C++ MemoryBuffer::getWidth方法代码示例

void *FastGaussianBlurValueOperation::initializeTileData(rcti *rect)
{
	lockMutex();
	if (!this->m_iirgaus) {
		MemoryBuffer *newBuf = (MemoryBuffer *)this->m_inputprogram->initializeTileData(rect);
		MemoryBuffer *copy = newBuf->duplicate();
		FastGaussianBlurOperation::IIR_gauss(copy, this->m_sigma, 0, 3);

		if (this->m_overlay == FAST_GAUSS_OVERLAY_MIN) {
			float *src = newBuf->getBuffer();
			float *dst = copy->getBuffer();
			for (int i = copy->getWidth() * copy->getHeight(); i != 0; i--, src += COM_NUM_CHANNELS_VALUE, dst += COM_NUM_CHANNELS_VALUE) {
				if (*src < *dst) {
					*dst = *src;
				}
			}
		}
		else if (this->m_overlay == FAST_GAUSS_OVERLAY_MAX) {
			float *src = newBuf->getBuffer();
			float *dst = copy->getBuffer();
			for (int i = copy->getWidth() * copy->getHeight(); i != 0; i--, src += COM_NUM_CHANNELS_VALUE, dst += COM_NUM_CHANNELS_VALUE) {
				if (*src > *dst) {
					*dst = *src;
				}
			}
		}

//		newBuf->

		this->m_iirgaus = copy;
	}
	unlockMutex();
	return this->m_iirgaus;
}

void WriteBufferOperation::executeRegion(rcti *rect, unsigned int tileNumber)
{
	MemoryBuffer *memoryBuffer = this->m_memoryProxy->getBuffer();
	float *buffer = memoryBuffer->getBuffer();
	if (this->m_input->isComplex()) {
		void *data = this->m_input->initializeTileData(rect);
		int x1 = rect->xmin;
		int y1 = rect->ymin;
		int x2 = rect->xmax;
		int y2 = rect->ymax;
		int x;
		int y;
		bool breaked = false;
		for (y = y1; y < y2 && (!breaked); y++) {
			int offset4 = (y * memoryBuffer->getWidth() + x1) * COM_NUMBER_OF_CHANNELS;
			for (x = x1; x < x2; x++) {
				this->m_input->read(&(buffer[offset4]), x, y, data);
				offset4 += COM_NUMBER_OF_CHANNELS;
			}
			if (isBreaked()) {
				breaked = true;
			}

		}
		if (data) {
			this->m_input->deinitializeTileData(rect, data);
			data = NULL;
		}
	}
	else {
		int x1 = rect->xmin;
		int y1 = rect->ymin;
		int x2 = rect->xmax;
		int y2 = rect->ymax;

		int x;
		int y;
		bool breaked = false;
		for (y = y1; y < y2 && (!breaked); y++) {
			int offset4 = (y * memoryBuffer->getWidth() + x1) * COM_NUMBER_OF_CHANNELS;
			for (x = x1; x < x2; x++) {
				this->m_input->readSampled(&(buffer[offset4]), x, y, COM_PS_NEAREST);
				offset4 += COM_NUMBER_OF_CHANNELS;
			}
			if (isBreaked()) {
				breaked = true;
			}
		}
	}
	memoryBuffer->setCreatedState();
}

void GaussianAlphaYBlurOperation::executePixel(float *color, int x, int y, void *data)
{
	const bool do_invert = this->m_do_subtract;
	MemoryBuffer *inputBuffer = (MemoryBuffer *)data;
	float *buffer = inputBuffer->getBuffer();
	int bufferwidth = inputBuffer->getWidth();
	int bufferstartx = inputBuffer->getRect()->xmin;
	int bufferstarty = inputBuffer->getRect()->ymin;

	int miny = y - this->m_rad;
	int maxy = y + this->m_rad;
	int minx = x;
	int maxx = x;
	miny = max(miny, inputBuffer->getRect()->ymin);
	minx = max(minx, inputBuffer->getRect()->xmin);
	maxy = min(maxy, inputBuffer->getRect()->ymax);
	maxx = min(maxx, inputBuffer->getRect()->xmax);

	/* *** this is the main part which is different to 'GaussianYBlurOperation'  *** */
	int step = getStep();

