Distorted Android

///////////////////////////////////////////////////////////////////////////////////////////////////

// Copyright 2016 Leszek Koltunski                                                               //

//                                                                                               //

// This file is part of Distorted.                                                               //

//                                                                                               //

// Distorted is free software: you can redistribute it and/or modify                             //

// it under the terms of the GNU General Public License as published by                          //

// the Free Software Foundation, either version 2 of the License, or                             //

// (at your option) any later version.                                                           //

//                                                                                               //

// Distorted is distributed in the hope that it will be useful,                                  //

// but WITHOUT ANY WARRANTY; without even the implied warranty of                                //

// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the                                 //

// GNU General Public License for more details.                                                  //

//                                                                                               //

// You should have received a copy of the GNU General Public License                             //

// along with Distorted.  If not, see <http://www.gnu.org/licenses/>.                            //

///////////////////////////////////////////////////////////////////////////////////////////////////

package org.distorted.library.effect;

import android.content.res.Resources;

import android.opengl.GLES30;

import org.distorted.library.Distorted;

import org.distorted.library.DistortedFramebuffer;

import org.distorted.library.DistortedOutputSurface;

import org.distorted.library.DistortedRenderState;

import org.distorted.library.R;

import org.distorted.library.message.EffectMessage;

import org.distorted.library.program.DistortedProgram;

import org.distorted.library.program.FragmentCompilationException;

import org.distorted.library.program.FragmentUniformsException;

import org.distorted.library.program.LinkingException;

import org.distorted.library.program.VertexCompilationException;

import org.distorted.library.program.VertexUniformsException;

import org.distorted.library.type.Static;

import org.distorted.library.type.Static1D;

import org.distorted.library.type.Static2D;

import org.distorted.library.type.Static3D;

import org.distorted.library.type.Static4D;

import org.distorted.library.type.Static5D;

import java.io.InputStream;

import java.nio.ByteBuffer;

import java.nio.ByteOrder;

import java.nio.FloatBuffer;

///////////////////////////////////////////////////////////////////////////////////////////////////

class EffectQueuePostprocess extends EffectQueue

  {

  private static final int MAX_HALO = 50;    // Support effects creating up to MAX_HALO pixels wide 'halo' around the object.

  private static final int POS_DATA_SIZE= 2; // Post Program: size of the position data in elements

  private static final int TEX_DATA_SIZE= 2; // Post Program: size of the texture coordinate data in elements.

  private static final int NUM_UNIFORMS = 5;

  private static final int NUM_CACHE    = 0;

  private static final int INDEX = Effect.POSTPROCESS;

  private static final FloatBuffer mQuadPositions, mQuadTexture, mQuadTextureInv;

  static

    {

    int dataLength      = 4;

    int bytes_per_float = 4;

    float[] position  = { -0.5f, -0.5f,  -0.5f, 0.5f,  0.5f,-0.5f,  0.5f, 0.5f };

    float[] textureNor= {  0.0f,  0.0f,   0.0f, 1.0f,  1.0f, 0.0f,  1.0f, 1.0f };

    float[] textureInv= {  0.0f,  0.0f,   1.0f, 0.0f,  0.0f, 1.0f,  1.0f, 1.0f };

    mQuadPositions = ByteBuffer.allocateDirect(POS_DATA_SIZE*dataLength*bytes_per_float).order(ByteOrder.nativeOrder()).asFloatBuffer();

    mQuadPositions.put(position).position(0);

    mQuadTexture= ByteBuffer.allocateDirect(TEX_DATA_SIZE*dataLength*bytes_per_float).order(ByteOrder.nativeOrder()).asFloatBuffer();

    mQuadTexture.put(textureNor).position(0);

    mQuadTextureInv= ByteBuffer.allocateDirect(TEX_DATA_SIZE*dataLength*bytes_per_float).order(ByteOrder.nativeOrder()).asFloatBuffer();

    mQuadTextureInv.put(textureInv).position(0);

    }

  int mQualityLevel;

  float mQualityScale;

  private int mHalo;

  /////////////////////////////////////////////////////////////////////////////////

  // BLUR effect

  private static final float GAUSSIAN[] =   // G(0.00), G(0.03), G(0.06), ..., G(3.00), 0

    {                                       // where G(x)= (1/(sqrt(2*PI))) * e^(-(x^2)/2). The last 0 terminates.

