ccv_resample.c

#include "ccv.h"
#include "ccv_internal.h"

/* area interpolation resample is adopted from OpenCV */

typedef struct {
	int si, di;
	unsigned int alpha;
} ccv_int_alpha;

static void _ccv_resample_area_8u(ccv_dense_matrix_t* a, ccv_dense_matrix_t* b, double rows_scale, double cols_scale)
{
	assert(a->cols > 0 && b->cols > 0);
	int ch = ccv_clamp(CCV_GET_CHANNEL(a->type), 1, 4);
	unsigned char* workspace = 0;
	ccv_int_alpha* xofs;
	unsigned int* buf;
	unsigned int* sum;
	if (a->cols > 1024 || b->cols * ch > 1024)
	{
		workspace = (unsigned char*)ccmalloc(sizeof(ccv_int_alpha) * a->cols * 2 + 2 * b->cols * ch * sizeof(unsigned int));
		xofs = (ccv_int_alpha*)workspace;
		buf = (unsigned int*)(workspace + sizeof(ccv_int_alpha) * a->cols * 2);
		sum = buf + b->cols * ch;
	} else {
		xofs = (ccv_int_alpha*)alloca(sizeof(ccv_int_alpha) * a->cols * 2);
		buf = (unsigned int*)alloca(b->cols * ch * sizeof(unsigned int));
		sum = (unsigned int*)alloca(b->cols * ch * sizeof(unsigned int));
	}
	double scale_x = (double)1 / cols_scale;
	double scale_y = (double)1 / rows_scale;
	// double scale = 1.f / (scale_x * scale_y);
	unsigned int inv_scale_256 = (int)(scale_x * scale_y * 0x10000);
	int dx, dy, sx, sy, i, k;
	for (dx = 0, k = 0; dx < b->cols; dx++)
	{
		double fsx1 = dx * scale_x, fsx2 = fsx1 + scale_x;
		int sx1 = (int)(fsx1 + 1.0 - 1e-6), sx2 = (int)(fsx2);

		if (sx1 > fsx1)
		{
			xofs[k].di = dx * ch;
			xofs[k].si = ccv_min(sx1 - 1, a->cols - 1) * ch;
			xofs[k++].alpha = (unsigned int)((sx1 - fsx1) * 0x100);
		}

		for (sx = sx1; sx < sx2; sx++)
		{
			xofs[k].di = dx * ch;
			xofs[k].si = ccv_min(sx, a->cols - 1) * ch;
			xofs[k++].alpha = 256;
		}

		if (fsx2 - sx2 > 1e-3)
		{
			xofs[k].di = dx * ch;
			xofs[k].si = ccv_min(sx2, a->cols - 1) * ch;
			xofs[k++].alpha = (unsigned int)((fsx2 - sx2) * 256);
		}
	}
	int xofs_count = k;
	for (dx = 0; dx < b->cols * ch; dx++)
		buf[dx] = sum[dx] = 0;
	dy = 0;
	int dy_weight_256 = 0;
	for (sy = 0; sy < a->rows; sy++)
	{
		unsigned char* a_ptr = a->data.u8 + a->step * sy;
		for (k = 0; k < xofs_count; k++)
		{
			int dxn = xofs[k].di;
			unsigned int alpha = xofs[k].alpha;
			for (i = 0; i < ch; i++)
				buf[dxn + i] += a_ptr[xofs[k].si + i] * alpha;
		}
		if ((dy + 1) * scale_y <= sy + 1)
		{
			unsigned int beta = (int)(ccv_max(sy + 1 - (dy + 1) * scale_y, 0.f) * 256);
			unsigned int beta1 = 256 - beta;
			unsigned char* b_ptr = b->data.u8 + b->step * dy;
			if (sy == a->rows - 1)
				beta = (int)(scale_y * 256);
			else
				dy_weight_256 = beta;
			if (beta <= 0)
			{
				for (dx = 0; dx < b->cols * ch; dx++)
				{
					b_ptr[dx] = ccv_clamp((sum[dx] + buf[dx] * 256) / inv_scale_256, 0, 255);
					sum[dx] = buf[dx] = 0;
				}
			} else {
				for (dx = 0; dx < b->cols * ch; dx++)
				{
					b_ptr[dx] = ccv_clamp((sum[dx] + buf[dx] * beta1) / inv_scale_256, 0, 255);
					sum[dx] = buf[dx] * beta;
					buf[dx] = 0;
				}
			}
			dy++;
		} else {
			if (sy == a->rows - 1)
			{
				dy_weight_256 = (int)(scale_y * 256) - dy_weight_256;
				for(dx = 0; dx < b->cols * ch; dx++)
				{
					sum[dx] += buf[dx] * dy_weight_256;
					buf[dx] = 0;
				}
			} else {
				dy_weight_256 += 256;
				for(dx = 0; dx < b->cols * ch; dx++)
				{
					sum[dx] += buf[dx] * 256;
					buf[dx] = 0;
				}
			}
		}
	}
	for (; dy < b->rows; dy++)
	{
		unsigned char* b_ptr = b->data.u8 + b->step * dy;
		for (dx = 0; dx < b->cols * ch; dx++)
			b_ptr[dx] = ccv_clamp(sum[dx] / inv_scale_256, 0, 255);
	}
	if (workspace)
		ccfree(workspace);
}

