1 /**************************************************************************
3 * Copyright 2011-2012 Advanced Micro Devices, Inc.
4 * Copyright 2009 VMware, Inc.
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 **************************************************************************/
31 * TGSI to LLVM IR translation.
33 * @author Jose Fonseca <jfonseca@vmware.com>
34 * @author Tom Stellard <thomas.stellard@amd.com>
40 #include "gallivm/lp_bld.h"
41 #include "gallivm/lp_bld_tgsi_action.h"
42 #include "gallivm/lp_bld_limits.h"
43 #include "gallivm/lp_bld_sample.h"
44 #include "lp_bld_type.h"
45 #include "pipe/p_compiler.h"
46 #include "pipe/p_state.h"
47 #include "tgsi/tgsi_exec.h"
48 #include "tgsi/tgsi_scan.h"
49 #include "tgsi/tgsi_info.h"
51 #define LP_CHAN_ALL ~0
53 #define LP_MAX_INSTRUCTIONS 256
55 struct tgsi_full_declaration
;
56 struct tgsi_full_immediate
;
57 struct tgsi_full_instruction
;
58 struct tgsi_full_src_register
;
59 struct tgsi_opcode_info
;
61 struct tgsi_shader_info
;
62 struct lp_build_mask_context
;
64 struct lp_derivatives
;
65 struct lp_build_tgsi_gs_iface
;
68 enum lp_build_tex_modifier
{
69 LP_BLD_TEX_MODIFIER_NONE
= 0,
70 LP_BLD_TEX_MODIFIER_PROJECTED
,
71 LP_BLD_TEX_MODIFIER_LOD_BIAS
,
72 LP_BLD_TEX_MODIFIER_EXPLICIT_LOD
,
73 LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV
,
74 LP_BLD_TEX_MODIFIER_LOD_ZERO
79 * Describe a channel of a register.
82 * - immediate value (i.e. derived from a IMM register)
85 * - undetermined (when .file == TGSI_FILE_NULL)
87 * This is one of the analysis results, and is used to described
88 * the output color in terms of inputs.
90 struct lp_tgsi_channel_info
92 unsigned file
:4; /* TGSI_FILE_* */
93 unsigned swizzle
:3; /* PIPE_SWIZZLE_x */
96 float value
; /* for TGSI_FILE_IMMEDIATE */
102 * Describe a texture sampler interpolator.
104 * The interpolation is described in terms of regular inputs.
106 struct lp_tgsi_texture_info
108 struct lp_tgsi_channel_info coord
[4];
109 unsigned target
:8; /* TGSI_TEXTURE_* */
110 unsigned sampler_unit
:8; /* Sampler unit */
111 unsigned texture_unit
:8; /* Texture unit */
112 unsigned modifier
:8; /* LP_BLD_TEX_MODIFIER_* */
118 struct tgsi_shader_info base
;
121 * Whether any of the texture opcodes access a register file other than
124 * We could also handle TGSI_FILE_CONST/IMMEDIATE here, but there is little
127 unsigned indirect_textures
:1;
130 * Whether any of the texture (sample) ocpodes use different sampler
131 * and sampler view unit.
133 unsigned sampler_texture_units_different
:1;
136 * Whether any immediate values are outside the range of 0 and 1
138 unsigned unclamped_immediates
:1;
141 * Texture opcode description. Aimed at detecting and described direct
145 struct lp_tgsi_texture_info tex
[PIPE_MAX_SAMPLERS
];
148 * Output description. Aimed at detecting and describing simple blit
151 struct lp_tgsi_channel_info output
[PIPE_MAX_SHADER_OUTPUTS
][4];
154 * Shortcut pointers into the above (for fragment shaders).
156 const struct lp_tgsi_channel_info
*cbuf
[PIPE_MAX_COLOR_BUFS
];
160 * Reference to system values.
162 struct lp_bld_tgsi_system_values
{
163 LLVMValueRef instance_id
;
164 LLVMValueRef vertex_id
;
165 LLVMValueRef vertex_id_nobase
;
166 LLVMValueRef prim_id
;
167 LLVMValueRef basevertex
;
168 LLVMValueRef invocation_id
;
173 * Sampler code generation interface.
