2 * Copyright © 2014 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Connor Abbott (cwabbott0@gmail.com)
28 #include "glsl_to_nir.h"
29 #include "nir_control_flow.h"
30 #include "ir_visitor.h"
31 #include "ir_hierarchical_visitor.h"
35 * pass to lower GLSL IR to NIR
37 * This will lower variable dereferences to loads/stores of corresponding
38 * variables in NIR - the variables will be converted to registers in a later
44 class nir_visitor
: public ir_visitor
47 nir_visitor(nir_shader
*shader
);
50 virtual void visit(ir_variable
*);
51 virtual void visit(ir_function
*);
52 virtual void visit(ir_function_signature
*);
53 virtual void visit(ir_loop
*);
54 virtual void visit(ir_if
*);
55 virtual void visit(ir_discard
*);
56 virtual void visit(ir_loop_jump
*);
57 virtual void visit(ir_return
*);
58 virtual void visit(ir_call
*);
59 virtual void visit(ir_assignment
*);
60 virtual void visit(ir_emit_vertex
*);
61 virtual void visit(ir_end_primitive
*);
62 virtual void visit(ir_expression
*);
63 virtual void visit(ir_swizzle
*);
64 virtual void visit(ir_texture
*);
65 virtual void visit(ir_constant
*);
66 virtual void visit(ir_dereference_variable
*);
67 virtual void visit(ir_dereference_record
*);
68 virtual void visit(ir_dereference_array
*);
69 virtual void visit(ir_barrier
*);
71 void create_function(ir_function
*ir
);
74 void create_overload(ir_function_signature
*ir
, nir_function
*function
);
75 void add_instr(nir_instr
*instr
, unsigned num_components
);
76 nir_src
evaluate_rvalue(ir_rvalue
*ir
);
78 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
);
79 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
);
80 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
,
82 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
,
83 nir_src src2
, nir_src src3
);
88 nir_function_impl
*impl
;
89 exec_list
*cf_node_list
;
90 nir_instr
*result
; /* result of the expression tree last visited */
92 nir_deref_var
*evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
);
94 /* the head of the dereference chain we're creating */
95 nir_deref_var
*deref_head
;
96 /* the tail of the dereference chain we're creating */
97 nir_deref
*deref_tail
;
99 nir_variable
*var
; /* variable created by ir_variable visitor */
101 /* whether the IR we're operating on is per-function or global */
104 /* map of ir_variable -> nir_variable */
105 struct hash_table
*var_table
;
107 /* map of ir_function_signature -> nir_function_overload */
108 struct hash_table
*overload_table
;
112 * This visitor runs before the main visitor, calling create_function() for
113 * each function so that the main visitor can resolve forward references in
117 class nir_function_visitor
: public ir_hierarchical_visitor
120 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
123 virtual ir_visitor_status
visit_enter(ir_function
*);
126 nir_visitor
*visitor
;
129 }; /* end of anonymous namespace */
132 glsl_to_nir(struct gl_shader
*sh
, const nir_shader_compiler_options
*options
)
134 nir_shader
*shader
= nir_shader_create(NULL
, sh
->Stage
, options
);
136 nir_visitor
v1(shader
);
137 nir_function_visitor
v2(&v1
);
139 visit_exec_list(sh
->ir
, &v1
);
141 nir_lower_outputs_to_temporaries(shader
);
146 nir_visitor::nir_visitor(nir_shader
*shader
)
148 this->supports_ints
= shader
->options
->native_integers
;
149 this->shader
= shader
;
150 this->is_global
= true;
151 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
152 _mesa_key_pointer_equal
);
153 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
154 _mesa_key_pointer_equal
);
157 nir_visitor::~nir_visitor()
159 _mesa_hash_table_destroy(this->var_table
, NULL
);
160 _mesa_hash_table_destroy(this->overload_table
, NULL
);
164 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
167 ralloc_steal(mem_ctx
, this->deref_head
);
168 return this->deref_head
;
171 static nir_constant
*
172 constant_copy(ir_constant
*ir
, void *mem_ctx
)
177 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
179 unsigned total_elems
= ir
->type
->components();
181 switch (ir
->type
->base_type
) {
183 for (i
= 0; i
< total_elems
; i
++)
184 ret
->value
.u
[i
] = ir
->value
.u
[i
];
188 for (i
= 0; i
< total_elems
; i
++)
189 ret
->value
.i
[i
] = ir
->value
.i
[i
];
192 case GLSL_TYPE_FLOAT
:
193 for (i
= 0; i
< total_elems
; i
++)
194 ret
->value
.f
[i
] = ir
->value
.f
[i
];
198 for (i
= 0; i
< total_elems
; i
++)
199 ret
->value
.b
[i
] = ir
->value
.b
[i
];
202 case GLSL_TYPE_STRUCT
:
203 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
206 foreach_in_list(ir_constant
, field
, &ir
->components
) {
207 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
212 case GLSL_TYPE_ARRAY
:
213 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
216 for (i
= 0; i
< ir
->type
->length
; i
++)
217 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
221 unreachable("not reached");
228 nir_visitor::visit(ir_variable
*ir
)
230 nir_variable
*var
= ralloc(shader
, nir_variable
);
231 var
->type
= ir
->type
;
232 var
->name
= ralloc_strdup(var
, ir
->name
);
234 if (ir
->is_interface_instance() && ir
->get_max_ifc_array_access() != NULL
) {
235 unsigned size
= ir
->get_interface_type()->length
;
236 var
->max_ifc_array_access
= ralloc_array(var
, unsigned, size
);
237 memcpy(var
->max_ifc_array_access
, ir
->get_max_ifc_array_access(),
238 size
* sizeof(unsigned));
240 var
->max_ifc_array_access
= NULL
;
243 var
->data
.read_only
= ir
->data
.read_only
;
244 var
->data
.centroid
= ir
->data
.centroid
;
245 var
->data
.sample
= ir
->data
.sample
;
246 var
->data
.invariant
= ir
->data
.invariant
;
247 var
->data
.location
= ir
->data
.location
;
249 switch(ir
->data
.mode
) {
251 case ir_var_temporary
:
253 var
->data
.mode
