1 /**************************************************************************
2 *
3 * Copyright 2009 VMware, Inc.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 /**
29 * The rast code is concerned with rasterization of command bins.
30 * Each screen tile has a bin associated with it. To render the
31 * scene we iterate over the tile bins and execute the commands
32 * in each bin.
33 * We'll do that with multiple threads...
34 */
35
36
37 #ifndef LP_RAST_H
38 #define LP_RAST_H
39
40 #include "pipe/p_compiler.h"
41 #include "util/u_pack_color.h"
42 #include "util/u_rect.h"
43 #include "lp_jit.h"
44
45
46 struct lp_rasterizer;
47 struct lp_scene;
48 struct lp_fence;
49 struct cmd_bin;
50
51 #define FIXED_TYPE_WIDTH 64
52 /** For sub-pixel positioning */
53 #define FIXED_ORDER 8
54 #define FIXED_ONE (1<<FIXED_ORDER)
55 #define FIXED_SHIFT (FIXED_TYPE_WIDTH - 1)
56 /** Maximum length of an edge in a primitive in pixels.
57 * If the framebuffer is large we have to think about fixed-point
58 * integer overflow. Coordinates need ((FIXED_TYPE_WIDTH/2) - 1) bits
59 * to be able to fit product of two such coordinates inside
60 * FIXED_TYPE_WIDTH, any larger and we could overflow a
61 * FIXED_TYPE_WIDTH_-bit int.
62 */
63 #define MAX_FIXED_LENGTH (1 << (((FIXED_TYPE_WIDTH/2) - 1) - FIXED_ORDER))
64
65 #define MAX_FIXED_LENGTH32 (1 << (((32/2) - 1) - FIXED_ORDER))
66
67 /* Rasterizer output size going to jit fs, width/height */
68 #define LP_RASTER_BLOCK_SIZE 4
69
70 #define LP_MAX_ACTIVE_BINNED_QUERIES 64
71
72 #define IMUL64(a, b) (((int64_t)(a)) * ((int64_t)(b)))
73
74 struct lp_rasterizer_task;
75
76 extern const float lp_sample_pos_4x[4][2];
77
78 /**
79 * Rasterization state.
80 * Objects of this type are put into the shared data bin and pointed
81 * to by commands in the per-tile bins.
82 */
83 struct lp_rast_state {
84 /* State for the shader. This also contains state which feeds into
85 * the fragment shader, such as blend color and alpha ref value.
86 */
87 struct lp_jit_context jit_context;
88
89 /* The shader itself. Probably we also need to pass a pointer to
90 * the tile color/z/stencil data somehow
91 */
92 struct lp_fragment_shader_variant *variant;
93 };
94
95
96 /**
97 * Texture blit offsets.
98 */
99 struct lp_rast_blit {
100 int16_t x0;
101 int16_t y0;
102 };
103
104
105 /**
106 * Coefficients necessary to run the shader at a given location.
107 * First coefficient is position.
108 * These pointers point into the bin data buffer.
109 */
110 struct lp_rast_shader_inputs {
111 unsigned frontfacing:1; /** True for front-facing */
112 unsigned disable:1; /** Partially binned, disable this command */
113 unsigned opaque:1; /** Is opaque */
114 unsigned is_blit:1; /* blit */
115 unsigned pad0:12; /* wasted space */
116 unsigned view_index:16;
117 unsigned stride; /* how much to advance data between a0, dadx, dady */
118 unsigned layer; /* the layer to render to (from gs, already clamped) */
119 unsigned viewport_index; /* the active viewport index (from gs, already clamped) */
120
121 /* followed by a0, dadx, dady and planes[] */
122 };
123
124 struct lp_rast_plane {
125 /* edge function values at minx,miny ?? */
126 int64_t c;
127
128 int32_t dcdx;
129 int32_t dcdy;
130
131 /* one-pixel sized trivial reject offsets for each plane */
132 uint32_t eo;
133 /*
134 * We rely on this struct being 64bit aligned (ideally it would be 128bit
135 * but that's quite the waste) and therefore on 32bit we need padding
136 * since otherwise (even with the 64bit number in there) it wouldn't be.
137 */
138 uint32_t pad;
139 };
140
141 /**
142 * Rasterization information for a triangle known to be in this bin,
143 * plus inputs to run the shader:
144 * These fields are tile- and bin-independent.
145 * Objects of this type are put into the lp_setup_context::data buffer.
146 */
147 struct lp_rast_triangle {
148 #ifdef DEBUG
149 float v[3][2];
150 float pad0;
151 float pad1;
152 #endif
153
154 /* inputs for the shader */
155 struct lp_rast_shader_inputs inputs;
156 /* planes are also allocated here */
157 };
158
159
160 #define RECT_PLANE_LEFT 0x1
161 #define RECT_PLANE_RIGHT 0x2
162 #define RECT_PLANE_TOP 0x4
163 #define RECT_PLANE_BOTTOM 0x8
164
165 /**
166 * Rasterization information for a screen-aligned rectangle known to
167 * be in this bin, plus inputs to run the shader:
168 * These fields are tile- and bin-independent.
