bs3-timing-1-32.c32@ 106743

Last change on this file since 106743 was 106061, checked in by vboxsync, 3 months ago
Copyright year updates by scm.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 13.1 KB

Line
1	/* $Id: bs3-timing-1-32.c32 106061 2024-09-16 14:03:52Z vboxsync $ */
2	/** @file
3	* BS3Kit - bs3-timinig-1, 32-bit C code.
4	*/
5
6	/*
7	* Copyright (C) 2007-2024 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	#ifndef STANDALONE_EXECUTABLE
42	# include <bs3kit.h>
43	#endif
44	#if defined(RT_ARcH_ARM) \|\| defined(RT_ARCH_ARM64)
45	# include <iprt/asm-arm.h>
46	#else
47	# include <iprt/asm-amd64-x86.h>
48	#endif
49	#include <iprt/asm-math.h>
50	#include <iprt/asm.h>
51	#include <iprt/uint128.h>
52
53
54	/*********************************************************************************************************************************
55	* Structures and Typedefs *
56	*********************************************************************************************************************************/
57	/** TSC timing results. */
58	typedef struct BS3TIMING1RESULT
59	{
60	/** Number of nanoseconds elapsed while testing. */
61	uint64_t cNsElapsed;
62	/** Number of CPU ticks elapsed while testing. */
63	uint64_t cTicksElapsed;
64	/** The minium number of ticks between TSC reads. */
65	uint64_t cTicksMin;
66	/** The maximum number of ticks between TSC reads. */
67	uint64_t cTicksMax;
68	/** The sum of all TSC read deltas. */
69	uint64_t cTicksSum;
70	/** Number of loops (TSC read deltas). */
71	uint64_t cTotalLoops;
72	/** Number of times TSC moved backwards. */
73	uint64_t cBackwards;
74	/** Approx log2(cTicks) distribution. */
75	uint64_t aDistribution[65];
76	} BS3TIMING1RESULT;
77
78
79	/*********************************************************************************************************************************
80	* Global Variables *
81	*********************************************************************************************************************************/
82	/** The total result. */
83	static BS3TIMING1RESULT g_TotalResult;
84
85	/** Set if history wrapped (i.e. table is full). */
86	static bool g_fBigHistoryWrapped = false;
87	/** The next history entry. */
88	static uint32_t g_iBigHistory;
89	/** History of large gaps. */
90	static struct { uint64_t uTsc, cTicksDelta; } g_aBigHistory[384];
91
92
93
94	/**
95	* Pretty prints
96	*/
97	static void bs3Timing1_PrintTicks(uint64_t cTicks, uint64_t uCpuFreq)
98	{
99	if (uCpuFreq > _128M)
100	{
101	if (cTicks >= uCpuFreq * 1000)
102	Bs3TestPrintf("%'RU64s", cTicks / uCpuFreq);
103	else
104	{
105	const char *pszUnit;
106	uint64_t uValue;
107	if (cTicks >= uCpuFreq)
108	{
109	pszUnit = "s";
110	uValue = (cTicks * RT_MS_1SEC) / uCpuFreq;
111	}
112	else if (cTicks * RT_MS_1SEC >= uCpuFreq)
113	{
114	pszUnit = "ms";
115	uValue = (cTicks * RT_US_1SEC) / uCpuFreq;
116	}
117	else if (cTicks * RT_US_1SEC >= uCpuFreq)
118	{
119	pszUnit = "us";
120	uValue = (cTicks * RT_NS_1SEC) / uCpuFreq;
121	}
122	else if (cTicks * RT_NS_1SEC >= uCpuFreq)
123	{
124	pszUnit = "ns";
125	uValue = (cTicks * UINT64_C(1000000000000)) / uCpuFreq;
126	}
127	else
128	{
129	Bs3TestPrintf("%'RU64ps", (cTicks * UINT64_C(1000000000000)) / uCpuFreq);
130	return;
131	}
132	Bs3TestPrintf("%u.%03u%s (%'RU64 ticks)", (uint32_t)uValue / 1000, (uint32_t)uValue % 1000, pszUnit, cTicks);
133	}
134	}
135	else
136	Bs3TestPrintf("%'RU64 ticks", cTicks);
137	}
138
139
140	/**
141	* Prints a result.
142	*
143	* @param pResult The result to print.
144	* @param iRun The run (loop in qpc parlance).
145	* @param uVerbosity The verbosity level.
146	*/
