Advertisement
Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- ; /*
- ; * Provide SSE luma mc functions for HEVC decoding
- ; * Copyright (c) 2013 Pierre-Edouard LEPERE
- ; *
- ; * This file is part of FFmpeg.
- ; *
- ; * FFmpeg 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.
- ; *
- ; * FFmpeg 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 FFmpeg; if not, write to the Free Software
- ; * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- ; */
- %include "libavutil/x86/x86util.asm"
- %if ARCH_X86_64
- SECTION_RODATA
- cextern hevc_epel_filters
- epel_extra_before: DB 1 ;corresponds to EPEL_EXTRA_BEFORE in hevc.h
- max_pb_size: DB 64 ;corresponds to MAX_PB_SIZE in hevc.h
- epel_h_shuffle1_m: DB 0, 1, 2, 3, 1, 2, 3, 4, 2, 3, 4, 5, 3, 4, 5, 6
- epel_h_shuffle2_m: DB 4, 5, 6, 7, 5, 6, 7, 8, 6, 7, 8, 9, 7, 8, 9, 10
- %ifdef PIC
- %define epel_h_shuffle1 picregq
- %define epel_h_shuffle2 picregq
- %define npicregs 1
- %else
- %define epel_h_shuffle1 epel_h_shuffle1_m
- %define epel_h_shuffle2 epel_h_shuffle2_m
- %define npicregs 0
- %endif
- SECTION .text
- INIT_XMM sse4 ; adds ff_ and _sse4 to function name
- ; ******************************
- ; void put_hevc_mc_pixels_8(int16_t *dst, ptrdiff_t dststride,
- ; uint8_t *_src, ptrdiff_t _srcstride,
- ; int width, int height, int mx, int my,
- ; int16_t* mcbuffer)
- ;
- ; r0 : *dst
- ; r1 : dststride
- ; r2 : *src
- ; r3 : srcstride
- ; r4 : width
- ; r5 : height
- ;
- ; ******************************
- ; 1 by 1. Can be done on any processor
- cglobal put_hevc_mc_pixels_2_8, 9, 12, 0 , dst, dststride, src, srcstride, width, height
- mov r6,0 ; height
- mc_pixels_2_h: ; for height
- mov r7, 0 ; width
- mc_pixels_2_w: ; for width
- mov r9,0
- mov r9b,[srcq+r7] ; get byte
- shl r9,6 ; shift
- mov [dstq+2*r7],r9w ; store
- inc r7
- cmp r7, widthq ; cmp width
- jl mc_pixels_2_w ; width loop
- lea dstq,[dstq+2*dststrideq] ; dst += dststride
- lea srcq,[srcq+srcstrideq] ; src += srcstride
- add r6,1
- cmp r6,heightq ; cmp height
- jl mc_pixels_2_h ; height loop
- RET
- ; 4 by 4
- cglobal put_hevc_mc_pixels_4_8, 9, 12, 0 , dst, dststride, src, srcstride,width,height
- pxor m0,m0 ; set register at zero
- mov r6,0 ; height
- mov r9,0
- ; 8 by 8
- mc_pixels_4_h: ; for height
- mov r7,0 ; width
- mc_pixels_4_w: ; for width
- pxor m1,m1
- movq m1,[srcq+r7] ; load 64 bits
- punpcklbw m2,m1,m0 ; unpack to 16 bits
- psllw m2,6 ; shift left 6 bits (14 - bit depth) each 16bit element
- movq [dstq+2*r7],m2 ; store 64 bits
- add r7,4 ; add 4 for width loop
- cmp r7, widthq ; cmp width
- jl mc_pixels_4_w ; width loop
- lea dstq,[dstq+2*dststrideq] ; dst += dststride
- lea srcq,[srcq+srcstrideq] ; src += srcstride
- add r6,1
- cmp r6,heightq ; cmp height
- jl mc_pixels_4_h ; height loop
- RET
- ; 8 by 8
