imagescaler.c

#include "imagescaler.h"

#include <cstdlib>
#include <cmath>

ImageScaler::ImageScaler() :
	m_memory(NULL),
	m_hor_filters(NULL),
	m_ver_filters(NULL),
	m_buffer(NULL),
	m_dst_image(NULL),
	m_dst_stride(0),
	m_dst_width(0),
	m_dst_height(0),
	m_src_width(0),
	m_src_height(0),
	m_src_x(0),
	m_src_y(0),
	m_dst_x(0),
	m_dst_y(0) {
}

ImageScaler::~ImageScaler() {
	if ( m_memory ) free( m_memory );
}

// sin(x)/(x)
static float sincf( float x ) {
	if ( fabsf(x) < 0.05f ) return 1.0f - (1.0f/6.0f)*x*x;  // taylor series approximation to avoid 0/0
	return sin(x)/x;
}

static void CalculateFilters( ImageScaler::Filter *filters, int dst_size, int src_size ) {
	const float fc = dst_size >= src_size ? 1.0f : ((float) dst_size)/((float) src_size);

	for (int i = 0; i < dst_size; i++) {
		const int    d          = 2*dst_size;                     // sample position denominator
		const int    e          = (2*i+1) * src_size - dst_size;  // sample position enumerator
		int          offset     =  e / d;                         // truncated sample position
		const float  sub_offset = ((float) (e - offset*d)) / ((float) d);  // exact sample position is (float) e/d = offset + sub_offset

		// calculate filter coefficients
		float  h[4];
		for (int j=0; j<4; j++) {
			const float t = 3.14159265359f * (sub_offset+(1-j));
			h[j] = sincf( fc * t ) * cosf( 0.25f * t );             // sinc-lowpass and cos-window
		}

		// ensure that filter does not reach out off image bounds:
		while ( offset < 1 ) {
			h[0] += h[1];
			h[1] = h[2];
			h[2] = h[3];
			h[3] = 0.0f;
			offset++;
		}

		while ( offset+3 > src_size ) {
			h[3] += h[2];
			h[2] = h[1];
			h[1] = h[0];
			h[0] = 0.0f;
			offset--;
		}

		// coefficients are normalized to sum up to 2048
		const float  norm = 2048.0f / ( h[0] + h[1] + h[2] + h[3] );

		offset--;  // offset of fist used pixel

		filters[i].m_offset = offset + 4;  // store offset of first unused pixel

		for (int j=0; j<4; j++) {
			const float t = norm * h[j];
			filters[i].m_coeff[(offset+j) & 3] = (int) ((t > 0.0f) ?  (t+0.5f) : (t-0.5f));  // consider ring buffer index permutations
		}
	}

	// set end marker
	filters[dst_size].m_offset = (unsigned) -1;

}

void ImageScaler::SetImageParameters( unsigned *dst_image, unsigned dst_stride, unsigned dst_width, unsigned dst_height, unsigned src_width, unsigned src_height ) {
	m_src_x = 0;
	m_src_y = 0;
	m_dst_x = 0;
	m_dst_y = 0;

	m_dst_image  = dst_image;
	m_dst_stride = dst_stride;

	// if image dimensions do not change we can keep the old filter coefficients
	if ( (src_width == m_src_width) && (src_height == m_src_height) && (dst_width == m_dst_width) && (dst_height == m_dst_height) ) return;

	m_dst_width  = dst_width;
	m_dst_height = dst_height;
	m_src_width  = src_width;
	m_src_height = src_height;

	if ( m_memory ) free( m_memory );

	const unsigned  hor_filters_size = (m_dst_width  + 1) * sizeof(Filter);  // reserve one extra position for end marker
	const unsigned  ver_filters_size = (m_dst_height + 1) * sizeof(Filter);
	const unsigned  buffer_size      = 4 * m_dst_width * sizeof(TmpPixel);

	char *p = (char *) malloc( hor_filters_size + ver_filters_size + buffer_size );

	m_memory = p;

	m_hor_filters = (Filter   *) p;  p += hor_filters_size;
	m_ver_filters = (Filter   *) p;  p += ver_filters_size;
	m_buffer      = (TmpPixel *) p;

	CalculateFilters( m_hor_filters, m_dst_width , m_src_width  );
	CalculateFilters( m_ver_filters, m_dst_height, m_src_height );
}

// shift range to 0..255 and clamp overflows
static unsigned shift_clamp( int x ) {
	x = ( x + (1<<21) ) >> 22;
	if ( x <   0 ) return   0;
	if ( x > 255 ) return 255;
	return x;
}

void ImageScaler::NextSourceLine() {
	m_dst_x = 0;
	m_src_x = 0;
	m_src_y++;

	while ( m_ver_filters[m_dst_y].m_offset == m_src_y ) {
		const int        h0  = m_ver_filters[m_dst_y].m_coeff[0];
		const int        h1  = m_ver_filters[m_dst_y].m_coeff[1];
		const int        h2  = m_ver_filters[m_dst_y].m_coeff[2];
		const int        h3  = m_ver_filters[m_dst_y].m_coeff[3];
		const TmpPixel  *src = m_buffer;
		unsigned        *dst = m_dst_image + m_dst_stride * m_dst_y;

		for (unsigned i=0; i<m_dst_width; i++) {
			const ImageScaler::TmpPixel t( src[0]*h0 + src[1]*h1 + src[2]*h2 + src[3]*h3 );
			src += 4;
			dst[i] = shift_clamp(t[0]) | (shift_clamp(t[1])<<8) | (shift_clamp(t[2])<<16) | (shift_clamp(t[3])<<24);
		}

		m_dst_y++;
	}
}