2 * Copyright © 2014 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Connor Abbott (cwabbott0@gmail.com)
28 #include "glsl_to_nir.h"
29 #include "nir_control_flow.h"
30 #include "ir_visitor.h"
31 #include "ir_hierarchical_visitor.h"
35 * pass to lower GLSL IR to NIR
37 * This will lower variable dereferences to loads/stores of corresponding
38 * variables in NIR - the variables will be converted to registers in a later
44 class nir_visitor
: public ir_visitor
47 nir_visitor(nir_shader
*shader
, gl_shader
*sh
);
50 virtual void visit(ir_variable
*);
51 virtual void visit(ir_function
*);
52 virtual void visit(ir_function_signature
*);
53 virtual void visit(ir_loop
*);
54 virtual void visit(ir_if
*);
55 virtual void visit(ir_discard
*);
56 virtual void visit(ir_loop_jump
*);
57 virtual void visit(ir_return
*);
58 virtual void visit(ir_call
*);
59 virtual void visit(ir_assignment
*);
60 virtual void visit(ir_emit_vertex
*);
61 virtual void visit(ir_end_primitive
*);
62 virtual void visit(ir_expression
*);
63 virtual void visit(ir_swizzle
*);
64 virtual void visit(ir_texture
*);
65 virtual void visit(ir_constant
*);
66 virtual void visit(ir_dereference_variable
*);
67 virtual void visit(ir_dereference_record
*);
68 virtual void visit(ir_dereference_array
*);
69 virtual void visit(ir_barrier
*);
71 void create_function(ir_function
*ir
);
74 void create_overload(ir_function_signature
*ir
, nir_function
*function
);
75 void add_instr(nir_instr
*instr
, unsigned num_components
);
76 nir_src
evaluate_rvalue(ir_rvalue
*ir
);
78 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
);
79 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
);
80 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
,
82 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
,
83 nir_src src2
, nir_src src3
);
90 nir_function_impl
*impl
;
91 exec_list
*cf_node_list
;
92 nir_instr
*result
; /* result of the expression tree last visited */
94 nir_deref_var
*evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
);
96 /* the head of the dereference chain we're creating */
97 nir_deref_var
*deref_head
;
98 /* the tail of the dereference chain we're creating */
99 nir_deref
*deref_tail
;
101 nir_variable
*var
; /* variable created by ir_variable visitor */
103 /* whether the IR we're operating on is per-function or global */
106 /* map of ir_variable -> nir_variable */
107 struct hash_table
*var_table
;
109 /* map of ir_function_signature -> nir_function_overload */
110 struct hash_table
*overload_table
;
114 * This visitor runs before the main visitor, calling create_function() for
115 * each function so that the main visitor can resolve forward references in
119 class nir_function_visitor
: public ir_hierarchical_visitor
122 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
125 virtual ir_visitor_status
visit_enter(ir_function
*);
128 nir_visitor
*visitor
;
131 }; /* end of anonymous namespace */
134 glsl_to_nir(struct gl_shader
*sh
, const nir_shader_compiler_options
*options
)
136 nir_shader
*shader
= nir_shader_create(NULL
, sh
->Stage
, options
);
138 nir_visitor
v1(shader
, sh
);
139 nir_function_visitor
v2(&v1
);
141 visit_exec_list(sh
->ir
, &v1
);
143 nir_lower_outputs_to_temporaries(shader
);
145 shader
->gs
.vertices_out
= sh
->Geom
.VerticesOut
;
146 shader
->gs
.invocations
= sh
->Geom
.Invocations
;
151 nir_visitor::nir_visitor(nir_shader
*shader
, gl_shader
*sh
)
153 this->supports_ints
= shader
->options
->native_integers
;
154 this->shader
= shader
;
156 this->is_global
= true;
157 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
158 _mesa_key_pointer_equal
);
159 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
160 _mesa_key_pointer_equal
);
163 nir_visitor::~nir_visitor()
165 _mesa_hash_table_destroy(this->var_table
, NULL
);
166 _mesa_hash_table_destroy(this->overload_table
, NULL
);
170 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
173 ralloc_steal(mem_ctx
, this->deref_head
);
174 return this->deref_head
;
177 static nir_constant
*
178 constant_copy(ir_constant
*ir
, void *mem_ctx
)
183 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
185 unsigned total_elems
= ir
->type
->components();
187 switch (ir
->type
->base_type
) {
189 for (i
= 0; i
< total_elems
; i
++)
190 ret
->value
.u
[i
] = ir
->value
.u
[i
];
194 for (i
= 0; i
< total_elems
; i
++)
195 ret
->value
.i
[i
] = ir
->value
.i
[i
];
198 case GLSL_TYPE_FLOAT
:
199 for (i
= 0; i
< total_elems
; i
++)
200 ret
->value
.f
[i
] = ir
->value
.f
[i
];
204 for (i
= 0; i
< total_elems
; i
++)
205 ret
->value
.b
[i
] = ir
->value
.b
[i
];
208 case GLSL_TYPE_STRUCT
:
209 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
212 foreach_in_list(ir_constant
, field
, &ir
->components
) {
213 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
218 case GLSL_TYPE_ARRAY
:
219 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
222 for (i
= 0; i
< ir
->type
->length
; i
++)
223 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
227 unreachable("not reached");
234 nir_visitor::visit(ir_variable
*ir
)
236 nir_variable
*var
= ralloc(shader
, nir_variable
);
237 var
->type
= ir
->type
;
238 var
->name
= ralloc_strdup(var
, ir
->name
);
240 if (ir
->is_interface_instance() && ir
->get_max_ifc_array_access() != NULL
) {
241 unsigned size
= ir
->get_interface_type()->length
;
242 var
->max_ifc_array_access
= ralloc_array(var
, unsigned, size
);
243 memcpy(var
->max_ifc_array_access
, ir
->get_max_ifc_array_access(),
244 size
* sizeof(unsigned));
246 var
->max_ifc_array_access
= NULL
;
249 var
->data
.read_only
= ir
->data
.read_only
;
250 var
->data
.centroid
= ir
->data
.centroid
;
251 var
->data
.sample
= ir
->data
.sample
;
252 var
->data
.invariant
= ir
->data
.invariant
;
253 var
->data
.location
= ir
->data
.location
;
255 switch(ir
->data
.mode
) {
257 case ir_var_temporary
:
259 var
->data
.mode
= nir_var_global
;
261 var
->data
.mode
= nir_var_local
;
264 case ir_var_function_in
:
265 case ir_var_function_out
:
266 case ir_var_function_inout
:
267 case ir_var_const_in
:
268 var
->data
.mode
= nir_var_local
;
271 case ir_var_shader_in
:
272 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
273 ir
->data
.location
== VARYING_SLOT_FACE
) {
274 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
275 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
276 var
->data
.mode
= nir_var_system_value
;
278 var
->data
.mode
= nir_var_shader_in
;
282 case ir_var_shader_out
:
283 var
->data
.mode
= nir_var_shader_out
;
287 var
->data
.mode
= nir_var_uniform
;
290 case ir_var_shader_storage
:
291 var
->data
.mode
= nir_var_shader_storage
;
294 case ir_var_system_value
:
295 var
->data
.mode
= nir_var_system_value
;
299 unreachable("not reached");
302 var
->data
.interpolation
= ir
->data
.interpolation
;
303 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
304 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
305 var
->data
.explicit_location
= ir
->data
.explicit_location
;
306 var
->data
.explicit_index
= ir
->data
.explicit_index
;
307 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
308 var
->data
.has_initializer
= ir
->data
.has_initializer
;
309 var
->data
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
310 var
->data
.location_frac
= ir
->data
.location_frac
;
311 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
312 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
314 switch (ir
->data
.depth_layout
) {
315 case ir_depth_layout_none
:
316 var
->data
.depth_layout
= nir_depth_layout_none
;
318 case ir_depth_layout_any
:
319 var
->data
.depth_layout
= nir_depth_layout_any
;
321 case ir_depth_layout_greater
:
322 var
->data
.depth_layout
= nir_depth_layout_greater
;
324 case ir_depth_layout_less
:
325 var
->data
.depth_layout
= nir_depth_layout_less
;
327 case ir_depth_layout_unchanged
:
328 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
331 unreachable("not reached");
334 var
->data
.index
= ir
->data
.index
;
335 var
->data
.descriptor_set
= 0;
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 exec_list_push_tail(&shader
->inputs
, &var
->node
);
380 case nir_var_shader_out
:
381 exec_list_push_tail(&shader
->outputs
, &var
->node
);
384 case nir_var_uniform
:
385 case nir_var_shader_storage
:
386 exec_list_push_tail(&shader
->uniforms
, &var
->node
);
389 case nir_var_system_value
:
390 exec_list_push_tail(&shader
->system_values
, &var
->node
);
394 unreachable("not reached");
397 _mesa_hash_table_insert(var_table
, ir
, var
);
402 nir_function_visitor::visit_enter(ir_function
*ir
)
404 visitor
->create_function(ir
);
405 return visit_continue_with_parent
;
410 nir_visitor::create_function(ir_function
*ir
)
412 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
413 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
414 create_overload(sig
, func
);
421 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
423 if (ir
->is_intrinsic
)
426 nir_function_overload
*overload
= nir_function_overload_create(function
);
428 unsigned num_params
= ir
->parameters
.length();
429 overload
->num_params
= num_params
;
430 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
433 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
434 switch (param
->data
.mode
) {
435 case ir_var_function_in
:
436 overload
->params
[i
].param_type
= nir_parameter_in
;
439 case ir_var_function_out
:
440 overload
->params
[i
].param_type
= nir_parameter_out
;
443 case ir_var_function_inout
:
444 overload
->params
[i
].param_type
= nir_parameter_inout
;
448 unreachable("not reached");
451 overload
->params
[i
].type
= param
->type
;
455 overload
->return_type
= ir
->return_type
;
457 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
461 nir_visitor::visit(ir_function
*ir
)
463 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
468 nir_visitor::visit(ir_function_signature
*ir
)
470 if (ir
->is_intrinsic
)
473 struct hash_entry
*entry
=
474 _mesa_hash_table_search(this->overload_table
, ir
);
477 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
479 if (ir
->is_defined
) {
480 nir_function_impl
*impl
= nir_function_impl_create(overload
);
483 unsigned num_params
= overload
->num_params
;
484 impl
->num_params
= num_params
;
485 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
487 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
489 impl
->params
[i
] = this->var
;
493 if (overload
->return_type
== glsl_type::void_type
) {
494 impl
->return_var
= NULL
;
496 impl
->return_var
= ralloc(this->shader
, nir_variable
);
497 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
499 impl
->return_var
->type
= overload
->return_type
;
502 this->is_global
= false;
504 this->cf_node_list
= &impl
->body
;
505 visit_exec_list(&ir
->body
, this);
507 this->is_global
= true;
509 overload
->impl
= NULL
;
514 nir_visitor::visit(ir_loop
*ir
)
516 exec_list
*old_list
= this->cf_node_list
;
518 nir_loop
*loop
= nir_loop_create(this->shader
);
519 nir_cf_node_insert_end(old_list
, &loop
->cf_node
);
520 this->cf_node_list
= &loop
->body
;
521 visit_exec_list(&ir
->body_instructions
, this);
523 this->cf_node_list
= old_list
;
527 nir_visitor::visit(ir_if
*ir
)
529 nir_src condition
= evaluate_rvalue(ir
->condition
);
531 exec_list
*old_list
= this->cf_node_list
;
533 nir_if
*if_stmt
= nir_if_create(this->shader
);
534 if_stmt
->condition
= condition
;
535 nir_cf_node_insert_end(old_list
, &if_stmt
->cf_node
);
537 this->cf_node_list
= &if_stmt
->then_list
;
538 visit_exec_list(&ir
->then_instructions
, this);
540 this->cf_node_list
= &if_stmt
->else_list
;
541 visit_exec_list(&ir
->else_instructions
, this);
543 this->cf_node_list
= old_list
;
547 nir_visitor::visit(ir_discard
*ir
)
550 * discards aren't treated as control flow, because before we lower them
551 * they can appear anywhere in the shader and the stuff after them may still
552 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
553 * discards will be immediately followed by a return.
556 nir_intrinsic_instr
*discard
;
558 discard
= nir_intrinsic_instr_create(this->shader
,
559 nir_intrinsic_discard_if
);
560 discard
->src
[0] = evaluate_rvalue(ir
->condition
);
562 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
564 nir_instr_insert_after_cf_list(this->cf_node_list
, &discard
->instr
);
568 nir_visitor::visit(ir_emit_vertex
*ir
)
570 nir_intrinsic_instr
*instr
=
571 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
572 instr
->const_index
[0] = ir
->stream_id();
573 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
577 nir_visitor::visit(ir_end_primitive
*ir
)
579 nir_intrinsic_instr
*instr
=
580 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
581 instr
->const_index
[0] = ir
->stream_id();
582 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
586 nir_visitor::visit(ir_loop_jump
*ir
)
590 case ir_loop_jump::jump_break
:
591 type
= nir_jump_break
;
593 case ir_loop_jump::jump_continue
:
594 type
= nir_jump_continue
;
597 unreachable("not reached");
600 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
601 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
605 nir_visitor::visit(ir_return
*ir
)
607 if (ir
->value
!= NULL
) {
608 nir_intrinsic_instr
*copy
=
609 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
611 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
612 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
615 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
616 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
620 nir_visitor::visit(ir_call
*ir
)
622 if (ir
->callee
->is_intrinsic
) {
624 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
625 op
= nir_intrinsic_atomic_counter_read_var
;
626 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
627 op
= nir_intrinsic_atomic_counter_inc_var
;
628 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
629 op
= nir_intrinsic_atomic_counter_dec_var
;
630 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
631 op
= nir_intrinsic_image_load
;
632 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
633 op
= nir_intrinsic_image_store
;
634 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
635 op
= nir_intrinsic_image_atomic_add
;
636 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
637 op
= nir_intrinsic_image_atomic_min
;
638 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
639 op
= nir_intrinsic_image_atomic_max
;
640 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
641 op
= nir_intrinsic_image_atomic_and
;
642 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
643 op
= nir_intrinsic_image_atomic_or
;
644 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
645 op
= nir_intrinsic_image_atomic_xor
;
646 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
647 op
= nir_intrinsic_image_atomic_exchange
;
648 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
649 op
= nir_intrinsic_image_atomic_comp_swap
;
650 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
651 op
= nir_intrinsic_memory_barrier
;
652 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_size") == 0) {
653 op
= nir_intrinsic_image_size
;
654 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_samples") == 0) {
655 op
= nir_intrinsic_image_samples
;
656 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_ssbo") == 0) {
657 op
= nir_intrinsic_store_ssbo
;
658 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_ssbo") == 0) {
659 op
= nir_intrinsic_load_ssbo
;
660 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_add_internal") == 0) {
661 op
= nir_intrinsic_ssbo_atomic_add
;
662 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_and_internal") == 0) {
663 op
= nir_intrinsic_ssbo_atomic_and
;
664 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_or_internal") == 0) {
665 op
= nir_intrinsic_ssbo_atomic_or
;
666 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_xor_internal") == 0) {
667 op
= nir_intrinsic_ssbo_atomic_xor
;
668 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_min_internal") == 0) {
669 op
= nir_intrinsic_ssbo_atomic_min
;
670 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_max_internal") == 0) {
671 op
= nir_intrinsic_ssbo_atomic_max
;
672 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_exchange_internal") == 0) {
673 op
= nir_intrinsic_ssbo_atomic_exchange
;
674 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_comp_swap_internal") == 0) {
675 op
= nir_intrinsic_ssbo_atomic_comp_swap
;
677 unreachable("not reached");
680 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
681 nir_dest
*dest
= &instr
->dest
;
684 case nir_intrinsic_atomic_counter_read_var
:
685 case nir_intrinsic_atomic_counter_inc_var
:
686 case nir_intrinsic_atomic_counter_dec_var
: {
687 ir_dereference
*param
=
688 (ir_dereference
*) ir
->actual_parameters
.get_head();
689 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
690 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
691 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
694 case nir_intrinsic_image_load
:
695 case nir_intrinsic_image_store
:
696 case nir_intrinsic_image_atomic_add
:
697 case nir_intrinsic_image_atomic_min
:
698 case nir_intrinsic_image_atomic_max
:
699 case nir_intrinsic_image_atomic_and
:
700 case nir_intrinsic_image_atomic_or
:
701 case nir_intrinsic_image_atomic_xor
:
702 case nir_intrinsic_image_atomic_exchange
:
703 case nir_intrinsic_image_atomic_comp_swap
:
704 case nir_intrinsic_image_samples
:
705 case nir_intrinsic_image_size
: {
706 nir_ssa_undef_instr
*instr_undef
=
707 nir_ssa_undef_instr_create(shader
, 1);
708 nir_instr_insert_after_cf_list(this->cf_node_list
,
709 &instr_undef
->instr
);
711 /* Set the image variable dereference. */
712 exec_node
*param
= ir
->actual_parameters
.get_head();
713 ir_dereference
*image
= (ir_dereference
*)param
;
714 const glsl_type
*type
=
715 image
->variable_referenced()->type
->without_array();
717 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
718 param
= param
->get_next();
720 /* Set the intrinsic destination. */
721 if (ir
->return_deref
) {
722 const nir_intrinsic_info
*info
=
723 &nir_intrinsic_infos
[instr
->intrinsic
];
724 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
725 info
->dest_components
, NULL
);
728 if (op
== nir_intrinsic_image_size
||
729 op
== nir_intrinsic_image_samples
) {
730 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
734 /* Set the address argument, extending the coordinate vector to four
737 const nir_src src_addr
= evaluate_rvalue((ir_dereference
*)param
);
738 nir_alu_instr
*instr_addr
= nir_alu_instr_create(shader
, nir_op_vec4
);
739 nir_ssa_dest_init(&instr_addr
->instr
, &instr_addr
->dest
.dest
, 4, NULL
);
741 for (int i
= 0; i
< 4; i
++) {
742 if (i
< type
->coordinate_components()) {
743 instr_addr
->src
[i
].src
= src_addr
;
744 instr_addr
->src
[i
].swizzle
[0] = i
;
746 instr_addr
->src
[i
].src
= nir_src_for_ssa(&instr_undef
->def
);
750 nir_instr_insert_after_cf_list(cf_node_list
, &instr_addr
->instr
);
751 instr
->src
[0] = nir_src_for_ssa(&instr_addr
->dest
.dest
.ssa
);
752 param
= param
->get_next();
754 /* Set the sample argument, which is undefined for single-sample
757 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
758 instr
->src
[1] = evaluate_rvalue((ir_dereference
*)param
);
759 param
= param
->get_next();
761 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
764 /* Set the intrinsic parameters. */
765 if (!param
->is_tail_sentinel()) {
766 instr
->src
[2] = evaluate_rvalue((ir_dereference
*)param
);
767 param
= param
->get_next();
770 if (!param
->is_tail_sentinel()) {
771 instr
->src
[3] = evaluate_rvalue((ir_dereference
*)param
);
772 param
= param
->get_next();
774 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
777 case nir_intrinsic_memory_barrier
:
778 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
780 case nir_intrinsic_store_ssbo
: {
781 exec_node
*param
= ir
->actual_parameters
.get_head();
782 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
784 param
= param
->get_next();
785 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
787 param
= param
->get_next();
788 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
790 param
= param
->get_next();
791 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
794 /* Check if we need the indirect version */
795 ir_constant
*const_offset
= offset
->as_constant();
797 op
= nir_intrinsic_store_ssbo_indirect
;
799 instr
= nir_intrinsic_instr_create(shader
, op
);
800 instr
->src
[2] = evaluate_rvalue(offset
);
801 instr
->const_index
[0] = 0;
804 instr
->const_index
[0] = const_offset
->value
.u
[0];
807 instr
->const_index
[1] = write_mask
->value
.u
[0];
809 instr
->src
[0] = evaluate_rvalue(val
);
810 instr
->num_components
= val
->type
->vector_elements
;
812 instr
->src
[1] = evaluate_rvalue(block
);
813 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
816 case nir_intrinsic_load_ssbo
: {
817 exec_node
*param
= ir
->actual_parameters
.get_head();
818 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
820 param
= param
->get_next();
821 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
823 /* Check if we need the indirect version */
824 ir_constant
*const_offset
= offset
->as_constant();
826 op
= nir_intrinsic_load_ssbo_indirect
;
828 instr
= nir_intrinsic_instr_create(shader
, op
);
829 instr
->src
[1] = evaluate_rvalue(offset
);
830 instr
->const_index
[0] = 0;
833 instr
->const_index
[0] = const_offset
->value
.u
[0];
836 instr
->src
[0] = evaluate_rvalue(block
);
838 const glsl_type
*type
= ir
->return_deref
->var
->type
;
839 instr
->num_components
= type
->vector_elements
;
841 /* Setup destination register */
842 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
843 type
->vector_elements
, NULL
);
845 /* Insert the created nir instruction now since in the case of boolean
846 * result we will need to emit another instruction after it
848 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
851 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
852 * consider a true boolean to be ~0. Fix this up with a != 0
855 if (type
->base_type
== GLSL_TYPE_BOOL
) {
856 nir_load_const_instr
*const_zero
=
857 nir_load_const_instr_create(shader
, 1);
858 const_zero
->value
.u
[0] = 0;
859 nir_instr_insert_after_cf_list(this->cf_node_list
,
862 nir_alu_instr
*load_ssbo_compare
=
863 nir_alu_instr_create(shader
, nir_op_ine
);
864 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
865 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
866 load_ssbo_compare
->src
[1].src
.is_ssa
= true;
867 load_ssbo_compare
->src
[1].src
.ssa
= &const_zero
->def
;
868 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
869 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
870 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
871 &load_ssbo_compare
->dest
.dest
,
872 type
->vector_elements
, NULL
);
873 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
874 nir_instr_insert_after_cf_list(this->cf_node_list
,
875 &load_ssbo_compare
->instr
);
876 dest
= &load_ssbo_compare
->dest
.dest
;
880 case nir_intrinsic_ssbo_atomic_add
:
881 case nir_intrinsic_ssbo_atomic_min
:
882 case nir_intrinsic_ssbo_atomic_max
:
883 case nir_intrinsic_ssbo_atomic_and
:
884 case nir_intrinsic_ssbo_atomic_or
:
885 case nir_intrinsic_ssbo_atomic_xor
:
886 case nir_intrinsic_ssbo_atomic_exchange
:
887 case nir_intrinsic_ssbo_atomic_comp_swap
: {
888 int param_count
= ir
->actual_parameters
.length();
889 assert(param_count
== 3 || param_count
== 4);
892 exec_node
*param
= ir
->actual_parameters
.get_head();
893 ir_instruction
*inst
= (ir_instruction
*) param
;
894 instr
->src
[0] = evaluate_rvalue(inst
->as_rvalue());
897 param
= param
->get_next();
898 inst
= (ir_instruction
*) param
;
899 instr
->src
[1] = evaluate_rvalue(inst
->as_rvalue());
901 /* data1 parameter (this is always present) */
902 param
= param
->get_next();
903 inst
= (ir_instruction
*) param
;
904 instr
->src
[2] = evaluate_rvalue(inst
->as_rvalue());
906 /* data2 parameter (only with atomic_comp_swap) */
907 if (param_count
== 4) {
908 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
909 param
= param
->get_next();
910 inst
= (ir_instruction
*) param
;
911 instr
->src
[3] = evaluate_rvalue(inst
->as_rvalue());
915 assert(ir
->return_deref
);
916 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
917 ir
->return_deref
->type
->vector_elements
, NULL
);
918 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
922 unreachable("not reached");
925 if (ir
->return_deref
) {
926 nir_intrinsic_instr
*store_instr
=
927 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
928 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
930 store_instr
->variables
[0] =
931 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
932 store_instr
->src
[0] = nir_src_for_ssa(&dest
->ssa
);
934 nir_instr_insert_after_cf_list(this->cf_node_list
,
935 &store_instr
->instr
);
941 struct hash_entry
*entry
=
942 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
944 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
946 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
949 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
950 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
954 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
955 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
959 nir_visitor::visit(ir_assignment
*ir
)
961 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
963 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
964 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
965 /* We're doing a plain-as-can-be copy, so emit a copy_var */
966 nir_intrinsic_instr
*copy
=
967 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
969 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
970 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
973 nir_if
*if_stmt
= nir_if_create(this->shader
);
974 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
975 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
976 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
978 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
983 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
985 ir
->lhs
->accept(this);
986 nir_deref_var
*lhs_deref
= this->deref_head
;
987 nir_src src
= evaluate_rvalue(ir
->rhs
);
989 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
991 * We have no good way to update only part of a variable, so just load
992 * the LHS and do a vec operation to combine the old with the new, and
994 * back into the LHS. Copy propagation should get rid of the mess.