	/* gauss */
	float alpha_accum = 0.0f;
	float multiplier_accum = 0.0f;

	/* dilate */
	float value_max = finv_test(buffer[(x * 4) + (y * 4 * bufferwidth)], do_invert); /* init with the current color to avoid unneeded lookups */
	float distfacinv_max = 1.0f; /* 0 to 1 */

	for (int ny = miny; ny < maxy; ny += step) {
		int bufferindex = ((minx - bufferstartx) * 4) + ((ny - bufferstarty) * 4 * bufferwidth);

		const int index = (ny - y) + this->m_rad;
		float value = finv_test(buffer[bufferindex], do_invert);
		float multiplier;

		/* gauss */
		{
			multiplier = this->m_gausstab[index];
			alpha_accum += value * multiplier;
			multiplier_accum += multiplier;
		}

		/* dilate - find most extreme color */
		if (value > value_max) {
			multiplier = this->m_distbuf_inv[index];
			value *= multiplier;
			if (value > value_max) {
				value_max = value;
				distfacinv_max = multiplier;
			}
		}

	}

	/* blend between the max value and gauss blue - gives nice feather */
	const float value_blur  = alpha_accum / multiplier_accum;
	const float value_final = (value_max * distfacinv_max) + (value_blur * (1.0f - distfacinv_max));
	color[0] = finv_test(value_final, do_invert);
}

void GaussianXBlurOperation::executePixel(float output[4], int x, int y, void *data)
{
	float color_accum[4] = {0.0f, 0.0f, 0.0f, 0.0f};
	float multiplier_accum = 0.0f;
	MemoryBuffer *inputBuffer = (MemoryBuffer *)data;
	float *buffer = inputBuffer->getBuffer();
	int bufferwidth = inputBuffer->getWidth();
	int bufferstartx = inputBuffer->getRect()->xmin;
	int bufferstarty = inputBuffer->getRect()->ymin;

	int miny = y;
	int minx = x - this->m_rad;
	int maxx = x + this->m_rad;
	miny = max(miny, inputBuffer->getRect()->ymin);
	minx = max(minx, inputBuffer->getRect()->xmin);
	maxx = min(maxx, inputBuffer->getRect()->xmax - 1);

	int step = getStep();
	int offsetadd = getOffsetAdd();
	int bufferindex = ((minx - bufferstartx) * 4) + ((miny - bufferstarty) * 4 * bufferwidth);
	for (int nx = minx, index = (minx - x) + this->m_rad; nx <= maxx; nx += step, index += step) {
		const float multiplier = this->m_gausstab[index];
		madd_v4_v4fl(color_accum, &buffer[bufferindex], multiplier);
		multiplier_accum += multiplier;
		bufferindex += offsetadd;
	}
	mul_v4_v4fl(output, color_accum, 1.0f / multiplier_accum);
}

void GaussianYBlurOperation::executePixel(float output[4], int x, int y, void *data)
{
	float color_accum[4] = {0.0f, 0.0f, 0.0f, 0.0f};
	float multiplier_accum = 0.0f;
	MemoryBuffer *inputBuffer = (MemoryBuffer *)data;
	float *buffer = inputBuffer->getBuffer();
	int bufferwidth = inputBuffer->getWidth();
	int bufferstartx = inputBuffer->getRect()->xmin;
	int bufferstarty = inputBuffer->getRect()->ymin;

	rcti &rect = *inputBuffer->getRect();
	int xmin = max_ii(x,                    rect.xmin);
	int ymin = max_ii(y - m_filtersize,     rect.ymin);
	int ymax = min_ii(y + m_filtersize + 1, rect.ymax);

	int index;
	int step = getStep();
	const int bufferIndexx = ((xmin - bufferstartx) * 4);
	for (int ny = ymin; ny < ymax; ny += step) {
		index = (ny - y) + this->m_filtersize;
		int bufferindex = bufferIndexx + ((ny - bufferstarty) * 4 * bufferwidth);
		const float multiplier = this->m_gausstab[index];
		madd_v4_v4fl(color_accum, &buffer[bufferindex], multiplier);
		multiplier_accum += multiplier;
	}
	mul_v4_v4fl(output, color_accum, 1.0f / multiplier_accum);
}

