Пример #1
0
class GPUFFT(DebugObject):

    """ This is a collection of compute shaders to generate the inverse
    fft efficiently on the gpu, with butterfly FFT and precomputed weights """

    def __init__(self, N, sourceTex, normalizationFactor):
        """ Creates a new fft instance. The source texture has to specified
        from the begining, as the shaderAttributes are pregenerated for
        performance reasons """
        DebugObject.__init__(self, "GPU-FFT")

        self.size = N
        self.log2Size = int(math.log(N, 2))
        self.normalizationFactor = normalizationFactor

        # Create a ping and a pong texture, because we can't write to the
        # same texture while reading to it (that would lead to unexpected
        # behaviour, we could solve that by using an appropriate thread size,
        # but it works fine so far)
        self.pingTexture = Texture("FFTPing")
        self.pingTexture.setup2dTexture(
            self.size, self.size, Texture.TFloat, Texture.FRgba32)
        self.pongTexture = Texture("FFTPong")
        self.pongTexture.setup2dTexture(
            self.size, self.size, Texture.TFloat, Texture.FRgba32)
        self.sourceTex = sourceTex

        for tex in [self.pingTexture, self.pongTexture, sourceTex]:
            tex.setMinfilter(Texture.FTNearest)
            tex.setMagfilter(Texture.FTNearest)
            tex.setWrapU(Texture.WMClamp)
            tex.setWrapV(Texture.WMClamp)

        # Pregenerate weights & indices for the shaders
        self._computeWeighting()

        # Pre generate the shaders, we have 2 passes: Horizontal and Vertical
        # which both execute log2(N) times with varying radii
        self.horizontalFFTShader = BetterShader.loadCompute(
            "Shader/Water/HorizontalFFT.compute")
        self.horizontalFFT = NodePath("HorizontalFFT")
        self.horizontalFFT.setShader(self.horizontalFFTShader)
        self.horizontalFFT.setShaderInput(
            "precomputedWeights", self.weightsLookupTex)
        self.horizontalFFT.setShaderInput("N", LVecBase2i(self.size))

        self.verticalFFTShader = BetterShader.loadCompute(
            "Shader/Water/VerticalFFT.compute")
        self.verticalFFT = NodePath("VerticalFFT")
        self.verticalFFT.setShader(self.verticalFFTShader)
        self.verticalFFT.setShaderInput(
            "precomputedWeights", self.weightsLookupTex)
        self.verticalFFT.setShaderInput("N", LVecBase2i(self.size))

        # Create a texture where the result is stored
        self.resultTexture = Texture("Result")
        self.resultTexture.setup2dTexture(
            self.size, self.size, Texture.TFloat, Texture.FRgba16)
        self.resultTexture.setMinfilter(Texture.FTLinear)
        self.resultTexture.setMagfilter(Texture.FTLinear)

        # Prepare the shader attributes, so we don't have to regenerate them
        # every frame -> That is VERY slow (3ms per fft instance)
        self._prepareAttributes()

    def getResultTexture(self):
        """ Returns the result texture, only contains valid data after execute
        was called at least once """
        return self.resultTexture

    def _generateIndices(self, storageA, storageB):
        """ This method generates the precompute indices, see
        http://cnx.org/content/m12012/latest/image1.png """
        numIter = self.size
        offset = 1
        step = 0
        for i in xrange(self.log2Size):
            numIter = numIter >> 1
            step = offset
            for j in xrange(self.size):
                goLeft = (j / step) % 2 == 1
                indexA, indexB = 0, 0
                if goLeft:
                    indexA, indexB = j - step, j
                else:
                    indexA, indexB = j, j + step

                storageA[i][j] = indexA
                storageB[i][j] = indexB
            offset = offset << 1

    def _generateWeights(self, storage):
        """ This method generates the precomputed weights """

