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 public class HaddShort { 21 22 private static final int N = 64 * 1024; 23 private static final int M = N + 31; 24 25 static short[] sB1 = new short[M]; 26 static short[] sB2 = new short[M]; 27 static short[] sBo = new short[M]; 28 $inline$mone()29 private static int $inline$mone() { 30 return -1; 31 } 32 33 /// CHECK-START: void HaddShort.halving_add_signed(short[], short[], short[]) loop_optimization (before) 34 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 35 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 36 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 37 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 38 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 39 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 40 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 41 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 42 // 43 /// CHECK-START-{ARM,ARM64}: void HaddShort.halving_add_signed(short[], short[], short[]) loop_optimization (after) 44 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 45 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 46 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none 47 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none halving_add_signed(short[] b1, short[] b2, short[] bo)48 private static void halving_add_signed(short[] b1, short[] b2, short[] bo) { 49 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 50 for (int i = 0; i < min_length; i++) { 51 bo[i] = (short) ((b1[i] + b2[i]) >> 1); 52 } 53 } 54 55 /// CHECK-START: void HaddShort.halving_add_signed_alt(short[], short[], short[]) loop_optimization (before) 56 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 57 /// CHECK-DAG: <<I10:i\d+>> IntConstant 10 loop:none 58 /// CHECK-DAG: <<M10:i\d+>> IntConstant -10 loop:none 59 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 60 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 61 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 62 /// CHECK-DAG: <<Add1:i\d+>> Add [<<Get1>>,<<I10>>] loop:<<Loop>> outer_loop:none 63 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Get2>>,<<M10>>] loop:<<Loop>> outer_loop:none 64 /// CHECK-DAG: <<Add3:i\d+>> Add [<<Add1>>,<<Add2>>] loop:<<Loop>> outer_loop:none 65 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add3>>,<<I1>>] loop:<<Loop>> outer_loop:none 66 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 67 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 68 // 69 /// CHECK-START-{ARM,ARM64}: void HaddShort.halving_add_signed_alt(short[], short[], short[]) loop_optimization (after) 70 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 71 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 72 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none 73 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none halving_add_signed_alt(short[] b1, short[] b2, short[] bo)74 private static void halving_add_signed_alt(short[] b1, short[] b2, short[] bo) { 75 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 76 for (int i = 0; i < min_length; i++) { 77 // Cancelling constant computations do not confuse recognition. 78 bo[i] = (short) (((b1[i] + 10) + (b2[i] - 10)) >> 1); 79 } 80 } 81 82 /// CHECK-START: void HaddShort.halving_add_unsigned(short[], short[], short[]) instruction_simplifier (before) 83 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 84 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 85 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 86 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 87 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 88 /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<UMAX>>] loop:<<Loop>> outer_loop:none 89 /// CHECK-DAG: <<And2:i\d+>> And [<<UMAX>>,<<Get2>>] loop:<<Loop>> outer_loop:none 90 /// CHECK-DAG: <<Add:i\d+>> Add [<<And1>>,<<And2>>] loop:<<Loop>> outer_loop:none 91 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 92 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 93 /// CHECK-DAG: ArraySet [{{l\d+}},{{i\d+}},<<Cnv>>] loop:<<Loop>> outer_loop:none 94 // 95 /// CHECK-START: void HaddShort.halving_add_unsigned(short[], short[], short[]) loop_optimization (before) 96 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 97 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 98 /// CHECK-DAG: <<Get1:c\d+>> ArrayGet loop:<<Loop>> outer_loop:none 99 /// CHECK-DAG: <<Get2:c\d+>> ArrayGet loop:<<Loop>> outer_loop:none 100 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 101 