997 nir_intrinsic_instr
*load
=
998 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
999 load
->num_components
= ir
->lhs
->type
->vector_elements
;
1000 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
1001 load
->variables
[0] = lhs_deref
;
1002 ralloc_steal(load
, load
->variables
[0]);
1003 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
1006 switch (ir
->lhs
->type
->vector_elements
) {
1007 case 1: vec_op
= nir_op_imov
; break;
1008 case 2: vec_op
= nir_op_vec2
; break;
1009 case 3: vec_op
= nir_op_vec3
; break;
1010 case 4: vec_op
= nir_op_vec4
; break;
1011 default: unreachable("Invalid number of components"); break;
1013 nir_alu_instr
*vec
= nir_alu_instr_create(this->shader
, vec_op
);
1014 nir_ssa_dest_init(&vec
->instr
, &vec
->dest
.dest
, num_components
, NULL
);
1015 vec
->dest
.write_mask
= (1 << num_components
) - 1;
1017 unsigned component
= 0;
1018 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
1019 if (ir
->write_mask
& (1 << i
)) {
1020 vec
->src
[i
].src
= src
;
1022 /* GLSL IR will give us the input to the write-masked assignment
1023 * in a single packed vector. So, for example, if the
1024 * writemask is xzw, then we have to swizzle x -> x, y -> z,
1025 * and z -> w and get the y component from the load.
1027 vec
->src
[i
].swizzle
[0] = component
++;
1029 vec
->src
[i
].src
.is_ssa
= true;
1030 vec
->src
[i
].src
.ssa
= &load
->dest
.ssa
;
1031 vec
->src
[i
].swizzle
[0] = i
;
1035 nir_instr_insert_after_cf_list(this->cf_node_list
, &vec
->instr
);
1038 src
.ssa
= &vec
->dest
.dest
.ssa
;
1041 nir_intrinsic_instr
*store
=
1042 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1043 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1044 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
1045 store
->variables
[0] = nir_deref_as_var(store_deref
);
1046 store
->src
[0] = src
;
1048 if (ir
->condition
) {
1049 nir_if
*if_stmt
= nir_if_create(this->shader
);
1050 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
1051 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
1052 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
1054 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
1059 * Given an instruction, returns a pointer to its destination or NULL if there
1060 * is no destination.
1062 * Note that this only handles instructions we generate at this level.
1065 get_instr_dest(nir_instr
*instr
)
1067 nir_alu_instr
*alu_instr
;
1068 nir_intrinsic_instr
*intrinsic_instr
;
1069 nir_tex_instr
*tex_instr
;
1071 switch (instr
->type
) {
1072 case nir_instr_type_alu
:
1073 alu_instr
= nir_instr_as_alu(instr
);
1074 return &alu_instr
->dest
.dest
;
1076 case nir_instr_type_intrinsic
:
1077 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1078 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1079 return &intrinsic_instr
->dest
;
1083 case nir_instr_type_tex
:
1084 tex_instr
= nir_instr_as_tex(instr
);
1085 return &tex_instr
->dest
;
1088 unreachable("not reached");
1095 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
1097 nir_dest
*dest
= get_instr_dest(instr
);
1100 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
1102 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
1103 this->result
= instr
;
1107 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1110 if (ir
->as_dereference() || ir
->as_constant()) {
1112 * A dereference is being used on the right hand side, which means we
1113 * must emit a variable load.
1116 nir_intrinsic_instr
*load_instr
=
1117 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1118 load_instr
->num_components
= ir
->type
->vector_elements
;
1119 load_instr
->variables
[0] = this->deref_head
;
1120 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1121 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
1124 nir_dest
*dest
= get_instr_dest(this->result
);
1125 assert(dest
->is_ssa
);
1127 return nir_src_for_ssa(&dest
->ssa
);
1131 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
1133 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
1134 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
1135 instr
->src
[i
].src
= srcs
[i
];
1136 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
1137 add_instr(&instr
->instr
, dest_size
);
1142 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
1144 assert(nir_op_infos
[op
].num_inputs
== 1);
1145 return emit(op
, dest_size
, &src1
);
1149 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
1152 assert(nir_op_infos
[op
].num_inputs
== 2);
1153 nir_src srcs
[] = { src1
, src2
};
1154 return emit(op
, dest_size
, srcs
);
1158 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
1159 nir_src src2
, nir_src src3
)
1161 assert(nir_op_infos
[op
].num_inputs
== 3);
1162 nir_src srcs
[] = { src1
, src2
, src3
};
1163 return emit(op
, dest_size
, srcs
);
1167 nir_visitor::visit(ir_expression
*ir
)
1169 /* Some special cases */
1170 switch (ir
->operation
) {
1171 case ir_binop_ubo_load
: {
1172 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
1174 nir_intrinsic_op op
;
1176 op
= nir_intrinsic_load_ubo
;
1178 op
= nir_intrinsic_load_ubo_indirect
;
1181 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
1182 load
->num_components
= ir
->type
->vector_elements
;
1183 load
->const_index
[0] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
1184 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
1186 load
->src
[1] = evaluate_rvalue(ir
->operands
[1]);
1187 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1190 * In UBO's, a true boolean value is any non-zero value, but we consider
1191 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1194 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
1195 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
, 1);
1196 const_zero
->value
.u
[0] = 0;
1197 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
1199 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
1200 compare
->src
[0].src
.is_ssa
= true;
1201 compare
->src
[0].src
.ssa
= &load
->dest
.ssa
;
1202 compare
->src
[1].src
.is_ssa
= true;
1203 compare
->src
[1].src
.ssa
= &const_zero
->def
;
1204 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1205 compare
->src
[1].swizzle
[i
] = 0;
1206 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
1208 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
1214 case ir_unop_interpolate_at_centroid
:
1215 case ir_binop_interpolate_at_offset
:
1216 case ir_binop_interpolate_at_sample
: {
1217 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1218 ir_swizzle
*swizzle
= NULL
;
1220 /* the api does not allow a swizzle here, but the varying packing code
1221 * may have pushed one into here.