typedef struct {
	int si, di;
	float alpha;
} ccv_area_alpha_t;

static void _ccv_resample_area(ccv_dense_matrix_t* a, ccv_dense_matrix_t* b, double rows_scale, double cols_scale)
{
	assert(a->cols > 0 && b->cols > 0);
	int ch = CCV_GET_CHANNEL(a->type);
	unsigned char* workspace = 0;
	ccv_area_alpha_t* xofs;
	float* buf;
	float* sum;
	if (a->cols > 1024 || b->cols * ch > 1024)
	{
		workspace = (unsigned char*)ccmalloc(sizeof(ccv_area_alpha_t) * a->cols * 2 + 2 * b->cols * ch * sizeof(float));
		xofs = (ccv_area_alpha_t*)workspace;
		buf = (float*)(workspace + sizeof(ccv_area_alpha_t) * a->cols * 2);
		sum = buf + b->cols * ch;
	} else {
		xofs = (ccv_area_alpha_t*)alloca(sizeof(ccv_area_alpha_t) * a->cols * 2);
		buf = (float*)alloca(b->cols * ch * sizeof(float));
		sum = (float*)alloca(b->cols * ch * sizeof(float));
	}
	double scale_x = (double)1 / cols_scale;
	double scale_y = (double)1 / rows_scale;
	double scale = 1.f / (scale_x * scale_y);
	int dx, dy, sx, sy, i, k;
	for (dx = 0, k = 0; dx < b->cols; dx++)
	{
		double fsx1 = dx * scale_x, fsx2 = fsx1 + scale_x;
		int sx1 = (int)(fsx1 + 1.0 - 1e-6), sx2 = (int)(fsx2);

		if (sx1 > fsx1)
		{
			xofs[k].di = dx * ch;
			xofs[k].si = ccv_min(sx1 - 1, a->cols - 1) * ch;
			xofs[k++].alpha = (float)((sx1 - fsx1) * scale);
		}

		for (sx = sx1; sx < sx2; sx++)
		{
			xofs[k].di = dx * ch;
			xofs[k].si = ccv_min(sx, a->cols - 1) * ch;
			xofs[k++].alpha = (float)scale;
		}

		if (fsx2 - sx2 > 1e-3)
		{
			xofs[k].di = dx * ch;
			xofs[k].si = ccv_min(sx2, a->cols - 1) * ch;
			xofs[k++].alpha = (float)((fsx2 - sx2) * scale);
		}
	}
	int xofs_count = k;
	for (dx = 0; dx < b->cols * ch; dx++)
		buf[dx] = sum[dx] = 0;
	dy = 0;
	float dy_weight = 0;
#define for_block(_for_get, _for_set) \
	for (sy = 0; sy < a->rows; sy++) \
	{ \
		unsigned char* a_ptr = a->data.u8 + a->step * sy; \
		for (k = 0; k < xofs_count; k++) \
		{ \
			int dxn = xofs[k].di; \
			float alpha = xofs[k].alpha; \
			for (i = 0; i < ch; i++) \
				buf[dxn + i] += _for_get(a_ptr, xofs[k].si + i) * alpha; \
		} \
		if ((dy + 1) * scale_y <= sy + 1) \
		{ \
			float beta = ccv_max(sy + 1 - (dy + 1) * scale_y, 0.f); \
			float beta1 = 1 - beta; \
			unsigned char* b_ptr = b->data.u8 + b->step * dy; \
			if (sy == a->rows - 1) \
				beta = scale_y; /* Such that if there are any residue, we will scale it up. */ \
			else \
				dy_weight = beta; \
			if (fabs(beta) < 1e-3) \
			{ \
				for (dx = 0; dx < b->cols * ch; dx++) \
				{ \
					_for_set(b_ptr, dx, sum[dx] + buf[dx]); \
					sum[dx] = buf[dx] = 0; \
				} \
			} else { \
				for (dx = 0; dx < b->cols * ch; dx++) \
				{ \
					_for_set(b_ptr, dx, sum[dx] + buf[dx] * beta1); \
					sum[dx] = buf[dx] * beta; \
					buf[dx] = 0; \
				} \
			} \
			dy++; \
		} else { \
			if (sy == a->rows - 1) \
			{ \
				dy_weight = scale_y - dy_weight; \
				for(dx = 0; dx < b->cols * ch; dx++) \
				{ \
					sum[dx] += buf[dx] * dy_weight; \
					buf[dx] = 0; \
				} \
			} else { \
				dy_weight += 1; \
				for(dx = 0; dx < b->cols * ch; dx++) \
				{ \
					sum[dx] += buf[dx]; \
					buf[dx] = 0; \
				} \
			} \
		} \
	} \
	for (; dy < b->rows; dy++) \
	{ \
		unsigned char* b_ptr = b->data.u8 + b->step * dy; \
		for (dx = 0; dx < b->cols * ch; dx++) \
			_for_set(b_ptr, dx, sum[dx]); \
	}
	ccv_matrix_getter(a->type, ccv_matrix_setter, b->type, for_block);
#undef for_block
	if (workspace)
		ccfree(workspace);
}

