1 //===- subzero/crosstest/test_sync_atomic_main.cpp - Driver for tests -----===//
2 //
3 // The Subzero Code Generator
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // Driver for cross testing atomic intrinsics, via the sync builtins.
11 //
12 //===----------------------------------------------------------------------===//
13
14 /* crosstest.py --test=test_sync_atomic.cpp --crosstest-bitcode=0 \
15 --driver=test_sync_atomic_main.cpp --prefix=Subzero_ \
16 --output=test_sync_atomic */
17
18 #include <pthread.h>
19 #include <stdint.h>
20
21 #include <cerrno>
22 #include <climits>
23 #include <cstdlib>
24 #include <cstring>
25 #include <iostream>
26
27 // Include test_sync_atomic.h twice - once normally, and once within the
28 // Subzero_ namespace, corresponding to the llc and Subzero translated
29 // object files, respectively.
30 #include "test_sync_atomic.h"
31 #include "xdefs.h"
32 namespace Subzero_ {
33 #include "test_sync_atomic.h"
34 }
35
36 volatile uint64 Values[] = {
37 0, 1, 0x7e, 0x7f, 0x80, 0x81, 0xfe, 0xff, 0x7ffe, 0x7fff, 0x8000, 0x8001,
38 0xfffe, 0xffff, 0x007fffff /*Max subnormal + */, 0x00800000 /*Min+ */,
39 0x7f7fffff /*Max+ */, 0x7f800000 /*+Inf*/, 0xff800000 /*-Inf*/,
40 0x7fa00000 /*SNaN*/, 0x7fc00000 /*QNaN*/, 0x7ffffffe, 0x7fffffff,
41 0x80000000, 0x80000001, 0xfffffffe, 0xffffffff, 0x100000000ll,
42 0x100000001ll, 0x000fffffffffffffll /*Max subnormal + */,
43 0x0010000000000000ll /*Min+ */, 0x7fefffffffffffffll /*Max+ */,
44 0x7ff0000000000000ll /*+Inf*/, 0xfff0000000000000ll /*-Inf*/,
45 0x7ff0000000000001ll /*SNaN*/, 0x7ff8000000000000ll /*QNaN*/,
46 0x7ffffffffffffffell, 0x7fffffffffffffffll, 0x8000000000000000ll,
47 0x8000000000000001ll, 0xfffffffffffffffell, 0xffffffffffffffffll};
48
49 const static size_t NumValues = sizeof(Values) / sizeof(*Values);
50
51 struct {
52 volatile uint8_t l8;
53 volatile uint16_t l16;
54 volatile uint32_t l32;
55 volatile uint64 l64;
56 } AtomicLocs;
57
58 template <typename Type>
testAtomicRMW(volatile Type * AtomicLoc,size_t & TotalTests,size_t & Passes,size_t & Failures)59 void testAtomicRMW(volatile Type *AtomicLoc, size_t &TotalTests, size_t &Passes,
60 size_t &Failures) {
61 typedef Type (*FuncType)(bool, volatile Type *, Type);
62 static struct {
63 const char *Name;
64 FuncType FuncLlc;
65 FuncType FuncSz;
66 } Funcs[] = {
67 #define X(inst) \
68 { STR(inst), test_##inst, Subzero_::test_##inst } \
69 , {STR(inst) "_alloca", test_alloca_##inst, Subzero_::test_alloca_##inst}, \
70 {STR(inst) "_const", test_const_##inst, Subzero_::test_const_##inst},
71 RMWOP_TABLE
72 #undef X
73 };
74 const static size_t NumFuncs = sizeof(Funcs) / sizeof(*Funcs);
75
76 for (size_t f = 0; f < NumFuncs; ++f) {
77 for (size_t i = 0; i < NumValues; ++i) {
78 Type Value1 = static_cast<Type>(Values[i]);
79 for (size_t j = 0; j < NumValues; ++j) {
80 Type Value2 = static_cast<Type>(Values[j]);
81 for (size_t k = 0; k < 2; ++k) {
82 bool fetch_first = k;
83 ++TotalTests;
84 *AtomicLoc = Value1;
85 Type ResultSz1 = Funcs[f].FuncSz(fetch_first, AtomicLoc, Value2);
86 Type ResultSz2 = *AtomicLoc;
87 *AtomicLoc = Value1;
88 Type ResultLlc1 = Funcs[f].FuncLlc(fetch_first, AtomicLoc, Value2);
89 Type ResultLlc2 = *AtomicLoc;
90 if (ResultSz1 == ResultLlc1 && ResultSz2 == ResultLlc2) {
91 ++Passes;
92 } else {
93 ++Failures;
94 std::cout << "test_" << Funcs[f].Name << (CHAR_BIT * sizeof(Type))
95 << "(" << fetch_first << ", "
96 << static_cast<uint64>(Value1) << ", "
97 << static_cast<uint64>(Value2)
98 << "): sz1=" << static_cast<uint64>(ResultSz1)
