1 /**************************************************************************
2 *
3 * Copyright 2007-2008 VMware, Inc.
4 * All Rights Reserved.
5 * Copyright 2009-2010 VMware, Inc. All rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
17 * of the Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 *
27 **************************************************************************/
28
29 #ifndef TGSI_EXEC_H
30 #define TGSI_EXEC_H
31
32 #include "util/compiler.h"
33 #include "pipe/p_state.h"
34 #include "pipe/p_shader_tokens.h"
35
36 #if defined __cplusplus
37 extern "C" {
38 #endif
39
40 #define TGSI_CHAN_X 0
41 #define TGSI_CHAN_Y 1
42 #define TGSI_CHAN_Z 2
43 #define TGSI_CHAN_W 3
44
45 #define TGSI_NUM_CHANNELS 4 /* R,G,B,A */
46 #define TGSI_QUAD_SIZE 4 /* 4 pixel/quad */
47
48 #define TGSI_FOR_EACH_CHANNEL( CHAN )\
49 for (CHAN = 0; CHAN < TGSI_NUM_CHANNELS; CHAN++)
50
51 #define TGSI_IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
52 ((INST)->Dst[0].Register.WriteMask & (1 << (CHAN)))
53
54 #define TGSI_IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
55 if (TGSI_IS_DST0_CHANNEL_ENABLED( INST, CHAN ))
56
57 #define TGSI_FOR_EACH_DST0_ENABLED_CHANNEL( INST, CHAN )\
58 TGSI_FOR_EACH_CHANNEL( CHAN )\
59 TGSI_IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )
60
61 #define TGSI_IS_DST1_CHANNEL_ENABLED( INST, CHAN )\
62 ((INST)->Dst[1].Register.WriteMask & (1 << (CHAN)))
63
64 #define TGSI_IF_IS_DST1_CHANNEL_ENABLED( INST, CHAN )\
65 if (TGSI_IS_DST1_CHANNEL_ENABLED( INST, CHAN ))
66
67 #define TGSI_FOR_EACH_DST1_ENABLED_CHANNEL( INST, CHAN )\
68 TGSI_FOR_EACH_CHANNEL( CHAN )\
69 TGSI_IF_IS_DST1_CHANNEL_ENABLED( INST, CHAN )
70
71 /**
72 * Registers may be treated as float, signed int or unsigned int.
73 */
74 union tgsi_exec_channel
75 {
76 alignas(16)
77 float f[TGSI_QUAD_SIZE];
78 int32_t i[TGSI_QUAD_SIZE];
79 uint32_t u[TGSI_QUAD_SIZE];
80 };
81
82 /**
83 * A vector[RGBA] of channels[4 pixels]
84 */
85 struct tgsi_exec_vector
86 {
87 alignas(16) union tgsi_exec_channel xyzw[TGSI_NUM_CHANNELS];
88 };
89
90 /**
91 * For fragment programs, information for computing fragment input
92 * values from plane equation of the triangle/line.
93 */
94 struct tgsi_interp_coef
95 {
96 float a0[TGSI_NUM_CHANNELS]; /* in an xyzw layout */
97 float dadx[TGSI_NUM_CHANNELS];
98 float dady[TGSI_NUM_CHANNELS];
99 };
100
101 enum tgsi_sampler_control
102 {
103 TGSI_SAMPLER_LOD_NONE,
104 TGSI_SAMPLER_LOD_BIAS,
105 TGSI_SAMPLER_LOD_EXPLICIT,
106 TGSI_SAMPLER_LOD_ZERO,
107 TGSI_SAMPLER_DERIVS_EXPLICIT,
108 TGSI_SAMPLER_GATHER,
109 };
110
111 struct tgsi_image_params {
112 unsigned unit;
113 unsigned tgsi_tex_instr;
114 enum pipe_format format;
115 unsigned execmask;
116 };
117
118 struct tgsi_image {
119 /* image interfaces */
120 void (*load)(const struct tgsi_image *image,
121 const struct tgsi_image_params *params,
122 const int s[TGSI_QUAD_SIZE],
123 const int t[TGSI_QUAD_SIZE],
124 const int r[TGSI_QUAD_SIZE],
125 const int sample[TGSI_QUAD_SIZE],
126 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
127
128 void (*store)(const struct tgsi_image *image,
129 const struct tgsi_image_params *params,
130 const int s[TGSI_QUAD_SIZE],
131 const int t[TGSI_QUAD_SIZE],
132 const int r[TGSI_QUAD_SIZE],
133 const int sample[TGSI_QUAD_SIZE],
134 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
135
136 void (*op)(const struct tgsi_image *image,
137 const struct tgsi_image_params *params,
138 enum tgsi_opcode opcode,
139 const int s[TGSI_QUAD_SIZE],
140 const int t[TGSI_QUAD_SIZE],
141 const int r[TGSI_QUAD_SIZE],
142 const int sample[TGSI_QUAD_SIZE],
143 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
144 float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
145
146 void (*get_dims)(const struct tgsi_image *image,
147 const struct tgsi_image_params *params,
148 int dims[4]);
149 };
150
151 struct tgsi_buffer_params {
152 unsigned unit;
153 unsigned execmask;
154 unsigned writemask;
155 };
156
157 /* SSBO interfaces */
158 struct tgsi_buffer {
159 void *(*lookup)(const struct tgsi_buffer *buffer,
160 uint32_t unit, uint32_t *size);
161 };
162
163 /**
164 * Information for sampling textures, which must be implemented
165 * by code outside the TGSI executor.
