| OLD | NEW |
|---|
| (Empty) | |
| 1 /* |
| 2 * Copyright (c) 2015 The WebM project authors. All Rights Reserved. |
| 3 * |
| 4 * Use of this source code is governed by a BSD-style license |
| 5 * that can be found in the LICENSE file in the root of the source |
| 6 * tree. An additional intellectual property rights grant can be found |
| 7 * in the file PATENTS. All contributing project authors may |
| 8 * be found in the AUTHORS file in the root of the source tree. |
| 9 */ |
| 10 |
| 11 #include <assert.h> |
| 12 |
| 13 #include "./vp9_rtcd.h" |
| 14 #include "vp9/encoder/mips/msa/vp9_fdct_msa.h" |
| 15 |
| 16 static void fdct8x16_1d_column(const int16_t *input, int16_t *tmp_ptr, |
| 17 int32_t src_stride) { |
| 18 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; |
| 19 v8i16 in0, in1, in2, in3, in4, in5, in6, in7; |
| 20 v8i16 in8, in9, in10, in11, in12, in13, in14, in15; |
| 21 v8i16 stp21, stp22, stp23, stp24, stp25, stp26, stp30; |
| 22 v8i16 stp31, stp32, stp33, stp34, stp35, stp36, stp37; |
| 23 v8i16 vec0, vec1, vec2, vec3, vec4, vec5, cnst0, cnst1, cnst4, cnst5; |
| 24 v8i16 coeff = { cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64, |
| 25 -cospi_8_64, -cospi_24_64, cospi_12_64, cospi_20_64 }; |
| 26 v8i16 coeff1 = { cospi_2_64, cospi_30_64, cospi_14_64, cospi_18_64, |
| 27 cospi_10_64, cospi_22_64, cospi_6_64, cospi_26_64 }; |
| 28 v8i16 coeff2 = { -cospi_2_64, -cospi_10_64, -cospi_18_64, -cospi_26_64, |
| 29 0, 0, 0, 0 }; |
| 30 |
| 31 LD_SH16(input, src_stride, |
| 32 in0, in1, in2, in3, in4, in5, in6, in7, |
| 33 in8, in9, in10, in11, in12, in13, in14, in15); |
| 34 SLLI_4V(in0, in1, in2, in3, 2); |
| 35 SLLI_4V(in4, in5, in6, in7, 2); |
| 36 SLLI_4V(in8, in9, in10, in11, 2); |
| 37 SLLI_4V(in12, in13, in14, in15, 2); |
| 38 ADD4(in0, in15, in1, in14, in2, in13, in3, in12, tmp0, tmp1, tmp2, tmp3); |
| 39 ADD4(in4, in11, in5, in10, in6, in9, in7, in8, tmp4, tmp5, tmp6, tmp7); |
| 40 VP9_FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, |
| 41 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7); |
| 42 ST_SH8(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp_ptr, 32); |
| 43 SUB4(in0, in15, in1, in14, in2, in13, in3, in12, in15, in14, in13, in12); |
| 44 SUB4(in4, in11, in5, in10, in6, in9, in7, in8, in11, in10, in9, in8); |
| 45 |
| 46 tmp_ptr += 16; |
| 47 |
| 48 /* stp 1 */ |
| 49 ILVL_H2_SH(in10, in13, in11, in12, vec2, vec4); |
| 50 ILVR_H2_SH(in10, in13, in11, in12, vec3, vec5); |
| 51 |
| 52 cnst4 = __msa_splati_h(coeff, 0); |
| 53 stp25 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst4); |
| 54 |
| 55 cnst5 = __msa_splati_h(coeff, 1); |
| 56 cnst5 = __msa_ilvev_h(cnst5, cnst4); |
| 57 stp22 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst5); |
| 58 stp24 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst4); |
| 59 stp23 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst5); |
| 60 |
| 61 /* stp2 */ |
| 62 BUTTERFLY_4(in8, in9, stp22, stp23, stp30, stp31, stp32, stp33); |
| 63 BUTTERFLY_4(in15, in14, stp25, stp24, stp37, stp36, stp35, stp34); |
| 64 ILVL_H2_SH(stp36, stp31, stp35, stp32, vec2, vec4); |
| 65 ILVR_H2_SH(stp36, stp31, stp35, stp32, vec3, vec5); |
| 66 SPLATI_H2_SH(coeff, 2, 3, cnst0, cnst1); |
| 67 cnst0 = __msa_ilvev_h(cnst0, cnst1); |
| 68 stp26 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst0); |
| 69 |
| 70 cnst0 = __msa_splati_h(coeff, 4); |
| 71 cnst1 = __msa_ilvev_h(cnst1, cnst0); |
| 72 stp21 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst1); |
| 73 |
| 74 BUTTERFLY_4(stp30, stp37, stp26, stp21, in8, in15, in14, in9); |
