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
, gl_shader_stage stage
);
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 gl_shader_stage stage
;
87 nir_function_impl
*impl
;
88 exec_list
*cf_node_list
;
89 nir_instr
*result
; /* result of the expression tree last visited */
91 /* the head of the dereference chain we're creating */
92 nir_deref_var
*deref_head
;
93 /* the tail of the dereference chain we're creating */
94 nir_deref
*deref_tail
;
96 nir_variable
*var
; /* variable created by ir_variable visitor */
98 /* whether the IR we're operating on is per-function or global */
101 /* map of ir_variable -> nir_variable */
102 struct hash_table
*var_table
;
104 /* map of ir_function_signature -> nir_function_overload */
105 struct hash_table
*overload_table
;
109 * This visitor runs before the main visitor, calling create_function() for
110 * each function so that the main visitor can resolve forward references in
114 class nir_function_visitor
: public ir_hierarchical_visitor
117 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
120 virtual ir_visitor_status
visit_enter(ir_function
*);
123 nir_visitor
*visitor
;
126 }; /* end of anonymous namespace */
129 glsl_to_nir(struct gl_shader
*sh
, const nir_shader_compiler_options
*options
)
131 nir_shader
*shader
= nir_shader_create(NULL
, options
);
133 nir_visitor
v1(shader
, sh
->Stage
);
134 nir_function_visitor
v2(&v1
);
136 visit_exec_list(sh
->ir
, &v1
);
141 nir_visitor::nir_visitor(nir_shader
*shader
, gl_shader_stage stage
)
143 this->supports_ints
= shader
->options
->native_integers
;
144 this->shader
= shader
;
146 this->is_global
= true;
147 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
148 _mesa_key_pointer_equal
);
149 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
150 _mesa_key_pointer_equal
);
153 nir_visitor::~nir_visitor()
155 _mesa_hash_table_destroy(this->var_table
, NULL
);
156 _mesa_hash_table_destroy(this->overload_table
, NULL
);
159 static nir_constant
*
160 constant_copy(ir_constant
*ir
, void *mem_ctx
)
165 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
167 unsigned total_elems
= ir
->type
->components();
169 switch (ir
->type
->base_type
) {
171 for (i
= 0; i
< total_elems
; i
++)
172 ret
->value
.u
[i
] = ir
->value
.u
[i
];
176 for (i
= 0; i
< total_elems
; i
++)
177 ret
->value
.i
[i
] = ir
->value
.i
[i
];
180 case GLSL_TYPE_FLOAT
:
181 for (i
= 0; i
< total_elems
; i
++)
182 ret
->value
.f
[i
] = ir
->value
.f
[i
];
186 for (i
= 0; i
< total_elems
; i
++)
187 ret
->value
.b
[i
] = ir
->value
.b
[i
];
190 case GLSL_TYPE_STRUCT
:
191 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
194 foreach_in_list(ir_constant
, field
, &ir
->components
) {
195 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
200 case GLSL_TYPE_ARRAY
:
201 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
204 for (i
= 0; i
< ir
->type
->length
; i
++)
205 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
209 unreachable("not reached");
216 nir_visitor::visit(ir_variable
*ir
)
218 nir_variable
*var
= ralloc(shader
, nir_variable
);
219 var
->type
= ir
->type
;
220 var
->name
= ralloc_strdup(var
, ir
->name
);
222 if (ir
->is_interface_instance() && ir
->get_max_ifc_array_access() != NULL
) {
223 unsigned size
= ir
->get_interface_type()->length
;
224 var
->max_ifc_array_access
= ralloc_array(var
, unsigned, size
);
225 memcpy(var
->max_ifc_array_access
, ir
->get_max_ifc_array_access(),
226 size
* sizeof(unsigned));
228 var
->max_ifc_array_access
= NULL
;
231 var
->data
.read_only
= ir
->data
.read_only
;
232 var
->data
.centroid
= ir
->data
.centroid
;
233 var
->data
.sample
= ir
->data
.sample
;
234 var
->data
.invariant
= ir
->data
.invariant
;
235 var
->data
.location
= ir
->data
.location
;
237 switch(ir
->data
.mode
) {
239 case ir_var_temporary
:
241 var
->data
.mode
= nir_var_global
;
243 var
->data
.mode
= nir_var_local
;
246 case ir_var_function_in
:
247 case ir_var_function_out
:
248 case ir_var_function_inout
:
249 case ir_var_const_in
:
250 var
->data
.mode
= nir_var_local
;
253 case ir_var_shader_in
:
254 if (ir
->data
.location
== VARYING_SLOT_FACE
) {
255 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
256 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
257 var
->data
.mode
= nir_var_system_value
;
259 var
->data
.mode
= nir_var_shader_in
;
263 case ir_var_shader_out
:
264 var
->data
.mode
= nir_var_shader_out
;
268 var
->data
.mode
= nir_var_uniform
;
272 case ir_var_system_value
:
273 var
->data
.mode
= nir_var_system_value
;
277 unreachable("not reached");
280 var
->data
.interpolation
= ir
->data
.interpolation
;
281 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
282 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
283 var
->data
.explicit_location
= ir
->data
.explicit_location
;
284 var
->data
.explicit_index
= ir
->data
.explicit_index
;
285 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
286 var
->data
.has_initializer
= ir
->data
.has_initializer
;
287 var
->data
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
288 var
->data
.location_frac
= ir
->data
.location_frac
;
289 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
290 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
292 switch (ir
->data
.depth_layout
) {
293 case ir_depth_layout_none
:
294 var
->data
.depth_layout
= nir_depth_layout_none
;
296 case ir_depth_layout_any
:
297 var
->data
.depth_layout
= nir_depth_layout_any
;
299 case ir_depth_layout_greater
:
300 var
->data
.depth_layout
= nir_depth_layout_greater
