niansa-misc/dxvk
1
0
Fork 0
mirror of https://github.com/doitsujin/dxvk.git synced 2025-03-06 20:58:37 +01:00
Code Issues Projects Releases Packages Wiki Activity

[dxvk] Improve low-latency frame pacing when gpuStart gets signaled earlier than expected

Browse source
This commit is contained in:
netborg 2025-02-28 08:39:40 +01:00
parent 91c6793b55
commit 988dcdc08c
1 changed files with 6 additions and 26 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

32
src/dxvk/framepacer/dxvk_framepacer_mode_low_latency.h
View file

@ -79,8 +79,10 @@ namespace dxvk {
// and calculate backwards when we want to start this frame
const SyncProps props = getSyncPrediction();
int32_t gpuReadyPrediction = duration_cast<microseconds>(
m->start + microseconds(m->gpuStart+getGpuStartToFinishPrediction()) - now).count();
int32_t lastFrameStart = duration_cast<microseconds>( m->start - now ).count();
int32_t gpuReadyPrediction = lastFrameStart
+ std::max( props.cpuUntilGpuStart, m->gpuStart )
+ props.optimizedGpuTime;
int32_t targetGpuSync = gpuReadyPrediction + props.gpuSync;
int32_t gpuDelay = targetGpuSync - props.cpuUntilGpuSync;
@ -144,6 +146,7 @@ namespace dxvk {
SyncProps& props = m_props[frameId % m_props.size()];
props.gpuSync = gpuRun[numLoop-1];
props.cpuUntilGpuSync = offset + duration_cast<microseconds>( m->gpuSubmit[numLoop-1] - m->start ).count();
props.cpuUntilGpuStart = props.cpuUntilGpuSync - props.gpuSync;
props.optimizedGpuTime = optimizedGpuTime;
props.csStart = m->csStart;
props.csFinished = m->csFinished;
@ -176,6 +179,7 @@ namespace dxvk {
int32_t optimizedGpuTime; // gpu executing packed submits in one go
int32_t gpuSync; // us after gpuStart
int32_t cpuUntilGpuSync;
int32_t cpuUntilGpuStart;
int32_t csStart;
int32_t csFinished;
bool isOutlier;
@ -203,30 +207,6 @@ namespace dxvk {
};
int32_t getGpuStartToFinishPrediction() {
uint64_t id = m_propsFinished;
if (id < DXGI_MAX_SWAP_CHAIN_BUFFERS+7)
return 0;
for (size_t i=0; i<7; ++i) {
const SyncProps& props = m_props[ (id-i) % m_props.size() ];
if (!props.isOutlier) {
const LatencyMarkers* m = m_latencyMarkersStorage->getConstMarkers(id-i);
if (m->gpuReady.empty() || m->gpuSubmit.empty())
return m->gpuFinished - m->gpuStart;
time_point t = std::max( m->gpuReady[0], m->gpuSubmit[0] );
return std::chrono::duration_cast<microseconds>( t - m->start ).count()
+ props.optimizedGpuTime
- m->gpuStart;
}
}
const LatencyMarkers* m = m_latencyMarkersStorage->getConstMarkers(id);
return m->gpuFinished - m->gpuStart;
};
bool isOutlier( uint64_t frameId ) {
constexpr size_t numLoop = 7;
int32_t totalCpuTime = 0;