Updated with PR feedback. Created a common function to calculate the number of worker threads.

Signed-off-by: rgba16f <82187279+rgba16f@users.noreply.github.com>
This commit is contained in:
rgba16f
2021-10-07 15:24:51 -05:00
parent a2c16fa24f
commit 4b92aa34b7
5 changed files with 56 additions and 11 deletions
@@ -20,6 +20,8 @@
#include <AzCore/Console/IConsole.h>
#include <AzCore/Threading/ThreadUtils.h>
AZ_CVAR(float, cl_jobThreadsConcurrencyRatio, 0.6f, nullptr, AZ::ConsoleFunctorFlags::Null, "Legacy Job system multiplier on the number of hw threads the machine creates at initialization");
AZ_CVAR(uint32_t, cl_jobThreadsNumReserved, 2, nullptr, AZ::ConsoleFunctorFlags::Null, "Legacy Job system number of hardware threads that are reserved for O3DE system threads");
AZ_CVAR(uint32_t, cl_jobThreadsMinNumber, 2, nullptr, AZ::ConsoleFunctorFlags::Null, "Legacy Job system minimum number of worker threads to create after scaling the number of hw threads");
@@ -54,10 +56,7 @@ namespace AZ
int numberOfWorkerThreads = m_numberOfWorkerThreads;
if (numberOfWorkerThreads <= 0) // spawn default number of threads
{
// calc number of job threads = cl_jobThreadsConcurrencyRatio * (number of hardware threads - cl_jobThreadsNumReserved),
// min = cl_jobThreadsMinNumber,
// max = number of hardware threads - cl_jobThreadsNumReserved
uint32_t scaledHardwareThreads = AZ::GetMax<uint32_t>( cl_jobThreadsMinNumber, static_cast<uint32_t>(AZStd::floor( 0.5f + AZ::GetClamp<float>(cl_jobThreadsConcurrencyRatio, 0.0f, 1.0f) * static_cast<float>(AZStd::thread::hardware_concurrency() - cl_jobThreadsNumReserved))));
uint32_t scaledHardwareThreads = Threading::CalcNumWorkerThreads(cl_jobThreadsConcurrencyRatio, cl_jobThreadsMinNumber, cl_jobThreadsNumReserved);
numberOfWorkerThreads = AZ::GetMin(static_cast<unsigned int>(desc.m_workerThreads.capacity()), scaledHardwareThreads);
#if (AZ_TRAIT_MAX_JOB_MANAGER_WORKER_THREADS)
numberOfWorkerThreads = AZ::GetMin(numberOfWorkerThreads, AZ_TRAIT_MAX_JOB_MANAGER_WORKER_THREADS);
@@ -12,11 +12,12 @@
#include <AzCore/Serialization/SerializeContext.h>
#include <AzCore/Serialization/EditContext.h>
#include <AzCore/Math/MathUtils.h>
#include <AzCore/Threading/ThreadUtils.h>
// Create a cvar as a central location for experimentation with switching from the Job system to TaskGraph system.
AZ_CVAR(bool, cl_activateTaskGraph, false, nullptr, AZ::ConsoleFunctorFlags::Null, "Flag clients of TaskGraph to switch between jobs/taskgraph (Note does not disable task graph system)");
AZ_CVAR(float, cl_taskGraphThreadsConcurrencyRatio, 1.0f, nullptr, AZ::ConsoleFunctorFlags::Null, "TaskGraph calculate the number of worker threads to spawn by scaling the number of hw threads, value is clamped between 0.0f and 1.0f");
AZ_CVAR(uint32_t, cl_taskGraphThreadsNumReserved, 2, nullptr, AZ::ConsoleFunctorFlags::Null, "TaskGraph number of hardware threads that are reserved for O3DE system threads");
AZ_CVAR(uint32_t, cl_taskGraphThreadsNumReserved, 2, nullptr, AZ::ConsoleFunctorFlags::Null, "TaskGraph number of hardware threads that are reserved for O3DE system threads. Value is clamped between 0 and the number of logical cores in the system");
AZ_CVAR(uint32_t, cl_taskGraphThreadsMinNumber, 2, nullptr, AZ::ConsoleFunctorFlags::Null, "TaskGraph minimum number of worker threads to create after scaling the number of hw threads");
static constexpr uint32_t TaskExecutorServiceCrc = AZ_CRC_CE("TaskExecutorService");
@@ -29,13 +30,9 @@ namespace AZ
if (Interface<TaskGraphActiveInterface>::Get() == nullptr)
{
// calc number of worker threads = cl_taskGraphThreadsConcurrencyRatio * (number of hardware threads - cl_taskGraphThreadsNumReserved),
// min = cl_taskGraphThreadsMinNumber,
// max = number of hardware threads - cl_taskGraphThreadsNumReserved
uint32_t scaledHardwareThreads = AZ::GetMax<uint32_t>( cl_taskGraphThreadsMinNumber, static_cast<uint32_t>(AZStd::floor( 0.5f + AZ::GetClamp<float>(cl_taskGraphThreadsConcurrencyRatio, 0.0f, 1.0f) * static_cast<float>(AZStd::thread::hardware_concurrency() - cl_taskGraphThreadsNumReserved))));
m_taskExecutor = aznew TaskExecutor(scaledHardwareThreads);
Interface<TaskGraphActiveInterface>::Register(this); // small window that another thread can try to use taskgraph between this line and the set instance.
