;/************************************************************************** ; * ; * XVID MPEG-4 VIDEO CODEC ; * 3dnow 8x8 block-based halfpel interpolation ; * ; * This program is free software; you can redistribute it and/or modify ; * it under the terms of the GNU General Public License as published by ; * the Free Software Foundation; either version 2 of the License, or ; * (at your option) any later version. ; * ; * This program 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 General Public License for more details. ; * ; * You should have received a copy of the GNU General Public License ; * along with this program; if not, write to the Free Software ; * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. ; * ; *************************************************************************/ ;/************************************************************************** ; * ; * History: ; * ; * 06.07.2002 applied Skal's XMM changes to the 3dnow code (Isibaar) ; * 04.02.2002 additional 3dnow optimizations (Isibaar) ; * 22.12.2001 inital version; (c)2001 peter ross ; * ; *************************************************************************/ bits 32 %macro cglobal 1 %ifdef PREFIX global _%1 %define %1 _%1 %else global %1 %endif %endmacro section .data align 16 mmx_one times 8 db 1 section .text cglobal interpolate8x8_halfpel_h_3dn cglobal interpolate8x8_halfpel_v_3dn cglobal interpolate8x8_halfpel_hv_3dn ;=========================================================================== ; ; void interpolate8x8_halfpel_h_3dn(uint8_t * const dst, ; const uint8_t * const src, ; const uint32_t stride, ; const uint32_t rounding); ; ;=========================================================================== %macro COPY_H_3DN_RND0 0 movq mm0, [eax] pavgusb mm0, [eax+1] movq mm1, [eax+edx] pavgusb mm1, [eax+edx+1] lea eax,[eax+2*edx] movq [ecx],mm0 movq [ecx+edx],mm1 %endmacro %macro COPY_H_3DN_RND1 0 movq mm0, [eax] movq mm1, [eax+edx] movq mm4, mm0 movq mm5, mm1 movq mm2, [eax+1] movq mm3, [eax+edx+1] pavgusb mm0, mm2 pxor mm2, mm4 pavgusb mm1, mm3 lea eax,[eax+2*edx] pxor mm3, mm5 pand mm2, mm7 pand mm3, mm7 psubb mm0, mm2 movq [ecx], mm0 psubb mm1, mm3 movq [ecx+edx], mm1 %endmacro align 16 interpolate8x8_halfpel_h_3dn: mov eax, [esp+16] ; rounding mov ecx, [esp+ 4] ; Dst test eax,eax mov eax, [esp+ 8] ; Src mov edx, [esp+12] ; stride jnz near .rounding1 COPY_H_3DN_RND0 lea ecx,[ecx+2*edx] COPY_H_3DN_RND0 lea ecx,[ecx+2*edx] COPY_H_3DN_RND0 lea ecx,[ecx+2*edx] COPY_H_3DN_RND0 ret .rounding1 ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 movq mm7, [mmx_one] COPY_H_3DN_RND1 lea ecx, [ecx+2*edx] COPY_H_3DN_RND1 lea ecx,[ecx+2*edx] COPY_H_3DN_RND1 lea ecx,[ecx+2*edx] COPY_H_3DN_RND1 ret ;=========================================================================== ; ; void interpolate8x8_halfpel_v_3dn(uint8_t * const dst, ; const uint8_t * const src, ; const uint32_t stride, ; const uint32_t rounding); ; ;=========================================================================== %macro COPY_V_3DN_RND0 0 movq mm0, [eax] movq mm1, [eax+edx] pavgusb mm0, mm1 pavgusb mm1, [eax+2*edx] lea eax,[eax+2*edx] movq [ecx],mm0 movq [ecx+edx],mm1 %endmacro %macro COPY_V_3DN_RND1 0 movq mm0, mm2 movq mm1, [eax] movq mm2, [eax+edx] lea eax,[eax+2*edx] movq mm4, mm0 movq mm5, mm1 pavgusb mm0, mm1 pxor mm4, mm1 pavgusb mm1, mm2 pxor mm5, mm2 pand mm4, mm7 ; lsb's of (i^j)... pand mm5, mm7 ; lsb's of (i^j)... psubb