libogre-1.9-dev 1.9.0+dfsg1-7 / usr / include / OGRE / OgreInstanceManager.h

/*
-----------------------------------------------------------------------------
This source file is part of OGRE
    (Object-oriented Graphics Rendering Engine)
For the latest info, see http://www.ogre3d.org/

Copyright (c) 2000-2013 Torus Knot Software Ltd

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
-----------------------------------------------------------------------------
*/
#ifndef __InstanceManager_H__
#define __InstanceManager_H__

#include "OgrePrerequisites.h"
#include "OgreMesh.h"
#include "OgreRenderOperation.h"
#include "OgreHeaderPrefix.h"

namespace Ogre
{
    /** \addtogroup Core
    *  @{
    */
    /** \addtogroup Scene
    *  @{
    */

    /** This is the main starting point for the new instancing system.
        Each InstanceManager can control one technique and one mesh, but it can manage
        multiple materials at the same time.
        @see SceneManager::createInstanceManager, which creates this InstanceManager. Each one
        must have a unique name. It's wasteless to create two InstanceManagers with the same
        mesh reference, instancing technique and instances per batch count.
        This class takes care of managing batches automatically, so that more are created when
        needed, and reuse existing ones as much as possible; thus the user doesn't have to worry
        of managing all those low level issues.
        @see InstanceBatch & @see InstanceEntity for more information.

    @remarks
        Design discussion webpage: http://www.ogre3d.org/forums/viewtopic.php?f=4&t=59902
    @author
        Matias N. Goldberg ("dark_sylinc")
    @version
        1.0
     */
    class _OgreExport InstanceManager : public FactoryAlloc
    {
    public:
        enum InstancingTechnique
        {
            ShaderBased,            ///< Any SM 2.0+ @see InstanceBatchShader
            TextureVTF,             ///< Needs Vertex Texture Fetch & SM 3.0+ @see InstanceBatchVTF
            HWInstancingBasic,      ///< Needs SM 3.0+ and HW instancing support @see InstanceBatchHW
            HWInstancingVTF,        ///< Needs SM 3.0+, HW instancing support & VTF @see InstanceBatchHW_VTF
            InstancingTechniquesCount
        };

        /** Values to be used in setSetting() & BatchSettings::setting */
        enum BatchSettingId
        {
            /// Makes all batches from same material cast shadows
            CAST_SHADOWS        = 0,
            /// Makes each batch to display it's bounding box. Useful for debugging or profiling
            SHOW_BOUNDINGBOX,

            NUM_SETTINGS
        };

    private:
        struct BatchSettings
        {
            //These are all per material
            bool setting[NUM_SETTINGS];

            BatchSettings()
            {
                setting[CAST_SHADOWS]     = true;
                setting[SHOW_BOUNDINGBOX] = false;
            }
        };

        typedef vector<InstanceBatch*>::type        InstanceBatchVec;   //vec[batchN] = Batch
        typedef map<String, InstanceBatchVec>::type InstanceBatchMap;   //map[materialName] = Vec

        typedef map<String, BatchSettings>::type    BatchSettingsMap;

        const String            mName;                  //Not the name of the mesh
        MeshPtr                 mMeshReference;
        InstanceBatchMap        mInstanceBatches;
        size_t                  mIdCount;

        InstanceBatchVec        mDirtyBatches;

        RenderOperation         mSharedRenderOperation;

        size_t                  mInstancesPerBatch;
        InstancingTechnique     mInstancingTechnique;
        uint16                  mInstancingFlags;       ///< @see InstanceManagerFlags
        unsigned short          mSubMeshIdx;
        
        BatchSettingsMap        mBatchSettings;
        SceneManager*           mSceneManager;

        size_t                  mMaxLookupTableInstances;
        unsigned char			mNumCustomParams;		//Number of custom params per instance.