void BokehBlurOperation::executePixel(float output[4], int x, int y, void *data)
{
  float color_accum[4];
  float tempBoundingBox[4];
  float bokeh[4];

  this->m_inputBoundingBoxReader->readSampled(tempBoundingBox, x, y, COM_PS_NEAREST);
  if (tempBoundingBox[0] > 0.0f) {
    float multiplier_accum[4] = {0.0f, 0.0f, 0.0f, 0.0f};
    MemoryBuffer *inputBuffer = (MemoryBuffer *)data;
    float *buffer = inputBuffer->getBuffer();
    int bufferwidth = inputBuffer->getWidth();
    int bufferstartx = inputBuffer->getRect()->xmin;
    int bufferstarty = inputBuffer->getRect()->ymin;
    const float max_dim = max(this->getWidth(), this->getHeight());
    int pixelSize = this->m_size * max_dim / 100.0f;
    zero_v4(color_accum);

    if (pixelSize < 2) {
      this->m_inputProgram->readSampled(color_accum, x, y, COM_PS_NEAREST);
      multiplier_accum[0] = 1.0f;
      multiplier_accum[1] = 1.0f;
      multiplier_accum[2] = 1.0f;
      multiplier_accum[3] = 1.0f;
    }
    int miny = y - pixelSize;
    int maxy = y + pixelSize;
    int minx = x - pixelSize;
    int maxx = x + pixelSize;
    miny = max(miny, inputBuffer->getRect()->ymin);
    minx = max(minx, inputBuffer->getRect()->xmin);
    maxy = min(maxy, inputBuffer->getRect()->ymax);
    maxx = min(maxx, inputBuffer->getRect()->xmax);

    int step = getStep();
    int offsetadd = getOffsetAdd() * COM_NUM_CHANNELS_COLOR;

    float m = this->m_bokehDimension / pixelSize;
    for (int ny = miny; ny < maxy; ny += step) {
      int bufferindex = ((minx - bufferstartx) * COM_NUM_CHANNELS_COLOR) +
                        ((ny - bufferstarty) * COM_NUM_CHANNELS_COLOR * bufferwidth);
      for (int nx = minx; nx < maxx; nx += step) {
        float u = this->m_bokehMidX - (nx - x) * m;
        float v = this->m_bokehMidY - (ny - y) * m;
        this->m_inputBokehProgram->readSampled(bokeh, u, v, COM_PS_NEAREST);
        madd_v4_v4v4(color_accum, bokeh, &buffer[bufferindex]);
        add_v4_v4(multiplier_accum, bokeh);
        bufferindex += offsetadd;
      }
    }
    output[0] = color_accum[0] * (1.0f / multiplier_accum[0]);
    output[1] = color_accum[1] * (1.0f / multiplier_accum[1]);
    output[2] = color_accum[2] * (1.0f / multiplier_accum[2]);
    output[3] = color_accum[3] * (1.0f / multiplier_accum[3]);
  }
  else {
    this->m_inputProgram->readSampled(output, x, y, COM_PS_NEAREST);
  }
}

void KeyingClipOperation::executePixel(float *color, int x, int y, void *data)
{
	const int delta = this->m_kernelRadius;
	const float tolerance = this->m_kernelTolerance;

	MemoryBuffer *inputBuffer = (MemoryBuffer *)data;
	float *buffer = inputBuffer->getBuffer();

	int bufferWidth = inputBuffer->getWidth();
	int bufferHeight = inputBuffer->getHeight();

	int i, j, count = 0, totalCount = 0;

	float value = buffer[(y * bufferWidth + x) * 4];

	bool ok = false;

	for (i = -delta + 1; i < delta; i++) {
		for (j = -delta + 1; j < delta; j++) {
			int cx = x + j, cy = y + i;

			if (i == 0 && j == 0)
				continue;

			if (cx >= 0 && cx < bufferWidth && cy >= 0 && cy < bufferHeight) {
				int bufferIndex = (cy * bufferWidth + cx) * 4;
				float currentValue = buffer[bufferIndex];

				if (fabsf(currentValue - value) < tolerance) {
					count++;
				}

				totalCount++;
			}
		}
	}

	ok = count >= (float) totalCount * 0.9f;

	if (this->m_isEdgeMatte) {
		if (ok)
			color[0] = 0.0f;
		else
			color[0] = 1.0f;
	}
	else {
		color[0] = value;

