utils.c

    c->chrSrcVSubSample += c->vChrDrop;

    /* drop every other pixel for chroma calculation unless user
     * wants full chroma */
    if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&
        srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
        srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
        srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
        srcFormat != AV_PIX_FMT_GBRP9BE   && srcFormat != AV_PIX_FMT_GBRP9LE  &&
        srcFormat != AV_PIX_FMT_GBRP10BE  && srcFormat != AV_PIX_FMT_GBRP10LE &&
        srcFormat != AV_PIX_FMT_GBRP16BE  && srcFormat != AV_PIX_FMT_GBRP16LE &&
        ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
         (flags & SWS_FAST_BILINEAR)))
        c->chrSrcHSubSample = 1;

    // Note the -((-x)>>y) is so that we always round toward +inf.
    c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);
    c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);
    c->chrDstW = -((-dstW) >> c->chrDstHSubSample);
    c->chrDstH = -((-dstH) >> c->chrDstVSubSample);

    /* unscaled special cases */
    if (unscaled && !usesHFilter && !usesVFilter &&
        (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
        ff_get_unscaled_swscale(c);

        if (c->swscale) {
            if (flags & SWS_PRINT_INFO)
                av_log(c, AV_LOG_INFO,
                       "using unscaled %s -> %s special converter\n",
                       sws_format_name(srcFormat), sws_format_name(dstFormat));
            return 0;
        }
    }

    c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
    if (c->srcBpc < 8)
        c->srcBpc = 8;
    c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
    if (c->dstBpc < 8)
        c->dstBpc = 8;
    if (c->dstBpc == 16)
        dst_stride <<= 1;
    FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,
                     (FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3) + 16,
                     fail);
    if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 10) {
        c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
                              (srcW & 15) == 0) ? 1 : 0;
        if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0
            && (flags & SWS_FAST_BILINEAR)) {
            if (flags & SWS_PRINT_INFO)
                av_log(c, AV_LOG_INFO,
                       "output width is not a multiple of 32 -> no MMXEXT scaler\n");
        }
        if (usesHFilter)
            c->canMMXEXTBeUsed = 0;
    } else
        c->canMMXEXTBeUsed = 0;

    c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
    c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

    /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
     * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
     * correct scaling.
     * n-2 is the last chrominance sample available.
     * This is not perfect, but no one should notice the difference, the more
     * correct variant would be like the vertical one, but that would require
     * some special code for the first and last pixel */
    if (flags & SWS_FAST_BILINEAR) {
        if (c->canMMXEXTBeUsed) {
            c->lumXInc += 20;
            c->chrXInc += 20;
        }
        // we don't use the x86 asm scaler if MMX is available
        else if (INLINE_MMX(cpu_flags)) {
            c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;
            c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
        }
    }

#define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

    /* precalculate horizontal scaler filter coefficients */
    {
#if HAVE_MMXEXT_INLINE
// can't downscale !!!
        if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
            c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
                                                             NULL, NULL, 8);
            c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
                                                             NULL, NULL, NULL, 4);

#if USE_MMAP
            c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
                                          PROT_READ | PROT_WRITE,
                                          MAP_PRIVATE | MAP_ANONYMOUS,
                                          -1, 0);
            c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
                                          PROT_READ | PROT_WRITE,
                                          MAP_PRIVATE | MAP_ANONYMOUS,
                                          -1, 0);
#elif HAVE_VIRTUALALLOC
            c->lumMmxextFilterCode = VirtualAlloc(NULL,
                                                  c->lumMmxextFilterCodeSize,
                                                  MEM_COMMIT,
                                                  PAGE_EXECUTE_READWRITE);
            c->chrMmxextFilterCode = VirtualAlloc(NULL,
                                                  c->chrMmxextFilterCodeSize,
                                                  MEM_COMMIT,
                                                  PAGE_EXECUTE_READWRITE);
#else
            c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
            c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
#endif

            if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
                return AVERROR(ENOMEM);
            FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);
            FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);
            FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);
            FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

            init_hscaler_mmxext(dstW, c->lumXInc, c->lumMmxextFilterCode,
                                c->hLumFilter, c->hLumFilterPos, 8);
            init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
                                c->hChrFilter, c->hChrFilterPos, 4);

