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
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
|
/*
* Copyright 2014 Google Inc.
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include <ctype.h>
#include "Benchmark.h"
#include "CrashHandler.h"
#include "GMBench.h"
#include "ProcStats.h"
#include "ResultsWriter.h"
#include "RecordingBench.h"
#include "SKPBench.h"
#include "Stats.h"
#include "Timer.h"
#include "SkBBHFactory.h"
#include "SkCanvas.h"
#include "SkCommonFlags.h"
#include "SkForceLinking.h"
#include "SkGraphics.h"
#include "SkOSFile.h"
#include "SkPictureRecorder.h"
#include "SkString.h"
#include "SkSurface.h"
#if SK_SUPPORT_GPU
#include "gl/GrGLDefines.h"
#include "GrContextFactory.h"
SkAutoTDelete<GrContextFactory> gGrFactory;
#endif
__SK_FORCE_IMAGE_DECODER_LINKING;
static const int kAutoTuneLoops = 0;
static const int kDefaultLoops =
#ifdef SK_DEBUG
1;
#else
kAutoTuneLoops;
#endif
static SkString loops_help_txt() {
SkString help;
help.printf("Number of times to run each bench. Set this to %d to auto-"
"tune for each bench. Timings are only reported when auto-tuning.",
kAutoTuneLoops);
return help;
}
DEFINE_int32(loops, kDefaultLoops, loops_help_txt().c_str());
DEFINE_int32(samples, 10, "Number of samples to measure for each bench.");
DEFINE_int32(overheadLoops, 100000, "Loops to estimate timer overhead.");
DEFINE_double(overheadGoal, 0.0001,
"Loop until timer overhead is at most this fraction of our measurments.");
DEFINE_double(gpuMs, 5, "Target bench time in millseconds for GPU.");
DEFINE_int32(gpuFrameLag, 5, "Overestimate of maximum number of frames GPU allows to lag.");
DEFINE_bool(gpuCompressAlphaMasks, false, "Compress masks generated from falling back to "
"software path rendering.");
DEFINE_string(outResultsFile, "", "If given, write results here as JSON.");
DEFINE_int32(maxCalibrationAttempts, 3,
"Try up to this many times to guess loops for a bench, or skip the bench.");
DEFINE_int32(maxLoops, 1000000, "Never run a bench more times than this.");
DEFINE_string(clip, "0,0,1000,1000", "Clip for SKPs.");
DEFINE_string(scales, "1.0", "Space-separated scales for SKPs.");
DEFINE_bool(bbh, true, "Build a BBH for SKPs?");
DEFINE_int32(benchTile, 256, "Tile dimension used for SKP playback.");
DEFINE_int32(flushEvery, 10, "Flush --outResultsFile every Nth run.");
static SkString humanize(double ms) {
if (FLAGS_verbose) return SkStringPrintf("%llu", (uint64_t)(ms*1e6));
if (ms > 1e+3) return SkStringPrintf("%.3gs", ms/1e3);
if (ms < 1e-3) return SkStringPrintf("%.3gns", ms*1e6);
#ifdef SK_BUILD_FOR_WIN
if (ms < 1) return SkStringPrintf("%.3gus", ms*1e3);
#else
if (ms < 1) return SkStringPrintf("%.3gµs", ms*1e3);
#endif
return SkStringPrintf("%.3gms", ms);
}
#define HUMANIZE(ms) humanize(ms).c_str()
static double time(int loops, Benchmark* bench, SkCanvas* canvas, SkGLContext* gl) {
if (canvas) {
canvas->clear(SK_ColorWHITE);
}
WallTimer timer;
timer.start();
if (bench) {
bench->draw(loops, canvas);
}
if (canvas) {
canvas->flush();
}
#if SK_SUPPORT_GPU
if (gl) {
SK_GL(*gl, Flush());
gl->swapBuffers();
}
#endif
timer.end();
return timer.fWall;
}
static double estimate_timer_overhead() {
double overhead = 0;
for (int i = 0; i < FLAGS_overheadLoops; i++) {
overhead += time(1, NULL, NULL, NULL);
}
return overhead / FLAGS_overheadLoops;
}
static int detect_forever_loops(int loops) {
// look for a magic run-forever value
if (loops < 0) {
loops = SK_MaxS32;
}
return loops;
}
