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
,
101 struct tgsi_image_params
{
103 unsigned tgsi_tex_instr
;
104 enum pipe_format format
;
109 /* image interfaces */
110 void (*load
)(const struct tgsi_image
*image
,
111 const struct tgsi_image_params
*params
,
112 const int s
[TGSI_QUAD_SIZE
],
113 const int t
[TGSI_QUAD_SIZE
],
114 const int r
[TGSI_QUAD_SIZE
],
115 const int sample
[TGSI_QUAD_SIZE
],
116 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
118 void (*store
)(const struct tgsi_image
*image
,
119 const struct tgsi_image_params
*params
,
120 const int s
[TGSI_QUAD_SIZE
],
121 const int t
[TGSI_QUAD_SIZE
],
122 const int r
[TGSI_QUAD_SIZE
],
123 const int sample
[TGSI_QUAD_SIZE
],
124 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
126 void (*op
)(const struct tgsi_image
*image
,
127 const struct tgsi_image_params
*params
,
129 const int s
[TGSI_QUAD_SIZE
],
130 const int t
[TGSI_QUAD_SIZE
],
131 const int r
[TGSI_QUAD_SIZE
],
132 const int sample
[TGSI_QUAD_SIZE
],
133 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
],
134 float rgba2
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
136 void (*get_dims
)(const struct tgsi_image
*image
,
137 const struct tgsi_image_params
*params
,
141 struct tgsi_buffer_params
{
148 /* buffer interfaces */
149 void (*load
)(const struct tgsi_buffer
*buffer
,
150 const struct tgsi_buffer_params
*params
,
151 const int s
[TGSI_QUAD_SIZE
],
152 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
154 void (*store
)(const struct tgsi_buffer
*buffer
,
155 const struct tgsi_buffer_params
*params
,
156 const int s
[TGSI_QUAD_SIZE
],
157 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
159 void (*op
)(const struct tgsi_buffer
*buffer
,
160 const struct tgsi_buffer_params
*params
,
162 const int s
[TGSI_QUAD_SIZE
],
163 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
],
164 float rgba2
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
166 void (*get_dims
)(const struct tgsi_buffer
*buffer
,
167 const struct tgsi_buffer_params
*params
,
172 * Information for sampling textures, which must be implemented
173 * by code outside the TGSI executor.
177 /** Get samples for four fragments in a quad */
178 /* this interface contains 5 sets of channels that vary
179 * depending on the sampler.
180 * s - the first texture coordinate for sampling.
181 * t - the second texture coordinate for sampling - unused for 1D,
183 * r - the third coordinate for sampling for 3D, cube, cube arrays,
184 * layer for 2D arrays. Compare value for 1D/2D shadows.
185 * c0 - Compare value for shadow cube and shadow 2d arrays,
186 * layer for cube arrays.
187 * derivs - explicit derivatives.
188 * offset - texel offsets
189 * lod - lod value, except for shadow cube arrays (compare value there).
