1 | /* $Id: tstPrfRT.cpp 96407 2022-08-22 17:43:14Z vboxsync $ */
|
---|
2 | /** @file
|
---|
3 | * IPRT testcase - profile some of the important functions.
|
---|
4 | */
|
---|
5 |
|
---|
6 | /*
|
---|
7 | * Copyright (C) 2006-2022 Oracle and/or its affiliates.
|
---|
8 | *
|
---|
9 | * This file is part of VirtualBox base platform packages, as
|
---|
10 | * available from https://www.virtualbox.org.
|
---|
11 | *
|
---|
12 | * This program is free software; you can redistribute it and/or
|
---|
13 | * modify it under the terms of the GNU General Public License
|
---|
14 | * as published by the Free Software Foundation, in version 3 of the
|
---|
15 | * License.
|
---|
16 | *
|
---|
17 | * This program is distributed in the hope that it will be useful, but
|
---|
18 | * WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
19 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
---|
20 | * General Public License for more details.
|
---|
21 | *
|
---|
22 | * You should have received a copy of the GNU General Public License
|
---|
23 | * along with this program; if not, see <https://www.gnu.org/licenses>.
|
---|
24 | *
|
---|
25 | * The contents of this file may alternatively be used under the terms
|
---|
26 | * of the Common Development and Distribution License Version 1.0
|
---|
27 | * (CDDL), a copy of it is provided in the "COPYING.CDDL" file included
|
---|
28 | * in the VirtualBox distribution, in which case the provisions of the
|
---|
29 | * CDDL are applicable instead of those of the GPL.
|
---|
30 | *
|
---|
31 | * You may elect to license modified versions of this file under the
|
---|
32 | * terms and conditions of either the GPL or the CDDL or both.
|
---|
33 | *
|
---|
34 | * SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0
|
---|
35 | */
|
---|
36 |
|
---|
37 |
|
---|
38 | /*********************************************************************************************************************************
|
---|
39 | * Header Files *
|
---|
40 | *********************************************************************************************************************************/
|
---|
41 | #include <iprt/initterm.h>
|
---|
42 | #include <iprt/time.h>
|
---|
43 | #include <iprt/log.h>
|
---|
44 | #include <iprt/test.h>
|
---|
45 | #include <iprt/thread.h>
|
---|
46 | #if defined(RT_ARCH_AMD64) || defined(RT_ARCH_X86)
|
---|
47 | # include <iprt/asm-amd64-x86.h>
|
---|
48 | #endif
|
---|
49 |
|
---|
50 |
|
---|
51 | /*********************************************************************************************************************************
|
---|
52 | * Internal Functions *
|
---|
53 | *********************************************************************************************************************************/
|
---|
54 | #if defined(RT_ARCH_AMD64) || defined(RT_ARCH_X86)
|
---|
55 | DECLASM(void) tstRTPRfAMemoryAccess(void);
|
---|
56 | DECLASM(void) tstRTPRfARegisterAccess(void);
|
---|
57 | DECLASM(void) tstRTPRfAMemoryUnalignedAccess(void);
|
---|
58 | #endif
|
---|
59 |
|
---|
60 |
|
---|
61 | /*********************************************************************************************************************************
|
---|
62 | * Global Variables *
|
---|
63 | *********************************************************************************************************************************/
|
---|
64 | static RTTEST g_hTest;
|
---|
65 |
|
---|
66 |
|
---|
67 | #if defined(RT_ARCH_AMD64) || defined(RT_ARCH_X86)
|
---|
68 |
|
---|
69 | void PrintResult(uint64_t u64Ticks, uint64_t u64MaxTicks, uint64_t u64MinTicks, unsigned cTimes, const char *pszOperation)
|
---|
70 | {
|
---|
71 | //RTTestPrintf(g_hTest, RTTESTLVL_ALWAYS,
|
---|
72 | // "%-32s %5lld / %5lld / %5lld ticks per call (%u calls %lld ticks)\n",
|
---|
73 | // pszOperation, u64MinTicks, u64Ticks / (uint64_t)cTimes, u64MaxTicks, cTimes, u64Ticks);
|
---|
