1 /* 2 * Copyright (C) 2017 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17 /** 18 * Tests for SAD (sum of absolute differences). 19 */ 20 public class SimdSadShort { 21 $inline$seven()22 private static int $inline$seven() { 23 return 7; 24 } 25 26 // TODO: lower precision still coming, b/64091002 27 sadShort2Short(short[] s1, short[] s2)28 private static short sadShort2Short(short[] s1, short[] s2) { 29 int min_length = Math.min(s1.length, s2.length); 30 short sad = 0; 31 for (int i = 0; i < min_length; i++) { 32 sad += Math.abs(s1[i] - s2[i]); 33 } 34 return sad; 35 } 36 sadShort2ShortAlt(short[] s1, short[] s2)37 private static short sadShort2ShortAlt(short[] s1, short[] s2) { 38 int min_length = Math.min(s1.length, s2.length); 39 short sad = 0; 40 for (int i = 0; i < min_length; i++) { 41 short s = s1[i]; 42 short p = s2[i]; 43 sad += s >= p ? s - p : p - s; 44 } 45 return sad; 46 } 47 sadShort2ShortAlt2(short[] s1, short[] s2)48 private static short sadShort2ShortAlt2(short[] s1, short[] s2) { 49 int min_length = Math.min(s1.length, s2.length); 50 short sad = 0; 51 for (int i = 0; i < min_length; i++) { 52 short s = s1[i]; 53 short p = s2[i]; 54 int x = s - p; 55 if (x < 0) x = -x; 56 sad += x; 57 } 58 return sad; 59 } 60 61 /// CHECK-START: int SimdSadShort.sadShort2Int(short[], short[]) loop_optimization (before) 62 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 63 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 64 /// CHECK-DAG: <<Phi2:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 65 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop>> outer_loop:none 66 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 67 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 68 /// CHECK-DAG: <<Sub:i\d+>> Sub [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 69 /// CHECK-DAG: <<Intrin:i\d+>> Abs [<<Sub>>] loop:<<Loop>> outer_loop:none 70 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 71 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 72 // 73 /// CHECK-START-ARM64: int SimdSadShort.sadShort2Int(short[], short[]) loop_optimization (after) 74 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 75 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 76 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 77 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 78 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 79 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 80 /// CHECK-DAG: <<Load2:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 81 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Load2>>] loop:<<Loop>> outer_loop:none 82 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none sadShort2Int(short[] s1, short[] s2)83 private static int sadShort2Int(short[] s1, short[] s2) { 84 int min_length = Math.min(s1.length, s2.length); 85 int sad = 0; 86 for (int i = 0; i < min_length; i++) { 87 sad += Math.abs(s1[i] - s2[i]); 88 } 89 return sad; 90 } 91 92 /// CHECK-START: int SimdSadShort.sadShort2IntAlt(short[], short[]) loop_optimization (before) 93 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 94 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 95 /// CHECK-DAG: <<Phi2:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 96 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop>> outer_loop:none 97 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 98 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 99 /// CHECK-DAG: <<Sub:i\d+>> Sub [<<Get2>>,<<Get1>>] loop:<<Loop>> outer_loop:none 100 /// CHECK-DAG: <<Intrin:i\d+>> Abs [<<Sub>>] loop:<<Loop>> outer_loop:none 101 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 102 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 103 // 104 /// CHECK-START-ARM64: int SimdSadShort.sadShort2IntAlt(short[], short[]) loop_optimization (after) 105 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 106 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 107 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 108 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 109 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 110 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 111 /// CHECK-DAG: <<Load2:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 112 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load2>>,<<Load1>>] loop:<<Loop>> outer_loop:none 113 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none sadShort2IntAlt(short[] s1, short[] s2)114 private static int sadShort2IntAlt(short[] s1, short[] s2) { 115 int min_length = Math.min(s1.length, s2.length); 116 int sad = 0; 117 for (int i = 0; i < min_length; i++) { 118 short s = s1[i]; 119 short p = s2[i]; 120 sad += s >= p ? s - p : p - s; 121 } 122 return sad; 123 } 