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 * Functional tests for SIMD vectorization. 19 */ 20 public class SimdShort { 21 22 static short[] a; 23 24 // 25 // Arithmetic operations. 26 // 27 28 /// CHECK-START: void SimdShort.add(int) loop_optimization (before) 29 /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none 30 /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none 31 // 32 /// CHECK-START-{ARM,ARM64}: void SimdShort.add(int) loop_optimization (after) 33 /// CHECK-DAG: VecLoad loop:<<Loop:B\d+>> outer_loop:none 34 /// CHECK-DAG: VecAdd loop:<<Loop>> outer_loop:none 35 /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none add(int x)36 static void add(int x) { 37 for (int i = 0; i < 128; i++) 38 a[i] += x; 39 } 40 41 /// CHECK-START: void SimdShort.sub(int) loop_optimization (before) 42 /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none 43 /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none 44 // 45 /// CHECK-START-{ARM,ARM64}: void SimdShort.sub(int) loop_optimization (after) 46 /// CHECK-DAG: VecLoad loop:<<Loop:B\d+>> outer_loop:none 47 /// CHECK-DAG: VecSub loop:<<Loop>> outer_loop:none 48 /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none sub(int x)49 static void sub(int x) { 50 for (int i = 0; i < 128; i++) 51 a[i] -= x; 52 } 53 54 /// CHECK-START: void SimdShort.mul(int) loop_optimization (before) 55 /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none 56 /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none 57 // 58 /// CHECK-START-{ARM,ARM64}: void SimdShort.mul(int) loop_optimization (after) 59 /// CHECK-DAG: VecLoad loop:<<Loop:B\d+>> outer_loop:none 60 /// CHECK-DAG: VecMul loop:<<Loop>> outer_loop:none 61 /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none mul(int x)62 static void mul(int x) { 63 for (int i = 0; i < 128; i++) 64 a[i] *= x; 65 } 66 67 /// CHECK-START: void SimdShort.div(int) loop_optimization (before) 68 /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none 69 /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none 70 // 71 /// CHECK-START: void SimdShort.div(int) loop_optimization (after) 72 /// CHECK-NOT: VecDiv 73 // 74 // Not supported on any architecture. 75 // div(int x)76 static void div(int x) { 77 for (int i = 0; i < 128; i++) 78 a[i] /= x; 79 } 80 81 /// CHECK-START: void SimdShort.neg() loop_optimization (before) 82 /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none 83 /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none 84 // 85 /// CHECK-START-{ARM,ARM64}: void SimdShort.neg() loop_optimization (after) 86 /// CHECK-DAG: VecLoad loop:<<Loop:B\d+>> outer_loop:none 87 /// CHECK-DAG: VecNeg loop:<<Loop>> outer_loop:none 88 /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none neg()89 static void neg() { 90 for (int i = 0; i < 128; i++) 91 a[i] = (short) -a[i]; 92 } 93 94 /// CHECK-START: void SimdShort.not() loop_optimization (before) 95 /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none 96 /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none 97 // 98 /// CHECK-START-{ARM,ARM64}: void SimdShort.not() loop_optimization (after) 99 /// CHECK-DAG: VecLoad loop:<<Loop:B\d+>> outer_loop:none 100 /// CHECK-DAG: VecNot loop:<<Loop>> outer_loop:none 101 /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none not()102 static void not() { 103 for (int i = 0; i < 128; i++) 104 a[i] = (short) ~a[i]; 105 } 106 107 /// CHECK-START: void SimdShort.shl4() loop_optimization (before) 108 /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none 109 /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none 110 // 111 /// CHECK-START-{ARM,ARM64}: void SimdShort.shl4() loop_optimization (after) 112 /// CHECK-DAG: VecLoad loop:<<Loop:B\d+>> outer_loop:none 113 /// CHECK-DAG: VecShl loop:<<Loop>> outer_loop:none 114 /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none