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
;
63 struct lp_derivatives
;
64 struct lp_build_tgsi_gs_iface
;
67 enum lp_build_tex_modifier
{
68 LP_BLD_TEX_MODIFIER_NONE
= 0,
69 LP_BLD_TEX_MODIFIER_PROJECTED
,
70 LP_BLD_TEX_MODIFIER_LOD_BIAS
,
71 LP_BLD_TEX_MODIFIER_EXPLICIT_LOD
,
72 LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV
,
73 LP_BLD_TEX_MODIFIER_LOD_ZERO
78 * Describe a channel of a register.
81 * - immediate value (i.e. derived from a IMM register)
84 * - undetermined (when .file == TGSI_FILE_NULL)
86 * This is one of the analysis results, and is used to described
87 * the output color in terms of inputs.
89 struct lp_tgsi_channel_info
91 unsigned file
:4; /* TGSI_FILE_* */
92 unsigned swizzle
:3; /* PIPE_SWIZZLE_x */
95 float value
; /* for TGSI_FILE_IMMEDIATE */
101 * Describe a texture sampler interpolator.
103 * The interpolation is described in terms of regular inputs.
105 struct lp_tgsi_texture_info
107 struct lp_tgsi_channel_info coord
[4];
108 unsigned target
:8; /* TGSI_TEXTURE_* */
109 unsigned sampler_unit
:8; /* Sampler unit */
110 unsigned texture_unit
:8; /* Texture unit */
111 unsigned modifier
:8; /* LP_BLD_TEX_MODIFIER_* */
117 struct tgsi_shader_info base
;
120 * Whether any of the texture opcodes access a register file other than
123 * We could also handle TGSI_FILE_CONST/IMMEDIATE here, but there is little
126 unsigned indirect_textures
:1;
129 * Whether any immediate values are outside the range of 0 and 1
131 unsigned unclamped_immediates
:1;
134 * Texture opcode description. Aimed at detecting and described direct
138 struct lp_tgsi_texture_info tex
[PIPE_MAX_SAMPLERS
];
141 * Output description. Aimed at detecting and describing simple blit
144 struct lp_tgsi_channel_info output
[PIPE_MAX_SHADER_OUTPUTS
][4];
147 * Shortcut pointers into the above (for fragment shaders).
149 const struct lp_tgsi_channel_info
*cbuf
[PIPE_MAX_COLOR_BUFS
];
153 * Reference to system values.
155 struct lp_bld_tgsi_system_values
{
156 LLVMValueRef instance_id
;
157 LLVMValueRef vertex_id
;
162 * Sampler code generation interface.
164 * Although texture sampling is a requirement for TGSI translation, it is
165 * a very different problem with several different approaches to it. This
166 * structure establishes an interface for texture sampling code generation, so
167 * that we can easily use different texture sampling strategies.
169 struct lp_build_sampler_soa
172 (*destroy
)( struct lp_build_sampler_soa
*sampler
);
175 (*emit_fetch_texel
)( const struct lp_build_sampler_soa
*sampler
,
176 struct gallivm_state
*gallivm
,
179 unsigned texture_index
,
180 unsigned sampler_index
,
181 const LLVMValueRef
*coords
,
182 const LLVMValueRef
*offsets
,
183 const struct lp_derivatives
*derivs
,
184 LLVMValueRef lod_bias
, /* optional */
185 LLVMValueRef explicit_lod
, /* optional */
186 LLVMValueRef
*texel
);
189 (*emit_size_query
)( const struct lp_build_sampler_soa
*sampler
,
190 struct gallivm_state
*gallivm
,
193 boolean need_nr_mips
,
194 LLVMValueRef explicit_lod
, /* optional */
195 LLVMValueRef
*sizes_out
);
199 struct lp_build_sampler_aos
202 (*emit_fetch_texel
)( struct lp_build_sampler_aos
*sampler
,
203 struct lp_build_context
*bld
,
204 unsigned target
, /* TGSI_TEXTURE_* */
207 const struct lp_derivatives derivs