		if (ok) {
			if (color[0] < this->m_clipBlack)
				color[0] = 0.0f;
			else if (color[0] >= this->m_clipWhite)
				color[0] = 1.0f;
			else
				color[0] = (color[0] - this->m_clipBlack) / (this->m_clipWhite - this->m_clipBlack);
		}
	}
}

void GaussianAlphaXBlurOperation::executePixel(float output[4], int x, int y, void *data)
{
	const bool do_invert = this->m_do_subtract;
	MemoryBuffer *inputBuffer = (MemoryBuffer *)data;
	float *buffer = inputBuffer->getBuffer();
	int bufferwidth = inputBuffer->getWidth();
	int bufferstartx = inputBuffer->getRect()->xmin;
	int bufferstarty = inputBuffer->getRect()->ymin;

	rcti &rect = *inputBuffer->getRect();
	int xmin = max_ii(x - m_filtersize,     rect.xmin);
	int xmax = min_ii(x + m_filtersize + 1, rect.xmax);
	int ymin = max_ii(y,                    rect.ymin);

	/* *** this is the main part which is different to 'GaussianXBlurOperation'  *** */
	int step = getStep();
	int offsetadd = getOffsetAdd();
	int bufferindex = ((xmin - bufferstartx) * 4) + ((ymin - bufferstarty) * 4 * bufferwidth);

	/* gauss */
	float alpha_accum = 0.0f;
	float multiplier_accum = 0.0f;

	/* dilate */
	float value_max = finv_test(buffer[(x * 4) + (y * 4 * bufferwidth)], do_invert); /* init with the current color to avoid unneeded lookups */
	float distfacinv_max = 1.0f; /* 0 to 1 */

	for (int nx = xmin; nx < xmax; nx += step) {
		const int index = (nx - x) + this->m_filtersize;
		float value = finv_test(buffer[bufferindex], do_invert);
		float multiplier;

		/* gauss */
		{
			multiplier = this->m_gausstab[index];
			alpha_accum += value * multiplier;
			multiplier_accum += multiplier;
		}

		/* dilate - find most extreme color */
		if (value > value_max) {
			multiplier = this->m_distbuf_inv[index];
			value *= multiplier;
			if (value > value_max) {
				value_max = value;
				distfacinv_max = multiplier;
			}
		}
		bufferindex += offsetadd;
	}

	/* blend between the max value and gauss blue - gives nice feather */
	const float value_blur  = alpha_accum / multiplier_accum;
	const float value_final = (value_max * distfacinv_max) + (value_blur * (1.0f - distfacinv_max));
	output[0] = finv_test(value_final, do_invert);
}

void *AntiAliasOperation::initializeTileData(rcti *rect)
{
	if (this->m_buffer) { return this->m_buffer; }
	lockMutex();
	if (this->m_buffer == NULL) {
		MemoryBuffer *tile = (MemoryBuffer *)this->m_valueReader->initializeTileData(rect);
		int size = tile->getHeight() * tile->getWidth();
		float *input = tile->getBuffer();
		char *valuebuffer = (char *)MEM_mallocN(sizeof(char) * size, __func__);
		for (int i = 0; i < size; i++) {
			float in = input[i * COM_NUMBER_OF_CHANNELS];
			valuebuffer[i] = FTOCHAR(in);
		}
		antialias_tagbuf(tile->getWidth(), tile->getHeight(), valuebuffer);
		this->m_buffer = valuebuffer;
	}
	unlockMutex();
	return this->m_buffer;
}

void *ErodeStepOperation::initializeTileData(rcti *rect)
{
	if (this->m_cached_buffer != NULL) {
		return this->m_cached_buffer;
	}
	lockMutex();
	if (this->m_cached_buffer == NULL) {
		MemoryBuffer *buffer = (MemoryBuffer *)this->m_inputProgram->initializeTileData(NULL);
		float *rectf = buffer->convertToValueBuffer();
		int x, y, i;
		float *p;
		int bwidth = buffer->getWidth();
		int bheight = buffer->getHeight();
		for (i = 0; i < this->m_iterations; i++) {
			for (y = 0; y < bheight; y++) {
				for (x = 0; x < bwidth - 1; x++) {
					p = rectf + (bwidth * y + x);
					*p = MIN2(*p, *(p + 1));
				}
			}
		
