/*
 * Copyright 1993-2015 NVIDIA Corporation.  All rights reserved.
 *
 * Please refer to the NVIDIA end user license agreement (EULA) associated
 * with this source code for terms and conditions that govern your use of
 * this software. Any use, reproduction, disclosure, or distribution of
 * this software and related documentation outside the terms of the EULA
 * is strictly prohibited.
 *
 */



////////////////////////////////////////////////////////////////////////////////
// KNN kernel
////////////////////////////////////////////////////////////////////////////////
__global__ void KNN(
    TColor *dst,
    int imageW,
    int imageH,
    float Noise,
    float lerpC,
    cudaTextureObject_t texImage
)
{
    const int ix = blockDim.x * blockIdx.x + threadIdx.x;
    const int iy = blockDim.y * blockIdx.y + threadIdx.y;
    //Add half of a texel to always address exact texel centers
    const float x = (float)ix + 0.5f;
    const float y = (float)iy + 0.5f;

    if (ix < imageW && iy < imageH)
    {
        //Normalized counter for the weight threshold
        float fCount = 0;
        //Total sum of pixel weights
        float sumWeights = 0;
        //Result accumulator
        float3 clr = {0, 0, 0};
        //Center of the KNN window
        float4 clr00 = tex2D<float4>(texImage, x, y);

        //Cycle through KNN window, surrounding (x, y) texel
        for (float i = -KNN_WINDOW_RADIUS; i <= KNN_WINDOW_RADIUS; i++)
            for (float j = -KNN_WINDOW_RADIUS; j <= KNN_WINDOW_RADIUS; j++)
            {
                float4     clrIJ = tex2D<float4>(texImage, x + j, y + i);
                float distanceIJ = vecLen(clr00, clrIJ);

                //Derive final weight from color distance
                float   weightIJ = __expf(- (distanceIJ * Noise + (i * i + j * j) * INV_KNN_WINDOW_AREA));

                //Accumulate (x + j, y + i) texel color with computed weight
                clr.x += clrIJ.x * weightIJ;
                clr.y += clrIJ.y * weightIJ;
                clr.z += clrIJ.z * weightIJ;

                //Sum of weights for color normalization to [0..1] range
                sumWeights     += weightIJ;

                //Update weight counter, if KNN weight for current window texel
                //exceeds the weight threshold
                fCount         += (weightIJ > KNN_WEIGHT_THRESHOLD) ? INV_KNN_WINDOW_AREA : 0;
            }

        //Normalize result color by sum of weights
        sumWeights = 1.0f / sumWeights;
        clr.x *= sumWeights;
        clr.y *= sumWeights;
        clr.z *= sumWeights;

        //Choose LERP quotient basing on how many texels
        //within the KNN window exceeded the weight threshold
        float lerpQ = (fCount > KNN_LERP_THRESHOLD) ? lerpC : 1.0f - lerpC;

        //Write final result to global memory
        clr.x = lerpf(clr.x, clr00.x, lerpQ);
        clr.y = lerpf(clr.y, clr00.y, lerpQ);
        clr.z = lerpf(clr.z, clr00.z, lerpQ);
        dst[imageW * iy + ix] = make_color(clr.x, clr.y, clr.z, 0);
    };
}

extern "C"
void cuda_KNN(
    TColor *d_dst,
    int imageW,
    int imageH,
    float Noise,
    float lerpC,
    cudaTextureObject_t texImage
)
{
    dim3 threads(BLOCKDIM_X, BLOCKDIM_Y);
    dim3 grid(iDivUp(imageW, BLOCKDIM_X), iDivUp(imageH, BLOCKDIM_Y));

    KNN<<<grid, threads>>>(d_dst, imageW, imageH, Noise, lerpC, texImage);
}


////////////////////////////////////////////////////////////////////////////////
// Stripped KNN kernel, only highlighting areas with different LERP directions
////////////////////////////////////////////////////////////////////////////////
__global__ void KNNdiag(
    TColor *dst,
    int imageW,
    int imageH,
    float Noise,
    float lerpC,
    cudaTextureObject_t texImage
)
{
    const int ix = blockDim.x * blockIdx.x + threadIdx.x;
    const int iy = blockDim.y * blockIdx.y + threadIdx.y;
    //Add half of a texel to always address exact texel centers
    const float x = (float)ix + 0.5f;
    const float y = (float)iy + 0.5f;

    if (ix < imageW && iy < imageH)
    {
        //Normalized counter for the weight threshold
        float  fCount = 0;
        //Center of the KNN window
        float4  clr00 = tex2D<float4>(texImage, x, y);

        //Cycle through KNN window, surrounding (x, y) texel
        for (float i = -KNN_WINDOW_RADIUS; i <= KNN_WINDOW_RADIUS; i++)
            for (float j = -KNN_WINDOW_RADIUS; j <= KNN_WINDOW_RADIUS; j++)
            {
                float4     clrIJ = tex2D<float4>(texImage, x + j, y + i);
                float distanceIJ = vecLen(clr00, clrIJ);

                //Derive final weight from color and geometric distance
                float weightIJ  = __expf(- (distanceIJ * Noise + (i * i + j * j) * INV_KNN_WINDOW_AREA));

                //Update weight counter, if KNN weight for current window texel
                //exceeds the weight threshold
                fCount += (weightIJ > KNN_WEIGHT_THRESHOLD) ? INV_KNN_WINDOW_AREA : 0.0f;
            }

        //Choose LERP quotient basing on how many texels
        //within the KNN window exceeded the weight threshold
        float lerpQ = (fCount > KNN_LERP_THRESHOLD) ? 1.0f : 0;

        //Write final result to global memory
        dst[imageW * iy + ix] = make_color(lerpQ, 0, (1.0f - lerpQ), 0);
    };
}

extern "C"
void cuda_KNNdiag(
    TColor *d_dst,
    int imageW,
    int imageH,
    float Noise,
    float lerpC,
    cudaTextureObject_t texImage
)
{
    dim3 threads(BLOCKDIM_X, BLOCKDIM_Y);
    dim3 grid(iDivUp(imageW, BLOCKDIM_X), iDivUp(imageH, BLOCKDIM_Y));

    KNNdiag<<<grid, threads>>>(d_dst, imageW, imageH, Noise, lerpC, texImage);
}