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 "ir_visitor.h"
30 #include "ir_hierarchical_visitor.h"
34 * pass to lower GLSL IR to NIR
36 * This will lower variable dereferences to loads/stores of corresponding
37 * variables in NIR - the variables will be converted to registers in a later
43 class nir_visitor
: public ir_visitor
46 nir_visitor(nir_shader
*shader
, bool supports_ints
);
49 virtual void visit(ir_variable
*);
50 virtual void visit(ir_function
*);
51 virtual void visit(ir_function_signature
*);
52 virtual void visit(ir_loop
*);
53 virtual void visit(ir_if
*);
54 virtual void visit(ir_discard
*);
55 virtual void visit(ir_loop_jump
*);
56 virtual void visit(ir_return
*);
57 virtual void visit(ir_call
*);
58 virtual void visit(ir_assignment
*);
59 virtual void visit(ir_emit_vertex
*);
60 virtual void visit(ir_end_primitive
*);
61 virtual void visit(ir_expression
*);
62 virtual void visit(ir_swizzle
*);
63 virtual void visit(ir_texture
*);
64 virtual void visit(ir_constant
*);
65 virtual void visit(ir_dereference_variable
*);
66 virtual void visit(ir_dereference_record
*);
67 virtual void visit(ir_dereference_array
*);
69 void create_function(ir_function
*ir
);
72 void create_overload(ir_function_signature
*ir
, nir_function
*function
);
73 void add_instr(nir_instr
*instr
, unsigned num_components
);
74 nir_src
evaluate_rvalue(ir_rvalue
*ir
);
76 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
);
77 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
);
78 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
,
81 nir_src src2
, nir_src src3
);
86 nir_function_impl
*impl
;
87 exec_list
*cf_node_list
;
88 nir_instr
*result
; /* result of the expression tree last visited */
90 /* the head of the dereference chain we're creating */
91 nir_deref_var
*deref_head
;
92 /* the tail of the dereference chain we're creating */
93 nir_deref
*deref_tail
;
95 nir_variable
*var
; /* variable created by ir_variable visitor */
97 /* whether the IR we're operating on is per-function or global */
100 /* map of ir_variable -> nir_variable */
101 struct hash_table
*var_table
;
103 /* map of ir_function_signature -> nir_function_overload */
104 struct hash_table
*overload_table
;
108 * This visitor runs before the main visitor, calling create_function() for
109 * each function so that the main visitor can resolve forward references in
113 class nir_function_visitor
: public ir_hierarchical_visitor
116 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
119 virtual ir_visitor_status
visit_enter(ir_function
*);
122 nir_visitor
*visitor
;
125 }; /* end of anonymous namespace */
128 glsl_to_nir(exec_list
*ir
, _mesa_glsl_parse_state
*state
,
129 bool native_integers
)
131 nir_shader
*shader
= nir_shader_create(NULL
);
134 shader
->num_user_structures
= state
->num_user_structures
;
135 shader
->user_structures
= ralloc_array(shader
, glsl_type
*,
136 shader
->num_user_structures
);
137 memcpy(shader
->user_structures
, state
->user_structures
,
138 shader
->num_user_structures
* sizeof(glsl_type
*));
140 shader
->num_user_structures
= 0;
141 shader
->user_structures
= NULL
;
144 nir_visitor
v1(shader
, native_integers
);
145 nir_function_visitor
v2(&v1
);
147 visit_exec_list(ir
, &v1
);
152 nir_visitor::nir_visitor(nir_shader
*shader
, bool supports_ints
)
154 this->supports_ints
= supports_ints
;
155 this->shader
= shader
;
156 this->is_global
= true;
157 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
158 _mesa_key_pointer_equal
);
159 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
160 _mesa_key_pointer_equal
);
163 nir_visitor::~nir_visitor()
165 _mesa_hash_table_destroy(this->var_table
, NULL
);
166 _mesa_hash_table_destroy(this->overload_table
, NULL
);
169 static nir_constant
*
170 constant_copy(ir_constant
*ir
, void *mem_ctx
)
175 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
177 unsigned total_elems
= ir
->type
->components();
179 switch (ir
->type
->base_type
) {
181 for (i
= 0; i
< total_elems
; i
++)
182 ret
->value
.u
[i
] = ir
->value
.u
[i
];
186 for (i
= 0; i
< total_elems
; i
++)
187 ret
->value
.i
[i
] = ir
->value
.i
[i
];
190 case GLSL_TYPE_FLOAT
:
191 for (i
= 0; i
< total_elems
; i
++)
192 ret
->value
.f
[i
] = ir
->value
.f
[i
];
196 for (i
= 0; i
< total_elems
; i
++)
197 ret
->value
.b
[i
] = ir
->value
.b
[i
];
200 case GLSL_TYPE_STRUCT
:
201 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
204 foreach_in_list(ir_constant
, field
, &ir
->components
) {
205 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
210 case GLSL_TYPE_ARRAY
:
211 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
214 for (i
= 0; i
< ir
->type
->length
; i
++)
215 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
227 nir_visitor::visit(ir_variable
*ir
)
229 nir_variable
*var
= ralloc(shader
, nir_variable
);
230 var
->type
= ir
->type
;
231 var
->name
= ralloc_strdup(var
, ir
->name
);
233 if (ir
->is_interface_instance() && ir
->get_max_ifc_array_access() != NULL
) {
234 unsigned size
= ir
->get_interface_type()->length
;
235 var
->max_ifc_array_access
= ralloc_array(var
, unsigned, size
);
236 memcpy(var
->max_ifc_array_access
, ir
->get_max_ifc_array_access(),
237 size
* sizeof(unsigned));
239 var
->max_ifc_array_access
