void SStopWatch::MarkInternal(Root* root)
{
FlushICache();
root->suspendTime = 0;
if ((root->extensions&EXTENSION_PERFORMANCE) != 0) {
IOSFastIoctl(CounterFD(), PERFCNT_READ_COUNTERS(NUM_COUNTERS+1),
0, NULL, sizeof(m_node->counters), m_node->counters, &root->perfError);
}
#if TARGET_PLATFORM == TARGET_PLATFORM_PALMSIM_WIN32
if ((root->extensions&EXTENSION_QUANTIFY_RUNNING) != 0) {
QuantifyStopRecordingData();
}
if ((root->flags&B_STOP_WATCH_QUANTIFY) != 0) {
QuantifyStartRecordingData();
root->extensions |= EXTENSION_QUANTIFY_RUNNING;
}
#endif
if ((root->flags&B_STOP_WATCH_NO_TIME) == 0) {
m_node->time = KALGetTime(B_TIMEBASE_RUN_TIME);
}
#if TARGET_PLATFORM == TARGET_PLATFORM_DEVICE_ARM
if ((root->extensions&EXTENSION_PERFORMANCE_HIGH) != 0) {
m_node->counters[0] = ReadPerformanceCounter(0);
m_node->counters[1] = ReadPerformanceCounter(1);
m_node->counters[2] = ReadPerformanceCounter(2);
}
#endif
}
SPerformanceSample::SPerformanceSample()
{
SStopWatch::FlushICache();
#if TARGET_PLATFORM == TARGET_PLATFORM_DEVICE_ARM
m_counters[0] = ReadPerformanceCounter(0);
m_counters[1] = ReadPerformanceCounter(1);
m_counters[2] = ReadPerformanceCounter(2);
#endif
}
SPerformanceSample::~SPerformanceSample()
{
#if TARGET_PLATFORM == TARGET_PLATFORM_DEVICE_ARM
uint32_t cur[3];
cur[0] = ReadPerformanceCounter(0);
cur[1] = ReadPerformanceCounter(1);
cur[2] = ReadPerformanceCounter(2);
printf("%lu %lu %lu\n", cur[0]-m_counters[0], cur[1]-m_counters[1], cur[2]-m_counters[2]);
#endif
}
void SStopWatch::RestartInternal(Root* root)
{
if (m_type != TYPE_CHILD) {
if ((root->extensions&(EXTENSION_PERFORMANCE|EXTENSION_PERFORMANCE_HIGH)) != 0) {
IOSFastIoctl(CounterFD(), PERFCNT_CONFIGURE_COUNTERS(NUM_COUNTERS),
sizeof(root->perfCounters), root->perfCounters, 0, NULL, &root->perfError);
//if (root->perfError != B_OK) root->extensions &= ~EXTENSION_PERFORMANCE;
}
if ((root->extensions&EXTENSION_PROFILING) != 0) {
// Stopping returns an error code if already stopped, so ignore this result.
IOSFastIoctl(ProfileFD(), profilerStopProfiling, 0, NULL, 0, NULL, &root->profError);
root->profError = B_OK;
if ((root->profFlags&B_PROFILE_KEEP_SAMPLES) == 0) {
IOSFastIoctl(ProfileFD(), profilerResetSamples, 0, NULL, 0, NULL, &root->perfError);
}
if (root->profError == B_OK) {
IOSFastIoctl(ProfileFD(), profilerStartProfiling, 0, NULL, 0, NULL, &root->perfError);
}
if (root->profError != B_OK) root->extensions &= ~EXTENSION_PROFILING;
}
if ((root->flags&B_STOP_WATCH_CLEAR_CACHE) != 0) {
// A "big" size says to just blindly flush the whole cache.
