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 HaddAltShort { 24 25 private static final int N = 64 * 1024; 26 private static final int M = N + 31; 27 28 static short[] sB1 = new short[M]; 29 static short[] sB2 = new short[M]; 30 static short[] sBo = new short[M]; 31 32 /// CHECK-START: void HaddAltShort.halving_add_signed(short[], short[], short[]) 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:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 36 /// CHECK-DAG: <<Get2:s\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:s\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 HaddAltShort.halving_add_signed(short[], short[], short[]) 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:Int16 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 HaddAltShort.halving_add_signed(short[], short[], short[]) 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:Int16 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(short[] b1, short[] b2, short[] bo)62 private static void halving_add_signed(short[] b1, short[] b2, short[] 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] = (short) ((b1[i] + b2[i]) >>> 1); 66 } 67 } 68 69 /// CHECK-START: void HaddAltShort.halving_add_unsigned(short[], short[], short[]) instruction_simplifier (before) 70 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 71 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 72 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 73 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 74 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 75 /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<UMAX>>] loop:<<Loop>> outer_loop:none 76 /// CHECK-DAG: <<And2:i\d+>> And [<<UMAX>>,<<Get2>>] 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:s\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 HaddAltShort.halving_add_unsigned(short[], short[], short[]) 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:c\d+>> ArrayGet loop:<<Loop>> outer_loop:none 86 /// CHECK-DAG: <<Get2:c\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:s\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 HaddAltShort.halving_add_unsigned(short[], short[], short[]) 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:Uint16 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 HaddAltShort.halving_add_unsigned(short[], short[], short[]) 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:Uint16 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(short[] b1, short[] b2, short[] bo)112 private static void halving_add_unsigned(short[] b1, short[] b2, short[] 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 int v1 = b1[i] & 0xffff; 116 int v2 = b2[i] & 0xffff; 117 bo[i] = (short) ((v1 + v2) >>> 1); 118 } 119 } 120 121 /// CHECK-START: void HaddAltShort.rounding_halving_add_signed(short[], short[], short[]) loop_optimization (before) 122 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 123 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 124 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 125 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 126 /// CHECK-DAG: <<Add1:i\d+>> Add [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 127 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<I1>>] loop:<<Loop>> outer_loop:none 128 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 129 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 130 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 131 // 132 /// CHECK-START-ARM: void HaddAltShort.rounding_halving_add_signed(short[], short[], short[]) loop_optimization (after) 133 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 134 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 135 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none 136 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 137 // 138 /// CHECK-START-ARM64: void HaddAltShort.rounding_halving_add_signed(short[], short[], short[]) loop_optimization (after) 139 /// CHECK-IF: hasIsaFeature("sve") 140 // 141 // HalvingAdd idiom is not supported for SVE. 142 /// CHECK-NOT: VecHalvingAdd 143 // 144 /// CHECK-ELSE: 145 // 146 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 147 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 148 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none 149 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 150 // 151 /// CHECK-FI: rounding_halving_add_signed(short[] b1, short[] b2, short[] bo)152 private static void rounding_halving_add_signed(short[] b1, short[] b2, short[] bo) { 153 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 154 for (int i = 0; i < min_length; i++) { 155 bo[i] = (short) ((b1[i] + b2[i] + 1) >>> 1); 156 } 157 } 158 159 /// CHECK-START: void HaddAltShort.rounding_halving_add_unsigned(short[], short[], short[]) instruction_simplifier (before) 160 