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
, bool native_integers
,
129 const nir_shader_compiler_options
*options
)
131 nir_shader
*shader
= nir_shader_create(NULL
, options
);
133 nir_visitor
v1(shader
, native_integers
);
134 nir_function_visitor
v2(&v1
);
136 visit_exec_list(ir
, &v1
);
141 nir_visitor::nir_visitor(nir_shader
*shader
, bool supports_ints
)
143 this->supports_ints
= supports_ints
;
144 this->shader
= shader
;
145 this->is_global
= true;
146 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
147 _mesa_key_pointer_equal
);
148 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
149 _mesa_key_pointer_equal
);
152 nir_visitor::~nir_visitor()
154 _mesa_hash_table_destroy(this->var_table
, NULL
);
155 _mesa_hash_table_destroy(this->overload_table
, NULL
);
158 static nir_constant
*
159 constant_copy(ir_constant
*ir
, void *mem_ctx
)
164 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
166 unsigned total_elems
= ir
->type
->components();
168 switch (ir
->type
->base_type
) {
170 for (i
= 0; i
< total_elems
; i
++)
171 ret
->value
.u
[i
] = ir
->value
.u
[i
];
175 for (i
= 0; i
< total_elems
; i
++)
176 ret
->value
.i
[i
] = ir
->value
.i
[i
];
179 case GLSL_TYPE_FLOAT
:
180 for (i
= 0; i
< total_elems
; i
++)
181 ret
->value
.f
[i
] = ir
->value
.f
[i
];
185 for (i
= 0; i
< total_elems
; i
++)
186 ret
->value
.b
[i
] = ir
->value
.b
[i
];
189 case GLSL_TYPE_STRUCT
:
190 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
193 foreach_in_list(ir_constant
, field
, &ir
->components
) {
194 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
199 case GLSL_TYPE_ARRAY
:
200 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
203 for (i
= 0; i
< ir
->type
->length
; i
++)
204 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
208 unreachable("not reached");
215 nir_visitor::visit(ir_variable
*ir
)
217 nir_variable
*var
= ralloc(shader
, nir_variable
);
218 var
->type
= ir
->type
;
219 var
->name
= ralloc_strdup(var
, ir
->name
);
221 if (ir
->is_interface_instance() && ir
->get_max_ifc_array_access() != NULL
) {
222 unsigned size
= ir
->get_interface_type()->length
;
223 var
->max_ifc_array_access
= ralloc_array(var
, unsigned, size
);
224 memcpy(var
->max_ifc_array_access
, ir
->get_max_ifc_array_access(),
225 size
* sizeof(unsigned));
227 var
->max_ifc_array_access
= NULL
;
230 var
->data
.read_only
= ir
->data
.read_only
;
231 var
->data
.centroid
= ir
->data
.centroid
;
232 var
->data
.sample
= ir
->data
.sample
;
233 var
->data
.invariant
= ir
->data
.invariant
;
234 var
->data
.location
= ir
->data
.location
;
236 switch(ir
->data
.mode
) {
238 case ir_var_temporary
:
240 var
->data
.mode
= nir_var_global
;
242 var
->data
.mode
= nir_var_local
;
245 case ir_var_function_in
:
246 case ir_var_function_out
:
247 case ir_var_function_inout
:
248 case ir_var_const_in
:
249 var
->data
.mode
= nir_var_local
;
252 case ir_var_shader_in
:
253 if (ir
->data
.location
== VARYING_SLOT_FACE
) {
254 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
255 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
256 var
->data
.mode
= nir_var_system_value
;
258 var
->data
.mode
= nir_var_shader_in
;
262 case ir_var_shader_out
:
263 var
->data
.mode
= nir_var_shader_out
;
267 var
->data
.mode
= nir_var_uniform
;
271 case ir_var_system_value
:
272 var
->data
.mode
= nir_var_system_value
;
276 unreachable("not reached");
279 var
->data
.interpolation
= ir
->data
.interpolation
;
280 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
281 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
282 var
->data
.explicit_location
= ir
->data
.explicit_location
;
283 var
->data
.explicit_index
= ir
->data
.explicit_index
;
284 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
285 var
->data
.has_initializer
= ir
->data
.has_initializer
;
286 var
->data
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
287 var
->data
.location_frac
= ir
->data
.location_frac
;
288 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
289 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
291 switch (ir
->data
.depth_layout
) {
292 case ir_depth_layout_none
:
293 var
->data
.depth_layout
= nir_depth_layout_none
;
295 case ir_depth_layout_any
:
296 var
->data
.depth_layout
= nir_depth_layout_any
;
298 case ir_depth_layout_greater
:
299 var
->data
.depth_layout
= nir_depth_layout_greater
;
301 case ir_depth_layout_less
:
302 var
->data
.depth_layout
= nir_depth_layout_less
;
304 case ir_depth_layout_unchanged
:
305 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
308 unreachable("not reached");
311 var
->data
.index
= ir
->data
.index
;
312 var
->data
.binding
= ir
->data
.binding
;
313 /* XXX Get rid of buffer_index */
314 var
->data
.atomic
.buffer_index
= ir
->data
.binding
;
315 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
316 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
317 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