        # Using a custom pi variable should force the calculations to use
        # high precision (I hope so)
        pi = 3.141592653589793238462643383
        numIter = self.size / 2
        numK = 1
        resolutionFloat = float(self.size)
        for i in xrange(self.log2Size):
            start = 0
            end = 2 * numK
            for b in xrange(numIter):
                K = 0
                for k in xrange(start, end, 2):
                    fK = float(K)
                    fNumIter = float(numIter)
                    weightA = Vec2(
                        math.cos(2.0 * pi * fK * fNumIter / resolutionFloat),
                        -math.sin(2.0 * pi * fK * fNumIter / resolutionFloat))
                    weightB = Vec2(
                        -math.cos(2.0 * pi * fK * fNumIter / resolutionFloat),
                        math.sin(2.0 * pi * fK * fNumIter / resolutionFloat))
                    storage[i][k / 2] = weightA
                    storage[i][k / 2 + numK] = weightB
                    K += 1
                start += 4 * numK
                end = start + 2 * numK
            numIter = numIter >> 1
            numK = numK << 1

    def _reverseRow(self, indices):
        """ Reverses the bits in the given row. This is required for inverse
        fft (actually we perform a normal fft, but reversing the bits gives
        us an inverse fft) """
        mask = 0x1
        for j in xrange(self.size):
            val = 0x0
            temp = int(indices[j])  # Int is required, for making a copy
            for i in xrange(self.log2Size):
                t = mask & temp
                val = (val << 1) | t
                temp = temp >> 1
            indices[j] = val

    def _computeWeighting(self):
        """ Precomputes the weights & indices, and stores them in a texture """
        indicesA = [[0 for i in xrange(self.size)]
                    for k in xrange(self.log2Size)]
        indicesB = [[0 for i in xrange(self.size)]
                    for k in xrange(self.log2Size)]
        weights = [[Vec2(0.0) for i in xrange(self.size)]
                   for k in xrange(self.log2Size)]

        self.debug("Pre-Generating indices ..")
        self._generateIndices(indicesA, indicesB)
        self._reverseRow(indicesA[0])
        self._reverseRow(indicesB[0])

        self.debug("Pre-Generating weights ..")
        self._generateWeights(weights)

        # Create storage for the weights & indices
        self.weightsLookup = PNMImage(self.size, self.log2Size, 4)
        self.weightsLookup.setMaxval((2 ** 16) - 1)
        self.weightsLookup.fill(0.0)

        # Populate storage
        for x in xrange(self.size):
            for y in xrange(self.log2Size):
                indexA = indicesA[y][x]
                indexB = indicesB[y][x]
                weight = weights[y][x]

                self.weightsLookup.setRed(x, y, indexA / float(self.size))
                self.weightsLookup.setGreen(x, y, indexB / float(self.size))
                self.weightsLookup.setBlue(x, y, weight.x * 0.5 + 0.5)
                self.weightsLookup.setAlpha(x, y, weight.y * 0.5 + 0.5)

        # Convert storage to texture so we can use it in a shader
        self.weightsLookupTex = Texture("Weights Lookup")
        self.weightsLookupTex.load(self.weightsLookup)
        self.weightsLookupTex.setFormat(Texture.FRgba16)
        self.weightsLookupTex.setMinfilter(Texture.FTNearest)
        self.weightsLookupTex.setMagfilter(Texture.FTNearest)
        self.weightsLookupTex.setWrapU(Texture.WMClamp)
        self.weightsLookupTex.setWrapV(Texture.WMClamp)

    def _prepareAttributes(self):
        """ Prepares all shaderAttributes, so that we have a list of
        ShaderAttributes we can simply walk through in the update method,
        that is MUCH faster than using setShaderInput, as each call to
        setShaderInput forces the generation of a new ShaderAttrib """
        self.attributes = []
        textures = [self.pingTexture, self.pongTexture]

        currentIndex = 0
        firstPass = True

        # Horizontal
        for step in xrange(self.log2Size):
            source = textures[currentIndex]
            dest = textures[1 - currentIndex]

            if firstPass:
                source = self.sourceTex
                firstPass = False

            index = self.log2Size - step - 1
            self.horizontalFFT.setShaderInput("source", source)
            self.horizontalFFT.setShaderInput("dest", dest)
            self.horizontalFFT.setShaderInput(
                "butterflyIndex", LVecBase2i(index))
            self._queueShader(self.horizontalFFT)
            currentIndex = 1 - currentIndex

        # Vertical
        for step in xrange(self.log2Size):
            source = textures[currentIndex]
            dest = textures[1 - currentIndex]
            isLastPass = step == self.log2Size - 1
            if isLastPass:
                dest = self.resultTexture
            index = self.log2Size - step - 1
            self.verticalFFT.setShaderInput("source", source)
            self.verticalFFT.setShaderInput("dest", dest)
            self.verticalFFT.setShaderInput(
                "isLastPass", isLastPass)
            self.verticalFFT.setShaderInput(
                "normalizationFactor", self.normalizationFactor)
            self.verticalFFT.setShaderInput(
                "butterflyIndex", LVecBase2i(index))
            self._queueShader(self.verticalFFT)

            currentIndex = 1 - currentIndex

    def execute(self):
        """ Executes the inverse fft once """
        for attr in self.attributes:
            self._executeShader(attr)

    def _queueShader(self, node):
        """ Internal method to fetch the ShaderAttrib of a node and store it
        in the update queue """
        sattr = node.getAttrib(ShaderAttrib)
        self.attributes.append(sattr)

    def _executeShader(self, sattr):
        """ Internal method to execute a shader by a given ShaderAttrib """
        Globals.base.graphicsEngine.dispatch_compute(
            (self.size / 16, self.size / 16, 1), sattr,
            Globals.base.win.get_gsg())
Пример #2
0
class WaterManager(DebugObject):
    """ Simple wrapper arround WaterDisplacement which combines 3 displacement
    maps into one, and also generates a normal map """
    def __init__(self):
        DebugObject.__init__(self, "WaterManager")
        self.options = OceanOptions()
        self.options.size = 512
        self.options.windDir.normalize()
        self.options.waveAmplitude *= 1e-7

        self.displacementTex = Texture("Displacement")
        self.displacementTex.setup2dTexture(self.options.size,
                                            self.options.size, Texture.TFloat,
                                            Texture.FRgba16)

        self.normalTex = Texture("Normal")
        self.normalTex.setup2dTexture(self.options.size, self.options.size,
                                      Texture.TFloat, Texture.FRgba16)

        self.combineShader = BetterShader.loadCompute(
            "Shader/Water/Combine.compute")

        self.ptaTime = PTAFloat.emptyArray(1)

        # Create a gaussian random texture, as shaders aren't well suited
        # for that
        setRandomSeed(523)
        self.randomStorage = PNMImage(self.options.size, self.options.size, 4)
        self.randomStorage.setMaxval((2**16) - 1)

        for x in xrange(self.options.size):
            for y in xrange(self.options.size):
                rand1 = self._getGaussianRandom() / 10.0 + 0.5
                rand2 = self._getGaussianRandom() / 10.0 + 0.5
                self.randomStorage.setXel(x, y, LVecBase3d(rand1, rand2, 0))
                self.randomStorage.setAlpha(x, y, 1.0)

        self.randomStorageTex = Texture("RandomStorage")
        self.randomStorageTex.load(self.randomStorage)
        self.randomStorageTex.setFormat(Texture.FRgba16)
        self.randomStorageTex.setMinfilter(Texture.FTNearest)
        self.randomStorageTex.setMagfilter(Texture.FTNearest)

        # Create the texture wwhere the intial height (H0 + Omega0) is stored.
        self.texInitialHeight = Texture("InitialHeight")
        self.texInitialHeight.setup2dTexture(self.options.size,
                                             self.options.size, Texture.TFloat,
                                             Texture.FRgba16)
        self.texInitialHeight.setMinfilter(Texture.FTNearest)
        self.texInitialHeight.setMagfilter(Texture.FTNearest)