#if USE_MMAP
            mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
            mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
#endif
        } else
#endif /* HAVE_MMXEXT_INLINE */
        {
            const int filterAlign = X86_MMX(cpu_flags)     ? 4 :
                                    PPC_ALTIVEC(cpu_flags) ? 8 : 1;

            if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
                           &c->hLumFilterSize, c->lumXInc,
                           srcW, dstW, filterAlign, 1 << 14,
                           (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
                           cpu_flags, srcFilter->lumH, dstFilter->lumH,
                           c->param, 1) < 0)
                goto fail;
            if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
                           &c->hChrFilterSize, c->chrXInc,
                           c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
                           (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
                           cpu_flags, srcFilter->chrH, dstFilter->chrH,
                           c->param, 1) < 0)
                goto fail;
        }
    } // initialize horizontal stuff

    /* precalculate vertical scaler filter coefficients */
    {
        const int filterAlign = X86_MMX(cpu_flags)     ? 2 :
                                PPC_ALTIVEC(cpu_flags) ? 8 : 1;

        if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
                       c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
                       (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
                       cpu_flags, srcFilter->lumV, dstFilter->lumV,
                       c->param, 0) < 0)
            goto fail;
        if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
                       c->chrYInc, c->chrSrcH, c->chrDstH,
                       filterAlign, (1 << 12),
                       (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
                       cpu_flags, srcFilter->chrV, dstFilter->chrV,
                       c->param, 0) < 0)
            goto fail;

#if HAVE_ALTIVEC
        FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);
        FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

        for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
            int j;
            short *p = (short *)&c->vYCoeffsBank[i];
            for (j = 0; j < 8; j++)
                p[j] = c->vLumFilter[i];
        }

        for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
            int j;
            short *p = (short *)&c->vCCoeffsBank[i];
            for (j = 0; j < 8; j++)
                p[j] = c->vChrFilter[i];
        }
#endif
    }

    // calculate buffer sizes so that they won't run out while handling these damn slices
    c->vLumBufSize = c->vLumFilterSize;
    c->vChrBufSize = c->vChrFilterSize;
    for (i = 0; i < dstH; i++) {
        int chrI      = (int64_t)i * c->chrDstH / dstH;
        int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
                              ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
                               << c->chrSrcVSubSample));

        nextSlice >>= c->chrSrcVSubSample;
        nextSlice <<= c->chrSrcVSubSample;
        if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
            c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
        if (c->vChrFilterPos[chrI] + c->vChrBufSize <
            (nextSlice >> c->chrSrcVSubSample))
            c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
                             c->vChrFilterPos[chrI];
    }

    /* Allocate pixbufs (we use dynamic allocation because otherwise we would
     * need to allocate several megabytes to handle all possible cases) */
    FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);
    FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
    FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
    if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
        FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
    /* Note we need at least one pixel more at the end because of the MMX code
     * (just in case someone wants to replace the 4000/8000). */
    /* align at 16 bytes for AltiVec */
    for (i = 0; i < c->vLumBufSize; i++) {
        FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
                          dst_stride + 16, fail);
        c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
    }
    // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)
    c->uv_off_px   = dst_stride_px + 64 / (c->dstBpc & ~7);
    c->uv_off_byte = dst_stride + 16;
    for (i = 0; i < c->vChrBufSize; i++) {
        FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
                         dst_stride * 2 + 32, fail);
        c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
        c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
                         = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
    }
    if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
        for (i = 0; i < c->vLumBufSize; i++) {
            FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
                              dst_stride + 16, fail);
            c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
        }

    // try to avoid drawing green stuff between the right end and the stride end
    for (i = 0; i < c->vChrBufSize; i++)
        memset(c->chrUPixBuf[i], 64, dst_stride * 2 + 1);

    assert(c->chrDstH <= dstH);

    if (flags & SWS_PRINT_INFO) {
        if (flags & SWS_FAST_BILINEAR)
            av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
        else if (flags & SWS_BILINEAR)
            av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
        else if (flags & SWS_BICUBIC)
            av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
        else if (flags & SWS_X)
            av_log(c, AV_LOG_INFO, "Experimental scaler, ");
        else if (flags & SWS_POINT)
            av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
        else if (flags & SWS_AREA)
            av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
        else if (flags & SWS_BICUBLIN)
            av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
        else if (flags & SWS_GAUSS)
            av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
        else if (flags & SWS_SINC)
            av_log(c, AV_LOG_INFO, "Sinc scaler, ");
        else if (flags & SWS_LANCZOS)
            av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
        else if (flags & SWS_SPLINE)
            av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
        else
            av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