			for (y = 0; y < bheight; y++) {
				for (x = bwidth - 1; x >= 1; x--) {
					p = rectf + (bwidth * y + x);
					*p = MIN2(*p, *(p - 1));
				}
			}
		
			for (x = 0; x < bwidth; x++) {
				for (y = 0; y < bheight - 1; y++) {
					p = rectf + (bwidth * y + x);
					*p = MIN2(*p, *(p + bwidth));
				}
			}
		
			for (x = 0; x < bwidth; x++) {
				for (y = bheight - 1; y >= 1; y--) {
					p = rectf + (bwidth * y + x);
					*p = MIN2(*p, *(p - bwidth));
				}
			}
		}
		this->m_cached_buffer = rectf;
	}
	unlockMutex();
	return this->m_cached_buffer;
}

void GaussianXBlurOperation::executePixel(float output[4], int x, int y, void *data)
{
	float color_accum[4] = {0.0f, 0.0f, 0.0f, 0.0f};
	float multiplier_accum = 0.0f;
	MemoryBuffer *inputBuffer = (MemoryBuffer *)data;
	float *buffer = inputBuffer->getBuffer();
	int bufferwidth = inputBuffer->getWidth();
	int bufferstartx = inputBuffer->getRect()->xmin;
	int bufferstarty = inputBuffer->getRect()->ymin;

	rcti &rect = *inputBuffer->getRect();
	int xmin = max_ii(x - m_filtersize,     rect.xmin);
	int xmax = min_ii(x + m_filtersize + 1, rect.xmax);
	int ymin = max_ii(y,                    rect.ymin);

	int step = getStep();
	int offsetadd = getOffsetAdd();
	int bufferindex = ((xmin - bufferstartx) * 4) + ((ymin - bufferstarty) * 4 * bufferwidth);

#ifdef __SSE2__
	__m128 accum_r = _mm_load_ps(color_accum);
	for (int nx = xmin, index = (xmin - x) + this->m_filtersize; nx < xmax; nx += step, index += step) {
		__m128 reg_a = _mm_load_ps(&buffer[bufferindex]);
		reg_a = _mm_mul_ps(reg_a, this->m_gausstab_sse[index]);
		accum_r = _mm_add_ps(accum_r, reg_a);
		multiplier_accum += this->m_gausstab[index];
		bufferindex += offsetadd;
	}
	_mm_store_ps(color_accum, accum_r);
#else
	for (int nx = xmin, index = (xmin - x) + this->m_filtersize; nx < xmax; nx += step, index += step) {
		const float multiplier = this->m_gausstab[index];
		madd_v4_v4fl(color_accum, &buffer[bufferindex], multiplier);
		multiplier_accum += multiplier;
		bufferindex += offsetadd;
	}
#endif
	mul_v4_v4fl(output, color_accum, 1.0f / multiplier_accum);
}

cl_mem OpenCLDevice::COM_clAttachMemoryBufferToKernelParameter(cl_kernel kernel,
                                                               int parameterIndex,
                                                               int offsetIndex,
                                                               list<cl_mem> *cleanup,
                                                               MemoryBuffer **inputMemoryBuffers,
                                                               ReadBufferOperation *reader)
{
  cl_int error;

  MemoryBuffer *result = reader->getInputMemoryBuffer(inputMemoryBuffers);

  const cl_image_format *imageFormat = determineImageFormat(result);

  cl_mem clBuffer = clCreateImage2D(this->m_context,
                                    CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
                                    imageFormat,
                                    result->getWidth(),
                                    result->getHeight(),
                                    0,
                                    result->getBuffer(),
                                    &error);

  if (error != CL_SUCCESS) {
    printf("CLERROR[%d]: %s\n", error, clewErrorString(error));
  }
  if (error == CL_SUCCESS) {
    cleanup->push_back(clBuffer);
  }

  error = clSetKernelArg(kernel, parameterIndex, sizeof(cl_mem), &clBuffer);
  if (error != CL_SUCCESS) {
    printf("CLERROR[%d]: %s\n", error, clewErrorString(error));
  }

