aboutsummaryrefslogtreecommitdiffhomepage
path: root/bench/benchmain.cpp
blob: 3080a0f022e8a20e522ba6332cb2afa84e80dd8a (plain)
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
/*
 * Copyright 2011 Google Inc.
 *
 * Use of this source code is governed by a BSD-style license that can be
 * found in the LICENSE file.
 */

#include "BenchTimer.h"
#include "ResultsWriter.h"
#include "SkBenchLogger.h"
#include "SkBenchmark.h"
#include "SkBitmapDevice.h"
#include "SkCanvas.h"
#include "SkColorPriv.h"
#include "SkCommandLineFlags.h"
#include "SkData.h"
#include "SkDeferredCanvas.h"
#include "SkGMBench.h"
#include "SkGraphics.h"
#include "SkImageEncoder.h"
#include "SkOSFile.h"
#include "SkPicture.h"
#include "SkPictureRecorder.h"
#include "SkString.h"
#include "SkSurface.h"

#if SK_SUPPORT_GPU
#include "GrContext.h"
#include "GrContextFactory.h"
#include "GrRenderTarget.h"
#include "SkGpuDevice.h"
#include "gl/GrGLDefines.h"
#else
class GrContext;
#endif // SK_SUPPORT_GPU

#include <limits>

enum BenchMode {
    kNormal_BenchMode,
    kDeferred_BenchMode,
    kDeferredSilent_BenchMode,
    kRecord_BenchMode,
    kPictureRecord_BenchMode
};
const char* BenchMode_Name[] = {
    "normal", "deferred", "deferredSilent", "record", "picturerecord"
};

static const char kDefaultsConfigStr[] = "defaults";

///////////////////////////////////////////////////////////////////////////////

class Iter {
public:
    Iter() : fBenches(BenchRegistry::Head()), fGMs(skiagm::GMRegistry::Head()) {}

    SkBenchmark* next() {
        if (fBenches) {
            BenchRegistry::Factory f = fBenches->factory();
            fBenches = fBenches->next();
            return (*f)(NULL);
        }

        while (fGMs) {
            SkAutoTDelete<skiagm::GM> gm(fGMs->factory()(NULL));
            fGMs = fGMs->next();
            if (gm->getFlags() & skiagm::GM::kAsBench_Flag) {
                return SkNEW_ARGS(SkGMBench, (gm.detach()));
            }
        }

        return NULL;
    }

private:
    const BenchRegistry* fBenches;
    const skiagm::GMRegistry* fGMs;
};

class AutoPrePostDraw {
public:
    AutoPrePostDraw(SkBenchmark* bench) : fBench(bench) {
        fBench->preDraw();
    }
    ~AutoPrePostDraw() {
        fBench->postDraw();
    }
private:
    SkBenchmark* fBench;
};

static void make_filename(const char name[], SkString* path) {
    path->set(name);
    for (int i = 0; name[i]; i++) {
        switch (name[i]) {
            case '/':
            case '\\':
            case ' ':
            case ':':
                path->writable_str()[i] = '-';
                break;
            default:
                break;
        }
    }
}

static void saveFile(const char name[], const char config[], const char dir[],
                     const SkImage* image) {
    SkAutoTUnref<SkData> data(image->encode(SkImageEncoder::kPNG_Type, 100));
    if (NULL == data.get()) {
        return;
    }

    SkString filename;
    make_filename(name, &filename);
    filename.appendf("_%s.png", config);
    SkString path = SkOSPath::SkPathJoin(dir, filename.c_str());
    ::remove(path.c_str());