318 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
319 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
320 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
321 var
->data
.image
.format
= ir
->data
.image_format
;
322 var
->data
.max_array_access
= ir
->data
.max_array_access
;
324 var
->num_state_slots
= ir
->get_num_state_slots();
325 if (var
->num_state_slots
> 0) {
326 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
327 var
->num_state_slots
);
329 ir_state_slot
*state_slots
= ir
->get_state_slots();
330 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
331 for (unsigned j
= 0; j
< 5; j
++)
332 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
333 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
336 var
->state_slots
= NULL
;
339 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
341 var
->interface_type
= ir
->get_interface_type();
343 switch (var
->data
.mode
) {
345 exec_list_push_tail(&impl
->locals
, &var
->node
);
349 exec_list_push_tail(&shader
->globals
, &var
->node
);
352 case nir_var_shader_in
:
353 _mesa_hash_table_insert(shader
->inputs
, var
->name
, var
);
356 case nir_var_shader_out
:
357 _mesa_hash_table_insert(shader
->outputs
, var
->name
, var
);
360 case nir_var_uniform
:
361 _mesa_hash_table_insert(shader
->uniforms
, var
->name
, var
);
364 case nir_var_system_value
:
365 exec_list_push_tail(&shader
->system_values
, &var
->node
);
369 unreachable("not reached");
372 _mesa_hash_table_insert(var_table
, ir
, var
);
377 nir_function_visitor::visit_enter(ir_function
*ir
)
379 visitor
->create_function(ir
);
380 return visit_continue_with_parent
;
385 nir_visitor::create_function(ir_function
*ir
)
387 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
388 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
389 create_overload(sig
, func
);
396 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
398 if (ir
->is_intrinsic
)
401 nir_function_overload
*overload
= nir_function_overload_create(function
);
403 unsigned num_params
= ir
->parameters
.length();
404 overload
->num_params
= num_params
;
405 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
408 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
409 switch (param
->data
.mode
) {
410 case ir_var_function_in
:
411 overload
->params
[i
].param_type
= nir_parameter_in
;
414 case ir_var_function_out
:
415 overload
->params
[i
].param_type
= nir_parameter_out
;
418 case ir_var_function_inout
:
419 overload
->params
[i
].param_type
= nir_parameter_inout
;
423 unreachable("not reached");
426 overload
->params
[i
].type
= param
->type
;
430 overload
->return_type
= ir
->return_type
;
432 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
436 nir_visitor::visit(ir_function
*ir
)
438 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
443 nir_visitor::visit(ir_function_signature
*ir
)
445 if (ir
->is_intrinsic
)
448 struct hash_entry
*entry
=
449 _mesa_hash_table_search(this->overload_table
, ir
);
452 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
454 if (ir
->is_defined
) {
455 nir_function_impl
*impl
= nir_function_impl_create(overload
);
458 unsigned num_params
= overload
->num_params
;
459 impl
->num_params
= num_params
;
460 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
462 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
464 impl
->params
[i
] = this->var
;
468 if (overload
->return_type
== glsl_type::void_type
) {
469 impl
->return_var
= NULL
;
471 impl
->return_var
= ralloc(this->shader
, nir_variable
);
472 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
474 impl
->return_var
->type
= overload
->return_type
;
477 this->is_global
= false;
479 this->cf_node_list
= &impl
->body
;
480 visit_exec_list(&ir
->body
, this);
482 this->is_global
= true;
484 overload
->impl
= NULL
;
489 nir_visitor::visit(ir_loop
*ir
)
491 exec_list
*old_list
= this->cf_node_list
;
493 nir_loop
*loop
= nir_loop_create(this->shader
);
494 nir_cf_node_insert_end(old_list
, &loop
->cf_node
);
495 this->cf_node_list
= &loop
->body
;
496 visit_exec_list(&ir
->body_instructions
, this);
498 this->cf_node_list
= old_list
;
502 nir_visitor::visit(ir_if
*ir
)
504 nir_src condition
= evaluate_rvalue(ir
->condition
);
506 exec_list
*old_list
= this->cf_node_list
;
508 nir_if
*if_stmt
= nir_if_create(this->shader
);
509 if_stmt
->condition
= condition
;
510 nir_cf_node_insert_end(old_list
, &if_stmt
->cf_node
);
512 this->cf_node_list
= &if_stmt
->then_list
;
513 visit_exec_list(&ir
->then_instructions
, this);
515 this->cf_node_list
= &if_stmt
->else_list
;
516 visit_exec_list(&ir
->else_instructions
, this);
518 this->cf_node_list
= old_list
;
522 nir_visitor::visit(ir_discard
*ir
)
525 * discards aren't treated as control flow, because before we lower them
526 * they can appear anywhere in the shader and the stuff after them may still
527 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
528 * discards will be immediately followed by a return.