typedef struct {
	int si[4];
	float coeffs[4];
} ccv_cubic_coeffs_t;

typedef struct {
	int si[4];
	int coeffs[4];
} ccv_cubic_integer_coeffs_t;

static void _ccv_init_cubic_coeffs(int si, int sz, float s, ccv_cubic_coeffs_t* coeff)
{
	const float A = -0.75f;
	coeff->si[0] = ccv_min(ccv_max(si - 1, 0), sz - 1);
	coeff->si[1] = ccv_min(ccv_max(si, 0), sz - 1);
	coeff->si[2] = ccv_min(ccv_max(si + 1, 0), sz - 1);
	coeff->si[3] = ccv_min(ccv_max(si + 2, 0), sz - 1);
	float x = s - si;
	coeff->coeffs[0] = ((A * (x + 1) - 5 * A) * (x + 1) + 8 * A) * (x + 1) - 4 * A;
	coeff->coeffs[1] = ((A + 2) * x - (A + 3)) * x * x + 1;
	coeff->coeffs[2] = ((A + 2) * (1 - x) - (A + 3)) * (1 - x) * (1 - x) + 1;
	coeff->coeffs[3] = 1.f - coeff->coeffs[0] - coeff->coeffs[1] - coeff->coeffs[2];
}

static void _ccv_resample_cubic_float_only(ccv_dense_matrix_t* a, ccv_dense_matrix_t* b, double rows_scale, double cols_scale)
{
	assert(CCV_GET_DATA_TYPE(b->type) == CCV_32F || CCV_GET_DATA_TYPE(b->type) == CCV_64F);
	int i, j, k, ch = CCV_GET_CHANNEL(a->type);
	assert(b->cols > 0 && b->step > 0);
	unsigned char* workspace = 0;
	ccv_cubic_coeffs_t* xofs;
	unsigned char* buf;
	if (b->step > 2048)
	{
		workspace = (unsigned char*)ccmalloc(sizeof(ccv_cubic_coeffs_t) * b->cols + b->step * 4);
		xofs = (ccv_cubic_coeffs_t*)workspace;
		buf = (unsigned char*)(workspace + sizeof(ccv_cubic_coeffs_t) * b->cols);
	} else {
		xofs = (ccv_cubic_coeffs_t*)alloca(sizeof(ccv_cubic_coeffs_t) * b->cols);
		buf = (unsigned char*)alloca(b->step * 4);
	}
	double scale_x = (double)1 / cols_scale;
	for (i = 0; i < b->cols; i++)
	{
		double sx = (i + 0.5) * scale_x - 0.5;
		_ccv_init_cubic_coeffs((int)sx, a->cols, (float)sx, xofs + i);
	}
	double scale_y = (double)1 / rows_scale;
#ifdef __clang_analyzer__
	memset(buf, 0, b->step * 4);
#endif
	unsigned char* a_ptr = a->data.u8;
	unsigned char* b_ptr = b->data.u8;
	int psi = -1, siy = 0;
#define for_block(_for_get, _for_set_b, _for_get_b) \
	for (i = 0; i < b->rows; i++) \
	{ \
		ccv_cubic_coeffs_t yofs; \
		double sy = (i + 0.5) * scale_y - 0.5; \
		_ccv_init_cubic_coeffs((int)sy, a->rows, (float)sy, &yofs); \
		if (yofs.si[3] > psi) \
		{ \
			for (; siy <= yofs.si[3]; siy++) \
			{ \
				unsigned char* row = buf + (siy & 0x3) * b->step; \
				for (j = 0; j < b->cols; j++) \
					for (k = 0; k < ch; k++) \
						_for_set_b(row, j * ch + k, _for_get(a_ptr, xofs[j].si[0] * ch + k) * xofs[j].coeffs[0] + \
													_for_get(a_ptr, xofs[j].si[1] * ch + k) * xofs[j].coeffs[1] + \
													_for_get(a_ptr, xofs[j].si[2] * ch + k) * xofs[j].coeffs[2] + \
													_for_get(a_ptr, xofs[j].si[3] * ch + k) * xofs[j].coeffs[3]); \
				a_ptr += a->step; \
			} \
			psi = yofs.si[3]; \
		} \
		unsigned char* row[4] = { \
			buf + (yofs.si[0] & 0x3) * b->step, \
			buf + (yofs.si[1] & 0x3) * b->step, \
			buf + (yofs.si[2] & 0x3) * b->step, \
			buf + (yofs.si[3] & 0x3) * b->step, \
		}; \
		for (j = 0; j < b->cols * ch; j++) \
			_for_set_b(b_ptr, j, _for_get_b(row[0], j, 0) * yofs.coeffs[0] + _for_get_b(row[1], j) * yofs.coeffs[1] + \
								 _for_get_b(row[2], j, 0) * yofs.coeffs[2] + _for_get_b(row[3], j) * yofs.coeffs[3]); \
		b_ptr += b->step; \
	}
	ccv_matrix_getter(a->type, ccv_matrix_setter_getter_float_only, b->type, for_block);
#undef for_block
	if (workspace)
		ccfree(workspace);
}

