Commit | Line | Data |
---|---|---|
75e28f62 E |
1 | /* |
2 | * Copyright (C) 2011 Gilead Kutnick "Exophase" <exophase@gmail.com> | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License as | |
6 | * published by the Free Software Foundation; either version 2 of | |
7 | * the License, or (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
12 | * General Public License for more details. | |
13 | */ | |
14 | ||
15 | #ifndef PSX_GPU_H | |
16 | #define PSX_GPU_H | |
17 | ||
2d658c89 | 18 | #define MAX_SPANS 512 |
19 | #define MAX_BLOCKS 64 | |
20 | #define MAX_BLOCKS_PER_ROW 128 | |
21 | ||
22 | #define SPAN_DATA_BLOCKS_SIZE 32 | |
23 | ||
ee060c58 | 24 | #define AHACK_TEXTURE_ADJ_U (1 << 0) |
25 | #define AHACK_TEXTURE_ADJ_V (1 << 1) | |
26 | ||
2d658c89 | 27 | #ifndef __ASSEMBLER__ |
28 | ||
a4021361 | 29 | #include "vector_types.h" |
30 | ||
ee060c58 | 31 | #ifndef unlikely |
32 | #define unlikely(x) __builtin_expect((x), 0) | |
33 | #endif | |
34 | ||
f4ce97e1 | 35 | #define sign_extend_11bit(value) \ |
36 | (((s32)((value) << 21)) >> 21) | |
37 | ||
75e28f62 E |
38 | typedef enum |
39 | { | |
40 | PRIMITIVE_TYPE_TRIANGLE = 0, | |
41 | PRIMITIVE_TYPE_SPRITE = 1, | |
42 | PRIMITIVE_TYPE_LINE = 2, | |
43 | PRIMITIVE_TYPE_UNKNOWN = 3 | |
44 | } primitive_type_enum; | |
45 | ||
46 | typedef enum | |
47 | { | |
48 | TEXTURE_MODE_4BPP = 0, | |
49 | TEXTURE_MODE_8BPP = 1, | |
50 | TEXTURE_MODE_16BPP = 2 | |
51 | } texture_mode_enum; | |
52 | ||
53 | typedef enum | |
54 | { | |
55 | BLEND_MODE_AVERAGE = 0, | |
56 | BLEND_MODE_ADD = 1, | |
57 | BLEND_MODE_SUBTRACT = 2, | |
58 | BLEND_MODE_ADD_FOURTH = 3 | |
59 | } blend_mode_enum; | |
60 | ||
61 | typedef enum | |
62 | { | |
63 | RENDER_FLAGS_MODULATE_TEXELS = 0x1, | |
64 | RENDER_FLAGS_BLEND = 0x2, | |
65 | RENDER_FLAGS_TEXTURE_MAP = 0x4, | |
66 | RENDER_FLAGS_QUAD = 0x8, | |
67 | RENDER_FLAGS_SHADE = 0x10, | |
68 | } render_flags_enum; | |
69 | ||
70 | typedef enum | |
71 | { | |
72 | RENDER_STATE_DITHER = 0x8, | |
73 | RENDER_STATE_MASK_EVALUATE = 0x20, | |
74 | } render_state_enum; | |
75 | ||
69b09c0d E |
76 | typedef enum |
77 | { | |
78 | RENDER_INTERLACE_ENABLED = 0x1, | |
f1359c57 | 79 | RENDER_INTERLACE_ODD = 0x2, |
f1359c57 | 80 | } render_mode_enum; |
69b09c0d | 81 | |
75e28f62 E |
82 | typedef struct |
83 | { | |
84 | u16 left_x; | |
85 | u16 num_blocks; | |
86 | u16 right_mask; | |
87 | u16 y; | |
88 | } edge_data_struct; | |
89 | ||
b0d96051 | 90 | // 64 (72) bytes total |
75e28f62 E |
91 | typedef struct |
92 | { | |
93 | // 16 bytes | |
94 | union | |
95 | { | |
96 | vec_8x16u uv; | |
97 | vec_8x16u texels; | |
98 | vec_8x16u draw_mask; | |
99 | }; | |
100 | ||
101 | // 24 bytes | |
102 | union | |
103 | { | |
104 | struct | |
105 | { | |
106 | vec_8x8u r; | |
107 | vec_8x8u g; | |
108 | vec_8x8u b; | |
109 | }; | |
110 | ||
111 | vec_8x16u pixels; | |
112 | }; | |
113 | ||
b0d96051 | 114 | // 8 (16) bytes |
75e28f62 E |
115 | u32 draw_mask_bits; |
116 | u16 *fb_ptr; | |
117 | ||
118 | // 16 bytes | |
119 | vec_8x16u dither_offsets; | |
120 | } block_struct; | |
121 | ||
75e28f62 E |