    0.398948f, 0.398769f, 0.398231f, 0.397336f, 0.396086f, 0.394485f, 0.392537f, 0.390247f, 0.387622f, 0.384668f,

    0.381393f, 0.377806f, 0.373916f, 0.369733f, 0.365268f, 0.360532f, 0.355538f, 0.350297f, 0.344823f, 0.339129f,

    0.333229f, 0.327138f, 0.320868f, 0.314436f, 0.307856f, 0.301142f, 0.294309f, 0.287373f, 0.280348f, 0.273248f,

    0.266089f, 0.258884f, 0.251648f, 0.244394f, 0.237135f, 0.229886f, 0.222657f, 0.215461f, 0.208311f, 0.201217f,

    0.194189f, 0.187238f, 0.180374f, 0.173605f, 0.166940f, 0.160386f, 0.153951f, 0.147641f, 0.141462f, 0.135420f,

    0.129520f, 0.123765f, 0.118159f, 0.112706f, 0.107408f, 0.102266f, 0.097284f, 0.092461f, 0.087797f, 0.083294f,

    0.078951f, 0.074767f, 0.070741f, 0.066872f, 0.063158f, 0.059596f, 0.056184f, 0.052920f, 0.049801f, 0.046823f,

    0.043984f, 0.041280f, 0.038707f, 0.036262f, 0.033941f, 0.031740f, 0.029655f, 0.027682f, 0.025817f, 0.024056f,

    0.022395f, 0.020830f, 0.019357f, 0.017971f, 0.016670f, 0.015450f, 0.014305f, 0.013234f, 0.012232f, 0.011295f,

    0.010421f, 0.009606f, 0.008847f, 0.008140f, 0.007483f, 0.006873f, 0.006307f, 0.005782f, 0.005296f, 0.004847f,

    0.004432f, 0.000000f

    };

  private static final int NUM_GAUSSIAN = GAUSSIAN.length-2;

  // The (fixed-function-sampled) Gaussian Blur kernels are of the size k0=1, k1=2, k2=2, k3=3, k4=3, k5=4, k6=4,...

  // i.e. k(i)=floor((i+3)/2).  (the 'i' in k(i) means 'blur taking into account the present pixel and 'i' pixels

  // in all 4 directions)

  // We need room for MAX_BLUR of them, and sum(i=0...N, floor((i+3)/2)) = N + floor(N*N/4)

  private static float[] weightsCache = new float[MAX_HALO + MAX_HALO*MAX_HALO/4];

  private static float[] offsetsCache = new float[MAX_HALO + MAX_HALO*MAX_HALO/4];

  private static DistortedProgram mBlur1Program, mBlur2Program;

  private static int mRadius1H,mOffsets1H,mWeights1H,mDepth1H, mColorTexture1H;

  private static int mRadius2H,mOffsets2H,mWeights2H,mDepth2H, mColorTexture2H;

  private static float[] mWeights = new float[MAX_HALO];

  private static float[] mOffsets = new float[MAX_HALO];

  /////////////////////////////////////////////////////////////////////////////////

  // GLOW effect

///////////////////////////////////////////////////////////////////////////////////////////////////

  EffectQueuePostprocess(long id)

    { 

    super(id,NUM_UNIFORMS,NUM_CACHE,INDEX );

    mQualityLevel = 0;

    mQualityScale = 1.0f;

    }

///////////////////////////////////////////////////////////////////////////////////////////////////

  static void createProgram(Resources resources)

  throws FragmentCompilationException,VertexCompilationException,VertexUniformsException,FragmentUniformsException,LinkingException

    {

    final InputStream blur1VertexStream   = resources.openRawResource(R.raw.blur_vertex_shader);

    final InputStream blur1FragmentStream = resources.openRawResource(R.raw.blur1_fragment_shader);

    try

      {

      mBlur1Program = new DistortedProgram(blur1VertexStream,blur1FragmentStream,

                                          Distorted.GLSL_VERSION,

                                          Distorted.GLSL_VERSION + "#define MAX_BLUR "+MAX_HALO, Distorted.GLSL);

      }

    catch(Exception e)

      {

      android.util.Log.e("EFFECTS", "exception trying to compile BLUR1 program: "+e.getMessage());

      throw new RuntimeException(e.getMessage());

