--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/hostsupport/hostopenvg/src/src/sfDynamicPixelPipe.cpp Wed Oct 06 17:59:01 2010 +0100
@@ -0,0 +1,1533 @@
+/* Copyright (c) 2010 Nokia Corporation and/or its subsidiary(-ies).
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and /or associated documentation files
+ * (the "Materials "), to deal in the Materials without restriction,
+ * including without limitation the rights to use, copy, modify, merge,
+ * publish, distribute, sublicense, and/or sell copies of the Materials,
+ * and to permit persons to whom the Materials are 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 Materials.
+ *
+ * THE MATERIALS ARE 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 MATERIALS OR
+ * THE USE OR OTHER DEALINGS IN THE MATERIALS.
+ */
+
+// This file contains the generated pixel-pipeline code and provides
+// interface to compile and run them.
+
+#ifndef __RIRASTERIZER_H
+# include "riRasterizer.h"
+#endif
+
+#ifndef __RIPIXELPIPE_H
+# include "riPixelPipe.h"
+#endif
+
+#ifndef __SFDYNAMICPIXELPIPE_H
+# include "sfDynamicPixelPipe.h"
+#endif
+
+#ifndef __RIUTILS_H
+# include "riUtils.h"
+#endif
+
+#ifndef __SFMASK_H
+# include "sfMask.h"
+#endif
+
+#ifndef __RIIMAGE_H
+# include "riImage.h"
+#endif
+
+#if defined(RI_DEBUG)
+# include <stdio.h>
+#endif
+
+namespace OpenVGRI {
+
+RI_INLINE static bool alwaysLoadDst(const PixelPipe::SignatureState& state)
+{
+ if (!state.isRenderToMask)
+ {
+ if (state.hasImage)
+ return true;
+
+ VGBlendMode bm = state.blendMode;
+
+ if (bm == VG_BLEND_SRC_IN ||
+ bm == VG_BLEND_DST_OVER ||
+ bm == VG_BLEND_DST_IN ||
+ bm == VG_BLEND_ADDITIVE ||
+ bm == VG_BLEND_MULTIPLY ||
+ bm == VG_BLEND_SCREEN ||
+ bm == VG_BLEND_DARKEN ||
+ bm == VG_BLEND_LIGHTEN)
+ {
+ return true;
+ } else
+ {
+ return false;
+ }
+ }
+ else
+ {
+ switch (state.maskOperation)
+ {
+ case VG_SET_MASK:
+ return false;
+ default:
+ return true;
+ }
+ }
+}
+
+RI_INLINE static bool canSolidFill(const PixelPipe::SignatureState& state)
+{
+ if (state.isRenderToMask)
+ {
+ if (state.maskOperation == VG_SET_MASK ||
+ state.maskOperation == VG_UNION_MASK)
+ return true;
+ // \note SUBTRACT is also possible.
+ return false;
+ }
+
+ if (state.paintType != VG_PAINT_TYPE_COLOR)
+ return false;
+
+ if (state.hasImage)
+ return false;
+
+ // Some blendmodes can use dst color even if coverage == 1.0
+ if (state.blendMode != VG_BLEND_SRC && state.blendMode != VG_BLEND_SRC_OVER)
+ return false;
+
+ if (state.hasMasking)
+ return false;
+
+ if (state.fillColorTransparent)
+ return false;
+
+ if (state.hasColorTransform)
+ return false; // \todo Trace solid color alpha -> 1.0
+
+ return true;
+}
+
+RI_INLINE static int intReflectRepeat(int n, int bits)
+{
+ const int mask = (1<<bits)-1;
+ return (n ^ (n << (31 - bits) >> 31)) & mask;
+}
+
+RI_INLINE static void applyGradientRepeat(int& sx0, int& sx1, PixelPipe::TilingMode sm)
+{
+ switch (sm)
+ {
+ case PixelPipe::TILING_MODE_PAD:
+ sx0 = RI_INT_CLAMP(sx0, 0, PixelPipe::SAMPLE_MASK);
+ sx1 = RI_INT_CLAMP(sx1, 0, PixelPipe::SAMPLE_MASK);
+ break;
+ case PixelPipe::TILING_MODE_REFLECT:
+ sx0 = intReflectRepeat(sx0, PixelPipe::SAMPLE_BITS);
+ sx1 = intReflectRepeat(sx1, PixelPipe::SAMPLE_BITS);
+ break;
+ default:
+ RI_ASSERT(sm == PixelPipe::TILING_MODE_REPEAT);
+
+ sx0 = sx0 & PixelPipe::SAMPLE_MASK;
+ sx1 = sx1 & PixelPipe::SAMPLE_MASK;
+ break;
+ }
+
+ RI_ASSERT(sx0 >= 0 && sx0 < (1<<Paint::GRADIENT_LUT_BITS));
+ RI_ASSERT(sx1 >= 0 && sx1 < (1<<Paint::GRADIENT_LUT_BITS));
+
+}
+
+RI_INLINE static IntegerColor readLUTColor(const PixelPipe::PPUniforms& uniforms, int i)
+{
+ RI_ASSERT(i >= 0 && i <= Paint::GRADIENT_LUT_MASK);
+ return uniforms.gradientLookup[i];
+}
+
+
+/**
+ * \brief Sample linear gradient using integer-arithmetic.
+ * \note The actual gradient computation is done piecewise within the
+ * pixel-pipeline.
+ */
+RI_INLINE static IntegerColor intLinearGradient(const PixelPipe::SignatureState& state, const PixelPipe::PPUniforms& u, const PixelPipe::PPVariants& v)
+{
+ RIint32 sx0 = v.sx >> (PixelPipe::GRADIENT_BITS - PixelPipe::SAMPLE_BITS);
+ RIint32 sx1 = sx0 + 1;
+
+ applyGradientRepeat(sx0, sx1, state.paintTilingMode);
+
+ IntegerColor ic0 = readLUTColor(u, sx0 >> (PixelPipe::SAMPLE_BITS - Paint::GRADIENT_LUT_BITS));
+
+ if (true)
+ {
+ return ic0;
+ } else
+ {
+ // bilinear interpolation
+ //RIint32 f1 = sx0;
+ readLUTColor(u, sx1 >> (PixelPipe::SAMPLE_BITS - Paint::GRADIENT_LUT_BITS));
+ RI_ASSERT(false);
+ return IntegerColor(0,0,0,0);
+ }
+}
+
+/**
+ * \brief Radial gradient implementation for the integer-pipeline. Will use float at least
+ * for the square-root. Will return integer-color always.
+ */
+RI_INLINE static IntegerColor intRadialGradient(const PixelPipe::SignatureState& state, const PixelPipe::PPUniforms& u, const PixelPipe::PPVariants& v)
+{
+ RGScalar a = (v.rx * u.rfxp) + (v.ry * u.rfyp);
+ RGScalar b = u.rsqrp * (RI_SQR(v.rx) + RI_SQR(v.ry));
+ RGScalar c = RI_SQR((v.rx * u.rfyp) - (v.ry * u.rfxp));
+ RGScalar d = b - c;
+ RI_ASSERT(!RI_ISNAN(d) ? d >= 0.0f : true);
+ RGScalar g = (a + sqrtf(d));
+
+ int sx0 = RI_FLOAT_TO_FX(g, PixelPipe::SAMPLE_BITS);
+ int sx1 = sx0 + 1;
+
+ applyGradientRepeat(sx0, sx1, state.paintTilingMode);
+
+ IntegerColor ic0 = readLUTColor(u, sx0 >> (PixelPipe::SAMPLE_BITS - Paint::GRADIENT_LUT_BITS));
+ RI_ASSERT(ic0.r <= 255);
+ RI_ASSERT(ic0.g <= 255);
+ RI_ASSERT(ic0.b <= 255);
+ RI_ASSERT(ic0.a <= 255);
+
+ if (false)
+ {
+ // Linear interpolation of 2 gradient samples.