static int clamp_loops(int loops) {
if (loops < 1) {
SkDebugf("ERROR: clamping loops from %d to 1.\n", loops);
return 1;
}
if (loops > FLAGS_maxLoops) {
SkDebugf("WARNING: clamping loops from %d to FLAGS_maxLoops, %d.\n", loops, FLAGS_maxLoops);
return FLAGS_maxLoops;
}
return loops;
}
static bool write_canvas_png(SkCanvas* canvas, const SkString& filename) {
if (filename.isEmpty()) {
return false;
}
if (kUnknown_SkColorType == canvas->imageInfo().colorType()) {
return false;
}
SkBitmap bmp;
bmp.setInfo(canvas->imageInfo());
if (!canvas->readPixels(&bmp, 0, 0)) {
SkDebugf("Can't read canvas pixels.\n");
return false;
}
SkString dir = SkOSPath::Dirname(filename.c_str());
if (!sk_mkdir(dir.c_str())) {
SkDebugf("Can't make dir %s.\n", dir.c_str());
return false;
}
SkFILEWStream stream(filename.c_str());
if (!stream.isValid()) {
SkDebugf("Can't write %s.\n", filename.c_str());
return false;
}
if (!SkImageEncoder::EncodeStream(&stream, bmp, SkImageEncoder::kPNG_Type, 100)) {
SkDebugf("Can't encode a PNG.\n");
return false;
}
return true;
}
static int kFailedLoops = -2;
static int cpu_bench(const double overhead, Benchmark* bench, SkCanvas* canvas, double* samples) {
// First figure out approximately how many loops of bench it takes to make overhead negligible.
double bench_plus_overhead = 0.0;
int round = 0;
if (kAutoTuneLoops == FLAGS_loops) {
while (bench_plus_overhead < overhead) {
if (round++ == FLAGS_maxCalibrationAttempts) {
SkDebugf("WARNING: Can't estimate loops for %s (%s vs. %s); skipping.\n",
bench->getUniqueName(), HUMANIZE(bench_plus_overhead), HUMANIZE(overhead));
return kFailedLoops;
}
bench_plus_overhead = time(1, bench, canvas, NULL);
}
}
// Later we'll just start and stop the timer once but loop N times.
// We'll pick N to make timer overhead negligible:
//
// overhead
// ------------------------- < FLAGS_overheadGoal
// overhead + N * Bench Time
//
// where bench_plus_overhead ≈ overhead + Bench Time.
//
// Doing some math, we get:
//
// (overhead / FLAGS_overheadGoal) - overhead
// ------------------------------------------ < N
// bench_plus_overhead - overhead)
//
// Luckily, this also works well in practice. :)
int loops = FLAGS_loops;
if (kAutoTuneLoops == loops) {
const double numer = overhead / FLAGS_overheadGoal - overhead;
const double denom = bench_plus_overhead - overhead;
loops = (int)ceil(numer / denom);
loops = clamp_loops(loops);
} else {
loops = detect_forever_loops(loops);
}
for (int i = 0; i < FLAGS_samples; i++) {
samples[i] = time(loops, bench, canvas, NULL) / loops;
}
return loops;
}
#if SK_SUPPORT_GPU
static int gpu_bench(SkGLContext* gl,
Benchmark* bench,
SkCanvas* canvas,
double* samples) {
gl->makeCurrent();
// Make sure we're done with whatever came before.
SK_GL(*gl, Finish());
// First, figure out how many loops it'll take to get a frame up to FLAGS_gpuMs.
int loops = FLAGS_loops;
if (kAutoTuneLoops == loops) {
loops = 1;
double elapsed = 0;
do {
loops *= 2;
// If the GPU lets frames lag at all, we need to make sure we're timing
// _this_ round, not still timing last round. We force this by looping
// more times than any reasonable GPU will allow frames to lag.
for (int i = 0; i < FLAGS_gpuFrameLag; i++) {
elapsed = time(loops, bench, canvas, gl);
}
} while (elapsed < FLAGS_gpuMs);
// We've overshot at least a little. Scale back linearly.
loops = (int)ceil(loops * FLAGS_gpuMs / elapsed);
loops = clamp_loops(loops);
// Might as well make sure we're not still timing our calibration.
