--- interpolate8x8.h 2002/11/17 00:20:30 1.7 +++ interpolate8x8.h 2004/08/10 21:58:55 1.13 @@ -1,59 +1,31 @@ /***************************************************************************** * * XVID MPEG-4 VIDEO CODEC - * - 8x8 block-based halfpel interpolation - headers + * - Interpolation related header - * - * Copyright(C) 2002 Peter Ross + * Copyright(C) 2001-2003 Peter Ross * - * This file is part of XviD, a free MPEG-4 video encoder/decoder - * - * XviD 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 + * 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 + * 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 + * along with this program ; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * - * Under section 8 of the GNU General Public License, the copyright - * holders of XVID explicitly forbid distribution in the following - * countries: - * - * - Japan - * - United States of America - * - * Linking XviD statically or dynamically with other modules is making a - * combined work based on XviD. Thus, the terms and conditions of the - * GNU General Public License cover the whole combination. - * - * As a special exception, the copyright holders of XviD give you - * permission to link XviD with independent modules that communicate with - * XviD solely through the VFW1.1 and DShow interfaces, regardless of the - * license terms of these independent modules, and to copy and distribute - * the resulting combined work under terms of your choice, provided that - * every copy of the combined work is accompanied by a complete copy of - * the source code of XviD (the version of XviD used to produce the - * combined work), being distributed under the terms of the GNU General - * Public License plus this exception. An independent module is a module - * which is not derived from or based on XviD. - * - * Note that people who make modified versions of XviD are not obligated - * to grant this special exception for their modified versions; it is - * their choice whether to do so. The GNU General Public License gives - * permission to release a modified version without this exception; this - * exception also makes it possible to release a modified version which - * carries forward this exception. - * - * $Id: interpolate8x8.h,v 1.7 2002/11/17 00:20:30 edgomez Exp $ + * $Id: interpolate8x8.h,v 1.13 2004/08/10 21:58:55 edgomez Exp $ * ****************************************************************************/ +#ifndef _INTERPOLATE8X8_H_ +#define _INTERPOLATE8X8_H_ + #include "../utils/mem_transfer.h" typedef void (INTERPOLATE8X8) (uint8_t * const dst, @@ -62,44 +34,190 @@ const uint32_t rounding); typedef INTERPOLATE8X8 *INTERPOLATE8X8_PTR; +typedef void (INTERPOLATE8X8_AVG2) (uint8_t *dst, + const uint8_t *src1, + const uint8_t *src2, + const uint32_t stride, + const uint32_t rounding, + const uint32_t height); +typedef INTERPOLATE8X8_AVG2 *INTERPOLATE8X8_AVG2_PTR; + +typedef void (INTERPOLATE8X8_AVG4) (uint8_t *dst, + const uint8_t *src1, + const uint8_t *src2, + const uint8_t *src3, + const uint8_t *src4, + const uint32_t stride, + const uint32_t rounding); +typedef INTERPOLATE8X8_AVG4 *INTERPOLATE8X8_AVG4_PTR; + +typedef void (INTERPOLATE_LOWPASS) (uint8_t *dst, + uint8_t *src, + int32_t stride, + int32_t rounding); + +typedef INTERPOLATE_LOWPASS *INTERPOLATE_LOWPASS_PTR; + +typedef void (INTERPOLATE_LOWPASS_HV) (uint8_t *dst1, + uint8_t *dst2, + uint8_t *src, + int32_t stride, + int32_t rounding); + +typedef INTERPOLATE_LOWPASS_HV *INTERPOLATE_LOWPASS_HV_PTR; + +typedef void (INTERPOLATE8X8_6TAP_LOWPASS) (uint8_t *dst, + uint8_t *src, + int32_t stride, + int32_t rounding); + +typedef INTERPOLATE8X8_6TAP_LOWPASS *INTERPOLATE8X8_6TAP_LOWPASS_PTR; + +/* These function do: dst = interpolate(src) */ extern INTERPOLATE8X8_PTR interpolate8x8_halfpel_h; extern INTERPOLATE8X8_PTR interpolate8x8_halfpel_v; extern INTERPOLATE8X8_PTR interpolate8x8_halfpel_hv; +/* These functions do: dst = (dst+interpolate(src) + 1)/2 + * Suitable for direct/interpolated bvop prediction block + * building w/o the need for intermediate interpolated result + * storing/reading + * NB: the rounding applies to the interpolation, but not + * the averaging step which will always use rounding=0 */ +extern INTERPOLATE8X8_PTR interpolate8x8_halfpel_add; +extern INTERPOLATE8X8_PTR interpolate8x8_halfpel_h_add; +extern INTERPOLATE8X8_PTR interpolate8x8_halfpel_v_add; +extern INTERPOLATE8X8_PTR interpolate8x8_halfpel_hv_add; + +extern INTERPOLATE8X8_AVG2_PTR interpolate8x8_avg2; +extern INTERPOLATE8X8_AVG4_PTR interpolate8x8_avg4; + +extern INTERPOLATE_LOWPASS_PTR interpolate8x8_lowpass_h; +extern INTERPOLATE_LOWPASS_PTR interpolate8x8_lowpass_v; + +extern INTERPOLATE_LOWPASS_PTR interpolate16x16_lowpass_h; +extern INTERPOLATE_LOWPASS_PTR interpolate16x16_lowpass_v; + +extern INTERPOLATE_LOWPASS_HV_PTR interpolate8x8_lowpass_hv; +extern INTERPOLATE_LOWPASS_HV_PTR interpolate16x16_lowpass_hv; + +extern INTERPOLATE8X8_6TAP_LOWPASS_PTR interpolate8x8_6tap_lowpass_h; +extern INTERPOLATE8X8_6TAP_LOWPASS_PTR interpolate8x8_6tap_lowpass_v; + INTERPOLATE8X8 interpolate8x8_halfpel_h_c; INTERPOLATE8X8 interpolate8x8_halfpel_v_c; INTERPOLATE8X8 interpolate8x8_halfpel_hv_c; +INTERPOLATE8X8 interpolate8x8_halfpel_add_c; +INTERPOLATE8X8 interpolate8x8_halfpel_h_add_c; +INTERPOLATE8X8 interpolate8x8_halfpel_v_add_c; +INTERPOLATE8X8 interpolate8x8_halfpel_hv_add_c; +#ifdef ARCH_IS_IA32 INTERPOLATE8X8 interpolate8x8_halfpel_h_mmx; INTERPOLATE8X8 interpolate8x8_halfpel_v_mmx; INTERPOLATE8X8 interpolate8x8_halfpel_hv_mmx; +INTERPOLATE8X8 interpolate8x8_halfpel_add_mmx; +INTERPOLATE8X8 interpolate8x8_halfpel_h_add_mmx; +INTERPOLATE8X8 interpolate8x8_halfpel_v_add_mmx; +INTERPOLATE8X8 interpolate8x8_halfpel_hv_add_mmx; + INTERPOLATE8X8 interpolate8x8_halfpel_h_xmm; INTERPOLATE8X8 interpolate8x8_halfpel_v_xmm; INTERPOLATE8X8 interpolate8x8_halfpel_hv_xmm; +INTERPOLATE8X8 interpolate8x8_halfpel_add_xmm; +INTERPOLATE8X8 interpolate8x8_halfpel_h_add_xmm; +INTERPOLATE8X8 interpolate8x8_halfpel_v_add_xmm; +INTERPOLATE8X8 interpolate8x8_halfpel_hv_add_xmm; + INTERPOLATE8X8 interpolate8x8_halfpel_h_3dn; INTERPOLATE8X8 interpolate8x8_halfpel_v_3dn; INTERPOLATE8X8 interpolate8x8_halfpel_hv_3dn; +INTERPOLATE8X8 interpolate8x8_halfpel_h_3dne; +INTERPOLATE8X8 interpolate8x8_halfpel_v_3dne; +INTERPOLATE8X8 interpolate8x8_halfpel_hv_3dne; +#endif + +#ifdef ARCH_IS_IA64 INTERPOLATE8X8 interpolate8x8_halfpel_h_ia64; INTERPOLATE8X8 interpolate8x8_halfpel_v_ia64; INTERPOLATE8X8 interpolate8x8_halfpel_hv_ia64; +#endif -void interpolate8x8_lowpass_h(uint8_t *dst, uint8_t *src, int32_t dst_stride, int32_t src_stride, int32_t rounding); -void interpolate8x8_lowpass_v(uint8_t *dst, uint8_t *src, int32_t dst_stride, int32_t src_stride, int32_t rounding); -void