tgsi_exec.h revision 2083a276eb270b748d1c2668eb9faa5aadc8e700 
1/**************************************************************************
2 *
3 * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
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 TUNGSTEN GRAPHICS 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 "pipe/p_compiler.h"
33#include "pipe/p_state.h"
34#include "pipe/p_shader_tokens.h"
35
36#if defined __cplusplus
37extern "C" {
38#endif
39
40
41#define NUM_CHANNELS 4  /* R,G,B,A */
42#define QUAD_SIZE    4  /* 4 pixel/quad */
43
44
45/**
46  * Registers may be treated as float, signed int or unsigned int.
47  */
48union tgsi_exec_channel
49{
50   float    f[QUAD_SIZE];
51   int      i[QUAD_SIZE];
52   unsigned u[QUAD_SIZE];
53};
54
55/**
56  * A vector[RGBA] of channels[4 pixels]
57  */
58struct tgsi_exec_vector
59{
60   union tgsi_exec_channel xyzw[NUM_CHANNELS];
61};
62
63/**
64 * For fragment programs, information for computing fragment input
65 * values from plane equation of the triangle/line.
66 */
67struct tgsi_interp_coef
68{
69   float a0[NUM_CHANNELS];	/* in an xyzw layout */
70   float dadx[NUM_CHANNELS];
71   float dady[NUM_CHANNELS];
72};
73
74enum tgsi_sampler_control {
75   tgsi_sampler_lod_bias,
76   tgsi_sampler_lod_explicit
77};
78
79/**
80 * Information for sampling textures, which must be implemented
81 * by code outside the TGSI executor.
82 */
83struct tgsi_sampler
84{
85   /** Get samples for four fragments in a quad */
86   void (*get_samples)(struct tgsi_sampler *sampler,
87                       const float s[QUAD_SIZE],
88                       const float t[QUAD_SIZE],
89                       const float p[QUAD_SIZE],
90                       const float c0[QUAD_SIZE],
91                       enum tgsi_sampler_control control,
92                       float rgba[NUM_CHANNELS][QUAD_SIZE]);
93   void (*get_dims)(struct tgsi_sampler *sampler, int level,
94		    int dims[4]);
95   void (*get_texel)(struct tgsi_sampler *sampler, const int i[QUAD_SIZE],
96		     const int j[QUAD_SIZE], const int k[QUAD_SIZE],
97		     const int lod[QUAD_SIZE], const int8_t offset[3],
98		     float rgba[NUM_CHANNELS][QUAD_SIZE]);
99};
100
101#define TGSI_EXEC_NUM_TEMPS       128
102#define TGSI_EXEC_NUM_IMMEDIATES  256
103#define TGSI_EXEC_NUM_TEMP_ARRAYS 8
104
105/*
106 * Locations of various utility registers (_I = Index, _C = Channel)
107 */
108#define TGSI_EXEC_TEMP_00000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
109#define TGSI_EXEC_TEMP_00000000_C   0
110
111#define TGSI_EXEC_TEMP_7FFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
112#define TGSI_EXEC_TEMP_7FFFFFFF_C   1
113
114#define TGSI_EXEC_TEMP_80000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
115#define TGSI_EXEC_TEMP_80000000_C   2
116
117#define TGSI_EXEC_TEMP_FFFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
118#define TGSI_EXEC_TEMP_FFFFFFFF_C   3
119
120#define TGSI_EXEC_TEMP_ONE_I        (TGSI_EXEC_NUM_TEMPS + 1)
121#define TGSI_EXEC_TEMP_ONE_C        0
122
123#define TGSI_EXEC_TEMP_TWO_I        (TGSI_EXEC_NUM_TEMPS + 1)
124#define TGSI_EXEC_TEMP_TWO_C        1
125
126#define TGSI_EXEC_TEMP_128_I        (TGSI_EXEC_NUM_TEMPS + 1)
127#define TGSI_EXEC_TEMP_128_C        2
128
129#define TGSI_EXEC_TEMP_MINUS_128_I  (TGSI_EXEC_NUM_TEMPS + 1)
130#define TGSI_EXEC_TEMP_MINUS_128_C  3
131
132#define TGSI_EXEC_TEMP_KILMASK_I    (TGSI_EXEC_NUM_TEMPS + 2)
133#define TGSI_EXEC_TEMP_KILMASK_C    0
134
135#define TGSI_EXEC_TEMP_OUTPUT_I     (TGSI_EXEC_NUM_TEMPS + 2)
136#define TGSI_EXEC_TEMP_OUTPUT_C     1
137
138#define TGSI_EXEC_TEMP_PRIMITIVE_I  (TGSI_EXEC_NUM_TEMPS + 2)
139#define TGSI_EXEC_TEMP_PRIMITIVE_C  2
140
141#define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 2)
142#define TGSI_EXEC_TEMP_THREE_C      3
143
144#define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
145#define TGSI_EXEC_TEMP_HALF_C       0
146
147/* execution mask, each value is either 0 or ~0 */
148#define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
149#define TGSI_EXEC_MASK_C            1
150
151/* 4 register buffer for various purposes */
152#define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
153#define TGSI_EXEC_NUM_TEMP_R        4
154
155#define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 8)
