2 * Copyright © 2014 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Connor Abbott (cwabbott0@gmail.com)
28 #include "glsl_to_nir.h"
29 #include "ir_visitor.h"
30 #include "ir_hierarchical_visitor.h"
34 * pass to lower GLSL IR to NIR
36 * This will lower variable dereferences to loads/stores of corresponding
37 * variables in NIR - the variables will be converted to registers in a later
43 class nir_visitor
: public ir_visitor
46 nir_visitor(nir_shader
*shader
, gl_shader_stage stage
);
49 virtual void visit(ir_variable
*);
50 virtual void visit(ir_function
*);
51 virtual void visit(ir_function_signature
*);
52 virtual void visit(ir_loop
*);
53 virtual void visit(ir_if
*);
54 virtual void visit(ir_discard
*);
55 virtual void visit(ir_loop_jump
*);
56 virtual void visit(ir_return
*);
57 virtual void visit(ir_call
*);
58 virtual void visit(ir_assignment
*);
59 virtual void visit(ir_emit_vertex
*);
60 virtual void visit(ir_end_primitive
*);
61 virtual void visit(ir_expression
*);
62 virtual void visit(ir_swizzle
*);
63 virtual void visit(ir_texture
*);
64 virtual void visit(ir_constant
*);
65 virtual void visit(ir_dereference_variable
*);
66 virtual void visit(ir_dereference_record
*);
67 virtual void visit(ir_dereference_array
*);
68 virtual void visit(ir_barrier
*);
70 void create_function(ir_function
*ir
);
73 void create_overload(ir_function_signature
*ir
, nir_function
*function
);
74 void add_instr(nir_instr
*instr
, unsigned num_components
);
75 nir_src
evaluate_rvalue(ir_rvalue
*ir
);
77 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
);
78 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
);
79 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
,
81 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
,
82 nir_src src2
, nir_src src3
);
87 gl_shader_stage stage
;
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
, options
);
136 nir_visitor
v1(shader
, sh
->Stage
);
137 nir_function_visitor
v2(&v1
);
139 visit_exec_list(sh
->ir
, &v1
);
144 nir_visitor::nir_visitor(nir_shader
*shader
, gl_shader_stage stage
)
146 this->supports_ints
= shader
->options
->native_integers
;
147 this->shader
= shader
;
149 this->is_global
= true;
150 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
151 _mesa_key_pointer_equal
);
152 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
153 _mesa_key_pointer_equal
);
156 nir_visitor::~nir_visitor()
158 _mesa_hash_table_destroy(this->var_table
, NULL
);
159 _mesa_hash_table_destroy(this->overload_table
, NULL
);
163 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
166 ralloc_steal(mem_ctx
, this->deref_head
);
167 return this->deref_head
;
170 static nir_constant
*
171 constant_copy(ir_constant
*ir
, void *mem_ctx
)
176 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
178 unsigned total_elems
= ir
->type
->components();
180 switch (ir
->type
->base_type
) {
182 for (i
= 0; i
< total_elems
; i
++)
183 ret
->value
.u
[i
] = ir
->value
.u
[i
];
187 for (i
= 0; i
< total_elems
; i
++)
188 ret
->value
.i
[i
] = ir
->value
.i
[i
];
191 case GLSL_TYPE_FLOAT
:
192 for (i
= 0; i
< total_elems
; i
++)
193 ret
->value
.f
[i
] = ir
->value
.f
[i
];
197 for (i
= 0; i
< total_elems
; i
++)
198 ret
->value
.b
[i
] = ir
->value
.b
[i
];
201 case GLSL_TYPE_STRUCT
:
202 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
205 foreach_in_list(ir_constant
, field
, &ir
->components
) {
206 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
211 case GLSL_TYPE_ARRAY
:
212 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
215 for (i
= 0; i
< ir
->type
->length
; i
++)
216 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
220 unreachable("not reached");
227 nir_visitor::visit(ir_variable
*ir
)
229 nir_variable
*var
= ralloc(shader
, nir_variable
);
230 var
->type
= ir
->type
;
231 var
->name
= ralloc_strdup(var
, ir
->name
);
233 if (ir
->is_interface_instance() && ir
->get_max_ifc_array_access() != NULL
) {
234 unsigned size
= ir
->get_interface_type()->length
;
235 var
->max_ifc_array_access
= ralloc_array(var
, unsigned, size
);
236 memcpy(var
->max_ifc_array_access
, ir
->get_max_ifc_array_access(),
237 size
* sizeof(unsigned));
239 var
->max_ifc_array_access
= NULL
;
242 var
->data
.read_only
= ir
->data
.read_only
;
243 var
->data
.centroid
= ir
->data
.centroid
;
244 var
->data
.sample
= ir
->data
.sample
;
245 var
->data
.invariant
= ir
->data
.invariant
;
246 var
->data
.location
= ir
->data
.location
;
248 switch(ir
->data
.mode
) {
250 case ir_var_temporary
:
252 var
->data
.mode
= nir_var_global
;
254 var
->data
.mode
= nir_var_local
;
257 case ir_var_function_in
:
258 case ir_var_function_out
:
259 case ir_var_function_inout
:
260 case ir_var_const_in
:
261 var
->data
.mode
= nir_var_local
;
264 case ir_var_shader_in
:
265 if (stage
== MESA_SHADER_FRAGMENT
&&
266 ir
->data
.location
== VARYING_SLOT_FACE
) {
267 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
268 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
269 var
->data
.mode
= nir_var_system_value
;
271 var
->data
.mode
= nir_var_shader_in
;
275 case ir_var_shader_out
:
276 var
->data
.mode
= nir_var_shader_out
;
280 var
->data
.mode
= nir_var_uniform
;
284 case ir_var_system_value
:
285 var
->data
.mode
= nir_var_system_value
;
289 unreachable("not reached");
292 var
->data
.interpolation
= ir
->data
.interpolation
;
293 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
294 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
295 var
->data
.explicit_location
= ir
->data
.explicit_location
;
296 var
->data
.explicit_index
= ir
->data
.explicit_index
;
297 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
298 var
->data
.has_initializer
= ir
->data
.has_initializer
;
299 var
->data
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
300 var
->data
.location_frac
= ir
->data
.location_frac
;
301 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