+ IntegerColor ic1 = readLUTColor(u, sx1 >> (PixelPipe::SAMPLE_BITS - Paint::GRADIENT_LUT_BITS));
+ //int fx0 = sx0 & PixelPipe::SAMPLE_MASK;
+ //int fx1 = PixelPipe::SAMPLE_MASK - fx0;
+
+ }
+
+ return ic0;
+}
+
+RI_INLINE static bool applyPatternRepeat(int &x, int &y, PixelPipe::TilingMode tilingMode)
+{
+ switch (tilingMode)
+ {
+ case PixelPipe::TILING_MODE_PAD:
+ x = RI_INT_CLAMP(x, 0, PixelPipe::GRADIENT_MASK);
+ y = RI_INT_CLAMP(y, 0, PixelPipe::GRADIENT_MASK);
+ break;
+ case PixelPipe::TILING_MODE_REPEAT:
+ x = x & PixelPipe::GRADIENT_MASK;
+ y = y & PixelPipe::GRADIENT_MASK;
+ break;
+ case PixelPipe::TILING_MODE_REFLECT:
+ x = intReflectRepeat(x, PixelPipe::GRADIENT_BITS);
+ y = intReflectRepeat(y, PixelPipe::GRADIENT_BITS);
+ break;
+ default:
+ RI_ASSERT(tilingMode == PixelPipe::TILING_MODE_FILL);
+ // Do nothing -> Fill is checked on integer coordinates.
+ break;
+ }
+ return false;
+}
+
+/**
+ * \brief Same as applyPatternRepeat, but with pattern-space integer coordinates without
+ * fractional part.
+ * \note Assumes that the coordinate is in range [0,width or height].
+ */
+RI_INLINE static bool applyPatternSampleRepeat(int &x, int &y, int w, int h, PixelPipe::TilingMode tilingMode)
+{
+
+ switch (tilingMode)
+ {
+ case PixelPipe::TILING_MODE_PAD:
+ RI_ASSERT(x >= 0 && x <= w);
+ RI_ASSERT(y >= 0 && y <= h);
+ if (x >= w) x = w-1;
+ if (y >= h) y = h-1;
+ break;
+ case PixelPipe::TILING_MODE_REPEAT:
+ RI_ASSERT(x >= 0 && x <= w);
+ RI_ASSERT(y >= 0 && y <= h);
+ if (x >= w) x = 0;
+ if (y >= h) y = 0;
+ break;
+ case PixelPipe::TILING_MODE_REFLECT:
+ RI_ASSERT(x >= 0 && x <= w);
+ RI_ASSERT(y >= 0 && y <= h);
+ if (x >= w) x = w-1; // w-2?
+ if (y >= h) y = h-1; // h-2?
+ break;
+ default:
+ RI_ASSERT(tilingMode == PixelPipe::TILING_MODE_FILL);
+ if (x < 0 || x >= w) return true;
+ if (y < 0 || y >= h) return true;
+ break;
+ }
+
+ return false;
+}
+
+RI_INLINE IntegerColor readPattern(const void* basePtr, int stride, const Color::Descriptor& desc, int ix, int iy, const IntegerColor* fillColor, bool fill)
+{
+ const void* ptr = Image::calculateAddress(basePtr, desc.bitsPerPixel, ix, iy, stride);
+
+ if (!fill)
+ return IntegerColor(Image::readPackedPixelFromAddress(ptr, desc.bitsPerPixel, ix), desc);
+ else
+ {
+ RI_ASSERT(fillColor);
+ return *fillColor;
+ }
+
+}
+
+/**
+ * \brief Rescale the result of bilinear interpolation.
+ * \todo See if this or individual shifts and rounds are faster on x86
+ */
+RI_INLINE static RIuint32 bilinearDiv(unsigned int c)
+{
+ RIuint32 rcp = 33026;
+
+ RIuint64 m = (RIuint64)c * rcp;
+ RIuint32 d = (RIuint32)(m >> 30);
+ return (d >> 1) + (d & 1);
+}
+
+/**
+ * \brief Read an optionally filtered sample from an image. For multiple samples, apply repeat
+ * for all the generated sampling points. This only implements a simple sampling: nearest
+ * or Linear filtering and is much simpler than the original RI.
+ * \param image Image to sample from
+ * \param sx0 Sample x in .8 fixed point. MUST be within the image except for FILL.
+ * \param sy0 Sample y in .8 fixed point. MUST be within the image except for FILL.
+ * \param samplerType Type of the sampler used.
+ * \param tilingMode Tiling mode for generated sample points, if required.
+ * \param fillColor Color to use for TILING_MODE_FILL
+ * \todo Where should we determine if a NN-sample needs to be unpacked?
+ * -> It is also easy to just read that sample separately.
+ */
+RI_INLINE static IntegerColor intSampleImage(
+ const void* ptr,
+ int stride,
+ int w,
+ int h,
+ const Color::Descriptor& desc,
+ RIint32 sx0,
+ RIint32 sy0,
+ PixelPipe::SamplerType samplerType,
+ PixelPipe::TilingMode tilingMode,
+ const IntegerColor* fillColor)
+{
+ RI_ASSERT(fillColor || (tilingMode != PixelPipe::TILING_MODE_FILL));
+
+ // \todo The following code is between low- and high-level representation of sampling.
+ // It should probably be modified to appear fully as low-level, since we want as many
+ // optimizations as possible.
+
+ const bool bilinear = samplerType == PixelPipe::SAMPLER_TYPE_LINEAR;
+
+ IntegerColor retColor;
+ bool maybeFill = tilingMode == PixelPipe::TILING_MODE_FILL;
+ bool fillSample = false;
+
+ RIint32 ix, iy;
+
+ IntegerColor ic00;
+
+ RIint32 fx = sx0 & 0xff;
+ RIint32 fy = sy0 & 0xff;
+
+ ix = sx0 >> PixelPipe::SAMPLE_BITS;
+ iy = sy0 >> PixelPipe::SAMPLE_BITS;
+
+ if (maybeFill)
+ {
+ if (ix < 0 || ix >= w)
+ fillSample = true;
+ if (iy < 0 || iy >= h)
+ fillSample = true;
+ }
+
+ ic00 = readPattern(ptr, stride, desc, ix, iy, fillColor, fillSample);
+
+ if (!bilinear)
+ {
+ retColor = ic00;
+ retColor.expandColor(desc); // \todo Handling of bilinear?
+ }
+ else
+ {
+ // Bilinear filtering.
+
+ IntegerColor ic01, ic10, ic11;
+ IntegerColor t0, t1;
+
+ int xs = ix + 1;
+ int ys = iy;
+
+ fillSample = applyPatternSampleRepeat(xs, ys, w, h, tilingMode);
+ ic01 = readPattern(ptr, stride, desc, xs, ys, fillColor, fillSample);
+
+ t0 = IntegerColor::linearBlendNS(ic00, ic01, fx);
+
+ xs = ix;
+ ys = iy+1;
+ fillSample = applyPatternSampleRepeat(xs, ys, w, h, tilingMode);
+ ic10 = readPattern(ptr, stride, desc, xs, ys, fillColor, fillSample);
+
+ xs = ix+1;
+ ys = iy+1;
+ fillSample = applyPatternSampleRepeat(xs, ys, w, h, tilingMode);
+ ic11 = readPattern(ptr, stride, desc, xs, ys, fillColor, fillSample);
+
+ t1 = IntegerColor::linearBlendNS(ic10, ic11, fx);
+
+ retColor = IntegerColor::linearBlendNS(t0, t1, fy);
+
+ retColor.r = bilinearDiv(retColor.r);
+ retColor.g = bilinearDiv(retColor.g);
+ retColor.b = bilinearDiv(retColor.b);
+ retColor.a = bilinearDiv(retColor.a);
+
+ return retColor;
+ }
+
+ return retColor;
+}
+
+RI_INLINE static RIint32 gradientToFixedCoords(RIint32 gradCoord, RIint32 dim)
+{
+ return (RIint32)(((RIint64)dim * gradCoord) >> (PixelPipe::GRADIENT_BITS - PixelPipe::SAMPLE_BITS));
+}
+
+
+RI_INLINE static IntegerColor intPattern(const PixelPipe::SignatureState &state, const PixelPipe::PPUniforms& u, const PixelPipe::PPVariants& v)
+{
+ // \todo The following code is between low- and high-level representation of sampling.
+ // It should probably be modified to appear fully as low-level, since we want as many
+ // optimizations as possible.
+
+ // "External" variables
+ const PixelPipe::TilingMode tilingMode = state.paintTilingMode;
+ const IntegerColor fillColor = u.tileFillColor;
+ const int w = u.paint_width;
+ const int h = u.paint_height;
+
+ IntegerColor retColor;
+
+ RIint32 sx0 = v.sx;
+ RIint32 sy0 = v.sy;
+
+ IntegerColor ic00;
+
+ applyPatternRepeat(sx0, sy0, tilingMode);
+ sx0 = gradientToFixedCoords(sx0, w);
+ sy0 = gradientToFixedCoords(sy0, h);
+ //sx0 = (RIint32)(((RIint64)w * sx0) >> (PixelPipe::GRADIENT_BITS - PixelPipe::SAMPLE_BITS));
+ //sy0 = (RIint32)(((RIint64)h * sy0) >> (PixelPipe::GRADIENT_BITS - PixelPipe::SAMPLE_BITS));
+
+ const void* ptr = u.patternPtr;
+ const int stride = u.patternStride;
+ const Color::Descriptor& desc = state.patternDesc;
+
+ return intSampleImage(ptr, stride, w, h, desc, sx0, sy0, state.paintSampler, tilingMode, &fillColor);
+}
+
+RI_INLINE static bool formatPremultipliedAfterSampling(const Color::Descriptor& desc, PixelPipe::SamplerType samplerType, PixelPipe::ImageGradientType gradientType)
+{
+ // Sampled at pixel centers -> no processing of colors -> does not get premultiplied
+ if (gradientType == PixelPipe::GRADIENT_TYPE_INTEGER)
+ return desc.isPremultiplied();
+
+ if (samplerType != PixelPipe::SAMPLER_TYPE_NEAREST)
+ return true;
+
+ return desc.isPremultiplied();
+}
+
+RI_INLINE static bool imagePremultipliedAfterSampling(const PixelPipe::SignatureState& state)
+{
+ RI_ASSERT(state.hasImage);
+
+ return formatPremultipliedAfterSampling(state.imageDesc, state.imageSampler, state.imageGradientType);
+}
+
+RI_INLINE static bool gradientPremultipliedAfterSampling(const PixelPipe::SignatureState& state)
+{
+ if (state.paintSampler != PixelPipe::SAMPLER_TYPE_NEAREST)
+ return true;
+
+ return true;
+
+ // Otherwise, the gradient value is a single sample, and should be in the destination
+ // color-space:
+ //return state.dstDesc.isPremultiplied();
+}
+
+RI_INLINE static bool patternPremultipliedAfterSampling(const PixelPipe::SignatureState& state)
+{
+ RI_ASSERT(state.paintType == VG_PAINT_TYPE_PATTERN);
+
+ return formatPremultipliedAfterSampling(state.patternDesc, state.paintSampler, PixelPipe::GRADIENT_TYPE_FIXED);
+}
+
+/**
+ * \brief Returns true if generated paint will be in RGB, false if luminance.
+ */
+RI_INLINE static bool paintInRGB(const PixelPipe::SignatureState& state)
+{
+ if (state.paintType != VG_PAINT_TYPE_PATTERN)
+ return true;
+
+ return !state.patternDesc.isLuminance();
+}
+
+
+/**
+ * \brief Applies color transform to input color
+ * \param isNonlinear "true" if input is nonlinear. This only affects luminance -> RGB conversion,
+ * other conversions happen in the input color-space.
+ * \note Leaves the color unpremultiplied, in source color-space and converts luminance to RGB
+ * \todo isNonlinear is not needed. It can be deduced from the state information!
+ */
+RI_INLINE static IntegerColor maybeColorTransform(const PixelPipe::SignatureState& state, const IntegerColor& c, const RIint32* colorTransformValues, bool isNonlinear)
+{
+ if (!state.hasColorTransform)
+ return c;
+
+ RI_ASSERT(state.hasImage || state.paintType == VG_PAINT_TYPE_PATTERN);
+
+ IntegerColor r = c;
+
+ if (state.imageMode == VG_DRAW_IMAGE_MULTIPLY)
+ {
+ r.unpremultiply();
+ }
+ else if (state.imageMode == VG_DRAW_IMAGE_STENCIL || state.paintType == VG_PAINT_TYPE_PATTERN)
+ {
+ // -> Check pattern
+ if (patternPremultipliedAfterSampling(state))
+ r.unpremultiply();
+ }
+ else
+ {
+ // -> Check image
+ if (imagePremultipliedAfterSampling(state))
+ r.unpremultiply();
+ }
+
+ // Check if it is necessary to convert to RGB:
+ if (state.imageMode == VG_DRAW_IMAGE_MULTIPLY)
+ {
+ if (state.imageDesc.isLuminance() && !paintInRGB(state))
+ {
+ r.fullLuminanceToRGB(false, isNonlinear, false, isNonlinear);
+ }
+ }
+ else if (state.imageMode == VG_DRAW_IMAGE_STENCIL)
+ {
+ if (state.patternDesc.isLuminance())
+ r.fullLuminanceToRGB(false, isNonlinear, false, isNonlinear);
+ }
+
+ // \todo Use lookup-tables in some cases?
+ r.r = (((RIint32)r.r * colorTransformValues[0]) >> PixelPipe::COLOR_TRANSFORM_BITS) + colorTransformValues[4];
+ r.g = (((RIint32)r.g * colorTransformValues[1]) >> PixelPipe::COLOR_TRANSFORM_BITS) + colorTransformValues[5];
+ r.b = (((RIint32)r.b * colorTransformValues[2]) >> PixelPipe::COLOR_TRANSFORM_BITS) + colorTransformValues[6];
+ r.a = (((RIint32)r.a * colorTransformValues[3]) >> PixelPipe::COLOR_TRANSFORM_BITS) + colorTransformValues[7];
+
+ // Clamp (integerColor?)
+ r.r = (RIuint32)RI_INT_CLAMP((int)r.r, 0, 255);
+ r.g = (RIuint32)RI_INT_CLAMP((int)r.g, 0, 255);
+ r.b = (RIuint32)RI_INT_CLAMP((int)r.b, 0, 255);
+ r.a = (RIuint32)RI_INT_CLAMP((int)r.a, 0, 255);
+
+
+ return r;
+}
+
+/// Some rounding multiplications for blends:
+
+/**
+ * \brief Multiply with rounding.
+ */
+RI_INLINE static RIuint32 rMul2(RIuint32 c0, RIuint32 c1, RIuint32 k0, RIuint32 k1)
+{
+ RIuint32 t = c0 * k0 + c1 * k1;
+ //RIuint32 r = (t + (t>>9)) >> 8;
+ RIuint32 r = (t + (1>>7))>>8;
+ RI_ASSERT(r <= 255);
+ return r;
+}
+
+/**
+ * \brief Returns rounding color-multiplication: c0 + c1 * k
+ */
+RI_INLINE static RIuint32 rMul1(RIuint32 c0, RIuint32 c1, RIuint32 k)
+{
+ RIuint32 t = c1 * k;
+ RIuint32 r = c0 + ((t + (t >> 7)) >> 8);
+ RI_ASSERT(r <= 255);
+ return r;
+}
+
+/**
+ * \brief Fixed-point multiplication
+ */
+RI_INLINE static RIuint32 rMul(RIuint32 c0, RIuint32 f)
+{
+ RIuint32 t = c0 * f;
+ return (t + (1<<7))>>8;
+}
+
+/**
+ * \brief Multiply two colors [0, 255]
+ */
+RI_INLINE static RIuint32 cMul(RIuint32 c0, RIuint32 c1)
+{
+ RIuint32 t = c0 * c1;
+ RIuint32 r = (t + (t >> 9)) >> 8;
+ //RIuint32 t = c0 * c1;
+ //RIuint32 r = (t + (t >> 7))>>8;
+ RI_ASSERT(r <= 255);
+ return r;
+}
+
+// \todo Are signed versions required?