- cglobal put_hevc_mc_pixels_8_8, 9, 12, 0 , dst, dststride, src, srcstride,width,height
- pxor m0,m0 ; set register at zero
- mov r6,0 ; height
- mov r9,0
- ; 8 by 8
- mc_pixels_8_h: ; for height
- mov r7,0 ; width
- mc_pixels_8_w: ; for width
- pxor m1,m1
- movq m1,[srcq+r7] ; load 64 bits
- punpcklbw m2,m1,m0 ; unpack to 16 bits
- psllw m2,6 ; shift left 6 bits (14 - bit depth) each 16bit element
- movdqu [dstq+2*r7],m2 ; store 128 bits
- add r7,8 ; add 8 for width loop
- cmp r7, r4 ; cmp width
- cmp r7, widthq ; cmp width
- jl mc_pixels_8_w ; width loop
- lea dstq,[dstq+2*dststrideq] ; dst += dststride
- lea srcq,[srcq+srcstrideq] ; src += srcstride
- add r6,1
- cmp r6,heightq ; cmp height
- jl mc_pixels_8_h ; height loop
- RET
- ; 16 by 16
- cglobal put_hevc_mc_pixels_16_8, 9, 12, 0 , dst, dststride, src, srcstride,width,height
- pxor m0,m0 ; set register at zero
- mov r6,0 ; height
- mov r9,0
- ; 8 by 8
- mc_pixels_16_h: ; for height
- mov r7,0 ; width
- mc_pixels_16_w: ; for width
- pxor m1,m1
- movdqu m1,[srcq+r7] ; load 128 bits
- punpcklbw m2,m1,m0 ; unpack low to 16 bits
- punpckhbw m3,m1,m0 ; unpack high to 16 bits
- psllw m2,6 ; shift left 6 bits (14 - bit depth) each 16bit element
- psllw m3,6 ; shift left 6 bits (14 - bit depth) each 16bit element
- movdqu [dstq+2*r7],m2 ; store 128 bits
- movdqu [dstq+2*r7+16],m3 ; store 128 bits
- add r7,16 ; add 16 for width loop
- cmp r7, widthq ; cmp width
- jl mc_pixels_16_w ; width loop
- lea dstq,[dstq+2*dststrideq] ; dst += dststride
- lea srcq,[srcq+srcstrideq] ; src += srcstride
- add r6,1
- cmp r6,heightq ; cmp height
- jl mc_pixels_16_h ; height loop
- RET
- ;function to call other mc_pixels functions according to width value
- cglobal put_hevc_mc_pixels_master_8, 9, 12, 0 , dst, dststride, src, srcstride,width,height
- cmp r4,8
- je goto_mc_pixels_8_8
- cmp r4,16
- je goto_mc_pixels_16_8
- cmp r4,4
- je goto_mc_pixels_4_8
- cmp r4,32
- je goto_mc_pixels_16_8
- cmp r4,12
- je goto_mc_pixels_4_8
- cmp r4,24
- je goto_mc_pixels_8_8
- cmp r4,6
- je goto_mc_pixels_2_8
- cmp r4,64
- je goto_mc_pixels_16_8
- cmp r4,2
- je goto_mc_pixels_2_8
- jmp goto_mc_pixels_2_8
- goto_mc_pixels_2_8:
- call put_hevc_mc_pixels_2_8
- RET
- goto_mc_pixels_4_8:
- call put_hevc_mc_pixels_4_8
- RET
- goto_mc_pixels_8_8:
- call put_hevc_mc_pixels_8_8
- RET
- goto_mc_pixels_16_8:
- call put_hevc_mc_pixels_16_8
- RET
- ; ******************************
- ; void put_hevc_epel_h_8(int16_t *dst, ptrdiff_t dststride,
- ; uint8_t *_src, ptrdiff_t _srcstride,
- ; int width, int height, int mx, int my,
- ; int16_t* mcbuffer)
- ;
- ; r0 : *dst
- ; r1 : dststride
- ; r2 : *src
- ; r3 : srcstride
- ; r4 : width
- ; r5 : height
- ;
- ; ******************************
- ;8 by 8