;
302 case ir_depth_layout_less
:
303 var
->data
.depth_layout
= nir_depth_layout_less
;
305 case ir_depth_layout_unchanged
:
306 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
309 unreachable("not reached");
312 var
->data
.index
= ir
->data
.index
;
313 var
->data
.binding
= ir
->data
.binding
;
314 /* XXX Get rid of buffer_index */
315 var
->data
.atomic
.buffer_index
= ir
->data
.binding
;
316 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
317 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
318 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
319 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
320 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
321 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
322 var
->data
.image
.format
= ir
->data
.image_format
;
323 var
->data
.max_array_access
= ir
->data
.max_array_access
;
325 var
->num_state_slots
= ir
->get_num_state_slots();
326 if (var
->num_state_slots
> 0) {
327 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
328 var
->num_state_slots
);
330 ir_state_slot
*state_slots
= ir
->get_state_slots();
331 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
332 for (unsigned j
= 0; j
< 5; j
++)
333 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
334 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
337 var
->state_slots
= NULL
;
340 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
342 var
->interface_type
= ir
->get_interface_type();
344 switch (var
->data
.mode
) {
346 exec_list_push_tail(&impl
->locals
, &var
->node
);
350 exec_list_push_tail(&shader
->globals
, &var
->node
);
353 case nir_var_shader_in
:
354 _mesa_hash_table_insert(shader
->inputs
, var
->name
, var
);
357 case nir_var_shader_out
:
358 _mesa_hash_table_insert(shader
->outputs
, var
->name
, var
);
361 case nir_var_uniform
:
362 _mesa_hash_table_insert(shader
->uniforms
, var
->name
, var
);
365 case nir_var_system_value
:
366 exec_list_push_tail(&shader
->system_values
, &var
->node
);
370 unreachable("not reached");
373 _mesa_hash_table_insert(var_table
, ir
, var
);
378 nir_function_visitor::visit_enter(ir_function
*ir
)
380 visitor
->create_function(ir
);
381 return visit_continue_with_parent
;
386 nir_visitor::create_function(ir_function
*ir
)
388 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
389 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
390 create_overload(sig
, func
);
397 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
399 if (ir
->is_intrinsic
)
402 nir_function_overload
*overload
= nir_function_overload_create(function
);
404 unsigned num_params
= ir
->parameters
.length();
405 overload
->num_params
= num_params
;
406 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
409 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
410 switch (param
->data
.mode
) {
411 case ir_var_function_in
:
412 overload
->params
[i
].param_type
= nir_parameter_in
;
415 case ir_var_function_out
:
416 overload
->params
[i
].param_type
= nir_parameter_out
;
419 case ir_var_function_inout
:
420 overload
->params
[i
].param_type
= nir_parameter_inout
;
424 unreachable("not reached");
427 overload
->params
[i
].type
= param
->type
;
431 overload
->return_type
= ir
->return_type
;
433 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
437 nir_visitor::visit(ir_function
*ir
)
439 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
444 nir_visitor::visit(ir_function_signature
*ir
)
446 if (ir
->is_intrinsic
)
449 struct hash_entry
*entry
=
450 _mesa_hash_table_search(this->overload_table
, ir
);
453 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
455 if (ir
->is_defined
) {
456 nir_function_impl
*impl
= nir_function_impl_create(overload
);
459 unsigned num_params
= overload
->num_params
;
460 impl
->num_params
= num_params
;
461 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
463 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
465 impl
->params
[i
] = this->var
;
469 if (overload
->return_type
== glsl_type::void_type
) {
470 impl
->return_var
= NULL
;
472 impl
->return_var
= ralloc(this->shader
, nir_variable
);
473 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
475 impl
->return_var
->type
= overload
->return_type
;
478 this->is_global
= false;
480 this->cf_node_list
= &impl
->body
;
481 visit_exec_list(&ir
->body
, this);
483 this->is_global
= true;
485 overload
->impl
= NULL
;
490 nir_visitor::visit(ir_loop
*ir
)
492 exec_list
*old_list
= this->cf_node_list
;
494 nir_loop
*loop
= nir_loop_create(this->shader
);
495 nir_cf_node_insert_end(old_list
, &loop
->cf_node
);
496 this->cf_node_list
= &loop
->body
;
497 visit_exec_list(&ir
->body_instructions
, this);
499 this->cf_node_list
= old_list
;
503 nir_visitor::visit(ir_if
*ir
)
505 nir_src condition
= evaluate_rvalue(ir
->condition
);
507 exec_list
*old_list
= this->cf_node_list
;
509 nir_if
*if_stmt
= nir_if_create(this->shader
);
510 if_stmt
->condition
= condition
;
511 nir_cf_node_insert_end(old_list
, &if_stmt
->cf_node
);
513 this->cf_node_list
= &if_stmt
->then_list
;
514 visit_exec_list(&ir
->then_instructions
, this);
516 this->cf_node_list
= &if_stmt
->else_list
;
517 visit_exec_list(&ir
->else_instructions
, this);
519 this->cf_node_list
= old_list
;
523 nir_visitor::visit(ir_discard
*ir
)
526 * discards aren't treated as control flow, because before we lower them
527 * they can appear anywhere in the shader and the stuff after them may still
528 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
529 * discards will be immediately followed by a return.