+RI_INLINE static RIuint32 cMin(RIuint32 c0, RIuint32 c1)
+{
+ return c0 <= c1 ? c0 : c1;
+}
+
+RI_INLINE static RIuint32 cMax(RIuint32 c0, RIuint32 c1)
+{
+ return c0 >= c1 ? c0 : c1;
+}
+
+/**
+ * \brief Blends two integer colors. Only considers the alpha-channels within
+ * the colors themselves. There should be a separate function to do
+ * blending with individual channel-alphas.
+ * \note It is also possible that LLVM is able to detect, whether individual alpha-
+ * channels contain a single/multi alpha
+ * \todo Overall, check how much and how fast LLVM is able to optimize out unused
+ * expressions.
+ */
+RI_INLINE static IntegerColor blendIntegerColors(const IntegerColor& s, const IntegerColor& d, VGBlendMode blendMode)
+{
+ IntegerColor r;
+
+ switch(blendMode)
+ {
+ case VG_BLEND_SRC:
+ r = s;
+ break;
+
+ case VG_BLEND_SRC_OVER:
+ {
+ RIuint32 ia = 255 - s.a;
+ r.r = rMul1(s.r, d.r, ia);
+ r.g = rMul1(s.g, d.g, ia);
+ r.b = rMul1(s.b, d.b, ia);
+ r.a = rMul1(s.a, d.a, ia);
+ break;
+ }
+ case VG_BLEND_DST_OVER:
+ {
+ RIuint32 ia = 255 - d.a;
+ r.r = rMul1(d.r, s.r, ia);
+ r.g = rMul1(d.g, s.g, ia);
+ r.b = rMul1(d.b, s.b, ia);
+ r.a = rMul1(d.a, s.a, ia);
+ break;
+ }
+ case VG_BLEND_SRC_IN:
+ {
+ r.r = cMul(s.r, d.a);
+ r.g = cMul(s.g, d.a);
+ r.b = cMul(s.b, d.a);
+ r.a = cMul(s.a, d.a);
+ break;
+ }
+ case VG_BLEND_DST_IN:
+ {
+ r.r = cMul(d.r, s.a);
+ r.g = cMul(d.g, s.a);
+ r.b = cMul(d.b, s.a);
+ r.a = cMul(d.a, s.a);
+ break;
+ }
+ case VG_BLEND_MULTIPLY:
+ {
+ RIuint32 iasrc, iadst;
+ iasrc = 255 - s.a;
+ iadst = 255 - d.a;
+ r.r = rMul2(s.r, d.r, iadst + d.r, iasrc);
+ r.g = rMul2(s.g, d.g, iadst + d.g, iasrc);
+ r.b = rMul2(s.b, d.b, iadst + d.b, iasrc);
+ r.a = rMul1(s.a, d.a, iasrc);
+ break;
+ }
+ case VG_BLEND_SCREEN:
+ {
+ r.r = rMul1(s.r, d.r, 255 - s.r);
+ r.g = rMul1(s.g, d.g, 255 - s.g);
+ r.b = rMul1(s.b, d.b, 255 - s.b);
+ r.a = rMul1(s.a, d.a, 255 - s.a);
+ break;
+ }
+ case VG_BLEND_DARKEN:
+ {
+ RIuint32 iasrc = 255 - s.a;
+ RIuint32 iadst = 255 - d.a;
+ r.r = cMin(rMul1(s.r, d.r, iasrc), rMul1(d.r, s.r, iadst));
+ r.g = cMin(rMul1(s.g, d.g, iasrc), rMul1(d.g, s.g, iadst));
+ r.b = cMin(rMul1(s.b, d.b, iasrc), rMul1(d.b, s.b, iadst));
+ r.a = rMul1(s.a, d.a, iasrc);
+ break;
+ }
+ case VG_BLEND_LIGHTEN:
+ {
+ // \todo Compact darken w/r lighten?
+ RIuint32 iasrc = 255 - s.a;
+ RIuint32 iadst = 255 - d.a;
+ r.r = cMax(rMul1(s.r, d.r, iasrc), rMul1(d.r, s.r, iadst));
+ r.g = cMax(rMul1(s.g, d.g, iasrc), rMul1(d.g, s.g, iadst));
+ r.b = cMax(rMul1(s.b, d.b, iasrc), rMul1(d.b, s.b, iadst));
+ //although the statement below is equivalent to r.a = s.a + d.a * (1.0f - s.a)
+ //in practice there can be a very slight difference because
+ //of the max operation in the blending formula that may cause color to exceed alpha.
+ //Because of this, we compute the result both ways and return the maximum.
+ r.a = cMax(rMul1(s.a, d.a, iasrc), rMul1(d.a, s.a, iadst));
+ break;
+ }
+ default:
+ {
+ RI_ASSERT(blendMode == VG_BLEND_ADDITIVE);
+ r.r = cMin(s.r + d.r, 255);
+ r.g = cMin(s.g + d.g, 255);
+ r.b = cMin(s.b + d.b, 255);
+ r.a = cMin(s.a + d.a, 255);
+ break;
+ }
+ }
+ return r;
+
+}
+
+RI_INLINE static IntegerColor blendIntegerStencil(const IntegerColor& s, const IntegerColor& im, const IntegerColor& d, VGBlendMode blendMode)
+{
+ IntegerColor r;
+
+ switch(blendMode)
+ {
+ case VG_BLEND_SRC:
+ r = s;
+ break;
+
+ case VG_BLEND_SRC_OVER:
+ {
+ r.r = rMul1(s.r, d.r, 255 - im.r);
+ r.g = rMul1(s.g, d.g, 255 - im.g);
+ r.b = rMul1(s.b, d.b, 255 - im.b);
+ r.a = rMul1(s.a, d.a, 255 - s.a);
+ break;
+ }
+ case VG_BLEND_DST_OVER:
+ {
+ r = blendIntegerColors(s, d, blendMode);
+ break;
+ }
+ case VG_BLEND_SRC_IN:
+ {
+ r = blendIntegerColors(s, d, blendMode);
+ break;
+ }
+ case VG_BLEND_DST_IN:
+ {
+ r.r = cMul(d.r, im.r);
+ r.g = cMul(d.g, im.g);
+ r.b = cMul(d.b, im.b);
+ r.a = cMul(d.a, s.a);
+ break;
+ }
+ case VG_BLEND_MULTIPLY:
+ {
+ RIuint32 iadst;
+ iadst = 255 - d.a;
+ r.r = rMul2(s.r, d.r, iadst + d.r, 255 - im.r);
+ r.g = rMul2(s.g, d.g, iadst + d.g, 255 - im.g);
+ r.b = rMul2(s.b, d.b, iadst + d.b, 255 - im.b);
+ r.a = rMul1(s.a, d.a, 255 - s.a);
+ break;
+ }
+ case VG_BLEND_SCREEN:
+ {
+ r = blendIntegerColors(s, d, blendMode);
+ break;
+ }
+ case VG_BLEND_DARKEN:
+ {
+ RIuint32 iadst = 255 - d.a;
+ r.r = cMin(rMul1(s.r, d.r, 255 - im.r), rMul1(d.r, s.r, iadst));
+ r.g = cMin(rMul1(s.g, d.g, 255 - im.g), rMul1(d.g, s.g, iadst));
+ r.b = cMin(rMul1(s.b, d.b, 255 - im.b), rMul1(d.b, s.b, iadst));
+ r.a = rMul1(s.a, d.a, 255 - s.a);
+ break;
+ }
+ case VG_BLEND_LIGHTEN:
+ {
+ // \todo Compact darken w/r lighten?
+ RIuint32 iadst = 255 - d.a;
+ r.r = cMax(rMul1(s.r, d.r, 255 - im.r), rMul1(d.r, s.r, iadst));
+ r.g = cMax(rMul1(s.g, d.g, 255 - im.g), rMul1(d.g, s.g, iadst));
+ r.b = cMax(rMul1(s.b, d.b, 255 - im.b), rMul1(d.b, s.b, iadst));
+ //although the statement below is equivalent to r.a = s.a + d.a * (1.0f - s.a)
+ //in practice there can be a very slight difference because
+ //of the max operation in the blending formula that may cause color to exceed alpha.
+ //Because of this, we compute the result both ways and return the maximum.
+ r.a = cMax(rMul1(s.a, d.a, 255 - s.a), rMul1(d.a, s.a, iadst));
+ break;
+ }
+ default:
+ {
+ RI_ASSERT(blendMode == VG_BLEND_ADDITIVE);
+ return blendIntegerColors(s, d, blendMode);
+ break;
+ }
+ }
+ return r;
+
+}
+
+/**
+ * \brief Perform SRC_OVER and apply coverage in a single operation.
+ * \note It is possible to do optimizations like this for other blending operations,
+ * but they are not as widely used -> optimize if there is a requirement.
+ * \note Prints are included because GDB is confused about the value of r.
+ */
+static RI_INLINE IntegerColor srcOverCoverage(const IntegerColor& s, const IntegerColor& d, RIuint32 cov)
+{
+ IntegerColor r;
+ RIuint32 ac = ((s.a + (s.a>>7)) * cov);
+ ac = (ac + (1<<7))>>8;
+ RIuint32 ia = 256 - ac;
+
+ r.r = rMul2(s.r, d.r, cov, ia);
+ r.g = rMul2(s.g, d.g, cov, ia);
+ r.b = rMul2(s.b, d.b, cov, ia);
+ r.a = rMul2(s.a, d.a, cov, ia);
+ //r.r = (s.r * cov + d.r * ia) >> 8;
+ //r.g = (s.g * cov + d.g * ia) >> 8;
+ //r.b = (s.b * cov + d.b * ia) >> 8;
+ //r.a = (s.a * cov + d.a * ia) >> 8;
+
+#if defined(RI_DEBUG)
+ if (!(r.r <= r.a && r.g <= r.a && r.b <= r.a && r.a <= 255))
+ {
+ printf("r: %d, g: %d, b: %d, a: %d\n",r.r,r.g,r.b,r.a);
+ RI_ASSERT(false);
+ }
+ //RI_ASSERT(r.r <= 255 && r.g <= 255 && r.b <= 255 && r.a <= 255);
+#endif
+
+ return r;
+}
+
+/**
+ * \brief Check if converting between two color formats requires a gamma-conversion.
+ * \todo Move this to descriptor class.
+ */
+static RI_INLINE bool needGammaConvert(const Color::Descriptor& srcDesc, const Color::Descriptor& dstDesc)
+{
+ //if ((!srcDesc.isAlphaOnly()) && (srcDesc.isNonlinear() != dstDesc.isNonlinear()))
+ //return true;
+ if ((srcDesc.isNonlinear() != dstDesc.isNonlinear()))
+ return true;
+
+ return false;
+}
+
+
+RI_INLINE static bool preBlendPremultiplication(const PixelPipe::SignatureState& state)
+{
+ // \todo Simplify the rules (see the corresponding places in the pixelpipe
+ const bool colorTransform = state.hasColorTransform;
+
+ if (PixelPipe::isImageOnly(state))
+ {
+ if (colorTransform)
+ return true;
+
+ // Gamma conversion will leave the result premultiplied
+ if (needGammaConvert(state.imageDesc, state.dstDesc))
+ return true;
+ //if (state.imageDesc.isAlphaOnly())
+ //return false;
+
+ return !imagePremultipliedAfterSampling(state);
+ }
+
+ if (state.hasImage)
+ {
+ if (state.imageMode == VG_DRAW_IMAGE_NORMAL)
+ return !imagePremultipliedAfterSampling(state);
+ // Image color has been combined with the paint color and that requires premultiplication
+ if (state.imageMode == VG_DRAW_IMAGE_MULTIPLY)
+ return false; // Always results in a premultiplied output color
+
+ return false; // ?
+ }
+
+ if (state.paintType == VG_PAINT_TYPE_COLOR)
+ return false;
+
+ if (state.paintType != VG_PAINT_TYPE_PATTERN)
+ return !gradientPremultipliedAfterSampling(state);
+
+ // Must be pattern
+ RI_ASSERT(state.paintType == VG_PAINT_TYPE_PATTERN);
+
+ if (state.hasColorTransform)
+ return true;
+
+ if (needGammaConvert(state.patternDesc, state.dstDesc))
+ return true;
+
+ return !patternPremultipliedAfterSampling(state);
+}
+
+/**
+ * \brief Apply coverage [0 .. 256] on color
+ * \note This is actually "just coverage".
+ */
+RI_INLINE static IntegerColor srcCoverage(const IntegerColor& s, const IntegerColor& d, RIuint32 cov)
+{
+ IntegerColor r;
+ RIuint32 icov = 256-cov;
+ // Make function for multiplication between fixed point values (coverage is
+ // a proper [0 .. 1] value.
+ r.r = (s.r * cov + d.r * icov) >> 8;
+ r.g = (s.g * cov + d.g * icov) >> 8;
+ r.b = (s.b * cov + d.b * icov) >> 8;
+ r.a = (s.a * cov + d.a * icov) >> 8;
+
+ RI_ASSERT(r.r <= 255 && r.g <= 255 && r.b <= 255 && r.a <= 255);
+
+ return r;
+}
+
+/**
+ * \brief Converts color gamma only. Care must be taken concerning luminance color formats.
+ * \return Converted color in "color". This will always be unpremultiplied if gamma conversion
+ * takes place, i.e, tries to minimize the amount of further conversions.
+ */
+RI_INLINE static void maybeGammaConvert(const Color::Descriptor& srcDesc, const Color::Descriptor& dstDesc, IntegerColor& color, bool inputPremultiplied)
+{
+ if (needGammaConvert(srcDesc, dstDesc))
+ {
+ if (inputPremultiplied)
+ color.unpremultiply();
+ //color.unpremultiply(srcDesc.isLuminance());
+
+ if (dstDesc.isNonlinear())
+ color.linearToGamma();
+ else
+ color.gammaToLinear();
+ }
+ // Output always unpremultiplied if gamma conversion takes place
+}
+
+/**
+ * \brief Integer pixel-pipeline.
+ * \note See internal_formats.txt for info on how the data is passed within the pipeline
+ */
+RI_INLINE static void intPixelPipe(const PixelPipe::SignatureState& signatureState, const PixelPipe::PPUniforms &uniforms, PixelPipe::PPVariants& variants)
+{
+ const RIuint32 ppMaxCoverage = Rasterizer::MAX_COVERAGE << (8 - Rasterizer::SAMPLE_BITS);
+ RIuint32 coverage = variants.coverage << (8 - Rasterizer::SAMPLE_BITS);
+ IntegerColor out;
+ IntegerColor imageColor; // imagemode != normal
+ const Color::Descriptor& dstDesc = signatureState.dstDesc;
+ const Color::Descriptor& patternDesc = signatureState.patternDesc;
+ const Color::Descriptor& imageDesc = signatureState.imageDesc;
+
+ if (!PixelPipe::isImageOnly(signatureState))
+ {
+ switch(signatureState.paintType)
+ {
+ case VG_PAINT_TYPE_COLOR:
+ out = uniforms.solidColor;
+ break;
+ case VG_PAINT_TYPE_LINEAR_GRADIENT:
+ out = intLinearGradient(signatureState, uniforms, variants);
+ variants.sx += uniforms.dgdx;
+ // \todo Optimize this so that the lookup is in premultiplied dst format!
+ // How about image-operations?
+ if ((signatureState.imageMode != VG_DRAW_IMAGE_MULTIPLY) && dstDesc.isLuminance())
+ {
+ out.fullRGBToLuminance(true, dstDesc.isNonlinear(), true, dstDesc.isNonlinear());
+ }
+ break;
+ case VG_PAINT_TYPE_RADIAL_GRADIENT:
+ out = intRadialGradient(signatureState, uniforms, variants);
+ variants.rx += uniforms.rdxdx;
+ variants.ry += uniforms.rdydx;
+
+ // \todo Optimize this so that the lookup is in premultiplied dst format!
+ if ((signatureState.imageMode != VG_DRAW_IMAGE_MULTIPLY) && dstDesc.isLuminance())
+ {
+ out.fullRGBToLuminance(true, dstDesc.isNonlinear(), true, dstDesc.isNonlinear());
+ }
+ break;
+ default:
+ RI_ASSERT(signatureState.paintType == VG_PAINT_TYPE_PATTERN);
+ out = intPattern(signatureState, uniforms, variants);
+ // color-space == pattern color-space, not always premultiplied, expanded
+ //
+ // \todo Only increment the proper pixel-counters. This requires detecting the
+ // transform type before generating the pixel-pipeline.
+ // \note Implement fastpaths for at least identity transform with image edges coinciding
+ // with the pixel edges. <- This has been done for images.
+ variants.sx += uniforms.paint_dxdx;
+ variants.sy += uniforms.paint_dydx;
+
+ if (!patternDesc.hasAlpha())
+ out.a = 255;
+
+ if (!signatureState.hasImage)
+ {
+ out = maybeColorTransform(signatureState, out, uniforms.colorTransformValues, patternDesc.isNonlinear());
+ const bool tmpPre = patternPremultipliedAfterSampling(signatureState) && !signatureState.hasColorTransform;
+ const bool outLuminance = !signatureState.hasColorTransform && imageDesc.isLuminance();
+
+ if (outLuminance != dstDesc.isLuminance())
+ {
+ if (outLuminance)
+ out.fullLuminanceToRGB(tmpPre, patternDesc.isNonlinear(), tmpPre, patternDesc.isNonlinear());
+ else
+ out.fullRGBToLuminance(tmpPre, patternDesc.isNonlinear(), tmpPre, patternDesc.isNonlinear());
+ }
+ maybeGammaConvert(patternDesc, dstDesc, out, tmpPre);
+ }
+
+ break;
+ }
+ }
+
+ if (signatureState.hasImage)
+ {
+ switch (signatureState.imageGradientType)
+ {
+ case PixelPipe::GRADIENT_TYPE_INTEGER:
+ {
+ void* addr = Image::calculateAddress(uniforms.imagePtr, imageDesc.bitsPerPixel, variants.iImageX, variants.iImageY, uniforms.imageStride);
+ RIuint32 packedImageColor = Image::readPackedPixelFromAddress(addr, imageDesc.bitsPerPixel, variants.iImageX);
+ imageColor.fromPackedColor(packedImageColor, imageDesc);
+ imageColor.expandColor(imageDesc);
+ // color-space == image color-space, not always premultiplied, expanded
+
+ // Only integer image-gradient can have unsafe image data as an input at the moment.
+ if (signatureState.unsafeImageInput)
+ {
+ if (imageDesc.hasAlpha() && imageDesc.isPremultiplied())
+ imageColor.clampToAlpha();
+ }
+
+ variants.iImageX += uniforms.image_idxdx;
+ variants.iImageY += uniforms.image_idydx;
+ break;
+ }
+ case PixelPipe::GRADIENT_TYPE_FIXED:
+ {
+ RI_ASSERT(!signatureState.unsafeImageInput);
+
+ RIint32 sx, sy;
+ sx = variants.iImageX;
+ sy = variants.iImageY;
+ applyPatternRepeat(sx, sy, PixelPipe::TILING_MODE_PAD);
+ sx = gradientToFixedCoords(sx, uniforms.image_iWidth);
+ sy = gradientToFixedCoords(sy, uniforms.image_iHeight);
+ imageColor = intSampleImage(
+ uniforms.imagePtr,
+ uniforms.imageStride,
+ uniforms.image_iWidth,
+ uniforms.image_iHeight,
+ imageDesc,
+ sx, sy, signatureState.imageSampler, PixelPipe::TILING_MODE_PAD, NULL);
+
+ variants.iImageX += uniforms.image_idxdx;
+ variants.iImageY += uniforms.image_idydx;
+ break;
+ }
+ default:
+ {
+ RI_ASSERT(signatureState.imageGradientType == PixelPipe::GRADIENT_TYPE_FLOAT);
+ RI_ASSERT(!signatureState.unsafeImageInput);
+
+ RIfloat fx, fy, fw, rw;
+ fx = variants.fImageX;
+ fy = variants.fImageY;
+ fw = variants.fImageW;
+ rw = 1.0f / fw;
+ RIint32 sx0, sy0;
+ fx = RI_CLAMP(fx * rw, 0.0f, uniforms.image_fWidth - 1.0f); // \todo fImageMaxX
+ fy = RI_CLAMP(fy * rw, 0.0f, uniforms.image_fHeight - 1.0f);
+ sx0 = RI_ROUND_TO_INT(fx * (1<<PixelPipe::SAMPLE_BITS));
+ sy0 = RI_ROUND_TO_INT(fy * (1<<PixelPipe::SAMPLE_BITS));
+
+ imageColor = intSampleImage(
+ uniforms.imagePtr,
+ uniforms.imageStride,
+ uniforms.image_iWidth,
+ uniforms.image_iHeight,
+ imageDesc,
+ sx0, sy0, signatureState.imageSampler, PixelPipe::TILING_MODE_PAD, NULL);
+
+ variants.fImageX += uniforms.image_fdxdx;
+ variants.fImageY += uniforms.image_fdydx;
+ variants.fImageW += uniforms.image_fdwdx;
+ break;
+ }
+ }
+
+ if (!imageDesc.hasAlpha())
+ imageColor.a = 255;
+
+ if (PixelPipe::isImageOnly(signatureState))
+ {
+ RI_ASSERT(signatureState.imageMode == VG_DRAW_IMAGE_NORMAL);
+ out = maybeColorTransform(signatureState, imageColor, uniforms.colorTransformValues, imageDesc.isNonlinear());
+
+ const bool tmpPre = imagePremultipliedAfterSampling(signatureState) && !signatureState.hasColorTransform;
+ const bool outLuminance = !signatureState.hasColorTransform && imageDesc.isLuminance();
+
+ // Color-format conversion to dst before blending.
+ if (outLuminance != dstDesc.isLuminance())
+ {
+ if (outLuminance)
+ out.fullLuminanceToRGB(tmpPre, imageDesc.isNonlinear(), tmpPre, imageDesc.isNonlinear());
+ else
+ out.fullRGBToLuminance(tmpPre, imageDesc.isNonlinear(), tmpPre, imageDesc.isNonlinear());
+ }
+ maybeGammaConvert(imageDesc, dstDesc, out, tmpPre);
+
+ //if (!signatureState.hasColorTransform)
+ //out.premultiply();
+ }
+ else
+ {
+ RI_ASSERT(signatureState.imageMode != VG_DRAW_IMAGE_NORMAL);
+
+ if (!imagePremultipliedAfterSampling(signatureState))
+ imageColor.premultiply();
+
+ if (signatureState.imageMode == VG_DRAW_IMAGE_MULTIPLY)
+ {
+ if (signatureState.paintType == VG_PAINT_TYPE_PATTERN &&
+ !patternPremultipliedAfterSampling(signatureState))
+ {
+ out.premultiply();
+ }
+
+ out.r = cMul(out.r, imageColor.r);
+ out.g = cMul(out.g, imageColor.g);
+ out.b = cMul(out.b, imageColor.b);
+ out.a = cMul(out.a, imageColor.a);
+
+ out = maybeColorTransform(signatureState, out, uniforms.colorTransformValues, imageDesc.isNonlinear());
+ //const bool outLuminance = !signatureState.hasColorTransform && imageDesc.isLuminance();
+ // Color transform will always result in RGB, regardless of input.
+ const bool outLuminance = (imageDesc.isLuminance() && !paintInRGB(signatureState)) && !signatureState.hasColorTransform;
+ if (!outLuminance && dstDesc.isLuminance())
+ {
+ // Convert to destination (luminance)
+ out.fullRGBToLuminance(!signatureState.hasColorTransform, imageDesc.isNonlinear(), true, dstDesc.isNonlinear());
+ }
+ else if (imageDesc.isNonlinear() != dstDesc.isNonlinear())
+ {
+ // Non-luminance gamma
+ if (!signatureState.hasColorTransform)
+ out.unpremultiply();
+
+ if (dstDesc.isNonlinear())
+ out.linearToGamma();
+ else
+ out.gammaToLinear();
+
+ out.premultiply();
+ }
+ else if (signatureState.hasColorTransform)
+ out.premultiply();
+
+ // Output dst and premultiplied.