FlushICache();
}
// Compute overhead -- baseline
if ((root->flags&B_STOP_WATCH_NO_TIME) == 0) {
root->timeOverhead = KALGetTime(B_TIMEBASE_RUN_TIME);
}
if ((root->extensions&EXTENSION_PERFORMANCE) != 0) {
memset(root->counterOverhead, 0, sizeof(root->counterOverhead));
IOSFastIoctl(CounterFD(), PERFCNT_READ_COUNTERS(NUM_COUNTERS+1),
0, NULL, sizeof(root->counterOverhead), root->counterOverhead, &root->perfError);
}
#if TARGET_PLATFORM == TARGET_PLATFORM_DEVICE_ARM
if ((root->extensions&EXTENSION_PERFORMANCE_HIGH) != 0) {
root->counterOverhead[0] = ReadPerformanceCounter(0);
root->counterOverhead[1] = ReadPerformanceCounter(1);
root->counterOverhead[2] = ReadPerformanceCounter(2);
}
#endif
// Some internal stuff we always have to do
if ((root->flags&B_STOP_WATCH_NESTING) != 0) {
Root* foo = CurrentRoot();
}
// Compute overhead -- total up measurement overhead
if ((root->flags&B_STOP_WATCH_NO_TIME) == 0) {
root->timeOverhead = KALGetTime(B_TIMEBASE_RUN_TIME)-root->timeOverhead;
}
if ((root->extensions&EXTENSION_PERFORMANCE) != 0) {
status_t err;
IOSFastIoctl(CounterFD(), PERFCNT_READ_COUNTERS(NUM_COUNTERS+1), 0, NULL, sizeof(root->tmpPerf), root->tmpPerf, &err);
for (int i=0; i<NUM_COUNTERS+1; i++) {
root->counterOverhead[i] = root->tmpPerf[i] - root->counterOverhead[i];
}
}
#if TARGET_PLATFORM == TARGET_PLATFORM_DEVICE_ARM
if ((root->extensions&EXTENSION_PERFORMANCE_HIGH) != 0) {
root->tmpPerf[0] = ReadPerformanceCounter(0);
root->tmpPerf[1] = ReadPerformanceCounter(1);
root->tmpPerf[2] = ReadPerformanceCounter(2);
for (int i=0; i<NUM_COUNTERS+1; i++) {
root->counterOverhead[i] = root->tmpPerf[i] - root->counterOverhead[i];
}
}
#endif
}
}
void SStopWatch::StopInternal(Root* root)
{
#if TARGET_PLATFORM == TARGET_PLATFORM_DEVICE_ARM
if ((root->extensions&EXTENSION_PERFORMANCE_HIGH) != 0) {
root->tmpPerf[0] = ReadPerformanceCounter(0);
root->tmpPerf[1] = ReadPerformanceCounter(1);
root->tmpPerf[2] = ReadPerformanceCounter(2);
}
#endif
Node* const node = m_node;
if ((root->flags&B_STOP_WATCH_NO_TIME) == 0) {
node->time = ElapsedTime();
}
#if TARGET_PLATFORM == TARGET_PLATFORM_PALMSIM_WIN32
if ((root->extensions&EXTENSION_QUANTIFY_RUNNING) != 0) {
QuantifyStopRecordingData();
}
#endif
if ((root->extensions&EXTENSION_PERFORMANCE) != 0) {
status_t err;
IOSFastIoctl(CounterFD(), PERFCNT_READ_COUNTERS(NUM_COUNTERS+1),
0, NULL, sizeof(root->tmpPerf), root->tmpPerf, &err);
}
if ((root->extensions&(EXTENSION_PERFORMANCE|EXTENSION_PERFORMANCE_HIGH)) != 0) {
for (int i=0; i<NUM_COUNTERS+1; i++) {
if (root->tmpPerf[i] <= node->counters[i])
node->counters[i] = (root->tmpPerf[i]+0x10000)-node->counters[i];
else
node->counters[i] = root->tmpPerf[i]-node->counters[i];
}
}
if (m_type == TYPE_CHILD) {
root->currentNode = m_parent;
return;
}
if ((root->extensions&EXTENSION_PROFILING) != 0) {
status_t err;
IOSFastIoctl(ProfileFD(), profilerStopProfiling, 0, NULL, 0, NULL, &err);
}
if ((root->flags&B_STOP_WATCH_HIGH_PRIORITY) != 0) {
KALThreadChangePriority((KeyID)SysCurrentThread(), root->origPriority);
}
if ((root->flags&B_STOP_WATCH_SILENT) == 0) {
root->PrintReport(bout);
}
if ((root->extensions&EXTENSION_PROFILING) != 0
&& (root->profFlags&B_PROFILE_UPLOAD_SAMPLES) != 0) {
status_t err;
IOSFastIoctl(ProfileFD(), profilerSendSamplesToPOD, 0, NULL, 0, NULL, &err);
}
SetCurrentRoot(NULL);
root->extensions |= EXTENSION_STOPPED;
}
void RestartPerfCntrs(PerfObj *PerfCntrs)
{
DWORD returnLength = 0; //Lint
DWORD returnNumCPUs; //Lint
DWORD i;
DWORD status;
SYSTEM_INFO SystemInfo;
GetSystemInfo(&SystemInfo);
// Move previous data from EndInfo to StartInfo.