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 102 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 103 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 104 // 105 /// CHECK-START-{ARM,ARM64}: void HaddShort.halving_add_unsigned(short[], short[], short[]) loop_optimization (after) 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 halving_add_unsigned(short[] b1, short[] b2, short[] bo)110 private static void halving_add_unsigned(short[] b1, short[] b2, short[] bo) { 111 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 112 for (int i = 0; i < min_length; i++) { 113 int v1 = b1[i] & 0xffff; 114 int v2 = b2[i] & 0xffff; 115 bo[i] = (short) ((v1 + v2) >> 1); 116 } 117 } 118 119 /// CHECK-START: void HaddShort.rounding_halving_add_signed(short[], short[], short[]) 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:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 123 /// CHECK-DAG: <<Get2:s\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: <<Shr:i\d+>> Shr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 127 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 128 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 129 // 130 /// CHECK-START-{ARM,ARM64}: void HaddShort.rounding_halving_add_signed(short[], short[], short[]) 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:Int16 rounded:true loop:<<Loop>> outer_loop:none 134 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none rounding_halving_add_signed(short[] b1, short[] b2, short[] bo)135 private static void rounding_halving_add_signed(short[] b1, short[] b2, short[] bo) { 136 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 137 for (int i = 0; i < min_length; i++) { 138 bo[i] = (short) ((b1[i] + b2[i] + 1) >> 1); 139 } 140 } 141 142 /// CHECK-START: void HaddShort.rounding_halving_add_signed_alt(short[], short[], short[]) loop_optimization (before) 143 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 144 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 145 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 146 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 147 /// CHECK-DAG: <<Add1:i\d+>> Add [<<Get1>>,<<I1>>] loop:<<Loop>> outer_loop:none 148 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 149 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 150 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 151 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 152 // 153 /// CHECK-START-{ARM,ARM64}: void HaddShort.rounding_halving_add_signed_alt(short[], short[], short[]) loop_optimization (after) 154 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 155 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 156 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none 157 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none rounding_halving_add_signed_alt(short[] b1, short[] b2, short[] bo)158 private static void rounding_halving_add_signed_alt(short[] b1, short[] b2, short[] bo) { 159 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 160 for (int i = 0; i < min_length; i++) { 161 // Slightly different order in idiom does not confuse recognition. 162 bo[i] = (short) (((1 + b1[i]) + b2[i]) >> 1); 163 } 164 } 165 166 /// CHECK-START: void HaddShort.rounding_halving_add_signed_alt2(short[], short[], short[]) loop_optimization (before) 167 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 168 /// CHECK-DAG: <<I10:i\d+>> IntConstant 10 loop:none 169 /// CHECK-DAG: <<M9:i\d+>> IntConstant -9 loop:none 170 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 171 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 172 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 173 /// CHECK-DAG: <<Add1:i\d+>> Add [<<Get1>>,<<I10>>] loop:<<Loop>> outer_loop:none 174 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Get2>>,<<M9>>] loop:<<Loop>> outer_loop:none 175 /// CHECK-DAG: <<Add3:i\d+>> Add [<<Add1>>,<<Add2>>] loop:<<Loop>> outer_loop:none 176 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add3>>,<<I1>>] loop:<<Loop>> outer_loop:none 177 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 178 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 179 // 180 /// CHECK-START-{ARM,ARM64}: void HaddShort.rounding_halving_add_signed_alt2(short[], short[], short[]) loop_optimization (after) 181 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 182 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 183 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none 184 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none rounding_halving_add_signed_alt2(short[] b1, short[] b2, short[] bo)185 private static void rounding_halving_add_signed_alt2(short[] b1, short[] b2, short[] bo) { 186 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 187 for (int i = 0; i < min_length; i++) { 188 // Computations that cancel to adding 1 also do not confuse recognition. 