166 */
167 struct tgsi_sampler
168 {
169 /** Get samples for four fragments in a quad */
170 /* this interface contains 5 sets of channels that vary
171 * depending on the sampler.
172 * s - the first texture coordinate for sampling.
173 * t - the second texture coordinate for sampling - unused for 1D,
174 layer for 1D arrays.
175 * r - the third coordinate for sampling for 3D, cube, cube arrays,
176 * layer for 2D arrays. Compare value for 1D/2D shadows.
177 * c0 - Compare value for shadow cube and shadow 2d arrays,
178 * layer for cube arrays.
179 * derivs - explicit derivatives.
180 * offset - texel offsets
181 * lod - lod value, except for shadow cube arrays (compare value there).
182 */
183 void (*get_samples)(struct tgsi_sampler *sampler,
184 const unsigned sview_index,
185 const unsigned sampler_index,
186 const float s[TGSI_QUAD_SIZE],
187 const float t[TGSI_QUAD_SIZE],
188 const float r[TGSI_QUAD_SIZE],
189 const float c0[TGSI_QUAD_SIZE],
190 const float c1[TGSI_QUAD_SIZE],
191 float derivs[3][2][TGSI_QUAD_SIZE],
192 const int8_t offset[3],
193 enum tgsi_sampler_control control,
194 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
195 void (*get_dims)(struct tgsi_sampler *sampler,
196 const unsigned sview_index,
197 int level, int dims[4]);
198 void (*get_texel)(struct tgsi_sampler *sampler,
199 const unsigned sview_index,
200 const int i[TGSI_QUAD_SIZE],
201 const int j[TGSI_QUAD_SIZE], const int k[TGSI_QUAD_SIZE],
202 const int lod[TGSI_QUAD_SIZE], const int8_t offset[3],
203 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
204 void (*query_lod)(const struct tgsi_sampler *tgsi_sampler,
205 const unsigned sview_index,
206 const unsigned sampler_index,
207 const float s[TGSI_QUAD_SIZE],
208 const float t[TGSI_QUAD_SIZE],
209 const float p[TGSI_QUAD_SIZE],
210 const float c0[TGSI_QUAD_SIZE],
211 const enum tgsi_sampler_control control,
212 float mipmap[TGSI_QUAD_SIZE],
213 float lod[TGSI_QUAD_SIZE]);
214 };
215
216 #define TGSI_EXEC_NUM_TEMPS 4096
217
218 #define TGSI_EXEC_MAX_NESTING 32
219 #define TGSI_EXEC_MAX_COND_NESTING TGSI_EXEC_MAX_NESTING
220 #define TGSI_EXEC_MAX_LOOP_NESTING TGSI_EXEC_MAX_NESTING
221 #define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
222 #define TGSI_EXEC_MAX_CALL_NESTING TGSI_EXEC_MAX_NESTING
223
224 /* The maximum number of input attributes per vertex. For 2D
225 * input register files, this is the stride between two 1D
226 * arrays.
227 */
228 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 32
229
230 /* The maximum number of bytes per constant buffer.