SK_GL(*gl, Finish());
} else {
loops = detect_forever_loops(loops);
}
// Pretty much the same deal as the calibration: do some warmup to make
// sure we're timing steady-state pipelined frames.
for (int i = 0; i < FLAGS_gpuFrameLag; i++) {
time(loops, bench, canvas, gl);
}
// Now, actually do the timing!
for (int i = 0; i < FLAGS_samples; i++) {
samples[i] = time(loops, bench, canvas, gl) / loops;
}
return loops;
}
#endif
static SkString to_lower(const char* str) {
SkString lower(str);
for (size_t i = 0; i < lower.size(); i++) {
lower[i] = tolower(lower[i]);
}
return lower;
}
struct Config {
const char* name;
Benchmark::Backend backend;
SkColorType color;
SkAlphaType alpha;
int samples;
#if SK_SUPPORT_GPU
GrContextFactory::GLContextType ctxType;
#else
int bogusInt;
#endif
};
struct Target {
explicit Target(const Config& c) : config(c) {}
const Config config;
SkAutoTDelete<SkSurface> surface;
#if SK_SUPPORT_GPU
SkGLContext* gl;
#endif
};
static bool is_cpu_config_allowed(const char* name) {
for (int i = 0; i < FLAGS_config.count(); i++) {
if (to_lower(FLAGS_config[i]).equals(name)) {
return true;
}
}
return false;
}
#if SK_SUPPORT_GPU
static bool is_gpu_config_allowed(const char* name, GrContextFactory::GLContextType ctxType,
int sampleCnt) {
if (!is_cpu_config_allowed(name)) {
return false;
}
if (const GrContext* ctx = gGrFactory->get(ctxType)) {
return sampleCnt <= ctx->getMaxSampleCount();
}
return false;
}
#endif
#if SK_SUPPORT_GPU
#define kBogusGLContextType GrContextFactory::kNative_GLContextType
#else
#define kBogusGLContextType 0
#endif
// Append all configs that are enabled and supported.
static void create_configs(SkTDArray<Config>* configs) {
#define CPU_CONFIG(name, backend, color, alpha) \
if (is_cpu_config_allowed(#name)) { \
Config config = { #name, Benchmark::backend, color, alpha, 0, kBogusGLContextType }; \
configs->push(config); \
}
if (FLAGS_cpu) {
CPU_CONFIG(nonrendering, kNonRendering_Backend, kUnknown_SkColorType, kUnpremul_SkAlphaType)
CPU_CONFIG(8888, kRaster_Backend, kN32_SkColorType, kPremul_SkAlphaType)
CPU_CONFIG(565, kRaster_Backend, kRGB_565_SkColorType, kOpaque_SkAlphaType)
}
#if SK_SUPPORT_GPU
#define GPU_CONFIG(name, ctxType, samples) \
if (is_gpu_config_allowed(#name, GrContextFactory::ctxType, samples)) { \
Config config = { \
#name, \
Benchmark::kGPU_Backend, \
kN32_SkColorType, \
kPremul_SkAlphaType, \
samples, \
GrContextFactory::ctxType }; \
configs->push(config); \
}
if (FLAGS_gpu) {
GPU_CONFIG(gpu, kNative_GLContextType, 0)
GPU_CONFIG(msaa4, kNative_GLContextType, 4)
GPU_CONFIG(msaa16, kNative_GLContextType, 16)
GPU_CONFIG(nvprmsaa4, kNVPR_GLContextType, 4)
GPU_CONFIG(nvprmsaa16, kNVPR_GLContextType, 16)
GPU_CONFIG(debug, kDebug_GLContextType, 0)
GPU_CONFIG(nullgpu, kNull_GLContextType, 0)
#ifdef SK_ANGLE
GPU_CONFIG(angle, kANGLE_GLContextType, 0)
#endif
}
#endif
}
// If bench is enabled for config, returns a Target* for it, otherwise NULL.
static Target* is_enabled(Benchmark* bench, const Config& config) {
if (!bench->isSuitableFor(config.backend)) {
return NULL;
}
SkImageInfo info = SkImageInfo::Make(bench->getSize().fX, bench->getSize().fY,
config.color, config.alpha);
Target* target = new Target(config);
if (Benchmark::kRaster_Backend == config.backend) {
target->surface.reset(SkSurface::NewRaster(info));
}
#if SK_SUPPORT_GPU
else if (Benchmark::kGPU_Backend == config.backend) {
target->surface.reset(SkSurface::NewRenderTarget(gGrFactory->get(config.ctxType), info,
config.samples));
target->gl = gGrFactory->getGLContext(config.ctxType);
}
#endif
if (Benchmark::kNonRendering_Backend != config.backend && !target->surface.get()) {
delete target;
return NULL;
}
return target;
}
// Creates targets for a benchmark and a set of configs.
