SaI2xScaler.cc

Go to the documentation of this file.

// 2xSaI is Copyright (c) 1999-2001 by Derek Liauw Kie Fa.
//   http://elektron.its.tudelft.nl/~dalikifa/
// 2xSaI is free under GPL.
//
// Modified for use in openMSX by Maarten ter Huurne.

#include "SaI2xScaler.hh"
#include "FrameSource.hh"
#include "ScalerOutput.hh"
#include "vla.hh"
#include "build-info.hh"
#include <cassert>
#include <cstdint>

namespace openmsx {

template <class Pixel>
SaI2xScaler<Pixel>::SaI2xScaler(const PixelOperations<Pixel>& pixelOps_)
        : Scaler2<Pixel>(pixelOps_)
        , pixelOps(pixelOps_)
{
}

template <class Pixel>
void SaI2xScaler<Pixel>::scaleBlank1to2(
                FrameSource& src, unsigned srcStartY, unsigned srcEndY,
                ScalerOutput<Pixel>& dst, unsigned dstStartY, unsigned dstEndY)
{
        unsigned dstHeight = dst.getHeight();
        unsigned stopDstY = (dstEndY == dstHeight)
                          ? dstEndY : dstEndY - 2;
        unsigned srcY = srcStartY, dstY = dstStartY;
        for (/* */; dstY < stopDstY; srcY += 1, dstY += 2) {
                auto color = src.getLineColor<Pixel>(srcY);
                dst.fillLine(dstY + 0, color);
                dst.fillLine(dstY + 1, color);
        }
        if (dstY != dstHeight) {
                unsigned nextLineWidth = src.getLineWidth(srcY + 1);
                assert(src.getLineWidth(srcY) == 1);
                assert(nextLineWidth != 1);
                this->dispatchScale(src, srcY, srcEndY, nextLineWidth,
                                    dst, dstY, dstEndY);
        }
}

template <class Pixel>
inline Pixel SaI2xScaler<Pixel>::blend(Pixel p1, Pixel p2)
{
        return pixelOps.template blend<1, 1>(p1, p2);
}

template <class Pixel>
void SaI2xScaler<Pixel>::scaleLine1on2(
        const Pixel* __restrict srcLine0, const Pixel* __restrict srcLine1,
        const Pixel* __restrict srcLine2, const Pixel* __restrict srcLine3,
        Pixel* __restrict dstUpper, Pixel* __restrict dstLower,
        unsigned srcWidth)
{
        // TODO: Scale border pixels as well.
        for (unsigned x = 0; x < srcWidth; x++) {
                // Map of the pixels:
                //   I|E F|J
                //   G|A B|K
                //   H|C D|L
                //   M|N O|P

                unsigned xl  = x ? x - 1 : 0;
                unsigned xr  = std::min(x + 1, srcWidth - 1);
                unsigned xrr = std::min(x + 2, srcWidth - 1);

                // TODO: Possible performance improvements:
                // - Play with order of fetching (effect on data cache).
                // - Try not fetching at all (using srcLineN[x] in algorithm).
                // - Try rotating the fetched colors (either in vars or in array).
                Pixel colorI = srcLine0[xl];
                Pixel colorE = srcLine0[x];
                Pixel colorF = srcLine0[xr];
                Pixel colorJ = srcLine0[xrr];

                Pixel colorG = srcLine1[xl];
                Pixel colorA = srcLine1[x];
                Pixel colorB = srcLine1[xr];
                Pixel colorK = srcLine1[xrr];

                Pixel colorH = srcLine2[xl];
                Pixel colorC = srcLine2[x];
                Pixel colorD = srcLine2[xr];
                Pixel colorL = srcLine2[xrr];

                Pixel colorM = srcLine3[xl];
                Pixel colorN = srcLine3[x];
                Pixel colorO = srcLine3[xr];
                //Pixel colorP = srcLine3[xrr];