74 | //RTTestValueF(g_hTest, u64MinTicks, RTTESTUNIT_NONE, "%s min ticks", pszOperation);
|
---|
75 | RTTestValueF(g_hTest, u64Ticks / (uint64_t)cTimes, RTTESTUNIT_NONE, "%s avg ticks", pszOperation);
|
---|
76 | //RTTestValueF(g_hTest, u64MaxTicks, RTTESTUNIT_NONE, "%s max ticks", pszOperation);
|
---|
77 | RT_NOREF_PV(u64MaxTicks); RT_NOREF_PV(u64MinTicks);
|
---|
78 | }
|
---|
79 |
|
---|
80 | # define ITERATE(preexpr, expr, postexpr, cIterations) \
|
---|
81 | AssertCompile(((cIterations) % 8) == 0); \
|
---|
82 | /* Min and max value. */ \
|
---|
83 | for (i = 0, u64MinTS = UINT64_MAX, u64MaxTS = 0; i < (cIterations); i++) \
|
---|
84 | { \
|
---|
85 | { preexpr } \
|
---|
86 | uint64_t u64StartTS = ASMReadTSC(); \
|
---|
87 | { expr } \
|
---|
88 | uint64_t u64ElapsedTS = ASMReadTSC() - u64StartTS; \
|
---|
89 | { postexpr } \
|
---|
90 | if (u64ElapsedTS > u64MinTS * 32) \
|
---|
91 | { \
|
---|
92 | i--; \
|
---|
93 | continue; \
|
---|
94 | } \
|
---|
95 | if (u64ElapsedTS < u64MinTS) \
|
---|
96 | u64MinTS = u64ElapsedTS; \
|
---|
97 | if (u64ElapsedTS > u64MaxTS) \
|
---|
98 | u64MaxTS = u64ElapsedTS; \
|
---|
99 | } \
|
---|
100 | { \
|
---|
101 | /* Calculate a good average value (may be smaller than min). */ \
|
---|
102 | i = (cIterations); \
|
---|
103 | AssertRelease((i % 8) == 0); \
|
---|
104 | { preexpr } \
|
---|
105 | uint64_t u64StartTS = ASMReadTSC(); \
|
---|
106 | while (i != 0) \
|
---|
107 | { \
|
---|
108 | { expr } \
|
---|
109 | { expr } \
|
---|
110 | { expr } \
|
---|
111 | { expr } \
|
---|
112 | { expr } \
|
---|
113 | { expr } \
|
---|
114 | { expr } \
|
---|
115 | { expr } \
|
---|
116 | i -= 8; \
|
---|
117 | } \
|
---|
118 | u64TotalTS = ASMReadTSC() - u64StartTS; \
|
---|
119 | { postexpr } \
|
---|
120 | i = (cIterations); \
|
---|
121 | }
|
---|
122 |
|
---|
123 | #else /* !AMD64 && !X86 */
|
---|
124 |
|
---|
125 | void PrintResult(uint64_t cNs, uint64_t cNsMax, uint64_t cNsMin, unsigned cTimes, const char *pszOperation)
|
---|
126 | {
|
---|
127 | //RTTestPrintf(g_hTest, RTTESTLVL_ALWAYS,
|
---|
128 | // "%-32s %5lld / %5lld / %5lld ns per call (%u calls %lld ns)\n",
|
---|
129 | // pszOperation, cNsMin, cNs / (uint64_t)cTimes, cNsMax, cTimes, cNs);
|
---|
130 | //RTTestValueF(g_hTest, cNsMin, RTTESTUNIT_NS_PER_CALL, "%s min", pszOperation);
|
---|
131 | RTTestValueF(g_hTest, cNs / (uint64_t)cTimes, RTTESTUNIT_NS_PER_CALL, "%s avg", pszOperation);
|
---|
132 | //RTTestValueF(g_hTest, cNsMax, RTTESTUNIT_NS_PER_CALL, "%s max", pszOperation);
|
---|
133 | }
|
---|
134 |
|
---|
135 | # define ITERATE(preexpr, expr, postexpr, cIterations) \
|
---|
136 | for (i = 0, u64TotalTS = 0, u64MinTS = UINT64_MAX, u64MaxTS = 0; i < (cIterations); i++) \
|
---|
137 | { \
|
---|
138 | { preexpr } \
|
---|
139 | uint64_t u64StartTS = RTTimeNanoTS(); \
|
---|
140 | { expr } \
|
---|
141 | uint64_t u64ElapsedTS = RTTimeNanoTS() - u64StartTS; \
|
---|
142 | { postexpr } \
|
---|
143 | if (u64ElapsedTS > u64MinTS * 32) \
|
---|
144 | { \
|
---|
145 | i--; \
|
---|
146 | continue; \
|
---|
147 | } \
|
---|
148 | if (u64ElapsedTS < u64MinTS) \
|
---|
149 | u64MinTS = u64ElapsedTS; \
|
---|
150 | if (u64ElapsedTS > u64MaxTS) \
|
---|
151 | u64MaxTS = u64ElapsedTS; \
|
---|
152 | u64TotalTS += u64ElapsedTS; \
|
---|
153 | }
|
---|
154 |
|
---|
155 | #endif /* !AMD64 && !X86 */
|
---|
156 |
|
---|
157 |
|
---|
158 | int main(int argc, char **argv)
|
---|
159 | {
|
---|
160 | uint64_t u64TotalTS;
|
---|
161 | uint64_t u64MinTS;
|
---|
162 | uint64_t u64MaxTS;
|
---|
163 | uint32_t i;
|
---|
164 |
|
---|
165 | RTEXITCODE rcExit = RTTestInitExAndCreate(argc, &argv, argc == 2 ? RTR3INIT_FLAGS_SUPLIB : 0, "tstRTPrf", &g_hTest);
|
---|
166 | if (rcExit != RTEXITCODE_SUCCESS)
|
---|
167 | return rcExit;
|
---|
168 | RTTestBanner(g_hTest);
|
---|
169 |
|
---|
170 | /*
|
---|
171 | * RTTimeNanoTS, RTTimeProgramNanoTS, RTTimeMilliTS, and RTTimeProgramMilliTS.