+ }
+ else
+ {
+ RI_ASSERT(signatureState.imageMode == VG_DRAW_IMAGE_STENCIL);
+ IntegerColor alphas, pr;
+
+ if (signatureState.paintType == VG_PAINT_TYPE_PATTERN)
+ {
+ out = maybeColorTransform(signatureState, out, uniforms.colorTransformValues, patternDesc.isNonlinear());
+ const bool isLuminance = patternDesc.isLuminance() && !signatureState.hasColorTransform;
+ // If using pattern, convert to destination color-space
+ // \todo If not, handle this when the lookups are generated.
+ if (isLuminance != dstDesc.isLuminance())
+ {
+ out.fullRGBToLuminance(patternPremultipliedAfterSampling(signatureState) && !signatureState.hasColorTransform, patternDesc.isNonlinear(), true, dstDesc.isNonlinear());
+ }
+ else if (patternDesc.isNonlinear() != dstDesc.isNonlinear())
+ {
+ if (patternPremultipliedAfterSampling(signatureState) && !signatureState.hasColorTransform)
+ out.unpremultiply();
+
+ if (dstDesc.isNonlinear())
+ out.linearToGamma();
+ else
+ out.gammaToLinear();
+
+ out.premultiply();
+ } else if (signatureState.hasColorTransform || !patternPremultipliedAfterSampling(signatureState))
+ out.premultiply();
+ }
+
+ if (dstDesc.isLuminance() && !imageDesc.isLuminance())
+ {
+ // Convert image to luminance
+ imageColor.rgbToLuminance();
+ imageColor.r = imageColor.b = imageColor.b = RI_INT_MIN(imageColor.r, imageColor.a);
+ }
+
+#if defined(RI_DEBUG) && 0
+ printf("stencil r: %d, g: %d, b: %d, a: %d\n",imageColor.r,imageColor.g,imageColor.b,imageColor.a);
+ printf("input r: %d, g: %d, b: %d, a: %d\n",out.r,out.g,out.b,out.a);
+#endif
+ if (signatureState.paintType == VG_PAINT_TYPE_COLOR)
+ {
+ // Better precision for solid color input.
+ // Compute alpha channels
+ alphas.r = rMul(out.a, imageColor.r);
+ alphas.g = rMul(out.a, imageColor.g);
+ alphas.b = rMul(out.a, imageColor.b);
+ // Premultiply
+ pr.r = rMul(out.r, imageColor.r);
+ pr.g = rMul(out.g, imageColor.g);
+ pr.b = rMul(out.b, imageColor.b);
+ pr.a = rMul(out.a, imageColor.a);
+ }
+ else
+ {
+ // Compute alpha channels
+ alphas.r = cMul(out.a, imageColor.r);
+ alphas.g = cMul(out.a, imageColor.g);
+ alphas.b = cMul(out.a, imageColor.b);
+ // Premultiply
+ pr.r = cMul(out.r, imageColor.r);
+ pr.g = cMul(out.g, imageColor.g);
+ pr.b = cMul(out.b, imageColor.b);
+ pr.a = cMul(out.a, imageColor.a);
+ }
+#if defined(RI_DEBUG) && 0
+ printf("alphas r: %d, g: %d, b: %d, a: %d\n",alphas.r,alphas.g,alphas.b,alphas.a);
+ printf("pr r: %d, g: %d, b: %d, a: %d\n",pr.r,pr.g,pr.b,pr.a);
+#endif
+ out = pr;
+ imageColor = alphas;
+ }
+ }
+ }
+
+ if (signatureState.hasMasking)
+ {
+ // \todo Read and process only the proper component of the mask pixel.
+ const int maskBpp = signatureState.maskDesc.bitsPerPixel;
+
+ RIuint32 packedMaskColor = Image::readPackedPixelFromAddress(variants.maskPtr, maskBpp, variants.dstX);
+ IntegerColor maskColor;
+ maskColor.fromPackedMask(packedMaskColor, signatureState.maskDesc);
+ maskColor.expandMask(signatureState.maskDesc);
+
+ RIuint32 maskCoverage = maskColor.a + (maskColor.a >> 7);
+ coverage = (coverage * maskCoverage) >> 8;
+
+ variants.maskPtr = (void*)Image::incrementPointer(variants.maskPtr, maskBpp, variants.dstX);
+ }
+
+#if defined(RI_DEBUG)
+ IntegerColor preblend = out;
+#endif
+ // \todo Coverage check for pixelpipes != solid color with solid output colors?
+
+ IntegerColor d(0,0,0,0);
+
+ // All operations that depend on DST are done next. Keep it organized like that.
+ if ((coverage < ppMaxCoverage) || (out.a < 255) || alwaysLoadDst(signatureState))
+ {
+ d = IntegerColor(Image::readPackedPixelFromAddress(
+ variants.dst, dstDesc.bitsPerPixel, variants.dstX), dstDesc);
+ d.expandColor(dstDesc);
+
+ if (!dstDesc.isPremultiplied())
+ {
+ d.premultiply();
+ }
+
+ // Premultiply output
+#if 0
+ if (!PixelPipe::isImageOnly(signatureState))
+ {
+ if (signatureState.paintType == VG_PAINT_TYPE_PATTERN && !patternPremultipliedAfterSampling(signatureState))
+ out.premultiply();
+ else if (signatureState.hasImage && !imagePremultipliedAfterSampling(signatureState))
+ out.premultiply();
+ }
+#endif
+
+ if (!signatureState.isRenderToMask)
+ {
+ VGBlendMode bm = signatureState.blendMode;
+
+ // Currently SRC requires premultiplication even when only applying coverage.
+ //if (bm != VG_BLEND_SRC)
+ {
+ // If the src color has not been premultiplied before, now's the time.
+ // \todo Fast path for src alpha == 255 and SRC_OVER? Others?
+ if (preBlendPremultiplication(signatureState))
+ out.premultiply();
+ }
+
+ if (signatureState.hasImage && signatureState.imageMode == VG_DRAW_IMAGE_STENCIL)
+ {
+ out = blendIntegerStencil(out, imageColor, d, bm);
+ }
+ else
+ {
+ switch(bm)
+ {
+ case VG_BLEND_SRC_OVER:
+ out = srcOverCoverage(out, d, coverage);
+ break;
+ case VG_BLEND_SRC:
+ out = srcCoverage(out, d, coverage);
+ break;
+ default:
+ out = blendIntegerColors(out, d, bm);
+ out = srcCoverage(out, d, coverage);
+ break;
+ }
+ }
+
+#if defined(RI_DEBUG)
+ if (dstDesc.isPremultiplied())
+ {
+ RI_ASSERT(out.r <= out.a);
+ RI_ASSERT(out.g <= out.a);
+ RI_ASSERT(out.b <= out.a);
+ }
+#endif
+
+ }
+ else
+ {
+ // Mask operation
+ out = intMaskOperation(coverage, d, signatureState.maskOperation);
+ }
+
+ // out is always premultiplied at this point. Must be in destination color-space
+ if (!dstDesc.isPremultiplied())
+ {
+ // Unpremultiply if output is not premultiplied
+ out.unpremultiply();
+ }
+ }
+ else
+ {
+ // Unpremultiply, ...
+ if (!dstDesc.isPremultiplied())
+ out.unpremultiply();
+ }
+
+ // VG_SET_MASK does not require dst load:
+ if (signatureState.isRenderToMask && signatureState.maskOperation == VG_SET_MASK)
+ out = intMaskOperation(coverage, d, VG_SET_MASK);
+
+ out.truncateColor(dstDesc);
+ Image::writePackedPixelToAddress(
+ variants.dst, dstDesc.bitsPerPixel, variants.dstX, out.getPackedColor(dstDesc));
+
+ // \todo X for bpp < 8
+ variants.dst = (void*)Image::incrementPointer(variants.dst, dstDesc.bitsPerPixel, variants.dstX);
+ //variants.dst = colorBuffer->advancePointer(variants.dst);
+ variants.dstX++;
+}
+
+RI_INLINE static void fillSolidSpan(const PixelPipe::SignatureState& state, const PixelPipe::PPUniforms& uniforms, int startX, int y, int nPixels, RIuint32 packedColor)
+{
+ Image::fillPackedPixels((void*)uniforms.dstPtr, state.dstDesc.bitsPerPixel, startX, y, uniforms.dstStride, nPixels, packedColor);
+}
+
+/**
+ * \brief This will calculate all the pixel-pipeline variants that need to be updated per-pixel.
