tgsi_exec.h revision ff2a0faba068ac8bc891f4a6427ad3e241c5f09f 
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};
94
95#define TGSI_EXEC_NUM_TEMPS       128
96#define TGSI_EXEC_NUM_IMMEDIATES  256
97#define TGSI_EXEC_NUM_TEMP_ARRAYS 8
98
99/*
100 * Locations of various utility registers (_I = Index, _C = Channel)
101 */
102#define TGSI_EXEC_TEMP_00000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
103#define TGSI_EXEC_TEMP_00000000_C   0
104
105#define TGSI_EXEC_TEMP_7FFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
106#define TGSI_EXEC_TEMP_7FFFFFFF_C   1
107
108#define TGSI_EXEC_TEMP_80000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
109#define TGSI_EXEC_TEMP_80000000_C   2
110
111#define TGSI_EXEC_TEMP_FFFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
112#define TGSI_EXEC_TEMP_FFFFFFFF_C   3
113
114#define TGSI_EXEC_TEMP_ONE_I        (TGSI_EXEC_NUM_TEMPS + 1)
115#define TGSI_EXEC_TEMP_ONE_C        0
116
117#define TGSI_EXEC_TEMP_TWO_I        (TGSI_EXEC_NUM_TEMPS + 1)
118#define TGSI_EXEC_TEMP_TWO_C        1
119
120#define TGSI_EXEC_TEMP_128_I        (TGSI_EXEC_NUM_TEMPS + 1)
121#define TGSI_EXEC_TEMP_128_C        2
122
123#define TGSI_EXEC_TEMP_MINUS_128_I  (TGSI_EXEC_NUM_TEMPS + 1)
124#define TGSI_EXEC_TEMP_MINUS_128_C  3
125
126#define TGSI_EXEC_TEMP_KILMASK_I    (TGSI_EXEC_NUM_TEMPS + 2)
127#define TGSI_EXEC_TEMP_KILMASK_C    0
128
129#define TGSI_EXEC_TEMP_OUTPUT_I     (TGSI_EXEC_NUM_TEMPS + 2)
130#define TGSI_EXEC_TEMP_OUTPUT_C     1
131
132#define TGSI_EXEC_TEMP_PRIMITIVE_I  (TGSI_EXEC_NUM_TEMPS + 2)
133#define TGSI_EXEC_TEMP_PRIMITIVE_C  2
134
135#define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 2)
136#define TGSI_EXEC_TEMP_THREE_C      3
137
138#define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
139#define TGSI_EXEC_TEMP_HALF_C       0
140
141/* execution mask, each value is either 0 or ~0 */
142#define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
143#define TGSI_EXEC_MASK_C            1
144
145/* 4 register buffer for various purposes */
146#define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
147#define TGSI_EXEC_NUM_TEMP_R        4
148
149#define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 8)
150#define TGSI_EXEC_NUM_ADDRS         1
151
152/* predicate register */
153#define TGSI_EXEC_TEMP_P0           (TGSI_EXEC_NUM_TEMPS + 9)
154#define TGSI_EXEC_NUM_PREDS         1
155
156#define TGSI_EXEC_NUM_TEMP_EXTRAS   10
157
158
159
160#define TGSI_EXEC_MAX_NESTING  32
161#define TGSI_EXEC_MAX_COND_NESTING  TGSI_EXEC_MAX_NESTING
162#define TGSI_EXEC_MAX_LOOP_NESTING  TGSI_EXEC_MAX_NESTING
163#define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
164#define TGSI_EXEC_MAX_CALL_NESTING  TGSI_EXEC_MAX_NESTING
165
166/* The maximum number of input attributes per vertex. For 2D
167 * input register files, this is the stride between two 1D
168 * arrays.
169 */
170#define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
171
172/* The maximum number of constant vectors per constant buffer.
173 */
174#define TGSI_EXEC_MAX_CONST_BUFFER  4096
175
176/* The maximum number of vertices per primitive */
177#define TGSI_MAX_PRIM_VERTICES 6
178
179/* The maximum number of primitives to be generated */
180#define TGSI_MAX_PRIMITIVES 64
181
182/* The maximum total number of vertices */
183#define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
184
185#define TGSI_MAX_MISC_INPUTS 8
186
187/** function call/activation record */
188struct tgsi_call_record
189{
190   uint CondStackTop;
191   uint LoopStackTop;
192   uint ContStackTop;
193   int SwitchStackTop;
194   int BreakStackTop;
195   uint ReturnAddr;
196};
197
198
199/* Switch-case block state. */
200struct tgsi_switch_record {
201   uint mask;                          /**< execution mask */
202   union tgsi_exec_channel selector;   /**< a value case statements are compared to */
203   uint defaultMask;                   /**< non-execute mask for default case */
204};
205
206
207enum tgsi_break_type {
208   TGSI_EXEC_BREAK_INSIDE_LOOP,
209   TGSI_EXEC_BREAK_INSIDE_SWITCH
210};
211
212
213#define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
214
215
216/**
217 * Run-time virtual machine state for executing TGSI shader.