531 nir_intrinsic_instr
*discard
;
533 discard
= nir_intrinsic_instr_create(this->shader
,
534 nir_intrinsic_discard_if
);
535 discard
->src
[0] = evaluate_rvalue(ir
->condition
);
537 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
539 nir_instr_insert_after_cf_list(this->cf_node_list
, &discard
->instr
);
543 nir_visitor::visit(ir_emit_vertex
*ir
)
545 nir_intrinsic_instr
*instr
=
546 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
547 instr
->const_index
[0] = ir
->stream_id();
548 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
552 nir_visitor::visit(ir_end_primitive
*ir
)
554 nir_intrinsic_instr
*instr
=
555 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
556 instr
->const_index
[0] = ir
->stream_id();
557 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
561 nir_visitor::visit(ir_loop_jump
*ir
)
565 case ir_loop_jump::jump_break
:
566 type
= nir_jump_break
;
568 case ir_loop_jump::jump_continue
:
569 type
= nir_jump_continue
;
572 unreachable("not reached");
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
;
608 unreachable("not reached");
611 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
612 ir_dereference
*param
=
613 (ir_dereference
*) ir
->actual_parameters
.get_head();
615 instr
->variables
[0] = this->deref_head
;
616 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
618 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
620 nir_intrinsic_instr
*store_instr
=
621 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
622 store_instr
->num_components
= 1;
624 ir
->return_deref
->accept(this);
625 store_instr
->variables
[0] = this->deref_head
;
626 store_instr
->src
[0].is_ssa
= true;
627 store_instr
->src
[0].ssa
= &instr
->dest
.ssa
;
629 nir_instr_insert_after_cf_list(this->cf_node_list
, &store_instr
->instr
);
634 struct hash_entry
*entry
=
635 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
637 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
639 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
642 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
644 instr
->params
[i
] = this->deref_head
;
648 ir
->return_deref
->accept(this);
649 instr
->return_deref
= this->deref_head
;
650 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
654 nir_visitor::visit(ir_assignment
*ir
)
656 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
658 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
659 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
660 /* We're doing a plain-as-can-be copy, so emit a copy_var */
661 nir_intrinsic_instr
*copy
=
662 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
664 ir
->lhs
->accept(this);
665 copy
->variables
[0] = this->deref_head
;
667 ir
->rhs
->accept(this);
668 copy
->variables
[1] = this->deref_head
;
672 nir_if
*if_stmt
= nir_if_create(this->shader
);
673 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
674 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
675 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
677 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
682 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
684 ir
->lhs
->accept(this);
685 nir_deref_var
*lhs_deref
= this->deref_head
;
686 nir_src src
= evaluate_rvalue(ir
->rhs
);
688 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
690 * We have no good way to update only part of a variable, so just load
691 * the LHS and do a vec operation to combine the old with the new, and
693 * back into the LHS. Copy propagation should get rid of the mess.
696 nir_intrinsic_instr
*load
=
697 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
698 load
->num_components
= ir
->lhs
->type
->vector_elements
;
699 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
700 load
->variables
[0] = lhs_deref
;
701 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
704 switch (ir
->lhs
->type
->vector_elements
) {
705 case 1: vec_op
= nir_op_imov
; break;
706 case 2: vec_op
= nir_op_vec2
; break;
707 case 3: vec_op
= nir_op_vec3
; break;
708 case 4: vec_op
= nir_op_vec4
; break;
709 default: unreachable("Invalid number of components"); break;
711 nir_alu_instr
*vec
= nir_alu_instr_create(this->shader
, vec_op
);
712 nir_ssa_dest_init(&vec
->instr
, &vec
->dest
.dest
, num_components
, NULL
);
713 vec
->dest
.write_mask
= (1 << num_components
) - 1;
715 unsigned component
= 0;
716 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
717 if (ir
->write_mask
& (1 << i
)) {
718 vec
->src
[i
].src
= src
;
720 /* GLSL IR will give us the input to the write-masked assignment
721 * in a single packed vector. So, for example, if the
722 * writemask is xzw, then we have to swizzle x -> x, y -> z,
723 * and z -> w and get the y component from the load.