static void create_targets(SkTDArray<Target*>* targets, Benchmark* b,
const SkTDArray<Config>& configs) {
for (int i = 0; i < configs.count(); ++i) {
if (Target* t = is_enabled(b, configs[i])) {
targets->push(t);
}
}
}
#if SK_SUPPORT_GPU
static void fill_gpu_options(ResultsWriter* log, SkGLContext* ctx) {
const GrGLubyte* version;
SK_GL_RET(*ctx, version, GetString(GR_GL_VERSION));
log->configOption("GL_VERSION", (const char*)(version));
SK_GL_RET(*ctx, version, GetString(GR_GL_RENDERER));
log->configOption("GL_RENDERER", (const char*) version);
SK_GL_RET(*ctx, version, GetString(GR_GL_VENDOR));
log->configOption("GL_VENDOR", (const char*) version);
SK_GL_RET(*ctx, version, GetString(GR_GL_SHADING_LANGUAGE_VERSION));
log->configOption("GL_SHADING_LANGUAGE_VERSION", (const char*) version);
}
#endif
class BenchmarkStream {
public:
BenchmarkStream() : fBenches(BenchRegistry::Head())
, fGMs(skiagm::GMRegistry::Head())
, fCurrentRecording(0)
, fCurrentScale(0)
, fCurrentSKP(0) {
for (int i = 0; i < FLAGS_skps.count(); i++) {
if (SkStrEndsWith(FLAGS_skps[i], ".skp")) {
fSKPs.push_back() = FLAGS_skps[i];
} else {
SkOSFile::Iter it(FLAGS_skps[i], ".skp");
SkString path;
while (it.next(&path)) {
fSKPs.push_back() = SkOSPath::Join(FLAGS_skps[0], path.c_str());
}
}
}
if (4 != sscanf(FLAGS_clip[0], "%d,%d,%d,%d",
&fClip.fLeft, &fClip.fTop, &fClip.fRight, &fClip.fBottom)) {
SkDebugf("Can't parse %s from --clip as an SkIRect.\n", FLAGS_clip[0]);
exit(1);
}
for (int i = 0; i < FLAGS_scales.count(); i++) {
if (1 != sscanf(FLAGS_scales[i], "%f", &fScales.push_back())) {
SkDebugf("Can't parse %s from --scales as an SkScalar.\n", FLAGS_scales[i]);
exit(1);
}
}
}
static bool ReadPicture(const char* path, SkAutoTUnref<SkPicture>* pic) {
// Not strictly necessary, as it will be checked again later,
// but helps to avoid a lot of pointless work if we're going to skip it.
if (SkCommandLineFlags::ShouldSkip(FLAGS_match, path)) {
return false;
}
SkAutoTUnref<SkStream> stream(SkStream::NewFromFile(path));
if (stream.get() == NULL) {
SkDebugf("Could not read %s.\n", path);
return false;
}
pic->reset(SkPicture::CreateFromStream(stream.get()));
if (pic->get() == NULL) {
SkDebugf("Could not read %s as an SkPicture.\n", path);
return false;
}
return true;
}
Benchmark* next() {
if (fBenches) {
Benchmark* bench = fBenches->factory()(NULL);
fBenches = fBenches->next();
fSourceType = "bench";
fBenchType = "micro";
return bench;
}
while (fGMs) {
SkAutoTDelete<skiagm::GM> gm(fGMs->factory()(NULL));
fGMs = fGMs->next();
if (gm->getFlags() & skiagm::GM::kAsBench_Flag) {
fSourceType = "gm";
fBenchType = "micro";
return SkNEW_ARGS(GMBench, (gm.detach()));
}
}
// First add all .skps as RecordingBenches.
while (fCurrentRecording < fSKPs.count()) {
const SkString& path = fSKPs[fCurrentRecording++];
SkAutoTUnref<SkPicture> pic;
if (!ReadPicture(path.c_str(), &pic)) {
continue;
}
SkString name = SkOSPath::Basename(path.c_str());
fSourceType = "skp";
fBenchType = "recording";
return SkNEW_ARGS(RecordingBench, (name.c_str(), pic.get(), FLAGS_bbh));
}
// Then once each for each scale as SKPBenches (playback).
