Mercurial > cgi-bin > hgwebdir.cgi > PR > Applications > VSs > VSs__H264__App
diff libavcodec/ppc/h264_altivec.c @ 2:897f711a7157
rearrange to work with autoconf
| author | Nina Engelhardt <nengel@mailbox.tu-berlin.de> |
|---|---|
| date | Tue, 25 Sep 2012 15:55:33 +0200 |
| parents | |
| children |
line diff
1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000 1.2 +++ b/libavcodec/ppc/h264_altivec.c Tue Sep 25 15:55:33 2012 +0200 1.3 @@ -0,0 +1,1021 @@ 1.4 +/* 1.5 + * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> 1.6 + * 1.7 + * This file is part of FFmpeg. 1.8 + * 1.9 + * FFmpeg is free software; you can redistribute it and/or 1.10 + * modify it under the terms of the GNU Lesser General Public 1.11 + * License as published by the Free Software Foundation; either 1.12 + * version 2.1 of the License, or (at your option) any later version. 1.13 + * 1.14 + * FFmpeg is distributed in the hope that it will be useful, 1.15 + * but WITHOUT ANY WARRANTY; without even the implied warranty of 1.16 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 1.17 + * Lesser General Public License for more details. 1.18 + * 1.19 + * You should have received a copy of the GNU Lesser General Public 1.20 + * License along with FFmpeg; if not, write to the Free Software 1.21 + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 1.22 + */ 1.23 + 1.24 +#include "libavcodec/dsputil.h" 1.25 +#include "libavcodec/h264_data.h" 1.26 +#include "libavcodec/h264_dsp.h" 1.27 + 1.28 +#include "dsputil_ppc.h" 1.29 +#include "dsputil_altivec.h" 1.30 +#include "util_altivec.h" 1.31 +#include "types_altivec.h" 1.32 + 1.33 +#define PUT_OP_U8_ALTIVEC(d, s, dst) d = s 1.34 +#define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s) 1.35 + 1.36 +#define OP_U8_ALTIVEC PUT_OP_U8_ALTIVEC 1.37 +#define PREFIX_h264_chroma_mc8_altivec put_h264_chroma_mc8_altivec 1.38 +#define PREFIX_no_rnd_vc1_chroma_mc8_altivec put_no_rnd_vc1_chroma_mc8_altivec 1.39 +#define PREFIX_h264_chroma_mc8_num altivec_put_h264_chroma_mc8_num 1.40 +#define PREFIX_h264_qpel16_h_lowpass_altivec put_h264_qpel16_h_lowpass_altivec 1.41 +#define PREFIX_h264_qpel16_h_lowpass_num altivec_put_h264_qpel16_h_lowpass_num 1.42 +#define PREFIX_h264_qpel16_v_lowpass_altivec put_h264_qpel16_v_lowpass_altivec 1.43 +#define PREFIX_h264_qpel16_v_lowpass_num altivec_put_h264_qpel16_v_lowpass_num 1.44 +#define PREFIX_h264_qpel16_hv_lowpass_altivec put_h264_qpel16_hv_lowpass_altivec 1.45 +#define PREFIX_h264_qpel16_hv_lowpass_num altivec_put_h264_qpel16_hv_lowpass_num 1.46 +#include "h264_template_altivec.c" 1.47 +#undef OP_U8_ALTIVEC 1.48 +#undef PREFIX_h264_chroma_mc8_altivec 1.49 +#undef PREFIX_no_rnd_vc1_chroma_mc8_altivec 1.50 +#undef PREFIX_h264_chroma_mc8_num 1.51 +#undef PREFIX_h264_qpel16_h_lowpass_altivec 1.52 +#undef PREFIX_h264_qpel16_h_lowpass_num 1.53 +#undef PREFIX_h264_qpel16_v_lowpass_altivec 1.54 +#undef PREFIX_h264_qpel16_v_lowpass_num 1.55 +#undef PREFIX_h264_qpel16_hv_lowpass_altivec 1.56 +#undef PREFIX_h264_qpel16_hv_lowpass_num 1.57 + 1.58 +#define OP_U8_ALTIVEC AVG_OP_U8_ALTIVEC 1.59 +#define PREFIX_h264_chroma_mc8_altivec avg_h264_chroma_mc8_altivec 1.60 +#define PREFIX_no_rnd_vc1_chroma_mc8_altivec avg_no_rnd_vc1_chroma_mc8_altivec 1.61 +#define PREFIX_h264_chroma_mc8_num altivec_avg_h264_chroma_mc8_num 1.62 +#define PREFIX_h264_qpel16_h_lowpass_altivec avg_h264_qpel16_h_lowpass_altivec 1.63 +#define PREFIX_h264_qpel16_h_lowpass_num altivec_avg_h264_qpel16_h_lowpass_num 1.64 +#define PREFIX_h264_qpel16_v_lowpass_altivec avg_h264_qpel16_v_lowpass_altivec 1.65 +#define PREFIX_h264_qpel16_v_lowpass_num altivec_avg_h264_qpel16_v_lowpass_num 1.66 +#define PREFIX_h264_qpel16_hv_lowpass_altivec avg_h264_qpel16_hv_lowpass_altivec 1.67 +#define PREFIX_h264_qpel16_hv_lowpass_num altivec_avg_h264_qpel16_hv_lowpass_num 1.68 +#include "h264_template_altivec.c" 1.69 +#undef OP_U8_ALTIVEC 1.70 +#undef PREFIX_h264_chroma_mc8_altivec 1.71 +#undef PREFIX_no_rnd_vc1_chroma_mc8_altivec 1.72 +#undef PREFIX_h264_chroma_mc8_num 1.73 +#undef PREFIX_h264_qpel16_h_lowpass_altivec 1.74 +#undef PREFIX_h264_qpel16_h_lowpass_num 1.75 +#undef PREFIX_h264_qpel16_v_lowpass_altivec 1.76 +#undef PREFIX_h264_qpel16_v_lowpass_num 1.77 +#undef PREFIX_h264_qpel16_hv_lowpass_altivec 1.78 +#undef PREFIX_h264_qpel16_hv_lowpass_num 1.79 + 1.80 +#define H264_MC(OPNAME, SIZE, CODETYPE) \ 1.81 +static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\ 1.82 + OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\ 1.83 +}\ 1.84 +\ 1.85 +static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \ 1.86 + DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ 1.87 + put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ 1.88 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\ 1.89 +}\ 1.90 +\ 1.91 +static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.92 + OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\ 1.93 +}\ 1.94 +\ 1.95 +static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.96 + DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ 1.97 + put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ 1.98 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\ 1.99 +}\ 1.100 +\ 1.101 +static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.102 + DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ 1.103 + put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ 1.104 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\ 1.105 +}\ 1.106 +\ 1.107 +static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.108 + OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\ 1.109 +}\ 1.110 +\ 1.111 +static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.112 + DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ 1.113 + put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ 1.114 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\ 1.115 +}\ 1.116 +\ 1.117 +static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.118 + DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ 1.119 + DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ 1.120 + put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ 1.121 + put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ 1.122 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ 1.123 +}\ 1.124 +\ 1.125 +static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.126 + DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ 1.127 + DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ 1.128 + put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ 1.129 + put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ 1.130 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ 1.131 +}\ 1.132 +\ 1.133 +static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.134 + DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ 1.135 + DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ 1.136 + put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ 1.137 + put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ 1.138 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ 1.139 +}\ 1.140 +\ 1.141 +static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.142 + DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ 1.143 + DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ 1.144 + put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ 1.145 + put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ 1.146 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ 1.147 +}\ 1.148 +\ 1.149 +static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.150 + DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ 1.151 + OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\ 1.152 +}\ 1.153 +\ 1.154 +static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.155 + DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ 1.156 + DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ 1.157 + DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ 1.158 + put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ 1.159 + put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ 1.160 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\ 1.161 +}\ 1.162 +\ 1.163 +static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.164 + DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ 1.165 + DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ 1.166 + DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ 1.167 + put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ 1.168 + put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ 1.169 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\ 1.170 +}\ 1.171 +\ 1.172 +static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.173 + DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ 1.174 + DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ 1.175 + DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ 1.176 + put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ 1.177 + put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ 1.178 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\ 1.179 +}\ 1.180 +\ 1.181 +static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ 1.182 + DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ 1.183 + DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ 1.184 + DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ 1.185 + put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ 1.186 + put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ 1.187 + OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\ 1.188 +}\ 1.189 + 1.190 +static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, 1.191 + const uint8_t * src2, int dst_stride, 1.192 + int src_stride1, int h) 1.193 +{ 1.194 + int i; 1.195 + vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align; 1.196 + 1.197 + mask_ = vec_lvsl(0, src2); 1.198 + 1.199 + for (i = 0; i < h; i++) { 1.200 + 1.201 + tmp1 = vec_ld(i * src_stride1, src1); 1.202 + mask = vec_lvsl(i * src_stride1, src1); 1.203 + tmp2 = vec_ld(i * src_stride1 + 15, src1); 1.204 + 1.205 + a = vec_perm(tmp1, tmp2, mask); 1.206 + 1.207 + tmp1 = vec_ld(i * 16, src2); 1.208 + tmp2 = vec_ld(i * 16 + 15, src2); 1.209 + 1.210 + b = vec_perm(tmp1, tmp2, mask_); 1.211 + 1.212 + tmp1 = vec_ld(0, dst); 1.213 + mask = vec_lvsl(0, dst); 1.214 + tmp2 = vec_ld(15, dst); 1.215 + 1.216 + d = vec_avg(a, b); 1.217 + 1.218 + edges = vec_perm(tmp2, tmp1, mask); 1.219 + 1.220 + align = vec_lvsr(0, dst); 1.221 + 1.222 + tmp2 = vec_perm(d, edges, align); 1.223 + tmp1 = vec_perm(edges, d, align); 1.224 + 1.225 + vec_st(tmp2, 15, dst); 1.226 + vec_st(tmp1, 0 , dst); 1.227 + 1.228 + dst += dst_stride; 1.229 + } 1.230 +} 1.231 + 1.232 +static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, 1.233 + const uint8_t * src2, int dst_stride, 1.234 + int src_stride1, int h) 1.235 +{ 1.236 + int i; 1.237 + vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align; 1.238 + 1.239 + mask_ = vec_lvsl(0, src2); 1.240 + 1.241 + for (i = 0; i < h; i++) { 1.242 + 1.243 + tmp1 = vec_ld(i * src_stride1, src1); 1.244 + mask = vec_lvsl(i * src_stride1, src1); 1.245 + tmp2 = vec_ld(i * src_stride1 + 15, src1); 1.246 + 1.247 + a = vec_perm(tmp1, tmp2, mask); 1.248 + 1.249 + tmp1 = vec_ld(i * 16, src2); 1.250 + tmp2 = vec_ld(i * 16 + 15, src2); 1.251 + 1.252 + b = vec_perm(tmp1, tmp2, mask_); 1.253 + 1.254 + tmp1 = vec_ld(0, dst); 1.255 + mask = vec_lvsl(0, dst); 1.256 + tmp2 = vec_ld(15, dst); 1.257 + 1.258 + d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b)); 1.259 + 1.260 + edges = vec_perm(tmp2, tmp1, mask); 1.261 + 1.262 + align = vec_lvsr(0, dst); 1.263 + 1.264 + tmp2 = vec_perm(d, edges, align); 1.265 + tmp1 = vec_perm(edges, d, align); 1.266 + 1.267 + vec_st(tmp2, 15, dst); 1.268 + vec_st(tmp1, 0 , dst); 1.269 + 1.270 + dst += dst_stride; 1.271 + } 1.272 +} 1.273 + 1.274 +/* Implemented but could be faster 1.275 +#define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h) 1.276 +#define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h) 1.277 + */ 1.278 + 1.279 +H264_MC(put_, 16, altivec) 1.280 +H264_MC(avg_, 16, altivec) 1.281 + 1.282 + 1.283 +/**************************************************************************** 1.284 + * IDCT transform: 1.285 + ****************************************************************************/ 1.286 + 1.287 +#define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3) \ 1.288 + /* 1st stage */ \ 1.289 + vz0 = vec_add(vb0,vb2); /* temp[0] = Y[0] + Y[2] */ \ 1.290 + vz1 = vec_sub(vb0,vb2); /* temp[1] = Y[0] - Y[2] */ \ 1.291 + vz2 = vec_sra(vb1,vec_splat_u16(1)); \ 1.292 + vz2 = vec_sub(vz2,vb3); /* temp[2] = Y[1].1/2 - Y[3] */ \ 1.293 + vz3 = vec_sra(vb3,vec_splat_u16(1)); \ 1.294 + vz3 = vec_add(vb1,vz3); /* temp[3] = Y[1] + Y[3].1/2 */ \ 1.295 + /* 2nd stage: output */ \ 1.296 + va0 = vec_add(vz0,vz3); /* x[0] = temp[0] + temp[3] */ \ 1.297 + va1 = vec_add(vz1,vz2); /* x[1] = temp[1] + temp[2] */ \ 1.298 + va2 = vec_sub(vz1,vz2); /* x[2] = temp[1] - temp[2] */ \ 1.299 + va3 = vec_sub(vz0,vz3) /* x[3] = temp[0] - temp[3] */ 1.300 + 1.301 +#define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \ 1.302 + b0 = vec_mergeh( a0, a0 ); \ 1.303 + b1 = vec_mergeh( a1, a0 ); \ 1.304 + b2 = vec_mergeh( a2, a0 ); \ 1.305 + b3 = vec_mergeh( a3, a0 ); \ 1.306 + a0 = vec_mergeh( b0, b2 ); \ 1.307 + a1 = vec_mergel( b0, b2 ); \ 1.308 + a2 = vec_mergeh( b1, b3 ); \ 1.309 + a3 = vec_mergel( b1, b3 ); \ 1.310 + b0 = vec_mergeh( a0, a2 ); \ 1.311 + b1 = vec_mergel( a0, a2 ); \ 1.312 + b2 = vec_mergeh( a1, a3 ); \ 1.313 + b3 = vec_mergel( a1, a3 ) 1.314 + 1.315 +#define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \ 1.316 + vdst_orig = vec_ld(0, dst); \ 1.317 + vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); \ 1.318 + vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst); \ 1.319 + va = vec_add(va, vdst_ss); \ 1.320 + va_u8 = vec_packsu(va, zero_s16v); \ 1.321 + va_u32 = vec_splat((vec_u32)va_u8, 0); \ 1.322 + vec_ste(va_u32, element, (uint32_t*)dst); 1.323 + 1.324 +static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride) 1.325 +{ 1.326 + vec_s16 va0, va1, va2, va3; 1.327 + vec_s16 vz0, vz1, vz2, vz3; 1.328 + vec_s16 vtmp0, vtmp1, vtmp2, vtmp3; 1.329 + vec_u8 va_u8; 1.330 + vec_u32 va_u32; 1.331 + vec_s16 vdst_ss; 1.332 + const vec_u16 v6us = vec_splat_u16(6); 1.333 + vec_u8 vdst, vdst_orig; 1.334 + vec_u8 vdst_mask = vec_lvsl(0, dst); 1.335 + int element = ((unsigned long)dst & 0xf) >> 2; 1.336 + LOAD_ZERO; 1.337 + 1.338 + block[0] += 32; /* add 32 as a DC-level for rounding */ 1.339 + 1.340 + vtmp0 = vec_ld(0,block); 1.341 + vtmp1 = vec_sld(vtmp0, vtmp0, 8); 1.342 + vtmp2 = vec_ld(16,block); 1.343 + vtmp3 = vec_sld(vtmp2, vtmp2, 8); 1.344 + 1.345 + VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3); 1.346 + VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3); 1.347 + VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3); 1.348 + 1.349 + va0 = vec_sra(va0,v6us); 1.350 + va1 = vec_sra(va1,v6us); 1.351 + va2 = vec_sra(va2,v6us); 1.352 + va3 = vec_sra(va3,v6us); 1.353 + 1.354 + VEC_LOAD_U8_ADD_S16_STORE_U8(va0); 1.355 + dst += stride; 1.356 + VEC_LOAD_U8_ADD_S16_STORE_U8(va1); 1.357 + dst += stride; 1.358 + VEC_LOAD_U8_ADD_S16_STORE_U8(va2); 1.359 + dst += stride; 1.360 + VEC_LOAD_U8_ADD_S16_STORE_U8(va3); 1.361 +} 1.362 + 1.363 +#define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\ 1.364 + /* a0 = SRC(0) + SRC(4); */ \ 1.365 + vec_s16 a0v = vec_add(s0, s4); \ 1.366 + /* a2 = SRC(0) - SRC(4); */ \ 1.367 + vec_s16 a2v = vec_sub(s0, s4); \ 1.368 + /* a4 = (SRC(2)>>1) - SRC(6); */ \ 1.369 + vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6); \ 1.370 + /* a6 = (SRC(6)>>1) + SRC(2); */ \ 1.371 + vec_s16 a6v = vec_add(vec_sra(s6, onev), s2); \ 1.372 + /* b0 = a0 + a6; */ \ 1.373 + vec_s16 b0v = vec_add(a0v, a6v); \ 1.374 + /* b2 = a2 + a4; */ \ 1.375 + vec_s16 b2v = vec_add(a2v, a4v); \ 1.376 + /* b4 = a2 - a4; */ \ 1.377 + vec_s16 b4v = vec_sub(a2v, a4v); \ 1.378 + /* b6 = a0 - a6; */ \ 1.379 + vec_s16 b6v = vec_sub(a0v, a6v); \ 1.380 + /* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \ 1.381 + /* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \ 1.382 + vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \ 1.383 + /* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \ 1.384 + /* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \ 1.385 + vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\ 1.386 + /* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \ 1.387 + /* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \ 1.388 + vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\ 1.389 + /* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \ 1.390 + vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\ 1.391 + /* b1 = (a7>>2) + a1; */ \ 1.392 + vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \ 1.393 + /* b3 = a3 + (a5>>2); */ \ 1.394 + vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \ 1.395 + /* b5 = (a3>>2) - a5; */ \ 1.396 + vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \ 1.397 + /* b7 = a7 - (a1>>2); */ \ 1.398 + vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \ 1.399 + /* DST(0, b0 + b7); */ \ 1.400 + d0 = vec_add(b0v, b7v); \ 1.401 + /* DST(1, b2 + b5); */ \ 1.402 + d1 = vec_add(b2v, b5v); \ 1.403 + /* DST(2, b4 + b3); */ \ 1.404 + d2 = vec_add(b4v, b3v); \ 1.405 + /* DST(3, b6 + b1); */ \ 1.406 + d3 = vec_add(b6v, b1v); \ 1.407 + /* DST(4, b6 - b1); */ \ 1.408 + d4 = vec_sub(b6v, b1v); \ 1.409 + /* DST(5, b4 - b3); */ \ 1.410 + d5 = vec_sub(b4v, b3v); \ 1.411 + /* DST(6, b2 - b5); */ \ 1.412 + d6 = vec_sub(b2v, b5v); \ 1.413 + /* DST(7, b0 - b7); */ \ 1.414 + d7 = vec_sub(b0v, b7v); \ 1.415 +} 1.416 + 1.417 +#define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \ 1.418 + /* unaligned load */ \ 1.419 + vec_u8 hv = vec_ld( 0, dest ); \ 1.420 + vec_u8 lv = vec_ld( 7, dest ); \ 1.421 + vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv ); \ 1.422 + vec_s16 idct_sh6 = vec_sra(idctv, sixv); \ 1.423 + vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv); \ 1.424 + vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16); \ 1.425 + vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum); \ 1.426 + vec_u8 edgehv; \ 1.427 + /* unaligned store */ \ 1.428 + vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\ 1.429 + vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \ 1.430 + lv = vec_sel( lv, bodyv, edgelv ); \ 1.431 + vec_st( lv, 7, dest ); \ 1.432 + hv = vec_ld( 0, dest ); \ 1.433 + edgehv = vec_perm( zero_u8v, sel, perm_stv ); \ 1.434 + hv = vec_sel( hv, bodyv, edgehv ); \ 1.435 + vec_st( hv, 0, dest ); \ 1.436 + } 1.437 + 1.438 +static void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) { 1.439 + vec_s16 s0, s1, s2, s3, s4, s5, s6, s7; 1.440 + vec_s16 d0, d1, d2, d3, d4, d5, d6, d7; 1.441 + vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7; 1.442 + 1.443 + vec_u8 perm_ldv = vec_lvsl(0, dst); 1.444 + vec_u8 perm_stv = vec_lvsr(8, dst); 1.445 + 1.446 + const vec_u16 onev = vec_splat_u16(1); 1.447 + const vec_u16 twov = vec_splat_u16(2); 1.448 + const vec_u16 sixv = vec_splat_u16(6); 1.449 + 1.450 + const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1}; 1.451 + LOAD_ZERO; 1.452 + 1.453 + dct[0] += 32; // rounding for the >>6 at the end 1.454 + 1.455 + s0 = vec_ld(0x00, (int16_t*)dct); 1.456 + s1 = vec_ld(0x10, (int16_t*)dct); 1.457 + s2 = vec_ld(0x20, (int16_t*)dct); 1.458 + s3 = vec_ld(0x30, (int16_t*)dct); 1.459 + s4 = vec_ld(0x40, (int16_t*)dct); 1.460 + s5 = vec_ld(0x50, (int16_t*)dct); 1.461 + s6 = vec_ld(0x60, (int16_t*)dct); 1.462 + s7 = vec_ld(0x70, (int16_t*)dct); 1.463 + 1.464 + IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, 1.465 + d0, d1, d2, d3, d4, d5, d6, d7); 1.466 + 1.467 + TRANSPOSE8( d0, d1, d2, d3, d4, d5, d6, d7 ); 1.468 + 1.469 + IDCT8_1D_ALTIVEC(d0, d1, d2, d3, d4, d5, d6, d7, 1.470 + idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7); 1.471 + 1.472 + ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel); 1.473 + ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel); 1.474 + ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel); 1.475 + ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel); 1.476 + ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel); 1.477 + ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel); 1.478 + ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel); 1.479 + ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel); 1.480 +} 1.481 + 1.482 +static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, DCTELEM *block, int stride, int size) 1.483 +{ 1.484 + vec_s16 dc16; 1.485 + vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner; 1.486 + LOAD_ZERO; 1.487 + DECLARE_ALIGNED(16, int, dc); 1.488 + int i; 1.489 + 1.490 + dc = (block[0] + 32) >> 6; 1.491 + dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1); 1.492 + 1.493 + if (size == 4) 1.494 + dc16 = vec_sld(dc16, zero_s16v, 8); 1.495 + dcplus = vec_packsu(dc16, zero_s16v); 1.496 + dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v); 1.497 + 1.498 + aligner = vec_lvsr(0, dst); 1.499 + dcplus = vec_perm(dcplus, dcplus, aligner); 1.500 + dcminus = vec_perm(dcminus, dcminus, aligner); 1.501 + 1.502 + for (i = 0; i < size; i += 4) { 1.503 + v0 = vec_ld(0, dst+0*stride); 1.504 + v1 = vec_ld(0, dst+1*stride); 1.505 + v2 = vec_ld(0, dst+2*stride); 1.506 + v3 = vec_ld(0, dst+3*stride); 1.507 + 1.508 + v0 = vec_adds(v0, dcplus); 1.509 + v1 = vec_adds(v1, dcplus); 1.510 + v2 = vec_adds(v2, dcplus); 1.511 + v3 = vec_adds(v3, dcplus); 1.512 + 1.513 + v0 = vec_subs(v0, dcminus); 1.514 + v1 = vec_subs(v1, dcminus); 1.515 + v2 = vec_subs(v2, dcminus); 1.516 + v3 = vec_subs(v3, dcminus); 1.517 + 1.518 + vec_st(v0, 0, dst+0*stride); 1.519 + vec_st(v1, 0, dst+1*stride); 1.520 + vec_st(v2, 0, dst+2*stride); 1.521 + vec_st(v3, 0, dst+3*stride); 1.522 + 1.523 + dst += 4*stride; 1.524 + } 1.525 +} 1.526 + 1.527 +static void h264_idct_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride) 1.528 +{ 1.529 + h264_idct_dc_add_internal(dst, block, stride, 4); 1.530 +} 1.531 + 1.532 +static void ff_h264_idct8_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride) 1.533 +{ 1.534 + h264_idct_dc_add_internal(dst, block, stride, 8); 1.535 +} 1.536 + 1.537 +static void ff_h264_idct_add16_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){ 1.538 + int i; 1.539 + for(i=0; i<16; i++){ 1.540 + int nnz = nnzc[ scan8[i] ]; 1.541 + if(nnz){ 1.542 + if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride); 1.543 + else ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride); 1.544 + } 1.545 + } 1.546 +} 1.547 + 1.548 +static void ff_h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){ 1.549 + int i; 1.550 + for(i=0; i<16; i++){ 1.551 + if(nnzc[ scan8[i] ]) ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride); 1.552 + else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride); 1.553 + } 1.554 +} 1.555 + 1.556 +static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){ 1.557 + int i; 1.558 + for(i=0; i<16; i+=4){ 1.559 + int nnz = nnzc[ scan8[i] ]; 1.560 + if(nnz){ 1.561 + if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride); 1.562 + else ff_h264_idct8_add_altivec (dst + block_offset[i], block + i*16, stride); 1.563 + } 1.564 + } 1.565 +} 1.566 + 1.567 +static void ff_h264_idct_add8_altivec(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){ 1.568 + int i; 1.569 + for(i=16; i<16+8; i++){ 1.570 + if(nnzc[ scan8[i] ]) 1.571 + ff_h264_idct_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride); 1.572 + else if(block[i*16]) 1.573 + h264_idct_dc_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride); 1.574 + } 1.575 +} 1.576 + 1.577 +#define transpose4x16(r0, r1, r2, r3) { \ 1.578 + register vec_u8 r4; \ 1.579 + register vec_u8 r5; \ 1.580 + register vec_u8 r6; \ 1.581 + register vec_u8 r7; \ 1.582 + \ 1.583 + r4 = vec_mergeh(r0, r2); /*0, 2 set 0*/ \ 1.584 + r5 = vec_mergel(r0, r2); /*0, 2 set 1*/ \ 1.585 + r6 = vec_mergeh(r1, r3); /*1, 3 set 0*/ \ 1.586 + r7 = vec_mergel(r1, r3); /*1, 3 set 1*/ \ 1.587 + \ 1.588 + r0 = vec_mergeh(r4, r6); /*all set 0*/ \ 1.589 + r1 = vec_mergel(r4, r6); /*all set 1*/ \ 1.590 + r2 = vec_mergeh(r5, r7); /*all set 2*/ \ 1.591 + r3 = vec_mergel(r5, r7); /*all set 3*/ \ 1.592 +} 1.593 + 1.594 +static inline void write16x4(uint8_t *dst, int dst_stride, 1.595 + register vec_u8 r0, register vec_u8 r1, 1.596 + register vec_u8 r2, register vec_u8 r3) { 1.597 + DECLARE_ALIGNED(16, unsigned char, result)[64]; 1.598 + uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst; 1.599 + int int_dst_stride = dst_stride/4; 1.600 + 1.601 + vec_st(r0, 0, result); 1.602 + vec_st(r1, 16, result); 1.603 + vec_st(r2, 32, result); 1.604 + vec_st(r3, 48, result); 1.605 + /* FIXME: there has to be a better way!!!! */ 1.606 + *dst_int = *src_int; 1.607 + *(dst_int+ int_dst_stride) = *(src_int + 1); 1.608 + *(dst_int+ 2*int_dst_stride) = *(src_int + 2); 1.609 + *(dst_int+ 3*int_dst_stride) = *(src_int + 3); 1.610 + *(dst_int+ 4*int_dst_stride) = *(src_int + 4); 1.611 + *(dst_int+ 5*int_dst_stride) = *(src_int + 5); 1.612 + *(dst_int+ 6*int_dst_stride) = *(src_int + 6); 1.613 + *(dst_int+ 7*int_dst_stride) = *(src_int + 7); 1.614 + *(dst_int+ 8*int_dst_stride) = *(src_int + 8); 1.615 + *(dst_int+ 9*int_dst_stride) = *(src_int + 9); 1.616 + *(dst_int+10*int_dst_stride) = *(src_int + 10); 1.617 + *(dst_int+11*int_dst_stride) = *(src_int + 11); 1.618 + *(dst_int+12*int_dst_stride) = *(src_int + 12); 1.619 + *(dst_int+13*int_dst_stride) = *(src_int + 13); 1.620 + *(dst_int+14*int_dst_stride) = *(src_int + 14); 1.621 + *(dst_int+15*int_dst_stride) = *(src_int + 15); 1.622 +} 1.623 + 1.624 +/** \brief performs a 6x16 transpose of data in src, and stores it to dst 1.625 + \todo FIXME: see if we can't spare some vec_lvsl() by them factorizing 1.626 + out of unaligned_load() */ 1.627 +#define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\ 1.628 + register vec_u8 r0 = unaligned_load(0, src); \ 1.629 + register vec_u8 r1 = unaligned_load( src_stride, src); \ 1.630 + register vec_u8 r2 = unaligned_load(2* src_stride, src); \ 1.631 + register vec_u8 r3 = unaligned_load(3* src_stride, src); \ 1.632 + register vec_u8 r4 = unaligned_load(4* src_stride, src); \ 