aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/go/_std_1.21/src/internal/profile/merge.go
blob: 3ea7d4cf42b921e8195d916e1f7e545715f07bbd (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
// Copyright 2019 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

package profile

import (
	"fmt"
	"sort"
	"strconv"
	"strings"
)

// Merge merges all the profiles in profs into a single Profile.
// Returns a new profile independent of the input profiles. The merged
// profile is compacted to eliminate unused samples, locations,
// functions and mappings. Profiles must have identical profile sample
// and period types or the merge will fail. profile.Period of the
// resulting profile will be the maximum of all profiles, and
// profile.TimeNanos will be the earliest nonzero one.
func Merge(srcs []*Profile) (*Profile, error) {
	if len(srcs) == 0 {
		return nil, fmt.Errorf("no profiles to merge")
	}
	p, err := combineHeaders(srcs)
	if err != nil {
		return nil, err
	}

	pm := &profileMerger{
		p:         p,
		samples:   make(map[sampleKey]*Sample, len(srcs[0].Sample)),
		locations: make(map[locationKey]*Location, len(srcs[0].Location)),
		functions: make(map[functionKey]*Function, len(srcs[0].Function)),
		mappings:  make(map[mappingKey]*Mapping, len(srcs[0].Mapping)),
	}

	for _, src := range srcs {
		// Clear the profile-specific hash tables
		pm.locationsByID = make(map[uint64]*Location, len(src.Location))
		pm.functionsByID = make(map[uint64]*Function, len(src.Function))
		pm.mappingsByID = make(map[uint64]mapInfo, len(src.Mapping))

		if len(pm.mappings) == 0 && len(src.Mapping) > 0 {
			// The Mapping list has the property that the first mapping
			// represents the main binary. Take the first Mapping we see,
			// otherwise the operations below will add mappings in an
			// arbitrary order.
			pm.mapMapping(src.Mapping[0])
		}

		for _, s := range src.Sample {
			if !isZeroSample(s) {
				pm.mapSample(s)
			}
		}
	}

	for _, s := range p.Sample {
		if isZeroSample(s) {
			// If there are any zero samples, re-merge the profile to GC
			// them.
			return Merge([]*Profile{p})
		}
	}

	return p, nil
}

// Normalize normalizes the source profile by multiplying each value in profile by the
// ratio of the sum of the base profile's values of that sample type to the sum of the
// source profile's value of that sample type.
func (p *Profile) Normalize(pb *Profile) error {

	if err := p.compatible(pb); err != nil {
		return err
	}

	baseVals := make([]int64, len(p.SampleType))
	for _, s := range pb.Sample {
		for i, v := range s.Value {
			baseVals[i] += v
		}
	}

	srcVals := make([]int64, len(p.SampleType))
	for _, s := range p.Sample {
		for i, v := range s.Value {
			srcVals[i] += v
		}
	}

	normScale := make([]float64, len(baseVals))
	for i := range baseVals {
		if srcVals[i] == 0 {
			normScale[i] = 0.0
		} else {
			normScale[i] = float64(baseVals[i]) / float64(srcVals[i])
		}
	}
	p.ScaleN(normScale)
	return nil
}

func isZeroSample(s *Sample) bool {
	for _, v := range s.Value {
		if v != 0 {
			return false
		}
	}
	return true
}

type profileMerger struct {
	p *Profile

	// Memoization tables within a profile.
	locationsByID map[uint64]*Location
	functionsByID map[uint64]*Function
	mappingsByID  map[uint64]mapInfo

	// Memoization tables for profile entities.
	samples   map[sampleKey]*Sample
	locations map[locationKey]*Location
	functions map[functionKey]*Function
	mappings  map[mappingKey]*Mapping
}

type mapInfo struct {
	m      *Mapping
	offset int64
}

func (pm *profileMerger) mapSample(src *Sample) *Sample {
	s := &Sample{
		Location: make([]*Location, len(src.Location)),
		Value:    make([]int64, len(src.Value)),
		Label:    make(map[string][]string, len(src.Label)),
		NumLabel: make(map[string][]int64, len(src.NumLabel)),
		NumUnit:  make(map[string][]string, len(src.NumLabel)),
	}
	for i, l := range src.Location {
		s.Location[i] = pm.mapLocation(l)
	}
	for k, v := range src.Label {
		vv := make([]string, len(v))
		copy(vv, v)
		s.Label[k] = vv
	}
	for k, v := range src.NumLabel {
		u := src.NumUnit[k]
		vv := make([]int64, len(v))
		uu := make([]string, len(u))
		copy(vv, v)
		copy(uu, u)
		s.NumLabel[k] = vv
		s.NumUnit[k] = uu
	}
	// Check memoization table. Must be done on the remapped location to
	// account for the remapped mapping. Add current values to the
	// existing sample.
	k := s.key()
	if ss, ok := pm.samples[k]; ok {
		for i, v := range src.Value {
			ss.Value[i] += v
		}
		return ss
	}
	copy(s.Value, src.Value)
	pm.samples[k] = s
	pm.p.Sample = append(pm.p.Sample, s)
	return s
}

