Sync v13.5-public from internal repo

This commit is contained in:
cuda-samples sync
2026-11-04 20:44:00 +00:00
parent 5443602d89
commit a9315c55fb
53 changed files with 1750 additions and 183 deletions

View File

@@ -162,6 +162,8 @@ static int fpsLimit = 5;
cudaEvent_t startEvent, stopEvent;
cudaEvent_t hostMemSyncEvent;
cudaStream_t nbodyTimingStream();
template <typename T> class NBodyDemo
{
public:
@@ -188,6 +190,11 @@ public:
static void runBenchmark(int iterations) { m_singleton->_runBenchmark(iterations); }
static cudaStream_t timingStream()
{
return (m_singleton && m_singleton->m_nbodyCuda) ? m_singleton->m_nbodyCuda->getComputeStream() : 0;
}
static void updateParams()
{
m_singleton->m_nbody->setSoftening(activeParams.m_softening);
@@ -440,7 +447,7 @@ private:
sdkStartTimer(&timer);
}
else {
checkCudaErrors(cudaEventRecord(startEvent, 0));
checkCudaErrors(cudaEventRecord(startEvent, nbodyTimingStream()));
}
for (int i = 0; i < iterations; ++i) {
@@ -455,7 +462,7 @@ private:
sdkStartTimer(&timer);
}
else {
checkCudaErrors(cudaEventRecord(stopEvent, 0));
checkCudaErrors(cudaEventRecord(stopEvent, nbodyTimingStream()));
checkCudaErrors(cudaEventSynchronize(stopEvent));
checkCudaErrors(cudaEventElapsedTime(&milliseconds, startEvent, stopEvent));
}
@@ -473,6 +480,19 @@ private:
}
};
cudaStream_t nbodyTimingStream()
{
if (useCpu) {
return 0;
}
if (fp64) {
return NBodyDemo<double>::timingStream();
}
return NBodyDemo<float>::timingStream();
}
void finalize()
{
if (!useCpu) {
@@ -642,7 +662,7 @@ void display()
if (!useCpu) {
cudaEventRecord(hostMemSyncEvent,
0); // insert an event to wait on before rendering
nbodyTimingStream()); // insert an event to wait on before rendering
}
}
@@ -720,7 +740,7 @@ void display()
sdkResetTimer(&timer);
}
else {
checkCudaErrors(cudaEventRecord(stopEvent, 0));
checkCudaErrors(cudaEventRecord(stopEvent, nbodyTimingStream()));
checkCudaErrors(cudaEventSynchronize(stopEvent));
checkCudaErrors(cudaEventElapsedTime(&milliseconds, startEvent, stopEvent));
}
@@ -748,7 +768,7 @@ void display()
// restart timer
if (!useCpu) {
checkCudaErrors(cudaEventRecord(startEvent, 0));
checkCudaErrors(cudaEventRecord(startEvent, nbodyTimingStream()));
}
}
@@ -1086,26 +1106,13 @@ int main(int argc, char **argv)
}
if (!useCpu) {
// Now choose the CUDA Device
// Either without GL interop:
if (benchmark || compareToCPU || useHostMem) {
// Note if we are using host memory for the body system, we
// don't use CUDA-GL interop.
if (checkCmdLineFlag(argc, (const char **)argv, "device")) {
customGPU = true;
}
devID = findCudaDevice(argc, (const char **)argv);
// Device selection does not depend on GL interop. Benchmark, compare,
// and host memory skip interop by leaving the PBO argument false below.
if (checkCmdLineFlag(argc, (const char **)argv, "device")) {
customGPU = true;
}
else // or with GL interop:
{
if (checkCmdLineFlag(argc, (const char **)argv, "device")) {
customGPU = true;
}
devID = findCudaDevice(argc, (const char **)argv);
}
devID = findCudaDevice(argc, (const char **)argv);
checkCudaErrors(cudaGetDevice(&devID));
checkCudaErrors(cudaGetDeviceProperties(&props, devID));
@@ -1147,10 +1154,7 @@ int main(int argc, char **argv)
exit(EXIT_SUCCESS);
}
if (numDevsRequested > 1) {
checkCudaErrors(cudaSetDevice(i));
}
checkCudaErrors(cudaSetDevice(i));
checkCudaErrors(cudaSetDeviceFlags(cudaDeviceMapHost));
#else
fprintf(stderr,
@@ -1322,7 +1326,7 @@ int main(int argc, char **argv)
glutIdleFunc(idle);
if (!useCpu) {
checkCudaErrors(cudaEventRecord(startEvent, 0));
checkCudaErrors(cudaEventRecord(startEvent, nbodyTimingStream()));
}
glutMainLoop();
@@ -1349,7 +1353,7 @@ int main(int argc, char **argv)
glutIdleFunc(idle);
if (!useCpu) {
checkCudaErrors(cudaEventRecord(startEvent, 0));
checkCudaErrors(cudaEventRecord(startEvent, nbodyTimingStream()));
}
glutMainLoop();