189 bo[i] = (short) (((b1[i] + 10) + (b2[i] - 9)) >> 1); 190 } 191 } 192 193 /// CHECK-START: void HaddShort.rounding_halving_add_signed_alt3(short[], short[], short[]) loop_optimization (before) 194 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 195 /// CHECK-DAG: <<M1:i\d+>> IntConstant -1 loop:none 196 /// CHECK-DAG: <<I9:i\d+>> IntConstant 9 loop:none 197 /// CHECK-DAG: <<M9:i\d+>> IntConstant -9 loop:none 198 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 199 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 200 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 201 /// CHECK-DAG: <<Add1:i\d+>> Add [<<Get1>>,<<I9>>] loop:<<Loop>> outer_loop:none 202 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Get2>>,<<M9>>] loop:<<Loop>> outer_loop:none 203 /// CHECK-DAG: <<Add3:i\d+>> Add [<<Add1>>,<<Add2>>] loop:<<Loop>> outer_loop:none 204 /// CHECK-DAG: <<Sub:i\d+>> Sub [<<Add3>>,<<M1>>] loop:<<Loop>> outer_loop:none 205 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Sub>>,<<I1>>] loop:<<Loop>> outer_loop:none 206 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 207 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 208 // 209 /// CHECK-START-{ARM,ARM64}: void HaddShort.rounding_halving_add_signed_alt3(short[], short[], short[]) loop_optimization (after) 210 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 211 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 212 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none 213 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none rounding_halving_add_signed_alt3(short[] b1, short[] b2, short[] bo)214 private static void rounding_halving_add_signed_alt3(short[] b1, short[] b2, short[] bo) { 215 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 216 for (int i = 0; i < min_length; i++) { 217 // Computations that cancel to adding 1 also do not confuse recognition. 218 bo[i] = (short) (((b1[i] + 9) + (b2[i] - 9) - $inline$mone()) >> 1); 219 } 220 } 221 222 /// CHECK-START: void HaddShort.rounding_halving_add_unsigned(short[], short[], short[]) instruction_simplifier (before) 223 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 224 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 225 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 226 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 227 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 228 /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<UMAX>>] loop:<<Loop>> outer_loop:none 229 /// CHECK-DAG: <<And2:i\d+>> And [<<UMAX>>,<<Get2>>] loop:<<Loop>> outer_loop:none 230 /// CHECK-DAG: <<Add1:i\d+>> Add [<<And1>>,<<And2>>] loop:<<Loop>> outer_loop:none 231 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<I1>>] loop:<<Loop>> outer_loop:none 232 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 233 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 234 /// CHECK-DAG: ArraySet [{{l\d+}},{{i\d+}},<<Cnv>>] loop:<<Loop>> outer_loop:none 235 // 236 /// CHECK-START: void HaddShort.rounding_halving_add_unsigned(short[], short[], short[]) loop_optimization (before) 237 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 238 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 239 /// CHECK-DAG: <<Get1:c\d+>> ArrayGet loop:<<Loop>> outer_loop:none 240 /// CHECK-DAG: <<Get2:c\d+>> ArrayGet loop:<<Loop>> outer_loop:none 241 /// CHECK-DAG: <<Add1:i\d+>> Add [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none 242 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<I1>>] loop:<<Loop>> outer_loop:none 243 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 244 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 245 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 246 // 247 /// CHECK-START-{ARM,ARM64}: void HaddShort.rounding_halving_add_unsigned(short[], short[], short[]) loop_optimization (after) 248 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 249 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 250 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none 251 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none rounding_halving_add_unsigned(short[] b1, short[] b2, short[] bo)252 private static void rounding_halving_add_unsigned(short[] b1, short[] b2, short[] bo) { 253 