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 nir_deref_var
*evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
);
93 /* the head of the dereference chain we're creating */
94 nir_deref_var
*deref_head
;
95 /* the tail of the dereference chain we're creating */
96 nir_deref
*deref_tail
;
98 nir_variable
*var
; /* variable created by ir_variable visitor */
100 /* whether the IR we're operating on is per-function or global */
103 /* map of ir_variable -> nir_variable */
104 struct hash_table
*var_table
;
106 /* map of ir_function_signature -> nir_function_overload */
107 struct hash_table
*overload_table
;
111 * This visitor runs before the main visitor, calling create_function() for
112 * each function so that the main visitor can resolve forward references in
116 class nir_function_visitor
: public ir_hierarchical_visitor
119 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
122 virtual ir_visitor_status
visit_enter(ir_function
*);
125 nir_visitor
*visitor
;
128 }; /* end of anonymous namespace */
131 glsl_to_nir(struct gl_shader
*sh
, const nir_shader_compiler_options
*options
)
133 nir_shader
*shader
= nir_shader_create(NULL
, options
);
135 nir_visitor
v1(shader
, sh
->Stage
);
136 nir_function_visitor
v2(&v1
);
138 visit_exec_list(sh
->ir
, &v1
);
143 nir_visitor::nir_visitor(nir_shader
*shader
, gl_shader_stage stage
)
145 this->supports_ints
= shader
->options
->native_integers
;
146 this->shader
= shader
;
148 this->is_global
= true;
149 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
150 _mesa_key_pointer_equal
);
151 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
152 _mesa_key_pointer_equal
);
155 nir_visitor::~nir_visitor()
157 _mesa_hash_table_destroy(this->var_table
, NULL
);
158 _mesa_hash_table_destroy(this->overload_table
, NULL
);
162 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
165 ralloc_steal(mem_ctx
, this->deref_head
);
166 return this->deref_head
;
169 static nir_constant
*
170 constant_copy(ir_constant
*ir
, void *mem_ctx
)
175 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
177 unsigned total_elems
= ir
->type
->components();
179 switch (ir
->type
->base_type
) {
181 for (i
= 0; i
< total_elems
; i
++)
182 ret
->value
.u
[i
] = ir
->value
.u
[i
];
186 for (i
= 0; i
< total_elems
; i
++)
187 ret
->value
.i
[i
] = ir
->value
.i
[i
];
190 case GLSL_TYPE_FLOAT
:
191 for (i
= 0; i
< total_elems
; i
++)
192 ret
->value
.f
[i
] = ir
->value
.f
[i
];
196 for (i
= 0; i
< total_elems
; i
++)
197 ret
->value
.b
[i
] = ir
->value
.b
[i
];
200 case GLSL_TYPE_STRUCT
:
201 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
204 foreach_in_list(ir_constant
, field
, &ir
->components
) {
205 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
210 case GLSL_TYPE_ARRAY
:
211 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
214 for (i
= 0; i
< ir
->type
->length
; i
++)
215 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
219 unreachable("not reached");
226 nir_visitor::visit(ir_variable
*ir
)
228 nir_variable
*var
= ralloc(shader
, nir_variable
);
229 var
->type
= ir
->type
;
230 var
->name
= ralloc_strdup(var
, ir
->name
);
232 if (ir
->is_interface_instance() && ir
->get_max_ifc_array_access() != NULL
) {
233 unsigned size
= ir
->get_interface_type()->length
;
234 var
->max_ifc_array_access
= ralloc_array(var
, unsigned, size
);
235 memcpy(var
->max_ifc_array_access
, ir
->get_max_ifc_array_access(),
236 size
* sizeof(unsigned));
238 var
->max_ifc_array_access
= NULL
;
241 var
->data
.read_only
= ir
->data
.read_only
;
242 var
->data
.centroid
= ir
->data
.centroid
;
243 var
->data
.sample
= ir
->data
.sample
;
244 var
->data
.invariant
= ir
->data
.invariant
;
245 var
->data
.location
= ir
->data
.location
;
247 switch(ir
->data
.mode
) {
249 case ir_var_temporary
:
251 var
->data
.mode
= nir_var_global
;
253 var
->data
.mode
= nir_var_local
;
256 case ir_var_function_in
:
257 case ir_var_function_out
:
258 case ir_var_function_inout
:
259 case ir_var_const_in
:
260 var
->data
.mode
= nir_var_local
;
263 case ir_var_shader_in
:
264 if (stage
== MESA_SHADER_FRAGMENT
&&
265 ir
->data
.location
== VARYING_SLOT_FACE
) {
266 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
267 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
268 var
->data
.mode
= nir_var_system_value
;
270 var
->data
.mode
= nir_var_shader_in
;
274 case ir_var_shader_out
:
275 var
->data
.mode
= nir_var_shader_out
;
279 var
->data
.mode
= nir_var_uniform
;
283 case ir_var_system_value
:
284 var
->data
.mode
= nir_var_system_value
;
288 unreachable("not reached");
291 var
->data
.interpolation
= ir
->data
.interpolation
;
292 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
293 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
294 var
->data
.explicit_location
= ir
->data
.explicit_location
;
295 var
->data
.explicit_index
= ir
->data
.explicit_index
;
296 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
297 var
->data
.has_initializer
= ir
->data
.has_initializer
;
298 var
->data
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
299 var
->data
.location_frac
= ir
->data
.location_frac