99 << " llc1=" << static_cast<uint64>(ResultLlc1)
100 << " sz2=" << static_cast<uint64>(ResultSz2)
101 << " llc2=" << static_cast<uint64>(ResultLlc2) << "\n";
102 }
103 }
104 }
105 }
106 }
107 }
108
109 template <typename Type>
testValCompareAndSwap(volatile Type * AtomicLoc,size_t & TotalTests,size_t & Passes,size_t & Failures)110 void testValCompareAndSwap(volatile Type *AtomicLoc, size_t &TotalTests,
111 size_t &Passes, size_t &Failures) {
112 typedef Type (*FuncType)(volatile Type *, Type, Type);
113 static struct {
114 const char *Name;
115 FuncType FuncLlc;
116 FuncType FuncSz;
117 } Funcs[] = {{"val_cmp_swap", test_val_cmp_swap, Subzero_::test_val_cmp_swap},
118 {"val_cmp_swap_loop", test_val_cmp_swap_loop,
119 Subzero_::test_val_cmp_swap_loop}};
120 const static size_t NumFuncs = sizeof(Funcs) / sizeof(*Funcs);
121 for (size_t f = 0; f < NumFuncs; ++f) {
122 for (size_t i = 0; i < NumValues; ++i) {
123 Type Value1 = static_cast<Type>(Values[i]);
124 for (size_t j = 0; j < NumValues; ++j) {
125 Type Value2 = static_cast<Type>(Values[j]);
126 for (size_t f = 0; f < 2; ++f) {
127 bool flip = f;
128 ++TotalTests;
129 *AtomicLoc = Value1;
130 Type ResultSz1 =
131 Funcs[f].FuncSz(AtomicLoc, flip ? Value2 : Value1, Value2);
132 Type ResultSz2 = *AtomicLoc;
133 *AtomicLoc = Value1;
134 Type ResultLlc1 =
135 Funcs[f].FuncLlc(AtomicLoc, flip ? Value2 : Value1, Value2);
136 Type ResultLlc2 = *AtomicLoc;
137 if (ResultSz1 == ResultLlc1 && ResultSz2 == ResultLlc2) {
138 ++Passes;
139 } else {
140 ++Failures;
141 std::cout << "test_" << Funcs[f].Name << (CHAR_BIT * sizeof(Type))
142 << "(" << static_cast<uint64>(Value1) << ", "
143 << static_cast<uint64>(Value2) << ", flip=" << flip
144 << "): sz1=" << static_cast<uint64>(ResultSz1)
145 << " llc1=" << static_cast<uint64>(ResultLlc1)
146 << " sz2=" << static_cast<uint64>(ResultSz2)
147 << " llc2=" << static_cast<uint64>(ResultLlc2) << "\n";
148 }
149 }
150 }
151 }
152 }
153 }
154
155 template <typename Type> struct ThreadData {
156 Type (*FuncPtr)(bool, volatile Type *, Type);
157 bool Fetch;
158 volatile Type *Ptr;
159 Type Adjustment;
160 };
161
threadWrapper(void * Data)162 template <typename Type> void *threadWrapper(void *Data) {
163 #if defined(ARM32) || defined(MIPS32)
164 // Given that most of times these crosstests for ARM are run under qemu, we
165 // set a lower NumReps to allow crosstests to complete within a reasonable
166 // amount of time.
167 static const size_t NumReps = 1000;
168 #else // ARM32 || MIPS32
169 static const size_t NumReps = 8000;
170 #endif // ARM32 || MIPS32
171
172 ThreadData<Type> *TData = reinterpret_cast<ThreadData<Type> *>(Data);
173 for (size_t i = 0; i < NumReps; ++i) {
174 (void)TData->FuncPtr(TData->Fetch, TData->Ptr, TData->Adjustment);
175 }
176 return NULL;
177 }
178
179 template <typename Type>
testAtomicRMWThreads(volatile Type * AtomicLoc,size_t & TotalTests,size_t & Passes,size_t & Failures)180 void testAtomicRMWThreads(volatile Type *AtomicLoc, size_t &TotalTests,
181 size_t &Passes, size_t &Failures) {
182 typedef Type (*FuncType)(bool, volatile Type *, Type);
183 static struct {
184 const char *Name;
185 FuncType FuncLlc;
186 FuncType FuncSz;
187 } Funcs[] = {
188 #define X(inst) \
189 { STR(inst), test_##inst, Subzero_::test_##inst } \
190 , {STR(inst) "_alloca", test_alloca_##inst, Subzero_::test_alloca_##inst},
191 RMWOP_TABLE
192 #undef X
193 };
194 const static size_t NumFuncs = sizeof(Funcs) / sizeof(*Funcs);
195
196 // Just test a few values, otherwise it takes a *really* long time.