1223 swizzle
= ir
->operands
[0]->as_swizzle();
1225 deref
= swizzle
->val
->as_dereference();
1229 deref
->accept(this);
1231 nir_intrinsic_op op
;
1232 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1233 switch (ir
->operation
) {
1234 case ir_unop_interpolate_at_centroid
:
1235 op
= nir_intrinsic_interp_var_at_centroid
;
1237 case ir_binop_interpolate_at_offset
:
1238 op
= nir_intrinsic_interp_var_at_offset
;
1240 case ir_binop_interpolate_at_sample
:
1241 op
= nir_intrinsic_interp_var_at_sample
;
1244 unreachable("Invalid interpolation intrinsic");
1247 /* This case can happen if the vertex shader does not write the
1248 * given varying. In this case, the linker will lower it to a
1249 * global variable. Since interpolating a variable makes no
1250 * sense, we'll just turn it into a load which will probably
1251 * eventually end up as an SSA definition.
1253 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1254 op
= nir_intrinsic_load_var
;
1257 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1258 intrin
->num_components
= deref
->type
->vector_elements
;
1259 intrin
->variables
[0] = this->deref_head
;
1260 ralloc_steal(intrin
, intrin
->variables
[0]);
1262 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1263 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1264 intrin
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
1266 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
1269 nir_alu_instr
*mov
= nir_alu_instr_create(shader
, nir_op_imov
);
1270 mov
->dest
.write_mask
= (1 << swizzle
->type
->vector_elements
) - 1;
1271 mov
->src
[0].src
.is_ssa
= true;
1272 mov
->src
[0].src
.ssa
= &intrin
->dest
.ssa
;
1274 mov
->src
[0].swizzle
[0] = swizzle
->mask
.x
;
1275 mov
->src
[0].swizzle
[1] = swizzle
->mask
.y
;
1276 mov
->src
[0].swizzle
[2] = swizzle
->mask
.z
;
1277 mov
->src
[0].swizzle
[3] = swizzle
->mask
.w
;
1278 for (unsigned i
= deref
->type
->vector_elements
; i
< 4; i
++)
1279 mov
->src
[0].swizzle
[i
] = 0;
1281 add_instr(&mov
->instr
, swizzle
->type
->vector_elements
);
1292 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1293 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1295 glsl_base_type types
[4];
1296 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1298 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1300 types
[i
] = GLSL_TYPE_FLOAT
;
1302 glsl_base_type out_type
;
1304 out_type
= ir
->type
->base_type
;
1306 out_type
= GLSL_TYPE_FLOAT
;
1308 unsigned dest_size
= ir
->type
->vector_elements
;
1310 nir_alu_instr
*instr
;
1313 switch (ir
->operation
) {
1314 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
1315 case ir_unop_logic_not
:
1316 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1319 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fneg
: nir_op_ineg
,
1323 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fabs
: nir_op_iabs
,
1326 case ir_unop_saturate
:
1327 assert(types
[0] == GLSL_TYPE_FLOAT
);
1328 instr
= emit(nir_op_fsat
, dest_size
, srcs
);
1331 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1334 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1335 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1336 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1337 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1338 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1339 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1340 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1342 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1345 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1348 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1350 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1351 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1352 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1353 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1354 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1357 case ir_unop_bitcast_i2f
:
1358 case ir_unop_bitcast_f2i
:
1359 case ir_unop_bitcast_u2f
:
1360 case ir_unop_bitcast_f2u
:
1361 case ir_unop_subroutine_to_int
:
1363 emit(nir_op_imov
, dest_size
, srcs
);
1366 switch (ir
->operands
[0]->type
->vector_elements
) {
1368 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1372 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1376 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1380 unreachable("not reached");
1383 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1384 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1385 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1386 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1387 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1388 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1389 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1390 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1391 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1392 case ir_unop_dFdx_fine
: emit(nir_op_fddx_fine
, dest_size
, srcs
); break;
1393 case ir_unop_dFdy_fine
: emit(nir_op_fddy_fine
, dest_size
, srcs
); break;
1394 case ir_unop_dFdx_coarse
: emit(nir_op_fddx_coarse
, dest_size
, srcs
); break;
1395 case ir_unop_dFdy_coarse
: emit(nir_op_fddy_coarse
, dest_size
, srcs
); break;
1396 case ir_unop_pack_snorm_2x16
:
1397 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1399 case ir_unop_pack_snorm_4x8
:
1400 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1402 case ir_unop_pack_unorm_2x16
:
1403 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1405 case ir_unop_pack_unorm_4x8
:
1406 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1408 case ir_unop_pack_half_2x16
:
1409 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1411 case ir_unop_unpack_snorm_2x16
:
1412 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1414 case ir_unop_unpack_snorm_4x8
:
1415 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1417 case ir_unop_unpack_unorm_2x16
:
1418 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1420 case ir_unop_unpack_unorm_4x8
:
1421 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1423 case ir_unop_unpack_half_2x16
:
1424 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1426 case ir_unop_unpack_half_2x16_split_x
:
1427 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1429 case ir_unop_unpack_half_2x16_split_y
:
1430 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1432 case ir_unop_bitfield_reverse
:
1433 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1435 case ir_unop_bit_count
:
1436 emit(nir_op_bit_count
, dest_size
, srcs
);
1438 case ir_unop_find_msb
:
1440 case GLSL_TYPE_UINT
:
1441 emit(nir_op_ufind_msb
, dest_size
, srcs
);
1444 emit(nir_op_ifind_msb
, dest_size
, srcs
);
1447 unreachable("Invalid type for findMSB()");
1450 case ir_unop_find_lsb
:
1451 emit(nir_op_find_lsb
, dest_size
, srcs
);
1455 switch (ir
->type
->vector_elements
) {
1457 switch (ir
->operands
[0]->type
->vector_elements
) {
1458 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1459 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1460 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1461 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1462 default: unreachable("not reached");
1466 switch (ir
->operands
[0]->type
->vector_elements
) {