1.633 + register vec_u8 r5 = unaligned_load(5* src_stride, src); \ 1.634 + register vec_u8 r6 = unaligned_load(6* src_stride, src); \ 1.635 + register vec_u8 r7 = unaligned_load(7* src_stride, src); \ 1.636 + register vec_u8 r14 = unaligned_load(14*src_stride, src); \ 1.637 + register vec_u8 r15 = unaligned_load(15*src_stride, src); \ 1.638 + \ 1.639 + r8 = unaligned_load( 8*src_stride, src); \ 1.640 + r9 = unaligned_load( 9*src_stride, src); \ 1.641 + r10 = unaligned_load(10*src_stride, src); \ 1.642 + r11 = unaligned_load(11*src_stride, src); \ 1.643 + r12 = unaligned_load(12*src_stride, src); \ 1.644 + r13 = unaligned_load(13*src_stride, src); \ 1.645 + \ 1.646 + /*Merge first pairs*/ \ 1.647 + r0 = vec_mergeh(r0, r8); /*0, 8*/ \ 1.648 + r1 = vec_mergeh(r1, r9); /*1, 9*/ \ 1.649 + r2 = vec_mergeh(r2, r10); /*2,10*/ \ 1.650 + r3 = vec_mergeh(r3, r11); /*3,11*/ \ 1.651 + r4 = vec_mergeh(r4, r12); /*4,12*/ \ 1.652 + r5 = vec_mergeh(r5, r13); /*5,13*/ \ 1.653 + r6 = vec_mergeh(r6, r14); /*6,14*/ \ 1.654 + r7 = vec_mergeh(r7, r15); /*7,15*/ \ 1.655 + \ 1.656 + /*Merge second pairs*/ \ 1.657 + r8 = vec_mergeh(r0, r4); /*0,4, 8,12 set 0*/ \ 1.658 + r9 = vec_mergel(r0, r4); /*0,4, 8,12 set 1*/ \ 1.659 + r10 = vec_mergeh(r1, r5); /*1,5, 9,13 set 0*/ \ 1.660 + r11 = vec_mergel(r1, r5); /*1,5, 9,13 set 1*/ \ 1.661 + r12 = vec_mergeh(r2, r6); /*2,6,10,14 set 0*/ \ 1.662 + r13 = vec_mergel(r2, r6); /*2,6,10,14 set 1*/ \ 1.663 + r14 = vec_mergeh(r3, r7); /*3,7,11,15 set 0*/ \ 1.664 + r15 = vec_mergel(r3, r7); /*3,7,11,15 set 1*/ \ 1.665 + \ 1.666 + /*Third merge*/ \ 1.667 + r0 = vec_mergeh(r8, r12); /*0,2,4,6,8,10,12,14 set 0*/ \ 1.668 + r1 = vec_mergel(r8, r12); /*0,2,4,6,8,10,12,14 set 1*/ \ 1.669 + r2 = vec_mergeh(r9, r13); /*0,2,4,6,8,10,12,14 set 2*/ \ 1.670 + r4 = vec_mergeh(r10, r14); /*1,3,5,7,9,11,13,15 set 0*/ \ 1.671 + r5 = vec_mergel(r10, r14); /*1,3,5,7,9,11,13,15 set 1*/ \ 1.672 + r6 = vec_mergeh(r11, r15); /*1,3,5,7,9,11,13,15 set 2*/ \ 1.673 + /* Don't need to compute 3 and 7*/ \ 1.674 + \ 1.675 + /*Final merge*/ \ 1.676 + r8 = vec_mergeh(r0, r4); /*all set 0*/ \ 1.677 + r9 = vec_mergel(r0, r4); /*all set 1*/ \ 1.678 + r10 = vec_mergeh(r1, r5); /*all set 2*/ \ 1.679 + r11 = vec_mergel(r1, r5); /*all set 3*/ \ 1.680 + r12 = vec_mergeh(r2, r6); /*all set 4*/ \ 1.681 + r13 = vec_mergel(r2, r6); /*all set 5*/ \ 1.682 + /* Don't need to compute 14 and 15*/ \ 1.683 + \ 1.684 +} 1.685 + 1.686 +// out: o = |x-y| < a 1.687 +static inline vec_u8 diff_lt_altivec ( register vec_u8 x, 1.688 + register vec_u8 y, 1.689 + register vec_u8 a) { 1.690 + 1.691 + register vec_u8 diff = vec_subs(x, y); 1.692 + register vec_u8 diffneg = vec_subs(y, x); 1.693 + register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */ 1.694 + o = (vec_u8)vec_cmplt(o, a); 1.695 + return o; 1.696 +} 1.697 + 1.698 +static inline vec_u8 h264_deblock_mask ( register vec_u8 p0, 1.699 + register vec_u8 p1, 1.700 + register vec_u8 q0, 1.701 + register vec_u8 q1, 1.702 + register vec_u8 alpha, 1.703 + register vec_u8 beta) { 1.704 + 1.705 + register vec_u8 mask; 1.706 + register vec_u8 tempmask; 1.707 + 1.708 + mask = diff_lt_altivec(p0, q0, alpha); 1.709 + tempmask = diff_lt_altivec(p1, p0, beta); 1.710 + mask = vec_and(mask, tempmask); 1.711 + tempmask = diff_lt_altivec(q1, q0, beta); 1.712 + mask = vec_and(mask, tempmask); 1.713 + 1.714 + return mask; 1.715 +} 1.716 + 1.717 +// out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0) 1.718 +static inline vec_u8 h264_deblock_q1(register vec_u8 p0, 1.719 + register vec_u8 p1, 1.720 + register vec_u8 p2, 1.721 + register vec_u8 q0, 1.722 + register vec_u8 tc0) { 1.723 + 1.724 + register vec_u8 average = vec_avg(p0, q0); 1.725 + register vec_u8 temp; 1.726 + register vec_u8 uncliped; 1.727 + register vec_u8 ones; 1.728 + register vec_u8 max; 1.729 + register vec_u8 min; 1.730 + register vec_u8 newp1; 1.731 + 1.732 + temp = vec_xor(average, p2); 1.733 + average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */ 1.734 + ones = vec_splat_u8(1); 1.735 + temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */ 1.736 + uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */ 1.737 + max = vec_adds(p1, tc0); 1.738 + min = vec_subs(p1, tc0); 1.739 + newp1 = vec_max(min, uncliped); 1.740 + newp1 = vec_min(max, newp1); 1.741 + return newp1; 1.742 +} 1.743 + 1.744 +#define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \ 1.745 + \ 1.746 + const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \ 1.747 + \ 1.748 + register vec_u8 pq0bit = vec_xor(p0,q0); \ 1.749 + register vec_u8 q1minus; \ 1.750 + register vec_u8 p0minus; \ 1.751 + register vec_u8 stage1; \ 1.752 + register vec_u8 stage2; \ 1.753 + register vec_u8 vec160; \ 1.754 + register vec_u8 delta; \ 1.755 + register vec_u8 deltaneg; \ 1.756 + \ 1.757 + q1minus = vec_nor(q1, q1); /* 255 - q1 */ \ 1.758 + stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \ 1.759 + stage2 = vec_sr(stage1, vec_splat_u8(1)); /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */ \ 1.760 + p0minus = vec_nor(p0, p0); /* 255 - p0 */ \ 1.761 + stage1 = vec_avg(q0, p0minus); /* (q0 - p0 + 256)>>1 */ \ 1.762 + pq0bit = vec_and(pq0bit, vec_splat_u8(1)); \ 1.763 + stage2 = vec_avg(stage2, pq0bit); /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \ 1.764 + stage2 = vec_adds(stage2, stage1); /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */ \ 1.765 + vec160 = vec_ld(0, &A0v); \ 1.766 + deltaneg = vec_subs(vec160, stage2); /* -d */ \ 1.767 + delta = vec_subs(stage2, vec160); /* d */ \ 1.768 + deltaneg = vec_min(tc0masked, deltaneg); \ 1.769 + delta = vec_min(tc0masked, delta); \ 1.770 + p0 = vec_subs(p0, deltaneg); \ 1.771 + q0 = vec_subs(q0, delta); \ 1.772 + p0 = vec_adds(p0, delta); \ 1.773 + q0 = vec_adds(q0, deltaneg); \ 1.774 +} 1.775 + 1.776 +#define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) { \ 1.777 + DECLARE_ALIGNED(16, unsigned char, temp)[16]; \ 1.778 + register vec_u8 alphavec; \ 1.779 + register vec_u8 betavec; \ 1.780 + register vec_u8 mask; \ 1.781 + register vec_u8 p1mask; \ 1.782 + register vec_u8 q1mask; \ 1.783 + register vector signed char tc0vec; \ 1.784 + register vec_u8 finaltc0; \ 1.785 + register vec_u8 tc0masked; \ 1.786 + register vec_u8 newp1; \ 1.787 + register vec_u8 newq1; \ 1.788 + \ 1.789 + temp[0] = alpha; \ 1.790 + temp[1] = beta; \ 1.791 + alphavec = vec_ld(0, temp); \ 1.792 + betavec = vec_splat(alphavec, 0x1); \ 1.793 + alphavec = vec_splat(alphavec, 0x0); \ 1.794 + mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */ \ 1.795 + \ 1.796 + *((int *)temp) = *((int *)tc0); \ 1.797 + tc0vec = vec_ld(0, (signed char*)temp); \ 1.798 + tc0vec = vec_mergeh(tc0vec, tc0vec); \ 1.799 + tc0vec = vec_mergeh(tc0vec, tc0vec); \ 1.800 + mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1))); /* if tc0[i] >= 0 */ \ 1.801 + finaltc0 = vec_and((vec_u8)tc0vec, mask); /* tc = tc0 */ \ 1.802 + \ 1.803 + p1mask = diff_lt_altivec(p2, p0, betavec); \ 1.804 + p1mask = vec_and(p1mask, mask); /* if ( |p2 - p0| < beta) */ \ 1.805 + tc0masked = vec_and(p1mask, (vec_u8)tc0vec); \ 1.806 + finaltc0 = vec_sub(finaltc0, p1mask); /* tc++ */ \ 1.807 + newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked); \ 1.808 + /*end if*/ \ 1.809 + \ 1.810 + q1mask = diff_lt_altivec(q2, q0, betavec); \ 1.811 + q1mask = vec_and(q1mask, mask); /* if ( |q2 - q0| < beta ) */\ 1.812 + tc0masked = vec_and(q1mask, (vec_u8)tc0vec); \ 1.813 + finaltc0 = vec_sub(finaltc0, q1mask); /* tc++ */ \ 1.814 + newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked); \ 1.815 + /*end if*/ \ 1.816 + \ 1.817 + h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0); \ 1.818 + p1 = newp1; \ 1.819 + q1 = newq1; \ 1.820 +} 1.821 + 1.822 +static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) { 1.823 + 1.824 + if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) { 1.825 + register vec_u8 p2 = vec_ld(-3*stride, pix); 1.826 + register vec_u8 p1 = vec_ld(-2*stride, pix); 1.827 + register vec_u8 p0 = vec_ld(-1*stride, pix); 1.828 + register vec_u8 q0 = vec_ld(0, pix); 1.829 + register vec_u8 q1 = vec_ld(stride, pix); 1.830 + register vec_u8 q2 = vec_ld(2*stride, pix); 1.831 + h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0); 1.832 + vec_st(p1, -2*stride, pix); 1.833 + vec_st(p0, -1*stride, pix); 1.834 + vec_st(q0, 0, pix); 1.835 + vec_st(q1, stride, pix); 1.836 + } 1.837 +} 1.838 + 1.839 +static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) { 1.840 + 1.841 + register vec_u8 line0, line1, line2, line3, line4, line5; 1.842 + if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0) 1.843 + return; 1.844 + readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5); 1.845 + h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0); 1.846 + transpose4x16(line1, line2, line3, line4); 1.847 + write16x4(pix-2, stride, line1, line2, line3, line4); 1.848 +} 1.849 + 1.850 +static av_always_inline 1.851 +void weight_h264_WxH_altivec(uint8_t *block, int stride, int log2_denom, int weight, int offset, int w, int h) 1.852 +{ 1.853 + int y, aligned; 1.854 + vec_u8 vblock; 1.855 + vec_s16 vtemp, vweight, voffset, v0, v1; 1.856 + vec_u16 vlog2_denom; 1.857 + DECLARE_ALIGNED(16, int32_t, temp)[4]; 1.858 + LOAD_ZERO; 1.859 + 1.860 + offset <<= log2_denom; 1.861 + if(log2_denom) offset += 1<<(log2_denom-1); 1.862 + temp[0] = log2_denom; 1.863 + temp[1] = weight; 1.864 + temp[2] = offset; 1.865 + 1.866 + vtemp = (vec_s16)vec_ld(0, temp); 1.867 + vlog2_denom = (vec_u16)vec_splat(vtemp, 1); 1.868 + vweight = vec_splat(vtemp, 3); 1.869 + voffset = vec_splat(vtemp, 5); 1.870 + aligned = !((unsigned long)block & 0xf); 1.871 + 1.872 + for (y=0; y<h; y++) { 1.873 + vblock = vec_ld(0, block); 1.874 + 1.875 + v0 = (vec_s16)vec_mergeh(zero_u8v, vblock); 1.876 + v1 = (vec_s16)vec_mergel(zero_u8v, vblock); 1.877 + 1.878 + if (w == 16 || aligned) { 1.879 + v0 = vec_mladd(v0, vweight, zero_s16v); 1.880 + v0 = vec_adds(v0, voffset); 1.881 + v0 = vec_sra(v0, vlog2_denom); 1.882 + } 1.883 + if (w == 16 || !aligned) { 1.884 + v1 = vec_mladd(v1, vweight, zero_s16v); 1.885 + v1 = vec_adds(v1, voffset); 1.886 + v1 = vec_sra(v1, vlog2_denom); 1.887 + } 1.888 + vblock = vec_packsu(v0, v1); 1.889 + vec_st(vblock, 0, block); 1.890 + 1.891 + block += stride; 1.892 + } 1.893 +} 1.894 + 1.895 +static av_always_inline 1.896 +void biweight_h264_WxH_altivec(uint8_t *dst, uint8_t *src, int stride, int log2_denom, 1.897 + int weightd, int weights, int offset, int w, int h) 1.898 +{ 1.899 + int y, dst_aligned, src_aligned; 1.900 + vec_u8 vsrc, vdst; 1.901 + vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3; 1.902 + vec_u16 vlog2_denom; 1.903 + DECLARE_ALIGNED(16, int32_t, temp)[4]; 1.904 + LOAD_ZERO; 1.905 + 1.906 + offset = ((offset + 1) | 1) << log2_denom; 1.907 + temp[0] = log2_denom+1; 1.908 + temp[1] = weights; 1.909 + temp[2] = weightd; 1.910 + temp[3] = offset; 1.911 + 1.912 + vtemp = (vec_s16)vec_ld(0, temp); 1.913 + vlog2_denom = (vec_u16)vec_splat(vtemp, 1); 1.914 + vweights = vec_splat(vtemp, 3); 1.915 + vweightd = vec_splat(vtemp, 5); 1.916 + voffset = vec_splat(vtemp, 7); 1.917 + dst_aligned = !