725 vec
->src
[i
].swizzle
[0] = component
++;
727 vec
->src
[i
].src
.is_ssa
= true;
728 vec
->src
[i
].src
.ssa
= &load
->dest
.ssa
;
729 vec
->src
[i
].swizzle
[0] = i
;
733 nir_instr_insert_after_cf_list(this->cf_node_list
, &vec
->instr
);
736 src
.ssa
= &vec
->dest
.dest
.ssa
;
739 nir_intrinsic_instr
*store
=
740 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
741 store
->num_components
= ir
->lhs
->type
->vector_elements
;
742 nir_deref
*store_deref
= nir_copy_deref(this->shader
, &lhs_deref
->deref
);
743 store
->variables
[0] = nir_deref_as_var(store_deref
);
747 nir_if
*if_stmt
= nir_if_create(this->shader
);
748 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
749 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
750 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
752 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
757 * Given an instruction, returns a pointer to its destination or NULL if there
760 * Note that this only handles instructions we generate at this level.
763 get_instr_dest(nir_instr
*instr
)
765 nir_alu_instr
*alu_instr
;
766 nir_intrinsic_instr
*intrinsic_instr
;
767 nir_tex_instr
*tex_instr
;
769 switch (instr
->type
) {
770 case nir_instr_type_alu
:
771 alu_instr
= nir_instr_as_alu(instr
);
772 return &alu_instr
->dest
.dest
;
774 case nir_instr_type_intrinsic
:
775 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
776 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
777 return &intrinsic_instr
->dest
;
781 case nir_instr_type_tex
:
782 tex_instr
= nir_instr_as_tex(instr
);
783 return &tex_instr
->dest
;
786 unreachable("not reached");
793 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
795 nir_dest
*dest
= get_instr_dest(instr
);
797 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
799 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
800 this->result
= instr
;
804 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
807 if (ir
->as_dereference() || ir
->as_constant()) {
809 * A dereference is being used on the right hand side, which means we
810 * must emit a variable load.
813 nir_intrinsic_instr
*load_instr
=
814 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
815 load_instr
->num_components
= ir
->type
->vector_elements
;
816 load_instr
->variables
[0] = this->deref_head
;
817 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
820 nir_dest
*dest
= get_instr_dest(this->result
);
822 assert(dest
->is_ssa
);
825 src
.ssa
= &dest
->ssa
;
831 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
833 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
834 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
835 instr
->src
[i
].src
= srcs
[i
];
836 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
837 add_instr(&instr
->instr
, dest_size
);
842 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
844 assert(nir_op_infos
[op
].num_inputs
== 1);
845 return emit(op
, dest_size
, &src1
);
849 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
852 assert(nir_op_infos
[op
].num_inputs
== 2);
853 nir_src srcs
[] = { src1
, src2
};
854 return emit(op
, dest_size
, srcs
);
858 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
859 nir_src src2
, nir_src src3
)
861 assert(nir_op_infos
[op
].num_inputs
== 3);
862 nir_src srcs
[] = { src1
, src2
, src3
};
863 return emit(op
, dest_size
, srcs
);
867 nir_visitor::visit(ir_expression
*ir
)
869 /* Some special cases */
870 switch (ir
->operation
) {
871 case ir_binop_ubo_load
: {
872 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
876 op
= nir_intrinsic_load_ubo
;
878 op
= nir_intrinsic_load_ubo_indirect
;
880 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
881 load
->num_components
= ir
->type
->vector_elements
;
882 load
->const_index
[0] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
883 load
->const_index
[1] = 1; /* number of vec4's */
884 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
886 load
->src
[1] = evaluate_rvalue(ir
->operands
[1]);
887 add_instr(&load
->instr
, ir
->type
->vector_elements
);
890 * In UBO's, a true boolean value is any non-zero value, but we consider
891 * a true boolean to be ~0. Fix this up with a != 0 comparison.
894 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
895 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
, 1);
896 const_zero
->value
.u
[0] = 0;
897 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
899 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
900 compare
->src
[0].src
.is_ssa
= true;
901 compare
->src
[0].src
.ssa
= &load
->dest
.ssa
;
902 compare
->src
[1].src
.is_ssa
= true;
903 compare
->src
[1].src
.ssa
= &const_zero
->def
;
904 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
905 compare
->src
[1].swizzle
[i
] = 0;
906 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
908 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
914 case ir_unop_interpolate_at_centroid
:
915 case ir_binop_interpolate_at_offset
:
916 case ir_binop_interpolate_at_sample
: {
917 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
918 ir_swizzle
*swizzle
= NULL
;
920 /* the api does not allow a swizzle here, but the varying packing code
921 * may have pushed one into here.