532 nir_intrinsic_instr
*discard
;
534 discard
= nir_intrinsic_instr_create(this->shader
,
535 nir_intrinsic_discard_if
);
536 discard
->src
[0] = evaluate_rvalue(ir
->condition
);
538 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
540 nir_instr_insert_after_cf_list(this->cf_node_list
, &discard
->instr
);
544 nir_visitor::visit(ir_emit_vertex
*ir
)
546 nir_intrinsic_instr
*instr
=
547 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
548 instr
->const_index
[0] = ir
->stream_id();
549 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
553 nir_visitor::visit(ir_end_primitive
*ir
)
555 nir_intrinsic_instr
*instr
=
556 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
557 instr
->const_index
[0] = ir
->stream_id();
558 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
562 nir_visitor::visit(ir_loop_jump
*ir
)
566 case ir_loop_jump::jump_break
:
567 type
= nir_jump_break
;
569 case ir_loop_jump::jump_continue
:
570 type
= nir_jump_continue
;
573 unreachable("not reached");
576 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
577 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
581 nir_visitor::visit(ir_return
*ir
)
583 if (ir
->value
!= NULL
) {
584 ir
->value
->accept(this);
585 nir_intrinsic_instr
*copy
=
586 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
588 copy
->variables
[0] = nir_deref_var_create(this->shader
,
589 this->impl
->return_var
);
590 copy
->variables
[1] = this->deref_head
;
593 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
594 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
598 nir_visitor::visit(ir_call
*ir
)
600 if (ir
->callee
->is_intrinsic
) {
602 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
603 op
= nir_intrinsic_atomic_counter_read_var
;
604 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
605 op
= nir_intrinsic_atomic_counter_inc_var
;
606 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
607 op
= nir_intrinsic_atomic_counter_dec_var
;
609 unreachable("not reached");
612 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
613 ir_dereference
*param
=
614 (ir_dereference
*) ir
->actual_parameters
.get_head();
616 instr
->variables
[0] = this->deref_head
;
617 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
619 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
621 nir_intrinsic_instr
*store_instr
=
622 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
623 store_instr
->num_components
= 1;
625 ir
->return_deref
->accept(this);
626 store_instr
->variables
[0] = this->deref_head
;
627 store_instr
->src
[0].is_ssa
= true;
628 store_instr
->src
[0].ssa
= &instr
->dest
.ssa
;
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 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
659 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
660 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
661 /* We're doing a plain-as-can-be copy, so emit a copy_var */
662 nir_intrinsic_instr
*copy
=
663 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
665 ir
->lhs
->accept(this);
666 copy
->variables
[0] = this->deref_head
;
668 ir
->rhs
->accept(this);
669 copy
->variables
[1] = this->deref_head
;
673 nir_if
*if_stmt
= nir_if_create(this->shader
);
674 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
675 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
676 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
678 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
683 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
685 ir
->lhs
->accept(this);
686 nir_deref_var
*lhs_deref
= this->deref_head
;
687 nir_src src
= evaluate_rvalue(ir
->rhs
);
689 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
691 * We have no good way to update only part of a variable, so just load
692 * the LHS and do a vec operation to combine the old with the new, and
694 * back into the LHS. Copy propagation should get rid of the mess.
