*1023 + 16*1022 + 16 unpad: times 8 dw 16*1022/32 ; needs to be mod 16 tap1: times 4 dw 1, -5 tap2: times 4 dw 20, 20 tap3: times 4 dw -5, 1 pd_0f: times 4 dd 0xffff SECTION .text %macro AVG_MOV 2 pavgw %2, %1 mova %1, %2 %endmacro %macro ADDW 3 %if mmsize == 8 paddw %1, %2 %else movu %3, %2 paddw %1, %3 %endif %endmacro %macro FILT_H 4 paddw %1, %4 psubw %1, %2 ; a-b psraw %1, 2 ; (a-b)/4 psubw %1, %2 ; (a-b)/4-b paddw %1, %3 ; (a-b)/4-b+c psraw %1, 2 ; ((a-b)/4-b+c)/4 paddw %1, %3 ; ((a-b)/4-b+c)/4+c = (a-5*b+20*c)/16 %endmacro %macro PRELOAD_V 0 lea r3, [r2*3] sub r1, r3 movu m0, [r1+r2] movu m1, [r1+r2*2] add r1, r3 movu m2, [r1] movu m3, [r1+r2] movu m4, [r1+r2*2] add r1, r3 %endmacro %macro FILT_V 8 movu %6, [r1] paddw %1, %6 mova %7, %2 paddw %7, %5 mova %8, %3 paddw %8, %4 FILT_H %1, %7, %8, [pw_16] psraw %1, 1 CLIPW %1, [pb_0], [pw_pixel_max] %endmacro %macro MC 1 %define OP_MOV mova INIT_MMX %1 mmxext, put, 4 INIT_XMM %1 sse2 , put, 8 %define OP_MOV AVG_MOV INIT_MMX %1 mmxext, avg, 4 INIT_XMM %1 sse2 , avg, 8 %endmacro %macro MCAxA 8 %ifdef ARCH_X86_64 %ifnidn %1,mmxext MCAxA_OP %1,%2,%3,%4,%5,%6,%7,%8 %endif %else MCAxA_OP %1,%2,%3,%4,%5,%6,%7,%8 %endif %endmacro %macro MCAxA_OP 8 cglobal %2_h264_qpel%5_%3_10_%1, %6,%7,%8 %ifdef ARCH_X86_32 call stub_%2_h264_qpel%4_%3_10_%1 mov r0, r0m mov r1, r1m add r0, %4*2 add r1, %4*2 call stub_%2_h264_qpel%4_%3_10_%1 mov r0, r0m mov r1, r1m lea r0, [r0+r2*%4] lea r1, [r1+r2*%4] call stub_%2_h264_qpel%4_%3_10_%1 mov r0, r0m mov r1, r1m lea r0, [r0+r2*%4+%4*2] lea r1, [r1+r2*%4+%4*2] call stub_%2_h264_qpel%4_%3_10_%1 RET %else ; ARCH_X86_64 mov r10, r0 mov r11, r1 call stub_%2_h264_qpel%4_%3_10_%1 lea r0, [r10+%4*2] lea r1, [r11+%4*2] call stub_%2_h264_qpel%4_%3_10_%1 lea r0, [r10+r2*%4] lea r1, [r11+r2*%4] call stub_%2_h264_qpel%4_%3_10_%1 lea r0, [r10+r2*%4+%4*2] lea r1, [r11+r2*%4+%4*2] %ifndef UNIX64 ; fall through to function call stub_%2_h264_qpel%4_%3_10_%1 RET %endif %endif %endmacro ;cpu, put/avg, mc, 4/8, ... %macro cglobal_mc 7 %assign i %4*2 MCAxA %1, %2, %3, %4, i, %5,%6,%7 cglobal %2_h264_qpel%4_%3_10_%1, %5,%6,%7 %ifndef UNIX64 ; no prologue or epilogue for UNIX64 call stub_%2_h264_qpel%4_%3_10_%1 RET %endif stub_%2_h264_qpel%4_%3_10_%1: %endmacro ;----------------------------------------------------------------------------- ; void h264_qpel_mc00(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro COPY4 0 movu m0, [r1 ] OP_MOV [r0 ], m0 movu m0, [r1+r2 ] OP_MOV [r0+r2 ], m0 movu m0, [r1+r2*2] OP_MOV [r0+r2*2], m0 movu m0, [r1+r3 ] OP_MOV [r0+r3 ], m0 %endmacro %macro MC00 1 INIT_MMX cglobal_mc mmxext, %1, mc00, 4, 3,4,0 lea r3, [r2*3] COPY4 ret INIT_XMM cglobal %1_h264_qpel8_mc00_10_sse2, 3,4 lea r3, [r2*3] COPY4 lea r0, [r0+r2*4] lea r1, [r1+r2*4] COPY4 RET cglobal %1_h264_qpel16_mc00_10_sse2, 3,4 mov r3d, 8 .loop: movu m0, [r1 ] movu m1, [r1 +16] OP_MOV [r0 ], m0 OP_MOV [r0 +16], m1 movu m0, [r1+r2 ] movu m1, [r1+r2+16] OP_MOV [r0+r2 ], m0 OP_MOV [r0+r2+16], m1 lea r0, [r0+r2*2] lea r1, [r1+r2*2] dec r3d jg .loop REP_RET %endmacro %define OP_MOV mova MC00 put %define OP_MOV AVG_MOV MC00 avg ;----------------------------------------------------------------------------- ; void h264_qpel_mc20(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro MC_CACHE 1 %define OP_MOV mova %define PALIGNR PALIGNR_MMX INIT_MMX %1 mmxext , put, 4 INIT_XMM %1 sse2_cache64 , put, 8 %define PALIGNR PALIGNR_SSSE3 %1 ssse3_cache64, put, 8 %1 sse2 , put, 8, 0 %define OP_MOV AVG_MOV %define PALIGNR PALIGNR_MMX INIT_MMX %1 mmxext , avg, 4 INIT_XMM %1 sse2_cache64 , avg, 8 %define PALIGNR PALIGNR_SSSE3 %1 ssse3_cache64, avg, 8 %1 sse2 , avg, 8, 0 %endmacro %macro MC20 3-4 cglobal_mc %1, %2, mc20, %3, 3,4,9 mov r3d, %3 mova m1, [pw_pixel_max] %if num_mmregs > 8 mova m8, [pw_16] %define p16 m8 %else %define p16 [pw_16] %endif .nextrow %if %0 == 4 movu m2, [r1-4] movu m3, [r1-2] movu m4, [r1+0] ADDW m2, [r1+6], m5 ADDW m3, [r1+4], m5 ADDW m4, [r1+2], m5 %else ; movu is slow on these processors %if mmsize==16 movu m2, [r1-4] movu m0, [r1+6] mova m6, m0 psrldq m0, 6 paddw m6, m2 PALIGNR m3, m0, m2, 2, m5 PALIGNR m7, m0, m2, 8, m5 paddw m3, m7 PALIGNR m4, m0, m2, 4, m5 PALIGNR m7, m0, m2, 6, m5 paddw m4, m7 SWAP 2, 6 %else movu m2, [r1-4] movu m6, [r1+4] PALIGNR m3, m6, m2, 2, m5 paddw m3, m6 PALIGNR m4, m6, m2, 4, m5 PALIGNR m7, m6, m2, 6, m5 paddw m4, m7 paddw m2, [r1+6] %endif %endif FILT_H m2, m3, m4, p16 psraw m2, 1 pxor m0, m0 CLIPW m2, m0, m1 OP_MOV [r0], m2 add r0, r2 add r1, r2 dec r3d jg .nextrow rep ret %endmacro MC_CACHE MC20 ;----------------------------------------------------------------------------- ; void h264_qpel_mc30(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro MC30 3-4 cglobal_mc %1, %2, mc30, %3, 3,5,9 lea r4, [r1+2] jmp stub_%2_h264_qpel%3_mc10_10_%1.body %endmacro MC_CACHE MC30 ;----------------------------------------------------------------------------- ; void h264_qpel_mc10(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro MC10 3-4 cglobal_mc %1, %2, mc10, %3, 3,5,9 mov r4, r1 .body mov r3d, %3 mova m1, [pw_pixel_max] %if num_mmregs > 8 mova m8, [pw_16] %define p16 m8 %else %define p16 [pw_16] %endif .nextrow %if %0 == 4 movu m2, [r1-4] movu m3, [r1-2] movu m4, [r1+0] ADDW m2, [r1+6], m5 ADDW m3, [r1+4], m5 ADDW m4, [r1+2], m5 %else ; movu is slow on these processors %if mmsize==16 movu m2, [r1-4] movu m0, [r1+6] mova m6, m0 psrldq m0, 6 paddw m6, m2 PALIGNR m3, m0, m2, 2, m5 PALIGNR m7, m0, m2, 8, m5 paddw m3, m7 PALIGNR m4, m0, m2, 4, m5 PALIGNR m7, m0, m2, 6, m5 paddw m4, m7 SWAP 2, 6 %else movu m2, [r1-4] movu m6, [r1+4] PALIGNR m3, m6, m2, 2, m5 paddw m3, m6 PALIGNR m4, m6, m2, 4, m5 PALIGNR m7, m6, m2, 6, m5 paddw m4, m7 paddw m2, [r1+6] %endif %endif FILT_H m2, m3, m4, p16 psraw m2, 1 pxor m0, m0 CLIPW m2, m0, m1 movu m3, [r4] pavgw m2, m3 OP_MOV [r0], m2 add r0, r2 add r1, r2 add r4, r2 dec r3d jg .nextrow rep ret %endmacro MC_CACHE MC10 ;----------------------------------------------------------------------------- ; void h264_qpel_mc02(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro V_FILT 11 v_filt%9_%10_10_%11: add r4, r2 .no_addr4: FILT_V m0, m1, m2, m3, m4, m5, m6, m7 add r1, r2 add r0, r2 ret %endmacro INIT_MMX RESET_MM_PERMUTATION %assign i 0 %rep 4 V_FILT m0, m1, m2, m3, m4, m5, m6, m7, 4, i, mmxext SWAP 0,1,2,3,4,5 %assign i i+1 %endrep INIT_XMM RESET_MM_PERMUTATION %assign i 0 %rep 6 V_FILT m0, m1, m2, m3, m4, m5, m6, m7, 8, i, sse2 SWAP 0,1,2,3,4,5 %assign i i+1 %endrep %macro MC02 3 cglobal_mc %1, %2, mc02, %3, 3,4,8 PRELOAD_V sub r0, r2 %assign j 0 %rep %3 %assign i (j % 6) call v_filt%3_ %+ i %+ _10_%1.no_addr4 OP_MOV [r0], m0 SWAP 0,1,2,3,4,5 %assign j j+1 %endrep ret %endmacro MC MC02 ;----------------------------------------------------------------------------- ; void h264_qpel_mc01(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro MC01 3 cglobal_mc %1, %2, mc01, %3, 3,5,8 mov r4, r1 .body PRELOAD_V sub r4, r2 sub r0, r2 %assign j 0 %rep %3 %assign i (j % 6) call v_filt%3_ %+ i %+ _10_%1 movu m7, [r4] pavgw m0, m7 OP_MOV [r0], m0 SWAP 0,1,2,3,4,5 %assign j j+1 %endrep ret %endmacro MC MC01 ;----------------------------------------------------------------------------- ; void h264_qpel_mc03(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro MC03 3 cglobal_mc %1, %2, mc03, %3, 3,5,8 lea r4, [r1+r2] jmp stub_%2_h264_qpel%3_mc01_10_%1.body %endmacro MC MC03 ;----------------------------------------------------------------------------- ; void h264_qpel_mc11(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro H_FILT_AVG 3-4 h_filt%2_%3_10_%1: ;FILT_H with fewer registers and averaged with the FILT_V result ;m6,m7 are tmp registers, m0 is the FILT_V result, the rest are to be used next in the next iteration ;unfortunately I need three registers, so m5 will have to be re-read from memory movu m5, [r4-4] ADDW m5, [r4+6], m7 movu m6, [r4-2] ADDW m6, [r4+4], m7 paddw m5, [pw_16] psubw m5, m6 ; a-b psraw m5, 2 ; (a-b)/4 psubw m5, m6 ; (a-b)/4-b movu m6, [r4+0] ADDW m6, [r4+2], m7 paddw m5, m6 ; (a-b)/4-b+c psraw m5, 2 ; ((a-b)/4-b+c)/4 paddw m5, m6 ; ((a-b)/4-b+c)/4+c = (a-5*b+20*c)/16 psraw m5, 1 CLIPW m5, [pb_0], [pw_pixel_max] ;avg FILT_V, FILT_H pavgw m0, m5 %if %0!=4 movu m5, [r1+r5] %endif ret %endmacro INIT_MMX RESET_MM_PERMUTATION %assign i 0 %rep 3 H_FILT_AVG mmxext, 4, i SWAP 0,1,2,3,4,5 %assign i i+1 %endrep H_FILT_AVG mmxext, 4, i, 0 INIT_XMM RESET_MM_PERMUTATION %assign i 0 %rep 6 %if i==1 H_FILT_AVG sse2, 8, i, 0 %else H_FILT_AVG sse2, 8, i %endif SWAP 0,1,2,3,4,5 %assign i i+1 %endrep %macro MC11 3 ; this REALLY needs x86_64 cglobal_mc %1, %2, mc11, %3, 3,6,8 mov r4, r1 .body PRELOAD_V sub r0, r2 sub r4, r2 mov r5, r2 neg r5 %assign j 0 %rep %3 %assign i (j % 6) call v_filt%3_ %+ i %+ _10_%1 call h_filt%3_ %+ i %+ _10_%1 %if %3==8 && i==1 movu m5, [r1+r5] %endif OP_MOV [r0], m0 