1 /**************************************************************************
2 *
3 * Copyright 2003 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
29 #ifndef I915_FPC_H
30 #define I915_FPC_H
31
32
33 #include "i915_context.h"
34 #include "i915_reg.h"
35
36 #include "pipe/p_shader_tokens.h"
37
38 #include "tgsi/tgsi_parse.h"
39
40 #define I915_PROGRAM_SIZE 192
41
42 /* Use those indices for pos/face routing, must be >= num of inputs */
43 #define I915_SEMANTIC_POS 100
44 #define I915_SEMANTIC_FACE 101
45
46
47 /**
48 * Program translation state
49 */
50 struct i915_fp_compile {
51 struct i915_fragment_shader *shader; /* the shader we're compiling */
52
53 boolean used_constants[I915_MAX_CONSTANT];
54
55 /** maps TGSI immediate index to constant slot */
56 uint num_immediates;
57 uint immediates_map[I915_MAX_CONSTANT];
58 float immediates[I915_MAX_CONSTANT][4];
59
60 boolean first_instruction;
61
62 uint declarations[I915_PROGRAM_SIZE];
63 uint program[I915_PROGRAM_SIZE];
64
65 uint *csr; /**< Cursor, points into program. */
66
67 uint *decl; /**< Cursor, points into declarations. */
68
69 uint decl_s; /**< flags for which s regs need to be decl'd */
70 uint decl_t; /**< flags for which t regs need to be decl'd */
71
72 uint temp_flag; /**< Tracks temporary regs which are in use */
73 uint utemp_flag; /**< Tracks TYPE_U temporary regs which are in use */
74
75 uint register_phases[16];
76 uint nr_tex_indirect;
77 uint nr_tex_insn;
78 uint nr_alu_insn;
79 uint nr_decl_insn;
80
81 boolean error; /**< Set if i915_program_error() is called */
82 uint NumNativeInstructions;
83 uint NumNativeAluInstructions;
84 uint NumNativeTexInstructions;
85 uint NumNativeTexIndirections;
86 };
87
88
89 /* Having zero and one in here makes the definition of swizzle a lot
90 * easier.
91 */
92 #define UREG_TYPE_SHIFT 29
93 #define UREG_NR_SHIFT 24
94 #define UREG_CHANNEL_X_NEGATE_SHIFT 23
95 #define UREG_CHANNEL_X_SHIFT 20
96 #define UREG_CHANNEL_Y_NEGATE_SHIFT 19
97 #define UREG_CHANNEL_Y_SHIFT 16
98 #define UREG_CHANNEL_Z_NEGATE_SHIFT 15
99 #define UREG_CHANNEL_Z_SHIFT 12
100 #define UREG_CHANNEL_W_NEGATE_SHIFT 11
101 #define UREG_CHANNEL_W_SHIFT 8
102 #define UREG_CHANNEL_ZERO_NEGATE_MBZ 5
103 #define UREG_CHANNEL_ZERO_SHIFT 4
104 #define UREG_CHANNEL_ONE_NEGATE_MBZ 1
105 #define UREG_CHANNEL_ONE_SHIFT 0
106
107 #define UREG_BAD 0xffffffff /* not a valid ureg */
108
109 #define X SRC_X
110 #define Y SRC_Y
111 #define Z SRC_Z
112 #define W SRC_W
113 #define ZERO SRC_ZERO
114 #define ONE SRC_ONE
115
116 /* Construct a ureg:
117 */
118 #define UREG( type, nr ) (((type)<< UREG_TYPE_SHIFT) | \
119 ((nr) << UREG_NR_SHIFT) | \
120 (X << UREG_CHANNEL_X_SHIFT) | \
121 (Y << UREG_CHANNEL_Y_SHIFT) | \
122 (Z << UREG_CHANNEL_Z_SHIFT) | \
123 (W << UREG_CHANNEL_W_SHIFT) | \
124 (ZERO << UREG_CHANNEL_ZERO_SHIFT) | \
125 (ONE << UREG_CHANNEL_ONE_SHIFT))
126
127 #define GET_CHANNEL_SRC( reg, channel ) ((reg<<(channel*4)) & (0xf<<20))
128 #define CHANNEL_SRC( src, channel ) (src>>(channel*4))
129