;
300 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
301 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
303 switch (ir
->data
.depth_layout
) {
304 case ir_depth_layout_none
:
305 var
->data
.depth_layout
= nir_depth_layout_none
;
307 case ir_depth_layout_any
:
308 var
->data
.depth_layout
= nir_depth_layout_any
;
310 case ir_depth_layout_greater
:
311 var
->data
.depth_layout
= nir_depth_layout_greater
;
313 case ir_depth_layout_less
:
314 var
->data
.depth_layout
= nir_depth_layout_less
;
316 case ir_depth_layout_unchanged
:
317 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
320 unreachable("not reached");
323 var
->data
.index
= ir
->data
.index
;
324 var
->data
.binding
= ir
->data
.binding
;
325 /* XXX Get rid of buffer_index */
326 var
->data
.atomic
.buffer_index
= ir
->data
.binding
;
327 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
328 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
329 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
330 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
331 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
332 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
333 var
->data
.image
.format
= ir
->data
.image_format
;
334 var
->data
.max_array_access
= ir
->data
.max_array_access
;
336 var
->num_state_slots
= ir
->get_num_state_slots();
337 if (var
->num_state_slots
> 0) {
338 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
339 var
->num_state_slots
);
341 ir_state_slot
*state_slots
= ir
->get_state_slots();
342 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
343 for (unsigned j
= 0; j
< 5; j
++)
344 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
345 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
348 var
->state_slots
= NULL
;
351 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
353 var
->interface_type
= ir
->get_interface_type();
355 switch (var
->data
.mode
) {
357 exec_list_push_tail(&impl
->locals
, &var
->node
);
361 exec_list_push_tail(&shader
->globals
, &var
->node
);
364 case nir_var_shader_in
:
365 exec_list_push_tail(&shader
->inputs
, &var
->node
);
368 case nir_var_shader_out
:
369 exec_list_push_tail(&shader
->outputs
, &var
->node
);
372 case nir_var_uniform
:
373 exec_list_push_tail(&shader
->uniforms
, &var
->node
);
376 case nir_var_system_value
:
377 exec_list_push_tail(&shader
->system_values
, &var
->node
);
381 unreachable("not reached");
384 _mesa_hash_table_insert(var_table
, ir
, var
);
389 nir_function_visitor::visit_enter(ir_function
*ir
)
391 visitor
->create_function(ir
);
392 return visit_continue_with_parent
;
397 nir_visitor::create_function(ir_function
*ir
)
399 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
400 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
401 create_overload(sig
, func
);
408 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
410 if (ir
->is_intrinsic
)
413 nir_function_overload
*overload
= nir_function_overload_create(function
);
415 unsigned num_params
= ir
->parameters
.length();
416 overload
->num_params
= num_params
;
417 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
420 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
421 switch (param
->data
.mode
) {
422 case ir_var_function_in
:
423 overload
->params
[i
].param_type
= nir_parameter_in
;
426 case ir_var_function_out
:
427 overload
->params
[i
].param_type
= nir_parameter_out
;
430 case ir_var_function_inout
:
431 overload
->params
[i
].param_type
= nir_parameter_inout
;
435 unreachable("not reached");
438 overload
->params
[i
].type
= param
->type
;
442 overload
->return_type
= ir
->return_type
;
444 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
448 nir_visitor::visit(ir_function
*ir
)
450 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
455 nir_visitor::visit(ir_function_signature
*ir
)
457 if (ir
->is_intrinsic
)
460 struct hash_entry
*entry
=
461 _mesa_hash_table_search(this->overload_table
, ir
);
464 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
466 if (ir
->is_defined
) {
467 nir_function_impl
*impl
= nir_function_impl_create(overload
);
470 unsigned num_params
= overload
->num_params
;
471 impl
->num_params
= num_params
;
472 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
474 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
476 impl
->params
[i
] = this->var
;
480 if (overload
->return_type
== glsl_type::void_type
) {
481 impl
->return_var
= NULL
;
483 impl
->return_var
= ralloc(this->shader
, nir_variable
);
484 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
486 impl
->return_var
->type
= overload
->return_type
;
489 this->is_global
= false;
491 this->cf_node_list
= &impl
->body
;
492 visit_exec_list(&ir
->body
, this);
494 this->is_global
= true;
496 overload
->impl
= NULL
;
501 nir_visitor::visit(ir_loop
*ir
)
503 exec_list
*old_list
= this->cf_node_list
;
505 nir_loop
*loop
= nir_loop_create(this->shader
);
506 nir_cf_node_insert_end(old_list
, &loop
->cf_node
);
507 this->cf_node_list
= &loop
->body
;
508 visit_exec_list(&ir
->body_instructions
, this);
510 this->cf_node_list
= old_list
;
514 nir_visitor::visit(ir_if
*ir
