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 SimdByte {
21 
22   static byte[] a;
23 
24   //
25   // Arithmetic operations.
26   //
27 
28   /// CHECK-START: void SimdByte.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 SimdByte.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 SimdByte.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 SimdByte.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 SimdByte.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 SimdByte.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 SimdByte.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 SimdByte.div(int) loop_optimization (after)
72   //
73   //  Not supported on any architecture.
74   //
div(int x)75   static void div(int x) {
76     for (int i = 0; i < 128; i++)
77       a[i] /= x;
78   }
79 
80   /// CHECK-START: void SimdByte.neg() loop_optimization (before)
81   /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none
82   /// CHECK-DAG: ArraySet loop:<<Loop>>      outer_loop:none
83   //
84   /// CHECK-START-{ARM,ARM64}: void SimdByte.neg() loop_optimization (after)
85   /// CHECK-DAG: VecLoad  loop:<<Loop:B\d+>> outer_loop:none
86   /// CHECK-DAG: VecNeg   loop:<<Loop>>      outer_loop:none
87   /// CHECK-DAG: VecStore loop:<<Loop>>      outer_loop:none
neg()88   static void neg() {
89     for (int i = 0; i < 128; i++)
90       a[i] = (byte) -a[i];
91   }
92 
93   /// CHECK-START: void SimdByte.not() loop_optimization (before)
94   /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none
95   /// CHECK-DAG: ArraySet loop:<<Loop>>      outer_loop:none
96   //
97   /// CHECK-START-{ARM,ARM64}: void SimdByte.not() loop_optimization (after)
98   /// CHECK-DAG: VecLoad  loop:<<Loop:B\d+>> outer_loop:none
99   /// CHECK-DAG: VecNot   loop:<<Loop>>      outer_loop:none
100   /// CHECK-DAG: VecStore loop:<<Loop>>      outer_loop:none
not()101   static void not() {
102     for (int i = 0; i < 128; i++)
103       a[i] = (byte) ~a[i];
104   }
105 
106   /// CHECK-START: void SimdByte.shl4() loop_optimization (before)
107   /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none
108   /// CHECK-DAG: ArraySet loop:<<Loop>>      outer_loop:none
109   //
110   /// CHECK-START-{ARM,ARM64}: void SimdByte.shl4() loop_optimization (after)
111   /// CHECK-DAG: VecLoad  loop:<<Loop:B\d+>> outer_loop:none
112   /// CHECK-DAG: VecShl   loop:<<Loop>>      outer_loop:none
113   /// CHECK-DAG: VecStore loop:<<Loop>>      outer_loop:none
shl4()114   static void shl4() {
115     for (int i = 0; i < 128; i++)
116       a[i] <<= 4;
117   }
118 
119   /// CHECK-START: void SimdByte.sar2() loop_optimization (before)
120   /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none
121   /// CHECK-DAG: ArraySet loop:<<Loop>>      outer_loop:none
122   //
123   /// CHECK-START-{ARM,ARM64}: void SimdByte.sar2() loop_optimization (after)
124   /// CHECK-DAG: VecLoad  loop:<<Loop:B\d+>> outer_loop:none
125   /// CHECK-DAG: VecShr   loop:<<Loop>>      outer_loop:none
126   /// CHECK-DAG: VecStore loop:<<Loop>>      outer_loop:none
sar2()127   static void sar2() {
128     for (int i = 0; i < 128; i++)
129       a[i] >>= 2;
130   }
131 
132   /// CHECK-START: void SimdByte.shr2() loop_optimization (before)
133   /// CHECK-DAG: ArrayGet loop:<<Loop:B\d+>> outer_loop:none
134   /// CHECK-DAG: ArraySet loop:<<Loop>>      outer_loop:none
135   //
136   // TODO: would need signess flip.
137   /// CHECK-START: void SimdByte.shr2() loop_optimization (after)
138   /// CHECK-NOT: VecUShr
shr2()139   static void shr2() {
140     for (int i = 0; i < 128; i++)
141       a[i] >>>= 2;
142   }
143 
144   //
145   // Shift sanity.