923 swizzle
= ir
->operands
[0]->as_swizzle();
925 deref
= swizzle
->val
->as_dereference();
932 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
933 switch (ir
->operation
) {
934 case ir_unop_interpolate_at_centroid
:
935 op
= nir_intrinsic_interp_var_at_centroid
;
937 case ir_binop_interpolate_at_offset
:
938 op
= nir_intrinsic_interp_var_at_offset
;
940 case ir_binop_interpolate_at_sample
:
941 op
= nir_intrinsic_interp_var_at_sample
;
944 unreachable("Invalid interpolation intrinsic");
947 /* This case can happen if the vertex shader does not write the
948 * given varying. In this case, the linker will lower it to a
949 * global variable. Since interpolating a variable makes no
950 * sense, we'll just turn it into a load which will probably
951 * eventually end up as an SSA definition.
953 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
954 op
= nir_intrinsic_load_var
;
957 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
958 intrin
->num_components
= deref
->type
->vector_elements
;
959 intrin
->variables
[0] = this->deref_head
;
961 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
962 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
963 intrin
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
965 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
968 nir_alu_instr
*mov
= nir_alu_instr_create(shader
, nir_op_imov
);
969 mov
->dest
.write_mask
= (1 << swizzle
->type
->vector_elements
) - 1;
970 mov
->src
[0].src
.is_ssa
= true;
971 mov
->src
[0].src
.ssa
= &intrin
->dest
.ssa
;
973 mov
->src
[0].swizzle
[0] = swizzle
->mask
.x
;
974 mov
->src
[0].swizzle
[1] = swizzle
->mask
.y
;
975 mov
->src
[0].swizzle
[2] = swizzle
->mask
.z
;
976 mov
->src
[0].swizzle
[3] = swizzle
->mask
.w
;
977 for (unsigned i
= deref
->type
->vector_elements
; i
< 4; i
++)
978 mov
->src
[0].swizzle
[i
] = 0;
980 add_instr(&mov
->instr
, swizzle
->type
->vector_elements
);
991 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
992 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
994 glsl_base_type types
[4];
995 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
997 types
[i
] = ir
->operands
[i
]->type
->base_type
;
999 types
[i
] = GLSL_TYPE_FLOAT
;
1001 glsl_base_type out_type
;
1003 out_type
= ir
->type
->base_type
;
1005 out_type
= GLSL_TYPE_FLOAT
;
1007 unsigned dest_size
= ir
->type
->vector_elements
;
1009 nir_alu_instr
*instr
;
1012 switch (ir
->operation
) {
1013 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
1014 case ir_unop_logic_not
:
1015 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1018 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fneg
: nir_op_ineg
,
1022 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fabs
: nir_op_iabs
,
1025 case ir_unop_saturate
:
1026 assert(types
[0] == GLSL_TYPE_FLOAT
);
1027 instr
= emit(nir_op_fsat
, dest_size
, srcs
);
1030 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1033 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1034 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1035 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1036 case ir_unop_exp
: emit(nir_op_fexp
, dest_size
, srcs
); break;
1037 case ir_unop_log
: emit(nir_op_flog
, dest_size
, srcs
); break;
1038 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1039 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1041 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1044 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1047 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1049 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1050 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1051 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1052 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1053 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1056 case ir_unop_bitcast_i2f
:
1057 case ir_unop_bitcast_f2i
:
1058 case ir_unop_bitcast_u2f
:
1059 case ir_unop_bitcast_f2u
:
1061 emit(nir_op_imov
, dest_size
, srcs
);
1064 switch (ir
->operands
[0]->type
->vector_elements
) {
1066 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1070 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1074 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1078 unreachable("not reached");
1081 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1082 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1083 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1084 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1085 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1086 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1087 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1088 case ir_unop_sin_reduced
:
1089 emit(nir_op_fsin_reduced
, dest_size
, srcs
);
1091 case ir_unop_cos_reduced
:
1092 emit(nir_op_fcos_reduced
, dest_size
, srcs
);
1094 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1095 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1096 case ir_unop_dFdx_fine
: emit(nir_op_fddx_fine
, dest_size
, srcs
); break;
1097 case ir_unop_dFdy_fine
: emit(nir_op_fddy_fine
, dest_size
, srcs
); break;
1098 case ir_unop_dFdx_coarse
: emit(nir_op_fddx_coarse
, dest_size
, srcs
); break;
1099 case ir_unop_dFdy_coarse
: emit(nir_op_fddy_coarse
, dest_size
, srcs
); break;
1100 case ir_unop_pack_snorm_2x16
:
1101 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1103 case ir_unop_pack_snorm_4x8
:
1104 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1106 case ir_unop_pack_unorm_2x16
:
1107 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1109 case ir_unop_pack_unorm_4x8
:
1110 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1112 case ir_unop_pack_half_2x16
:
1113 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1115 case ir_unop_unpack_snorm_2x16
:
1116 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1118 case ir_unop_unpack_snorm_4x8
:
1119 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1121 case ir_unop_unpack_unorm_2x16
:
1122 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1124 case ir_unop_unpack_unorm_4x8
:
1125 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1127 case ir_unop_unpack_half_2x16
:
1128 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1130 case ir_unop_unpack_half_2x16_split_x
:
1131 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1133 case ir_unop_unpack_half_2x16_split_y
:
1134 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1136 case ir_unop_bitfield_reverse
:
1137 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1139 case ir_unop_bit_count
:
1140 emit(nir_op_bit_count
, dest_size
, srcs
);
1142 case ir_unop_find_msb
:
1144 case GLSL_TYPE_UINT
:
1145 emit(nir_op_ufind_msb
, dest_size
, srcs
);
1148 emit(nir_op_ifind_msb
, dest_size
, srcs
);
1151 unreachable("Invalid type for findMSB()");
1154 case ir_unop_find_lsb
:
1155 emit(nir_op_find_lsb
, dest_size
, srcs
);
1159 switch (ir
->type
->vector_elements
) {
1161 switch (ir
->operands
[0]->type
->vector_elements
) {
1162 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1163 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1164 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1165 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1166 default: unreachable("not reached");
1170 switch (ir
->operands
[0]->type
->vector_elements
) {
1171 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1172 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1173 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1174 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1175 default: unreachable("not reached");
1179 switch (ir
->operands
[0]->type
->vector_elements
) {
1180 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1181 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1182 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1183 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1184 default: unreachable("not reached");
1188 switch (ir
->operands
[0]->type
->vector_elements
) {
1189 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1190 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1191 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1192 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1193 default: unreachable("not reached");
1197 unreachable("not reached");
1208 case ir_binop_bit_and
:
1209 case ir_binop_bit_or
:
1210 case ir_binop_bit_xor
:
1211 case ir_binop_lshift
:
1212 case ir_binop_rshift
:
1213 switch (ir
->operation
) {
1215 if (out_type
== GLSL_TYPE_FLOAT
)
1221 if (out_type
== GLSL_TYPE_FLOAT
)
1227 if (out_type
== GLSL_TYPE_FLOAT
)
1233 if (out_type
== GLSL_TYPE_FLOAT
)
1235 else if (out_type
== GLSL_TYPE_INT
)
1241 if (out_type
== GLSL_TYPE_FLOAT
)
1247 if (out_type
== GLSL_TYPE_FLOAT
)
1249 else if (out_type
== GLSL_TYPE_INT
)
1255 if (out_type
== GLSL_TYPE_FLOAT
)
1257 else if (out_type
== GLSL_TYPE_INT
)
1262 case ir_binop_bit_and
:
1265 case ir_binop_bit_or
:
1268 case ir_binop_bit_xor
:
1271 case ir_binop_lshift
:
1274 case ir_binop_rshift
:
1275 if (out_type
== GLSL_TYPE_INT
)
1285 unreachable("not reached");
1288 instr
= emit(op
, dest_size
, srcs
);
1290 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1291 ir
->operands
[1]->type
->vector_elements
== 1) {
1292 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1294 instr
->src
[1].swizzle
[i
] = 0;
1298 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1299 ir
->operands
[0]->type
->vector_elements
== 1) {
1300 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1302 instr
->src
[0].swizzle
[i
] = 0;
1307 case ir_binop_imul_high
:
1308 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1311 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1312 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1314 if (supports_ints
) {
1315 if (types
[0] == GLSL_TYPE_FLOAT
)
1316 emit(nir_op_flt
, dest_size
, srcs
);
1317 else if (types
[0] == GLSL_TYPE_INT
)
1318 emit(nir_op_ilt
, dest_size
, srcs
);
1320 emit(nir_op_ult
, dest_size
, srcs
);
1322 emit(nir_op_slt
, dest_size
, srcs
);
1325 case ir_binop_greater
:
1326 if (supports_ints
) {
1327 if (types
[0] == GLSL_TYPE_FLOAT
)
1328 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1329 else if (types
[0] == GLSL_TYPE_INT
)
1330 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1332 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1334 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1337 case ir_binop_lequal
:
1338 if (supports_ints
) {
1339 if (types
[0] == GLSL_TYPE_FLOAT
)
1340 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1341 else if (types
[0] == GLSL_TYPE_INT
)
1342 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1344 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1346 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1349 case ir_binop_gequal
:
1350 if (supports_ints
) {
1351 if (types