        # Create the shader which populates the initial height texture
        self.shaderInitialHeight = BetterShader.loadCompute(
            "Shader/Water/InitialHeight.compute")
        self.nodeInitialHeight = NodePath("initialHeight")
        self.nodeInitialHeight.setShader(self.shaderInitialHeight)
        self.nodeInitialHeight.setShaderInput("dest", self.texInitialHeight)
        self.nodeInitialHeight.setShaderInput("N",
                                              LVecBase2i(self.options.size))
        self.nodeInitialHeight.setShaderInput("patchLength",
                                              self.options.patchLength)
        self.nodeInitialHeight.setShaderInput("windDir", self.options.windDir)
        self.nodeInitialHeight.setShaderInput("windSpeed",
                                              self.options.windSpeed)
        self.nodeInitialHeight.setShaderInput("waveAmplitude",
                                              self.options.waveAmplitude)
        self.nodeInitialHeight.setShaderInput("windDependency",
                                              self.options.windDependency)
        self.nodeInitialHeight.setShaderInput("randomTex",
                                              self.randomStorageTex)

        self.attrInitialHeight = self.nodeInitialHeight.getAttrib(ShaderAttrib)

        self.heightTextures = []
        for i in xrange(3):

            tex = Texture("Height")
            tex.setup2dTexture(self.options.size, self.options.size,
                               Texture.TFloat, Texture.FRgba16)
            tex.setMinfilter(Texture.FTNearest)
            tex.setMagfilter(Texture.FTNearest)
            tex.setWrapU(Texture.WMClamp)
            tex.setWrapV(Texture.WMClamp)
            self.heightTextures.append(tex)

        # Also create the shader which updates the spectrum
        self.shaderUpdate = BetterShader.loadCompute(
            "Shader/Water/Update.compute")
        self.nodeUpdate = NodePath("update")
        self.nodeUpdate.setShader(self.shaderUpdate)
        self.nodeUpdate.setShaderInput("outH0x", self.heightTextures[0])
        self.nodeUpdate.setShaderInput("outH0y", self.heightTextures[1])
        self.nodeUpdate.setShaderInput("outH0z", self.heightTextures[2])
        self.nodeUpdate.setShaderInput("initialHeight", self.texInitialHeight)
        self.nodeUpdate.setShaderInput("N", LVecBase2i(self.options.size))
        self.nodeUpdate.setShaderInput("time", self.ptaTime)
        self.attrUpdate = self.nodeUpdate.getAttrib(ShaderAttrib)

        # Create 3 FFTs
        self.fftX = GPUFFT(self.options.size, self.heightTextures[0],
                           self.options.normalizationFactor)
        self.fftY = GPUFFT(self.options.size, self.heightTextures[1],
                           self.options.normalizationFactor)
        self.fftZ = GPUFFT(self.options.size, self.heightTextures[2],
                           self.options.normalizationFactor)

        self.combineNode = NodePath("Combine")
        self.combineNode.setShader(self.combineShader)
        self.combineNode.setShaderInput("displacementX",
                                        self.fftX.getResultTexture())
        self.combineNode.setShaderInput("displacementY",
                                        self.fftY.getResultTexture())
        self.combineNode.setShaderInput("displacementZ",
                                        self.fftZ.getResultTexture())
        self.combineNode.setShaderInput("normalDest", self.normalTex)
        self.combineNode.setShaderInput("displacementDest",
                                        self.displacementTex)
        self.combineNode.setShaderInput("N", LVecBase2i(self.options.size))
        self.combineNode.setShaderInput("choppyScale",
                                        self.options.choppyScale)
        self.combineNode.setShaderInput("gridLength", self.options.patchLength)
        # Store only the shader attrib as this is way faster
        self.attrCombine = self.combineNode.getAttrib(ShaderAttrib)

    def _getGaussianRandom(self):
        """ Returns a gaussian random number """
        u1 = generateRandom()
        u2 = generateRandom()
        if u1 < 1e-6:
            u1 = 1e-6
        return sqrt(-2 * log(u1)) * cos(2 * pi * u2)

    def setup(self):
        """ Setups the manager """

        Globals.base.graphicsEngine.dispatch_compute(
            (self.options.size / 16, self.options.size / 16, 1),
            self.attrInitialHeight, Globals.base.win.get_gsg())

    def getDisplacementTexture(self):
        """ Returns the displacement texture, storing the 3D Displacement in
        the RGB channels """
        return self.displacementTex

    def getNormalTexture(self):
        """ Returns the normal texture, storing the normal in world space in
        the RGB channels """
        return self.normalTex

    def update(self):
        """ Updates the displacement / normal map """

        self.ptaTime[0] = 1 + \
            Globals.clock.getFrameTime() * self.options.timeScale