= nir_var_global
;
255 var
->data
.mode
= nir_var_local
;
258 case ir_var_function_in
:
259 case ir_var_function_out
:
260 case ir_var_function_inout
:
261 case ir_var_const_in
:
262 var
->data
.mode
= nir_var_local
;
265 case ir_var_shader_in
:
266 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
267 ir
->data
.location
== VARYING_SLOT_FACE
) {
268 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
269 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
270 var
->data
.mode
= nir_var_system_value
;
272 var
->data
.mode
= nir_var_shader_in
;
276 case ir_var_shader_out
:
277 var
->data
.mode
= nir_var_shader_out
;
281 var
->data
.mode
= nir_var_uniform
;
284 case ir_var_shader_storage
:
285 var
->data
.mode
= nir_var_shader_storage
;
288 case ir_var_system_value
:
289 var
->data
.mode
= nir_var_system_value
;
293 unreachable("not reached");
296 var
->data
.interpolation
= ir
->data
.interpolation
;
297 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
298 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
299 var
->data
.explicit_location
= ir
->data
.explicit_location
;
300 var
->data
.explicit_index
= ir
->data
.explicit_index
;
301 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
302 var
->data
.has_initializer
= ir
->data
.has_initializer
;
303 var
->data
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
304 var
->data
.location_frac
= ir
->data
.location_frac
;
305 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
306 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
308 switch (ir
->data
.depth_layout
) {
309 case ir_depth_layout_none
:
310 var
->data
.depth_layout
= nir_depth_layout_none
;
312 case ir_depth_layout_any
:
313 var
->data
.depth_layout
= nir_depth_layout_any
;
315 case ir_depth_layout_greater
:
316 var
->data
.depth_layout
= nir_depth_layout_greater
;
318 case ir_depth_layout_less
:
319 var
->data
.depth_layout
= nir_depth_layout_less
;
321 case ir_depth_layout_unchanged
:
322 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
325 unreachable("not reached");
328 var
->data
.index
= ir
->data
.index
;
329 var
->data
.binding
= ir
->data
.binding
;
330 /* XXX Get rid of buffer_index */
331 var
->data
.atomic
.buffer_index
= ir
->data
.binding
;
332 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
333 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
334 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
335 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
336 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
337 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
338 var
->data
.image
.format
= ir
->data
.image_format
;
339 var
->data
.max_array_access
= ir
->data
.max_array_access
;
341 var
->num_state_slots
= ir
->get_num_state_slots();
342 if (var
->num_state_slots
> 0) {
343 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
344 var
->num_state_slots
);
346 ir_state_slot
*state_slots
= ir
->get_state_slots();
347 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
348 for (unsigned j
= 0; j
< 5; j
++)
349 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
350 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
353 var
->state_slots
= NULL
;
356 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
358 var
->interface_type
= ir
->get_interface_type();
360 switch (var
->data
.mode
) {
362 exec_list_push_tail(&impl
->locals
, &var
->node
);
366 exec_list_push_tail(&shader
->globals
, &var
->node
);
369 case nir_var_shader_in
:
370 exec_list_push_tail(&shader
->inputs
, &var
->node
);
373 case nir_var_shader_out
:
374 exec_list_push_tail(&shader
->outputs
, &var
->node
);
377 case nir_var_uniform
:
378 case nir_var_shader_storage
:
379 exec_list_push_tail(&shader
->uniforms
, &var
->node
);
382 case nir_var_system_value
:
383 exec_list_push_tail(&shader
->system_values
, &var
->node
);
387 unreachable("not reached");
390 _mesa_hash_table_insert(var_table
, ir
, var
);
395 nir_function_visitor::visit_enter(ir_function
*ir
)
397 visitor
->create_function(ir
);
398 return visit_continue_with_parent
;
403 nir_visitor::create_function(ir_function
*ir
)
405 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
406 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
407 create_overload(sig
, func
);
414 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
416 if (ir
->is_intrinsic
)
419 nir_function_overload
*overload
= nir_function_overload_create(function
);
421 unsigned num_params
= ir
->parameters
.length();
422 overload
->num_params
= num_params
;
423 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
426 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
427 switch (param
->data
.mode
) {
428 case ir_var_function_in
:
429 overload
->params
[i
].param_type
= nir_parameter_in
;
432 case ir_var_function_out
:
433 overload
->params
[i
].param_type
= nir_parameter_out
;
436 case ir_var_function_inout
:
437 overload
->params
[i
].param_type
= nir_parameter_inout
;
441 unreachable("not reached");
444 overload
->params
[i
].type
= param
->type
;
448 overload
->return_type
= ir
->return_type
;
450 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
454 nir_visitor::visit(ir_function
*ir
)
456 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
461 nir_visitor::visit(ir_function_signature
*ir
)
463 if (ir
->is_intrinsic
)
466 struct hash_entry
*entry
=
467 _mesa_hash_table_search(this->overload_table
, ir
);
470 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
472 if (ir
->is_defined
) {
473 nir_function_impl
*impl
= nir_function_impl_create(overload
);
476 unsigned num_params
= overload
->num_params
;
477 impl
->num_params
= num_params
;
478 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
