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 "nir_control_flow.h"
30 #include "ir_visitor.h"
31 #include "ir_hierarchical_visitor.h"
35 * pass to lower GLSL IR to NIR
37 * This will lower variable dereferences to loads/stores of corresponding
38 * variables in NIR - the variables will be converted to registers in a later
44 class nir_visitor
: public ir_visitor
47 nir_visitor(nir_shader
*shader
);
50 virtual void visit(ir_variable
*);
51 virtual void visit(ir_function
*);
52 virtual void visit(ir_function_signature
*);
53 virtual void visit(ir_loop
*);
54 virtual void visit(ir_if
*);
55 virtual void visit(ir_discard
*);
56 virtual void visit(ir_loop_jump
*);
57 virtual void visit(ir_return
*);
58 virtual void visit(ir_call
*);
59 virtual void visit(ir_assignment
*);
60 virtual void visit(ir_emit_vertex
*);
61 virtual void visit(ir_end_primitive
*);
62 virtual void visit(ir_expression
*);
63 virtual void visit(ir_swizzle
*);
64 virtual void visit(ir_texture
*);
65 virtual void visit(ir_constant
*);
66 virtual void visit(ir_dereference_variable
*);
67 virtual void visit(ir_dereference_record
*);
68 virtual void visit(ir_dereference_array
*);
69 virtual void visit(ir_barrier
*);
71 void create_function(ir_function
*ir
);
74 void create_overload(ir_function_signature
*ir
, nir_function
*function
);
75 void add_instr(nir_instr
*instr
, unsigned num_components
);
76 nir_src
evaluate_rvalue(ir_rvalue
*ir
);
78 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
);
79 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
,
82 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
,
83 nir_src src2
, nir_src src3
);
88 nir_function_impl
*impl
;
89 exec_list
*cf_node_list
;
90 nir_instr
*result
; /* result of the expression tree last visited */
92 nir_deref_var
*evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
);
94 /* the head of the dereference chain we're creating */
95 nir_deref_var
*deref_head
;
96 /* the tail of the dereference chain we're creating */
97 nir_deref
*deref_tail
;
99 nir_variable
*var
; /* variable created by ir_variable visitor */
101 /* whether the IR we're operating on is per-function or global */
104 /* map of ir_variable -> nir_variable */
105 struct hash_table
*var_table
;
107 /* map of ir_function_signature -> nir_function_overload */
108 struct hash_table
*overload_table
;
112 * This visitor runs before the main visitor, calling create_function() for
113 * each function so that the main visitor can resolve forward references in
117 class nir_function_visitor
: public ir_hierarchical_visitor
120 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
123 virtual ir_visitor_status
visit_enter(ir_function
*);
126 nir_visitor
*visitor
;
129 }; /* end of anonymous namespace */
132 glsl_to_nir(struct gl_shader
*sh
, const nir_shader_compiler_options
*options
)
134 nir_shader
*shader
= nir_shader_create(NULL
, sh
->Stage
, options
);
136 nir_visitor
v1(shader
);
137 nir_function_visitor
v2(&v1
);
139 visit_exec_list(sh
->ir
, &v1
);
141 nir_lower_outputs_to_temporaries(shader
);
143 shader
->gs
.vertices_out
= sh
->Geom
.VerticesOut
;
144 shader
->gs
.invocations
= sh
->Geom
.Invocations
;
149 nir_visitor::nir_visitor(nir_shader
*shader
)
151 this->supports_ints
= shader
->options
->native_integers
;
152 this->shader
= shader
;
153 this->is_global
= true;
154 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
155 _mesa_key_pointer_equal
);
156 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
157 _mesa_key_pointer_equal
);
160 nir_visitor::~nir_visitor()
162 _mesa_hash_table_destroy(this->var_table
, NULL
);
163 _mesa_hash_table_destroy(this->overload_table
, NULL
);
167 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
170 ralloc_steal(mem_ctx
, this->deref_head
);
171 return this->deref_head
;
174 static nir_constant
*
175 constant_copy(ir_constant
*ir
, void *mem_ctx
)
180 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
182 unsigned total_elems
= ir
->type
->components();
184 switch (ir
->type
->base_type
) {
186 for (i
= 0; i
< total_elems
; i
++)
187 ret
->value
.u
[i
] = ir
->value
.u
[i
];
191 for (i
= 0; i
< total_elems
; i
++)
192 ret
->value
.i
[i
] = ir
->value
.i
[i
];
195 case GLSL_TYPE_FLOAT
:
196 for (i
= 0; i
< total_elems
; i
++)
197 ret
->value
.f
[i
] = ir
->value
.f
[i
];
201 for (i
= 0; i
< total_elems
; i
++)
202 ret
->value
.b
[i
] = ir
->value
.b
[i
];
205 case GLSL_TYPE_STRUCT
:
206 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
209 foreach_in_list(ir_constant
, field
, &ir
->components
) {
210 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
215 case GLSL_TYPE_ARRAY
:
216 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
219 for (i
= 0; i
< ir
->type
->length
; i
++)
220 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
224 unreachable("not reached");
231 nir_visitor::visit(ir_variable
*ir
)
233 nir_variable
*var
= ralloc(shader
, nir_variable
);
234 var
->type
= ir
->type
;
235 var
->name
= ralloc_strdup(var
, ir
->name
);
237 if (ir
->is_interface_instance() && ir
->get_max_ifc_array_access() != NULL
) {
238 unsigned size
= ir
->get_interface_type()->length
;
239 var
->max_ifc_array_access
= ralloc_array(var
, unsigned, size
);
240 memcpy(var
->max_ifc_array_access
, ir
->get_max_ifc_array_access(),
241 size
* sizeof(unsigned));
243 var
->max_ifc_array_access
= NULL
;
246 var
->data
.read_only
= ir
->data
.read_only
;
247 var
->data
.centroid
= ir
->data
.centroid
;
248 var
->data
.sample
= ir
->data
.sample
;
249 var
->data
.invariant
= ir
->data
.invariant
;
250 var
->data
.location
= ir
->data
.location
;
252 switch(ir
->data
.mode
) {
254 case ir_var_temporary
:
256 var
->data
.mode
= nir_var_global
;
258 var
->data
.mode
= nir_var_local
;
261 case ir_var_function_in
:
262 case ir_var_function_out
:
263 case ir_var_function_inout
:
264 case ir_var_const_in
:
265 var
->data
.mode
= nir_var_local
;
268 case ir_var_shader_in
:
269 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
270 ir
->data
.location
== VARYING_SLOT_FACE
) {
271 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
272 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
273 var
->data
.mode
= nir_var_system_value
;
275 var
->data
.mode
= nir_var_shader_in
;
279 case ir_var_shader_out
:
280 var
->data
.mode
= nir_var_shader_out
;
284 var
->data
.mode
= nir_var_uniform
;
287 case ir_var_shader_storage
:
288 var
->data
.mode
= nir_var_shader_storage
;
291 case ir_var_system_value
:
292 var
->data
.mode
= nir_var_system_value
;
296 unreachable("not reached");
299 var
->data
.interpolation
= ir
->data
.interpolation
;
300 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
301 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
302 var
->data
.explicit_location
= ir
->data
.explicit_location
;
303 var
->data
.explicit_index
= ir
->data
.explicit_index
;
304 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
305 var
->data
.has_initializer
= ir
->data
.has_initializer
;
306 var
->data
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
307 var
->data
.location_frac
= ir
->data
.location_frac
;
308 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
309 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
311 switch (ir
->data
.depth_layout
) {
312 case ir_depth_layout_none
:
313 var
->data
.depth_layout
= nir_depth_layout_none
;
315 case ir_depth_layout_any
:
316 var
->data
.depth_layout
= nir_depth_layout_any
;
318 case ir_depth_layout_greater
:
319 var
->data
.depth_layout
= nir_depth_layout_greater
;
321 case ir_depth_layout_less
:
322 var
->data
.depth_layout
= nir_depth_layout_less
;
324 case ir_depth_layout_unchanged
:
325 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
328 unreachable("not reached");
331 var
->data
.index
= ir
->data
.index
;
332 var
->data
.binding
= ir
->data
.binding
;
333 /* XXX Get rid of buffer_index */
334 var
->data
.atomic
.buffer_index
= ir
->data
.binding
;
335 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
336 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
337 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
338 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
339 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
340 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
341 var
->data
.image
.format
= ir
->data
.image_format
;
342 var
->data
.max_array_access
= ir
->data
.max_array_access
;
344 var
->num_state_slots
= ir
->get_num_state_slots();
345 if (var
->num_state_slots
> 0) {
346 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
347 var
->num_state_slots
);
349 ir_state_slot
*state_slots
= ir
->get_state_slots();
350 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
351 for (unsigned j
= 0; j
< 5; j
++)
352 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
353 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
356 var
->state_slots
= NULL
;
359 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
361 var
->interface_type
= ir
->get_interface_type();
363 switch (var
->data
.mode
) {
365 exec_list_push_tail(&impl
->locals
, &var
->node
);
369 exec_list_push_tail(&shader
->globals
, &var
->node
);
372 case nir_var_shader_in
:
373 exec_list_push_tail(&shader
->inputs
, &var
->node
);
376 case nir_var_shader_out
:
377 exec_list_push_tail(&shader
->outputs
, &var
->node
);
380 case nir_var_uniform
:
381 case nir_var_shader_storage
:
382 exec_list_push_tail(&shader
->uniforms
, &var
->node
);
385 case nir_var_system_value
:
386 exec_list_push_tail(&shader
->system_values
, &var
->node
);
390 unreachable("not reached");
393 _mesa_hash_table_insert(var_table
, ir
, var
);
398 nir_function_visitor::visit_enter(ir_function
*ir
)
400 visitor
->create_function(ir
);
401 return visit_continue_with_parent
;
406 nir_visitor::create_function(ir_function
*ir
)
408 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
409 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
410 create_overload(sig
, func
);
417 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
419 if (ir
->is_intrinsic
)
422 nir_function_overload
*overload
= nir_function_overload_create(function
);
424 unsigned num_params
= ir
->parameters
.length();
425 overload
->num_params
= num_params
;
426 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
429 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
430 switch (param
->data
.mode
) {
431 case ir_var_function_in
:
432 overload
->params
[i
].param_type
= nir_parameter_in
;
435 case ir_var_function_out
:
436 overload
->params
[i
].param_type
= nir_parameter_out
;
439 case ir_var_function_inout
:
440 overload
->params
[i
].param_type
= nir_parameter_inout
;
444 unreachable("not reached");
447 overload
->params
[i
].type
= param
->type
;
451 overload
->return_type
= ir
->return_type
;
453 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
457 nir_visitor::visit(ir_function
*ir
)
459 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
464 nir_visitor::visit(ir_function_signature
*ir
)
466 if (ir
->is_intrinsic
)
469 struct hash_entry
*entry
=
470 _mesa_hash_table_search(this->overload_table
, ir
);
473 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
475 if (ir
->is_defined
) {
476 nir_function_impl
*impl
= nir_function_impl_create(overload
);
479 unsigned num_params
= overload
->num_params
;
480 impl
->num_params
= num_params
;
481 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
483 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
485 impl
->params
[i
] = this->var
;
489 if (overload
->return_type
== glsl_type::void_type
) {
490 impl
->return_var
= NULL
;
492 impl
->return_var
= ralloc(this->shader
, nir_variable
);
493 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
495 impl
->return_var
->type
= overload
->return_type
;
498 this->is_global
= false;
500 this->cf_node_list
= &impl
->body
;
501 visit_exec_list(&ir
->body
, this);
503 this->is_global
= true;
505 overload