697 nir_intrinsic_instr
*load
=
698 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
699 load
->num_components
= ir
->lhs
->type
->vector_elements
;
700 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
701 load
->variables
[0] = lhs_deref
;
702 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
705 switch (ir
->lhs
->type
->vector_elements
) {
706 case 1: vec_op
= nir_op_imov
; break;
707 case 2: vec_op
= nir_op_vec2
; break;
708 case 3: vec_op
= nir_op_vec3
; break;
709 case 4: vec_op
= nir_op_vec4
; break;
710 default: unreachable("Invalid number of components"); break;
712 nir_alu_instr
*vec
= nir_alu_instr_create(this->shader
, vec_op
);
713 nir_ssa_dest_init(&vec
->instr
, &vec
->dest
.dest
, num_components
, NULL
);
714 vec
->dest
.write_mask
= (1 << num_components
) - 1;
716 unsigned component
= 0;
717 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
718 if (ir
->write_mask
& (1 << i
)) {
719 vec
->src
[i
].src
= src
;
721 /* GLSL IR will give us the input to the write-masked assignment
722 * in a single packed vector. So, for example, if the
723 * writemask is xzw, then we have to swizzle x -> x, y -> z,
724 * and z -> w and get the y component from the load.
726 vec
->src
[i
].swizzle
[0] = component
++;
728 vec
->src
[i
].src
.is_ssa
= true;
729 vec
->src
[i
].src
.ssa
= &load
->dest
.ssa
;
730 vec
->src
[i
].swizzle
[0] = i
;
734 nir_instr_insert_after_cf_list(this->cf_node_list
, &vec
->instr
);
737 src
.ssa
= &vec
->dest
.dest
.ssa
;
740 nir_intrinsic_instr
*store
=
741 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
742 store
->num_components
= ir
->lhs
->type
->vector_elements
;
743 nir_deref
*store_deref
= nir_copy_deref(this->shader
, &lhs_deref
->deref
);
744 store
->variables
[0] = nir_deref_as_var(store_deref
);
748 nir_if
*if_stmt
= nir_if_create(this->shader
);
749 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
750 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
751 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
753 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
758 * Given an instruction, returns a pointer to its destination or NULL if there
761 * Note that this only handles instructions we generate at this level.
764 get_instr_dest(nir_instr
*instr
)
766 nir_alu_instr
*alu_instr
;
767 nir_intrinsic_instr
*intrinsic_instr
;
768 nir_tex_instr
*tex_instr
;
770 switch (instr
->type
) {
771 case nir_instr_type_alu
:
772 alu_instr
= nir_instr_as_alu(instr
);
773 return &alu_instr
->dest
.dest
;
775 case nir_instr_type_intrinsic
:
776 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
777 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
778 return &intrinsic_instr
->dest
;
782 case nir_instr_type_tex
:
783 tex_instr
= nir_instr_as_tex(instr
);
784 return &tex_instr
->dest
;
787 unreachable("not reached");
794 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
796 nir_dest
*dest
= get_instr_dest(instr
);
798 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
800 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
801 this->result
= instr
;
805 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
808 if (ir
->as_dereference() || ir
->as_constant()) {
810 * A dereference is being used on the right hand side, which means we
811 * must emit a variable load.
814 nir_intrinsic_instr
*load_instr
=
815 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
816 load_instr
->num_components
= ir
->type
->vector_elements
;
817 load_instr
->variables
[0] = this->deref_head
;
818 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
821 nir_dest
*dest
= get_instr_dest(this->result
);
823 assert(dest
->is_ssa
);
826 src
.ssa
= &dest
->ssa
;
832 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
834 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
835 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
836 instr
->src
[i
].src
= srcs
[i
];
837 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
838 add_instr(&instr
->instr
, dest_size
);
843 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
845 assert(nir_op_infos
[op
].num_inputs
== 1);
846 return emit(op
, dest_size
, &src1
);
850 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
853 assert(nir_op_infos
[op
].num_inputs
== 2);
854 nir_src srcs
[] = { src1
, src2
};
855 return emit(op
, dest_size
, srcs
);
859 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
860 nir_src src2
, nir_src src3
)
862 assert(nir_op_infos
[op
].num_inputs
== 3);
863 nir_src srcs
[] = { src1
, src2
, src3
};
864 return emit(op
, dest_size
, srcs
);
868 nir_visitor::visit(ir_expression
*ir
)
870 /* Some special cases */
871 switch (ir
->operation
) {
872 case ir_binop_ubo_load
: {
873 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
877 op
= nir_intrinsic_load_ubo
;
879 op
= nir_intrinsic_load_ubo_indirect
;
881 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
882 load
->num_components
= ir
->type
->vector_elements
;
883 load
->const_index
[0] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
884 load
->const_index
[1] = 1; /* number of vec4's */
885 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
887 load
->src
[1] = evaluate_rvalue(ir
->operands
[1]);
888 add_instr(&load
->instr
, ir
->type
->vector_elements
);
891 * In UBO's, a true boolean value is any non-zero value, but we consider
892 * a true boolean to be ~0. Fix this up with a != 0 comparison.