| 75 ILVRL_H2_SH(in15, in8, vec1, vec0); |
| 76 SPLATI_H2_SH(coeff1, 0, 1, cnst0, cnst1); |
| 77 cnst0 = __msa_ilvev_h(cnst0, cnst1); |
| 78 |
| 79 in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0); |
| 80 ST_SH(in8, tmp_ptr); |
| 81 |
| 82 cnst0 = __msa_splati_h(coeff2, 0); |
| 83 cnst0 = __msa_ilvev_h(cnst1, cnst0); |
| 84 in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0); |
| 85 ST_SH(in8, tmp_ptr + 224); |
| 86 |
| 87 ILVRL_H2_SH(in14, in9, vec1, vec0); |
| 88 SPLATI_H2_SH(coeff1, 2, 3, cnst0, cnst1); |
| 89 cnst1 = __msa_ilvev_h(cnst1, cnst0); |
| 90 |
| 91 in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1); |
| 92 ST_SH(in8, tmp_ptr + 128); |
| 93 |
| 94 cnst1 = __msa_splati_h(coeff2, 2); |
| 95 cnst0 = __msa_ilvev_h(cnst0, cnst1); |
| 96 in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0); |
| 97 ST_SH(in8, tmp_ptr + 96); |
| 98 |
| 99 SPLATI_H2_SH(coeff, 2, 5, cnst0, cnst1); |
| 100 cnst1 = __msa_ilvev_h(cnst1, cnst0); |
| 101 |
| 102 stp25 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1); |
| 103 |
| 104 cnst1 = __msa_splati_h(coeff, 3); |
| 105 cnst1 = __msa_ilvev_h(cnst0, cnst1); |
| 106 stp22 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1); |
| 107 |
| 108 /* stp4 */ |
| 109 ADD2(stp34, stp25, stp33, stp22, in13, in10); |
| 110 |
| 111 ILVRL_H2_SH(in13, in10, vec1, vec0); |
| 112 SPLATI_H2_SH(coeff1, 4, 5, cnst0, cnst1); |
| 113 cnst0 = __msa_ilvev_h(cnst0, cnst1); |
| 114 in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0); |
| 115 ST_SH(in8, tmp_ptr + 64); |
| 116 |
| 117 cnst0 = __msa_splati_h(coeff2, 1); |
| 118 cnst0 = __msa_ilvev_h(cnst1, cnst0); |
| 119 in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0); |
| 120 ST_SH(in8, tmp_ptr + 160); |
| 121 |
| 122 SUB2(stp34, stp25, stp33, stp22, in12, in11); |
| 123 ILVRL_H2_SH(in12, in11, vec1, vec0); |
| 124 SPLATI_H2_SH(coeff1, 6, 7, cnst0, cnst1); |
| 125 cnst1 = __msa_ilvev_h(cnst1, cnst0); |
| 126 |
| 127 in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1); |
| 128 ST_SH(in8, tmp_ptr + 192); |
| 129 |
| 130 cnst1 = __msa_splati_h(coeff2, 3); |
| 131 cnst0 = __msa_ilvev_h(cnst0, cnst1); |
| 132 in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0); |
| 133 ST_SH(in8, tmp_ptr + 32); |
| 134 } |
| 135 |
| 136 static void fdct16x8_1d_row(int16_t *input, int16_t *output) { |
| 137 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; |
| 138 v8i16 in0, in1, in2, in3, in4, in5, in6, in7; |
| 139 v8i16 in8, in9, in10, in11, in12, in13, in14, in15; |
| 140 |
| 141 LD_SH8(input, 16, in0, in1, in2, in3, in4, in5, in6, in7); |
| 142 LD_SH8((input + 8), 16, in8, in9, in10, in11, in12, in13, in14, in15); |
| 143 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, |
| 144 in0, in1, in2, in3, in4, in5, in6, in7); |
| 145 TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15, |
| 146 in8, in9, in10, in11, in12, in13, in14, in15); |
| 147 ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3); |
| 148 ADD4(in4, 1, in5, 1, in6, 1, in7, 1, in4, in5, in6, in7); |
| 149 ADD4(in8, 1, in9, 1, in10, 1, in11, 1, in8, in9, in10, in11); |
| 150 ADD4(in12, 1, in13, 1, in14, 1, in15, 1, in12, in13, in14, in15); |
| 151 SRA_4V(in0, in1, in2, in3, 2); |
| 152 SRA_4V(in4, in5, in6, in7, 2); |
| 153 SRA_4V(in8, in9, in10, in11, 2); |
| 154 SRA_4V(in12, in13, in14, in15, 2); |
| 155 BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11, |
| 156 in12, in13, in14, in15, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, |
| 157 tmp6, tmp7, in8, in9, in10, in11, in12, in13, in14, in15); |
