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 is_blit:1; /* blit */
114 unsigned viewport_index:4; /* viewport index */
115 unsigned layer:11;
116 unsigned view_index:14;
117 unsigned stride; /* how much to advance data between a0, dadx, dady */
118 unsigned pad[2];
119 /* followed by a0, dadx, dady and planes[] */
120 };
121
122
123 struct lp_rast_plane {
124 /* edge function values at minx,miny ?? */
125 int64_t c;
126
127 int32_t dcdx;
128 int32_t dcdy;
129
130 /* one-pixel sized trivial reject offsets for each plane */
131 uint32_t eo;
132 /*
133 * We rely on this struct being 64bit aligned (ideally it would be 128bit
134 * but that's quite the waste) and therefore on 32bit we need padding
135 * since otherwise (even with the 64bit number in there) it wouldn't be.
136 */
137 uint32_t pad;
138 };
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 /*
192 * Return the address (as float[][4]) of the FS input values which
193 * are immediately after the 'inputs' object.
194 */
195 static inline float(*
GET_A0(const struct lp_rast_shader_inputs * inputs)196 GET_A0(const struct lp_rast_shader_inputs *inputs))[4]
197 {
198 return (float (*)[4]) (inputs + 1);
199 }
200
201 /*
202 * Return the address (as float[][4]) of the FS input partial derivatives
203 * (w.r.t. X) which are after the 'inputs' object.
204 */
205 static inline float(*
GET_DADX(const struct lp_rast_shader_inputs * inputs)206 GET_DADX(const struct lp_rast_shader_inputs *inputs))[4]
207 {
208 const uint8_t *p = (const uint8_t *) (inputs + 1);
209 return (float (*)[4]) (p + 1 * inputs->stride);
210 }
211
212 /*
213 * Return the address (as float[][4]) of the FS input partial derivatives
214 * (w.r.t. Y) which are after the 'inputs' object.
215 */
216 static inline float(*
GET_DADY(const struct lp_rast_shader_inputs * inputs)217 GET_DADY(const struct lp_rast_shader_inputs *inputs))[4]
218 {
219 const uint8_t *p = (const uint8_t *) (inputs + 1);
220 return (float (*)[4]) (p + 2 * inputs->stride);
221 }
222
223 static inline struct lp_rast_plane *
GET_PLANES(const struct lp_rast_triangle * tri)224 GET_PLANES(const struct lp_rast_triangle *tri)
225 {
226 const uint8_t *p = (const uint8_t *) (&tri->inputs + 1);
227 return (struct lp_rast_plane *) (p + 3 * tri->inputs.stride);
228 }
229
230
231 struct lp_rasterizer *
232 lp_rast_create( unsigned num_threads );
233
234 void
235 lp_rast_destroy( struct lp_rasterizer * );
236
237 void
238 lp_rast_queue_scene( struct lp_rasterizer *rast,
239 struct lp_scene *scene );
240
241 void
242 lp_rast_finish( struct lp_rasterizer *rast );
243
244
245 union lp_rast_cmd_arg {
246 const struct lp_rast_shader_inputs *shade_tile;
247 struct {
248 const struct lp_rast_triangle *tri;
249 unsigned plane_mask;
250 } triangle;
251 const struct lp_rast_rectangle *rectangle;
252 const struct lp_rast_state *set_state;
253 const struct lp_rast_clear_rb *clear_rb;
254 struct {
255 uint64_t value;
256 uint64_t mask;
257 } clear_zstencil;
258 struct lp_fence *fence;
259 struct llvmpipe_query *query_obj;
260 };
261
262
263 /* Cast wrappers. Hopefully these compile to noops!
264 */
265 static inline union lp_rast_cmd_arg
lp_rast_arg_inputs(const struct lp_rast_shader_inputs * shade_tile)266 lp_rast_arg_inputs( const struct lp_rast_shader_inputs *shade_tile )
267 {
268 union lp_rast_cmd_arg arg;
269 arg.shade_tile = shade_tile;
270 return arg;
271 }
272
273
274 static inline union lp_rast_cmd_arg
lp_rast_arg_triangle(const struct lp_rast_triangle * triangle,unsigned plane_mask)275 lp_rast_arg_triangle( const struct lp_rast_triangle *triangle,
276 unsigned plane_mask)
277 {
278 union lp_rast_cmd_arg arg;
279 arg.triangle.tri = triangle;
280 arg.triangle.plane_mask = plane_mask;
281 return arg;
282 }
283
284
285 /**
286 * Build argument for a contained triangle.
287 *
288 * All planes are enabled, so instead of the plane mask we pass the upper
289 * left coordinates of the a block that fully encloses the triangle.
290 */
291 static inline union lp_rast_cmd_arg
lp_rast_arg_triangle_contained(const struct lp_rast_triangle * triangle,unsigned x,unsigned y)292 lp_rast_arg_triangle_contained( const struct lp_rast_triangle *triangle,
293 unsigned x, unsigned y)
294 {
295 union lp_rast_cmd_arg arg;
296 arg.triangle.tri = triangle;
297 arg.triangle.plane_mask = x | (y << 8);
298 return arg;
299 }
300
301
302 static inline union lp_rast_cmd_arg
lp_rast_arg_rectangle(const struct lp_rast_rectangle * rectangle)303 lp_rast_arg_rectangle( const struct lp_rast_rectangle *rectangle )
304 {
305 union lp_rast_cmd_arg arg;
306 arg.rectangle = rectangle;
307 return arg;
308 }
309
310
311 static inline union lp_rast_cmd_arg
lp_rast_arg_state(const struct lp_rast_state * state)312 lp_rast_arg_state( const struct lp_rast_state *state )
313 {
314 union lp_rast_cmd_arg arg;
315 arg.set_state = state;
316 return arg;
317 }
318
319
320 static inline union lp_rast_cmd_arg
lp_rast_arg_fence(struct lp_fence * fence)321 lp_rast_arg_fence( struct lp_fence *fence )
322 {
323 union lp_rast_cmd_arg arg;
324 arg.fence = fence;
325 return arg;
326 }
327
328
329 static inline union lp_rast_cmd_arg
lp_rast_arg_clearzs(uint64_t value,uint64_t mask)330 lp_rast_arg_clearzs( uint64_t value, uint64_t mask )
331 {
332 union lp_rast_cmd_arg arg;
333 arg.clear_zstencil.value = value;
334 arg.clear_zstencil.mask = mask;
335 return arg;
336 }
337
338
339 static inline union lp_rast_cmd_arg
lp_rast_arg_query(struct llvmpipe_query * pq)340 lp_rast_arg_query( struct llvmpipe_query *pq )
341 {
342 union lp_rast_cmd_arg arg;
343 arg.query_obj = pq;
344 return arg;
345 }
346
347
348 static inline union lp_rast_cmd_arg
lp_rast_arg_null(void)349 lp_rast_arg_null( void )
350 {
351 union lp_rast_cmd_arg arg;
352 arg.set_state = NULL;
353 return arg;
354 }
355
356
357 /**
358 * Binnable Commands.
359 * These get put into bins by the setup code and are called when
360 * the bins are executed.
