llvmpipe: reduce alignment requirement for resources from 64x64 to 4x4
[mesa.git] / src / gallium / drivers / llvmpipe / lp_rast.h
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 "lp_jit.h"
42
43
44 struct lp_rasterizer;
45 struct lp_scene;
46 struct lp_fence;
47 struct cmd_bin;
48
49 /** For sub-pixel positioning */
50 #define FIXED_ORDER 4
51 #define FIXED_ONE (1<<FIXED_ORDER)
52
53 /* Rasterizer output size going to jit fs, width/height */
54 #define LP_RASTER_BLOCK_SIZE 4
55
56
57 struct lp_rasterizer_task;
58
59
60 /**
61 * Rasterization state.
62 * Objects of this type are put into the shared data bin and pointed
63 * to by commands in the per-tile bins.
64 */
65 struct lp_rast_state {
66 /* State for the shader. This also contains state which feeds into
67 * the fragment shader, such as blend color and alpha ref value.
68 */
69 struct lp_jit_context jit_context;
70
71 /* The shader itself. Probably we also need to pass a pointer to
72 * the tile color/z/stencil data somehow
73 */
74 struct lp_fragment_shader_variant *variant;
75 };
76
77
78 /**
79 * Coefficients necessary to run the shader at a given location.
80 * First coefficient is position.
81 * These pointers point into the bin data buffer.
82 */
83 struct lp_rast_shader_inputs {
84 unsigned frontfacing:1; /** True for front-facing */
85 unsigned disable:1; /** Partially binned, disable this command */
86 unsigned opaque:1; /** Is opaque */
87 unsigned pad0:29; /* wasted space */
88 unsigned stride; /* how much to advance data between a0, dadx, dady */
89 unsigned pad2; /* wasted space */
90 unsigned pad3; /* wasted space */
91 /* followed by a0, dadx, dady and planes[] */
92 };
93
94 /* Note: the order of these values is important as they are loaded by
95 * sse code in rasterization:
96 */
97 struct lp_rast_plane {
98 /* edge function values at minx,miny ?? */
99 int c;
100
101 int dcdx;
102 int dcdy;
103
104 /* one-pixel sized trivial reject offsets for each plane */
105 int eo;
106 };
107
108 /**
109 * Rasterization information for a triangle known to be in this bin,
110 * plus inputs to run the shader:
111 * These fields are tile- and bin-independent.
112 * Objects of this type are put into the lp_setup_context::data buffer.
113 */
114 struct lp_rast_triangle {
115 #ifdef DEBUG
116 float v[3][2];
117 float pad0;
118 float pad1;
119 #endif
120
121 /* inputs for the shader */
122 struct lp_rast_shader_inputs inputs;
123 /* planes are also allocated here */
124 };
125
126
127 #define GET_A0(inputs) ((float (*)[4])((inputs)+1))
128 #define GET_DADX(inputs) ((float (*)[4])((char *)((inputs) + 1) + (inputs)->stride))
129 #define GET_DADY(inputs) ((float (*)[4])((char *)((inputs) + 1) + 2 * (inputs)->stride))
130 #define GET_PLANES(tri) ((struct lp_rast_plane *)((char *)(&(tri)->inputs + 1) + 3 * (tri)->inputs.stride))
131
132
133
134 struct lp_rasterizer *
135 lp_rast_create( unsigned num_threads );
136
137 void
138 lp_rast_destroy( struct lp_rasterizer * );
139
140 unsigned
141 lp_rast_get_num_threads( struct lp_rasterizer * );
142
143 void
144 lp_rast_queue_scene( struct lp_rasterizer *rast,
145 struct lp_scene *scene );
146
147 void
148 lp_rast_finish( struct lp_rasterizer *rast );
149
150
151 union lp_rast_cmd_arg {
152 const struct lp_rast_shader_inputs *shade_tile;
153 struct {
154 const struct lp_rast_triangle *tri;
155 unsigned plane_mask;
156 } triangle;
157 const struct lp_rast_state *set_state;
158 union pipe_color_union clear_color;
159 struct {
160 uint64_t value;
161 uint64_t mask;
162 } clear_zstencil;
163 const struct lp_rast_state *state;
164 struct lp_fence *fence;
165 struct llvmpipe_query *query_obj;
166 };
167
168
169 /* Cast wrappers. Hopefully these compile to noops!