130 #define GET_UREG_TYPE(reg) (((reg)>>UREG_TYPE_SHIFT)®_TYPE_MASK)
131 #define GET_UREG_NR(reg) (((reg)>>UREG_NR_SHIFT)®_NR_MASK)
132
133
134
135 #define UREG_XYZW_CHANNEL_MASK 0x00ffff00
136
137 /* One neat thing about the UREG representation:
138 */
139 static INLINE int
swizzle(int reg,uint x,uint y,uint z,uint w)140 swizzle(int reg, uint x, uint y, uint z, uint w)
141 {
142 assert(x <= SRC_ONE);
143 assert(y <= SRC_ONE);
144 assert(z <= SRC_ONE);
145 assert(w <= SRC_ONE);
146 return ((reg & ~UREG_XYZW_CHANNEL_MASK) |
147 CHANNEL_SRC(GET_CHANNEL_SRC(reg, x), 0) |
148 CHANNEL_SRC(GET_CHANNEL_SRC(reg, y), 1) |
149 CHANNEL_SRC(GET_CHANNEL_SRC(reg, z), 2) |
150 CHANNEL_SRC(GET_CHANNEL_SRC(reg, w), 3));
151 }
152
153
154 #define A0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
155 #define D0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
156 #define T0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
157 #define A0_SRC0( reg ) (((reg)&UREG_MASK)>>UREG_A0_SRC0_SHIFT_LEFT)
158 #define A1_SRC0( reg ) (((reg)&UREG_MASK)<<UREG_A1_SRC0_SHIFT_RIGHT)
159 #define A1_SRC1( reg ) (((reg)&UREG_MASK)>>UREG_A1_SRC1_SHIFT_LEFT)
160 #define A2_SRC1( reg ) (((reg)&UREG_MASK)<<UREG_A2_SRC1_SHIFT_RIGHT)
161 #define A2_SRC2( reg ) (((reg)&UREG_MASK)>>UREG_A2_SRC2_SHIFT_LEFT)
162
163 /* These are special, and don't have swizzle/negate bits.
164 */
165 #define T0_SAMPLER( reg ) (GET_UREG_NR(reg)<<T0_SAMPLER_NR_SHIFT)
166 #define T1_ADDRESS_REG( reg ) ((GET_UREG_NR(reg)<<T1_ADDRESS_REG_NR_SHIFT) | \
167 (GET_UREG_TYPE(reg)<<T1_ADDRESS_REG_TYPE_SHIFT))
168
169
170 /* Macros for translating UREG's into the various register fields used
171 * by the I915 programmable unit.
172 */
173 #define UREG_A0_DEST_SHIFT_LEFT (UREG_TYPE_SHIFT - A0_DEST_TYPE_SHIFT)
174 #define UREG_A0_SRC0_SHIFT_LEFT (UREG_TYPE_SHIFT - A0_SRC0_TYPE_SHIFT)
175 #define UREG_A1_SRC0_SHIFT_RIGHT (A1_SRC0_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
176 #define UREG_A1_SRC1_SHIFT_LEFT (UREG_TYPE_SHIFT - A1_SRC1_TYPE_SHIFT)
177 #define UREG_A2_SRC1_SHIFT_RIGHT (A2_SRC1_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
178 #define UREG_A2_SRC2_SHIFT_LEFT (UREG_TYPE_SHIFT - A2_SRC2_TYPE_SHIFT)
179
180 #define UREG_MASK 0xffffff00
181 #define UREG_TYPE_NR_MASK ((REG_TYPE_MASK << UREG_TYPE_SHIFT) | \
182 (REG_NR_MASK << UREG_NR_SHIFT))
183
184
185
186
187 /***********************************************************************
188 * Public interface for the compiler
189 */
190 extern void
191 i915_translate_fragment_program( struct i915_context *i915,
192 struct i915_fragment_shader *fs);
193
194
195
196 extern uint i915_get_temp(struct i915_fp_compile *p);
197 extern uint i915_get_utemp(struct i915_fp_compile *p);
198 extern void i915_release_utemps(struct i915_fp_compile *p);
199
200
201 extern uint i915_emit_texld(struct i915_fp_compile *p,
202 uint dest,
203 uint destmask,
204 uint sampler,
205 uint coord,
206 uint op,
207 uint num_coord);
208
209 extern uint i915_emit_arith(struct i915_fp_compile *p,
210 uint op,
211 uint dest,