| 158 ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, input, 16); |
| 159 VP9_FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, |
| 160 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7); |
| 161 LD_SH8(input, 16, in8, in9, in10, in11, in12, in13, in14, in15); |
| 162 VP9_FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15, |
| 163 in0, in1, in2, in3, in4, in5, in6, in7); |
| 164 TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, |
| 165 tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3); |
| 166 ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, output, 16); |
| 167 TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, |
| 168 tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7); |
| 169 ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, output + 8, 16); |
| 170 } |
| 171 |
| 172 void vp9_fdct16x16_msa(const int16_t *input, int16_t *output, |
| 173 int32_t src_stride) { |
| 174 int32_t i; |
| 175 DECLARE_ALIGNED(32, int16_t, tmp_buf[16 * 16]); |
| 176 |
| 177 /* column transform */ |
| 178 for (i = 0; i < 2; ++i) { |
| 179 fdct8x16_1d_column((input + 8 * i), (&tmp_buf[0] + 8 * i), src_stride); |
| 180 } |
| 181 |
| 182 /* row transform */ |
| 183 for (i = 0; i < 2; ++i) { |
| 184 fdct16x8_1d_row((&tmp_buf[0] + (128 * i)), (output + (128 * i))); |
| 185 } |
| 186 } |
| 187 |
| 188 void vp9_fdct16x16_1_msa(const int16_t *input, int16_t *out, int32_t stride) { |
| 189 out[1] = 0; |
| 190 |
| 191 out[0] = VP9_LD_HADD(input, stride); |
| 192 out[0] += VP9_LD_HADD(input + 8, stride); |
| 193 out[0] += VP9_LD_HADD(input + 16 * 8, stride); |
| 194 out[0] += VP9_LD_HADD(input + 16 * 8 + 8, stride); |
| 195 out[0] >>= 1; |
| 196 } |
| 197 |
| 198 static void fadst16_cols_step1_msa(const int16_t *input, int32_t stride, |
| 199 const int32_t *const0, int16_t *int_buf) { |
| 200 v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15; |
| 201 v8i16 tp0, tp1, tp2, tp3, g0, g1, g2, g3, g8, g9, g10, g11, h0, h1, h2, h3; |
| 202 v4i32 k0, k1, k2, k3; |
| 203 |
| 204 /* load input data */ |
| 205 r0 = LD_SH(input); |
| 206 r15 = LD_SH(input + 15 * stride); |
| 207 r7 = LD_SH(input + 7 * stride); |
| 208 r8 = LD_SH(input + 8 * stride); |
| 209 SLLI_4V(r0, r15, r7, r8, 2); |
| 210 |
| 211 /* stage 1 */ |
| 212 LD_SW2(const0, 4, k0, k1); |
| 213 LD_SW2(const0 + 8, 4, k2, k3); |
| 214 VP9_MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3); |
| 215 |
| 216 r3 = LD_SH(input + 3 * stride); |
| 217 r4 = LD_SH(input + 4 * stride); |
| 218 r11 = LD_SH(input + 11 * stride); |
| 219 r12 = LD_SH(input + 12 * stride); |
| 220 SLLI_4V(r3, r4, r11, r12, 2); |
| 221 |
| 222 LD_SW2(const0 + 4 * 4, 4, k0, k1); |
| 223 LD_SW2(const0 + 4 * 6, 4, k2, k3); |
| 224 VP9_MADD_BF(r11, r4, r3, r12, k0, k1, k2, k3, g8, g9, g10, g11); |
| 225 |
| 226 /* stage 2 */ |
| 227 BUTTERFLY_4(g0, g2, g10, g8, tp0, tp2, tp3, tp1); |
| 228 ST_SH2(tp0, tp2, int_buf, 8); |
| 229 ST_SH2(tp1, tp3, int_buf + 4 * 8, 8); |
| 230 |
| 231 LD_SW2(const0 + 4 * 8, 4, k0, k1); |
| 232 k2 = LD_SW(const0 + 4 * 10); |
| 233 VP9_MADD_BF(g1, g3, g9, g11, k0, k1, k2, k0, h0, h1, h2, h3); |
| 234 |
| 235 ST_SH2(h0, h1, int_buf + 8 * 8, 8); |
| 236 ST_SH2(h3, h2, int_buf + 12 * 8, 8); |
| 237 |
| 238 r9 = LD_SH(input + 9 * stride); |
| 239 r6 = LD_SH(input + 6 * stride); |
| 240 r1 = LD_SH(input + stride); |
| 241 r14 = LD_SH(input + 14 * stride); |
| 242 SLLI_4V(r9, r6, r1, r14, 2); |
| 243 |
| 244 LD_SW2(const0 + 4 * 11, 4, k0, k1); |
| 245 LD_SW2(const0 + 4 * 13, 4, k2, k3); |
| 246 VP9_MADD_BF(r9, r6, r1, r14, k0, k1, k2, k3, g0, g1, g2, g3); |
