skia2/bench/nanobench.cpp

806 lines
28 KiB
C++
Raw Normal View History

/*
* 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 "SkBBoxHierarchy.h"
#include "SkCanvas.h"
#include "SkCommonFlags.h"
#include "SkForceLinking.h"
#include "SkGraphics.h"
#include "SkOSFile.h"
#include "SkPictureRecorder.h"
#include "SkPictureUtils.h"
#include "SkString.h"
#include "SkSurface.h"
#include "SkTaskGroup.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_bool(mpd, true, "Use MultiPictureDraw for the SKPs?");
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. "
"There's probably something wrong with the bench.\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 void setup_gl(SkGLContext* gl) {
gl->makeCurrent();
// Make sure we're done with whatever came before.
SK_GL(*gl, Finish());
}
static int gpu_bench(SkGLContext* gl,
Benchmark* bench,
SkCanvas* canvas,
double* samples) {
// 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 {
if (1<<30 == loops) {
// We're about to wrap. Something's wrong with the bench.
loops = 0;
break;
}
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;
bool useDFText;
#else
int bogusInt;
bool bogusBool;
#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, false }; \
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, useDFText) \
if (is_gpu_config_allowed(#name, GrContextFactory::ctxType, samples)) { \
Config config = { \
#name, \
Benchmark::kGPU_Backend, \
kN32_SkColorType, \
kPremul_SkAlphaType, \
samples, \
GrContextFactory::ctxType, \
useDFText }; \
configs->push(config); \
}
if (FLAGS_gpu) {
GPU_CONFIG(gpu, kNative_GLContextType, 0, false)
GPU_CONFIG(msaa4, kNative_GLContextType, 4, false)
GPU_CONFIG(msaa16, kNative_GLContextType, 16, false)
GPU_CONFIG(nvprmsaa4, kNVPR_GLContextType, 4, false)
GPU_CONFIG(nvprmsaa16, kNVPR_GLContextType, 16, false)
GPU_CONFIG(gpudft, kNative_GLContextType, 0, true)
GPU_CONFIG(debug, kDebug_GLContextType, 0, false)
GPU_CONFIG(nullgpu, kNull_GLContextType, 0, false)
#ifdef SK_ANGLE
GPU_CONFIG(angle, kANGLE_GLContextType, 0, false)
#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) {
uint32_t flags = config.useDFText ? SkSurfaceProps::kUseDistanceFieldFonts_Flag : 0;
SkSurfaceProps props(flags, SkSurfaceProps::kLegacyFontHost_InitType);
target->surface.reset(SkSurface::NewRenderTarget(gGrFactory->get(config.ctxType), info,
config.samples, &props));
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)
, fCurrentUseMPD(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);
}
}
fUseMPDs.push_back() = false;
if (FLAGS_mpd) {
fUseMPDs.push_back() = true;
}
}
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;
}