895 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
896 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
, 1);
897 const_zero
->value
.u
[0] = 0;
898 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
900 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
901 compare
->src
[0].src
.is_ssa
= true;
902 compare
->src
[0].src
.ssa
= &load
->dest
.ssa
;
903 compare
->src
[1].src
.is_ssa
= true;
904 compare
->src
[1].src
.ssa
= &const_zero
->def
;
905 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
906 compare
->src
[1].swizzle
[i
] = 0;
907 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
909 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
915 case ir_unop_interpolate_at_centroid
:
916 case ir_binop_interpolate_at_offset
:
917 case ir_binop_interpolate_at_sample
: {
918 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
919 ir_swizzle
*swizzle
= NULL
;
921 /* the api does not allow a swizzle here, but the varying packing code
922 * may have pushed one into here.
924 swizzle
= ir
->operands
[0]->as_swizzle();
926 deref
= swizzle
->val
->as_dereference();
933 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
934 switch (ir
->operation
) {
935 case ir_unop_interpolate_at_centroid
:
936 op
= nir_intrinsic_interp_var_at_centroid
;
938 case ir_binop_interpolate_at_offset
:
939 op
= nir_intrinsic_interp_var_at_offset
;
941 case ir_binop_interpolate_at_sample
:
942 op
= nir_intrinsic_interp_var_at_sample
;
945 unreachable("Invalid interpolation intrinsic");
948 /* This case can happen if the vertex shader does not write the
949 * given varying. In this case, the linker will lower it to a
950 * global variable. Since interpolating a variable makes no
951 * sense, we'll just turn it into a load which will probably
952 * eventually end up as an SSA definition.
954 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
955 op
= nir_intrinsic_load_var
;
958 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
959 intrin
->num_components
= deref
->type
->vector_elements
;
960 intrin
->variables
[0] = this->deref_head
;
962 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
963 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
964 intrin
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
966 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
969 nir_alu_instr
*mov
= nir_alu_instr_create(shader
, nir_op_imov
);
970 mov
->dest
.write_mask
= (1 << swizzle
->type
->vector_elements
) - 1;
971 mov
->src
[0].src
.is_ssa
= true;
972 mov
->src
[0].src
.ssa
= &intrin
->dest
.ssa
;
974 mov
->src
[0].swizzle
[0] = swizzle
->mask
.x
;
975 mov
->src
[0].swizzle
[1] = swizzle
->mask
.y
;
976 mov
->src
[0].swizzle
[2] = swizzle
->mask
.z
;
977 mov
->src
[0].swizzle
[3] = swizzle
->mask
.w
;
978 for (unsigned i
= deref
->type
->vector_elements
; i
< 4; i
++)
979 mov
->src
[0].swizzle
[i
] = 0;
981 add_instr(&mov
->instr
, swizzle
->type
->vector_elements
);
992 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
993 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
995 glsl_base_type types
[4];
996 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
998 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1000 types
[i
] = GLSL_TYPE_FLOAT
;
1002 glsl_base_type out_type
;
1004 out_type
= ir
->type
->base_type
;
1006 out_type
= GLSL_TYPE_FLOAT
;
1008 unsigned dest_size
= ir
->type
->vector_elements
;
1010 nir_alu_instr
*instr
;
1013 switch (ir
->operation
) {
1014 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
1015 case ir_unop_logic_not
:
1016 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1019 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fneg
: nir_op_ineg
,
1023 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fabs
: nir_op_iabs
,
1026 case ir_unop_saturate
:
1027 assert(types
[0] == GLSL_TYPE_FLOAT
);
1028 instr
= emit(nir_op_fsat
, dest_size
, srcs
);
1031 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1034 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1035 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1036 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1037 case ir_unop_exp
: emit(nir_op_fexp
, dest_size
, srcs
); break;
1038 case ir_unop_log
: emit(nir_op_flog
, dest_size
, srcs
); break;
1039 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1040 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1042 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1045 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1048 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1050 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1051 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1052 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1053 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1054 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1057 case ir_unop_bitcast_i2f
:
1058 case ir_unop_bitcast_f2i
:
1059 case ir_unop_bitcast_u2f
:
1060 case ir_unop_bitcast_f2u
:
1062 emit(nir_op_imov
, dest_size
, srcs
);
1065 switch (ir
->operands
[0]->type
->vector_elements
) {
1067 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1071 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1075 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1079 unreachable("not reached");
1082 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1083 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1084 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1085 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1086 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1087 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1088 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1089 case ir_unop_sin_reduced