    SkFILEWStream   stream(path.c_str());
    stream.write(data->data(), data->size());
}

static void perform_clip(SkCanvas* canvas, int w, int h) {
    SkRect r;

    r.set(SkIntToScalar(10), SkIntToScalar(10),
          SkIntToScalar(w*2/3), SkIntToScalar(h*2/3));
    canvas->clipRect(r, SkRegion::kIntersect_Op);

    r.set(SkIntToScalar(w/3), SkIntToScalar(h/3),
          SkIntToScalar(w-10), SkIntToScalar(h-10));
    canvas->clipRect(r, SkRegion::kXOR_Op);
}

static void perform_rotate(SkCanvas* canvas, int w, int h) {
    const SkScalar x = SkIntToScalar(w) / 2;
    const SkScalar y = SkIntToScalar(h) / 2;

    canvas->translate(x, y);
    canvas->rotate(SkIntToScalar(35));
    canvas->translate(-x, -y);
}

static void perform_scale(SkCanvas* canvas, int w, int h) {
    const SkScalar x = SkIntToScalar(w) / 2;
    const SkScalar y = SkIntToScalar(h) / 2;

    canvas->translate(x, y);
    // just enough so we can't take the sprite case
    canvas->scale(SK_Scalar1 * 99/100, SK_Scalar1 * 99/100);
    canvas->translate(-x, -y);
}

static SkSurface* make_surface(SkColorType colorType, const SkIPoint& size,
                               SkBenchmark::Backend backend, int sampleCount,
                               GrContext* context) {
    SkSurface* surface = NULL;
    SkImageInfo info = SkImageInfo::Make(size.fX, size.fY, colorType,
                                         kPremul_SkAlphaType);

    switch (backend) {
        case SkBenchmark::kRaster_Backend:
            surface = SkSurface::NewRaster(info);
            surface->getCanvas()->clear(SK_ColorWHITE);
            break;
#if SK_SUPPORT_GPU
        case SkBenchmark::kGPU_Backend: {
            surface = SkSurface::NewRenderTarget(context, info, sampleCount);
            break;
        }
#endif
        case SkBenchmark::kPDF_Backend:
        default:
            SkDEBUGFAIL("unsupported");
    }
    return surface;
}

#if SK_SUPPORT_GPU
GrContextFactory gContextFactory;
typedef GrContextFactory::GLContextType GLContextType;
static const GLContextType kNative = GrContextFactory::kNative_GLContextType;
static const GLContextType kNVPR   = GrContextFactory::kNVPR_GLContextType;
#if SK_ANGLE
static const GLContextType kANGLE  = GrContextFactory::kANGLE_GLContextType;
#endif
static const GLContextType kDebug  = GrContextFactory::kDebug_GLContextType;
static const GLContextType kNull   = GrContextFactory::kNull_GLContextType;
#else
typedef int GLContextType;
static const GLContextType kNative = 0, kANGLE = 0, kDebug = 0, kNull = 0;
#endif

#ifdef SK_DEBUG
static const bool kIsDebug = true;
#else
static const bool kIsDebug = false;
#endif

static const struct Config {
    SkColorType         fColorType;
    const char*         name;
    int                 sampleCount;
    SkBenchmark::Backend backend;
    GLContextType       contextType;
    bool                runByDefault;
} gConfigs[] = {
    { kN32_SkColorType,     "NONRENDERING", 0, SkBenchmark::kNonRendering_Backend, kNative, true},
    { kN32_SkColorType,     "8888",         0, SkBenchmark::kRaster_Backend,       kNative, true},
    { kRGB_565_SkColorType, "565",          0, SkBenchmark::kRaster_Backend,       kNative, true},
#if SK_SUPPORT_GPU
    { kN32_SkColorType,     "GPU",          0, SkBenchmark::kGPU_Backend,          kNative, true},
    { kN32_SkColorType,     "MSAA4",        4, SkBenchmark::kGPU_Backend,          kNative, false},
    { kN32_SkColorType,     "MSAA16",      16, SkBenchmark::kGPU_Backend,          kNative, false},
    { kN32_SkColorType,     "NVPRMSAA4",    4, SkBenchmark::kGPU_Backend,          kNVPR,   true},
    { kN32_SkColorType,     "NVPRMSAA16",  16, SkBenchmark::kGPU_Backend,          kNVPR,   false},
#if SK_ANGLE
    { kN32_SkColorType,     "ANGLE",        0, SkBenchmark::kGPU_Backend,          kANGLE,  true},
#endif // SK_ANGLE
    { kN32_SkColorType,     "Debug",        0, SkBenchmark::kGPU_Backend,          kDebug,  kIsDebug},
    { kN32_SkColorType,     "NULLGPU",      0, SkBenchmark::kGPU_Backend,          kNull,   true},
#endif // SK_SUPPORT_GPU
};

