2 * Copyright © 2014 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Connor Abbott (cwabbott0@gmail.com)
28 #include "glsl_to_nir.h"
29 #include "ir_visitor.h"
30 #include "ir_hierarchical_visitor.h"
34 * pass to lower GLSL IR to NIR
36 * This will lower variable dereferences to loads/stores of corresponding
37 * variables in NIR - the variables will be converted to registers in a later
43 class nir_visitor
: public ir_visitor
46 nir_visitor(nir_shader
*shader
, bool supports_ints
);
49 virtual void visit(ir_variable
*);
50 virtual void visit(ir_function
*);
51 virtual void visit(ir_function_signature
*);
52 virtual void visit(ir_loop
*);
53 virtual void visit(ir_if
*);
54 virtual void visit(ir_discard
*);
55 virtual void visit(ir_loop_jump
*);
56 virtual void visit(ir_return
*);
57 virtual void visit(ir_call
*);
58 virtual void visit(ir_assignment
*);
59 virtual void visit(ir_emit_vertex
*);
60 virtual void visit(ir_end_primitive
*);
61 virtual void visit(ir_expression
*);
62 virtual void visit(ir_swizzle
*);
63 virtual void visit(ir_texture
*);
64 virtual void visit(ir_constant
*);
65 virtual void visit(ir_dereference_variable
*);
66 virtual void visit(ir_dereference_record
*);
67 virtual void visit(ir_dereference_array
*);
69 void create_function(ir_function
*ir
);
72 void create_overload(ir_function_signature
*ir
, nir_function
*function
);
73 void add_instr(nir_instr
*instr
, unsigned num_components
);
74 nir_src
evaluate_rvalue(ir_rvalue
*ir
);
76 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
);
77 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
);
78 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
,
80 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
,
81 nir_src src2
, nir_src src3
);
86 nir_function_impl
*impl
;
87 exec_list
*cf_node_list
;
88 nir_instr
*result
; /* result of the expression tree last visited */
90 /* the head of the dereference chain we're creating */
91 nir_deref_var
*deref_head
;
92 /* the tail of the dereference chain we're creating */
93 nir_deref
*deref_tail
;
95 nir_variable
*var
; /* variable created by ir_variable visitor */
97 /* whether the IR we're operating on is per-function or global */
100 /* map of ir_variable -> nir_variable */
101 struct hash_table
*var_table
;
103 /* map of ir_function_signature -> nir_function_overload */
104 struct hash_table
*overload_table
;
108 * This visitor runs before the main visitor, calling create_function() for
109 * each function so that the main visitor can resolve forward references in
113 class nir_function_visitor
: public ir_hierarchical_visitor
116 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
119 virtual ir_visitor_status
visit_enter(ir_function
*);
122 nir_visitor
*visitor
;
125 }; /* end of anonymous namespace */
127 static const nir_shader_compiler_options default_options
= {
131 glsl_to_nir(exec_list
*ir
, _mesa_glsl_parse_state
*state
,
132 bool native_integers
)
134 const nir_shader_compiler_options
*options
;
137 struct gl_context
*ctx
= state
->ctx
;
138 struct gl_shader_compiler_options
*gl_options
=
139 &ctx
->Const
.ShaderCompilerOptions
[state
->stage
];
141 if (!gl_options
->NirOptions
) {
142 nir_shader_compiler_options
*new_options
=
143 rzalloc(ctx
, nir_shader_compiler_options
);
144 options
= gl_options
->NirOptions
= new_options
;
146 if (gl_options
->EmitNoPow
)
147 new_options
->lower_fpow
= true;
149 options
= gl_options
->NirOptions
;
152 options
= &default_options
;
155 nir_shader
*shader
= nir_shader_create(NULL
, options
);
157 nir_visitor
v1(shader
, native_integers
);
158 nir_function_visitor
v2(&v1
);
160 visit_exec_list(ir
, &v1
);
165 nir_visitor::nir_visitor(nir_shader
*shader
, bool supports_ints
)
167 this->supports_ints
= supports_ints
;
168 this->shader
= shader
;
169 this->is_global
= true;
170 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
171 _mesa_key_pointer_equal
);
172 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
173 _mesa_key_pointer_equal
);
176 nir_visitor::~nir_visitor()
178 _mesa_hash_table_destroy(this->var_table
, NULL
);
179 _mesa_hash_table_destroy(this->overload_table
, NULL
);
182 static nir_constant
*
183 constant_copy(ir_constant
*ir
, void *mem_ctx
)
188 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
190 unsigned total_elems
= ir
->type
->components();
192 switch (ir
->type
->base_type
) {
194 for (i
= 0; i
< total_elems
; i
++)
195 ret
->value
.u
[i
] = ir
->value
.u
[i
];
199 for (i
= 0; i
< total_elems
; i
++)
200 ret
->value
.i
[i
] = ir
->value
.i
[i
];
203 case GLSL_TYPE_FLOAT
:
204 for (i
= 0; i
< total_elems
; i
++)
205 ret
->value
.f
[i
] = ir
->value
.f
[i
];
209 for (i
= 0; i
< total_elems
; i
++)
210 ret
->value
.b
[i
] = ir
->value
.b
[i
];
213 case GLSL_TYPE_STRUCT
:
214 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
217 foreach_in_list(ir_constant
, field
, &ir
->components
) {
218 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
223 case GLSL_TYPE_ARRAY
:
224 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
227 for (i
= 0; i
< ir
->type
->length
; i
++)
228 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
232 unreachable("not reached");
239 nir_visitor::visit(ir_variable
*ir
)
241 nir_variable
*var
= ralloc(shader