:
1090 emit(nir_op_fsin_reduced
, dest_size
, srcs
);
1092 case ir_unop_cos_reduced
:
1093 emit(nir_op_fcos_reduced
, dest_size
, srcs
);
1095 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1096 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1097 case ir_unop_dFdx_fine
: emit(nir_op_fddx_fine
, dest_size
, srcs
); break;
1098 case ir_unop_dFdy_fine
: emit(nir_op_fddy_fine
, dest_size
, srcs
); break;
1099 case ir_unop_dFdx_coarse
: emit(nir_op_fddx_coarse
, dest_size
, srcs
); break;
1100 case ir_unop_dFdy_coarse
: emit(nir_op_fddy_coarse
, dest_size
, srcs
); break;
1101 case ir_unop_pack_snorm_2x16
:
1102 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1104 case ir_unop_pack_snorm_4x8
:
1105 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1107 case ir_unop_pack_unorm_2x16
:
1108 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1110 case ir_unop_pack_unorm_4x8
:
1111 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1113 case ir_unop_pack_half_2x16
:
1114 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1116 case ir_unop_unpack_snorm_2x16
:
1117 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1119 case ir_unop_unpack_snorm_4x8
:
1120 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1122 case ir_unop_unpack_unorm_2x16
:
1123 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1125 case ir_unop_unpack_unorm_4x8
:
1126 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1128 case ir_unop_unpack_half_2x16
:
1129 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1131 case ir_unop_unpack_half_2x16_split_x
:
1132 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1134 case ir_unop_unpack_half_2x16_split_y
:
1135 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1137 case ir_unop_bitfield_reverse
:
1138 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1140 case ir_unop_bit_count
:
1141 emit(nir_op_bit_count
, dest_size
, srcs
);
1143 case ir_unop_find_msb
:
1145 case GLSL_TYPE_UINT
:
1146 emit(nir_op_ufind_msb
, dest_size
, srcs
);
1149 emit(nir_op_ifind_msb
, dest_size
, srcs
);
1152 unreachable("Invalid type for findMSB()");
1155 case ir_unop_find_lsb
:
1156 emit(nir_op_find_lsb
, dest_size
, srcs
);
1160 switch (ir
->type
->vector_elements
) {
1162 switch (ir
->operands
[0]->type
->vector_elements
) {
1163 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1164 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1165 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1166 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1167 default: unreachable("not reached");
1171 switch (ir
->operands
[0]->type
->vector_elements
) {
1172 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1173 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1174 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1175 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1176 default: unreachable("not reached");
1180 switch (ir
->operands
[0]->type
->vector_elements
) {
1181 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1182 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1183 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1184 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1185 default: unreachable("not reached");
1189 switch (ir
->operands
[0]->type
->vector_elements
) {
1190 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1191 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1192 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1193 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1194 default: unreachable("not reached");
1198 unreachable("not reached");
1209 case ir_binop_bit_and
:
1210 case ir_binop_bit_or
:
1211 case ir_binop_bit_xor
:
1212 case ir_binop_lshift
:
1213 case ir_binop_rshift
:
1214 switch (ir
->operation
) {
1216 if (out_type
== GLSL_TYPE_FLOAT
)
1222 if (out_type
== GLSL_TYPE_FLOAT
)
1228 if (out_type
== GLSL_TYPE_FLOAT
)
1234 if (out_type
== GLSL_TYPE_FLOAT
)
1236 else if (out_type
== GLSL_TYPE_INT
)
1242 if (out_type
== GLSL_TYPE_FLOAT
)
1248 if (out_type
== GLSL_TYPE_FLOAT
)
1250 else if (out_type
== GLSL_TYPE_INT
)
1256 if (out_type
== GLSL_TYPE_FLOAT
)
1258 else if (out_type
== GLSL_TYPE_INT
)
1263 case ir_binop_bit_and
:
1266 case ir_binop_bit_or
:
1269 case ir_binop_bit_xor
:
1272 case ir_binop_lshift
:
1275 case ir_binop_rshift
:
1276 if (out_type
== GLSL_TYPE_INT
)
1286 unreachable("not reached");
1289 instr
= emit(op
, dest_size
, srcs
);
1291 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1292 ir
->operands
[1]->type
->vector_elements
== 1) {
1293 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1295 instr
->src
[1].swizzle
[i
] = 0;
1299 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1300 ir
->operands
[0]->type
->vector_elements
== 1) {
1301 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1303 instr
->src
[0].swizzle
[i
] = 0;
1308 case ir_binop_imul_high
:
1309 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1312 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1313 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1315 if (supports_ints
) {
1316 if (types
[0] == GLSL_TYPE_FLOAT
)
1317 emit(nir_op_flt
, dest_size
, srcs
);
1318 else if (types
[0] == GLSL_TYPE_INT
)
1319 emit(nir_op_ilt
, dest_size
, srcs
);
1321 emit(nir_op_ult
, dest_size
, srcs
);
1323 emit(nir_op_slt
, dest_size
, srcs
);
1326 case ir_binop_greater
:
1327 if (supports_ints
) {
1328 if (types
[0] == GLSL_TYPE_FLOAT
)
1329 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1330 else if (types
[0] == GLSL_TYPE_INT
)
1331 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1333 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1335 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1338 case ir_binop_lequal