124 125 /// CHECK-START: int SimdSadShort.sadShort2IntAlt2(short[], short[]) loop_optimization (before) 126 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 127 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 128 /// CHECK-DAG: <<Phi2:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 129 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop>> outer_loop:none 130 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 131 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 132 /// CHECK-DAG: <<Sub:i\d+>> Sub [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 133 /// CHECK-DAG: <<Intrin:i\d+>> Abs [<<Sub>>] loop:<<Loop>> outer_loop:none 134 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 135 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 136 // 137 /// CHECK-START-ARM64: int SimdSadShort.sadShort2IntAlt2(short[], short[]) loop_optimization (after) 138 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 139 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 140 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 141 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 142 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 143 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 144 /// CHECK-DAG: <<Load2:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 145 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Load2>>] loop:<<Loop>> outer_loop:none 146 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none sadShort2IntAlt2(short[] s1, short[] s2)147 private static int sadShort2IntAlt2(short[] s1, short[] s2) { 148 int min_length = Math.min(s1.length, s2.length); 149 int sad = 0; 150 for (int i = 0; i < min_length; i++) { 151 short s = s1[i]; 152 short p = s2[i]; 153 int x = s - p; 154 if (x < 0) x = -x; 155 sad += x; 156 } 157 return sad; 158 } 159 160 /// CHECK-START: int SimdSadShort.sadShort2IntConstant1(short[]) loop_optimization (before) 161 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 162 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 163 /// CHECK-DAG: <<Cons:i\d+>> IntConstant -7 loop:none 164 /// CHECK-DAG: <<Phi2:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 165 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop>> outer_loop:none 166 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 167 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get1>>,<<Cons>>] loop:<<Loop>> outer_loop:none 168 /// CHECK-DAG: <<Intrin:i\d+>> Abs [<<Add>>] loop:<<Loop>> outer_loop:none 169 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 170 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 171 // 172 /// CHECK-START-ARM64: int SimdSadShort.sadShort2IntConstant1(short[]) loop_optimization (after) 173 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 174 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 175 /// CHECK-DAG: <<Cons:i\d+>> IntConstant 7 loop:none 176 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 177 /// CHECK-DAG: <<Rep:d\d+>> VecReplicateScalar [<<Cons>>] loop:none 178 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 179 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 180 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 181 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 182 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Rep>>] loop:<<Loop>> outer_loop:none 183 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none sadShort2IntConstant1(short[] s)184 private static int sadShort2IntConstant1(short[] s) { 185 int sad = 0; 186 for (int i = 0; i < s.length; i++) { 187 sad += Math.abs(s[i] - 7); // s[i] + -7 188 } 189 return sad; 190 } 191 192 /// CHECK-START: int SimdSadShort.sadShort2IntConstant2(short[]) loop_optimization (before) 193 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 194 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 195 /// CHECK-DAG: <<Cons:i\d+>> IntConstant 7 loop:none 196 /// CHECK-DAG: <<Phi2:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 197 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop>> outer_loop:none 198 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 199 /// CHECK-DAG: <<Sub:i\d+>> Sub [<<Get1>>,<<Cons>>] loop:<<Loop>> outer_loop:none 200 /// CHECK-DAG: <<Intrin:i\d+>> Abs [<<Sub>>] loop:<<Loop>> outer_loop:none 201 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 202 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 203 // 204 /// CHECK-START-ARM64: int SimdSadShort.sadShort2IntConstant2(short[]) loop_optimization (after) 205 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 206 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 207 /// CHECK-DAG: <<Cons:i\d+>> IntConstant 7 loop:none 208 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 209 /// CHECK-DAG: <<Rep:d\d+>> VecReplicateScalar [<<Cons>>] loop:none 210 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 211 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 212 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 213 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 214 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Rep>>] loop:<<Loop>> outer_loop:none 215 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none sadShort2IntConstant2(short[] s)216 private static int sadShort2IntConstant2(short[] s) { 217 int sad = 0; 218 for (int i = 0; i < s.length; i++) { 219 sad += Math.abs(s[i] - $inline$seven()); // s[i] - 7 220 } 221 return sad; 222 } 223 224 /// CHECK-START: int SimdSadShort.sadShort2IntConstant3(short[]) loop_optimization (before) 225 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 226 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 227 /// CHECK-DAG: <<Cons:i\d+>> IntConstant 7 loop:none 228 /// CHECK-DAG: <<Phi2:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 229 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop>> outer_loop:none 230 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 231 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get1>>,<<Cons>>] loop:<<Loop>> outer_loop:none 232 /// CHECK-DAG: <<Intrin:i\d+>> Abs [<<Add>>] loop:<<Loop>> outer_loop:none 233 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 234 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 235 // 236 /// CHECK-START-ARM64: int SimdSadShort.sadShort2IntConstant3(short[]) loop_optimization (after) 237 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 238 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 239 /// CHECK-DAG: <<Cons:i\d+>> IntConstant -7 loop:none 240 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 241 /// CHECK-DAG: <<Rep:d\d+>> VecReplicateScalar [<<Cons>>] loop:none 242 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 243 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 244 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 245 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 246 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Rep>>] loop:<<Loop>> outer_loop:none 247 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none sadShort2IntConstant3(short[] s)248 private static int sadShort2IntConstant3(short[] s) { 249 int sad = 0; 250 for (int i = 0; i < s.length; i++) { 251 sad += Math.abs(s[i] + $inline$seven()); // hidden s[i] - (-7) 252 } 253 return sad; 254 } 255 256 /// CHECK-START: long SimdSadShort.sadShort2Long(short[], short[]) loop_optimization (before) 257 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 258 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 259 /// CHECK-DAG: <<ConsL:j\d+>> LongConstant 0 loop:none 260 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 261 /// CHECK-DAG: <<Phi2:j\d+>> Phi [<<ConsL>>,{{j\d+}}] loop:<<Loop>> outer_loop:none 262 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 263 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 264 /// CHECK-DAG: <<Cnv1:j\d+>> TypeConversion [<<Get1>>] loop:<<Loop>> outer_loop:none 265 /// CHECK-DAG: <<Cnv2:j\d+>> TypeConversion [<<Get2>>] loop:<<Loop>> outer_loop:none 266 /// CHECK-DAG: <<Sub:j\d+>> Sub [<<Cnv1>>,<<Cnv2>>] loop:<<Loop>> outer_loop:none 267 /// CHECK-DAG: <<Intrin:j\d+>> Abs [<<Sub>>] loop:<<Loop>> outer_loop:none 268 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 269 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 270 // 271 /// CHECK-START-ARM64: long SimdSadShort.sadShort2Long(short[], short[]) loop_optimization (after) 272 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 273 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 274 /// CHECK-DAG: <<ConsL:j\d+>> LongConstant 0 loop:none 275 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<ConsL>>] loop:none 276 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 277 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 278 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 279 /// CHECK-DAG: <<Load2:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 280 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Load2>>] loop:<<Loop>> outer_loop:none 281 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none sadShort2Long(short[] s1, short[] s2)282 private static long sadShort2Long(short[] s1, short[] s2) { 283 int min_length = Math.min(s1.length, s2.length); 284 long sad = 0; 285 for (int i = 0; i < min_length; i++) { 286 long x = s1[i]; 287 long y = s2[i]; 288 sad += Math.abs(x - y); 289 } 290 return sad; 291 } 292 293 /// CHECK-START: long SimdSadShort.sadShort2LongAt1(short[], short[]) loop_optimization (before) 294 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 295 