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 HaddAltByte { 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 HaddAltByte.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 HaddAltByte.halving_add_signed(byte[], byte[], byte[]) loop_optimization (after) 43 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 44 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 45 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none 46 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 47 // 48 /// CHECK-START-ARM64: void HaddAltByte.halving_add_signed(byte[], byte[], byte[]) loop_optimization (after) 49 /// CHECK-IF: hasIsaFeature("sve") 50 // 51 // HalvingAdd idiom is not supported for SVE. 52 /// CHECK-NOT: VecHalvingAdd 53 // 54 /// CHECK-ELSE: 55 // 56 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 57 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 58 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none 59 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 60 // 61 /// CHECK-FI: 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 HaddAltByte.halving_add_unsigned(byte[], byte[], byte[]) instruction_simplifier (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+}},{{i\d+}},<<Cnv>>] loop:<<Loop>> outer_loop:none 81 // 82 /// CHECK-START: void HaddAltByte.halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (before) 83 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 84 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 85 /// CHECK-DAG: <<Get1:a\d+>> ArrayGet loop:<<Loop>> outer_loop:none 86 /// CHECK-DAG: <<Get2:a\d+>> ArrayGet loop:<<Loop>> outer_loop:none 87 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 88 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 89 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 90 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 91 // 92 /// CHECK-START-ARM: void HaddAltByte.halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after) 93 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 94 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 95 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none 96 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 97 // 98 /// CHECK-START-ARM64: void HaddAltByte.halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after) 99 /// CHECK-IF: hasIsaFeature("sve") 100 // 101 // HalvingAdd idiom is not supported for SVE. 102 /// CHECK-NOT: VecHalvingAdd 103 // 104 /// CHECK-ELSE: 105 // 106 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 107 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 108 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none 109 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 110 // 111 /// CHECK-FI: halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo)112 private static void halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo) { 113 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 114 for (int i = 0; i < min_length; i++) { 115 bo[i] = (byte) (((b1[i] & 0xff) + (b2[i] & 0xff)) >>> 1); 116 } 117 } 118 119 /// CHECK-START: void HaddAltByte.rounding_halving_add_signed(byte[], byte[], byte[]) loop_optimization (before) 120 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 121 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 122 /// CHECK-DAG: <<Get1:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 123 /// CHECK-DAG: <<Get2:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 124 /// CHECK-DAG: <<Add1:i\d+>> Add [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 125 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<I1>>] loop:<<Loop>> outer_loop:none 126 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 127 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 128 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 129 // 130 /// CHECK-START-ARM: void HaddAltByte.rounding_halving_add_signed(byte[], byte[], byte[]) loop_optimization (after) 131 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 132 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 133 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:true loop:<<Loop>> outer_loop:none 134 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 135 // 136 /// CHECK-START-ARM64: void HaddAltByte.rounding_halving_add_signed(byte[], byte[], byte[]) loop_optimization (after) 137 /// CHECK-IF: hasIsaFeature("sve") 138 // 139 // HalvingAdd idiom is not supported for SVE. 140 /// CHECK-NOT: VecHalvingAdd 141 // 142 /// CHECK-ELSE: 143 // 144 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 145 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 146 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:true loop:<<Loop>> outer_loop:none 147 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 148 // 149 /// CHECK-FI: rounding_halving_add_signed(byte[] b1, byte[] b2, byte[] bo)150 private static void rounding_halving_add_signed(byte[] b1, byte[] b2, byte[] bo) { 151 