146   //
147 
sar31()148   static void sar31() {
149     for (int i = 0; i < 128; i++)
150       a[i] >>= 31;
151   }
152 
shr31()153   static void shr31() {
154     for (int i = 0; i < 128; i++)
155       a[i] >>>= 31;
156   }
157 
shr32()158   static void shr32() {
159     for (int i = 0; i < 128; i++)
160       a[i] >>>= 32;  // 0, since & 31
161   }
162 
shr33()163   static void shr33() {
164     for (int i = 0; i < 128; i++)
165       a[i] >>>= 33;  // 1, since & 31
166   }
167 
shl9()168   static void shl9() {
169     for (int i = 0; i < 128; i++)
170       a[i] <<= 9;  // yields all-zeros
171   }
172 
173   //
174   // Loop bounds.
175   //
176 
bounds()177   static void bounds() {
178     for (int i = 1; i < 127; i++)
179       a[i] += 11;
180   }
181 
182   //
183   // Test Driver.
184   //
185 
main()186   public static void main() {
187     // Set up.
188     a = new byte[128];
189     for (int i = 0; i < 128; i++) {
190       a[i] = (byte) i;
191     }
192     // Arithmetic operations.
193     add(2);
194     for (int i = 0; i < 128; i++) {
195       expectEquals((byte)(i + 2), a[i], "add");
196     }
197     sub(2);
198     for (int i = 0; i < 128; i++) {
199       expectEquals(i, a[i], "sub");
200     }
201     mul(2);
202     for (int i = 0; i < 128; i++) {
203       expectEquals((byte)(i + i), a[i], "mul");
204     }
205     div(2);
206     for (int i = 0; i < 128; i++) {
207       expectEquals(((byte)(i + i)) >> 1, a[i], "div");
208       a[i] = (byte) i;  // undo arithmetic wrap-around effects
209     }
210     neg();
211     for (int i = 0; i < 128; i++) {
212       expectEquals(-i, a[i], "neg");
213     }
214     // Loop bounds.
215     bounds();
216     expectEquals(0, a[0], "bounds0");
217     for (int i = 1; i < 127; i++) {
218       expectEquals(11 - i, a[i], "bounds");
219     }
220     expectEquals(-127, a[127], "bounds127");
221     // Shifts.
222     for (int i = 0; i < 128; i++) {
223       a[i] = (byte) 0xff;
224     }
225     shl4();
226     for (int i = 0; i < 128; i++) {
227       expectEquals((byte) 0xf0, a[i], "shl4");
228     }
229     sar2();
230     for (int i = 0; i < 128; i++) {
231       expectEquals((byte) 0xfc, a[i], "sar2");
232     }
233     shr2();
234     for (int i = 0; i < 128; i++) {
235       expectEquals((byte) 0xff, a[i], "shr2");  // sic!
236     }
237     sar31();
238     for (int i = 0; i < 128; i++) {
239       expectEquals((byte) 0xff, a[i], "sar31");
240     }
241     shr31();
242     for (int i = 0; i < 128; i++) {
243       expectEquals(0x01, a[i], "shr31");
244       a[i] = (byte) 0x12;  // reset
245     }
246     shr32();
247     for (int i = 0; i < 128; i++) {
248       expectEquals((byte) 0x12, a[i], "shr32");
249     }
250     shr33();
251     for (int i = 0; i < 128; i++) {
252       expectEquals((byte) 0x09, a[i], "shr33");
253     }
254     shl9();
255     for (int i = 0; i < 128; i++) {
256       expectEquals((byte) 0x00, a[i], "shl9");
257       a[i] = (byte) 0xf0;  // reset
258     }
259     not();
260     for (int i = 0; i < 128; i++) {
261       expectEquals((byte) 0x0f, a[i], "not");
262     }
263     // Done.
264     System.out.println("SimdByte passed");
265   }
266 
expectEquals(int expected, int result, String action)267   private static void expectEquals(int expected, int result, String action) {
268     if (expected != result) {
269       throw new Error("Expected: " + expected + ", found: " + result + " for " + action);
270     }
271   }
272 }
273