// key generates sampleKey to be used as a key for maps.
func (sample *Sample) key() sampleKey {
	ids := make([]string, len(sample.Location))
	for i, l := range sample.Location {
		ids[i] = strconv.FormatUint(l.ID, 16)
	}

	labels := make([]string, 0, len(sample.Label))
	for k, v := range sample.Label {
		labels = append(labels, fmt.Sprintf("%q%q", k, v))
	}
	sort.Strings(labels)

	numlabels := make([]string, 0, len(sample.NumLabel))
	for k, v := range sample.NumLabel {
		numlabels = append(numlabels, fmt.Sprintf("%q%x%x", k, v, sample.NumUnit[k]))
	}
	sort.Strings(numlabels)

	return sampleKey{
		strings.Join(ids, "|"),
		strings.Join(labels, ""),
		strings.Join(numlabels, ""),
	}
}

type sampleKey struct {
	locations string
	labels    string
	numlabels string
}

func (pm *profileMerger) mapLocation(src *Location) *Location {
	if src == nil {
		return nil
	}

	if l, ok := pm.locationsByID[src.ID]; ok {
		pm.locationsByID[src.ID] = l
		return l
	}

	mi := pm.mapMapping(src.Mapping)
	l := &Location{
		ID:       uint64(len(pm.p.Location) + 1),
		Mapping:  mi.m,
		Address:  uint64(int64(src.Address) + mi.offset),
		Line:     make([]Line, len(src.Line)),
		IsFolded: src.IsFolded,
	}
	for i, ln := range src.Line {
		l.Line[i] = pm.mapLine(ln)
	}
	// Check memoization table. Must be done on the remapped location to
	// account for the remapped mapping ID.
	k := l.key()
	if ll, ok := pm.locations[k]; ok {
		pm.locationsByID[src.ID] = ll
		return ll
	}
	pm.locationsByID[src.ID] = l
	pm.locations[k] = l
	pm.p.Location = append(pm.p.Location, l)
	return l
}

// key generates locationKey to be used as a key for maps.
func (l *Location) key() locationKey {
	key := locationKey{
		addr:     l.Address,
		isFolded: l.IsFolded,
	}
	if l.Mapping != nil {
		// Normalizes address to handle address space randomization.
		key.addr -= l.Mapping.Start
		key.mappingID = l.Mapping.ID
	}
	lines := make([]string, len(l.Line)*2)
	for i, line := range l.Line {
		if line.Function != nil {
			lines[i*2] = strconv.FormatUint(line.Function.ID, 16)
		}
		lines[i*2+1] = strconv.FormatInt(line.Line, 16)
	}
	key.lines = strings.Join(lines, "|")
	return key
}

type locationKey struct {
	addr, mappingID uint64
	lines           string
	isFolded        bool
}

func (pm *profileMerger) mapMapping(src *Mapping) mapInfo {
	if src == nil {
		return mapInfo{}
	}

	if mi, ok := pm.mappingsByID[src.ID]; ok {
		return mi
	}

	// Check memoization tables.
	mk := src.key()
	if m, ok := pm.mappings[mk]; ok {
		mi := mapInfo{m, int64(m.Start) - int64(src.Start)}
		pm.mappingsByID[src.ID] = mi
		return mi
	}
	m := &Mapping{
		ID:              uint64(len(pm.p.Mapping) + 1),
		Start:           src.Start,
		Limit:           src.Limit,
		Offset:          src.Offset,
		File:            src.File,
		BuildID:         src.BuildID,
		HasFunctions:    src.HasFunctions,
		HasFilenames:    src.HasFilenames,
		HasLineNumbers:  src.HasLineNumbers,
		HasInlineFrames: src.HasInlineFrames,
	}
	pm.p.Mapping = append(pm.p.Mapping, m)

	// Update memoization tables.
	pm.mappings[mk] = m
	mi := mapInfo{m, 0}
	pm.mappingsByID[src.ID] = mi
	return mi
}