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 161 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 162 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 163 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 164 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 165 /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<UMAX>>] loop:<<Loop>> outer_loop:none 166 /// CHECK-DAG: <<And2:i\d+>> And [<<UMAX>>,<<Get2>>] loop:<<Loop>> outer_loop:none 167 /// CHECK-DAG: <<Add1:i\d+>> Add [<<And1>>,<<And2>>] loop:<<Loop>> outer_loop:none 168 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<I1>>] loop:<<Loop>> outer_loop:none 169 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 170 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 171 /// CHECK-DAG: ArraySet [{{l\d+}},{{i\d+}},<<Cnv>>] loop:<<Loop>> outer_loop:none 172 // 173 /// CHECK-START: void HaddAltShort.rounding_halving_add_unsigned(short[], short[], short[]) loop_optimization (before) 174 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 175 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 176 /// CHECK-DAG: <<Get1:c\d+>> ArrayGet loop:<<Loop>> outer_loop:none 177 /// CHECK-DAG: <<Get2:c\d+>> ArrayGet loop:<<Loop>> outer_loop:none 178 /// CHECK-DAG: <<Add1:i\d+>> Add [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 179 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<I1>>] loop:<<Loop>> outer_loop:none 180 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 181 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 182 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 183 // 184 /// CHECK-START-ARM: void HaddAltShort.rounding_halving_add_unsigned(short[], short[], short[]) loop_optimization (after) 185 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 186 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 187 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none 188 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 189 // 190 /// CHECK-START-ARM64: void HaddAltShort.rounding_halving_add_unsigned(short[], short[], short[]) loop_optimization (after) 191 /// CHECK-IF: hasIsaFeature("sve") 192 // 193 // HalvingAdd idiom is not supported for SVE. 194 /// CHECK-NOT: VecHalvingAdd 195 // 196 /// CHECK-ELSE: 197 // 198 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 199 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 200 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none 201 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 202 // 203 /// CHECK-FI: rounding_halving_add_unsigned(short[] b1, short[] b2, short[] bo)204 private static void rounding_halving_add_unsigned(short[] b1, short[] b2, short[] bo) { 205 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 206 for (int i = 0; i < min_length; i++) { 207 int v1 = b1[i] & 0xffff; 208 int v2 = b2[i] & 0xffff; 209 bo[i] = (short) ((v1 + v2 + 1) >>> 1); 210 } 211 } 212 213 /// CHECK-START: void HaddAltShort.halving_add_signed_constant(short[], short[]) loop_optimization (before) 214 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 215 /// CHECK-DAG: <<SMAX:i\d+>> IntConstant 32767 loop:none 216 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 217 /// CHECK-DAG: <<Get:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 218 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get>>,<<SMAX>>] loop:<<Loop>> outer_loop:none 219 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 220 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 221 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 222 // 223 /// CHECK-START-ARM: void HaddAltShort.halving_add_signed_constant(short[], short[]) loop_optimization (after) 224 /// CHECK-DAG: <<SMAX:i\d+>> IntConstant 32767 loop:none 225 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<SMAX>>] loop:none 226 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 227 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none 228 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 229 // 230 /// CHECK-START-ARM64: void HaddAltShort.halving_add_signed_constant(short[], short[]) loop_optimization (after) 231 /// CHECK-IF: hasIsaFeature("sve") 232 // 233 // HalvingAdd idiom is not supported for SVE. 234 /// CHECK-NOT: VecHalvingAdd 235 // 236 /// CHECK-ELSE: 237 // 238 /// CHECK-DAG: <<SMAX:i\d+>> IntConstant 32767 loop:none 239 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<SMAX>>] loop:none 240 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 241 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none 242 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 243 // 244 /// CHECK-FI: halving_add_signed_constant(short[] b1, short[] bo)245 private static void halving_add_signed_constant(short[] b1, short[] bo) { 246 int min_length = Math.min(bo.length, b1.length); 247 for (int i = 0; i < min_length; i++) { 248 bo[i] = (short) ((b1[i] + 0x7fff) >>> 1); 249 } 250 } 251 252 /// CHECK-START: void HaddAltShort.halving_add_unsigned_constant(short[], short[]) instruction_simplifier (before) 253 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 254 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 255 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 256 /// CHECK-DAG: <<Get:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 257 /// CHECK-DAG: <<And:i\d+>> And [<<Get>>,<<UMAX>>] loop:<<Loop>> outer_loop:none 258 /// CHECK-DAG: <<Add:i\d+>> Add [<<And>>,<<UMAX>>] loop:<<Loop>> outer_loop:none 259 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 260 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 261 /// CHECK-DAG: ArraySet [{{l\d+}},{{i\d+}},<<Cnv>>] loop:<<Loop>> outer_loop:none 262 // 263 /// CHECK-START: void HaddAltShort.halving_add_unsigned_constant(short[], short[]) loop_optimization (before) 264 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 265 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 266 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 267 /// CHECK-DAG: <<Get:c\d+>> ArrayGet loop:<<Loop>> outer_loop:none 268 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get>>,<<UMAX>>] loop:<<Loop>> outer_loop:none 269 /// CHECK-DAG: <<UShr:i\d+>> UShr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 270 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<UShr>>] loop:<<Loop>> outer_loop:none 271 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 272 // 273 /// CHECK-START-ARM: void HaddAltShort.halving_add_unsigned_constant(short[], short[]) loop_optimization (after) 274 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 275 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>] loop:none 276 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 277 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none 278 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 279 // 280 /// CHECK-START-ARM64: void HaddAltShort.halving_add_unsigned_constant(short[], short[]) loop_optimization (after) 281 /// CHECK-IF: hasIsaFeature("sve") 282 // 283 // HalvingAdd idiom is not supported for SVE. 284 /// CHECK-NOT: VecHalvingAdd 285 // 286 /// CHECK-ELSE: 287 // 288 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 289 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>] loop:none 290 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 291 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none 292 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none 293 // 294 /// CHECK-FI: halving_add_unsigned_constant(short[] b1, short[] bo)295 private static void halving_add_unsigned_constant(short[] b1, short[] bo) { 296 int min_length = Math.min(bo.length, b1.length); 297 for (int i = 0; i < min_length; i++) { 298 bo[i] = (short) (((b1[i] & 0xffff) + 0xffff) >>> 1); 299 } 300 } 301 main()302 public static void main() { 303 // Some interesting values. 304 short[] interesting = { 305 (short) 0x0000, 306 (short) 0x0001, 307 (short) 0x0002, 308 (short) 0x1234, 309 (short) 0x8000, 310 (short) 0x8001, 311 (short) 0x7fff, 312 (short) 0xffff 313 }; 314 // Initialize cross-values to test all cases, and also 315 // set up some extra values to exercise the cleanup loop. 316 for (int i = 0; i < M; i++) { 317 sB1[i] = (short) i; 318 sB2[i] = interesting[i & 7]; 319 } 320 321 // Test halving add idioms. Note that the expected result is computed 322 // with the arithmetic >> to demonstrate the computed narrower result 323 // does not depend on the wider >> or >>>. 324 halving_add_signed(sB1, sB2, sBo); 325 for (int i = 0; i < M; i++) { 326 short e = (short) ((sB1[i] + sB2[i]) >> 1); 327 expectEquals(e, sBo[i]); 328 } 329 halving_add_unsigned(sB1, sB2, sBo); 330 for (int i = 0; i < M; i++) { 331 short e = (short) (((sB1[i] & 0xffff) + (sB2[i] & 0xffff)) >> 1); 332 expectEquals(e, sBo[i]); 333 } 334 rounding_halving_add_signed(sB1, sB2, sBo); 335 for (int i = 0; i < M; i++) { 336 short e = (short) ((sB1[i] + sB2[i] + 1) >> 1); 337 expectEquals(e, sBo[i]); 338 } 339 rounding_halving_add_unsigned(sB1, sB2, sBo); 340 for (int i = 0; i < M; i++) { 341 short e = (short) (((sB1[i] & 0xffff) + (sB2[i] & 0xffff) + 1) >> 1); 342 expectEquals(e, sBo[i]); 343 } 344 halving_add_signed_constant(sB1, sBo); 345 for (int i = 0; i < M; i++) { 346 short e = (short) ((sB1[i] + 0x7fff) >> 1); 347 expectEquals(e, sBo[i]); 348 } 349 halving_add_unsigned_constant(sB1, sBo); 350 for (int i = 0; i < M; i++) { 351 short e = (short) (((sB1[i] & 0xffff) + 0xffff) >> 1); 352 expectEquals(e, sBo[i]); 353 } 354 355 System.out.println("HaddAltShort passed"); 356 } 357 expectEquals(int expected, int result)358 private static void expectEquals(int expected, int result) { 359 if (expected != result) { 360 throw new Error("Expected: " + expected + ", found: " + result); 361 } 362 } 363 } 364