                Pixel product, product1, product2;
                if (colorA == colorD && colorB != colorC) {
                        product  = ((colorA == colorE && colorB == colorL) ||
                                    (colorA == colorC && colorA == colorF &&
                                     colorB != colorE && colorB == colorJ))
                                 ? colorA
                                 : blend(colorA, colorB);
                        product1 = ((colorA == colorG && colorC == colorO) ||
                                    (colorA == colorB && colorA == colorH &&
                                     colorG != colorC && colorC == colorM))
                                 ? colorA
                                 : blend(colorA, colorC);
                        product2 = colorA;
                } else if (colorB == colorC && colorA != colorD) {
                        product  = ((colorB == colorF && colorA == colorH) ||
                                    (colorB == colorE && colorB == colorD &&
                                     colorA != colorF && colorA == colorI))
                                 ? colorB
                                 : blend(colorA, colorB);
                        product1 = ((colorC == colorH && colorA == colorF) ||
                                    (colorC == colorG && colorC == colorD &&
                                     colorA != colorH && colorA == colorI))
                                 ? colorC
                                 : blend(colorA, colorC);
                        product2 = colorB;
                } else if (colorA == colorD && colorB == colorC) {
                        if (colorA == colorB) {
                                product = product1 = product2 = colorA;
                        } else {
                                int r = 0;
                                if (colorE == colorG) {
                                        if      (colorA == colorE) r--;
                                        else if (colorB == colorE) r++;
                                }
                                if (colorF == colorK) {
                                        if      (colorA == colorF) r--;
                                        else if (colorB == colorF) r++;
                                }
                                if (colorH == colorN) {
                                        if      (colorA == colorH) r--;
                                        else if (colorB == colorH) r++;
                                }
                                if (colorL == colorO) {
                                        if      (colorA == colorL) r--;
                                        else if (colorB == colorL) r++;
                                }
                                product = product1 = blend(colorA, colorB);
                                product2 = r > 0 ? colorA : (r < 0 ? colorB : product);
                        }
                } else {
                        product  = (colorA == colorC && colorA == colorF &&
                                    colorB != colorE && colorB == colorJ)
                                 ? colorA
                                 : ((colorB == colorE && colorB == colorD &&
                                     colorA != colorF && colorA == colorI)
                                   ? colorB
                                   : blend(colorA, colorB));
                        product1 = (colorA == colorB && colorA == colorH &&
                                    colorG != colorC && colorC == colorM)
                                 ? colorA
                                 : ((colorC == colorG && colorC == colorD &&
                                     colorA != colorH && colorA == colorI)
                                   ? colorC
                                   : blend(colorA, colorC));
                        product2 = blend( // TODO: Quad-blend may be better?
                                blend(colorA, colorB),
                                blend(colorC, colorD));
                }

                dstUpper[x * 2 + 0] = colorA;
                dstUpper[x * 2 + 1] = product;
                dstLower[x * 2 + 0] = product1;
                dstLower[x * 2 + 1] = product2;
        }
}

template <class Pixel>
void SaI2xScaler<Pixel>::scaleLine1on1(
        const Pixel* __restrict srcLine0, const Pixel* __restrict srcLine1,
        const Pixel* __restrict srcLine2, const Pixel* __restrict srcLine3,
        Pixel* __restrict dstUpper, Pixel* __restrict dstLower,
        unsigned srcWidth)
{
        // Apply 2xSaI and keep the bottom-left pixel.
        // It's not great, but at least it looks better than doubling the pixel
        // like SimpleScaler does.
        dstUpper[0] = srcLine1[0];
        dstLower[0] = blend(srcLine1[0], srcLine2[0]);
        for (unsigned x = 1; x < srcWidth - 1; x++) {
                // Map of the pixels:
                //   I E F
                //   G A B
                //   H C D
                //   M N O

                Pixel colorI = srcLine0[x - 1];
                //Pixel colorE = srcLine0[x];
                Pixel colorF = srcLine0[x + 1];

                Pixel colorG = srcLine1[x - 1];
                Pixel colorA = srcLine1[x];
                Pixel colorB = srcLine1[x + 1];

                Pixel colorH = srcLine2[x - 1];
                Pixel colorC = srcLine2[x];
                Pixel colorD = srcLine2[x + 1];

                Pixel colorM = srcLine3[x - 1];
                //Pixel colorN = srcLine3[x];
                Pixel colorO = srcLine3[x + 1];

                Pixel product1;
                if (colorA == colorD && colorB != colorC) {
                        product1 = ((colorA == colorG && colorC == colorO) ||
                                    (colorA == colorB && colorA == colorH &&
                                     colorG != colorC && colorC == colorM))
                                 ? colorA
                                 : blend(colorA, colorC);
                } else if (colorB == colorC && colorA != colorD) {
                        product1 = ((colorC == colorH && colorA == colorF) ||
                                    (colorC == colorG && colorC == colorD &&
                                     colorA != colorH && colorA == colorI))
                                 ? colorC
                                 : blend(colorA, colorC);
                } else if (colorA == colorD && colorB == colorC) {
                        product1 = (colorA == colorC)
                                 ? colorA
                                 : blend(colorA, colorC);
                } else {
                        product1 = (colorA == colorB && colorA == colorH &&
                                    colorG != colorC && colorC == colorM)
                                 ? colorA
                                 : ((colorC == colorG && colorC == colorD &&
                                     colorA != colorH && colorA == colorI)
                                   ? colorC
                                   : blend(colorA, colorC));
                }