, nir_variable
);
242 var
->type
= ir
->type
;
243 var
->name
= ralloc_strdup(var
, ir
->name
);
245 if (ir
->is_interface_instance() && ir
->get_max_ifc_array_access() != NULL
) {
246 unsigned size
= ir
->get_interface_type()->length
;
247 var
->max_ifc_array_access
= ralloc_array(var
, unsigned, size
);
248 memcpy(var
->max_ifc_array_access
, ir
->get_max_ifc_array_access(),
249 size
* sizeof(unsigned));
251 var
->max_ifc_array_access
= NULL
;
254 var
->data
.read_only
= ir
->data
.read_only
;
255 var
->data
.centroid
= ir
->data
.centroid
;
256 var
->data
.sample
= ir
->data
.sample
;
257 var
->data
.invariant
= ir
->data
.invariant
;
258 var
->data
.location
= ir
->data
.location
;
260 switch(ir
->data
.mode
) {
262 case ir_var_temporary
:
264 var
->data
.mode
= nir_var_global
;
266 var
->data
.mode
= nir_var_local
;
269 case ir_var_function_in
:
270 case ir_var_function_out
:
271 case ir_var_function_inout
:
272 case ir_var_const_in
:
273 var
->data
.mode
= nir_var_local
;
276 case ir_var_shader_in
:
277 if (ir
->data
.location
== VARYING_SLOT_FACE
) {
278 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
279 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
280 var
->data
.mode
= nir_var_system_value
;
282 var
->data
.mode
= nir_var_shader_in
;
286 case ir_var_shader_out
:
287 var
->data
.mode
= nir_var_shader_out
;
291 var
->data
.mode
= nir_var_uniform
;
295 case ir_var_system_value
:
296 var
->data
.mode
= nir_var_system_value
;
300 unreachable("not reached");
303 var
->data
.interpolation
= ir
->data
.interpolation
;
304 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
305 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
306 var
->data
.explicit_location
= ir
->data
.explicit_location
;
307 var
->data
.explicit_index
= ir
->data
.explicit_index
;
308 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
309 var
->data
.has_initializer
= ir
->data
.has_initializer
;
310 var
->data
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
311 var
->data
.location_frac
= ir
->data
.location_frac
;
312 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
313 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
315 switch (ir
->data
.depth_layout
) {
316 case ir_depth_layout_none
:
317 var
->data
.depth_layout
= nir_depth_layout_none
;
319 case ir_depth_layout_any
:
320 var
->data
.depth_layout
= nir_depth_layout_any
;
322 case ir_depth_layout_greater
:
323 var
->data
.depth_layout
= nir_depth_layout_greater
;
325 case ir_depth_layout_less
:
326 var
->data
.depth_layout
= nir_depth_layout_less
;
328 case ir_depth_layout_unchanged
:
329 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
332 unreachable("not reached");
335 var
->data
.index
= ir
->data
.index
;
336 var
->data
.binding
= ir
->data
.binding
;
337 /* XXX Get rid of buffer_index */
338 var
->data
.atomic
.buffer_index
= ir
->data
.binding
;
339 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
340 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
341 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
342 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
343 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
344 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
345 var
->data
.image
.format
= ir
->data
.image_format
;
346 var
->data
.max_array_access
= ir
->data
.max_array_access
;
348 var
->num_state_slots
= ir
->get_num_state_slots();
349 if (var
->num_state_slots
> 0) {
350 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
351 var
->num_state_slots
);
353 ir_state_slot
*state_slots
= ir
->get_state_slots();
354 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
355 for (unsigned j
= 0; j
< 5; j
++)
356 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
357 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
360 var
->state_slots
= NULL
;
363 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
365 var
->interface_type
= ir
->get_interface_type();
367 switch (var
->data
.mode
) {
369 exec_list_push_tail(&impl
->locals
, &var
->node
);
373 exec_list_push_tail(&shader
->globals
, &var
->node
);
376 case nir_var_shader_in
:
377 _mesa_hash_table_insert(shader
->inputs
, var
->name
, var
);
380 case nir_var_shader_out
:
381 _mesa_hash_table_insert(shader
->outputs
, var
->name
, var
);
384 case nir_var_uniform
:
385 _mesa_hash_table_insert(shader
->uniforms
, var
->name
, var
);
388 case nir_var_system_value
:
389 exec_list_push_tail(&shader
->system_values
, &var
->node
);
393 unreachable("not reached");
396 _mesa_hash_table_insert(var_table
, ir
, var
);
401 nir_function_visitor::visit_enter(ir_function
*ir
)
403 visitor
->create_function(ir
);
404 return visit_continue_with_parent
;
409 nir_visitor::create_function(ir_function
*ir
)
411 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
412 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
413 create_overload(sig
, func
);
420 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
422 if (ir
->is_intrinsic
)
425 nir_function_overload
*overload
= nir_function_overload_create(function
);
427 unsigned num_params
= ir
->parameters
.length();
428 overload
->num_params
= num_params
;
429 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
432 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