480 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
482 impl
->params
[i
] = this->var
;
486 if (overload
->return_type
== glsl_type::void_type
) {
487 impl
->return_var
= NULL
;
489 impl
->return_var
= ralloc(this->shader
, nir_variable
);
490 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
492 impl
->return_var
->type
= overload
->return_type
;
495 this->is_global
= false;
497 this->cf_node_list
= &impl
->body
;
498 visit_exec_list(&ir
->body
, this);
500 this->is_global
= true;
502 overload
->impl
= NULL
;
507 nir_visitor::visit(ir_loop
*ir
)
509 exec_list
*old_list
= this->cf_node_list
;
511 nir_loop
*loop
= nir_loop_create(this->shader
);
512 nir_cf_node_insert_end(old_list
, &loop
->cf_node
);
513 this->cf_node_list
= &loop
->body
;
514 visit_exec_list(&ir
->body_instructions
, this);
516 this->cf_node_list
= old_list
;
520 nir_visitor::visit(ir_if
*ir
)
522 nir_src condition
= evaluate_rvalue(ir
->condition
);
524 exec_list
*old_list
= this->cf_node_list
;
526 nir_if
*if_stmt
= nir_if_create(this->shader
);
527 if_stmt
->condition
= condition
;
528 nir_cf_node_insert_end(old_list
, &if_stmt
->cf_node
);
530 this->cf_node_list
= &if_stmt
->then_list
;
531 visit_exec_list(&ir
->then_instructions
, this);
533 this->cf_node_list
= &if_stmt
->else_list
;
534 visit_exec_list(&ir
->else_instructions
, this);
536 this->cf_node_list
= old_list
;
540 nir_visitor::visit(ir_discard
*ir
)
543 * discards aren't treated as control flow, because before we lower them
544 * they can appear anywhere in the shader and the stuff after them may still
545 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
546 * discards will be immediately followed by a return.
549 nir_intrinsic_instr
*discard
;
551 discard
= nir_intrinsic_instr_create(this->shader
,
552 nir_intrinsic_discard_if
);
553 discard
->src
[0] = evaluate_rvalue(ir
->condition
);
555 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
557 nir_instr_insert_after_cf_list(this->cf_node_list
, &discard
->instr
);
561 nir_visitor::visit(ir_emit_vertex
*ir
)
563 nir_intrinsic_instr
*instr
=
564 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
565 instr
->const_index
[0] = ir
->stream_id();
566 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
570 nir_visitor::visit(ir_end_primitive
*ir
)
572 nir_intrinsic_instr
*instr
=
573 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
574 instr
->const_index
[0] = ir
->stream_id();
575 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
579 nir_visitor::visit(ir_loop_jump
*ir
)
583 case ir_loop_jump::jump_break
:
584 type
= nir_jump_break
;
586 case ir_loop_jump::jump_continue
:
587 type
= nir_jump_continue
;
590 unreachable("not reached");
593 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
594 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
598 nir_visitor::visit(ir_return
*ir
)
600 if (ir
->value
!= NULL
) {
601 nir_intrinsic_instr
*copy
=
602 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
604 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
605 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
608 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
609 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
613 nir_visitor::visit(ir_call
*ir
)
615 if (ir
->callee
->is_intrinsic
) {
617 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
618 op
= nir_intrinsic_atomic_counter_read_var
;
619 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
620 op
= nir_intrinsic_atomic_counter_inc_var
;
621 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
622 op
= nir_intrinsic_atomic_counter_dec_var
;
623 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
624 op
= nir_intrinsic_image_load
;
625 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
626 op
= nir_intrinsic_image_store
;
627 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
628 op
= nir_intrinsic_image_atomic_add
;
629 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
630 op
= nir_intrinsic_image_atomic_min
;
631 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
632 op
= nir_intrinsic_image_atomic_max
;
633 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
634 op
= nir_intrinsic_image_atomic_and
;
635 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
636 op
= nir_intrinsic_image_atomic_or
;
637 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
638 op
= nir_intrinsic_image_atomic_xor
;
639 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
640 op
= nir_intrinsic_image_atomic_exchange
;
641 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
642 op
= nir_intrinsic_image_atomic_comp_swap
;
643 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
644 op
= nir_intrinsic_memory_barrier
;
645 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_size") == 0) {
646 op
= nir_intrinsic_image_size
;
647 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_samples") == 0) {
648 op
= nir_intrinsic_image_samples
;
650 unreachable("not reached");
653 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
656 case nir_intrinsic_atomic_counter_read_var
:
657 case nir_intrinsic_atomic_counter_inc_var
:
658 case nir_intrinsic_atomic_counter_dec_var
: {
659 ir_dereference
*param
=
660 (ir_dereference
*) ir
->actual_parameters
.get_head();
661 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
662 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
665 case nir_intrinsic_image_load
:
666 case nir_intrinsic_image_store
:
667 case nir_intrinsic_image_atomic_add
:
668 case nir_intrinsic_image_atomic_min
:
669 case nir_intrinsic_image_atomic_max
:
670 case nir_intrinsic_image_atomic_and
:
671 case nir_intrinsic_image_atomic_or
:
672 case nir_intrinsic_image_atomic_xor
:
673 case nir_intrinsic_image_atomic_exchange
:
674 case nir_intrinsic_image_atomic_comp_swap
:
675 case nir_intrinsic_image_samples
:
676 case nir_intrinsic_image_size
: {
677 nir_ssa_undef_instr
*instr_undef
=
678 nir_ssa_undef_instr_create(shader
, 1);
679 nir_instr_insert_after_cf_list(this->cf_node_list
,
680 &instr_undef
->instr
);
682 /* Set the image variable dereference. */
683 exec_node
*param
= ir
->actual_parameters
.get_head();
684 ir_dereference
*image
= (ir_dereference
*)param
;
685 const glsl_type
*type
=
686 image
->variable_referenced()->type
->without_array();
688 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
689 param
= param
->get_next();
691 /* Set the intrinsic destination. */
692 if (ir
->return_deref
) {
693 const nir_intrinsic_info
*info
=
694 &nir_intrinsic_infos
[instr
->intrinsic
];
695 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
696 info
->dest_components
, NULL
);
699 if (op
== nir_intrinsic_image_size
||
700 op
== nir_intrinsic_image_samples
)
703 /* Set the address argument, extending the coordinate vector to four
706 const nir_src src_addr
= evaluate_rvalue((ir_dereference
*)param
);
707 nir_alu_instr
*instr_addr
= nir_alu_instr_create(shader
, nir_op_vec4
);
708 nir_ssa_dest_init(&instr_addr
->instr
, &instr_addr
->dest
.dest
, 4, NULL
);
710 for (int i
= 0; i
< 4; i
++) {
711 if (i
< type
->coordinate_components()) {
712 instr_addr
->src
[i
].src
= src_addr
;
713 instr_addr
->src
[i
].swizzle
[0] = i
;
715 instr_addr
->src
[i
].src
= nir_src_for_ssa(&instr_undef
->def
);
719 nir_instr_insert_after_cf_list(cf_node_list
, &instr_addr
->instr
);
720 instr
->src
[0] = nir_src_for_ssa(&instr_addr
->dest
.dest
.ssa
);
721 param
= param
->get_next();
723 /* Set the sample argument, which is undefined for single-sample
726 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
727 instr
->src
[1] = evaluate_rvalue((ir_dereference
*)param
);
728 param
= param
->get_next();
730 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
733 /* Set the intrinsic parameters. */
734 if (!param
->is_tail_sentinel()) {
735 instr
->src
[2] = evaluate_rvalue((ir_dereference
*)param
);
736 param
= param
->get_next();
739 if (!param
->is_tail_sentinel()) {
740 instr
->src
[3] = evaluate_rvalue((ir_dereference
*)param
);
741 param
= param
->get_next();
745 case nir_intrinsic_memory_barrier
:
748 unreachable("not reached");
751 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
753 if (ir
->return_deref
) {
754 nir_intrinsic_instr
*store_instr
=
755 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
756 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
758 store_instr
->variables
[0] =
759 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
760 store_instr
->src
[0] = nir_src_for_ssa(&instr
->dest
.ssa
);
762 nir_instr_insert_after_cf_list(this->cf_node_list
,
763 &store_instr
->instr
);
769 struct hash_entry
*entry
=
770 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
772 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
774 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
777 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
778 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
782 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
783 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
787 nir_visitor::visit(ir_assignment
*ir
)
789 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
791 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
792 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
793 /* We're doing a plain-as-can-be copy, so emit a copy_var */
794 nir_intrinsic_instr
*copy
=
795 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
797 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
798 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
801 nir_if
*if_stmt
= nir_if_create(this->shader
);
802 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
803 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
804 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
806 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
811 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
813 ir
->lhs
->accept(this);
814 nir_deref_var
*lhs_deref
= this->deref_head
;
815 nir_src src
= evaluate_rvalue(ir
->rhs
);
817 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
819 * We have no good way to update only part of a variable, so just load
820 * the LHS and do a vec operation to combine the old with the new, and
822 * back into the LHS. Copy propagation should get rid of the mess.
825 nir_intrinsic_instr
*load
=
826 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
827 load
->num_components
= ir
->lhs
->type
->vector_elements
;
828 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
829 load
->variables
[0] = lhs_deref
;
830 ralloc_steal(load
, load
->variables
[0]);
831 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
834 switch (ir
->lhs
->type
->vector_elements
) {
835 case 1: vec_op
= nir_op_imov
; break;
836 case 2: vec_op
= nir_op_vec2
; break;
837 case 3: vec_op
= nir_op_vec3
; break;
838 case 4: vec_op
= nir_op_vec4
; break;
839 default: unreachable("Invalid number of components"); break;
841 nir_alu_instr
*vec
= nir_alu_instr_create(this->shader
, vec_op
);
842 nir_ssa_dest_init(&vec
->instr
, &vec
->dest
.dest
, num_components
, NULL
);
843 vec
->dest
.write_mask
= (1 << num_components
) - 1;
845 unsigned component
= 0;
846 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
847 if (ir
->write_mask
& (1 << i
)) {
848 vec
->src
[i
].src
= src
;
850 /* GLSL IR will give us the input to the write-masked assignment
851 * in a single packed vector. So, for example, if the
852 * writemask is xzw, then we have to swizzle x -> x, y -> z,
853 * and z -> w and get the y component from the load.