      }

    int blur1ProgramH = mBlur1Program.getProgramHandle();

    mRadius1H       = GLES30.glGetUniformLocation( blur1ProgramH, "u_Radius");

    mOffsets1H      = GLES30.glGetUniformLocation( blur1ProgramH, "u_Offsets");

    mWeights1H      = GLES30.glGetUniformLocation( blur1ProgramH, "u_Weights");

    mDepth1H        = GLES30.glGetUniformLocation( blur1ProgramH, "u_Depth");

    mColorTexture1H = GLES30.glGetUniformLocation( blur1ProgramH, "u_ColorTexture");

    final InputStream blur2VertexStream   = resources.openRawResource(R.raw.blur_vertex_shader);

    final InputStream blur2FragmentStream = resources.openRawResource(R.raw.blur2_fragment_shader);

    try

      {

      mBlur2Program = new DistortedProgram(blur2VertexStream,blur2FragmentStream,

                                          Distorted.GLSL_VERSION,

                                          Distorted.GLSL_VERSION + "#define MAX_BLUR "+MAX_HALO, Distorted.GLSL);

      }

    catch(Exception e)

      {

      android.util.Log.e("EFFECTS", "exception trying to compile BLUR2 program: "+e.getMessage());

      // run anyway as compiling Blur2 WILL fail on OpenGL 2.0 contexts

      mBlur2Program = mBlur1Program;

      }

    int blur2ProgramH = mBlur2Program.getProgramHandle();

    mRadius2H       = GLES30.glGetUniformLocation( blur2ProgramH, "u_Radius");

    mOffsets2H      = GLES30.glGetUniformLocation( blur2ProgramH, "u_Offsets");

    mWeights2H      = GLES30.glGetUniformLocation( blur2ProgramH, "u_Weights");

    mDepth2H        = GLES30.glGetUniformLocation( blur2ProgramH, "u_Depth");

    mColorTexture2H = GLES30.glGetUniformLocation( blur2ProgramH, "u_ColorTexture");

    }

///////////////////////////////////////////////////////////////////////////////////////////////////

  private boolean compute(long currTime)

    {

    if( currTime==mTime ) return false;

    if( mTime==0 ) mTime = currTime;

    long step = (currTime-mTime);

    mHalo = 0;

    int halo;

    for(int i=0; i<mNumEffects; i++)

      {

      mCurrentDuration[i] += step;

      if( mInter[0][i]!=null && mInter[0][i].interpolateMain(mUniforms ,NUM_UNIFORMS*i, mCurrentDuration[i], step) )

        {

        for(int j=0; j<mNumListeners; j++)

          EffectMessageSender.newMessage( mListeners.elementAt(j),

                                          EffectMessage.EFFECT_FINISHED,

                                         (mID[i]<<Effect.LENGTH)+Effect.POSTPROCESS,

                                          mName[i],

                                          mObjectID);

        if( PostprocessEffect.isUnity(mName[i], mUniforms, NUM_UNIFORMS*i) )

          {

          remove(i);

          i--;

          continue;

          }

        else mInter[0][i] = null;

        }

      if( mInter[1][i]!=null ) mInter[1][i].interpolateMain( mUniforms, NUM_UNIFORMS*i+1, mCurrentDuration[i], step);

      halo = (int)mUniforms[NUM_UNIFORMS*i];

      if( halo>mHalo ) mHalo = halo;

      }

    mTime = currTime;

    return true;

    }  

///////////////////////////////////////////////////////////////////////////////////////////////////

  protected void moveEffect(int index)

    {

    mUniforms[NUM_UNIFORMS*index  ] = mUniforms[NUM_UNIFORMS*(index+1)  ];

    mUniforms[NUM_UNIFORMS*index+1] = mUniforms[NUM_UNIFORMS*(index+1)+1];

    mUniforms[NUM_UNIFORMS*index+2] = mUniforms[NUM_UNIFORMS*(index+1)+2];

    mUniforms[NUM_UNIFORMS*index+3] = mUniforms[NUM_UNIFORMS*(index+1)+3];

    mUniforms[NUM_UNIFORMS*index+4] = mUniforms[NUM_UNIFORMS*(index+1)+4];

    }

///////////////////////////////////////////////////////////////////////////////////////////////////

// This implements the 'Better separable implementation using GPU fixed function sampling' from

// https://software.intel.com/en-us/blogs/2014/07/15/an-investigation-of-fast-real-time-gpu-based-image-blur-algorithms

  private void computeGaussianKernel(int radius)

    {

    int offset = radius + radius*radius/4;

    if( weightsCache[offset]==0.0f )