147	static void bs3Timing1_PrintResult(BS3TIMING1RESULT const *pResult, unsigned iRun, unsigned uVerbosity)
148	{
149	uint64_t uCpuFreq;
150
151	/*
152	* Calc CPU frequency.
153	*/
154	if (pResult->cNsElapsed > 0 && pResult->cTicksElapsed > 0)
155	{
156	#if 1
157	RTUINT128U Tmp1, Divisor, Result;
158	RTUInt128Div(&Result,
159	RTUInt128MulU64ByU64(&Tmp1, pResult->cTicksElapsed, RT_NS_1SEC),
160	RTUInt128AssignU64(&Divisor, pResult->cNsElapsed));
161	uCpuFreq = Result.s.Lo;
162	#else
163	unsigned const cShift = pResult->cTicksElapsed < UINT64_C(0x000225C17D04) ? 0
164	: pResult->cTicksElapsed < UINT64_C(0x00225C17D04D) ? 4
165	: pResult->cTicksElapsed < UINT64_C(0x0225C17D04DA) ? 8
166	: pResult->cTicksElapsed < UINT64_C(0x225C1D940BF6) ? 12
167	: 16;
168	uCpuFreq = pResult->cTicksElapsed * ((uint64_t)RT_NS_1SEC >> cShift) / (pResult->cNsElapsed >> cShift);
169	#endif
170	}
171	else
172	uCpuFreq = 1;
173
174	/*
175	* Report results.
176	*
177	* Note! in 32-bit and 16-bit mode, values 4G or higher gets formatted as
178	* hexadecimal to avoid 64-bit division.
179	*/
180	Bs3TestPrintf("Loop #%u: %'RU64 tests: ", iRun, pResult->cTotalLoops);
181
182	Bs3TestPrintf("average ");
183	bs3Timing1_PrintTicks(pResult->cTicksSum / pResult->cTotalLoops, uCpuFreq);
184	Bs3TestPrintf(", min ");
185	bs3Timing1_PrintTicks(pResult->cTicksMin, uCpuFreq);
186	Bs3TestPrintf(", max ");
187	bs3Timing1_PrintTicks(pResult->cTicksMax, uCpuFreq);
188	Bs3TestPrintf("\n");
189
190	/* Distribution (tick delta log2-ish). */
191	if (uVerbosity > 0)
192	{
193	unsigned iItem = 0;
194	unsigned i;
195	for (i = uVerbosity > 1 ? 0 : 5; i < RT_ELEMENTS(pResult->aDistribution); i++)
196	if (pResult->aDistribution[i] != 0)
197	{
198	if (iItem >= 6)
199	{
200	iItem = 0;
201	Bs3TestPrintf("\n");
202	}
203	iItem++;
204	Bs3TestPrintf(" %'11RU64\|2^%-2u", pResult->aDistribution[i], i);
205	}
206	if (uVerbosity > 1)
207	Bs3TestPrintf(iItem < 6 ? " (%'RU64 Hz)\n" : "\n (%'RU64 Hz)\n", uCpuFreq);
208	else
209	Bs3TestPrintf("\n");
210	}
211	if (pResult->cBackwards != 0)
212	Bs3TestFailedF("TSC went backwards %'RU64 time(s)", pResult->cBackwards);
213	}
214
215
216	/**
217	* Do one TSC timing iteration.
218	*
219	* @param iRun The iteration number (loop).
220	* @param cSecs The number of seconds to sample TSCs.
221	* @param uVerbosity The noise level.
222	* @param iMinHistory The threshold level to put stuff in g_auTscHistory.
223	*/
224	static void bs3Timing1_Tsc_One(unsigned iRun, uint32_t cSecs, unsigned uVerbosity, unsigned iMinHistory)
225	{
226	uint64_t const nsStart = Bs3TestNow();
227	uint64_t const uTscStart = ASMReadTSC();
228	uint64_t const nsDeadline = nsStart + cSecs * RT_NS_1SEC_64;
229	uint64_t cNsElapsed;
230	BS3TIMING1RESULT Result;
231	unsigned i;
232
233	Bs3MemZero(&Result, sizeof(Result));
234	Result.cTicksMin = UINT64_MAX;
235
236	/*
237	* Test loop.
238	*/
239	do
240	{
241	unsigned cLoops = 100000 + 1;
242	Result.cTotalLoops += cLoops - 1;
243	while (--cLoops != 0)
244	{
245	uint64_t uTscPrev = ASMReadTSC();
246	uint64_t uTscNow = ASMReadTSC();
247	uint64_t cTicks = uTscNow - uTscPrev;
248	unsigned iBit;
249
250	/* check that it doesn't go backwards*/
251	if ((int64_t)cTicks < 0)
252	Result.cBackwards++;
253
254	/* min/max/avg */
255	Result.cTicksSum += cTicks;
256	if (cTicks < Result.cTicksMin)
257	Result.cTicksMin = cTicks;
258	if (cTicks > Result.cTicksMax)
259	Result.cTicksMax = cTicks;
260
261	/* result distribution by most significant bit. */
262	iBit = ASMBitLastSetU64(cTicks);
263	Result.aDistribution[iBit] += 1;
264	if (iBit < iMinHistory)
265	{ /* likely */ }
266	else
267	{
268	g_aBigHistory[g_iBigHistory].uTsc = uTscPrev;
269	g_aBigHistory[g_iBigHistory].cTicksDelta = cTicks;
