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 "lp_bld_type.h"
44 #include "pipe/p_compiler.h"
45 #include "pipe/p_state.h"
46 #include "tgsi/tgsi_exec.h"
47 #include "tgsi/tgsi_scan.h"
48 #include "tgsi/tgsi_info.h"
50 #define LP_CHAN_ALL ~0
52 #define LP_MAX_INSTRUCTIONS 256
54 struct tgsi_full_declaration
;
55 struct tgsi_full_immediate
;
56 struct tgsi_full_instruction
;
57 struct tgsi_full_src_register
;
58 struct tgsi_opcode_info
;
60 struct tgsi_shader_info
;
61 struct lp_build_mask_context
;
65 enum lp_build_tex_modifier
{
66 LP_BLD_TEX_MODIFIER_NONE
= 0,
67 LP_BLD_TEX_MODIFIER_PROJECTED
,
68 LP_BLD_TEX_MODIFIER_LOD_BIAS
,
69 LP_BLD_TEX_MODIFIER_EXPLICIT_LOD
,
70 LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV
75 * Describe a channel of a register.
78 * - immediate value (i.e. derived from a IMM register)
81 * - undetermined (when .file == TGSI_FILE_NULL)
83 * This is one of the analysis results, and is used to described
84 * the output color in terms of inputs.
86 struct lp_tgsi_channel_info
88 unsigned file
:4; /* TGSI_FILE_* */
89 unsigned swizzle
:3; /* PIPE_SWIZZLE_x */
92 float value
; /* for TGSI_FILE_IMMEDIATE */
98 * Describe a texture sampler interpolator.
100 * The interpolation is described in terms of regular inputs.
102 struct lp_tgsi_texture_info
104 struct lp_tgsi_channel_info coord
[4];
105 unsigned target
:8; /* TGSI_TEXTURE_* */
106 unsigned unit
:8; /* Sampler unit */
107 unsigned modifier
:8; /* LP_BLD_TEX_MODIFIER_* */
113 struct tgsi_shader_info base
;
116 * Whether any of the texture opcodes access a register file other than
119 * We could also handle TGSI_FILE_CONST/IMMEDIATE here, but there is little
122 unsigned indirect_textures
:1;
125 * Texture opcode description. Aimed at detecting and described direct
129 struct lp_tgsi_texture_info tex
[PIPE_MAX_SAMPLERS
];
132 * Output description. Aimed at detecting and describing simple blit
135 struct lp_tgsi_channel_info output
[PIPE_MAX_SHADER_OUTPUTS
][4];
138 * Shortcut pointers into the above (for fragment shaders).
140 const struct lp_tgsi_channel_info
*cbuf
[PIPE_MAX_COLOR_BUFS
];
144 * Sampler code generation interface.
146 * Although texture sampling is a requirement for TGSI translation, it is
147 * a very different problem with several different approaches to it. This
148 * structure establishes an interface for texture sampling code generation, so
149 * that we can easily use different texture sampling strategies.
151 struct lp_build_sampler_soa
154 (*destroy
)( struct lp_build_sampler_soa
*sampler
);
157 (*emit_fetch_texel
)( const struct lp_build_sampler_soa
*sampler
,
158 struct gallivm_state
*gallivm
,
162 const LLVMValueRef
*coords
,
163 const LLVMValueRef
*ddx
,
164 const LLVMValueRef
*ddy
,
165 LLVMValueRef lod_bias
, /* optional */
166 LLVMValueRef explicit_lod
, /* optional */
167 LLVMValueRef
*texel
);
171 struct lp_build_sampler_aos
174 (*emit_fetch_texel
)( struct lp_build_sampler_aos
*sampler
,
175 struct lp_build_context
*bld
,
176 unsigned target
, /* TGSI_TEXTURE_* */
181 enum lp_build_tex_modifier modifier
);
186 lp_build_tgsi_info(const struct tgsi_token
*tokens
,
187 struct lp_tgsi_info
*info
);
191 lp_build_tgsi_soa(struct gallivm_state
*gallivm
,
192 const struct tgsi_token
*tokens
,
194 struct lp_build_mask_context
*mask
,
195 LLVMValueRef consts_ptr
,
196 LLVMValueRef system_values_array
,
197 const LLVMValueRef
*pos
,
198 const LLVMValueRef (*inputs
)[4],
199 LLVMValueRef (*outputs
)[4],
200 struct lp_build_sampler_soa
*sampler
,
201 const struct tgsi_shader_info
*info
);
205 lp_build_tgsi_aos(struct gallivm_state
*gallivm
,
206 const struct tgsi_token
*tokens
,
208 const unsigned char swizzles
[4],
209 LLVMValueRef consts_ptr
,
210 const LLVMValueRef
*inputs
,
211 LLVMValueRef
*outputs
,
212 struct lp_build_sampler_aos
*sampler
,
213 const struct tgsi_shader_info
*info
);
217 lp_build_system_values_array(struct gallivm_state
*gallivm
,
218 const struct tgsi_shader_info
*info
,