= NULL
;
242 var
->data
.read_only
= ir
->data
.read_only
;
243 var
->data
.centroid
= ir
->data
.centroid
;
244 var
->data
.sample
= ir
->data
.sample
;
245 var
->data
.invariant
= ir
->data
.invariant
;
247 switch(ir
->data
.mode
) {
249 case ir_var_temporary
:
251 var
->data
.mode
= nir_var_global
;
253 var
->data
.mode
= nir_var_local
;
256 case ir_var_function_in
:
257 case ir_var_function_out
:
258 case ir_var_function_inout
:
259 case ir_var_const_in
:
260 var
->data
.mode
= nir_var_local
;
263 case ir_var_shader_in
:
264 var
->data
.mode
= nir_var_shader_in
;
267 case ir_var_shader_out
:
268 var
->data
.mode
= nir_var_shader_out
;
272 var
->data
.mode
= nir_var_uniform
;
276 case ir_var_system_value
:
277 var
->data
.mode
= nir_var_system_value
;
285 var
->data
.interpolation
= ir
->data
.interpolation
;
286 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
287 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
288 var
->data
.explicit_location
= ir
->data
.explicit_location
;
289 var
->data
.explicit_index
= ir
->data
.explicit_index
;
290 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
291 var
->data
.has_initializer
= ir
->data
.has_initializer
;
292 var
->data
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
293 var
->data
.location_frac
= ir
->data
.location_frac
;
294 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
295 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
297 switch (ir
->data
.depth_layout
) {
298 case ir_depth_layout_none
:
299 var
->data
.depth_layout
= nir_depth_layout_none
;
301 case ir_depth_layout_any
:
302 var
->data
.depth_layout
= nir_depth_layout_any
;
304 case ir_depth_layout_greater
:
305 var
->data
.depth_layout
= nir_depth_layout_greater
;
307 case ir_depth_layout_less
:
308 var
->data
.depth_layout
= nir_depth_layout_less
;
310 case ir_depth_layout_unchanged
:
311 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
318 var
->data
.location
= ir
->data
.location
;
319 var
->data
.index
= ir
->data
.index
;
320 var
->data
.binding
= ir
->data
.binding
;
321 /* XXX Get rid of buffer_index */
322 var
->data
.atomic
.buffer_index
= ir
->data
.binding
;
323 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
324 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
325 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
326 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
327 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
328 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
329 var
->data
.image
.format
= ir
->data
.image_format
;
330 var
->data
.max_array_access
= ir
->data
.max_array_access
;
332 var
->num_state_slots
= ir
->get_num_state_slots();
333 var
->state_slots
= ralloc_array(var
, nir_state_slot
, var
->num_state_slots
);
334 ir_state_slot
*state_slots
= ir
->get_state_slots();
335 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
336 for (unsigned j
= 0; j
< 5; j
++)
337 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
338 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
341 var
->constant_value
= constant_copy(ir
->constant_value
, var
);
342 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
344 var
->interface_type
= ir
->get_interface_type();
346 switch (var
->data
.mode
) {
348 exec_list_push_tail(&impl
->locals
, &var
->node
);
352 exec_list_push_tail(&shader
->globals
, &var
->node
);
355 case nir_var_shader_in
:
356 _mesa_hash_table_insert(shader
->inputs
, var
->name
, var
);
359 case nir_var_shader_out
:
360 _mesa_hash_table_insert(shader
->outputs
, var
->name
, var
);
363 case nir_var_uniform
:
364 _mesa_hash_table_insert(shader
->uniforms
, var
->name
, var
);
367 case nir_var_system_value
:
368 exec_list_push_tail(&shader
->system_values
, &var
->node
);
376 _mesa_hash_table_insert(var_table
, ir
, var
);
381 nir_function_visitor::visit_enter(ir_function
*ir
)
383 visitor
->create_function(ir
);
384 return visit_continue_with_parent
;
389 nir_visitor::create_function(ir_function
*ir
)
391 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
392 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
393 create_overload(sig
, func
);
400 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
402 if (ir
->is_intrinsic
)
405 nir_function_overload
*overload
= nir_function_overload_create(function
);
407 unsigned num_params
= ir
->parameters
.length();
408 overload
->num_params
= num_params
;
409 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
412 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
413 switch (param
->data
.mode
) {
414 case ir_var_function_in
:
415 overload
->params
[i
].param_type
= nir_parameter_in
;
418 case ir_var_function_out
:
419 overload
->params
[i
].param_type
= nir_parameter_out
;
422 case ir_var_function_inout
:
423 overload
->params
[i
].param_type
= nir_parameter_inout
;
431 overload
->params
[i
].type
= param
->type
;
435 overload
->return_type
= ir
->return_type
;
437 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
441 nir_visitor::visit(ir_function
*ir
)
443 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
448 nir_visitor::visit(ir_function_signature
*ir
)
450 if (ir
->is_intrinsic
)
453 struct hash_entry
*entry
=
454 _mesa_hash_table_search(this->overload_table
, ir
);