)
516 nir_src condition
= evaluate_rvalue(ir
->condition
);
518 exec_list
*old_list
= this->cf_node_list
;
520 nir_if
*if_stmt
= nir_if_create(this->shader
);
521 if_stmt
->condition
= condition
;
522 nir_cf_node_insert_end(old_list
, &if_stmt
->cf_node
);
524 this->cf_node_list
= &if_stmt
->then_list
;
525 visit_exec_list(&ir
->then_instructions
, this);
527 this->cf_node_list
= &if_stmt
->else_list
;
528 visit_exec_list(&ir
->else_instructions
, this);
530 this->cf_node_list
= old_list
;
534 nir_visitor::visit(ir_discard
*ir
)
537 * discards aren't treated as control flow, because before we lower them
538 * they can appear anywhere in the shader and the stuff after them may still
539 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
540 * discards will be immediately followed by a return.
543 nir_intrinsic_instr
*discard
;
545 discard
= nir_intrinsic_instr_create(this->shader
,
546 nir_intrinsic_discard_if
);
547 discard
->src
[0] = evaluate_rvalue(ir
->condition
);
549 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
551 nir_instr_insert_after_cf_list(this->cf_node_list
, &discard
->instr
);
555 nir_visitor::visit(ir_emit_vertex
*ir
)
557 nir_intrinsic_instr
*instr
=
558 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
559 instr
->const_index
[0] = ir
->stream_id();
560 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
564 nir_visitor::visit(ir_end_primitive
*ir
)
566 nir_intrinsic_instr
*instr
=
567 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
568 instr
->const_index
[0] = ir
->stream_id();
569 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
573 nir_visitor::visit(ir_loop_jump
*ir
)
577 case ir_loop_jump::jump_break
:
578 type
= nir_jump_break
;
580 case ir_loop_jump::jump_continue
:
581 type
= nir_jump_continue
;
584 unreachable("not reached");
587 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
588 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
592 nir_visitor::visit(ir_return
*ir
)
594 if (ir
->value
!= NULL
) {
595 nir_intrinsic_instr
*copy
=
596 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
598 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
599 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
602 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
603 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
607 nir_visitor::visit(ir_call
*ir
)
609 if (ir
->callee
->is_intrinsic
) {
611 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
612 op
= nir_intrinsic_atomic_counter_read_var
;
613 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
614 op
= nir_intrinsic_atomic_counter_inc_var
;
615 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
616 op
= nir_intrinsic_atomic_counter_dec_var
;
618 unreachable("not reached");
621 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
622 ir_dereference
*param
=
623 (ir_dereference
*) ir
->actual_parameters
.get_head();
624 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
625 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
627 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
629 nir_intrinsic_instr
*store_instr
=
630 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
631 store_instr
->num_components
= 1;
633 store_instr
->variables
[0] = evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
634 store_instr
->src
[0].is_ssa
= true;
635 store_instr
->src
[0].ssa
= &instr
->dest
.ssa
;
637 nir_instr_insert_after_cf_list(this->cf_node_list
, &store_instr
->instr
);
642 struct hash_entry
*entry
=
643 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
645 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
647 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
650 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
651 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
655 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
656 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
660 nir_visitor::visit(ir_assignment
*ir
)
662 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
664 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
665 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
666 /* We're doing a plain-as-can-be copy, so emit a copy_var */
667 nir_intrinsic_instr
*copy
=
668 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
670 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
671 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
674 nir_if
*if_stmt
= nir_if_create(this->shader
);
675 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
676 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
677 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
679 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
684 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
686 ir
->lhs
->accept(this);
687 nir_deref_var
*lhs_deref
= this->deref_head
;
688 nir_src src
= evaluate_rvalue(ir
->rhs
);
690 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
692 * We have no good way to update only part of a variable, so just load
693 * the LHS and do a vec operation to combine the old with the new, and
695 * back into the LHS. Copy propagation should get rid of the mess.
