Rework Performance Statistics

We used instantaneous values for FPS previously which led to a lot of variation in it and the inability to determine a proper FPS value due to constant fluctuations. All FPS values are now averaged to allow reading out a stable value and a deviation statistic has been added for the frame-time to judge judder and frame-pacing which allows for a significantly better measure of overall performance. The formatting for all the floating-point numbers is now fixed-point to prevent shifting of position due to decimal digits becoming 0.
This commit is contained in:
PixelyIon
2021-06-27 06:56:16 +05:30
committed by ◱ Mark
parent ce804254ab
commit 6595670a5d
5 changed files with 38 additions and 16 deletions

View File

@ -13,8 +13,9 @@
#include "native_window.h"
#include "texture/format.h"
extern skyline::i32 Fps;
extern skyline::i32 FrameTime;
extern jint Fps;
extern jfloat AverageFrametimeMs;
extern jfloat AverageFrametimeDeviationMs;
namespace skyline::gpu {
using namespace service::hosbinder;
@ -39,11 +40,12 @@ namespace skyline::gpu {
void PresentationEngine::ChoreographerCallback(long frameTimeNanos, PresentationEngine *engine) {
u64 cycleLength{frameTimeNanos - engine->lastChoreographerTime};
if (std::abs(static_cast<i64>(cycleLength - engine->refreshCycleDuration)) > (constant::NsInMillisecond / 2))
if (std::abs(static_cast<i64>(cycleLength - engine->refreshCycleDuration)) > (constant::NsInMillisecond / 2)) {
if (engine->window)
engine->window->perform(engine->window, NATIVE_WINDOW_GET_REFRESH_CYCLE_DURATION, &engine->refreshCycleDuration);
else
engine->refreshCycleDuration = cycleLength;
}
engine->lastChoreographerTime = frameTimeNanos;
engine->vsyncEvent->Signal();
@ -271,8 +273,17 @@ namespace skyline::gpu {
if (frameTimestamp) {
auto now{util::GetTimeNs()};
FrameTime = static_cast<u32>((now - frameTimestamp) / 10000); // frametime / 100 is the real ms value, this is to retain the first two decimals
Fps = static_cast<u16>(constant::NsInSecond / (now - frameTimestamp));
auto sampleWeight{swapInterval ? constant::NsInSecond / (refreshCycleDuration * swapInterval) : 10}; //!< The weight of each sample in calculating the average, we arbitrarily average 10 samples for unlocked FPS
u64 currentFrametime{now - frameTimestamp};
averageFrametimeNs = (((sampleWeight - 1) * averageFrametimeNs) / sampleWeight) + (currentFrametime / sampleWeight);
AverageFrametimeMs = static_cast<jfloat>(averageFrametimeNs) / constant::NsInMillisecond;
i64 currentFrametimeDeviation{std::abs(static_cast<i64>(averageFrametimeNs - currentFrametime))};
averageFrametimeDeviationNs = (((sampleWeight - 1) * averageFrametimeDeviationNs) / sampleWeight) + (currentFrametimeDeviation / sampleWeight);
AverageFrametimeDeviationMs = static_cast<jfloat>(averageFrametimeDeviationNs) / constant::NsInMillisecond;
Fps = std::round(static_cast<float>(constant::NsInSecond) / averageFrametimeNs);
TRACE_EVENT_INSTANT("gpu", "Present", presentationTrack, "FrameTimeNs", now - frameTimestamp, "Fps", Fps);

View File

@ -40,7 +40,9 @@ namespace skyline::gpu {
static constexpr size_t MaxSwapchainImageCount{6}; //!< The maximum amount of swapchain textures, this affects the amount of images that can be in the swapchain
std::array<std::shared_ptr<Texture>, MaxSwapchainImageCount> images; //!< All the swapchain textures in the same order as supplied by the host swapchain
u64 frameTimestamp{}; //!< The timestamp of the last frame being shown
u64 frameTimestamp{}; //!< The timestamp of the last frame being shown in nanoseconds
u64 averageFrametimeNs{}; //!< The average time between frames in nanoseconds
u64 averageFrametimeDeviationNs{}; //!< The average deviation of frametimes in nanoseconds
perfetto::Track presentationTrack; //!< Perfetto track used for presentation events
std::thread choreographerThread; //!< A thread for signalling the V-Sync event using AChoreographer