219 LLVMValueRef instance_id
,
220 LLVMValueRef facing
);
223 struct lp_exec_mask
{
224 struct lp_build_context
*bld
;
228 LLVMTypeRef int_vec_type
;
230 LLVMValueRef cond_stack
[LP_MAX_TGSI_NESTING
];
232 LLVMValueRef cond_mask
;
234 LLVMBasicBlockRef loop_block
;
235 LLVMValueRef cont_mask
;
236 LLVMValueRef break_mask
;
237 LLVMValueRef break_var
;
239 LLVMBasicBlockRef loop_block
;
240 LLVMValueRef cont_mask
;
241 LLVMValueRef break_mask
;
242 LLVMValueRef break_var
;
243 } loop_stack
[LP_MAX_TGSI_NESTING
];
246 LLVMValueRef ret_mask
;
249 LLVMValueRef ret_mask
;
250 } call_stack
[LP_MAX_TGSI_NESTING
];
253 LLVMValueRef exec_mask
;
256 struct lp_build_tgsi_inst_list
258 struct tgsi_full_instruction
*instructions
;
259 uint max_instructions
;
260 uint num_instructions
;
263 unsigned lp_bld_tgsi_list_init(struct lp_build_tgsi_context
* bld_base
);
266 unsigned lp_bld_tgsi_add_instruction(
267 struct lp_build_tgsi_context
* bld_base
,
268 struct tgsi_full_instruction
*inst_to_add
);
271 struct lp_build_tgsi_context
;
274 typedef LLVMValueRef (*lp_build_emit_fetch_fn
)(struct lp_build_tgsi_context
*,
275 const struct tgsi_full_src_register
*,
276 enum tgsi_opcode_type
,
279 struct lp_build_tgsi_context
281 struct lp_build_context base
;
283 struct lp_build_context uint_bld
;
284 struct lp_build_context int_bld
;
286 /** This array stores functions that are used to transform TGSI opcodes to
289 struct lp_build_tgsi_action op_actions
[TGSI_OPCODE_LAST
];
291 /* TGSI_OPCODE_RSQ is defined as 1 / sqrt( abs(src0.x) ), rsq_action
292 * should compute 1 / sqrt (src0.x) */
293 struct lp_build_tgsi_action rsq_action
;
295 const struct tgsi_shader_info
*info
;
297 lp_build_emit_fetch_fn emit_fetch_funcs
[TGSI_FILE_COUNT
];
299 LLVMValueRef (*emit_swizzle
)(struct lp_build_tgsi_context
*,
300 LLVMValueRef
, unsigned, unsigned, unsigned, unsigned);
302 void (*emit_store
)(struct lp_build_tgsi_context
*,
303 const struct tgsi_full_instruction
*,
304 const struct tgsi_opcode_info
*,
305 LLVMValueRef dst
[4]);
307 void (*emit_declaration
)(struct lp_build_tgsi_context
*,
308 const struct tgsi_full_declaration
*decl
);
310 void (*emit_immediate
)(struct lp_build_tgsi_context
*,
311 const struct tgsi_full_immediate
*imm
);
314 /* Allow the user to store data in this structure rather than passing it
315 * to every function. */
322 struct tgsi_full_instruction
*instructions
;
323 uint max_instructions
;
324 uint num_instructions
;
326 /** This function allows the user to insert some instructions at the
327 * beginning of the program. It is optional and does not need to be
330 void (*emit_prologue
)(struct lp_build_tgsi_context
*);
332 /** This function allows the user to insert some instructions at the end of
333 * the program. This callback is intended to be used for emitting
334 * instructions to handle the export for the output registers, but it can
335 * be used for any purpose. Implementing this function is optiona, but
338 void (*emit_epilogue
)(struct lp_build_tgsi_context
*);
341 struct lp_build_tgsi_soa_context
343 struct lp_build_tgsi_context bld_base
;
345 /* Builder for scalar elements of shader's data type (float) */
346 struct lp_build_context elem_bld
;
348 LLVMValueRef consts_ptr
;
349 const LLVMValueRef
*pos
;
350 const LLVMValueRef (*inputs
)[TGSI_NUM_CHANNELS
];
351 LLVMValueRef (*outputs
)[TGSI_NUM_CHANNELS
];
353 const struct lp_build_sampler_soa
*sampler
;
355 LLVMValueRef immediates
[LP_MAX_TGSI_IMMEDIATES
][TGSI_NUM_CHANNELS
];
356 LLVMValueRef temps
[LP_MAX_TGSI_TEMPS
][TGSI_NUM_CHANNELS
];
357 LLVMValueRef addr
[LP_MAX_TGSI_ADDRS
][TGSI_NUM_CHANNELS
];
358 LLVMValueRef preds
[LP_MAX_TGSI_PREDS
][TGSI_NUM_CHANNELS
];
360 /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
361 * set in the indirect_files field.
362 * The temps[] array above is unused then.
364 LLVMValueRef temps_array
;
366 /* We allocate/use this array of output if (1 << TGSI_FILE_OUTPUT) is
367 * set in the indirect_files field.
