gpsp/video.c

4487 lines
215 KiB
C

/* gameplaySP
*
* Copyright (C) 2006 Exophase <exophase@gmail.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
* published by the Free Software Foundation; either version 2 of
* the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "common.h"
u16* gba_screen_pixels = NULL;
#define get_screen_pixels() gba_screen_pixels
#define get_screen_pitch() GBA_SCREEN_PITCH
static void render_scanline_conditional_tile(u32 start, u32 end, u16 *scanline,
u32 enable_flags, u32 dispcnt, u32 bldcnt, const tile_layer_render_struct
*layer_renderers);
static void render_scanline_conditional_bitmap(u32 start, u32 end, u16 *scanline,
u32 enable_flags, u32 dispcnt, u32 bldcnt, const bitmap_layer_render_struct
*layer_renderers);
#define no_op \
// This old version is not necessary if the palette is either being converted
// transparently or the ABGR 1555 format is being used natively. The direct
// version (without conversion) is much faster.
#define tile_lookup_palette_full(palette, source) \
current_pixel = palette[source]; \
convert_palette(current_pixel) \
#define tile_lookup_palette(palette, source) \
current_pixel = palette[source]; \
#define tile_expand_base_normal(index) \
tile_lookup_palette(palette, current_pixel); \
dest_ptr[index] = current_pixel \
#define tile_expand_transparent_normal(index) \
tile_expand_base_normal(index) \
#define tile_expand_copy(index) \
dest_ptr[index] = copy_ptr[index] \
#define advance_dest_ptr_base(delta) \
dest_ptr += delta \
#define advance_dest_ptr_transparent(delta) \
advance_dest_ptr_base(delta) \
#define advance_dest_ptr_copy(delta) \
advance_dest_ptr_base(delta); \
copy_ptr += delta \
#define color_combine_mask_a(layer) \
((io_registers[REG_BLDCNT] >> layer) & 0x01) \
// For color blending operations, will create a mask that has in bit
// 10 if the layer is target B, and bit 9 if the layer is target A.
#define color_combine_mask(layer) \
(color_combine_mask_a(layer) | \
((io_registers[REG_BLDCNT] >> (layer + 7)) & 0x02)) << 9 \
// For alpha blending renderers, draw the palette index (9bpp) and
// layer bits rather than the raw RGB. For the base this should write to
// the 32bit location directly.
#define tile_expand_base_alpha(index) \
dest_ptr[index] = current_pixel | pixel_combine \
#define tile_expand_base_bg(index) \
dest_ptr[index] = bg_combine \
// For layered (transparent) writes this should shift the "stack" and write
// to the bottom. This will preserve the topmost pixel and the most recent
// one.
#define tile_expand_transparent_alpha(index) \
dest_ptr[index] = (dest_ptr[index] << 16) | current_pixel | pixel_combine \
// OBJ should only shift if the top isn't already OBJ
#define tile_expand_transparent_alpha_obj(index) \
dest = dest_ptr[index]; \
if(dest & 0x00000100) \
dest_ptr[index] = (dest & 0xFFFF0000) | current_pixel | pixel_combine; \
else \
dest_ptr[index] = (dest << 16) | current_pixel | pixel_combine; \
// For color effects that don't need to preserve the previous layer.
// The color32 version should be used with 32bit wide dest_ptr so as to be
// compatible with alpha combine on top of it.
#define tile_expand_base_color16(index) \
dest_ptr[index] = current_pixel | pixel_combine \
#define tile_expand_transparent_color16(index) \
tile_expand_base_color16(index) \
#define tile_expand_base_color32(index) \
tile_expand_base_color16(index) \
#define tile_expand_transparent_color32(index) \
tile_expand_base_color16(index) \
// Operations for isolation 8bpp pixels within 32bpp pixel blocks.
#define tile_8bpp_pixel_op_mask(op_param) \
current_pixel = current_pixels & 0xFF \
#define tile_8bpp_pixel_op_shift_mask(shift) \
current_pixel = (current_pixels >> shift) & 0xFF \
#define tile_8bpp_pixel_op_shift(shift) \
current_pixel = current_pixels >> shift \
#define tile_8bpp_pixel_op_none(shift) \
// Base should always draw raw in 8bpp mode; color 0 will be drawn where
// color 0 is.
#define tile_8bpp_draw_base_normal(index) \
tile_expand_base_normal(index) \
#define tile_8bpp_draw_base_alpha(index) \
if(current_pixel) \
{ \
tile_expand_base_alpha(index); \
} \
else \
{ \
tile_expand_base_bg(index); \
} \
#define tile_8bpp_draw_base_color16(index) \
tile_8bpp_draw_base_alpha(index) \
#define tile_8bpp_draw_base_color32(index) \
tile_8bpp_draw_base_alpha(index) \
#define tile_8bpp_draw_base(index, op, op_param, alpha_op) \
tile_8bpp_pixel_op_##op(op_param); \
tile_8bpp_draw_base_##alpha_op(index) \
// Transparent (layered) writes should only replace what is there if the
// pixel is not transparent (zero)
#define tile_8bpp_draw_transparent(index, op, op_param, alpha_op) \
tile_8bpp_pixel_op_##op(op_param); \
if(current_pixel) \
{ \
tile_expand_transparent_##alpha_op(index); \
} \
#define tile_8bpp_draw_copy(index, op, op_param, alpha_op) \
tile_8bpp_pixel_op_##op(op_param); \
if(current_pixel) \
{ \
tile_expand_copy(index); \
} \
// Get the current tile from the map in 8bpp mode
#define get_tile_8bpp() \
current_tile = *map_ptr; \
tile_ptr = tile_base + ((current_tile & 0x3FF) * 64) \
// Draw half of a tile in 8bpp mode, for base renderer
#define tile_8bpp_draw_four_noflip(index, combine_op, alpha_op) \
tile_8bpp_draw_##combine_op(index + 0, mask, 0, alpha_op); \
tile_8bpp_draw_##combine_op(index + 1, shift_mask, 8, alpha_op); \
tile_8bpp_draw_##combine_op(index + 2, shift_mask, 16, alpha_op); \
tile_8bpp_draw_##combine_op(index + 3, shift, 24, alpha_op) \
// Like the above, but draws the half-tile horizontally flipped
#define tile_8bpp_draw_four_flip(index, combine_op, alpha_op) \
tile_8bpp_draw_##combine_op(index + 3, mask, 0, alpha_op); \
tile_8bpp_draw_##combine_op(index + 2, shift_mask, 8, alpha_op); \
tile_8bpp_draw_##combine_op(index + 1, shift_mask, 16, alpha_op); \
tile_8bpp_draw_##combine_op(index + 0, shift, 24, alpha_op) \
#define tile_8bpp_draw_four_base(index, alpha_op, flip_op) \
tile_8bpp_draw_four_##flip_op(index, base, alpha_op) \
// Draw half of a tile in 8bpp mode, for transparent renderer; as an
// optimization the entire thing is checked against zero (in transparent
// capable renders it is more likely for the pixels to be transparent than
// opaque)
#define tile_8bpp_draw_four_transparent(index, alpha_op, flip_op) \
if(current_pixels != 0) \
{ \
tile_8bpp_draw_four_##flip_op(index, transparent, alpha_op); \
} \
#define tile_8bpp_draw_four_copy(index, alpha_op, flip_op) \
if(current_pixels != 0) \
{ \
tile_8bpp_draw_four_##flip_op(index, copy, alpha_op); \
} \
// Helper macro for drawing 8bpp tiles clipped against the edge of the screen
#define partial_tile_8bpp(combine_op, alpha_op) \
for(i = 0; i < partial_tile_run; i++) \
{ \
tile_8bpp_draw_##combine_op(0, mask, 0, alpha_op); \
current_pixels >>= 8; \
advance_dest_ptr_##combine_op(1); \
} \
// Draws 8bpp tiles clipped against the left side of the screen,
// partial_tile_offset indicates how much clipped in it is, partial_tile_run
// indicates how much it should draw.
#define partial_tile_right_noflip_8bpp(combine_op, alpha_op) \
if(partial_tile_offset >= 4) \
{ \
current_pixels = *((u32 *)(tile_ptr + 4)) >> \
((partial_tile_offset - 4) * 8); \
partial_tile_8bpp(combine_op, alpha_op); \
} \
else \
{ \
partial_tile_run -= 4; \
current_pixels = *((u32 *)tile_ptr) >> (partial_tile_offset * 8); \
partial_tile_8bpp(combine_op, alpha_op); \
current_pixels = *((u32 *)(tile_ptr + 4)); \
tile_8bpp_draw_four_##combine_op(0, alpha_op, noflip); \
advance_dest_ptr_##combine_op(4); \
} \
// Draws 8bpp tiles clipped against both the left and right side of the
// screen, IE, runs of less than 8 - partial_tile_offset.
#define partial_tile_mid_noflip_8bpp(combine_op, alpha_op) \
if(partial_tile_offset >= 4) \
{ \
current_pixels = *((u32 *)(tile_ptr + 4)) >> \
((partial_tile_offset - 4) * 8); \
} \
else \
{ \
current_pixels = *((u32 *)tile_ptr) >> (partial_tile_offset * 8); \
if((partial_tile_offset + partial_tile_run) > 4) \
{ \
u32 old_run = partial_tile_run; \
partial_tile_run = 4 - partial_tile_offset; \
partial_tile_8bpp(combine_op, alpha_op); \
partial_tile_run = old_run - partial_tile_run; \
current_pixels = *((u32 *)(tile_ptr + 4)); \
} \
} \
partial_tile_8bpp(combine_op, alpha_op); \
// Draws 8bpp tiles clipped against the right side of the screen,
// partial_tile_run indicates how much there is to draw.
#define partial_tile_left_noflip_8bpp(combine_op, alpha_op) \
if(partial_tile_run >= 4) \
{ \
current_pixels = *((u32 *)tile_ptr); \
tile_8bpp_draw_four_##combine_op(0, alpha_op, noflip); \
advance_dest_ptr_##combine_op(4); \
tile_ptr += 4; \
partial_tile_run -= 4; \
} \
\
current_pixels = *((u32 *)(tile_ptr)); \
partial_tile_8bpp(combine_op, alpha_op) \
// Draws a non-clipped (complete) 8bpp tile.
#define tile_noflip_8bpp(combine_op, alpha_op) \
current_pixels = *((u32 *)tile_ptr); \
tile_8bpp_draw_four_##combine_op(0, alpha_op, noflip); \
current_pixels = *((u32 *)(tile_ptr + 4)); \
tile_8bpp_draw_four_##combine_op(4, alpha_op, noflip) \
// Like the above versions but draws flipped tiles.
#define partial_tile_flip_8bpp(combine_op, alpha_op) \
for(i = 0; i < partial_tile_run; i++) \
{ \
tile_8bpp_draw_##combine_op(0, shift, 24, alpha_op); \
current_pixels <<= 8; \
advance_dest_ptr_##combine_op(1); \
} \
#define partial_tile_right_flip_8bpp(combine_op, alpha_op) \
if(partial_tile_offset >= 4) \
{ \
current_pixels = *((u32 *)tile_ptr) << ((partial_tile_offset - 4) * 8); \
partial_tile_flip_8bpp(combine_op, alpha_op); \
} \
else \
{ \
partial_tile_run -= 4; \
current_pixels = *((u32 *)(tile_ptr + 4)) << \
((partial_tile_offset - 4) * 8); \
partial_tile_flip_8bpp(combine_op, alpha_op); \
current_pixels = *((u32 *)tile_ptr); \
tile_8bpp_draw_four_##combine_op(0, alpha_op, flip); \
advance_dest_ptr_##combine_op(4); \
} \
#define partial_tile_mid_flip_8bpp(combine_op, alpha_op) \
if(partial_tile_offset >= 4) \
current_pixels = *((u32 *)tile_ptr) << ((partial_tile_offset - 4) * 8); \
else \
{ \
current_pixels = *((u32 *)(tile_ptr + 4)) << \
((partial_tile_offset - 4) * 8); \
\
if((partial_tile_offset + partial_tile_run) > 4) \
{ \
u32 old_run = partial_tile_run; \
partial_tile_run = 4 - partial_tile_offset; \
partial_tile_flip_8bpp(combine_op, alpha_op); \
partial_tile_run = old_run - partial_tile_run; \
current_pixels = *((u32 *)(tile_ptr)); \
} \
} \
partial_tile_flip_8bpp(combine_op, alpha_op); \
#define partial_tile_left_flip_8bpp(combine_op, alpha_op) \
if(partial_tile_run >= 4) \
{ \
current_pixels = *((u32 *)(tile_ptr + 4)); \
tile_8bpp_draw_four_##combine_op(0, alpha_op, flip); \
advance_dest_ptr_##combine_op(4); \
tile_ptr -= 4; \
partial_tile_run -= 4; \
} \
\
current_pixels = *((u32 *)(tile_ptr + 4)); \
partial_tile_flip_8bpp(combine_op, alpha_op) \
#define tile_flip_8bpp(combine_op, alpha_op) \
current_pixels = *((u32 *)(tile_ptr + 4)); \
tile_8bpp_draw_four_##combine_op(0, alpha_op, flip); \
current_pixels = *((u32 *)tile_ptr); \
tile_8bpp_draw_four_##combine_op(4, alpha_op, flip) \
// Operations for isolating 4bpp tiles in a 32bit block
#define tile_4bpp_pixel_op_mask(op_param) \
current_pixel = current_pixels & 0x0F \
#define tile_4bpp_pixel_op_shift_mask(shift) \
current_pixel = (current_pixels >> shift) & 0x0F \
#define tile_4bpp_pixel_op_shift(shift) \
current_pixel = current_pixels >> shift \
#define tile_4bpp_pixel_op_none(op_param) \
// Draws a single 4bpp pixel as base, normal renderer; checks to see if the
// pixel is zero because if so the current palette should not be applied.
// These ifs can be replaced with a lookup table, may or may not be superior
// this way, should be benchmarked. The lookup table would be from 0-255
// identity map except for multiples of 16, which would map to 0.