433 switch (param
->data
.mode
) {
434 case ir_var_function_in
:
435 overload
->params
[i
].param_type
= nir_parameter_in
;
438 case ir_var_function_out
:
439 overload
->params
[i
].param_type
= nir_parameter_out
;
442 case ir_var_function_inout
:
443 overload
->params
[i
].param_type
= nir_parameter_inout
;
447 unreachable("not reached");
450 overload
->params
[i
].type
= param
->type
;
454 overload
->return_type
= ir
->return_type
;
456 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
460 nir_visitor::visit(ir_function
*ir
)
462 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
467 nir_visitor::visit(ir_function_signature
*ir
)
469 if (ir
->is_intrinsic
)
472 struct hash_entry
*entry
=
473 _mesa_hash_table_search(this->overload_table
, ir
);
476 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
478 if (ir
->is_defined
) {
479 nir_function_impl
*impl
= nir_function_impl_create(overload
);
482 unsigned num_params
= overload
->num_params
;
483 impl
->num_params
= num_params
;
484 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
486 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
488 impl
->params
[i
] = this->var
;
492 if (overload
->return_type
== glsl_type::void_type
) {
493 impl
->return_var
= NULL
;
495 impl
->return_var
= ralloc(this->shader
, nir_variable
);
496 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
498 impl
->return_var
->type
= overload
->return_type
;
501 this->is_global
= false;
503 this->cf_node_list
= &impl
->body
;
504 visit_exec_list(&ir
->body
, this);
506 this->is_global
= true;
508 overload
->impl
= NULL
;
513 nir_visitor::visit(ir_loop
*ir
)
515 exec_list
*old_list
= this->cf_node_list
;
517 nir_loop
*loop
= nir_loop_create(this->shader
);
518 nir_cf_node_insert_end(old_list
, &loop
->cf_node
);
519 this->cf_node_list
= &loop
->body
;
520 visit_exec_list(&ir
->body_instructions
, this);
522 this->cf_node_list
= old_list
;
526 nir_visitor::visit(ir_if
*ir
)
528 nir_src condition
= evaluate_rvalue(ir
->condition
);
530 exec_list
*old_list
= this->cf_node_list
;
532 nir_if
*if_stmt
= nir_if_create(this->shader
);
533 if_stmt
->condition
= condition
;
534 nir_cf_node_insert_end(old_list
, &if_stmt
->cf_node
);
536 this->cf_node_list
= &if_stmt
->then_list
;
537 visit_exec_list(&ir
->then_instructions
, this);
539 this->cf_node_list
= &if_stmt
->else_list
;
540 visit_exec_list(&ir
->else_instructions
, this);
542 this->cf_node_list
= old_list
;
546 nir_visitor::visit(ir_discard
*ir
)
549 * discards aren't treated as control flow, because before we lower them
550 * they can appear anywhere in the shader and the stuff after them may still
551 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
552 * discards will be immediately followed by a return.
555 nir_intrinsic_instr
*discard
;
557 discard
= nir_intrinsic_instr_create(this->shader
,
558 nir_intrinsic_discard_if
);
559 discard
->src
[0] = evaluate_rvalue(ir
->condition
);
561 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
563 nir_instr_insert_after_cf_list(this->cf_node_list
, &discard
->instr
);
567 nir_visitor::visit(ir_emit_vertex
*ir
)
569 nir_intrinsic_instr
*instr
=
570 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
571 instr
->const_index
[0] = ir
->stream_id();
572 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
576 nir_visitor::visit(ir_end_primitive
*ir
)
578 nir_intrinsic_instr
*instr
=
579 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
580 instr
->const_index
[0] = ir
->stream_id();
581 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
585 nir_visitor::visit(ir_loop_jump
*ir
)
589 case ir_loop_jump::jump_break
:
590 type
= nir_jump_break
;
592 case ir_loop_jump::jump_continue
:
593 type
= nir_jump_continue
;
596 unreachable("not reached");
599 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
600 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
604 nir_visitor::visit(ir_return
*ir
)
606 if (ir
->value
!= NULL
) {
607 ir
->value
->accept(this);
608 nir_intrinsic_instr
*copy
=
609 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
611 copy
->variables
[0] = nir_deref_var_create(this->shader
,
612 this->impl
->return_var
);
613 copy
->variables
[1] = this->deref_head
;
616 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
617 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
621 nir_visitor::visit(ir_call
*ir
)
623 if (ir
->callee
->is_intrinsic
) {
625 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
626 op
= nir_intrinsic_atomic_counter_read_var
;
627 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
628 op
= nir_intrinsic_atomic_counter_inc_var
;
629 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
630 op
= nir_intrinsic_atomic_counter_dec_var
;
632 unreachable("not reached");
635 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
636 ir_dereference
*param
=
637 (ir_dereference
*) ir
->actual_parameters
.get_head();
639 instr
->variables
[0] = this->deref_head
;
640 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
642 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
644 nir_intrinsic_instr
*store_instr
=
645 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
646 store_instr
->num_components
= 1;
648 ir
->return_deref
->accept(this);