175 * Although texture sampling is a requirement for TGSI translation, it is
176 * a very different problem with several different approaches to it. This
177 * structure establishes an interface for texture sampling code generation, so
178 * that we can easily use different texture sampling strategies.
180 struct lp_build_sampler_soa
183 (*destroy
)( struct lp_build_sampler_soa
*sampler
);
186 (*emit_tex_sample
)(const struct lp_build_sampler_soa
*sampler
,
187 struct gallivm_state
*gallivm
,
188 const struct lp_sampler_params
*params
);
191 (*emit_size_query
)( const struct lp_build_sampler_soa
*sampler
,
192 struct gallivm_state
*gallivm
,
196 LLVMValueRef context_ptr
,
197 boolean need_nr_mips
,
198 enum lp_sampler_lod_property
,
199 LLVMValueRef explicit_lod
, /* optional */
200 LLVMValueRef
*sizes_out
);
204 struct lp_build_sampler_aos
207 (*emit_fetch_texel
)( struct lp_build_sampler_aos
*sampler
,
208 struct lp_build_context
*bld
,
209 unsigned target
, /* TGSI_TEXTURE_* */
212 const struct lp_derivatives derivs
,
213 enum lp_build_tex_modifier modifier
);
218 lp_build_tgsi_info(const struct tgsi_token
*tokens
,
219 struct lp_tgsi_info
*info
);
223 lp_build_tgsi_soa(struct gallivm_state
*gallivm
,
224 const struct tgsi_token
*tokens
,
226 struct lp_build_mask_context
*mask
,
227 LLVMValueRef consts_ptr
,
228 LLVMValueRef const_sizes_ptr
,
229 const struct lp_bld_tgsi_system_values
*system_values
,
230 const LLVMValueRef (*inputs
)[4],
231 LLVMValueRef (*outputs
)[4],
232 LLVMValueRef context_ptr
,
233 struct lp_build_sampler_soa
*sampler
,
234 const struct tgsi_shader_info
*info
,
235 const struct lp_build_tgsi_gs_iface
*gs_iface
);
239 lp_build_tgsi_aos(struct gallivm_state
*gallivm
,
240 const struct tgsi_token
*tokens
,
242 const unsigned char swizzles
[4],
243 LLVMValueRef consts_ptr
,
244 const LLVMValueRef
*inputs
,
245 LLVMValueRef
*outputs
,
246 struct lp_build_sampler_aos
*sampler
,
247 const struct tgsi_shader_info
*info
);
250 enum lp_exec_mask_break_type
{
251 LP_EXEC_MASK_BREAK_TYPE_LOOP
,
252 LP_EXEC_MASK_BREAK_TYPE_SWITCH
256 struct lp_exec_mask
{
257 struct lp_build_context
*bld
;
262 LLVMTypeRef int_vec_type
;
264 LLVMValueRef exec_mask
;
266 LLVMValueRef ret_mask
;
267 LLVMValueRef cond_mask
;
268 LLVMValueRef switch_mask
; /* current switch exec mask */
269 LLVMValueRef cont_mask
;
270 LLVMValueRef break_mask
;
272 struct function_ctx
{
274 LLVMValueRef ret_mask
;
276 LLVMValueRef cond_stack
[LP_MAX_TGSI_NESTING
];
279 /* keep track if break belongs to switch or loop */
280 enum lp_exec_mask_break_type break_type_stack
[LP_MAX_TGSI_NESTING
];
281 enum lp_exec_mask_break_type break_type
;
284 LLVMValueRef switch_val
;
285 LLVMValueRef switch_mask
;
286 LLVMValueRef switch_mask_default
;
287 boolean switch_in_default
;
289 } switch_stack
[LP_MAX_TGSI_NESTING
];
290 int switch_stack_size
;
291 LLVMValueRef switch_val
;
