COMPMID-3391: Implement Async interfaces
Change-Id: I8168cea5056ff48a0253ebb8c88ea549a3ea69a2
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3335
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
diff --git a/src/runtime/OMP/OMPScheduler.cpp b/src/runtime/OMP/OMPScheduler.cpp
index f67f06f..a1851f0 100644
--- a/src/runtime/OMP/OMPScheduler.cpp
+++ b/src/runtime/OMP/OMPScheduler.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2019 ARM Limited.
+ * Copyright (c) 2017-2020 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -83,6 +83,39 @@
}
}
+void OMPScheduler::schedule_op(ICPPKernel *kernel, const Hints &hints, std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs)
+{
+ ARM_COMPUTE_ERROR_ON_MSG(!kernel, "The child class didn't set the kernel");
+ ARM_COMPUTE_ERROR_ON_MSG(hints.strategy() == StrategyHint::DYNAMIC,
+ "Dynamic scheduling is not supported in OMPScheduler");
+
+ const Window &max_window = kernel->window();
+ const unsigned int num_iterations = max_window.num_iterations(hints.split_dimension());
+ const unsigned int num_threads = std::min(num_iterations, _num_threads);
+
+ if(!kernel->is_parallelisable() || num_threads == 1)
+ {
+ ThreadInfo info;
+ info.cpu_info = &_cpu_info;
+ kernel->run_op(inputs, outputs, max_window, info);
+ }
+ else
+ {
+ const unsigned int num_windows = num_threads;
+ std::vector<IScheduler::Workload> workloads(num_windows);
+ for(unsigned int t = 0; t < num_windows; t++)
+ {
+ //Capture 't' by copy, all the other variables by reference:
+ workloads[t] = [t, &hints, &max_window, &num_windows, &kernel, &inputs, &outputs](const ThreadInfo & info)
+ {
+ Window win = max_window.split_window(hints.split_dimension(), t, num_windows);
+ win.validate();
+ kernel->run_op(inputs, outputs, win, info);
+ };
+ }
+ run_workloads(workloads);
+ }
+}
#ifndef DOXYGEN_SKIP_THIS
void OMPScheduler::run_workloads(std::vector<arm_compute::IScheduler::Workload> &workloads)
{