->impl
= NULL
;
510 nir_visitor::visit(ir_loop
*ir
)
512 exec_list
*old_list
= this->cf_node_list
;
514 nir_loop
*loop
= nir_loop_create(this->shader
);
515 nir_cf_node_insert_end(old_list
, &loop
->cf_node
);
516 this->cf_node_list
= &loop
->body
;
517 visit_exec_list(&ir
->body_instructions
, this);
519 this->cf_node_list
= old_list
;
523 nir_visitor::visit(ir_if
*ir
)
525 nir_src condition
= evaluate_rvalue(ir
->condition
);
527 exec_list
*old_list
= this->cf_node_list
;
529 nir_if
*if_stmt
= nir_if_create(this->shader
);
530 if_stmt
->condition
= condition
;
531 nir_cf_node_insert_end(old_list
, &if_stmt
->cf_node
);
533 this->cf_node_list
= &if_stmt
->then_list
;
534 visit_exec_list(&ir
->then_instructions
, this);
536 this->cf_node_list
= &if_stmt
->else_list
;
537 visit_exec_list(&ir
->else_instructions
, this);
539 this->cf_node_list
= old_list
;
543 nir_visitor::visit(ir_discard
*ir
)
546 * discards aren't treated as control flow, because before we lower them
547 * they can appear anywhere in the shader and the stuff after them may still
548 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
549 * discards will be immediately followed by a return.
552 nir_intrinsic_instr
*discard
;
554 discard
= nir_intrinsic_instr_create(this->shader
,
555 nir_intrinsic_discard_if
);
556 discard
->src
[0] = evaluate_rvalue(ir
->condition
);
558 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
560 nir_instr_insert_after_cf_list(this->cf_node_list
, &discard
->instr
);
564 nir_visitor::visit(ir_emit_vertex
*ir
)
566 nir_intrinsic_instr
*instr
=
567 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
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_end_primitive
*ir
)
575 nir_intrinsic_instr
*instr
=
576 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
577 instr
->const_index
[0] = ir
->stream_id();
578 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
582 nir_visitor::visit(ir_loop_jump
*ir
)
586 case ir_loop_jump::jump_break
:
587 type
= nir_jump_break
;
589 case ir_loop_jump::jump_continue
:
590 type
= nir_jump_continue
;
593 unreachable("not reached");
596 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
597 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
601 nir_visitor::visit(ir_return
*ir
)
603 if (ir
->value
!= NULL
) {
604 nir_intrinsic_instr
*copy
=
605 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
607 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
608 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
611 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
612 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
616 nir_visitor::visit(ir_call
*ir
)
618 if (ir
->callee
->is_intrinsic
) {
620 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
621 op
= nir_intrinsic_atomic_counter_read_var
;
622 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
623 op
= nir_intrinsic_atomic_counter_inc_var
;
624 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
625 op
= nir_intrinsic_atomic_counter_dec_var
;
626 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
627 op
= nir_intrinsic_image_load
;
628 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
629 op
= nir_intrinsic_image_store
;
630 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
631 op
= nir_intrinsic_image_atomic_add
;
632 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
633 op
= nir_intrinsic_image_atomic_min
;
634 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
635 op
= nir_intrinsic_image_atomic_max
;
636 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
637 op
= nir_intrinsic_image_atomic_and
;
638 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
639 op
= nir_intrinsic_image_atomic_or
;
640 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
641 op
= nir_intrinsic_image_atomic_xor
;
642 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
643 op
= nir_intrinsic_image_atomic_exchange
;
644 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
645 op
= nir_intrinsic_image_atomic_comp_swap
;
646 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
647 op
= nir_intrinsic_memory_barrier
;
648 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_size") == 0) {
649 op
= nir_intrinsic_image_size
;
650 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_samples") == 0) {
651 op
= nir_intrinsic_image_samples
;
653 unreachable("not reached");
656 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
659 case nir_intrinsic_atomic_counter_read_var
:
660 case nir_intrinsic_atomic_counter_inc_var
:
661 case nir_intrinsic_atomic_counter_dec_var
: {
662 ir_dereference
*param
=
663 (ir_dereference
*) ir
->actual_parameters
.get_head();
664 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
665 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
668 case nir_intrinsic_image_load
:
669 case nir_intrinsic_image_store
:
670 case nir_intrinsic_image_atomic_add
:
671 case nir_intrinsic_image_atomic_min
:
672 case nir_intrinsic_image_atomic_max
:
673 case nir_intrinsic_image_atomic_and
:
674 case nir_intrinsic_image_atomic_or
:
675 case nir_intrinsic_image_atomic_xor
:
676 case nir_intrinsic_image_atomic_exchange
:
677 case nir_intrinsic_image_atomic_comp_swap
:
678 case nir_intrinsic_image_samples
:
679 case nir_intrinsic_image_size
: {
680 nir_ssa_undef_instr
*instr_undef
=
681 nir_ssa_undef_instr_create(shader
, 1);
682 nir_instr_insert_after_cf_list(this->cf_node_list
,
683 &instr_undef
->instr
);
685 /* Set the image variable dereference. */
686 exec_node
*param
= ir
->actual_parameters
.get_head();
687 ir_dereference
*image
= (ir_dereference
*)param
;
688 const glsl_type
*type
=
689 image
->variable_referenced()->type
->without_array();
691 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
692 param
= param
->get_next();
694 /* Set the intrinsic destination. */
695 if (ir
->return_deref
) {
696 const nir_intrinsic_info
*info
=
697 &nir_intrinsic_infos
[instr
->intrinsic
];
698 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
699 info
->dest_components
, NULL
);
702 if (op
== nir_intrinsic_image_size
||
703 op
== nir_intrinsic_image_samples
)
706 /* Set the address argument, extending the coordinate vector to four
709 const nir_src src_addr
= evaluate_rvalue((ir_dereference
*)param
);
710 nir_alu_instr
*instr_addr
= nir_alu_instr_create(shader
, nir_op_vec4
);
711 nir_ssa_dest_init(&instr_addr
->instr
, &instr_addr
->dest
.dest
, 4, NULL
);
713 for (int i
= 0; i
< 4; i
++) {
714 if (i
< type
->coordinate_components()) {
715 instr_addr
->src
[i
].src
= src_addr
;
716 instr_addr
->src
[i
].swizzle
[0] = i
;
718 instr_addr
->src
[i
].src
= nir_src_for_ssa(&instr_undef
->def
);
722 nir_instr_insert_after_cf_list(cf_node_list
, &instr_addr
->instr
);
723 instr
->src
[0] = nir_src_for_ssa(&instr_addr
->dest
.dest
.ssa
);
724 param
= param
->get_next();
726 /* Set the sample argument, which is undefined for single-sample
729 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
730 instr
->src
[1] = evaluate_rvalue((ir_dereference
*)param
);
731 param
= param
->get_next();
733 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
736 /* Set the intrinsic parameters. */
737 if (!param
->is_tail_sentinel()) {
738 instr
->src
[2] = evaluate_rvalue((ir_dereference
*)param
);
739 param
= param
->get_next();
742 if (!param
->is_tail_sentinel()) {
743 instr
->src
[3] = evaluate_rvalue((ir_dereference
*)param
);
744 param
= param
->get_next();
748 case nir_intrinsic_memory_barrier
:
751 unreachable("not reached");
754 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
756 if (ir
->return_deref
) {
757 nir_intrinsic_instr
*store_instr
=
758 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
759 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
761 store_instr
->variables
[0] =
762 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
763 store_instr
->src
[0] = nir_src_for_ssa(&instr
->dest
.ssa
);
765 nir_instr_insert_after_cf_list(this->cf_node_list
,
766 &store_instr
->instr
);
772 struct hash_entry
*entry
=
773 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
775 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
777 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
780 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
781 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
785 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
786 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
790 nir_visitor::visit(ir_assignment
*ir
)
792 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
794 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
795 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
796 /* We're doing a plain-as-can-be copy, so emit a copy_var */
797 nir_intrinsic_instr
*copy
=
798 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
800 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
801 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
804 nir_if
*if_stmt
= nir_if_create(this->shader
);
805 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
806 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
807 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
809 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
814 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
816 ir
->lhs
->accept(this);
817 nir_deref_var
*lhs_deref
= this->deref_head
;
818 nir_src src
= evaluate_rvalue(ir
->rhs
);
820 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
822 * We have no good way to update only part of a variable, so just load
823 * the LHS and do a vec operation to combine the old with the new, and
825 * back into the LHS. Copy propagation should get rid of the mess.
828 nir_intrinsic_instr
*load
=
829 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
830 load
->num_components
= ir
->lhs
->type
->vector_elements
;
831 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
832 load
->variables
[0] = lhs_deref
;
833 ralloc_steal(load
, load
->variables
[0]);
834 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
837 switch (ir
->lhs
->type
->vector_elements
) {
838 case 1: vec_op
= nir_op_imov
; break;
839 case 2: vec_op
= nir_op_vec2
; break;
840 case 3: vec_op
= nir_op_vec3
; break;
841 case 4: vec_op
= nir_op_vec4
; break;
842 default: unreachable("Invalid number of components"); break;
844 nir_alu_instr
*vec
= nir_alu_instr_create(this->shader
, vec_op
);
845 nir_ssa_dest_init(&vec
->instr
, &vec
->dest
.dest
, num_components
, NULL
);
846 vec
->dest
.write_mask
= (1 << num_components
) - 1;
848 unsigned component
= 0;
849 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
850 if (ir
->write_mask
& (1 << i
)) {
851 vec
->src
[i
].src
= src
;
853 /* GLSL IR will give us the input to the write-masked assignment
854 * in a single packed vector. So, for example, if the
855 * writemask is xzw, then we have to swizzle x -> x, y -> z,
856 * and z -> w and get the y component from the load.
858 vec
->src
[i
].swizzle
[0] = component
++;
860 vec
->src
[i
].src
.is_ssa
= true;
861 vec
->src
[i
].src
.ssa
= &load
->dest
.ssa
;
862 vec
->src
[i
].swizzle
[0] = i
;
866 nir_instr_insert_after_cf_list(this->cf_node_list
, &vec
->instr
);
869 src
.ssa
= &vec
->dest
.dest
.ssa
;
872 nir_intrinsic_instr
*store
=
873 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
874 store
->num_components
= ir
->lhs
->type
->vector_elements
;
875 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
876 store
->variables
[0] = nir_deref_as_var(store_deref
);
880 nir_if
*if_stmt
= nir_if_create(this->shader
);
881 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
882 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
883 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
885 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
890 * Given an instruction, returns a pointer to its destination or NULL if there
893 * Note that this only handles instructions we generate at this level.
896 get_instr_dest(nir_instr
*instr
)
898 nir_alu_instr
*alu_instr
;
899 nir_intrinsic_instr
*intrinsic_instr
;
900 nir_tex_instr
*tex_instr
;
902 switch (instr
->type
) {
903 case nir_instr_type_alu
:
904 alu_instr
= nir_instr_as_alu(instr
);
905 return &alu_instr
->dest
.dest
;
907 case nir_instr_type_intrinsic
:
908 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
909 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
910 return &intrinsic_instr
->dest
;
914 case nir_instr_type_tex
:
915 tex_instr
= nir_instr_as_tex(instr
);
916 return &tex_instr
->dest
;
919 unreachable("not reached");
926 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
928 nir_dest
*dest
= get_instr_dest(instr
);
930 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
932 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
933 this->result
= instr
;
937 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
940 if (ir
->as_dereference() || ir
->as_constant()) {
942 * A dereference is being used on the right hand side, which means we
943 * must emit a variable load.
946 nir_intrinsic_instr
*load_instr
=
947 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
948 load_instr
->num_components
= ir
->type
->vector_elements
;
949 load_instr
->variables
[0] = this->deref_head
;
950 ralloc_steal(load_instr
, load_instr
->variables
[0]);
951 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
954 nir_dest
*dest
= get_instr_dest(this->result
);
955 assert(dest
->is_ssa
);
957 return nir_src_for_ssa(&dest
->ssa
);
961 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
963 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
964 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
965 instr
->src
[i
].src
= srcs
[i
];
966 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
967 add_instr(&instr
->instr
, dest_size
);
972 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
974 assert(nir_op_infos
[op
].num_inputs
== 1);
975 return emit(op
, dest_size
, &src1
);
979 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
982 assert(nir_op_infos
[op
].num_inputs
== 2);
983 nir_src srcs
[] = { src1
, src2
};
984 return emit(op
, dest_size
, srcs
);
988 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
989 nir_src src2
, nir_src src3
)
991 assert(nir_op_infos
[op
].num_inputs
== 3);
992 nir_src srcs
[] = { src1
, src2
, src3
};
993 return emit(op
, dest_size
, srcs
);
997 nir_visitor::visit(ir_expression
*ir
)
999 /* Some special cases */
1000 switch (ir
->operation
) {
1001 case ir_binop_ubo_load
: {
1002 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
1004 nir_intrinsic_op op
;
1006 op
= nir_intrinsic_load_ubo
;
1008 op
= nir_intrinsic_load_ubo_indirect
;
1010 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
1011 load
->num_components
= ir
->type
->vector_elements
;
1012 load
->const_index
[0] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
1013 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
1015 load
->src
[1] = evaluate_rvalue(ir
->operands
[1]);
1016 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1019 * In UBO's, a true boolean value is any non-zero value, but we consider
1020 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1023 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
1024 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
, 1);
1025 const_zero
->value
.u
[0] = 0;
1026 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
1028 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
1029 compare
->src
[0].src
.is_ssa
= true;
1030 compare
->src
[0].src
.ssa
= &load
->dest
.ssa
;
1031 compare
->src
[1].src
.is_ssa
= true;
1032 compare
->src
[1].src
.ssa
= &const_zero
->def
;
1033 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1034 compare
->src
[1].swizzle
[i
] = 0;
1035 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
1037 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
1043 case ir_unop_interpolate_at_centroid
:
1044 case ir_binop_interpolate_at_offset
:
1045 case ir_binop_interpolate_at_sample
: {
1046 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1047 ir_swizzle
*swizzle
= NULL
;
1049 /* the api does not allow a swizzle here, but the varying packing code
1050 * may have pushed one into here.
1052 swizzle
= ir
->operands
[0]->as_swizzle();
1054 deref
= swizzle
->val
->as_dereference();
1058 deref
->accept(this);
1060 nir_intrinsic_op op
;
1061 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1062 switch (ir
->operation
) {
1063 case ir_unop_interpolate_at_centroid
:
1064 op
= nir_intrinsic_interp_var_at_centroid
;
1066 case ir_binop_interpolate_at_offset
:
1067 op
= nir_intrinsic_interp_var_at_offset
;
1069 case ir_binop_interpolate_at_sample
:
1070 op
= nir_intrinsic_interp_var_at_sample
;
1073 unreachable("Invalid interpolation intrinsic");
1076 /* This case can happen if the vertex shader does not write the
1077 * given varying. In this case, the linker will lower it to a
1078 * global variable. Since interpolating a variable makes no
1079 * sense, we'll just turn it into a load which will probably
1080 * eventually end up as an SSA definition.
1082 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1083 op
= nir_intrinsic_load_var
;
1086 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1087 intrin
->num_components
= deref
->type
->vector_elements
;
1088 intrin
->variables
[0] = this->deref_head
;
1089 ralloc_steal(intrin
, intrin
->variables
[0]);
1091 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1092 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1093 intrin
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
1095 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
1098 nir_alu_instr
*mov
= nir_alu_instr_create(shader
, nir_op_imov
);
1099 mov
->dest
.write_mask
= (1 << swizzle
->type
->vector_elements
) - 1;
1100 mov
->src
[0].src
.is_ssa
= true;
1101 mov
->src
[0].src
.ssa
= &intrin
->dest
.ssa
;
1103 mov
->src
[0].swizzle
[0] = swizzle
->mask
.x
;
1104 mov
->src
[0].swizzle
[1] = swizzle
->mask
.y
;
1105 mov
->src
[0].swizzle
[2] = swizzle
->mask
.z
;
1106 mov
->src
[0].swizzle
[3] = swizzle
->mask
.w
;
1107 for (unsigned i
= deref
->type
->vector_elements
; i
< 4; i
++)
1108 mov
->src
[0].swizzle
[i
] = 0;
1110 add_instr(&mov
->instr
, swizzle
->type
->vector_elements
);
1121 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1122 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1124 glsl_base_type types
[4];
1125 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1127 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1129 types
[i
] = GLSL_TYPE_FLOAT
;
1131 glsl_base_type out_type
;
1133 out_type
= ir
->type
->base_type
;
1135 out_type
= GLSL_TYPE_FLOAT
;
1137 unsigned dest_size
= ir
->type
->vector_elements
;
1139 nir_alu_instr
*instr
;
1142 switch (ir
->operation
) {
1143 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
1144 case ir_unop_logic_not
:
1145 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1148 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fneg
: nir_op_ineg
,
1152 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fabs
: nir_op_iabs
,
1155 case ir_unop_saturate
:
1156 assert(types
[0] == GLSL_TYPE_FLOAT
);
1157 instr
= emit(nir_op_fsat
, dest_size
, srcs
);
1160 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1163 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1164 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1165 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1166 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1167 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1168 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1169 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1171 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1174 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1177 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1179 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1180 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1181 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1182 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1183 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1186 case ir_unop_bitcast_i2f
:
1187 case ir_unop_bitcast_f2i
:
1188 case ir_unop_bitcast_u2f
:
1189 case ir_unop_bitcast_f2u
:
1190 case ir_unop_subroutine_to_int
:
1192 emit(nir_op_imov
, dest_size
, srcs
);
1195 switch (ir
->operands
[0]->type
->vector_elements
) {
1197 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1201 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1205 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1209 unreachable("not reached");
1212 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1213 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1214 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1215 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1216 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1217 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1218 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1219 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1220 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1221 case ir_unop_dFdx_fine
: emit(nir_op_fddx_fine
, dest_size
, srcs
); break;
1222 case ir_unop_dFdy_fine
: emit(nir_op_fddy_fine
, dest_size
, srcs
); break;
1223 case ir_unop_dFdx_coarse
: emit(nir_op_fddx_coarse
, dest_size
, srcs
); break;
1224 case ir_unop_dFdy_coarse
: emit(nir_op_fddy_coarse
, dest_size
, srcs
); break;
1225 case ir_unop_pack_snorm_2x16
:
1226 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1228 case ir_unop_pack_snorm_4x8
:
1229 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1231 case ir_unop_pack_unorm_2x16
:
1232 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1234 case ir_unop_pack_unorm_4x8
:
1235 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1237 case ir_unop_pack_half_2x16
:
1238 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1240 case ir_unop_unpack_snorm_2x16
:
1241 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1243 case ir_unop_unpack_snorm_4x8
:
1244 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1246 case ir_unop_unpack_unorm_2x16
:
1247 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1249 case ir_unop_unpack_unorm_4x8
:
1250 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1252 case ir_unop_unpack_half_2x16
:
1253 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1255 case ir_unop_unpack_half_2x16_split_x
:
1256 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1258 case ir_unop_unpack_half_2x16_split_y
:
1259 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1261 case ir_unop_bitfield_reverse
:
1262 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1264 case ir_unop_bit_count
:
1265 emit(nir_op_bit_count
, dest_size
, srcs
);
1267 case ir_unop_find_msb
:
1269 case GLSL_TYPE_UINT
:
1270 emit(nir_op_ufind_msb
, dest_size
, srcs
);
1273 emit(nir_op_ifind_msb
, dest_size
, srcs
);
1276 unreachable("Invalid type for findMSB()");
1279 case ir_unop_find_lsb
:
1280 emit(nir_op_find_lsb
, dest_size
, srcs
);
1284 switch (ir
->type
->vector_elements
) {
1286 switch (ir
->operands
[0]->type
->vector_elements
) {
1287 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1288 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1289 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1290 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1291 default: unreachable("not reached");
1295 switch (ir
->operands
[0]->type
->vector_elements
) {
1296 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1297 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1298 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1299 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1300 default: unreachable("not reached");
1304 switch (ir
->operands
[0]->type
->vector_elements
) {
1305 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1306 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1307 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1308 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1309 default: unreachable("not reached");