122 | typedef struct render_block_handler_struct render_block_handler_struct; |
123 | ||
124 | typedef struct | |
125 | { | |
126 | // 144 bytes | |
127 | vec_8x16u test_mask; | |
128 | ||
129 | vec_4x32u uvrg; | |
130 | vec_4x32u uvrg_dx; | |
131 | vec_4x32u uvrg_dy; | |
132 | ||
133 | vec_4x32u u_block_span; | |
134 | vec_4x32u v_block_span; | |
135 | vec_4x32u r_block_span; | |
136 | vec_4x32u g_block_span; | |
137 | vec_4x32u b_block_span; | |
138 | ||
75e28f62 E |
139 | u32 b; |
140 | u32 b_dy; | |
141 | ||
142 | u32 triangle_area; | |
143 | ||
144 | u32 texture_window_settings; | |
145 | u32 current_texture_mask; | |
146 | u32 viewport_mask; | |
147 | u32 dirty_textures_4bpp_mask; | |
148 | u32 dirty_textures_8bpp_mask; | |
149 | u32 dirty_textures_8bpp_alternate_mask; | |
150 | ||
151 | u32 triangle_color; | |
75e28f62 E |
152 | u32 dither_table[4]; |
153 | ||
154 | struct render_block_handler_struct *render_block_handler; | |
155 | void *texture_page_ptr; | |
3867c6ef | 156 | void *texture_page_base; |
75e28f62 E |
157 | u16 *clut_ptr; |
158 | u16 *vram_ptr; | |
c1817bd9 | 159 | u16 *vram_out_ptr; |
75e28f62 | 160 | |
2da2fc76 | 161 | u32 uvrgb_phase; |
162 | ||
75e28f62 E |
163 | u16 render_state_base; |
164 | u16 render_state; | |
165 | ||
166 | u16 num_spans; | |
167 | u16 num_blocks; | |
168 | ||
75e28f62 E |
169 | s16 viewport_start_x; |
170 | s16 viewport_start_y; | |
171 | s16 viewport_end_x; | |
172 | s16 viewport_end_y; | |
173 | ||
174 | u16 mask_msb; | |
175 | ||
75e28f62 E |
176 | u8 triangle_winding; |
177 | ||
178 | u8 display_area_draw_enable; | |
179 | ||
180 | u8 current_texture_page; | |
181 | u8 last_8bpp_texture_page; | |
182 | ||
183 | u8 texture_mask_width; | |
184 | u8 texture_mask_height; | |
185 | u8 texture_window_x; | |
186 | u8 texture_window_y; | |
187 | ||
188 | u8 primitive_type; | |
f1359c57 | 189 | u8 render_mode; |
75e28f62 | 190 | |
c6063f89 | 191 | s16 offset_x; |
192 | s16 offset_y; | |
193 | ||
194 | u16 clut_settings; | |
195 | u16 texture_settings; | |
196 | ||
ed0fd81d | 197 | u32 *reciprocal_table_ptr; |
198 | ||
c1817bd9 | 199 | // enhancement stuff |
0b4038f8 | 200 | u16 *enhancement_buf_ptr; // main alloc |
201 | u16 *enhancement_current_buf_ptr; // offset into above, 4 bufs | |
ee060c58 | 202 | u32 hacks_active; // AHACK_TEXTURE_ADJ_U ... |
0b4038f8 | 203 | u32 saved_hres; |
c1817bd9 | 204 | s16 saved_viewport_start_x; |
205 | s16 saved_viewport_start_y; | |
206 | s16 saved_viewport_end_x; | |
207 | s16 saved_viewport_end_y; | |
2da2fc76 | 208 | struct psx_gpu_scanout { |
209 | u16 x, y, w, h; | |
210 | } enhancement_scanouts[4]; // 0-3 specifying which buf to use | |
211 | u16 enhancement_scanout_eselect; // eviction selector | |
212 | u16 enhancement_current_buf; | |
c1817bd9 | 213 | |
ed8077c5 | 214 | u32 allow_dithering:1; |
215 | u32 force_dithering:1; | |
6ab6ab97 | 216 | u32 hack_disable_main:1; |
217 | u32 hack_texture_adj:1; | |
218 | ||
75e28f62 | 219 | // Align up to 64 byte boundary to keep the upcoming buffers cache line |
c1817bd9 | 220 | // aligned, also make reachable with single immediate addition |
