[GSoC 2013 - accepted] Ogre 2.0

Threads related to Google Summer of Code
User avatar
dark_sylinc
OGRE Team Member
OGRE Team Member
Posts: 3664
Joined: Sat Jul 21, 2007 4:55 pm
Location: Buenos Aires, Argentina
x 109
Contact:

Re: [GSoC 2013 - accepted] Ogre 2.0

Post by dark_sylinc » Thu Nov 21, 2013 3:09 am

Check last push, has already been done ;)
0 x

Transporter
Minaton
Posts: 933
Joined: Mon Mar 05, 2012 11:37 am
Location: Germany

Re: [GSoC 2013 - accepted] Ogre 2.0

Post by Transporter » Thu Nov 21, 2013 11:31 am

dark_sylinc wrote:Hi, I just realized I talk about this assert issue in page 13 (point 2.8) of the manual. See if that helps (and I've also updated the manual today, so redownload).
I use my Multiscreen-Example and move the main camera node (m_pEgoSceneNode). I use a simple camera manager to move the node:

Code: Select all

class OisNodeControl
{
private:
	Ogre::SceneNode* m_pTarget;
	Ogre::Real m_fTopSpeed;
	Ogre::Vector3 m_vVelocity;
	bool m_bGoingForward;
	bool m_bGoingBack;
	bool m_bGoingLeft;
	bool m_bGoingRight;
	bool m_bGoingUp;
	bool m_bGoingDown;
	bool m_bMouseLeft;
	bool m_bMouseRight;

public:
	OisNodeControl(void);

	~OisNodeControl(void);

	void setTarget(Ogre::SceneNode* target);

	Ogre::SceneNode* getTarget();

	void setTopSpeed(Ogre::Real topSpeed);

	Ogre::Real getTopSpeed();

	void manualStop();

	void frameRenderingQueued(const Ogre::FrameEvent& evt);

	void injectKeyDown(const OIS::KeyEvent& evt);

	void injectKeyUp(const OIS::KeyEvent& evt);

	void injectMouseMove(const OIS::MouseEvent& evt);

	void injectMouseDown(const OIS::MouseEvent& evt, OIS::MouseButtonID id);

	void injectMouseUp(const OIS::MouseEvent& evt, OIS::MouseButtonID id);
};

Code: Select all

#include "OisNodeControl.h"

OisNodeControl::OisNodeControl(void)
	: m_pTarget(NULL)
	, m_fTopSpeed(50.0f)
	, m_vVelocity(Ogre::Vector3::ZERO)
	, m_bGoingForward(false)
	, m_bGoingBack(false)
	, m_bGoingLeft(false)
	, m_bGoingRight(false)
	, m_bGoingUp(false)
	, m_bGoingDown(false)
	, m_bMouseLeft(false)
	, m_bMouseRight(false)
{
}

OisNodeControl::~OisNodeControl(void)
{
}

void OisNodeControl::setTarget(Ogre::SceneNode* target)
{
	m_pTarget = target;
}

Ogre::SceneNode* OisNodeControl::getTarget()
{
	return m_pTarget;
}

void OisNodeControl::setTopSpeed(Ogre::Real topSpeed)
{
	m_fTopSpeed = topSpeed;
}

Ogre::Real OisNodeControl::getTopSpeed()
{
	return m_fTopSpeed;
}

void OisNodeControl::manualStop()
{
	m_vVelocity = Ogre::Vector3::ZERO;
	m_bGoingForward = false;
	m_bGoingBack = false;
	m_bGoingLeft = false;
	m_bGoingRight = false;
	m_bGoingUp = false;
	m_bGoingDown = false;
	m_bMouseLeft = false;
	m_bMouseRight = false;
}

void OisNodeControl::frameRenderingQueued(const Ogre::FrameEvent& evt)
{
	if(m_pTarget == NULL) return;