|
---|
172 | */
|
---|
173 | ITERATE(RT_NOTHING, RTTimeNanoTS();, RT_NOTHING, _32M);
|
---|
174 | PrintResult(u64TotalTS, u64MaxTS, u64MinTS, i, "RTTimeNanoTS");
|
---|
175 |
|
---|
176 | ITERATE(RT_NOTHING, RTTimeProgramNanoTS();, RT_NOTHING, UINT32_C(1000000));
|
---|
177 | PrintResult(u64TotalTS, u64MaxTS, u64MinTS, i, "RTTimeProgramNanoTS");
|
---|
178 |
|
---|
179 | ITERATE(RT_NOTHING, RTTimeMilliTS();, RT_NOTHING, UINT32_C(1000000));
|
---|
180 | PrintResult(u64TotalTS, u64MaxTS, u64MinTS, i, "RTTimeMilliTS");
|
---|
181 |
|
---|
182 | ITERATE(RT_NOTHING, RTTimeProgramMilliTS();, RT_NOTHING, UINT32_C(1000000));
|
---|
183 | PrintResult(u64TotalTS, u64MaxTS, u64MinTS, i, "RTTimeProgramMilliTS");
|
---|
184 |
|
---|
185 | /*
|
---|
186 | * RTTimeNow
|
---|
187 | */
|
---|
188 | RTTIMESPEC Time;
|
---|
189 | ITERATE(RT_NOTHING, RTTimeNow(&Time);, RT_NOTHING, UINT32_C(1000000));
|
---|
190 | PrintResult(u64TotalTS, u64MaxTS, u64MinTS, i, "RTTimeNow");
|
---|
191 |
|
---|
192 | /*
|
---|
193 | * RTLogDefaultInstance()
|
---|
194 | */
|
---|
195 | ITERATE(RT_NOTHING, RTLogDefaultInstance();, RT_NOTHING, UINT32_C(1000000));
|
---|
196 | PrintResult(u64TotalTS, u64MaxTS, u64MinTS, i, "RTLogDefaultInstance");
|
---|
197 |
|
---|
198 | /*
|
---|
199 | * RTThreadSelf and RTThreadNativeSelf
|
---|
200 | */
|
---|
201 | ITERATE(RT_NOTHING, RTThreadSelf();, RT_NOTHING, UINT32_C(1000000));
|
---|
202 | PrintResult(u64TotalTS, u64MaxTS, u64MinTS, i, "RTThreadSelf");
|
---|
203 |
|
---|
204 | ITERATE(RT_NOTHING, RTThreadNativeSelf();, RT_NOTHING, UINT32_C(1000000));
|
---|
205 | PrintResult(u64TotalTS, u64MaxTS, u64MinTS, i, "RTThreadNativeSelf");
|
---|
206 |
|
---|
207 | #if defined(RT_ARCH_X86) || defined(RT_ARCH_AMD64)
|
---|
208 | /*
|
---|
209 | * Registers vs stack.
|
---|
210 | */
|
---|
211 | ITERATE(RT_NOTHING, tstRTPRfARegisterAccess();, RT_NOTHING, UINT32_C(1000));
|
---|
212 | uint64_t const cRegTotal = u64TotalTS;
|
---|
213 | //PrintResult(u64TotalTS, u64MaxTS, u64MinTS, i, "Register only algorithm");
|
---|
214 |
|
---|
215 | ITERATE(RT_NOTHING, tstRTPRfAMemoryAccess();, RT_NOTHING, UINT32_C(1000));
|
---|
216 | uint64_t const cMemTotal = u64TotalTS;
|
---|
217 | //PrintResult(u64TotalTS, u64MaxTS, u64MinTS, i, "Memory only algorithm");
|
---|
218 |
|
---|
219 | ITERATE(RT_NOTHING, tstRTPRfAMemoryUnalignedAccess();, RT_NOTHING, UINT32_C(1000));
|
---|
220 | uint64_t const cMemUnalignedTotal = u64TotalTS;
|
---|
221 | //PrintResult(u64TotalTS, u64MaxTS, u64MinTS, i, "Memory only algorithm");
|
---|
222 |
|
---|
223 | uint64_t const cSlower100 = cMemTotal * 100 / cRegTotal;
|
---|
224 | RTTestValue(g_hTest, "Memory instead of registers slowdown", cSlower100, RTTESTUNIT_PCT);
|
---|
225 | uint64_t const cUnalignedSlower100 = cMemUnalignedTotal * 100 / cRegTotal;
|
---|
226 | RTTestValue(g_hTest, "Unaligned memory instead of registers slowdown", cUnalignedSlower100, RTTESTUNIT_PCT);
|
---|
227 | #endif
|
---|
228 |
|
---|
229 | return RTTestSummaryAndDestroy(g_hTest);
|
---|
230 | }
|
---|