170 */
171 static INLINE union lp_rast_cmd_arg
172 lp_rast_arg_inputs( const struct lp_rast_shader_inputs *shade_tile )
173 {
174 union lp_rast_cmd_arg arg;
175 arg.shade_tile = shade_tile;
176 return arg;
177 }
178
179 static INLINE union lp_rast_cmd_arg
180 lp_rast_arg_triangle( const struct lp_rast_triangle *triangle,
181 unsigned plane_mask)
182 {
183 union lp_rast_cmd_arg arg;
184 arg.triangle.tri = triangle;
185 arg.triangle.plane_mask = plane_mask;
186 return arg;
187 }
188
189 /**
190 * Build argument for a contained triangle.
191 *
192 * All planes are enabled, so instead of the plane mask we pass the upper
193 * left coordinates of the a block that fully encloses the triangle.
194 */
195 static INLINE union lp_rast_cmd_arg
196 lp_rast_arg_triangle_contained( const struct lp_rast_triangle *triangle,
197 unsigned x, unsigned y)
198 {
199 union lp_rast_cmd_arg arg;
200 arg.triangle.tri = triangle;
201 arg.triangle.plane_mask = x | (y << 8);
202 return arg;
203 }
204
205 static INLINE union lp_rast_cmd_arg
206 lp_rast_arg_state( const struct lp_rast_state *state )
207 {
208 union lp_rast_cmd_arg arg;
209 arg.set_state = state;
210 return arg;
211 }
212
213 static INLINE union lp_rast_cmd_arg
214 lp_rast_arg_fence( struct lp_fence *fence )
215 {
216 union lp_rast_cmd_arg arg;
217 arg.fence = fence;
218 return arg;
219 }
220
221
222 static INLINE union lp_rast_cmd_arg
223 lp_rast_arg_clearzs( uint64_t value, uint64_t mask )
224 {
225 union lp_rast_cmd_arg arg;
226 arg.clear_zstencil.value = value;
227 arg.clear_zstencil.mask = mask;
228 return arg;
229 }
230
231
232 static INLINE union lp_rast_cmd_arg
233 lp_rast_arg_query( struct llvmpipe_query *pq )
234 {
235 union lp_rast_cmd_arg arg;
236 arg.query_obj = pq;
237 return arg;
238 }
239
240 static INLINE union lp_rast_cmd_arg
241 lp_rast_arg_null( void )
242 {
243 union lp_rast_cmd_arg arg;
244 arg.set_state = NULL;
245 return arg;
246 }
247
248
249 /**
250 * Binnable Commands.
251 * These get put into bins by the setup code and are called when
252 * the bins are executed.
253 */
254 #define LP_RAST_OP_CLEAR_COLOR 0x0
255 #define LP_RAST_OP_CLEAR_ZSTENCIL 0x1
256 #define LP_RAST_OP_TRIANGLE_1 0x2
257 #define LP_RAST_OP_TRIANGLE_2 0x3
258 #define LP_RAST_OP_TRIANGLE_3 0x4
259 #define LP_RAST_OP_TRIANGLE_4 0x5
260 #define LP_RAST_OP_TRIANGLE_5 0x6
261 #define LP_RAST_OP_TRIANGLE_6 0x7
262 #define LP_RAST_OP_TRIANGLE_7 0x8
263 #define LP_RAST_OP_TRIANGLE_8 0x9
264 #define LP_RAST_OP_TRIANGLE_3_4 0xa
265 #define LP_RAST_OP_TRIANGLE_3_16 0xb
266 #define LP_RAST_OP_TRIANGLE_4_16 0xc
267 #define LP_RAST_OP_SHADE_TILE 0xd
268 #define LP_RAST_OP_SHADE_TILE_OPAQUE 0xe
269 #define LP_RAST_OP_BEGIN_QUERY 0xf
270 #define LP_RAST_OP_END_QUERY 0x10
271 #define LP_RAST_OP_SET_STATE 0x11
272
273 #define LP_RAST_OP_MAX 0x12
274 #define LP_RAST_OP_MASK 0xff
275
276 void
277 lp_debug_bins( struct lp_scene *scene );
278 void
279 lp_debug_draw_bins_by_cmd_length( struct lp_scene *scene );
280 void
281 lp_debug_draw_bins_by_coverage( struct lp_scene *scene );
282
283
284 #endif