1313 switch (ir
->operands
[0]->type
->vector_elements
) {
1314 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1315 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1316 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1317 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1318 default: unreachable("not reached");
1322 unreachable("not reached");
1333 case ir_binop_bit_and
:
1334 case ir_binop_bit_or
:
1335 case ir_binop_bit_xor
:
1336 case ir_binop_logic_and
:
1337 case ir_binop_logic_or
:
1338 case ir_binop_logic_xor
:
1339 case ir_binop_lshift
:
1340 case ir_binop_rshift
:
1341 switch (ir
->operation
) {
1343 if (out_type
== GLSL_TYPE_FLOAT
)
1349 if (out_type
== GLSL_TYPE_FLOAT
)
1355 if (out_type
== GLSL_TYPE_FLOAT
)
1361 if (out_type
== GLSL_TYPE_FLOAT
)
1363 else if (out_type
== GLSL_TYPE_INT
)
1369 if (out_type
== GLSL_TYPE_FLOAT
)
1375 if (out_type
== GLSL_TYPE_FLOAT
)
1377 else if (out_type
== GLSL_TYPE_INT
)
1383 if (out_type
== GLSL_TYPE_FLOAT
)
1385 else if (out_type
== GLSL_TYPE_INT
)
1390 case ir_binop_bit_and
:
1393 case ir_binop_bit_or
:
1396 case ir_binop_bit_xor
:
1399 case ir_binop_logic_and
:
1405 case ir_binop_logic_or
:
1411 case ir_binop_logic_xor
:
1417 case ir_binop_lshift
:
1420 case ir_binop_rshift
:
1421 if (out_type
== GLSL_TYPE_INT
)
1431 unreachable("not reached");
1434 instr
= emit(op
, dest_size
, srcs
);
1436 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1437 ir
->operands
[1]->type
->vector_elements
== 1) {
1438 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1440 instr
->src
[1].swizzle
[i
] = 0;
1444 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1445 ir
->operands
[0]->type
->vector_elements
== 1) {
1446 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1448 instr
->src
[0].swizzle
[i
] = 0;
1453 case ir_binop_imul_high
:
1454 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1457 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1458 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1460 if (supports_ints
) {
1461 if (types
[0] == GLSL_TYPE_FLOAT
)
1462 emit(nir_op_flt
, dest_size
, srcs
);
1463 else if (types
[0] == GLSL_TYPE_INT
)
1464 emit(nir_op_ilt
, dest_size
, srcs
);
1466 emit(nir_op_ult
, dest_size
, srcs
);
1468 emit(nir_op_slt
, dest_size
, srcs
);
1471 case ir_binop_greater
:
1472 if (supports_ints
) {
1473 if (types
[0] == GLSL_TYPE_FLOAT
)
1474 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1475 else if (types
[0] == GLSL_TYPE_INT
)
1476 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1478 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1480 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1483 case ir_binop_lequal
:
1484 if (supports_ints
) {
1485 if (types
[0] == GLSL_TYPE_FLOAT
)
1486 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1487 else if (types
[0] == GLSL_TYPE_INT
)
1488 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1490 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1492 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1495 case ir_binop_gequal
:
1496 if (supports_ints
) {
1497 if (types
[0] == GLSL_TYPE_FLOAT
)
1498 emit(nir_op_fge
, dest_size
, srcs
);
1499 else if (types
[0] == GLSL_TYPE_INT
)
1500 emit(nir_op_ige
, dest_size
, srcs
);
1502 emit(nir_op_uge
, dest_size
, srcs
);
1504 emit(nir_op_slt
, dest_size
, srcs
);
1507 case ir_binop_equal
:
1508 if (supports_ints
) {
1509 if (types
[0] == GLSL_TYPE_FLOAT
)
1510 emit(nir_op_feq
, dest_size
, srcs
);
1512 emit(nir_op_ieq
, dest_size
, srcs
);
1514 emit(nir_op_seq
, dest_size
, srcs
);
1517 case ir_binop_nequal
:
1518 if (supports_ints
) {
1519 if (types
[0] == GLSL_TYPE_FLOAT
)
1520 emit(nir_op_fne
, dest_size
, srcs
);
1522 emit(nir_op_ine
, dest_size
, srcs
);
1524 emit(nir_op_sne
, dest_size
, srcs
);
1527 case ir_binop_all_equal
:
1528 if (supports_ints
) {
1529 if (types
[0] == GLSL_TYPE_FLOAT
) {
1530 switch (ir
->operands
[0]->type
->vector_elements
) {
1531 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1532 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1533 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1534 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1536 unreachable("not reached");
1539 switch (ir
->operands
[0]->type
->vector_elements
) {
1540 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1541 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1542 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1543 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1545 unreachable("not reached");
1549 switch (ir
->operands
[0]->type
->vector_elements
) {
1550 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1551 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1552 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1553 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1555 unreachable("not reached");
1559 case ir_binop_any_nequal
:
1560 if (supports_ints
) {
1561 if (types
[0] == GLSL_TYPE_FLOAT
) {
1562 switch (ir
->operands
[0]->type
->vector_elements
) {
1563 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1564 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1565 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1566 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1568 unreachable("not reached");
1571 switch (ir
->operands
[0]->type
->vector_elements
) {
1572 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1573 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1574 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1575 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1577 unreachable("not reached");
1581 switch (ir
->operands
[0]->type
->vector_elements
) {
1582 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1583 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1584 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1585 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1587 unreachable("not reached");
1592 switch (ir
->operands
[0]->type
->vector_elements
) {
1593 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1594 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1595 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1597 unreachable("not reached");
1601 case ir_binop_pack_half_2x16_split
:
1602 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1604 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1605 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1606 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1608 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1609 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1610 ir
->operands
[2]->type
->vector_elements
== 1) {
1611 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1613 instr
->src
[2].swizzle
[i
] = 0;
1619 emit(nir_op_bcsel
, dest_size
, srcs
);
1621 emit(nir_op_fcsel
, dest_size
, srcs
);
1624 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1625 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1626 instr
->src
[0].swizzle
[i
] = 0;
1629 case ir_triop_bitfield_extract
:
1630 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1631 nir_op_ubitfield_extract
, dest_size
, srcs
);
1632 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1633 instr
->src
[1].swizzle
[i
] = 0;
1634 instr
->src
[2].swizzle
[i
] = 0;
1637 case ir_quadop_bitfield_insert
:
1638 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1639 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1640 instr
->src
[2].swizzle
[i
] = 0;
1641 instr
->src
[3].swizzle
[i
] = 0;
1644 case ir_quadop_vector
:
1645 switch (ir
->type
->vector_elements
) {
1646 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1647 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1648 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1649 default: unreachable("not reached");
1654 unreachable("not reached");
1659 nir_visitor::visit(ir_swizzle
*ir
)
1661 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1662 ir
->type
->vector_elements
,
1663 evaluate_rvalue(ir
->val
));
1665 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1666 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1667 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1671 nir_visitor::visit(ir_texture
*ir
)
1678 num_srcs
= 1; /* coordinate */
1683 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1684 num_srcs
= 2; /* coordinate, bias/lod */
1688 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1694 if (ir
->lod_info
.lod
!= NULL
)
1695 num_srcs
= 2; /* coordinate, lod */
1697 num_srcs
= 1; /* coordinate */
1701 op
= nir_texop_txf_ms
;
1702 num_srcs
= 2; /* coordinate, sample_index */
1707 if (ir
->lod_info
.lod
!= NULL
)
1708 num_srcs
= 1; /* lod */
1715 num_srcs
= 1; /* coordinate */
1720 num_srcs
= 1; /* coordinate */
1723 case ir_query_levels
:
1724 op
= nir_texop_query_levels
;
1728 case ir_texture_samples
:
1729 op
= nir_texop_texture_samples
;
1734 unreachable("not reached");
1737 if (ir
->projector
!= NULL
)
1739 if (ir
->shadow_comparitor
!= NULL
)
1741 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1744 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1747 instr
->sampler_dim
=
1748 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1749 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1750 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1751 if (instr
->is_shadow
)
1752 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1753 switch (ir
->type
->base_type
) {
1754 case GLSL_TYPE_FLOAT
:
1755 instr
->dest_type
= nir_type_float
;
1758 instr
->dest_type
= nir_type_int
;
1760 case GLSL_TYPE_UINT
:
1761 instr
->dest_type
= nir_type_unsigned
;
1764 unreachable("not reached");
1767 instr
->sampler
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1769 unsigned src_number
= 0;
1771 if (ir
->coordinate
!= NULL
) {
1772 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1773 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->coordinate
);
1774 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1778 if (ir
->projector
!= NULL
) {
1779 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->projector
);
1780 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1784 if (ir
->shadow_comparitor
!= NULL
) {
1785 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->shadow_comparitor
);
1786 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1790 if (ir
->offset
!= NULL
) {
1791 /* we don't support multiple offsets yet */
1792 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1794 ir_constant
*const_offset
= ir
->offset
->as_constant();
1795 if (const_offset
!= NULL
) {
1796 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1797 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1799 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->offset
);
1800 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1807 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.bias
);
1808 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1815 if (ir
->lod_info
.lod
!= NULL
) {
1816 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.lod
);
1817 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1823 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdx
);
1824 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1826 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdy
);
1827 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1832 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.sample_index
);
1833 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1838 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1845 assert(src_number
== num_srcs
);
1847 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
1851 nir_visitor::visit(ir_constant
*ir
)
1854 * We don't know if this variable is an an array or struct that gets
1855 * dereferenced, so do the safe thing an make it a variable with a
1856 * constant initializer and return a dereference.
1859 nir_variable
*var
= ralloc(this->shader
, nir_variable
);
1860 var
->name
= ralloc_strdup(var
, "const_temp");
1861 var
->type
= ir
->type
;
1862 var
->data
.mode
= nir_var_local
;
1863 var
->data
.read_only
= true;
1864 var
->constant_initializer
= constant_copy(ir
, var
);
1865 exec_list_push_tail(&this->impl
->locals
, &var
->node
);
1867 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1868 this->deref_tail
= &this->deref_head
->deref
;
1872 nir_visitor::visit(ir_dereference_variable
*ir
)
1874 struct hash_entry
*entry
=
1875 _mesa_hash_table_search(this->var_table
, ir
->var
);
1877 nir_variable
*var
= (nir_variable
*) entry
->data
;
1879 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
1880 this->deref_head
= deref
;
1881 this->deref_tail
= &deref
->deref
;
1885 nir_visitor::visit(ir_dereference_record
*ir
)
1887 ir
->record
->accept(this);
1889 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
1890 assert(field_index
>= 0);
1892 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
1893 deref
->deref
.type
= ir
->type
;
1894 this->deref_tail
->child
= &deref
->deref
;
1895 this->deref_tail
= &deref
->deref
;
1899 nir_visitor::visit(ir_dereference_array
*ir
)
1901 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
1902 deref
->deref
.type
= ir
->type
;
1904 ir_constant
*const_index
= ir
->array_index
->as_constant();
1905 if (const_index
!= NULL
) {
1906 deref
->deref_array_type
= nir_deref_array_type_direct
;
1907 deref
->base_offset
= const_index
->value
.u
[0];
1909 deref
->deref_array_type
= nir_deref_array_type_indirect
;
1910 deref
->indirect
= evaluate_rvalue(ir
->array_index
);
1913 ir
->array
->accept(this);
1915 this->deref_tail
->child
= &deref
->deref
;
1916 ralloc_steal(this->deref_tail
, deref
);
1917 this->deref_tail
= &deref
->deref
;
1921 nir_visitor::visit(ir_barrier
*ir
)
1923 nir_intrinsic_instr
*instr
=
1924 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
1925 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);