1 /**************************************************************************
3 * Copyright 2007-2008 VMware, Inc.
5 * Copyright 2009-2010 VMware, Inc. All rights Reserved.
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:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
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.
27 **************************************************************************/
32 #include "pipe/p_compiler.h"
33 #include "pipe/p_state.h"
34 #include "pipe/p_shader_tokens.h"
36 #if defined __cplusplus
45 #define TGSI_NUM_CHANNELS 4 /* R,G,B,A */
46 #define TGSI_QUAD_SIZE 4 /* 4 pixel/quad */
48 #define TGSI_FOR_EACH_CHANNEL( CHAN )\
49 for (CHAN = 0; CHAN < TGSI_NUM_CHANNELS; CHAN++)
51 #define TGSI_IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
52 ((INST)->Dst[0].Register.WriteMask & (1 << (CHAN)))
54 #define TGSI_IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
55 if (TGSI_IS_DST0_CHANNEL_ENABLED( INST, CHAN ))
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 )
63 * Registers may be treated as float, signed int or unsigned int.
65 union tgsi_exec_channel
67 float f
[TGSI_QUAD_SIZE
];
68 int i
[TGSI_QUAD_SIZE
];
69 unsigned u
[TGSI_QUAD_SIZE
];
73 * A vector[RGBA] of channels[4 pixels]
75 struct tgsi_exec_vector
77 union tgsi_exec_channel xyzw
[TGSI_NUM_CHANNELS
];
81 * For fragment programs, information for computing fragment input
82 * values from plane equation of the triangle/line.
84 struct tgsi_interp_coef
86 float a0
[TGSI_NUM_CHANNELS
]; /* in an xyzw layout */
87 float dadx
[TGSI_NUM_CHANNELS
];
88 float dady
[TGSI_NUM_CHANNELS
];
91 enum tgsi_sampler_control
93 TGSI_SAMPLER_LOD_NONE
,
94 TGSI_SAMPLER_LOD_BIAS
,
95 TGSI_SAMPLER_LOD_EXPLICIT
,
96 TGSI_SAMPLER_LOD_ZERO
,
97 TGSI_SAMPLER_DERIVS_EXPLICIT
,
102 * Information for sampling textures, which must be implemented
103 * by code outside the TGSI executor.
107 /** Get samples for four fragments in a quad */
108 /* this interface contains 5 sets of channels that vary
109 * depending on the sampler.
110 * s - the first texture coordinate for sampling.
111 * t - the second texture coordinate for sampling - unused for 1D,
113 * r - the third coordinate for sampling for 3D, cube, cube arrays,
114 * layer for 2D arrays. Compare value for 1D/2D shadows.
115 * c0 - Compare value for shadow cube and shadow 2d arrays,
116 * layer for cube arrays.
117 * derivs - explicit derivatives.
118 * offset - texel offsets
119 * lod - lod value, except for shadow cube arrays (compare value there).
