| Index: tools/skpbench/skpbench.cpp
|
| diff --git a/tools/skpbench/skpbench.cpp b/tools/skpbench/skpbench.cpp
|
| index 6d0381a28d75dcd3cfde5658bb737ea75b584930..adb6af0b146669f6a1c44fa253f2ad7a5e7614a6 100644
|
| --- a/tools/skpbench/skpbench.cpp
|
| +++ b/tools/skpbench/skpbench.cpp
|
| @@ -5,7 +5,6 @@
|
| * found in the LICENSE file.
|
| */
|
|
|
| -#include "GpuTimer.h"
|
| #include "GrContextFactory.h"
|
| #include "SkCanvas.h"
|
| #include "SkOSFile.h"
|
| @@ -34,9 +33,12 @@
|
| * Currently, only GPU configs are supported.
|
| */
|
|
|
| +using sk_gpu_test::PlatformFence;
|
| +using sk_gpu_test::kInvalidPlatformFence;
|
| +using sk_gpu_test::FenceSync;
|
| +
|
| DEFINE_int32(duration, 5000, "number of milliseconds to run the benchmark");
|
| DEFINE_int32(sampleMs, 50, "minimum duration of a sample");
|
| -DEFINE_bool(gpuClock, false, "time on the gpu clock (gpu work only)");
|
| DEFINE_bool(fps, false, "use fps instead of ms");
|
| DEFINE_string(skp, "", "path to a single .skp file to benchmark");
|
| DEFINE_string(png, "", "if set, save a .png proof to disk at this file location");
|
| @@ -44,13 +46,13 @@
|
| DEFINE_bool(suppressHeader, false, "don't print a header row before the results");
|
|
|
| static const char* header =
|
| -" accum median max min stddev samples sample_ms clock metric config bench";
|
| + " accum median max min stddev samples sample_ms metric config bench";
|
|
|
| static const char* resultFormat =
|
| -"%8.4g %8.4g %8.4g %8.4g %6.3g%% %7li %9i %-5s %-6s %-9s %s";
|
| + "%8.4g %8.4g %8.4g %8.4g %6.3g%% %7li %9i %-6s %-9s %s";
|
|
|
| struct Sample {
|
| - using duration = std::chrono::nanoseconds;
|
| + using clock = std::chrono::high_resolution_clock;
|
|
|
| Sample() : fFrames(0), fDuration(0) {}
|
| double seconds() const { return std::chrono::duration<double>(fDuration).count(); }
|
| @@ -58,13 +60,13 @@
|
| double value() const { return FLAGS_fps ? fFrames / this->seconds() : this->ms() / fFrames; }
|
| static const char* metric() { return FLAGS_fps ? "fps" : "ms"; }
|
|
|
| - int fFrames;
|
| - duration fDuration;
|
| + int fFrames;
|
| + clock::duration fDuration;
|
| };
|
|
|
| class GpuSync {
|
| public:
|
| - GpuSync(const sk_gpu_test::FenceSync* fenceSync);
|
| + GpuSync(const FenceSync* fenceSync);
|
| ~GpuSync();
|
|
|
| void syncToPreviousFrame();
|
| @@ -72,8 +74,8 @@
|
| private:
|
| void updateFence();
|
|
|
| - const sk_gpu_test::FenceSync* const fFenceSync;
|
| - sk_gpu_test::PlatformFence fFence;
|
| + const FenceSync* const fFenceSync;
|
| + PlatformFence fFence;
|
| };
|
|
|
| enum class ExitErr {
|
| @@ -90,10 +92,10 @@
|
| static SkString join(const SkCommandLineFlags::StringArray&);
|
| static void exitf(ExitErr, const char* format, ...);
|
|
|
| -static void run_benchmark(const sk_gpu_test::FenceSync* fenceSync, SkCanvas* canvas,
|
| - const SkPicture* skp, std::vector<Sample>* samples) {
|
| - using clock = std::chrono::high_resolution_clock;
|
| - const Sample::duration sampleDuration = std::chrono::milliseconds(FLAGS_sampleMs);
|
| +static void run_benchmark(const FenceSync* fenceSync, SkCanvas* canvas, const SkPicture* skp,
|
| + std::vector<Sample>* samples) {
|
| + using clock = Sample::clock;
|
| + const clock::duration sampleDuration = std::chrono::milliseconds(FLAGS_sampleMs);
|
| const clock::duration benchDuration = std::chrono::milliseconds(FLAGS_duration);
|
|
|
| draw_skp_and_flush(canvas, skp);
|
| @@ -119,66 +121,6 @@
|
| ++sample.fFrames;
|
| } while (sample.fDuration < sampleDuration);
|
| } while (now < endTime || 0 == samples->size() % 2);
|
| -}
|
| -
|
| -static void run_gpu_time_benchmark(sk_gpu_test::GpuTimer* gpuTimer,
|
| - const sk_gpu_test::FenceSync* fenceSync, SkCanvas* canvas,
|
| - const SkPicture* skp, std::vector<Sample>* samples) {
|
| - using sk_gpu_test::PlatformTimerQuery;
|
| - using clock = std::chrono::steady_clock;
|
| - const clock::duration sampleDuration = std::chrono::milliseconds(FLAGS_sampleMs);
