yading@10: ;*****************************************************************************
yading@10: ;* MMX/SSE2/AVX-optimized 10-bit H.264 weighted prediction code
yading@10: ;*****************************************************************************
yading@10: ;* Copyright (C) 2005-2011 x264 project
yading@10: ;*
yading@10: ;* Authors: Daniel Kang <daniel.d.kang@gmail.com>
yading@10: ;*
yading@10: ;* This file is part of Libav.
yading@10: ;*
yading@10: ;* Libav 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: ;* Libav 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 Libav; 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: %include "libavutil/x86/x86util.asm"
yading@10: 
yading@10: SECTION_RODATA 32
yading@10: 
yading@10: pw_pixel_max: times 8 dw ((1 << 10)-1)
yading@10: sq_1: dq 1
yading@10:       dq 0
yading@10: 
yading@10: cextern pw_1
yading@10: 
yading@10: SECTION .text
yading@10: 
yading@10: ;-----------------------------------------------------------------------------
yading@10: ; void h264_weight(uint8_t *dst, int stride, int height, int log2_denom,
yading@10: ;                  int weight, int offset);
yading@10: ;-----------------------------------------------------------------------------
yading@10: %macro WEIGHT_PROLOGUE 0
yading@10: .prologue:
yading@10:     PROLOGUE 0,6,8
yading@10:     movifnidn  r0, r0mp
yading@10:     movifnidn r1d, r1m
yading@10:     movifnidn r2d, r2m
yading@10:     movifnidn r4d, r4m
yading@10:     movifnidn r5d, r5m
yading@10: %endmacro
yading@10: 
yading@10: %macro WEIGHT_SETUP 0
yading@10:     mova       m0, [pw_1]
yading@10:     movd       m2, r3m
yading@10:     pslld      m0, m2       ; 1<<log2_denom
yading@10:     SPLATW     m0, m0
yading@10:     shl        r5, 19       ; *8, move to upper half of dword
yading@10:     lea        r5, [r5+r4*2+0x10000]
yading@10:     movd       m3, r5d      ; weight<<1 | 1+(offset<<(3))
yading@10:     pshufd     m3, m3, 0
yading@10:     mova       m4, [pw_pixel_max]
yading@10:     paddw      m2, [sq_1]   ; log2_denom+1
yading@10: %if notcpuflag(sse4)
yading@10:     pxor       m7, m7
yading@10: %endif
yading@10: %endmacro
yading@10: 
yading@10: %macro WEIGHT_OP 1-2
yading@10: %if %0==1
yading@10:     mova        m5, [r0+%1]
yading@10:     punpckhwd   m6, m5, m0
yading@10:     punpcklwd   m5, m0
yading@10: %else
yading@10:     movq        m5, [r0+%1]
yading@10:     movq        m6, [r0+%2]
yading@10:     punpcklwd   m5, m0
yading@10:     punpcklwd   m6, m0
yading@10: %endif
yading@10:     pmaddwd     m5, m3
yading@10:     pmaddwd     m6, m3
yading@10:     psrad       m5, m2
yading@10:     psrad       m6, m2
yading@10: %if cpuflag(sse4)
yading@10:     packusdw    m5, m6
yading@10:     pminsw      m5, m4
yading@10: %else
yading@10:     packssdw    m5, m6
yading@10:     CLIPW       m5, m7, m4
yading@10: %endif
yading@10: %endmacro
yading@10: 
yading@10: %macro WEIGHT_FUNC_DBL 0
yading@10: cglobal h264_weight_16_10
yading@10:     WEIGHT_PROLOGUE
yading@10:     WEIGHT_SETUP
yading@10: .nextrow:
yading@10:     WEIGHT_OP  0
yading@10:     mova [r0   ], m5
yading@10:     WEIGHT_OP 16
yading@10:     mova [r0+16], m5
yading@10:     add       r0, r1
yading@10:     dec       r2d
yading@10:     jnz .nextrow
yading@10:     REP_RET
yading@10: %endmacro
yading@10: 
yading@10: INIT_XMM sse2
yading@10: WEIGHT_FUNC_DBL
yading@10: INIT_XMM sse4
yading@10: WEIGHT_FUNC_DBL
yading@10: 
yading@10: 
yading@10: %macro WEIGHT_FUNC_MM 0
yading@10: cglobal h264_weight_8_10
yading@10:     WEIGHT_PROLOGUE
yading@10:     WEIGHT_SETUP
yading@10: .nextrow:
yading@10:     WEIGHT_OP   0
yading@10:     mova     [r0], m5
yading@10:     add        r0, r1
yading@10:     dec        r2d
yading@10:     jnz .nextrow
yading@10:     REP_RET
yading@10: %endmacro
yading@10: 
yading@10: INIT_XMM sse2
yading@10: WEIGHT_FUNC_MM
yading@10: INIT_XMM sse4
yading@10: WEIGHT_FUNC_MM
yading@10: 
yading@10: 
yading@10: %macro WEIGHT_FUNC_HALF_MM 0
yading@10: cglobal h264_weight_4_10
yading@10:     WEIGHT_PROLOGUE
yading@10:     sar         r2d, 1
yading@10:     WEIGHT_SETUP
yading@10:     lea         r3, [r1*2]
yading@10: .nextrow:
yading@10:     WEIGHT_OP    0, r1
yading@10:     movh      [r0], m5
yading@10:     movhps [r0+r1], m5
yading@10:     add         r0, r3
yading@10:     dec         r2d
yading@10:     jnz .nextrow