121 void (*get_samples
)(struct tgsi_sampler
*sampler
,
122 const unsigned sview_index
,
123 const unsigned sampler_index
,
124 const float s
[TGSI_QUAD_SIZE
],
125 const float t
[TGSI_QUAD_SIZE
],
126 const float r
[TGSI_QUAD_SIZE
],
127 const float c0
[TGSI_QUAD_SIZE
],
128 const float c1
[TGSI_QUAD_SIZE
],
129 float derivs
[3][2][TGSI_QUAD_SIZE
],
130 const int8_t offset
[3],
131 enum tgsi_sampler_control control
,
132 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
133 void (*get_dims
)(struct tgsi_sampler
*sampler
,
134 const unsigned sview_index
,
135 int level
, int dims
[4]);
136 void (*get_texel
)(struct tgsi_sampler
*sampler
,
137 const unsigned sview_index
,
138 const int i
[TGSI_QUAD_SIZE
],
139 const int j
[TGSI_QUAD_SIZE
], const int k
[TGSI_QUAD_SIZE
],
140 const int lod
[TGSI_QUAD_SIZE
], const int8_t offset
[3],
141 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
142 void (*query_lod
)(const struct tgsi_sampler
*tgsi_sampler
,
143 const unsigned sview_index
,
144 const unsigned sampler_index
,
145 const float s
[TGSI_QUAD_SIZE
],
146 const float t
[TGSI_QUAD_SIZE
],
147 const float p
[TGSI_QUAD_SIZE
],
148 const float c0
[TGSI_QUAD_SIZE
],
149 const enum tgsi_sampler_control control
,
150 float mipmap
[TGSI_QUAD_SIZE
],
151 float lod
[TGSI_QUAD_SIZE
]);
154 #define TGSI_EXEC_NUM_TEMPS 4096
155 #define TGSI_EXEC_NUM_IMMEDIATES 256
158 * Locations of various utility registers (_I = Index, _C = Channel)
160 #define TGSI_EXEC_TEMP_00000000_I (TGSI_EXEC_NUM_TEMPS + 0)
161 #define TGSI_EXEC_TEMP_00000000_C 0
163 #define TGSI_EXEC_TEMP_7FFFFFFF_I (TGSI_EXEC_NUM_TEMPS + 0)
164 #define TGSI_EXEC_TEMP_7FFFFFFF_C 1
166 #define TGSI_EXEC_TEMP_80000000_I (TGSI_EXEC_NUM_TEMPS + 0)
167 #define TGSI_EXEC_TEMP_80000000_C 2
169 #define TGSI_EXEC_TEMP_FFFFFFFF_I (TGSI_EXEC_NUM_TEMPS + 0)
170 #define TGSI_EXEC_TEMP_FFFFFFFF_C 3
172 #define TGSI_EXEC_TEMP_ONE_I (TGSI_EXEC_NUM_TEMPS + 1)
173 #define TGSI_EXEC_TEMP_ONE_C 0
175 #define TGSI_EXEC_TEMP_TWO_I (TGSI_EXEC_NUM_TEMPS + 1)
176 #define TGSI_EXEC_TEMP_TWO_C 1
178 #define TGSI_EXEC_TEMP_128_I (TGSI_EXEC_NUM_TEMPS + 1)
179 #define TGSI_EXEC_TEMP_128_C 2
181 #define TGSI_EXEC_TEMP_MINUS_128_I (TGSI_EXEC_NUM_TEMPS + 1)
182 #define TGSI_EXEC_TEMP_MINUS_128_C 3
184 #define TGSI_EXEC_TEMP_KILMASK_I (TGSI_EXEC_NUM_TEMPS + 2)
185 #define TGSI_EXEC_TEMP_KILMASK_C 0
187 #define TGSI_EXEC_TEMP_OUTPUT_I (TGSI_EXEC_NUM_TEMPS + 2)
188 #define TGSI_EXEC_TEMP_OUTPUT_C 1
190 #define TGSI_EXEC_TEMP_PRIMITIVE_I (TGSI_EXEC_NUM_TEMPS + 2)
191 #define TGSI_EXEC_TEMP_PRIMITIVE_C 2
193 #define TGSI_EXEC_TEMP_THREE_I (TGSI_EXEC_NUM_TEMPS + 2)
194 #define TGSI_EXEC_TEMP_THREE_C 3
196 #define TGSI_EXEC_TEMP_HALF_I (TGSI_EXEC_NUM_TEMPS + 3)
197 #define TGSI_EXEC_TEMP_HALF_C 0
199 /* 4 register buffer for various purposes */
200 #define TGSI_EXEC_TEMP_R0 (TGSI_EXEC_NUM_TEMPS + 4)
201 #define TGSI_EXEC_NUM_TEMP_R 4
203 #define TGSI_EXEC_TEMP_ADDR (TGSI_EXEC_NUM_TEMPS + 8)
204 #define TGSI_EXEC_NUM_ADDRS 3
206 /* predicate register */
207 #define TGSI_EXEC_TEMP_P0 (TGSI_EXEC_NUM_TEMPS + 11)
208 #define TGSI_EXEC_NUM_PREDS 1
210 #define TGSI_EXEC_NUM_TEMP_EXTRAS 12
214 #define TGSI_EXEC_MAX_NESTING 32
215 #define TGSI_EXEC_MAX_COND_NESTING TGSI_EXEC_MAX_NESTING
216 #define TGSI_EXEC_MAX_LOOP_NESTING TGSI_EXEC_MAX_NESTING
217 #define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
218 #define TGSI_EXEC_MAX_CALL_NESTING TGSI_EXEC_MAX_NESTING
220 /* The maximum number of input attributes per vertex. For 2D
221 * input register files, this is the stride between two 1D
224 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 32
226 /* The maximum number of bytes per constant buffer.