649 store_instr
->variables
[0] = this->deref_head
;
650 store_instr
->src
[0].is_ssa
= true;
651 store_instr
->src
[0].ssa
= &instr
->dest
.ssa
;
653 nir_instr_insert_after_cf_list(this->cf_node_list
, &store_instr
->instr
);
658 struct hash_entry
*entry
=
659 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
661 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
663 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
666 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
668 instr
->params
[i
] = this->deref_head
;
672 ir
->return_deref
->accept(this);
673 instr
->return_deref
= this->deref_head
;
674 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
678 nir_visitor::visit(ir_assignment
*ir
)
680 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
682 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
683 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
684 /* We're doing a plain-as-can-be copy, so emit a copy_var */
685 nir_intrinsic_instr
*copy
=
686 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
688 ir
->lhs
->accept(this);
689 copy
->variables
[0] = this->deref_head
;
691 ir
->rhs
->accept(this);
692 copy
->variables
[1] = this->deref_head
;
696 nir_if
*if_stmt
= nir_if_create(this->shader
);
697 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
698 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
699 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
701 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
706 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
708 ir
->lhs
->accept(this);
709 nir_deref_var
*lhs_deref
= this->deref_head
;
710 nir_src src
= evaluate_rvalue(ir
->rhs
);
712 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
714 * We have no good way to update only part of a variable, so just load
715 * the LHS and do a vec operation to combine the old with the new, and
717 * back into the LHS. Copy propagation should get rid of the mess.
720 nir_intrinsic_instr
*load
=
721 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
722 load
->num_components
= ir
->lhs
->type
->vector_elements
;
723 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
724 load
->variables
[0] = lhs_deref
;
725 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
728 switch (ir
->lhs
->type
->vector_elements
) {
729 case 1: vec_op
= nir_op_imov
; break;
730 case 2: vec_op
= nir_op_vec2
; break;
731 case 3: vec_op
= nir_op_vec3
; break;
732 case 4: vec_op
= nir_op_vec4
; break;
733 default: unreachable("Invalid number of components"); break;
735 nir_alu_instr
*vec
= nir_alu_instr_create(this->shader
, vec_op
);
736 nir_ssa_dest_init(&vec
->instr
, &vec
->dest
.dest
, num_components
, NULL
);
737 vec
->dest
.write_mask
= (1 << num_components
) - 1;
739 unsigned component
= 0;
740 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
741 if (ir
->write_mask
& (1 << i
)) {
742 vec
->src
[i
].src
= src
;
744 /* GLSL IR will give us the input to the write-masked assignment
745 * in a single packed vector. So, for example, if the
746 * writemask is xzw, then we have to swizzle x -> x, y -> z,
747 * and z -> w and get the y component from the load.
749 vec
->src
[i
].swizzle
[0] = component
++;
751 vec
->src
[i
].src
.is_ssa
= true;
752 vec
->src
[i
].src
.ssa
= &load
->dest
.ssa
;
753 vec
->src
[i
].swizzle
[0] = i
;
757 nir_instr_insert_after_cf_list(this->cf_node_list
, &vec
->instr
);
760 src
.ssa
= &vec
->dest
.dest
.ssa
;
763 nir_intrinsic_instr
*store
=
764 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
765 store
->num_components
= ir
->lhs
->type
->vector_elements
;
766 nir_deref
*store_deref
= nir_copy_deref(this->shader
, &lhs_deref
->deref
);
767 store
->variables
[0] = nir_deref_as_var(store_deref
);
771 nir_if
*if_stmt
= nir_if_create(this->shader
);
772 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
773 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
774 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
776 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
781 * Given an instruction, returns a pointer to its destination or NULL if there
784 * Note that this only handles instructions we generate at this level.
787 get_instr_dest(nir_instr
*instr
)
789 nir_alu_instr
*alu_instr
;
790 nir_intrinsic_instr
*intrinsic_instr
;
791 nir_tex_instr
*tex_instr
;
793 switch (instr
->type
) {
794 case nir_instr_type_alu
:
795 alu_instr
= nir_instr_as_alu(instr
);
796 return &alu_instr
->dest
.dest
;
798 case nir_instr_type_intrinsic
:
799 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
800 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
801 return &intrinsic_instr
->dest
;
805 case nir_instr_type_tex
:
806 tex_instr
= nir_instr_as_tex(instr
);
807 return &tex_instr
->dest
;
810 unreachable("not reached");
817 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
819 nir_dest
*dest
= get_instr_dest(instr
);
821 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
823 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
824 this->result
= instr
;
828 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
831 if (ir
->as_dereference() || ir
->as_constant()) {
833 * A dereference is being used on the right hand side, which means we
834 * must emit a variable load.