302 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
304 switch (ir
->data
.depth_layout
) {
305 case ir_depth_layout_none
:
306 var
->data
.depth_layout
= nir_depth_layout_none
;
308 case ir_depth_layout_any
:
309 var
->data
.depth_layout
= nir_depth_layout_any
;
311 case ir_depth_layout_greater
:
312 var
->data
.depth_layout
= nir_depth_layout_greater
;
314 case ir_depth_layout_less
:
315 var
->data
.depth_layout
= nir_depth_layout_less
;
317 case ir_depth_layout_unchanged
:
318 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
321 unreachable("not reached");
324 var
->data
.index
= ir
->data
.index
;
325 var
->data
.binding
= ir
->data
.binding
;
326 /* XXX Get rid of buffer_index */
327 var
->data
.atomic
.buffer_index
= ir
->data
.binding
;
328 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
329 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
330 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
331 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
332 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
333 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
334 var
->data
.image
.format
= ir
->data
.image_format
;
335 var
->data
.max_array_access
= ir
->data
.max_array_access
;
337 var
->num_state_slots
= ir
->get_num_state_slots();
338 if (var
->num_state_slots
> 0) {
339 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
340 var
->num_state_slots
);
342 ir_state_slot
*state_slots
= ir
->get_state_slots();
343 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
344 for (unsigned j
= 0; j
< 5; j
++)
345 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
346 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
349 var
->state_slots
= NULL
;
352 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
354 var
->interface_type
= ir
->get_interface_type();
356 switch (var
->data
.mode
) {
358 exec_list_push_tail(&impl
->locals
, &var
->node
);
362 exec_list_push_tail(&shader
->globals
, &var
->node
);
365 case nir_var_shader_in
:
366 exec_list_push_tail(&shader
->inputs
, &var
->node
);
369 case nir_var_shader_out
:
370 exec_list_push_tail(&shader
->outputs
, &var
->node
);
373 case nir_var_uniform
:
374 exec_list_push_tail(&shader
->uniforms
, &var
->node
);
377 case nir_var_system_value
:
378 exec_list_push_tail(&shader
->system_values
, &var
->node
);
382 unreachable("not reached");
385 _mesa_hash_table_insert(var_table
, ir
, var
);
390 nir_function_visitor::visit_enter(ir_function
*ir
)
392 visitor
->create_function(ir
);
393 return visit_continue_with_parent
;
398 nir_visitor::create_function(ir_function
*ir
)
400 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
401 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
402 create_overload(sig
, func
);
409 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
411 if (ir
->is_intrinsic
)
414 nir_function_overload
*overload
= nir_function_overload_create(function
);
416 unsigned num_params
= ir
->parameters
.length();
417 overload
->num_params
= num_params
;
418 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
421 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
422 switch (param
->data
.mode
) {
423 case ir_var_function_in
:
424 overload
->params
[i
].param_type
= nir_parameter_in
;
427 case ir_var_function_out
:
428 overload
->params
[i
].param_type
= nir_parameter_out
;
431 case ir_var_function_inout
:
432 overload
->params
[i
].param_type
= nir_parameter_inout
;
436 unreachable("not reached");
439 overload
->params
[i
].type
= param
->type
;
443 overload
->return_type
= ir
->return_type
;
445 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
449 nir_visitor::visit(ir_function
*ir
)
451 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
456 nir_visitor::visit(ir_function_signature
*ir
)
458 if (ir
->is_intrinsic
)
461 struct hash_entry
*entry
=
462 _mesa_hash_table_search(this->overload_table
, ir
);
465 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
467 if (ir
->is_defined
) {
468 nir_function_impl
*impl
= nir_function_impl_create(overload
);
471 unsigned num_params
= overload
->num_params
;
472 impl
->num_params
= num_params
;
473 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
475 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
477 impl
->params
[i
] = this->var
;
481 if (overload
->return_type
== glsl_type::void_type
) {
482 impl
->return_var
= NULL
;
484 impl
->return_var
= ralloc(this->shader
, nir_variable
);
485 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
487 impl
->return_var
->type
= overload
->return_type
;
490 this->is_global
= false;
492 this->cf_node_list
= &impl
->body
;
493 visit_exec_list(&ir
->body
, this);
495 this->is_global
= true;
497 overload
->impl
= NULL
;
502 nir_visitor::visit(ir_loop
*ir
)
504 exec_list
*old_list
= this->cf_node_list
;
506 nir_loop
*loop
= nir_loop_create(this->shader
);
507 nir_cf_node_insert_end(old_list
, &loop
->cf_node
);
508 this->cf_node_list
= &loop
->body
;
509 visit_exec_list(&ir
->body_instructions
, this);
511 this->cf_node_list
= old_list
;
515 nir_visitor::visit(ir_if
*ir
)
517 nir_src condition
= evaluate_rvalue(ir
->condition
);
519 exec_list
*old_list
= this->cf_node_list
;
521 nir_if
*if_stmt
= nir_if_create(this->shader
);
522 if_stmt
->condition
= condition
;
523 nir_cf_node_insert_end(old_list
, &if_stmt
->cf_node
);
525 this->cf_node_list
= &if_stmt
->then_list
;
526 visit_exec_list(&ir
->then_instructions
, this);
528 this->cf_node_list
= &if_stmt
->else_list
;
529 visit_exec_list(&ir
->else_instructions
, this);
531 this->cf_node_list
= old_list
;
535 nir_visitor::visit(ir_discard
*ir
)
538 * discards aren't treated as control flow, because before we lower them
539 * they can appear anywhere in the shader and the stuff after them may still
540 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
541 * discards will be immediately followed by a return.