| 247 |
| 248 ST_SH2(g1, g3, int_buf + 3 * 8, 4 * 8); |
| 249 |
| 250 r13 = LD_SH(input + 13 * stride); |
| 251 r2 = LD_SH(input + 2 * stride); |
| 252 r5 = LD_SH(input + 5 * stride); |
| 253 r10 = LD_SH(input + 10 * stride); |
| 254 SLLI_4V(r13, r2, r5, r10, 2); |
| 255 |
| 256 LD_SW2(const0 + 4 * 15, 4, k0, k1); |
| 257 LD_SW2(const0 + 4 * 17, 4, k2, k3); |
| 258 VP9_MADD_BF(r13, r2, r5, r10, k0, k1, k2, k3, h0, h1, h2, h3); |
| 259 |
| 260 ST_SH2(h1, h3, int_buf + 11 * 8, 4 * 8); |
| 261 |
| 262 BUTTERFLY_4(h0, h2, g2, g0, tp0, tp1, tp2, tp3); |
| 263 ST_SH4(tp0, tp1, tp2, tp3, int_buf + 2 * 8, 4 * 8); |
| 264 } |
| 265 |
| 266 static void fadst16_cols_step2_msa(int16_t *int_buf, const int32_t *const0, |
| 267 int16_t *out) { |
| 268 int16_t *out_ptr = out + 128; |
| 269 v8i16 tp0, tp1, tp2, tp3, g5, g7, g13, g15; |
| 270 v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h10, h11; |
| 271 v8i16 out0, out1, out2, out3, out4, out5, out6, out7; |
| 272 v8i16 out8, out9, out10, out11, out12, out13, out14, out15; |
| 273 v4i32 k0, k1, k2, k3; |
| 274 |
| 275 LD_SH2(int_buf + 3 * 8, 4 * 8, g13, g15); |
| 276 LD_SH2(int_buf + 11 * 8, 4 * 8, g5, g7); |
| 277 LD_SW2(const0 + 4 * 19, 4, k0, k1); |
| 278 k2 = LD_SW(const0 + 4 * 21); |
| 279 VP9_MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7); |
| 280 |
| 281 tp0 = LD_SH(int_buf + 4 * 8); |
| 282 tp1 = LD_SH(int_buf + 5 * 8); |
| 283 tp3 = LD_SH(int_buf + 10 * 8); |
| 284 tp2 = LD_SH(int_buf + 14 * 8); |
| 285 LD_SW2(const0 + 4 * 22, 4, k0, k1); |
| 286 k2 = LD_SW(const0 + 4 * 24); |
| 287 VP9_MADD_BF(tp0, tp1, tp2, tp3, k0, k1, k2, k0, out4, out6, out5, out7); |
| 288 out4 = -out4; |
| 289 ST_SH(out4, (out + 3 * 16)); |
| 290 ST_SH(out5, (out_ptr + 4 * 16)); |
| 291 |
| 292 h1 = LD_SH(int_buf + 9 * 8); |
| 293 h3 = LD_SH(int_buf + 12 * 8); |
| 294 VP9_MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15); |
| 295 out13 = -out13; |
| 296 ST_SH(out12, (out + 2 * 16)); |
| 297 ST_SH(out13, (out_ptr + 5 * 16)); |
| 298 |
| 299 tp0 = LD_SH(int_buf); |
| 300 tp1 = LD_SH(int_buf + 8); |
| 301 tp2 = LD_SH(int_buf + 2 * 8); |
| 302 tp3 = LD_SH(int_buf + 6 * 8); |
| 303 |
| 304 BUTTERFLY_4(tp0, tp1, tp3, tp2, out0, out1, h11, h10); |
| 305 out1 = -out1; |
| 306 ST_SH(out0, (out)); |
| 307 ST_SH(out1, (out_ptr + 7 * 16)); |
| 308 |
| 309 h0 = LD_SH(int_buf + 8 * 8); |
| 310 h2 = LD_SH(int_buf + 13 * 8); |
| 311 |
| 312 BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10); |
| 313 out8 = -out8; |
| 314 ST_SH(out8, (out + 16)); |
| 315 ST_SH(out9, (out_ptr + 6 * 16)); |
| 316 |
| 317 /* stage 4 */ |
| 318 LD_SW2(const0 + 4 * 25, 4, k0, k1); |
| 319 LD_SW2(const0 + 4 * 27, 4, k2, k3); |
| 320 VP9_MADD_SHORT(h10, h11, k1, k2, out2, out3); |
| 321 ST_SH(out2, (out + 7 * 16)); |
| 322 ST_SH(out3, (out_ptr)); |
| 323 |
| 324 VP9_MADD_SHORT(out6, out7, k0, k3, out6, out7); |
| 325 ST_SH(out6, (out + 4 * 16)); |
| 326 ST_SH(out7, (out_ptr + 3 * 16)); |
| 327 |
| 328 VP9_MADD_SHORT(out10, out11, k0, k3, out10, out11); |
| 329 ST_SH(out10, (out + 6 * 16)); |
| 330 ST_SH(out11, (out_ptr + 16)); |
| 331 |
| 332 VP9_MADD_SHORT(out14, out15, k1, k2, out14, out15); |
| 333 ST_SH(out14, (out + 5 * 16)); |
| 334 ST_SH(out15, (out_ptr + 2 * 16)); |
| 335 } |
| 336 |
| 337 static void fadst16_transpose_postproc_msa(int16_t *input, int16_t *out) { |
| 338 v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15; |
| 339 v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15; |
| 340 |
| 341 /* load input data */ |
| 342 LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7); |