yading@10:     REP_RET
yading@10: %endmacro
yading@10: 
yading@10: INIT_XMM sse2
yading@10: WEIGHT_FUNC_HALF_MM
yading@10: INIT_XMM sse4
yading@10: WEIGHT_FUNC_HALF_MM
yading@10: 
yading@10: 
yading@10: ;-----------------------------------------------------------------------------
yading@10: ; void h264_biweight(uint8_t *dst, uint8_t *src, int stride, int height,
yading@10: ;                    int log2_denom, int weightd, int weights, int offset);
yading@10: ;-----------------------------------------------------------------------------
yading@10: %if ARCH_X86_32
yading@10: DECLARE_REG_TMP 3
yading@10: %else
yading@10: DECLARE_REG_TMP 7
yading@10: %endif
yading@10: 
yading@10: %macro BIWEIGHT_PROLOGUE 0
yading@10: .prologue:
yading@10:     PROLOGUE 0,8,8
yading@10:     movifnidn  r0, r0mp
yading@10:     movifnidn  r1, r1mp
yading@10:     movifnidn r2d, r2m
yading@10:     movifnidn r5d, r5m
yading@10:     movifnidn r6d, r6m
yading@10:     movifnidn t0d, r7m
yading@10: %endmacro
yading@10: 
yading@10: %macro BIWEIGHT_SETUP 0
yading@10:     lea        t0, [t0*4+1] ; (offset<<2)+1
yading@10:     or         t0, 1
yading@10:     shl        r6, 16
yading@10:     or         r5, r6
yading@10:     movd       m4, r5d      ; weightd | weights
yading@10:     movd       m5, t0d      ; (offset+1)|1
yading@10:     movd       m6, r4m      ; log2_denom
yading@10:     pslld      m5, m6       ; (((offset<<2)+1)|1)<<log2_denom
yading@10:     paddd      m6, [sq_1]
yading@10:     pshufd     m4, m4, 0
yading@10:     pshufd     m5, m5, 0
yading@10:     mova       m3, [pw_pixel_max]
yading@10:     movifnidn r3d, r3m
yading@10: %if notcpuflag(sse4)
yading@10:     pxor       m7, m7
yading@10: %endif
yading@10: %endmacro
yading@10: 
yading@10: %macro BIWEIGHT 1-2
yading@10: %if %0==1
yading@10:     mova       m0, [r0+%1]
yading@10:     mova       m1, [r1+%1]
yading@10:     punpckhwd  m2, m0, m1
yading@10:     punpcklwd  m0, m1
yading@10: %else
yading@10:     movq       m0, [r0+%1]
yading@10:     movq       m1, [r1+%1]
yading@10:     punpcklwd  m0, m1
yading@10:     movq       m2, [r0+%2]
yading@10:     movq       m1, [r1+%2]
yading@10:     punpcklwd  m2, m1
yading@10: %endif
yading@10:     pmaddwd    m0, m4
yading@10:     pmaddwd    m2, m4
yading@10:     paddd      m0, m5
yading@10:     paddd      m2, m5
yading@10:     psrad      m0, m6
yading@10:     psrad      m2, m6
yading@10: %if cpuflag(sse4)
yading@10:     packusdw   m0, m2
yading@10:     pminsw     m0, m3
yading@10: %else
yading@10:     packssdw   m0, m2
yading@10:     CLIPW      m0, m7, m3
yading@10: %endif
yading@10: %endmacro
yading@10: 
yading@10: %macro BIWEIGHT_FUNC_DBL 0
yading@10: cglobal h264_biweight_16_10
yading@10:     BIWEIGHT_PROLOGUE
yading@10:     BIWEIGHT_SETUP
yading@10: .nextrow:
yading@10:     BIWEIGHT   0
yading@10:     mova [r0   ], m0
yading@10:     BIWEIGHT  16
yading@10:     mova [r0+16], m0
yading@10:     add       r0, r2
yading@10:     add       r1, r2
yading@10:     dec       r3d
yading@10:     jnz .nextrow
yading@10:     REP_RET
yading@10: %endmacro
yading@10: 
yading@10: INIT_XMM sse2
yading@10: BIWEIGHT_FUNC_DBL
yading@10: INIT_XMM sse4
yading@10: BIWEIGHT_FUNC_DBL
yading@10: 
yading@10: %macro BIWEIGHT_FUNC 0
yading@10: cglobal h264_biweight_8_10
yading@10:     BIWEIGHT_PROLOGUE
yading@10:     BIWEIGHT_SETUP
yading@10: .nextrow:
yading@10:     BIWEIGHT  0
yading@10:     mova   [r0], m0
yading@10:     add      r0, r2
yading@10:     add      r1, r2
yading@10:     dec      r3d
yading@10:     jnz .nextrow
yading@10:     REP_RET
yading@10: %endmacro
yading@10: 
yading@10: INIT_XMM sse2
yading@10: BIWEIGHT_FUNC
yading@10: INIT_XMM sse4
yading@10: BIWEIGHT_FUNC
yading@10: 
yading@10: %macro BIWEIGHT_FUNC_HALF 0
yading@10: cglobal h264_biweight_4_10
yading@10:     BIWEIGHT_PROLOGUE
yading@10:     BIWEIGHT_SETUP
yading@10:     sar        r3d, 1
yading@10:     lea        r4, [r2*2]
yading@10: .nextrow:
yading@10:     BIWEIGHT     0, r2
yading@10:     movh   [r0   ], m0
yading@10:     movhps [r0+r2], m0
yading@10:     add         r0, r4
yading@10:     add         r1, r4
yading@10:     dec         r3d
yading@10:     jnz .nextrow
yading@10:     REP_RET
yading@10: %endmacro
yading@10: 
yading@10: INIT_XMM sse2
yading@10: BIWEIGHT_FUNC_HALF
yading@10: INIT_XMM sse4
yading@10: BIWEIGHT_FUNC_HALF