457 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
459 if (ir
->is_defined
) {
460 nir_function_impl
*impl
= nir_function_impl_create(overload
);
463 unsigned num_params
= overload
->num_params
;
464 impl
->num_params
= num_params
;
465 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
467 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
469 impl
->params
[i
] = this->var
;
473 if (overload
->return_type
== glsl_type::void_type
) {
474 impl
->return_var
= NULL
;
476 impl
->return_var
= ralloc(this->shader
, nir_variable
);
477 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
479 impl
->return_var
->type
= overload
->return_type
;
482 this->is_global
= false;
484 this->cf_node_list
= &impl
->body
;
485 visit_exec_list(&ir
->body
, this);
487 this->is_global
= true;
489 overload
->impl
= NULL
;
494 nir_visitor::visit(ir_loop
*ir
)
496 exec_list
*old_list
= this->cf_node_list
;
498 nir_loop
*loop
= nir_loop_create(this->shader
);
499 nir_cf_node_insert_end(old_list
, &loop
->cf_node
);
500 this->cf_node_list
= &loop
->body
;
501 visit_exec_list(&ir
->body_instructions
, this);
503 this->cf_node_list
= old_list
;
507 nir_visitor::visit(ir_if
*ir
)
509 nir_src condition
= evaluate_rvalue(ir
->condition
);
511 exec_list
*old_list
= this->cf_node_list
;
513 nir_if
*if_stmt
= nir_if_create(this->shader
);
514 if_stmt
->condition
= condition
;
515 nir_cf_node_insert_end(old_list
, &if_stmt
->cf_node
);
517 this->cf_node_list
= &if_stmt
->then_list
;
518 visit_exec_list(&ir
->then_instructions
, this);
520 this->cf_node_list
= &if_stmt
->else_list
;
521 visit_exec_list(&ir
->else_instructions
, this);
523 this->cf_node_list
= old_list
;
527 nir_visitor::visit(ir_discard
*ir
)
530 * discards aren't treated as control flow, because before we lower them
531 * they can appear anywhere in the shader and the stuff after them may still
532 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
533 * discards will be immediately followed by a return.
536 nir_intrinsic_instr
*discard
=
537 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
538 nir_instr_insert_after_cf_list(this->cf_node_list
, &discard
->instr
);
542 nir_visitor::visit(ir_emit_vertex
*ir
)
544 nir_intrinsic_instr
*instr
=
545 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
546 instr
->const_index
[0] = ir
->stream_id();
547 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
551 nir_visitor::visit(ir_end_primitive
*ir
)
553 nir_intrinsic_instr
*instr
=
554 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
555 instr
->const_index
[0] = ir
->stream_id();
556 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
560 nir_visitor::visit(ir_loop_jump
*ir
)
564 case ir_loop_jump::jump_break
:
565 type
= nir_jump_break
;
567 case ir_loop_jump::jump_continue
:
568 type
= nir_jump_continue
;
575 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
576 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
580 nir_visitor::visit(ir_return
*ir
)
582 if (ir
->value
!= NULL
) {
583 ir
->value
->accept(this);
584 nir_intrinsic_instr
*copy
=
585 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
587 copy
->variables
[0] = nir_deref_var_create(this->shader
,
588 this->impl
->return_var
);
589 copy
->variables
[1] = this->deref_head
;
592 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
593 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
597 nir_visitor::visit(ir_call
*ir
)
599 if (ir
->callee
->is_intrinsic
) {
601 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
602 op
= nir_intrinsic_atomic_counter_read_var
;
603 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
604 op
= nir_intrinsic_atomic_counter_inc_var
;
605 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
606 op
= nir_intrinsic_atomic_counter_dec_var
;
611 nir_register
*reg
= nir_local_reg_create(impl
);
612 reg
->num_components
= 1;
614 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
615 ir_dereference
*param
=
616 (ir_dereference
*) ir
->actual_parameters
.get_head();
618 instr
->variables
[0] = this->deref_head
;
619 instr
->dest
.reg
.reg
= reg
;
621 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
623 nir_intrinsic_instr
*store_instr
=
624 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var_vec1
);
626 ir
->return_deref
->accept(this);
627 store_instr
->variables
[0] = this->deref_head
;
628 store_instr
->src
[0].reg
.reg
= reg
;
630 nir_instr_insert_after_cf_list(this->cf_node_list
, &store_instr
->instr
);
635 struct hash_entry
*entry
=
636 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
638 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
640 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
643 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
645 instr
->params
[i
] = this->deref_head
;
649 ir
->return_deref
->accept(this);
650 instr
->return_deref
= this->deref_head
;
651 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
655 nir_visitor::visit(ir_assignment
*ir
)
657 if (ir
->write_mask
!= (1 << ir
->lhs
->type
->vector_elements
) - 1 &&
658 ir
->write_mask
!= 0) {
660 * We have no good way to update only part of a variable, so just load
661 * the LHS into a register, do a writemasked move, and then store it
662 * back into the LHS. Copy propagation should get rid of the mess.
