Merge remote branch 'origin/gallium-0.1' into nouveau-gallium-0.1
[mesa.git] / src / gallium / auxiliary / tgsi / tgsi_exec.h
1 /**************************************************************************
2 *
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
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 TUNGSTEN GRAPHICS 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 #if !defined TGSI_EXEC_H
29 #define TGSI_EXEC_H
30
31 #include "pipe/p_compiler.h"
32
33 #if defined __cplusplus
34 extern "C" {
35 #endif
36
37 #define MAX_LABELS 1024
38
39 #define NUM_CHANNELS 4 /* R,G,B,A */
40 #define QUAD_SIZE 4 /* 4 pixel/quad */
41
42 /**
43 * Registers may be treated as float, signed int or unsigned int.
44 */
45 union tgsi_exec_channel
46 {
47 float f[QUAD_SIZE];
48 int i[QUAD_SIZE];
49 unsigned u[QUAD_SIZE];
50 };
51
52 /**
53 * A vector[RGBA] of channels[4 pixels]
54 */
55 struct tgsi_exec_vector
56 {
57 union tgsi_exec_channel xyzw[NUM_CHANNELS];
58 };
59
60 /**
61 * For fragment programs, information for computing fragment input
62 * values from plane equation of the triangle/line.
63 */
64 struct tgsi_interp_coef
65 {
66 float a0[NUM_CHANNELS]; /* in an xyzw layout */
67 float dadx[NUM_CHANNELS];
68 float dady[NUM_CHANNELS];
69 };
70
71
72 struct softpipe_tile_cache; /**< Opaque to TGSI */
73
74 /**
75 * Information for sampling textures, which must be implemented
76 * by code outside the TGSI executor.
77 */
78 struct tgsi_sampler
79 {
80 const struct pipe_sampler_state *state;
81 struct pipe_texture *texture;
82 /** Get samples for four fragments in a quad */
83 void (*get_samples)(struct tgsi_sampler *sampler,
84 const float s[QUAD_SIZE],
85 const float t[QUAD_SIZE],
86 const float p[QUAD_SIZE],
87 float lodbias,
88 float rgba[NUM_CHANNELS][QUAD_SIZE]);
89 void *pipe; /*XXX temporary*/
90 struct softpipe_tile_cache *cache;
91 };
92
93 /**
94 * For branching/calling subroutines.
95 */
96 struct tgsi_exec_labels
97 {
98 unsigned labels[MAX_LABELS][2];
99 unsigned count;
100 };
101
102
103 #define TGSI_EXEC_NUM_TEMPS 128
104 #define TGSI_EXEC_NUM_TEMP_EXTRAS 6
105 #define TGSI_EXEC_NUM_IMMEDIATES 256
106
107 /*
108 * Locations of various utility registers (_I = Index, _C = Channel)
109 */
110 #define TGSI_EXEC_TEMP_00000000_I (TGSI_EXEC_NUM_TEMPS + 0)
111 #define TGSI_EXEC_TEMP_00000000_C 0
112
113 #define TGSI_EXEC_TEMP_7FFFFFFF_I (TGSI_EXEC_NUM_TEMPS + 0)
114 #define TGSI_EXEC_TEMP_7FFFFFFF_C 1
115
116 #define TGSI_EXEC_TEMP_80000000_I (TGSI_EXEC_NUM_TEMPS + 0)
117 #define TGSI_EXEC_TEMP_80000000_C 2
118
119 #define TGSI_EXEC_TEMP_FFFFFFFF_I (TGSI_EXEC_NUM_TEMPS + 0)
120 #define TGSI_EXEC_TEMP_FFFFFFFF_C 3
121
122 #define TGSI_EXEC_TEMP_ONE_I (TGSI_EXEC_NUM_TEMPS + 1)
123 #define TGSI_EXEC_TEMP_ONE_C 0
124
125 #define TGSI_EXEC_TEMP_TWO_I (TGSI_EXEC_NUM_TEMPS + 1)
126 #define TGSI_EXEC_TEMP_TWO_C 1
127
128 #define TGSI_EXEC_TEMP_128_I (TGSI_EXEC_NUM_TEMPS + 1)
129 #define TGSI_EXEC_TEMP_128_C 2
130
131 #define TGSI_EXEC_TEMP_MINUS_128_I (TGSI_EXEC_NUM_TEMPS + 1)
132 #define TGSI_EXEC_TEMP_MINUS_128_C 3
133
134 #define TGSI_EXEC_TEMP_KILMASK_I (TGSI_EXEC_NUM_TEMPS + 2)
135 #define TGSI_EXEC_TEMP_KILMASK_C 0
136
137 #define TGSI_EXEC_TEMP_OUTPUT_I (TGSI_EXEC_NUM_TEMPS + 2)
138 #define TGSI_EXEC_TEMP_OUTPUT_C 1