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 152 for (int i = 0; i < min_length; i++) { 153 bo[i] = (byte) ((b1[i] + b2[i] + 1) >>> 1); 154 } 155 } 156 157 /// CHECK-START: void HaddAltByte.rounding_halving_add_unsigned(byte[], byte[], byte[]) instruction_simplifier (before) 158 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 159 /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none 160 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 161 /// CHECK-DAG: <<Get1:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 162 /// CHECK-DAG: <<Get2:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 163 /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<I255>>] loop:<<Loop>> outer_loop:none 164 /// CHECK-DAG: <<And2:i\d+>> And [<<Get2>>,<<I255>>] loop:<<Loop>> outer_loop:none 165 /// CHECK-DAG: <<Add1:i\d+>> Add [<<And1>>,<<And2>>] loop:<<Loop>> outer_loop:none 166 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<I1>>] loop:<<Loop>> outer_loop:none 167 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 168 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 169 /// CHECK-DAG: ArraySet [{{l\d+}},{{i\d+}},<<Cnv>>] loop:<<Loop>> outer_loop:none 170 // 171 /// CHECK-START: void HaddAltByte.rounding_halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (before) 172 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 173 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 174 /// CHECK-DAG: <<Get1:a\d+>> ArrayGet loop:<<Loop>> outer_loop:none 175 /// CHECK-DAG: <<Get2:a\d+>> ArrayGet loop:<<Loop>> outer_loop:none 176 /// CHECK-DAG: <<Add1:i\d+>> Add [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 177 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<I1>>] loop:<<Loop>> outer_loop:none 178 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 179 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 180 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 181 // 182 /// CHECK-START-ARM: void HaddAltByte.rounding_halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after) 183 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 184 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 185 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint8 rounded:true loop:<<Loop>> outer_loop:none 186 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 187 // 188 /// CHECK-START-ARM64: void HaddAltByte.rounding_halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after) 189 /// CHECK-IF: hasIsaFeature("sve") 190 // 191 // HalvingAdd idiom is not supported for SVE. 192 /// CHECK-NOT: VecHalvingAdd 193 // 194 /// CHECK-ELSE: 195 // 196 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 197 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 198 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint8 rounded:true loop:<<Loop>> outer_loop:none 199 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 200 // 201 /// CHECK-FI: rounding_halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo)202 private static void rounding_halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo) { 203 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 204 for (int i = 0; i < min_length; i++) { 205 bo[i] = (byte) (((b1[i] & 0xff) + (b2[i] & 0xff) + 1) >>> 1); 206 } 207 } 208 209 /// CHECK-START: void HaddAltByte.halving_add_signed_constant(byte[], byte[]) loop_optimization (before) 210 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 211 /// CHECK-DAG: <<I127:i\d+>> IntConstant 127 loop:none 212 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 213 /// CHECK-DAG: <<Get:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 214 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get>>,<<I127>>] loop:<<Loop>> outer_loop:none 215 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 216 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 217 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 218 // 219 /// CHECK-START-ARM: void HaddAltByte.halving_add_signed_constant(byte[], byte[]) loop_optimization (after) 220 /// CHECK-DAG: <<I127:i\d+>> IntConstant 127 loop:none 221 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I127>>] loop:none 222 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 223 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none 224 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 225 // 226 /// CHECK-START-ARM64: void HaddAltByte.halving_add_signed_constant(byte[], byte[]) loop_optimization (after) 227 /// CHECK-IF: hasIsaFeature("sve") 228 // 229 // HalvingAdd idiom is not supported for SVE. 230 /// CHECK-NOT: VecHalvingAdd 231 // 232 /// CHECK-ELSE: 233 // 234 /// CHECK-DAG: <<I127:i\d+>> IntConstant 127 loop:none 235 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I127>>] loop:none 236 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 237 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none 238 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 239 // 240 /// CHECK-FI: halving_add_signed_constant(byte[] b1, byte[] bo)241 private static void halving_add_signed_constant(byte[] b1, byte[] bo) { 242 int min_length = Math.min(bo.length, b1.length); 243 for (int i = 0; i < min_length; i++) { 244 bo[i] = (byte) ((b1[i] + 0x7f) >>> 1); 245 } 246 } 247 248 /// CHECK-START: void HaddAltByte.halving_add_unsigned_constant(byte[], byte[]) instruction_simplifier (before) 249 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 250 /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none 251 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 252 /// CHECK-DAG: <<Get:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none 253 /// CHECK-DAG: <<And:i\d+>> And [<<Get>>,<<I255>>] loop:<<Loop>> outer_loop:none 254 /// CHECK-DAG: <<Add:i\d+>> Add [<<And>>,<<I255>>] loop:<<Loop>> outer_loop:none 255 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 256 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 257 /// CHECK-DAG: ArraySet [{{l\d+}},{{i\d+}},<<Cnv>>] loop:<<Loop>> outer_loop:none 258 // 259 /// CHECK-START: void HaddAltByte.halving_add_unsigned_constant(byte[], byte[]) loop_optimization (before) 260 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 261 /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none 262 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 263 /// CHECK-DAG: <<Get:a\d+>> ArrayGet loop:<<Loop>> outer_loop:none 264 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get>>,<<I255>>] loop:<<Loop>> outer_loop:none 265 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 266 /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 267 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 268 // 269 /// CHECK-START-ARM: void HaddAltByte.halving_add_unsigned_constant(byte[], byte[]) loop_optimization (after) 270 /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none 271 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I255>>] loop:none 272 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 273 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none 274 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 275 // 276 /// CHECK-START-ARM64: void HaddAltByte.halving_add_unsigned_constant(byte[], byte[]) loop_optimization (after) 277 /// CHECK-IF: hasIsaFeature("sve") 278 // 279 // HalvingAdd idiom is not supported for SVE. 280 /// CHECK-NOT: VecHalvingAdd 281 // 282 /// CHECK-ELSE: 283 // 284 /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none 285 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I255>>] loop:none 286 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 287 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none 288 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 289 // 290 /// CHECK-FI: halving_add_unsigned_constant(byte[] b1, byte[] bo)291 private static void halving_add_unsigned_constant(byte[] b1, byte[] bo) { 292 int min_length = Math.min(bo.length, b1.length); 293 for (int i = 0; i < min_length; i++) { 294 bo[i] = (byte) (((b1[i] & 0xff) + 0xff) >>> 1); 295 } 296 } 297 main()298 public static void main() { 299 // Initialize cross-values to test all cases, and also 300 // set up some extra values to exercise the cleanup loop. 301 int k = 0; 302 for (int i = 0; i < N; i++) { 303 for (int j = 0; j < N; j++) { 304 sB1[k] = (byte) i; 305 sB2[k] = (byte) j; 306 k++; 307 } 308 } 309 for (int i = 0; i < 15; i++) { 310 sB1[k] = (byte) i; 311 sB2[k] = 100; 312 k++; 313 } 314 expectEquals(k, M); 315 316 // Test halving add idioms. Note that the expected result is computed 317 // with the arithmetic >> to demonstrate the computed narrower result 318 // does not depend on the wider >> or >>>. 319 halving_add_signed(sB1, sB2, sBo); 320 for (int i = 0; i < M; i++) { 321 byte e = (byte) ((sB1[i] + sB2[i]) >> 1); 322 expectEquals(e, sBo[i]); 323 } 324 halving_add_unsigned(sB1, sB2, sBo); 325 for (int i = 0; i < M; i++) { 326 byte e = (byte) (((sB1[i] & 0xff) + (sB2[i] & 0xff)) >> 1); 327 expectEquals(e, sBo[i]); 328 } 329 rounding_halving_add_signed(sB1, sB2, sBo); 330 for (int i = 0; i < M; i++) { 331 byte e = (byte) ((sB1[i] + sB2[i] + 1) >> 1); 332 expectEquals(e, sBo[i]); 333 } 334 rounding_halving_add_unsigned(sB1, sB2, sBo); 335 for (int i = 0; i < M; i++) { 336 byte e = (byte) (((sB1[i] & 0xff) + (sB2[i] & 0xff) + 1) >> 1); 337 expectEquals(e, sBo[i]); 338 } 339 halving_add_signed_constant(sB1, sBo); 340 for (int i = 0; i < M; i++) { 341 byte e = (byte) ((sB1[i] + 0x7f) >> 1); 342 expectEquals(e, sBo[i]); 343 } 344 halving_add_unsigned_constant(sB1, sBo); 345 for (int i = 0; i < M; i++) { 346 byte e = (byte) (((sB1[i] & 0xff) + 0xff) >> 1); 347 expectEquals(e, sBo[i]); 348 } 349 350 System.out.println("HaddAltByte passed"); 351 } 352 expectEquals(int expected, int result)353 private static void expectEquals(int expected, int result) { 354 if (expected != result) { 355 throw new Error("Expected: " + expected + ", found: " + result); 356 } 357 } 358 } 359