368 * The outputs[] array above is unused then.
370 LLVMValueRef outputs_array
;
372 /* We allocate/use this array of inputs if (1 << TGSI_FILE_INPUT) is
373 * set in the indirect_files field.
374 * The inputs[] array above is unused then.
376 LLVMValueRef inputs_array
;
378 LLVMValueRef system_values_array
;
380 /** bitmask indicating which register files are accessed indirectly */
381 unsigned indirect_files
;
383 struct lp_build_mask_context
*mask
;
384 struct lp_exec_mask exec_mask
;
391 lp_emit_declaration_soa(
392 struct lp_build_tgsi_context
*bld
,
393 const struct tgsi_full_declaration
*decl
);
395 void lp_emit_immediate_soa(
396 struct lp_build_tgsi_context
*bld_base
,
397 const struct tgsi_full_immediate
*imm
);
400 lp_emit_instruction_soa(
401 struct lp_build_tgsi_soa_context
*bld
,
402 const struct tgsi_full_instruction
*inst
,
403 const struct tgsi_opcode_info
*info
);
408 struct lp_build_tgsi_soa_context
*bld
,
414 struct lp_build_tgsi_soa_context
*bld
,
418 struct lp_build_tgsi_aos_context
420 struct lp_build_tgsi_context bld_base
;
422 /* Builder for integer masks and indices */
423 struct lp_build_context int_bld
;
427 * - swizzles[0] = red index
428 * - swizzles[1] = green index
429 * - swizzles[2] = blue index
430 * - swizzles[3] = alpha index
432 unsigned char swizzles
[4];
433 unsigned char inv_swizzles
[4];
435 LLVMValueRef consts_ptr
;
436 const LLVMValueRef
*inputs
;
437 LLVMValueRef
*outputs
;
439 struct lp_build_sampler_aos
*sampler
;
441 LLVMValueRef immediates
[LP_MAX_TGSI_IMMEDIATES
];
442 LLVMValueRef temps
[LP_MAX_TGSI_TEMPS
];
443 LLVMValueRef addr
[LP_MAX_TGSI_ADDRS
];
444 LLVMValueRef preds
[LP_MAX_TGSI_PREDS
];
446 /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
447 * set in the indirect_files field.
448 * The temps[] array above is unused then.
450 LLVMValueRef temps_array
;
452 /** bitmask indicating which register files are accessed indirectly */
453 unsigned indirect_files
;
457 static INLINE
struct lp_build_tgsi_soa_context
*
458 lp_soa_context(struct lp_build_tgsi_context
*bld_base
)
460 return (struct lp_build_tgsi_soa_context
*)bld_base
;
463 static INLINE
struct lp_build_tgsi_aos_context
*
464 lp_aos_context(struct lp_build_tgsi_context
*bld_base
)
466 return (struct lp_build_tgsi_aos_context
*)bld_base
;
470 lp_emit_declaration_aos(
471 struct lp_build_tgsi_aos_context
*bld
,
472 const struct tgsi_full_declaration
*decl
);
476 lp_emit_instruction_aos(
477 struct lp_build_tgsi_aos_context
*bld
,
478 const struct tgsi_full_instruction
*inst
,
479 const struct tgsi_opcode_info
*info
,
484 struct lp_build_tgsi_aos_context
*bld
,
485 const struct tgsi_full_instruction
*inst
,
489 void lp_build_fetch_args(
490 struct lp_build_tgsi_context
* bld_base
,
491 struct lp_build_emit_data
* emit_data
);
494 lp_build_tgsi_inst_llvm_aos(
495 struct lp_build_tgsi_context
* bld_base
,
496 const struct tgsi_full_instruction
*inst
);
499 lp_build_tgsi_intrinsic(
500 const struct lp_build_tgsi_action
* action
,
501 struct lp_build_tgsi_context
* bld_base
,
502 struct lp_build_emit_data
* emit_data
);
506 struct lp_build_tgsi_context
*bld_base
,
507 unsigned tgsi_opcode
,
508 struct lp_build_emit_data
* emit_data
);
511 lp_build_emit_llvm_unary(
512 struct lp_build_tgsi_context
*bld_base
,
513 unsigned tgsi_opcode
,
517 lp_build_emit_llvm_binary(
518 struct lp_build_tgsi_context
*bld_base
,
519 unsigned tgsi_opcode
,
524 lp_build_emit_llvm_ternary(
525 struct lp_build_tgsi_context
*bld_base
,
526 unsigned tgsi_opcode
,
532 lp_build_tgsi_inst_llvm(
533 struct lp_build_tgsi_context
* bld_base
,
534 const struct tgsi_full_instruction
*inst
);
538 struct lp_build_tgsi_context
*bld_base
,
539 const struct tgsi_full_instruction
*inst
,
541 const unsigned chan_index
);
545 struct lp_build_tgsi_context
* bld_base
,
546 const struct tgsi_token
*tokens
);
548 #endif /* LP_BLD_TGSI_H */