139
140 #define TGSI_EXEC_TEMP_PRIMITIVE_I (TGSI_EXEC_NUM_TEMPS + 2)
141 #define TGSI_EXEC_TEMP_PRIMITIVE_C 2
142
143 #define TGSI_EXEC_TEMP_THREE_I (TGSI_EXEC_NUM_TEMPS + 2)
144 #define TGSI_EXEC_TEMP_THREE_C 3
145
146 #define TGSI_EXEC_TEMP_HALF_I (TGSI_EXEC_NUM_TEMPS + 3)
147 #define TGSI_EXEC_TEMP_HALF_C 0
148
149 #define TGSI_EXEC_TEMP_R0 (TGSI_EXEC_NUM_TEMPS + 4)
150
151 #define TGSI_EXEC_TEMP_ADDR (TGSI_EXEC_NUM_TEMPS + 5)
152
153
154 #define TGSI_EXEC_MAX_COND_NESTING 20
155 #define TGSI_EXEC_MAX_LOOP_NESTING 20
156 #define TGSI_EXEC_MAX_CALL_NESTING 20
157
158 /**
159 * Run-time virtual machine state for executing TGSI shader.
160 */
161 struct tgsi_exec_machine
162 {
163 /* Total = program temporaries + internal temporaries
164 * + 1 padding to align to 16 bytes
165 */
166 struct tgsi_exec_vector _Temps[TGSI_EXEC_NUM_TEMPS +
167 TGSI_EXEC_NUM_TEMP_EXTRAS + 1];
168
169 /*
170 * This will point to _Temps after aligning to 16B boundary.
171 */
172 struct tgsi_exec_vector *Temps;
173 struct tgsi_exec_vector *Addrs;
174
175 struct tgsi_sampler *Samplers;
176
177 float Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
178 unsigned ImmLimit;
179 const float (*Consts)[4];
180 struct tgsi_exec_vector *Inputs;
181 struct tgsi_exec_vector *Outputs;
182 const struct tgsi_token *Tokens;
183 unsigned Processor;
184
185 /* GEOMETRY processor only. */
186 unsigned *Primitives;
187
188 /* FRAGMENT processor only. */
189 const struct tgsi_interp_coef *InterpCoefs;
190 struct tgsi_exec_vector QuadPos;
191
192 /* Conditional execution masks */
193 uint CondMask; /**< For IF/ELSE/ENDIF */
194 uint LoopMask; /**< For BGNLOOP/ENDLOOP */
195 uint ContMask; /**< For loop CONT statements */
196 uint FuncMask; /**< For function calls */
197 uint ExecMask; /**< = CondMask & LoopMask */
198
199 /** Condition mask stack (for nested conditionals) */
200 uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
201 int CondStackTop;
202
203 /** Loop mask stack (for nested loops) */
204 uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
205 int LoopStackTop;
206
207 /** Loop continue mask stack (see comments in tgsi_exec.c) */
208 uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
209 int ContStackTop;
210
211 /** Function execution mask stack (for executing subroutine code) */
212 uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
213 int FuncStackTop;
214
215 /** Function call stack for saving/restoring the program counter */
216 uint CallStack[TGSI_EXEC_MAX_CALL_NESTING];
217 int CallStackTop;
218
219 struct tgsi_full_instruction *Instructions;
220 uint NumInstructions;
221
222 struct tgsi_full_declaration *Declarations;
223 uint NumDeclarations;
224
225 struct tgsi_exec_labels Labels;
226 };
227
228 void
229 tgsi_exec_machine_init(
230 struct tgsi_exec_machine *mach );
231
232
233 void
234 tgsi_exec_machine_bind_shader(
235 struct tgsi_exec_machine *mach,
236 const struct tgsi_token *tokens,
237 uint numSamplers,
238 struct tgsi_sampler *samplers);
239
240 uint
241 tgsi_exec_machine_run(
242 struct tgsi_exec_machine *mach );
243
244
245 void
246 tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
247
248
249 #if defined __cplusplus
250 } /* extern "C" */
251 #endif
252
253 #endif /* TGSI_EXEC_H */