191 void (*get_samples
)(struct tgsi_sampler
*sampler
,
192 const unsigned sview_index
,
193 const unsigned sampler_index
,
194 const float s
[TGSI_QUAD_SIZE
],
195 const float t
[TGSI_QUAD_SIZE
],
196 const float r
[TGSI_QUAD_SIZE
],
197 const float c0
[TGSI_QUAD_SIZE
],
198 const float c1
[TGSI_QUAD_SIZE
],
199 float derivs
[3][2][TGSI_QUAD_SIZE
],
200 const int8_t offset
[3],
201 enum tgsi_sampler_control control
,
202 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
203 void (*get_dims
)(struct tgsi_sampler
*sampler
,
204 const unsigned sview_index
,
205 int level
, int dims
[4]);
206 void (*get_texel
)(struct tgsi_sampler
*sampler
,
207 const unsigned sview_index
,
208 const int i
[TGSI_QUAD_SIZE
],
209 const int j
[TGSI_QUAD_SIZE
], const int k
[TGSI_QUAD_SIZE
],
210 const int lod
[TGSI_QUAD_SIZE
], const int8_t offset
[3],
211 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
212 void (*query_lod
)(const struct tgsi_sampler
*tgsi_sampler
,
213 const unsigned sview_index
,
214 const unsigned sampler_index
,
215 const float s
[TGSI_QUAD_SIZE
],
216 const float t
[TGSI_QUAD_SIZE
],
217 const float p
[TGSI_QUAD_SIZE
],
218 const float c0
[TGSI_QUAD_SIZE
],
219 const enum tgsi_sampler_control control
,
220 float mipmap
[TGSI_QUAD_SIZE
],
221 float lod
[TGSI_QUAD_SIZE
]);
224 #define TGSI_EXEC_NUM_TEMPS 4096
225 #define TGSI_EXEC_NUM_IMMEDIATES 256
228 * Locations of various utility registers (_I = Index, _C = Channel)
230 #define TGSI_EXEC_TEMP_00000000_I (TGSI_EXEC_NUM_TEMPS + 0)
231 #define TGSI_EXEC_TEMP_00000000_C 0
233 #define TGSI_EXEC_TEMP_7FFFFFFF_I (TGSI_EXEC_NUM_TEMPS + 0)
234 #define TGSI_EXEC_TEMP_7FFFFFFF_C 1
236 #define TGSI_EXEC_TEMP_80000000_I (TGSI_EXEC_NUM_TEMPS + 0)
237 #define TGSI_EXEC_TEMP_80000000_C 2
239 #define TGSI_EXEC_TEMP_FFFFFFFF_I (TGSI_EXEC_NUM_TEMPS + 0)
240 #define TGSI_EXEC_TEMP_FFFFFFFF_C 3
242 #define TGSI_EXEC_TEMP_ONE_I (TGSI_EXEC_NUM_TEMPS + 1)
243 #define TGSI_EXEC_TEMP_ONE_C 0
245 #define TGSI_EXEC_TEMP_TWO_I (TGSI_EXEC_NUM_TEMPS + 1)
246 #define TGSI_EXEC_TEMP_TWO_C 1
248 #define TGSI_EXEC_TEMP_128_I (TGSI_EXEC_NUM_TEMPS + 1)
249 #define TGSI_EXEC_TEMP_128_C 2
251 #define TGSI_EXEC_TEMP_MINUS_128_I (TGSI_EXEC_NUM_TEMPS + 1)
252 #define TGSI_EXEC_TEMP_MINUS_128_C 3
254 #define TGSI_EXEC_TEMP_KILMASK_I (TGSI_EXEC_NUM_TEMPS + 2)
255 #define TGSI_EXEC_TEMP_KILMASK_C 0
257 #define TGSI_EXEC_TEMP_OUTPUT_I (TGSI_EXEC_NUM_TEMPS + 2)
258 #define TGSI_EXEC_TEMP_OUTPUT_C 1
260 #define TGSI_EXEC_TEMP_PRIMITIVE_I (TGSI_EXEC_NUM_TEMPS + 2)
261 #define TGSI_EXEC_TEMP_PRIMITIVE_C 2
263 #define TGSI_EXEC_TEMP_THREE_I (TGSI_EXEC_NUM_TEMPS + 2)
264 #define TGSI_EXEC_TEMP_THREE_C 3
266 #define TGSI_EXEC_TEMP_HALF_I (TGSI_EXEC_NUM_TEMPS + 3)
267 #define TGSI_EXEC_TEMP_HALF_C 0
269 /* 4 register buffer for various purposes */
270 #define TGSI_EXEC_TEMP_R0 (TGSI_EXEC_NUM_TEMPS + 4)
271 #define TGSI_EXEC_NUM_TEMP_R 4
273 #define TGSI_EXEC_TEMP_ADDR (TGSI_EXEC_NUM_TEMPS + 8)
274 #define TGSI_EXEC_NUM_ADDRS 3
276 #define TGSI_EXEC_NUM_TEMP_EXTRAS 12
280 #define TGSI_EXEC_MAX_NESTING 32
281 #define TGSI_EXEC_MAX_COND_NESTING TGSI_EXEC_MAX_NESTING
282 #define TGSI_EXEC_MAX_LOOP_NESTING TGSI_EXEC_MAX_NESTING
283 #define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
284 #define TGSI_EXEC_MAX_CALL_NESTING TGSI_EXEC_MAX_NESTING
286 /* The maximum number of input attributes per vertex. For 2D
287 * input register files, this is the stride between two 1D
290 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 32
292 /* The maximum number of bytes per constant buffer.
294 #define TGSI_EXEC_MAX_CONST_BUFFER_SIZE (4096 * sizeof(float[4]))
296 /* The maximum number of vertices per primitive */
297 #define TGSI_MAX_PRIM_VERTICES 6
299 /* The maximum number of primitives to be generated */
300 #define TGSI_MAX_PRIMITIVES 64
302 /* The maximum total number of vertices */
303 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
305 #define TGSI_MAX_MISC_INPUTS 8
307 /** function call/activation record */
308 struct tgsi_call_record
319 /* Switch-case block state. */
320 struct tgsi_switch_record
{
321 uint mask
; /**< execution mask */
322 union tgsi_exec_channel selector
; /**< a value case statements are compared to */
323 uint defaultMask
; /**< non-execute mask for default case */
327 enum tgsi_break_type
{
328 TGSI_EXEC_BREAK_INSIDE_LOOP
,
329 TGSI_EXEC_BREAK_INSIDE_SWITCH
333 #define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
337 * Run-time virtual machine state for executing TGSI shader.
339 struct tgsi_exec_machine
341 /* Total = program temporaries + internal temporaries
343 struct tgsi_exec_vector Temps
[TGSI_EXEC_NUM_TEMPS
+
344 TGSI_EXEC_NUM_TEMP_EXTRAS
];
346 float Imms
[TGSI_EXEC_NUM_IMMEDIATES
][4];
348 float ImmArray
[TGSI_EXEC_NUM_IMMEDIATES
][4];
350 struct tgsi_exec_vector
*Inputs
;
351 struct tgsi_exec_vector
*Outputs
;
354 unsigned SysSemanticToIndex
[TGSI_SEMANTIC_COUNT
];
355 struct tgsi_exec_vector SystemValue
[TGSI_MAX_MISC_INPUTS
];
357 struct tgsi_exec_vector
*Addrs
;
359 struct tgsi_sampler
*Sampler
;
361 struct tgsi_image
*Image
;
362 struct tgsi_buffer
*Buffer
;
365 const void *Consts
[PIPE_MAX_CONSTANT_BUFFERS
];
366 unsigned ConstsSize
[PIPE_MAX_CONSTANT_BUFFERS
];
368 const struct tgsi_token
*Tokens
; /**< Declarations, instructions */
369 enum pipe_shader_type ShaderType
; /**< PIPE_SHADER_x */
371 /* GEOMETRY processor only. */
372 unsigned *Primitives
;
374 unsigned MaxGeometryShaderOutputs
;
375 unsigned MaxOutputVertices
;
377 /* FRAGMENT processor only. */
378 const struct tgsi_interp_coef
*InterpCoefs
;
379 struct tgsi_exec_vector QuadPos
;
380 float Face
; /**< +1 if front facing, -1 if back facing */
381 bool flatshade_color
;
385 unsigned LocalMemSize
;
387 /* See GLSL 4.50 specification for definition of helper invocations */
388 uint NonHelperMask
; /**< non-helpers */
389 /* Conditional execution masks */
390 uint CondMask
; /**< For IF/ELSE/ENDIF */
391 uint LoopMask
; /**< For BGNLOOP/ENDLOOP */
392 uint ContMask
; /**< For loop CONT statements */
393 uint FuncMask
; /**< For function calls */
394 uint ExecMask
; /**< = CondMask & LoopMask */
396 /* Current switch-case state. */
397 struct tgsi_switch_record Switch
;
399 /* Current break type. */
400 enum tgsi_break_type BreakType
;
402 /** Condition mask stack (for nested conditionals) */
403 uint CondStack
[TGSI_EXEC_MAX_COND_NESTING
];
406 /** Loop mask stack (for nested loops) */
407 uint LoopStack
[TGSI_EXEC_MAX_LOOP_NESTING
];
410 /** Loop label stack */
411 uint LoopLabelStack
[TGSI_EXEC_MAX_LOOP_NESTING
];
412 int LoopLabelStackTop
;
414 /** Loop continue mask stack (see comments in tgsi_exec.c) */
415 uint ContStack
[TGSI_EXEC_MAX_LOOP_NESTING
];
418 /** Switch case stack */
419 struct tgsi_switch_record SwitchStack
[TGSI_EXEC_MAX_SWITCH_NESTING
];
422 enum tgsi_break_type BreakStack
[TGSI_EXEC_MAX_BREAK_STACK
];
425 /** Function execution mask stack (for executing subroutine code) */
426 uint FuncStack
[TGSI_EXEC_MAX_CALL_NESTING
];
429 /** Function call stack for saving/restoring the program counter */
430 struct tgsi_call_record CallStack
[TGSI_EXEC_MAX_CALL_NESTING
];
433 struct tgsi_full_instruction
*Instructions
;
434 uint NumInstructions
;
436 struct tgsi_full_declaration
*Declarations
;
437 uint NumDeclarations
;
439 struct tgsi_declaration_sampler_view
440 SamplerViews
[PIPE_MAX_SHADER_SAMPLER_VIEWS
];
442 boolean UsedGeometryShader
;
447 struct tgsi_exec_machine
*
448 tgsi_exec_machine_create(enum pipe_shader_type shader_type
);
451 tgsi_exec_machine_destroy(struct tgsi_exec_machine
*mach
);
455 tgsi_exec_machine_bind_shader(
456 struct tgsi_exec_machine
*mach
,
457 const struct tgsi_token
*tokens
,
458 struct tgsi_sampler
*sampler
,
459 struct tgsi_image
*image
,
460 struct tgsi_buffer
*buffer
);
463 tgsi_exec_machine_run(
464 struct tgsi_exec_machine
*mach
, int start_pc
);
468 tgsi_exec_machine_free_data(struct tgsi_exec_machine
*mach
);
472 tgsi_check_soa_dependencies(const struct tgsi_full_instruction
*inst
);
476 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine
*mach
,
479 const unsigned *buf_sizes
);
483 tgsi_exec_get_shader_param(enum pipe_shader_cap param
)
486 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
487 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
488 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
489 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
491 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
492 return TGSI_EXEC_MAX_NESTING
;
493 case PIPE_SHADER_CAP_MAX_INPUTS
:
494 return TGSI_EXEC_MAX_INPUT_ATTRIBS
;
495 case PIPE_SHADER_CAP_MAX_OUTPUTS
:
497 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE
:
498 return TGSI_EXEC_MAX_CONST_BUFFER_SIZE
;
499 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
500 return PIPE_MAX_CONSTANT_BUFFERS
;
501 case PIPE_SHADER_CAP_MAX_TEMPS
:
502 return TGSI_EXEC_NUM_TEMPS
;
503 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
505 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR
:
506 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR
:
507 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR
:
508 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR
:
510 case PIPE_SHADER_CAP_SUBROUTINES
:
512 case PIPE_SHADER_CAP_INTEGERS
:
514 case PIPE_SHADER_CAP_FP16
:
516 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
517 return PIPE_MAX_SAMPLERS
;
518 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS
:
519 return PIPE_MAX_SHADER_SAMPLER_VIEWS
;
520 case PIPE_SHADER_CAP_PREFERRED_IR
:
521 return PIPE_SHADER_IR_TGSI
;
522 case PIPE_SHADER_CAP_SUPPORTED_IRS
:
523 return 1 << PIPE_SHADER_IR_TGSI
;
524 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED
:
526 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED
:
527 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE
:
529 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED
:
530 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED
:
531 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD
:
532 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS
:
534 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS
:
535 return PIPE_MAX_SHADER_BUFFERS
;
536 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES
:
537 return PIPE_MAX_SHADER_IMAGES
;
539 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT
:
542 /* if we get here, we missed a shader cap above (and should have seen
543 * a compiler warning.)
548 #if defined __cplusplus
552 #endif /* TGSI_EXEC_H */