        av_log(c, AV_LOG_INFO, "from %s to %s%s ",
               sws_format_name(srcFormat),
#ifdef DITHER1XBPP
               dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||
               dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
               dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
                                                             "dithered " : "",
#else
               "",
#endif
               sws_format_name(dstFormat));

        if (INLINE_MMXEXT(cpu_flags))
            av_log(c, AV_LOG_INFO, "using MMXEXT\n");
        else if (INLINE_AMD3DNOW(cpu_flags))
            av_log(c, AV_LOG_INFO, "using 3DNOW\n");
        else if (INLINE_MMX(cpu_flags))
            av_log(c, AV_LOG_INFO, "using MMX\n");
        else if (PPC_ALTIVEC(cpu_flags))
            av_log(c, AV_LOG_INFO, "using AltiVec\n");
        else
            av_log(c, AV_LOG_INFO, "using C\n");

        av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
        av_log(c, AV_LOG_DEBUG,
               "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
               c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
        av_log(c, AV_LOG_DEBUG,
               "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
               c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
               c->chrXInc, c->chrYInc);
    }

    c->swscale = ff_getSwsFunc(c);
    return 0;
fail: // FIXME replace things by appropriate error codes
    return -1;
}

#if FF_API_SWS_GETCONTEXT
SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
                           int dstW, int dstH, enum AVPixelFormat dstFormat,
                           int flags, SwsFilter *srcFilter,
                           SwsFilter *dstFilter, const double *param)
{
    SwsContext *c;

    if (!(c = sws_alloc_context()))
        return NULL;

    c->flags     = flags;
    c->srcW      = srcW;
    c->srcH      = srcH;
    c->dstW      = dstW;
    c->dstH      = dstH;
    c->srcRange  = handle_jpeg(&srcFormat);
    c->dstRange  = handle_jpeg(&dstFormat);
    c->srcFormat = srcFormat;
    c->dstFormat = dstFormat;

    if (param) {
        c->param[0] = param[0];
        c->param[1] = param[1];
    }
    sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
                             ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
                             c->dstRange, 0, 1 << 16, 1 << 16);

    if (sws_init_context(c, srcFilter, dstFilter) < 0) {
        sws_freeContext(c);
        return NULL;
    }

    return c;
}
#endif

SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
                                float lumaSharpen, float chromaSharpen,
                                float chromaHShift, float chromaVShift,
                                int verbose)
{
    SwsFilter *filter = av_malloc(sizeof(SwsFilter));
    if (!filter)
        return NULL;

    if (lumaGBlur != 0.0) {
        filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
        filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
    } else {
        filter->lumH = sws_getIdentityVec();
        filter->lumV = sws_getIdentityVec();
    }

    if (chromaGBlur != 0.0) {
        filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
        filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
    } else {
        filter->chrH = sws_getIdentityVec();
        filter->chrV = sws_getIdentityVec();
    }

    if (chromaSharpen != 0.0) {
        SwsVector *id = sws_getIdentityVec();
        sws_scaleVec(filter->chrH, -chromaSharpen);
        sws_scaleVec(filter->chrV, -chromaSharpen);
        sws_addVec(filter->chrH, id);
        sws_addVec(filter->chrV, id);
        sws_freeVec(id);
    }

    if (lumaSharpen != 0.0) {
        SwsVector *id = sws_getIdentityVec();
        sws_scaleVec(filter->lumH, -lumaSharpen);
        sws_scaleVec(filter->lumV, -lumaSharpen);
        sws_addVec(filter->lumH, id);
        sws_addVec(filter->lumV, id);
        sws_freeVec(id);
    }

    if (chromaHShift != 0.0)
        sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

    if (chromaVShift != 0.0)
        sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

    sws_normalizeVec(filter->chrH, 1.0);
    sws_normalizeVec(filter->chrV, 1.0);
    sws_normalizeVec(filter->lumH, 1.0);
    sws_normalizeVec(filter->lumV, 1.0);

    if (verbose)
        sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
    if (verbose)
        sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

    return filter;
}

SwsVector *sws_allocVec(int length)
{
    SwsVector *vec = av_malloc(sizeof(SwsVector));
    if (!vec)
        return NULL;
    vec->length = length;
    vec->coeff  = av_malloc(sizeof(double) * length);
    if (!vec->coeff)
        av_freep(&vec);
    return vec;
}