292 LLVMValueRef switch_mask_default
; /* reverse of switch mask used for default */
293 boolean switch_in_default
; /* if switch exec is currently in default */
294 unsigned switch_pc
; /* when used points to default or endswitch-1 */
296 LLVMValueRef loop_limiter
;
297 LLVMBasicBlockRef loop_block
;
298 LLVMValueRef break_var
;
300 LLVMBasicBlockRef loop_block
;
301 LLVMValueRef cont_mask
;
302 LLVMValueRef break_mask
;
303 LLVMValueRef break_var
;
304 } loop_stack
[LP_MAX_TGSI_NESTING
];
308 int function_stack_size
;
311 struct lp_build_tgsi_inst_list
313 struct tgsi_full_instruction
*instructions
;
314 uint max_instructions
;
315 uint num_instructions
;
318 unsigned lp_bld_tgsi_list_init(struct lp_build_tgsi_context
* bld_base
);
321 unsigned lp_bld_tgsi_add_instruction(
322 struct lp_build_tgsi_context
* bld_base
,
323 const struct tgsi_full_instruction
*inst_to_add
);
326 struct lp_build_tgsi_context
;
329 typedef LLVMValueRef (*lp_build_emit_fetch_fn
)(struct lp_build_tgsi_context
*,
330 const struct tgsi_full_src_register
*,
331 enum tgsi_opcode_type
,
334 struct lp_build_tgsi_context
336 struct lp_build_context base
;
338 struct lp_build_context uint_bld
;
339 struct lp_build_context int_bld
;
341 /** This array stores functions that are used to transform TGSI opcodes to
344 struct lp_build_tgsi_action op_actions
[TGSI_OPCODE_LAST
];
346 /* TGSI_OPCODE_RSQ is defined as 1 / sqrt( abs(src0.x) ), rsq_action
347 * should compute 1 / sqrt (src0.x) */
348 struct lp_build_tgsi_action rsq_action
;
350 struct lp_build_tgsi_action sqrt_action
;
352 const struct tgsi_shader_info
*info
;
354 lp_build_emit_fetch_fn emit_fetch_funcs
[TGSI_FILE_COUNT
];
356 LLVMValueRef (*emit_swizzle
)(struct lp_build_tgsi_context
*,
357 LLVMValueRef
, unsigned, unsigned, unsigned, unsigned);
360 void (*emit_debug
)(struct lp_build_tgsi_context
*,
361 const struct tgsi_full_instruction
*,
362 const struct tgsi_opcode_info
*);
364 void (*emit_store
)(struct lp_build_tgsi_context
*,
365 const struct tgsi_full_instruction
*,
366 const struct tgsi_opcode_info
*,
367 LLVMValueRef dst
[4]);
369 void (*emit_declaration
)(struct lp_build_tgsi_context
*,
370 const struct tgsi_full_declaration
*decl
);
372 void (*emit_immediate
)(struct lp_build_tgsi_context
*,
373 const struct tgsi_full_immediate
*imm
);
376 /* Allow the user to store data in this structure rather than passing it
377 * to every function. */
384 struct tgsi_full_instruction
*instructions
;
385 uint max_instructions
;
386 uint num_instructions
;
388 /** This function allows the user to insert some instructions at the
389 * beginning of the program. It is optional and does not need to be
392 void (*emit_prologue
)(struct lp_build_tgsi_context
*);
394 /** This function allows the user to insert some instructions at the end of
395 * the program. This callback is intended to be used for emitting