156#define TGSI_EXEC_NUM_ADDRS         1
157
158/* predicate register */
159#define TGSI_EXEC_TEMP_P0           (TGSI_EXEC_NUM_TEMPS + 9)
160#define TGSI_EXEC_NUM_PREDS         1
161
162#define TGSI_EXEC_NUM_TEMP_EXTRAS   10
163
164
165
166#define TGSI_EXEC_MAX_NESTING  32
167#define TGSI_EXEC_MAX_COND_NESTING  TGSI_EXEC_MAX_NESTING
168#define TGSI_EXEC_MAX_LOOP_NESTING  TGSI_EXEC_MAX_NESTING
169#define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
170#define TGSI_EXEC_MAX_CALL_NESTING  TGSI_EXEC_MAX_NESTING
171
172/* The maximum number of input attributes per vertex. For 2D
173 * input register files, this is the stride between two 1D
174 * arrays.
175 */
176#define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
177
178/* The maximum number of constant vectors per constant buffer.
179 */
180#define TGSI_EXEC_MAX_CONST_BUFFER  4096
181
182/* The maximum number of vertices per primitive */
183#define TGSI_MAX_PRIM_VERTICES 6
184
185/* The maximum number of primitives to be generated */
186#define TGSI_MAX_PRIMITIVES 64
187
188/* The maximum total number of vertices */
189#define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
190
191#define TGSI_MAX_MISC_INPUTS 8
192
193/** function call/activation record */
194struct tgsi_call_record
195{
196   uint CondStackTop;
197   uint LoopStackTop;
198   uint ContStackTop;
199   int SwitchStackTop;
200   int BreakStackTop;
201   uint ReturnAddr;
202};
203
204
205/* Switch-case block state. */
206struct tgsi_switch_record {
207   uint mask;                          /**< execution mask */
208   union tgsi_exec_channel selector;   /**< a value case statements are compared to */
209   uint defaultMask;                   /**< non-execute mask for default case */
210};
211
212
213enum tgsi_break_type {
214   TGSI_EXEC_BREAK_INSIDE_LOOP,
215   TGSI_EXEC_BREAK_INSIDE_SWITCH
216};
217
218
219#define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
220
221
222/**
223 * Run-time virtual machine state for executing TGSI shader.
224 */
225struct tgsi_exec_machine
226{
227   /* Total = program temporaries + internal temporaries
228    */
229   struct tgsi_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS +
230                                       TGSI_EXEC_NUM_TEMP_EXTRAS];
231   struct tgsi_exec_vector       TempArray[TGSI_EXEC_NUM_TEMP_ARRAYS][TGSI_EXEC_NUM_TEMPS];
232
233   float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
234
235   float                         ImmArray[TGSI_EXEC_NUM_IMMEDIATES][4];
236
237   struct tgsi_exec_vector       *Inputs;
238   struct tgsi_exec_vector       *Outputs;
239
240   /* System values */
241   unsigned                      SysSemanticToIndex[TGSI_SEMANTIC_COUNT];
242   float                         SystemValue[TGSI_MAX_MISC_INPUTS][4];
243
244   struct tgsi_exec_vector       *Addrs;
245   struct tgsi_exec_vector       *Predicates;
246
247   struct tgsi_sampler           **Samplers;
248
249   unsigned                      ImmLimit;
250
251   const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
252   unsigned ConstsSize[PIPE_MAX_CONSTANT_BUFFERS];
253
254   const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
255   unsigned                      Processor; /**< TGSI_PROCESSOR_x */
256
257   /* GEOMETRY processor only. */
258   unsigned                      *Primitives;
259   unsigned                       NumOutputs;
260   unsigned                       MaxGeometryShaderOutputs;
261
262   /* FRAGMENT processor only. */
263   const struct tgsi_interp_coef *InterpCoefs;
264   struct tgsi_exec_vector       QuadPos;
265   float                         Face;    /**< +1 if front facing, -1 if back facing */
266
267   /* Conditional execution masks */
268   uint CondMask;  /**< For IF/ELSE/ENDIF */
269   uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
270   uint ContMask;  /**< For loop CONT statements */
271   uint FuncMask;  /**< For function calls */
272   uint ExecMask;  /**< = CondMask & LoopMask */
273
274   /* Current switch-case state. */
275   struct tgsi_switch_record Switch;
276
277   /* Current break type. */
278   enum tgsi_break_type BreakType;
279
280   /** Condition mask stack (for nested conditionals) */
281   uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
282   int CondStackTop;