SwsVector *sws_getGaussianVec(double variance, double quality)
{
    const int length = (int)(variance * quality + 0.5) | 1;
    int i;
    double middle  = (length - 1) * 0.5;
    SwsVector *vec = sws_allocVec(length);

    if (!vec)
        return NULL;

    for (i = 0; i < length; i++) {
        double dist = i - middle;
        vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
                        sqrt(2 * variance * M_PI);
    }

    sws_normalizeVec(vec, 1.0);

    return vec;
}

SwsVector *sws_getConstVec(double c, int length)
{
    int i;
    SwsVector *vec = sws_allocVec(length);

    if (!vec)
        return NULL;

    for (i = 0; i < length; i++)
        vec->coeff[i] = c;

    return vec;
}

SwsVector *sws_getIdentityVec(void)
{
    return sws_getConstVec(1.0, 1);
}

static double sws_dcVec(SwsVector *a)
{
    int i;
    double sum = 0;

    for (i = 0; i < a->length; i++)
        sum += a->coeff[i];

    return sum;
}

void sws_scaleVec(SwsVector *a, double scalar)
{
    int i;

    for (i = 0; i < a->length; i++)
        a->coeff[i] *= scalar;
}

void sws_normalizeVec(SwsVector *a, double height)
{
    sws_scaleVec(a, height / sws_dcVec(a));
}

static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
{
    int length = a->length + b->length - 1;
    int i, j;
    SwsVector *vec = sws_getConstVec(0.0, length);

    if (!vec)
        return NULL;

    for (i = 0; i < a->length; i++) {
        for (j = 0; j < b->length; j++) {
            vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
        }
    }

    return vec;
}

static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
{
    int length = FFMAX(a->length, b->length);
    int i;
    SwsVector *vec = sws_getConstVec(0.0, length);

    if (!vec)
        return NULL;

    for (i = 0; i < a->length; i++)
        vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
    for (i = 0; i < b->length; i++)
        vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

    return vec;
}

static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
{
    int length = FFMAX(a->length, b->length);
    int i;
    SwsVector *vec = sws_getConstVec(0.0, length);

    if (!vec)
        return NULL;

    for (i = 0; i < a->length; i++)
        vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
    for (i = 0; i < b->length; i++)
        vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

    return vec;
}

/* shift left / or right if "shift" is negative */
static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
{
    int length = a->length + FFABS(shift) * 2;
    int i;
    SwsVector *vec = sws_getConstVec(0.0, length);

    if (!vec)
        return NULL;

    for (i = 0; i < a->length; i++) {
        vec->coeff[i + (length    - 1) / 2 -
                       (a->length - 1) / 2 - shift] = a->coeff[i];
    }

    return vec;
}

void sws_shiftVec(SwsVector *a, int shift)
{
    SwsVector *shifted = sws_getShiftedVec(a, shift);
    av_free(a->coeff);
    a->coeff  = shifted->coeff;
    a->length = shifted->length;
    av_free(shifted);
}

void sws_addVec(SwsVector *a, SwsVector *b)
{
    SwsVector *sum = sws_sumVec(a, b);
    av_free(a->coeff);
    a->coeff  = sum->coeff;
    a->length = sum->length;
    av_free(sum);
}

void sws_subVec(SwsVector *a, SwsVector *b)
{
    SwsVector *diff = sws_diffVec(a, b);
    av_free(a->coeff);
    a->coeff  = diff->coeff;
    a->length = diff->length;
    av_free(diff);
}

void sws_convVec(SwsVector *a, SwsVector *b)
{
    SwsVector *conv = sws_getConvVec(a, b);
    av_free(a->coeff);
    a->coeff  = conv->coeff;
    a->length = conv->length;
    av_free(conv);
}