shl4()115 static void shl4() { 116 for (int i = 0; i < 128; i++) 117 a[i] <<= 4; 118 } 119 120 /// CHECK-START: void SimdShort.sar2() loop_optimization (before) 121 /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none 122 /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none 123 // 124 /// CHECK-START-{ARM,ARM64}: void SimdShort.sar2() loop_optimization (after) 125 /// CHECK-DAG: VecLoad loop:<<Loop:B\d+>> outer_loop:none 126 /// CHECK-DAG: VecShr loop:<<Loop>> outer_loop:none 127 /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none sar2()128 static void sar2() { 129 for (int i = 0; i < 128; i++) 130 a[i] >>= 2; 131 } 132 133 /// CHECK-START: void SimdShort.shr2() loop_optimization (before) 134 /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none 135 /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none 136 // 137 // TODO: would need signess flip. 138 /// CHECK-START: void SimdShort.shr2() loop_optimization (after) 139 /// CHECK-NOT: VecUShr shr2()140 static void shr2() { 141 for (int i = 0; i < 128; i++) 142 a[i] >>>= 2; 143 } 144 145 // 146 // Shift sanity. 147 // 148 sar31()149 static void sar31() { 150 for (int i = 0; i < 128; i++) 151 a[i] >>= 31; 152 } 153 shr31()154 static void shr31() { 155 for (int i = 0; i < 128; i++) 156 a[i] >>>= 31; 157 } 158 shr32()159 static void shr32() { 160 for (int i = 0; i < 128; i++) 161 a[i] >>>= 32; // 0, since & 31 162 } 163 164 shr33()165 static void shr33() { 166 for (int i = 0; i < 128; i++) 167 a[i] >>>= 33; // 1, since & 31 168 } 169 170 // 171 // Loop bounds. 172 // 173 add()174 static void add() { 175 for (int i = 1; i < 127; i++) 176 a[i] += 11; 177 } 178 179 // 180 // Test Driver. 181 // 182 main()183 public static void main() { 184 // Set up. 185 a = new short[128]; 186 for (int i = 0; i < 128; i++) { 187 a[i] = (short) i; 188 } 189 // Arithmetic operations. 190 add(2); 191 for (int i = 0; i < 128; i++) { 192 expectEquals(i + 2, a[i], "add"); 193 } 194 sub(2); 195 for (int i = 0; i < 128; i++) { 196 expectEquals(i, a[i], "sub"); 197 } 198 mul(2); 199 for (int i = 0; i < 128; i++) { 200 expectEquals(i + i, a[i], "mul"); 201 } 202 div(2); 203 for (int i = 0; i < 128; i++) { 204 expectEquals(i, a[i], "div"); 205 } 206 neg(); 207 for (int i = 0; i < 128; i++) { 208 expectEquals(-i, a[i], "neg"); 209 } 210 // Loop bounds. 211 add(); 212 expectEquals(0, a[0], "bounds0"); 213 for (int i = 1; i < 127; i++) { 214 expectEquals(11 - i, a[i], "bounds"); 215 } 216 expectEquals(-127, a[127], "bounds127"); 217 // Shifts. 218 for (int i = 0; i < 128; i++) { 219 a[i] = (short) 0xffff; 220 } 221 shl4(); 222 for (int i = 0; i < 128; i++) { 223 expectEquals((short) 0xfff0, a[i], "shl4"); 224 } 225 sar2(); 226 for (int i = 0; i < 128; i++) { 227 expectEquals((short) 0xfffc, a[i], "sar2"); 228 } 229 shr2(); 230 for (int i = 0; i < 128; i++) { 231 expectEquals((short) 0xffff, a[i], "shr2"); // sic! 232 } 233 sar31(); 234 for (int i = 0; i < 128; i++) { 235 expectEquals((short) 0xffff, a[i], "sar31"); 236 } 237 shr31(); 238 for (int i = 0; i < 128; i++) { 239 expectEquals(0x0001, a[i], "shr31"); 240 a[i] = (short) 0x1200; // reset 241 } 242 shr32(); 243 for (int i = 0; i < 128; i++) { 244 expectEquals((short) 0x1200, a[i], "shr32"); 245 } 246 shr33(); 247 for (int i = 0; i < 128; i++) { 248 expectEquals((short) 0x0900, a[i], "shr33"); 249 a[i] = (short) 0xf0f1; // reset 250 } 251 not(); 252 for (int i = 0; i < 128; i++) { 253 expectEquals((short) 0x0f0e, a[i], "not"); 254 } 255 // Done. 256 System.out.println("SimdShort passed"); 257 } 258 expectEquals(int expected, int result, String action)259 private static void expectEquals(int expected, int result, String action) { 260 if (expected != result) { 261 throw new Error("Expected: " + expected + ", found: " + result + " for " + action); 262 } 263 } 264 } 265