665 ir
->lhs
->accept(this);
666 nir_deref_var
*lhs_deref
= this->deref_head
;
667 nir_register
*reg
= nir_local_reg_create(this->impl
);
668 reg
->num_components
= ir
->lhs
->type
->vector_elements
;
671 switch (ir
->lhs
->type
->vector_elements
) {
672 case 1: op
= nir_intrinsic_load_var_vec1
; break;
673 case 2: op
= nir_intrinsic_load_var_vec2
; break;
674 case 3: op
= nir_intrinsic_load_var_vec3
; break;
675 case 4: op
= nir_intrinsic_load_var_vec4
; break;
676 default: assert(0); break;
679 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
680 load
->dest
.reg
.reg
= reg
;
681 load
->variables
[0] = lhs_deref
;
682 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
684 nir_alu_instr
*move
=
685 nir_alu_instr_create(this->shader
,
686 supports_ints
? nir_op_fmov
: nir_op_imov
);
687 move
->dest
.dest
.reg
.reg
= reg
;
688 move
->dest
.write_mask
= ir
->write_mask
;
689 move
->src
[0].src
= evaluate_rvalue(ir
->rhs
);
692 * GLSL IR will give us the input to the write-masked assignment in a
693 * single packed vector, whereas we expect each input component to be in
694 * the same channel as the writemask. So, for example, if the writemask
695 * is xzw, then we have to swizzle x -> x, y -> z, and z -> w.
698 unsigned component
= 0;
699 for (unsigned i
= 0; i
< 4; i
++) {
700 if ((ir
->write_mask
>> i
) & 1) {
701 move
->src
[0].swizzle
[i
] = component
++;
703 move
->src
[0].swizzle
[i
] = 0;
707 if (ir
->condition
!= NULL
) {
708 move
->has_predicate
= true;
709 move
->predicate
= evaluate_rvalue(ir
->condition
);
712 nir_instr_insert_after_cf_list(this->cf_node_list
, &move
->instr
);
714 switch (ir
->lhs
->type
->vector_elements
) {
715 case 1: op
= nir_intrinsic_store_var_vec1
; break;
716 case 2: op
= nir_intrinsic_store_var_vec2
; break;
717 case 3: op
= nir_intrinsic_store_var_vec3
; break;
718 case 4: op
= nir_intrinsic_store_var_vec4
; break;
719 default: assert(0); break;
722 nir_intrinsic_instr
*store
= nir_intrinsic_instr_create(this->shader
, op
);
723 nir_deref
*store_deref
= nir_copy_deref(this->shader
, &lhs_deref
->deref
);
724 store
->variables
[0] = nir_deref_as_var(store_deref
);
725 store
->src
[0].reg
.reg
= reg
;
726 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
730 if (ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) {
731 /* we're copying structs or arrays, so emit a copy_var */
732 nir_intrinsic_instr
*copy
=
733 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
735 ir
->lhs
->accept(this);
736 copy
->variables
[0] = this->deref_head
;
738 ir
->rhs
->accept(this);
739 copy
->variables
[1] = this->deref_head
;
741 if (ir
->condition
!= NULL
) {
742 copy
->has_predicate
= true;
743 copy
->predicate
= evaluate_rvalue(ir
->condition
);
745 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
749 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
752 switch (ir
->lhs
->type
->vector_elements
) {
753 case 1: op
= nir_intrinsic_store_var_vec1
; break;
754 case 2: op
= nir_intrinsic_store_var_vec2
; break;
755 case 3: op
= nir_intrinsic_store_var_vec3
; break;
756 case 4: op
= nir_intrinsic_store_var_vec4
; break;
757 default: assert(0); break;
760 nir_intrinsic_instr
*store
= nir_intrinsic_instr_create(this->shader
, op
);
762 ir
->lhs
->accept(this);
763 store
->variables
[0] = this->deref_head
;
764 store
->src
[0] = evaluate_rvalue(ir
->rhs
);
766 if (ir
->condition
!= NULL
) {
767 store
->has_predicate
= true;
768 store
->predicate
= evaluate_rvalue(ir
->condition
);
771 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
775 * Given an instruction, returns a pointer to its destination or NULL if there
778 * Note that this only handles instructions we generate at this level.
781 get_instr_dest(nir_instr
*instr
)
783 nir_alu_instr
*alu_instr
;
784 nir_intrinsic_instr
*intrinsic_instr
;
785 nir_tex_instr
*tex_instr
;
786 nir_load_const_instr
*load_const_instr
;
788 switch (instr
->type
) {
789 case nir_instr_type_alu
:
790 alu_instr
= nir_instr_as_alu(instr
);
791 return &alu_instr
->dest
.dest
;
793 case nir_instr_type_intrinsic
:
794 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
795 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
796 return &intrinsic_instr
->dest
;
800 case nir_instr_type_texture
:
801 tex_instr
= nir_instr_as_texture(instr
);
802 return &tex_instr
->dest
;
804 case nir_instr_type_load_const
:
805 load_const_instr
= nir_instr_as_load_const(instr
);
806 return &load_const_instr
->dest
;
817 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
819 nir_dest
*dest
= get_instr_dest(instr
);
821 dest
->reg
.reg
= nir_local_reg_create(this->impl
);
822 dest
->reg
.reg
->num_components
= num_components
;
824 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
825 this->result
= instr
;
829 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
832 if (ir
->as_dereference() || ir
->as_constant()) {
834 * A dereference is being used on the right hand side, which means we
835 * must emit a variable load.
