1 /* 2 * Copyright (c) 2012 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 "./vp9_rtcd.h" 12 #include "vpx_dsp/x86/inv_txfm_sse2.h" 13 #include "vpx_dsp/x86/txfm_common_sse2.h" 14 #include "vpx_ports/mem.h" 15 16 void vp9_iht4x4_16_add_sse2(const tran_low_t *input, uint8_t *dest, int stride, 17 int tx_type) { 18 __m128i in[2]; 19 const __m128i eight = _mm_set1_epi16(8); 20 21 in[0] = load_input_data8(input); 22 in[1] = load_input_data8(input + 8); 23 24 switch (tx_type) { 25 case 0: // DCT_DCT 26 idct4_sse2(in); 27 idct4_sse2(in); 28 break; 29 case 1: // ADST_DCT 30 idct4_sse2(in); 31 iadst4_sse2(in); 32 break; 33 case 2: // DCT_ADST 34 iadst4_sse2(in); 35 idct4_sse2(in); 36 break; 37 case 3: // ADST_ADST 38 iadst4_sse2(in); 39 iadst4_sse2(in); 40 break; 41 default: assert(0); break; 42 } 43 44 // Final round and shift 45 in[0] = _mm_add_epi16(in[0], eight); 46 in[1] = _mm_add_epi16(in[1], eight); 47 48 in[0] = _mm_srai_epi16(in[0], 4); 49 in[1] = _mm_srai_epi16(in[1], 4); 50 51 recon_and_store4x4_sse2(in, dest, stride); 52 } 53 54 void vp9_iht8x8_64_add_sse2(const tran_low_t *input, uint8_t *dest, int stride, 55 int tx_type) { 56 __m128i in[8]; 57 const __m128i final_rounding = _mm_set1_epi16(1 << 4); 58 59 // load input data 60 in[0] = load_input_data8(input); 61 in[1] = load_input_data8(input + 8 * 1); 62 in[2] = load_input_data8(input + 8 * 2); 63 in[3] = load_input_data8(input + 8 * 3); 64 in[4] = load_input_data8(input + 8 * 4); 65 in[5] = load_input_data8(input + 8 * 5); 66 in[6] = load_input_data8(input + 8 * 6); 67 in[7] = load_input_data8(input + 8 * 7); 68 69 switch (tx_type) { 70 case 0: // DCT_DCT 71 idct8_sse2(in); 72 idct8_sse2(in); 73 break; 74 case 1: // ADST_DCT 75 idct8_sse2(in); 76 iadst8_sse2(in); 77 break; 78 case 2: // DCT_ADST 79 iadst8_sse2(in); 80 idct8_sse2(in); 81 break; 82 case 3: // ADST_ADST 83 iadst8_sse2(in); 84 iadst8_sse2(in); 85 break; 86 default: assert(0); break; 87 } 88 89 // Final rounding and shift 90 in[0] = _mm_adds_epi16(in[0], final_rounding); 91 in[1] = _mm_adds_epi16(in[1], final_rounding); 92 in[2] = _mm_adds_epi16(in[2], final_rounding); 93 in[3] = _mm_adds_epi16(in[3], final_rounding); 94 in[4] = _mm_adds_epi16(in[4], final_rounding); 95 in[5] = _mm_adds_epi16(in[5], final_rounding); 96 in[6] = _mm_adds_epi16(in[6], final_rounding); 97 in[7] = _mm_adds_epi16(in[7], final_rounding); 98 99 in[0] = _mm_srai_epi16(in[0], 5); 100 in[1] = _mm_srai_epi16(in[1], 5); 101 in[2] = _mm_srai_epi16(in[2], 5); 102 in[3] = _mm_srai_epi16(in[3], 5); 103 in[4] = _mm_srai_epi16(in[4], 5); 104 in[5] = _mm_srai_epi16(in[5], 5); 105 in[6] = _mm_srai_epi16(in[6], 5); 106 in[7] = _mm_srai_epi16(in[7], 5); 107 108 recon_and_store(dest + 0 * stride, in[0]); 109 recon_and_store(dest + 1 * stride, in[1]); 110 recon_and_store(dest + 2 * stride, in[2]); 111 recon_and_store(dest + 3 * stride, in[3]); 112 recon_and_store(dest + 4 * stride, in[4]); 113 recon_and_store(dest + 5 * stride, in[5]); 114 recon_and_store(dest + 6 * stride, in[6]); 115 recon_and_store(dest + 7 * stride, in[7]); 116 } 117 118 static INLINE void load_buffer_8x16(const tran_low_t *const input, 119 __m128i *const in) { 120 in[0] = load_input_data8(input + 0 * 16); 121 in[1] = load_input_data8(input + 1 * 16); 122 in[2] = load_input_data8(input + 2 * 16); 123 in[3] = load_input_data8(input + 3 * 16); 124 in[4] = load_input_data8(input + 4 * 16); 125 in[5] = load_input_data8(input + 5 * 16); 126 in[6] = load_input_data8(input + 6 * 16); 127 in[7] = load_input_data8(input + 7 * 16); 128 129 