        /** Finds a batch with at least one free instanced entity we can use.
            If none found, creates one.
        */
        inline InstanceBatch* getFreeBatch( const String &materialName );

        /** Called when batches are fully exhausted (can't return more instances) so a new batch
            is created.
            For the first time use, it can take big build time.
            It takes care of getting the render operation which will be shared by further batches,
            which decreases their build time, and prevents GPU RAM from skyrocketing.
        @param materialName The material name, to know where to put this batch in the map
        @param firstTime True if this is the first time it is called
        @return The created InstancedManager for convenience
        */
        InstanceBatch* buildNewBatch( const String &materialName, bool firstTime );

        /** @see defragmentBatches overload, this takes care of an array of batches
            for a specific material */
        void defragmentBatches( bool optimizeCull, vector<InstancedEntity*>::type &entities,
								vector<Ogre::Vector4>::type &usedParams,
								InstanceBatchVec &fragmentedBatches );

        /** @see setSetting. This function helps it by setting the given parameter to all batches
            in container.
        */
        void applySettingToBatches( BatchSettingId id, bool value, const InstanceBatchVec &container );

		/** Called when we you use a mesh which has shared vertices, the function creates separate
			vertex/index buffers and also recreates the bone assignments.
		*/
		void unshareVertices(const Ogre::MeshPtr &mesh);

    public:
        InstanceManager( const String &customName, SceneManager *sceneManager,
                         const String &meshName, const String &groupName,
                         InstancingTechnique instancingTechnique, uint16 instancingFlags,
                         size_t instancesPerBatch, unsigned short subMeshIdx, bool useBoneMatrixLookup = false);
        virtual ~InstanceManager();

        const String& getName() const { return mName; }

        SceneManager* getSceneManager() const { return mSceneManager; }

        /** Raises an exception if trying to change it after creating the first InstancedEntity
        @remarks The actual value may be less if the technique doesn't support having so much
        @see getMaxOrBestNumInstancesPerBatches for the usefulness of this function
        @param instancesPerBatch New instances per batch number
        */
        void setInstancesPerBatch( size_t instancesPerBatch );

        /** Sets the size of the lookup table for techniques supporting bone lookup table.
            Raises an exception if trying to change it after creating the first InstancedEntity.
            Setting this value below the number of unique (non-sharing) entity instance animations
            will produce a crash during runtime. Setting this value above will increase memory
            consumption and reduce framerate.
        @remarks The value should be as close but not below the actual value. 
        @param maxLookupTableInstances New size of the lookup table
        */
        void setMaxLookupTableInstances( size_t maxLookupTableInstances );

		/** Sets the number of custom parameters per instance. Some techniques (i.e. HWInstancingBasic)
			support this, but not all of them. They also may have limitations to the max number. All
			instancing implementations assume each instance param is a Vector4 (4 floats).
		@remarks
			This function cannot be called after the first batch has been created. Otherwise
			it will raise an exception. If the technique doesn't support custom params, it will
			raise an exception at the time of building the first InstanceBatch.

			HWInstancingBasic:
				* Each custom params adds an additional float4 TEXCOORD.
			HWInstancingVTF:
				* Not implemented. (Recommendation: Implement this as an additional float4 VTF fetch)
			TextureVTF:
				* Not implemented. (see HWInstancingVTF's recommendation)
			ShaderBased:
				* Not supported.
		@param numCustomParams Number of custom parameters each instance will have. Default: 0
		*/
		void setNumCustomParams( unsigned char numCustomParams );

		unsigned char getNumCustomParams() const
		{ return mNumCustomParams; }

        /** @return Instancing technique this manager was created for. Can't be changed after creation */
        InstancingTechnique getInstancingTechnique() const
        { return mInstancingTechnique; }

        /** Calculates the maximum (or the best amount, depending on flags) of instances
            per batch given the suggested size for the technique this manager was created for.
        @remarks
            This is done automatically when creating an instanced entity, but this function in conjunction
            with @see setInstancesPerBatch allows more flexible control over the amount of instances
            per batch
        @param materialName Name of the material to base on
        @param suggestedSize Suggested amount of instances per batch
        @param flags Flags to pass to the InstanceManager. @see InstanceManagerFlags
        @return The max/best amount of instances per batch given the suggested size and flags
        */
        size_t getMaxOrBestNumInstancesPerBatch( String materialName, size_t suggestedSize, uint16 flags );

        /** @copydoc SceneManager::createInstancedEntity */
        InstancedEntity* createInstancedEntity( const String &materialName );

        /** This function can be useful to improve CPU speed after having too many instances
            created, which where now removed, thus freeing many batches with zero used Instanced Entities
            However the batches aren't automatically removed from memory until the InstanceManager is
            destroyed, or this function is called. This function removes those batches which are completely
            unused (only wasting memory).
        */
        void cleanupEmptyBatches(void);

        /** After creating many entities (which turns in many batches) and then removing entities that
            are in the middle of these batches, there might be many batches with many free entities.
            Worst case scenario, there could be left one batch per entity. Imagine there can be
            80 entities per batch, there are 80 batches, making a total of 6400 entities. Then
            6320 of those entities are removed in a very specific way, which leads to having
            80 batches, 80 entities, and GPU vertex shader still needs to process 6400!
            This is called fragmentation. This function reparents the InstancedEntities
            to fewer batches, in this case leaving only one batch with 80 entities

        @remarks
            This function takes time. Make sure to call this only when you're sure there's
            too much of fragmentation and you won't be creating more InstancedEntities soon
            Also in many cases cleanupEmptyBatches() ought to be enough
            Defragmentation is done per material
            Static batches won't be defragmented. If you want to degragment them, set them
            to dynamic again, and switch back to static after calling this function.

        @param optimizeCulling When true, entities close together will be reorganized
            in the same batch for more efficient CPU culling. This can take more CPU
            time. You want this to be false if you now you're entities are moving very
            randomly which tends them to get separated and spread all over the scene
            (which nullifies any CPU culling)
        */
        void defragmentBatches( bool optimizeCulling );

        /** Applies a setting for all batches using the same material_ existing ones and
            those that will be created in the future.
        @par
            For example setSetting( BatchSetting::CAST_SHADOWS, false ) disables shadow
            casting for all instanced entities (@see MovableObject::setCastShadow)
        @par
            For example setSetting( BatchSetting::SHOW_BOUNDINGBOX, true, "MyMat" )
            will display the bounding box of the batch (not individual InstancedEntities)
            from all batches using material "MyMat"
        @note If the material name hasn't been used, the settings are still stored
            This allows setting up batches before they get even created.
        @param id Setting Id to setup, @see BatchSettings::BatchSettingId
        @param enabled Boolean value. It's meaning depends on the id.
        @param materialName When Blank, the setting is applied to all existing materials
        */
        void setSetting( BatchSettingId id, bool enabled, const String &materialName = StringUtil::BLANK );

        /// If settings for the given material didn't exist, default value is returned
        bool getSetting( BatchSettingId id, const String &materialName ) const;

        /** Returns true if settings were already created for the given material name.
            If false is returned, it means getSetting will return default settings.
        */
        bool hasSettings( const String &materialName ) const
        { return mBatchSettings.find( materialName ) != mBatchSettings.end(); }

        /** @copydoc InstanceBatch::setStaticAndUpdate */
        void setBatchesAsStaticAndUpdate( bool bStatic );

        /** Called by an InstanceBatch when it requests their bounds to be updated for proper culling
        @param dirtyBatch The batch which is dirty, usually same as caller.
        */
        void _addDirtyBatch( InstanceBatch *dirtyBatch );

        /** Called by SceneManager when we told it we have at least one dirty batch */
        void _updateDirtyBatches(void);

        typedef ConstMapIterator<InstanceBatchMap> InstanceBatchMapIterator;
        typedef ConstVectorIterator<InstanceBatchVec> InstanceBatchIterator;

        /// Get non-updateable iterator over instance batches per material
        InstanceBatchMapIterator getInstanceBatchMapIterator(void) const
        { return InstanceBatchMapIterator( mInstanceBatches.begin(), mInstanceBatches.end() ); }

        /** Get non-updateable iterator over instance batches for given material
        @remarks
            Each InstanceBatch pointer may be modified for low level usage (i.e.
            setCustomParameter), but there's no synchronization mechanism when
            multithreading or creating more instances, that's up to the user.
        */
        InstanceBatchIterator getInstanceBatchIterator( const String &materialName ) const
        {
            InstanceBatchMap::const_iterator it = mInstanceBatches.find( materialName );
            if(it != mInstanceBatches.end())
                return InstanceBatchIterator( it->second.begin(), it->second.end() );
            else
                OGRE_EXCEPT(Exception::ERR_INVALID_STATE, "Cannot create instance batch iterator. "
                            "Material " + materialName + " cannot be found.", "InstanceManager::getInstanceBatchIterator");
        }
    };
} // namespace Ogre

#include "OgreHeaderSuffix.h"

#endif // __InstanceManager_H__