839 switch (ir
->type
->vector_elements
) {
841 op
= nir_intrinsic_load_var_vec1
;
844 op
= nir_intrinsic_load_var_vec2
;
847 op
= nir_intrinsic_load_var_vec3
;
850 op
= nir_intrinsic_load_var_vec4
;
854 nir_intrinsic_instr
*load_instr
=
855 nir_intrinsic_instr_create(this->shader
, op
);
856 load_instr
->variables
[0] = this->deref_head
;
857 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
861 * instr doesn't have a destination right now, give it one and then set up
862 * the source so that it points to it.
864 * TODO: once we support SSA plumb through a use_ssa boolean and use SSA
865 * here instead of creating a register.
867 nir_dest
*dest
= get_instr_dest(this->result
);
868 assert(dest
->reg
.reg
);
872 src
.reg
.base_offset
= 0;
873 src
.reg
.indirect
= NULL
;
874 src
.reg
.reg
= dest
->reg
.reg
;
880 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
882 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
883 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
884 instr
->src
[i
].src
= srcs
[i
];
885 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
886 add_instr(&instr
->instr
, dest_size
);
891 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
893 assert(nir_op_infos
[op
].num_inputs
== 1);
894 return emit(op
, dest_size
, &src1
);
898 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
901 assert(nir_op_infos
[op
].num_inputs
== 2);
902 nir_src srcs
[] = { src1
, src2
};
903 return emit(op
, dest_size
, srcs
);
907 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
908 nir_src src2
, nir_src src3
)
910 assert(nir_op_infos
[op
].num_inputs
== 3);
911 nir_src srcs
[] = { src1
, src2
, src3
};
912 return emit(op
, dest_size
, srcs
);
916 nir_visitor::visit(ir_expression
*ir
)
918 if (ir
->operation
== ir_binop_ubo_load
) {
919 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
923 switch (ir
->type
->vector_elements
) {
924 case 1: op
= nir_intrinsic_load_ubo_vec1
; break;
925 case 2: op
= nir_intrinsic_load_ubo_vec2
; break;
926 case 3: op
= nir_intrinsic_load_ubo_vec3
; break;
927 case 4: op
= nir_intrinsic_load_ubo_vec4
; break;
928 default: assert(0); break;
931 switch (ir
->type
->vector_elements
) {
932 case 1: op
= nir_intrinsic_load_ubo_vec1_indirect
; break;
933 case 2: op
= nir_intrinsic_load_ubo_vec2_indirect
; break;
934 case 3: op
= nir_intrinsic_load_ubo_vec3_indirect
; break;
935 case 4: op
= nir_intrinsic_load_ubo_vec4_indirect
; break;
936 default: assert(0); break;
939 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
940 load
->const_index
[0] = ir
->operands
[0]->as_constant()->value
.u
[0];
941 load
->const_index
[1] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
942 load
->const_index
[2] = 1; /* number of vec4's */
944 load
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
945 add_instr(&load
->instr
, ir
->type
->vector_elements
);
948 * In UBO's, a true boolean value is any non-zero value, but we consider
949 * a true boolean to be ~0. Fix this up with a != 0 comparison.