| 343 TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, |
| 344 r0, r1, r2, r3, r4, r5, r6, r7); |
| 345 VP9_FDCT_POSTPROC_2V_NEG_H(r0, r1); |
| 346 VP9_FDCT_POSTPROC_2V_NEG_H(r2, r3); |
| 347 VP9_FDCT_POSTPROC_2V_NEG_H(r4, r5); |
| 348 VP9_FDCT_POSTPROC_2V_NEG_H(r6, r7); |
| 349 ST_SH8(r0, r1, r2, r3, r4, r5, r6, r7, out, 8); |
| 350 out += 64; |
| 351 |
| 352 LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15); |
| 353 TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, |
| 354 r8, r9, r10, r11, r12, r13, r14, r15); |
| 355 VP9_FDCT_POSTPROC_2V_NEG_H(r8, r9); |
| 356 VP9_FDCT_POSTPROC_2V_NEG_H(r10, r11); |
| 357 VP9_FDCT_POSTPROC_2V_NEG_H(r12, r13); |
| 358 VP9_FDCT_POSTPROC_2V_NEG_H(r14, r15); |
| 359 ST_SH8(r8, r9, r10, r11, r12, r13, r14, r15, out, 8); |
| 360 out += 64; |
| 361 |
| 362 /* load input data */ |
| 363 input += 128; |
| 364 LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7); |
| 365 TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, |
| 366 r0, r1, r2, r3, r4, r5, r6, r7); |
| 367 VP9_FDCT_POSTPROC_2V_NEG_H(r0, r1); |
| 368 VP9_FDCT_POSTPROC_2V_NEG_H(r2, r3); |
| 369 VP9_FDCT_POSTPROC_2V_NEG_H(r4, r5); |
| 370 VP9_FDCT_POSTPROC_2V_NEG_H(r6, r7); |
| 371 ST_SH8(r0, r1, r2, r3, r4, r5, r6, r7, out, 8); |
| 372 out += 64; |
| 373 |
| 374 LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15); |
| 375 TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, |
| 376 r8, r9, r10, r11, r12, r13, r14, r15); |
| 377 VP9_FDCT_POSTPROC_2V_NEG_H(r8, r9); |
| 378 VP9_FDCT_POSTPROC_2V_NEG_H(r10, r11); |
| 379 VP9_FDCT_POSTPROC_2V_NEG_H(r12, r13); |
| 380 VP9_FDCT_POSTPROC_2V_NEG_H(r14, r15); |
| 381 ST_SH8(r8, r9, r10, r11, r12, r13, r14, r15, out, 8); |
| 382 } |
| 383 |
| 384 static void fadst16_rows_step1_msa(int16_t *input, const int32_t *const0, |
| 385 int16_t *int_buf) { |
| 386 v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15; |
| 387 v8i16 tp0, tp1, tp2, tp3, g0, g1, g2, g3, g8, g9, g10, g11, h0, h1, h2, h3; |
| 388 v4i32 k0, k1, k2, k3; |
| 389 |
| 390 /* load input data */ |
| 391 r0 = LD_SH(input); |
| 392 r7 = LD_SH(input + 7 * 8); |
| 393 r8 = LD_SH(input + 8 * 8); |
| 394 r15 = LD_SH(input + 15 * 8); |
| 395 |
| 396 /* stage 1 */ |
| 397 LD_SW2(const0, 4, k0, k1); |
| 398 LD_SW2(const0 + 4 * 2, 4, k2, k3); |
| 399 VP9_MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3); |
| 400 |
| 401 r3 = LD_SH(input + 3 * 8); |
| 402 r4 = LD_SH(input + 4 * 8); |
| 403 r11 = LD_SH(input + 11 * 8); |
| 404 r12 = LD_SH(input + 12 * 8); |
| 405 |
| 406 LD_SW2(const0 + 4 * 4, 4, k0, k1); |
| 407 LD_SW2(const0 + 4 * 6, 4, k2, k3); |
| 408 VP9_MADD_BF(r11, r4, r3, r12, k0, k1, k2, k3, g8, g9, g10, g11); |
| 409 |
| 410 /* stage 2 */ |
| 411 BUTTERFLY_4(g0, g2, g10, g8, tp0, tp2, tp3, tp1); |
| 412 ST_SH2(tp0, tp1, int_buf, 4 * 8); |
| 413 ST_SH2(tp2, tp3, int_buf + 8, 4 * 8); |
| 414 |
| 415 LD_SW2(const0 + 4 * 8, 4, k0, k1); |
| 416 k2 = LD_SW(const0 + 4 * 10); |
| 417 VP9_MADD_BF(g1, g3, g9, g11, k0, k1, k2, k0, h0, h1, h2, h3); |
| 418 ST_SH2(h0, h3, int_buf + 8 * 8, 4 * 8); |
| 419 ST_SH2(h1, h2, int_buf + 9 * 8, 4 * 8); |
| 420 |
| 421 r1 = LD_SH(input + 8); |
| 422 r6 = LD_SH(input + 6 * 8); |
| 423 r9 = LD_SH(input + 9 * 8); |
| 424 r14 = LD_SH(input + 14 * 8); |
| 425 |
| 426 LD_SW2(const0 + 4 * 11, 4, k0, k1); |
| 427 LD_SW2(const0 + 4 * 13, 4, k2, k3); |
| 428 VP9_MADD_BF(r9, r6, r1, r14, k0, k1, k2, k3, g0, g1, g2, g3); |
| 429 ST_SH2(g1, g3, int_buf + 3 * 8, 4 * 8); |
| 430 |
| 431 r2 = LD_SH(input + 2 * 8); |
| 432 r5 = LD_SH(input + 5 * 8); |
| 433 r10 = LD_SH(input + 10 * 8); |
| 434 r13 = LD_SH(input + 13 * 8); |
| 435 |
| 436 LD_SW2(const0 + 4 * 15, 4, k0, k1); |
| 437 LD_SW2(const0 + 4 * 17, 4, k2, k3); |
| 438 VP9_MADD_BF(r13, r2, r5, r10, k0, k1, k2, k3, h0, h1, h2, h3); |
| 439 ST_SH2(h1, h3, int_buf + 11 * 8, 4 * 8); |
| 440 BUTTERFLY_4(h0, h2, g2, g0, tp0, tp1, tp2, tp3); |
| 441 ST_SH4(tp0, tp1, tp2, tp3, int_buf + 2 * 8, 4 * 8); |
| 442 } |
| 443 |
| 444 static void fadst16_rows_step2_msa(int16_t *int_buf, const int32_t *const0, |
| 445 int16_t *out) { |
| 446 int16_t *out_ptr = out + 8; |
| 447 v8i16 tp0, tp1, tp2, tp3, g5, g7, g13, g15; |
| 448 v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h10, h11; |
| 449 v8i16 out0, out1, out2, out3, out4, out5, out6, out7; |
| 450 v8i16 out8, out9, out10, out11, out12, out13, out14, out15; |
| 451 v4i32 k0, k1, k2, k3; |
| 452 |
| 453 g13 = LD_SH(int_buf + 3 * 8); |
| 454 g15 = LD_SH(int_buf + 7 * 8); |
| 455 g5 = LD_SH(int_buf + 11 * 8); |
| 456 g7 = LD_SH(int_buf + 15 * 8); |
| 457 |
| 458 LD_SW2(const0 + 4 * 19, 4, k0, k1); |
| 459 k2 = LD_SW(const0 + 4 * 21); |
| 460 VP9_MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7); |
| 461 |
| 462 tp0 = LD_SH(int_buf + 4 * 8); |
| 463 tp1 = LD_SH(int_buf + 5 * 8); |
| 464 tp3 = LD_SH(int_buf + 10 * 8); |
| 465 tp2 = LD_SH(int_buf + 14 * 8); |
| 466 |
| 467 LD_SW2(const0 + 4 * 22, 4, k0, k1); |
| 468 k2 = LD_SW(const0 + 4 * 24); |
| 469 VP9_MADD_BF(tp0, tp1, tp2, tp3, k0, k1, k2, k0, out4, out6, out5, out7); |
| 470 out4 = -out4; |
| 471 ST_SH(out4, (out + 3 * 16)); |
| 472 ST_SH(out5, (out_ptr + 4 * 16)); |
| 473 |
| 474 h1 = LD_SH(int_buf + 9 * 8); |
| 475 h3 = LD_SH(int_buf + 12 * 8); |
| 476 VP9_MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15); |
| 477 out13 = -out13; |
| 478 ST_SH(out12, (out + 2 * 16)); |
| 479 ST_SH(out13, (out_ptr + 5 * 16)); |
| 480 |
| 481 tp0 = LD_SH(int_buf); |
| 482 tp1 = LD_SH(int_buf + 8); |
| 483 tp2 = LD_SH(int_buf + 2 * 8); |
| 484 tp3 = LD_SH(int_buf + 6 * 8); |
| 485 |
| 486 BUTTERFLY_4(tp0, tp1, tp3, tp2, out0, out1, h11, h10); |
| 487 out1 = -out1; |
| 488 ST_SH(out0, (out)); |
| 489 ST_SH(out1, (out_ptr + 7 * 16)); |
| 490 |
| 491 h0 = LD_SH(int_buf + 8 * 8); |
| 492 h2 = LD_SH(int_buf + 13 * 8); |
| 493 BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10); |
| 494 out8 = -out8; |
| 495 ST_SH(out8, (out + 16)); |
| 496 ST_SH(out9, (out_ptr + 6 * 16)); |
| 497 |
| 498 /* stage 4 */ |
| 499 LD_SW2(const0 + 4 * 25, 4, k0, k1); |
| 500 LD_SW2(const0 + 4 * 27, 4, k2, k3); |
| 501 VP9_MADD_SHORT(h10, h11, k1, k2, out2, out3); |
| 502 ST_SH(out2, (out + 7 * 16)); |
| 503 ST_SH(out3, (out_ptr)); |
| 504 |
| 505 VP9_MADD_SHORT(out6, out7, k0, k3, out6, out7); |
| 506 ST_SH(out6, (out + 4 * 16)); |
| 507 ST_SH(out7, (out_ptr + 3 * 16)); |
| 508 |
| 509 VP9_MADD_SHORT(out10, out11, k0, k3, out10, out11); |
| 510 ST_SH(out10, (out + 6 * 16)); |
| 511 ST_SH(out11, (out_ptr + 16)); |
| 512 |
| 513 VP9_MADD_SHORT(out14, out15, k1, k2, out14, out15); |
| 514 ST_SH(out14, (out + 5 * 16)); |
| 515 ST_SH(out15, (out_ptr + 2 * 16)); |
| 516 } |
| 517 |
| 518 static void fadst16_transpose_msa(int16_t *input, int16_t *out) { |
| 519 v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15; |
| 520 v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15; |
| 521 |
| 522 /* load input data */ |
| 523 LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11, |
| 524 l4, l12, l5, l13, l6, l14, l7, l15); |
| 525 TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, |
| 526 r0, r1, r2, r3, r4, r5, r6, r7); |