  COM_clAttachMemoryBufferOffsetToKernelParameter(kernel, offsetIndex, result);
  return clBuffer;
}

void VariableSizeBokehBlurOperation::executePixel(float output[4], int x, int y, void *data)
{
	VariableSizeBokehBlurTileData *tileData = (VariableSizeBokehBlurTileData *)data;
	MemoryBuffer *inputProgramBuffer = tileData->color;
	MemoryBuffer *inputBokehBuffer = tileData->bokeh;
	MemoryBuffer *inputSizeBuffer = tileData->size;
	float *inputSizeFloatBuffer = inputSizeBuffer->getBuffer();
	float *inputProgramFloatBuffer = inputProgramBuffer->getBuffer();
	float readColor[4];
	float bokeh[4];
	float tempSize[4];
	float multiplier_accum[4];
	float color_accum[4];

	const float max_dim = max(m_width, m_height);
	const float scalar = this->m_do_size_scale ? (max_dim / 100.0f) : 1.0f;
	int maxBlurScalar = tileData->maxBlurScalar;

	BLI_assert(inputBokehBuffer->getWidth()  == COM_BLUR_BOKEH_PIXELS);
	BLI_assert(inputBokehBuffer->getHeight() == COM_BLUR_BOKEH_PIXELS);

#ifdef COM_DEFOCUS_SEARCH
	float search[4];
	this->m_inputSearchProgram->read(search, x / InverseSearchRadiusOperation::DIVIDER, y / InverseSearchRadiusOperation::DIVIDER, NULL);
	int minx = search[0];
	int miny = search[1];
	int maxx = search[2];
	int maxy = search[3];
#else
	int minx = max(x - maxBlurScalar, 0);
	int miny = max(y - maxBlurScalar, 0);
	int maxx = min(x + maxBlurScalar, (int)m_width);
	int maxy = min(y + maxBlurScalar, (int)m_height);
#endif
	{
		inputSizeBuffer->readNoCheck(tempSize, x, y);
		inputProgramBuffer->readNoCheck(readColor, x, y);

		copy_v4_v4(color_accum, readColor);
		copy_v4_fl(multiplier_accum, 1.0f);
		float size_center = tempSize[0] * scalar;
		
		const int addXStepValue = QualityStepHelper::getStep();
		const int addYStepValue = addXStepValue;
		const int addXStepColor = addXStepValue * COM_NUM_CHANNELS_COLOR;

		if (size_center > this->m_threshold) {
			for (int ny = miny; ny < maxy; ny += addYStepValue) {
				float dy = ny - y;
				int offsetValueNy = ny * inputSizeBuffer->getWidth();
				int offsetValueNxNy = offsetValueNy + (minx);
				int offsetColorNxNy = offsetValueNxNy * COM_NUM_CHANNELS_COLOR;
				for (int nx = minx; nx < maxx; nx += addXStepValue) {
					if (nx != x || ny != y) {
						float size = min(inputSizeFloatBuffer[offsetValueNxNy] * scalar, size_center);
						if (size > this->m_threshold) {
							float dx = nx - x;
							if (size > fabsf(dx) && size > fabsf(dy)) {
								float uv[2] = {
									(float)(COM_BLUR_BOKEH_PIXELS / 2) + (dx / size) * (float)((COM_BLUR_BOKEH_PIXELS / 2) - 1),
									(float)(COM_BLUR_BOKEH_PIXELS / 2) + (dy / size) * (float)((COM_BLUR_BOKEH_PIXELS / 2) - 1)};
								inputBokehBuffer->read(bokeh, uv[0], uv[1]);
								madd_v4_v4v4(color_accum, bokeh, &inputProgramFloatBuffer[offsetColorNxNy]);
								add_v4_v4(multiplier_accum, bokeh);
							}
						}
					}
					offsetColorNxNy += addXStepColor;
					offsetValueNxNy += addXStepValue;				}
			}
		}

		output[0] = color_accum[0] / multiplier_accum[0];
		output[1] = color_accum[1] / multiplier_accum[1];
		output[2] = color_accum[2] / multiplier_accum[2];
		output[3] = color_accum[3] / multiplier_accum[3];

		/* blend in out values over the threshold, otherwise we get sharp, ugly transitions */
		if ((size_center > this->m_threshold) &&
		    (size_center < this->m_threshold * 2.0f))
		{
			/* factor from 0-1 */
			float fac = (size_center - this->m_threshold) / this->m_threshold;
			interp_v4_v4v4(output, readColor, output, fac);
		}
	}