1467 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1468 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1469 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1470 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1471 default: unreachable("not reached");
1475 switch (ir
->operands
[0]->type
->vector_elements
) {
1476 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1477 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1478 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1479 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1480 default: unreachable("not reached");
1484 switch (ir
->operands
[0]->type
->vector_elements
) {
1485 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1486 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1487 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1488 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1489 default: unreachable("not reached");
1493 unreachable("not reached");
1496 case ir_unop_get_buffer_size
: {
1497 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1499 nir_intrinsic_get_buffer_size
);
1500 load
->num_components
= ir
->type
->vector_elements
;
1501 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
1502 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1514 case ir_binop_bit_and
:
1515 case ir_binop_bit_or
:
1516 case ir_binop_bit_xor
:
1517 case ir_binop_logic_and
:
1518 case ir_binop_logic_or
:
1519 case ir_binop_logic_xor
:
1520 case ir_binop_lshift
:
1521 case ir_binop_rshift
:
1522 switch (ir
->operation
) {
1524 if (out_type
== GLSL_TYPE_FLOAT
)
1530 if (out_type
== GLSL_TYPE_FLOAT
)
1536 if (out_type
== GLSL_TYPE_FLOAT
)
1542 if (out_type
== GLSL_TYPE_FLOAT
)
1544 else if (out_type
== GLSL_TYPE_INT
)
1550 if (out_type
== GLSL_TYPE_FLOAT
)
1556 if (out_type
== GLSL_TYPE_FLOAT
)
1558 else if (out_type
== GLSL_TYPE_INT
)
1564 if (out_type
== GLSL_TYPE_FLOAT
)
1566 else if (out_type
== GLSL_TYPE_INT
)
1571 case ir_binop_bit_and
:
1574 case ir_binop_bit_or
:
1577 case ir_binop_bit_xor
:
1580 case ir_binop_logic_and
:
1586 case ir_binop_logic_or
:
1592 case ir_binop_logic_xor
:
1598 case ir_binop_lshift
:
1601 case ir_binop_rshift
:
1602 if (out_type
== GLSL_TYPE_INT
)
1612 unreachable("not reached");
1615 instr
= emit(op
, dest_size
, srcs
);
1617 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1618 ir
->operands
[1]->type
->vector_elements
== 1) {
1619 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1621 instr
->src
[1].swizzle
[i
] = 0;
1625 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1626 ir
->operands
[0]->type
->vector_elements
== 1) {
1627 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1629 instr
->src
[0].swizzle
[i
] = 0;
1634 case ir_binop_imul_high
:
1635 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1638 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1639 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1641 if (supports_ints
) {
1642 if (types
[0] == GLSL_TYPE_FLOAT
)
1643 emit(nir_op_flt
, dest_size
, srcs
);
1644 else if (types
[0] == GLSL_TYPE_INT
)
1645 emit(nir_op_ilt
, dest_size
, srcs
);
1647 emit(nir_op_ult
, dest_size
, srcs
);
1649 emit(nir_op_slt
, dest_size
, srcs
);
1652 case ir_binop_greater
:
1653 if (supports_ints
) {
1654 if (types
[0] == GLSL_TYPE_FLOAT
)
1655 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1656 else if (types
[0] == GLSL_TYPE_INT
)
1657 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1659 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1661 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1664 case ir_binop_lequal
:
1665 if (supports_ints
) {
1666 if (types
[0] == GLSL_TYPE_FLOAT
)
1667 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1668 else if (types
[0] == GLSL_TYPE_INT
)
1669 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1671 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1673 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1676 case ir_binop_gequal
:
1677 if (supports_ints
) {
1678 if (types
[0] == GLSL_TYPE_FLOAT
)
1679 emit(nir_op_fge
, dest_size
, srcs
);
1680 else if (types
[0] == GLSL_TYPE_INT
)
1681 emit(nir_op_ige
, dest_size
, srcs
);
1683 emit(nir_op_uge
, dest_size
, srcs
);
1685 emit(nir_op_slt
, dest_size
, srcs
);
1688 case ir_binop_equal
:
1689 if (supports_ints
) {
1690 if (types
[0] == GLSL_TYPE_FLOAT
)
1691 emit(nir_op_feq
, dest_size
, srcs
);
1693 emit(nir_op_ieq
, dest_size
, srcs
);
1695 emit(nir_op_seq
, dest_size
, srcs
);
1698 case ir_binop_nequal
:
1699 if (supports_ints
) {
1700 if (types
[0] == GLSL_TYPE_FLOAT
)
1701 emit(nir_op_fne
, dest_size
, srcs
);
1703 emit(nir_op_ine
, dest_size
, srcs
);
1705 emit(nir_op_sne
, dest_size
, srcs
);
1708 case ir_binop_all_equal
:
1709 if (supports_ints
) {
1710 if (types
[0] == GLSL_TYPE_FLOAT
) {
1711 switch (ir
->operands
[0]->type
->vector_elements
) {
1712 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1713 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1714 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1715 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1717 unreachable("not reached");
1720 switch (ir
->operands
[0]->type
->vector_elements
) {
1721 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1722 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1723 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1724 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1726 unreachable("not reached");
1730 switch (ir
->operands
[0]->type
->vector_elements
) {
1731 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1732 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1733 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1734 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1736 unreachable("not reached");
1740 case ir_binop_any_nequal
:
1741 if (supports_ints
) {
1742 if (types
[0] == GLSL_TYPE_FLOAT
) {
1743 switch (ir
->operands
[0]->type
->vector_elements
) {
1744 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1745 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1746 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1747 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1749 unreachable("not reached");
1752 switch (ir
->operands
[0]->type
->vector_elements
) {
1753 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1754 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1755 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1756 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1758 unreachable("not reached");
1762 switch (ir
->operands
[0]->type
->vector_elements
) {
1763 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1764 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1765 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1766 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1768 unreachable("not reached");
1773 switch (ir
->operands
[0]->type
->vector_elements
) {
1774 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1775 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1776 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1778 unreachable("not reached");
1782 case ir_binop_pack_half_2x16_split
:
1783 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1785 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1786 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1787 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1789 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1790 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1791 ir