:
1339 if (supports_ints
) {
1340 if (types
[0] == GLSL_TYPE_FLOAT
)
1341 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1342 else if (types
[0] == GLSL_TYPE_INT
)
1343 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1345 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1347 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1350 case ir_binop_gequal
:
1351 if (supports_ints
) {
1352 if (types
[0] == GLSL_TYPE_FLOAT
)
1353 emit(nir_op_fge
, dest_size
, srcs
);
1354 else if (types
[0] == GLSL_TYPE_INT
)
1355 emit(nir_op_ige
, dest_size
, srcs
);
1357 emit(nir_op_uge
, dest_size
, srcs
);
1359 emit(nir_op_slt
, dest_size
, srcs
);
1362 case ir_binop_equal
:
1363 if (supports_ints
) {
1364 if (types
[0] == GLSL_TYPE_FLOAT
)
1365 emit(nir_op_feq
, dest_size
, srcs
);
1367 emit(nir_op_ieq
, dest_size
, srcs
);
1369 emit(nir_op_seq
, dest_size
, srcs
);
1372 case ir_binop_nequal
:
1373 if (supports_ints
) {
1374 if (types
[0] == GLSL_TYPE_FLOAT
)
1375 emit(nir_op_fne
, dest_size
, srcs
);
1377 emit(nir_op_ine
, dest_size
, srcs
);
1379 emit(nir_op_sne
, dest_size
, srcs
);
1382 case ir_binop_all_equal
:
1383 if (supports_ints
) {
1384 if (types
[0] == GLSL_TYPE_FLOAT
) {
1385 switch (ir
->operands
[0]->type
->vector_elements
) {
1386 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1387 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1388 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1389 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1391 unreachable("not reached");
1394 switch (ir
->operands
[0]->type
->vector_elements
) {
1395 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1396 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1397 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1398 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1400 unreachable("not reached");
1404 switch (ir
->operands
[0]->type
->vector_elements
) {
1405 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1406 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1407 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1408 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1410 unreachable("not reached");
1414 case ir_binop_any_nequal
:
1415 if (supports_ints
) {
1416 if (types
[0] == GLSL_TYPE_FLOAT
) {
1417 switch (ir
->operands
[0]->type
->vector_elements
) {
1418 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1419 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1420 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1421 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1423 unreachable("not reached");
1426 switch (ir
->operands
[0]->type
->vector_elements
) {
1427 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1428 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1429 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1430 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1432 unreachable("not reached");
1436 switch (ir
->operands
[0]->type
->vector_elements
) {
1437 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1438 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1439 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1440 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1442 unreachable("not reached");
1446 case ir_binop_logic_and
:
1448 emit(nir_op_iand
, dest_size
, srcs
);
1450 emit(nir_op_fand
, dest_size
, srcs
);
1452 case ir_binop_logic_or
:
1454 emit(nir_op_ior
, dest_size
, srcs
);
1456 emit(nir_op_for
, dest_size
, srcs
);
1458 case ir_binop_logic_xor
:
1460 emit(nir_op_ixor
, dest_size
, srcs
);
1462 emit(nir_op_fxor
, dest_size
, srcs
);
1465 switch (ir
->operands
[0]->type
->vector_elements
) {
1466 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1467 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1468 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1470 unreachable("not reached");
1474 case ir_binop_pack_half_2x16_split
:
1475 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1477 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1478 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1479 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1481 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1482 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1483 ir
->operands
[2]->type
->vector_elements
== 1) {
1484 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1486 instr
->src
[2].swizzle
[i
] = 0;
1492 emit(nir_op_bcsel
, dest_size
, srcs
);
1494 emit(nir_op_fcsel
, dest_size
, srcs
);
1497 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1498 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1499 instr
->src
[0].swizzle
[i
] = 0;
1502 case ir_triop_bitfield_extract
:
1503 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1504 nir_op_ubitfield_extract
, dest_size
, srcs
);
1505 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1506 instr
->src
[1].swizzle
[i
] = 0;
1507 instr
->src
[2].swizzle
[i
] = 0;
1510 case ir_quadop_bitfield_insert
:
1511 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1512 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1513 instr
->src
[2].swizzle
[i
] = 0;
1514 instr
->src
[3].swizzle
[i
] = 0;
1517 case ir_quadop_vector
:
1518 switch (ir
->type
->vector_elements
) {
1519 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1520 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1521 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1522 default: unreachable("not reached");
1527 unreachable("not reached");
1532 nir_visitor::visit(ir_swizzle
*ir
)
1534 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1535 ir