228 #define TGSI_EXEC_MAX_CONST_BUFFER_SIZE (4096 * sizeof(float[4]))
230 /* The maximum number of vertices per primitive */
231 #define TGSI_MAX_PRIM_VERTICES 6
233 /* The maximum number of primitives to be generated */
234 #define TGSI_MAX_PRIMITIVES 64
236 /* The maximum total number of vertices */
237 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
239 #define TGSI_MAX_MISC_INPUTS 8
241 /** function call/activation record */
242 struct tgsi_call_record
253 /* Switch-case block state. */
254 struct tgsi_switch_record
{
255 uint mask
; /**< execution mask */
256 union tgsi_exec_channel selector
; /**< a value case statements are compared to */
257 uint defaultMask
; /**< non-execute mask for default case */
261 enum tgsi_break_type
{
262 TGSI_EXEC_BREAK_INSIDE_LOOP
,
263 TGSI_EXEC_BREAK_INSIDE_SWITCH
267 #define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
271 * Run-time virtual machine state for executing TGSI shader.
273 struct tgsi_exec_machine
275 /* Total = program temporaries + internal temporaries
277 struct tgsi_exec_vector Temps
[TGSI_EXEC_NUM_TEMPS
+
278 TGSI_EXEC_NUM_TEMP_EXTRAS
];
280 float Imms
[TGSI_EXEC_NUM_IMMEDIATES
][4];
282 float ImmArray
[TGSI_EXEC_NUM_IMMEDIATES
][4];
284 struct tgsi_exec_vector
*Inputs
;
285 struct tgsi_exec_vector
*Outputs
;
288 unsigned SysSemanticToIndex
[TGSI_SEMANTIC_COUNT
];
289 union tgsi_exec_channel SystemValue
[TGSI_MAX_MISC_INPUTS
];
291 struct tgsi_exec_vector
*Addrs
;
292 struct tgsi_exec_vector
*Predicates
;
294 struct tgsi_sampler
*Sampler
;
298 const void *Consts
[PIPE_MAX_CONSTANT_BUFFERS
];
299 unsigned ConstsSize
[PIPE_MAX_CONSTANT_BUFFERS
];
301 const struct tgsi_token
*Tokens
; /**< Declarations, instructions */
302 unsigned Processor
; /**< TGSI_PROCESSOR_x */
304 /* GEOMETRY processor only. */
305 unsigned *Primitives
;
307 unsigned MaxGeometryShaderOutputs
;
308 unsigned MaxOutputVertices
;
310 /* FRAGMENT processor only. */
311 const struct tgsi_interp_coef
*InterpCoefs
;
312 struct tgsi_exec_vector QuadPos
;
313 float Face
; /**< +1 if front facing, -1 if back facing */
314 bool flatshade_color
;
316 /* See GLSL 4.50 specification for definition of helper invocations */
317 uint NonHelperMask
; /**< non-helpers */
318 /* Conditional execution masks */
319 uint CondMask
; /**< For IF/ELSE/ENDIF */
320 uint LoopMask
; /**< For BGNLOOP/ENDLOOP */
321 uint ContMask
; /**< For loop CONT statements */
322 uint FuncMask
; /**< For function calls */
323 uint ExecMask
; /**< = CondMask & LoopMask */
325 /* Current switch-case state. */
326 struct tgsi_switch_record Switch
;
328 /* Current break type. */
329 enum tgsi_break_type BreakType
;
331 /** Condition mask stack (for nested conditionals) */
332 uint CondStack
[TGSI_EXEC_MAX_COND_NESTING
];
335 /** Loop mask stack (for nested loops) */
336 uint LoopStack
[TGSI_EXEC_MAX_LOOP_NESTING
];
339 /** Loop label stack */
340 uint LoopLabelStack
[TGSI_EXEC_MAX_LOOP_NESTING
];
341 int LoopLabelStackTop
;
343 /** Loop continue mask stack (see comments in tgsi_exec.c) */
344 uint ContStack
[TGSI_EXEC_MAX_LOOP_NESTING
];
347 /** Switch case stack */
348 struct tgsi_switch_record SwitchStack
[TGSI_EXEC_MAX_SWITCH_NESTING
];
351 enum tgsi_break_type BreakStack
[TGSI_EXEC_MAX_BREAK_STACK
];
354 /** Function execution mask stack (for executing subroutine code) */