#define tile_4bpp_draw_base_normal(index) \
if(current_pixel) \
current_pixel |= current_palette; \
tile_expand_base_normal(index); \
#define tile_4bpp_draw_base_alpha(index) \
if(current_pixel) \
{ \
current_pixel |= current_palette; \
tile_expand_base_alpha(index); \
} \
else \
{ \
tile_expand_base_bg(index); \
} \
#define tile_4bpp_draw_base_color16(index) \
tile_4bpp_draw_base_alpha(index) \
#define tile_4bpp_draw_base_color32(index) \
tile_4bpp_draw_base_alpha(index) \
#define tile_4bpp_draw_base(index, op, op_param, alpha_op) \
tile_4bpp_pixel_op_##op(op_param); \
tile_4bpp_draw_base_##alpha_op(index) \
// Draws a single 4bpp pixel as layered, if not transparent.
#define tile_4bpp_draw_transparent(index, op, op_param, alpha_op) \
tile_4bpp_pixel_op_##op(op_param); \
if(current_pixel) \
{ \
current_pixel |= current_palette; \
tile_expand_transparent_##alpha_op(index); \
} \
#define tile_4bpp_draw_copy(index, op, op_param, alpha_op) \
tile_4bpp_pixel_op_##op(op_param); \
if(current_pixel) \
{ \
current_pixel |= current_palette; \
tile_expand_copy(index); \
} \
// Draws eight background pixels in transparent mode, for alpha or normal
// renderers.
#define tile_4bpp_draw_eight_base_zero(value) \
dest_ptr[0] = value; \
dest_ptr[1] = value; \
dest_ptr[2] = value; \
dest_ptr[3] = value; \
dest_ptr[4] = value; \
dest_ptr[5] = value; \
dest_ptr[6] = value; \
dest_ptr[7] = value \
// Draws eight background pixels for the alpha renderer, basically color zero
// with the background flag high.
#define tile_4bpp_draw_eight_base_zero_alpha() \
tile_4bpp_draw_eight_base_zero(bg_combine) \
#define tile_4bpp_draw_eight_base_zero_color16() \
tile_4bpp_draw_eight_base_zero_alpha() \
#define tile_4bpp_draw_eight_base_zero_color32() \
tile_4bpp_draw_eight_base_zero_alpha() \
// Draws eight background pixels for the normal renderer, just a bunch of
// zeros.
#define tile_4bpp_draw_eight_base_zero_normal() \
current_pixel = palette[0]; \
tile_4bpp_draw_eight_base_zero(current_pixel) \
// Draws eight 4bpp pixels.
#define tile_4bpp_draw_eight_noflip(combine_op, alpha_op) \
tile_4bpp_draw_##combine_op(0, mask, 0, alpha_op); \
tile_4bpp_draw_##combine_op(1, shift_mask, 4, alpha_op); \
tile_4bpp_draw_##combine_op(2, shift_mask, 8, alpha_op); \
tile_4bpp_draw_##combine_op(3, shift_mask, 12, alpha_op); \
tile_4bpp_draw_##combine_op(4, shift_mask, 16, alpha_op); \
tile_4bpp_draw_##combine_op(5, shift_mask, 20, alpha_op); \
tile_4bpp_draw_##combine_op(6, shift_mask, 24, alpha_op); \
tile_4bpp_draw_##combine_op(7, shift, 28, alpha_op) \
// Draws eight 4bpp pixels in reverse order (for hflip).
#define tile_4bpp_draw_eight_flip(combine_op, alpha_op) \
tile_4bpp_draw_##combine_op(7, mask, 0, alpha_op); \
tile_4bpp_draw_##combine_op(6, shift_mask, 4, alpha_op); \
tile_4bpp_draw_##combine_op(5, shift_mask, 8, alpha_op); \
tile_4bpp_draw_##combine_op(4, shift_mask, 12, alpha_op); \
tile_4bpp_draw_##combine_op(3, shift_mask, 16, alpha_op); \
tile_4bpp_draw_##combine_op(2, shift_mask, 20, alpha_op); \
tile_4bpp_draw_##combine_op(1, shift_mask, 24, alpha_op); \
tile_4bpp_draw_##combine_op(0, shift, 28, alpha_op) \
// Draws eight 4bpp pixels in base mode, checks if all are zero, if so draws
// the appropriate background pixels.
#define tile_4bpp_draw_eight_base(alpha_op, flip_op) \
if(current_pixels != 0) \
{ \
tile_4bpp_draw_eight_##flip_op(base, alpha_op); \
} \
else \
{ \
tile_4bpp_draw_eight_base_zero_##alpha_op(); \
} \
// Draws eight 4bpp pixels in transparent (layered) mode, checks if all are
// zero and if so draws nothing.
#define tile_4bpp_draw_eight_transparent(alpha_op, flip_op) \
if(current_pixels != 0) \
{ \
tile_4bpp_draw_eight_##flip_op(transparent, alpha_op); \
} \
#define tile_4bpp_draw_eight_copy(alpha_op, flip_op) \
if(current_pixels != 0) \
{ \
tile_4bpp_draw_eight_##flip_op(copy, alpha_op); \
} \
// Gets the current tile in 4bpp mode, also getting the current palette and
// the pixel block.
#define get_tile_4bpp() \
current_tile = *map_ptr; \
current_palette = (current_tile >> 12) << 4; \
tile_ptr = tile_base + ((current_tile & 0x3FF) * 32); \
// Helper macro for drawing clipped 4bpp tiles.
#define partial_tile_4bpp(combine_op, alpha_op) \
for(i = 0; i < partial_tile_run; i++) \
{ \
tile_4bpp_draw_##combine_op(0, mask, 0, alpha_op); \
current_pixels >>= 4; \
advance_dest_ptr_##combine_op(1); \
} \
// Draws a 4bpp tile clipped against the left edge of the screen.
// partial_tile_offset is how far in it's clipped, partial_tile_run is
// how many to draw.
#define partial_tile_right_noflip_4bpp(combine_op, alpha_op) \
current_pixels = *((u32 *)tile_ptr) >> (partial_tile_offset * 4); \
partial_tile_4bpp(combine_op, alpha_op) \
// Draws a 4bpp tile clipped against both edges of the screen, same as right.
#define partial_tile_mid_noflip_4bpp(combine_op, alpha_op) \
partial_tile_right_noflip_4bpp(combine_op, alpha_op) \
// Draws a 4bpp tile clipped against the right edge of the screen.
// partial_tile_offset is how many to draw.
#define partial_tile_left_noflip_4bpp(combine_op, alpha_op) \
current_pixels = *((u32 *)tile_ptr); \
partial_tile_4bpp(combine_op, alpha_op) \
// Draws a complete 4bpp tile row (not clipped)
#define tile_noflip_4bpp(combine_op, alpha_op) \
current_pixels = *((u32 *)tile_ptr); \
tile_4bpp_draw_eight_##combine_op(alpha_op, noflip) \
// Like the above, but draws flipped tiles.
#define partial_tile_flip_4bpp(combine_op, alpha_op) \
for(i = 0; i < partial_tile_run; i++) \
{ \
tile_4bpp_draw_##combine_op(0, shift, 28, alpha_op); \
current_pixels <<= 4; \
advance_dest_ptr_##combine_op(1); \
} \
#define partial_tile_right_flip_4bpp(combine_op, alpha_op) \
current_pixels = *((u32 *)tile_ptr) << (partial_tile_offset * 4); \
partial_tile_flip_4bpp(combine_op, alpha_op) \
#define partial_tile_mid_flip_4bpp(combine_op, alpha_op) \
partial_tile_right_flip_4bpp(combine_op, alpha_op) \
#define partial_tile_left_flip_4bpp(combine_op, alpha_op) \
current_pixels = *((u32 *)tile_ptr); \
partial_tile_flip_4bpp(combine_op, alpha_op) \
#define tile_flip_4bpp(combine_op, alpha_op) \
current_pixels = *((u32 *)tile_ptr); \
tile_4bpp_draw_eight_##combine_op(alpha_op, flip) \
// Draws a single (partial or complete) tile from the tilemap, flipping
// as necessary.
#define single_tile_map(tile_type, combine_op, color_depth, alpha_op) \
get_tile_##color_depth(); \
if(current_tile & 0x800) \
tile_ptr += vertical_pixel_flip; \
\
if(current_tile & 0x400) \
{ \
tile_type##_flip_##color_depth(combine_op, alpha_op); \
} \
else \
{ \
tile_type##_noflip_##color_depth(combine_op, alpha_op); \
} \
#define single_tile_map_base_4bpp_color16(tile_type) \
get_tile_4bpp(); \
if(current_tile & 0x800) \
tile_ptr += vertical_pixel_flip; \
\
if(current_tile & 0x400) \
{ \
tile_type##_flip_4bpp(base, color16); \
} \
else \
{ \
tile_type##_noflip_4bpp(base, color16); \
} \
// Draws multiple sequential tiles from the tilemap, hflips and vflips as
// necessary.
#define multiple_tile_map(combine_op, color_depth, alpha_op) \
for(i = 0; i < tile_run; i++) \
{ \
single_tile_map(tile, combine_op, color_depth, alpha_op); \
advance_dest_ptr_##combine_op(8); \
map_ptr++; \
} \
#define multiple_tile_map_transparent_8bpp_color16() \
for(i = 0; i < tile_run; i++) \
{ \
single_tile_map(tile, transparent, 8bpp, color16); \
advance_dest_ptr_transparent(8); \
map_ptr++; \
} \
#define multiple_tile_map_transparent_4bpp_color16() \
for(i = 0; i < tile_run; i++) \
{ \
single_tile_map(tile, transparent, 4bpp, color16); \
advance_dest_ptr_transparent(8); \
map_ptr++; \
} \
#define multiple_tile_map_base_8bpp_color16() \
for(i = 0; i < tile_run; i++) \
{ \
single_tile_map(tile, base, 8bpp, color16); \
advance_dest_ptr_base(8); \
map_ptr++; \
} \
#define multiple_tile_map_base_4bpp_color16() \
for(i = 0; i < tile_run; i++) \
{ \
single_tile_map_base_4bpp_color16(tile); \
advance_dest_ptr_base(8); \
map_ptr++; \
} \
#define multiple_tile_map_transparent_8bpp_normal() \
for(i = 0; i < tile_run; i++) \
{ \
single_tile_map(tile, transparent, 8bpp, normal); \
advance_dest_ptr_transparent(8); \
map_ptr++; \
} \
#define multiple_tile_map_transparent_4bpp_normal() \
for(i = 0; i < tile_run; i++) \
{ \
single_tile_map(tile, transparent, 4bpp, normal); \
advance_dest_ptr_transparent(8); \
map_ptr++; \
} \
#define multiple_tile_map_base_8bpp_normal() \
for(i = 0; i < tile_run; i++) \
{ \
single_tile_map(tile, base, 8bpp, normal); \
advance_dest_ptr_base(8); \
map_ptr++; \
} \
#define multiple_tile_map_base_4bpp_normal() \
for(i = 0; i < tile_run; i++) \
{ \
single_tile_map(tile, base, 4bpp, normal); \
advance_dest_ptr_base(8); \
map_ptr++; \
} \
// Draws a partial tile from a tilemap clipped against the left edge of the
// screen.
#define partial_tile_right_map(combine_op, color_depth, alpha_op) \
single_tile_map(partial_tile_right, combine_op, color_depth, alpha_op); \
map_ptr++ \
// Draws a partial tile from a tilemap clipped against both edges of the
// screen.
#define partial_tile_mid_map(combine_op, color_depth, alpha_op) \
single_tile_map(partial_tile_mid, combine_op, color_depth, alpha_op) \
// Draws a partial tile from a tilemap clipped against the right edge of the
// screen.
#define partial_tile_left_map(combine_op, color_depth, alpha_op) \
single_tile_map(partial_tile_left, combine_op, color_depth, alpha_op) \
// Advances a non-flipped 4bpp obj to the next tile.
#define obj_advance_noflip_4bpp() \
tile_ptr += 32 \
// Advances a non-flipped 8bpp obj to the next tile.
#define obj_advance_noflip_8bpp() \
tile_ptr += 64 \
// Advances a flipped 4bpp obj to the next tile.
#define obj_advance_flip_4bpp() \
tile_ptr -= 32 \
// Advances a flipped 8bpp obj to the next tile.
#define obj_advance_flip_8bpp() \
tile_ptr -= 64 \
// Draws multiple sequential tiles from an obj, flip_op determines if it should
// be flipped or not (set to flip or noflip)
#define multiple_tile_obj(combine_op, color_depth, alpha_op, flip_op) \
for(i = 0; i < tile_run; i++) \
{ \
tile_##flip_op##_##color_depth(combine_op, alpha_op); \
obj_advance_##flip_op##_##color_depth(); \
advance_dest_ptr_##combine_op(8); \
} \
// Draws an obj's tile clipped against the left side of the screen
#define partial_tile_right_obj(combine_op, color_depth, alpha_op, flip_op) \
partial_tile_right_##flip_op##_##color_depth(combine_op, alpha_op); \
obj_advance_##flip_op##_##color_depth() \
// Draws an obj's tile clipped against both sides of the screen
#define partial_tile_mid_obj(combine_op, color_depth, alpha_op, flip_op) \
partial_tile_mid_##flip_op##_##color_depth(combine_op, alpha_op) \
// Draws an obj's tile clipped against the right side of the screen
#define partial_tile_left_obj(combine_op, color_depth, alpha_op, flip_op) \
partial_tile_left_##flip_op##_##color_depth(combine_op, alpha_op) \
// Extra variables specific for 8bpp/4bpp tile renderers.
#define tile_extra_variables_8bpp() \
#define tile_extra_variables_4bpp() \
u32 current_palette \
// Byte lengths of complete tiles and tile rows in 4bpp and 8bpp.
#define tile_width_4bpp 4
#define tile_size_4bpp 32
#define tile_width_8bpp 8
#define tile_size_8bpp 64
#define render_scanline_dest_normal u16
#define render_scanline_dest_alpha u32
#define render_scanline_dest_alpha_obj u32
#define render_scanline_dest_color16 u16
#define render_scanline_dest_color32 u32
#define render_scanline_dest_partial_alpha u32
#define render_scanline_dest_copy_tile u16
#define render_scanline_dest_copy_bitmap u16
// If rendering a scanline that is not a target A then there's no point in
// keeping what's underneath it because it can't blend with it.