Revert of nanobench: lazily decode bitmaps in .skps. (patchset #1 id:1 of https://codereview.chromium.org/743613005/) Reason for revert: Some bots crashing. Original issue's description: > nanobench: lazily decode bitmaps in .skps. > > This cuts down on tool overhead when running something like recording only, > $ out/Release/nanobench --match skp --config nonrendering > which doesn't usually ever need to decode the images. > > The actual measurements for recording don't change, as the decode is not in the timed section. It just skips irrelevant code, removing it from the profile and making the tool run faster. > > This does, however, make a significant difference for playback speed. Most skps draw faster with this patch, some slower. I don't really have a good intuition for what's going on here. There is a fixed clip acting as a viewport, so there are probably lots of images that don't ever need to be decoded. Ideas? Is this perhaps because we're now blitting from smaller, partially decoded source images? > > ~/skia (clean) $ compare clean.log lazy-decode-bitmaps.log > tabl_slashdot.skp_1 2.76ms -> 4.33ms 1.57x > tabl_slashdot.skp_1_mpd 2.79ms -> 4.07ms 1.46x > tabl_sahadan.skp_1 3.41ms -> 4.87ms 1.43x > tabl_googleblog.skp_1 1.52ms -> 2.05ms 1.35x > tabl_techmeme.skp_1_mpd 1.14ms -> 1.51ms 1.32x > tabl_transformice.skp_1 2.61ms -> 3.43ms 1.31x > tabl_sahadan.skp_1_mpd 3.54ms -> 4.48ms 1.26x > tabl_techmeme.skp_1 1.01ms -> 1.27ms 1.26x > tabl_nytimes.skp_1_mpd 1ms -> 1.23ms 1.23x > tabl_worldjournal.skp_1_mpd 1.98ms -> 2.43ms 1.23x > tabl_pravda.skp_1_mpd 2.05ms -> 2.51ms 1.22x > tabl_transformice.skp_1_mpd 2.75ms -> 3.19ms 1.16x > tabl_nytimes.skp_1 874us -> 1.01ms 1.15x > tabl_pravda.skp_1 1.83ms -> 1.99ms 1.09x > tabl_worldjournal.skp_1 1.76ms -> 1.91ms 1.09x > desk_wowwiki.skp_1_mpd 3.7ms -> 3.9ms 1.05x > tabl_digg.skp_1 3.99ms -> 4.16ms 1.04x > tabl_ukwsj.skp_1_mpd 3ms -> 3.12ms 1.04x > desk_booking.skp_1 3.74ms -> 3.81ms 1.02x > desk_googlespreadsheetdashed.skp_1 10.6ms -> 10.6ms 1x > tabl_ukwsj.skp_1 2.88ms -> 2.89ms 1x > desk_googlespreadsheetdashed.skp_1_mpd 11.8ms -> 11.8ms 1x > desk_jsfiddlehumperclip.skp_1_mpd 891us -> 888us 1x > desk_googlespreadsheet.skp_1 4.65ms -> 4.62ms 0.99x > tabl_gspro.skp_1_mpd 1.97ms -> 1.94ms 0.99x > desk_booking.skp_1_mpd 4.1ms -> 4ms 0.98x > desk_carsvg.skp_1 18.2ms -> 17.7ms 0.97x > desk_gmailthread.skp_1_mpd 2.81ms -> 2.73ms 0.97x > desk_tigersvg.skp_1_mpd 19.5ms -> 18.9ms 0.97x > desk_mapsvg.skp_1 88.4ms -> 85.6ms 0.97x > tabl_cnet.skp_1_mpd 1.43ms -> 1.38ms 0.97x > desk_jsfiddlebigcar.skp_1 1.26ms -> 1.22ms 0.96x > desk_gws.skp_1 1.87ms -> 1.8ms 0.96x > desk_linkedin.skp_1 2.07ms -> 1.98ms 0.96x > tabl_deviantart.skp_1_mpd 118ms -> 113ms 0.96x > tabl_cnet.skp_1 1.2ms -> 1.14ms 0.95x > tabl_androidpolice.skp_1_mpd 5.95ms -> 5.63ms 0.95x > desk_sfgate.skp_1 1.75ms -> 1.64ms 0.94x > desk_twitter.skp_1 74ms -> 69.6ms 0.94x > desk_youtube.skp_1_mpd 3.17ms -> 2.96ms 0.93x > desk_gmailthread.skp_1 2.73ms -> 2.54ms 0.93x > desk_silkfinance.skp_1_mpd 1.71ms -> 1.59ms 0.93x > desk_jsfiddlebigcar.skp_1_mpd 1.45ms -> 1.35ms 0.93x > desk_pokemonwiki.skp_1_mpd 2.72ms -> 2.51ms 0.92x > desk_gws.skp_1_mpd 2.14ms -> 1.98ms 0.92x > desk_googlehome.skp_1 563us -> 517us 0.92x > desk_espn.skp_1 4.24ms -> 3.89ms 0.92x > tabl_culturalsolutions.skp_1 12.7ms -> 11.6ms 0.91x > desk_sfgate.skp_1_mpd 1.91ms -> 1.74ms 0.91x > tabl_hsfi.skp_1 1.06ms -> 966us 0.91x > desk_samoasvg.skp_1_mpd 10.5ms -> 9.47ms 0.91x > desk_facebook.skp_1_mpd 3.8ms -> 3.43ms 0.9x > desk_youtube.skp_1 3.52ms -> 3.14ms 0.89x > desk_ebay.skp_1_mpd 2.95ms -> 2.62ms 0.89x > desk_samoasvg.skp_1 10.9ms -> 9.66ms 0.89x > desk_googlespreadsheet.skp_1_mpd 5.59ms -> 4.94ms 0.88x > desk_mapsvg.skp_1_mpd 100ms -> 87.9ms 0.88x > desk_espn.skp_1_mpd 4.7ms -> 4.12ms 0.88x > desk_wordpress.skp_1_mpd 1.92ms -> 1.68ms 0.87x > tabl_deviantart.skp_1 140ms -> 122ms 0.87x > tabl_cuteoverload.skp_1_mpd 4.41ms -> 3.83ms 0.87x > desk_tigersvg.skp_1 19.6ms -> 17ms 0.87x > tabl_googlecalendar.skp_1 4.01ms -> 3.44ms 0.86x > desk_blogger.skp_1 2.49ms -> 2.14ms 0.86x > desk_chalkboard.skp_1_mpd 52.7ms -> 45ms 0.85x > desk_weather.skp_1 2.88ms -> 2.46ms 0.85x > desk_chalkboard.skp_1 51ms -> 43.4ms 0.85x > desk_yahooanswers.skp_1 2.74ms -> 2.32ms 0.85x > desk_forecastio.skp_1_mpd 1.26ms -> 1.07ms 0.85x > tabl_androidpolice.skp_1 5.18ms -> 4.34ms 0.84x > desk_yahooanswers.skp_1_mpd 3.44ms -> 2.85ms 0.83x > tabl_cnn.skp_1_mpd 2.59ms -> 2.15ms 0.83x > desk_pinterest.skp_1 2.69ms -> 2.22ms 0.83x > tabl_hsfi.skp_1_mpd 1.6ms -> 1.32ms 0.82x > tabl_culturalsolutions.skp_1_mpd 13.8ms -> 11.3ms 0.82x > desk_twitter.skp_1_mpd 76.6ms -> 63ms 0.82x > desk_ebay.skp_1 3.11ms -> 2.51ms 0.81x > tabl_mlb.skp_1_mpd 3.17ms -> 2.53ms 0.8x > tabl_mozilla.skp_1 2.42ms -> 1.91ms 0.79x > desk_pokemonwiki.skp_1 2.84ms -> 2.22ms 0.78x > desk_carsvg.skp_1_mpd 23.3ms -> 17.8ms 0.77x > desk_wowwiki.skp_1 4.21ms -> 3.21ms 0.76x > desk_amazon.skp_1 963us -> 728us 0.76x > desk_css3gradients.skp_1 2.58ms -> 1.92ms 0.74x > tabl_cuteoverload.skp_1 4.55ms -> 3.38ms 0.74x > tabl_cnn.skp_1 3.13ms -> 2.29ms 0.73x > tabl_googleblog.skp_1_mpd 2.32ms -> 1.7ms 0.73x > desk_mobilenews.skp_1 3.65ms -> 2.61ms 0.71x > desk_googleplus.skp_1 3.76ms -> 2.66ms 0.71x > tabl_mozilla.skp_1_mpd 2.88ms -> 2.03ms 0.71x > desk_pinterest.skp_1_mpd 3.17ms -> 2.21ms 0.7x > desk_css3gradients.skp_1_mpd 2.98ms -> 2.07ms 0.69x > desk_silkfinance.skp_1 2.06ms -> 1.42ms 0.69x > desk_facebook.skp_1 4.5ms -> 3.07ms 0.68x > desk_mobilenews.skp_1_mpd 4.05ms -> 2.73ms 0.68x > desk_baidu.skp_1_mpd 2.73ms -> 1.81ms 0.66x > desk_weather.skp_1_mpd 3.93ms -> 2.5ms 0.64x > desk_wordpress.skp_1 2.15ms -> 1.36ms 0.63x > desk_googlehome.skp_1_mpd 1.02ms -> 605us 0.59x > desk_fontwipe.skp_1 722us -> 402us 0.56x > desk_fontwipe.skp_1_mpd 897us -> 486us 0.54x > desk_baidu.skp_1 3.02ms -> 1.6ms 0.53x > desk_forecastio.skp_1 2.01ms -> 999us 0.5x > desk_amazon.skp_1_mpd 1.77ms -> 860us 0.49x > > BUG=skia: > > Committed: https://skia.googlesource.com/skia/+/7e225bdb1f00ae4aed524ff8d0a61df3d3abb109 TBR=reed@google.com,robertphillips@google.com,scroggo@google.com,mtklein@chromium.org NOTREECHECKS=true NOTRY=true BUG=skia: Review URL: https://codereview.chromium.org/759753004
2014-11-25 22:57:26 +00:00
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";
fSKPBytes = static_cast<double>(SkPictureUtils::ApproximateBytesUsed(pic));
fSKPOps = pic->approximateOpCount();
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)) {
fCurrentSKP++;
continue;
}
while (fCurrentUseMPD < fUseMPDs.count()) {
if (FLAGS_bbh) {
// The SKP we read off disk doesn't have a BBH. Re-record so it grows one.
SkRTreeFactory factory;
SkPictureRecorder recorder;
static const int kFlags = SkPictureRecorder::kComputeSaveLayerInfo_RecordFlag;
pic->playback(recorder.beginRecording(pic->cullRect().width(),
pic->cullRect().height(),
&factory,
fUseMPDs[fCurrentUseMPD] ? kFlags : 0));
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], fUseMPDs[fCurrentUseMPD++]));
}
fCurrentUseMPD = 0;
fCurrentSKP++;
}
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());
if (fCurrentUseMPD > 0) {
SkASSERT(1 == fCurrentUseMPD || 2 == fCurrentUseMPD);
log->configOption("multi_picture_draw", fUseMPDs[fCurrentUseMPD-1] ? "true" : "false");
}
}
if (0 == strcmp(fBenchType, "recording")) {
log->metric("bytes", fSKPBytes);
log->metric("ops", fSKPOps);
}
}
private:
const BenchRegistry* fBenches;
const skiagm::GMRegistry* fGMs;
SkIRect fClip;
SkTArray<SkScalar> fScales;
SkTArray<SkString> fSKPs;
SkTArray<bool> fUseMPDs;
double fSKPBytes, fSKPOps;
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 fCurrentUseMPD;
};
int nanobench_main();
int nanobench_main() {
SetupCrashHandler();
SkAutoGraphics ag;
SkTaskGroup::Enabler enabled;
#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;
#if SK_SUPPORT_GPU
if (Benchmark::kGPU_Backend == targets[j]->config.backend) {
setup_gl(targets[j]->gl);
}
#endif
bench->perCanvasPreDraw(canvas);
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());
bench->perCanvasPostDraw(canvas);
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->metric("min_ms", stats.min);
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
}
log->bench("memory_usage", 0,0);
log->config("meta");
log->metric("max_rss_mb", sk_tools::getMaxResidentSetSizeMB());
return 0;
}
#if !defined SK_BUILD_FOR_IOS
int main(int argc, char** argv) {
SkCommandLineFlags::Parse(argc, argv);
return nanobench_main();
}
#endif