      {

      float z, x= 0.0f, P= (float)NUM_GAUSSIAN / (radius>3 ? radius:3);

      mWeights[0] = GAUSSIAN[0];

      float sum   = GAUSSIAN[0];

      int j;

      for(int i=1; i<=radius; i++)

        {

        x += P;

        j = (int)x;

        z = x-j;

        mWeights[i] = (1-z)*GAUSSIAN[j] + z*GAUSSIAN[j+1];

        sum += 2*mWeights[i];

        }

      for(int i=0; i<=radius; i++) mWeights[i] /= sum;

      int numloops = radius/2;

      weightsCache[offset] = mWeights[0];

      offsetsCache[offset] = 0.0f;

      for(int i=0; i<numloops; i++)

        {

        offsetsCache[offset+i+1] = mWeights[2*i+1]*(2*i+1) + mWeights[2*i+2]*(2*i+2);

        weightsCache[offset+i+1] = mWeights[2*i+1] + mWeights[2*i+2];

        offsetsCache[offset+i+1]/= weightsCache[offset+i+1];

        }

      if( radius%2 == 1 )

        {

        int index = offset + radius/2 +1;

        offsetsCache[index]=radius;

        weightsCache[index]=mWeights[radius];

        }

      }

    }

///////////////////////////////////////////////////////////////////////////////////////////////////

  int getHalo()

    {

    return mNumEffects>0 ? mHalo : 0;

    }

///////////////////////////////////////////////////////////////////////////////////////////////////

  int postprocess(long time, DistortedOutputSurface surface)

    {

    int numRenders = 0;

    if( mNumEffects>0 )

      {

      compute(time);

      for(int i=0; i<mNumEffects; i++)

        {

        if (mName[i] == PostprocessEffect.BLUR )

          {

          blur(NUM_UNIFORMS*i,surface);

          numRenders += 2;

          }

        else if (mName[i] == PostprocessEffect.GLOW )

          {

          glow(NUM_UNIFORMS*i,surface);

          numRenders += 2;

          }

        }

      }

    return numRenders;

    }

///////////////////////////////////////////////////////////////////////////////////////////////////

  private void blur(int index, DistortedOutputSurface surface)

    {

    DistortedFramebuffer buffer = surface.mBuffer[mQualityLevel];

    GLES30.glBindFramebuffer(GLES30.GL_FRAMEBUFFER, buffer.mFBOH[0]);

    float w1 = buffer.mWidth;

    float h1 = buffer.mHeight;

    int radius = (int)(mUniforms[index]*mQualityScale);

    if( radius>=MAX_HALO ) radius = MAX_HALO-1;

    computeGaussianKernel(radius);

    int offset = radius + radius*radius/4;

    radius = (radius+1)/2;

    // horizontal blur

    GLES30.glViewport(0, 0, (int)w1, (int)h1);

    mBlur1Program.useProgram();

    GLES30.glFramebufferTexture2D(GLES30.GL_FRAMEBUFFER, GLES30.GL_COLOR_ATTACHMENT0, GLES30.GL_TEXTURE_2D, buffer.mColorH[1], 0);

    GLES30.glActiveTexture(GLES30.GL_TEXTURE0);

    GLES30.glBindTexture(GLES30.GL_TEXTURE_2D, buffer.mColorH[0]);

    GLES30.glUniform1fv( mWeights1H, radius+1, weightsCache,offset);

    GLES30.glUniform1i( mRadius1H, radius);

    GLES30.glUniform1f( mDepth1H , 1.0f-surface.mNear);

    GLES30.glUniform1i( mColorTexture1H , 0 );

    for(int i=0; i<=radius; i++) mOffsets[i] = offsetsCache[offset+i]/h1;

    GLES30.glUniform1fv( mOffsets1H ,radius+1, mOffsets,0);

    GLES30.glVertexAttribPointer(mBlur1Program.mAttribute[0], POS_DATA_SIZE, GLES30.GL_FLOAT, false, 0, mQuadPositions);

    GLES30.glVertexAttribPointer(mBlur1Program.mAttribute[1], TEX_DATA_SIZE, GLES30.GL_FLOAT, false, 0, mQuadTexture);

    DistortedRenderState.useStencilMark();

    GLES30.glDrawArrays(GLES30.GL_TRIANGLE_STRIP, 0, 4);

    DistortedRenderState.unuseStencilMark();