544 nir_intrinsic_instr
*discard
;
546 discard
= nir_intrinsic_instr_create(this->shader
,
547 nir_intrinsic_discard_if
);
548 discard
->src
[0] = evaluate_rvalue(ir
->condition
);
550 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
552 nir_instr_insert_after_cf_list(this->cf_node_list
, &discard
->instr
);
556 nir_visitor::visit(ir_emit_vertex
*ir
)
558 nir_intrinsic_instr
*instr
=
559 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
560 instr
->const_index
[0] = ir
->stream_id();
561 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
565 nir_visitor::visit(ir_end_primitive
*ir
)
567 nir_intrinsic_instr
*instr
=
568 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
569 instr
->const_index
[0] = ir
->stream_id();
570 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
574 nir_visitor::visit(ir_loop_jump
*ir
)
578 case ir_loop_jump::jump_break
:
579 type
= nir_jump_break
;
581 case ir_loop_jump::jump_continue
:
582 type
= nir_jump_continue
;
585 unreachable("not reached");
588 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
589 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
593 nir_visitor::visit(ir_return
*ir
)
595 if (ir
->value
!= NULL
) {
596 nir_intrinsic_instr
*copy
=
597 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
599 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
600 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
603 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
604 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
608 nir_visitor::visit(ir_call
*ir
)
610 if (ir
->callee
->is_intrinsic
) {
612 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
613 op
= nir_intrinsic_atomic_counter_read_var
;
614 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
615 op
= nir_intrinsic_atomic_counter_inc_var
;
616 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
617 op
= nir_intrinsic_atomic_counter_dec_var
;
618 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
619 op
= nir_intrinsic_image_load
;
620 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
621 op
= nir_intrinsic_image_store
;
622 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
623 op
= nir_intrinsic_image_atomic_add
;
624 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
625 op
= nir_intrinsic_image_atomic_min
;
626 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
627 op
= nir_intrinsic_image_atomic_max
;
628 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
629 op
= nir_intrinsic_image_atomic_and
;
630 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
631 op
= nir_intrinsic_image_atomic_or
;
632 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
633 op
= nir_intrinsic_image_atomic_xor
;
634 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
635 op
= nir_intrinsic_image_atomic_exchange
;
636 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
637 op
= nir_intrinsic_image_atomic_comp_swap
;
638 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
639 op
= nir_intrinsic_memory_barrier
;
641 unreachable("not reached");
644 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
647 case nir_intrinsic_atomic_counter_read_var
:
648 case nir_intrinsic_atomic_counter_inc_var
:
649 case nir_intrinsic_atomic_counter_dec_var
: {
650 ir_dereference
*param
=
651 (ir_dereference
*) ir
->actual_parameters
.get_head();
652 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
653 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
656 case nir_intrinsic_image_load
:
657 case nir_intrinsic_image_store
:
658 case nir_intrinsic_image_atomic_add
:
659 case nir_intrinsic_image_atomic_min
:
660 case nir_intrinsic_image_atomic_max
:
661 case nir_intrinsic_image_atomic_and
:
662 case nir_intrinsic_image_atomic_or
:
663 case nir_intrinsic_image_atomic_xor
:
664 case nir_intrinsic_image_atomic_exchange
:
665 case nir_intrinsic_image_atomic_comp_swap
: {
666 nir_ssa_undef_instr
*instr_undef
=
667 nir_ssa_undef_instr_create(shader
, 1);
668 nir_instr_insert_after_cf_list(this->cf_node_list
,
669 &instr_undef
->instr
);
671 /* Set the image variable dereference. */
672 exec_node
*param
= ir
->actual_parameters
.get_head();
673 ir_dereference
*image
= (ir_dereference
*)param
;
674 const glsl_type
*type
=
675 image
->variable_referenced()->type
->without_array();
677 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
678 param
= param
->get_next();
680 /* Set the address argument, extending the coordinate vector to four
683 const nir_src src_addr
= evaluate_rvalue((ir_dereference
*)param
);
684 nir_alu_instr
*instr_addr
= nir_alu_instr_create(shader
, nir_op_vec4
);
685 nir_ssa_dest_init(&instr_addr
->instr
, &instr_addr
->dest
.dest
, 4, NULL
);
687 for (int i
= 0; i
< 4; i
++) {
688 if (i
< type
->coordinate_components()) {
689 instr_addr
->src
[i
].src
= src_addr
;
690 instr_addr
->src
[i
].swizzle
[0] = i
;
692 instr_addr
->src
[i
].src
= nir_src_for_ssa(&instr_undef
->def
);
696 nir_instr_insert_after_cf_list(cf_node_list
, &instr_addr
->instr
);
697 instr
->src
[0] = nir_src_for_ssa(&instr_addr
->dest
.dest
.ssa
);
698 param
= param
->get_next();
700 /* Set the sample argument, which is undefined for single-sample