[0] == GLSL_TYPE_FLOAT
)
1352 emit(nir_op_fge
, dest_size
, srcs
);
1353 else if (types
[0] == GLSL_TYPE_INT
)
1354 emit(nir_op_ige
, dest_size
, srcs
);
1356 emit(nir_op_uge
, dest_size
, srcs
);
1358 emit(nir_op_slt
, dest_size
, srcs
);
1361 case ir_binop_equal
:
1362 if (supports_ints
) {
1363 if (types
[0] == GLSL_TYPE_FLOAT
)
1364 emit(nir_op_feq
, dest_size
, srcs
);
1366 emit(nir_op_ieq
, dest_size
, srcs
);
1368 emit(nir_op_seq
, dest_size
, srcs
);
1371 case ir_binop_nequal
:
1372 if (supports_ints
) {
1373 if (types
[0] == GLSL_TYPE_FLOAT
)
1374 emit(nir_op_fne
, dest_size
, srcs
);
1376 emit(nir_op_ine
, dest_size
, srcs
);
1378 emit(nir_op_sne
, dest_size
, srcs
);
1381 case ir_binop_all_equal
:
1382 if (supports_ints
) {
1383 if (types
[0] == GLSL_TYPE_FLOAT
) {
1384 switch (ir
->operands
[0]->type
->vector_elements
) {
1385 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1386 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1387 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1388 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1390 unreachable("not reached");
1393 switch (ir
->operands
[0]->type
->vector_elements
) {
1394 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1395 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1396 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1397 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1399 unreachable("not reached");
1403 switch (ir
->operands
[0]->type
->vector_elements
) {
1404 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1405 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1406 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1407 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1409 unreachable("not reached");
1413 case ir_binop_any_nequal
:
1414 if (supports_ints
) {
1415 if (types
[0] == GLSL_TYPE_FLOAT
) {
1416 switch (ir
->operands
[0]->type
->vector_elements
) {
1417 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1418 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1419 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1420 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1422 unreachable("not reached");
1425 switch (ir
->operands
[0]->type
->vector_elements
) {
1426 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1427 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1428 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1429 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1431 unreachable("not reached");
1435 switch (ir
->operands
[0]->type
->vector_elements
) {
1436 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1437 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1438 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1439 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1441 unreachable("not reached");
1445 case ir_binop_logic_and
:
1447 emit(nir_op_iand
, dest_size
, srcs
);
1449 emit(nir_op_fand
, dest_size
, srcs
);
1451 case ir_binop_logic_or
:
1453 emit(nir_op_ior
, dest_size
, srcs
);
1455 emit(nir_op_for
, dest_size
, srcs
);
1457 case ir_binop_logic_xor
:
1459 emit(nir_op_ixor
, dest_size
, srcs
);
1461 emit(nir_op_fxor
, dest_size
, srcs
);
1464 switch (ir
->operands
[0]->type
->vector_elements
) {
1465 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1466 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1467 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1469 unreachable("not reached");
1473 case ir_binop_pack_half_2x16_split
:
1474 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1476 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1477 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1478 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1480 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1481 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1482 ir
->operands
[2]->type
->vector_elements
== 1) {
1483 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1485 instr
->src
[2].swizzle
[i
] = 0;
1491 emit(nir_op_bcsel
, dest_size
, srcs
);
1493 emit(nir_op_fcsel
, dest_size
, srcs
);
1496 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1497 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1498 instr
->src
[0].swizzle
[i
] = 0;
1501 case ir_triop_bitfield_extract
:
1502 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1503 nir_op_ubitfield_extract
, dest_size
, srcs
);
1504 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1505 instr
->src
[1].swizzle
[i
] = 0;
1506 instr
->src
[2].swizzle
[i
] = 0;
1509 case ir_quadop_bitfield_insert
:
1510 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1511 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1512 instr
->src
[2].swizzle
[i
] = 0;
1513 instr
->src
[3].swizzle
[i
] = 0;
1516 case ir_quadop_vector
:
1517 switch (ir
->type
->vector_elements
) {
1518 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1519 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1520 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1521 default: unreachable("not reached");
1526 unreachable("not reached");
1531 nir_visitor::visit(ir_swizzle
*ir
)
1533 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1534 ir
->type
->vector_elements
,
1535 evaluate_rvalue(ir
->val
));
1537 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1538 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1539 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1543 nir_visitor::visit(ir_texture