interpolate8x8_lowpass_hv(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int32_t dst1_stride, int32_t dst2_stride, int32_t src_stride, int32_t rounding); -void interpolate8x8_bilinear2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int32_t dst_stride, int32_t src_stride, int32_t rounding); -void interpolate8x8_bilinear4(uint8_t *dst, uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4, int32_t stride, int32_t rounding); +#ifdef ARCH_IS_PPC +INTERPOLATE8X8 interpolate8x8_halfpel_h_altivec_c; +INTERPOLATE8X8 interpolate8x8_halfpel_v_altivec_c; +INTERPOLATE8X8 interpolate8x8_halfpel_hv_altivec_c; +#endif + +INTERPOLATE8X8_AVG2 interpolate8x8_avg2_c; +INTERPOLATE8X8_AVG4 interpolate8x8_avg4_c; + +#ifdef ARCH_IS_IA32 +INTERPOLATE8X8_AVG2 interpolate8x8_avg2_mmx; +INTERPOLATE8X8_AVG4 interpolate8x8_avg4_mmx; +#endif + +#ifdef ARCH_IS_PPC +INTERPOLATE8X8_AVG2 interpolate8x8_avg2_altivec_c; +INTERPOLATE8X8_AVG4 interpolate8x8_avg4_altivec_c; +#endif + +INTERPOLATE_LOWPASS interpolate8x8_lowpass_h_c; +INTERPOLATE_LOWPASS interpolate8x8_lowpass_v_c; + +INTERPOLATE_LOWPASS interpolate16x16_lowpass_h_c; +INTERPOLATE_LOWPASS interpolate16x16_lowpass_v_c; + +INTERPOLATE_LOWPASS_HV interpolate8x8_lowpass_hv_c; +INTERPOLATE_LOWPASS_HV interpolate16x16_lowpass_hv_c; + +INTERPOLATE8X8_6TAP_LOWPASS interpolate8x8_6tap_lowpass_h_c; +INTERPOLATE8X8_6TAP_LOWPASS interpolate8x8_6tap_lowpass_v_c; + +#ifdef ARCH_IS_IA32 +INTERPOLATE8X8_6TAP_LOWPASS interpolate8x8_6tap_lowpass_h_mmx; +INTERPOLATE8X8_6TAP_LOWPASS interpolate8x8_6tap_lowpass_v_mmx; +#endif + +#ifdef ARCH_IS_PPC +INTERPOLATE8X8_6TAP_LOWPASS interpolate8x8_6tap_lowpass_h_altivec_c; +#endif -void interpolate8x8_c(uint8_t * const dst, - const uint8_t * const src, +static __inline void +interpolate8x8_switch(uint8_t * const cur, + const uint8_t * const refn, const uint32_t x, const uint32_t y, - const uint32_t stride); + const int32_t dx, + const int dy, + const uint32_t stride, + const uint32_t rounding) +{ + + const uint8_t * const src = refn + (int)((y + (dy>>1)) * stride + x + (dx>>1)); + uint8_t * const dst = cur + (int)(y * stride + x); + + switch (((dx & 1) << 1) + (dy & 1)) { /* ((dx%2)?2:0)+((dy%2)?1:0) */ + case 0: + transfer8x8_copy(dst, src, stride); + break; + case 1: + interpolate8x8_halfpel_v(dst, src, stride, rounding); + break; + case 2: + interpolate8x8_halfpel_h(dst, src, stride, rounding); + break; + default: + interpolate8x8_halfpel_hv(dst, src, stride, rounding); + break; + } +} static __inline void -interpolate8x8_switch(uint8_t * const cur, +interpolate8x8_add_switch(uint8_t * const cur, const uint8_t * const refn, const uint32_t x, const uint32_t y, @@ -108,158 +226,117 @@ const uint32_t stride, const uint32_t rounding) { - int32_t ddx, ddy; - switch (((dx & 1) << 1) + (dy & 1)) // ((dx%2)?2:0)+((dy%2)?1:0) - { + const uint8_t * const src = refn + (int)((y + (dy>>1)) * stride + x + (dx>>1)); + uint8_t * const dst = cur + (int)(y * stride + x); + + switch (((dx & 1) << 1) + (dy & 1)) { /* ((dx%2)?2:0)+((dy%2)?1:0) */ case 0: - ddx = dx / 2; - ddy = dy / 2; - transfer8x8_copy(cur + y * stride + x, - refn + (int)((y + ddy) * stride + x + ddx), stride); + interpolate8x8_halfpel_add(dst, src, stride, rounding); break; - case 1: - ddx = dx / 2; - ddy = (dy - 1) / 2; - interpolate8x8_halfpel_v(cur + y * stride + x, - refn + (int)((y + ddy) * stride + x + ddx), stride, - rounding); + interpolate8x8_halfpel_v_add(dst, src, stride, rounding); break; - case 