	Ogre::Vector3 accel = Ogre::Vector3::ZERO;
	Ogre::Vector3 Right = m_pTarget->getOrientation() * Ogre::Vector3::UNIT_Z;
	Right.normalise();
	Ogre::Vector3 Direction = Right.crossProduct(Ogre::Vector3::NEGATIVE_UNIT_Y);
	Direction.normalise();

	if (m_bGoingForward || m_bMouseLeft) accel += Direction;
	if (m_bGoingBack || m_bMouseRight) accel -= Direction;
	if (m_bGoingRight) accel += Right;
	if (m_bGoingLeft) accel -= Right;
	if (m_bGoingUp) accel += Ogre::Vector3::UNIT_Y;
	if (m_bGoingDown) accel -= Ogre::Vector3::UNIT_Y;

	if (accel.squaredLength() != 0)
	{
		accel.normalise();
		m_vVelocity += accel * m_fTopSpeed * evt.timeSinceLastFrame * 10;
	}
	else m_vVelocity -= m_vVelocity * evt.timeSinceLastFrame * 10;

	Ogre::Real tooSmall = std::numeric_limits<Ogre::Real>::epsilon();
	if (m_vVelocity.squaredLength() > m_fTopSpeed * m_fTopSpeed)
	{
		m_vVelocity.normalise();
		m_vVelocity *= m_fTopSpeed;
	}
	else if (m_vVelocity.squaredLength() < tooSmall * tooSmall)
		m_vVelocity = Ogre::Vector3::ZERO;

	if (m_vVelocity != Ogre::Vector3::ZERO)
	{
		m_pTarget->translate(m_vVelocity * evt.timeSinceLastFrame);
		m_pTarget->_getDerivedPositionUpdated();
	}
}

void OisNodeControl::injectKeyDown(const OIS::KeyEvent& evt)
{
	if (evt.key == OIS::KC_W || evt.key == OIS::KC_UP) m_bGoingForward = true;
	else if (evt.key == OIS::KC_S || evt.key == OIS::KC_DOWN) m_bGoingBack = true;
	else if (evt.key == OIS::KC_A || evt.key == OIS::KC_LEFT) m_bGoingLeft = true;
	else if (evt.key == OIS::KC_D || evt.key == OIS::KC_RIGHT) m_bGoingRight = true;
	else if (evt.key == OIS::KC_PGUP) m_bGoingUp = true;
	else if (evt.key == OIS::KC_PGDOWN) m_bGoingDown = true;
}

void OisNodeControl::injectKeyUp(const OIS::KeyEvent& evt)
{
	if (evt.key == OIS::KC_W || evt.key == OIS::KC_UP) m_bGoingForward = false;
	else if (evt.key == OIS::KC_S || evt.key == OIS::KC_DOWN) m_bGoingBack = false;
	else if (evt.key == OIS::KC_A || evt.key == OIS::KC_LEFT) m_bGoingLeft = false;
	else if (evt.key == OIS::KC_D || evt.key == OIS::KC_RIGHT) m_bGoingRight = false;
	else if (evt.key == OIS::KC_PGUP) m_bGoingUp = false;
	else if (evt.key == OIS::KC_PGDOWN) m_bGoingDown = false;
}

void OisNodeControl::injectMouseMove(const OIS::MouseEvent& evt)
{
	m_pTarget->yaw(Ogre::Degree(-evt.state.X.rel * 0.15f));
	//m_pTarget->roll(Ogre::Degree(-evt.state.Y.rel * 0.15f));
	m_pTarget->_getDerivedOrientationUpdated();
	m_fTopSpeed += evt.state.Z.rel * 0.01f;
	if(m_fTopSpeed < 1.0f) m_fTopSpeed = 1.0f;
}

void OisNodeControl::injectMouseDown(const OIS::MouseEvent& evt, OIS::MouseButtonID id)
{
	if(id == OIS::MB_Left) m_bMouseLeft = true;
	else if(id == OIS::MB_Right) m_bMouseRight = true;
}

void OisNodeControl::injectMouseUp(const OIS::MouseEvent& evt, OIS::MouseButtonID id)
{
	m_bMouseLeft = false;
	m_bMouseRight = false;
}
I have an empty scene with a sky box.
0 x