,
208 enum lp_build_tex_modifier modifier
);
213 lp_build_tgsi_info(const struct tgsi_token
*tokens
,
214 struct lp_tgsi_info
*info
);
218 lp_build_tgsi_soa(struct gallivm_state
*gallivm
,
219 const struct tgsi_token
*tokens
,
221 struct lp_build_mask_context
*mask
,
222 LLVMValueRef consts_ptr
,
223 const struct lp_bld_tgsi_system_values
*system_values
,
224 const LLVMValueRef
*pos
,
225 const LLVMValueRef (*inputs
)[4],
226 LLVMValueRef (*outputs
)[4],
227 struct lp_build_sampler_soa
*sampler
,
228 const struct tgsi_shader_info
*info
,
229 const struct lp_build_tgsi_gs_iface
*gs_iface
);
233 lp_build_tgsi_aos(struct gallivm_state
*gallivm
,
234 const struct tgsi_token
*tokens
,
236 const unsigned char swizzles
[4],
237 LLVMValueRef consts_ptr
,
238 const LLVMValueRef
*inputs
,
239 LLVMValueRef
*outputs
,
240 struct lp_build_sampler_aos
*sampler
,
241 const struct tgsi_shader_info
*info
);
244 struct lp_exec_mask
{
245 struct lp_build_context
*bld
;
250 LLVMTypeRef int_vec_type
;
252 LLVMValueRef cond_stack
[LP_MAX_TGSI_NESTING
];
254 LLVMValueRef cond_mask
;
256 LLVMBasicBlockRef loop_block
;
257 LLVMValueRef cont_mask
;
258 LLVMValueRef break_mask
;
259 LLVMValueRef break_var
;
261 LLVMBasicBlockRef loop_block
;
262 LLVMValueRef cont_mask
;
263 LLVMValueRef break_mask
;
264 LLVMValueRef break_var
;
265 } loop_stack
[LP_MAX_TGSI_NESTING
];
268 LLVMValueRef ret_mask
;
271 LLVMValueRef ret_mask
;
272 } call_stack
[LP_MAX_TGSI_NESTING
];
275 LLVMValueRef exec_mask
;
276 LLVMValueRef loop_limiter
;
279 struct lp_build_tgsi_inst_list
281 struct tgsi_full_instruction
*instructions
;
282 uint max_instructions
;
283 uint num_instructions
;
286 unsigned lp_bld_tgsi_list_init(struct lp_build_tgsi_context
* bld_base
);
289 unsigned lp_bld_tgsi_add_instruction(
290 struct lp_build_tgsi_context
* bld_base
,
291 struct tgsi_full_instruction
*inst_to_add
);
294 struct lp_build_tgsi_context
;
297 typedef LLVMValueRef (*lp_build_emit_fetch_fn
)(struct lp_build_tgsi_context
*,
298 const struct tgsi_full_src_register
*,
299 enum tgsi_opcode_type
,
302 struct lp_build_tgsi_context
304 struct lp_build_context base
;
306 struct lp_build_context uint_bld
;
307 struct lp_build_context int_bld
;
309 /** This array stores functions that are used to transform TGSI opcodes to
312 struct lp_build_tgsi_action op_actions
[TGSI_OPCODE_LAST
];
314 /* TGSI_OPCODE_RSQ is defined as 1 / sqrt( abs(src0.x) ), rsq_action
315 * should compute 1 / sqrt (src0.x) */
316 struct lp_build_tgsi_action rsq_action
;
318 struct lp_build_tgsi_action sqrt_action
;
320 const struct tgsi_shader_info
*info
;
322 lp_build_emit_fetch_fn emit_fetch_funcs
[TGSI_FILE_COUNT
];
324 LLVMValueRef (*emit_swizzle
)(struct lp_build_tgsi_context
*,
325 LLVMValueRef
, unsigned, unsigned, unsigned, unsigned);
327 void (*emit_store
)(struct lp_build_tgsi_context
*,
328 const struct tgsi_full_instruction
*,
329 const struct tgsi_opcode_info
*,
330 LLVMValueRef dst
[4]);
332 void (*emit_declaration
)(struct lp_build_tgsi_context
*,
333 const struct tgsi_full_declaration
*decl
);
335 void (*emit_immediate
)(struct lp_build_tgsi_context
*,
336 const struct tgsi_full_immediate
*imm
);
339 /* Allow the user to store data in this structure rather than passing it
340 * to every function. */