        Globals.base.graphicsEngine.dispatch_compute(
            (self.options.size / 16, self.options.size / 16, 1),
            self.attrUpdate, Globals.base.win.get_gsg())

        self.fftX.execute()
        self.fftY.execute()
        self.fftZ.execute()

        # Execute the shader which combines the 3 displacement maps into
        # 1 displacement texture and 1 normal texture. We could use dFdx in
        # the fragment shader, however that gives no accurate results as
        # dFdx returns the same value for a 2x2 pixel block
        Globals.base.graphicsEngine.dispatch_compute(
            (self.options.size / 16, self.options.size / 16, 1),
            self.attrCombine, Globals.base.win.get_gsg())
Пример #3
0
class WaterManager(DebugObject):

    """ Simple wrapper arround WaterDisplacement which combines 3 displacement
    maps into one, and also generates a normal map """

    def __init__(self):
        DebugObject.__init__(self, "WaterManager")
        self.options = OceanOptions()
        self.options.size = 512
        self.options.windDir.normalize()
        self.options.waveAmplitude *= 1e-7

        self.displacementTex = Texture("Displacement")
        self.displacementTex.setup2dTexture(
            self.options.size, self.options.size,
            Texture.TFloat, Texture.FRgba16)

        self.normalTex = Texture("Normal")
        self.normalTex.setup2dTexture(
            self.options.size, self.options.size,
            Texture.TFloat, Texture.FRgba16)

        self.combineShader = Shader.loadCompute(Shader.SLGLSL,
            "Shader/WaterFFT/Combine.compute")

        self.ptaTime = PTAFloat.emptyArray(1)

        # Create a gaussian random texture, as shaders aren't well suited
        # for that
        setRandomSeed(523)
        self.randomStorage = PNMImage(self.options.size, self.options.size, 4)
        self.randomStorage.setMaxval((2 ** 16) - 1)

        for x in xrange(self.options.size):
            for y in xrange(self.options.size):
                rand1 = self._getGaussianRandom() / 10.0 + 0.5
                rand2 = self._getGaussianRandom() / 10.0 + 0.5
                self.randomStorage.setXel(x, y, float(rand1), float(rand2), 0)
                self.randomStorage.setAlpha(x, y, 1.0)

        self.randomStorageTex = Texture("RandomStorage")
        self.randomStorageTex.load(self.randomStorage)
        self.randomStorageTex.setFormat(Texture.FRgba16)
        self.randomStorageTex.setMinfilter(Texture.FTNearest)
        self.randomStorageTex.setMagfilter(Texture.FTNearest)

        # Create the texture wwhere the intial height (H0 + Omega0) is stored.
        self.texInitialHeight = Texture("InitialHeight")
        self.texInitialHeight.setup2dTexture(
            self.options.size, self.options.size,
            Texture.TFloat, Texture.FRgba16)
        self.texInitialHeight.setMinfilter(Texture.FTNearest)
        self.texInitialHeight.setMagfilter(Texture.FTNearest)

        # Create the shader which populates the initial height texture
        self.shaderInitialHeight = Shader.loadCompute(Shader.SLGLSL,
            "Shader/WaterFFT/InitialHeight.compute")
        self.nodeInitialHeight = NodePath("initialHeight")
        self.nodeInitialHeight.setShader(self.shaderInitialHeight)
        self.nodeInitialHeight.setShaderInput("dest", self.texInitialHeight)
        self.nodeInitialHeight.setShaderInput(
            "N", LVecBase2i(self.options.size))
        self.nodeInitialHeight.setShaderInput(
            "patchLength", self.options.patchLength)
        self.nodeInitialHeight.setShaderInput("windDir", self.options.windDir)
        self.nodeInitialHeight.setShaderInput(
            "windSpeed", self.options.windSpeed)
        self.nodeInitialHeight.setShaderInput(
            "waveAmplitude", self.options.waveAmplitude)
        self.nodeInitialHeight.setShaderInput(
            "windDependency", self.options.windDependency)
        self.nodeInitialHeight.setShaderInput(
            "randomTex", self.randomStorageTex)