DEFINE_string(outDir, "", "If given, image of each bench will be put in outDir.");
DEFINE_string(timers, "cg", "Timers to display. "
              "Options: w(all) W(all, truncated) c(pu) C(pu, truncated) g(pu)");

DEFINE_bool(rotate, false,  "Rotate canvas before bench run?");
DEFINE_bool(scale,  false,  "Scale canvas before bench run?");
DEFINE_bool(clip,   false,  "Clip canvas before bench run?");

DEFINE_bool(forceAA,        true,     "Force anti-aliasing?");
DEFINE_bool(forceFilter,    false,    "Force bitmap filtering?");
DEFINE_string(forceDither, "default", "Force dithering: true, false, or default?");
DEFINE_bool(forceBlend,     false,    "Force alpha blending?");

DEFINE_int32(gpuCacheBytes, -1, "GPU cache size limit in bytes.  0 to disable cache.");
DEFINE_int32(gpuCacheCount, -1, "GPU cache size limit in object count.  0 to disable cache.");

DEFINE_bool2(leaks, l, false, "show leaked ref cnt'd objects.");
DEFINE_string(match, "",  "[~][^]substring[$] [...] of test name to run.\n"
                          "Multiple matches may be separated by spaces.\n"
                          "~ causes a matching test to always be skipped\n"
                          "^ requires the start of the test to match\n"
                          "$ requires the end of the test to match\n"
                          "^ and $ requires an exact match\n"
                          "If a test does not match any list entry,\n"
                          "it is skipped unless some list entry starts with ~\n");
DEFINE_string(mode, "normal",
             "normal:         draw to a normal canvas;\n"
             "deferred:       draw to a deferred canvas;\n"
             "deferredSilent: deferred with silent playback;\n"
             "record:         draw to an SkPicture;\n"
             "picturerecord:  draw from an SkPicture to an SkPicture.\n");
DEFINE_string(config, kDefaultsConfigStr,
              "Run configs given.  By default, runs the configs marked \"runByDefault\" in gConfigs.");
DEFINE_string(logFile, "", "Also write stdout here.");
DEFINE_int32(minMs, 20,  "Shortest time we'll allow a benchmark to run.");
DEFINE_int32(maxMs, 4000, "Longest time we'll allow a benchmark to run.");
DEFINE_bool(runOnce, kIsDebug, "Run each bench exactly once and don't report timings.");
DEFINE_double(error, 0.01,
              "Ratio of subsequent bench measurements must drop within 1±error to converge.");
DEFINE_string(timeFormat, "%9.2f", "Format to print results, in milliseconds per 1000 loops.");
DEFINE_bool2(verbose, v, false, "Print more.");
DEFINE_string2(resourcePath, i, "resources", "directory for test resources.");
DEFINE_string(outResultsFile, "", "If given, the results will be written to the file in JSON format.");

DEFINE_bool(dryRun, false, "Don't actually run the tests, just print what would have been done.");

// Has this bench converged?  First arguments are milliseconds / loop iteration,
// last is overall runtime in milliseconds.
static bool HasConverged(double prevPerLoop, double currPerLoop, double currRaw) {
    if (currRaw < FLAGS_minMs) {
        return false;
    }
    const double low = 1 - FLAGS_error, high = 1 + FLAGS_error;
    const double ratio = currPerLoop / prevPerLoop;
    return low < ratio && ratio < high;
}

int tool_main(int argc, char** argv);
int tool_main(int argc, char** argv) {
    SkCommandLineFlags::Parse(argc, argv);
#if SK_ENABLE_INST_COUNT
    if (FLAGS_leaks) {
        gPrintInstCount = true;
    }
#endif
    SkAutoGraphics ag;