((unsigned long)dst & 0xf); 1.918 + src_aligned = !((unsigned long)src & 0xf); 1.919 + 1.920 + for (y=0; y<h; y++) { 1.921 + vdst = vec_ld(0, dst); 1.922 + vsrc = vec_ld(0, src); 1.923 + 1.924 + v0 = (vec_s16)vec_mergeh(zero_u8v, vdst); 1.925 + v1 = (vec_s16)vec_mergel(zero_u8v, vdst); 1.926 + v2 = (vec_s16)vec_mergeh(zero_u8v, vsrc); 1.927 + v3 = (vec_s16)vec_mergel(zero_u8v, vsrc); 1.928 + 1.929 + if (w == 8) { 1.930 + if (src_aligned) 1.931 + v3 = v2; 1.932 + else 1.933 + v2 = v3; 1.934 + } 1.935 + 1.936 + if (w == 16 || dst_aligned) { 1.937 + v0 = vec_mladd(v0, vweightd, zero_s16v); 1.938 + v2 = vec_mladd(v2, vweights, zero_s16v); 1.939 + 1.940 + v0 = vec_adds(v0, voffset); 1.941 + v0 = vec_adds(v0, v2); 1.942 + v0 = vec_sra(v0, vlog2_denom); 1.943 + } 1.944 + if (w == 16 || !dst_aligned) { 1.945 + v1 = vec_mladd(v1, vweightd, zero_s16v); 1.946 + v3 = vec_mladd(v3, vweights, zero_s16v); 1.947 + 1.948 + v1 = vec_adds(v1, voffset); 1.949 + v1 = vec_adds(v1, v3); 1.950 + v1 = vec_sra(v1, vlog2_denom); 1.951 + } 1.952 + vdst = vec_packsu(v0, v1); 1.953 + vec_st(vdst, 0, dst); 1.954 + 1.955 + dst += stride; 1.956 + src += stride; 1.957 + } 1.958 +} 1.959 + 1.960 +#define H264_WEIGHT(W,H) \ 1.961 +static void ff_weight_h264_pixels ## W ## x ## H ## _altivec(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \ 1.962 + weight_h264_WxH_altivec(block, stride, log2_denom, weight, offset, W, H); \ 1.963 +}\ 1.964 +static void ff_biweight_h264_pixels ## W ## x ## H ## _altivec(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset){ \ 1.965 + biweight_h264_WxH_altivec(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \ 1.966 +} 1.967 + 1.968 +H264_WEIGHT(16,16) 1.969 +H264_WEIGHT(16, 8) 1.970 +H264_WEIGHT( 8,16) 1.971 +H264_WEIGHT( 8, 8) 1.972 +H264_WEIGHT( 8, 4) 1.973 + 1.974 +void dsputil_h264_init_ppc(DSPContext* c) { 1.975 + c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec; 1.976 + c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec; 1.977 + 1.978 +#define dspfunc(PFX, IDX, NUM) \ 1.979 + c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \ 1.980 + c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \ 1.981 + c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \ 1.982 + c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \ 1.983 + c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \ 1.984 + c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \ 1.985 + c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \ 1.986 + c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \ 1.987 + c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \ 1.988 + c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \ 1.989 + c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \ 1.990 + c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \ 1.991 + c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \ 1.992 + c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \ 1.993 + c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \ 1.994 + c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec 1.995 + 1.996 + dspfunc(put_h264_qpel, 0, 16); 1.997 + dspfunc(avg_h264_qpel, 0, 16); 1.998 +#undef dspfunc 1.999 +} 1.1000 + 1.1001 +void ff_h264dsp_init_ppc(H264DSPContext *c){ 1.1002 + c->h264_idct_dc_add= h264_idct_dc_add_altivec; 1.1003 + c->h264_idct_add = ff_h264_idct_add_altivec; 1.1004 + c->h264_idct_add8 = ff_h264_idct_add8_altivec; 1.1005 + c->h264_idct_add16 = ff_h264_idct_add16_altivec; 1.1006 + c->h264_idct_add16intra = ff_h264_idct_add16intra_altivec; 1.1007 + 1.1008 + c->h264_idct8_dc_add = ff_h264_idct8_dc_add_altivec; 1.1009 + c->h264_idct8_add = ff_h264_idct8_add_altivec; 1.1010 + c->h264_idct8_add4 = ff_h264_idct8_add4_altivec; 1.1011 + c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec; 1.1012 + c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec; 1.1013 + 1.1014 + c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels16x16_altivec; 1.1015 + c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels16x8_altivec; 1.1016 + c->weight_h264_pixels_tab[2] = ff_weight_h264_pixels8x16_altivec; 1.1017 + c->weight_h264_pixels_tab[3] = ff_weight_h264_pixels8x8_altivec; 1.1018 + c->weight_h264_pixels_tab[4] = ff_weight_h264_pixels8x4_altivec; 1.1019 + c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16x16_altivec; 1.1020 + c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels16x8_altivec; 1.1021 + c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels8x16_altivec; 1.1022 + c->biweight_h264_pixels_tab[3] = ff_biweight_h264_pixels8x8_altivec; 1.1023 + c->biweight_h264_pixels_tab[4] = ff_biweight_h264_pixels8x4_altivec; 1.1024 +}