169 * Objects of this type are put into the lp_setup_context::data buffer.
170 */
171 struct lp_rast_rectangle {
172 #ifdef DEBUG
173 float v[2][2]; /**< diagonal corners */
174 #endif
175
176 /* Rectangle boundaries in integer pixels:
177 */
178 struct u_rect box;
179
180 /* inputs for the shader */
181 struct lp_rast_shader_inputs inputs;
182 };
183
184
185 struct lp_rast_clear_rb {
186 union util_color color_val;
187 unsigned cbuf;
188 };
189
190
191 #define GET_A0(inputs) ((float (*)[4])((inputs)+1))
192 #define GET_DADX(inputs) ((float (*)[4])((char *)((inputs) + 1) + (inputs)->stride))
193 #define GET_DADY(inputs) ((float (*)[4])((char *)((inputs) + 1) + 2 * (inputs)->stride))
194 #define GET_PLANES(tri) ((struct lp_rast_plane *)((char *)(&(tri)->inputs + 1) + 3 * (tri)->inputs.stride))
195
196
197
198 struct lp_rasterizer *
199 lp_rast_create( unsigned num_threads );
200
201 void
202 lp_rast_destroy( struct lp_rasterizer * );
203
204 void
205 lp_rast_queue_scene( struct lp_rasterizer *rast,
206 struct lp_scene *scene );
207
208 void
209 lp_rast_finish( struct lp_rasterizer *rast );
210
211
212 union lp_rast_cmd_arg {
213 const struct lp_rast_shader_inputs *shade_tile;
214 struct {
215 const struct lp_rast_triangle *tri;
216 unsigned plane_mask;
217 } triangle;
218 const struct lp_rast_rectangle *rectangle;
219 const struct lp_rast_state *set_state;
220 const struct lp_rast_clear_rb *clear_rb;
221 struct {
222 uint64_t value;
223 uint64_t mask;
224 } clear_zstencil;
225 const struct lp_rast_state *state;
226 struct lp_fence *fence;
227 struct llvmpipe_query *query_obj;
228 };
229
230
231 /* Cast wrappers. Hopefully these compile to noops!
232 */
233 static inline union lp_rast_cmd_arg
lp_rast_arg_inputs(const struct lp_rast_shader_inputs * shade_tile)234 lp_rast_arg_inputs( const struct lp_rast_shader_inputs *shade_tile )
235 {
236 union lp_rast_cmd_arg arg;
237 arg.shade_tile = shade_tile;
238 return arg;
239 }
240
241 static inline union lp_rast_cmd_arg
lp_rast_arg_triangle(const struct lp_rast_triangle * triangle,unsigned plane_mask)242 lp_rast_arg_triangle( const struct lp_rast_triangle *triangle,
243 unsigned plane_mask)
244 {
245 union lp_rast_cmd_arg arg;
246 arg.triangle.tri = triangle;
247 arg.triangle.plane_mask = plane_mask;
248 return arg;
249 }
250
251 /**
252 * Build argument for a contained triangle.
253 *
254 * All planes are enabled, so instead of the plane mask we pass the upper
255 * left coordinates of the a block that fully encloses the triangle.
256 */
257 static inline union lp_rast_cmd_arg
lp_rast_arg_triangle_contained(const struct lp_rast_triangle * triangle,unsigned x,unsigned y)258 lp_rast_arg_triangle_contained( const struct lp_rast_triangle *triangle,
259 unsigned x, unsigned y)
260 {
261 union lp_rast_cmd_arg arg;
262 arg.triangle.tri = triangle;
263 arg.triangle.plane_mask = x | (y << 8);
264 return arg;
265 }
266
267 static inline union lp_rast_cmd_arg
lp_rast_arg_rectangle(const struct lp_rast_rectangle * rectangle)268 lp_rast_arg_rectangle( const struct lp_rast_rectangle *rectangle )
269 {
270 union lp_rast_cmd_arg arg;
271 arg.rectangle = rectangle;
272 return arg;
273 }
274
275 static inline union lp_rast_cmd_arg
lp_rast_arg_state(const struct lp_rast_state * state)276 lp_rast_arg_state( const struct lp_rast_state *state )
277 {
278 union lp_rast_cmd_arg arg;
279 arg.set_state = state;
280 return arg;
281 }
282
283 static inline union lp_rast_cmd_arg
lp_rast_arg_fence(struct lp_fence * fence)284 lp_rast_arg_fence( struct lp_fence *fence )
285 {
286 union lp_rast_cmd_arg arg;
287 arg.fence = fence;
288 return arg;
289 }
290
291
292 static inline union lp_rast_cmd_arg
lp_rast_arg_clearzs(uint64_t value,uint64_t mask)293 lp_rast_arg_clearzs( uint64_t value, uint64_t mask )
294 {
295 union lp_rast_cmd_arg arg;
296 arg.clear_zstencil.value = value;
297 arg.clear_zstencil.mask = mask;
298 return arg;
299 }
300
301
302 static inline union lp_rast_cmd_arg
lp_rast_arg_query(struct llvmpipe_query * pq)303 lp_rast_arg_query( struct llvmpipe_query *pq )
304 {
305 union lp_rast_cmd_arg arg;
306 arg.query_obj = pq;
307 return arg;
308 }
309
310 static inline union lp_rast_cmd_arg
lp_rast_arg_null(void)311 lp_rast_arg_null( void )
312 {
313 union lp_rast_cmd_arg arg;
314 arg.set_state = NULL;
315 return arg;
316 }
317
318
319 /**
320 * Binnable Commands.