| 527 TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, |
| 528 r8, r9, r10, r11, r12, r13, r14, r15); |
| 529 ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8); |
| 530 ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8); |
| 531 out += 16 * 8; |
| 532 |
| 533 /* load input data */ |
| 534 input += 128; |
| 535 LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11, |
| 536 l4, l12, l5, l13, l6, l14, l7, l15); |
| 537 TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, |
| 538 r0, r1, r2, r3, r4, r5, r6, r7); |
| 539 TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, |
| 540 r8, r9, r10, r11, r12, r13, r14, r15); |
| 541 ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8); |
| 542 ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8); |
| 543 } |
| 544 |
| 545 static void postproc_fdct16x8_1d_row(int16_t *intermediate, int16_t *output) { |
| 546 int16_t *temp = intermediate; |
| 547 int16_t *out = output; |
| 548 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; |
| 549 v8i16 in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11; |
| 550 v8i16 in12, in13, in14, in15; |
| 551 |
| 552 LD_SH8(temp, 16, in0, in1, in2, in3, in4, in5, in6, in7); |
| 553 temp = intermediate + 8; |
| 554 LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15); |
| 555 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, |
| 556 in0, in1, in2, in3, in4, in5, in6, in7); |
| 557 TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15, |
| 558 in8, in9, in10, in11, in12, in13, in14, in15); |
| 559 VP9_FDCT_POSTPROC_2V_NEG_H(in0, in1); |
| 560 VP9_FDCT_POSTPROC_2V_NEG_H(in2, in3); |
| 561 VP9_FDCT_POSTPROC_2V_NEG_H(in4, in5); |
| 562 VP9_FDCT_POSTPROC_2V_NEG_H(in6, in7); |
| 563 VP9_FDCT_POSTPROC_2V_NEG_H(in8, in9); |
| 564 VP9_FDCT_POSTPROC_2V_NEG_H(in10, in11); |
| 565 VP9_FDCT_POSTPROC_2V_NEG_H(in12, in13); |
| 566 VP9_FDCT_POSTPROC_2V_NEG_H(in14, in15); |
| 567 BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, |
| 568 in8, in9, in10, in11, in12, in13, in14, in15, |
| 569 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, |
| 570 in8, in9, in10, in11, in12, in13, in14, in15); |
| 571 temp = intermediate; |
| 572 ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, temp, 16); |
| 573 VP9_FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, |
| 574 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7); |
| 575 temp = intermediate; |
| 576 LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15); |
| 577 VP9_FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15, |
| 578 in0, in1, in2, in3, in4, in5, in6, in7); |
| 579 TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, |
| 580 tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3); |
| 581 ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, out, 16); |
| 582 TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, |
| 583 tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7); |
| 584 out = output + 8; |
| 585 ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, out, 16); |
| 586 } |
| 587 |
| 588 void vp9_fht16x16_msa(const int16_t *input, int16_t *output, |
| 589 int32_t stride, int32_t tx_type) { |
| 590 DECLARE_ALIGNED(32, int16_t, tmp[256]); |
| 591 DECLARE_ALIGNED(32, int16_t, trans_buf[256]); |
| 592 DECLARE_ALIGNED(32, int16_t, tmp_buf[128]); |
| 593 int32_t i; |
| 594 int16_t *ptmpbuf = &tmp_buf[0]; |
| 595 int16_t *trans = &trans_buf[0]; |
| 596 const int32_t const_arr[29 * 4] = { |
| 597 52707308, 52707308, 52707308, 52707308, |
| 598 -1072430300, -1072430300, -1072430300, -1072430300, |
