From 85fa502b8d69f72eecee95fe3692344ae4035989 Mon Sep 17 00:00:00 2001 From: Guenter Bartsch Date: Wed, 5 Dec 2001 15:12:02 +0000 Subject: altivec support by Jeffrey W. Baker CVS patchset: 1171 CVS date: 2001/12/05 15:12:02 --- src/libmpeg2/Makefile.am | 3 +- src/libmpeg2/idct.c | 8 + src/libmpeg2/idct_altivec.c | 668 ++++++++++++ src/libmpeg2/motion_comp.c | 6 + src/libmpeg2/motion_comp_altivec.c | 2024 ++++++++++++++++++++++++++++++++++++ src/libmpeg2/mpeg2_internal.h | 10 + src/xine-utils/cpu_accel.c | 34 +- src/xine-utils/xineutils.h | 4 +- 8 files changed, 2737 insertions(+), 20 deletions(-) create mode 100644 src/libmpeg2/idct_altivec.c create mode 100644 src/libmpeg2/motion_comp_altivec.c (limited to 'src') diff --git a/src/libmpeg2/Makefile.am b/src/libmpeg2/Makefile.am index cf093d802..eec705b15 100644 --- a/src/libmpeg2/Makefile.am +++ b/src/libmpeg2/Makefile.am @@ -14,7 +14,8 @@ lib_LTLIBRARIES = xineplug_decode_mpeg2.la xineplug_decode_mpeg2_la_SOURCES = slice.c header.c stats.c idct.c \ motion_comp.c decode.c idct_mmx.c motion_comp_mmx.c \ - idct_mlib.c motion_comp_mlib.c xine_decoder.c + idct_mlib.c motion_comp_mlib.c idct_altivec.c \ + motion_comp_altivec.c xine_decoder.c xineplug_decode_mpeg2_la_LDFLAGS = -avoid-version -module noinst_HEADERS = vlc.h mpeg2.h mpeg2_internal.h diff --git a/src/libmpeg2/idct.c b/src/libmpeg2/idct.c index 2aaf6eaf9..0b77a62ed 100644 --- a/src/libmpeg2/idct.c +++ b/src/libmpeg2/idct.c @@ -84,6 +84,14 @@ void idct_init (void) idct_block_add = idct_block_add_mlib; } else #endif +#ifdef ENABLE_ALTIVEC + if (config.flags & MM_ACCEL_PPC_ALTIVEC) { + fprintf (stderr, "Using altivec for IDCT transform\n"); + idct_block_copy = idct_block_copy_altivec; + idct_block_add = idct_block_add_altivec; + idct_altivec_init (); + } else +#endif { int i; diff --git a/src/libmpeg2/idct_altivec.c b/src/libmpeg2/idct_altivec.c new file mode 100644 index 000000000..96be2ebc3 --- /dev/null +++ b/src/libmpeg2/idct_altivec.c @@ -0,0 +1,668 @@ +/* + * idct_altivec.c + * Copyright (C) 2000-2001 Michel Lespinasse + * Copyright (C) 1999-2000 Aaron Holtzman + * + * This file is part of mpeg2dec, a free MPEG-2 video stream decoder. + * See http://libmpeg2.sourceforge.net/ for updates. + * + * mpeg2dec 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. + * + * mpeg2dec 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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +#ifndef __ALTIVEC__ + +#include "config.h" + +#ifdef ARCH_PPC +#ifdef ENABLE_ALTIVEC + +#include + +#include "mpeg2_internal.h" +#include "xineutils.h" + +static int16_t constants[5][8] ATTR_ALIGN(16) = { + {23170, 13573, 6518, 21895, -23170, -21895, 32, 31}, + {16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725}, + {22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521}, + {21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692}, + {19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722} +}; + +/* + * The asm code is generated with: + * + * gcc-2.95 -fvec -D__ALTIVEC__ -O9 -fomit-frame-pointer -mregnames -S + * idct_altivec.c + * + * awk '{args=""; len=split ($2, arg, ","); + * for (i=1; i<=len; i++) { a=arg[i]; if (i> 3) | ((j & 7) << 3); + j = scan_alt[i]; + scan_alt[i] = (j >> 3) | ((j & 7) << 3); + } +} + +#endif /* ENABLE_ALTIVEC */ +#endif /* ARCH_PPC */ + +#else /* __ALTIVEC__ */ + +#define vector_s16_t vector signed short +#define vector_u16_t vector unsigned short +#define vector_s8_t vector signed char +#define vector_u8_t vector unsigned char +#define vector_s32_t vector signed int +#define vector_u32_t vector unsigned int + +#define IDCT_HALF \ + /* 1st stage */ \ + t1 = vec_mradds (a1, vx7, vx1 ); \ + t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7)); \ + t7 = vec_mradds (a2, vx5, vx3); \ + t3 = vec_mradds (ma2, vx3, vx5); \ + \ + /* 2nd stage */ \ + t5 = vec_adds (vx0, vx4); \ + t0 = vec_subs (vx0, vx4); \ + t2 = vec_mradds (a0, vx6, vx2); \ + t4 = vec_mradds (a0, vx2, vec_subs (zero,vx6)); \ + t6 = vec_adds (t8, t3); \ + t3 = vec_subs (t8, t3); \ + t8 = vec_subs (t1, t7); \ + t1 = vec_adds (t1, t7); \ + \ + /* 3rd stage */ \ + t7 = vec_adds (t5, t2); \ + t2 = vec_subs (t5, t2); \ + t5 = vec_adds (t0, t4); \ + t0 = vec_subs (t0, t4); \ + t4 = vec_subs (t8, t3); \ + t3 = vec_adds (t8, t3); \ + \ + /* 4th stage */ \ + vy0 = vec_adds (t7, t1); \ + vy7 = vec_subs (t7, t1); \ + vy1 = vec_mradds (c4, t3, t5); \ + vy6 = vec_mradds (mc4, t3, t5); \ + vy2 = vec_mradds (c4, t4, t0); \ + vy5 = vec_mradds (mc4, t4, t0); \ + vy3 = vec_adds (t2, t6); \ + vy4 = vec_subs (t2, t6); + +#define IDCT \ + vector_s16_t vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7; \ + vector_s16_t vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7; \ + vector_s16_t a0, a1, a2, ma2, c4, mc4, zero, bias; \ + vector_s16_t t0, t1, t2, t3, t4, t5, t6, t7, t8; \ + vector_u16_t shift; \ + \ + c4 = vec_splat (constants[0], 0); \ + a0 = vec_splat (constants[0], 1); \ + a1 = vec_splat (constants[0], 2); \ + a2 = vec_splat (constants[0], 3); \ + mc4 = vec_splat (constants[0], 4); \ + ma2 = vec_splat (constants[0], 5); \ + bias = (vector_s16_t)vec_splat ((vector_s32_t)constants[0], 3); \ + \ + zero = vec_splat_s16 (0); \ + shift = vec_splat_u16 (4); \ + \ + vx0 = vec_mradds (vec_sl (block[0], shift), constants[1], zero); \ + vx1 = vec_mradds (vec_sl (block[1], shift), constants[2], zero); \ + vx2 = vec_mradds (vec_sl (block[2], shift), constants[3], zero); \ + vx3 = vec_mradds (vec_sl (block[3], shift), constants[4], zero); \ + vx4 = vec_mradds (vec_sl (block[4], shift), constants[1], zero); \ + vx5 = vec_mradds (vec_sl (block[5], shift), constants[4], zero); \ + vx6 = vec_mradds (vec_sl (block[6], shift), constants[3], zero); \ + vx7 = vec_mradds (vec_sl (block[7], shift), constants[2], zero); \ + \ + IDCT_HALF \ + \ + vx0 = vec_mergeh (vy0, vy4); \ + vx1 = vec_mergel (vy0, vy4); \ + vx2 = vec_mergeh (vy1, vy5); \ + vx3 = vec_mergel (vy1, vy5); \ + vx4 = vec_mergeh (vy2, vy6); \ + vx5 = vec_mergel (vy2, vy6); \ + vx6 = vec_mergeh (vy3, vy7); \ + vx7 = vec_mergel (vy3, vy7); \ + \ + vy0 = vec_mergeh (vx0, vx4); \ + vy1 = vec_mergel (vx0, vx4); \ + vy2 = vec_mergeh (vx1, vx5); \ + vy3 = vec_mergel (vx1, vx5); \ + vy4 = vec_mergeh (vx2, vx6); \ + vy5 = vec_mergel (vx2, vx6); \ + vy6 = vec_mergeh (vx3, vx7); \ + vy7 = vec_mergel (vx3, vx7); \ + \ + vx0 = vec_adds (vec_mergeh (vy0, vy4), bias); \ + vx1 = vec_mergel (vy0, vy4); \ + vx2 = vec_mergeh (vy1, vy5); \ + vx3 = vec_mergel (vy1, vy5); \ + vx4 = vec_mergeh (vy2, vy6); \ + vx5 = vec_mergel (vy2, vy6); \ + vx6 = vec_mergeh (vy3, vy7); \ + vx7 = vec_mergel (vy3, vy7); \ + \ + IDCT_HALF \ + \ + shift = vec_splat_u16 (6); \ + vx0 = vec_sra (vy0, shift); \ + vx1 = vec_sra (vy1, shift); \ + vx2 = vec_sra (vy2, shift); \ + vx3 = vec_sra (vy3, shift); \ + vx4 = vec_sra (vy4, shift); \ + vx5 = vec_sra (vy5, shift); \ + vx6 = vec_sra (vy6, shift); \ + vx7 = vec_sra (vy7, shift); + +static vector_s16_t constants[5] = { + (vector_s16_t)(23170, 13573, 6518, 21895, -23170, -21895, 32, 31), + (vector_s16_t)(16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725), + (vector_s16_t)(22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521), + (vector_s16_t)(21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692), + (vector_s16_t)(19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722) +}; + +void idct_block_copy_altivec (vector_s16_t * block, unsigned char * dest, + int stride) +{ + vector_u8_t tmp; + + IDCT + +#define COPY(dest,src) \ + tmp = vec_packsu (src, src); \ + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); \ + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + + COPY (dest, vx0) dest += stride; + COPY (dest, vx1) dest += stride; + COPY (dest, vx2) dest += stride; + COPY (dest, vx3) dest += stride; + COPY (dest, vx4) dest += stride; + COPY (dest, vx5) dest += stride; + COPY (dest, vx6) dest += stride; + COPY (dest, vx7) +} + +void idct_block_add_altivec (vector_s16_t * block, unsigned char * dest, + int stride) +{ + vector_u8_t tmp; + vector_s16_t tmp2, tmp3; + vector_u8_t perm0; + vector_u8_t perm1; + vector_u8_t p0, p1, p; + + IDCT + + p0 = vec_lvsl (0, dest); + p1 = vec_lvsl (stride, dest); + p = vec_splat_u8 (-1); + perm0 = vec_mergeh (p, p0); + perm1 = vec_mergeh (p, p1); + +#define ADD(dest,src,perm) \ + /* *(uint64_t *)&tmp = *(uint64_t *)dest; */ \ + tmp = vec_ld (0, dest); \ + tmp2 = (vector_s16_t)vec_perm (tmp, (vector_u8_t)zero, perm); \ + tmp3 = vec_adds (tmp2, src); \ + tmp = vec_packsu (tmp3, tmp3); \ + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); \ + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + + ADD (dest, vx0, perm0) dest += stride; + ADD (dest, vx1, perm1) dest += stride; + ADD (dest, vx2, perm0) dest += stride; + ADD (dest, vx3, perm1) dest += stride; + ADD (dest, vx4, perm0) dest += stride; + ADD (dest, vx5, perm1) dest += stride; + ADD (dest, vx6, perm0) dest += stride; + ADD (dest, vx7, perm1) +} + +#endif /* __ALTIVEC__ */ diff --git a/src/libmpeg2/motion_comp.c b/src/libmpeg2/motion_comp.c index b8f308525..9ea10c532 100644 --- a/src/libmpeg2/motion_comp.c +++ b/src/libmpeg2/motion_comp.c @@ -49,6 +49,12 @@ void motion_comp_init (void) fprintf (stderr, "Using mlib for motion compensation\n"); mc_functions = mc_functions_mlib; } else +#endif +#ifdef ENABLE_ALTIVEC + if (config.flags & MM_ACCEL_PPC_ALTIVEC) { + fprintf (stderr, "Using altivec for motion compensation\n"); + mc_functions = mc_functions_altivec; + } else #endif { fprintf (stderr, "No accelerated motion compensation found\n"); diff --git a/src/libmpeg2/motion_comp_altivec.c b/src/libmpeg2/motion_comp_altivec.c new file mode 100644 index 000000000..e4b3a234c --- /dev/null +++ b/src/libmpeg2/motion_comp_altivec.c @@ -0,0 +1,2024 @@ +/* + * motion_comp_altivec.c + * Copyright (C) 2000-2001 Michel Lespinasse + * Copyright (C) 1999-2000 Aaron Holtzman + * + * This file is part of mpeg2dec, a free MPEG-2 video stream decoder. + * See http://libmpeg2.sourceforge.net/ for updates. + * + * mpeg2dec 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. + * + * mpeg2dec 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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +#ifndef __ALTIVEC__ + +#include "config.h" + +#ifdef ARCH_PPC +#ifdef ENABLE_ALTIVEC + +#include + +#include "mpeg2_internal.h" + +/* + * The asm code is generated with: + * + * gcc-2.95 -fvec -D__ALTIVEC__ -O9 -fomit-frame-pointer -mregnames -S + * motion_comp_altivec.c + * + * sed 's/.L/._L/g' motion_comp_altivec.s | + * awk '{args=""; len=split ($2, arg, ","); + * for (i=1; i<=len; i++) { a=arg[i]; if (i> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + tmp = vec_perm (ref0, ref1, perm); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + vec_st (tmp, 0, dest); + tmp = vec_perm (ref0, ref1, perm); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + vec_st (tmp, stride, dest); + dest += 2*stride; + tmp = vec_perm (ref0, ref1, perm); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + vec_st (tmp, 0, dest); + tmp = vec_perm (ref0, ref1, perm); + vec_st (tmp, stride, dest); +} + +void MC_put_o_8_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t perm0, perm1, tmp0, tmp1, ref0, ref1; + + tmp0 = vec_lvsl (0, ref); + tmp0 = vec_mergeh (tmp0, tmp0); + perm0 = vec_pack ((vector_u16_t)tmp0, (vector_u16_t)tmp0); + tmp1 = vec_lvsl (stride, ref); + tmp1 = vec_mergeh (tmp1, tmp1); + perm1 = vec_pack ((vector_u16_t)tmp1, (vector_u16_t)tmp1); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + tmp0 = vec_perm (ref0, ref1, perm0); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + vec_ste ((vector_u32_t)tmp0, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp0, 4, (unsigned int *)dest); + dest += stride; + tmp1 = vec_perm (ref0, ref1, perm1); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + vec_ste ((vector_u32_t)tmp1, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp1, 4, (unsigned int *)dest); + dest += stride; + tmp0 = vec_perm (ref0, ref1, perm0); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + vec_ste ((vector_u32_t)tmp0, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp0, 4, (unsigned int *)dest); + dest += stride; + tmp1 = vec_perm (ref0, ref1, perm1); + vec_ste ((vector_u32_t)tmp1, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp1, 4, (unsigned int *)dest); +} + +void MC_put_x_16_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t permA, permB, ref0, ref1, tmp; + + permA = vec_lvsl (0, ref); + permB = vec_add (permA, vec_splat_u8 (1)); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + tmp = vec_avg (vec_perm (ref0, ref1, permA), + vec_perm (ref0, ref1, permB)); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + vec_st (tmp, 0, dest); + tmp = vec_avg (vec_perm (ref0, ref1, permA), + vec_perm (ref0, ref1, permB)); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + vec_st (tmp, stride, dest); + dest += 2*stride; + tmp = vec_avg (vec_perm (ref0, ref1, permA), + vec_perm (ref0, ref1, permB)); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + vec_st (tmp, 0, dest); + tmp = vec_avg (vec_perm (ref0, ref1, permA), + vec_perm (ref0, ref1, permB)); + vec_st (tmp, stride, dest); +} + +void MC_put_x_8_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t perm0A, perm0B, perm1A, perm1B, ones, tmp0, tmp1, ref0, ref1; + + ones = vec_splat_u8 (1); + tmp0 = vec_lvsl (0, ref); + tmp0 = vec_mergeh (tmp0, tmp0); + perm0A = vec_pack ((vector_u16_t)tmp0, (vector_u16_t)tmp0); + perm0B = vec_add (perm0A, ones); + tmp1 = vec_lvsl (stride, ref); + tmp1 = vec_mergeh (tmp1, tmp1); + perm1A = vec_pack ((vector_u16_t)tmp1, (vector_u16_t)tmp1); + perm1B = vec_add (perm1A, ones); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (8, ref); + ref += stride; + tmp0 = vec_avg (vec_perm (ref0, ref1, perm0A), + vec_perm (ref0, ref1, perm0B)); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (8, ref); + ref += stride; + vec_ste ((vector_u32_t)tmp0, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp0, 4, (unsigned int *)dest); + dest += stride; + tmp1 = vec_avg (vec_perm (ref0, ref1, perm1A), + vec_perm (ref0, ref1, perm1B)); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (8, ref); + ref += stride; + vec_ste ((vector_u32_t)tmp1, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp1, 4, (unsigned int *)dest); + dest += stride; + tmp0 = vec_avg (vec_perm (ref0, ref1, perm0A), + vec_perm (ref0, ref1, perm0B)); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (8, ref); + vec_ste ((vector_u32_t)tmp0, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp0, 4, (unsigned int *)dest); + dest += stride; + tmp1 = vec_avg (vec_perm (ref0, ref1, perm1A), + vec_perm (ref0, ref1, perm1B)); + vec_ste ((vector_u32_t)tmp1, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp1, 4, (unsigned int *)dest); +} + +void MC_put_y_16_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t perm, ref0, ref1, tmp0, tmp1, tmp; + + perm = vec_lvsl (0, ref); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + tmp0 = vec_perm (ref0, ref1, perm); + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + tmp1 = vec_perm (ref0, ref1, perm); + tmp = vec_avg (tmp0, tmp1); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + vec_st (tmp, 0, dest); + tmp0 = vec_perm (ref0, ref1, perm); + tmp = vec_avg (tmp0, tmp1); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + vec_st (tmp, stride, dest); + dest += 2*stride; + tmp1 = vec_perm (ref0, ref1, perm); + tmp = vec_avg (tmp0, tmp1); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + vec_st (tmp, 0, dest); + tmp0 = vec_perm (ref0, ref1, perm); + tmp = vec_avg (tmp0, tmp1); + vec_st (tmp, stride, dest); +} + +void MC_put_y_8_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t perm0, perm1, tmp0, tmp1, tmp, ref0, ref1; + + tmp0 = vec_lvsl (0, ref); + tmp0 = vec_mergeh (tmp0, tmp0); + perm0 = vec_pack ((vector_u16_t)tmp0, (vector_u16_t)tmp0); + tmp1 = vec_lvsl (stride, ref); + tmp1 = vec_mergeh (tmp1, tmp1); + perm1 = vec_pack ((vector_u16_t)tmp1, (vector_u16_t)tmp1); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + tmp0 = vec_perm (ref0, ref1, perm0); + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + tmp1 = vec_perm (ref0, ref1, perm1); + tmp = vec_avg (tmp0, tmp1); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + tmp0 = vec_perm (ref0, ref1, perm0); + tmp = vec_avg (tmp0, tmp1); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + tmp1 = vec_perm (ref0, ref1, perm1); + tmp = vec_avg (tmp0, tmp1); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + tmp0 = vec_perm (ref0, ref1, perm0); + tmp = vec_avg (tmp0, tmp1); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); +} + +void MC_put_xy_16_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t permA, permB, ref0, ref1, A, B, avg0, avg1, xor0, xor1, tmp; + vector_u8_t ones; + + ones = vec_splat_u8 (1); + permA = vec_lvsl (0, ref); + permB = vec_add (permA, ones); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + A = vec_perm (ref0, ref1, permA); + B = vec_perm (ref0, ref1, permB); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + A = vec_perm (ref0, ref1, permA); + B = vec_perm (ref0, ref1, permB); + avg1 = vec_avg (A, B); + xor1 = vec_xor (A, B); + tmp = vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1))); + + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + vec_st (tmp, 0, dest); + A = vec_perm (ref0, ref1, permA); + B = vec_perm (ref0, ref1, permB); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + tmp = vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1))); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + vec_st (tmp, stride, dest); + dest += 2*stride; + A = vec_perm (ref0, ref1, permA); + B = vec_perm (ref0, ref1, permB); + avg1 = vec_avg (A, B); + xor1 = vec_xor (A, B); + tmp = vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1))); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + vec_st (tmp, 0, dest); + A = vec_perm (ref0, ref1, permA); + B = vec_perm (ref0, ref1, permB); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + tmp = vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1))); + vec_st (tmp, stride, dest); +} + +void MC_put_xy_8_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t perm0A, perm0B, perm1A, perm1B, ref0, ref1, A, B; + vector_u8_t avg0, avg1, xor0, xor1, tmp, ones; + + ones = vec_splat_u8 (1); + perm0A = vec_lvsl (0, ref); + perm0A = vec_mergeh (perm0A, perm0A); + perm0A = vec_pack ((vector_u16_t)perm0A, (vector_u16_t)perm0A); + perm0B = vec_add (perm0A, ones); + perm1A = vec_lvsl (stride, ref); + perm1A = vec_mergeh (perm1A, perm1A); + perm1A = vec_pack ((vector_u16_t)perm1A, (vector_u16_t)perm1A); + perm1B = vec_add (perm1A, ones); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + A = vec_perm (ref0, ref1, perm0A); + B = vec_perm (ref0, ref1, perm0B); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + A = vec_perm (ref0, ref1, perm1A); + B = vec_perm (ref0, ref1, perm1B); + avg1 = vec_avg (A, B); + xor1 = vec_xor (A, B); + tmp = vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1))); + + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + A = vec_perm (ref0, ref1, perm0A); + B = vec_perm (ref0, ref1, perm0B); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + tmp = vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1))); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + A = vec_perm (ref0, ref1, perm1A); + B = vec_perm (ref0, ref1, perm1B); + avg1 = vec_avg (A, B); + xor1 = vec_xor (A, B); + tmp = vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1))); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + A = vec_perm (ref0, ref1, perm0A); + B = vec_perm (ref0, ref1, perm0B); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + tmp = vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1))); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); +} + +#if 0 +void MC_put_xy_8_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t permA, permB, ref0, ref1, A, B, C, D, tmp, zero, ones; + vector_u16_t splat2, temp; + + ones = vec_splat_u8 (1); + permA = vec_lvsl (0, ref); + permB = vec_add (permA, ones); + + zero = vec_splat_u8 (0); + splat2 = vec_splat_u16 (2); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + A = vec_perm (ref0, ref1, permA); + B = vec_perm (ref0, ref1, permB); + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + C = vec_perm (ref0, ref1, permA); + D = vec_perm (ref0, ref1, permB); + + temp = vec_add (vec_add ((vector_u16_t)vec_mergeh (zero, A), + (vector_u16_t)vec_mergeh (zero, B)), + vec_add ((vector_u16_t)vec_mergeh (zero, C), + (vector_u16_t)vec_mergeh (zero, D))); + temp = vec_sr (vec_add (temp, splat2), splat2); + tmp = vec_pack (temp, temp); + + vec_st (tmp, 0, dest); + dest += stride; + tmp = vec_avg (vec_perm (ref0, ref1, permA), + vec_perm (ref0, ref1, permB)); + } while (--height); +} +#endif + +void MC_avg_o_16_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t perm, ref0, ref1, tmp, prev; + + perm = vec_lvsl (0, ref); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + prev = vec_ld (0, dest); + tmp = vec_avg (prev, vec_perm (ref0, ref1, perm)); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_st (tmp, 0, dest); + tmp = vec_avg (prev, vec_perm (ref0, ref1, perm)); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + prev = vec_ld (2*stride, dest); + vec_st (tmp, stride, dest); + dest += 2*stride; + tmp = vec_avg (prev, vec_perm (ref0, ref1, perm)); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + prev = vec_ld (stride, dest); + vec_st (tmp, 0, dest); + tmp = vec_avg (prev, vec_perm (ref0, ref1, perm)); + vec_st (tmp, stride, dest); +} + +void