m_taskExecutor = aznew TaskExecutor(Threading::CalcNumWorkerThreads(cl_taskGraphThreadsConcurrencyRatio, cl_taskGraphThreadsMinNumber, cl_taskGraphThreadsNumReserved));
TaskExecutor::SetInstance(m_taskExecutor);
Interface<TaskGraphActiveInterface>::Register(this);
}
}
@@ -0,0 +1,25 @@
/*
* Copyright (c) Contributors to the Open 3D Engine Project.
* For complete copyright and license terms please see the LICENSE at the root of this distribution.
*
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
*/
#include <AzCore/Threading/ThreadUtils.h>
#include <AzCore/std/parallel/thread.h>
#include <AzCore/Math/MathUtils.h>
namespace AZ::Threading
{
uint32_t CalcNumWorkerThreads(float workerThreadsRatio, uint32_t minNumWorkerThreads, uint32_t reservedNumThreads)
{
const uint32_t maxHardwareThreads = AZStd::thread::hardware_concurrency();
const uint32_t numReservedThreads = AZ::GetMin<uint32_t>(reservedNumThreads, maxHardwareThreads); // protect against num reserved being bigger than the number of hw threads
const uint32_t maxWorkerThreads = maxHardwareThreads - numReservedThreads;
const float requestedWorkerThreads = AZ::GetClamp<float>(workerThreadsRatio, 0.0f, 1.0f) * static_cast<float>(maxWorkerThreads);
const uint32_t requestedWorkerThreadsRounded = AZStd::lround(requestedWorkerThreads);
const uint32_t numWorkerThreads = AZ::GetMax<uint32_t>(minNumWorkerThreads, requestedWorkerThreadsRounded);
return numWorkerThreads;
}
};
@@ -0,0 +1,22 @@
/*
* Copyright (c) Contributors to the Open 3D Engine Project.
* For complete copyright and license terms please see the LICENSE at the root of this distribution.
*
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
*/
#pragma once
#include <AzCore/base.h>
namespace AZ::Threading
{
//! Calculates the number of worker threads a system should use based on the number of hardware threads a device has.
//! result = max (minNumWorkerThreads, workerThreadsRatio * (num_hardware_threads - reservedNumThreads))
//! @param workerThreadsRatio scale applied to the calculated maximum number of threads available after reserved threads have been accounted for. Clamped between 0 and 1.
//! @param minNumWorkerThreads minimum value that will be returned. Value is unclamped and can be more than num_hardware_threads.
//! @param reservedNumThreads number of hardware threads to reserve for O3DE system threads. Value clamped to num_hardware_threads.
//! @return number of worker threads for the calling system to allocate
uint32_t CalcNumWorkerThreads(float workerThreadsRatio, uint32_t minNumWorkerThreads, uint32_t reservedNumThreads);
};
@@ -639,6 +639,8 @@ set(FILES
Threading/ThreadSafeDeque.inl
Threading/ThreadSafeObject.h
Threading/ThreadSafeObject.inl
Threading/ThreadUtils.h
Threading/ThreadUtils.cpp
Time/ITime.h
Time/TimeSystemComponent.cpp
Time/TimeSystemComponent.h