197 volatile uint64 ValuesSubset[] = {1, 0x7e, 0x000fffffffffffffffll};
198 const size_t NumValuesSubset = sizeof(ValuesSubset) / sizeof(*ValuesSubset);
199
200 for (size_t f = 0; f < NumFuncs; ++f) {
201 for (size_t i = 0; i < NumValuesSubset; ++i) {
202 Type Value1 = static_cast<Type>(ValuesSubset[i]);
203 for (size_t j = 0; j < NumValuesSubset; ++j) {
204 Type Value2 = static_cast<Type>(ValuesSubset[j]);
205 bool fetch_first = true;
206 ThreadData<Type> TDataSz = {Funcs[f].FuncSz, fetch_first, AtomicLoc,
207 Value2};
208 ThreadData<Type> TDataLlc = {Funcs[f].FuncLlc, fetch_first, AtomicLoc,
209 Value2};
210 ++TotalTests;
211 const size_t NumThreads = 4;
212 pthread_t t[NumThreads];
213 pthread_attr_t attr[NumThreads];
214
215 // Try N threads w/ just Llc.
216 *AtomicLoc = Value1;
217 for (size_t m = 0; m < NumThreads; ++m) {
218 pthread_attr_init(&attr[m]);
219 if (pthread_create(&t[m], &attr[m], &threadWrapper<Type>,
220 reinterpret_cast<void *>(&TDataLlc)) != 0) {
221 std::cout << "pthread_create failed w/ " << strerror(errno) << "\n";
222 abort();
223 }
224 }
225 for (size_t m = 0; m < NumThreads; ++m) {
226 pthread_join(t[m], NULL);
227 }
228 Type ResultLlc = *AtomicLoc;
229
230 // Try N threads w/ both Sz and Llc.
231 *AtomicLoc = Value1;
232 for (size_t m = 0; m < NumThreads; ++m) {
233 pthread_attr_init(&attr[m]);
234 if (pthread_create(&t[m], &attr[m], &threadWrapper<Type>,
235 m % 2 == 0
236 ? reinterpret_cast<void *>(&TDataLlc)
237 : reinterpret_cast<void *>(&TDataSz)) != 0) {
238 ++Failures;
239 std::cout << "pthread_create failed w/ " << strerror(errno) << "\n";
240 abort();
241 }
242 }
243 for (size_t m = 0; m < NumThreads; ++m) {
244 if (pthread_join(t[m], NULL) != 0) {
245 ++Failures;
246 std::cout << "pthread_join failed w/ " << strerror(errno) << "\n";
247 abort();
248 }
249 }
250 Type ResultMixed = *AtomicLoc;
251
252 if (ResultLlc == ResultMixed) {
253 ++Passes;
254 } else {
255 ++Failures;
256 std::cout << "test_with_threads_" << Funcs[f].Name
257 << (8 * sizeof(Type)) << "(" << static_cast<uint64>(Value1)
258 << ", " << static_cast<uint64>(Value2)
259 << "): llc=" << static_cast<uint64>(ResultLlc)
260 << " mixed=" << static_cast<uint64>(ResultMixed) << "\n";
261 }
262 }
263 }
264 }
265 }
266
main(int argc,char * argv[])267 int main(int argc, char *argv[]) {
268 size_t TotalTests = 0;
269 size_t Passes = 0;
270 size_t Failures = 0;
271
272 testAtomicRMW<uint8_t>(&AtomicLocs.l8, TotalTests, Passes, Failures);
273 testAtomicRMW<uint16_t>(&AtomicLocs.l16, TotalTests, Passes, Failures);
274 testAtomicRMW<uint32_t>(&AtomicLocs.l32, TotalTests, Passes, Failures);
275 testAtomicRMW<uint64>(&AtomicLocs.l64, TotalTests, Passes, Failures);
276 testValCompareAndSwap<uint8_t>(&AtomicLocs.l8, TotalTests, Passes, Failures);
277 testValCompareAndSwap<uint16_t>(&AtomicLocs.l16, TotalTests, Passes,
278 Failures);
279 testValCompareAndSwap<uint32_t>(&AtomicLocs.l32, TotalTests, Passes,
280 Failures);
281 testValCompareAndSwap<uint64>(&AtomicLocs.l64, TotalTests, Passes, Failures);
282 testAtomicRMWThreads<uint8_t>(&AtomicLocs.l8, TotalTests, Passes, Failures);
283 testAtomicRMWThreads<uint16_t>(&AtomicLocs.l16, TotalTests, Passes, Failures);
284 testAtomicRMWThreads<uint32_t>(&AtomicLocs.l32, TotalTests, Passes, Failures);
285 testAtomicRMWThreads<uint64>(&AtomicLocs.l64, TotalTests, Passes, Failures);
286
287 std::cout << "TotalTests=" << TotalTests << " Passes=" << Passes
288 << " Failures=" << Failures << "\n";
289 return Failures;
290 }
291