    // First, parse some flags.
    SkBenchLogger logger;
    if (FLAGS_logFile.count()) {
        logger.SetLogFile(FLAGS_logFile[0]);
    }

    LoggerResultsWriter logWriter(logger, FLAGS_timeFormat[0]);
    MultiResultsWriter writer;
    writer.add(&logWriter);
    SkAutoTDelete<JSONResultsWriter> jsonWriter;
    if (FLAGS_outResultsFile.count()) {
        jsonWriter.reset(SkNEW(JSONResultsWriter(FLAGS_outResultsFile[0])));
        writer.add(jsonWriter.get());
    }
    // Instantiate after all the writers have been added to writer so that we
    // call close() before their destructors are called on the way out.
    CallEnd<MultiResultsWriter> ender(writer);

    const uint8_t alpha = FLAGS_forceBlend ? 0x80 : 0xFF;
    SkTriState::State dither = SkTriState::kDefault;
    for (size_t i = 0; i < 3; i++) {
        if (strcmp(SkTriState::Name[i], FLAGS_forceDither[0]) == 0) {
            dither = static_cast<SkTriState::State>(i);
        }
    }

    BenchMode benchMode = kNormal_BenchMode;
    for (size_t i = 0; i < SK_ARRAY_COUNT(BenchMode_Name); i++) {
        if (strcmp(FLAGS_mode[0], BenchMode_Name[i]) == 0) {
            benchMode = static_cast<BenchMode>(i);
        }
    }

    SkTDArray<int> configs;
    bool runDefaultConfigs = false;
    // Try user-given configs first.
    for (int i = 0; i < FLAGS_config.count(); i++) {
        for (int j = 0; j < static_cast<int>(SK_ARRAY_COUNT(gConfigs)); ++j) {
            if (0 == strcmp(FLAGS_config[i], gConfigs[j].name)) {
                *configs.append() = j;
            } else if (0 == strcmp(FLAGS_config[i], kDefaultsConfigStr)) {
                runDefaultConfigs = true;
            }
        }
    }
    // If there weren't any, fill in with defaults.
    if (runDefaultConfigs) {
        for (int i = 0; i < static_cast<int>(SK_ARRAY_COUNT(gConfigs)); ++i) {
            if (gConfigs[i].runByDefault) {
                *configs.append() = i;
            }
        }
    }
    // Filter out things we can't run.
    if (kNormal_BenchMode != benchMode) {
        // Non-rendering configs only run in normal mode
        for (int i = 0; i < configs.count(); ++i) {
            const Config& config = gConfigs[configs[i]];
            if (SkBenchmark::kNonRendering_Backend == config.backend) {
                configs.remove(i, 1);
                --i;
            }
        }
    }
    // Set the resource path.
    if (!FLAGS_resourcePath.isEmpty()) {
        SkBenchmark::SetResourcePath(FLAGS_resourcePath[0]);
    }

#if SK_SUPPORT_GPU
    for (int i = 0; i < configs.count(); ++i) {
        const Config& config = gConfigs[configs[i]];

        if (SkBenchmark::kGPU_Backend == config.backend) {
            GrContext* context = gContextFactory.get(config.contextType);
            if (NULL == context) {
                SkDebugf("GrContext could not be created for config %s. Config will be skipped.\n",
                    config.name);
                configs.remove(i);
                --i;
                continue;
            }
            if (config.sampleCount > context->getMaxSampleCount()){
                SkDebugf(
                    "Sample count (%d) for config %s is not supported. Config will be skipped.\n",
                    config.sampleCount, config.name);
                configs.remove(i);
                --i;
                continue;
            }
        }
    }
#endif