396 * instructions to handle the export for the output registers, but it can
397 * be used for any purpose. Implementing this function is optiona, but
400 void (*emit_epilogue
)(struct lp_build_tgsi_context
*);
403 struct lp_build_tgsi_gs_iface
405 LLVMValueRef (*fetch_input
)(const struct lp_build_tgsi_gs_iface
*gs_iface
,
406 struct lp_build_tgsi_context
* bld_base
,
407 boolean is_vindex_indirect
,
408 LLVMValueRef vertex_index
,
409 boolean is_aindex_indirect
,
410 LLVMValueRef attrib_index
,
411 LLVMValueRef swizzle_index
);
412 void (*emit_vertex
)(const struct lp_build_tgsi_gs_iface
*gs_iface
,
413 struct lp_build_tgsi_context
* bld_base
,
414 LLVMValueRef (*outputs
)[4],
415 LLVMValueRef emitted_vertices_vec
);
416 void (*end_primitive
)(const struct lp_build_tgsi_gs_iface
*gs_iface
,
417 struct lp_build_tgsi_context
* bld_base
,
418 LLVMValueRef verts_per_prim_vec
,
419 LLVMValueRef emitted_prims_vec
);
420 void (*gs_epilogue
)(const struct lp_build_tgsi_gs_iface
*gs_iface
,
421 struct lp_build_tgsi_context
* bld_base
,
422 LLVMValueRef total_emitted_vertices_vec
,
423 LLVMValueRef emitted_prims_vec
);
426 struct lp_build_tgsi_soa_context
428 struct lp_build_tgsi_context bld_base
;
430 /* Builder for scalar elements of shader's data type (float) */
431 struct lp_build_context elem_bld
;
433 const struct lp_build_tgsi_gs_iface
*gs_iface
;
434 LLVMValueRef emitted_prims_vec_ptr
;
435 LLVMValueRef total_emitted_vertices_vec_ptr
;
436 LLVMValueRef emitted_vertices_vec_ptr
;
437 LLVMValueRef max_output_vertices_vec
;
439 LLVMValueRef consts_ptr
;
440 LLVMValueRef const_sizes_ptr
;
441 LLVMValueRef consts
[LP_MAX_TGSI_CONST_BUFFERS
];
442 LLVMValueRef consts_sizes
[LP_MAX_TGSI_CONST_BUFFERS
];
443 const LLVMValueRef (*inputs
)[TGSI_NUM_CHANNELS
];
444 LLVMValueRef (*outputs
)[TGSI_NUM_CHANNELS
];
445 LLVMValueRef context_ptr
;
447 const struct lp_build_sampler_soa
*sampler
;
449 struct tgsi_declaration_sampler_view sv
[PIPE_MAX_SHADER_SAMPLER_VIEWS
];
451 LLVMValueRef immediates
[LP_MAX_INLINED_IMMEDIATES
][TGSI_NUM_CHANNELS
];
452 LLVMValueRef temps
[LP_MAX_INLINED_TEMPS
][TGSI_NUM_CHANNELS
];
453 LLVMValueRef addr
[LP_MAX_TGSI_ADDRS
][TGSI_NUM_CHANNELS
];
454 LLVMValueRef preds
[LP_MAX_TGSI_PREDS
][TGSI_NUM_CHANNELS
];
456 /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
457 * set in the indirect_files field.
458 * The temps[] array above is unused then.
460 LLVMValueRef temps_array
;
462 /* We allocate/use this array of output if (1 << TGSI_FILE_OUTPUT) is
463 * set in the indirect_files field.
464 * The outputs[] array above is unused then.
466 LLVMValueRef outputs_array
;
468 /* We allocate/use this array of inputs if (1 << TGSI_FILE_INPUT) is
469 * set in the indirect_files field.
470 * The inputs[] array above is unused then.
472 LLVMValueRef inputs_array
;
474 /* We allocate/use this array of temps if (1 << TGSI_FILE_IMMEDIATE) is
475 * set in the indirect_files field.