952 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
953 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
);
954 const_zero
->num_components
= 1;
955 const_zero
->value
.u
[0] = 0;
956 const_zero
->dest
.reg
.reg
= nir_local_reg_create(this->impl
);
957 const_zero
->dest
.reg
.reg
->num_components
= 1;
958 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
960 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
961 compare
->src
[0].src
.reg
.reg
= load
->dest
.reg
.reg
;
962 compare
->src
[1].src
.reg
.reg
= const_zero
->dest
.reg
.reg
;
963 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
964 compare
->src
[1].swizzle
[i
] = 0;
965 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
967 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
974 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
975 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
977 glsl_base_type types
[4];
978 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
980 types
[i
] = ir
->operands
[i
]->type
->base_type
;
982 types
[i
] = GLSL_TYPE_FLOAT
;
984 glsl_base_type out_type
;
986 out_type
= ir
->type
->base_type
;
988 out_type
= GLSL_TYPE_FLOAT
;
990 unsigned dest_size
= ir
->type
->vector_elements
;
992 nir_alu_instr
*instr
;
995 switch (ir
->operation
) {
996 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
997 case ir_unop_logic_not
:
998 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1001 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fmov
: nir_op_imov
,
1003 instr
->src
[0].negate
= true;
1006 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fmov
: nir_op_imov
,
1008 instr
->src
[0].abs
= true;
1011 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1014 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1015 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1016 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1017 case ir_unop_exp
: emit(nir_op_fexp
, dest_size
, srcs
); break;
1018 case ir_unop_log
: emit(nir_op_flog
, dest_size
, srcs
); break;
1019 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1020 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1022 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1025 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1028 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1030 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1031 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1032 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1033 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1034 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1037 case ir_unop_bitcast_i2f
:
1038 case ir_unop_bitcast_f2i
:
1039 case ir_unop_bitcast_u2f
:
1040 case ir_unop_bitcast_f2u
:
1042 emit(nir_op_imov
, dest_size
, srcs
);
1045 switch (ir
->operands
[0]->type
->vector_elements
) {
1047 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1051 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1055 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1063 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1064 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1065 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1066 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1067 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1068 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1069 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1070 case ir_unop_sin_reduced
:
1071 emit(nir_op_fsin_reduced
, dest_size
, srcs
);
1073 case ir_unop_cos_reduced
:
1074 emit(nir_op_fcos_reduced
, dest_size
, srcs
);
1076 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1077 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1078 case ir_unop_pack_snorm_2x16
:
1079 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1081 case ir_unop_pack_snorm_4x8
:
1082 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1084 case ir_unop_pack_unorm_2x16
:
1085 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1087 case ir_unop_pack_unorm_4x8
:
1088 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1090 case ir_unop_pack_half_2x16
:
1091 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1093 case ir_unop_unpack_snorm_2x16
:
1094 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1096 case ir_unop_unpack_snorm_4x8
:
1097 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1099 case ir_unop_unpack_unorm_2x16
:
1100 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1102 case ir_unop_unpack_unorm_4x8
:
1103 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1105 case ir_unop_unpack_half_2x16
:
1106 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1108 case ir_unop_unpack_half_2x16_split_x
:
1109 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1111 case ir_unop_unpack_half_2x16_split_y
:
1112 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1114 case ir_unop_bitfield_reverse
:
1115 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1117 case ir_unop_bit_count
: emit(nir_op_bit_count
, dest_size
, srcs
); break;
1118 case ir_unop_find_msb
: emit(nir_op_find_msb
, dest_size
, srcs
); break;
1119 case ir_unop_find_lsb
: emit(nir_op_find_lsb
, dest_size
, srcs
); break;
1121 switch (ir
->type
->vector_elements
) {
1123 switch (ir
->operands
[0]->type
->vector_elements
) {
1124 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1125 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1126 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1127 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1128 default: assert(0); break;
1132 switch (ir
->operands
[0]->type
->vector_elements
) {
1133 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1134 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1135 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1136 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1137 default: assert(0); break;
1141 switch (ir
->operands
[0]->type
->vector_elements
) {
1142 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1143 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1144 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1145 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1146 default: assert(0); break;
1150 switch (ir
->operands
[0]->type
->vector_elements
) {
1151 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1152 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1153 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1154 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1155 default: assert(0); break;