837 nir_intrinsic_instr
*load_instr
=
838 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
839 load_instr
->num_components
= ir
->type
->vector_elements
;
840 load_instr
->variables
[0] = this->deref_head
;
841 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
844 nir_dest
*dest
= get_instr_dest(this->result
);
846 assert(dest
->is_ssa
);
849 src
.ssa
= &dest
->ssa
;
855 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
857 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
858 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
859 instr
->src
[i
].src
= srcs
[i
];
860 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
861 add_instr(&instr
->instr
, dest_size
);
866 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
868 assert(nir_op_infos
[op
].num_inputs
== 1);
869 return emit(op
, dest_size
, &src1
);
873 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
876 assert(nir_op_infos
[op
].num_inputs
== 2);
877 nir_src srcs
[] = { src1
, src2
};
878 return emit(op
, dest_size
, srcs
);
882 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
883 nir_src src2
, nir_src src3
)
885 assert(nir_op_infos
[op
].num_inputs
== 3);
886 nir_src srcs
[] = { src1
, src2
, src3
};
887 return emit(op
, dest_size
, srcs
);
891 nir_visitor::visit(ir_expression
*ir
)
893 /* Some special cases */
894 switch (ir
->operation
) {
895 case ir_binop_ubo_load
: {
896 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
900 op
= nir_intrinsic_load_ubo
;
902 op
= nir_intrinsic_load_ubo_indirect
;
904 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
905 load
->num_components
= ir
->type
->vector_elements
;
906 load
->const_index
[0] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
907 load
->const_index
[1] = 1; /* number of vec4's */
908 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
910 load
->src
[1] = evaluate_rvalue(ir
->operands
[1]);
911 add_instr(&load
->instr
, ir
->type
->vector_elements
);
914 * In UBO's, a true boolean value is any non-zero value, but we consider
915 * a true boolean to be ~0. Fix this up with a != 0 comparison.
918 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
919 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
, 1);
920 const_zero
->value
.u
[0] = 0;
921 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
923 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
924 compare
->src
[0].src
.is_ssa
= true;
925 compare
->src
[0].src
.ssa
= &load
->dest
.ssa
;
926 compare
->src
[1].src
.is_ssa
= true;
927 compare
->src
[1].src
.ssa
= &const_zero
->def
;
928 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
929 compare
->src
[1].swizzle
[i
] = 0;
930 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
932 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
938 case ir_unop_interpolate_at_centroid
:
939 case ir_binop_interpolate_at_offset
:
940 case ir_binop_interpolate_at_sample
: {
941 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
942 ir_swizzle
*swizzle
= NULL
;
944 /* the api does not allow a swizzle here, but the varying packing code
945 * may have pushed one into here.
947 swizzle
= ir
->operands
[0]->as_swizzle();
949 deref
= swizzle
->val
->as_dereference();
956 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
957 switch (ir
->operation
) {
958 case ir_unop_interpolate_at_centroid
:
959 op
= nir_intrinsic_interp_var_at_centroid
;
961 case ir_binop_interpolate_at_offset
:
962 op
= nir_intrinsic_interp_var_at_offset
;
964 case ir_binop_interpolate_at_sample
:
965 op
= nir_intrinsic_interp_var_at_sample
;
968 unreachable("Invalid interpolation intrinsic");
971 /* This case can happen if the vertex shader does not write the
972 * given varying. In this case, the linker will lower it to a
973 * global variable. Since interpolating a variable makes no
974 * sense, we'll just turn it into a load which will probably
975 * eventually end up as an SSA definition.
977 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
978 op
= nir_intrinsic_load_var
;
981 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
982 intrin
->num_components
= deref
->type
->vector_elements
;
983 intrin
->variables
[0] = this->deref_head
;
985 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
986 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
987 intrin
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
989 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
992 nir_alu_instr
*mov
= nir_alu_instr_create(shader
, nir_op_imov
);
993 mov
->dest
.write_mask
= (1 << swizzle
->type
->vector_elements
) - 1;
994 mov
->src
[0].src
.is_ssa
= true;
995 mov
->src
[0].src
.ssa
= &intrin
->dest
.ssa
;
997 mov
->src
[0].swizzle
[0] = swizzle
->mask
.x
;
998 mov
->src
[0].swizzle
[1] = swizzle
->mask
.y
;
999 mov
->src
[0].swizzle
[2] = swizzle
->mask
.z
;
1000 mov
->src
[0].swizzle
[3] = swizzle
->mask
.w
;
1001 for (unsigned i
= deref
->type
->vector_elements
; i
< 4; i
++)
1002 mov
->src
[0].swizzle
[i
] = 0;
1004 add_instr(&mov
->instr
, swizzle
->type
->vector_elements
);
1015 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1016 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1018 glsl_base_type types
[4];
1019 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1021 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1023 types
[i
] = GLSL_TYPE_FLOAT
;
1025 glsl_base_type out_type
;
1027 out_type
= ir
->type
->base_type
;
1029 out_type
= GLSL_TYPE_FLOAT
;
1031 unsigned dest_size
= ir
->type
->vector_elements
;
1033 nir_alu_instr
*instr
;
1036 switch (ir
->operation
) {
1037 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
1038 case ir_unop_logic_not
:
1039 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1042 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fneg
: nir_op_ineg
,
1046 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fabs
: nir_op_iabs
,
1049 case ir_unop_saturate
:
1050 assert(types
[0] == GLSL_TYPE_FLOAT
);
1051 instr
= emit(nir_op_fsat
, dest_size
, srcs
);
1054 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1057 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1058 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1059 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1060 case ir_unop_exp