CopyMemory((PCHAR)&PerfCntrs->StartInfo[0],
(PCHAR)&PerfCntrs->EndInfo[0],
sizeof(SYSTEM_PROCESSOR_PERFORMANCE_INFORMATION)*(MAXCPUS +1));
PerfCntrs->StartTime = PerfCntrs->EndTime;
// get the current CPUTIME information
if ( (status = NtQuerySystemInformation( SystemProcessorPerformanceInformation,
(PCHAR)&PerfCntrs->EndInfo[0], sizeof(SYSTEM_PROCESSOR_PERFORMANCE_INFORMATION)*MAXCPUS,
&returnLength )) != 0)
{
fprintf(stderr, "NtQuery failed, status: %X\n", status);
exit(1);
}
PerfCntrs->EndTime = ReadPerformanceCounter();
// Validate that NtQuery returned a reasonable amount of data
if ((returnLength % sizeof(SYSTEM_PROCESSOR_PERFORMANCE_INFORMATION)) != 0)
{
fprintf(stderr, "NtQuery didn't return expected amount of data\n");
fprintf(stderr, "Expected a multiple of %i, returned %i\n",
sizeof(SYSTEM_PROCESSOR_PERFORMANCE_INFORMATION), returnLength);
exit(1);
}
returnNumCPUs = returnLength / sizeof(SYSTEM_PROCESSOR_PERFORMANCE_INFORMATION);
if (returnNumCPUs != (int)SystemInfo.dwNumberOfProcessors)
{
fprintf(stderr, "NtQuery didn't return expected amount of data\n");
fprintf(stderr, "Expected data for %i CPUs, returned %i\n",
(int)SystemInfo.dwNumberOfProcessors, returnNumCPUs);
exit(1);
}
// Zero entries not returned by NtQuery
ZeroMemory((PCHAR)&PerfCntrs->EndInfo[returnNumCPUs],
sizeof(SYSTEM_PROCESSOR_PERFORMANCE_INFORMATION)*
(MAXCPUS +1 - returnNumCPUs));
// Total all of the CPUs
// KernelTime needs to be fixed-up; it includes both idle &
// true kernel time
// Note that kernel time also includes DpcTime & InterruptTime, but
// I like this.
for (i=0; i < returnNumCPUs; i++)
{
PerfCntrs->EndInfo[i].KernelTime.QuadPart -= PerfCntrs->EndInfo[i].IdleTime.QuadPart;
PerfCntrs->EndInfo[MAXCPUS].IdleTime.QuadPart += PerfCntrs->EndInfo[i].IdleTime.QuadPart;
PerfCntrs->EndInfo[MAXCPUS].KernelTime.QuadPart += PerfCntrs->EndInfo[i].KernelTime.QuadPart;
PerfCntrs->EndInfo[MAXCPUS].UserTime.QuadPart += PerfCntrs->EndInfo[i].UserTime.QuadPart;
PerfCntrs->EndInfo[MAXCPUS].DpcTime.QuadPart += PerfCntrs->EndInfo[i].DpcTime.QuadPart;
PerfCntrs->EndInfo[MAXCPUS].InterruptTime.QuadPart += PerfCntrs->EndInfo[i].InterruptTime.QuadPart;
PerfCntrs->EndInfo[MAXCPUS].InterruptCount += PerfCntrs->EndInfo[i].InterruptCount;
}
} /* RestartPerfCntrs */