- cglobal put_hevc_epel_h_8_8, 7, 12+ npicregs, 0 , dst, dststride, src, srcstride,width,height,mx,my,picreg
- %ifdef PIC
- lea picregq, [epel_h_shuffle2_m]
- %endif
- movdqu m4,[epel_h_shuffle2]
- epel_h_8_h:
- mov r8,0 ; width counter
- epel_h_8_w:
- movdqu m2,[srcq+r8-1] ; load data from source
- pshufb m5,m4,m0 ; shuffle2
- pshufb m2,m2,m0 ; shuffle
- pmaddubsw m2,m1 ; maddubs (see SSE instruction set for details)
- pmaddubsw m5,m1 ; maddubs (see SSE instruction set for details)
- phaddw m2,m5 ; horizontal add
- movdqu [dstq+2*r8],m2 ; store data to dst
- add r8,4
- cmp r8,widthq ;
- jl epel_h_4_w
- lea dstq,[dstq+2*dststrideq] ; dst += dststride
- lea srcq,[srcq+srcstrideq] ; src += srcstride
- add r9,1
- cmp r9,heightq ; cmp height
- jl epel_h_4_h ; height loop
- RET
- ;4 by 4
- cglobal put_hevc_epel_h_4_8, 9, 12, 0 , dst, dststride, src, srcstride,width,height,mx,my
- epel_h_4_h:
- mov r8,0 ; width counter
- epel_h_4_w:
- movdqu m2,[srcq+r8-1] ; load data from source
- pshufb m2,m2,m0 ; shuffle
- pmaddubsw m2,m1 ; maddubs (see SSE instruction set for details)
- phaddw m2,m3 ; horizontal add
- movq [dstq+2*r8],m2 ; store data to dst
- add r8,4
- cmp r8,widthq ;
- jl epel_h_8_w
- lea dstq,[dstq+2*dststrideq] ; dst += dststride
- lea srcq,[srcq+srcstrideq] ; src += srcstride
- add r9,1
- cmp r9,heightq ; cmp height
- jl epel_h_8_h ; height loop
- RET
- cglobal put_hevc_epel_h_master_8, 9, 12, 0 , dst, dststride, src, srcstride,width,height,mx,my,picreg
- %ifdef PIC
- lea picregq, [epel_h_shuffle1_m]
- %endif
- movsxd mxq,mxd ; extend sign
- pxor m3,m3 ; set zero
- lea r10,[hevc_epel_filters]
- sub mxq,1
- shl mxq,4 ; multiply by 16
- movq m0,[r10+mxq] ; get 4 first values of filters
- pshufd m1,m0,0 ; cast 32 bit on all register.
- movdqu m0,[epel_h_shuffle1] ; register containing shuffle
- mov r9,0 ; height counter
- cmp widthq,2
- je goto_epel_h_4_8
- cmp widthq,4
- je goto_epel_h_4_8
- cmp widthq,6
- je goto_epel_h_4_8
- cmp widthq,8
- je goto_epel_h_8_8
- cmp widthq,12
- je goto_epel_h_4_8
- cmp widthq,16
- je goto_epel_h_8_8
- cmp widthq,24
- je goto_epel_h_8_8
- cmp widthq,32
- je goto_epel_h_8_8
- cmp widthq,64
- je goto_epel_h_8_8
- jmp goto_epel_h_4_8
- RET
- goto_epel_h_4_8:
- call put_hevc_epel_h_4_8
- RET
- goto_epel_h_8_8:
- call put_hevc_epel_h_8_8
- RET
- ;TODO : epel_h_2
- ; ******************************
- ; void put_hevc_epel_v_8(int16_t *dst, ptrdiff_t dststride,
- ; uint8_t *_src, ptrdiff_t _srcstride,
- ; int width, int height, int mx, int my,
- ; int16_t* mcbuffer)
- ;
- ; r0 : *dst
- ; r1 : dststride
- ; r2 : *src
- ; r3 : srcstride
- ; r4 : width
- ; r5 : height
- ;
- ; ******************************
- cglobal put_hevc_epel_v_16_8, 8, 12, 0 , dst, dststride, src, srcstride, width, height, mx, my
- ;filters and initialization is done in master
- epel_v_16_h:
- mov r9,0 ;set width counter