}

void GlareGhostOperation::generateGlare(float *data, MemoryBuffer *inputTile, NodeGlare *settings)
{
	const int qt = 1 << settings->quality;
	const float s1 = 4.0f / (float)qt, s2 = 2.0f * s1;
	int x, y, n, p, np;
	fRGB c, tc, cm[64];
	float sc, isc, u, v, sm, s, t, ofs, scalef[64];
	const float cmo = 1.0f - settings->colmod;

	MemoryBuffer *gbuf = inputTile->duplicate();
	MemoryBuffer *tbuf1 = inputTile->duplicate();

	bool breaked = false;

	FastGaussianBlurOperation::IIR_gauss(tbuf1, s1, 0, 3);
	if (!breaked) FastGaussianBlurOperation::IIR_gauss(tbuf1, s1, 1, 3);
	if (isBreaked()) breaked = true;
	if (!breaked) FastGaussianBlurOperation::IIR_gauss(tbuf1, s1, 2, 3);

	MemoryBuffer *tbuf2 = tbuf1->duplicate();

	if (isBreaked()) breaked = true;
	if (!breaked) FastGaussianBlurOperation::IIR_gauss(tbuf2, s2, 0, 3);
	if (isBreaked()) breaked = true;
	if (!breaked) FastGaussianBlurOperation::IIR_gauss(tbuf2, s2, 1, 3);
	if (isBreaked()) breaked = true;
	if (!breaked) FastGaussianBlurOperation::IIR_gauss(tbuf2, s2, 2, 3);

	ofs = (settings->iter & 1) ? 0.5f : 0.0f;
	for (x = 0; x < (settings->iter * 4); x++) {
		y = x & 3;
		cm[x][0] = cm[x][1] = cm[x][2] = 1;
		if (y == 1) fRGB_rgbmult(cm[x], 1.0f, cmo, cmo);
		if (y == 2) fRGB_rgbmult(cm[x], cmo, cmo, 1.0f);
		if (y == 3) fRGB_rgbmult(cm[x], cmo, 1.0f, cmo);
		scalef[x] = 2.1f * (1.0f - (x + ofs) / (float)(settings->iter * 4));
		if (x & 1) scalef[x] = -0.99f / scalef[x];
	}

	sc = 2.13;
	isc = -0.97;
	for (y = 0; y < gbuf->getHeight() && (!breaked); y++) {
		v = ((float)y + 0.5f) / (float)gbuf->getHeight();
		for (x = 0; x < gbuf->getWidth(); x++) {
			u = ((float)x + 0.5f) / (float)gbuf->getWidth();
			s = (u - 0.5f) * sc + 0.5f, t = (v - 0.5f) * sc + 0.5f;
			tbuf1->readBilinear(c, s * gbuf->getWidth(), t * gbuf->getHeight());
			sm = smoothMask(s, t);
			mul_v3_fl(c, sm);
			s = (u - 0.5f) * isc + 0.5f, t = (v - 0.5f) * isc + 0.5f;
			tbuf2->readBilinear(tc, s * gbuf->getWidth() - 0.5f, t * gbuf->getHeight() - 0.5f);
			sm = smoothMask(s, t);
			madd_v3_v3fl(c, tc, sm);

			gbuf->writePixel(x, y, c);
		}
		if (isBreaked()) breaked = true;

	}

	memset(tbuf1->getBuffer(), 0, tbuf1->getWidth() * tbuf1->getHeight() * COM_NUM_CHANNELS_COLOR * sizeof(float));
	for (n = 1; n < settings->iter && (!breaked); n++) {
		for (y = 0; y < gbuf->getHeight() && (!breaked); y++) {
			v = ((float)y + 0.5f) / (float)gbuf->getHeight();
			for (x = 0; x < gbuf->getWidth(); x++) {
				u = ((float)x + 0.5f) / (float)gbuf->getWidth();
				tc[0] = tc[1] = tc[2] = 0.0f;
				for (p = 0; p < 4; p++) {
					np = (n << 2) + p;
					s = (u - 0.5f) * scalef[np] + 0.5f;
					t = (v - 0.5f) * scalef[np] + 0.5f;
					gbuf->readBilinear(c, s * gbuf->getWidth() - 0.5f, t * gbuf->getHeight() - 0.5f);
					mul_v3_v3(c, cm[np]);
					sm = smoothMask(s, t) * 0.25f;
					madd_v3_v3fl(tc, c, sm);
				}
				tbuf1->addPixel(x, y, tc);
			}
			if (isBreaked()) breaked = true;
		}
		memcpy(gbuf->getBuffer(), tbuf1->getBuffer(), tbuf1->getWidth() * tbuf1->getHeight() * COM_NUM_CHANNELS_COLOR * sizeof(float));
	}
	memcpy(data, gbuf->getBuffer(), gbuf->getWidth() * gbuf->getHeight() * COM_NUM_CHANNELS_COLOR * sizeof(float));