while (fCurrentScale < fScales.count()) {
while (fCurrentSKP < fSKPs.count()) {
const SkString& path = fSKPs[fCurrentSKP++];
SkAutoTUnref<SkPicture> pic;
if (!ReadPicture(path.c_str(), &pic)) {
continue;
}
if (FLAGS_bbh) {
// The SKP we read off disk doesn't have a BBH. Re-record so it grows one.
const SkTileGridFactory::TileGridInfo info = {
SkISize::Make(FLAGS_benchTile, FLAGS_benchTile), // tile interval
SkISize::Make(0,0), // margin
SkIPoint::Make(0,0), // offset
};
SkTileGridFactory factory(info);
SkPictureRecorder recorder;
pic->playback(recorder.beginRecording(pic->cullRect().width(),
pic->cullRect().height(),
&factory));
pic.reset(recorder.endRecording());
}
SkString name = SkOSPath::Basename(path.c_str());
fSourceType = "skp";
fBenchType = "playback";
return SkNEW_ARGS(SKPBench,
(name.c_str(), pic.get(), fClip, fScales[fCurrentScale]));
}
fCurrentSKP = 0;
fCurrentScale++;
}
return NULL;
}
void fillCurrentOptions(ResultsWriter* log) const {
log->configOption("source_type", fSourceType);
log->configOption("bench_type", fBenchType);
if (0 == strcmp(fSourceType, "skp")) {
log->configOption("clip",
SkStringPrintf("%d %d %d %d", fClip.fLeft, fClip.fTop,
fClip.fRight, fClip.fBottom).c_str());
log->configOption("scale", SkStringPrintf("%.2g", fScales[fCurrentScale]).c_str());
}
}
private:
const BenchRegistry* fBenches;
const skiagm::GMRegistry* fGMs;
SkIRect fClip;
SkTArray<SkScalar> fScales;
SkTArray<SkString> fSKPs;
const char* fSourceType; // What we're benching: bench, GM, SKP, ...
const char* fBenchType; // How we bench it: micro, recording, playback, ...
int fCurrentRecording;
int fCurrentScale;
int fCurrentSKP;
};
int nanobench_main();
int nanobench_main() {
SetupCrashHandler();
SkAutoGraphics ag;
#if SK_SUPPORT_GPU
GrContext::Options grContextOpts;
grContextOpts.fDrawPathToCompressedTexture = FLAGS_gpuCompressAlphaMasks;
gGrFactory.reset(SkNEW_ARGS(GrContextFactory, (grContextOpts)));
#endif
if (FLAGS_veryVerbose) {
FLAGS_verbose = true;
}
if (kAutoTuneLoops != FLAGS_loops) {
FLAGS_samples = 1;
FLAGS_gpuFrameLag = 0;
}
if (!FLAGS_writePath.isEmpty()) {
SkDebugf("Writing files to %s.\n", FLAGS_writePath[0]);
if (!sk_mkdir(FLAGS_writePath[0])) {
SkDebugf("Could not create %s. Files won't be written.\n", FLAGS_writePath[0]);
FLAGS_writePath.set(0, NULL);
}
}
SkAutoTDelete<ResultsWriter> log(SkNEW(ResultsWriter));
if (!FLAGS_outResultsFile.isEmpty()) {
log.reset(SkNEW(NanoJSONResultsWriter(FLAGS_outResultsFile[0])));
}
if (1 == FLAGS_properties.count() % 2) {
SkDebugf("ERROR: --properties must be passed with an even number of arguments.\n");
return 1;
}
for (int i = 1; i < FLAGS_properties.count(); i += 2) {
log->property(FLAGS_properties[i-1], FLAGS_properties[i]);
}
if (1 == FLAGS_key.count() % 2) {
SkDebugf("ERROR: --key must be passed with an even number of arguments.\n");
return 1;
}
for (int i = 1; i < FLAGS_key.count(); i += 2) {
log->key(FLAGS_key[i-1], FLAGS_key[i]);
}
const double overhead = estimate_timer_overhead();
SkDebugf("Timer overhead: %s\n", HUMANIZE(overhead));
SkAutoTMalloc<double> samples(FLAGS_samples);
if (kAutoTuneLoops != FLAGS_loops) {
SkDebugf("Fixed number of loops; times would only be misleading so we won't print them.\n");
} else if (FLAGS_verbose) {
// No header.