->operands
[2]->type
->vector_elements
== 1) {
1792 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1794 instr
->src
[2].swizzle
[i
] = 0;
1800 emit(nir_op_bcsel
, dest_size
, srcs
);
1802 emit(nir_op_fcsel
, dest_size
, srcs
);
1805 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1806 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1807 instr
->src
[0].swizzle
[i
] = 0;
1810 case ir_triop_bitfield_extract
:
1811 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1812 nir_op_ubitfield_extract
, dest_size
, srcs
);
1813 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1814 instr
->src
[1].swizzle
[i
] = 0;
1815 instr
->src
[2].swizzle
[i
] = 0;
1818 case ir_quadop_bitfield_insert
:
1819 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1820 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1821 instr
->src
[2].swizzle
[i
] = 0;
1822 instr
->src
[3].swizzle
[i
] = 0;
1825 case ir_quadop_vector
:
1826 switch (ir
->type
->vector_elements
) {
1827 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1828 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1829 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1830 default: unreachable("not reached");
1835 unreachable("not reached");
1840 nir_visitor::visit(ir_swizzle
*ir
)
1842 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1843 ir
->type
->vector_elements
,
1844 evaluate_rvalue(ir
->val
));
1846 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1847 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1848 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1852 nir_visitor::visit(ir_texture
*ir
)
1859 num_srcs
= 1; /* coordinate */
1864 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1865 num_srcs
= 2; /* coordinate, bias/lod */
1869 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1875 if (ir
->lod_info
.lod
!= NULL
)
1876 num_srcs
= 2; /* coordinate, lod */
1878 num_srcs
= 1; /* coordinate */
1882 op
= nir_texop_txf_ms
;
1883 num_srcs
= 2; /* coordinate, sample_index */
1888 if (ir
->lod_info
.lod
!= NULL
)
1889 num_srcs
= 1; /* lod */
1896 num_srcs
= 1; /* coordinate */
1901 num_srcs
= 1; /* coordinate */
1904 case ir_query_levels
:
1905 op
= nir_texop_query_levels
;
1909 case ir_texture_samples
:
1910 op
= nir_texop_texture_samples
;
1915 unreachable("not reached");
1918 if (ir
->projector
!= NULL
)
1920 if (ir
->shadow_comparitor
!= NULL
)
1922 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1925 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1928 instr
->sampler_dim
=
1929 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1930 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1931 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1932 if (instr
->is_shadow
)
1933 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1934 switch (ir
->type
->base_type
) {
1935 case GLSL_TYPE_FLOAT
:
1936 instr
->dest_type
= nir_type_float
;
1939 instr
->dest_type
= nir_type_int
;
1941 case GLSL_TYPE_UINT
:
1942 instr
->dest_type
= nir_type_unsigned
;
1945 unreachable("not reached");
1948 instr
->sampler
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1950 unsigned src_number
= 0;
1952 if (ir
->coordinate
!= NULL
) {
1953 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1954 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->coordinate
);
1955 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1959 if (ir
->projector
!= NULL
) {
1960 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->projector
);
1961 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1965 if (ir
->shadow_comparitor
!= NULL
) {
1966 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->shadow_comparitor
);
1967 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1971 if (ir
->offset
!= NULL
) {
1972 /* we don't support multiple offsets yet */
1973 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1975 ir_constant
*const_offset
= ir
->offset
->as_constant();
1976 if (const_offset
!= NULL
) {
1977 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1978 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1980 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->offset
);
1981 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1988 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.bias
);
1989 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1996 if (ir
->lod_info
.lod
!= NULL
) {
1997 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.lod
);
1998 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
2004 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdx
);
2005 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
2007 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdy
);
2008 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
2013 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.sample_index
);
2014 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
2019 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
2026 assert(src_number
== num_srcs
);
2028 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
2032 nir_visitor::visit(ir_constant
*ir
)
2035 * We don't know if this variable is an an array or struct that gets
2036 * dereferenced, so do the safe thing an make it a variable with a
2037 * constant initializer and return a dereference.
2040 nir_variable
*var
= ralloc(this->shader
, nir_variable
);
2041 var
->name
= ralloc_strdup(var
, "const_temp");
2042 var
->type
= ir
->type
;
2043 var
->data
.mode
= nir_var_local
;
2044 var
->data
.read_only
= true;
2045 var
->constant_initializer
= constant_copy(ir
, var
);
2046 exec_list_push_tail(&this->impl
->locals
, &var
->node
);
2048 this->deref_head
= nir_deref_var_create(this->shader
, var
);
2049 this->deref_tail
= &this->deref_head
->deref
;
2053 nir_visitor::visit(ir_dereference_variable
*ir
)
2055 struct hash_entry
*entry
=
2056 _mesa_hash_table_search(this->var_table
, ir
->var
);
2058 nir_variable
*var
= (nir_variable
*) entry
->data
;
2060 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
2061 this->deref_head
= deref
;
2062 this->deref_tail
= &deref
->deref
;
2066 nir_visitor::visit(ir_dereference_record
*ir
)
2068 ir
->record
->accept(this);
2070 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
2071 assert(field_index
>= 0);
2073 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
2074 deref
->deref
.type
= ir
->type
;
2075 this->deref_tail
->child
= &deref
->deref
;
2076 this->deref_tail
= &deref
->deref
;
2080 nir_visitor::visit(ir_dereference_array
*ir
)
2082 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
2083 deref
->deref
.type
= ir
->type
;
2085 ir_constant
*const_index
= ir
->array_index
->as_constant();
2086 if (const_index
!= NULL
) {
2087 deref
->deref_array_type
= nir_deref_array_type_direct
;
2088 deref
->base_offset
= const_index
->value
.u
[0];
2090 deref
->deref_array_type
= nir_deref_array_type_indirect
;
2091 deref
->indirect
= evaluate_rvalue(ir
->array_index
);
2094 ir
->array
->accept(this);
2096 this->deref_tail
->child
= &deref
->deref
;
2097 ralloc_steal(this->deref_tail
, deref
);
2098 this->deref_tail
= &deref
->deref
;
2102 nir_visitor::visit(ir_barrier
*ir
)
2104 nir_intrinsic_instr
*instr
=
2105 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
2106 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);