- epel_v_16_v:
- lea r10,[srcq+r9]
- mov r11,r10
- sub r11,srcstrideq
- movdqu m5,[r11] ;load 64bit of x-stride
- movdqu m6,[r10] ;load 64bit of x
- movdqu m7,[r10+srcstrideq] ;load 64bit of x+stride
- movdqu m8,[r10+2*srcstrideq] ;load 64bit of x+2*stride
- punpckhbw m9,m5,m0 ;unpack 8bit to 16bit
- punpckhbw m10,m6,m0 ;unpack 8bit to 16bit
- punpckhbw m11,m7,m0 ;unpack 8bit to 16bit
- punpckhbw m12,m8,m0 ;unpack 8bit to 16bit
- punpcklbw m5,m0 ;unpack 8bit to 16bit
- punpcklbw m6,m0 ;unpack 8bit to 16bit
- punpcklbw m7,m0 ;unpack 8bit to 16bit
- punpcklbw m8,m0 ;unpack 8bit to 16bit
- pmullw m9,m1 ;multiply values with filter
- pmullw m10,m2
- pmullw m11,m3
- pmullw m12,m4
- pmullw m5,m1 ;multiply values with filter
- pmullw m6,m2
- pmullw m7,m3
- pmullw m8,m4
- paddsw m9,m10 ;add the different values
- paddsw m11,m12
- paddsw m9,m11
- paddsw m5,m6 ;add the different values
- paddsw m7,m8
- paddsw m5,m7
- movdqu [dstq+2*r9],m5 ;store 128bit to dst
- movdqu [dstq+2*r9+16],m9 ;store 128bit to dst
- add r9,16 ; add 16 for width loop
- cmp r9, widthq ; cmp width
- jl epel_v_16_v ; width loop
- lea dstq,[dstq+2*dststrideq] ; dst += dststride
- lea srcq,[srcq+srcstrideq] ; src += srcstride
- add r8,1
- cmp r8,heightq ; cmp height
- jl epel_v_16_h ; height loop
- RET
- cglobal put_hevc_epel_v_8_8, 8, 12, 0 , dst, dststride, src, srcstride,width,height,mx,my
- ;filters and initialization is done in master
- epel_v_8_h:
- mov r9,0 ;set width counter
- epel_v_8_v:
- lea r10,[srcq+r9]
- mov r11,r10
- sub r11,srcstrideq
- movq m5,[r11] ;load 64bit of x-stride
- movq m6,[r10] ;load 64bit of x
- movq m7,[r10+srcstrideq] ;load 64bit of x+stride
- movq m8,[r10+2*srcstrideq] ;load 64bit of x+2*stride
- punpcklbw m5,m0 ;unpack 8bit to 16bit
- punpcklbw m6,m0 ;unpack 8bit to 16bit
- punpcklbw m7,m0 ;unpack 8bit to 16bit
- punpcklbw m8,m0 ;unpack 8bit to 16bit
- pmullw m5,m1 ;multiply values with filter
- pmullw m6,m2
- pmullw m7,m3
- pmullw m8,m4
- paddsw m5,m6 ;add the different values
- paddsw m7,m8
- paddsw m5,m7
- movdqu [dstq+2*r9],m5 ;store 64bit to dst
- add r9,8 ; add 4 for width loop
- cmp r9, widthq ; cmp width
- jl epel_v_8_v ; width loop
- lea dstq,[dstq+2*dststrideq] ; dst += dststride
- lea srcq,[srcq+srcstrideq] ; src += srcstride
- add r8,1
- cmp r8,heightq ; cmp height
- jl epel_v_8_h ; height loop
- RET
- cglobal put_hevc_epel_v_4_8, 8, 12, 0, dst, dststride, src, srcstride, width, height, mx, my
- epel_v_4_h:
- mov r9,0 ;set width counter
- epel_v_4_v:
- lea r10,[srcq+r9]
- mov r11,r10
- sub r11,srcstrideq
- movq m5,[r11] ;load 64bit of x-stride
- movq m6,[r10] ;load 64bit of x
- movq m7,[r10+srcstrideq] ;load 64bit of x+stride
- movq m8,[r10+2*srcstrideq] ;load 64bit of x+2*stride
- punpcklbw m5,m0 ;unpack 8bit to 16bit