477 LLVMValueRef imms_array
;
480 struct lp_bld_tgsi_system_values system_values
;
482 /** bitmask indicating which register files are accessed indirectly */
483 unsigned indirect_files
;
485 struct lp_build_mask_context
*mask
;
486 struct lp_exec_mask exec_mask
;
489 boolean use_immediates_array
;
493 lp_emit_declaration_soa(
494 struct lp_build_tgsi_context
*bld
,
495 const struct tgsi_full_declaration
*decl
);
497 void lp_emit_immediate_soa(
498 struct lp_build_tgsi_context
*bld_base
,
499 const struct tgsi_full_immediate
*imm
);
502 lp_emit_instruction_soa(
503 struct lp_build_tgsi_soa_context
*bld
,
504 const struct tgsi_full_instruction
*inst
,
505 const struct tgsi_opcode_info
*info
);
510 struct lp_build_tgsi_soa_context
*bld
,
516 struct lp_build_tgsi_soa_context
*bld
,
520 struct lp_build_tgsi_aos_context
522 struct lp_build_tgsi_context bld_base
;
524 /* Builder for integer masks and indices */
525 struct lp_build_context int_bld
;
529 * - swizzles[0] = red index
530 * - swizzles[1] = green index
531 * - swizzles[2] = blue index
532 * - swizzles[3] = alpha index
534 unsigned char swizzles
[4];
535 unsigned char inv_swizzles
[4];
537 LLVMValueRef consts_ptr
;
538 const LLVMValueRef
*inputs
;
539 LLVMValueRef
*outputs
;
541 struct lp_build_sampler_aos
*sampler
;
543 struct tgsi_declaration_sampler_view sv
[PIPE_MAX_SHADER_SAMPLER_VIEWS
];
545 LLVMValueRef immediates
[LP_MAX_INLINED_IMMEDIATES
];
546 LLVMValueRef temps
[LP_MAX_INLINED_TEMPS
];
547 LLVMValueRef addr
[LP_MAX_TGSI_ADDRS
];
548 LLVMValueRef preds
[LP_MAX_TGSI_PREDS
];
550 /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
551 * set in the indirect_files field.
552 * The temps[] array above is unused then.
554 LLVMValueRef temps_array
;
556 /** bitmask indicating which register files are accessed indirectly */
557 unsigned indirect_files
;
561 static INLINE
struct lp_build_tgsi_soa_context
*
562 lp_soa_context(struct lp_build_tgsi_context
*bld_base
)
564 return (struct lp_build_tgsi_soa_context
*)bld_base
;
567 static INLINE
struct lp_build_tgsi_aos_context
*
568 lp_aos_context(struct lp_build_tgsi_context
*bld_base
)
570 return (struct lp_build_tgsi_aos_context
*)bld_base
;
574 lp_emit_declaration_aos(
575 struct lp_build_tgsi_aos_context
*bld
,
576 const struct tgsi_full_declaration
*decl
);
580 lp_emit_instruction_aos(
581 struct lp_build_tgsi_aos_context
*bld
,
582 const struct tgsi_full_instruction
*inst
,
583 const struct tgsi_opcode_info
*info
,
588 struct lp_build_tgsi_aos_context
*bld
,
589 const struct tgsi_full_instruction
*inst
,
593 void lp_build_fetch_args(
594 struct lp_build_tgsi_context
* bld_base
,
595 struct lp_build_emit_data
* emit_data
);
598 lp_build_tgsi_inst_llvm_aos(
599 struct lp_build_tgsi_context
* bld_base
,
600 const struct tgsi_full_instruction
*inst
);
603 lp_build_tgsi_intrinsic(
604 const struct lp_build_tgsi_action
* action
,
605 struct lp_build_tgsi_context
* bld_base
,
606 struct lp_build_emit_data
* emit_data
);
610 struct lp_build_tgsi_context
*bld_base
,
611 unsigned tgsi_opcode
,
612 struct lp_build_emit_data
* emit_data
);
615 lp_build_emit_llvm_unary(
616 struct lp_build_tgsi_context
*bld_base
,
617 unsigned tgsi_opcode
,
621 lp_build_emit_llvm_binary(
622 struct lp_build_tgsi_context
*bld_base
,
623 unsigned tgsi_opcode
,
628 lp_build_emit_llvm_ternary(
629 struct lp_build_tgsi_context
*bld_base
,
630 unsigned tgsi_opcode
,
636 lp_build_tgsi_inst_llvm(
637 struct lp_build_tgsi_context
* bld_base
,
638 const struct tgsi_full_instruction
*inst
);
642 struct lp_build_tgsi_context
*bld_base
,
643 const struct tgsi_full_instruction
*inst
,
645 const unsigned chan_index
);
649 lp_build_emit_fetch_texoffset(
650 struct lp_build_tgsi_context
*bld_base
,
651 const struct tgsi_full_instruction
*inst
,
653 const unsigned chan_index
);
657 struct lp_build_tgsi_context
* bld_base
,
658 const struct tgsi_token
*tokens
);
660 #endif /* LP_BLD_TGSI_H */