SWAP 0,1,2,3,4,5 %assign j j+1 %endrep ret %endmacro MC MC11 ;----------------------------------------------------------------------------- ; void h264_qpel_mc31(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro MC31 3 cglobal_mc %1, %2, mc31, %3, 3,6,8 mov r4, r1 add r1, 2 jmp stub_%2_h264_qpel%3_mc11_10_%1.body %endmacro MC MC31 ;----------------------------------------------------------------------------- ; void h264_qpel_mc13(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro MC13 3 cglobal_mc %1, %2, mc13, %3, 3,7,12 lea r4, [r1+r2] jmp stub_%2_h264_qpel%3_mc11_10_%1.body %endmacro MC MC13 ;----------------------------------------------------------------------------- ; void h264_qpel_mc33(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro MC33 3 cglobal_mc %1, %2, mc33, %3, 3,6,8 lea r4, [r1+r2] add r1, 2 jmp stub_%2_h264_qpel%3_mc11_10_%1.body %endmacro MC MC33 ;----------------------------------------------------------------------------- ; void h264_qpel_mc22(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro FILT_H2 3 psubw %1, %2 ; a-b psubw %2, %3 ; b-c psllw %2, 2 psubw %1, %2 ; a-5*b+4*c psllw %3, 4 paddw %1, %3 ; a-5*b+20*c %endmacro %macro FILT_VNRD 8 movu %6, [r1] paddw %1, %6 mova %7, %2 paddw %7, %5 mova %8, %3 paddw %8, %4 FILT_H2 %1, %7, %8 %endmacro %macro HV 2 %ifidn %1,sse2 %define PAD 12 %define COUNT 2 %else %define PAD 0 %define COUNT 3 %endif put_hv%2_10_%1: neg r2 ; This actually saves instructions lea r1, [r1+r2*2-mmsize+PAD] lea r4, [rsp+PAD+gprsize] mov r3d, COUNT .v_loop: movu m0, [r1] sub r1, r2 movu m1, [r1] sub r1, r2 movu m2, [r1] sub r1, r2 movu m3, [r1] sub r1, r2 movu m4, [r1] sub r1, r2 %assign i 0 %rep %2-1 FILT_VNRD m0, m1, m2, m3, m4, m5, m6, m7 psubw m0, [pad20] movu [r4+i*mmsize*3], m0 sub r1, r2 SWAP 0,1,2,3,4,5 %assign i i+1 %endrep FILT_VNRD m0, m1, m2, m3, m4, m5, m6, m7 psubw m0, [pad20] movu [r4+i*mmsize*3], m0 add r4, mmsize lea r1, [r1+r2*8+mmsize] %if %2==8 lea r1, [r1+r2*4] %endif dec r3d jg .v_loop neg r2 ret %endmacro INIT_MMX HV mmxext, 4 INIT_XMM HV sse2 , 8 %macro H_LOOP 2 %if num_mmregs > 8 %define s1 m8 %define s2 m9 %define s3 m10 %define d1 m11 %else %define s1 [tap1] %define s2 [tap2] %define s3 [tap3] %define d1 [depad] %endif h%2_loop_op_%1: movu m1, [r1+mmsize-4] movu m2, [r1+mmsize-2] mova m3, [r1+mmsize+0] movu m4, [r1+mmsize+2] movu m5, [r1+mmsize+4] movu m6, [r1+mmsize+6] %if num_mmregs > 8 pmaddwd m1, s1 pmaddwd m2, s1 pmaddwd m3, s2 pmaddwd m4, s2 pmaddwd m5, s3 pmaddwd m6, s3 paddd m1, d1 paddd m2, d1 %else mova m0, s1 pmaddwd m1, m0 pmaddwd m2, m0 mova m0, s2 pmaddwd m3, m0 pmaddwd m4, m0 mova m0, s3 pmaddwd m5, m0 pmaddwd m6, m0 mova m0, d1 paddd m1, m0 paddd m2, m0 %endif paddd m3, m5 paddd m4, m6 paddd m1, m3 paddd m2, m4 psrad m1, 10 psrad m2, 10 pslld m2, 16 pand m1, [pd_0f] por m1, m2 %if num_mmregs <= 8 pxor m0, m0 %endif CLIPW m1, m0, m7 add r1, mmsize*3 ret %endmacro INIT_MMX H_LOOP mmxext, 4 INIT_XMM