231 */
232 #define TGSI_EXEC_MAX_CONST_BUFFER_SIZE (4096 * sizeof(float[4]))
233
234 /* The maximum number of vertices per primitive */
235 #define TGSI_MAX_PRIM_VERTICES 6
236
237 /* The maximum number of primitives to be generated */
238 #define TGSI_MAX_PRIMITIVES 64
239
240 /* The maximum total number of vertices */
241 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
242
243 #define TGSI_MAX_MISC_INPUTS 8
244
245 #define TGSI_MAX_VERTEX_STREAMS 4
246
247 /** function call/activation record */
248 struct tgsi_call_record
249 {
250 unsigned CondStackTop;
251 unsigned LoopStackTop;
252 unsigned ContStackTop;
253 int SwitchStackTop;
254 int BreakStackTop;
255 unsigned ReturnAddr;
256 };
257
258 /* should match draw_buffer_info */
259 struct tgsi_exec_consts_info {
260 const void *ptr;
261 unsigned size;
262 };
263
264 /* Switch-case block state. */
265 struct tgsi_switch_record {
266 unsigned mask; /**< execution mask */
267 union tgsi_exec_channel selector; /**< a value case statements are compared to */
268 unsigned defaultMask; /**< non-execute mask for default case */
269 };
270
271
272 enum tgsi_break_type {
273 TGSI_EXEC_BREAK_INSIDE_LOOP,
274 TGSI_EXEC_BREAK_INSIDE_SWITCH
275 };
276
277
278 #define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
279
280 typedef float float4[4];
281
282 struct tgsi_exec_machine;
283
284 typedef void (* apply_sample_offset_func)(
285 const struct tgsi_exec_machine *mach,
286 unsigned attrib,
287 unsigned chan,
288 float ofs_x,
289 float ofs_y,
290 union tgsi_exec_channel *out_chan);
291
292 /**
293 * Run-time virtual machine state for executing TGSI shader.
294 */
295 struct tgsi_exec_machine
296 {
297 /* Total = program temporaries + internal temporaries
298 */
299 alignas(16)
300 struct tgsi_exec_vector Temps[TGSI_EXEC_NUM_TEMPS];
301
302 unsigned ImmsReserved;
303 float4 *Imms;
304
305 struct tgsi_exec_vector *Inputs;
306 struct tgsi_exec_vector *Outputs;
307 apply_sample_offset_func *InputSampleOffsetApply;
308
309 /* System values */
310 unsigned SysSemanticToIndex[TGSI_SEMANTIC_COUNT];
311 struct tgsi_exec_vector SystemValue[TGSI_MAX_MISC_INPUTS];
312
313 struct tgsi_exec_vector Addrs[3];
314
315 struct tgsi_sampler *Sampler;
316
317 struct tgsi_image *Image;
318 struct tgsi_buffer *Buffer;
319 unsigned ImmLimit;
320
321 const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
322 unsigned ConstsSize[PIPE_MAX_CONSTANT_BUFFERS];
323
324 const struct tgsi_token *Tokens; /**< Declarations, instructions */
325 enum pipe_shader_type ShaderType; /**< PIPE_SHADER_x */
326
327 /* GEOMETRY processor only. */
328 /* Number of vertices emitted per emitted primitive. */
329 unsigned *Primitives[TGSI_MAX_VERTEX_STREAMS];
330 /* Offsets in ->Outputs of the primitives' vertex output data */
331 unsigned *PrimitiveOffsets[TGSI_MAX_VERTEX_STREAMS];
332 unsigned NumOutputs;
333 unsigned MaxOutputVertices;
334 /* Offset in ->Outputs for the current vertex to be emitted. */
335 unsigned OutputVertexOffset;
336 /* Number of primitives emitted. */
337 unsigned OutputPrimCount[TGSI_MAX_VERTEX_STREAMS];
338
339 /* FRAGMENT processor only. */
340 const struct tgsi_interp_coef *InterpCoefs;
341 struct tgsi_exec_vector QuadPos;
342 float Face; /**< +1 if front facing, -1 if back facing */
343 bool flatshade_color;
344
345 /* Compute Only */
346 void *LocalMem;
347 unsigned LocalMemSize;
348
349 /* See GLSL 4.50 specification for definition of helper invocations */
350 unsigned NonHelperMask; /**< non-helpers */
351 /* Conditional execution masks */
352 unsigned CondMask; /**< For IF/ELSE/ENDIF */
353 unsigned LoopMask; /**< For BGNLOOP/ENDLOOP */
354 unsigned ContMask; /**< For loop CONT statements */
355 unsigned FuncMask; /**< For function calls */
356 unsigned ExecMask; /**< = CondMask & LoopMask */
357 unsigned KillMask; /**< Mask of channels killed in the current shader execution */
358
359 /* Current switch-case state. */
360 struct tgsi_switch_record Switch;
361
362 /* Current break type. */
363 enum tgsi_break_type BreakType;
364
365 /** Condition mask stack (for nested conditionals) */
366 unsigned CondStack[TGSI_EXEC_MAX_COND_NESTING];
367 int CondStackTop;
368
369 /** Loop mask stack (for nested loops) */
370 unsigned LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
371 int LoopStackTop;