    // All flags should be parsed now.  Report our settings.
    if (FLAGS_runOnce) {
        logger.logError("bench was run with --runOnce, so we're going to hide the times."
                        " It's for your own good!\n");
    }
    writer.option("mode", FLAGS_mode[0]);
    writer.option("alpha", SkStringPrintf("0x%02X", alpha).c_str());
    writer.option("antialias", SkStringPrintf("%d", FLAGS_forceAA).c_str());
    writer.option("filter", SkStringPrintf("%d", FLAGS_forceFilter).c_str());
    writer.option("dither",  SkTriState::Name[dither]);

    writer.option("rotate", SkStringPrintf("%d", FLAGS_rotate).c_str());
    writer.option("scale", SkStringPrintf("%d", FLAGS_scale).c_str());
    writer.option("clip", SkStringPrintf("%d", FLAGS_clip).c_str());

#if defined(SK_BUILD_FOR_WIN32)
    writer.option("system", "WIN32");
#elif defined(SK_BUILD_FOR_MAC)
    writer.option("system", "MAC");
#elif defined(SK_BUILD_FOR_ANDROID)
    writer.option("system", "ANDROID");
#elif defined(SK_BUILD_FOR_UNIX)
    writer.option("system", "UNIX");
#else
    writer.option("system", "other");
#endif

#if defined(SK_DEBUG)
    writer.option("build", "DEBUG");
#else
    writer.option("build", "RELEASE");
#endif

    // Set texture cache limits if non-default.
    for (size_t i = 0; i < SK_ARRAY_COUNT(gConfigs); ++i) {
#if SK_SUPPORT_GPU
        const Config& config = gConfigs[i];
        if (SkBenchmark::kGPU_Backend != config.backend) {
            continue;
        }
        GrContext* context = gContextFactory.get(config.contextType);
        if (NULL == context) {
            continue;
        }

        size_t bytes;
        int count;
        context->getTextureCacheLimits(&count, &bytes);
        if (-1 != FLAGS_gpuCacheBytes) {
            bytes = static_cast<size_t>(FLAGS_gpuCacheBytes);
        }
        if (-1 != FLAGS_gpuCacheCount) {
            count = FLAGS_gpuCacheCount;
        }
        context->setTextureCacheLimits(count, bytes);
#endif
    }

    // Run each bench in each configuration it supports and we asked for.
    Iter iter;
    SkBenchmark* bench;
    while ((bench = iter.next()) != NULL) {
        SkAutoTUnref<SkBenchmark> benchUnref(bench);
        if (SkCommandLineFlags::ShouldSkip(FLAGS_match, bench->getName())) {
            continue;
        }

        bench->setForceAlpha(alpha);
        bench->setForceAA(FLAGS_forceAA);
        bench->setForceFilter(FLAGS_forceFilter);
        bench->setDither(dither);
        AutoPrePostDraw appd(bench);

        bool loggedBenchName = false;
        for (int i = 0; i < configs.count(); ++i) {
            const int configIndex = configs[i];
            const Config& config = gConfigs[configIndex];

            if (!bench->isSuitableFor(config.backend)) {
                continue;
            }

            GrContext* context = NULL;
#if SK_SUPPORT_GPU
            SkGLContextHelper* glContext = NULL;
            if (SkBenchmark::kGPU_Backend == config.backend) {
                context = gContextFactory.get(config.contextType);
                if (NULL == context) {
                    continue;
                }
                glContext = gContextFactory.getGLContext(config.contextType);
            }
#endif

            SkAutoTUnref<SkCanvas> canvas;
            SkAutoTUnref<SkPicture> recordFrom;
            SkPictureRecorder recorderTo;
            const SkIPoint dim = bench->getSize();

            const SkPicture::RecordingFlags kRecordFlags =
                SkPicture::kUsePathBoundsForClip_RecordingFlag;

