nengel@2: /* nengel@2: * Copyright (c) 2001 Michel Lespinasse nengel@2: * nengel@2: * This file is part of FFmpeg. nengel@2: * nengel@2: * FFmpeg is free software; you can redistribute it and/or nengel@2: * modify it under the terms of the GNU Lesser General Public nengel@2: * License as published by the Free Software Foundation; either nengel@2: * version 2.1 of the License, or (at your option) any later version. nengel@2: * nengel@2: * FFmpeg is distributed in the hope that it will be useful, nengel@2: * but WITHOUT ANY WARRANTY; without even the implied warranty of nengel@2: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU nengel@2: * Lesser General Public License for more details. nengel@2: * nengel@2: * You should have received a copy of the GNU Lesser General Public nengel@2: * License along with FFmpeg; if not, write to the Free Software nengel@2: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA nengel@2: */ nengel@2: nengel@2: /* nengel@2: * NOTE: This code is based on GPL code from the libmpeg2 project. The nengel@2: * author, Michel Lespinasses, has given explicit permission to release nengel@2: * under LGPL as part of FFmpeg. nengel@2: */ nengel@2: nengel@2: /* nengel@2: * FFmpeg integration by Dieter Shirley nengel@2: * nengel@2: * This file is a direct copy of the AltiVec IDCT module from the libmpeg2 nengel@2: * project. I've deleted all of the libmpeg2-specific code, renamed the nengel@2: * functions and reordered the function parameters. The only change to the nengel@2: * IDCT function itself was to factor out the partial transposition, and to nengel@2: * perform a full transpose at the end of the function. nengel@2: */ nengel@2: nengel@2: nengel@2: #include /* malloc(), free() */ nengel@2: #include nengel@2: #include "config.h" nengel@2: #if HAVE_ALTIVEC_H nengel@2: #include nengel@2: #endif nengel@2: #include "libavcodec/dsputil.h" nengel@2: #include "types_altivec.h" nengel@2: #include "dsputil_ppc.h" nengel@2: #include "dsputil_altivec.h" nengel@2: nengel@2: #define IDCT_HALF \ nengel@2: /* 1st stage */ \ nengel@2: t1 = vec_mradds (a1, vx7, vx1 ); \ nengel@2: t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7)); \ nengel@2: t7 = vec_mradds (a2, vx5, vx3); \ nengel@2: t3 = vec_mradds (ma2, vx3, vx5); \ nengel@2: \ nengel@2: /* 2nd stage */ \ nengel@2: t5 = vec_adds (vx0, vx4); \ nengel@2: t0 = vec_subs (vx0, vx4); \ nengel@2: t2 = vec_mradds (a0, vx6, vx2); \ nengel@2: t4 = vec_mradds (a0, vx2, vec_subs (zero, vx6)); \ nengel@2: t6 = vec_adds (t8, t3); \ nengel@2: t3 = vec_subs (t8, t3); \ nengel@2: t8 = vec_subs (t1, t7); \ nengel@2: t1 = vec_adds (t1, t7); \ nengel@2: \ nengel@2: /* 3rd stage */ \ nengel@2: t7 = vec_adds (t5, t2); \ nengel@2: t2 = vec_subs (t5, t2); \ nengel@2: t5 = vec_adds (t0, t4); \ nengel@2: t0 = vec_subs (t0, t4); \ nengel@2: t4 = vec_subs (t8, t3); \ nengel@2: t3 = vec_adds (t8, t3); \ nengel@2: \ nengel@2: /* 4th stage */ \ nengel@2: vy0 = vec_adds (t7, t1); \ nengel@2: vy7 = vec_subs (t7, t1); \ nengel@2: vy1 = vec_mradds (c4, t3, t5); \ nengel@2: vy6 = vec_mradds (mc4, t3, t5); \ nengel@2: vy2 = vec_mradds (c4, t4, t0); \ nengel@2: vy5 = vec_mradds (mc4, t4, t0); \ nengel@2: vy3 = vec_adds (t2, t6); \ nengel@2: vy4 = vec_subs (t2, t6); nengel@2: nengel@2: nengel@2: #define IDCT \ nengel@2: vec_s16 vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7; \ nengel@2: vec_s16 vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7; \ nengel@2: vec_s16 a0, a1, a2, ma2, c4, mc4, zero, bias; \ nengel@2: vec_s16 t0, t1, t2, t3, t4, t5, t6, t7, t8; \ nengel@2: vec_u16 shift; \ nengel@2: \ nengel@2: c4 = vec_splat (constants[0], 0); \ nengel@2: a0 = vec_splat (constants[0], 1); \ nengel@2: a1 = vec_splat (constants[0], 2); \ nengel@2: a2 = vec_splat (constants[0], 3); \ nengel@2: mc4 = vec_splat (constants[0], 4); \ nengel@2: ma2 = vec_splat (constants[0], 5); \ nengel@2: bias = (vec_s16)vec_splat ((vec_s32)constants[0], 3); \ nengel@2: \ nengel@2: zero = vec_splat_s16 (0); \ nengel@2: shift = vec_splat_u16 (4); \ nengel@2: \ nengel@2: vx0 = vec_mradds (vec_sl (block[0], shift), constants[1], zero); \ nengel@2: vx1 = vec_mradds (vec_sl (block[1], shift), constants[2], zero); \ nengel@2: vx2 = vec_mradds (vec_sl (block[2], shift), constants[3], zero); \ nengel@2: vx3 = vec_mradds (vec_sl (block[3], shift), constants[4], zero); \ nengel@2: vx4 = vec_mradds (vec_sl (block[4], shift), constants[1], zero); \ nengel@2: vx5 = vec_mradds (vec_sl (block[5], shift), constants[4], zero); \ nengel@2: vx6 = vec_mradds (vec_sl (block[6], shift), constants[3], zero); \ nengel@2: vx7 = vec_mradds (vec_sl (block[7], shift), constants[2], zero); \ nengel@2: \ nengel@2: IDCT_HALF \ nengel@2: \ nengel@2: vx0 = vec_mergeh (vy0, vy4); \ nengel@2: vx1 = vec_mergel (vy0, vy4); \ nengel@2: vx2 = vec_mergeh (vy1, vy5); \ nengel@2: vx3 = vec_mergel (vy1, vy5); \ nengel@2: vx4 = vec_mergeh (vy2, vy6); \ nengel@2: vx5 = vec_mergel (vy2, vy6); \ nengel@2: vx6 = vec_mergeh (vy3, vy7); \ nengel@2: vx7 = vec_mergel (vy3, vy7); \ nengel@2: \ nengel@2: vy0 = vec_mergeh (vx0, vx4); \ nengel@2: vy1 = vec_mergel (vx0, vx4); \ nengel@2: vy2 = vec_mergeh (vx1, vx5); \ nengel@2: vy3 = vec_mergel (vx1, vx5); \ nengel@2: vy4 = vec_mergeh (vx2, vx6); \ nengel@2: vy5 = vec_mergel (vx2, vx6); \ nengel@2: vy6 = vec_mergeh (vx3, vx7); \ nengel@2: vy7 = vec_mergel (vx3, vx7); \ nengel@2: \ nengel@2: vx0 = vec_adds (vec_mergeh (vy0, vy4), bias); \ nengel@2: vx1 = vec_mergel (vy0, vy4); \ nengel@2: vx2 = vec_mergeh (vy1, vy5); \ nengel@2: vx3 = vec_mergel (vy1, vy5); \ nengel@2: vx4 = vec_mergeh (vy2, vy6); \ nengel@2: vx5 = vec_mergel (vy2, vy6); \ nengel@2: vx6 = vec_mergeh (vy3, vy7); \ nengel@2: vx7 = vec_mergel (vy3, vy7); \ nengel@2: \ nengel@2: IDCT_HALF \ nengel@2: \ nengel@2: shift = vec_splat_u16 (6); \ nengel@2: vx0 = vec_sra (vy0, shift); \ nengel@2: vx1 = vec_sra (vy1, shift); \ nengel@2: vx2 = vec_sra (vy2, shift); \ nengel@2: vx3 = vec_sra (vy3, shift); \ nengel@2: vx4 = vec_sra (vy4, shift); \ nengel@2: vx5 = vec_sra (vy5, shift); \ nengel@2: vx6 = vec_sra (vy6, shift); \ nengel@2: vx7 = vec_sra (vy7, shift); nengel@2: nengel@2: nengel@2: static const vec_s16 constants[5] = { nengel@2: {23170, 13573, 6518, 21895, -23170, -21895, 32, 31}, nengel@2: {16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725}, nengel@2: {22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521}, nengel@2: {21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692}, nengel@2: {19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722} nengel@2: }; nengel@2: nengel@2: void idct_put_altivec(uint8_t* dest, int stride, int16_t *blk) nengel@2: { nengel@2: POWERPC_PERF_DECLARE(altivec_idct_put_num, 1); nengel@2: vec_s16 *block = (vec_s16*)blk; nengel@2: vec_u8 tmp; nengel@2: nengel@2: #if CONFIG_POWERPC_PERF nengel@2: POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1); nengel@2: #endif nengel@2: IDCT nengel@2: nengel@2: #define COPY(dest,src) \ nengel@2: tmp = vec_packsu (src, src); \ nengel@2: vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest); \ nengel@2: vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest); nengel@2: nengel@2: COPY (dest, vx0) dest += stride; nengel@2: COPY (dest, vx1) dest += stride; nengel@2: COPY (dest, vx2) dest += stride; nengel@2: COPY (dest, vx3) dest += stride; nengel@2: COPY (dest, vx4) dest += stride; nengel@2: COPY (dest, vx5) dest += stride; nengel@2: COPY (dest, vx6) dest += stride; nengel@2: COPY (dest, vx7) nengel@2: nengel@2: POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1); nengel@2: } nengel@2: nengel@2: void idct_add_altivec(uint8_t* dest, int stride, int16_t *blk) nengel@2: { nengel@2: POWERPC_PERF_DECLARE(altivec_idct_add_num, 1); nengel@2: vec_s16 *block = (vec_s16*)blk; nengel@2: vec_u8 tmp; nengel@2: vec_s16 tmp2, tmp3; nengel@2: vec_u8 perm0; nengel@2: vec_u8 perm1; nengel@2: vec_u8 p0, p1, p; nengel@2: nengel@2: #if CONFIG_POWERPC_PERF nengel@2: POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1); nengel@2: #endif nengel@2: nengel@2: IDCT nengel@2: nengel@2: p0 = vec_lvsl (0, dest); nengel@2: p1 = vec_lvsl (stride, dest); nengel@2: p = vec_splat_u8 (-1); nengel@2: perm0 = vec_mergeh (p, p0); nengel@2: perm1 = vec_mergeh (p, p1); nengel@2: nengel@2: #define ADD(dest,src,perm) \ nengel@2: /* *(uint64_t *)&tmp = *(uint64_t *)dest; */ \ nengel@2: tmp = vec_ld (0, dest); \ nengel@2: tmp2 = (vec_s16)vec_perm (tmp, (vec_u8)zero, perm); \ nengel@2: tmp3 = vec_adds (tmp2, src); \ nengel@2: tmp = vec_packsu (tmp3, tmp3); \ nengel@2: vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest); \ nengel@2: vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest); nengel@2: nengel@2: ADD (dest, vx0, perm0) dest += stride; nengel@2: ADD (dest, vx1, perm1) dest += stride; nengel@2: ADD (dest, vx2, perm0) dest += stride; nengel@2: ADD (dest, vx3, perm1) dest += stride; nengel@2: ADD (dest, vx4, perm0) dest += stride; nengel@2: ADD (dest, vx5, perm1) dest += stride; nengel@2: ADD (dest, vx6, perm0) dest += stride; nengel@2: ADD (dest, vx7, perm1) nengel@2: nengel@2: POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1); nengel@2: } nengel@2: