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 halving-add idiomatic vectorization. 19 * 20 * Alternative version expressed with logical shift right 21 * in the higher precision (has no impact on idiom). 22 */ 23 public class Main { 24 25 private static final int N = 256; 26 private static final int M = N * N + 15; 27 28 static byte[] sB1 = new byte[M]; 29 static byte[] sB2 = new byte[M]; 30 static byte[] sBo = new byte[M]; 31 32 /// CHECK-START: void Main.halving_add_signed(byte[], byte[], byte[]) loop_optimization (before) 33 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 34 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 35 /// CHECK-DAG: <<Get1:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 36 /// CHECK-DAG: <<Get2:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 37 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 38 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 39 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 40 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 41 // 42 /// CHECK-START-ARM: void Main.halving_add_signed(byte[], byte[], byte[]) loop_optimization (after) 43 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 44 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 45 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 46 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none 47 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 48 // 49 /// CHECK-START-ARM64: void Main.halving_add_signed(byte[], byte[], byte[]) loop_optimization (after) 50 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 51 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 52 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 53 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none 54 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 55 // 56 /// CHECK-START-MIPS64: void Main.halving_add_signed(byte[], byte[], byte[]) loop_optimization (after) 57 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 58 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 59 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 60 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none 61 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none halving_add_signed(byte[] b1, byte[] b2, byte[] bo)62 private static void halving_add_signed(byte[] b1, byte[] b2, byte[] bo) { 63 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 64 for (int i = 0; i < min_length; i++) { 65 bo[i] = (byte) ((b1[i] + b2[i]) >>> 1); 66 } 67 } 68 69 /// CHECK-START: void Main.halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (before) 70 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 71 /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none 72 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 73 /// CHECK-DAG: <<Get1:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 74 /// CHECK-DAG: <<Get2:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 75 /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<I255>>] loop:<<Loop>> outer_loop:none 76 /// CHECK-DAG: <<And2:i\d+>> And [<<Get2>>,<<I255>>] loop:<<Loop>> outer_loop:none 77 /// CHECK-DAG: <<Add:i\d+>> Add [<<And1>>,<<And2>>] loop:<<Loop>> outer_loop:none 78 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 79 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 80 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 81 // 82 /// CHECK-START-ARM: void Main.halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after) 83 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 84 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 85 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 86 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none 87 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 88 // 89 /// CHECK-START-ARM64: void Main.halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after) 90 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 91 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 92 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 93 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none 94 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 95 // 96 /// CHECK-START-MIPS64: void Main.halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after) 97 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 98 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 99 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 100 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none 101 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo)102 private static void halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo) { 103 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 104 for (int i = 0; i < min_length; i++) { 105 bo[i] = (byte) (((b1[i] & 0xff) + (b2[i] & 0xff)) >>> 1); 106 } 107 } 108 109 /// CHECK-START: void Main.rounding_halving_add_signed(byte[], byte[], byte[]) loop_optimization (before) 110 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 111 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 112 /// CHECK-DAG: <<Get1:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 113 /// CHECK-DAG: <<Get2:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 114 /// CHECK-DAG: <<Add1:i\d+>> Add [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 115 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<I1>>] loop:<<Loop>> outer_loop:none 116 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 117 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 118 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 119 // 120 /// CHECK-START-ARM: void Main.rounding_halving_add_signed(byte[], byte[], byte[]) loop_optimization (after) 121 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 122 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 123 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 124 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none 125 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 126 // 127 /// CHECK-START-ARM64: void Main.rounding_halving_add_signed(byte[], byte[], byte[]) loop_optimization (after) 128 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 129 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 130 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 131 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none 132 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 133 // 134 /// CHECK-START-MIPS64: void Main.rounding_halving_add_signed(byte[], byte[], byte[]) loop_optimization (after) 135 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 136 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 137 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 138 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none 139 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none rounding_halving_add_signed(byte[] b1, byte[] b2, byte[] bo)140 private static void rounding_halving_add_signed(byte[] b1, byte[] b2, byte[] bo) { 141 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 142 for (int i = 0; i < min_length; i++) { 143 bo[i] = (byte) ((b1[i] + b2[i] + 1) >>> 1); 144 } 145 } 146 147 /// CHECK-START: void Main.rounding_halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (before) 148 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 149 /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none 150 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 151 /// CHECK-DAG: <<Get1:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 152 /// CHECK-DAG: <<Get2:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 153 /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<I255>>] loop:<<Loop>> outer_loop:none 154 /// CHECK-DAG: <<And2:i\d+>> And [<<Get2>>,<<I255>>] loop:<<Loop>> outer_loop:none 155 /// CHECK-DAG: <<Add1:i\d+>> Add [<<And1>>,<<And2>>] loop:<<Loop>> outer_loop:none 156 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<I1>>] loop:<<Loop>> outer_loop:none 157 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 158 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 159 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 160 // 161 /// CHECK-START-ARM: void Main.rounding_halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after) 162 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 163 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 164 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 165 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none 166 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 167 // 168 /// CHECK-START-ARM64: void Main.rounding_halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after) 169 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 170 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 171 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 172 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none 173 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 174 // 175 /// CHECK-START-MIPS64: void Main.rounding_halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after) 176 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 177 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 178 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 179 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none 180 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none rounding_halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo)181 private static void rounding_halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo) { 182 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 183 for (int i = 0; i < min_length; i++) { 184 bo[i] = (byte) (((b1[i] & 0xff) + (b2[i] & 0xff) + 1) >>> 1); 185 } 186 } 187 188 /// CHECK-START: void Main.halving_add_signed_constant(byte[], byte[]) loop_optimization (before) 189 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 190 /// CHECK-DAG: <<I127:i\d+>> IntConstant 127 loop:none 191 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 192 /// CHECK-DAG: <<Get:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 193 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get>>,<<I127>>] loop:<<Loop>> outer_loop:none 194 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 195 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 196 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 197 // 198 /// CHECK-START-ARM: void Main.halving_add_signed_constant(byte[], byte[]) loop_optimization (after) 199 /// CHECK-DAG: <<I127:i\d+>> IntConstant 127 loop:none 200 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I127>>] loop:none 201 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 202 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 203 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none 204 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 205 // 206 /// CHECK-START-ARM64: void Main.halving_add_signed_constant(byte[], byte[]) loop_optimization (after) 207 /// CHECK-DAG: <<I127:i\d+>> IntConstant 127 loop:none 208 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I127>>] loop:none 209 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 210 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 211 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none 212 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 213 // 214 /// CHECK-START-MIPS64: void Main.halving_add_signed_constant(byte[], byte[]) loop_optimization (after) 215 /// CHECK-DAG: <<I127:i\d+>> IntConstant 127 loop:none 216 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I127>>] loop:none 217 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 218 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 219 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none 220 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none halving_add_signed_constant(byte[] b1, byte[] bo)221 private static void halving_add_signed_constant(byte[] b1, byte[] bo) { 222 int min_length = Math.min(bo.length, b1.length); 223 for (int i = 0; i < min_length; i++) { 224 bo[i] = (byte) ((b1[i] + 0x7f) >>> 1); 225 } 226 } 227 228 /// CHECK-START: void Main.halving_add_unsigned_constant(byte[], byte[]) loop_optimization (before) 229 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 230 /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none 231 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 232 /// CHECK-DAG: <<Get:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 233 /// CHECK-DAG: <<And:i\d+>> And [<<Get>>,<<I255>>] loop:<<Loop>> outer_loop:none 234 /// CHECK-DAG: <<Add:i\d+>> Add [<<And>>,<<I255>>] loop:<<Loop>> outer_loop:none 235 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 236 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 237 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 238 // 239 /// CHECK-START-ARM: void Main.halving_add_unsigned_constant(byte[], byte[]) loop_optimization (after) 240 /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none 241 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I255>>] loop:none 242 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 243 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 244 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none 245 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 246 // 247 /// CHECK-START-ARM64: void Main.halving_add_unsigned_constant(byte[], byte[]) loop_optimization (after) 248 /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none 249 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I255>>] loop:none 250 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 251 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 252 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none 253 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none 254 // 255 /// CHECK-START-MIPS64: void Main.halving_add_unsigned_constant(byte[], byte[]) loop_optimization (after) 256 /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none 257 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I255>>] loop:none 258 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 259 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 260 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none 261 /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>> outer_loop:none halving_add_unsigned_constant(byte[] b1, byte[] bo)262 private static void halving_add_unsigned_constant(byte[] b1, byte[] bo) { 263 int min_length = Math.min(bo.length, b1.length); 264 for (int i = 0; i < min_length; i++) { 265 bo[i] = (byte) (((b1[i] & 0xff) + 0xff) >>> 1); 266 } 267 } 268 main(String[] args)269 public static void main(String[] args) { 270 // Initialize cross-values to test all cases, and also 271 // set up some extra values to exercise the cleanup loop. 272 int k = 0; 273 for (int i = 0; i < N; i++) { 274 for (int j = 0; j < N; j++) { 275 sB1[k] = (byte) i; 276 sB2[k] = (byte) j; 277 k++; 278 } 279 } 280 for (int i = 0; i < 15; i++) { 281 sB1[k] = (byte) i; 282 sB2[k] = 100; 283 k++; 284 } 285 expectEquals(k, M); 286 287 // Test halving add idioms. Note that the expected result is computed 288 // with the arithmetic >> to demonstrate the computed narrower result 289 // does not depend on the wider >> or >>>. 290 halving_add_signed(sB1, sB2, sBo); 291 for (int i = 0; i < M; i++) { 292 byte e = (byte) ((sB1[i] + sB2[i]) >> 1); 293 expectEquals(e, sBo[i]); 294 } 295 halving_add_unsigned(sB1, sB2, sBo); 296 for (int i = 0; i < M; i++) { 297 byte e = (byte) (((sB1[i] & 0xff) + (sB2[i] & 0xff)) >> 1); 298 expectEquals(e, sBo[i]); 299 } 300 rounding_halving_add_signed(sB1, sB2, sBo); 301 for (int i = 0; i < M; i++) { 302 byte e = (byte) ((sB1[i] + sB2[i] + 1) >> 1); 303 expectEquals(e, sBo[i]); 304 } 305 rounding_halving_add_unsigned(sB1, sB2, sBo); 306 for (int i = 0; i < M; i++) { 307 byte e = (byte) (((sB1[i] & 0xff) + (sB2[i] & 0xff) + 1) >> 1); 308 expectEquals(e, sBo[i]); 309 } 310 halving_add_signed_constant(sB1, sBo); 311 for (int i = 0; i < M; i++) { 312 byte e = (byte) ((sB1[i] + 0x7f) >> 1); 313 expectEquals(e, sBo[i]); 314 } 315 halving_add_unsigned_constant(sB1, sBo); 316 for (int i = 0; i < M; i++) { 317 byte e = (byte) (((sB1[i] & 0xff) + 0xff) >> 1); 318 expectEquals(e, sBo[i]); 319 } 320 321 System.out.println("passed"); 322 } 323 expectEquals(int expected, int result)324 private static void expectEquals(int expected, int result) { 325 if (expected != result) { 326 throw new Error("Expected: " + expected + ", found: " + result); 327 } 328 } 329 } 330