            SkAutoTUnref<SkSurface> surface;
            if (SkBenchmark::kNonRendering_Backend != config.backend) {
                surface.reset(make_surface(config.fColorType,
                                           dim,
                                           config.backend,
                                           config.sampleCount,
                                           context));
                if (!surface.get()) {
                    logger.logError(SkStringPrintf(
                        "Device creation failure for config %s. Will skip.\n", config.name));
                    continue;
                }

                switch(benchMode) {
                    case kDeferredSilent_BenchMode:
                    case kDeferred_BenchMode:
                        canvas.reset(SkDeferredCanvas::Create(surface.get()));
                        break;
                    case kRecord_BenchMode:
                        canvas.reset(SkRef(recorderTo.beginRecording(dim.fX, dim.fY,
                                                                     NULL, kRecordFlags)));
                        break;
                    case kPictureRecord_BenchMode: {
                        SkPictureRecorder recorderFrom;
                        bench->draw(1, recorderFrom.beginRecording(dim.fX, dim.fY,
                                                                   NULL, kRecordFlags));
                        recordFrom.reset(recorderFrom.endRecording());
                        canvas.reset(SkRef(recorderTo.beginRecording(dim.fX, dim.fY,
                                                                     NULL, kRecordFlags)));
                        break;
                    }
                    case kNormal_BenchMode:
                        canvas.reset(SkRef(surface->getCanvas()));
                        break;
                    default:
                        SkASSERT(false);
                }
            }

            if (NULL != canvas) {
                canvas->clear(SK_ColorWHITE);
                if (FLAGS_clip)   {
                    perform_clip(canvas, dim.fX, dim.fY);
                }
                if (FLAGS_scale)  {
                    perform_scale(canvas, dim.fX, dim.fY);
                }
                if (FLAGS_rotate) {
                    perform_rotate(canvas, dim.fX, dim.fY);
                }
            }

            if (!loggedBenchName) {
                loggedBenchName = true;
                writer.bench(bench->getName(), dim.fX, dim.fY);
            }

#if SK_SUPPORT_GPU
            SkGLContextHelper* contextHelper = NULL;
            if (SkBenchmark::kGPU_Backend == config.backend) {
                contextHelper = gContextFactory.getGLContext(config.contextType);
            }
            BenchTimer timer(contextHelper);
#else
            BenchTimer timer;
#endif

            double previous = std::numeric_limits<double>::infinity();
            bool converged = false;

            // variables used to compute loopsPerFrame
            double frameIntervalTime = 0.0f;
            int frameIntervalTotalLoops = 0;

            bool frameIntervalComputed = false;
            int loopsPerFrame = 0;
            int loopsPerIter = 0;
            if (FLAGS_verbose) { SkDebugf("%s %s: ", bench->getName(), config.name); }
            if (!FLAGS_dryRun) {
                do {
                    // Ramp up 1 -> 2 -> 4 -> 8 -> 16 -> ... -> ~1 billion.
                    loopsPerIter = (loopsPerIter == 0) ? 1 : loopsPerIter * 2;
                    if (loopsPerIter >= (1<<30) || timer.fWall > FLAGS_maxMs) {
                        // If you find it takes more than a billion loops to get up to 20ms of runtime,
                        // you've got a computer clocked at several THz or have a broken benchmark.  ;)
                        //     "1B ought to be enough for anybody."
                        logger.logError(SkStringPrintf(
                            "\nCan't get %s %s to converge in %dms (%d loops)",
                             bench->getName(), config.name, FLAGS_maxMs, loopsPerIter));
                        break;
                    }

                    if ((benchMode == kRecord_BenchMode || benchMode == kPictureRecord_BenchMode)) {
                        // Clear the recorded commands so that they do not accumulate.
                        canvas.reset(SkRef(recorderTo.beginRecording(dim.fX, dim.fY,
                                                                     NULL, kRecordFlags)));
                    }

                    timer.start();
                    // Inner loop that allows us to break the run into smaller
                    // chunks (e.g. frames). This is especially useful for the GPU
                    // as we can flush and/or swap buffers to keep the GPU from
                    // queuing up too much work.
                    for (int loopCount = loopsPerIter; loopCount > 0; ) {
                        // Save and restore around each call to draw() to guarantee a pristine canvas.
                        SkAutoCanvasRestore saveRestore(canvas, true/*also save*/);