User avatar
dark_sylinc
OGRE Team Member
OGRE Team Member
Posts: 3664
Joined: Sat Jul 21, 2007 4:55 pm
Location: Buenos Aires, Argentina
x 109
Contact:

Re: [GSoC 2013 - accepted] Ogre 2.0

Post by dark_sylinc » Thu Nov 21, 2013 4:27 pm

frameRenderingQueued happens after updateAllTransform which explains the problem. Put the code in a listener that comes before (admittedly I have yet to see what happens with each listeners, some are redundant now, while others are undocumented for stuff like this).

You can safely replace the dummy calls to _getDerivedOrientationUpdated for _getDerivedPositionUpdated() may(all *Updated functions will update all transforms, not just position)

Is it too much if you zip the other files as well?
0 x

User avatar
scrawl
OGRE Expert User
OGRE Expert User
Posts: 1119
Joined: Sat Jan 01, 2011 7:57 pm
Location: Germany
x 2

Re: [GSoC 2013 - accepted] Ogre 2.0

Post by scrawl » Sun Nov 24, 2013 7:06 am

Oh nice! The sample browser works again.
Performance for me is still unchanged though (significantly worse than 1.9)
0 x

User avatar
dark_sylinc
OGRE Team Member
OGRE Team Member
Posts: 3664
Joined: Sat Jul 21, 2007 4:55 pm
Location: Buenos Aires, Argentina
x 109
Contact:

Re: [GSoC 2013 - accepted] Ogre 2.0

Post by dark_sylinc » Sun Nov 24, 2013 4:19 pm

Out of curiosity I ran Ogre 1.9 vs 2.0 in a new real Ubuntu 13.11 box (I've been using VirtualBoxes) and the performance was more or less the same (slightly lower for 2.0 though).

Few things of the top of my head that may affect your measurements:
  • Kernel Tick Rate: We need to switch contexts far more often than 1.9 did, (even if numThreads = 1). Perhaps this can be improved in the future; but right now this means that if your kernel was compiled with a very low tick rate, you may get consistently lower framerates (however Ogre 2.0 will still scale much better when stressing/adding more objects). If this is the problem, power consumption is not affected because the lower framerate is caused by us sleeping.
  • Graphics drivers: We changed the order of some API operations, which some GPU drivers may like more/less. Linux GPU drivers are always a hot topic.
Edit: Just checked, I'm using a tickless kernel:

Code: Select all

cat /boot/config-3.11.0-12-generic | grep HZCONFIG_NO_HZ_COMMON=y
# CONFIG_HZ_PERIODIC is not set
CONFIG_NO_HZ_IDLE=y
# CONFIG_NO_HZ_FULL is not set
CONFIG_NO_HZ=y
CONFIG_RCU_FAST_NO_HZ=y
# CONFIG_HZ_100 is not set
CONFIG_HZ_250=y
# CONFIG_HZ_300 is not set
# CONFIG_HZ_1000 is not set
CONFIG_HZ=250
CONFIG_MACHZ_WDT=m
Executing this test: http://www.advenage.com/topics/linux-ti ... quency.php returned me (not sure how accurate this test could be):
kernel timer interrupt frequency is approx. 4016 Hz or higher
0 x

User avatar
scrawl
OGRE Expert User
OGRE Expert User
Posts: 1119
Joined: Sat Jan 01, 2011 7:57 pm
Location: Germany
x 2

Re: [GSoC 2013 - accepted] Ogre 2.0

Post by scrawl » Mon Nov 25, 2013 1:45 am

The thing is, not even the scaling seems to be working properly for me.
I set the "new instancing" demo to 50x50 entities with No instancing, Linear Skinning, shadows off, Move & animate enabled.
Ogre 2.0 gives me 40 FPS. In 1.9 I have 50 FPS.
Or is this due to more bugs in the instancing demo? Should I write my own profiling test with just Entities?