static void _ccv_init_cubic_integer_coeffs(int si, int sz, float s, ccv_cubic_integer_coeffs_t* coeff)
{
	const float A = -0.75f;
	coeff->si[0] = ccv_min(ccv_max(si - 1, 0), sz - 1);
	coeff->si[1] = ccv_min(ccv_max(si, 0), sz - 1);
	coeff->si[2] = ccv_min(ccv_max(si + 1, 0), sz - 1);
	coeff->si[3] = ccv_min(ccv_max(si + 2, 0), sz - 1);
	float x = s - si;
	const int W_BITS = 1 << 6;
	coeff->coeffs[0] = (int)((((A * (x + 1) - 5 * A) * (x + 1) + 8 * A) * (x + 1) - 4 * A) * W_BITS + 0.5);
	coeff->coeffs[1] = (int)((((A + 2) * x - (A + 3)) * x * x + 1) * W_BITS + 0.5);
	coeff->coeffs[2] = (int)((((A + 2) * (1 - x) - (A + 3)) * (1 - x) * (1 - x) + 1) * W_BITS + 0.5);
	coeff->coeffs[3] = W_BITS - coeff->coeffs[0] - coeff->coeffs[1] - coeff->coeffs[2];
}

static void _ccv_resample_cubic_integer_only(ccv_dense_matrix_t* a, ccv_dense_matrix_t* b, double rows_scale, double cols_scale)
{
	assert(CCV_GET_DATA_TYPE(b->type) == CCV_8U || CCV_GET_DATA_TYPE(b->type) == CCV_32S || CCV_GET_DATA_TYPE(b->type) == CCV_64S);
	int i, j, k, ch = CCV_GET_CHANNEL(a->type);
	int no_8u_type = (b->type & CCV_8U) ? CCV_32S : b->type;
	assert(b->cols > 0);
	unsigned char* workspace = 0;
	ccv_cubic_integer_coeffs_t* xofs;
	unsigned char* buf;
	int bufstep = b->cols * ch * CCV_GET_DATA_TYPE_SIZE(no_8u_type);
	if (b->step > 2048)
	{
		workspace = (unsigned char*)ccmalloc(sizeof(ccv_cubic_integer_coeffs_t) * b->cols + bufstep * 4);
		xofs = (ccv_cubic_integer_coeffs_t*)workspace;
		buf = (unsigned char*)(workspace + sizeof(ccv_cubic_integer_coeffs_t) * b->cols);
	} else {
		xofs = (ccv_cubic_integer_coeffs_t*)alloca(sizeof(ccv_cubic_integer_coeffs_t) * b->cols);
		buf = (unsigned char*)alloca(bufstep * 4);
	}
	double scale_x = (double)1 / cols_scale;
	for (i = 0; i < b->cols; i++)
	{
		double sx = (i + 0.5) * scale_x - 0.5;
		_ccv_init_cubic_integer_coeffs((int)sx, a->cols, (float)sx, xofs + i);
	}
	double scale_y = (double)1 / rows_scale;
#ifdef __clang_analyzer__
	memset(buf, 0, bufstep * 4);
#endif
	unsigned char* a_ptr = a->data.u8;
	unsigned char* b_ptr = b->data.u8;
	int psi = -1, siy = 0;
#define for_block(_for_get_a, _for_set, _for_get, _for_set_b) \
	for (i = 0; i < b->rows; i++) \
	{ \
		ccv_cubic_integer_coeffs_t yofs; \
		double sy = (i + 0.5) * scale_y - 0.5; \
		_ccv_init_cubic_integer_coeffs((int)sy, a->rows, (float)sy, &yofs); \
		if (yofs.si[3] > psi) \
		{ \
			for (; siy <= yofs.si[3]; siy++) \
			{ \
				unsigned char* row = buf + (siy & 0x3) * bufstep; \
				for (j = 0; j < b->cols; j++) \
					for (k = 0; k < ch; k++) \
						_for_set(row, j * ch + k, _for_get_a(a_ptr, xofs[j].si[0] * ch + k) * xofs[j].coeffs[0] + \
												  _for_get_a(a_ptr, xofs[j].si[1] * ch + k) * xofs[j].coeffs[1] + \
												  _for_get_a(a_ptr, xofs[j].si[2] * ch + k) * xofs[j].coeffs[2] + \
												  _for_get_a(a_ptr, xofs[j].si[3] * ch + k) * xofs[j].coeffs[3]); \
				a_ptr += a->step; \
			} \
			psi = yofs.si[3]; \
		} \
		unsigned char* row[4] = { \
			buf + (yofs.si[0] & 0x3) * bufstep, \
			buf + (yofs.si[1] & 0x3) * bufstep, \
			buf + (yofs.si[2] & 0x3) * bufstep, \
			buf + (yofs.si[3] & 0x3) * bufstep, \
		}; \
		for (j = 0; j < b->cols * ch; j++) \
			_for_set_b(b_ptr, j, ccv_descale(_for_get(row[0], j) * yofs.coeffs[0] + _for_get(row[1], j) * yofs.coeffs[1] + \
											 _for_get(row[2], j) * yofs.coeffs[2] + _for_get(row[3], j) * yofs.coeffs[3], 12)); \
		b_ptr += b->step; \
	}
	ccv_matrix_getter(a->type, ccv_matrix_setter_getter_integer_only, no_8u_type, ccv_matrix_setter_integer_only, b->type, for_block);
#undef for_block
	if (workspace)
		ccfree(workspace);
}

void ccv_resample(ccv_dense_matrix_t* a, ccv_dense_matrix_t** b, int btype, double rows_scale, double cols_scale, int type)
{
	assert(rows_scale > 0 && cols_scale > 0);
	ccv_declare_derived_signature(sig, a->sig != 0, ccv_sign_with_format(64, "ccv_resample(%f,%f,%d)", rows_scale, cols_scale, type), a->sig, CCV_EOF_SIGN);
	btype = (btype == 0) ? CCV_GET_DATA_TYPE(a->type) | CCV_GET_CHANNEL(a->type) : CCV_GET_DATA_TYPE(btype) | CCV_GET_CHANNEL(a->type);
	int rows, cols;
	if (*b)
	{
		rows = (*b)->rows;
		cols = (*b)->cols;
	} else {
		rows = (int)(a->rows * rows_scale + 0.5);
		cols = (int)(a->cols * cols_scale + 0.5);
	}
	ccv_dense_matrix_t* db = *b = ccv_dense_matrix_renew(*b, rows, cols, CCV_ALL_DATA_TYPE | CCV_GET_CHANNEL(a->type), btype, sig);
	ccv_object_return_if_cached(, db);
	if (a->rows == db->rows && a->cols == db->cols)
	{
		if (CCV_GET_CHANNEL(a->type) == CCV_GET_CHANNEL(db->type) && CCV_GET_DATA_TYPE(db->type) == CCV_GET_DATA_TYPE(a->type))
			memcpy(db->data.u8, a->data.u8, a->rows * a->step);
		else {
			ccv_shift(a, (ccv_matrix_t**)&db, 0, 0, 0);
		}
		return;
	}
	if ((type & CCV_INTER_AREA) && a->rows >= db->rows && a->cols >= db->cols)
	{
		/* using the fast alternative (fix point scale, 0x100 to avoid overflow) */
		if (CCV_GET_DATA_TYPE(a->type) == CCV_8U && CCV_GET_DATA_TYPE(db->type) == CCV_8U && a->rows * a->cols / (db->rows * db->cols) < 0x100)
			_ccv_resample_area_8u(a, db, rows_scale, cols_scale);
		else
			_ccv_resample_area(a, db, rows_scale, cols_scale);
	} else if (type & CCV_INTER_CUBIC) {
		if (CCV_GET_DATA_TYPE(db->type) == CCV_32F || CCV_GET_DATA_TYPE(db->type) == CCV_64F)
			_ccv_resample_cubic_float_only(a, db, rows_scale, cols_scale);
		else
			_ccv_resample_cubic_integer_only(a, db, rows_scale, cols_scale);
	} else if (type & CCV_INTER_LINEAR) {
		assert(0 && "CCV_INTER_LINEAR is not implemented");
	} else if (type & CCV_INTER_LANCZOS) {
		assert(0 && "CCV_INTER_LANCZOS is not implemented");
	} else {
		assert(0 && "Not implemented");
	}
}