| 599 795618043, 795618043, 795618043, 795618043, |
| 600 -721080468, -721080468, -721080468, -721080468, |
| 601 459094491, 459094491, 459094491, 459094491, |
| 602 -970646691, -970646691, -970646691, -970646691, |
| 603 1010963856, 1010963856, 1010963856, 1010963856, |
| 604 -361743294, -361743294, -361743294, -361743294, |
| 605 209469125, 209469125, 209469125, 209469125, |
| 606 -1053094788, -1053094788, -1053094788, -1053094788, |
| 607 1053160324, 1053160324, 1053160324, 1053160324, |
| 608 639644520, 639644520, 639644520, 639644520, |
| 609 -862444000, -862444000, -862444000, -862444000, |
| 610 1062144356, 1062144356, 1062144356, 1062144356, |
| 611 -157532337, -157532337, -157532337, -157532337, |
| 612 260914709, 260914709, 260914709, 260914709, |
| 613 -1041559667, -1041559667, -1041559667, -1041559667, |
| 614 920985831, 920985831, 920985831, 920985831, |
| 615 -551995675, -551995675, -551995675, -551995675, |
| 616 596522295, 596522295, 596522295, 596522295, |
| 617 892853362, 892853362, 892853362, 892853362, |
| 618 -892787826, -892787826, -892787826, -892787826, |
| 619 410925857, 410925857, 410925857, 410925857, |
| 620 -992012162, -992012162, -992012162, -992012162, |
| 621 992077698, 992077698, 992077698, 992077698, |
| 622 759246145, 759246145, 759246145, 759246145, |
| 623 -759180609, -759180609, -759180609, -759180609, |
| 624 -759222975, -759222975, -759222975, -759222975, |
| 625 759288511, 759288511, 759288511, 759288511 }; |
| 626 |
| 627 switch (tx_type) { |
| 628 case DCT_DCT: |
| 629 /* column transform */ |
| 630 for (i = 0; i < 2; ++i) { |
| 631 fdct8x16_1d_column(input + 8 * i, tmp + 8 * i, stride); |
| 632 } |
| 633 |
| 634 /* row transform */ |
| 635 for (i = 0; i < 2; ++i) { |
| 636 fdct16x8_1d_row(tmp + (128 * i), output + (128 * i)); |
| 637 } |
| 638 break; |
| 639 case ADST_DCT: |
| 640 /* column transform */ |
| 641 for (i = 0; i < 2; ++i) { |
| 642 fadst16_cols_step1_msa(input + (i << 3), stride, const_arr, ptmpbuf); |
| 643 fadst16_cols_step2_msa(ptmpbuf, const_arr, tmp + (i << 3)); |
| 644 } |
| 645 |
| 646 /* row transform */ |
| 647 for (i = 0; i < 2; ++i) { |
| 648 postproc_fdct16x8_1d_row(tmp + (128 * i), output + (128 * i)); |
| 649 } |
| 650 break; |
| 651 case DCT_ADST: |
| 652 /* column transform */ |
| 653 for (i = 0; i < 2; ++i) { |
| 654 fdct8x16_1d_column(input + 8 * i, tmp + 8 * i, stride); |
| 655 } |
| 656 |
| 657 fadst16_transpose_postproc_msa(tmp, trans); |
| 658 |
| 659 /* row transform */ |
| 660 for (i = 0; i < 2; ++i) { |
| 661 fadst16_rows_step1_msa(trans + (i << 7), const_arr, ptmpbuf); |
| 662 fadst16_rows_step2_msa(ptmpbuf, const_arr, tmp + (i << 7)); |
| 663 } |
| 664 |
| 665 fadst16_transpose_msa(tmp, output); |
| 666 break; |
| 667 case ADST_ADST: |
| 668 /* column transform */ |
| 669 for (i = 0; i < 2; ++i) { |
| 670 fadst16_cols_step1_msa(input + (i << 3), stride, const_arr, ptmpbuf); |
| 671 fadst16_cols_step2_msa(ptmpbuf, const_arr, tmp + (i << 3)); |
| 672 } |
| 673 |
| 674 fadst16_transpose_postproc_msa(tmp, trans); |
| 675 |
| 676 /* row transform */ |
| 677 for (i = 0; i < 2; ++i) { |
| 678 fadst16_rows_step1_msa(trans + (i << 7), const_arr, ptmpbuf); |
| 679 fadst16_rows_step2_msa(ptmpbuf, const_arr, tmp + (i << 7)); |
| 680 } |
| 681 |
| 682 fadst16_transpose_msa(tmp, output); |
| 683 break; |
| 684 default: |
| 685 assert(0); |
| 686 break; |
| 687 } |
| 688 } |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 1217053002:
libvpx: Pull from upstream (Closed)
Base URL: https://chromium.googlesource.com/chromium/deps/libvpx.git@master