+ * \note There may be a need for a different, faster function for image rendering, where
+ * there are faster methods of updating the variants.
+ */
+RI_INLINE static void prepareSpanVariants(const PixelPipe::SignatureState& state, const PixelPipe::PPUniforms& uniforms, const Span& span, PixelPipe::PPVariants& variants)
+{
+ //variants.dst = uniforms.dst->calculateAddress(span.x0, span.y);
+ variants.dst = Image::calculateAddress(uniforms.dstPtr, state.dstDesc.bitsPerPixel, span.x0, span.y, uniforms.dstStride);
+ variants.dstX = span.x0;
+ variants.coverage = span.coverage;
+
+ if (state.paintType != VG_PAINT_TYPE_COLOR)
+ {
+ if (state.paintType == VG_PAINT_TYPE_LINEAR_GRADIENT)
+ {
+ // \todo Adjust pixel-center.
+ int x = uniforms.dgdx * span.x0 + uniforms.dgdy * span.y + uniforms.lgc;
+ variants.sx = x;
+ }
+ else if (state.paintType == VG_PAINT_TYPE_RADIAL_GRADIENT)
+ {
+ RGScalar x = uniforms.rdxdx * (RGScalar)span.x0 + uniforms.rdxdy * (RGScalar)span.y;
+ RGScalar y = uniforms.rdydy * (RGScalar)span.y + uniforms.rdydx * (RGScalar)span.x0;
+
+ variants.rx = x + uniforms.rx0;
+ variants.ry = y + uniforms.ry0;
+ }
+ else
+ {
+ RI_ASSERT(state.paintType == VG_PAINT_TYPE_PATTERN);
+ variants.sx = uniforms.paint_dxdx * span.x0 + uniforms.paint_dxdy * span.y + uniforms.paint_x0;
+ variants.sy = uniforms.paint_dydy * span.y + uniforms.paint_dydx * span.x0 + uniforms.paint_y0;
+ }
+ }
+
+ if (state.hasMasking)
+ {
+ variants.maskPtr = Image::calculateAddress(uniforms.maskPtr, state.maskDesc.bitsPerPixel, span.x0, span.y, uniforms.maskStride);
+ }
+
+ if (state.hasImage)
+ {
+ switch (state.imageGradientType)
+ {
+ case PixelPipe::GRADIENT_TYPE_INTEGER:
+ case PixelPipe::GRADIENT_TYPE_FIXED:
+ variants.iImageX = uniforms.image_ix0 + span.x0 * uniforms.image_idxdx + span.y * uniforms.image_idxdy;
+ variants.iImageY = uniforms.image_iy0 + span.y * uniforms.image_idydy + span.x0 * uniforms.image_idydx;
+ break;
+ default:
+ RI_ASSERT(state.imageGradientType == PixelPipe::GRADIENT_TYPE_FLOAT);
+ variants.fImageX = uniforms.image_fx0 + span.x0 * uniforms.image_fdxdx + span.y * uniforms.image_fdxdy;
+ variants.fImageY = uniforms.image_fy0 + span.y * uniforms.image_fdydy + span.x0 * uniforms.image_fdydx;
+ variants.fImageW = uniforms.image_fw0 + span.x0 * uniforms.image_fdwdx + span.y * uniforms.image_fdwdy;
+ break;
+ }
+ }
+}
+
+void executePixelPipeline(const PixelPipe::SignatureState& state, const PixelPipe::PPUniforms& uniforms, PixelPipe::PPVariants& variants, const Span* spans, int nSpans)
+{
+ RI_ASSERT(nSpans > 0);
+ for (int i = 0; i < nSpans; i++)
+ {
+ const Span& s = spans[i];
+
+ if (s.coverage != Rasterizer::MAX_COVERAGE || !canSolidFill(state))
+ {
+ int n = s.len;
+ RI_ASSERT(n);
+ prepareSpanVariants(state, uniforms, s, variants);
+
+ do {
+ intPixelPipe(state, uniforms, variants);
+ } while (--n);
+ } else
+ {
+ fillSolidSpan(state, uniforms, s.x0, s.y, s.len, uniforms.packedSolidColor);
+ }
+ }
+
+}
+
+void calculatePPHash(PixelPipeHash& hash, const PixelPipe::SignatureState& derivedState)
+{
+ const RIuint32 blendModeBits = 4;
+ const RIuint32 imageModeBits = 2;
+ const RIuint32 paintTypeBits = 2;
+ const RIuint32 tilingModeBits = 2;
+ const RIuint32 samplerBits = 1;
+ const RIuint32 imageGradientTypeBits = 2;
+ const RIuint32 boolBits = 1;
+ const RIuint32 descBits = 10;
+ const RIuint32 maskOperationBits = 3;
+
+ RIuint32 blendMode = ((RIuint32)derivedState.blendMode) - ((RIuint32)VG_BLEND_SRC);
+ RIuint32 imageMode = ((RIuint32)derivedState.imageMode) - ((RIuint32)VG_DRAW_IMAGE_NORMAL);
+ RIuint32 paintType = ((RIuint32)derivedState.paintType) - ((RIuint32)VG_PAINT_TYPE_COLOR);
+ RIuint32 maskOperation = ((RIuint32)derivedState.maskOperation) - ((RIuint32)VG_CLEAR_MASK);
+ RIuint32 paintTilingMode = ((RIuint32)derivedState.paintTilingMode);
+ RIuint32 paintSampler = ((RIuint32)derivedState.paintSampler);
+ RIuint32 imageSampler = ((RIuint32)derivedState.imageSampler);
+
+ RIuint32 imageGradientType = ((RIuint32)derivedState.imageGradientType);
+
+ RIuint32 dstFormat = (RIuint32)(derivedState.dstDesc.toIndex());
+ RIuint32 maskFormat = (RIuint32)(derivedState.maskDesc.toIndex());
+ RIuint32 imageFormat = (RIuint32)(derivedState.imageDesc.toIndex());
+ RIuint32 patternFormat = (RIuint32)(derivedState.patternDesc.toIndex());
+
+ RIuint32 hasMasking = derivedState.hasMasking ? 1 : 0;
+ RIuint32 hasImage = derivedState.hasImage ? 1 : 0;
+ RIuint32 hasColorTransform = derivedState.hasColorTransform ? 1 : 0;
+ RIuint32 isMaskOperation = derivedState.isRenderToMask ? 1 : 0;
+ RIuint32 fillColorTransparent = derivedState.fillColorTransparent ? 1 : 0;
+ RIuint32 unsafeImageInput = derivedState.unsafeImageInput ? 1 : 0;
+
+ // Modify hashes according to relevant state:
+ int b = 0;
+ b = riInsertBits32(hash.value, sizeof(hash.value), blendMode, blendModeBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), imageMode, imageModeBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), paintType, paintTypeBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), maskOperation, maskOperationBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), paintTilingMode, tilingModeBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), paintSampler, samplerBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), imageSampler, samplerBits, b);
+
+ b = riInsertBits32(hash.value, sizeof(hash.value), imageGradientType, imageGradientTypeBits, b);
+
+ b = riInsertBits32(hash.value, sizeof(hash.value), dstFormat, descBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), maskFormat, descBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), imageFormat, descBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), patternFormat, descBits, b);
+
+ b = riInsertBits32(hash.value, sizeof(hash.value), hasMasking, boolBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), hasImage, boolBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), hasColorTransform, boolBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), isMaskOperation, boolBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), fillColorTransparent, boolBits, b);
+ b = riInsertBits32(hash.value, sizeof(hash.value), unsafeImageInput, boolBits, b);
+}
+
+}
+