212 uint mask,
213 uint saturate,
214 uint src0, uint src1, uint src2);
215
216 extern uint i915_emit_decl(struct i915_fp_compile *p,
217 uint type, uint nr, uint d0_flags);
218
219
220 extern uint i915_emit_const1f(struct i915_fp_compile *p, float c0);
221
222 extern uint i915_emit_const2f(struct i915_fp_compile *p,
223 float c0, float c1);
224
225 extern uint i915_emit_const4fv(struct i915_fp_compile *p,
226 const float * c);
227
228 extern uint i915_emit_const4f(struct i915_fp_compile *p,
229 float c0, float c1,
230 float c2, float c3);
231
232
233 /*======================================================================
234 * i915_fpc_translate.c
235 */
236
237 extern void
238 i915_program_error(struct i915_fp_compile *p, const char *msg, ...);
239
240
241 /*======================================================================
242 * i915_fpc_optimize.c
243 */
244
245
246 struct i915_src_register
247 {
248 unsigned File : 4; /* TGSI_FILE_ */
249 unsigned Indirect : 1; /* BOOL */
250 unsigned Dimension : 1; /* BOOL */
251 int Index : 16; /* SINT */
252 unsigned SwizzleX : 3; /* TGSI_SWIZZLE_ */
253 unsigned SwizzleY : 3; /* TGSI_SWIZZLE_ */
254 unsigned SwizzleZ : 3; /* TGSI_SWIZZLE_ */
255 unsigned SwizzleW : 3; /* TGSI_SWIZZLE_ */
256 unsigned Absolute : 1; /* BOOL */
257 unsigned Negate : 1; /* BOOL */
258 };
259
260 /* Additional swizzle supported in i915 */
261 #define TGSI_SWIZZLE_ZERO 4
262 #define TGSI_SWIZZLE_ONE 5
263
264 struct i915_dst_register
265 {
266 unsigned File : 4; /* TGSI_FILE_ */
267 unsigned WriteMask : 4; /* TGSI_WRITEMASK_ */
268 unsigned Indirect : 1; /* BOOL */
269 unsigned Dimension : 1; /* BOOL */
270 int Index : 16; /* SINT */
271 unsigned Padding : 6;
272 };
273
274
275 struct i915_full_dst_register
276 {
277 struct i915_dst_register Register;
278 /*
279 struct tgsi_src_register Indirect;
280 struct tgsi_dimension Dimension;
281 struct tgsi_src_register DimIndirect;
282 */
283 };
284
285 struct i915_full_src_register
286 {
287 struct i915_src_register Register;
288 /*
289 struct tgsi_src_register Indirect;
290 struct tgsi_dimension Dimension;
291 struct tgsi_src_register DimIndirect;
292 */
293 };
294
295 struct i915_full_instruction
296 {
297 struct tgsi_instruction Instruction;
298 /*
299 struct tgsi_instruction_predicate Predicate;
300 struct tgsi_instruction_label Label;
301 */
302 struct tgsi_instruction_texture Texture;
303 struct i915_full_dst_register Dst[1];
304 struct i915_full_src_register Src[3];
305 };
306
307
308 union i915_full_token
309 {
310 struct tgsi_token Token;
311 struct tgsi_full_declaration FullDeclaration;
312 struct tgsi_full_immediate FullImmediate;
313 struct i915_full_instruction FullInstruction;
314 struct tgsi_full_property FullProperty;
315 };
316
317 struct i915_token_list
318 {
319 union i915_full_token* Tokens;
320 unsigned NumTokens;
321 };
322
323 extern struct i915_token_list* i915_optimize(const struct tgsi_token *tokens);
324
325 extern void i915_optimize_free(struct i915_token_list* tokens);
326
327 #endif
328