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 254 for (int i = 0; i < min_length; i++) { 255 int v1 = b1[i] & 0xffff; 256 int v2 = b2[i] & 0xffff; 257 bo[i] = (short) ((v1 + v2 + 1) >> 1); 258 } 259 } 260 261 /// CHECK-START: void HaddShort.rounding_halving_add_unsigned_alt(short[], short[], short[]) instruction_simplifier (before) 262 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 263 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 264 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 265 /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 266 /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 267 /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<UMAX>>] loop:<<Loop>> outer_loop:none 268 /// CHECK-DAG: <<And2:i\d+>> And [<<UMAX>>,<<Get2>>] loop:<<Loop>> outer_loop:none 269 /// CHECK-DAG: <<Add1:i\d+>> Add [<<And2>>,<<I1>>] loop:<<Loop>> outer_loop:none 270 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Add1>>,<<And1>>] loop:<<Loop>> outer_loop:none 271 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 272 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 273 /// CHECK-DAG: ArraySet [{{l\d+}},{{i\d+}},<<Cnv>>] loop:<<Loop>> outer_loop:none 274 // 275 /// CHECK-START: void HaddShort.rounding_halving_add_unsigned_alt(short[], short[], short[]) loop_optimization (before) 276 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 277 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 278 /// CHECK-DAG: <<Get1:c\d+>> ArrayGet loop:<<Loop>> outer_loop:none 279 /// CHECK-DAG: <<Get2:c\d+>> ArrayGet loop:<<Loop>> outer_loop:none 280 /// CHECK-DAG: <<Add1:i\d+>> Add [<<Get2>>,<<I1>>] loop:<<Loop>> outer_loop:none 281 /// CHECK-DAG: <<Add2:i\d+>> Add [<<Get1>>,<<Add1>>] loop:<<Loop>> outer_loop:none 282 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add2>>,<<I1>>] loop:<<Loop>> outer_loop:none 283 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 284 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 285 // 286 /// CHECK-START-{ARM,ARM64}: void HaddShort.rounding_halving_add_unsigned_alt(short[], short[], short[]) loop_optimization (after) 287 /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 288 /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none 289 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none 290 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none rounding_halving_add_unsigned_alt(short[] b1, short[] b2, short[] bo)291 private static void rounding_halving_add_unsigned_alt(short[] b1, short[] b2, short[] bo) { 292 int min_length = Math.min(bo.length, Math.min(b1.length, b2.length)); 293 for (int i = 0; i < min_length; i++) { 294 // Slightly different order in idiom does not confuse recognition. 295 int v1 = b1[i] & 0xffff; 296 int v2 = b2[i] & 0xffff; 297 bo[i] = (short) (v1 + (v2 + 1) >> 1); 298 } 299 } 300 301 /// CHECK-START: void HaddShort.halving_add_signed_constant(short[], short[]) loop_optimization (before) 302 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 303 /// CHECK-DAG: <<SMAX:i\d+>> IntConstant 32767 loop:none 304 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 305 /// CHECK-DAG: <<Get:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 306 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get>>,<<SMAX>>] loop:<<Loop>> outer_loop:none 307 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 308 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 309 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 310 // 311 /// CHECK-START-{ARM,ARM64}: void HaddShort.halving_add_signed_constant(short[], short[]) loop_optimization (after) 312 /// CHECK-DAG: <<SMAX:i\d+>> IntConstant 32767 loop:none 313 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<SMAX>>] loop:none 314 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 315 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none 316 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none halving_add_signed_constant(short[] b1, short[] bo)317 private static void halving_add_signed_constant(short[] b1, short[] bo) { 318 int min_length = Math.min(bo.length, b1.length); 319 for (int i = 0; i < min_length; i++) { 320 bo[i] = (short) ((b1[i] + 0x7fff) >> 1); 321 } 322 } 323 324 /// CHECK-START: void HaddShort.halving_add_unsigned_constant(short[], short[]) instruction_simplifier (before) 325 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 326 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 327 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 328 /// CHECK-DAG: <<Get:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none 