2: - ddx = (dx - 1) / 2; - ddy = dy / 2; - interpolate8x8_halfpel_h(cur + y * stride + x, - refn + (int)((y + ddy) * stride + x + ddx), stride, - rounding); + interpolate8x8_halfpel_h_add(dst, src, stride, rounding); break; - default: - ddx = (dx - 1) / 2; - ddy = (dy - 1) / 2; - interpolate8x8_halfpel_hv(cur + y * stride + x, - refn + (int)((y + ddy) * stride + x + ddx), stride, - rounding); + interpolate8x8_halfpel_hv_add(dst, src, stride, rounding); break; } } +static __inline void +interpolate16x16_switch(uint8_t * const cur, + const uint8_t * const refn, + const uint32_t x, + const uint32_t y, + const int32_t dx, + const int dy, + const uint32_t stride, + const uint32_t rounding) +{ + interpolate8x8_switch(cur, refn, x, y, dx, dy, stride, rounding); + interpolate8x8_switch(cur, refn, x+8, y, dx, dy, stride, rounding); + interpolate8x8_switch(cur, refn, x, y+8, dx, dy, stride, rounding); + interpolate8x8_switch(cur, refn, x+8, y+8, dx, dy, stride, rounding); +} -static __inline void interpolate8x8_quarterpel(uint8_t * const cur, - uint8_t * const refn, - const uint32_t x, const uint32_t y, - const int32_t dx, const int dy, - const uint32_t stride, - const uint32_t rounding) +static __inline void +interpolate16x16_add_switch(uint8_t * const cur, + const uint8_t * const refn, + const uint32_t x, + const uint32_t y, + const int32_t dx, + const int dy, + const uint32_t stride, + const uint32_t rounding) { - const int32_t xRef = x*4 + dx; - const int32_t yRef = y*4 + dy; + interpolate8x8_add_switch(cur, refn, x, y, dx, dy, stride, rounding); + interpolate8x8_add_switch(cur, refn, x+8, y, dx, dy, stride, rounding); + interpolate8x8_add_switch(cur, refn, x, y+8, dx, dy, stride, rounding); + interpolate8x8_add_switch(cur, refn, x+8, y+8, dx, dy, stride, rounding); +} - uint8_t *src, *dst; - int32_t x_int, y_int, x_frac, y_frac; +static __inline void +interpolate32x32_switch(uint8_t * const cur, + const uint8_t * const refn, + const uint32_t x, + const uint32_t y, + const int32_t dx, + const int dy, + const uint32_t stride, + const uint32_t rounding) +{ + interpolate16x16_switch(cur, refn, x, y, dx, dy, stride, rounding); + interpolate16x16_switch(cur, refn, x+16, y, dx, dy, stride, rounding); + interpolate16x16_switch(cur, refn, x, y+16, dx, dy, stride, rounding); + interpolate16x16_switch(cur, refn, x+16, y+16, dx, dy, stride, rounding); +} - uint8_t halfpel_h[72]; - uint8_t halfpel_v[64]; - uint8_t halfpel_hv[64]; - - x_int = xRef/4; - if (xRef < 0 && xRef % 4) - x_int--; - - x_frac = xRef - (4*x_int); - - y_int = yRef/4; - if (yRef < 0 && yRef % 4) - y_int--; - - y_frac = yRef - (4*y_int); - - src = refn + y_int * stride + x_int; - dst = cur + y * stride + x; - - switch((y_frac << 2) | (x_frac)) { +static __inline void +interpolate32x32_add_switch(uint8_t * const cur, + const uint8_t * const refn, + const uint32_t x, + const uint32_t y, + const int32_t dx, + const int dy, + const uint32_t stride, + const uint32_t rounding) +{ + interpolate16x16_add_switch(cur, refn, x, y, dx, dy, stride, rounding); + interpolate16x16_add_switch(cur, refn, x+16, y, dx, dy, stride, rounding); + interpolate16x16_add_switch(cur, refn, x, y+16, dx, dy, stride, rounding); + interpolate16x16_add_switch(cur, refn, x+16, y+16, dx, dy, stride, rounding); +} - case 0: - transfer8x8_copy(dst, src, stride); - break; +static __inline uint8_t * +interpolate8x8_switch2(uint8_t * const buffer, + const uint8_t * const refn, + const int x, + const int y, + const int dx, + const int dy, + const uint32_t stride, + const uint32_t rounding) +{ + + const uint8_t * const src = refn + (int)((y + (dy>>1)) * stride + x + (dx>>1)); + switch (((dx & 1) << 1) + (dy & 1)) { /* ((dx%2)?2:0)+((dy%2)?1:0) */ + case 0: + return (uint8_t *)src; case 1: - interpolate8x8_lowpass_h(halfpel_h, src, 8, stride, rounding); - interpolate8x8_bilinear2(dst, src, halfpel_h, stride, stride, rounding); - break; - + interpolate8x8_halfpel_v(buffer, src, stride, rounding); + break; case 2: - interpolate8x8_lowpass_h(dst, src, stride, stride, rounding); - break; - - case 3: - interpolate8x8_lowpass_h(halfpel_h, src, 8, stride, rounding); - interpolate8x8_bilinear2(dst, src+1, halfpel_h, stride, stride, rounding); - break; - - case 4: - interpolate8x8_lowpass_v(halfpel_v, src, 8, stride, rounding); - interpolate8x8_bilinear2(dst, src, halfpel_v, stride, stride, rounding); - break; - - case 5: - interpolate8x8_lowpass_v(halfpel_v, src, 8, stride, rounding); - interpolate8x8_lowpass_hv(halfpel_hv, halfpel_h, src, 8, 8, stride, rounding); - interpolate8x8_bilinear4(dst, src, halfpel_h, halfpel_v, halfpel_hv, stride, rounding); - break; - - case 6: - interpolate8x8_lowpass_hv(halfpel_hv, halfpel_h, src, 8, 8, stride, rounding); - interpolate8x8_bilinear2(dst, halfpel_h, halfpel_hv, stride, 8, 1-rounding); - break; - - case 7: - interpolate8x8_lowpass_v(halfpel_v, src+1, 8, stride, 16-rounding); - interpolate8x8_lowpass_hv(halfpel_hv, halfpel_h, src, 8, 8, stride, rounding); - interpolate8x8_bilinear4(dst, src+1, halfpel_h, halfpel_v, halfpel_hv, stride, rounding); - break; - - case 8: - interpolate8x8_lowpass_v(dst, src, stride, stride, rounding); - break; - - case 9: - interpolate8x8_lowpass_v(halfpel_v, src, 8, stride, 16-rounding); - interpolate8x8_lowpass_hv(halfpel_hv, halfpel_h, src, 8, 8, stride, rounding); - interpolate8x8_bilinear2(dst, halfpel_v, halfpel_hv, stride, 8, rounding); - break; - - case 10: - interpolate8x8_lowpass_hv(dst, halfpel_h, src, stride, 8, stride, rounding); - break; - - case 11: - interpolate8x8_lowpass_v(halfpel_v, src+1, 8, stride, 16-rounding); - interpolate8x8_lowpass_hv(halfpel_hv, halfpel_h, src, 8, 8, stride, rounding); - interpolate8x8_bilinear2(dst, halfpel_v, halfpel_hv, stride, 8, rounding); - break; - - case 12: - interpolate8x8_lowpass_v(halfpel_v, src, 8, stride, rounding); - interpolate8x8_bilinear2(dst, src+stride, halfpel_v, stride, stride, rounding); - break; - - case 13: - interpolate8x8_lowpass_v(halfpel_v, src, 8, stride, rounding); - interpolate8x8_lowpass_hv(halfpel_hv, halfpel_h, src, 8, 8, stride, rounding); - interpolate8x8_bilinear4(dst, src+stride, halfpel_h+8, halfpel_v, halfpel_hv, stride, rounding); - break; - - case 14: - interpolate8x8_lowpass_hv(halfpel_hv, halfpel_h, src, 8, 8, stride, rounding); - interpolate8x8_bilinear2(dst, halfpel_h+8, halfpel_hv, stride, 8, rounding); - break; - - case 15: - interpolate8x8_lowpass_v(halfpel_v, src+1, 8, stride, rounding); - interpolate8x8_lowpass_hv(halfpel_hv, halfpel_h, src, 8, 8, stride, rounding); - interpolate8x8_bilinear4(dst, src+stride+1, halfpel_h+8, halfpel_v, halfpel_hv, stride, rounding); + interpolate8x8_halfpel_h(buffer, src, stride, rounding); + break; + default: + interpolate8x8_halfpel_hv(buffer, src, stride, rounding); break; } + return buffer; } + +#endif