                dstUpper[x] = colorA;
                dstLower[x] = product1;
        }
        dstUpper[srcWidth - 1] = srcLine1[srcWidth - 1];
        dstLower[srcWidth - 1] =
                blend(srcLine1[srcWidth - 1], srcLine2[srcWidth - 1]);
}

template <class Pixel>
void SaI2xScaler<Pixel>::scale1x1to2x2(FrameSource& src,
        unsigned srcStartY, unsigned /*srcEndY*/, unsigned srcWidth,
        ScalerOutput<Pixel>& dst, unsigned dstStartY, unsigned dstEndY)
{
        assert(dst.getWidth() == srcWidth * 2);

        VLA_SSE_ALIGNED(Pixel, buf0_, srcWidth); auto* buf0 = buf0_;
        VLA_SSE_ALIGNED(Pixel, buf1_, srcWidth); auto* buf1 = buf1_;
        VLA_SSE_ALIGNED(Pixel, buf2_, srcWidth); auto* buf2 = buf2_;
        VLA_SSE_ALIGNED(Pixel, buf3_, srcWidth); auto* buf3 = buf3_;

        int srcY = srcStartY;
        auto* srcLine0 = src.getLinePtr(srcY - 1, srcWidth, buf0);
        auto* srcLine1 = src.getLinePtr(srcY + 0, srcWidth, buf1);
        auto* srcLine2 = src.getLinePtr(srcY + 1, srcWidth, buf2);

        for (unsigned dstY = dstStartY; dstY < dstEndY; srcY += 1, dstY += 2) {
                auto* srcLine3 = src.getLinePtr(srcY + 2, srcWidth, buf3);
                auto* dstUpper = dst.acquireLine(dstY + 0);
                auto* dstLower = dst.acquireLine(dstY + 1);
                scaleLine1on2(srcLine0, srcLine1, srcLine2, srcLine3,
                              dstUpper, dstLower, srcWidth);
                dst.releaseLine(dstY + 0, dstUpper);
                dst.releaseLine(dstY + 1, dstLower);
                srcLine0 = srcLine1;
                srcLine1 = srcLine2;
                srcLine2 = srcLine3;
                std::swap(buf0, buf1);
                std::swap(buf1, buf2);
                std::swap(buf2, buf3);
        }
}

template <class Pixel>
void SaI2xScaler<Pixel>::scale1x1to1x2(FrameSource& src,
        unsigned srcStartY, unsigned /*srcEndY*/, unsigned srcWidth,
        ScalerOutput<Pixel>& dst, unsigned dstStartY, unsigned dstEndY)
{
        assert(dst.getWidth() == srcWidth);

        VLA_SSE_ALIGNED(Pixel, buf0_, srcWidth); auto* buf0 = buf0_;
        VLA_SSE_ALIGNED(Pixel, buf1_, srcWidth); auto* buf1 = buf1_;
        VLA_SSE_ALIGNED(Pixel, buf2_, srcWidth); auto* buf2 = buf2_;
        VLA_SSE_ALIGNED(Pixel, buf3_, srcWidth); auto* buf3 = buf3_;

        int srcY = srcStartY;
        auto* srcLine0 = src.getLinePtr(srcY - 1, srcWidth, buf0);
        auto* srcLine1 = src.getLinePtr(srcY + 0, srcWidth, buf1);
        auto* srcLine2 = src.getLinePtr(srcY + 1, srcWidth, buf2);

        for (unsigned dstY = dstStartY; dstY < dstEndY; srcY += 1, dstY += 2) {
                auto* srcLine3 = src.getLinePtr(srcY + 2, srcWidth, buf3);
                auto* dstUpper = dst.acquireLine(dstY + 0);
                auto* dstLower = dst.acquireLine(dstY + 1);
                scaleLine1on1(srcLine0, srcLine1, srcLine2, srcLine3,
                              dstUpper, dstLower, srcWidth);
                dst.releaseLine(dstY + 0, dstUpper);
                dst.releaseLine(dstY + 1, dstLower);
                srcLine0 = srcLine1;
                srcLine1 = srcLine2;
                srcLine2 = srcLine3;
                std::swap(buf0, buf1);
                std::swap(buf1, buf2);
                std::swap(buf2, buf3);
        }
}


// Force template instantiation.
#if HAVE_16BPP
template class SaI2xScaler<uint16_t>;
#endif
#if HAVE_32BPP
template class SaI2xScaler<uint32_t>;
#endif

} // namespace openmsx