855 vec
->src
[i
].swizzle
[0] = component
++;
857 vec
->src
[i
].src
.is_ssa
= true;
858 vec
->src
[i
].src
.ssa
= &load
->dest
.ssa
;
859 vec
->src
[i
].swizzle
[0] = i
;
863 nir_instr_insert_after_cf_list(this->cf_node_list
, &vec
->instr
);
866 src
.ssa
= &vec
->dest
.dest
.ssa
;
869 nir_intrinsic_instr
*store
=
870 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
871 store
->num_components
= ir
->lhs
->type
->vector_elements
;
872 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
873 store
->variables
[0] = nir_deref_as_var(store_deref
);
877 nir_if
*if_stmt
= nir_if_create(this->shader
);
878 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
879 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
880 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
882 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
887 * Given an instruction, returns a pointer to its destination or NULL if there
890 * Note that this only handles instructions we generate at this level.
893 get_instr_dest(nir_instr
*instr
)
895 nir_alu_instr
*alu_instr
;
896 nir_intrinsic_instr
*intrinsic_instr
;
897 nir_tex_instr
*tex_instr
;
899 switch (instr
->type
) {
900 case nir_instr_type_alu
:
901 alu_instr
= nir_instr_as_alu(instr
);
902 return &alu_instr
->dest
.dest
;
904 case nir_instr_type_intrinsic
:
905 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
906 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
907 return &intrinsic_instr
->dest
;
911 case nir_instr_type_tex
:
912 tex_instr
= nir_instr_as_tex(instr
);
913 return &tex_instr
->dest
;
916 unreachable("not reached");
923 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
925 nir_dest
*dest
= get_instr_dest(instr
);
927 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
929 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
930 this->result
= instr
;
934 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
937 if (ir
->as_dereference() || ir
->as_constant()) {
939 * A dereference is being used on the right hand side, which means we
940 * must emit a variable load.
943 nir_intrinsic_instr
*load_instr
=
944 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
945 load_instr
->num_components
= ir
->type
->vector_elements
;
946 load_instr
->variables
[0] = this->deref_head
;
947 ralloc_steal(load_instr
, load_instr
->variables
[0]);
948 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
951 nir_dest
*dest
= get_instr_dest(this->result
);
952 assert(dest
->is_ssa
);
954 return nir_src_for_ssa(&dest
->ssa
);
958 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
960 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
961 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
962 instr
->src
[i
].src
= srcs
[i
];
963 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
964 add_instr(&instr
->instr
, dest_size
);
969 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
971 assert(nir_op_infos
[op
].num_inputs
== 1);
972 return emit(op
, dest_size
, &src1
);
976 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
979 assert(nir_op_infos
[op
].num_inputs
== 2);
980 nir_src srcs
[] = { src1
, src2
};
981 return emit(op
, dest_size
, srcs
);
985 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
986 nir_src src2
, nir_src src3
)
988 assert(nir_op_infos
[op
].num_inputs
== 3);
989 nir_src srcs
[] = { src1
, src2
, src3
};
990 return emit(op
, dest_size
, srcs
);
994 nir_visitor::visit(ir_expression
*ir
)
996 /* Some special cases */
997 switch (ir
->operation
) {
998 case ir_binop_ubo_load
: {
999 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
1001 nir_intrinsic_op op
;
1003 op
= nir_intrinsic_load_ubo
;
1005 op
= nir_intrinsic_load_ubo_indirect
;
1007 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
1008 load
->num_components
= ir
->type
->vector_elements
;
1009 load
->const_index
[0] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
1010 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
1012 load
->src
[1] = evaluate_rvalue(ir
->operands
[1]);
1013 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1016 * In UBO's, a true boolean value is any non-zero value, but we consider
1017 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1020 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
1021 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
, 1);
1022 const_zero
->value
.u
[0] = 0;
1023 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
1025 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
1026 compare
->src
[0].src
.is_ssa
= true;
1027 compare
->src
[0].src
.ssa
= &load
->dest
.ssa
;
1028 compare
->src
[1].src
.is_ssa
= true;
1029 compare
->src
[1].src
.ssa
= &const_zero
->def
;
1030 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1031 compare
->src
[1].swizzle
[i
] = 0;
1032 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
1034 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
1040 case ir_unop_interpolate_at_centroid
:
1041 case ir_binop_interpolate_at_offset
:
1042 case ir_binop_interpolate_at_sample
: {
1043 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1044 ir_swizzle
*swizzle
= NULL
;
1046 /* the api does not allow a swizzle here, but the varying packing code
1047 * may have pushed one into here.
1049 swizzle
= ir
->operands
[0]->as_swizzle();
1051 deref
= swizzle
->val
->as_dereference();
1055 deref
->accept(this);
1057 nir_intrinsic_op op
;
1058 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1059 switch (ir
->operation
) {
1060 case ir_unop_interpolate_at_centroid
:
1061 op
= nir_intrinsic_interp_var_at_centroid
;
1063 case ir_binop_interpolate_at_offset
:
1064 op
= nir_intrinsic_interp_var_at_offset
;
1066 case ir_binop_interpolate_at_sample
:
1067 op
= nir_intrinsic_interp_var_at_sample
;
1070 unreachable("Invalid interpolation intrinsic");
1073 /* This case can happen if the vertex shader does not write the
1074 * given varying. In this case, the linker will lower it to a
1075 * global variable. Since interpolating a variable makes no
1076 * sense, we'll just turn it into a load which will probably
1077 * eventually end up as an SSA definition.
