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-IF: hasIsaFeature("sve") 76 // 77 // SAD idiom is not supported for SVE. 78 /// CHECK-NOT: VecSADAccumulate 79 // 80 /// CHECK-ELSE: 81 // 82 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 83 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 84 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 85 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 86 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 87 /// CHECK-DAG: <<Load2:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 88 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Load2>>] loop:<<Loop>> outer_loop:none 89 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none 90 // 91 /// CHECK-FI: sadShort2Int(short[] s1, short[] s2)92 private static int sadShort2Int(short[] s1, short[] s2) { 93 int min_length = Math.min(s1.length, s2.length); 94 int sad = 0; 95 for (int i = 0; i < min_length; i++) { 96 sad += Math.abs(s1[i] - s2[i]); 97 } 98 return sad; 99 } 100 101 /// CHECK-START: int SimdSadShort.sadShort2IntAlt(short[], short[]) loop_optimization (before) 102 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 103 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 104 /// CHECK-DAG: <<Phi2:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 105 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop>> outer_loop:none 106 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 107 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 108 /// CHECK-DAG: <<Sub:i\d+>> Sub [<<Get2>>,<<Get1>>] loop:<<Loop>> outer_loop:none 109 /// CHECK-DAG: <<Intrin:i\d+>> Abs [<<Sub>>] loop:<<Loop>> outer_loop:none 110 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 111 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 112 // 113 /// CHECK-START-ARM64: int SimdSadShort.sadShort2IntAlt(short[], short[]) loop_optimization (after) 114 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 115 /// CHECK-IF: hasIsaFeature("sve") 116 // 117 // SAD idiom is not supported for SVE. 118 /// CHECK-NOT: VecSADAccumulate 119 // 120 /// CHECK-ELSE: 121 // 122 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 123 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 124 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 125 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 126 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 127 /// CHECK-DAG: <<Load2:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 128 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load2>>,<<Load1>>] loop:<<Loop>> outer_loop:none 129 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none 130 // 131 /// CHECK-FI: sadShort2IntAlt(short[] s1, short[] s2)132 private static int sadShort2IntAlt(short[] s1, short[] s2) { 133 int min_length = Math.min(s1.length, s2.length); 134 int sad = 0; 135 for (int i = 0; i < min_length; i++) { 136 short s = s1[i]; 137 short p = s2[i]; 138 sad += s >= p ? s - p : p - s; 139 } 140 return sad; 141 } 142 143 /// CHECK-START: int SimdSadShort.sadShort2IntAlt2(short[], short[]) loop_optimization (before) 144 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 145 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 146 /// CHECK-DAG: <<Phi2:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 147 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop>> outer_loop:none 148 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 149 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 150 /// CHECK-DAG: <<Sub:i\d+>> Sub [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 151 /// CHECK-DAG: <<Intrin:i\d+>> Abs [<<Sub>>] loop:<<Loop>> outer_loop:none 152 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 153 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 154 // 155 /// CHECK-START-ARM64: int SimdSadShort.sadShort2IntAlt2(short[], short[]) loop_optimization (after) 156 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 157 /// CHECK-IF: hasIsaFeature("sve") 158 // 159 // SAD idiom is not supported for SVE. 160 /// CHECK-NOT: VecSADAccumulate 161 // 162 /// CHECK-ELSE: 163 // 164 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 165 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 166 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 167 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 168 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 169 /// CHECK-DAG: <<Load2:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 170 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Load2>>] loop:<<Loop>> outer_loop:none 171 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none 172 // 173 /// CHECK-FI: sadShort2IntAlt2(short[] s1, short[] s2)174 private static int sadShort2IntAlt2(short[] s1, short[] s2) { 175 int min_length = Math.min(s1.length, s2.length); 176 int sad = 0; 177 for (int i = 0; i < min_length; i++) { 178 short s = s1[i]; 179 short p = s2[i]; 180 int x = s - p; 181 if (x < 0) x = -x; 182 sad += x; 183 } 184 return sad; 185 } 186 187 /// CHECK-START: int SimdSadShort.sadShort2IntConstant1(short[]) loop_optimization (before) 188 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 189 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 190 /// CHECK-DAG: <<Cons:i\d+>> IntConstant -7 loop:none 191 /// CHECK-DAG: <<Phi2:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 192 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop>> outer_loop:none 193 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 194 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get1>>,<<Cons>>] loop:<<Loop>> outer_loop:none 195 /// CHECK-DAG: <<Intrin:i\d+>> Abs [<<Add>>] loop:<<Loop>> outer_loop:none 196 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 197 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 198 // 199 /// CHECK-START-ARM64: int SimdSadShort.sadShort2IntConstant1(short[]) loop_optimization (after) 200 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 201 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 202 /// CHECK-DAG: <<Cons:i\d+>> IntConstant 7 loop:none 203 /// CHECK-IF: hasIsaFeature("sve") 204 // 205 // SAD idiom is not supported for SVE. 206 /// CHECK-NOT: VecSADAccumulate 207 // 208 /// CHECK-ELSE: 209 // 210 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 211 /// CHECK-DAG: <<Rep:d\d+>> VecReplicateScalar [<<Cons>>] loop:none 212 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 213 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 214 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 215 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 216 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Rep>>] loop:<<Loop>> outer_loop:none 217 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none 218 // 219 /// CHECK-FI: sadShort2IntConstant1(short[] s)220 private static int sadShort2IntConstant1(short[] s) { 221 int sad = 0; 222 for (int i = 0; i < s.length; i++) { 223 sad += Math.abs(s[i] - 7); // s[i] + -7 224 } 225 return sad; 226 } 227 228 /// CHECK-START: int SimdSadShort.sadShort2IntConstant2(short[]) loop_optimization (before) 229 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 230 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 231 /// CHECK-DAG: <<Cons:i\d+>> IntConstant 7 loop:none 232 /// CHECK-DAG: <<Phi2:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 233 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop>> outer_loop:none 234 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 235 /// CHECK-DAG: <<Sub:i\d+>> Sub [<<Get1>>,<<Cons>>] loop:<<Loop>> outer_loop:none 236 /// CHECK-DAG: <<Intrin:i\d+>> Abs [<<Sub>>] loop:<<Loop>> outer_loop:none 237 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 238 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 239 // 240 /// CHECK-START-ARM64: int SimdSadShort.sadShort2IntConstant2(short[]) loop_optimization (after) 241 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 242 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 243 /// CHECK-DAG: <<Cons:i\d+>> IntConstant 7 loop:none 244 /// CHECK-IF: hasIsaFeature("sve") 245 // 246 // SAD idiom is not supported for SVE. 247 /// CHECK-NOT: VecSADAccumulate 248 // 249 /// CHECK-ELSE: 250 // 251 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 252 /// CHECK-DAG: <<Rep:d\d+>> VecReplicateScalar [<<Cons>>] loop:none 253 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 254 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 255 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 256 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 257 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Rep>>] loop:<<Loop>> outer_loop:none 258 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none 259 // 260 /// CHECK-FI: sadShort2IntConstant2(short[] s)261 private static int sadShort2IntConstant2(short[] s) { 262 int sad = 0; 263 for (int i = 0; i < s.length; i++) { 264 sad += Math.abs(s[i] - $inline$seven()); // s[i] - 7 265 } 266 return sad; 267 } 268 269 /// CHECK-START: int SimdSadShort.sadShort2IntConstant3(short[]) loop_optimization (before) 270 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 271 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 272 /// CHECK-DAG: <<Cons:i\d+>> IntConstant 7 loop:none 273 /// CHECK-DAG: <<Phi2:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 274 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop>> outer_loop:none 275 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 276 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get1>>,<<Cons>>] loop:<<Loop>> outer_loop:none 277 /// CHECK-DAG: <<Intrin:i\d+>> Abs [<<Add>>] loop:<<Loop>> outer_loop:none 278 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 279 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 280 // 281 /// CHECK-START-ARM64: int SimdSadShort.sadShort2IntConstant3(short[]) loop_optimization (after) 282 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 283 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 284 /// CHECK-DAG: <<Cons:i\d+>> IntConstant -7 loop:none 285 /// CHECK-IF: hasIsaFeature("sve") 286 // 287 // SAD idiom is not supported for SVE. 288 /// CHECK-NOT: VecSADAccumulate 289 // 290 /// CHECK-ELSE: 291 // 292 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 293 /// CHECK-DAG: <<Rep:d\d+>> VecReplicateScalar [<<Cons>>] loop:none 294 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<Cons0>>] loop:none 295 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 296 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 297 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 298 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Rep>>] loop:<<Loop>> outer_loop:none 299 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none 300 // 301 /// CHECK-FI: sadShort2IntConstant3(short[] s)302 private static int sadShort2IntConstant3(short[] s) { 303 int sad = 0; 304 for (int i = 0; i < s.length; i++) { 305 sad += Math.abs(s[i] + $inline$seven()); // hidden s[i] - (-7) 306 } 307 return sad; 308 } 309 310 /// CHECK-START: long SimdSadShort.sadShort2Long(short[], short[]) loop_optimization (before) 311 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 312 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 313 /// CHECK-DAG: <<ConsL:j\d+>> LongConstant 0 loop:none 314 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 315 /// CHECK-DAG: <<Phi2:j\d+>> Phi [<<ConsL>>,{{j\d+}}] loop:<<Loop>> outer_loop:none 316 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 317 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 318 /// CHECK-DAG: <<Cnv1:j\d+>> TypeConversion [<<Get1>>] loop:<<Loop>> outer_loop:none 319 /// CHECK-DAG: <<Cnv2:j\d+>> TypeConversion [<<Get2>>] loop:<<Loop>> outer_loop:none 320 /// CHECK-DAG: <<Sub:j\d+>> Sub [<<Cnv1>>,<<Cnv2>>] loop:<<Loop>> outer_loop:none 321 /// CHECK-DAG: <<Intrin:j\d+>> Abs [<<Sub>>] loop:<<Loop>> outer_loop:none 322 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 323 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 324 // 325 /// CHECK-START-ARM64: long SimdSadShort.sadShort2Long(short[], short[]) loop_optimization (after) 326 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 327 /// CHECK-DAG: <<ConsL:j\d+>> LongConstant 0 loop:none 328 /// CHECK-IF: hasIsaFeature("sve") 329 // 330 // SAD idiom is not supported for SVE. 331 /// CHECK-NOT: VecSADAccumulate 332 // 333 /// CHECK-ELSE: 334 // 335 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 336 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<ConsL>>] loop:none 337 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 338 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 339 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 340 /// CHECK-DAG: <<Load2:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 341 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Load2>>] loop:<<Loop>> outer_loop:none 342 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none 343 // 344 /// CHECK-FI: sadShort2Long(short[] s1, short[] s2)345 private static long sadShort2Long(short[] s1, short[] s2) { 346 int min_length = Math.min(s1.length, s2.length); 347 long sad = 0; 348 for (int i = 0; i < min_length; i++) { 349 long x = s1[i]; 350 long y = s2[i]; 351 sad += Math.abs(x - y); 352 } 353 return sad; 354 } 355 356 /// CHECK-START: long SimdSadShort.sadShort2LongAt1(short[], short[]) loop_optimization (before) 357 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 358 /// CHECK-DAG: <<Cons1:i\d+>> IntConstant 1 loop:none 359 /// CHECK-DAG: <<ConsL:j\d+>> LongConstant 1 loop:none 360 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 361 /// CHECK-DAG: <<Phi2:j\d+>> Phi [<<ConsL>>,{{j\d+}}] loop:<<Loop>> outer_loop:none 362 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 363 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 364 /// CHECK-DAG: <<Cnv1:j\d+>> TypeConversion [<<Get1>>] loop:<<Loop>> outer_loop:none 365 /// CHECK-DAG: <<Cnv2:j\d+>> TypeConversion [<<Get2>>] loop:<<Loop>> outer_loop:none 366 /// CHECK-DAG: <<Sub:j\d+>> Sub [<<Cnv1>>,<<Cnv2>>] loop:<<Loop>> outer_loop:none 367 /// CHECK-DAG: <<Intrin:j\d+>> Abs [<<Sub>>] loop:<<Loop>> outer_loop:none 368 /// CHECK-DAG: Add [<<Phi2>>,<<Intrin>>] loop:<<Loop>> outer_loop:none 369 /// CHECK-DAG: Add [<<Phi1>>,<<Cons1>>] loop:<<Loop>> outer_loop:none 370 // 371 /// CHECK-START-ARM64: long SimdSadShort.sadShort2LongAt1(short[], short[]) loop_optimization (after) 372 /// CHECK-DAG: <<Cons0:i\d+>> IntConstant 0 loop:none 373 /// CHECK-DAG: <<ConsL:j\d+>> LongConstant 1 loop:none 374 /// CHECK-IF: hasIsaFeature("sve") 375 // 376 // SAD idiom is not supported for SVE. 377 /// CHECK-NOT: VecSADAccumulate 378 // 379 /// CHECK-ELSE: 380 // 381 /// CHECK-DAG: <<Cons8:i\d+>> IntConstant 8 loop:none 382 /// CHECK-DAG: <<Set:d\d+>> VecSetScalars [<<ConsL>>] loop:none 383 /// CHECK-DAG: <<Phi1:i\d+>> Phi [<<Cons0>>,{{i\d+}}] loop:<<Loop:B\d+>> outer_loop:none 384 /// CHECK-DAG: <<Phi2:d\d+>> Phi [<<Set>>,{{d\d+}}] loop:<<Loop>> outer_loop:none 385 /// CHECK-DAG: <<Load1:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 386 /// CHECK-DAG: <<Load2:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop>> outer_loop:none 387 /// CHECK-DAG: <<SAD:d\d+>> VecSADAccumulate [<<Phi2>>,<<Load1>>,<<Load2>>] loop:<<Loop>> outer_loop:none 388 /// CHECK-DAG: Add [<<Phi1>>,<<Cons8>>] loop:<<Loop>> outer_loop:none 389 // 390 /// CHECK-FI: sadShort2LongAt1(short[] s1, short[] s2)391 private static long sadShort2LongAt1(short[] s1, short[] s2) { 392 int min_length = Math.min(s1.length, s2.length); 393 long sad = 1; // starts at 1 394 for (int i = 0; i < min_length; i++) { 395 long x = s1[i]; 396 long y = s2[i]; 397 sad += Math.abs(x - y); 398 } 399 return sad; 400 } 401 main()402 public static void main() { 403 // Cross-test the two most extreme values individually. 404 short[] s1 = { 0, -32768, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; 405 short[] s2 = { 0, 32767, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; 406 expectEquals(-1, sadShort2Short(s1, s2)); 407 expectEquals(-1, sadShort2Short(s2, s1)); 408 expectEquals(-1, sadShort2ShortAlt(s1, s2)); 409 expectEquals(-1, sadShort2ShortAlt(s2, s1)); 410 expectEquals(-1, sadShort2ShortAlt2(s1, s2)); 411 expectEquals(-1, sadShort2ShortAlt2(s2, s1)); 412 expectEquals(65535, sadShort2Int(s1, s2)); 413 expectEquals(65535, sadShort2Int(s2, s1)); 414 expectEquals(65535, sadShort2IntAlt(s1, s2)); 415 expectEquals(65535, sadShort2IntAlt(s2, s1)); 416 expectEquals(65535, sadShort2IntAlt2(s1, s2)); 417 expectEquals(65535, sadShort2IntAlt2(s2, s1)); 418 expectEquals(32880, sadShort2IntConstant1(s1)); 419 expectEquals(32880, sadShort2IntConstant2(s1)); 420 expectEquals(32866, sadShort2IntConstant3(s1)); 421 expectEquals(65535L, sadShort2Long(s1, s2)); 422 expectEquals(65535L, sadShort2Long(s2, s1)); 423 expectEquals(65536L, sadShort2LongAt1(s1, s2)); 424 expectEquals(65536L, sadShort2LongAt1(s2, s1)); 425 426 // Use cross-values to test all cases. 427 short[] interesting = { 428 (short) 0x0000, 429 (short) 0x0001, 430 (short) 0x0002, 431 (short) 0x1234, 432 (short) 0x8000, 433 (short) 0x8001, 434 (short) 0x7fff, 435 (short) 0xffff 436 }; 437 int n = interesting.length; 438 int m = n * n + 1; 439 s1 = new short[m]; 440 s2 = new short[m]; 441 int k = 0; 442 for (int i = 0; i < n; i++) { 443 for (int j = 0; j < n; j++) { 444 s1[k] = interesting[i]; 445 s2[k] = interesting[j]; 446 k++; 447 } 448 } 449 s1[k] = 10; 450 s2[k] = 2; 451 expectEquals(-18932, sadShort2Short(s1, s2)); 452 expectEquals(-18932, sadShort2ShortAlt(s1, s2)); 453 expectEquals(-18932, sadShort2ShortAlt2(s1, s2)); 454 expectEquals(1291788, sadShort2Int(s1, s2)); 455 expectEquals(1291788, sadShort2IntAlt(s1, s2)); 456 expectEquals(1291788, sadShort2IntAlt2(s1, s2)); 457 expectEquals(823907, sadShort2IntConstant1(s1)); 458 expectEquals(823907, sadShort2IntConstant2(s1)); 459 expectEquals(823953, sadShort2IntConstant3(s1)); 460 expectEquals(1291788L, sadShort2Long(s1, s2)); 461 expectEquals(1291789L, sadShort2LongAt1(s1, s2)); 462 463 System.out.println("SimdSadShort passed"); 464 } 465 expectEquals(int expected, int result)466 private static void expectEquals(int expected, int result) { 467 if (expected != result) { 468 throw new Error("Expected: " + expected + ", found: " + result); 469 } 470 } 471 expectEquals(long expected, long result)472 private static void expectEquals(long expected, long result) { 473 if (expected != result) { 474 throw new Error("Expected: " + expected + ", found: " + result); 475 } 476 } 477 } 478