db2804fb | 221 | u8 reserved_a[68 + 9*4 - 9*sizeof(void *)]; |
222 | ||
223 | // space for saving regs on c call to flush_render_block_buffer() and asm | |
224 | u32 saved_tmp[48 / sizeof(u32)]; | |
225 | u32 saved_q4_q7[64 / sizeof(u32)]; | |
75e28f62 E |
226 | |
227 | // 8KB | |
228 | block_struct blocks[MAX_BLOCKS_PER_ROW]; | |
229 | ||
230 | // 14336 bytes | |
231 | vec_4x32u span_uvrg_offset[MAX_SPANS]; | |
232 | edge_data_struct span_edge_data[MAX_SPANS]; | |
233 | u32 span_b_offset[MAX_SPANS]; | |
234 | ||
75e28f62 E |
235 | u8 texture_4bpp_cache[32][256 * 256]; |
236 | u8 texture_8bpp_even_cache[16][256 * 256]; | |
237 | u8 texture_8bpp_odd_cache[16][256 * 256]; | |
75e28f62 E |
238 | } psx_gpu_struct; |
239 | ||
240 | typedef struct __attribute__((aligned(16))) | |
241 | { | |
242 | u8 u; | |
243 | u8 v; | |
244 | ||
245 | u8 r; | |
246 | u8 g; | |
247 | u8 b; | |
248 | ||
249 | u8 reserved[3]; | |
250 | ||
251 | s16 x; | |
252 | s16 y; | |
a4021361 | 253 | |
254 | u32 padding; | |
75e28f62 E |
255 | } vertex_struct; |
256 | ||
f4ce97e1 | 257 | typedef struct |
258 | { | |
259 | vertex_struct *vertexes[3]; | |
260 | s16 offset_x; | |
261 | s16 offset_y; | |
262 | } prepared_triangle; | |
263 | ||
75e28f62 E |
264 | void render_block_fill(psx_gpu_struct *psx_gpu, u32 color, u32 x, u32 y, |
265 | u32 width, u32 height); | |
266 | void render_block_copy(psx_gpu_struct *psx_gpu, u16 *source, u32 x, u32 y, | |
267 | u32 width, u32 height, u32 pitch); | |
268 | void render_block_move(psx_gpu_struct *psx_gpu, u32 source_x, u32 source_y, | |
269 | u32 dest_x, u32 dest_y, u32 width, u32 height); | |
270 | ||
271 | void render_triangle(psx_gpu_struct *psx_gpu, vertex_struct *vertexes, | |
272 | u32 flags); | |
273 | void render_sprite(psx_gpu_struct *psx_gpu, s32 x, s32 y, u32 u, u32 v, | |
1cec4719 | 274 | s32 *width, s32 *height, u32 flags, u32 color); |
75e28f62 | 275 | void render_line(psx_gpu_struct *gpu, vertex_struct *vertexes, u32 flags, |
3b3dee71 | 276 | u32 color, int double_resolution); |
75e28f62 E |
277 | |
278 | u32 texture_region_mask(s32 x1, s32 y1, s32 x2, s32 y2); | |
279 | ||
a2cb152a | 280 | void update_texture_8bpp_cache(psx_gpu_struct *psx_gpu); |
75e28f62 E |
281 | void flush_render_block_buffer(psx_gpu_struct *psx_gpu); |
282 | ||
ee060c58 | 283 | void setup_blocks_uv_adj_hack(psx_gpu_struct *psx_gpu, block_struct *block, |
284 | edge_data_struct *span_edge_data, vec_4x32u *span_uvrg_offset); | |
285 | ||
e8c0e0bb | 286 | void initialize_psx_gpu(psx_gpu_struct *psx_gpu, u16 *vram); |
90ac6fed | 287 | u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size, |
8412166f | 288 | s32 *cpu_cycles_sum_out, s32 *cpu_cycles_last, u32 *last_command); |
75e28f62 E |
289 | |
290 | void triangle_benchmark(psx_gpu_struct *psx_gpu); | |
291 | ||
a4021361 | 292 | void compute_all_gradients(psx_gpu_struct * __restrict__ psx_gpu, |
293 | const vertex_struct * __restrict__ a, const vertex_struct * __restrict__ b, | |
294 | const vertex_struct * __restrict__ c); | |
295 | ||
2d658c89 | 296 | #endif // __ASSEMBLER__ |
297 | #endif // PSX_GPU_H |