    GLES30.glActiveTexture(GLES30.GL_TEXTURE0);

    GLES30.glBindTexture(GLES30.GL_TEXTURE_2D, 0);

    // vertical blur

    mBlur2Program.useProgram();

    GLES30.glFramebufferTexture2D(GLES30.GL_FRAMEBUFFER, GLES30.GL_COLOR_ATTACHMENT0, GLES30.GL_TEXTURE_2D, buffer.mColorH[0], 0);

    GLES30.glActiveTexture(GLES30.GL_TEXTURE0);

    GLES30.glBindTexture(GLES30.GL_TEXTURE_2D, buffer.mColorH[1]);

    GLES30.glColorMask(true,true,true,true);

    GLES30.glClearColor(0.0f,0.0f,0.0f,0.0f);

    GLES30.glClear(GLES30.GL_COLOR_BUFFER_BIT);

    GLES30.glUniform1fv( mWeights2H, radius+1, weightsCache,offset);

    GLES30.glUniform1i( mRadius2H, radius);

    GLES30.glUniform1f( mDepth2H , 1.0f-surface.mNear);

    GLES30.glUniform1i( mColorTexture2H , 0 );

    for(int i=0; i<=radius; i++) mOffsets[i] = offsetsCache[offset+i]/w1;

    GLES30.glUniform1fv( mOffsets2H ,radius+1, mOffsets,0);

    GLES30.glVertexAttribPointer(mBlur2Program.mAttribute[0], POS_DATA_SIZE, GLES30.GL_FLOAT, false, 0, mQuadPositions);

    GLES30.glVertexAttribPointer(mBlur2Program.mAttribute[1], TEX_DATA_SIZE, GLES30.GL_FLOAT, false, 0, mQuadTexture);

    DistortedRenderState.useStencilMark();

    GLES30.glDrawArrays(GLES30.GL_TRIANGLE_STRIP, 0, 4);

    DistortedRenderState.unuseStencilMark();

    GLES30.glActiveTexture(GLES30.GL_TEXTURE0);

    GLES30.glBindTexture(GLES30.GL_TEXTURE_2D, 0);

    }

///////////////////////////////////////////////////////////////////////////////////////////////////

  private void glow(int index, DistortedOutputSurface surface)

    {

    }

///////////////////////////////////////////////////////////////////////////////////////////////////

  synchronized long add(PostprocessEffect pe)

    {

    if( mMax[INDEX]>mNumEffects )

      {

      int dim0 = 0;

      if( pe.mDynamic0 != null )

        {

        mInter[0][mNumEffects] = pe.mDynamic0;

        dim0 = pe.mDynamic0.getDimension();

        }

      else

        {

        mInter[0][mNumEffects] = null;

        if( pe.mStatic0 != null )

          {

          Static s = pe.mStatic0;

          dim0 = s.getDimension();

          switch( dim0 )

            {

            case 5 : mUniforms[NUM_UNIFORMS*mNumEffects + 4] = ((Static5D)s).getV();

            case 4 : mUniforms[NUM_UNIFORMS*mNumEffects + 3] = ((Static4D)s).getW();

            case 3 : mUniforms[NUM_UNIFORMS*mNumEffects + 2] = ((Static3D)s).getZ();

            case 2 : mUniforms[NUM_UNIFORMS*mNumEffects + 1] = ((Static2D)s).getY();

            case 1 : mUniforms[NUM_UNIFORMS*mNumEffects    ] = ((Static1D)s).getX();

            }

          }

        }

      mInter[1][mNumEffects] = null;

      if( pe.mStatic1 != null )

        {

        Static s = pe.mStatic1;

        int dim1 = s.getDimension();

        switch( dim1 )

          {

          case 5 : mUniforms[NUM_UNIFORMS*mNumEffects + dim0 + 4] = ((Static5D)s).getV();

          case 4 : mUniforms[NUM_UNIFORMS*mNumEffects + dim0 + 3] = ((Static4D)s).getW();

          case 3 : mUniforms[NUM_UNIFORMS*mNumEffects + dim0 + 2] = ((Static3D)s).getZ();

          case 2 : mUniforms[NUM_UNIFORMS*mNumEffects + dim0 + 1] = ((Static2D)s).getY();

          case 1 : mUniforms[NUM_UNIFORMS*mNumEffects + dim0    ] = ((Static1D)s).getX();

          }

        }

      return addBase(pe);

      }

    return -1;

    }

  }