372
373 /** Loop label stack */
374 unsigned LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
375 int LoopLabelStackTop;
376
377 /** Loop continue mask stack (see comments in tgsi_exec.c) */
378 unsigned ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
379 int ContStackTop;
380
381 /** Switch case stack */
382 struct tgsi_switch_record SwitchStack[TGSI_EXEC_MAX_SWITCH_NESTING];
383 int SwitchStackTop;
384
385 enum tgsi_break_type BreakStack[TGSI_EXEC_MAX_BREAK_STACK];
386 int BreakStackTop;
387
388 /** Function execution mask stack (for executing subroutine code) */
389 unsigned FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
390 int FuncStackTop;
391
392 /** Function call stack for saving/restoring the program counter */
393 struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
394 int CallStackTop;
395
396 struct tgsi_full_instruction *Instructions;
397 unsigned NumInstructions;
398
399 struct tgsi_full_declaration *Declarations;
400 unsigned NumDeclarations;
401
402 struct tgsi_declaration_sampler_view
403 SamplerViews[PIPE_MAX_SHADER_SAMPLER_VIEWS];
404
405 bool UsedGeometryShader;
406
407 int pc;
408 };
409
410 struct tgsi_exec_machine *
411 tgsi_exec_machine_create(enum pipe_shader_type shader_type);
412
413 void
414 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
415
416
417 void
418 tgsi_exec_machine_bind_shader(
419 struct tgsi_exec_machine *mach,
420 const struct tgsi_token *tokens,
421 struct tgsi_sampler *sampler,
422 struct tgsi_image *image,
423 struct tgsi_buffer *buffer);
424
425 uint
426 tgsi_exec_machine_run(
427 struct tgsi_exec_machine *mach, int start_pc );
428
429
430 extern void
431 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach,
432 unsigned num_bufs,
433 const struct tgsi_exec_consts_info *bufs);
434
435
436 static inline int
tgsi_exec_get_shader_param(enum pipe_shader_cap param)437 tgsi_exec_get_shader_param(enum pipe_shader_cap param)
438 {
439 switch(param) {
440 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
441 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
442 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
443 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
444 return INT_MAX;
445 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
446 return TGSI_EXEC_MAX_NESTING;
447 case PIPE_SHADER_CAP_MAX_INPUTS:
448 return TGSI_EXEC_MAX_INPUT_ATTRIBS;
449 case PIPE_SHADER_CAP_MAX_OUTPUTS:
450 return 32;
451 case PIPE_SHADER_CAP_MAX_CONST_BUFFER0_SIZE:
452 return TGSI_EXEC_MAX_CONST_BUFFER_SIZE;
453 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
454 return PIPE_MAX_CONSTANT_BUFFERS;
455 case PIPE_SHADER_CAP_MAX_TEMPS:
456 return TGSI_EXEC_NUM_TEMPS;
457 case PIPE_SHADER_CAP_CONT_SUPPORTED:
458 return 1;
459 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
460 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
461 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
462 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
463 return 1;
464 case PIPE_SHADER_CAP_SUBROUTINES:
465 return 1;
466 case PIPE_SHADER_CAP_INTEGERS:
467 return 1;
468 case PIPE_SHADER_CAP_INT64_ATOMICS:
469 case PIPE_SHADER_CAP_FP16:
470 case PIPE_SHADER_CAP_FP16_DERIVATIVES:
471 case PIPE_SHADER_CAP_FP16_CONST_BUFFERS:
472 case PIPE_SHADER_CAP_INT16:
473 case PIPE_SHADER_CAP_GLSL_16BIT_CONSTS:
474 return 0;
475 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
476 return PIPE_MAX_SAMPLERS;
477 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
478 return PIPE_MAX_SHADER_SAMPLER_VIEWS;
479 case PIPE_SHADER_CAP_SUPPORTED_IRS:
480 return 1 << PIPE_SHADER_IR_TGSI;
481 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
482 return 1;
483 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
484 return 1;
485 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
486 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
487 return 0;
488 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
489 return PIPE_MAX_SHADER_BUFFERS;
490 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
491 return PIPE_MAX_SHADER_IMAGES;
492 }
493 /* if we get here, we missed a shader cap above (and should have seen
494 * a compiler warning.)
495 */
496 return 0;
497 }
498
499 #if defined __cplusplus
500 } /* extern "C" */
501 #endif
502
503 #endif /* TGSI_EXEC_H */
504