321 * These get put into bins by the setup code and are called when
322 * the bins are executed.
323 */
324 #define LP_RAST_OP_CLEAR_COLOR 0x0
325 #define LP_RAST_OP_CLEAR_ZSTENCIL 0x1
326 #define LP_RAST_OP_TRIANGLE_1 0x2
327 #define LP_RAST_OP_TRIANGLE_2 0x3
328 #define LP_RAST_OP_TRIANGLE_3 0x4
329 #define LP_RAST_OP_TRIANGLE_4 0x5
330 #define LP_RAST_OP_TRIANGLE_5 0x6
331 #define LP_RAST_OP_TRIANGLE_6 0x7
332 #define LP_RAST_OP_TRIANGLE_7 0x8
333 #define LP_RAST_OP_TRIANGLE_8 0x9
334 #define LP_RAST_OP_TRIANGLE_3_4 0xa
335 #define LP_RAST_OP_TRIANGLE_3_16 0xb
336 #define LP_RAST_OP_TRIANGLE_4_16 0xc
337 #define LP_RAST_OP_SHADE_TILE 0xd
338 #define LP_RAST_OP_SHADE_TILE_OPAQUE 0xe
339 #define LP_RAST_OP_BEGIN_QUERY 0xf
340 #define LP_RAST_OP_END_QUERY 0x10
341 #define LP_RAST_OP_SET_STATE 0x11
342 #define LP_RAST_OP_TRIANGLE_32_1 0x12
343 #define LP_RAST_OP_TRIANGLE_32_2 0x13
344 #define LP_RAST_OP_TRIANGLE_32_3 0x14
345 #define LP_RAST_OP_TRIANGLE_32_4 0x15
346 #define LP_RAST_OP_TRIANGLE_32_5 0x16
347 #define LP_RAST_OP_TRIANGLE_32_6 0x17
348 #define LP_RAST_OP_TRIANGLE_32_7 0x18
349 #define LP_RAST_OP_TRIANGLE_32_8 0x19
350 #define LP_RAST_OP_TRIANGLE_32_3_4 0x1a
351 #define LP_RAST_OP_TRIANGLE_32_3_16 0x1b
352 #define LP_RAST_OP_TRIANGLE_32_4_16 0x1c
353
354 #define LP_RAST_OP_MS_TRIANGLE_1 0x1d
355 #define LP_RAST_OP_MS_TRIANGLE_2 0x1e
356 #define LP_RAST_OP_MS_TRIANGLE_3 0x1f
357 #define LP_RAST_OP_MS_TRIANGLE_4 0x20
358 #define LP_RAST_OP_MS_TRIANGLE_5 0x21
359 #define LP_RAST_OP_MS_TRIANGLE_6 0x22
360 #define LP_RAST_OP_MS_TRIANGLE_7 0x23
361 #define LP_RAST_OP_MS_TRIANGLE_8 0x24
362 #define LP_RAST_OP_MS_TRIANGLE_3_4 0x25
363 #define LP_RAST_OP_MS_TRIANGLE_3_16 0x26
364 #define LP_RAST_OP_MS_TRIANGLE_4_16 0x27
365 #define LP_RAST_OP_RECTANGLE 0x28 /* Keep at end */
366 #define LP_RAST_OP_BLIT 0x29 /* Keep at end */
367
368 #define LP_RAST_OP_MAX 0x2a
369 #define LP_RAST_OP_MASK 0xff
370
371 /* Returned by characterize_bin:
372 */
373 #define LP_RAST_FLAGS_TRI (0x1)
374 #define LP_RAST_FLAGS_RECT (0x2)
375 #define LP_RAST_FLAGS_TILE (0x4)
376 #define LP_RAST_FLAGS_BLIT (0x8)
377
378 struct lp_bin_info {
379 unsigned type:8;
380 unsigned count:24;
381 };
382
383 struct lp_bin_info
384 lp_characterize_bin(const struct cmd_bin *bin);
385
386 void
387 lp_debug_bins( struct lp_scene *scene );
388 void
389 lp_debug_draw_bins_by_cmd_length( struct lp_scene *scene );
390 void
391 lp_debug_draw_bins_by_coverage( struct lp_scene *scene );
392
393
394 #endif
395