- punpcklbw m6,m0 ;unpack 8bit to 16bit
- punpcklbw m7,m0 ;unpack 8bit to 16bit
- punpcklbw m8,m0 ;unpack 8bit to 16bit
- pmullw m5,m1 ;multiply values with filter
- pmullw m6,m2
- pmullw m7,m3
- pmullw m8,m4
- paddsw m5,m6 ;add the different values
- paddsw m7,m8
- paddsw m5,m7
- movq [dstq+2*r9],m5 ;store 64bit to dst
- add r9,4 ; add 4 for width loop
- cmp r9, widthq ; cmp width
- jl epel_v_4_v ; width loop
- lea dstq,[dstq+2*dststrideq] ; dst += dststride
- lea srcq,[srcq+srcstrideq] ; src += srcstride
- add r8,1
- cmp r8,heightq ; cmp height
- jl epel_v_4_h ; height loop
- RET
- ;function to call other epel_v functions according to width value
- cglobal put_hevc_epel_v_master_8, 8, 12, 0 , dst, dststride, src, srcstride,width,height,mx,my
- pxor m1,m1 ;set zero
- movsxd myq,myd ;extend sign
- sub myq,1 ;my-1
- shl myq,4 ;multiply by 16
- lea r10,[hevc_epel_filters]
- movq m0,[r10+myq] ;filters
- punpcklbw m1,m0 ;unpack to 16 bit
- psraw m1,8 ;shift for bit-sign
- punpcklwd m1,m1 ;put double values to 32bit.
- psrldq m2,m1,4 ;filter 1
- psrldq m3,m1,8 ;filter 2
- psrldq m4,m1,12 ;filter 3
- pshufd m1,m1,0 ;extend 32bit to whole register
- pshufd m2,m2,0
- pshufd m3,m3,0
- pshufd m4,m4,0
- pxor m0,m0 ;set zero
- mov r8,0 ;set height counter
- cmp r4,8
- je goto_epel_v_8_8
- cmp r4,16
- je goto_epel_v_16_8
- cmp r4,4
- je goto_epel_v_4_8
- cmp r4,32
- je goto_epel_v_16_8
- cmp r4,12
- je goto_epel_v_4_8
- cmp r4,24
- je goto_epel_v_8_8
- cmp r4,6
- je goto_epel_v_2_8
- cmp r4,64
- je goto_epel_v_16_8
- cmp r4,2
- je goto_epel_v_2_8
- jmp goto_epel_v_2_8
- RET
- goto_epel_v_4_8:
- call put_hevc_epel_v_4_8
- RET
- goto_epel_v_8_8:
- call put_hevc_epel_v_8_8
- RET
- goto_epel_v_16_8:
- call put_hevc_epel_v_16_8
- RET
- goto_epel_v_2_8:
- call put_hevc_epel_v_4_8
- RET
- ;TODO : put_hevc_epel_v_2_8.
- ; ******************************
- ; void put_hevc_epel_hv_8(int16_t *dst, ptrdiff_t dststride,
- ; uint8_t *_src, ptrdiff_t _srcstride,
- ; int width, int height, int mx, int my,
- ; int16_t* mcbuffer)
- ;
- ; r0 : *dst
- ; r1 : dststride
- ; r2 : *src
- ; r3 : srcstride
- ; r4 : width
- ; r5 : height
- ; r6 : mx
- ; r7 : my
- ; r8 : mcbuffer
- ;
- ; ******************************
- cglobal put_hevc_epel_hv_4_8, 9, 12, 0 , dst, dststride, src, srcstride, width, height, mx, my, mcbuffer,picreg
- %ifdef PIC
- lea picregq, [epel_h_shuffle1_m]
- %endif
- movsxd mxq,mxd ; extend sign
- movsxd myq,myd ; extend sign
- movsxd srcstrideq,srcstrided ; extend sign
- movsxd dststrideq,dststrided ; extend sign
- movsxd widthq,widthd ; extend sign
- movsxd heightq,heightd ; extend sign
- pxor m3,m3 ; set zero
- lea r10,[hevc_epel_filters]
- sub mxq,1
- shl mxq,4 ; multiply by 16
- movq m0,[r10+mxq] ; get 4 first values of filters
- pshufd m1,m0,0 ; cast 32 bit on all register.
- movdqu m0,[epel_h_shuffle1] ; register containing shuffle1
- mov r15, srcstrideq
- imul r15, epel_extra_before
- ; sub srcq,srcstrideq ; src -= EPEL_EXTRA_BEFORE * srcstride
- mov r14,max_pb_size
- shl r14,2 ;double because it's used for 16bit adressing
- lea r13,[mcbufferq]
- mov r9,0 ; height counter
- epel_hv_4_h_h:
- xor r10,r10 ; width counter
- epel_hv_4_h_w:
- movdqu m2,[srcq+r10] ; load data from source
- pshufb m2,m2,m0 ; shuffle
- pmaddubsw m2,m1 ; maddubs (see SSE instruction set for details)
- phaddw m2,m3 ; horizontal add
- movq [r13+2*r10],m2 ; store data to dst
- add r10,4
- cmp r10,widthq ;
- jl epel_hv_4_h_w
- lea dstq,[r13+r14] ; dst += dststride
- lea srcq,[srcq+srcstrideq] ; src += srcstride
- add r9,1
- cmp r9,heightq ; cmp height
- jl epel_hv_4_h_h ; height loop
- ;end H treatment
- ; imul r14,r15;
- add mcbufferq,r14 ; mcbuffer+= EPEL_EXTRA_BEFORE * MAX_PB_SIZE
- pxor m1,m1 ;set zero
- movsxd myq,myd ;extend sign
- sub myq,1 ;my-1
- shl myq,4 ;multiply by 16
- lea r10,[hevc_epel_filters]
- movq m0,[r10+myq] ;filters
- punpcklbw m1,m0 ;unpack to 16 bit
- psraw m1,8 ;shift for bit-sign
- punpcklwd m1,m1 ;put double values to 32bit.
- psrldq m2,m1,4 ;filter 1
- psrldq m3,m1,8 ;filter 2
- psrldq m4,m1,12 ;filter 3
- pshufd m1,m1,0 ;extend 32bit to whole register
- pshufd m2,m2,0
- pshufd m3,m3,0
- pshufd m4,m4,0
- pxor m0,m0 ;set zero
- mov r8,0 ;set height counter
- epel_hv_4_v_h:
- mov r9,0 ;set width counter
- epel_hv_4_v_w:
- lea r10,[mcbufferq+r9]
- mov r11,r10
- sub r11,srcstrideq
- ; movq m5,[r11] ;load 64bit of x-stride
- ; movq m6,[r10] ;load 64bit of x
- ; movq m7,[r10+srcstrideq] ;load 64bit of x+stride
- ; movq m8,[r10+2*srcstrideq] ;load 64bit of x+2*stride
- punpcklbw m5,m0 ;unpack to 16 bit
- psraw m5,16 ;shift for bit-sign
- punpcklbw m6,m0 ;unpack to 16 bit
- psraw m6,16 ;shift for bit-sign
- punpcklbw m7,m0 ;unpack to 16 bit
- psraw m7,16 ;shift for bit-sign
- punpcklbw m8,m0 ;unpack to 16 bit
- psraw m8,16 ;shift for bit-sign
- pmulld m5,m1 ;multiply values with filter - SSE4.2 function
- pmulld m6,m2
- pmulld m7,m3
- pmulld m8,m4
- paddd m5,m6 ;add the different values
- paddd m7,m8
- paddd m5,m7
- psrld m5,2 ; >> BIT_DEPTH-2
- packssdw m5,m0 ;back to 16bit
- pxor m5,m5
- ; movdqu [dstq+2*r9],m5 ;store 128bit to dst
- add r9,4 ; add 4 for width loop
- cmp r9, widthq ; cmp width
- jl epel_hv_4_v_w ; width loop
- lea dstq,[dstq+2*dststrideq] ; dst += dststride
- lea srcq,[mcbufferq+r15] ; src += srcstride
- add r8,1
- cmp r8,heightq ; cmp height
- jl epel_hv_4_v_h ; height loop
- RET
- %endif ; ARCH_X86_64
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