*ir
)
1550 num_srcs
= 1; /* coordinate */
1555 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1556 num_srcs
= 2; /* coordinate, bias/lod */
1560 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1566 if (ir
->lod_info
.lod
!= NULL
)
1567 num_srcs
= 2; /* coordinate, lod */
1569 num_srcs
= 1; /* coordinate */
1573 op
= nir_texop_txf_ms
;
1574 num_srcs
= 2; /* coordinate, sample_index */
1579 if (ir
->lod_info
.lod
!= NULL
)
1580 num_srcs
= 1; /* lod */
1587 num_srcs
= 1; /* coordinate */
1592 num_srcs
= 1; /* coordinate */
1595 case ir_query_levels
:
1596 op
= nir_texop_query_levels
;
1601 unreachable("not reached");
1604 if (ir
->projector
!= NULL
)
1606 if (ir
->shadow_comparitor
!= NULL
)
1608 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1611 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1614 instr
->sampler_dim
=
1615 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1616 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1617 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1618 if (instr
->is_shadow
)
1619 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1620 switch (ir
->type
->base_type
) {
1621 case GLSL_TYPE_FLOAT
:
1622 instr
->dest_type
= nir_type_float
;
1625 instr
->dest_type
= nir_type_int
;
1627 case GLSL_TYPE_UINT
:
1628 instr
->dest_type
= nir_type_unsigned
;
1631 unreachable("not reached");
1634 ir
->sampler
->accept(this);
1635 instr
->sampler
= this->deref_head
;
1637 unsigned src_number
= 0;
1639 if (ir
->coordinate
!= NULL
) {
1640 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1641 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->coordinate
);
1642 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1646 if (ir
->projector
!= NULL
) {
1647 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->projector
);
1648 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1652 if (ir
->shadow_comparitor
!= NULL
) {
1653 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->shadow_comparitor
);
1654 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1658 if (ir
->offset
!= NULL
) {
1659 /* we don't support multiple offsets yet */
1660 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1662 ir_constant
*const_offset
= ir
->offset
->as_constant();
1663 if (const_offset
!= NULL
) {
1664 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1665 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1667 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->offset
);
1668 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1675 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.bias
);
1676 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1683 if (ir
->lod_info
.lod
!= NULL
) {
1684 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.lod
);
1685 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1691 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdx
);
1692 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1694 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdy
);
1695 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1700 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.sample_index
);
1701 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1706 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1713 assert(src_number
== num_srcs
);
1715 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
1719 nir_visitor::visit(ir_constant
*ir
)
1722 * We don't know if this variable is an an array or struct that gets
1723 * dereferenced, so do the safe thing an make it a variable with a
1724 * constant initializer and return a dereference.
1727 nir_variable
*var
= ralloc(this->shader
, nir_variable
);
1728 var
->name
= ralloc_strdup(var
, "const_temp");
1729 var
->type
= ir
->type
;
1730 var
->data
.mode
= nir_var_local
;
1731 var
->data
.read_only
= true;
1732 var
->constant_initializer
= constant_copy(ir
, var
);
1733 exec_list_push_tail(&this->impl
->locals
, &var
->node
);
1735 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1736 this->deref_tail
= &this->deref_head
->deref
;
1740 nir_visitor::visit(ir_dereference_variable
*ir
)
1742 struct hash_entry
*entry
=
1743 _mesa_hash_table_search(this->var_table
, ir
->var
);
1745 nir_variable
*var
= (nir_variable
*) entry
->data
;
1747 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
1748 this->deref_head
= deref
;
1749 this->deref_tail
= &deref
->deref
;
1753 nir_visitor::visit(ir_dereference_record
*ir
)
1755 ir
->record
->accept(this);
1757 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
1758 assert(field_index
>= 0);
1760 nir_deref_struct
*deref
= nir_deref_struct_create(this->shader
, field_index
);
1761 deref
->deref
.type
= ir
->type
;
1762 this->deref_tail
->child
= &deref
->deref
;
1763 this->deref_tail
= &deref
->deref
;
1767 nir_visitor::visit(ir_dereference_array
*ir
)
1769 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
1770 deref
->deref
.type
= ir
->type
;
1772 ir_constant
*const_index
= ir
->array_index
->as_constant();
1773 if (const_index
!= NULL
) {
1774 deref
->deref_array_type
= nir_deref_array_type_direct
;
1775 deref
->base_offset
= const_index
->value
.u
[0];
1777 deref
->deref_array_type
= nir_deref_array_type_indirect
;
1778 deref
->indirect
= evaluate_rvalue(ir
->array_index
);
1781 ir
->array
->accept(this);
1783 this->deref_tail
->child
= &deref
->deref
;
1784 this->deref_tail
= &deref
->deref
;