michael@0: #ifdef HAVE_CONFIG_H
michael@0: #include <config.h>
michael@0: #endif
michael@0: 
michael@0: #include <math.h>
michael@0: #include <string.h>
michael@0: 
michael@0: #include "pixman-private.h"
michael@0: 
michael@0: #include "pixman-combine32.h"
michael@0: 
michael@0: static force_inline uint32_t
michael@0: combine_mask (const uint32_t src, const uint32_t mask)
michael@0: {
michael@0:     uint32_t s, m;
michael@0: 
michael@0:     m = mask >> A_SHIFT;
michael@0: 
michael@0:     if (!m)
michael@0: 	return 0;
michael@0:     s = src;
michael@0: 
michael@0:     UN8x4_MUL_UN8 (s, m);
michael@0: 
michael@0:     return s;
michael@0: }
michael@0: 
michael@0: static void
michael@0: combine_src_u (pixman_implementation_t *imp,
michael@0:                pixman_op_t              op,
michael@0:                uint32_t *               dest,
michael@0:                const uint32_t *         src,
michael@0:                const uint32_t *         mask,
michael@0:                int                      width)
michael@0: {
michael@0:     int i;
michael@0: 
michael@0:     if (!mask)
michael@0: 	memcpy (dest, src, width * sizeof (uint16_t));
michael@0:     else
michael@0:     {
michael@0: 	uint16_t *d = (uint16_t*)dest;
michael@0: 	uint16_t *src16 = (uint16_t*)src;
michael@0: 	for (i = 0; i < width; ++i)
michael@0: 	{
michael@0: 	    if ((*mask & 0xff000000) == 0xff000000) {
michael@0: 		// it's likely worth special casing
michael@0: 		// fully opaque because it avoids
michael@0: 		// the cost of conversion as well the multiplication
michael@0: 		*(d + i) = *src16;
michael@0: 	    } else {
michael@0: 		// the mask is still 32bits
michael@0: 		uint32_t s = combine_mask (convert_0565_to_8888(*src16), *mask);
michael@0: 		*(d + i) = convert_8888_to_0565(s);
michael@0: 	    }
michael@0: 	    mask++;
michael@0: 	    src16++;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0: }
michael@0: 
michael@0: static void
michael@0: combine_over_u (pixman_implementation_t *imp,
michael@0:                pixman_op_t              op,
michael@0:                uint32_t *                dest,
michael@0:                const uint32_t *          src,
michael@0:                const uint32_t *          mask,
michael@0:                int                      width)
michael@0: {
michael@0:     int i;
michael@0: 
michael@0:     if (!mask)
michael@0: 	memcpy (dest, src, width * sizeof (uint16_t));
michael@0:     else
michael@0:     {
michael@0: 	uint16_t *d = (uint16_t*)dest;
michael@0: 	uint16_t *src16 = (uint16_t*)src;
michael@0: 	for (i = 0; i < width; ++i)
michael@0: 	{
michael@0: 	    if ((*mask & 0xff000000) == 0xff000000) {
michael@0: 		// it's likely worth special casing
michael@0: 		// fully opaque because it avoids
michael@0: 		// the cost of conversion as well the multiplication
michael@0: 		*(d + i) = *src16;
michael@0: 	    } else if ((*mask & 0xff000000) == 0x00000000) {
michael@0: 		// keep the dest the same
michael@0: 	    } else {
michael@0: 		// the mask is still 32bits
michael@0: 		uint32_t s = combine_mask (convert_0565_to_8888(*src16), *mask);
michael@0: 		uint32_t ia = ALPHA_8 (~s);
michael@0: 		uint32_t d32 = convert_0565_to_8888(*(d + i));
michael@0: 		UN8x4_MUL_UN8_ADD_UN8x4 (d32, ia, s);
michael@0: 		*(d + i) = convert_8888_to_0565(d32);
michael@0: 	    }
michael@0: 	    mask++;
michael@0: 	    src16++;
michael@0: 	}
michael@0:     }
michael@0: 
michael@0: }
michael@0: 
michael@0: 
michael@0: void
michael@0: _pixman_setup_combiner_functions_16 (pixman_implementation_t *imp)
michael@0: {
michael@0:     int i;
michael@0:     for (i = 0; i < PIXMAN_N_OPERATORS; i++) {
michael@0: 	imp->combine_16[i] = NULL;
michael@0:     }
michael@0:     imp->combine_16[PIXMAN_OP_SRC] = combine_src_u;
michael@0:     imp->combine_16[PIXMAN_OP_OVER] = combine_over_u;
michael@0: }
michael@0: