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 )
61 #define TGSI_IS_DST1_CHANNEL_ENABLED( INST, CHAN )\
62 ((INST)->Dst[1].Register.WriteMask & (1 << (CHAN)))
64 #define TGSI_IF_IS_DST1_CHANNEL_ENABLED( INST, CHAN )\
65 if (TGSI_IS_DST1_CHANNEL_ENABLED( INST, CHAN ))
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 )
72 * Registers may be treated as float, signed int or unsigned int.
74 union tgsi_exec_channel
76 float f
[TGSI_QUAD_SIZE
];
77 int i
[TGSI_QUAD_SIZE
];
78 unsigned u
[TGSI_QUAD_SIZE
];
82 * A vector[RGBA] of channels[4 pixels]
84 struct tgsi_exec_vector
86 union tgsi_exec_channel xyzw
[TGSI_NUM_CHANNELS
];
90 * For fragment programs, information for computing fragment input
91 * values from plane equation of the triangle/line.
93 struct tgsi_interp_coef
95 float a0
[TGSI_NUM_CHANNELS
]; /* in an xyzw layout */
96 float dadx
[TGSI_NUM_CHANNELS
];
97 float dady
[TGSI_NUM_CHANNELS
];
100 enum tgsi_sampler_control
102 TGSI_SAMPLER_LOD_NONE
,
103 TGSI_SAMPLER_LOD_BIAS
,
104 TGSI_SAMPLER_LOD_EXPLICIT
,
105 TGSI_SAMPLER_LOD_ZERO
,
106 TGSI_SAMPLER_DERIVS_EXPLICIT
,
110 struct tgsi_image_params
{
112 unsigned tgsi_tex_instr
;
113 enum pipe_format format
;
118 /* image interfaces */
119 void (*load
)(const struct tgsi_image
*image
,
120 const struct tgsi_image_params
*params
,
121 const int s
[TGSI_QUAD_SIZE
],
122 const int t
[TGSI_QUAD_SIZE
],
123 const int r
[TGSI_QUAD_SIZE
],
124 const int sample
[TGSI_QUAD_SIZE
],
125 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
127 void (*store
)(const struct tgsi_image
*image
,
128 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
]);
135 void (*op
)(const struct tgsi_image
*image
,
136 const struct tgsi_image_params
*params
,
137 enum tgsi_opcode opcode
,
138 const int s
[TGSI_QUAD_SIZE
],
139 const int t
[TGSI_QUAD_SIZE
],
140 const int r
[TGSI_QUAD_SIZE
],
141 const int sample
[TGSI_QUAD_SIZE
],
142 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
],
143 float rgba2
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
145 void (*get_dims
)(const struct tgsi_image
*image
,
146 const struct tgsi_image_params
*params
,
150 struct tgsi_buffer_params
{
157 /* buffer interfaces */
158 void (*load
)(const struct tgsi_buffer
*buffer
,
159 const struct tgsi_buffer_params
*params
,
160 const int s
[TGSI_QUAD_SIZE
],
161 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
163 void (*store
)(const struct tgsi_buffer
*buffer
,
164 const struct tgsi_buffer_params
*params
,
165 const int s
[TGSI_QUAD_SIZE
],
166 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
168 void (*op
)(const struct tgsi_buffer
*buffer
,
169 const struct tgsi_buffer_params
*params
,
170 enum tgsi_opcode opcode
,
171 const int s
[TGSI_QUAD_SIZE
],
172 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
],
173 float rgba2
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
175 void (*get_dims
)(const struct tgsi_buffer
*buffer
,
176 const struct tgsi_buffer_params
*params
,
181 * Information for sampling textures, which must be implemented
182 * by code outside the TGSI executor.
186 /** Get samples for four fragments in a quad */
187 /* this interface contains 5 sets of channels that vary
188 * depending on the sampler.
189 * s - the first texture coordinate for sampling.
190 * t - the second texture coordinate for sampling - unused for 1D,
192 * r - the third coordinate for sampling for 3D, cube, cube arrays,
193 * layer for 2D arrays. Compare value for 1D/2D shadows.
194 * c0 - Compare value for shadow cube and shadow 2d arrays,
195 * layer for cube arrays.
196 * derivs - explicit derivatives.
197 * offset - texel offsets
198 * lod - lod value, except for shadow cube arrays (compare value there).
200 void (*get_samples
)(struct tgsi_sampler
*sampler
,
201 const unsigned sview_index
,
202 const unsigned sampler_index
,
203 const float s
[TGSI_QUAD_SIZE
],
204 const float t
[TGSI_QUAD_SIZE
],
205 const float r
[TGSI_QUAD_SIZE
],
206 const float c0
[TGSI_QUAD_SIZE
],
207 const float c1
[TGSI_QUAD_SIZE
],
208 float derivs
[3][2][TGSI_QUAD_SIZE
],
209 const int8_t offset
[3],
210 enum tgsi_sampler_control control
,
211 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
212 void (*get_dims
)(struct tgsi_sampler
*sampler
,
213 const unsigned sview_index
,
214 int level
, int dims
[4]);
215 void (*get_texel
)(struct tgsi_sampler
*sampler
,
216 const unsigned sview_index
,
217 const int i
[TGSI_QUAD_SIZE
],
218 const int j
[TGSI_QUAD_SIZE
], const int k
[TGSI_QUAD_SIZE
],
219 const int lod
[TGSI_QUAD_SIZE
], const int8_t offset
[3],
220 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
]);
221 void (*query_lod
)(const struct tgsi_sampler
*tgsi_sampler
,
222 const unsigned sview_index
,
223 const unsigned sampler_index
,
224 const float s
[TGSI_QUAD_SIZE
],
225 const float t
[TGSI_QUAD_SIZE
],
226 const float p
[TGSI_QUAD_SIZE
],
227 const float c0
[TGSI_QUAD_SIZE
],
228 const enum tgsi_sampler_control control
,
229 float mipmap
[TGSI_QUAD_SIZE
],
230 float lod
[TGSI_QUAD_SIZE
]);
233 #define TGSI_EXEC_NUM_TEMPS 4096
236 * Locations of various utility registers (_I = Index, _C = Channel)
238 #define TGSI_EXEC_TEMP_00000000_I (TGSI_EXEC_NUM_TEMPS + 0)
239 #define TGSI_EXEC_TEMP_00000000_C 0
241 #define TGSI_EXEC_TEMP_7FFFFFFF_I (TGSI_EXEC_NUM_TEMPS + 0)
242 #define TGSI_EXEC_TEMP_7FFFFFFF_C 1
244 #define TGSI_EXEC_TEMP_80000000_I (TGSI_EXEC_NUM_TEMPS + 0)
245 #define TGSI_EXEC_TEMP_80000000_C 2
247 #define TGSI_EXEC_TEMP_FFFFFFFF_I (TGSI_EXEC_NUM_TEMPS + 0)
248 #define TGSI_EXEC_TEMP_FFFFFFFF_C 3
250 #define TGSI_EXEC_TEMP_ONE_I (TGSI_EXEC_NUM_TEMPS + 1)
251 #define TGSI_EXEC_TEMP_ONE_C 0
253 #define TGSI_EXEC_TEMP_TWO_I (TGSI_EXEC_NUM_TEMPS + 1)
254 #define TGSI_EXEC_TEMP_TWO_C 1
256 #define TGSI_EXEC_TEMP_128_I (TGSI_EXEC_NUM_TEMPS + 1)
257 #define TGSI_EXEC_TEMP_128_C 2
259 #define TGSI_EXEC_TEMP_MINUS_128_I (TGSI_EXEC_NUM_TEMPS + 1)
260 #define TGSI_EXEC_TEMP_MINUS_128_C 3
262 #define TGSI_EXEC_TEMP_KILMASK_I (TGSI_EXEC_NUM_TEMPS + 2)
263 #define TGSI_EXEC_TEMP_KILMASK_C 0
265 #define TGSI_EXEC_TEMP_OUTPUT_I (TGSI_EXEC_NUM_TEMPS + 2)
266 #define TGSI_EXEC_TEMP_OUTPUT_C 1
268 #define TGSI_EXEC_TEMP_PRIMITIVE_I (TGSI_EXEC_NUM_TEMPS + 2)
269 #define TGSI_EXEC_TEMP_PRIMITIVE_C 2
271 #define TGSI_EXEC_TEMP_THREE_I (TGSI_EXEC_NUM_TEMPS + 2)
272 #define TGSI_EXEC_TEMP_THREE_C 3
274 #define TGSI_EXEC_TEMP_HALF_I (TGSI_EXEC_NUM_TEMPS + 3)
275 #define TGSI_EXEC_TEMP_HALF_C 0
277 /* 4 register buffer for various purposes */
278 #define TGSI_EXEC_TEMP_R0 (TGSI_EXEC_NUM_TEMPS + 4)
279 #define TGSI_EXEC_NUM_TEMP_R 4
281 #define TGSI_EXEC_TEMP_ADDR (TGSI_EXEC_NUM_TEMPS + 8)
282 #define TGSI_EXEC_NUM_ADDRS 3
284 #define TGSI_EXEC_NUM_TEMP_EXTRAS 12
288 #define TGSI_EXEC_MAX_NESTING 32
289 #define TGSI_EXEC_MAX_COND_NESTING TGSI_EXEC_MAX_NESTING
290 #define TGSI_EXEC_MAX_LOOP_NESTING TGSI_EXEC_MAX_NESTING
291 #define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
292 #define TGSI_EXEC_MAX_CALL_NESTING TGSI_EXEC_MAX_NESTING
294 /* The maximum number of input attributes per vertex. For 2D
295 * input register files, this is the stride between two 1D
298 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 32
300 /* The maximum number of bytes per constant buffer.
302 #define TGSI_EXEC_MAX_CONST_BUFFER_SIZE (4096 * sizeof(float[4]))
304 /* The maximum number of vertices per primitive */
305 #define TGSI_MAX_PRIM_VERTICES 6
307 /* The maximum number of primitives to be generated */
308 #define TGSI_MAX_PRIMITIVES 64
310 /* The maximum total number of vertices */
311 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
313 #define TGSI_MAX_MISC_INPUTS 8
315 /** function call/activation record */
316 struct tgsi_call_record
327 /* Switch-case block state. */
328 struct tgsi_switch_record
{
329 uint mask
; /**< execution mask */
330 union tgsi_exec_channel selector
; /**< a value case statements are compared to */
331 uint defaultMask
; /**< non-execute mask for default case */
335 enum tgsi_break_type
{
336 TGSI_EXEC_BREAK_INSIDE_LOOP
,
337 TGSI_EXEC_BREAK_INSIDE_SWITCH
341 #define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
343 typedef float float4
[4];
346 * Run-time virtual machine state for executing TGSI shader.
348 struct tgsi_exec_machine
350 /* Total = program temporaries + internal temporaries
352 struct tgsi_exec_vector Temps
[TGSI_EXEC_NUM_TEMPS
+
353 TGSI_EXEC_NUM_TEMP_EXTRAS
];
355 unsigned ImmsReserved
;
358 struct tgsi_exec_vector
*Inputs
;
359 struct tgsi_exec_vector
*Outputs
;
362 unsigned SysSemanticToIndex
[TGSI_SEMANTIC_COUNT
];
363 struct tgsi_exec_vector SystemValue
[TGSI_MAX_MISC_INPUTS
];
365 struct tgsi_exec_vector
*Addrs
;
367 struct tgsi_sampler
*Sampler
;
369 struct tgsi_image
*Image
;
370 struct tgsi_buffer
*Buffer
;
373 const void *Consts
[PIPE_MAX_CONSTANT_BUFFERS
];
374 unsigned ConstsSize
[PIPE_MAX_CONSTANT_BUFFERS
];
376 const struct tgsi_token
*Tokens
; /**< Declarations, instructions */
377 enum pipe_shader_type ShaderType
; /**< PIPE_SHADER_x */
379 /* GEOMETRY processor only. */
380 unsigned *Primitives
;
382 unsigned MaxGeometryShaderOutputs
;
383 unsigned MaxOutputVertices
;
385 /* FRAGMENT processor only. */
386 const struct tgsi_interp_coef
*InterpCoefs
;
387 struct tgsi_exec_vector QuadPos
;
388 float Face
; /**< +1 if front facing, -1 if back facing */
389 bool flatshade_color
;
393 unsigned LocalMemSize
;
395 /* See GLSL 4.50 specification for definition of helper invocations */
396 uint NonHelperMask
; /**< non-helpers */
397 /* Conditional execution masks */
398 uint CondMask
; /**< For IF/ELSE/ENDIF */
399 uint LoopMask
; /**< For BGNLOOP/ENDLOOP */
400 uint ContMask
; /**< For loop CONT statements */
401 uint FuncMask
; /**< For function calls */
402 uint ExecMask
; /**< = CondMask & LoopMask */
404 /* Current switch-case state. */
405 struct tgsi_switch_record Switch
;
407 /* Current break type. */
408 enum tgsi_break_type BreakType
;
410 /** Condition mask stack (for nested conditionals) */
411 uint CondStack
[TGSI_EXEC_MAX_COND_NESTING
];
414 /** Loop mask stack (for nested loops) */
415 uint LoopStack
[TGSI_EXEC_MAX_LOOP_NESTING
];
418 /** Loop label stack */
419 uint LoopLabelStack
[TGSI_EXEC_MAX_LOOP_NESTING
];
420 int LoopLabelStackTop
;
422 /** Loop continue mask stack (see comments in tgsi_exec.c) */
423 uint ContStack
[TGSI_EXEC_MAX_LOOP_NESTING
];
426 /** Switch case stack */
427 struct tgsi_switch_record SwitchStack
[TGSI_EXEC_MAX_SWITCH_NESTING
];
430 enum tgsi_break_type BreakStack
[TGSI_EXEC_MAX_BREAK_STACK
];
433 /** Function execution mask stack (for executing subroutine code) */
434 uint FuncStack
[TGSI_EXEC_MAX_CALL_NESTING
];
437 /** Function call stack for saving/restoring the program counter */
438 struct tgsi_call_record CallStack
[TGSI_EXEC_MAX_CALL_NESTING
];
441 struct tgsi_full_instruction
*Instructions
;
442 uint NumInstructions
;
444 struct tgsi_full_declaration
*Declarations
;
445 uint NumDeclarations
;
447 struct tgsi_declaration_sampler_view
448 SamplerViews
[PIPE_MAX_SHADER_SAMPLER_VIEWS
];
450 boolean UsedGeometryShader
;
455 struct tgsi_exec_machine
*
456 tgsi_exec_machine_create(enum pipe_shader_type shader_type
);
459 tgsi_exec_machine_destroy(struct tgsi_exec_machine
*mach
);
463 tgsi_exec_machine_bind_shader(
464 struct tgsi_exec_machine
*mach
,
465 const struct tgsi_token
*tokens
,
466 struct tgsi_sampler
*sampler
,
467 struct tgsi_image
*image
,
468 struct tgsi_buffer
*buffer
);
471 tgsi_exec_machine_run(
472 struct tgsi_exec_machine
*mach
, int start_pc
);
476 tgsi_exec_machine_free_data(struct tgsi_exec_machine
*mach
);
480 tgsi_check_soa_dependencies(const struct tgsi_full_instruction
*inst
);
484 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine
*mach
,
487 const unsigned *buf_sizes
);
491 tgsi_exec_get_shader_param(enum pipe_shader_cap param
)
494 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
495 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
496 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
497 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
499 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
500 return TGSI_EXEC_MAX_NESTING
;
501 case PIPE_SHADER_CAP_MAX_INPUTS
:
502 return TGSI_EXEC_MAX_INPUT_ATTRIBS
;
503 case PIPE_SHADER_CAP_MAX_OUTPUTS
:
505 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE
:
506 return TGSI_EXEC_MAX_CONST_BUFFER_SIZE
;
507 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
508 return PIPE_MAX_CONSTANT_BUFFERS
;
509 case PIPE_SHADER_CAP_MAX_TEMPS
:
510 return TGSI_EXEC_NUM_TEMPS
;
511 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
513 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR
:
514 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR
:
515 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR
:
516 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR
:
518 case PIPE_SHADER_CAP_SUBROUTINES
:
520 case PIPE_SHADER_CAP_INTEGERS
:
522 case PIPE_SHADER_CAP_INT64_ATOMICS
:
523 case PIPE_SHADER_CAP_FP16
:
525 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
526 return PIPE_MAX_SAMPLERS
;
527 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS
:
528 return PIPE_MAX_SHADER_SAMPLER_VIEWS
;
529 case PIPE_SHADER_CAP_PREFERRED_IR
:
530 return PIPE_SHADER_IR_TGSI
;
531 case PIPE_SHADER_CAP_SUPPORTED_IRS
:
532 return 1 << PIPE_SHADER_IR_TGSI
;
533 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED
:
535 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED
:
536 case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED
:
537 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE
:
539 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED
:
540 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED
:
541 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD
:
542 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS
:
543 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS
:
544 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS
:
546 case PIPE_SHADER_CAP_SCALAR_ISA
:
548 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS
:
549 return PIPE_MAX_SHADER_BUFFERS
;
550 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES
:
551 return PIPE_MAX_SHADER_IMAGES
;
553 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT
:
556 /* if we get here, we missed a shader cap above (and should have seen
557 * a compiler warning.)
562 #if defined __cplusplus
566 #endif /* TGSI_EXEC_H */