: emit(nir_op_fexp
, dest_size
, srcs
); break;
1061 case ir_unop_log
: emit(nir_op_flog
, dest_size
, srcs
); break;
1062 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1063 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1065 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1068 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1071 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1073 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1074 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1075 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1076 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1077 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1080 case ir_unop_bitcast_i2f
:
1081 case ir_unop_bitcast_f2i
:
1082 case ir_unop_bitcast_u2f
:
1083 case ir_unop_bitcast_f2u
:
1085 emit(nir_op_imov
, dest_size
, srcs
);
1088 switch (ir
->operands
[0]->type
->vector_elements
) {
1090 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1094 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1098 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1102 unreachable("not reached");
1105 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1106 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1107 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1108 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1109 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1110 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1111 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1112 case ir_unop_sin_reduced
:
1113 emit(nir_op_fsin_reduced
, dest_size
, srcs
);
1115 case ir_unop_cos_reduced
:
1116 emit(nir_op_fcos_reduced
, dest_size
, srcs
);
1118 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1119 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1120 case ir_unop_dFdx_fine
: emit(nir_op_fddx_fine
, dest_size
, srcs
); break;
1121 case ir_unop_dFdy_fine
: emit(nir_op_fddy_fine
, dest_size
, srcs
); break;
1122 case ir_unop_dFdx_coarse
: emit(nir_op_fddx_coarse
, dest_size
, srcs
); break;
1123 case ir_unop_dFdy_coarse
: emit(nir_op_fddy_coarse
, dest_size
, srcs
); break;
1124 case ir_unop_pack_snorm_2x16
:
1125 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1127 case ir_unop_pack_snorm_4x8
:
1128 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1130 case ir_unop_pack_unorm_2x16
:
1131 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1133 case ir_unop_pack_unorm_4x8
:
1134 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1136 case ir_unop_pack_half_2x16
:
1137 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1139 case ir_unop_unpack_snorm_2x16
:
1140 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1142 case ir_unop_unpack_snorm_4x8
:
1143 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1145 case ir_unop_unpack_unorm_2x16
:
1146 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1148 case ir_unop_unpack_unorm_4x8
:
1149 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1151 case ir_unop_unpack_half_2x16
:
1152 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1154 case ir_unop_unpack_half_2x16_split_x
:
1155 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1157 case ir_unop_unpack_half_2x16_split_y
:
1158 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1160 case ir_unop_bitfield_reverse
:
1161 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1163 case ir_unop_bit_count
:
1164 emit(nir_op_bit_count
, dest_size
, srcs
);
1166 case ir_unop_find_msb
:
1168 case GLSL_TYPE_UINT
:
1169 emit(nir_op_ufind_msb
, dest_size
, srcs
);
1172 emit(nir_op_ifind_msb
, dest_size
, srcs
);
1175 unreachable("Invalid type for findMSB()");
1178 case ir_unop_find_lsb
:
1179 emit(nir_op_find_lsb
, dest_size
, srcs
);
1183 switch (ir
->type
->vector_elements
) {
1185 switch (ir
->operands
[0]->type
->vector_elements
) {
1186 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1187 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1188 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1189 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1190 default: unreachable("not reached");
1194 switch (ir
->operands
[0]->type
->vector_elements
) {
1195 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1196 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1197 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1198 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1199 default: unreachable("not reached");
1203 switch (ir
->operands
[0]->type
->vector_elements
) {
1204 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1205 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1206 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1207 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1208 default: unreachable("not reached");
1212 switch (ir
->operands
[0]->type
->vector_elements
) {
1213 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1214 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1215 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1216 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1217 default: unreachable("not reached");
1221 unreachable("not reached");
1232 case ir_binop_bit_and
:
1233 case ir_binop_bit_or
:
1234 case ir_binop_bit_xor
:
1235 case ir_binop_lshift
:
1236 case ir_binop_rshift
:
1237 switch (ir
->operation
) {
1239 if (out_type
== GLSL_TYPE_FLOAT
)
1245 if (out_type
== GLSL_TYPE_FLOAT
)
1251 if (out_type
== GLSL_TYPE_FLOAT
)
1257 if (out_type
== GLSL_TYPE_FLOAT
)
1259 else if (out_type
== GLSL_TYPE_INT
)
1265 if (out_type
== GLSL_TYPE_FLOAT
)
1271 if (out_type
== GLSL_TYPE_FLOAT
)
1273 else if (out_type
== GLSL_TYPE_INT
)
1279 if (out_type
== GLSL_TYPE_FLOAT
)
1281 else if (out_type
== GLSL_TYPE_INT
)
1286 case ir_binop_bit_and
:
1289 case ir_binop_bit_or
:
1292 case ir_binop_bit_xor
:
1295 case ir_binop_lshift
:
1298 case ir_binop_rshift
:
1299 if (out_type
== GLSL_TYPE_INT
)
1309 unreachable("not reached");
1312 instr
= emit(op
, dest_size
, srcs
);
1314 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1315 ir
->operands
[1]->type
->vector_elements
== 1) {
1316 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1318 instr
->src
[1].swizzle
[i
] = 0;
1322 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1323 ir
->operands
[0]->type
->vector_elements
== 1) {
1324 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1326 instr
->src
[0].swizzle
[i
] = 0;
1331 case ir_binop_imul_high
:
1332 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1335 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1336 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1338 if (supports_ints
) {
1339 if (types
[0] == GLSL_TYPE_FLOAT
)
1340 emit(nir_op_flt
, dest_size
, srcs
);
1341 else if (types
[0] == GLSL_TYPE_INT
)
1342 emit(nir_op_ilt
, dest_size
, srcs
);
1344 emit(nir_op_ult
, dest_size
, srcs
);
1346 emit(nir_op_slt
, dest_size
, srcs
);
1349 case ir_binop_greater
:
1350 if (supports_ints
) {
1351 if (types
[0] == GLSL_TYPE_FLOAT
)
1352 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1353 else if (types
[0] == GLSL_TYPE_INT
)
1354 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1356 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1358 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1361 case ir_binop_lequal
:
1362 if (supports_ints
) {
1363 if (types
[0] == GLSL_TYPE_FLOAT
)
1364 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1365 else if (types
[0] == GLSL_TYPE_INT
)
1366 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1368 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1370 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1373 case ir_binop_gequal
:
1374 if (supports_ints
) {
1375 if (types
[0] == GLSL_TYPE_FLOAT
)
1376 emit(nir_op_fge
, dest_size
, srcs
);
1377 else if (types
[0] == GLSL_TYPE_INT
)
1378 emit(nir_op_ige
, dest_size
, srcs
);
1380 emit(nir_op_uge
, dest_size
, srcs
);
1382 emit(nir_op_slt
, dest_size
, srcs
);
1385 case ir_binop_equal
:
1386 if (supports_ints
) {
1387 if (types
[0] == GLSL_TYPE_FLOAT
)
1388 emit(nir_op_feq
, dest_size
, srcs
);
1390 emit(nir_op_ieq
, dest_size
, srcs
);
1392 emit(nir_op_seq
, dest_size
, srcs
);
1395 case ir_binop_nequal
:
1396 if (supports_ints
) {
1397 if (types
[0] == GLSL_TYPE_FLOAT
)
1398 emit(nir_op_fne
, dest_size
, srcs
);
1400 emit(nir_op_ine
, dest_size
, srcs
);
1402 emit(nir_op_sne
, dest_size
, srcs
);
1405 case ir_binop_all_equal
:
1406 if (supports_ints
) {
1407 if (types
[0] == GLSL_TYPE_FLOAT
) {
1408 switch (ir
->operands
[0]->type
->vector_elements
) {
1409 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1410 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1411 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1412 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1414 unreachable("not reached");
1417 switch (ir
->operands
[0]->type
->vector_elements
) {
1418 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1419 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1420 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1421 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1423 unreachable("not reached");
1427 switch (ir
->operands
[0]->type
->vector_elements
) {
1428 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1429 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1430 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1431 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1433 unreachable("not reached");
1437 case ir_binop_any_nequal
:
1438 if (supports_ints
) {
1439 if (types
[0] == GLSL_TYPE_FLOAT
) {
1440 switch (ir
->operands
[0]->type
->vector_elements
) {
1441 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1442 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1443 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1444 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1446 unreachable("not reached");
1449 switch (ir
->operands
[0]->type
->vector_elements
) {
1450 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1451 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1452 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1453 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1455 unreachable("not reached");
1459 switch (ir
->operands
[0]->type
->vector_elements
) {
1460 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1461 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1462 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1463 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1465 unreachable("not reached");
1469 case ir_binop_logic_and
:
1471 emit(nir_op_iand
, dest_size
, srcs
);
1473 emit(nir_op_fand
, dest_size
, srcs
);
1475 case ir_binop_logic_or
:
1477 emit(nir_op_ior
, dest_size
, srcs
);
1479 emit(nir_op_for
, dest_size
, srcs
);
1481 case ir_binop_logic_xor
:
1483 emit(nir_op_ixor
, dest_size
, srcs
);
1485 emit(nir_op_fxor
, dest_size
, srcs
);
1488 switch (ir
->operands
[0]->type
->vector_elements
) {
1489 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1490 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1491 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1493 unreachable("not reached");
1497 case ir_binop_pack_half_2x16_split
:
1498 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1500 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1501 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1502 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1504 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1505 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1506 ir
->operands
[2]->type
->vector_elements
== 1) {
1507 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1509 instr
->src
[2].swizzle
[i
] = 0;
1515 emit(nir_op_bcsel
, dest_size
, srcs
);
1517 emit(nir_op_fcsel
, dest_size
, srcs
);
1520 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1521 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1522 instr
->src
[0].swizzle
[i
] = 0;
1525 case ir_triop_bitfield_extract
:
1526 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1527 nir_op_ubitfield_extract
, dest_size
, srcs
);
1528 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1529 instr
->src
[1].swizzle
[i
] = 0;
1530 instr
->src
[2].swizzle
[i
] = 0;
1533 case ir_quadop_bitfield_insert
:
1534 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1535 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1536 instr
