aboutsummaryrefslogtreecommitdiff
path: root/src/armnn/NeonTimer.cpp
diff options
context:
space:
mode:
authorDerek Lamberti <derek.lamberti@arm.com>2018-10-01 09:28:57 +0100
committerMatthew Bentham <matthew.bentham@arm.com>2018-10-10 16:16:57 +0100
commite4ba53a85c559d4fe574305276ac815cf7995762 (patch)
treee3f8a3e1b54b2a9cf1e21909f5e4ad9e285be49d /src/armnn/NeonTimer.cpp
parentfcb382af87e79ae4bdc6a604241efbf2533e1737 (diff)
downloadarmnn-e4ba53a85c559d4fe574305276ac815cf7995762.tar.gz
IVGCVSW-1824 Fix slow profiling of neon. (~50% reduced end-to-end time)
Change-Id: I58295c298934317a2b365887bd9f9f6705cd0a21
Diffstat (limited to 'src/armnn/NeonTimer.cpp')
-rw-r--r--src/armnn/NeonTimer.cpp11
1 files changed, 9 insertions, 2 deletions
diff --git a/src/armnn/NeonTimer.cpp b/src/armnn/NeonTimer.cpp
index 1ee0c64103..219edc9680 100644
--- a/src/armnn/NeonTimer.cpp
+++ b/src/armnn/NeonTimer.cpp
@@ -13,24 +13,31 @@
namespace armnn
{
+namespace
+{
+static thread_local auto g_Interceptor = std::make_shared<NeonInterceptorScheduler>(arm_compute::Scheduler::get());
+}
void NeonTimer::Start()
{
m_Kernels.clear();
+ BOOST_ASSERT(g_Interceptor->GetKernels() == nullptr);
+ g_Interceptor->SetKernels(&m_Kernels);
+
m_RealSchedulerType = arm_compute::Scheduler::get_type();
//Note: We can't currently replace a custom scheduler
if(m_RealSchedulerType != arm_compute::Scheduler::Type::CUSTOM)
{
// Keep the real schedule and add NeonInterceptorScheduler as an interceptor
m_RealScheduler = &arm_compute::Scheduler::get();
- auto interceptor = std::make_shared<NeonInterceptorScheduler>(m_Kernels, *m_RealScheduler);
- arm_compute::Scheduler::set(std::static_pointer_cast<arm_compute::IScheduler>(interceptor));
+ arm_compute::Scheduler::set(std::static_pointer_cast<arm_compute::IScheduler>(g_Interceptor));
}
}
void NeonTimer::Stop()
{
// Restore real scheduler
+ g_Interceptor->SetKernels(nullptr);
arm_compute::Scheduler::set(m_RealSchedulerType);
m_RealScheduler = nullptr;
}