698 nir_intrinsic_instr
*load
=
699 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
700 load
->num_components
= ir
->lhs
->type
->vector_elements
;
701 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
702 load
->variables
[0] = lhs_deref
;
703 ralloc_steal(load
, load
->variables
[0]);
704 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
707 switch (ir
->lhs
->type
->vector_elements
) {
708 case 1: vec_op
= nir_op_imov
; break;
709 case 2: vec_op
= nir_op_vec2
; break;
710 case 3: vec_op
= nir_op_vec3
; break;
711 case 4: vec_op
= nir_op_vec4
; break;
712 default: unreachable("Invalid number of components"); break;
714 nir_alu_instr
*vec
= nir_alu_instr_create(this->shader
, vec_op
);
715 nir_ssa_dest_init(&vec
->instr
, &vec
->dest
.dest
, num_components
, NULL
);
716 vec
->dest
.write_mask
= (1 << num_components
) - 1;
718 unsigned component
= 0;
719 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
720 if (ir
->write_mask
& (1 << i
)) {
721 vec
->src
[i
].src
= src
;
723 /* GLSL IR will give us the input to the write-masked assignment
724 * in a single packed vector. So, for example, if the
725 * writemask is xzw, then we have to swizzle x -> x, y -> z,
726 * and z -> w and get the y component from the load.
728 vec
->src
[i
].swizzle
[0] = component
++;
730 vec
->src
[i
].src
.is_ssa
= true;
731 vec
->src
[i
].src
.ssa
= &load
->dest
.ssa
;
732 vec
->src
[i
].swizzle
[0] = i
;
736 nir_instr_insert_after_cf_list(this->cf_node_list
, &vec
->instr
);
739 src
.ssa
= &vec
->dest
.dest
.ssa
;
742 nir_intrinsic_instr
*store
=
743 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
744 store
->num_components
= ir
->lhs
->type
->vector_elements
;
745 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
746 store
->variables
[0] = nir_deref_as_var(store_deref
);
750 nir_if
*if_stmt
= nir_if_create(this->shader
);
751 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
752 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
753 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
755 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
760 * Given an instruction, returns a pointer to its destination or NULL if there
763 * Note that this only handles instructions we generate at this level.
766 get_instr_dest(nir_instr
*instr
)
768 nir_alu_instr
*alu_instr
;
769 nir_intrinsic_instr
*intrinsic_instr
;
770 nir_tex_instr
*tex_instr
;
772 switch (instr
->type
) {
773 case nir_instr_type_alu
:
774 alu_instr
= nir_instr_as_alu(instr
);
775 return &alu_instr
->dest
.dest
;
777 case nir_instr_type_intrinsic
:
778 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
779 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
780 return &intrinsic_instr
->dest
;
784 case nir_instr_type_tex
:
785 tex_instr
= nir_instr_as_tex(instr
);
786 return &tex_instr
->dest
;
789 unreachable("not reached");
796 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
798 nir_dest
*dest
= get_instr_dest(instr
);
800 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
802 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
803 this->result
= instr
;
807 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
810 if (ir
->as_dereference() || ir
->as_constant()) {
812 * A dereference is being used on the right hand side, which means we
813 * must emit a variable load.