218 */
219struct tgsi_exec_machine
220{
221   /* Total = program temporaries + internal temporaries
222    */
223   struct tgsi_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS +
224                                       TGSI_EXEC_NUM_TEMP_EXTRAS];
225   struct tgsi_exec_vector       TempArray[TGSI_EXEC_NUM_TEMP_ARRAYS][TGSI_EXEC_NUM_TEMPS];
226
227   float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
228
229   float                         ImmArray[TGSI_EXEC_NUM_IMMEDIATES][4];
230
231   struct tgsi_exec_vector       *Inputs;
232   struct tgsi_exec_vector       *Outputs;
233
234   /* System values */
235   unsigned                      SysSemanticToIndex[TGSI_SEMANTIC_COUNT];
236   float                         SystemValue[TGSI_MAX_MISC_INPUTS][4];
237
238   struct tgsi_exec_vector       *Addrs;
239   struct tgsi_exec_vector       *Predicates;
240
241   struct tgsi_sampler           **Samplers;
242
243   unsigned                      ImmLimit;
244
245   const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
246   unsigned ConstsSize[PIPE_MAX_CONSTANT_BUFFERS];
247
248   const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
249   unsigned                      Processor; /**< TGSI_PROCESSOR_x */
250
251   /* GEOMETRY processor only. */
252   unsigned                      *Primitives;
253   unsigned                       NumOutputs;
254   unsigned                       MaxGeometryShaderOutputs;
255
256   /* FRAGMENT processor only. */
257   const struct tgsi_interp_coef *InterpCoefs;
258   struct tgsi_exec_vector       QuadPos;
259   float                         Face;    /**< +1 if front facing, -1 if back facing */
260
261   /* Conditional execution masks */
262   uint CondMask;  /**< For IF/ELSE/ENDIF */
263   uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
264   uint ContMask;  /**< For loop CONT statements */
265   uint FuncMask;  /**< For function calls */
266   uint ExecMask;  /**< = CondMask & LoopMask */
267
268   /* Current switch-case state. */
269   struct tgsi_switch_record Switch;
270
271   /* Current break type. */
272   enum tgsi_break_type BreakType;
273
274   /** Condition mask stack (for nested conditionals) */
275   uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
276   int CondStackTop;
277
278   /** Loop mask stack (for nested loops) */
279   uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
280   int LoopStackTop;
281
282   /** Loop label stack */
283   uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
284   int LoopLabelStackTop;
285
286   /** Loop continue mask stack (see comments in tgsi_exec.c) */
287   uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
288   int ContStackTop;
289
290   /** Switch case stack */
291   struct tgsi_switch_record SwitchStack[TGSI_EXEC_MAX_SWITCH_NESTING];
292   int SwitchStackTop;
293
294   enum tgsi_break_type BreakStack[TGSI_EXEC_MAX_BREAK_STACK];
295   int BreakStackTop;
296
297   /** Function execution mask stack (for executing subroutine code) */
298   uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
299   int FuncStackTop;
300
301   /** Function call stack for saving/restoring the program counter */
302   struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
303   int CallStackTop;
304
305   struct tgsi_full_instruction *Instructions;
306   uint NumInstructions;
307
308   struct tgsi_full_declaration *Declarations;
309   uint NumDeclarations;
310
311   struct tgsi_declaration_resource Resources[PIPE_MAX_SHADER_RESOURCES];
312
313   boolean UsedGeometryShader;
314};
315
316struct tgsi_exec_machine *
317tgsi_exec_machine_create( void );
318
319void
320tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
321
322
323void
324tgsi_exec_machine_bind_shader(
325   struct tgsi_exec_machine *mach,
326   const struct tgsi_token *tokens,
327   uint numSamplers,
328   struct tgsi_sampler **samplers);
329
330uint
331tgsi_exec_machine_run(
332   struct tgsi_exec_machine *mach );
333
334
335void
336tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
337
338
339boolean
340tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst);
341
342
343static INLINE void
344tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
345{
346   mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
347      mask;
348}
349
350
351/** Set execution mask values prior to executing the shader */
352static INLINE void
353tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
354                   boolean ch0, boolean ch1, boolean ch2, boolean ch3)
355{
356   int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
357   mask[0] = ch0 ? ~0 : 0;
358   mask[1] = ch1 ? ~0 : 0;
359   mask[2] = ch2 ? ~0 : 0;
360   mask[3] = ch3 ? ~0 : 0;
361}
362
363
364extern void
365tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach,
366                               unsigned num_bufs,
367                               const void **bufs,
368                               const unsigned *buf_sizes);
369
370
371static INLINE int
372tgsi_exec_get_shader_param(enum pipe_shader_cap param)
373{
374   switch(param) {
375   case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
376   case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
377   case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
378   case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
379      return INT_MAX;
380   case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
381      return TGSI_EXEC_MAX_NESTING;
382   case PIPE_SHADER_CAP_MAX_INPUTS:
383      return TGSI_EXEC_MAX_INPUT_ATTRIBS;
384   case PIPE_SHADER_CAP_MAX_CONSTS:
385      return TGSI_EXEC_MAX_CONST_BUFFER;
386   case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
387      return PIPE_MAX_CONSTANT_BUFFERS;
388   case PIPE_SHADER_CAP_MAX_TEMPS:
389      return TGSI_EXEC_NUM_TEMPS;
390   case PIPE_SHADER_CAP_MAX_ADDRS:
391      return TGSI_EXEC_NUM_ADDRS;
392   case PIPE_SHADER_CAP_MAX_PREDS:
393      return TGSI_EXEC_NUM_PREDS;
394   case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
395      return 1;
396   case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
397   case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
398   case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
399   case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
400      return 1;
401   case PIPE_SHADER_CAP_SUBROUTINES:
402      return 1;
403   default:
404      return 0;
405   }
406}
407
408#if defined __cplusplus
409} /* extern "C" */
410#endif
411
412#endif /* TGSI_EXEC_H */
413