        self.attrInitialHeight = self.nodeInitialHeight.getAttrib(ShaderAttrib)

        self.heightTextures = []
        for i in xrange(3):

            tex = Texture("Height")
            tex.setup2dTexture(self.options.size, self.options.size,
                               Texture.TFloat, Texture.FRgba16)
            tex.setMinfilter(Texture.FTNearest)
            tex.setMagfilter(Texture.FTNearest)
            tex.setWrapU(Texture.WMClamp)
            tex.setWrapV(Texture.WMClamp)
            self.heightTextures.append(tex)

        # Also create the shader which updates the spectrum
        self.shaderUpdate = Shader.loadCompute(Shader.SLGLSL,
            "Shader/WaterFFT/Update.compute")
        self.nodeUpdate = NodePath("update")
        self.nodeUpdate.setShader(self.shaderUpdate)
        self.nodeUpdate.setShaderInput("outH0x", self.heightTextures[0])
        self.nodeUpdate.setShaderInput("outH0y", self.heightTextures[1])
        self.nodeUpdate.setShaderInput("outH0z", self.heightTextures[2])
        self.nodeUpdate.setShaderInput("initialHeight", self.texInitialHeight)
        self.nodeUpdate.setShaderInput("N", LVecBase2i(self.options.size))
        self.nodeUpdate.setShaderInput("time", self.ptaTime)
        self.attrUpdate = self.nodeUpdate.getAttrib(ShaderAttrib)

        # Create 3 FFTs
        self.fftX = GPUFFT(self.options.size, self.heightTextures[0],
                           self.options.normalizationFactor)
        self.fftY = GPUFFT(self.options.size, self.heightTextures[1],
                           self.options.normalizationFactor)
        self.fftZ = GPUFFT(self.options.size, self.heightTextures[2],
                           self.options.normalizationFactor)

        self.combineNode = NodePath("Combine")
        self.combineNode.setShader(self.combineShader)
        self.combineNode.setShaderInput(
            "displacementX", self.fftX.getResultTexture())
        self.combineNode.setShaderInput(
            "displacementY", self.fftY.getResultTexture())
        self.combineNode.setShaderInput(
            "displacementZ", self.fftZ.getResultTexture())
        self.combineNode.setShaderInput("normalDest", self.normalTex)
        self.combineNode.setShaderInput(
            "displacementDest", self.displacementTex)
        self.combineNode.setShaderInput(
            "N", LVecBase2i(self.options.size))
        self.combineNode.setShaderInput(
            "choppyScale", self.options.choppyScale)
        self.combineNode.setShaderInput(
            "gridLength", self.options.patchLength)
        # Store only the shader attrib as this is way faster
        self.attrCombine = self.combineNode.getAttrib(ShaderAttrib)

    def _getGaussianRandom(self):
        """ Returns a gaussian random number """
        u1 = generateRandom()
        u2 = generateRandom()
        if u1 < 1e-6:
            u1 = 1e-6
        return sqrt(-2 * log(u1)) * cos(2 * pi * u2)

    def setup(self):
        """ Setups the manager """

        Globals.base.graphicsEngine.dispatch_compute(
            (self.options.size / 16,
             self.options.size / 16, 1), self.attrInitialHeight,
            Globals.base.win.get_gsg())

    def getDisplacementTexture(self):
        """ Returns the displacement texture, storing the 3D Displacement in
        the RGB channels """
        return self.displacementTex

    def getNormalTexture(self):
        """ Returns the normal texture, storing the normal in world space in
        the RGB channels """
        return self.normalTex

    def update(self):
        """ Updates the displacement / normal map """

        self.ptaTime[0] = 1 + \
            Globals.clock.getFrameTime() * self.options.timeScale