MC_avg_o_8_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t perm0, perm1, tmp0, tmp1, ref0, ref1, prev; + + tmp0 = vec_lvsl (0, ref); + tmp0 = vec_mergeh (tmp0, tmp0); + perm0 = vec_pack ((vector_u16_t)tmp0, (vector_u16_t)tmp0); + tmp1 = vec_lvsl (stride, ref); + tmp1 = vec_mergeh (tmp1, tmp1); + perm1 = vec_pack ((vector_u16_t)tmp1, (vector_u16_t)tmp1); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + prev = vec_ld (0, dest); + tmp0 = vec_avg (prev, vec_perm (ref0, ref1, perm0)); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp0, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp0, 4, (unsigned int *)dest); + dest += stride; + tmp1 = vec_avg (prev, vec_perm (ref0, ref1, perm1)); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp1, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp1, 4, (unsigned int *)dest); + dest += stride; + tmp0 = vec_avg (prev, vec_perm (ref0, ref1, perm0)); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp0, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp0, 4, (unsigned int *)dest); + dest += stride; + tmp1 = vec_avg (prev, vec_perm (ref0, ref1, perm1)); + vec_ste ((vector_u32_t)tmp1, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp1, 4, (unsigned int *)dest); +} + +void MC_avg_x_16_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t permA, permB, ref0, ref1, tmp, prev; + + permA = vec_lvsl (0, ref); + permB = vec_add (permA, vec_splat_u8 (1)); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + prev = vec_ld (0, dest); + ref += stride; + tmp = vec_avg (prev, vec_avg (vec_perm (ref0, ref1, permA), + vec_perm (ref0, ref1, permB))); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_st (tmp, 0, dest); + tmp = vec_avg (prev, vec_avg (vec_perm (ref0, ref1, permA), + vec_perm (ref0, ref1, permB))); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + prev = vec_ld (2*stride, dest); + vec_st (tmp, stride, dest); + dest += 2*stride; + tmp = vec_avg (prev, vec_avg (vec_perm (ref0, ref1, permA), + vec_perm (ref0, ref1, permB))); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + prev = vec_ld (stride, dest); + vec_st (tmp, 0, dest); + tmp = vec_avg (prev, vec_avg (vec_perm (ref0, ref1, permA), + vec_perm (ref0, ref1, permB))); + vec_st (tmp, stride, dest); +} + +void MC_avg_x_8_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t perm0A, perm0B, perm1A, perm1B, ones, tmp0, tmp1, ref0, ref1; + vector_u8_t prev; + + ones = vec_splat_u8 (1); + tmp0 = vec_lvsl (0, ref); + tmp0 = vec_mergeh (tmp0, tmp0); + perm0A = vec_pack ((vector_u16_t)tmp0, (vector_u16_t)tmp0); + perm0B = vec_add (perm0A, ones); + tmp1 = vec_lvsl (stride, ref); + tmp1 = vec_mergeh (tmp1, tmp1); + perm1A = vec_pack ((vector_u16_t)tmp1, (vector_u16_t)tmp1); + perm1B = vec_add (perm1A, ones); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (8, ref); + prev = vec_ld (0, dest); + ref += stride; + tmp0 = vec_avg (prev, vec_avg (vec_perm (ref0, ref1, perm0A), + vec_perm (ref0, ref1, perm0B))); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (8, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp0, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp0, 4, (unsigned int *)dest); + dest += stride; + tmp1 = vec_avg (prev, vec_avg (vec_perm (ref0, ref1, perm1A), + vec_perm (ref0, ref1, perm1B))); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (8, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp1, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp1, 4, (unsigned int *)dest); + dest += stride; + tmp0 = vec_avg (prev, vec_avg (vec_perm (ref0, ref1, perm0A), + vec_perm (ref0, ref1, perm0B))); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (8, ref); + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp0, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp0, 4, (unsigned int *)dest); + dest += stride; + tmp1 = vec_avg (prev, vec_avg (vec_perm (ref0, ref1, perm1A), + vec_perm (ref0, ref1, perm1B))); + vec_ste ((vector_u32_t)tmp1, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp1, 4, (unsigned int *)dest); +} + +void MC_avg_y_16_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t perm, ref0, ref1, tmp0, tmp1, tmp, prev; + + perm = vec_lvsl (0, ref); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + tmp0 = vec_perm (ref0, ref1, perm); + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + prev = vec_ld (0, dest); + tmp1 = vec_perm (ref0, ref1, perm); + tmp = vec_avg (prev, vec_avg (tmp0, tmp1)); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_st (tmp, 0, dest); + tmp0 = vec_perm (ref0, ref1, perm); + tmp = vec_avg (prev, vec_avg (tmp0, tmp1)); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + ref += stride; + prev = vec_ld (2*stride, dest); + vec_st (tmp, stride, dest); + dest += 2*stride; + tmp1 = vec_perm (ref0, ref1, perm); + tmp = vec_avg (prev, vec_avg (tmp0, tmp1)); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (15, ref); + prev = vec_ld (stride, dest); + vec_st (tmp, 0, dest); + tmp0 = vec_perm (ref0, ref1, perm); + tmp = vec_avg (prev, vec_avg (tmp0, tmp1)); + vec_st (tmp, stride, dest); +} + +void MC_avg_y_8_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t perm0, perm1, tmp0, tmp1, tmp, ref0, ref1, prev; + + tmp0 = vec_lvsl (0, ref); + tmp0 = vec_mergeh (tmp0, tmp0); + perm0 = vec_pack ((vector_u16_t)tmp0, (vector_u16_t)tmp0); + tmp1 = vec_lvsl (stride, ref); + tmp1 = vec_mergeh (tmp1, tmp1); + perm1 = vec_pack ((vector_u16_t)tmp1, (vector_u16_t)tmp1); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + tmp0 = vec_perm (ref0, ref1, perm0); + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + prev = vec_ld (0, dest); + tmp1 = vec_perm (ref0, ref1, perm1); + tmp = vec_avg (prev, vec_avg (tmp0, tmp1)); + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + tmp0 = vec_perm (ref0, ref1, perm0); + tmp = vec_avg (prev, vec_avg (tmp0, tmp1)); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + tmp1 = vec_perm (ref0, ref1, perm1); + tmp = vec_avg (prev, vec_avg (tmp0, tmp1)); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (7, ref); + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + tmp0 = vec_perm (ref0, ref1, perm0); + tmp = vec_avg (prev, vec_avg (tmp0, tmp1)); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); +} + +void MC_avg_xy_16_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t permA, permB, ref0, ref1, A, B, avg0, avg1, xor0, xor1, tmp; + vector_u8_t ones, prev; + + ones = vec_splat_u8 (1); + permA = vec_lvsl (0, ref); + permB = vec_add (permA, ones); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + A = vec_perm (ref0, ref1, permA); + B = vec_perm (ref0, ref1, permB); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + prev = vec_ld (0, dest); + A = vec_perm (ref0, ref1, permA); + B = vec_perm (ref0, ref1, permB); + avg1 = vec_avg (A, B); + xor1 = vec_xor (A, B); + tmp = vec_avg (prev, vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1)))); + + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_st (tmp, 0, dest); + A = vec_perm (ref0, ref1, permA); + B = vec_perm (ref0, ref1, permB); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + tmp = vec_avg (prev, + vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1)))); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + prev = vec_ld (2*stride, dest); + vec_st (tmp, stride, dest); + dest += 2*stride; + A = vec_perm (ref0, ref1, permA); + B = vec_perm (ref0, ref1, permB); + avg1 = vec_avg (A, B); + xor1 = vec_xor (A, B); + tmp = vec_avg (prev, + vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1)))); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + prev = vec_ld (stride, dest); + vec_st (tmp, 0, dest); + A = vec_perm (ref0, ref1, permA); + B = vec_perm (ref0, ref1, permB); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + tmp = vec_avg (prev, vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1)))); + vec_st (tmp, stride, dest); +} + +void MC_avg_xy_8_altivec (unsigned char * dest, unsigned char * ref, + int stride, int height) +{ + vector_u8_t perm0A, perm0B, perm1A, perm1B, ref0, ref1, A, B; + vector_u8_t avg0, avg1, xor0, xor1, tmp, ones, prev; + + ones = vec_splat_u8 (1); + perm0A = vec_lvsl (0, ref); + perm0A = vec_mergeh (perm0A, perm0A); + perm0A = vec_pack ((vector_u16_t)perm0A, (vector_u16_t)perm0A); + perm0B = vec_add (perm0A, ones); + perm1A = vec_lvsl (stride, ref); + perm1A = vec_mergeh (perm1A, perm1A); + perm1A = vec_pack ((vector_u16_t)perm1A, (vector_u16_t)perm1A); + perm1B = vec_add (perm1A, ones); + + height = (height >> 1) - 1; + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + A = vec_perm (ref0, ref1, perm0A); + B = vec_perm (ref0, ref1, perm0B); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + prev = vec_ld (0, dest); + A = vec_perm (ref0, ref1, perm1A); + B = vec_perm (ref0, ref1, perm1B); + avg1 = vec_avg (A, B); + xor1 = vec_xor (A, B); + tmp = vec_avg (prev, vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1)))); + + + do { + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + A = vec_perm (ref0, ref1, perm0A); + B = vec_perm (ref0, ref1, perm0B); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + tmp = vec_avg (prev, + vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1)))); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + ref += stride; + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + A = vec_perm (ref0, ref1, perm1A); + B = vec_perm (ref0, ref1, perm1B); + avg1 = vec_avg (A, B); + xor1 = vec_xor (A, B); + tmp = vec_avg (prev, + vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1)))); + } while (--height); + + ref0 = vec_ld (0, ref); + ref1 = vec_ld (16, ref); + prev = vec_ld (stride, dest); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); + dest += stride; + A = vec_perm (ref0, ref1, perm0A); + B = vec_perm (ref0, ref1, perm0B); + avg0 = vec_avg (A, B); + xor0 = vec_xor (A, B); + tmp = vec_avg (prev, vec_sub (vec_avg (avg0, avg1), + vec_and (vec_and (ones, vec_or (xor0, xor1)), + vec_xor (avg0, avg1)))); + vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); + vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); +} + +#endif /* __ALTIVEC__ */ diff --git a/src/libmpeg2/mpeg2_internal.h b/src/libmpeg2/mpeg2_internal.h index 8e7ce0d11..d9bc71c62 100644 --- a/src/libmpeg2/mpeg2_internal.h +++ b/src/libmpeg2/mpeg2_internal.h @@ -170,6 +170,13 @@ void idct_block_copy_mmx (int16_t *block, uint8_t * dest, int stride); void idct_block_add_mmx (int16_t *block, uint8_t * dest, int stride); void idct_mmx_init (void); +#ifdef ENABLE_ALTIVEC +/* idct_altivec.c */ +void idct_block_copy_altivec (int16_t * block, uint8_t * dest, int stride); +void idct_block_add_altivec (int16_t * block, uint8_t * dest, int stride); +void idct_altivec_init (void); +#endif + /* motion_comp.c */ void motion_comp_init (void); @@ -191,6 +198,9 @@ extern mc_functions_t mc_functions_c; extern mc_functions_t mc_functions_mmx; extern mc_functions_t mc_functions_mmxext; extern mc_functions_t mc_functions_3dnow; +#ifdef ENABLE_ALTIVEC +extern mc_functions_t mc_functions_altivec; +#endif extern mc_functions_t mc_functions_mlib; /* slice.c */ diff --git a/src/xine-utils/cpu_accel.c b/src/xine-utils/cpu_accel.c index 5fa5a45ed..41baa5d5e 100644 --- a/src/xine-utils/cpu_accel.c +++ b/src/xine-utils/cpu_accel.c @@ -121,33 +121,31 @@ static void sigill_handler (int n) { uint32_t xine_mm_accel (void) { #ifdef ARCH_X86 - static int got_accel = 0; static uint32_t accel; - if (!got_accel) { - got_accel = 1; - - accel = x86_accel (); - - /* test OS support for SSE */ - if( accel & MM_ACCEL_X86_SSE ) { - if (setjmp(sigill_return)) { - accel &= ~(MM_ACCEL_X86_SSE|MM_ACCEL_X86_SSE2); - } else { - signal (SIGILL, sigill_handler); - __asm __volatile ("xorps %xmm0, %xmm0"); - signal (SIGILL, SIG_DFL); - } + accel = x86_accel (); + + /* test OS support for SSE */ + if( accel & MM_ACCEL_X86_SSE ) { + if (setjmp(sigill_return)) { + accel &= ~(MM_ACCEL_X86_SSE|MM_ACCEL_X86_SSE2); + } else { + signal (SIGILL, sigill_handler); + __asm __volatile ("xorps %xmm0, %xmm0"); + signal (SIGILL, SIG_DFL); } } return accel; -#else +#endif #ifdef HAVE_MLIB return MM_ACCEL_MLIB; -#else - return 0; +#endif +#ifdef ARCH_PPC +#ifdef ENABLE_ALTIVEC + return MM_ACCEL_PPC_ALTIVEC; #endif #endif + return 0; } diff --git a/src/xine-utils/xineutils.h b/src/xine-utils/xineutils.h index a4a689f4a..634cf0c74 100644 --- a/src/xine-utils/xineutils.h +++ b/src/xine-utils/xineutils.h @@ -17,7 +17,7 @@ * along with this program; if not, write to the Free Software * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA * - * $Id: xineutils.h,v 1.4 2001/11/30 00:53:51 f1rmb Exp $ + * $Id: xineutils.h,v 1.5 2001/12/05 15:12:03 guenter Exp $ * */ #ifndef XINEUTILS_H @@ -49,6 +49,8 @@ extern "C" { #define MM_ACCEL_X86_MMXEXT 0x20000000 #define MM_ACCEL_X86_SSE 0x10000000 #define MM_ACCEL_X86_SSE2 0x08000000 +/* powerpc accelerations */ +#define MM_ACCEL_PPC_ALTIVEC 0x04000000 /* x86 compat defines */ #define MM_MMX MM_ACCEL_X86_MMX #define MM_3DNOW MM_ACCEL_X86_3DNOW -- cgit v1.2.3