816 nir_intrinsic_instr
*load_instr
=
817 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
818 load_instr
->num_components
= ir
->type
->vector_elements
;
819 load_instr
->variables
[0] = this->deref_head
;
820 ralloc_steal(load_instr
, load_instr
->variables
[0]);
821 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
824 nir_dest
*dest
= get_instr_dest(this->result
);
826 assert(dest
->is_ssa
);
829 src
.ssa
= &dest
->ssa
;
835 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
837 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
838 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
839 instr
->src
[i
].src
= srcs
[i
];
840 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
841 add_instr(&instr
->instr
, dest_size
);
846 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
848 assert(nir_op_infos
[op
].num_inputs
== 1);
849 return emit(op
, dest_size
, &src1
);
853 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
856 assert(nir_op_infos
[op
].num_inputs
== 2);
857 nir_src srcs
[] = { src1
, src2
};
858 return emit(op
, dest_size
, srcs
);
862 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
863 nir_src src2
, nir_src src3
)
865 assert(nir_op_infos
[op
].num_inputs
== 3);
866 nir_src srcs
[] = { src1
, src2
, src3
};
867 return emit(op
, dest_size
, srcs
);
871 nir_visitor::visit(ir_expression
*ir
)
873 /* Some special cases */
874 switch (ir
->operation
) {
875 case ir_binop_ubo_load
: {
876 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
880 op
= nir_intrinsic_load_ubo
;
882 op
= nir_intrinsic_load_ubo_indirect
;
884 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
885 load
->num_components
= ir
->type
->vector_elements
;
886 load
->const_index
[0] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
887 load
->const_index
[1] = 1; /* number of vec4's */
888 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
890 load
->src
[1] = evaluate_rvalue(ir
->operands
[1]);
891 add_instr(&load
->instr
, ir
->type
->vector_elements
);
894 * In UBO's, a true boolean value is any non-zero value, but we consider
895 * a true boolean to be ~0. Fix this up with a != 0 comparison.
898 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
899 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
, 1);
900 const_zero
->value
.u
[0] = 0;
901 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
903 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
904 compare
->src
[0].src
.is_ssa
= true;
905 compare
->src
[0].src
.ssa
= &load
->dest
.ssa
;
906 compare
->src
[1].src
.is_ssa
= true;
907 compare
->src
[1].src
.ssa
= &const_zero
->def
;
908 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
909 compare
->src
[1].swizzle
[i
] = 0;
910 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
912 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
918 case ir_unop_interpolate_at_centroid
:
919 case ir_binop_interpolate_at_offset
:
920 case ir_binop_interpolate_at_sample
: {
921 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
922 ir_swizzle
*swizzle
= NULL
;
924 /* the api does not allow a swizzle here, but the varying packing code
925 * may have pushed one into here.
927 swizzle
= ir
->operands
[0]->as_swizzle();
929 deref
= swizzle
->val
->as_dereference();
936 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
937 switch (ir
->operation
) {
938 case ir_unop_interpolate_at_centroid
:
939 op
= nir_intrinsic_interp_var_at_centroid
;
941 case ir_binop_interpolate_at_offset
:
942 op
= nir_intrinsic_interp_var_at_offset
;
944 case ir_binop_interpolate_at_sample
:
945 op
= nir_intrinsic_interp_var_at_sample
;
948 unreachable("Invalid interpolation intrinsic");
951 /* This case can happen if the vertex shader does not write the
952 * given varying. In this case, the linker will lower it to a
953 * global variable. Since interpolating a variable makes no
954 * sense, we'll just turn it into a load which will probably
955 * eventually end up as an SSA definition.
957 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
958 op
= nir_intrinsic_load_var
;
961 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
962 intrin
->num_components
= deref
->type
->vector_elements
;
963 intrin
->variables
[0] = this->deref_head
;
964 ralloc_steal(intrin
, intrin
->variables
[0]);
966 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
967 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
968 intrin
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
970 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
973 nir_alu_instr
*mov
= nir_alu_instr_create(shader
, nir_op_imov
);
974 mov
->dest
.write_mask
= (1 << swizzle
->type
->vector_elements
) - 1;
975 mov
->src
[0].src
.is_ssa
= true;
976 mov
->src
[0].src
.ssa
= &intrin
->dest
.ssa
;
978 mov
->src
[0].swizzle
[0] = swizzle
->mask
.x
;
979 mov
->src
[0].swizzle
[1] = swizzle
->mask
.y
;
980 mov
->src
[0].swizzle
[2] = swizzle
->mask
.z
;
981 mov
->src
[0].swizzle
[3] = swizzle
->mask
.w
;
982 for (unsigned i
= deref
->type
->vector_elements
; i
< 4; i
++)
983 mov
->src
[0].swizzle
[i
] = 0;
985 add_instr(&mov
->instr
, swizzle
->type
->vector_elements
);
996 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
997 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
999 glsl_base_type types
[4];
1000 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1002 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1004 types
[i
] = GLSL_TYPE_FLOAT
;
1006 glsl_base_type out_type
;
1008 out_type
= ir
->type
->base_type
;
1010 out_type
= GLSL_TYPE_FLOAT
;
1012 unsigned dest_size
= ir
->type
->vector_elements
;
1014 nir_alu_instr
*instr
;
1017 switch (ir
->operation
) {
1018 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
1019 case ir_unop_logic_not
:
1020 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1023 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fneg
: nir_op_ineg
,
1027 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fabs
: nir_op_iabs
,
1030 case ir_unop_saturate
:
1031 assert(types
[0] == GLSL_TYPE_FLOAT
);
1032 instr
= emit(nir_op_fsat
, dest_size
, srcs
);
1035 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1038 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1039 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1040 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1041 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1042 case ir_unop_log
: emit(nir_op_flog
, dest_size
, srcs
); break;
1043 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1044 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1046 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1049 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1052 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1054 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1055 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1056 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1057 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1058 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1061 case ir_unop_bitcast_i2f
:
1062 case ir_unop_bitcast_f2i
:
1063 case ir_unop_bitcast_u2f
:
1064 case ir_unop_bitcast_f2u
:
1066 emit(nir_op_imov
, dest_size
, srcs
);
1069 switch (ir
->operands
[0]->type
->vector_elements
) {
1071 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1075 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1079 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1083 unreachable("not reached");
1086 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1087 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1088 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1089 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1090 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1091 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1092 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1093 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1094 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1095 case ir_unop_dFdx_fine
: emit(nir_op_fddx_fine
, dest_size
, srcs
); break;
1096 case ir_unop_dFdy_fine
: emit(nir_op_fddy_fine
, dest_size
, srcs
); break;
1097 case ir_unop_dFdx_coarse
: emit(nir_op_fddx_coarse
, dest_size
, srcs
); break;
1098 case ir_unop_dFdy_coarse
: emit(nir_op_fddy_coarse
, dest_size
, srcs
); break;
1099 case ir_unop_pack_snorm_2x16
:
1100 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1102 case ir_unop_pack_snorm_4x8
:
1103 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1105 case ir_unop_pack_unorm_2x16
:
1106 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1108 case ir_unop_pack_unorm_4x8
:
1109 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1111 case ir_unop_pack_half_2x16
:
1112 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1114 case ir_unop_unpack_snorm_2x16
:
1115 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1117 case ir_unop_unpack_snorm_4x8
:
1118 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1120 case ir_unop_unpack_unorm_2x16
:
1121 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1123 case ir_unop_unpack_unorm_4x8
:
1124 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1126 case ir_unop_unpack_half_2x16
:
1127 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1129 case ir_unop_unpack_half_2x16_split_x
:
1130 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1132 case ir_unop_unpack_half_2x16_split_y
:
1133 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1135 case ir_unop_bitfield_reverse
:
1136 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1138 case ir_unop_bit_count
:
1139 emit(nir_op_bit_count
, dest_size
, srcs
);
1141 case ir_unop_find_msb
:
1143 case GLSL_TYPE_UINT
:
1144 emit(nir_op_ufind_msb
, dest_size
, srcs
);
1147 emit(nir_op_ifind_msb
, dest_size
, srcs
);
1150 unreachable("Invalid type for findMSB()");
1153 case ir_unop_find_lsb
:
1154 emit(nir_op_find_lsb
, dest_size
, srcs
);
1158 switch (ir
->type
->vector_elements
) {
1160 switch (ir
->operands
[0]->type
->vector_elements
) {
1161 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1162 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1163 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1164 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1165 default: unreachable("not reached");
1169 switch (ir
->operands
[0]->type
->vector_elements
) {
1170 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1171 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1172 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1173 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1174 default: unreachable("not reached");
1178 switch (ir
->operands
[0]->type
->vector_elements
) {
1179 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1180 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1181 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1182 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1183 default: unreachable("not reached");
1187 switch (ir
->operands
[0]->type
->vector_elements
) {
1188 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1189 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1190 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1191 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1192 default: unreachable("not reached");
1196 unreachable("not reached");
1207 case ir_binop_bit_and
:
1208 case ir_binop_bit_or
:
1209 case ir_binop_bit_xor
:
1210 case ir_binop_logic_and
:
1211 case ir_binop_logic_or
:
1212 case ir_binop_logic_xor
:
1213 case ir_binop_lshift
:
1214 case ir_binop_rshift
:
1215 switch (ir
->operation
) {
1217 if (out_type
== GLSL_TYPE_FLOAT
)
1223 if (out_type
== GLSL_TYPE_FLOAT
)
1229 if (out_type
== GLSL_TYPE_FLOAT
)
1235 if (out_type
== GLSL_TYPE_FLOAT
)
1237 else if (out_type
== GLSL_TYPE_INT
)
1243 if (out_type
== GLSL_TYPE_FLOAT
)
1249 if (out_type
== GLSL_TYPE_FLOAT
)
1251 else if (out_type
== GLSL_TYPE_INT
)
1257 if (out_type
== GLSL_TYPE_FLOAT
)
1259 else if (out_type
== GLSL_TYPE_INT
)
1264 case ir_binop_bit_and
:
1267 case ir_binop_bit_or
:
1270 case ir_binop_bit_xor
:
1273 case ir_binop_logic_and
:
1279 case ir_binop_logic_or
:
1285 case ir_binop_logic_xor
:
1291 case ir_binop_lshift
:
1294 case ir_binop_rshift
:
1295 if (out_type
== GLSL_TYPE_INT
)
1305 unreachable("not reached");
1308 instr
= emit(op
, dest_size
, srcs
);
1310 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1311 ir
->operands
[1]->type
->vector_elements
== 1) {
1312 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1314 instr
->src
[1].swizzle
[i
] = 0;
1318 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1319 ir
->operands
[0]->type
->vector_elements
== 1) {
1320 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1322 instr
->src
[0].swizzle
[i
] = 0;
1327 case ir_binop_imul_high
:
1328 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1331 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1332 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1334 if (supports_ints
) {
1335 if (types
[0] == GLSL_TYPE_FLOAT
)
1336 emit(nir_op_flt
, dest_size
, srcs
);
1337 else if (types
[0] == GLSL_TYPE_INT
)
1338 emit(nir_op_ilt
, dest_size
, srcs
);
1340 emit(nir_op_ult
, dest_size
, srcs
);
1342 emit(nir_op_slt
, dest_size
, srcs
);
1345 case ir_binop_greater
:
1346 if (supports_ints
) {
1347 if (types
[0] == GLSL_TYPE_FLOAT
)
1348 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1349 else if (types
[0] == GLSL_TYPE_INT
)
1350 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1352 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1354 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1357 case ir_binop_lequal
:
1358 if (supports_ints
) {
1359 if (types
[0] == GLSL_TYPE_FLOAT
)
1360 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1361 else if (types
[0] == GLSL_TYPE_INT
)
1362 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1364 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1366 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1369 case ir_binop_gequal
:
1370 if (supports_ints
) {
1371 if (types
[0] == GLSL_TYPE_FLOAT
)
1372 emit(nir_op_fge
, dest_size
, srcs
);
1373 else if (types
[0] == GLSL_TYPE_INT
)
1374 emit(nir_op_ige
, dest_size
, srcs
);
1376 emit(nir_op_uge
, dest_size
, srcs
);
1378 emit(nir_op_slt
, dest_size
, srcs
);
1381 case ir_binop_equal
:
1382 if (supports_ints
) {
1383 if (types
[0] == GLSL_TYPE_FLOAT
)
1384 emit(nir_op_feq
, dest_size
, srcs
);
1386 emit(nir_op_ieq
, dest_size
, srcs
);
1388 emit(nir_op_seq
, dest_size
, srcs
);
1391 case ir_binop_nequal
:
1392 if (supports_ints
) {
1393 if (types
[0] == GLSL_TYPE_FLOAT
)
1394 emit(nir_op_fne
, dest_size
, srcs
);
1396 emit(nir_op_ine
, dest_size
, srcs
);
1398 emit(nir_op_sne
, dest_size
, srcs
);
1401 case ir_binop_all_equal
:
1402 if (supports_ints
) {
1403 if (types
[0] == GLSL_TYPE_FLOAT
) {
1404 switch (ir
->operands
[0]->type
->vector_elements
) {
1405 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1406 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1407 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1408 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1410 unreachable("not reached");
1413 switch (ir
->operands
[0]->type
->vector_elements
) {
1414 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1415 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1416 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1417 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1419 unreachable("not reached");
1423 switch (ir
->operands
[0]->type
->vector_elements
) {
1424 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1425 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1426 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1427 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1429 unreachable("not reached");
1433 case ir_binop_any_nequal
:
1434 if (supports_ints
) {
1435 if (types
[0] == GLSL_TYPE_FLOAT
) {
1436 switch (ir
->operands
[0]->type
->vector_elements
) {
1437 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1438 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1439 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1440 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1442 unreachable("not reached");
1445 switch (ir
->operands
[0]->type
->vector_elements
) {
1446 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1447 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1448 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1449 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1451 unreachable("not reached");
1455 switch (ir
->operands
[0]->type
->vector_elements
) {
1456 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1457 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1458 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1459 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1461 unreachable("not reached");
1466 switch (ir
->operands
[0]->type
->vector_elements
) {
1467 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1468 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1469 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1471 unreachable("not reached");
1475 case ir_binop_pack_half_2x16_split
:
1476 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1478 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1479 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1480 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1482 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1483 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1484 ir
->operands
[2]->type
->vector_elements
== 1) {
1485 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1487 instr
->src
[2].swizzle
[i
] = 0;
1493 emit(nir_op_bcsel
, dest_size
, srcs
);
1495 emit(nir_op_fcsel
, dest_size
, srcs
);
1498 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1499 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1500 instr
->src
[0].swizzle
[i
] = 0;
1503 case ir_triop_bitfield_extract
:
1504 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1505 nir_op_ubitfield_extract
, dest_size
, srcs
);
1506 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1507 instr
->src
[1].swizzle
[i
] = 0;
1508 instr
->src
[2].swizzle
[i
] = 0;
1511 case ir_quadop_bitfield_insert
:
1512 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1513 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1514 instr
->src
[2].swizzle
[i
] = 0;
1515 instr
->src
[3].swizzle
[i
] = 0;
1518 case ir_quadop_vector
:
1519 switch (ir
->type
->vector_elements
) {
1520 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1521 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1522 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1523 default: unreachable("not reached");
1528 unreachable("not reached");
1533 nir_visitor::visit(ir_swizzle
*ir
)
1535 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1536 ir
->type
->vector_elements
,
1537 evaluate_rvalue(ir
->val
));
1539 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1540 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1541 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1545 nir_visitor::visit(ir_texture
*ir
)
1552 num_srcs
= 1; /* coordinate */
1557 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1558 num_srcs
= 2; /* coordinate, bias/lod */
1562 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1568 if (ir
->lod_info
.lod
!= NULL
)
1569 num_srcs
= 2; /* coordinate, lod */
1571 num_srcs
= 1; /* coordinate */
1575 op
= nir_texop_txf_ms
;
1576 num_srcs
= 2; /* coordinate, sample_index */
1581 if (ir
->lod_info
.lod
!= NULL
)
1582 num_srcs
= 1; /* lod */
1589 num_srcs
= 1; /* coordinate */
1594 num_srcs
= 1; /* coordinate */
1597 case ir_query_levels
:
1598 op
= nir_texop_query_levels
;
1603 unreachable("not reached");
1606 if (ir
->projector
!= NULL
)
1608 if (ir
->shadow_comparitor
!= NULL
)
1610 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1613 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1616 instr
->sampler_dim
=
1617 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1618 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1619 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1620 if (instr
->is_shadow
)
1621 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1622 switch (ir
->type
->base_type
) {
1623 case GLSL_TYPE_FLOAT
:
1624 instr
->dest_type
= nir_type_float
;
1627 instr
->dest_type
= nir_type_int
;
1629 case GLSL_TYPE_UINT
:
1630 instr
->dest_type
= nir_type_unsigned
;
1633 unreachable("not reached");
1636 instr
->sampler
= evaluate_deref(&instr
->instr
, ir
->sampler
);
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->deref_tail
, 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 ralloc_steal(this->deref_tail
, deref
);
1786 this->deref_tail
= &deref
->deref
;