H_LOOP sse2 , 8 %macro MC22 3 cglobal_mc %1, %2, mc22, %3, 3,7,12 %define PAD mmsize*8*4*2 ; SIZE*16*4*sizeof(pixel) mov r6, rsp ; backup stack pointer and rsp, ~(mmsize-1) ; align stack sub rsp, PAD call put_hv%3_10_%1 mov r3d, %3 mova m7, [pw_pixel_max] %if num_mmregs > 8 pxor m0, m0 mova m8, [tap1] mova m9, [tap2] mova m10, [tap3] mova m11, [depad] %endif mov r1, rsp .h_loop: call h%3_loop_op_%1 OP_MOV [r0], m1 add r0, r2 dec r3d jg .h_loop mov rsp, r6 ; restore stack pointer ret %endmacro MC MC22 ;----------------------------------------------------------------------------- ; void h264_qpel_mc12(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro MC12 3 cglobal_mc %1, %2, mc12, %3, 3,7,12 %define PAD mmsize*8*4*2 ; SIZE*16*4*sizeof(pixel) mov r6, rsp ; backup stack pointer and rsp, ~(mmsize-1) ; align stack sub rsp, PAD call put_hv%3_10_%1 xor r4d, r4d .body mov r3d, %3 pxor m0, m0 mova m7, [pw_pixel_max] %if num_mmregs > 8 mova m8, [tap1] mova m9, [tap2] mova m10, [tap3] mova m11, [depad] %endif mov r1, rsp .h_loop: call h%3_loop_op_%1 movu m3, [r1+r4-2*mmsize] ; movu needed for mc32, etc paddw m3, [depad2] psrlw m3, 5 psubw m3, [unpad] CLIPW m3, m0, m7 pavgw m1, m3 OP_MOV [r0], m1 add r0, r2 dec r3d jg .h_loop mov rsp, r6 ; restore stack pointer ret %endmacro MC MC12 ;----------------------------------------------------------------------------- ; void h264_qpel_mc32(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro MC32 3 cglobal_mc %1, %2, mc32, %3, 3,7,12 %define PAD mmsize*8*3*2 ; SIZE*16*4*sizeof(pixel) mov r6, rsp ; backup stack pointer and rsp, ~(mmsize-1) ; align stack sub rsp, PAD call put_hv%3_10_%1 mov r4d, 2 ; sizeof(pixel) jmp stub_%2_h264_qpel%3_mc12_10_%1.body %endmacro MC MC32 ;----------------------------------------------------------------------------- ; void h264_qpel_mc21(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro H_NRD 2 put_h%2_10_%1: add rsp, gprsize mov r3d, %2 xor r4d, r4d mova m6, [pad20] .nextrow movu m2, [r5-4] movu m3, [r5-2] movu m4, [r5+0] ADDW m2, [r5+6], m5 ADDW m3, [r5+4], m5 ADDW m4, [r5+2], m5 FILT_H2 m2, m3, m4 psubw m2, m6 mova [rsp+r4], m2 add r4d, mmsize*3 add r5, r2 dec r3d jg .nextrow sub rsp, gprsize ret %endmacro INIT_MMX H_NRD mmxext, 4 INIT_XMM H_NRD sse2 , 8 %macro MC21 3 cglobal_mc %1, %2, mc21, %3, 3,7,12 mov r5, r1 .body %define PAD mmsize*8*3*2 ; SIZE*16*4*sizeof(pixel) mov r6, rsp ; backup stack pointer and rsp, ~(mmsize-1) ; align stack sub rsp, PAD call put_h%3_10_%1 sub rsp, PAD call put_hv%3_10_%1 mov r4d, PAD-mmsize ; H buffer jmp stub_%2_h264_qpel%3_mc12_10_%1.body %endmacro MC MC21 ;----------------------------------------------------------------------------- ; void h264_qpel_mc23(uint8_t *dst, uint8_t *src, int stride) ;----------------------------------------------------------------------------- %macro MC23 3 cglobal_mc %1, %2, mc23, %3, 3,7,12 lea r5, [r1+r2] jmp stub_%2_h264_qpel%3_mc21_10_%1.body %endmacro MC MC23