283
284   /** Loop mask stack (for nested loops) */
285   uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
286   int LoopStackTop;
287
288   /** Loop label stack */
289   uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
290   int LoopLabelStackTop;
291
292   /** Loop continue mask stack (see comments in tgsi_exec.c) */
293   uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
294   int ContStackTop;
295
296   /** Switch case stack */
297   struct tgsi_switch_record SwitchStack[TGSI_EXEC_MAX_SWITCH_NESTING];
298   int SwitchStackTop;
299
300   enum tgsi_break_type BreakStack[TGSI_EXEC_MAX_BREAK_STACK];
301   int BreakStackTop;
302
303   /** Function execution mask stack (for executing subroutine code) */
304   uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
305   int FuncStackTop;
306
307   /** Function call stack for saving/restoring the program counter */
308   struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
309   int CallStackTop;
310
311   struct tgsi_full_instruction *Instructions;
312   uint NumInstructions;
313
314   struct tgsi_full_declaration *Declarations;
315   uint NumDeclarations;
316
317   struct tgsi_declaration_resource Resources[PIPE_MAX_SHADER_RESOURCES];
318
319   boolean UsedGeometryShader;
320};
321
322struct tgsi_exec_machine *
323tgsi_exec_machine_create( void );
324
325void
326tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
327
328
329void
330tgsi_exec_machine_bind_shader(
331   struct tgsi_exec_machine *mach,
332   const struct tgsi_token *tokens,
333   uint numSamplers,
334   struct tgsi_sampler **samplers);
335
336uint
337tgsi_exec_machine_run(
338   struct tgsi_exec_machine *mach );
339
340
341void
342tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
343
344
345boolean
346tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst);
347
348
349static INLINE void
350tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
351{
352   mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
353      mask;
354}
355
356
357/** Set execution mask values prior to executing the shader */
358static INLINE void
359tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
360                   boolean ch0, boolean ch1, boolean ch2, boolean ch3)
361{
362   int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
363   mask[0] = ch0 ? ~0 : 0;
364   mask[1] = ch1 ? ~0 : 0;
365   mask[2] = ch2 ? ~0 : 0;
366   mask[3] = ch3 ? ~0 : 0;
367}
368
369
370extern void
371tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach,
372                               unsigned num_bufs,
373                               const void **bufs,
374                               const unsigned *buf_sizes);
375
376
377static INLINE int
378tgsi_exec_get_shader_param(enum pipe_shader_cap param)
379{
380   switch(param) {
381   case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
382   case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
383   case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
384   case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
385      return INT_MAX;
386   case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
387      return TGSI_EXEC_MAX_NESTING;
388   case PIPE_SHADER_CAP_MAX_INPUTS:
389      return TGSI_EXEC_MAX_INPUT_ATTRIBS;
390   case PIPE_SHADER_CAP_MAX_CONSTS:
391      return TGSI_EXEC_MAX_CONST_BUFFER;
392   case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
393      return PIPE_MAX_CONSTANT_BUFFERS;
394   case PIPE_SHADER_CAP_MAX_TEMPS:
395      return TGSI_EXEC_NUM_TEMPS;
396   case PIPE_SHADER_CAP_MAX_ADDRS:
397      return TGSI_EXEC_NUM_ADDRS;
398   case PIPE_SHADER_CAP_MAX_PREDS:
399      return TGSI_EXEC_NUM_PREDS;
400   case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
401      return 1;
402   case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
403   case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
404   case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
405   case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
406      return 1;
407   case PIPE_SHADER_CAP_SUBROUTINES:
408      return 1;
409   case PIPE_SHADER_CAP_INTEGERS:
410      return 1;
411   default:
412      return 0;
413   }
414}
415
416#if defined __cplusplus
417} /* extern "C" */
418#endif
419
420#endif /* TGSI_EXEC_H */
421