1171 case ir_binop_bit_and
:
1172 case ir_binop_bit_or
:
1173 case ir_binop_bit_xor
:
1174 case ir_binop_lshift
:
1175 case ir_binop_rshift
:
1176 switch (ir
->operation
) {
1178 if (out_type
== GLSL_TYPE_FLOAT
)
1184 if (out_type
== GLSL_TYPE_FLOAT
)
1190 if (out_type
== GLSL_TYPE_FLOAT
)
1196 if (out_type
== GLSL_TYPE_FLOAT
)
1198 else if (out_type
== GLSL_TYPE_INT
)
1204 if (out_type
== GLSL_TYPE_FLOAT
)
1210 if (out_type
== GLSL_TYPE_FLOAT
)
1212 else if (out_type
== GLSL_TYPE_INT
)
1218 if (out_type
== GLSL_TYPE_FLOAT
)
1220 else if (out_type
== GLSL_TYPE_INT
)
1225 case ir_binop_bit_and
:
1228 case ir_binop_bit_or
:
1231 case ir_binop_bit_xor
:
1234 case ir_binop_lshift
:
1237 case ir_binop_rshift
:
1238 if (out_type
== GLSL_TYPE_INT
)
1252 instr
= emit(op
, dest_size
, srcs
);
1254 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1255 ir
->operands
[1]->type
->vector_elements
== 1) {
1256 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1258 instr
->src
[1].swizzle
[i
] = 0;
1262 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1263 ir
->operands
[0]->type
->vector_elements
== 1) {
1264 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1266 instr
->src
[0].swizzle
[i
] = 0;
1271 case ir_binop_imul_high
:
1272 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1275 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1276 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1278 if (supports_ints
) {
1279 if (types
[0] == GLSL_TYPE_FLOAT
)
1280 emit(nir_op_flt
, dest_size
, srcs
);
1281 else if (types
[0] == GLSL_TYPE_INT
)
1282 emit(nir_op_ilt
, dest_size
, srcs
);
1284 emit(nir_op_ult
, dest_size
, srcs
);
1286 emit(nir_op_slt
, dest_size
, srcs
);
1289 case ir_binop_greater
:
1290 if (supports_ints
) {
1291 if (types
[0] == GLSL_TYPE_FLOAT
)
1292 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1293 else if (types
[0] == GLSL_TYPE_INT
)
1294 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1296 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1298 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1301 case ir_binop_lequal
:
1302 if (supports_ints
) {
1303 if (types
[0] == GLSL_TYPE_FLOAT
)
1304 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1305 else if (types
[0] == GLSL_TYPE_INT
)
1306 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1308 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1310 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1313 case ir_binop_gequal
:
1314 if (supports_ints
) {
1315 if (types
[0] == GLSL_TYPE_FLOAT
)
1316 emit(nir_op_fge
, dest_size
, srcs
);
1317 else if (types
[0] == GLSL_TYPE_INT
)
1318 emit(nir_op_ige
, dest_size
, srcs
);
1320 emit(nir_op_uge
, dest_size
, srcs
);
1322 emit(nir_op_slt
, dest_size
, srcs
);
1325 case ir_binop_equal
:
1326 if (supports_ints
) {
1327 if (types
[0] == GLSL_TYPE_FLOAT
)
1328 emit(nir_op_feq
, dest_size
, srcs
);
1330 emit(nir_op_ieq
, dest_size
, srcs
);
1332 emit(nir_op_seq
, dest_size
, srcs
);
1335 case ir_binop_nequal
:
1336 if (supports_ints
) {
1337 if (types
[0] == GLSL_TYPE_FLOAT
)
1338 emit(nir_op_fne
, dest_size
, srcs
);
1340 emit(nir_op_ine
, dest_size
, srcs
);
1342 emit(nir_op_sne
, dest_size
, srcs
);
1345 case ir_binop_all_equal
:
1346 if (supports_ints
) {
1347 if (types
[0] == GLSL_TYPE_FLOAT
) {
1348 switch (ir
->operands
[0]->type
->vector_elements
) {
1349 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1350 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1351 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1352 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1358 switch (ir
->operands
[0]->type
->vector_elements
) {
1359 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1360 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1361 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1362 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1369 switch (ir
->operands
[0]->type
->vector_elements
) {
1370 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1371 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1372 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1373 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1380 case ir_binop_any_nequal
:
1381 if (supports_ints
) {
1382 if (types
[0] == GLSL_TYPE_FLOAT
) {
1383 switch (ir
->operands
[0]->type
->vector_elements
) {
1384 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1385 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1386 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1387 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1393 switch (ir
->operands
[0]->type
->vector_elements
) {
1394 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1395 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1396 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1397 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1404 switch (ir
->operands
[0]->type
->vector_elements
) {
1405 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1406 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1407 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1408 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1415 case ir_binop_logic_and
:
1417 emit(nir_op_iand
, dest_size
, srcs
);
1419 emit(nir_op_fand
, dest_size
, srcs
);
1421 case ir_binop_logic_or
:
1423 emit(nir_op_ior
, dest_size
, srcs
);
1425 emit(nir_op_for
, dest_size
, srcs
);
1427 case ir_binop_logic_xor
:
1429 emit(nir_op_ixor
, dest_size
, srcs
);
1431 emit(nir_op_fxor
, dest_size
, srcs
);
1434 switch (ir
->operands
[0]->type
->vector_elements
) {
1435 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1436 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1437 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1444 case ir_binop_pack_half_2x16_split
:
1445 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1447 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1448 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1449 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1451 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1452 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1453 ir
->operands
[2]->type
->vector_elements
== 1) {
1454 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1456 instr
->src
[2].swizzle
[i
] = 0;
1462 emit(nir_op_bcsel
, dest_size
, srcs
);
1464 emit(nir_op_fcsel
, dest_size
, srcs
);