// key generates encoded strings of Mapping to be used as a key for
// maps.
func (m *Mapping) key() mappingKey {
	// Normalize addresses to handle address space randomization.
	// Round up to next 4K boundary to avoid minor discrepancies.
	const mapsizeRounding = 0x1000

	size := m.Limit - m.Start
	size = size + mapsizeRounding - 1
	size = size - (size % mapsizeRounding)
	key := mappingKey{
		size:   size,
		offset: m.Offset,
	}

	switch {
	case m.BuildID != "":
		key.buildIDOrFile = m.BuildID
	case m.File != "":
		key.buildIDOrFile = m.File
	default:
		// A mapping containing neither build ID nor file name is a fake mapping. A
		// key with empty buildIDOrFile is used for fake mappings so that they are
		// treated as the same mapping during merging.
	}
	return key
}

type mappingKey struct {
	size, offset  uint64
	buildIDOrFile string
}

func (pm *profileMerger) mapLine(src Line) Line {
	ln := Line{
		Function: pm.mapFunction(src.Function),
		Line:     src.Line,
	}
	return ln
}

func (pm *profileMerger) mapFunction(src *Function) *Function {
	if src == nil {
		return nil
	}
	if f, ok := pm.functionsByID[src.ID]; ok {
		return f
	}
	k := src.key()
	if f, ok := pm.functions[k]; ok {
		pm.functionsByID[src.ID] = f
		return f
	}
	f := &Function{
		ID:         uint64(len(pm.p.Function) + 1),
		Name:       src.Name,
		SystemName: src.SystemName,
		Filename:   src.Filename,
		StartLine:  src.StartLine,
	}
	pm.functions[k] = f
	pm.functionsByID[src.ID] = f
	pm.p.Function = append(pm.p.Function, f)
	return f
}

// key generates a struct to be used as a key for maps.
func (f *Function) key() functionKey {
	return functionKey{
		f.StartLine,
		f.Name,
		f.SystemName,
		f.Filename,
	}
}

type functionKey struct {
	startLine                  int64
	name, systemName, fileName string
}

// combineHeaders checks that all profiles can be merged and returns
// their combined profile.
func combineHeaders(srcs []*Profile) (*Profile, error) {
	for _, s := range srcs[1:] {
		if err := srcs[0].compatible(s); err != nil {
			return nil, err
		}
	}

	var timeNanos, durationNanos, period int64
	var comments []string
	seenComments := map[string]bool{}
	var defaultSampleType string
	for _, s := range srcs {
		if timeNanos == 0 || s.TimeNanos < timeNanos {
			timeNanos = s.TimeNanos
		}
		durationNanos += s.DurationNanos
		if period == 0 || period < s.Period {
			period = s.Period
		}
		for _, c := range s.Comments {
			if seen := seenComments[c]; !seen {
				comments = append(comments, c)
				seenComments[c] = true
			}
		}
		if defaultSampleType == "" {
			defaultSampleType = s.DefaultSampleType
		}
	}

	p := &Profile{
		SampleType: make([]*ValueType, len(srcs[0].SampleType)),

		DropFrames: srcs[0].DropFrames,
		KeepFrames: srcs[0].KeepFrames,

		TimeNanos:     timeNanos,
		DurationNanos: durationNanos,
		PeriodType:    srcs[0].PeriodType,
		Period:        period,

		Comments:          comments,
		DefaultSampleType: defaultSampleType,
	}
	copy(p.SampleType, srcs[0].SampleType)
	return p, nil
}

// compatible determines if two profiles can be compared/merged.
// returns nil if the profiles are compatible; otherwise an error with
// details on the incompatibility.
func (p *Profile) compatible(pb *Profile) error {
	if !equalValueType(p.PeriodType, pb.PeriodType) {
		return fmt.Errorf("incompatible period types %v and %v", p.PeriodType, pb.PeriodType)
	}

	if len(p.SampleType) != len(pb.SampleType) {
		return fmt.Errorf("incompatible sample types %v and %v", p.SampleType, pb.SampleType)
	}

	for i := range p.SampleType {
		if !equalValueType(p.SampleType[i], pb.SampleType[i]) {
			return fmt.Errorf("incompatible sample types %v and %v", p.SampleType, pb.SampleType)
		}
	}
	return nil
}

// equalValueType returns true if the two value types are semantically
// equal. It ignores the internal fields used during encode/decode.
func equalValueType(st1, st2 *ValueType) bool {
	return st1.Type == st2.Type && st1.Unit == st2.Unit
}