->type
->vector_elements
,
1536 evaluate_rvalue(ir
->val
));
1538 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1539 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1540 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1544 nir_visitor::visit(ir_texture
*ir
)
1551 num_srcs
= 1; /* coordinate */
1556 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1557 num_srcs
= 2; /* coordinate, bias/lod */
1561 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1567 if (ir
->lod_info
.lod
!= NULL
)
1568 num_srcs
= 2; /* coordinate, lod */
1570 num_srcs
= 1; /* coordinate */
1574 op
= nir_texop_txf_ms
;
1575 num_srcs
= 2; /* coordinate, sample_index */
1580 if (ir
->lod_info
.lod
!= NULL
)
1581 num_srcs
= 1; /* lod */
1588 num_srcs
= 1; /* coordinate */
1593 num_srcs
= 1; /* coordinate */
1596 case ir_query_levels
:
1597 op
= nir_texop_query_levels
;
1602 unreachable("not reached");
1605 if (ir
->projector
!= NULL
)
1607 if (ir
->shadow_comparitor
!= NULL
)
1609 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1612 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1615 instr
->sampler_dim
=
1616 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1617 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1618 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1619 if (instr
->is_shadow
)
1620 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1621 switch (ir
->type
->base_type
) {
1622 case GLSL_TYPE_FLOAT
:
1623 instr
->dest_type
= nir_type_float
;
1626 instr
->dest_type
= nir_type_int
;
1628 case GLSL_TYPE_UINT
:
1629 instr
->dest_type
= nir_type_unsigned
;
1632 unreachable("not reached");
1635 ir
->sampler
->accept(this);
1636 instr
->sampler
= this->deref_head
;
1638 unsigned src_number
= 0;
1640 if (ir
->coordinate
!= NULL
) {
1641 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1642 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->coordinate
);
1643 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1647 if (ir
->projector
!= NULL
) {
1648 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->projector
);
1649 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1653 if (ir
->shadow_comparitor
!= NULL
) {
1654 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->shadow_comparitor
);
1655 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1659 if (ir
->offset
!= NULL
) {
1660 /* we don't support multiple offsets yet */
1661 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1663 ir_constant
*const_offset
= ir
->offset
->as_constant();
1664 if (const_offset
!= NULL
) {
1665 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1666 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1668 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->offset
);
1669 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1676 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.bias
);
1677 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1684 if (ir
->lod_info
.lod
!= NULL
) {
1685 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.lod
);
1686 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1692 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdx
);
1693 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1695 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdy
);
1696 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1701 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.sample_index
);
1702 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1707 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1714 assert(src_number
== num_srcs
);
1716 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
1720 nir_visitor::visit(ir_constant
*ir
)
1723 * We don't know if this variable is an an array or struct that gets
1724 * dereferenced, so do the safe thing an make it a variable with a
1725 * constant initializer and return a dereference.
1728 nir_variable
*var
= ralloc(this->shader
, nir_variable
);
1729 var
->name
= ralloc_strdup(var
, "const_temp");
1730 var
->type
= ir
->type
;
1731 var
->data
.mode
= nir_var_local
;
1732 var
->data
.read_only
= true;
1733 var
->constant_initializer
= constant_copy(ir
, var
);
1734 exec_list_push_tail(&this->impl
->locals
, &var
->node
);
1736 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1737 this->deref_tail
= &this->deref_head
->deref
;
1741 nir_visitor::visit(ir_dereference_variable
*ir
)
1743 struct hash_entry
*entry
=
1744 _mesa_hash_table_search(this->var_table
, ir
->var
);
1746 nir_variable
*var
= (nir_variable
*) entry
->data
;
1748 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
1749 this->deref_head
= deref
;
1750 this->deref_tail
= &deref
->deref
;
1754 nir_visitor::visit(ir_dereference_record
*ir
)
1756 ir
->record
->accept(this);
1758 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
1759 assert(field_index
>= 0);
1761 nir_deref_struct
*deref
= nir_deref_struct_create(this->shader
, field_index
);
1762 deref
->deref
.type
= ir
->type
;
1763 this->deref_tail
->child
= &deref
->deref
;
1764 this->deref_tail
= &deref
->deref
;
1768 nir_visitor::visit(ir_dereference_array
*ir
)
1770 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
1771 deref
->deref
.type
= ir
->type
;
1773 ir_constant
*const_index
= ir
->array_index
->as_constant();
1774 if (const_index
!= NULL
) {
1775 deref
->deref_array_type
= nir_deref_array_type_direct
;
1776 deref
->base_offset
= const_index
->value
.u
[0];
1778 deref
->deref_array_type
= nir_deref_array_type_indirect
;
1779 deref
->indirect
= evaluate_rvalue(ir
->array_index
);
1782 ir
->array
->accept(this);
1784 this->deref_tail
->child
= &deref
->deref
;
1785 this->deref_tail
= &deref
->deref
;