270	if (++g_iBigHistory >= RT_ELEMENTS(g_aBigHistory))
271	{
272	g_iBigHistory = 0;
273	g_fBigHistoryWrapped = true;
274	}
275	}
276	}
277	} while ((cNsElapsed = Bs3TestNow()) < nsDeadline);
278
279	Result.cTicksElapsed = ASMReadTSC() - uTscStart;
280	Result.cNsElapsed = cNsElapsed - nsStart;
281
282	bs3Timing1_PrintResult(&Result, iRun, uVerbosity);
283
284	/* Add to total. */
285	g_TotalResult.cNsElapsed += Result.cNsElapsed;
286	g_TotalResult.cTicksElapsed += Result.cTicksElapsed;
287	if (Result.cTicksMin < g_TotalResult.cTicksMin \|\| g_TotalResult.cTicksMin == 0)
288	g_TotalResult.cTicksMin = Result.cTicksMin;
289	if (Result.cTicksMax > g_TotalResult.cTicksMax)
290	g_TotalResult.cTicksMax += Result.cTicksMax;
291	g_TotalResult.cTicksSum += Result.cTicksSum;
292	g_TotalResult.cTotalLoops += Result.cTotalLoops;
293	g_TotalResult.cBackwards += Result.cBackwards;
294	for (i = 0; i < RT_ELEMENTS(Result.aDistribution); i++)
295	g_TotalResult.aDistribution[i] += Result.aDistribution[i];
296	}
297
298
299	/**
300	* The TSC test driver.
301	*
302	* @param cLoops Number of test iterations.
303	* @param cSecs The number of seconds per iteration.
304	* @param uVerbosity How noisy we should be.
305	* @param iMinHistory The threshold for big gap history.
306	*/
307	static void bs3Timing1_Tsc_Driver(unsigned cLoops, unsigned cSecs, unsigned uVerbosity, unsigned iMinHistory)
308	{
309	unsigned iLoop;
310
311	#if 1
312	/*
313	* Verify that the first/last bit in U64 works (didn't).
314	*/
315	iLoop = ASMBitLastSetU64( UINT64_C(0x1000100010001000)); if (iLoop != 61) Bs3TestFailedF("%d: iLoop=%d\n", __LINE__, iLoop);
316	iLoop = ASMBitFirstSetU64(UINT64_C(0x1000100010001000)); if (iLoop != 13) Bs3TestFailedF("%d: iLoop=%d\n", __LINE__, iLoop);
317	iLoop = ASMBitLastSetU64( UINT64_C(0x000ffff000000000)); if (iLoop != 52) Bs3TestFailedF("%d: iLoop=%d\n", __LINE__, iLoop);
318	iLoop = ASMBitFirstSetU64(UINT64_C(0x000ffff000000000)); if (iLoop != 37) Bs3TestFailedF("%d: iLoop=%d\n", __LINE__, iLoop);
319	#endif
320
321	/*
322	* Do the work.
323	*/
324	Bs3TestPrintf("Running %u loops, %u second%s each...\n", cLoops, cSecs, cSecs != 1 ? "s" : "");
325	for (iLoop = 1; iLoop <= cLoops; iLoop++)
326	bs3Timing1_Tsc_One(iLoop, cSecs, uVerbosity, iMinHistory);
327
328	/*
329	* Report the total.
330	*/
331	Bs3TestPrintf("Total:\n");
332	bs3Timing1_PrintResult(&g_TotalResult, iLoop, uVerbosity + 1);
333
334	/*
335	* Dump the large gap history, if any.
336	*/
337	if (g_fBigHistoryWrapped \|\| g_iBigHistory > 0)
338	{
339	uint32_t const iFirst = g_fBigHistoryWrapped ? g_iBigHistory : 0;
340	uint32_t const iEnd = g_iBigHistory;
341	uint64_t uTscPrev = g_aBigHistory[iFirst].uTsc;
342	uint32_t i = iFirst;
343	Bs3TestPrintf("Big gap history (TSC, prev delta, test delta\|level):\n");
344	do
345	{
346	Bs3TestPrintf(" %'RU64: %'14RU64 - %'14RU64\|%u\n", g_aBigHistory[i].uTsc, g_aBigHistory[i].uTsc - uTscPrev,
347	g_aBigHistory[i].cTicksDelta, ASMBitLastSetU64(g_aBigHistory[i].cTicksDelta));
348	uTscPrev = g_aBigHistory[i].uTsc;
349	if (++i >= RT_ELEMENTS(g_aBigHistory))
350	i = 0;
351	} while (i != iEnd);
352	}
353	else
354	Bs3TestPrintf("No big gap history.\n");
355	}
356
357
358	#ifndef STANDALONE_EXECUTABLE
359	BS3_DECL(void) bs3Timing1_Tsc_pe32(void)
360	{
361	Bs3TestPrintf("bs3Timing1_Tsc_pe32\n");
362	bs3Timing1_Tsc_Driver(60, 10 /sec/, 1 /uVerbosity/, 17);
363	}
364	#endif
365
366	/* P.S. don't forget: VBoxManage setextradata bs3-timing-1 VBoxInternal/Devices/VMMDev/0/Config/TestingEnabled 1 */
367

Note: See TracBrowser for help on using the repository browser.

Download in other formats:

Original Format