mm0, mm4 ; ...are substracted from result of pavgusb movq [ecx], mm0 psubb mm1, mm5 ; ...are substracted from result of pavgusb movq [ecx+edx], mm1 %endmacro align 16 interpolate8x8_halfpel_v_3dn: mov eax, [esp+16] ; rounding mov ecx, [esp+ 4] ; Dst test eax,eax mov eax, [esp+ 8] ; Src mov edx, [esp+12] ; stride ; we process 2 line at a time jnz near .rounding1 COPY_V_3DN_RND0 lea ecx, [ecx+2*edx] COPY_V_3DN_RND0 lea ecx, [ecx+2*edx] COPY_V_3DN_RND0 lea ecx, [ecx+2*edx] COPY_V_3DN_RND0 ret .rounding1 ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 movq mm7, [mmx_one] movq mm2, [eax] ; loop invariant add eax, edx COPY_V_3DN_RND1 lea ecx,[ecx+2*edx] COPY_V_3DN_RND1 lea ecx,[ecx+2*edx] COPY_V_3DN_RND1 lea ecx,[ecx+2*edx] COPY_V_3DN_RND1 ret ;=========================================================================== ; ; void interpolate8x8_halfpel_hv_3dn(uint8_t * const dst, ; const uint8_t * const src, ; const uint32_t stride, ; const uint32_t rounding); ; ; ;=========================================================================== ; The trick is to correct the result of 'pavgusb' with some combination of the ; lsb's of the 4 input values i,j,k,l, and their intermediate 'pavgusb' (s and t). ; The boolean relations are: ; (i+j+k+l+3)/4 = (s+t+1)/2 - (ij&kl)&st ; (i+j+k+l+2)/4 = (s+t+1)/2 - (ij|kl)&st ; (i+j+k+l+1)/4 = (s+t+1)/2 - (ij&kl)|st ; (i+j+k+l+0)/4 = (s+t+1)/2 - (ij|kl)|st ; with s=(i+j+1)/2, t=(k+l+1)/2, ij = i^j, kl = k^l, st = s^t. ; Moreover, we process 2 lines at a times, for better overlapping (~15% faster). %macro COPY_HV_3DN_RND0 0 lea eax,[eax+edx] movq mm0, [eax] movq mm1, [eax+1] movq mm6, mm0 pavgusb mm0, mm1 ; mm0=(j+k+1)/2. preserved for next step lea eax,[eax+edx] pxor mm1, mm6 ; mm1=(j^k). preserved for next step por mm3, mm1 ; ij |= jk movq mm6, mm2 pxor mm6, mm0 ; mm6 = s^t pand mm3, mm6 ; (ij|jk) &= st pavgusb mm2, mm0 ; mm2 = (s+t+1)/2 pand mm3, mm7 ; mask lsb psubb mm2, mm3 ; apply. movq [ecx], mm2 movq mm2, [eax] movq mm3, [eax+1] movq mm6, mm2 pavgusb mm2, mm3 ; preserved for next iteration lea ecx,[ecx+edx] pxor mm3, mm6 ; preserved for next iteration por mm1, mm3 movq mm6, mm0 pxor mm6, mm2 pand mm1, mm6 pavgusb mm0, mm2 pand mm1, mm7 psubb mm0, mm1 movq [ecx], mm0 %endmacro %macro COPY_HV_3DN_RND1 0 lea eax,[eax+edx] movq mm0, [eax] movq mm1, [eax+1] movq mm6, mm0 pavgusb mm0, mm1 ; mm0=(j+k+1)/2. preserved for next step lea eax,[eax+edx] pxor mm1, mm6 ; mm1=(j^k). preserved for next step pand mm3, mm1 movq mm6, mm2 pxor mm6, mm0 por mm3, mm6 pavgusb mm2, mm0 pand mm3, mm7 psubb mm2, mm3 movq [ecx], mm2 movq mm2, [eax] movq mm3, [eax+1] movq mm6, mm2 pavgusb mm2, mm3 ; preserved for next iteration lea ecx,[ecx+edx] pxor mm3, mm6 ; preserved for next iteration pand mm1, mm3 movq mm6, mm0 pxor mm6, mm2 por mm1, mm6 pavgusb mm0, mm2 pand mm1, mm7 psubb mm0, mm1 movq [ecx], mm0 %endmacro align 16 interpolate8x8_halfpel_hv_3dn mov eax, [esp+16] ; rounding mov ecx, [esp+ 4] ; Dst test eax,eax mov eax, [esp+ 8] ; Src mov edx, [esp+12] ; stride movq mm7, [mmx_one] ; loop invariants: mm2=(i+j+1)/2 and mm3= i^j movq mm2, [eax] movq mm3, [eax+1] movq mm6, mm2 pavgusb mm2, mm3 pxor mm3, mm6 ; mm2/mm3 ready jnz near .rounding1 COPY_HV_3DN_RND0 add ecx, edx COPY_HV_3DN_RND0 add ecx, edx COPY_HV_3DN_RND0 add ecx, edx COPY_HV_3DN_RND0 ret .rounding1 COPY_HV_3DN_RND1 add ecx, edx COPY_HV_3DN_RND1 add ecx, edx COPY_HV_3DN_RND1 add ecx, edx COPY_HV_3DN_RND1 ret