#define render_scanline_skip_alpha(bg_type, combine_op) \
if((pixel_combine & 0x00000200) == 0) \
{ \
render_scanline_##bg_type##_##combine_op##_color32(layer, \
start, end, scanline); \
return; \
} \
#define render_scanline_extra_variables_base_normal(bg_type) \
u16 *palette = palette_ram_converted \
#define render_scanline_extra_variables_base_alpha(bg_type) \
u32 bg_combine = color_combine_mask(5); \
u32 pixel_combine = color_combine_mask(layer) | (bg_combine << 16); \
render_scanline_skip_alpha(bg_type, base) \
#define render_scanline_extra_variables_base_color() \
u32 bg_combine = color_combine_mask(5); \
u32 pixel_combine = color_combine_mask(layer) \
#define render_scanline_extra_variables_base_color16(bg_type) \
render_scanline_extra_variables_base_color() \
#define render_scanline_extra_variables_base_color32(bg_type) \
render_scanline_extra_variables_base_color() \
#define render_scanline_extra_variables_transparent_normal(bg_type) \
render_scanline_extra_variables_base_normal(bg_type) \
#define render_scanline_extra_variables_transparent_alpha(bg_type) \
u32 pixel_combine = color_combine_mask(layer); \
render_scanline_skip_alpha(bg_type, transparent) \
#define render_scanline_extra_variables_transparent_color() \
u32 pixel_combine = color_combine_mask(layer) \
#define render_scanline_extra_variables_transparent_color16(bg_type) \
render_scanline_extra_variables_transparent_color() \
#define render_scanline_extra_variables_transparent_color32(bg_type) \
render_scanline_extra_variables_transparent_color() \
static const u32 map_widths[] = { 256, 512, 256, 512 };
static void render_scanline_text_base_normal(u32 layer,
u32 start, u32 end, void *scanline)
{
render_scanline_extra_variables_base_normal(text);
u32 bg_control = io_registers[REG_BG0CNT + layer];
u32 map_size = (bg_control >> 14) & 0x03;
u32 map_width = map_widths[map_size];
u32 horizontal_offset =
(io_registers[REG_BG0HOFS + (layer * 2)] + start) % 512;
u32 vertical_offset = (io_registers[REG_VCOUNT] +
io_registers[REG_BG0VOFS + (layer * 2)]) % 512;
u32 current_pixel;
u32 current_pixels;
u32 partial_tile_run = 0;
u32 partial_tile_offset;
u32 tile_run;
u32 i;
render_scanline_dest_normal *dest_ptr =
((render_scanline_dest_normal *)scanline) + start;
u16 *map_base = (u16 *)(vram + ((bg_control >> 8) & 0x1F) * (1024 * 2));
u16 *map_ptr, *second_ptr;
u8 *tile_ptr;
end -= start;
if((map_size & 0x02) && (vertical_offset >= 256))
{
map_base += ((map_width / 8) * 32) +
(((vertical_offset - 256) / 8) * 32);
}
else
{
map_base += (((vertical_offset % 256) / 8) * 32);
}
if(map_size & 0x01)
{
if(horizontal_offset >= 256)
{
horizontal_offset -= 256;
map_ptr = map_base + (32 * 32) + (horizontal_offset / 8);
second_ptr = map_base;
}
else
{
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base + (32 * 32);
}
}
else
{
horizontal_offset %= 256;
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base;
}
if(bg_control & 0x80)
{
/* color depth: 8bpp
* combine: base
* alpha : normal
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_8bpp;
s32 vertical_pixel_flip =
((tile_size_8bpp - tile_width_8bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_8bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(base, 8bpp, normal);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(base, 8bpp, normal);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(base, 8bpp, normal);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map_base_8bpp_normal();
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map_base_8bpp_normal();
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(base, 8bpp, normal);
}
}
else
{
/* color depth: 4bpp
* combine: base
* alpha : normal
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_4bpp;
s32 vertical_pixel_flip =
((tile_size_4bpp - tile_width_4bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_4bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(base, 4bpp, normal);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(base, 4bpp, normal);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(base, 4bpp, normal);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map_base_4bpp_normal();
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map_base_4bpp_normal();
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(base, 4bpp, normal);
}
}
}
static void render_scanline_text_transparent_normal(u32 layer,
u32 start, u32 end, void *scanline)
{
render_scanline_extra_variables_transparent_normal(text);
u32 bg_control = io_registers[REG_BG0CNT + layer];
u32 map_size = (bg_control >> 14) & 0x03;
u32 map_width = map_widths[map_size];
u32 horizontal_offset =
(io_registers[REG_BG0HOFS + (layer * 2)] + start) % 512;
u32 vertical_offset = (io_registers[REG_VCOUNT] +
io_registers[REG_BG0VOFS + (layer * 2)]) % 512;
u32 current_pixel;
u32 current_pixels;
u32 partial_tile_run = 0;
u32 partial_tile_offset;
u32 tile_run;
u32 i;
render_scanline_dest_normal *dest_ptr =
((render_scanline_dest_normal *)scanline) + start;
u16 *map_base = (u16 *)(vram + ((bg_control >> 8) & 0x1F) * (1024 * 2));
u16 *map_ptr, *second_ptr;
u8 *tile_ptr;
end -= start;
if((map_size & 0x02) && (vertical_offset >= 256))
{
map_base += ((map_width / 8) * 32) +
(((vertical_offset - 256) / 8) * 32);
}
else
{
map_base += (((vertical_offset % 256) / 8) * 32);
}
if(map_size & 0x01)
{
if(horizontal_offset >= 256)
{
horizontal_offset -= 256;
map_ptr = map_base + (32 * 32) + (horizontal_offset / 8);
second_ptr = map_base;
}
else
{
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base + (32 * 32);
}
}
else
{
horizontal_offset %= 256;
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base;
}
if(bg_control & 0x80)
{
/* color depth: 8bpp
* combine: transparent
* alpha : normal
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_8bpp;
s32 vertical_pixel_flip =
((tile_size_8bpp - tile_width_8bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_8bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(transparent, 8bpp, normal);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(transparent, 8bpp, normal);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(transparent, 8bpp, normal);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map_transparent_8bpp_normal();
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map_transparent_8bpp_normal();
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(transparent, 8bpp, normal);
}
}
else
{
/* color depth: 4bpp
* combine: transparent
* alpha : normal
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_4bpp;
s32 vertical_pixel_flip =
((tile_size_4bpp - tile_width_4bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_4bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(transparent, 4bpp, normal);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(transparent, 4bpp, normal);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(transparent, 4bpp, normal);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map_transparent_4bpp_normal();
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map_transparent_4bpp_normal();
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(transparent, 4bpp, normal);
}
}
}
static void render_scanline_text_base_color16(u32 layer,
u32 start, u32 end, void *scanline)
{
render_scanline_extra_variables_base_color16(text);
u32 bg_control = io_registers[REG_BG0CNT + layer];
u32 map_size = (bg_control >> 14) & 0x03;
u32 map_width = map_widths[map_size];
u32 horizontal_offset =
(io_registers[REG_BG0HOFS + (layer * 2)] + start) % 512;
u32 vertical_offset = (io_registers[REG_VCOUNT] +
io_registers[REG_BG0VOFS + (layer * 2)]) % 512;
u32 current_pixel;
u32 current_pixels;
u32 partial_tile_run = 0;
u32 partial_tile_offset;
u32 tile_run;
u32 i;
render_scanline_dest_color16 *dest_ptr =
((render_scanline_dest_color16 *)scanline) + start;
u16 *map_base = (u16 *)(vram + ((bg_control >> 8) & 0x1F) * (1024 * 2));
u16 *map_ptr, *second_ptr;
u8 *tile_ptr;
end -= start;
if((map_size & 0x02) && (vertical_offset >= 256))
{
map_base += ((map_width / 8) * 32) +
(((vertical_offset - 256) / 8) * 32);
}
else
{
map_base += (((vertical_offset % 256) / 8) * 32);
}
if(map_size & 0x01)
{
if(horizontal_offset >= 256)
{
horizontal_offset -= 256;
map_ptr = map_base + (32 * 32) + (horizontal_offset / 8);
second_ptr = map_base;
}
else
{
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base + (32 * 32);
}
}
else
{
horizontal_offset %= 256;
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base;
}
if(bg_control & 0x80)
{
/* color depth: 8bpp
* combine: base
* alpha :color16
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_8bpp;
s32 vertical_pixel_flip =
((tile_size_8bpp - tile_width_8bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_8bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(base, 8bpp, color16);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(base, 8bpp, color16);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(base, 8bpp, color16);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map_base_8bpp_color16();
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map_base_8bpp_color16();
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(base, 8bpp, color16);
}
}
else
{
/* color depth: 4bpp
* combine: base
* alpha :color16
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_4bpp;
s32 vertical_pixel_flip =
((tile_size_4bpp - tile_width_4bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_4bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(base, 4bpp, color16);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(base, 4bpp, color16);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(base, 4bpp, color16);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map_base_4bpp_color16();
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map_base_4bpp_color16();
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(base, 4bpp, color16);
}
}
}
static void render_scanline_text_transparent_color16(u32 layer,
u32 start, u32 end, void *scanline)
{
render_scanline_extra_variables_transparent_color16(text);
u32 bg_control = io_registers[REG_BG0CNT + layer];
u32 map_size = (bg_control >> 14) & 0x03;
u32 map_width = map_widths[map_size];
u32 horizontal_offset =
(io_registers[REG_BG0HOFS + (layer * 2)] + start) % 512;
u32 vertical_offset = (io_registers[REG_VCOUNT] +
io_registers[REG_BG0VOFS + (layer * 2)]) % 512;
u32 current_pixel;
u32 current_pixels;
u32 partial_tile_run = 0;
u32 partial_tile_offset;
u32 tile_run;
u32 i;
render_scanline_dest_color16 *dest_ptr =
((render_scanline_dest_color16 *)scanline) + start;
u16 *map_base = (u16 *)(vram + ((bg_control >> 8) & 0x1F) * (1024 * 2));
u16 *map_ptr, *second_ptr;
u8 *tile_ptr;
end -= start;
if((map_size & 0x02) && (vertical_offset >= 256))
{
map_base += ((map_width / 8) * 32) +
(((vertical_offset - 256) / 8) * 32);
}
else
{
map_base += (((vertical_offset % 256) / 8) * 32);
}
if(map_size & 0x01)
{
if(horizontal_offset >= 256)
{
horizontal_offset -= 256;
map_ptr = map_base + (32 * 32) + (horizontal_offset / 8);
second_ptr = map_base;
}
else
{
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base + (32 * 32);
}
}
else
{
horizontal_offset %= 256;
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base;
}
if(bg_control & 0x80)
{
/* color depth: 8bpp
* combine: transparent
* alpha :color16
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_8bpp;
s32 vertical_pixel_flip =
((tile_size_8bpp - tile_width_8bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_8bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(transparent, 8bpp, color16);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(transparent, 8bpp, color16);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(transparent, 8bpp, color16);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map_transparent_8bpp_color16();
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map_transparent_8bpp_color16();
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(transparent, 8bpp, color16);
}
}
else
{
/* color depth: 4bpp
* combine: transparent
* alpha :color16
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_4bpp;
s32 vertical_pixel_flip =
((tile_size_4bpp - tile_width_4bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_4bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(transparent, 4bpp, color16);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(transparent, 4bpp, color16);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(transparent, 4bpp, color16);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map_transparent_4bpp_color16();
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map_transparent_4bpp_color16();
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(transparent, 4bpp, color16);
}
}
}
static void render_scanline_text_base_color32(u32 layer,
u32 start, u32 end, void *scanline)
{
render_scanline_extra_variables_base_color32(text);
u32 bg_control = io_registers[REG_BG0CNT + layer];
u32 map_size = (bg_control >> 14) & 0x03;
u32 map_width = map_widths[map_size];
u32 horizontal_offset =
(io_registers[REG_BG0HOFS + (layer * 2)] + start) % 512;
u32 vertical_offset = (io_registers[REG_VCOUNT] +
io_registers[REG_BG0VOFS + (layer * 2)]) % 512;
u32 current_pixel;
u32 current_pixels;
u32 partial_tile_run = 0;
u32 partial_tile_offset;
u32 tile_run;
u32 i;
render_scanline_dest_color32 *dest_ptr =
((render_scanline_dest_color32 *)scanline) + start;
u16 *map_base = (u16 *)(vram + ((bg_control >> 8) & 0x1F) * (1024 * 2));
u16 *map_ptr, *second_ptr;
u8 *tile_ptr;
end -= start;
if((map_size & 0x02) && (vertical_offset >= 256))
{
map_base += ((map_width / 8) * 32) +
(((vertical_offset - 256) / 8) * 32);
}
else
{
map_base += (((vertical_offset % 256) / 8) * 32);
}
if(map_size & 0x01)
{
if(horizontal_offset >= 256)
{
horizontal_offset -= 256;
map_ptr = map_base + (32 * 32) + (horizontal_offset / 8);
second_ptr = map_base;
}
else
{
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base + (32 * 32);
}
}
else
{
horizontal_offset %= 256;
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base;
}
if(bg_control & 0x80)
{
/* color depth: 8bpp
* combine: base
* alpha :color32
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_8bpp;
s32 vertical_pixel_flip =
((tile_size_8bpp - tile_width_8bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_8bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(base, 8bpp, color32);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(base, 8bpp, color32);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(base, 8bpp, color32);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map(base, 8bpp, color32);
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map(base, 8bpp, color32);
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(base, 8bpp, color32);
}
}
else
{
/* color depth: 4bpp
* combine: base
* alpha :color32
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_4bpp;
s32 vertical_pixel_flip =
((tile_size_4bpp - tile_width_4bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_4bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(base, 4bpp, color32);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(base, 4bpp, color32);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(base, 4bpp, color32);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map(base, 4bpp, color32);
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map(base, 4bpp, color32);
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(base, 4bpp, color32);
}
}
}
static void render_scanline_text_transparent_color32(u32 layer,
u32 start, u32 end, void *scanline)
{
render_scanline_extra_variables_transparent_color32(text);
u32 bg_control = io_registers[REG_BG0CNT + layer];
u32 map_size = (bg_control >> 14) & 0x03;
u32 map_width = map_widths[map_size];
u32 horizontal_offset =
(io_registers[REG_BG0HOFS + (layer * 2)] + start) % 512;
u32 vertical_offset = (io_registers[REG_VCOUNT] +
io_registers[REG_BG0VOFS + (layer * 2)]) % 512;
u32 current_pixel;
u32 current_pixels;
u32 partial_tile_run = 0;
u32 partial_tile_offset;
u32 tile_run;
u32 i;
render_scanline_dest_color32 *dest_ptr =
((render_scanline_dest_color32 *)scanline) + start;
u16 *map_base = (u16 *)(vram + ((bg_control >> 8) & 0x1F) * (1024 * 2));
u16 *map_ptr, *second_ptr;
u8 *tile_ptr;
end -= start;
if((map_size & 0x02) && (vertical_offset >= 256))
{
map_base += ((map_width / 8) * 32) +
(((vertical_offset - 256) / 8) * 32);
}
else
{
map_base += (((vertical_offset % 256) / 8) * 32);
}
if(map_size & 0x01)
{
if(horizontal_offset >= 256)
{
horizontal_offset -= 256;
map_ptr = map_base + (32 * 32) + (horizontal_offset / 8);
second_ptr = map_base;
}
else
{
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base + (32 * 32);
}
}
else
{
horizontal_offset %= 256;
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base;
}
if(bg_control & 0x80)
{
/* color depth: 8bpp
* combine: transparent
* alpha :color32
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_8bpp;
s32 vertical_pixel_flip =
((tile_size_8bpp - tile_width_8bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_8bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(transparent, 8bpp, color32);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(transparent, 8bpp, color32);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(transparent, 8bpp, color32);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map(transparent, 8bpp, color32);
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map(transparent, 8bpp, color32);
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(transparent, 8bpp, color32);
}
}
else
{
/* color depth: 4bpp
* combine: transparent
* alpha :color32
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_4bpp;
s32 vertical_pixel_flip =
((tile_size_4bpp - tile_width_4bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_4bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(transparent, 4bpp, color32);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(transparent, 4bpp, color32);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(transparent, 4bpp, color32);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map(transparent, 4bpp, color32);
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map(transparent, 4bpp, color32);
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(transparent, 4bpp, color32);
}
}
}
static void render_scanline_text_base_alpha(u32 layer,
u32 start, u32 end, void *scanline)
{
render_scanline_extra_variables_base_alpha(text);
u32 bg_control = io_registers[REG_BG0CNT + layer];
u32 map_size = (bg_control >> 14) & 0x03;
u32 map_width = map_widths[map_size];
u32 horizontal_offset =
(io_registers[REG_BG0HOFS + (layer * 2)] + start) % 512;
u32 vertical_offset = (io_registers[REG_VCOUNT] +
io_registers[REG_BG0VOFS + (layer * 2)]) % 512;
u32 current_pixel;
u32 current_pixels;
u32 partial_tile_run = 0;
u32 partial_tile_offset;
u32 tile_run;
u32 i;
render_scanline_dest_alpha *dest_ptr =
((render_scanline_dest_alpha *)scanline) + start;
u16 *map_base = (u16 *)(vram + ((bg_control >> 8) & 0x1F) * (1024 * 2));
u16 *map_ptr, *second_ptr;
u8 *tile_ptr;
end -= start;
if((map_size & 0x02) && (vertical_offset >= 256))
{
map_base += ((map_width / 8) * 32) +
(((vertical_offset - 256) / 8) * 32);
}
else
{
map_base += (((vertical_offset % 256) / 8) * 32);
}
if(map_size & 0x01)
{
if(horizontal_offset >= 256)
{
horizontal_offset -= 256;
map_ptr = map_base + (32 * 32) + (horizontal_offset / 8);
second_ptr = map_base;
}
else
{
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base + (32 * 32);
}
}
else
{
horizontal_offset %= 256;
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base;
}
if(bg_control & 0x80)
{
/* color depth: 8bpp
* combine: base
* alpha : alpha
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_8bpp;
s32 vertical_pixel_flip =
((tile_size_8bpp - tile_width_8bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_8bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(base, 8bpp, alpha);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(base, 8bpp, alpha);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(base, 8bpp, alpha);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map(base, 8bpp, alpha);
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map(base, 8bpp, alpha);
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(base, 8bpp, alpha);
}
}
else
{
/* color depth: 4bpp
* combine: base
* alpha : alpha
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_4bpp;
s32 vertical_pixel_flip =
((tile_size_4bpp - tile_width_4bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_4bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(base, 4bpp, alpha);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(base, 4bpp, alpha);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(base, 4bpp, alpha);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map(base, 4bpp, alpha);
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map(base, 4bpp, alpha);
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(base, 4bpp, alpha);
}
}
}
static void render_scanline_text_transparent_alpha(u32 layer,
u32 start, u32 end, void *scanline)
{
render_scanline_extra_variables_transparent_alpha(text);
u32 bg_control = io_registers[REG_BG0CNT + layer];
u32 map_size = (bg_control >> 14) & 0x03;
u32 map_width = map_widths[map_size];
u32 horizontal_offset =
(io_registers[REG_BG0HOFS + (layer * 2)] + start) % 512;
u32 vertical_offset = (io_registers[REG_VCOUNT] +
io_registers[REG_BG0VOFS + (layer * 2)]) % 512;
u32 current_pixel;
u32 current_pixels;
u32 partial_tile_run = 0;
u32 partial_tile_offset;
u32 tile_run;
u32 i;
render_scanline_dest_alpha *dest_ptr =
((render_scanline_dest_alpha *)scanline) + start;
u16 *map_base = (u16 *)(vram + ((bg_control >> 8) & 0x1F) * (1024 * 2));
u16 *map_ptr, *second_ptr;
u8 *tile_ptr;
end -= start;
if((map_size & 0x02) && (vertical_offset >= 256))
{
map_base += ((map_width / 8) * 32) +
(((vertical_offset - 256) / 8) * 32);
}
else
{
map_base += (((vertical_offset % 256) / 8) * 32);
}
if(map_size & 0x01)
{
if(horizontal_offset >= 256)
{
horizontal_offset -= 256;
map_ptr = map_base + (32 * 32) + (horizontal_offset / 8);
second_ptr = map_base;
}
else
{
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base + (32 * 32);
}
}
else
{
horizontal_offset %= 256;
map_ptr = map_base + (horizontal_offset / 8);
second_ptr = map_base;
}
if(bg_control & 0x80)
{
/* color depth: 8bpp
* combine: transparent
* alpha : alpha
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_8bpp;
s32 vertical_pixel_flip =
((tile_size_8bpp - tile_width_8bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_8bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(transparent, 8bpp, alpha);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(transparent, 8bpp, alpha);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(transparent, 8bpp, alpha);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map(transparent, 8bpp, alpha);
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map(transparent, 8bpp, alpha);
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(transparent, 8bpp, alpha);
}
}
else
{
/* color depth: 4bpp
* combine: transparent
* alpha : alpha
*/
/* Render a single scanline of text tiles */
u32 vertical_pixel_offset = (vertical_offset % 8) *
tile_width_4bpp;
s32 vertical_pixel_flip =
((tile_size_4bpp - tile_width_4bpp) -
vertical_pixel_offset) - vertical_pixel_offset;
tile_extra_variables_4bpp();
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)) +
vertical_pixel_offset;
u32 pixel_run = 256 - (horizontal_offset % 256);
u32 current_tile;
map_base += ((vertical_offset % 256) / 8) * 32;
partial_tile_offset = (horizontal_offset % 8);
if(pixel_run >= end)
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
if(end < partial_tile_run)
{
partial_tile_run = end;
partial_tile_mid_map(transparent, 4bpp, alpha);
return;
}
else
{
end -= partial_tile_run;
partial_tile_right_map(transparent, 4bpp, alpha);
}
}
}
else
{
if(partial_tile_offset)
{
partial_tile_run = 8 - partial_tile_offset;
partial_tile_right_map(transparent, 4bpp, alpha);
}
tile_run = (pixel_run - partial_tile_run) / 8;
multiple_tile_map(transparent, 4bpp, alpha);
map_ptr = second_ptr;
end -= pixel_run;
}
tile_run = end / 8;
multiple_tile_map(transparent, 4bpp, alpha);
partial_tile_run = end % 8;
if(partial_tile_run)
{
partial_tile_left_map(transparent, 4bpp, alpha);
}
}
}
s32 affine_reference_x[2];
s32 affine_reference_y[2];
#define affine_render_bg_pixel_normal() \
current_pixel = palette_ram_converted[0] \
#define affine_render_bg_pixel_alpha() \
current_pixel = bg_combine \
#define affine_render_bg_pixel_color16() \
affine_render_bg_pixel_alpha() \
#define affine_render_bg_pixel_color32() \
affine_render_bg_pixel_alpha() \
#define affine_render_bg_pixel_base(alpha_op) \
affine_render_bg_pixel_##alpha_op() \
#define affine_render_bg_pixel_transparent(alpha_op) \
#define affine_render_bg_pixel_copy(alpha_op) \
#define affine_render_bg_base(alpha_op) \
dest_ptr[0] = current_pixel
#define affine_render_bg_transparent(alpha_op) \
#define affine_render_bg_copy(alpha_op) \
#define affine_render_bg_remainder_base(alpha_op) \
affine_render_bg_pixel_##alpha_op(); \
for(; i < end; i++) \
{ \
affine_render_bg_base(alpha_op); \
advance_dest_ptr_base(1); \
} \
#define affine_render_bg_remainder_transparent(alpha_op) \
#define affine_render_bg_remainder_copy(alpha_op) \
#define affine_render_next(combine_op) \
source_x += dx; \
source_y += dy; \
advance_dest_ptr_##combine_op(1) \
#define affine_render_scale_offset() \
tile_base += ((pixel_y % 8) * 8); \
map_base += (pixel_y / 8) << map_pitch \
#define affine_render_scale_pixel(combine_op, alpha_op) \
map_offset = (pixel_x / 8); \
if(map_offset != last_map_offset) \
{ \
tile_ptr = tile_base + (map_base[map_offset] * 64); \
last_map_offset = map_offset; \
} \
tile_ptr = tile_base + (map_base[(pixel_x / 8)] * 64); \
current_pixel = tile_ptr[(pixel_x % 8)]; \
tile_8bpp_draw_##combine_op(0, none, 0, alpha_op); \
affine_render_next(combine_op) \
#define affine_render_scale(combine_op, alpha_op) \
{ \
pixel_y = source_y >> 8; \
u32 i = 0; \
affine_render_bg_pixel_##combine_op(alpha_op); \
if((u32)pixel_y < (u32)width_height) \
{ \
affine_render_scale_offset(); \
for(; i < end; i++) \
{ \
pixel_x = source_x >> 8; \
\
if((u32)pixel_x < (u32)width_height) \
{ \
break; \
} \
\
affine_render_bg_##combine_op(alpha_op); \
affine_render_next(combine_op); \
} \
\
for(; i < end; i++) \
{ \
pixel_x = source_x >> 8; \
\
if((u32)pixel_x >= (u32)width_height) \
break; \
\
affine_render_scale_pixel(combine_op, alpha_op); \
} \
} \
affine_render_bg_remainder_##combine_op(alpha_op); \
} \
#define affine_render_scale_wrap(combine_op, alpha_op) \
{ \
u32 wrap_mask = width_height - 1; \
pixel_y = (source_y >> 8) & wrap_mask; \
if((u32)pixel_y < (u32)width_height) \
{ \
affine_render_scale_offset(); \
for(i = 0; i < end; i++) \
{ \
pixel_x = (source_x >> 8) & wrap_mask; \
affine_render_scale_pixel(combine_op, alpha_op); \
} \
} \
} \
#define affine_render_rotate_pixel(combine_op, alpha_op) \
map_offset = (pixel_x / 8) + ((pixel_y / 8) << map_pitch); \
if(map_offset != last_map_offset) \
{ \
tile_ptr = tile_base + (map_base[map_offset] * 64); \
last_map_offset = map_offset; \
} \
\
current_pixel = tile_ptr[(pixel_x % 8) + ((pixel_y % 8) * 8)]; \
tile_8bpp_draw_##combine_op(0, none, 0, alpha_op); \
affine_render_next(combine_op) \
#define affine_render_rotate(combine_op, alpha_op) \
{ \
affine_render_bg_pixel_##combine_op(alpha_op); \
for(i = 0; i < end; i++) \
{ \
pixel_x = source_x >> 8; \
pixel_y = source_y >> 8; \
\
if(((u32)pixel_x < (u32)width_height) && \
((u32)pixel_y < (u32)width_height)) \
{ \
break; \
} \
affine_render_bg_##combine_op(alpha_op); \
affine_render_next(combine_op); \
} \
\
for(; i < end; i++) \
{ \
pixel_x = source_x >> 8; \
pixel_y = source_y >> 8; \
\
if(((u32)pixel_x >= (u32)width_height) || \
((u32)pixel_y >= (u32)width_height)) \
{ \
affine_render_bg_remainder_##combine_op(alpha_op); \
break; \
} \
\
affine_render_rotate_pixel(combine_op, alpha_op); \
} \
} \
#define affine_render_rotate_wrap(combine_op, alpha_op) \
{ \
u32 wrap_mask = width_height - 1; \
for(i = 0; i < end; i++) \
{ \
pixel_x = (source_x >> 8) & wrap_mask; \
pixel_y = (source_y >> 8) & wrap_mask; \
\
affine_render_rotate_pixel(combine_op, alpha_op); \
} \
} \
// Build affine background renderers.
#define render_scanline_affine_builder(combine_op, alpha_op) \
void render_scanline_affine_##combine_op##_##alpha_op(u32 layer, \
u32 start, u32 end, void *scanline) \
{ \
render_scanline_extra_variables_##combine_op##_##alpha_op(affine); \
u32 bg_control = io_registers[REG_BG0CNT + layer]; \
u32 current_pixel; \
s32 source_x, source_y; \
u32 pixel_x, pixel_y; \
u32 layer_offset = (layer - 2) * 8; \
s32 dx, dy; \
u32 map_size = (bg_control >> 14) & 0x03; \
u32 width_height = 1 << (7 + map_size); \
u32 map_pitch = map_size + 4; \
u8 *map_base = vram + (((bg_control >> 8) & 0x1F) * (1024 * 2)); \
u8 *tile_base = vram + (((bg_control >> 2) & 0x03) * (1024 * 16)); \
u8 *tile_ptr = NULL; \
u32 map_offset, last_map_offset = (u32)-1; \
u32 i; \
render_scanline_dest_##alpha_op *dest_ptr = \
((render_scanline_dest_##alpha_op *)scanline) + start; \
\
dx = (s16)io_registers[REG_BG2PA + layer_offset]; \
dy = (s16)io_registers[REG_BG2PC + layer_offset]; \
source_x = affine_reference_x[layer - 2] + (start * dx); \
source_y = affine_reference_y[layer - 2] + (start * dy); \
\
end -= start; \
\
switch(((bg_control >> 12) & 0x02) | (dy != 0)) \
{ \
case 0x00: \
affine_render_scale(combine_op, alpha_op); \
break; \
\
case 0x01: \
affine_render_rotate(combine_op, alpha_op); \
break; \
\
case 0x02: \
affine_render_scale_wrap(combine_op, alpha_op); \
break; \
\
case 0x03: \
affine_render_rotate_wrap(combine_op, alpha_op); \
break; \
} \
} \
render_scanline_affine_builder(base, normal);
render_scanline_affine_builder(transparent, normal);
render_scanline_affine_builder(base, color16);
render_scanline_affine_builder(transparent, color16);
render_scanline_affine_builder(base, color32);
render_scanline_affine_builder(transparent, color32);
render_scanline_affine_builder(base, alpha);
render_scanline_affine_builder(transparent, alpha);
#define bitmap_render_pixel_mode3(alpha_op) \
convert_palette(current_pixel); \
*dest_ptr = current_pixel \
#define bitmap_render_pixel_mode4(alpha_op) \
tile_expand_base_##alpha_op(0) \
#define bitmap_render_pixel_mode5(alpha_op) \
bitmap_render_pixel_mode3(alpha_op) \
#define bitmap_render_scale(type, alpha_op, width, height) \
pixel_y = (source_y >> 8); \
if((u32)pixel_y < (u32)height) \
{ \
pixel_x = (source_x >> 8); \
src_ptr += (pixel_y * width); \
if(dx == 0x100) \
{ \
if(pixel_x < 0) \
{ \
end += pixel_x; \
dest_ptr -= pixel_x; \
pixel_x = 0; \
} \
else if(pixel_x > 0) \
src_ptr += pixel_x; \
\
if((pixel_x + end) >= width) \
end = (width - pixel_x); \
\
for(i = 0; (s32)i < (s32)end; i++) \
{ \
current_pixel = *src_ptr; \
bitmap_render_pixel_##type(alpha_op); \
src_ptr++; \
dest_ptr++; \
} \
} \
else \
{ \
if((u32)(source_y >> 8) < (u32)height) \
{ \
for(i = 0; i < end; i++) \
{ \
pixel_x = (source_x >> 8); \
\
if((u32)pixel_x < (u32)width) \
break; \
\
source_x += dx; \
dest_ptr++; \
} \
\
for(; i < end; i++) \
{ \
pixel_x = (source_x >> 8); \
\
if((u32)pixel_x >= (u32)width) \
break; \
\
current_pixel = src_ptr[pixel_x]; \
bitmap_render_pixel_##type(alpha_op); \
\
source_x += dx; \
dest_ptr++; \
} \
} \
} \
} \
#define bitmap_render_rotate(type, alpha_op, width, height) \
for(i = 0; i < end; i++) \
{ \
pixel_x = source_x >> 8; \
pixel_y = source_y >> 8; \
\
if(((u32)pixel_x < (u32)width) && ((u32)pixel_y < (u32)height)) \
break; \
\
source_x += dx; \
source_y += dy; \
dest_ptr++; \
} \
\
for(; i < end; i++) \
{ \
pixel_x = (source_x >> 8); \
pixel_y = (source_y >> 8); \
\
if(((u32)pixel_x >= (u32)width) || ((u32)pixel_y >= (u32)height)) \
break; \
\
current_pixel = src_ptr[pixel_x + (pixel_y * width)]; \
bitmap_render_pixel_##type(alpha_op); \
\
source_x += dx; \
source_y += dy; \
dest_ptr++; \
} \
#define render_scanline_vram_setup_mode3() \
u16 *src_ptr = (u16 *)vram \
#define render_scanline_vram_setup_mode5() \
u16 *src_ptr = (u16 *)vram; \
if(io_registers[REG_DISPCNT] & 0x10) \
src_ptr = (u16 *)(vram + 0xA000); \
#define render_scanline_vram_setup_mode4() \
u16 *palette = palette_ram_converted; \
u8 *src_ptr = vram; \
if(io_registers[REG_DISPCNT] & 0x10) \
src_ptr = vram + 0xA000; \
// Build bitmap scanline rendering functions.
#define render_scanline_bitmap_builder(type, alpha_op, width, height) \
static void render_scanline_bitmap_##type##_##alpha_op(u32 start, u32 end, \
void *scanline) \
{ \
u32 current_pixel; \
s32 source_x, source_y; \
s32 pixel_x, pixel_y; \
\
s32 dx = (s16)io_registers[REG_BG2PA]; \
s32 dy = (s16)io_registers[REG_BG2PC]; \
\
u32 i; \
\
render_scanline_dest_##alpha_op *dest_ptr = \
((render_scanline_dest_##alpha_op *)scanline) + start; \
render_scanline_vram_setup_##type(); \
\
end -= start; \
\
source_x = affine_reference_x[0] + (start * dx); \
source_y = affine_reference_y[0] + (start * dy); \
\
if(dy == 0) \
{ \
bitmap_render_scale(type, alpha_op, width, height); \
} \
else \
{ \
bitmap_render_rotate(type, alpha_op, width, height); \
} \
} \
render_scanline_bitmap_builder(mode3, normal, 240, 160);
render_scanline_bitmap_builder(mode4, normal, 240, 160);
render_scanline_bitmap_builder(mode5, normal, 160, 128);
// Fill in the renderers for a layer based on the mode type,
#define tile_layer_render_functions(type) \
{ \
render_scanline_##type##_base_normal, \
render_scanline_##type##_transparent_normal, \
render_scanline_##type##_base_alpha, \
render_scanline_##type##_transparent_alpha, \
render_scanline_##type##_base_color16, \
render_scanline_##type##_transparent_color16, \
render_scanline_##type##_base_color32, \
render_scanline_##type##_transparent_color32 \
} \
// Use if a layer is unsupported for that mode.
#define tile_layer_render_null() \
{ \
NULL, NULL, NULL, NULL \
} \
#define bitmap_layer_render_functions(type) \
{ \
render_scanline_bitmap_##type##_normal \
} \
// Structs containing functions to render the layers for each mode, for
// each render type.
static const tile_layer_render_struct tile_mode_renderers[3][4] =
{
{
tile_layer_render_functions(text), tile_layer_render_functions(text),
tile_layer_render_functions(text), tile_layer_render_functions(text)
},
{
tile_layer_render_functions(text), tile_layer_render_functions(text),
tile_layer_render_functions(affine), tile_layer_render_functions(text)
},
{
tile_layer_render_functions(text), tile_layer_render_functions(text),
tile_layer_render_functions(affine), tile_layer_render_functions(affine)
}
};
static const bitmap_layer_render_struct bitmap_mode_renderers[3] =
{
bitmap_layer_render_functions(mode3),
bitmap_layer_render_functions(mode4),
bitmap_layer_render_functions(mode5)
};
#define render_scanline_layer_functions_tile() \
const tile_layer_render_struct *layer_renderers = \
tile_mode_renderers[dispcnt & 0x07] \
#define render_scanline_layer_functions_bitmap() \
const bitmap_layer_render_struct *layer_renderers = \
bitmap_mode_renderers + ((dispcnt & 0x07) - 3) \
// Adjust a flipped obj's starting position
#define obj_tile_offset_noflip(color_depth) \
#define obj_tile_offset_flip(color_depth) \
+ (tile_size_##color_depth * ((obj_width - 8) / 8)) \
// Adjust the obj's starting point if it goes too far off the left edge of
// the screen.
#define obj_tile_right_offset_noflip(color_depth) \
tile_ptr += (partial_tile_offset / 8) * tile_size_##color_depth \
#define obj_tile_right_offset_flip(color_depth) \
tile_ptr -= (partial_tile_offset / 8) * tile_size_##color_depth \
// Get the current row offset into an obj in 1D map space
#define obj_tile_offset_1D(color_depth, flip_op) \
tile_ptr = tile_base + ((obj_attribute_2 & 0x3FF) * 32) \
+ ((vertical_offset / 8) * (obj_width / 8) * tile_size_##color_depth) \
+ ((vertical_offset % 8) * tile_width_##color_depth) \
obj_tile_offset_##flip_op(color_depth) \
// Get the current row offset into an obj in 2D map space
#define obj_tile_offset_2D(color_depth, flip_op) \
tile_ptr = tile_base + ((obj_attribute_2 & 0x3FF) * 32) \
+ ((vertical_offset / 8) * 1024) \
+ ((vertical_offset % 8) * tile_width_##color_depth) \
obj_tile_offset_##flip_op(color_depth) \
// Get the palette for 4bpp obj.
#define obj_get_palette_4bpp() \
current_palette = (obj_attribute_2 >> 8) & 0xF0 \
#define obj_get_palette_8bpp() \
// Render the current row of an obj.
#define obj_render(combine_op, color_depth, alpha_op, map_space, flip_op) \
{ \
obj_get_palette_##color_depth(); \
obj_tile_offset_##map_space(color_depth, flip_op); \
\
if(obj_x < (s32)start) \
{ \
dest_ptr = scanline + start; \
pixel_run = obj_width - (start - obj_x); \
if((s32)pixel_run > 0) \
{ \
if((obj_x + obj_width) >= end) \
{ \
pixel_run = end - start; \
partial_tile_offset = start - obj_x; \
obj_tile_right_offset_##flip_op(color_depth); \
partial_tile_offset %= 8; \
\
if(partial_tile_offset) \
{ \
partial_tile_run = 8 - partial_tile_offset; \
if((s32)pixel_run < (s32)partial_tile_run) \
{ \
if((s32)pixel_run > 0) \
{ \
partial_tile_run = pixel_run; \
partial_tile_mid_obj(combine_op, color_depth, alpha_op, \
flip_op); \
} \
continue; \
} \
else \
{ \
pixel_run -= partial_tile_run; \
partial_tile_right_obj(combine_op, color_depth, alpha_op, \
flip_op); \
} \
} \
tile_run = pixel_run / 8; \
multiple_tile_obj(combine_op, color_depth, alpha_op, flip_op); \
partial_tile_run = pixel_run % 8; \
if(partial_tile_run) \
{ \
partial_tile_left_obj(combine_op, color_depth, alpha_op, \
flip_op); \
} \
} \
else \
{ \
partial_tile_offset = start - obj_x; \
obj_tile_right_offset_##flip_op(color_depth); \
partial_tile_offset %= 8; \
if(partial_tile_offset) \
{ \
partial_tile_run = 8 - partial_tile_offset; \
partial_tile_right_obj(combine_op, color_depth, alpha_op, \
flip_op); \
} \
tile_run = pixel_run / 8; \
multiple_tile_obj(combine_op, color_depth, alpha_op, flip_op); \
} \
} \
} \
else \
\
if((obj_x + obj_width) >= end) \
{ \
pixel_run = end - obj_x; \
if((s32)pixel_run > 0) \
{ \
dest_ptr = scanline + obj_x; \
tile_run = pixel_run / 8; \
multiple_tile_obj(combine_op, color_depth, alpha_op, flip_op); \
partial_tile_run = pixel_run % 8; \
if(partial_tile_run) \
{ \
partial_tile_left_obj(combine_op, color_depth, alpha_op, flip_op); \
} \
} \
} \
else \
{ \
dest_ptr = scanline + obj_x; \
tile_run = obj_width / 8; \
multiple_tile_obj(combine_op, color_depth, alpha_op, flip_op); \
} \
} \
#define obj_scale_offset_1D(color_depth) \
tile_ptr = tile_base + ((obj_attribute_2 & 0x3FF) * 32) \
+ ((vertical_offset / 8) * (max_x / 8) * tile_size_##color_depth) \
+ ((vertical_offset % 8) * tile_width_##color_depth) \
// Get the current row offset into an obj in 2D map space
#define obj_scale_offset_2D(color_depth) \
tile_ptr = tile_base + ((obj_attribute_2 & 0x3FF) * 32) \
+ ((vertical_offset / 8) * 1024) \
+ ((vertical_offset % 8) * tile_width_##color_depth) \
#define obj_render_scale_pixel_4bpp(combine_op, alpha_op) \
current_pixel = \
tile_ptr[tile_map_offset + ((tile_x >> 1) & 0x03)]; \
if(tile_x & 0x01) \
current_pixel >>= 4; \
else \
current_pixel &= 0x0F; \
\
tile_4bpp_draw_##combine_op(0, none, 0, alpha_op) \
#define obj_render_scale_pixel_8bpp(combine_op, alpha_op) \
current_pixel = tile_ptr[tile_map_offset + (tile_x & 0x07)]; \
tile_8bpp_draw_##combine_op(0, none, 0, alpha_op); \
#define obj_render_scale(combine_op, color_depth, alpha_op, map_space) \
{ \
u32 vertical_offset; \
source_y += (y_delta * dmy); \
vertical_offset = (source_y >> 8); \
if((u32)vertical_offset < (u32)max_y) \
{ \
obj_scale_offset_##map_space(color_depth); \
source_x += (y_delta * dmx) - (middle_x * dx); \
\
for(i = 0; i < obj_width; i++) \
{ \
tile_x = (source_x >> 8); \
\
if((u32)tile_x < (u32)max_x) \
break; \
\
source_x += dx; \
advance_dest_ptr_##combine_op(1); \
} \
\
for(; i < obj_width; i++) \
{ \
tile_x = (source_x >> 8); \
\
if((u32)tile_x >= (u32)max_x) \
break; \
\
tile_map_offset = (tile_x >> 3) * tile_size_##color_depth; \
obj_render_scale_pixel_##color_depth(combine_op, alpha_op); \
\
source_x += dx; \
advance_dest_ptr_##combine_op(1); \
} \
} \
} \
#define obj_rotate_offset_1D(color_depth) \
obj_tile_pitch = (max_x / 8) * tile_size_##color_depth \
#define obj_rotate_offset_2D(color_depth) \
obj_tile_pitch = 1024 \
#define obj_render_rotate_pixel_4bpp(combine_op, alpha_op) \
current_pixel = tile_ptr[tile_map_offset + \
((tile_x >> 1) & 0x03) + ((tile_y & 0x07) * obj_pitch)]; \
if(tile_x & 0x01) \
current_pixel >>= 4; \
else \
current_pixel &= 0x0F; \
\
tile_4bpp_draw_##combine_op(0, none, 0, alpha_op) \
#define obj_render_rotate_pixel_8bpp(combine_op, alpha_op) \
current_pixel = tile_ptr[tile_map_offset + \
(tile_x & 0x07) + ((tile_y & 0x07) * obj_pitch)]; \
\
tile_8bpp_draw_##combine_op(0, none, 0, alpha_op) \
#define obj_render_rotate(combine_op, color_depth, alpha_op, map_space) \
{ \
tile_ptr = tile_base + ((obj_attribute_2 & 0x3FF) * 32); \
obj_rotate_offset_##map_space(color_depth); \
\
source_x += (y_delta * dmx) - (middle_x * dx); \
source_y += (y_delta * dmy) - (middle_x * dy); \
\
for(i = 0; i < obj_width; i++) \
{ \
tile_x = (source_x >> 8); \
tile_y = (source_y >> 8); \
\
if(((u32)tile_x < (u32)max_x) && ((u32)tile_y < (u32)max_y)) \
break; \
\
source_x += dx; \
source_y += dy; \
advance_dest_ptr_##combine_op(1); \
} \
\
for(; i < obj_width; i++) \
{ \
tile_x = (source_x >> 8); \
tile_y = (source_y >> 8); \
\
if(((u32)tile_x >= (u32)max_x) || ((u32)tile_y >= (u32)max_y)) \
break; \
\
tile_map_offset = ((tile_x >> 3) * tile_size_##color_depth) + \
((tile_y >> 3) * obj_tile_pitch); \
obj_render_rotate_pixel_##color_depth(combine_op, alpha_op); \
\
source_x += dx; \
source_y += dy; \
advance_dest_ptr_##combine_op(1); \
} \
} \
// Render the current row of an affine transformed OBJ.
#define obj_render_affine(combine_op, color_depth, alpha_op, map_space) \
{ \
s16 *params = (s16 *)oam_ram + (((obj_attribute_1 >> 9) & 0x1F) * 16); \
s32 dx = params[3]; \
s32 dmx = params[7]; \
s32 dy = params[11]; \
s32 dmy = params[15]; \
s32 source_x, source_y; \
s32 tile_x, tile_y; \
u32 tile_map_offset; \
s32 middle_x; \
s32 middle_y; \
s32 max_x = obj_width; \
s32 max_y = obj_height; \
s32 y_delta; \
u32 obj_pitch = tile_width_##color_depth; \
u32 obj_tile_pitch; \
\
middle_x = (obj_width / 2); \
middle_y = (obj_height / 2); \
\
source_x = (middle_x << 8); \
source_y = (middle_y << 8); \
\
\
if(obj_attribute_0 & 0x200) \
{ \
obj_width *= 2; \
obj_height *= 2; \
middle_x *= 2; \
middle_y *= 2; \
} \
\
if((s32)obj_x < (s32)start) \
{ \
u32 x_delta = start - obj_x; \
middle_x -= x_delta; \
obj_width -= x_delta; \
obj_x = start; \
\
if((s32)obj_width <= 0) \
continue; \
} \
\
if((s32)(obj_x + obj_width) >= (s32)end) \
{ \
obj_width = end - obj_x; \
\
if((s32)obj_width <= 0) \
continue; \
} \
dest_ptr = scanline + obj_x; \
\
y_delta = vcount - (obj_y + middle_y); \
\
obj_get_palette_##color_depth(); \
\
if(dy == 0) \
{ \
obj_render_scale(combine_op, color_depth, alpha_op, map_space); \
} \
else \
{ \
obj_render_rotate(combine_op, color_depth, alpha_op, map_space); \
} \
} \
static const u32 obj_width_table[] =
{ 8, 16, 32, 64, 16, 32, 32, 64, 8, 8, 16, 32 };
static const u32 obj_height_table[] =
{ 8, 16, 32, 64, 8, 8, 16, 32, 16, 32, 32, 64 };
static u8 obj_priority_list[5][160][128];
static u32 obj_priority_count[5][160];
static u32 obj_alpha_count[160];
// Build obj rendering functions
#define render_scanline_obj_extra_variables_normal(bg_type) \
u16 *palette = palette_ram_converted + 256 \
#define render_scanline_obj_extra_variables_color() \
u32 pixel_combine = color_combine_mask(4) | (1 << 8) \
#define render_scanline_obj_extra_variables_alpha_obj(map_space) \
render_scanline_obj_extra_variables_color(); \
u32 dest; \
if((pixel_combine & 0x00000200) == 0) \
{ \
render_scanline_obj_color32_##map_space(priority, start, end, scanline); \
return; \
} \
#define render_scanline_obj_extra_variables_color16(map_space) \
render_scanline_obj_extra_variables_color() \
#define render_scanline_obj_extra_variables_color32(map_space) \
render_scanline_obj_extra_variables_color() \
#define render_scanline_obj_extra_variables_partial_alpha(map_space) \
render_scanline_obj_extra_variables_color(); \
u32 base_pixel_combine = pixel_combine; \
u32 dest \
#define render_scanline_obj_extra_variables_copy(type) \
u32 bldcnt = io_registers[REG_BLDCNT]; \
u32 dispcnt = io_registers[REG_DISPCNT]; \
u32 obj_enable = io_registers[REG_WINOUT] >> 8; \
render_scanline_layer_functions_##type(); \
u32 copy_start, copy_end; \
u16 copy_buffer[240]; \
u16 *copy_ptr \
#define render_scanline_obj_extra_variables_copy_tile(map_space) \
render_scanline_obj_extra_variables_copy(tile) \
#define render_scanline_obj_extra_variables_copy_bitmap(map_space) \
render_scanline_obj_extra_variables_copy(bitmap) \
#define render_scanline_obj_main(combine_op, alpha_op, map_space) \
if(obj_attribute_0 & 0x100) \
{ \
if((obj_attribute_0 >> 13) & 0x01) \
{ \
obj_render_affine(combine_op, 8bpp, alpha_op, map_space); \
} \
else \
{ \
obj_render_affine(combine_op, 4bpp, alpha_op, map_space); \
} \
} \
else \
{ \
vertical_offset = vcount - obj_y; \
\
if((obj_attribute_1 >> 13) & 0x01) \
vertical_offset = obj_height - vertical_offset - 1; \
\
switch(((obj_attribute_0 >> 12) & 0x02) | \
((obj_attribute_1 >> 12) & 0x01)) \
{ \
case 0x0: \
obj_render(combine_op, 4bpp, alpha_op, map_space, noflip); \
break; \
\
case 0x1: \
obj_render(combine_op, 4bpp, alpha_op, map_space, flip); \
break; \
\
case 0x2: \
obj_render(combine_op, 8bpp, alpha_op, map_space, noflip); \
break; \
\
case 0x3: \
obj_render(combine_op, 8bpp, alpha_op, map_space, flip); \
break; \
} \
} \
#define render_scanline_obj_no_partial_alpha(combine_op, alpha_op, map_space) \
render_scanline_obj_main(combine_op, alpha_op, map_space) \
#define render_scanline_obj_partial_alpha(combine_op, alpha_op, map_space) \
if((obj_attribute_0 >> 10) & 0x03) \
{ \
pixel_combine = 0x00000300; \
render_scanline_obj_main(combine_op, alpha_obj, map_space); \
} \
else \
{ \
pixel_combine = base_pixel_combine; \
render_scanline_obj_main(combine_op, color32, map_space); \
} \
#define render_scanline_obj_prologue_transparent(alpha_op) \
#define render_scanline_obj_prologue_copy_body(type) \
copy_start = obj_x; \
copy_end = obj_x + obj_width; \
if(obj_attribute_0 & 0x200) \
copy_end += obj_width; \
\
if(copy_start < start) \
copy_start = start; \
if(copy_end > end) \
copy_end = end; \
\
if((copy_start < end) && (copy_end > start)) \
{ \
render_scanline_conditional_##type(copy_start, copy_end, copy_buffer, \
obj_enable, dispcnt, bldcnt, layer_renderers); \
copy_ptr = copy_buffer + copy_start; \
} \
else \
{ \
continue; \
} \
#define render_scanline_obj_prologue_copy_tile() \
render_scanline_obj_prologue_copy_body(tile) \
#define render_scanline_obj_prologue_copy_bitmap() \
render_scanline_obj_prologue_copy_body(bitmap) \
#define render_scanline_obj_prologue_copy(alpha_op) \
render_scanline_obj_prologue_##alpha_op() \
#define render_scanline_obj_builder(combine_op, alpha_op, map_space, \
partial_alpha_op) \
static void render_scanline_obj_##alpha_op##_##map_space(u32 priority, \
u32 start, u32 end, render_scanline_dest_##alpha_op *scanline) \
{ \
render_scanline_obj_extra_variables_##alpha_op(map_space); \
s32 obj_num, i; \
s32 obj_x, obj_y; \
s32 obj_size; \
s32 obj_width, obj_height; \
u32 obj_attribute_0, obj_attribute_1, obj_attribute_2; \
s32 vcount = io_registers[REG_VCOUNT]; \
u32 tile_run; \
u32 current_pixels; \
u32 current_pixel; \
u32 current_palette; \
u32 vertical_offset; \
u32 partial_tile_run, partial_tile_offset; \
u32 pixel_run; \
u16 *oam_ptr; \
render_scanline_dest_##alpha_op *dest_ptr; \
u8 *tile_base = vram + 0x10000; \
u8 *tile_ptr; \
u32 obj_count = obj_priority_count[priority][vcount]; \
u8 *obj_list = obj_priority_list[priority][vcount]; \
\
for(obj_num = 0; obj_num < obj_count; obj_num++) \
{ \
oam_ptr = oam_ram + (obj_list[obj_num] * 4); \
obj_attribute_0 = oam_ptr[0]; \
obj_attribute_1 = oam_ptr[1]; \
obj_attribute_2 = oam_ptr[2]; \
obj_size = ((obj_attribute_0 >> 12) & 0x0C) | (obj_attribute_1 >> 14); \
\
obj_x = (s32)(obj_attribute_1 << 23) >> 23; \
obj_width = obj_width_table[obj_size]; \
\
render_scanline_obj_prologue_##combine_op(alpha_op); \
\
obj_y = obj_attribute_0 & 0xFF; \
\
if(obj_y > 160) \
obj_y -= 256; \
\
obj_height = obj_height_table[obj_size]; \
render_scanline_obj_##partial_alpha_op(combine_op, alpha_op, map_space); \
} \
} \
render_scanline_obj_builder(transparent, normal, 1D, no_partial_alpha);
render_scanline_obj_builder(transparent, normal, 2D, no_partial_alpha);
render_scanline_obj_builder(transparent, color16, 1D, no_partial_alpha);
render_scanline_obj_builder(transparent, color16, 2D, no_partial_alpha);
render_scanline_obj_builder(transparent, color32, 1D, no_partial_alpha);
render_scanline_obj_builder(transparent, color32, 2D, no_partial_alpha);
render_scanline_obj_builder(transparent, alpha_obj, 1D, no_partial_alpha);
render_scanline_obj_builder(transparent, alpha_obj, 2D, no_partial_alpha);
render_scanline_obj_builder(transparent, partial_alpha, 1D, partial_alpha);
render_scanline_obj_builder(transparent, partial_alpha, 2D, partial_alpha);
render_scanline_obj_builder(copy, copy_tile, 1D, no_partial_alpha);
render_scanline_obj_builder(copy, copy_tile, 2D, no_partial_alpha);
render_scanline_obj_builder(copy, copy_bitmap, 1D, no_partial_alpha);
render_scanline_obj_builder(copy, copy_bitmap, 2D, no_partial_alpha);
static void order_obj(u32 video_mode)
{
s32 obj_num, priority, row;
s32 obj_x, obj_y;
s32 obj_size, obj_mode;
s32 obj_width, obj_height;
u32 obj_priority;
u32 obj_attribute_0, obj_attribute_1, obj_attribute_2;
u32 current_count;
u16 *oam_ptr = oam_ram + 508;
for(priority = 0; priority < 5; priority++)
{
for(row = 0; row < 160; row++)
obj_priority_count[priority][row] = 0;
}
for(row = 0; row < 160; row++)
obj_alpha_count[row] = 0;
for(obj_num = 127; obj_num >= 0; obj_num--, oam_ptr -= 4)
{
obj_attribute_0 = oam_ptr[0];
obj_attribute_2 = oam_ptr[2];
obj_size = obj_attribute_0 & 0xC000;
obj_priority = (obj_attribute_2 >> 10) & 0x03;
obj_mode = (obj_attribute_0 >> 10) & 0x03;
if(((obj_attribute_0 & 0x0300) != 0x0200) && (obj_size != 0xC000) &&
(obj_mode != 3) && ((video_mode < 3) ||
((obj_attribute_2 & 0x3FF) >= 512)))
{
obj_y = obj_attribute_0 & 0xFF;
if(obj_y > 160)
obj_y -= 256;
obj_attribute_1 = oam_ptr[1];
obj_size = ((obj_size >> 12) & 0x0C) | (obj_attribute_1 >> 14);
obj_height = obj_height_table[obj_size];
obj_width = obj_width_table[obj_size];
if(obj_attribute_0 & 0x200)
{
obj_height *= 2;
obj_width *= 2;
}
if(((obj_y + obj_height) > 0) && (obj_y < 160))
{
obj_x = (s32)(obj_attribute_1 << 23) >> 23;
if(((obj_x + obj_width) > 0) && (obj_x < 240))
{
if(obj_y < 0)
{
obj_height += obj_y;
obj_y = 0;
}
if((obj_y + obj_height) >= 160)
obj_height = 160 - obj_y;
if(obj_mode == 1)
{
for(row = obj_y; row < obj_y + obj_height; row++)
{
current_count = obj_priority_count[obj_priority][row];
obj_priority_list[obj_priority][row][current_count] = obj_num;
obj_priority_count[obj_priority][row] = current_count + 1;
obj_alpha_count[row]++;
}
}
else
{
if(obj_mode == 2)
obj_priority = 4;
for(row = obj_y; row < obj_y + obj_height; row++)
{
current_count = obj_priority_count[obj_priority][row];
obj_priority_list[obj_priority][row][current_count] = obj_num;
obj_priority_count[obj_priority][row] = current_count + 1;
}
}
}
}
}
}
}
u32 layer_order[16];
u32 layer_count;
static void order_layers(u32 layer_flags)
{
s32 priority, layer_number;
layer_count = 0;
for(priority = 3; priority >= 0; priority--)
{
for(layer_number = 3; layer_number >= 0; layer_number--)
{
if(((layer_flags >> layer_number) & 1) &&
((io_registers[REG_BG0CNT + layer_number] & 0x03) == priority))
{
layer_order[layer_count] = layer_number;
layer_count++;
}
}
if((obj_priority_count[priority][io_registers[REG_VCOUNT]] > 0)
&& (layer_flags & 0x10))
{
layer_order[layer_count] = priority | 0x04;
layer_count++;
}
}
}
#define fill_line(_start, _end) \
u32 i; \
\
for(i = _start; i < _end; i++) \
dest_ptr[i] = color; \
#define fill_line_color_normal() \
color = palette_ram_converted[color] \
#define fill_line_color_alpha() \
#define fill_line_color_color16() \
#define fill_line_color_color32() \
#define fill_line_builder(type) \
static void fill_line_##type(u16 color, render_scanline_dest_##type *dest_ptr,\
u32 start, u32 end) \
{ \
fill_line_color_##type(); \
fill_line(start, end); \
} \
fill_line_builder(normal);
fill_line_builder(alpha);
fill_line_builder(color16);
fill_line_builder(color32);
// Alpha blend two pixels (pixel_top and pixel_bottom).
#define blend_pixel() \
pixel_bottom = palette_ram_converted[(pixel_pair >> 16) & 0x1FF]; \
pixel_bottom = (pixel_bottom | (pixel_bottom << 16)) & 0x07E0F81F; \
pixel_top = ((pixel_top * blend_a) + (pixel_bottom * blend_b)) >> 4 \
// Alpha blend two pixels, allowing for saturation (individual channels > 31).
// The operation is optimized towards saturation not occuring.
#define blend_saturate_pixel() \
pixel_bottom = palette_ram_converted[(pixel_pair >> 16) & 0x1FF]; \
pixel_bottom = (pixel_bottom | (pixel_bottom << 16)) & 0x07E0F81F; \
pixel_top = ((pixel_top * blend_a) + (pixel_bottom * blend_b)) >> 4; \
if(pixel_top & 0x08010020) \
{ \
if(pixel_top & 0x08000000) \
pixel_top |= 0x07E00000; \
\
if(pixel_top & 0x00010000) \
pixel_top |= 0x0000F800; \
\
if(pixel_top & 0x00000020) \
pixel_top |= 0x0000001F; \
} \
#define brighten_pixel() \
pixel_top = upper + ((pixel_top * blend) >> 4); \
#define darken_pixel() \
pixel_top = (pixel_top * blend) >> 4; \
#define effect_condition_alpha \
((pixel_pair & 0x04000200) == 0x04000200) \
#define effect_condition_fade(pixel_source) \
((pixel_source & 0x00000200) == 0x00000200) \
#define expand_pixel_no_dest(expand_type, pixel_source) \
pixel_top = (pixel_top | (pixel_top << 16)) & 0x07E0F81F; \
expand_type##_pixel(); \
pixel_top &= 0x07E0F81F; \
pixel_top = (pixel_top >> 16) | pixel_top \
#define expand_pixel(expand_type, pixel_source) \
pixel_top = palette_ram_converted[pixel_source & 0x1FF]; \
expand_pixel_no_dest(expand_type, pixel_source); \
*screen_dest_ptr = pixel_top \
#define expand_loop(expand_type, effect_condition, pixel_source) \
screen_src_ptr += start; \
screen_dest_ptr += start; \
\
end -= start; \
\
for(i = 0; i < end; i++) \
{ \
pixel_source = *screen_src_ptr; \
if(effect_condition) \
{ \
expand_pixel(expand_type, pixel_source); \
} \
else \
{ \
*screen_dest_ptr = \
palette_ram_converted[pixel_source & 0x1FF]; \
} \
\
screen_src_ptr++; \
screen_dest_ptr++; \
} \
#define expand_loop_partial_alpha(alpha_expand, expand_type) \
screen_src_ptr += start; \
screen_dest_ptr += start; \
\
end -= start; \
\
for(i = 0; i < end; i++) \
{ \
pixel_pair = *screen_src_ptr; \
if(effect_condition_fade(pixel_pair)) \
{ \
if(effect_condition_alpha) \
{ \
expand_pixel(alpha_expand, pixel_pair); \
} \
else \
{ \
expand_pixel(expand_type, pixel_pair); \
} \
} \
else \
{ \
*screen_dest_ptr = \
palette_ram_converted[pixel_pair & 0x1FF]; \
} \
\
screen_src_ptr++; \
screen_dest_ptr++; \
} \
#define expand_partial_alpha(expand_type) \
if((blend_a + blend_b) > 16) \
{ \
expand_loop_partial_alpha(blend_saturate, expand_type); \
} \
else \
{ \
expand_loop_partial_alpha(blend, expand_type); \
} \
// Blend top two pixels of scanline with each other.
#define expand_normal(screen_ptr, start, end)
void expand_blend(u32 *screen_src_ptr, u16 *screen_dest_ptr,
u32 start, u32 end);
#ifndef ARM_ARCH_BLENDING_OPTS
void expand_blend(u32 *screen_src_ptr, u16 *screen_dest_ptr,
u32 start, u32 end)
{
u32 pixel_pair;
u32 pixel_top, pixel_bottom;
u32 bldalpha = io_registers[REG_BLDALPHA];
u32 blend_a = bldalpha & 0x1F;
u32 blend_b = (bldalpha >> 8) & 0x1F;
u32 i;
if(blend_a > 16)
blend_a = 16;
if(blend_b > 16)
blend_b = 16;
// The individual colors can saturate over 31, this should be taken
// care of in an alternate pass as it incurs a huge additional speedhit.
if((blend_a + blend_b) > 16)
{
expand_loop(blend_saturate, effect_condition_alpha, pixel_pair);
}
else
{
expand_loop(blend, effect_condition_alpha, pixel_pair);
}
}
#endif
// Blend scanline with white.
static void expand_darken(u16 *screen_src_ptr, u16 *screen_dest_ptr,
u32 start, u32 end)
{
u32 pixel_top;
s32 blend = 16 - (io_registers[REG_BLDY] & 0x1F);
u32 i;
if(blend < 0)
blend = 0;
expand_loop(darken, effect_condition_fade(pixel_top), pixel_top);
}
// Blend scanline with black.
static void expand_brighten(u16 *screen_src_ptr, u16 *screen_dest_ptr,
u32 start, u32 end)
{
u32 pixel_top;
u32 blend = io_registers[REG_BLDY] & 0x1F;
u32 upper;
u32 i;
if(blend > 16)
blend = 16;
upper = ((0x07E0F81F * blend) >> 4) & 0x07E0F81F;
blend = 16 - blend;
expand_loop(brighten, effect_condition_fade(pixel_top), pixel_top);
}
// Expand scanline such that if both top and bottom pass it's alpha,
// if only top passes it's as specified, and if neither pass it's normal.
static void expand_darken_partial_alpha(u32 *screen_src_ptr, u16 *screen_dest_ptr,
u32 start, u32 end)
{
s32 blend = 16 - (io_registers[REG_BLDY] & 0x1F);
u32 pixel_pair;
u32 pixel_top, pixel_bottom;
u32 bldalpha = io_registers[REG_BLDALPHA];
u32 blend_a = bldalpha & 0x1F;
u32 blend_b = (bldalpha >> 8) & 0x1F;
u32 i;
if(blend < 0)
blend = 0;
if(blend_a > 16)
blend_a = 16;
if(blend_b > 16)
blend_b = 16;
expand_partial_alpha(darken);
}
static void expand_brighten_partial_alpha(u32 *screen_src_ptr, u16 *screen_dest_ptr,
u32 start, u32 end)
{
s32 blend = io_registers[REG_BLDY] & 0x1F;
u32 pixel_pair;
u32 pixel_top, pixel_bottom;
u32 bldalpha = io_registers[REG_BLDALPHA];
u32 blend_a = bldalpha & 0x1F;
u32 blend_b = (bldalpha >> 8) & 0x1F;
u32 upper;
u32 i;
if(blend > 16)
blend = 16;
upper = ((0x07E0F81F * blend) >> 4) & 0x07E0F81F;
blend = 16 - blend;
if(blend_a > 16)
blend_a = 16;
if(blend_b > 16)
blend_b = 16;
expand_partial_alpha(brighten);
}
// Render an OBJ layer from start to end, depending on the type (1D or 2D)
// stored in dispcnt.
#define render_obj_layer(type, dest, _start, _end) \
current_layer &= ~0x04; \
if(dispcnt & 0x40) \
render_scanline_obj_##type##_1D(current_layer, _start, _end, dest); \
else \
render_scanline_obj_##type##_2D(current_layer, _start, _end, dest) \
// Render a target all the way with the background color as taken from the
// palette.
#define fill_line_bg(type, dest, _start, _end) \
fill_line_##type(0, dest, _start, _end) \
// Render all layers as they appear in the layer order.
#define render_layers(tile_alpha, obj_alpha, dest) \
{ \
current_layer = layer_order[0]; \
if(current_layer & 0x04) \
{ \
/* If the first one is OBJ render the background then render it. */ \
fill_line_bg(tile_alpha, dest, 0, 240); \
render_obj_layer(obj_alpha, dest, 0, 240); \
} \
else \
{ \
/* Otherwise render a base layer. */ \
layer_renderers[current_layer].tile_alpha##_render_base(current_layer, \
0, 240, dest); \
} \
\
/* Render the rest of the layers. */ \
for(layer_order_pos = 1; layer_order_pos < layer_count; layer_order_pos++) \
{ \
current_layer = layer_order[layer_order_pos]; \
if(current_layer & 0x04) \
{ \
render_obj_layer(obj_alpha, dest, 0, 240); \
} \
else \
{ \
layer_renderers[current_layer]. \
tile_alpha##_render_transparent(current_layer, 0, 240, dest); \
} \
} \
} \
#define render_condition_alpha \
(((io_registers[REG_BLDALPHA] & 0x1F1F) != 0x001F) && \
((io_registers[REG_BLDCNT] & 0x3F) != 0) && \
((io_registers[REG_BLDCNT] & 0x3F00) != 0)) \
#define render_condition_fade \
(((io_registers[REG_BLDY] & 0x1F) != 0) && \
((io_registers[REG_BLDCNT] & 0x3F) != 0)) \
#define render_layers_color_effect(renderer, layer_condition, \
alpha_condition, fade_condition, _start, _end) \
{ \
if(layer_condition) \
{ \
if(obj_alpha_count[io_registers[REG_VCOUNT]] > 0) \
{ \
/* Render based on special effects mode. */ \
u32 screen_buffer[240]; \
switch((bldcnt >> 6) & 0x03) \
{ \
/* Alpha blend */ \
case 0x01: \
{ \
if(alpha_condition) \
{ \
renderer(alpha, alpha_obj, screen_buffer); \
expand_blend(screen_buffer, scanline, _start, _end); \
return; \
} \
break; \
} \
\
/* Fade to white */ \
case 0x02: \
{ \
if(fade_condition) \
{ \
renderer(color32, partial_alpha, screen_buffer); \
expand_brighten_partial_alpha(screen_buffer, scanline, \
_start, _end); \
return; \
} \
break; \
} \
\
/* Fade to black */ \
case 0x03: \
{ \
if(fade_condition) \
{ \
renderer(color32, partial_alpha, screen_buffer); \
expand_darken_partial_alpha(screen_buffer, scanline, \
_start, _end); \
return; \
} \
break; \
} \
} \
\
renderer(color32, partial_alpha, screen_buffer); \
expand_blend(screen_buffer, scanline, _start, _end); \
} \
else \
{ \
/* Render based on special effects mode. */ \
switch((bldcnt >> 6) & 0x03) \
{ \
/* Alpha blend */ \
case 0x01: \
{ \
if(alpha_condition) \
{ \
u32 screen_buffer[240]; \
renderer(alpha, alpha_obj, screen_buffer); \
expand_blend(screen_buffer, scanline, _start, _end); \
return; \
} \
break; \
} \
\
/* Fade to white */ \
case 0x02: \
{ \
if(fade_condition) \
{ \
renderer(color16, color16, scanline); \
expand_brighten(scanline, scanline, _start, _end); \
return; \
} \
break; \
} \
\
/* Fade to black */ \
case 0x03: \
{ \
if(fade_condition) \
{ \
renderer(color16, color16, scanline); \
expand_darken(scanline, scanline, _start, _end); \
return; \
} \
break; \
} \
} \
\
renderer(normal, normal, scanline); \
expand_normal(scanline, _start, _end); \
} \
} \
else \
{ \
u32 pixel_top = palette_ram_converted[0]; \
switch((bldcnt >> 6) & 0x03) \
{ \
/* Fade to white */ \
case 0x02: \
{ \
if(color_combine_mask_a(5)) \
{ \
u32 blend = io_registers[REG_BLDY] & 0x1F; \
u32 upper; \
\
if(blend > 16) \
blend = 16; \
\
upper = ((0x07E0F81F * blend) >> 4) & 0x07E0F81F; \
blend = 16 - blend; \
\
expand_pixel_no_dest(brighten, pixel_top); \
} \
break; \
} \
\
/* Fade to black */ \
case 0x03: \
{ \
if(color_combine_mask_a(5)) \
{ \
s32 blend = 16 - (io_registers[REG_BLDY] & 0x1F); \
\
if(blend < 0) \
blend = 0; \
\
expand_pixel_no_dest(darken, pixel_top); \
} \
break; \
} \
} \
fill_line_color16(pixel_top, scanline, _start, _end); \
} \
} \
// Renders an entire scanline from 0 to 240, based on current color mode.
static void render_scanline_tile(u16 *scanline, u32 dispcnt)
{
u32 current_layer;
u32 layer_order_pos;
u32 bldcnt = io_registers[REG_BLDCNT];
render_scanline_layer_functions_tile();
render_layers_color_effect(render_layers, layer_count,
render_condition_alpha, render_condition_fade, 0, 240);
}
static void render_scanline_bitmap(u16 *scanline, u32 dispcnt)
{
render_scanline_layer_functions_bitmap();
u32 current_layer;
u32 layer_order_pos;
fill_line_bg(normal, scanline, 0, 240);
for(layer_order_pos = 0; layer_order_pos < layer_count; layer_order_pos++)
{
current_layer = layer_order[layer_order_pos];
if(current_layer & 0x04)
{
render_obj_layer(normal, scanline, 0, 240);
}
else
{
layer_renderers->normal_render(0, 240, scanline);
}
}
}
// Render layers from start to end based on if they're allowed in the
// enable flags.
#define render_layers_conditional(tile_alpha, obj_alpha, dest) \
{ \
__label__ skip; \
current_layer = layer_order[layer_order_pos]; \
/* If OBJ aren't enabled skip to the first non-OBJ layer */ \
if(!(enable_flags & 0x10)) \
{ \
while((current_layer & 0x04) || !((1 << current_layer) & enable_flags)) \
{ \
layer_order_pos++; \
current_layer = layer_order[layer_order_pos]; \
\
/* Oops, ran out of layers, render the background. */ \
if(layer_order_pos == layer_count) \
{ \
fill_line_bg(tile_alpha, dest, start, end); \
goto skip; \
} \
} \
\
/* Render the first valid layer */ \
layer_renderers[current_layer].tile_alpha##_render_base(current_layer, \
start, end, dest); \
\
layer_order_pos++; \
\
/* Render the rest of the layers if active, skipping OBJ ones. */ \
for(; layer_order_pos < layer_count; layer_order_pos++) \
{ \
current_layer = layer_order[layer_order_pos]; \
if(!(current_layer & 0x04) && ((1 << current_layer) & enable_flags)) \
{ \
layer_renderers[current_layer]. \
tile_alpha##_render_transparent(current_layer, start, end, dest); \
} \
} \
} \
else \
{ \
/* Find the first active layer, skip all of the inactive ones */ \
while(!((current_layer & 0x04) || ((1 << current_layer) & enable_flags))) \
{ \
layer_order_pos++; \
current_layer = layer_order[layer_order_pos]; \
\
/* Oops, ran out of layers, render the background. */ \
if(layer_order_pos == layer_count) \
{ \
fill_line_bg(tile_alpha, dest, start, end); \
goto skip; \
} \
} \
\
if(current_layer & 0x04) \
{ \
/* If the first one is OBJ render the background then render it. */ \
fill_line_bg(tile_alpha, dest, start, end); \
render_obj_layer(obj_alpha, dest, start, end); \
} \
else \
{ \
/* Otherwise render a base layer. */ \
layer_renderers[current_layer]. \
tile_alpha##_render_base(current_layer, start, end, dest); \
} \
\
layer_order_pos++; \
\
/* Render the rest of the layers. */ \
for(; layer_order_pos < layer_count; layer_order_pos++) \
{ \
current_layer = layer_order[layer_order_pos]; \
if(current_layer & 0x04) \
{ \
render_obj_layer(obj_alpha, dest, start, end); \
} \
else \
{ \
if(enable_flags & (1 << current_layer)) \
{ \
layer_renderers[current_layer]. \
tile_alpha##_render_transparent(current_layer, start, end, dest); \
} \
} \
} \
} \
\
skip: \
; \
} \
// Render all of the BG and OBJ in a tiled scanline from start to end ONLY if
// enable_flag allows that layer/OBJ. Also conditionally render color effects.
static void render_scanline_conditional_tile(u32 start, u32 end, u16 *scanline,
u32 enable_flags, u32 dispcnt, u32 bldcnt, const tile_layer_render_struct
*layer_renderers)
{
u32 current_layer;
u32 layer_order_pos = 0;
render_layers_color_effect(render_layers_conditional,
(layer_count && (enable_flags & 0x1F)),
((enable_flags & 0x20) && render_condition_alpha),
((enable_flags & 0x20) && render_condition_fade), start, end);
}
// Render the BG and OBJ in a bitmap scanline from start to end ONLY if
// enable_flag allows that layer/OBJ. Also conditionally render color effects.
static void render_scanline_conditional_bitmap(u32 start, u32 end, u16 *scanline,
u32 enable_flags, u32 dispcnt, u32 bldcnt, const bitmap_layer_render_struct
*layer_renderers)
{
u32 current_layer;
u32 layer_order_pos;
fill_line_bg(normal, scanline, start, end);
for(layer_order_pos = 0; layer_order_pos < layer_count; layer_order_pos++)
{
current_layer = layer_order[layer_order_pos];
if(current_layer & 0x04)
{
if(enable_flags & 0x10)
{
render_obj_layer(normal, scanline, start, end);
}
}
else
{
if(enable_flags & 0x04)
layer_renderers->normal_render(start, end, scanline);
}
}
}
#define window_x_coords(window_number) \
window_##window_number##_x1 = \
io_registers[REG_WIN##window_number##H] >> 8; \
window_##window_number##_x2 = \
io_registers[REG_WIN##window_number##H] & 0xFF; \
window_##window_number##_enable = \
(winin >> (window_number * 8)) & 0x3F; \
\
if(window_##window_number##_x1 > 240) \
window_##window_number##_x1 = 240; \
\
if(window_##window_number##_x2 > 240) \
window_##window_number##_x2 = 240 \
#define window_coords(window_number) \
u32 window_##window_number##_x1, window_##window_number##_x2; \
u32 window_##window_number##_y1, window_##window_number##_y2; \
u32 window_##window_number##_enable = 0; \
window_##window_number##_y1 = \
io_registers[REG_WIN##window_number##V] >> 8; \
window_##window_number##_y2 = \
io_registers[REG_WIN##window_number##V] & 0xFF; \
\
if(window_##window_number##_y1 > window_##window_number##_y2) \
{ \
if((((vcount <= window_##window_number##_y2) || \
(vcount > window_##window_number##_y1)) || \
(window_##window_number##_y2 > 227)) && \
(window_##window_number##_y1 <= 227)) \
{ \
window_x_coords(window_number); \
} \
else \
{ \
window_##window_number##_x1 = 240; \
window_##window_number##_x2 = 240; \
} \
} \
else \
{ \
if((((vcount >= window_##window_number##_y1) && \
(vcount < window_##window_number##_y2)) || \
(window_##window_number##_y2 > 227)) && \
(window_##window_number##_y1 <= 227)) \
{ \
window_x_coords(window_number); \
} \
else \
{ \
window_##window_number##_x1 = 240; \
window_##window_number##_x2 = 240; \
} \
} \
#define render_window_segment(type, start, end, window_type) \
if(start != end) \
{ \
render_scanline_conditional_##type(start, end, scanline, \
window_##window_type##_enable, dispcnt, bldcnt, layer_renderers); \
} \
#define render_window_segment_unequal(type, start, end, window_type) \
render_scanline_conditional_##type(start, end, scanline, \
window_##window_type##_enable, dispcnt, bldcnt, layer_renderers) \
#define render_window_segment_clip(type, clip_start, clip_end, start, end, \
window_type) \
{ \
if(start != end) \
{ \
if(start < clip_start) \
{ \
if(end > clip_start) \
{ \
if(end > clip_end) \
{ \
render_window_segment_unequal(type, clip_start, clip_end, \
window_type); \
} \
else \
{ \
render_window_segment_unequal(type, clip_start, end, window_type); \
} \
} \
} \
else \
\
if(end > clip_end) \
{ \
if(start < clip_end) \
render_window_segment_unequal(type, start, clip_end, window_type); \
} \
else \
{ \
render_window_segment_unequal(type, start, end, window_type); \
} \
} \
} \
#define render_window_clip_1(type, start, end) \
if(window_1_x1 != 240) \
{ \
if(window_1_x1 > window_1_x2) \
{ \
render_window_segment_clip(type, start, end, 0, window_1_x2, 1); \
render_window_segment_clip(type, start, end, window_1_x2, window_1_x1, \
out); \
render_window_segment_clip(type, start, end, window_1_x1, 240, 1); \
} \
else \
{ \
render_window_segment_clip(type, start, end, 0, window_1_x1, out); \
render_window_segment_clip(type, start, end, window_1_x1, window_1_x2, \
1); \
render_window_segment_clip(type, start, end, window_1_x2, 240, out); \
} \
} \
else \
{ \
render_window_segment(type, start, end, out); \
} \
#define render_window_clip_obj(type, start, end); \
render_window_segment(type, start, end, out); \
if(dispcnt & 0x40) \
render_scanline_obj_copy_##type##_1D(4, start, end, scanline); \
else \
render_scanline_obj_copy_##type##_2D(4, start, end, scanline) \
#define render_window_segment_clip_obj(type, clip_start, clip_end, start, \
end) \
{ \
if(start != end) \
{ \
if(start < clip_start) \
{ \
if(end > clip_start) \
{ \
if(end > clip_end) \
{ \
render_window_clip_obj(type, clip_start, clip_end); \
} \
else \
{ \
render_window_clip_obj(type, clip_start, end); \
} \
} \
} \
else \
\
if(end > clip_end) \
{ \
if(start < clip_end) \
{ \
render_window_clip_obj(type, start, clip_end); \
} \
} \
else \
{ \
render_window_clip_obj(type, start, end); \
} \
} \
} \
#define render_window_clip_1_obj(type, start, end) \
if(window_1_x1 != 240) \
{ \
if(window_1_x1 > window_1_x2) \
{ \
render_window_segment_clip(type, start, end, 0, window_1_x2, 1); \
render_window_segment_clip_obj(type, start, end, window_1_x2, \
window_1_x1); \
render_window_segment_clip(type, start, end, window_1_x1, 240, 1); \
} \
else \
{ \
render_window_segment_clip_obj(type, start, end, 0, window_1_x1); \
render_window_segment_clip(type, start, end, window_1_x1, window_1_x2, \
1); \
render_window_segment_clip_obj(type, start, end, window_1_x2, 240); \
} \
} \
else \
{ \
render_window_clip_obj(type, start, end); \
} \
#define render_window_single(type, window_number) \
u32 winin = io_registers[REG_WININ]; \
window_coords(window_number); \
if(window_##window_number##_x1 > window_##window_number##_x2) \
{ \
render_window_segment(type, 0, window_##window_number##_x2, \
window_number); \
render_window_segment(type, window_##window_number##_x2, \
window_##window_number##_x1, out); \
render_window_segment(type, window_##window_number##_x1, 240, \
window_number); \
} \
else \
{ \
render_window_segment(type, 0, window_##window_number##_x1, out); \
render_window_segment(type, window_##window_number##_x1, \
window_##window_number##_x2, window_number); \
render_window_segment(type, window_##window_number##_x2, 240, out); \
} \
#define render_window_multi(type, front, back) \
if(window_##front##_x1 > window_##front##_x2) \
{ \
render_window_segment(type, 0, window_##front##_x2, front); \
render_window_clip_##back(type, window_##front##_x2, \
window_##front##_x1); \
render_window_segment(type, window_##front##_x1, 240, front); \
} \
else \
{ \
render_window_clip_##back(type, 0, window_##front##_x1); \
render_window_segment(type, window_##front##_x1, window_##front##_x2, \
front); \
render_window_clip_##back(type, window_##front##_x2, 240); \
} \
#define render_scanline_window_builder(type) \
static void render_scanline_window_##type(u16 *scanline, u32 dispcnt) \
{ \
u32 vcount = io_registers[REG_VCOUNT]; \
u32 winout = io_registers[REG_WINOUT]; \
u32 bldcnt = io_registers[REG_BLDCNT]; \
u32 window_out_enable = winout & 0x3F; \
\
render_scanline_layer_functions_##type(); \
\
switch(dispcnt >> 13) \
{ \
/* Just window 0 */ \
case 0x01: \
{ \
render_window_single(type, 0); \
break; \
} \
\
/* Just window 1 */ \
case 0x02: \
{ \
render_window_single(type, 1); \
break; \
} \
\
/* Windows 1 and 2 */ \
case 0x03: \
{ \
u32 winin = io_registers[REG_WININ]; \
window_coords(0); \
window_coords(1); \
render_window_multi(type, 0, 1); \
break; \
} \
\
/* Just OBJ windows */ \
case 0x04: \
{ \
render_window_clip_obj(type, 0, 240); \
break; \
} \
\
/* Window 0 and OBJ window */ \
case 0x05: \
{ \
u32 winin = io_registers[REG_WININ]; \
window_coords(0); \
render_window_multi(type, 0, obj); \
break; \
} \
\
/* Window 1 and OBJ window */ \
case 0x06: \
{ \
u32 winin = io_registers[REG_WININ]; \
window_coords(1); \
render_window_multi(type, 1, obj); \
break; \
} \
\
/* Window 0, 1, and OBJ window */ \
case 0x07: \
{ \
u32 winin = io_registers[REG_WININ]; \
window_coords(0); \
window_coords(1); \
render_window_multi(type, 0, 1_obj); \
break; \
} \
} \
} \
render_scanline_window_builder(tile);
render_scanline_window_builder(bitmap);
static const u32 active_layers[6] = { 0x1F, 0x17, 0x1C, 0x14, 0x14, 0x14 };
void update_scanline(void)
{
u32 pitch = get_screen_pitch();
u32 dispcnt = io_registers[REG_DISPCNT];
u32 vcount = io_registers[REG_VCOUNT];
u16 *screen_offset = get_screen_pixels() + (vcount * pitch);
u32 video_mode = dispcnt & 0x07;
// If OAM has been modified since the last scanline has been updated then
// reorder and reprofile the OBJ lists.
if(oam_update)
{
order_obj(video_mode);
oam_update = 0;
}
order_layers((dispcnt >> 8) & active_layers[video_mode]);
if(skip_next_frame)
return;
// If the screen is in in forced blank draw pure white.
if(dispcnt & 0x80)
{
fill_line_color16(0xFFFF, screen_offset, 0, 240);
}
else
{
if(video_mode < 3)
{
if(dispcnt >> 13)
{
render_scanline_window_tile(screen_offset, dispcnt);
}
else
{
render_scanline_tile(screen_offset, dispcnt);
}
}
else
{
if(dispcnt >> 13)
render_scanline_window_bitmap(screen_offset, dispcnt);
else
render_scanline_bitmap(screen_offset, dispcnt);
}
}
affine_reference_x[0] += (s16)io_registers[REG_BG2PB];
affine_reference_y[0] += (s16)io_registers[REG_BG2PD];
affine_reference_x[1] += (s16)io_registers[REG_BG3PB];
affine_reference_y[1] += (s16)io_registers[REG_BG3PD];
}
#define video_savestate_builder(type) \
void video_##type##_savestate(void) \
{ \
state_mem_##type##_array(affine_reference_x); \
state_mem_##type##_array(affine_reference_y); \
}
video_savestate_builder(read)
video_savestate_builder(write)