703 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
704 instr
->src
[1] = evaluate_rvalue((ir_dereference
*)param
);
705 param
= param
->get_next();
707 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
710 /* Set the intrinsic parameters. */
711 if (!param
->is_tail_sentinel()) {
712 instr
->src
[2] = evaluate_rvalue((ir_dereference
*)param
);
713 param
= param
->get_next();
716 if (!param
->is_tail_sentinel()) {
717 instr
->src
[3] = evaluate_rvalue((ir_dereference
*)param
);
718 param
= param
->get_next();
721 /* Set the intrinsic destination. */
722 if (ir
->return_deref
)
723 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
724 ir
->return_deref
->type
->vector_elements
, NULL
);
727 case nir_intrinsic_memory_barrier
:
730 unreachable("not reached");
733 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
735 if (ir
->return_deref
) {
736 nir_intrinsic_instr
*store_instr
=
737 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
738 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
740 store_instr
->variables
[0] =
741 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
742 store_instr
->src
[0] = nir_src_for_ssa(&instr
->dest
.ssa
);
744 nir_instr_insert_after_cf_list(this->cf_node_list
,
745 &store_instr
->instr
);
751 struct hash_entry
*entry
=
752 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
754 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
756 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
759 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
760 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
764 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
765 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
769 nir_visitor::visit(ir_assignment
*ir
)
771 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
773 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
774 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
775 /* We're doing a plain-as-can-be copy, so emit a copy_var */
776 nir_intrinsic_instr
*copy
=
777 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
779 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
780 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
783 nir_if
*if_stmt
= nir_if_create(this->shader
);
784 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
785 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
786 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
788 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
793 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
795 ir
->lhs
->accept(this);
796 nir_deref_var
*lhs_deref
= this->deref_head
;
797 nir_src src
= evaluate_rvalue(ir
->rhs
);
799 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
801 * We have no good way to update only part of a variable, so just load
802 * the LHS and do a vec operation to combine the old with the new, and
804 * back into the LHS. Copy propagation should get rid of the mess.
807 nir_intrinsic_instr
*load
=
808 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
809 load
->num_components
= ir
->lhs
->type
->vector_elements
;
810 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
811 load
->variables
[0] = lhs_deref
;
812 ralloc_steal(load
, load
->variables
[0]);
813 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
816 switch (ir
->lhs
->type
->vector_elements
) {
817 case 1: vec_op
= nir_op_imov
; break;
818 case 2: vec_op
= nir_op_vec2
; break;
819 case 3: vec_op
= nir_op_vec3
; break;
820 case 4: vec_op
= nir_op_vec4
; break;
821 default: unreachable("Invalid number of components"); break;
823 nir_alu_instr
*vec
= nir_alu_instr_create(this->shader
, vec_op
);
824 nir_ssa_dest_init(&vec
->instr
, &vec
->dest
.dest
, num_components
, NULL
);
825 vec
->dest
.write_mask
= (1 << num_components
) - 1;
827 unsigned component
= 0;
828 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
829 if (ir
->write_mask
& (1 << i
)) {
830 vec
->src
[i
].src
= src
;
832 /* GLSL IR will give us the input to the write-masked assignment
833 * in a single packed vector. So, for example, if the
834 * writemask is xzw, then we have to swizzle x -> x, y -> z,
835 * and z -> w and get the y component from the load.
837 vec
->src
[i
].swizzle
[0] = component
++;
839 vec
->src
[i
].src
.is_ssa
= true;
840 vec
->src
[i
].src
.ssa
= &load
->dest
.ssa
;
841 vec
->src
[i
].swizzle
[0] = i
;
845 nir_instr_insert_after_cf_list(this->cf_node_list
, &vec
->instr
);
848 src
.ssa
= &vec
->dest
.dest
.ssa
;
851 nir_intrinsic_instr
*store
=
852 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
853 store
->num_components
= ir
->lhs
->type
->vector_elements
;
854 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
855 store
->variables
[0] = nir_deref_as_var(store_deref
);
859 nir_if
*if_stmt
= nir_if_create(this->shader
);
860 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
861 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
862 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
864 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
869 * Given an instruction, returns a pointer to its destination or NULL if there
872 * Note that this only handles instructions we generate at this level.