/* the following code is adopted from OpenCV cvPyrDown */
void ccv_sample_down(ccv_dense_matrix_t* a, ccv_dense_matrix_t** b, int type, int src_x, int src_y)
{
	assert(src_x >= 0 && src_y >= 0);
	ccv_declare_derived_signature(sig, a->sig != 0, ccv_sign_with_format(64, "ccv_sample_down(%d,%d)", src_x, src_y), a->sig, CCV_EOF_SIGN);
	type = (type == 0) ? CCV_GET_DATA_TYPE(a->type) | CCV_GET_CHANNEL(a->type) : CCV_GET_DATA_TYPE(type) | CCV_GET_CHANNEL(a->type);
	ccv_dense_matrix_t* db = *b = ccv_dense_matrix_renew(*b, a->rows / 2, a->cols / 2, CCV_ALL_DATA_TYPE | CCV_GET_CHANNEL(a->type), type, sig);
	ccv_object_return_if_cached(, db);
	int ch = CCV_GET_CHANNEL(a->type);
	int cols0 = db->cols - 1 - src_x;
	int dy, sy = -2 + src_y, sx = src_x * ch, dx, k;
	unsigned char* workspace = 0;
	int* tab;
	unsigned char* buf;
	if (a->cols + src_x + 2 > 1024 || db->cols * ch > 1024)
	{
		workspace = (unsigned char*)ccmalloc((a->cols + src_x + 2) * ch * sizeof(int) + 5 * db->cols * ch * ccv_max(CCV_GET_DATA_TYPE_SIZE(db->type), sizeof(int)));
		tab = (int*)workspace;
		buf = (unsigned char*)(workspace + (a->cols + src_x + 2) * ch * sizeof(int));
	} else {
		tab = (int*)alloca((a->cols + src_x + 2) * ch * sizeof(int));
		buf = (unsigned char*)alloca(5 * db->cols * ch * ccv_max(CCV_GET_DATA_TYPE_SIZE(db->type), sizeof(int)));
	}
	for (dx = 0; dx < a->cols + src_x + 2; dx++)
		for (k = 0; k < ch; k++)
			tab[dx * ch + k] = ((dx >= a->cols) ? a->cols * 2 - 1 - dx : dx) * ch + k;
	int bufstep = db->cols * ch * ccv_max(CCV_GET_DATA_TYPE_SIZE(db->type), sizeof(int));
#ifdef __clang_analyzer__
	memset(buf, 0, 5 * bufstep);
#endif
	unsigned char* b_ptr = db->data.u8;
	/* why is src_y * 4 in computing the offset of row?
	 * Essentially, it means sy - src_y but in a manner that doesn't result negative number.
	 * notice that we added src_y before when computing sy in the first place, however,
	 * it is not desirable to have that offset when we try to wrap it into our 5-row buffer (
	 * because in later rearrangement, we have no src_y to backup the arrangement). In
	 * such micro scope, we managed to stripe 5 addition into one shift and addition. */
#define for_block(_for_get_a, _for_set, _for_get, _for_set_b) \
	for (dy = 0; dy < db->rows; dy++) \
	{ \
		for(; sy <= dy * 2 + 2 + src_y; sy++) \
		{ \
			unsigned char* row = buf + ((sy + src_y * 4 + 2) % 5) * bufstep; \
			int _sy = (sy < 0) ? -1 - sy : (sy >= a->rows) ? a->rows * 2 - 1 - sy : sy; \
			unsigned char* a_ptr = a->data.u8 + a->step * _sy; \
			for (k = 0; k < ch; k++) \
				_for_set(row, k, _for_get_a(a_ptr, sx + k) * 10 + _for_get_a(a_ptr, ch + sx + k) * 5 + _for_get_a(a_ptr, 2 * ch + sx + k)); \
			for(dx = ch; dx < cols0 * ch; dx += ch) \
				for (k = 0; k < ch; k++) \
					_for_set(row, dx + k, _for_get_a(a_ptr, dx * 2 + sx + k) * 6 + (_for_get_a(a_ptr, dx * 2 + sx + k - ch) + _for_get_a(a_ptr, dx * 2 + sx + k + ch)) * 4 + _for_get_a(a_ptr, dx * 2 + sx + k - ch * 2) + _for_get_a(a_ptr, dx * 2 + sx + k + ch * 2)); \
			x_block(_for_get_a, _for_set, _for_get, _for_set_b); \
		} \
		unsigned char* rows[5]; \
		for(k = 0; k < 5; k++) \
			rows[k] = buf + ((dy * 2 + k) % 5) * bufstep; \
		for(dx = 0; dx < db->cols * ch; dx++) \
			_for_set_b(b_ptr, dx, (_for_get(rows[2], dx) * 6 + (_for_get(rows[1], dx) + _for_get(rows[3], dx)) * 4 + _for_get(rows[0], dx) + _for_get(rows[4], dx)) / 256); \
		b_ptr += db->step; \
	}
	int no_8u_type = (a->type & CCV_8U) ? CCV_32S : a->type;
	if (src_x > 0)
	{
#define x_block(_for_get_a, _for_set, _for_get, _for_set_b) \
		for (dx = cols0 * ch; dx < db->cols * ch; dx += ch) \
			for (k = 0; k < ch; k++) \
				_for_set(row, dx + k, _for_get_a(a_ptr, tab[dx * 2 + sx + k]) * 6 + (_for_get_a(a_ptr, tab[dx * 2 + sx + k - ch]) + _for_get_a(a_ptr, tab[dx * 2 + sx + k + ch])) * 4 + _for_get_a(a_ptr, tab[dx * 2 + sx + k - ch * 2]) + _for_get_a(a_ptr, tab[dx * 2 + sx + k + ch * 2]));
		ccv_matrix_getter_a(a->type, ccv_matrix_setter_getter, no_8u_type, ccv_matrix_setter_b, db->type, for_block);
#undef x_block
	} else {
#define x_block(_for_get_a, _for_set, _for_get, _for_set_b) \
		for (k = 0; k < ch; k++) \
			_for_set(row, (db->cols - 1) * ch + k, _for_get_a(a_ptr, a->cols * ch + sx - ch + k) * 10 + _for_get_a(a_ptr, (a->cols - 2) * ch + sx + k) * 5 + _for_get_a(a_ptr, (a->cols - 3) * ch + sx + k));
		ccv_matrix_getter_a(a->type, ccv_matrix_setter_getter, no_8u_type, ccv_matrix_setter_b, db->type, for_block);
#undef x_block
	}
#undef for_block
	if (workspace)
		ccfree(workspace);
}

void ccv_sample_up(ccv_dense_matrix_t* a, ccv_dense_matrix_t** b, int type, int src_x, int src_y)
{
	assert(src_x >= 0 && src_y >= 0);
	ccv_declare_derived_signature(sig, a->sig != 0, ccv_sign_with_format(64, "ccv_sample_up(%d,%d)", src_x, src_y), a->sig, CCV_EOF_SIGN);
	type = (type == 0) ? CCV_GET_DATA_TYPE(a->type) | CCV_GET_CHANNEL(a->type) : CCV_GET_DATA_TYPE(type) | CCV_GET_CHANNEL(a->type);
	ccv_dense_matrix_t* db = *b = ccv_dense_matrix_renew(*b, a->rows * 2, a->cols * 2, CCV_ALL_DATA_TYPE | CCV_GET_CHANNEL(a->type), type, sig);
	ccv_object_return_if_cached(, db);
	int ch = CCV_GET_CHANNEL(a->type);
	int cols0 = a->cols - 1 - src_x;
	assert(a->cols > 0 && cols0 > 0);
	int y, x, sy = -1 + src_y, sx = src_x * ch, k;
	unsigned char* workspace = 0;
	int* tab;
	unsigned char* buf;
	if (a->cols + src_x + 2 > 1024 || db->cols * ch > 1024)
	{
		workspace = (unsigned char*)ccmalloc((a->cols + src_x + 2) * ch * sizeof(int) + 3 * db->cols * ch * ccv_max(CCV_GET_DATA_TYPE_SIZE(db->type), sizeof(int)));
		tab = (int*)workspace;
		buf = (unsigned char*)(workspace + (a->cols + src_x + 2) * ch * sizeof(int));
	} else {
		tab = (int*)alloca((a->cols + src_x + 2) * ch * sizeof(int));
		buf = (unsigned char*)alloca(3 * db->cols * ch * ccv_max(CCV_GET_DATA_TYPE_SIZE(db->type), sizeof(int)));
	}
	for (x = 0; x < a->cols + src_x + 2; x++)
		for (k = 0; k < ch; k++)
			tab[x * ch + k] = ((x >= a->cols) ? a->cols * 2 - 1 - x : x) * ch + k;
	int bufstep = db->cols * ch * ccv_max(CCV_GET_DATA_TYPE_SIZE(db->type), sizeof(int));
#ifdef __clang_analyzer__
	memset(buf, 0, 3 * bufstep);
#endif
	unsigned char* b_ptr = db->data.u8;
	/* why src_y * 2: the same argument as in ccv_sample_down */
#define for_block(_for_get_a, _for_set, _for_get, _for_set_b) \
	for (y = 0; y < a->rows; y++) \
	{ \
		for (; sy <= y + 1 + src_y; sy++) \
		{ \
			unsigned char* row = buf + ((sy + src_y * 2 + 1) % 3) * bufstep; \
			int _sy = (sy < 0) ? -1 - sy : (sy >= a->rows) ? a->rows * 2 - 1 - sy : sy; \
			unsigned char* a_ptr = a->data.u8 + a->step * _sy; \
			if (a->cols == 1) \
			{ \
				for (k = 0; k < ch; k++) \
				{ \
					_for_set(row, k, _for_get_a(a_ptr, k) * (G025 + G075 + G125)); \
					_for_set(row, k + ch, _for_get_a(a_ptr, k) * (G025 + G075 + G125)); \
				} \
				continue; \
			} \
			if (sx == 0) \
			{ \
				for (k = 0; k < ch; k++) \
				{ \
					_for_set(row, k, _for_get_a(a_ptr, k + sx) * (G025 + G075) + _for_get_a(a_ptr, k + sx + ch) * G125); \
					_for_set(row, k + ch, _for_get_a(a_ptr, k + sx) * (G125 + G025) + _for_get_a(a_ptr, k + sx + ch) * G075); \
				} \
			} \
			/* some serious flaw in computing Gaussian weighting in previous version
			 * specially, we are doing perfect upsampling (2x) so, it concerns a grid like:
			 * XXYY
			 * XXYY
			 * in this case, to upsampling, the weight should be from distance 0.25 and 1.25, and 0.25 and 0.75
			 * previously, it was mistakingly be 0.0 1.0, 0.5 0.5 (imperfect upsampling (2x - 1)) */ \
			for (x = (sx == 0) ? ch : 0; x < cols0 * ch; x += ch) \
			{ \
				for (k = 0; k < ch; k++) \
				{ \
					_for_set(row, x * 2 + k, _for_get_a(a_ptr, x + sx - ch + k) * G075 + _for_get_a(a_ptr, x + sx + k) * G025 + _for_get_a(a_ptr, x + sx + ch + k) * G125); \
					_for_set(row, x * 2 + ch + k, _for_get_a(a_ptr, x + sx - ch + k) * G125 + _for_get_a(a_ptr, x + sx + k) * G025 + _for_get_a(a_ptr, x + sx + ch + k) * G075); \
				} \
			} \
			x_block(_for_get_a, _for_set, _for_get, _for_set_b); \
		} \
		unsigned char* rows[3]; \
		for (k = 0; k < 3; k++) \
			rows[k] = buf + ((y + k) % 3) * bufstep; \
		for (x = 0; x < db->cols * ch; x++) \
		{ \
			_for_set_b(b_ptr, x, (_for_get(rows[0], x) * G075 + _for_get(rows[1], x) * G025 + _for_get(rows[2], x) * G125) / GALL); \
			_for_set_b(b_ptr + db->step, x, (_for_get(rows[0], x) * G125 + _for_get(rows[1], x) * G025 + _for_get(rows[2], x) * G075) / GALL); \
		} \
		b_ptr += 2 * db->step; \
	}
	int no_8u_type = (a->type & CCV_8U) ? CCV_32S : a->type;
	/* unswitch if condition in manual way */
	if ((a->type & CCV_8U) || (a->type & CCV_32S) || (a->type & CCV_64S))
	{
#define G025 (23)
#define G075 (8)
#define G125 (1)
#define GALL (1024)
		if (src_x > 0)
		{
#define x_block(_for_get_a, _for_set, _for_get, _for_set_b) \
			for (x = cols0 * ch; x < a->cols * ch; x += ch) \
				for (k = 0; k < ch; k++) \
				{ \
					_for_set(row, x * 2 + k, _for_get_a(a_ptr, tab[x + sx - ch + k]) * G075 + _for_get_a(a_ptr, tab[x + sx + k]) * G025 + _for_get_a(a_ptr, tab[x + sx + ch + k]) * G125); \
					_for_set(row, x * 2 + ch + k, _for_get_a(a_ptr, tab[x + sx - ch + k]) * G125 + _for_get_a(a_ptr, tab[x + sx + k]) * G025 + _for_get_a(a_ptr, tab[x + sx + ch + k], 0) * G075); \
			}
			ccv_matrix_getter_integer_only(a->type, ccv_matrix_setter_getter_integer_only, no_8u_type, ccv_matrix_setter_b, db->type, for_block);
#undef x_block
		} else {
#define x_block(_for_get_a, _for_set, _for_get, _for_set_b) \
			for (k = 0; k < ch; k++) \
			{ \
				_for_set(row, (a->cols - 1) * 2 * ch + k, _for_get_a(a_ptr, (a->cols - 2) * ch + k) * G075 + _for_get_a(a_ptr, (a->cols - 1) * ch + k) * (G025 + G125)); \
				_for_set(row, (a->cols - 1) * 2 * ch + ch + k, _for_get_a(a_ptr, (a->cols - 2) * ch + k) * G125 + _for_get_a(a_ptr, (a->cols - 1) * ch + k) * (G025 + G075)); \
			}
			ccv_matrix_getter_integer_only(a->type, ccv_matrix_setter_getter_integer_only, no_8u_type, ccv_matrix_setter_b, db->type, for_block);
#undef x_block
		}
#undef GALL
#undef G125
#undef G075
#undef G025
	} else {
#define G025 (0.705385)
#define G075 (0.259496)
#define G125 (0.035119)
#define GALL (1)
		if (src_x > 0)
		{
#define x_block(_for_get_a, _for_set, _for_get, _for_set_b) \
			for (x = cols0 * ch; x < a->cols * ch; x += ch) \
				for (k = 0; k < ch; k++) \
				{ \
					_for_set(row, x * 2 + k, _for_get_a(a_ptr, tab[x + sx - ch + k]) * G075 + _for_get_a(a_ptr, tab[x + sx + k]) * G025 + _for_get_a(a_ptr, tab[x + sx + ch + k]) * G125); \
					_for_set(row, x * 2 + ch + k, _for_get_a(a_ptr, tab[x + sx - ch + k]) * G125 + _for_get_a(a_ptr, tab[x + sx + k]) * G025 + _for_get_a(a_ptr, tab[x + sx + ch + k]) * G075); \
			}
			ccv_matrix_getter_float_only(a->type, ccv_matrix_setter_getter_float_only, no_8u_type, ccv_matrix_setter_b, db->type, for_block);
#undef x_block
		} else {
#define x_block(_for_get_a, _for_set, _for_get, _for_set_b) \
			for (k = 0; k < ch; k++) \
			{ \
				_for_set(row, (a->cols - 1) * 2 * ch + k, _for_get_a(a_ptr, (a->cols - 2) * ch + k) * G075 + _for_get_a(a_ptr, (a->cols - 1) * ch + k) * (G025 + G125)); \
				_for_set(row, (a->cols - 1) * 2 * ch + ch + k, _for_get_a(a_ptr, (a->cols - 2) * ch + k) * G125 + _for_get_a(a_ptr, (a->cols - 1) * ch + k) * (G025 + G075)); \
			}
			ccv_matrix_getter_float_only(a->type, ccv_matrix_setter_getter_float_only, no_8u_type, ccv_matrix_setter_b, db->type, for_block);
#undef x_block
		}
#undef GALL
#undef G125
#undef G075
#undef G025
	}
#undef for_block
	if (workspace)
		ccfree(workspace);
}