        Globals.base.graphicsEngine.dispatch_compute(
            (self.options.size / 16,
             self.options.size / 16, 1), self.attrUpdate,
            Globals.base.win.get_gsg())

        self.fftX.execute()
        self.fftY.execute()
        self.fftZ.execute()

        # Execute the shader which combines the 3 displacement maps into
        # 1 displacement texture and 1 normal texture. We could use dFdx in
        # the fragment shader, however that gives no accurate results as
        # dFdx returns the same value for a 2x2 pixel block
        Globals.base.graphicsEngine.dispatch_compute(
            (self.options.size / 16,
             self.options.size / 16, 1), self.attrCombine,
            Globals.base.win.get_gsg())
Пример #4
0
class WaterManager:
    """ Simple wrapper around WaterDisplacement which combines 3 displacement
    maps into one, and also generates a normal map """

    def __init__(self, water_options):
        self.options = water_options
        self.options.size = 512
        self.options.wind_dir.normalize()
        self.options.wave_amplitude *= 1e-7

        self.displacement_tex = Texture("Displacement")
        self.displacement_tex.setup_2d_texture(
            self.options.size, self.options.size,
            Texture.TFloat, Texture.FRgba16)

        self.normal_tex = Texture("Normal")
        self.normal_tex.setup_2d_texture(
            self.options.size, self.options.size,
            Texture.TFloat, Texture.FRgba16)

        self.combine_shader = Shader.load_compute(Shader.SLGLSL,
                                                  "/$$rp/rpcore/water/shader/combine.compute")

        self.pta_time = PTAFloat.emptyArray(1)

        # Create a gaussian random texture, as shaders aren't well suited
        # for that
        setRandomSeed(523)
        self.random_storage = PNMImage(self.options.size, self.options.size, 4)
        self.random_storage.setMaxval((2 ** 16) - 1)

        for x in range(self.options.size):
            for y in range(self.options.size):
                rand1 = self._get_gaussian_random() / 10.0 + 0.5
                rand2 = self._get_gaussian_random() / 10.0 + 0.5
                self.random_storage.setXel(x, y, float(rand1), float(rand2), 0)
                self.random_storage.setAlpha(x, y, 1.0)

        self.random_storage_tex = Texture("RandomStorage")
        self.random_storage_tex.load(self.random_storage)
        self.random_storage_tex.set_format(Texture.FRgba16)
        self.random_storage_tex.set_minfilter(Texture.FTNearest)
        self.random_storage_tex.set_magfilter(Texture.FTNearest)

        # Create the texture wwhere the intial height (H0 + Omega0) is stored.
        self.tex_initial_height = Texture("InitialHeight")
        self.tex_initial_height.setup_2d_texture(
            self.options.size, self.options.size,
            Texture.TFloat, Texture.FRgba16)
        self.tex_initial_height.set_minfilter(Texture.FTNearest)
        self.tex_initial_height.set_magfilter(Texture.FTNearest)

        # Create the shader which populates the initial height texture
        self.shader_initial_height = Shader.load_compute(Shader.SLGLSL,
                                                         "/$$rp/rpcore/water/shader/initial_height.compute")
        self.node_initial_height = NodePath("initialHeight")
        self.node_initial_height.set_shader(self.shader_initial_height)
        self.node_initial_height.set_shader_input("dest", self.tex_initial_height)
        self.node_initial_height.set_shader_input(
            "N", LVecBase2i(self.options.size))
        self.node_initial_height.set_shader_input(
            "patchLength", self.options.patch_length)
        self.node_initial_height.set_shader_input("windDir", self.options.wind_dir)
        self.node_initial_height.set_shader_input(
            "windSpeed", self.options.wind_speed)
        self.node_initial_height.set_shader_input(
            "waveAmplitude", self.options.wave_amplitude)
        self.node_initial_height.set_shader_input(
            "windDependency", self.options.wind_dependency)
        self.node_initial_height.set_shader_input(
            "randomTex", self.random_storage_tex)

        self.attr_initial_height = self.node_initial_height.get_attrib(ShaderAttrib)

        self.height_textures = []
        for i in range(3):
            tex = Texture("Height")
            tex.setup_2d_texture(self.options.size, self.options.size,
                                 Texture.TFloat, Texture.FRgba16)
            tex.set_minfilter(Texture.FTNearest)
            tex.set_magfilter(Texture.FTNearest)
            tex.set_wrap_u(Texture.WMClamp)
            tex.set_wrap_v(Texture.WMClamp)
            self.height_textures.append(tex)