355 uint FuncStack
[TGSI_EXEC_MAX_CALL_NESTING
];
358 /** Function call stack for saving/restoring the program counter */
359 struct tgsi_call_record CallStack
[TGSI_EXEC_MAX_CALL_NESTING
];
362 struct tgsi_full_instruction
*Instructions
;
363 uint NumInstructions
;
365 struct tgsi_full_declaration
*Declarations
;
366 uint NumDeclarations
;
368 struct tgsi_declaration_sampler_view
369 SamplerViews
[PIPE_MAX_SHADER_SAMPLER_VIEWS
];
371 boolean UsedGeometryShader
;
374 struct tgsi_exec_machine
*
375 tgsi_exec_machine_create( void );
378 tgsi_exec_machine_destroy(struct tgsi_exec_machine
*mach
);
382 tgsi_exec_machine_bind_shader(
383 struct tgsi_exec_machine
*mach
,
384 const struct tgsi_token
*tokens
,
385 struct tgsi_sampler
*sampler
);
388 tgsi_exec_machine_run(
389 struct tgsi_exec_machine
*mach
);
393 tgsi_exec_machine_free_data(struct tgsi_exec_machine
*mach
);
397 tgsi_check_soa_dependencies(const struct tgsi_full_instruction
*inst
);
401 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine
*mach
,
404 const unsigned *buf_sizes
);
408 tgsi_exec_get_shader_param(enum pipe_shader_cap param
)
411 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
412 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
413 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
414 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
416 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
417 return TGSI_EXEC_MAX_NESTING
;
418 case PIPE_SHADER_CAP_MAX_INPUTS
:
419 return TGSI_EXEC_MAX_INPUT_ATTRIBS
;
420 case PIPE_SHADER_CAP_MAX_OUTPUTS
:
422 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE
:
423 return TGSI_EXEC_MAX_CONST_BUFFER_SIZE
;
424 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
425 return PIPE_MAX_CONSTANT_BUFFERS
;
426 case PIPE_SHADER_CAP_MAX_TEMPS
:
427 return TGSI_EXEC_NUM_TEMPS
;
428 case PIPE_SHADER_CAP_MAX_PREDS
:
429 return TGSI_EXEC_NUM_PREDS
;
430 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
432 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR
:
433 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR
:
434 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR
:
435 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR
:
437 case PIPE_SHADER_CAP_SUBROUTINES
:
439 case PIPE_SHADER_CAP_INTEGERS
:
441 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
442 return PIPE_MAX_SAMPLERS
;
443 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS
:
444 return PIPE_MAX_SHADER_SAMPLER_VIEWS
;
445 case PIPE_SHADER_CAP_PREFERRED_IR
:
446 return PIPE_SHADER_IR_TGSI
;
447 case PIPE_SHADER_CAP_SUPPORTED_IRS
:
448 return 1 << PIPE_SHADER_IR_TGSI
;
449 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED
:
451 case PIPE_SHADER_CAP_DOUBLES
:
452 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED
:
453 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE
:
455 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED
:
456 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED
:
457 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS
:
458 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES
:
460 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT
:
463 /* if we get here, we missed a shader cap above (and should have seen
464 * a compiler warning.)
469 #if defined __cplusplus
473 #endif /* TGSI_EXEC_H */