                        int loops;
                        if (frameIntervalComputed && loopCount > loopsPerFrame) {
                            loops = loopsPerFrame;
                            loopCount -= loopsPerFrame;
                        } else {
                            loops = loopCount;
                            loopCount = 0;
                        }

                        if (benchMode == kPictureRecord_BenchMode) {
                            recordFrom->draw(canvas);
                        } else {
                            bench->draw(loops, canvas);
                        }

                        if (kDeferredSilent_BenchMode == benchMode) {
                            static_cast<SkDeferredCanvas*>(canvas.get())->silentFlush();
                        } else if (NULL != canvas) {
                            canvas->flush();
                        }

    #if SK_SUPPORT_GPU
                        // swap drawing buffers on each frame to prevent the GPU
                        // from queuing up too much work
                        if (NULL != glContext) {
                            glContext->swapBuffers();
                        }
    #endif
                    }



                    // Stop truncated timers before GL calls complete, and stop the full timers after.
                    timer.truncatedEnd();
    #if SK_SUPPORT_GPU
                    if (NULL != glContext) {
                        context->flush();
                        SK_GL(*glContext, Finish());
                    }
    #endif
                    timer.end();

                    // setup the frame interval for subsequent iterations
                    if (!frameIntervalComputed) {
                        frameIntervalTime += timer.fWall;
                        frameIntervalTotalLoops += loopsPerIter;
                        if (frameIntervalTime >= FLAGS_minMs) {
                            frameIntervalComputed = true;
                            loopsPerFrame =
                              (int)(((double)frameIntervalTotalLoops / frameIntervalTime) * FLAGS_minMs);
                            if (loopsPerFrame < 1) {
                                loopsPerFrame = 1;
                            }
    //                        SkDebugf("  %s has %d loops in %f ms (normalized to %d)\n",
    //                                 bench->getName(), frameIntervalTotalLoops,
    //                                 timer.fWall, loopsPerFrame);
                        }
                    }

                    const double current = timer.fWall / loopsPerIter;
                    if (FLAGS_verbose && current > previous) { SkDebugf("↑"); }
                    if (FLAGS_verbose) { SkDebugf("%.3g ", current); }
                    converged = HasConverged(previous, current, timer.fWall);
                    previous = current;
                } while (!FLAGS_runOnce && !converged);
            }
            if (FLAGS_verbose) { SkDebugf("\n"); }

            if (!FLAGS_dryRun && FLAGS_outDir.count() && SkBenchmark::kNonRendering_Backend != config.backend) {
                SkAutoTUnref<SkImage> image(surface->newImageSnapshot());
                if (image.get()) {
                    saveFile(bench->getName(), config.name, FLAGS_outDir[0],
                             image);
                }
            }

            if (FLAGS_runOnce) {
                // Let's not mislead ourselves by looking at Debug build or single iteration bench times!
                continue;
            }

            // Normalize to ms per 1000 iterations.
            const double normalize = 1000.0 / loopsPerIter;
            const struct { char shortName; const char* longName; double ms; } times[] = {
                {'w', "msecs",  normalize * timer.fWall},
                {'W', "Wmsecs", normalize * timer.fTruncatedWall},
                {'c', "cmsecs", normalize * timer.fCpu},
                {'C', "Cmsecs", normalize * timer.fTruncatedCpu},
                {'g', "gmsecs", normalize * timer.fGpu},
            };

            writer.config(config.name);
            for (size_t i = 0; i < SK_ARRAY_COUNT(times); i++) {
                if (strchr(FLAGS_timers[0], times[i].shortName) && times[i].ms > 0) {
                    writer.timer(times[i].longName, times[i].ms);
                }
            }
        }
    }
#if SK_SUPPORT_GPU
    gContextFactory.destroyContexts();
#endif
    return 0;
}

#if !defined(SK_BUILD_FOR_IOS) && !defined(SK_BUILD_FOR_NACL)
int main(int argc, char * const argv[]) {
    return tool_main(argc, (char**) argv);
}
#endif