1467 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1468 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1469 instr
->src
[0].swizzle
[i
] = 0;
1472 case ir_triop_bitfield_extract
:
1473 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1474 nir_op_ubitfield_extract
, dest_size
, srcs
);
1475 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1476 instr
->src
[1].swizzle
[i
] = 0;
1477 instr
->src
[2].swizzle
[i
] = 0;
1480 case ir_quadop_bitfield_insert
:
1481 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1482 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1483 instr
->src
[2].swizzle
[i
] = 0;
1484 instr
->src
[3].swizzle
[i
] = 0;
1487 case ir_quadop_vector
:
1488 switch (ir
->type
->vector_elements
) {
1489 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1490 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1491 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1492 default: assert(0); break;
1503 nir_visitor::visit(ir_swizzle
*ir
)
1505 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1506 ir
->type
->vector_elements
,
1507 evaluate_rvalue(ir
->val
));
1509 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1510 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1511 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1515 nir_visitor::visit(ir_texture
*ir
)
1522 num_srcs
= 1; /* coordinate */
1527 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1528 num_srcs
= 2; /* coordinate, bias/lod */
1532 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1538 if (ir
->lod_info
.lod
!= NULL
)
1539 num_srcs
= 2; /* coordinate, lod */
1541 num_srcs
= 1; /* coordinate */
1545 op
= nir_texop_txf_ms
;
1546 num_srcs
= 2; /* coordinate, sample_index */
1551 if (ir
->lod_info
.lod
!= NULL
)
1552 num_srcs
= 1; /* lod */
1559 num_srcs
= 1; /* coordinate */
1564 num_srcs
= 1; /* coordinate */
1567 case ir_query_levels
:
1568 op
= nir_texop_query_levels
;
1577 if (ir
->projector
!= NULL
)
1579 if (ir
->shadow_comparitor
!= NULL
)
1581 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1584 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1587 instr
->sampler_dim
=
1588 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1589 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1590 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1591 if (instr
->is_shadow
)
1592 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1593 switch (ir
->type
->base_type
) {
1594 case GLSL_TYPE_FLOAT
:
1595 instr
->dest_type
= nir_type_float
;
1598 instr
->dest_type
= nir_type_int
;
1600 case GLSL_TYPE_UINT
:
1601 instr
->dest_type
= nir_type_unsigned
;
1607 ir
->sampler
->accept(this);
1608 instr
->sampler
= this->deref_head
;
1610 unsigned src_number
= 0;
1612 if (ir
->coordinate
!= NULL
) {
1613 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1614 instr
->src
[src_number
] = evaluate_rvalue(ir
->coordinate
);
1615 instr
->src_type
[src_number
] = nir_tex_src_coord
;
1619 if (ir
->projector
!= NULL
) {
1620 instr
->src
[src_number
] = evaluate_rvalue(ir
->projector
);
1621 instr
->src_type
[src_number
] = nir_tex_src_projector
;
1625 if (ir
->shadow_comparitor
!= NULL
) {
1626 instr
->src
[src_number
] = evaluate_rvalue(ir
->shadow_comparitor
);
1627 instr
->src_type
[src_number
] = nir_tex_src_comparitor
;
1631 if (ir
->offset
!= NULL
) {
1632 /* we don't support multiple offsets yet */
1633 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1635 ir_constant
*const_offset
= ir
->offset
->as_constant();
1636 if (const_offset
!= NULL
) {
1637 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1638 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1640 instr
->src
[src_number
] = evaluate_rvalue(ir
->offset
);
1641 instr
->src_type
[src_number
] = nir_tex_src_offset
;
1648 instr
->src
[src_number
] = evaluate_rvalue(ir
->lod_info
.bias
);
1649 instr
->src_type
[src_number
] = nir_tex_src_bias
;
1656 if (ir
->lod_info
.lod
!= NULL
) {
1657 instr
->src
[src_number
] = evaluate_rvalue(ir
->lod_info
.lod
);
1658 instr
->src_type
[src_number
] = nir_tex_src_lod
;
1664 instr
->src
[src_number
] = evaluate_rvalue(ir
->lod_info
.grad
.dPdx
);
1665 instr
->src_type
[src_number
] = nir_tex_src_ddx
;
1667 instr
->src
[src_number
] = evaluate_rvalue(ir
->lod_info
.grad
.dPdy
);
1668 instr
->src_type
[src_number
] = nir_tex_src_ddy
;
1673 instr
->src
[src_number
] = evaluate_rvalue(ir
->lod_info
.sample_index
);
1674 instr
->src_type
[src_number
] = nir_tex_src_ms_index
;
1679 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1686 assert(src_number
== num_srcs
);
1688 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
1692 nir_visitor::visit(ir_constant
*ir
)
1695 * We don't know if this variable is an an array or struct that gets
1696 * dereferenced, so do the safe thing an make it a variable and return a
1700 nir_variable
*var
= ralloc(this->shader
, nir_variable
);
1701 var
->name
= ralloc_strdup(var
, "const_temp");
1702 var
->type
= ir
->type
;
1703 var
->data
.mode
= nir_var_local
;
1704 var
->data
.read_only
= true;
1705 var
->constant_value
= constant_copy(ir
, var
);
1706 var
->constant_initializer
= constant_copy(ir
, var
);
1707 exec_list_push_tail(&this->impl
->locals
, &var
->node
);
1709 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1710 this->deref_tail
= &this->deref_head
->deref
;
1714 nir_visitor::visit(ir_dereference_variable
*ir
)
1716 struct hash_entry
*entry
=
1717 _mesa_hash_table_search(this->var_table
, ir
->var
);
1719 nir_variable
*var
= (nir_variable
*) entry
->data
;
1721 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
1722 this->deref_head
= deref
;
1723 this->deref_tail
= &deref
->deref
;
1727 nir_visitor::visit(ir_dereference_record
*ir
)
1729 ir
->record
->accept(this);
1731 nir_deref_struct
*deref
= nir_deref_struct_create(this->shader
, ir
->field
);
1732 deref
->deref
.type
= ir
->type
;
1733 this->deref_tail
->child
= &deref
->deref
;
1734 this->deref_tail
= &deref
->deref
;
1738 nir_visitor::visit(ir_dereference_array
*ir
)
1740 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
1741 deref
->deref
.type
= ir
->type
;
1743 ir_constant
*const_index
= ir
->array_index
->as_constant();
1744 if (const_index
!= NULL
) {
1745 deref
->base_offset
= const_index
->value
.u
[0];
1747 deref
->has_indirect
= true;
1748 deref
->indirect
= evaluate_rvalue(ir
->array_index
);
1751 ir
->array
->accept(this);
1753 this->deref_tail
->child
= &deref
->deref
;
1754 this->deref_tail
= &deref
->deref
;