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
;
617 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
618 op
= nir_intrinsic_image_load
;
619 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
620 op
= nir_intrinsic_image_store
;
621 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
622 op
= nir_intrinsic_image_atomic_add
;
623 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
624 op
= nir_intrinsic_image_atomic_min
;
625 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
626 op
= nir_intrinsic_image_atomic_max
;
627 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
628 op
= nir_intrinsic_image_atomic_and
;
629 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
630 op
= nir_intrinsic_image_atomic_or
;
631 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
632 op
= nir_intrinsic_image_atomic_xor
;
633 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
634 op
= nir_intrinsic_image_atomic_exchange
;
635 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
636 op
= nir_intrinsic_image_atomic_comp_swap
;
637 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
638 op
= nir_intrinsic_memory_barrier
;
640 unreachable("not reached");
643 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
646 case nir_intrinsic_atomic_counter_read_var
:
647 case nir_intrinsic_atomic_counter_inc_var
:
648 case nir_intrinsic_atomic_counter_dec_var
: {
649 ir_dereference
*param
=
650 (ir_dereference
*) ir
->actual_parameters
.get_head();
651 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
652 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
655 case nir_intrinsic_image_load
:
656 case nir_intrinsic_image_store
:
657 case nir_intrinsic_image_atomic_add
:
658 case nir_intrinsic_image_atomic_min
:
659 case nir_intrinsic_image_atomic_max
:
660 case nir_intrinsic_image_atomic_and
:
661 case nir_intrinsic_image_atomic_or
:
662 case nir_intrinsic_image_atomic_xor
:
663 case nir_intrinsic_image_atomic_exchange
:
664 case nir_intrinsic_image_atomic_comp_swap
: {
665 nir_ssa_undef_instr
*instr_undef
=
666 nir_ssa_undef_instr_create(shader
, 1);
667 nir_instr_insert_after_cf_list(this->cf_node_list
,
668 &instr_undef
->instr
);
670 /* Set the image variable dereference. */
671 exec_node
*param
= ir
->actual_parameters
.get_head();
672 ir_dereference
*image
= (ir_dereference
*)param
;
673 const glsl_type
*type
=
674 image
->variable_referenced()->type
->without_array();
676 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
677 param
= param
->get_next();
679 /* Set the address argument, extending the coordinate vector to four
682 const nir_src src_addr
= evaluate_rvalue((ir_dereference
*)param
);
683 nir_alu_instr
*instr_addr
= nir_alu_instr_create(shader
, nir_op_vec4
);
684 nir_ssa_dest_init(&instr_addr
->instr
, &instr_addr
->dest
.dest
, 4, NULL
);
686 for (int i
= 0; i
< 4; i
++) {
687 if (i
< type
->coordinate_components()) {
688 instr_addr
->src
[i
].src
= src_addr
;
689 instr_addr
->src
[i
].swizzle
[0] = i
;
691 instr_addr
->src
[i
].src
= nir_src_for_ssa(&instr_undef
->def
);
695 nir_instr_insert_after_cf_list(cf_node_list
, &instr_addr
->instr
);
696 instr
->src
[0] = nir_src_for_ssa(&instr_addr
->dest
.dest
.ssa
);
697 param
= param
->get_next();
699 /* Set the sample argument, which is undefined for single-sample
702 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
703 instr
->src
[1] = evaluate_rvalue((ir_dereference
*)param
);
704 param
= param
->get_next();
706 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
709 /* Set the intrinsic parameters. */
710 if (!param
->is_tail_sentinel()) {
711 instr
->src
[2] = evaluate_rvalue((ir_dereference
*)param
);
712 param
= param
->get_next();
715 if (!param
->is_tail_sentinel()) {
716 instr
->src
[3] = evaluate_rvalue((ir_dereference
*)param
);
717 param
= param
->get_next();
720 /* Set the intrinsic destination. */
721 if (ir
->return_deref
)
722 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
723 ir
->return_deref
->type
->vector_elements
, NULL
);
726 case nir_intrinsic_memory_barrier
:
729 unreachable("not reached");
732 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
734 if (ir
->return_deref
) {
735 nir_intrinsic_instr
*store_instr
=
736 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
737 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
739 store_instr
->variables
[0] =
740 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
741 store_instr
->src
[0] = nir_src_for_ssa(&instr
->dest
.ssa
);
743 nir_instr_insert_after_cf_list(this->cf_node_list
,
744 &store_instr
->instr
);
750 struct hash_entry
*entry
=
751 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
753 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
755 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
758 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
759 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
763 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
764 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
768 nir_visitor::visit(ir_assignment
*ir
)
770 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
772 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
773 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
774 /* We're doing a plain-as-can-be copy, so emit a copy_var */
775 nir_intrinsic_instr
*copy
=
776 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
778 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
779 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
782 nir_if
*if_stmt
= nir_if_create(this->shader
);
783 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
784 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
785 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
787 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
792 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
794 ir
->lhs
->accept(this);
795 nir_deref_var
*lhs_deref
= this->deref_head
;
796 nir_src src
= evaluate_rvalue(ir
->rhs
);
798 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
800 * We have no good way to update only part of a variable, so just load
801 * the LHS and do a vec operation to combine the old with the new, and
803 * back into the LHS. Copy propagation should get rid of the mess.