I tried your commands, here is the output:

Code: Select all

cat /boot/config-`uname -r` | grep HZ
CONFIG_RCU_FAST_NO_HZ=y
CONFIG_NO_HZ=y
# CONFIG_HZ_100 is not set
CONFIG_HZ_250=y
# CONFIG_HZ_300 is not set
# CONFIG_HZ_1000 is not set
CONFIG_HZ=250
CONFIG_MACHZ_WDT=m

Code: Select all

$ ./a.out 
kernel timer interrupt frequency is approx. 4016 Hz or higher
0 x

User avatar
dark_sylinc
OGRE Team Member
OGRE Team Member
Posts: 3664
Joined: Sat Jul 21, 2007 4:55 pm
Location: Buenos Aires, Argentina
x 109
Contact:

Re: [GSoC 2013 - accepted] Ogre 2.0

Post by dark_sylinc » Mon Nov 25, 2013 3:22 pm

Most of the scaling is thrown out of the window when the entity has skeletal animation (for now), though every one else seems to have 2.0>1.9 by a small margin in that sample.

HWBasic is a good way to compare since skeletons are stripped, though I've had issue running the glsl code for Linux.
Since you're the only one having this odd behavior of 2.0<1.9, I'm affraid I can't do much w/out access to your machine/setup. Try debugging and profiling tools to get the culprit.
0 x

User avatar
volca
Gnome
Posts: 393
Joined: Thu Dec 08, 2005 9:57 pm
Contact:

Re: [GSoC 2013 - accepted] Ogre 2.0

Post by volca » Thu Mar 06, 2014 2:07 pm

Hey,

just browsing through the commits - is this intentional?
https://bitbucket.org/sinbad/ogre/commi ... de/Grass.h

in think the vMax construction is wrong. Perhaps change:

Code: Select all

Vector3 vMax( -vMax );
to

Code: Select all

Vector3 vMax( -vMin );
0 x
Image

User avatar
dark_sylinc
OGRE Team Member
OGRE Team Member
Posts: 3664
Joined: Sat Jul 21, 2007 4:55 pm
Location: Buenos Aires, Argentina
x 109
Contact:

Re: [GSoC 2013 - accepted] Ogre 2.0

Post by dark_sylinc » Thu Mar 06, 2014 3:55 pm

volca wrote:Hey,

just browsing through the commits - is this intentional?
https://bitbucket.org/sinbad/ogre/commi ... de/Grass.h

in think the vMax construction is wrong. Perhaps change:

Code: Select all

Vector3 vMax( -vMax );
to

Code: Select all

Vector3 vMax( -vMin );
Fixed in the same day a 1:30hs later ;)
Thanks for keeping an eye on my commits though.
0 x

User avatar
c6burns
Beholder
Posts: 1511
Joined: Fri Feb 22, 2013 4:44 am
Location: Deep behind enemy lines

Re: [GSoC 2013 - accepted] Ogre 2.0

Post by c6burns » Thu Mar 06, 2014 4:00 pm

Image
Alllllways watching ...
0 x

User avatar
volca
Gnome
Posts: 393
Joined: Thu Dec 08, 2005 9:57 pm
Contact:

Re: [GSoC 2013 - accepted] Ogre 2.0

Post by volca » Fri Mar 07, 2014 3:02 pm

Okay, sorry for the false alarm :) I still find it alarming that the self-copy construction seems to be valid (g++ didn't even spit a warning when I tried it).

EDIT: At least clang identifies the problem:

Code: Select all

a.cc:11:8: warning: variable 'a' is uninitialized when used within its own initialization [-Wuninitialized]
        ynt a(a);
            ~ ^
0 x
Image

Post Reply