        # Also create the shader which updates the spectrum
        self.shader_update = Shader.load_compute(Shader.SLGLSL,
                                                 "/$$rp/rpcore/water/shader/update.compute")
        self.node_update = NodePath("update")
        self.node_update.set_shader(self.shader_update)
        self.node_update.set_shader_input("outH0x", self.height_textures[0])
        self.node_update.set_shader_input("outH0y", self.height_textures[1])
        self.node_update.set_shader_input("outH0z", self.height_textures[2])
        self.node_update.set_shader_input("initialHeight", self.tex_initial_height)
        self.node_update.set_shader_input("N", LVecBase2i(self.options.size))
        self.node_update.set_shader_input("time", self.pta_time)
        self.attr_update = self.node_update.get_attrib(ShaderAttrib)

        # Create 3 FFTs
        self.fftX = GPUFFT(self.options.size, self.height_textures[0],
                           self.options.normalization_factor)
        self.fftY = GPUFFT(self.options.size, self.height_textures[1],
                           self.options.normalization_factor)
        self.fftZ = GPUFFT(self.options.size, self.height_textures[2],
                           self.options.normalization_factor)

        self.combine_node = NodePath("Combine")
        self.combine_node.set_shader(self.combine_shader)
        self.combine_node.set_shader_input(
            "displacementX", self.fftX.get_result_texture())
        self.combine_node.set_shader_input(
            "displacementY", self.fftY.get_result_texture())
        self.combine_node.set_shader_input(
            "displacementZ", self.fftZ.get_result_texture())
        self.combine_node.set_shader_input("normalDest", self.normal_tex)
        self.combine_node.set_shader_input(
            "displacementDest", self.displacement_tex)
        self.combine_node.set_shader_input(
            "N", LVecBase2i(self.options.size))
        self.combine_node.set_shader_input(
            "choppyScale", self.options.choppy_scale)
        self.combine_node.set_shader_input(
            "gridLength", self.options.patch_length)
        # Store only the shader attrib as this is way faster
        self.attr_combine = self.combine_node.get_attrib(ShaderAttrib)

    def _get_gaussian_random(self):
        """ Returns a gaussian random number """
        u1 = generateRandom()
        u2 = generateRandom()
        if u1 < 1e-6:
            u1 = 1e-6
        return sqrt(-2 * log(u1)) * cos(2 * pi * u2)

    def setup(self):
        """ Setups the manager """

        Globals.base.graphicsEngine.dispatch_compute(
            (self.options.size // 16, self.options.size // 16, 1),
            self.attr_initial_height,
            Globals.base.win.get_gsg())

    def get_displacement_texture(self):
        """ Returns the displacement texture, storing the 3D Displacement in
        the RGB channels """
        return self.displacement_tex

    def get_normal_texture(self):
        """ Returns the normal texture, storing the normal in world space in
        the RGB channels """
        return self.normal_tex

    def update(self):
        """ Updates the displacement / normal map """

        self.pta_time[0] = 1 + Globals.clock.get_frame_time() * self.options.time_scale

        Globals.base.graphicsEngine.dispatch_compute(
            (self.options.size // 16, self.options.size // 16, 1),
            self.attr_update,
            Globals.base.win.get_gsg())

        self.fftX.execute()
        self.fftY.execute()
        self.fftZ.execute()

        # Execute the shader which combines the 3 displacement maps into
        # 1 displacement texture and 1 normal texture. We could use dFdx in
        # the fragment shader, however that gives no accurate results as
        # dFdx returns the same value for a 2x2 pixel block
        Globals.base.graphicsEngine.dispatch_compute(
            (self.options.size // 16, self.options.size // 16, 1),
            self.attr_combine,
            Globals.base.win.get_gsg())