329 /// CHECK-DAG: <<And:i\d+>> And [<<Get>>,<<UMAX>>] loop:<<Loop>> outer_loop:none 330 /// CHECK-DAG: <<Add:i\d+>> Add [<<And>>,<<UMAX>>] loop:<<Loop>> outer_loop:none 331 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 332 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 333 /// CHECK-DAG: ArraySet [{{l\d+}},{{i\d+}},<<Cnv>>] loop:<<Loop>> outer_loop:none 334 // 335 /// CHECK-START: void HaddShort.halving_add_unsigned_constant(short[], short[]) loop_optimization (before) 336 /// CHECK-DAG: <<I1:i\d+>> IntConstant 1 loop:none 337 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 338 /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none 339 /// CHECK-DAG: <<Get:c\d+>> ArrayGet loop:<<Loop>> outer_loop:none 340 /// CHECK-DAG: <<Add:i\d+>> Add [<<Get>>,<<UMAX>>] loop:<<Loop>> outer_loop:none 341 /// CHECK-DAG: <<Shr:i\d+>> Shr [<<Add>>,<<I1>>] loop:<<Loop>> outer_loop:none 342 /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Shr>>] loop:<<Loop>> outer_loop:none 343 /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none 344 // 345 /// CHECK-START-{ARM,ARM64}: void HaddShort.halving_add_unsigned_constant(short[], short[]) loop_optimization (after) 346 /// CHECK-DAG: <<UMAX:i\d+>> IntConstant 65535 loop:none 347 /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>] loop:none 348 /// CHECK-DAG: <<Get:d\d+>> VecLoad loop:<<Loop:B\d+>> outer_loop:none 349 /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none 350 /// CHECK-DAG: VecStore [{{l\d+}},{{i\d+}},<<HAdd>>] loop:<<Loop>> outer_loop:none halving_add_unsigned_constant(short[] b1, short[] bo)351 private static void halving_add_unsigned_constant(short[] b1, short[] bo) { 352 int min_length = Math.min(bo.length, b1.length); 353 for (int i = 0; i < min_length; i++) { 354 bo[i] = (short) (((b1[i] & 0xffff) + 0xffff) >> 1); 355 } 356 } 357 main()358 public static void main() { 359 // Some interesting values. 360 short[] interesting = { 361 (short) 0x0000, 362 (short) 0x0001, 363 (short) 0x0002, 364 (short) 0x1234, 365 (short) 0x8000, 366 (short) 0x8001, 367 (short) 0x7fff, 368 (short) 0xffff 369 }; 370 // Initialize cross-values to test all cases, and also 371 // set up some extra values to exercise the cleanup loop. 372 for (int i = 0; i < M; i++) { 373 sB1[i] = (short) i; 374 sB2[i] = interesting[i & 7]; 375 } 376 377 // Test halving add idioms. 378 halving_add_signed(sB1, sB2, sBo); 379 for (int i = 0; i < M; i++) { 380 short e = (short) ((sB1[i] + sB2[i]) >> 1); 381 expectEquals(e, sBo[i]); 382 } 383 halving_add_signed_alt(sB1, sB2, sBo); 384 for (int i = 0; i < M; i++) { 385 short e = (short) ((sB1[i] + sB2[i]) >> 1); 386 expectEquals(e, sBo[i]); 387 } 388 halving_add_unsigned(sB1, sB2, sBo); 389 for (int i = 0; i < M; i++) { 390 short e = (short) (((sB1[i] & 0xffff) + (sB2[i] & 0xffff)) >> 1); 391 expectEquals(e, sBo[i]); 392 } 393 rounding_halving_add_signed(sB1, sB2, sBo); 394 for (int i = 0; i < M; i++) { 395 short e = (short) ((sB1[i] + sB2[i] + 1) >> 1); 396 expectEquals(e, sBo[i]); 397 } 398 rounding_halving_add_signed_alt(sB1, sB2, sBo); 399 for (int i = 0; i < M; i++) { 400 short e = (short) ((sB1[i] + sB2[i] + 1) >> 1); 401 expectEquals(e, sBo[i]); 402 } 403 rounding_halving_add_signed_alt2(sB1, sB2, sBo); 404 for (int i = 0; i < M; i++) { 405 short e = (short) ((sB1[i] + sB2[i] + 1) >> 1); 406 expectEquals(e, sBo[i]); 407 } 408 rounding_halving_add_signed_alt3(sB1, sB2, sBo); 409 for (int i = 0; i < M; i++) { 410 short e = (short) ((sB1[i] + sB2[i] + 1) >> 1); 411 expectEquals(e, sBo[i]); 412 } 413 rounding_halving_add_unsigned(sB1, sB2, sBo); 414 for (int i = 0; i < M; i++) { 415 short e = (short) (((sB1[i] & 0xffff) + (sB2[i] & 0xffff) + 1) >> 1); 416 expectEquals(e, sBo[i]); 417 } 418 rounding_halving_add_unsigned_alt(sB1, sB2, sBo); 419 for (int i = 0; i < M; i++) { 420 short e = (short) (((sB1[i] & 0xffff) + (sB2[i] & 0xffff) + 1) >> 1); 421 expectEquals(e, sBo[i]); 422 } 423 halving_add_signed_constant(sB1, sBo); 424 for (int i = 0; i < M; i++) { 425 short e = (short) ((sB1[i] + 0x7fff) >> 1); 426 expectEquals(e, sBo[i]); 427 } 428 halving_add_unsigned_constant(sB1, sBo); 429 for (int i = 0; i < M; i++) { 430 short e = (short) (((sB1[i] & 0xffff) + 0xffff) >> 1); 431 expectEquals(e, sBo[i]); 432 } 433 434 System.out.println("HaddShort passed"); 435 } 436 expectEquals(int expected, int result)437 private static void expectEquals(int expected, int result) { 438 if (expected != result) { 439 throw new Error("Expected: " + expected + ", found: " + result); 440 } 441 } 442 } 443