	delete gbuf;
	delete tbuf1;
	delete tbuf2;
}

void GlareStreaksOperation::generateGlare(float *data, MemoryBuffer *inputTile, NodeGlare *settings)
{
	int x, y, n;
	unsigned int nump = 0;
	float c1[4], c2[4], c3[4], c4[4];
	float a, ang = DEG2RADF(360.0f) / (float)settings->streaks;

	int size = inputTile->getWidth() * inputTile->getHeight();
	int size4 = size * 4;

	bool breaked = false;

	MemoryBuffer *tsrc = inputTile->duplicate();
	MemoryBuffer *tdst = new MemoryBuffer(COM_DT_COLOR, inputTile->getRect());
	tdst->clear();
	memset(data, 0, size4 * sizeof(float));

	for (a = 0.0f; a < DEG2RADF(360.0f) && (!breaked); a += ang) {
		const float an = a + settings->angle_ofs;
		const float vx = cos((double)an), vy = sin((double)an);
		for (n = 0; n < settings->iter && (!breaked); ++n) {
			const float p4 = pow(4.0, (double)n);
			const float vxp = vx * p4, vyp = vy * p4;
			const float wt = pow((double)settings->fade, (double)p4);
			const float cmo = 1.0f - (float)pow((double)settings->colmod, (double)n + 1);  // colormodulation amount relative to current pass
			float *tdstcol = tdst->getBuffer();
			for (y = 0; y < tsrc->getHeight() && (!breaked); ++y) {
				for (x = 0; x < tsrc->getWidth(); ++x, tdstcol += 4) {
					// first pass no offset, always same for every pass, exact copy,
					// otherwise results in uneven brightness, only need once
					if (n == 0) tsrc->read(c1, x, y); else c1[0] = c1[1] = c1[2] = 0;
					tsrc->readBilinear(c2, x + vxp, y + vyp);
					tsrc->readBilinear(c3, x + vxp * 2.0f, y + vyp * 2.0f);
					tsrc->readBilinear(c4, x + vxp * 3.0f, y + vyp * 3.0f);
					// modulate color to look vaguely similar to a color spectrum
					c2[1] *= cmo;
					c2[2] *= cmo;

					c3[0] *= cmo;
					c3[1] *= cmo;

					c4[0] *= cmo;
					c4[2] *= cmo;

					tdstcol[0] = 0.5f * (tdstcol[0] + c1[0] + wt * (c2[0] + wt * (c3[0] + wt * c4[0])));
					tdstcol[1] = 0.5f * (tdstcol[1] + c1[1] + wt * (c2[1] + wt * (c3[1] + wt * c4[1])));
					tdstcol[2] = 0.5f * (tdstcol[2] + c1[2] + wt * (c2[2] + wt * (c3[2] + wt * c4[2])));
					tdstcol[3] = 1.0f;
				}
				if (isBreaked()) {
					breaked = true;
				}
			}
			memcpy(tsrc->getBuffer(), tdst->getBuffer(), sizeof(float) * size4);
		}

		float *sourcebuffer = tsrc->getBuffer();
		float factor = 1.0f / (float)(6 - settings->iter);
		for (int i = 0; i < size4; i += 4) {
			madd_v3_v3fl(&data[i], &sourcebuffer[i], factor);
			data[i + 3] =  1.0f;
		}

		tdst->clear();
		memcpy(tsrc->getBuffer(), inputTile->getBuffer(), sizeof(float) * size4);
		nump++;
	}

	delete tsrc;
	delete tdst;
}