1079 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1080 op
= nir_intrinsic_load_var
;
1083 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1084 intrin
->num_components
= deref
->type
->vector_elements
;
1085 intrin
->variables
[0] = this->deref_head
;
1086 ralloc_steal(intrin
, intrin
->variables
[0]);
1088 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1089 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1090 intrin
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
1092 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
1095 nir_alu_instr
*mov
= nir_alu_instr_create(shader
, nir_op_imov
);
1096 mov
->dest
.write_mask
= (1 << swizzle
->type
->vector_elements
) - 1;
1097 mov
->src
[0].src
.is_ssa
= true;
1098 mov
->src
[0].src
.ssa
= &intrin
->dest
.ssa
;
1100 mov
->src
[0].swizzle
[0] = swizzle
->mask
.x
;
1101 mov
->src
[0].swizzle
[1] = swizzle
->mask
.y
;
1102 mov
->src
[0].swizzle
[2] = swizzle
->mask
.z
;
1103 mov
->src
[0].swizzle
[3] = swizzle
->mask
.w
;
1104 for (unsigned i
= deref
->type
->vector_elements
; i
< 4; i
++)
1105 mov
->src
[0].swizzle
[i
] = 0;
1107 add_instr(&mov
->instr
, swizzle
->type
->vector_elements
);
1118 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1119 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1121 glsl_base_type types
[4];
1122 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1124 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1126 types
[i
] = GLSL_TYPE_FLOAT
;
1128 glsl_base_type out_type
;
1130 out_type
= ir
->type
->base_type
;
1132 out_type
= GLSL_TYPE_FLOAT
;
1134 unsigned dest_size
= ir
->type
->vector_elements
;
1136 nir_alu_instr
*instr
;
1139 switch (ir
->operation
) {
1140 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
1141 case ir_unop_logic_not
:
1142 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1145 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fneg
: nir_op_ineg
,
1149 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fabs
: nir_op_iabs
,
1152 case ir_unop_saturate
:
1153 assert(types
[0] == GLSL_TYPE_FLOAT
);
1154 instr
= emit(nir_op_fsat
, dest_size
, srcs
);
1157 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1160 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1161 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1162 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1163 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1164 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1165 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1166 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1168 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1171 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1174 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1176 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1177 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1178 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1179 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1180 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1183 case ir_unop_bitcast_i2f
:
1184 case ir_unop_bitcast_f2i
:
1185 case ir_unop_bitcast_u2f
:
1186 case ir_unop_bitcast_f2u
:
1187 case ir_unop_subroutine_to_int
:
1189 emit(nir_op_imov
, dest_size
, srcs
);
1192 switch (ir
->operands
[0]->type
->vector_elements
) {
1194 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1198 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1202 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1206 unreachable("not reached");
1209 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1210 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1211 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1212 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1213 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1214 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1215 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1216 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1217 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1218 case ir_unop_dFdx_fine
: emit(nir_op_fddx_fine
, dest_size
, srcs
); break;
1219 case ir_unop_dFdy_fine
: emit(nir_op_fddy_fine
, dest_size
, srcs
); break;
1220 case ir_unop_dFdx_coarse
: emit(nir_op_fddx_coarse
, dest_size
, srcs
); break;
1221 case ir_unop_dFdy_coarse
: emit(nir_op_fddy_coarse
, dest_size
, srcs
); break;
1222 case ir_unop_pack_snorm_2x16
:
1223 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1225 case ir_unop_pack_snorm_4x8
:
1226 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1228 case ir_unop_pack_unorm_2x16
:
1229 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1231 case ir_unop_pack_unorm_4x8
:
1232 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1234 case ir_unop_pack_half_2x16
:
1235 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1237 case ir_unop_unpack_snorm_2x16
:
1238 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1240 case ir_unop_unpack_snorm_4x8
:
1241 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1243 case ir_unop_unpack_unorm_2x16
:
1244 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1246 case ir_unop_unpack_unorm_4x8
:
1247 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1249 case ir_unop_unpack_half_2x16
:
1250 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1252 case ir_unop_unpack_half_2x16_split_x
:
1253 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1255 case ir_unop_unpack_half_2x16_split_y
:
1256 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1258 case ir_unop_bitfield_reverse
:
1259 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1261 case ir_unop_bit_count
:
1262 emit(nir_op_bit_count
, dest_size
, srcs
);
1264 case ir_unop_find_msb
:
1266 case GLSL_TYPE_UINT
:
1267 emit(nir_op_ufind_msb
, dest_size
, srcs
);
1270 emit(nir_op_ifind_msb
, dest_size
, srcs
);
1273 unreachable("Invalid type for findMSB()");
1276 case ir_unop_find_lsb
:
1277 emit(nir_op_find_lsb
, dest_size
, srcs
);
1281 switch (ir
->type
->vector_elements
) {
1283 switch (ir
->operands
[0]->type
->vector_elements
) {
1284 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1285 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1286 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1287 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1288 default: unreachable("not reached");
1292 switch (ir
->operands
[0]->type
->vector_elements
) {
1293 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1294 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1295 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1296 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1297 default: unreachable("not reached");
1301 switch (ir
->operands
[0]->type
->vector_elements
) {
1302 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1303 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1304 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1305 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1306 default: unreachable("not reached");