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 296 /// CHECK-DAG: <<ConsL:j\d+>> LongConstant 1 loop:none 297 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 298 /// CHECK-DAG: <<Phi2:j\d+>> Phi [<<ConsL>>,{{j\d+}}] loop:<<Loop>> outer_loop:none 299 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 300 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 301 /// CHECK-DAG: <<Cnv1:j\d+>> TypeConversion [<<Get1>>] loop:<<Loop>> outer_loop:none 302 /// CHECK-DAG: <<Cnv2:j\d+>> TypeConversion [<<Get2>>] loop:<<Loop>> outer_loop:none 303 /// CHECK-DAG: <<Sub:j\d+>> Sub [<<Cnv1>>,<<Cnv2>>] loop:<<Loop>> outer_loop:none 304 /// CHECK-DAG: <<Intrin:j\d+>> Abs [<<Sub>>] loop:<<Loop>> outer_loop:none 305 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 306 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 307 // 308 /// CHECK-START-ARM64: long SimdSadShort.sadShort2LongAt1(short[], short[]) loop_optimization (after) 309 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 310 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 311 /// CHECK-DAG: <<ConsL:j\d+>> LongConstant 1 loop:none 312 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<ConsL>>] loop:none 313 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 314 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 315 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 316 /// CHECK-DAG: <<Load2:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 317 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Load2>>] loop:<<Loop>> outer_loop:none 318 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none sadShort2LongAt1(short[] s1, short[] s2)319 private static long sadShort2LongAt1(short[] s1, short[] s2) { 320 int min_length = Math.min(s1.length, s2.length); 321 long sad = 1; // starts at 1 322 for (int i = 0; i < min_length; i++) { 323 long x = s1[i]; 324 long y = s2[i]; 325 sad += Math.abs(x - y); 326 } 327 return sad; 328 } 329 main()330 public static void main() { 331 // Cross-test the two most extreme values individually. 332 short[] s1 = { 0, -32768, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; 333 short[] s2 = { 0, 32767, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; 334 expectEquals(-1, sadShort2Short(s1, s2)); 335 expectEquals(-1, sadShort2Short(s2, s1)); 336 expectEquals(-1, sadShort2ShortAlt(s1, s2)); 337 expectEquals(-1, sadShort2ShortAlt(s2, s1)); 338 expectEquals(-1, sadShort2ShortAlt2(s1, s2)); 339 expectEquals(-1, sadShort2ShortAlt2(s2, s1)); 340 expectEquals(65535, sadShort2Int(s1, s2)); 341 expectEquals(65535, sadShort2Int(s2, s1)); 342 expectEquals(65535, sadShort2IntAlt(s1, s2)); 343 expectEquals(65535, sadShort2IntAlt(s2, s1)); 344 expectEquals(65535, sadShort2IntAlt2(s1, s2)); 345 expectEquals(65535, sadShort2IntAlt2(s2, s1)); 346 expectEquals(32880, sadShort2IntConstant1(s1)); 347 expectEquals(32880, sadShort2IntConstant2(s1)); 348 expectEquals(32866, sadShort2IntConstant3(s1)); 349 expectEquals(65535L, sadShort2Long(s1, s2)); 350 expectEquals(65535L, sadShort2Long(s2, s1)); 351 expectEquals(65536L, sadShort2LongAt1(s1, s2)); 352 expectEquals(65536L, sadShort2LongAt1(s2, s1)); 353 354 // Use cross-values to test all cases. 355 short[] interesting = { 356 (short) 0x0000, 357 (short) 0x0001, 358 (short) 0x0002, 359 (short) 0x1234, 360 (short) 0x8000, 361 (short) 0x8001, 362 (short) 0x7fff, 363 (short) 0xffff 364 }; 365 int n = interesting.length; 366 int m = n * n + 1; 367 s1 = new short[m]; 368 s2 = new short[m]; 369 int k = 0; 370 for (int i = 0; i < n; i++) { 371 for (int j = 0; j < n; j++) { 372 s1[k] = interesting[i]; 373 s2[k] = interesting[j]; 374 k++; 375 } 376 } 377 s1[k] = 10; 378 s2[k] = 2; 379 expectEquals(-18932, sadShort2Short(s1, s2)); 380 expectEquals(-18932, sadShort2ShortAlt(s1, s2)); 381 expectEquals(-18932, sadShort2ShortAlt2(s1, s2)); 382 expectEquals(1291788, sadShort2Int(s1, s2)); 383 expectEquals(1291788, sadShort2IntAlt(s1, s2)); 384 expectEquals(1291788, sadShort2IntAlt2(s1, s2)); 385 expectEquals(823907, sadShort2IntConstant1(s1)); 386 expectEquals(823907, sadShort2IntConstant2(s1)); 387 expectEquals(823953, sadShort2IntConstant3(s1)); 388 expectEquals(1291788L, sadShort2Long(s1, s2)); 389 expectEquals(1291789L, sadShort2LongAt1(s1, s2)); 390 391 System.out.println("SimdSadShort passed"); 392 } 393 expectEquals(int expected, int result)394 private static void expectEquals(int expected, int result) { 395 if (expected != result) { 396 throw new Error("Expected: " + expected + ", found: " + result); 397 } 398 } 399 expectEquals(long expected, long result)400 private static void expectEquals(long expected, long result) { 401 if (expected != result) { 402 throw new Error("Expected: " + expected + ", found: " + result); 403 } 404 } 405 } 406