SwsVector *sws_cloneVec(SwsVector *a)
{
    int i;
    SwsVector *vec = sws_allocVec(a->length);

    if (!vec)
        return NULL;

    for (i = 0; i < a->length; i++)
        vec->coeff[i] = a->coeff[i];

    return vec;
}

void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
{
    int i;
    double max = 0;
    double min = 0;
    double range;

    for (i = 0; i < a->length; i++)
        if (a->coeff[i] > max)
            max = a->coeff[i];

    for (i = 0; i < a->length; i++)
        if (a->coeff[i] < min)
            min = a->coeff[i];

    range = max - min;

    for (i = 0; i < a->length; i++) {
        int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
        av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
        for (; x > 0; x--)
            av_log(log_ctx, log_level, " ");
        av_log(log_ctx, log_level, "|\n");
    }
}

void sws_freeVec(SwsVector *a)
{
    if (!a)
        return;
    av_freep(&a->coeff);
    a->length = 0;
    av_free(a);
}

void sws_freeFilter(SwsFilter *filter)
{
    if (!filter)
        return;

    if (filter->lumH)
        sws_freeVec(filter->lumH);
    if (filter->lumV)
        sws_freeVec(filter->lumV);
    if (filter->chrH)
        sws_freeVec(filter->chrH);
    if (filter->chrV)
        sws_freeVec(filter->chrV);
    av_free(filter);
}

void sws_freeContext(SwsContext *c)
{
    int i;
    if (!c)
        return;

    if (c->lumPixBuf) {
        for (i = 0; i < c->vLumBufSize; i++)
            av_freep(&c->lumPixBuf[i]);
        av_freep(&c->lumPixBuf);
    }

    if (c->chrUPixBuf) {
        for (i = 0; i < c->vChrBufSize; i++)
            av_freep(&c->chrUPixBuf[i]);
        av_freep(&c->chrUPixBuf);
        av_freep(&c->chrVPixBuf);
    }

    if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
        for (i = 0; i < c->vLumBufSize; i++)
            av_freep(&c->alpPixBuf[i]);
        av_freep(&c->alpPixBuf);
    }

    av_freep(&c->vLumFilter);
    av_freep(&c->vChrFilter);
    av_freep(&c->hLumFilter);
    av_freep(&c->hChrFilter);
#if HAVE_ALTIVEC
    av_freep(&c->vYCoeffsBank);
    av_freep(&c->vCCoeffsBank);
#endif

    av_freep(&c->vLumFilterPos);
    av_freep(&c->vChrFilterPos);
    av_freep(&c->hLumFilterPos);
    av_freep(&c->hChrFilterPos);

#if HAVE_MMX_INLINE
#if USE_MMAP
    if (c->lumMmxextFilterCode)
        munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
    if (c->chrMmxextFilterCode)
        munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
#elif HAVE_VIRTUALALLOC
    if (c->lumMmxextFilterCode)
        VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
    if (c->chrMmxextFilterCode)
        VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
#else
    av_free(c->lumMmxextFilterCode);
    av_free(c->chrMmxextFilterCode);
#endif
    c->lumMmxextFilterCode = NULL;
    c->chrMmxextFilterCode = NULL;
#endif /* HAVE_MMX_INLINE */

    av_freep(&c->yuvTable);
    av_free(c->formatConvBuffer);

    av_free(c);
}

struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
                                        int srcH, enum AVPixelFormat srcFormat,
                                        int dstW, int dstH,
                                        enum AVPixelFormat dstFormat, int flags,
                                        SwsFilter *srcFilter,
                                        SwsFilter *dstFilter,
                                        const double *param)
{
    static const double default_param[2] = { SWS_PARAM_DEFAULT,
                                             SWS_PARAM_DEFAULT };

    if (!param)
        param = default_param;

    if (context &&
        (context->srcW      != srcW      ||
         context->srcH      != srcH      ||
         context->srcFormat != srcFormat ||
         context->dstW      != dstW      ||
         context->dstH      != dstH      ||
         context->dstFormat != dstFormat ||
         context->flags     != flags     ||
         context->param[0]  != param[0]  ||
         context->param[1]  != param[1])) {
        sws_freeContext(context);
        context = NULL;
    }

    if (!context) {
        if (!(context = sws_alloc_context()))
            return NULL;
        context->srcW      = srcW;
        context->srcH      = srcH;
        context->srcRange  = handle_jpeg(&srcFormat);
        context->srcFormat = srcFormat;
        context->dstW      = dstW;
        context->dstH      = dstH;
        context->dstRange  = handle_jpeg(&dstFormat);
        context->dstFormat = dstFormat;
        context->flags     = flags;
        context->param[0]  = param[0];
        context->param[1]  = param[1];
        sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
                                 context->srcRange,
                                 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
                                 context->dstRange, 0, 1 << 16, 1 << 16);
        if (sws_init_context(context, srcFilter, dstFilter) < 0) {
            sws_freeContext(context);
            return NULL;
        }
    }
    return context;
}