347 struct tgsi_full_instruction
*instructions
;
348 uint max_instructions
;
349 uint num_instructions
;
351 /** This function allows the user to insert some instructions at the
352 * beginning of the program. It is optional and does not need to be
355 void (*emit_prologue
)(struct lp_build_tgsi_context
*);
357 /** This function allows the user to insert some instructions at the end of
358 * the program. This callback is intended to be used for emitting
359 * instructions to handle the export for the output registers, but it can
360 * be used for any purpose. Implementing this function is optiona, but
363 void (*emit_epilogue
)(struct lp_build_tgsi_context
*);
366 struct lp_build_tgsi_gs_iface
368 LLVMValueRef (*fetch_input
)(const struct lp_build_tgsi_gs_iface
*gs_iface
,
369 struct lp_build_tgsi_context
* bld_base
,
370 LLVMValueRef vertex_index
,
371 LLVMValueRef attrib_index
,
372 LLVMValueRef swizzle_index
);
373 void (*emit_vertex
)(const struct lp_build_tgsi_gs_iface
*gs_iface
,
374 struct lp_build_tgsi_context
* bld_base
,
375 LLVMValueRef (*outputs
)[4],
376 LLVMValueRef emitted_vertices_vec
);
377 void (*end_primitive
)(const struct lp_build_tgsi_gs_iface
*gs_iface
,
378 struct lp_build_tgsi_context
* bld_base
,
379 LLVMValueRef verts_per_prim_vec
,
380 LLVMValueRef emitted_prims_vec
);
381 void (*gs_epilogue
)(const struct lp_build_tgsi_gs_iface
*gs_iface
,
382 struct lp_build_tgsi_context
* bld_base
,
383 LLVMValueRef total_emitted_vertices_vec
,
384 LLVMValueRef emitted_prims_vec
);
387 struct lp_build_tgsi_soa_context
389 struct lp_build_tgsi_context bld_base
;
391 /* Builder for scalar elements of shader's data type (float) */
392 struct lp_build_context elem_bld
;
394 const struct lp_build_tgsi_gs_iface
*gs_iface
;
395 LLVMValueRef emitted_prims_vec
;
396 LLVMValueRef total_emitted_vertices_vec
;
397 LLVMValueRef emitted_vertices_vec
;
398 /* if a shader doesn't have ENDPRIM instruction but it has
399 * a number of EMIT instructions it means the END instruction
400 * implicitly invokes ENDPRIM. handle this via a flag here
401 * in the future maybe we can enforce TGSI to always have
402 * an explicit ENDPRIM */
403 boolean pending_end_primitive
;
405 LLVMValueRef consts_ptr
;
406 const LLVMValueRef
*pos
;
407 const LLVMValueRef (*inputs
)[TGSI_NUM_CHANNELS
];
408 LLVMValueRef (*outputs
)[TGSI_NUM_CHANNELS
];
410 const struct lp_build_sampler_soa
*sampler
;
412 struct tgsi_declaration_sampler_view sv
[PIPE_MAX_SHADER_SAMPLER_VIEWS
];
414 LLVMValueRef immediates
[LP_MAX_TGSI_IMMEDIATES
][TGSI_NUM_CHANNELS
];
415 LLVMValueRef temps
[LP_MAX_TGSI_TEMPS
][TGSI_NUM_CHANNELS
];
416 LLVMValueRef addr
[LP_MAX_TGSI_ADDRS
][TGSI_NUM_CHANNELS
];
417 LLVMValueRef preds
[LP_MAX_TGSI_PREDS
][TGSI_NUM_CHANNELS
];
419 /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
420 * set in the indirect_files field.
421 * The temps[] array above is unused then.
423 LLVMValueRef temps_array
;
425 /* We allocate/use this array of output if (1 << TGSI_FILE_OUTPUT) is
426 * set in the indirect_files field.
427 * The outputs[] array above is unused then.
429 LLVMValueRef outputs_array
;
431 /* We allocate/use this array of inputs if (1 << TGSI_FILE_INPUT) is
432 * set in the indirect_files field.
433 * The inputs[] array above is unused then.
435 LLVMValueRef inputs_array
;
437 struct lp_bld_tgsi_system_values system_values
;
439 /** bitmask indicating which register files are accessed indirectly */
440 unsigned indirect_files
;
442 struct lp_build_mask_context
*mask
;
443 struct lp_exec_mask exec_mask
;
450 lp_emit_declaration_soa(
451 struct lp_build_tgsi_context
*bld
,
452 const struct tgsi_full_declaration
*decl
);
454 void lp_emit_immediate_soa(
455 struct lp_build_tgsi_context
*bld_base
,
456 const struct tgsi_full_immediate
*imm
);
459 lp_emit_instruction_soa(
460 struct lp_build_tgsi_soa_context
*bld
,
461 const struct tgsi_full_instruction
*inst
,
462 const struct tgsi_opcode_info
*info
);
467 struct lp_build_tgsi_soa_context
*bld
,
473 struct lp_build_tgsi_soa_context
*bld
,
477 struct lp_build_tgsi_aos_context
479 struct lp_build_tgsi_context bld_base
;
481 /* Builder for integer masks and indices */
482 struct lp_build_context int_bld
;
486 * - swizzles[0] = red index
487 * - swizzles[1] = green index
488 * - swizzles[2] = blue index
489 * - swizzles[3] = alpha index
491 unsigned char swizzles
[4];
492 unsigned char inv_swizzles
[4];
494 LLVMValueRef consts_ptr
;
495 const LLVMValueRef
*inputs
;
496 LLVMValueRef
*outputs
;
498 struct lp_build_sampler_aos
*sampler
;
500 LLVMValueRef immediates
[LP_MAX_TGSI_IMMEDIATES
];
501 LLVMValueRef temps
[LP_MAX_TGSI_TEMPS
];
502 LLVMValueRef addr
[LP_MAX_TGSI_ADDRS
];
503 LLVMValueRef preds
[LP_MAX_TGSI_PREDS
];
505 /* We allocate/use this array of temps if (1 << TGSI_FILE_TEMPORARY) is
506 * set in the indirect_files field.
507 * The temps[] array above is unused then.
509 LLVMValueRef temps_array
;
511 /** bitmask indicating which register files are accessed indirectly */
512 unsigned indirect_files
;
516 static INLINE
struct lp_build_tgsi_soa_context
*
517 lp_soa_context(struct lp_build_tgsi_context
*bld_base
)
519 return (struct lp_build_tgsi_soa_context
*)bld_base
;
522 static INLINE
struct lp_build_tgsi_aos_context
*
523 lp_aos_context(struct lp_build_tgsi_context
*bld_base
)
525 return (struct lp_build_tgsi_aos_context
*)bld_base
;
529 lp_emit_declaration_aos(
530 struct lp_build_tgsi_aos_context
*bld
,
531 const struct tgsi_full_declaration
*decl
);
535 lp_emit_instruction_aos(
536 struct lp_build_tgsi_aos_context
*bld
,
537 const struct tgsi_full_instruction
*inst
,
538 const struct tgsi_opcode_info
*info
,
543 struct lp_build_tgsi_aos_context
*bld
,
544 const struct tgsi_full_instruction
*inst
,
548 void lp_build_fetch_args(
549 struct lp_build_tgsi_context
* bld_base
,
550 struct lp_build_emit_data
* emit_data
);
553 lp_build_tgsi_inst_llvm_aos(
554 struct lp_build_tgsi_context
* bld_base
,
555 const struct tgsi_full_instruction
*inst
);
558 lp_build_tgsi_intrinsic(
559 const struct lp_build_tgsi_action
* action
,
560 struct lp_build_tgsi_context
* bld_base
,
561 struct lp_build_emit_data
* emit_data
);
565 struct lp_build_tgsi_context
*bld_base
,
566 unsigned tgsi_opcode
,
567 struct lp_build_emit_data
* emit_data
);
570 lp_build_emit_llvm_unary(
571 struct lp_build_tgsi_context
*bld_base
,
572 unsigned tgsi_opcode
,
576 lp_build_emit_llvm_binary(
577 struct lp_build_tgsi_context
*bld_base
,
578 unsigned tgsi_opcode
,
583 lp_build_emit_llvm_ternary(
584 struct lp_build_tgsi_context
*bld_base
,
585 unsigned tgsi_opcode
,
591 lp_build_tgsi_inst_llvm(
592 struct lp_build_tgsi_context
* bld_base
,
593 const struct tgsi_full_instruction
*inst
);
597 struct lp_build_tgsi_context
*bld_base
,
598 const struct tgsi_full_instruction
*inst
,
600 const unsigned chan_index
);
604 lp_build_emit_fetch_texoffset(
605 struct lp_build_tgsi_context
*bld_base
,
606 const struct tgsi_full_instruction
*inst
,
608 const unsigned chan_index
);
612 struct lp_build_tgsi_context
* bld_base
,
613 const struct tgsi_token
*tokens
);
615 #endif /* LP_BLD_TGSI_H */