1310 switch (ir
->operands
[0]->type
->vector_elements
) {
1311 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1312 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1313 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1314 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1315 default: unreachable("not reached");
1319 unreachable("not reached");
1330 case ir_binop_bit_and
:
1331 case ir_binop_bit_or
:
1332 case ir_binop_bit_xor
:
1333 case ir_binop_logic_and
:
1334 case ir_binop_logic_or
:
1335 case ir_binop_logic_xor
:
1336 case ir_binop_lshift
:
1337 case ir_binop_rshift
:
1338 switch (ir
->operation
) {
1340 if (out_type
== GLSL_TYPE_FLOAT
)
1346 if (out_type
== GLSL_TYPE_FLOAT
)
1352 if (out_type
== GLSL_TYPE_FLOAT
)
1358 if (out_type
== GLSL_TYPE_FLOAT
)
1360 else if (out_type
== GLSL_TYPE_INT
)
1366 if (out_type
== GLSL_TYPE_FLOAT
)
1372 if (out_type
== GLSL_TYPE_FLOAT
)
1374 else if (out_type
== GLSL_TYPE_INT
)
1380 if (out_type
== GLSL_TYPE_FLOAT
)
1382 else if (out_type
== GLSL_TYPE_INT
)
1387 case ir_binop_bit_and
:
1390 case ir_binop_bit_or
:
1393 case ir_binop_bit_xor
:
1396 case ir_binop_logic_and
:
1402 case ir_binop_logic_or
:
1408 case ir_binop_logic_xor
:
1414 case ir_binop_lshift
:
1417 case ir_binop_rshift
:
1418 if (out_type
== GLSL_TYPE_INT
)
1428 unreachable("not reached");
1431 instr
= emit(op
, dest_size
, srcs
);
1433 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1434 ir
->operands
[1]->type
->vector_elements
== 1) {
1435 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1437 instr
->src
[1].swizzle
[i
] = 0;
1441 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1442 ir
->operands
[0]->type
->vector_elements
== 1) {
1443 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1445 instr
->src
[0].swizzle
[i
] = 0;
1450 case ir_binop_imul_high
:
1451 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1454 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1455 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1457 if (supports_ints
) {
1458 if (types
[0] == GLSL_TYPE_FLOAT
)
1459 emit(nir_op_flt
, dest_size
, srcs
);
1460 else if (types
[0] == GLSL_TYPE_INT
)
1461 emit(nir_op_ilt
, dest_size
, srcs
);
1463 emit(nir_op_ult
, dest_size
, srcs
);
1465 emit(nir_op_slt
, dest_size
, srcs
);
1468 case ir_binop_greater
:
1469 if (supports_ints
) {
1470 if (types
[0] == GLSL_TYPE_FLOAT
)
1471 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1472 else if (types
[0] == GLSL_TYPE_INT
)
1473 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1475 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1477 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1480 case ir_binop_lequal
:
1481 if (supports_ints
) {
1482 if (types
[0] == GLSL_TYPE_FLOAT
)
1483 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1484 else if (types
[0] == GLSL_TYPE_INT
)
1485 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1487 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1489 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1492 case ir_binop_gequal
:
1493 if (supports_ints
) {
1494 if (types
[0] == GLSL_TYPE_FLOAT
)
1495 emit(nir_op_fge
, dest_size
, srcs
);
1496 else if (types
[0] == GLSL_TYPE_INT
)
1497 emit(nir_op_ige
, dest_size
, srcs
);
1499 emit(nir_op_uge
, dest_size
, srcs
);
1501 emit(nir_op_slt
, dest_size
, srcs
);
1504 case ir_binop_equal
:
1505 if (supports_ints
) {
1506 if (types
[0] == GLSL_TYPE_FLOAT
)
1507 emit(nir_op_feq
, dest_size
, srcs
);
1509 emit(nir_op_ieq
, dest_size
, srcs
);
1511 emit(nir_op_seq
, dest_size
, srcs
);
1514 case ir_binop_nequal
:
1515 if (supports_ints
) {
1516 if (types
[0] == GLSL_TYPE_FLOAT
)
1517 emit(nir_op_fne
, dest_size
, srcs
);
1519 emit(nir_op_ine
, dest_size
, srcs
);
1521 emit(nir_op_sne
, dest_size
, srcs
);
1524 case ir_binop_all_equal
:
1525 if (supports_ints
) {
1526 if (types
[0] == GLSL_TYPE_FLOAT
) {
1527 switch (ir
->operands
[0]->type
->vector_elements
) {
1528 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1529 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1530 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1531 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1533 unreachable("not reached");
1536 switch (ir
->operands
[0]->type
->vector_elements
) {
1537 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1538 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1539 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1540 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1542 unreachable("not reached");
1546 switch (ir
->operands
[0]->type
->vector_elements
) {
1547 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1548 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1549 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1550 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1552 unreachable("not reached");
1556 case ir_binop_any_nequal
:
1557 if (supports_ints
) {
1558 if (types
[0] == GLSL_TYPE_FLOAT
) {
1559 switch (ir
->operands
[0]->type
->vector_elements
) {
1560 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1561 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1562 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1563 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1565 unreachable("not reached");
1568 switch (ir
->operands
[0]->type
->vector_elements
) {
1569 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1570 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1571 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1572 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1574 unreachable("not reached");
1578 switch (ir
->operands
[0]->type
->vector_elements
) {
1579 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1580 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1581 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1582 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1584 unreachable("not reached");
1589 switch (ir
->operands
[0]->type
->vector_elements
) {
1590 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1591 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1592 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1594 unreachable("not reached");
1598 case ir_binop_pack_half_2x16_split
:
1599 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1601 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1602 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1603 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1605 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1606 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1607 ir
->operands
[2]->type
->vector_elements
== 1) {
1608 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1610 instr
->src
[2].swizzle
[i
] = 0;
1616 emit(nir_op_bcsel
, dest_size
, srcs
);
1618 emit(nir_op_fcsel
, dest_size
, srcs
);
1621 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1622 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1623 instr