in[8] = load_input_data8(input + 8 * 16); 130 in[9] = load_input_data8(input + 9 * 16); 131 in[10] = load_input_data8(input + 10 * 16); 132 in[11] = load_input_data8(input + 11 * 16); 133 in[12] = load_input_data8(input + 12 * 16); 134 in[13] = load_input_data8(input + 13 * 16); 135 in[14] = load_input_data8(input + 14 * 16); 136 in[15] = load_input_data8(input + 15 * 16); 137 } 138 139 static INLINE void write_buffer_8x16(uint8_t *const dest, __m128i *const in, 140 const int stride) { 141 const __m128i final_rounding = _mm_set1_epi16(1 << 5); 142 // Final rounding and shift 143 in[0] = _mm_adds_epi16(in[0], final_rounding); 144 in[1] = _mm_adds_epi16(in[1], final_rounding); 145 in[2] = _mm_adds_epi16(in[2], final_rounding); 146 in[3] = _mm_adds_epi16(in[3], final_rounding); 147 in[4] = _mm_adds_epi16(in[4], final_rounding); 148 in[5] = _mm_adds_epi16(in[5], final_rounding); 149 in[6] = _mm_adds_epi16(in[6], final_rounding); 150 in[7] = _mm_adds_epi16(in[7], final_rounding); 151 in[8] = _mm_adds_epi16(in[8], final_rounding); 152 in[9] = _mm_adds_epi16(in[9], final_rounding); 153 in[10] = _mm_adds_epi16(in[10], final_rounding); 154 in[11] = _mm_adds_epi16(in[11], final_rounding); 155 in[12] = _mm_adds_epi16(in[12], final_rounding); 156 in[13] = _mm_adds_epi16(in[13], final_rounding); 157 in[14] = _mm_adds_epi16(in[14], final_rounding); 158 in[15] = _mm_adds_epi16(in[15], final_rounding); 159 160 in[0] = _mm_srai_epi16(in[0], 6); 161 in[1] = _mm_srai_epi16(in[1], 6); 162 in[2] = _mm_srai_epi16(in[2], 6); 163 in[3] = _mm_srai_epi16(in[3], 6); 164 in[4] = _mm_srai_epi16(in[4], 6); 165 in[5] = _mm_srai_epi16(in[5], 6); 166 in[6] = _mm_srai_epi16(in[6], 6); 167 in[7] = _mm_srai_epi16(in[7], 6); 168 in[8] = _mm_srai_epi16(in[8], 6); 169 in[9] = _mm_srai_epi16(in[9], 6); 170 in[10] = _mm_srai_epi16(in[10], 6); 171 in[11] = _mm_srai_epi16(in[11], 6); 172 in[12] = _mm_srai_epi16(in[12], 6); 173 in[13] = _mm_srai_epi16(in[13], 6); 174 in[14] = _mm_srai_epi16(in[14], 6); 175 in[15] = _mm_srai_epi16(in[15], 6); 176 177 recon_and_store(dest + 0 * stride, in[0]); 178 recon_and_store(dest + 1 * stride, in[1]); 179 recon_and_store(dest + 2 * stride, in[2]); 180 recon_and_store(dest + 3 * stride, in[3]); 181 recon_and_store(dest + 4 * stride, in[4]); 182 recon_and_store(dest + 5 * stride, in[5]); 183 recon_and_store(dest + 6 * stride, in[6]); 184 recon_and_store(dest + 7 * stride, in[7]); 185 recon_and_store(dest + 8 * stride, in[8]); 186 recon_and_store(dest + 9 * stride, in[9]); 187 recon_and_store(dest + 10 * stride, in[10]); 188 recon_and_store(dest + 11 * stride, in[11]); 189 recon_and_store(dest + 12 * stride, in[12]); 190 recon_and_store(dest + 13 * stride, in[13]); 191 recon_and_store(dest + 14 * stride, in[14]); 192 recon_and_store(dest + 15 * stride, in[15]); 193 } 194 195 void vp9_iht16x16_256_add_sse2(const tran_low_t *input, uint8_t *dest, 196 int stride, int tx_type) { 197 __m128i in0[16], in1[16]; 198 199 load_buffer_8x16(input, in0); 200 input += 8; 201 load_buffer_8x16(input, in1); 202 203 switch (tx_type) { 204 case 0: // DCT_DCT 205 idct16_sse2(in0, in1); 206 idct16_sse2(in0, in1); 207 break; 208 case 1: // ADST_DCT 209 idct16_sse2(in0, in1); 210 iadst16_sse2(in0, in1); 211 break; 212 case 2: // DCT_ADST 213 iadst16_sse2(in0, in1); 214 idct16_sse2(in0, in1); 215 break; 216 case 3: // ADST_ADST 217 iadst16_sse2(in0, in1); 218 iadst16_sse2(in0, in1); 219 break; 220 default: assert(0); break; 221 } 222 223 write_buffer_8x16(dest, in0, stride); 224 dest += 8; 225 write_buffer_8x16(dest, in1, stride); 226 } 227