} else if (FLAGS_quiet) {
SkDebugf("median\tbench\tconfig\n");
} else {
SkDebugf("maxrss\tloops\tmin\tmedian\tmean\tmax\tstddev\t%-*s\tconfig\tbench\n",
FLAGS_samples, "samples");
}
SkTDArray<Config> configs;
create_configs(&configs);
int runs = 0;
BenchmarkStream benchStream;
while (Benchmark* b = benchStream.next()) {
SkAutoTDelete<Benchmark> bench(b);
if (SkCommandLineFlags::ShouldSkip(FLAGS_match, bench->getUniqueName())) {
continue;
}
SkTDArray<Target*> targets;
create_targets(&targets, bench.get(), configs);
if (!targets.isEmpty()) {
log->bench(bench->getUniqueName(), bench->getSize().fX, bench->getSize().fY);
bench->preDraw();
}
for (int j = 0; j < targets.count(); j++) {
SkCanvas* canvas = targets[j]->surface.get() ? targets[j]->surface->getCanvas() : NULL;
const char* config = targets[j]->config.name;
const int loops =
#if SK_SUPPORT_GPU
Benchmark::kGPU_Backend == targets[j]->config.backend
? gpu_bench(targets[j]->gl, bench.get(), canvas, samples.get())
:
#endif
cpu_bench( overhead, bench.get(), canvas, samples.get());
if (canvas && !FLAGS_writePath.isEmpty() && FLAGS_writePath[0]) {
SkString pngFilename = SkOSPath::Join(FLAGS_writePath[0], config);
pngFilename = SkOSPath::Join(pngFilename.c_str(), bench->getUniqueName());
pngFilename.append(".png");
write_canvas_png(canvas, pngFilename);
}
if (kFailedLoops == loops) {
// Can't be timed. A warning note has already been printed.
continue;
}
Stats stats(samples.get(), FLAGS_samples);
log->config(config);
log->configOption("name", bench->getName());
benchStream.fillCurrentOptions(log.get());
#if SK_SUPPORT_GPU
if (Benchmark::kGPU_Backend == targets[j]->config.backend) {
fill_gpu_options(log.get(), targets[j]->gl);
}
#endif
log->timer("min_ms", stats.min);
log->timer("median_ms", stats.median);
log->timer("mean_ms", stats.mean);
log->timer("max_ms", stats.max);
log->timer("stddev_ms", sqrt(stats.var));
if (runs++ % FLAGS_flushEvery == 0) {
log->flush();
}
if (kAutoTuneLoops != FLAGS_loops) {
if (targets.count() == 1) {
config = ""; // Only print the config if we run the same bench on more than one.
}
SkDebugf("%4dM\t%s\t%s\n"
, sk_tools::getMaxResidentSetSizeMB()
, bench->getUniqueName()
, config);
} else if (FLAGS_verbose) {
for (int i = 0; i < FLAGS_samples; i++) {
SkDebugf("%s ", HUMANIZE(samples[i]));
}
SkDebugf("%s\n", bench->getUniqueName());
} else if (FLAGS_quiet) {
if (targets.count() == 1) {
config = ""; // Only print the config if we run the same bench on more than one.
}
SkDebugf("%s\t%s\t%s\n", HUMANIZE(stats.median), bench->getUniqueName(), config);
} else {
const double stddev_percent = 100 * sqrt(stats.var) / stats.mean;
SkDebugf("%4dM\t%d\t%s\t%s\t%s\t%s\t%.0f%%\t%s\t%s\t%s\n"
, sk_tools::getMaxResidentSetSizeMB()
, loops
, HUMANIZE(stats.min)
, HUMANIZE(stats.median)
, HUMANIZE(stats.mean)
, HUMANIZE(stats.max)
, stddev_percent
, stats.plot.c_str()
, config
, bench->getUniqueName()
);
}
#if SK_SUPPORT_GPU && GR_CACHE_STATS
if (FLAGS_veryVerbose &&
Benchmark::kGPU_Backend == targets[j]->config.backend) {
gGrFactory->get(targets[j]->config.ctxType)->printCacheStats();
}
#endif
}
targets.deleteAll();
#if SK_SUPPORT_GPU
if (FLAGS_abandonGpuContext) {
gGrFactory->abandonContexts();
}
if (FLAGS_resetGpuContext || FLAGS_abandonGpuContext) {
gGrFactory->destroyContexts();
}
#endif
}
return 0;
}
#if !defined SK_BUILD_FOR_IOS
int main(int argc, char** argv) {
SkCommandLineFlags::Parse(argc, argv);
return nanobench_main();
}
#endif
|