361 */
362 enum lp_rast_op {
363 LP_RAST_OP_CLEAR_COLOR = 0x0,
364 LP_RAST_OP_CLEAR_ZSTENCIL = 0x1,
365 LP_RAST_OP_TRIANGLE_1 = 0x2,
366 LP_RAST_OP_TRIANGLE_2 = 0x3,
367 LP_RAST_OP_TRIANGLE_3 = 0x4,
368 LP_RAST_OP_TRIANGLE_4 = 0x5,
369 LP_RAST_OP_TRIANGLE_5 = 0x6,
370 LP_RAST_OP_TRIANGLE_6 = 0x7,
371 LP_RAST_OP_TRIANGLE_7 = 0x8,
372 LP_RAST_OP_TRIANGLE_8 = 0x9,
373 LP_RAST_OP_TRIANGLE_3_4 = 0xa,
374 LP_RAST_OP_TRIANGLE_3_16 = 0xb,
375 LP_RAST_OP_TRIANGLE_4_16 = 0xc,
376 LP_RAST_OP_SHADE_TILE = 0xd,
377 LP_RAST_OP_SHADE_TILE_OPAQUE = 0xe,
378 LP_RAST_OP_BEGIN_QUERY = 0xf,
379 LP_RAST_OP_END_QUERY = 0x10,
380 LP_RAST_OP_SET_STATE = 0x11,
381 LP_RAST_OP_TRIANGLE_32_1 = 0x12,
382 LP_RAST_OP_TRIANGLE_32_2 = 0x13,
383 LP_RAST_OP_TRIANGLE_32_3 = 0x14,
384 LP_RAST_OP_TRIANGLE_32_4 = 0x15,
385 LP_RAST_OP_TRIANGLE_32_5 = 0x16,
386 LP_RAST_OP_TRIANGLE_32_6 = 0x17,
387 LP_RAST_OP_TRIANGLE_32_7 = 0x18,
388 LP_RAST_OP_TRIANGLE_32_8 = 0x19,
389 LP_RAST_OP_TRIANGLE_32_3_4 = 0x1a,
390 LP_RAST_OP_TRIANGLE_32_3_16 = 0x1b,
391 LP_RAST_OP_TRIANGLE_32_4_16 = 0x1c,
392 LP_RAST_OP_MS_TRIANGLE_1 = 0x1d,
393 LP_RAST_OP_MS_TRIANGLE_2 = 0x1e,
394 LP_RAST_OP_MS_TRIANGLE_3 = 0x1f,
395 LP_RAST_OP_MS_TRIANGLE_4 = 0x20,
396 LP_RAST_OP_MS_TRIANGLE_5 = 0x21,
397 LP_RAST_OP_MS_TRIANGLE_6 = 0x22,
398 LP_RAST_OP_MS_TRIANGLE_7 = 0x23,
399 LP_RAST_OP_MS_TRIANGLE_8 = 0x24,
400 LP_RAST_OP_MS_TRIANGLE_3_4 = 0x25,
401 LP_RAST_OP_MS_TRIANGLE_3_16 = 0x26,
402 LP_RAST_OP_MS_TRIANGLE_4_16 = 0x27,
403 LP_RAST_OP_RECTANGLE = 0x28, /* Keep at end */
404 LP_RAST_OP_BLIT = 0x29, /* Keep at end */
405 LP_RAST_OP_MAX = 0x2a,
406 LP_RAST_OP_MASK = 0xff
407 };
408
409
410 /* Returned by characterize_bin:
411 */
412 #define LP_RAST_FLAGS_TRI (0x1)
413 #define LP_RAST_FLAGS_RECT (0x2)
414 #define LP_RAST_FLAGS_TILE (0x4)
415 #define LP_RAST_FLAGS_BLIT (0x8)
416
417 struct lp_bin_info {
418 unsigned type:8; // bitmask of LP_RAST_FLAGS_x
419 unsigned count:24;
420 };
421
422 struct lp_bin_info
423 lp_characterize_bin(const struct cmd_bin *bin);
424
425 void
426 lp_debug_bins( struct lp_scene *scene );
427
428 void
429 lp_debug_draw_bins_by_cmd_length( struct lp_scene *scene );
430
431 void
432 lp_debug_draw_bins_by_coverage( struct lp_scene *scene );
433
434 void lp_rast_fence(struct lp_rasterizer *rast,
435 struct lp_fence **fence);
436 #endif
437