#pragma version(1) #include "../../../../scriptc/rs_types.rsh" #include "../../../../scriptc/rs_math.rsh" #include "ip.rsh" int height; int width; int radius; uchar4 * InPixel; uchar4 * OutPixel; uchar4 * ScratchPixel; float inBlack; float outBlack; float inWhite; float outWhite; float gamma; float saturation; static float inWMinInB; static float outWMinOutB; static float overInWMinInB; static FilterStruct filterStruct; #pragma rs export_var(height, width, radius, InPixel, OutPixel, ScratchPixel, inBlack, outBlack, inWhite, outWhite, gamma, saturation, InPixel, OutPixel, ScratchPixel, vBlurScript, hBlurScript) #pragma rs export_func(filter, filterBenchmark); rs_script vBlurScript; rs_script hBlurScript; // Store our coefficients here static float gaussian[MAX_RADIUS * 2 + 1]; static rs_matrix3x3 colorMat; static void computeColorMatrix() { // Saturation // Linear weights //float rWeight = 0.3086f; //float gWeight = 0.6094f; //float bWeight = 0.0820f; // Gamma 2.2 weights (we haven't converted our image to linear space yet for perf reasons) float rWeight = 0.299f; float gWeight = 0.587f; float bWeight = 0.114f; float oneMinusS = 1.0f - saturation; rsMatrixSet(&colorMat, 0, 0, oneMinusS * rWeight + saturation); rsMatrixSet(&colorMat, 0, 1, oneMinusS * rWeight); rsMatrixSet(&colorMat, 0, 2, oneMinusS * rWeight); rsMatrixSet(&colorMat, 1, 0, oneMinusS * gWeight); rsMatrixSet(&colorMat, 1, 1, oneMinusS * gWeight + saturation); rsMatrixSet(&colorMat, 1, 2, oneMinusS * gWeight); rsMatrixSet(&colorMat, 2, 0, oneMinusS * bWeight); rsMatrixSet(&colorMat, 2, 1, oneMinusS * bWeight); rsMatrixSet(&colorMat, 2, 2, oneMinusS * bWeight + saturation); inWMinInB = inWhite - inBlack; outWMinOutB = outWhite - outBlack; overInWMinInB = 1.f / inWMinInB; } static void computeGaussianWeights() { // Compute gaussian weights for the blur // e is the euler's number float e = 2.718281828459045f; float pi = 3.1415926535897932f; // g(x) = ( 1 / sqrt( 2 * pi ) * sigma) * e ^ ( -x^2 / 2 * sigma^2 ) // x is of the form [-radius .. 0 .. radius] // and sigma varies with radius. // Based on some experimental radius values and sigma's // we approximately fit sigma = f(radius) as // sigma = radius * 0.4 + 0.6 // The larger the radius gets, the more our gaussian blur // will resemble a box blur since with large sigma // the gaussian curve begins to lose its shape float sigma = 0.4f * (float)radius + 0.6f; // Now compute the coefficints // We will store some redundant values to save some math during // the blur calculations // precompute some values float coeff1 = 1.0f / (sqrt( 2.0f * pi ) * sigma); float coeff2 = - 1.0f / (2.0f * sigma * sigma); float normalizeFactor = 0.0f; float floatR = 0.0f; int r; for(r = -radius; r <= radius; r ++) { floatR = (float)r; gaussian[r + radius] = coeff1 * pow(e, floatR * floatR * coeff2); normalizeFactor += gaussian[r + radius]; } //Now we need to normalize the weights because all our coefficients need to add up to one normalizeFactor = 1.0f / normalizeFactor; for(r = -radius; r <= radius; r ++) { floatR = (float)r; gaussian[r + radius] *= normalizeFactor; } } // This needs to be inline static float4 levelsSaturation(float4 currentPixel) { float3 temp = rsMatrixMultiply(&colorMat, currentPixel.xyz); temp = (clamp(temp, 0.1f, 255.f) - inBlack) * overInWMinInB; temp = pow(temp, (float3)gamma); currentPixel.xyz = clamp(temp * outWMinOutB + outBlack, 0.1f, 255.f); return currentPixel; } static void processNoBlur() { int w, h, r; int count = 0; float inWMinInB = inWhite - inBlack; float outWMinOutB = outWhite - outBlack; float4 currentPixel = 0; for(h = 0; h < height; h ++) { for(w = 0; w < width; w ++) { uchar4 *input = InPixel + h*width + w; //currentPixel.xyz = convert_float3(input.xyz); currentPixel.x = (float)(input->x); currentPixel.y = (float)(input->y); currentPixel.z = (float)(input->z); currentPixel = levelsSaturation(currentPixel); uchar4 *output = OutPixel + h*width + w; //output.xyz = convert_uchar3(currentPixel.xyz); output->x = (uint8_t)currentPixel.x; output->y = (uint8_t)currentPixel.y; output->z = (uint8_t)currentPixel.z; output->w = input->w; } } rsSendToClient(&count, 1, 4, 0); } static void horizontalBlurLevels() { float4 blurredPixel = 0; float4 currentPixel = 0; // Horizontal blur int w, h, r; for(h = 0; h < height; h ++) { uchar4 *output = OutPixel + h*width; for(w = 0; w < width; w ++) { blurredPixel = 0; for(r = -radius; r <= radius; r ++) { // Stepping left and right away from the pixel int validW = w + r; // Clamp to zero and width max() isn't exposed for ints yet if(validW < 0) { validW = 0; } if(validW > width - 1) { validW = width - 1; } //int validW = rsClamp(w + r, 0, width - 1); uchar4 *input = InPixel + h*width + validW; float weight = gaussian[r + radius]; currentPixel.x = (float)(input->x); currentPixel.y = (float)(input->y); currentPixel.z = (float)(input->z); //currentPixel.w = (float)(input->a); blurredPixel.xyz += currentPixel.xyz * weight; } blurredPixel = levelsSaturation(blurredPixel); output->x = (uint8_t)blurredPixel.x; output->y = (uint8_t)blurredPixel.y; output->z = (uint8_t)blurredPixel.z; //output->a = (uint8_t)blurredPixel.w; output++; } } } static void initStructs() { filterStruct.gaussian = gaussian; filterStruct.width = width; filterStruct.height = height; filterStruct.radius = radius; } void filter() { RS_DEBUG(height); RS_DEBUG(width); RS_DEBUG(radius); initStructs(); computeColorMatrix(); if(radius == 0) { processNoBlur(); return; } computeGaussianWeights(); horizontalBlurLevels(); rsForEach(vBlurScript, rsGetAllocation(InPixel), rsGetAllocation(OutPixel), &filterStruct); int count = 0; rsSendToClient(&count, 1, 4, 0); } void filterBenchmark() { initStructs(); computeGaussianWeights(); rsForEach(hBlurScript, rsGetAllocation(InPixel), rsGetAllocation(OutPixel), &filterStruct); rsForEach(vBlurScript, rsGetAllocation(InPixel), rsGetAllocation(OutPixel), &filterStruct); int count = 0; rsSendToClient(&count, 1, 4, 0); }