|
| - const clock::duration benchDuration = std::chrono::milliseconds(FLAGS_duration);
|
| -
|
| - if (!gpuTimer->disjointSupport()) {
|
| - fprintf(stderr, "WARNING: GPU timer cannot detect disjoint operations; "
|
| - "results may be unreliable\n");
|
| - }
|
| -
|
| - draw_skp_and_flush(canvas, skp);
|
| - GpuSync gpuSync(fenceSync);
|
| -
|
| - gpuTimer->queueStart();
|
| - draw_skp_and_flush(canvas, skp);
|
| - PlatformTimerQuery previousTime = gpuTimer->queueStop();
|
| - gpuSync.syncToPreviousFrame();
|
| -
|
| - clock::time_point now = clock::now();
|
| - const clock::time_point endTime = now + benchDuration;
|
| -
|
| - do {
|
| - const clock::time_point sampleEndTime = now + sampleDuration;
|
| - samples->emplace_back();
|
| - Sample& sample = samples->back();
|
| -
|
| - do {
|
| - gpuTimer->queueStart();
|
| - draw_skp_and_flush(canvas, skp);
|
| - PlatformTimerQuery time = gpuTimer->queueStop();
|
| - gpuSync.syncToPreviousFrame();
|
| -
|
| - switch (gpuTimer->checkQueryStatus(previousTime)) {
|
| - using QueryStatus = sk_gpu_test::GpuTimer::QueryStatus;
|
| - case QueryStatus::kInvalid:
|
| - exitf(ExitErr::kUnavailable, "GPU timer failed");
|
| - case QueryStatus::kPending:
|
| - exitf(ExitErr::kUnavailable, "timer query still not ready after fence sync");
|
| - case QueryStatus::kDisjoint:
|
| - if (FLAGS_verbosity >= 4) {
|
| - fprintf(stderr, "discarding timer query due to disjoint operations.\n");
|
| - }
|
| - break;
|
| - case QueryStatus::kAccurate:
|
| - sample.fDuration += gpuTimer->getTimeElapsed(previousTime);
|
| - ++sample.fFrames;
|
| - break;
|
| - }
|
| - gpuTimer->deleteQuery(previousTime);
|
| - previousTime = time;
|
| - now = clock::now();
|
| - } while (now < sampleEndTime || 0 == sample.fFrames);
|
| - } while (now < endTime || 0 == samples->size() % 2);
|
| -
|
| - gpuTimer->deleteQuery(previousTime);
|
| }
|
|
|
| void print_result(const std::vector<Sample>& samples, const char* config, const char* bench) {
|
| @@ -207,8 +149,7 @@
|
| const double stddev = 100/*%*/ * sqrt(variance) / accumValue;
|
|
|
| printf(resultFormat, accumValue, values[values.size() / 2], values.back(), values.front(),
|
| - stddev, values.size(), FLAGS_sampleMs, FLAGS_gpuClock ? "gpu" : "cpu", Sample::metric(),
|
| - config, bench);
|
| + stddev, values.size(), FLAGS_sampleMs, Sample::metric(), config, bench);
|
| printf("\n");
|
| fflush(stdout);
|
| }
|
| @@ -306,15 +247,7 @@
|
| // Run the benchmark.
|
| SkCanvas* canvas = surface->getCanvas();
|
| canvas->translate(-skp->cullRect().x(), -skp->cullRect().y());
|
| - if (!FLAGS_gpuClock) {
|
| - run_benchmark(testCtx->fenceSync(), canvas, skp.get(), &samples);
|
| - } else {
|
| - if (!testCtx->gpuTimingSupport()) {
|
| - exitf(ExitErr::kUnavailable, "GPU does not support timing");
|
| - }
|
| - run_gpu_time_benchmark(testCtx->gpuTimer(), testCtx->fenceSync(), canvas, skp.get(),
|
| - &samples);
|
| - }
|
| + run_benchmark(testCtx->fenceSync(), canvas, skp.get(), &samples);
|
| print_result(samples, config->getTag().c_str(), SkOSPath::Basename(skpfile).c_str());
|
|
|
| // Save a proof (if one was requested).
|
| @@ -367,7 +300,7 @@
|
| exit((int)err);
|
| }
|
|
|
| -GpuSync::GpuSync(const sk_gpu_test::FenceSync* fenceSync)
|
| +GpuSync::GpuSync(const FenceSync* fenceSync)
|
| : fFenceSync(fenceSync) {
|
| this->updateFence();
|
| }
|
| @@ -377,7 +310,7 @@
|
| }
|
|
|
| void GpuSync::syncToPreviousFrame() {
|
| - if (sk_gpu_test::kInvalidFence == fFence) {
|
| + if (kInvalidPlatformFence == fFence) {
|
| exitf(ExitErr::kSoftware, "attempted to sync with invalid fence");
|
| }
|
| if (!fFenceSync->waitFence(fFence)) {
|
| @@ -389,7 +322,7 @@
|
|
|
| void GpuSync::updateFence() {
|
| fFence = fFenceSync->insertFence();
|
| - if (sk_gpu_test::kInvalidFence == fFence) {
|
| + if (kInvalidPlatformFence == fFence) {
|
| exitf(ExitErr::kUnavailable, "failed to insert fence");
|
| }
|
| }
|
|
|