806 nir_intrinsic_instr
*load
=
807 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
808 load
->num_components
= ir
->lhs
->type
->vector_elements
;
809 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
810 load
->variables
[0] = lhs_deref
;
811 ralloc_steal(load
, load
->variables
[0]);
812 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
815 switch (ir
->lhs
->type
->vector_elements
) {
816 case 1: vec_op
= nir_op_imov
; break;
817 case 2: vec_op
= nir_op_vec2
; break;
818 case 3: vec_op
= nir_op_vec3
; break;
819 case 4: vec_op
= nir_op_vec4
; break;
820 default: unreachable("Invalid number of components"); break;
822 nir_alu_instr
*vec
= nir_alu_instr_create(this->shader
, vec_op
);
823 nir_ssa_dest_init(&vec
->instr
, &vec
->dest
.dest
, num_components
, NULL
);
824 vec
->dest
.write_mask
= (1 << num_components
) - 1;
826 unsigned component
= 0;
827 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
828 if (ir
->write_mask
& (1 << i
)) {
829 vec
->src
[i
].src
= src
;
831 /* GLSL IR will give us the input to the write-masked assignment
832 * in a single packed vector. So, for example, if the
833 * writemask is xzw, then we have to swizzle x -> x, y -> z,
834 * and z -> w and get the y component from the load.
836 vec
->src
[i
].swizzle
[0] = component
++;
838 vec
->src
[i
].src
.is_ssa
= true;
839 vec
->src
[i
].src
.ssa
= &load
->dest
.ssa
;
840 vec
->src
[i
].swizzle
[0] = i
;
844 nir_instr_insert_after_cf_list(this->cf_node_list
, &vec
->instr
);
847 src
.ssa
= &vec
->dest
.dest
.ssa
;
850 nir_intrinsic_instr
*store
=
851 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
852 store
->num_components
= ir
->lhs
->type
->vector_elements
;
853 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
854 store
->variables
[0] = nir_deref_as_var(store_deref
);
858 nir_if
*if_stmt
= nir_if_create(this->shader
);
859 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
860 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
861 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
863 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
868 * Given an instruction, returns a pointer to its destination or NULL if there
871 * Note that this only handles instructions we generate at this level.
874 get_instr_dest(nir_instr
*instr
)
876 nir_alu_instr
*alu_instr
;
877 nir_intrinsic_instr
*intrinsic_instr
;
878 nir_tex_instr
*tex_instr
;
880 switch (instr
->type
) {
881 case nir_instr_type_alu
:
882 alu_instr
= nir_instr_as_alu(instr
);
883 return &alu_instr
->dest
.dest
;
885 case nir_instr_type_intrinsic
:
886 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
887 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
888 return &intrinsic_instr
->dest
;
892 case nir_instr_type_tex
:
893 tex_instr
= nir_instr_as_tex(instr
);
894 return &tex_instr
->dest
;
897 unreachable("not reached");
904 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
906 nir_dest
*dest
= get_instr_dest(instr
);
908 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
910 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
911 this->result
= instr
;
915 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
918 if (ir
->as_dereference() || ir
->as_constant()) {
920 * A dereference is being used on the right hand side, which means we
921 * must emit a variable load.
924 nir_intrinsic_instr
*load_instr
=
925 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
926 load_instr
->num_components
= ir
->type
->vector_elements
;
927 load_instr
->variables
[0] = this->deref_head
;
928 ralloc_steal(load_instr
, load_instr
->variables
[0]);
929 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
932 nir_dest
*dest
= get_instr_dest(this->result
);
934 assert(dest
->is_ssa
);
935 nir_src src
= NIR_SRC_INIT
;
937 src
.ssa
= &dest
->ssa
;
943 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
945 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
946 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
947 instr
->src
[i
].src
= srcs
[i
];
948 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
949 add_instr(&instr
->instr
, dest_size
);
954 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
956 assert(nir_op_infos
[op
].num_inputs
== 1);
957 return emit(op
, dest_size
, &src1
);
961 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
964 assert(nir_op_infos
[op
].num_inputs
== 2);
965 nir_src srcs
[] = { src1
, src2
};
966 return emit(op
, dest_size
, srcs
);
970 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
971 nir_src src2
, nir_src src3
)
973 assert(nir_op_infos
[op
].num_inputs
== 3);
974 nir_src srcs
[] = { src1
, src2
, src3
};
975 return emit(op
, dest_size
, srcs
);
979 nir_visitor::visit(ir_expression
*ir
)
981 /* Some special cases */
982 switch (ir
->operation
) {
983 case ir_binop_ubo_load
: {
984 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
988 op
= nir_intrinsic_load_ubo
;
990 op
= nir_intrinsic_load_ubo_indirect
;
992 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
993 load
->num_components
= ir
->type
->vector_elements
;
994 load
->const_index
[0] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
995 load
->const_index
[1] = 1; /* number of vec4's */
996 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
998 load
->src
[1] = evaluate_rvalue(ir
->operands
[1]);
999 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1002 * In UBO's, a true boolean value is any non-zero value, but we consider
1003 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1006 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
1007 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
, 1);
1008 const_zero
->value
.u
[0] = 0;
1009 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
1011 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
1012 compare
->src
[0].src
.is_ssa
= true;
1013 compare
->src
[0].src
.ssa
= &load
->dest
.ssa
;
1014 compare
->src
[1].src
.is_ssa
= true;
1015 compare
->src
[1].src
.ssa
= &const_zero
->def
;
1016 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1017 compare
->src
[1].swizzle
[i
] = 0;
1018 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
1020 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
1026 case ir_unop_interpolate_at_centroid
:
1027 case ir_binop_interpolate_at_offset
:
1028 case ir_binop_interpolate_at_sample
: {
1029 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1030 ir_swizzle
*swizzle
= NULL
;
1032 /* the api does not allow a swizzle here, but the varying packing code
1033 * may have pushed one into here.
1035 swizzle
= ir
->operands
[0]->as_swizzle();
1037 deref
= swizzle
->val
->as_dereference();
1041 deref
->accept(this);
1043 nir_intrinsic_op op
;
1044 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1045 switch (ir
->operation
) {
1046 case ir_unop_interpolate_at_centroid
:
1047 op
= nir_intrinsic_interp_var_at_centroid
;
1049 case ir_binop_interpolate_at_offset
:
1050 op
= nir_intrinsic_interp_var_at_offset
;
1052 case ir_binop_interpolate_at_sample
:
1053 op
= nir_intrinsic_interp_var_at_sample
;
1056 unreachable("Invalid interpolation intrinsic");
1059 /* This case can happen if the vertex shader does not write the
1060 * given varying. In this case, the linker will lower it to a
1061 * global variable. Since interpolating a variable makes no
1062 * sense, we'll just turn it into a load which will probably
1063 * eventually end up as an SSA definition.
1065 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1066 op
= nir_intrinsic_load_var
;
1069 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1070 intrin
->num_components
= deref
->type
->vector_elements
;
1071 intrin
->variables
[0] = this->deref_head
;
1072 ralloc_steal(intrin
, intrin
->variables
[0]);
1074 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1075 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1076 intrin
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
1078 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
1081 nir_alu_instr
*mov
= nir_alu_instr_create(shader
, nir_op_imov
);
1082 mov
->dest
.write_mask
= (1 << swizzle
->type
->vector_elements
) - 1;
1083 mov
->src
[0].src
.is_ssa
= true;
1084 mov
->src
[0].src
.ssa
= &intrin
->dest
.ssa
;
1086 mov
->src
[0].swizzle
[0] = swizzle
->mask
.x
;
1087 mov
->src
[0].swizzle
[1] = swizzle
->mask
.y
;
1088 mov
->src
[0].swizzle
[2] = swizzle
->mask
.z
;
1089 mov
->src
[0].swizzle
[3] = swizzle
->mask
.w
;
1090 for (unsigned i
= deref
->type
->vector_elements
; i
< 4; i
++)
1091 mov
->src
[0].swizzle
[i
] = 0;
1093 add_instr(&mov
->instr
, swizzle
->type
->vector_elements
);
1104 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1105 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1107 glsl_base_type types
[4];
1108 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1110 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1112 types
[i
] = GLSL_TYPE_FLOAT
;
1114 glsl_base_type out_type
;
1116 out_type
= ir
->type
->base_type
;
1118 out_type
= GLSL_TYPE_FLOAT
;
1120 unsigned dest_size
= ir
->type
->vector_elements
;
1122 nir_alu_instr
*instr
;
1125 switch (ir
->operation
) {
1126 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
1127 case ir_unop_logic_not
:
1128 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1131 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fneg
: nir_op_ineg
,
1135 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fabs
: nir_op_iabs
,
1138 case ir_unop_saturate
:
1139 assert(types
[0] == GLSL_TYPE_FLOAT
);
1140 instr
= emit(nir_op_fsat
, dest_size
, srcs
);
1143 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1146 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1147 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1148 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1149 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1150 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1151 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1152 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1154 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1157 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1160 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1162 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1163 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1164 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1165 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1166 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1169 case ir_unop_bitcast_i2f
:
1170 case ir_unop_bitcast_f2i
:
1171 case ir_unop_bitcast_u2f
:
1172 case ir_unop_bitcast_f2u
:
1174 emit(nir_op_imov
, dest_size
, srcs
);
1177 switch (ir
->operands
[0]->type
->vector_elements
) {
1179 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1183 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1187 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1191 unreachable("not reached");
1194 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1195 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1196 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1197 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1198 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1199 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1200 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1201 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1202 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1203 case ir_unop_dFdx_fine
: emit(nir_op_fddx_fine
, dest_size
, srcs
); break;
1204 case ir_unop_dFdy_fine
: emit(nir_op_fddy_fine
, dest_size
, srcs
); break;
1205 case ir_unop_dFdx_coarse
: emit(nir_op_fddx_coarse
, dest_size
, srcs
); break;
1206 case ir_unop_dFdy_coarse
: emit(nir_op_fddy_coarse
, dest_size
, srcs
); break;
1207 case ir_unop_pack_snorm_2x16
:
1208 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1210 case ir_unop_pack_snorm_4x8
:
1211 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1213 case ir_unop_pack_unorm_2x16
:
1214 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1216 case ir_unop_pack_unorm_4x8
:
1217 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1219 case ir_unop_pack_half_2x16
:
1220 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1222 case ir_unop_unpack_snorm_2x16
:
1223 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1225 case ir_unop_unpack_snorm_4x8
:
1226 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1228 case ir_unop_unpack_unorm_2x16
:
1229 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1231 case ir_unop_unpack_unorm_4x8
:
1232 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1234 case ir_unop_unpack_half_2x16
:
1235 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1237 case ir_unop_unpack_half_2x16_split_x
:
1238 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1240 case ir_unop_unpack_half_2x16_split_y
:
1241 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1243 case ir_unop_bitfield_reverse
:
1244 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1246 case ir_unop_bit_count
:
1247 emit(nir_op_bit_count
, dest_size
, srcs
);
1249 case ir_unop_find_msb
:
1251 case GLSL_TYPE_UINT
:
1252 emit(nir_op_ufind_msb
, dest_size
, srcs
);
1255 emit(nir_op_ifind_msb
, dest_size
, srcs
);
1258 unreachable("Invalid type for findMSB()");
1261 case ir_unop_find_lsb
:
1262 emit(nir_op_find_lsb
, dest_size
, srcs
);
1266 switch (ir
->type
->vector_elements
) {
1268 switch (ir
->operands
[0]->type
->vector_elements
) {
1269 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1270 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1271 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1272 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1273 default: unreachable("not reached");
1277 switch (ir
->operands
[0]->type
->vector_elements
) {
1278 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1279 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1280 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1281 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1282 default: unreachable("not reached");
1286 switch (ir
->operands
[0]->type
->vector_elements
) {
1287 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1288 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1289 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1290 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1291 default: unreachable("not reached");
1295 switch (ir
->operands
[0]->type
->vector_elements
) {
1296 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1297 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1298 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1299 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1300 default: unreachable("not reached");
1304 unreachable("not reached");
1315 case ir_binop_bit_and
:
1316 case ir_binop_bit_or
:
1317 case ir_binop_bit_xor
:
1318 case ir_binop_logic_and
:
1319 case ir_binop_logic_or
:
1320 case ir_binop_logic_xor
:
1321 case ir_binop_lshift
:
1322 case ir_binop_rshift
:
1323 switch (ir
->operation
) {
1325 if (out_type
== GLSL_TYPE_FLOAT
)
1331 if (out_type
== GLSL_TYPE_FLOAT
)
1337 if (out_type
== GLSL_TYPE_FLOAT
)
1343 if (out_type
== GLSL_TYPE_FLOAT
)
1345 else if (out_type
== GLSL_TYPE_INT
)
1351 if (out_type
== GLSL_TYPE_FLOAT
)
1357 if (out_type
== GLSL_TYPE_FLOAT
)
1359 else if (out_type
== GLSL_TYPE_INT
)
1365 if (out_type
== GLSL_TYPE_FLOAT
)
1367 else if (out_type
== GLSL_TYPE_INT
)
1372 case ir_binop_bit_and
:
1375 case ir_binop_bit_or
:
1378 case ir_binop_bit_xor
:
1381 case ir_binop_logic_and
:
1387 case ir_binop_logic_or
:
1393 case ir_binop_logic_xor
:
1399 case ir_binop_lshift
:
1402 case ir_binop_rshift
:
1403 if (out_type
== GLSL_TYPE_INT
)
1413 unreachable("not reached");
1416 instr
= emit(op
, dest_size
, srcs
);
1418 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1419 ir
->operands
[1]->type
->vector_elements
== 1) {
1420 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1422 instr
->src
[1].swizzle
[i
] = 0;
1426 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1427 ir
->operands
[0]->type
->vector_elements
== 1) {
1428 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1430 instr
->src
[0].swizzle
[i
] = 0;
1435 case ir_binop_imul_high
:
1436 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1439 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1440 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1442 if (supports_ints
) {
1443 if (types
[0] == GLSL_TYPE_FLOAT
)
1444 emit(nir_op_flt
, dest_size
, srcs
);
1445 else if (types
[0] == GLSL_TYPE_INT
)
1446 emit(nir_op_ilt
, dest_size
, srcs
);
1448 emit(nir_op_ult
, dest_size
, srcs
);
1450 emit(nir_op_slt
, dest_size
, srcs
);
1453 case ir_binop_greater
:
1454 if (supports_ints
) {
1455 if (types
[0] == GLSL_TYPE_FLOAT
)
1456 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1457 else if (types
[0] == GLSL_TYPE_INT
)
1458 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1460 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1462 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1465 case ir_binop_lequal
:
1466 if (supports_ints
) {
1467 if (types
[0] == GLSL_TYPE_FLOAT
)
1468 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1469 else if (types
[0] == GLSL_TYPE_INT
)
1470 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1472 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1474 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1477 case ir_binop_gequal
:
1478 if (supports_ints
) {
1479 if (types
[0] == GLSL_TYPE_FLOAT
)
1480 emit(nir_op_fge
, dest_size
, srcs
);
1481 else if (types
[0] == GLSL_TYPE_INT
)
1482 emit(nir_op_ige
, dest_size
, srcs
);
1484 emit(nir_op_uge
, dest_size
, srcs
);
1486 emit(nir_op_slt
, dest_size
, srcs
);
1489 case ir_binop_equal
:
1490 if (supports_ints
) {
1491 if (types
[0] == GLSL_TYPE_FLOAT
)
1492 emit(nir_op_feq
, dest_size
, srcs
);
1494 emit(nir_op_ieq
, dest_size
, srcs
);
1496 emit(nir_op_seq
, dest_size
, srcs
);
1499 case ir_binop_nequal
:
1500 if (supports_ints
) {
1501 if (types
[0] == GLSL_TYPE_FLOAT
)
1502 emit(nir_op_fne
, dest_size
, srcs
);
1504 emit(nir_op_ine
, dest_size
, srcs
);
1506 emit(nir_op_sne
, dest_size
, srcs
);
1509 case ir_binop_all_equal
:
1510 if (supports_ints
) {
1511 if (types
[0] == GLSL_TYPE_FLOAT
) {
1512 switch (ir
->operands
[0]->type
->vector_elements
) {
1513 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1514 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1515 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1516 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1518 unreachable("not reached");
1521 switch (ir
->operands
[0]->type
->vector_elements
) {
1522 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1523 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1524 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1525 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1527 unreachable("not reached");
1531 switch (ir
->operands
[0]->type
->vector_elements
) {
1532 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1533 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1534 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1535 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1537 unreachable("not reached");
1541 case ir_binop_any_nequal
:
1542 if (supports_ints
) {
1543 if (types
[0] == GLSL_TYPE_FLOAT
) {
1544 switch (ir
->operands
[0]->type
->vector_elements
) {
1545 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1546 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1547 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1548 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1550 unreachable("not reached");
1553 switch (ir
->operands
[0]->type
->vector_elements
) {
1554 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1555 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1556 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1557 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1559 unreachable("not reached");
1563 switch (ir
->operands
[0]->type
->vector_elements
) {
1564 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1565 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1566 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1567 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1569 unreachable("not reached");
1574 switch (ir
->operands
[0]->type
->vector_elements
) {
1575 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1576 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1577 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1579 unreachable("not reached");
1583 case ir_binop_pack_half_2x16_split
:
1584 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1586 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1587 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1588 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1590 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1591 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1592 ir
->operands
[2]->type
->vector_elements
== 1) {
1593 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1595 instr
->src
[2].swizzle
[i
] = 0;
1601 emit(nir_op_bcsel
, dest_size
, srcs
);
1603 emit(nir_op_fcsel
, dest_size
, srcs
);
1606 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1607 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1608 instr
->src
[0].swizzle
[i
] = 0;
1611 case ir_triop_bitfield_extract
:
1612 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1613 nir_op_ubitfield_extract
, dest_size
, srcs
);
1614 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1615 instr
->src
[1].swizzle
[i
] = 0;
1616 instr
->src
[2].swizzle
[i
] = 0;
1619 case ir_quadop_bitfield_insert
:
1620 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1621 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1622 instr
->src
[2].swizzle
[i
] = 0;
1623 instr
->src
[3].swizzle
[i
] = 0;
1626 case ir_quadop_vector
:
1627 switch (ir
->type
->vector_elements
) {
1628 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1629 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1630 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1631 default: unreachable("not reached");
1636 unreachable("not reached");
1641 nir_visitor::visit(ir_swizzle
*ir
)
1643 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1644 ir
->type
->vector_elements
,
1645 evaluate_rvalue(ir
->val
));
1647 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1648 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1649 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1653 nir_visitor::visit(ir_texture
*ir
)
1660 num_srcs
= 1; /* coordinate */
1665 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1666 num_srcs
= 2; /* coordinate, bias/lod */
1670 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1676 if (ir
->lod_info
.lod
!= NULL
)
1677 num_srcs
= 2; /* coordinate, lod */
1679 num_srcs
= 1; /* coordinate */
1683 op
= nir_texop_txf_ms
;
1684 num_srcs
= 2; /* coordinate, sample_index */
1689 if (ir
->lod_info
.lod
!= NULL
)
1690 num_srcs
= 1; /* lod */
1697 num_srcs
= 1; /* coordinate */
1702 num_srcs
= 1; /* coordinate */
1705 case ir_query_levels
:
1706 op
= nir_texop_query_levels
;
1711 unreachable("not reached");
1714 if (ir
->projector
!= NULL
)
1716 if (ir
->shadow_comparitor
!= NULL
)
1718 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1721 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1724 instr
->sampler_dim
=
1725 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1726 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1727 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1728 if (instr
->is_shadow
)
1729 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1730 switch (ir
->type
->base_type
) {
1731 case GLSL_TYPE_FLOAT
:
1732 instr
->dest_type
= nir_type_float
;
1735 instr
->dest_type
= nir_type_int
;
1737 case GLSL_TYPE_UINT
:
1738 instr
->dest_type
= nir_type_unsigned
;
1741 unreachable("not reached");
1744 instr
->sampler
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1746 unsigned src_number
= 0;
1748 if (ir
->coordinate
!= NULL
) {
1749 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1750 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->coordinate
);
1751 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1755 if (ir
->projector
!= NULL
) {
1756 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->projector
);
1757 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1761 if (ir
->shadow_comparitor
!= NULL
) {
1762 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->shadow_comparitor
);
1763 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1767 if (ir
->offset
!= NULL
) {
1768 /* we don't support multiple offsets yet */
1769 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1771 ir_constant
*const_offset
= ir
->offset
->as_constant();
1772 if (const_offset
!= NULL
) {
1773 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1774 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1776 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->offset
);
1777 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1784 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.bias
);
1785 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1792 if (ir
->lod_info
.lod
!= NULL
) {
1793 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.lod
);
1794 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1800 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdx
);
1801 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1803 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdy
);
1804 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1809 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.sample_index
);
1810 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1815 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1822 assert(src_number
== num_srcs
);
1824 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
1828 nir_visitor::visit(ir_constant
*ir
)
1831 * We don't know if this variable is an an array or struct that gets
1832 * dereferenced, so do the safe thing an make it a variable with a
1833 * constant initializer and return a dereference.
1836 nir_variable
*var
= ralloc(this->shader
, nir_variable
);
1837 var
->name
= ralloc_strdup(var
, "const_temp");
1838 var
->type
= ir
->type
;
1839 var
->data
.mode
= nir_var_local
;
1840 var
->data
.read_only
= true;
1841 var
->constant_initializer
= constant_copy(ir
, var
);
1842 exec_list_push_tail(&this->impl
->locals
, &var
->node
);
1844 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1845 this->deref_tail
= &this->deref_head
->deref
;
1849 nir_visitor::visit(ir_dereference_variable
*ir
)
1851 struct hash_entry
*entry
=
1852 _mesa_hash_table_search(this->var_table
, ir
->var
);
1854 nir_variable
*var
= (nir_variable
*) entry
->data
;
1856 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
1857 this->deref_head
= deref
;
1858 this->deref_tail
= &deref
->deref
;
1862 nir_visitor::visit(ir_dereference_record
*ir
)
1864 ir
->record
->accept(this);
1866 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
1867 assert(field_index
>= 0);
1869 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
1870 deref
->deref
.type
= ir
->type
;
1871 this->deref_tail
->child
= &deref
->deref
;
1872 this->deref_tail
= &deref
->deref
;
1876 nir_visitor::visit(ir_dereference_array
*ir
)
1878 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
1879 deref
->deref
.type
= ir
->type
;
1881 ir_constant
*const_index
= ir
->array_index
->as_constant();
1882 if (const_index
!= NULL
) {
1883 deref
->deref_array_type
= nir_deref_array_type_direct
;
1884 deref
->base_offset
= const_index
->value
.u
[0];
1886 deref
->deref_array_type
= nir_deref_array_type_indirect
;
1887 deref
->indirect
= evaluate_rvalue(ir
->array_index
);
1890 ir
->array
->accept(this);
1892 this->deref_tail
->child
= &deref
->deref
;
1893 ralloc_steal(this->deref_tail
, deref
);
1894 this->deref_tail
= &deref
->deref
;