->src
[0].swizzle
[i
] = 0;
1626 case ir_triop_bitfield_extract
:
1627 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1628 nir_op_ubitfield_extract
, dest_size
, srcs
);
1629 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1630 instr
->src
[1].swizzle
[i
] = 0;
1631 instr
->src
[2].swizzle
[i
] = 0;
1634 case ir_quadop_bitfield_insert
:
1635 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1636 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1637 instr
->src
[2].swizzle
[i
] = 0;
1638 instr
->src
[3].swizzle
[i
] = 0;
1641 case ir_quadop_vector
:
1642 switch (ir
->type
->vector_elements
) {
1643 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1644 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1645 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1646 default: unreachable("not reached");
1651 unreachable("not reached");
1656 nir_visitor::visit(ir_swizzle
*ir
)
1658 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1659 ir
->type
->vector_elements
,
1660 evaluate_rvalue(ir
->val
));
1662 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1663 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1664 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1668 nir_visitor::visit(ir_texture
*ir
)
1675 num_srcs
= 1; /* coordinate */
1680 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1681 num_srcs
= 2; /* coordinate, bias/lod */
1685 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1691 if (ir
->lod_info
.lod
!= NULL
)
1692 num_srcs
= 2; /* coordinate, lod */
1694 num_srcs
= 1; /* coordinate */
1698 op
= nir_texop_txf_ms
;
1699 num_srcs
= 2; /* coordinate, sample_index */
1704 if (ir
->lod_info
.lod
!= NULL
)
1705 num_srcs
= 1; /* lod */
1712 num_srcs
= 1; /* coordinate */
1717 num_srcs
= 1; /* coordinate */
1720 case ir_query_levels
:
1721 op
= nir_texop_query_levels
;
1725 case ir_texture_samples
:
1726 op
= nir_texop_texture_samples
;
1731 unreachable("not reached");
1734 if (ir
->projector
!= NULL
)
1736 if (ir
->shadow_comparitor
!= NULL
)
1738 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1741 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1744 instr
->sampler_dim
=
1745 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1746 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1747 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1748 if (instr
->is_shadow
)
1749 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1750 switch (ir
->type
->base_type
) {
1751 case GLSL_TYPE_FLOAT
:
1752 instr
->dest_type
= nir_type_float
;
1755 instr
->dest_type
= nir_type_int
;
1757 case GLSL_TYPE_UINT
:
1758 instr
->dest_type
= nir_type_unsigned
;
1761 unreachable("not reached");
1764 instr
->sampler
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1766 unsigned src_number
= 0;
1768 if (ir
->coordinate
!= NULL
) {
1769 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1770 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->coordinate
);
1771 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1775 if (ir
->projector
!= NULL
) {
1776 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->projector
);
1777 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1781 if (ir
->shadow_comparitor
!= NULL
) {
1782 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->shadow_comparitor
);
1783 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1787 if (ir
->offset
!= NULL
) {
1788 /* we don't support multiple offsets yet */
1789 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1791 ir_constant
*const_offset
= ir
->offset
->as_constant();
1792 if (const_offset
!= NULL
) {
1793 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1794 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1796 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->offset
);
1797 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1804 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.bias
);
1805 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1812 if (ir
->lod_info
.lod
!= NULL
) {
1813 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.lod
);
1814 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1820 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdx
);
1821 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1823 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdy
);
1824 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1829 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.sample_index
);
1830 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1835 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1842 assert(src_number
== num_srcs
);
1844 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
1848 nir_visitor::visit(ir_constant
*ir
)
1851 * We don't know if this variable is an an array or struct that gets
1852 * dereferenced, so do the safe thing an make it a variable with a
1853 * constant initializer and return a dereference.
1856 nir_variable
*var
= ralloc(this->shader
, nir_variable
);
1857 var
->name
= ralloc_strdup(var
, "const_temp");
1858 var
->type
= ir
->type
;
1859 var
->data
.mode
= nir_var_local
;
1860 var
->data
.read_only
= true;
1861 var
->constant_initializer
= constant_copy(ir
, var
);
1862 exec_list_push_tail(&this->impl
->locals
, &var
->node
);
1864 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1865 this->deref_tail
= &this->deref_head
->deref
;
1869 nir_visitor::visit(ir_dereference_variable
*ir
)
1871 struct hash_entry
*entry
=
1872 _mesa_hash_table_search(this->var_table
, ir
->var
);
1874 nir_variable
*var
= (nir_variable
*) entry
->data
;
1876 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
1877 this->deref_head
= deref
;
1878 this->deref_tail
= &deref
->deref
;
1882 nir_visitor::visit(ir_dereference_record
*ir
)
1884 ir
->record
->accept(this);
1886 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
1887 assert(field_index
>= 0);
1889 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
1890 deref
->deref
.type
= ir
->type
;
1891 this->deref_tail
->child
= &deref
->deref
;
1892 this->deref_tail
= &deref
->deref
;
1896 nir_visitor::visit(ir_dereference_array
*ir
)
1898 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
1899 deref
->deref
.type
= ir
->type
;
1901 ir_constant
*const_index
= ir
->array_index
->as_constant();
1902 if (const_index
!= NULL
) {
1903 deref
->deref_array_type
= nir_deref_array_type_direct
;
1904 deref
->base_offset
= const_index
->value
.u
[0];
1906 deref
->deref_array_type
= nir_deref_array_type_indirect
;
1907 deref
->indirect
= evaluate_rvalue(ir
->array_index
);
1910 ir
->array
->accept(this);
1912 this->deref_tail
->child
= &deref
->deref
;
1913 ralloc_steal(this->deref_tail
, deref
);
1914 this->deref_tail
= &deref
->deref
;
1918 nir_visitor::visit(ir_barrier
*ir
)
1920 nir_intrinsic_instr
*instr
=
1921 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
1922 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);