->src
[2].swizzle
[i
] = 0;
1537 instr
->src
[3].swizzle
[i
] = 0;
1540 case ir_quadop_vector
:
1541 switch (ir
->type
->vector_elements
) {
1542 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1543 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1544 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1545 default: unreachable("not reached");
1550 unreachable("not reached");
1555 nir_visitor::visit(ir_swizzle
*ir
)
1557 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1558 ir
->type
->vector_elements
,
1559 evaluate_rvalue(ir
->val
));
1561 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1562 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1563 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1567 nir_visitor::visit(ir_texture
*ir
)
1574 num_srcs
= 1; /* coordinate */
1579 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1580 num_srcs
= 2; /* coordinate, bias/lod */
1584 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1590 if (ir
->lod_info
.lod
!= NULL
)
1591 num_srcs
= 2; /* coordinate, lod */
1593 num_srcs
= 1; /* coordinate */
1597 op
= nir_texop_txf_ms
;
1598 num_srcs
= 2; /* coordinate, sample_index */
1603 if (ir
->lod_info
.lod
!= NULL
)
1604 num_srcs
= 1; /* lod */
1611 num_srcs
= 1; /* coordinate */
1616 num_srcs
= 1; /* coordinate */
1619 case ir_query_levels
:
1620 op
= nir_texop_query_levels
;
1625 unreachable("not reached");
1628 if (ir
->projector
!= NULL
)
1630 if (ir
->shadow_comparitor
!= NULL
)
1632 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1635 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1638 instr
->sampler_dim
=
1639 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1640 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1641 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1642 if (instr
->is_shadow
)
1643 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1644 switch (ir
->type
->base_type
) {
1645 case GLSL_TYPE_FLOAT
:
1646 instr
->dest_type
= nir_type_float
;
1649 instr
->dest_type
= nir_type_int
;
1651 case GLSL_TYPE_UINT
:
1652 instr
->dest_type
= nir_type_unsigned
;
1655 unreachable("not reached");
1658 ir
->sampler
->accept(this);
1659 instr
->sampler
= this->deref_head
;
1661 unsigned src_number
= 0;
1663 if (ir
->coordinate
!= NULL
) {
1664 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1665 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->coordinate
);
1666 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1670 if (ir
->projector
!= NULL
) {
1671 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->projector
);
1672 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1676 if (ir
->shadow_comparitor
!= NULL
) {
1677 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->shadow_comparitor
);
1678 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1682 if (ir
->offset
!= NULL
) {
1683 /* we don't support multiple offsets yet */
1684 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1686 ir_constant
*const_offset
= ir
->offset
->as_constant();
1687 if (const_offset
!= NULL
) {
1688 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1689 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1691 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->offset
);
1692 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1699 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.bias
);
1700 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1707 if (ir
->lod_info
.lod
!= NULL
) {
1708 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.lod
);
1709 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1715 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdx
);
1716 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1718 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdy
);
1719 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1724 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.sample_index
);
1725 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1730 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1737 assert(src_number
== num_srcs
);
1739 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
1743 nir_visitor::visit(ir_constant
*ir
)
1746 * We don't know if this variable is an an array or struct that gets
1747 * dereferenced, so do the safe thing an make it a variable with a
1748 * constant initializer and return a dereference.
1751 nir_variable
*var
= ralloc(this->shader
, nir_variable
);
1752 var
->name
= ralloc_strdup(var
, "const_temp");
1753 var
->type
= ir
->type
;
1754 var
->data
.mode
= nir_var_local
;
1755 var
->data
.read_only
= true;
1756 var
->constant_initializer
= constant_copy(ir
, var
);
1757 exec_list_push_tail(&this->impl
->locals
, &var
->node
);
1759 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1760 this->deref_tail
= &this->deref_head
->deref
;
1764 nir_visitor::visit(ir_dereference_variable
*ir
)
1766 struct hash_entry
*entry
=
1767 _mesa_hash_table_search(this->var_table
, ir
->var
);
1769 nir_variable
*var
= (nir_variable
*) entry
->data
;
1771 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
1772 this->deref_head
= deref
;
1773 this->deref_tail
= &deref
->deref
;
1777 nir_visitor::visit(ir_dereference_record
*ir
)
1779 ir
->record
->accept(this);
1781 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
1782 assert(field_index
>= 0);
1784 nir_deref_struct
*deref
= nir_deref_struct_create(this->shader
, field_index
);
1785 deref
->deref
.type
= ir
->type
;
1786 this->deref_tail
->child
= &deref
->deref
;
1787 this->deref_tail
= &deref
->deref
;
1791 nir_visitor::visit(ir_dereference_array
*ir
)
1793 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
1794 deref
->deref
.type
= ir
->type
;
1796 ir_constant
*const_index
= ir
->array_index
->as_constant();
1797 if (const_index
!= NULL
) {
1798 deref
->deref_array_type
= nir_deref_array_type_direct
;
1799 deref
->base_offset
= const_index
->value
.u
[0];
1801 deref
->deref_array_type
= nir_deref_array_type_indirect
;
1802 deref
->indirect
= evaluate_rvalue(ir
->array_index
);
1805 ir
->array
->accept(this);
1807 this->deref_tail
->child
= &deref
->deref
;
1808 this->deref_tail
= &deref
->deref
;