875 get_instr_dest(nir_instr
*instr
)
877 nir_alu_instr
*alu_instr
;
878 nir_intrinsic_instr
*intrinsic_instr
;
879 nir_tex_instr
*tex_instr
;
881 switch (instr
->type
) {
882 case nir_instr_type_alu
:
883 alu_instr
= nir_instr_as_alu(instr
);
884 return &alu_instr
->dest
.dest
;
886 case nir_instr_type_intrinsic
:
887 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
888 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
889 return &intrinsic_instr
->dest
;
893 case nir_instr_type_tex
:
894 tex_instr
= nir_instr_as_tex(instr
);
895 return &tex_instr
->dest
;
898 unreachable("not reached");
905 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
907 nir_dest
*dest
= get_instr_dest(instr
);
909 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
911 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
912 this->result
= instr
;
916 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
919 if (ir
->as_dereference() || ir
->as_constant()) {
921 * A dereference is being used on the right hand side, which means we
922 * must emit a variable load.
925 nir_intrinsic_instr
*load_instr
=
926 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
927 load_instr
->num_components
= ir
->type
->vector_elements
;
928 load_instr
->variables
[0] = this->deref_head
;
929 ralloc_steal(load_instr
, load_instr
->variables
[0]);
930 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
933 nir_dest
*dest
= get_instr_dest(this->result
);
934 assert(dest
->is_ssa
);
936 return nir_src_for_ssa(&dest
->ssa
);
940 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
942 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
943 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
944 instr
->src
[i
].src
= srcs
[i
];
945 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
946 add_instr(&instr
->instr
, dest_size
);
951 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
953 assert(nir_op_infos
[op
].num_inputs
== 1);
954 return emit(op
, dest_size
, &src1
);
958 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
961 assert(nir_op_infos
[op
].num_inputs
== 2);
962 nir_src srcs
[] = { src1
, src2
};
963 return emit(op
, dest_size
, srcs
);
967 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
968 nir_src src2
, nir_src src3
)
970 assert(nir_op_infos
[op
].num_inputs
== 3);
971 nir_src srcs
[] = { src1
, src2
, src3
};
972 return emit(op
, dest_size
, srcs
);
976 nir_visitor::visit(ir_expression
*ir
)
978 /* Some special cases */
979 switch (ir
->operation
) {
980 case ir_binop_ubo_load
: {
981 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
985 op
= nir_intrinsic_load_ubo
;
987 op
= nir_intrinsic_load_ubo_indirect
;
989 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
990 load
->num_components
= ir
->type
->vector_elements
;
991 load
->const_index
[0] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
992 load
->const_index
[1] = 1; /* number of vec4's */
993 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
995 load
->src
[1] = evaluate_rvalue(ir
->operands
[1]);
996 add_instr(&load
->instr
, ir
->type
->vector_elements
);
999 * In UBO's, a true boolean value is any non-zero value, but we consider
1000 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1003 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
1004 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
, 1);
1005 const_zero
->value
.u
[0] = 0;
1006 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
1008 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
1009 compare
->src
[0].src
.is_ssa
= true;
1010 compare
->src
[0].src
.ssa
= &load
->dest
.ssa
;
1011 compare
->src
[1].src
.is_ssa
= true;
1012 compare
->src
[1].src
.ssa
= &const_zero
->def
;
1013 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1014 compare
->src
[1].swizzle
[i
] = 0;
1015 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
1017 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
1023 case ir_unop_interpolate_at_centroid
:
1024 case ir_binop_interpolate_at_offset
:
1025 case ir_binop_interpolate_at_sample
: {
1026 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1027 ir_swizzle
*swizzle
= NULL
;
1029 /* the api does not allow a swizzle here, but the varying packing code
1030 * may have pushed one into here.
1032 swizzle
= ir
->operands
[0]->as_swizzle();
1034 deref
= swizzle
->val
->as_dereference();
1038 deref
->accept(this);
1040 nir_intrinsic_op op
;
1041 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1042 switch (ir
->operation
) {
1043 case ir_unop_interpolate_at_centroid
:
1044 op
= nir_intrinsic_interp_var_at_centroid
;
1046 case ir_binop_interpolate_at_offset
:
1047 op
= nir_intrinsic_interp_var_at_offset
;
1049 case ir_binop_interpolate_at_sample
:
1050 op
= nir_intrinsic_interp_var_at_sample
;
1053 unreachable("Invalid interpolation intrinsic");
1056 /* This case can happen if the vertex shader does not write the
1057 * given varying. In this case, the linker will lower it to a
1058 * global variable. Since interpolating a variable makes no
1059 * sense, we'll just turn it into a load which will probably
1060 * eventually end up as an SSA definition.
1062 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1063 op
= nir_intrinsic_load_var
;
1066 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1067 intrin
->num_components
= deref
->type
->vector_elements
;
1068 intrin
->variables
[0] = this->deref_head
;
1069 ralloc_steal(intrin
, intrin
->variables
[0]);
1071 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1072 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1073 intrin
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
1075 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
1078 nir_alu_instr
*mov
= nir_alu_instr_create(shader
, nir_op_imov
);
1079 mov
->dest
.write_mask
= (1 << swizzle
->type
->vector_elements
) - 1;
1080 mov
->src
[0].src
.is_ssa
= true;
1081 mov
->src
[0].src
.ssa
= &intrin
->dest
.ssa
;
1083 mov
->src
[0].swizzle
[0] = swizzle
->mask
.x
;
1084 mov
->src
[0].swizzle
[1] = swizzle
->mask
.y
;
1085 mov
->src
[0].swizzle
[2] = swizzle
->mask
.z
;
1086 mov
->src
[0].swizzle
[3] = swizzle
->mask
.w
;
1087 for (unsigned i
= deref
->type
->vector_elements
; i
< 4; i
++)
1088 mov
->src
[0].swizzle
[i
] = 0;
1090 add_instr(&mov
->instr
, swizzle
->type
->vector_elements
);
1101 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1102 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1104 glsl_base_type types
[4];
1105 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1107 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1109 types
[i
] = GLSL_TYPE_FLOAT
;
1111 glsl_base_type out_type
;
1113 out_type
= ir
->type
->base_type
;
1115 out_type
= GLSL_TYPE_FLOAT
;
1117 unsigned dest_size
= ir
->type
->vector_elements
;
1119 nir_alu_instr
*instr
;
1122 switch (ir
->operation
) {
1123 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
1124 case ir_unop_logic_not
:
1125 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1128 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fneg
: nir_op_ineg
,
1132 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fabs
: nir_op_iabs
,
1135 case ir_unop_saturate
:
1136 assert(types
[0] == GLSL_TYPE_FLOAT
);
1137 instr
= emit(nir_op_fsat
, dest_size
, srcs
);
1140 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1143 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1144 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1145 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1146 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1147 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1148 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1149 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1151 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1154 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1157 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1159 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1160 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1161 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1162 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1163 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1166 case ir_unop_bitcast_i2f
:
1167 case ir_unop_bitcast_f2i
:
1168 case ir_unop_bitcast_u2f
:
1169 case ir_unop_bitcast_f2u
:
1171 emit(nir_op_imov
, dest_size
, srcs
);
1174 switch (ir
->operands
[0]->type
->vector_elements
) {
1176 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1180 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1184 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1188 unreachable("not reached");
1191 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1192 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1193 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1194 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1195 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1196 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1197 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1198 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1199 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1200 case ir_unop_dFdx_fine
: emit(nir_op_fddx_fine
, dest_size
, srcs
); break;
1201 case ir_unop_dFdy_fine
: emit(nir_op_fddy_fine
, dest_size
, srcs
); break;
1202 case ir_unop_dFdx_coarse
: emit(nir_op_fddx_coarse
, dest_size
, srcs
); break;
1203 case ir_unop_dFdy_coarse
: emit(nir_op_fddy_coarse
, dest_size
, srcs
); break;
1204 case ir_unop_pack_snorm_2x16
:
1205 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1207 case ir_unop_pack_snorm_4x8
:
1208 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1210 case ir_unop_pack_unorm_2x16
:
1211 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1213 case ir_unop_pack_unorm_4x8
:
1214 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1216 case ir_unop_pack_half_2x16
:
1217 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1219 case ir_unop_unpack_snorm_2x16
:
1220 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1222 case ir_unop_unpack_snorm_4x8
:
1223 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1225 case ir_unop_unpack_unorm_2x16
:
1226 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1228 case ir_unop_unpack_unorm_4x8
:
1229 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1231 case ir_unop_unpack_half_2x16
:
1232 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1234 case ir_unop_unpack_half_2x16_split_x
:
1235 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1237 case ir_unop_unpack_half_2x16_split_y
:
1238 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1240 case ir_unop_bitfield_reverse
:
1241 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1243 case ir_unop_bit_count
:
1244 emit(nir_op_bit_count
, dest_size
, srcs
);
1246 case ir_unop_find_msb
:
1248 case GLSL_TYPE_UINT
:
1249 emit(nir_op_ufind_msb
, dest_size
, srcs
);
1252 emit(nir_op_ifind_msb
, dest_size
, srcs
);
1255 unreachable("Invalid type for findMSB()");
1258 case ir_unop_find_lsb
:
1259 emit(nir_op_find_lsb
, dest_size
, srcs
);
1263 switch (ir
->type
->vector_elements
) {
1265 switch (ir
->operands
[0]->type
->vector_elements
) {
1266 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1267 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1268 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1269 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1270 default: unreachable("not reached");
1274 switch (ir
->operands
[0]->type
->vector_elements
) {
1275 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1276 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1277 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1278 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1279 default: unreachable("not reached");
1283 switch (ir
->operands
[0]->type
->vector_elements
) {
1284 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1285 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1286 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1287 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1288 default: unreachable("not reached");
1292 switch (ir
->operands
[0]->type
->vector_elements
) {
1293 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1294 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1295 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1296 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1297 default: unreachable("not reached");
1301 unreachable("not reached");
1312 case ir_binop_bit_and
:
1313 case ir_binop_bit_or
:
1314 case ir_binop_bit_xor
:
1315 case ir_binop_logic_and
:
1316 case ir_binop_logic_or
:
1317 case ir_binop_logic_xor
:
1318 case ir_binop_lshift
:
1319 case ir_binop_rshift
:
1320 switch (ir
->operation
) {
1322 if (out_type
== GLSL_TYPE_FLOAT
)
1328 if (out_type
== GLSL_TYPE_FLOAT
)
1334 if (out_type
== GLSL_TYPE_FLOAT
)
1340 if (out_type
== GLSL_TYPE_FLOAT
)
1342 else if (out_type
== GLSL_TYPE_INT
)
1348 if (out_type
== GLSL_TYPE_FLOAT
)
1354 if (out_type
== GLSL_TYPE_FLOAT
)
1356 else if (out_type
== GLSL_TYPE_INT
)
1362 if (out_type
== GLSL_TYPE_FLOAT
)
1364 else if (out_type
== GLSL_TYPE_INT
)
1369 case ir_binop_bit_and
:
1372 case ir_binop_bit_or
:
1375 case ir_binop_bit_xor
:
1378 case ir_binop_logic_and
:
1384 case ir_binop_logic_or
:
1390 case ir_binop_logic_xor
:
1396 case ir_binop_lshift
:
1399 case ir_binop_rshift
:
1400 if (out_type
== GLSL_TYPE_INT
)
1410 unreachable("not reached");
1413 instr
= emit(op
, dest_size
, srcs
);
1415 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1416 ir
->operands
[1]->type
->vector_elements
== 1) {
1417 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1419 instr
->src
[1].swizzle
[i
] = 0;
1423 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1424 ir
->operands
[0]->type
->vector_elements
== 1) {
1425 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1427 instr
->src
[0].swizzle
[i
] = 0;
1432 case ir_binop_imul_high
:
1433 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1436 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1437 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1439 if (supports_ints
) {
1440 if (types
[0] == GLSL_TYPE_FLOAT
)
1441 emit(nir_op_flt
, dest_size
, srcs
);
1442 else if (types
[0] == GLSL_TYPE_INT
)
1443 emit(nir_op_ilt
, dest_size
, srcs
);
1445 emit(nir_op_ult
, dest_size
, srcs
);
1447 emit(nir_op_slt
, dest_size
, srcs
);
1450 case ir_binop_greater
:
1451 if (supports_ints
) {
1452 if (types
[0] == GLSL_TYPE_FLOAT
)
1453 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1454 else if (types
[0] == GLSL_TYPE_INT
)
1455 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1457 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1459 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1462 case ir_binop_lequal
:
1463 if (supports_ints
) {
1464 if (types
[0] == GLSL_TYPE_FLOAT
)
1465 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1466 else if (types
[0] == GLSL_TYPE_INT
)
1467 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1469 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1471 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1474 case ir_binop_gequal
:
1475 if (supports_ints
) {
1476 if (types
[0] == GLSL_TYPE_FLOAT
)
1477 emit(nir_op_fge
, dest_size
, srcs
);
1478 else if (types
[0] == GLSL_TYPE_INT
)
1479 emit(nir_op_ige
, dest_size
, srcs
);
1481 emit(nir_op_uge
, dest_size
, srcs
);
1483 emit(nir_op_slt
, dest_size
, srcs
);
1486 case ir_binop_equal
:
1487 if (supports_ints
) {
1488 if (types
[0] == GLSL_TYPE_FLOAT
)
1489 emit(nir_op_feq
, dest_size
, srcs
);
1491 emit(nir_op_ieq
, dest_size
, srcs
);
1493 emit(nir_op_seq
, dest_size
, srcs
);
1496 case ir_binop_nequal
:
1497 if (supports_ints
) {
1498 if (types
[0] == GLSL_TYPE_FLOAT
)
1499 emit(nir_op_fne
, dest_size
, srcs
);
1501 emit(nir_op_ine
, dest_size
, srcs
);
1503 emit(nir_op_sne
, dest_size
, srcs
);
1506 case ir_binop_all_equal
:
1507 if (supports_ints
) {
1508 if (types
[0] == GLSL_TYPE_FLOAT
) {
1509 switch (ir
->operands
[0]->type
->vector_elements
) {
1510 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1511 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1512 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1513 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1515 unreachable("not reached");
1518 switch (ir
->operands
[0]->type
->vector_elements
) {
1519 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1520 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1521 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1522 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1524 unreachable("not reached");
1528 switch (ir
->operands
[0]->type
->vector_elements
) {
1529 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1530 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1531 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1532 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1534 unreachable("not reached");
1538 case ir_binop_any_nequal
:
1539 if (supports_ints
) {
1540 if (types
[0] == GLSL_TYPE_FLOAT
) {
1541 switch (ir
->operands
[0]->type
->vector_elements
) {
1542 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1543 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1544 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1545 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1547 unreachable("not reached");
1550 switch (ir
->operands
[0]->type
->vector_elements
) {
1551 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1552 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1553 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1554 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1556 unreachable("not reached");
1560 switch (ir
->operands
[0]->type
->vector_elements
) {
1561 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1562 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1563 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1564 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1566 unreachable("not reached");
1571 switch (ir
->operands
[0]->type
->vector_elements
) {
1572 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1573 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1574 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1576 unreachable("not reached");
1580 case ir_binop_pack_half_2x16_split
:
1581 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1583 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1584 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1585 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1587 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1588 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1589 ir
->operands
[2]->type
->vector_elements
== 1) {
1590 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1592 instr
->src
[2].swizzle
[i
] = 0;
1598 emit(nir_op_bcsel
, dest_size
, srcs
);
1600 emit(nir_op_fcsel
, dest_size
, srcs
);
1603 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1604 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1605 instr
->src
[0].swizzle
[i
] = 0;
1608 case ir_triop_bitfield_extract
:
1609 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1610 nir_op_ubitfield_extract
, dest_size
, srcs
);
1611 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1612 instr
->src
[1].swizzle
[i
] = 0;
1613 instr
->src
[2].swizzle
[i
] = 0;
1616 case ir_quadop_bitfield_insert
:
1617 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1618 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1619 instr
->src
[2].swizzle
[i
] = 0;
1620 instr
->src
[3].swizzle
[i
] = 0;
1623 case ir_quadop_vector
:
1624 switch (ir
->type
->vector_elements
) {
1625 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1626 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1627 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1628 default: unreachable("not reached");
1633 unreachable("not reached");
1638 nir_visitor::visit(ir_swizzle
*ir
)
1640 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1641 ir
->type
->vector_elements
,
1642 evaluate_rvalue(ir
->val
));
1644 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1645 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1646 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1650 nir_visitor::visit(ir_texture
*ir
)
1657 num_srcs
= 1; /* coordinate */
1662 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1663 num_srcs
= 2; /* coordinate, bias/lod */
1667 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1673 if (ir
->lod_info
.lod
!= NULL
)
1674 num_srcs
= 2; /* coordinate, lod */
1676 num_srcs
= 1; /* coordinate */
1680 op
= nir_texop_txf_ms
;
1681 num_srcs
= 2; /* coordinate, sample_index */
1686 if (ir
->lod_info
.lod
!= NULL
)
1687 num_srcs
= 1; /* lod */
1694 num_srcs
= 1; /* coordinate */
1699 num_srcs
= 1; /* coordinate */
1702 case ir_query_levels
:
1703 op
= nir_texop_query_levels
;
1708 unreachable("not reached");
1711 if (ir
->projector
!= NULL
)
1713 if (ir
->shadow_comparitor
!= NULL
)
1715 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1718 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1721 instr
->sampler_dim
=
1722 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1723 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1724 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1725 if (instr
->is_shadow
)
1726 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1727 switch (ir
->type
->base_type
) {
1728 case GLSL_TYPE_FLOAT
:
1729 instr
->dest_type
= nir_type_float
;
1732 instr
->dest_type
= nir_type_int
;
1734 case GLSL_TYPE_UINT
:
1735 instr
->dest_type
= nir_type_unsigned
;
1738 unreachable("not reached");
1741 instr
->sampler
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1743 unsigned src_number
= 0;
1745 if (ir
->coordinate
!= NULL
) {
1746 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1747 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->coordinate
);
1748 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1752 if (ir
->projector
!= NULL
) {
1753 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->projector
);
1754 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1758 if (ir
->shadow_comparitor
!= NULL
) {
1759 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->shadow_comparitor
);
1760 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1764 if (ir
->offset
!= NULL
) {
1765 /* we don't support multiple offsets yet */
1766 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1768 ir_constant
*const_offset
= ir
->offset
->as_constant();
1769 if (const_offset
!= NULL
) {
1770 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1771 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1773 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->offset
);
1774 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1781 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.bias
);
1782 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1789 if (ir
->lod_info
.lod
!= NULL
) {
1790 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.lod
);
1791 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1797 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdx
);
1798 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1800 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdy
);
1801 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1806 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.sample_index
);
1807 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1812 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1819 assert(src_number
== num_srcs
);
1821 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
1825 nir_visitor::visit(ir_constant
*ir
)
1828 * We don't know if this variable is an an array or struct that gets
1829 * dereferenced, so do the safe thing an make it a variable with a
1830 * constant initializer and return a dereference.
1833 nir_variable
*var
= ralloc(this->shader
, nir_variable
);
1834 var
->name
= ralloc_strdup(var
, "const_temp");
1835 var
->type
= ir
->type
;
1836 var
->data
.mode
= nir_var_local
;
1837 var
->data
.read_only
= true;
1838 var
->constant_initializer
= constant_copy(ir
, var
);
1839 exec_list_push_tail(&this->impl
->locals
, &var
->node
);
1841 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1842 this->deref_tail
= &this->deref_head
->deref
;
1846 nir_visitor::visit(ir_dereference_variable
*ir
)
1848 struct hash_entry
*entry
=
1849 _mesa_hash_table_search(this->var_table
, ir
->var
);
1851 nir_variable
*var
= (nir_variable
*) entry
->data
;
1853 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
1854 this->deref_head
= deref
;
1855 this->deref_tail
= &deref
->deref
;
1859 nir_visitor::visit(ir_dereference_record
*ir
)
1861 ir
->record
->accept(this);
1863 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
1864 assert(field_index
>= 0);
1866 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
1867 deref
->deref
.type
= ir
->type
;
1868 this->deref_tail
->child
= &deref
->deref
;
1869 this->deref_tail
= &deref
->deref
;
1873 nir_visitor::visit(ir_dereference_array
*ir
)
1875 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
1876 deref
->deref
.type
= ir
->type
;
1878 ir_constant
*const_index
= ir
->array_index
->as_constant();
1879 if (const_index
!= NULL
) {
1880 deref
->deref_array_type
= nir_deref_array_type_direct
;
1881 deref
->base_offset
= const_index
->value
.u
[0];
1883 deref
->deref_array_type
= nir_deref_array_type_indirect
;
1884 deref
->indirect
= evaluate_rvalue(ir
->array_index
);
1887 ir
->array
->accept(this);
1889 this->deref_tail
->child
= &deref
->deref
;
1890 ralloc_steal(this->deref_tail
, deref
);
1891 this->deref_tail
= &deref
->deref
;
1895 nir_visitor::visit(ir_barrier
*ir
)
1897 unreachable("Not implemented!");