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"
32 #include "compiler/nir/nir_control_flow.h"
33 #include "compiler/nir/nir_builder.h"
34 #include "main/imports.h"
37 * pass to lower GLSL IR to NIR
39 * This will lower variable dereferences to loads/stores of corresponding
40 * variables in NIR - the variables will be converted to registers in a later
46 class nir_visitor
: public ir_visitor
49 nir_visitor(nir_shader
*shader
);
52 virtual void visit(ir_variable
*);
53 virtual void visit(ir_function
*);
54 virtual void visit(ir_function_signature
*);
55 virtual void visit(ir_loop
*);
56 virtual void visit(ir_if
*);
57 virtual void visit(ir_discard
*);
58 virtual void visit(ir_loop_jump
*);
59 virtual void visit(ir_return
*);
60 virtual void visit(ir_call
*);
61 virtual void visit(ir_assignment
*);
62 virtual void visit(ir_emit_vertex
*);
63 virtual void visit(ir_end_primitive
*);
64 virtual void visit(ir_expression
*);
65 virtual void visit(ir_swizzle
*);
66 virtual void visit(ir_texture
*);
67 virtual void visit(ir_constant
*);
68 virtual void visit(ir_dereference_variable
*);
69 virtual void visit(ir_dereference_record
*);
70 virtual void visit(ir_dereference_array
*);
71 virtual void visit(ir_barrier
*);
73 void create_function(ir_function_signature
*ir
);
76 void add_instr(nir_instr
*instr
, unsigned num_components
, unsigned bit_size
);
77 nir_ssa_def
*evaluate_rvalue(ir_rvalue
*ir
);
79 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
**srcs
);
80 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
);
81 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
,
83 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
,
84 nir_ssa_def
*src2
, nir_ssa_def
*src3
);
89 nir_function_impl
*impl
;
91 nir_ssa_def
*result
; /* result of the expression tree last visited */
93 nir_deref_var
*evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
);
95 /* the head of the dereference chain we're creating */
96 nir_deref_var
*deref_head
;
97 /* the tail of the dereference chain we're creating */
98 nir_deref
*deref_tail
;
100 nir_variable
*var
; /* variable created by ir_variable visitor */
102 /* whether the IR we're operating on is per-function or global */
105 /* map of ir_variable -> nir_variable */
106 struct hash_table
*var_table
;
108 /* map of ir_function_signature -> nir_function_overload */
109 struct hash_table
*overload_table
;
113 * This visitor runs before the main visitor, calling create_function() for
114 * each function so that the main visitor can resolve forward references in
118 class nir_function_visitor
: public ir_hierarchical_visitor
121 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
124 virtual ir_visitor_status
visit_enter(ir_function
*);
127 nir_visitor
*visitor
;
130 } /* end of anonymous namespace */
133 glsl_to_nir(const struct gl_shader_program
*shader_prog
,
134 gl_shader_stage stage
,
135 const nir_shader_compiler_options
*options
)
137 struct gl_linked_shader
*sh
= shader_prog
->_LinkedShaders
[stage
];
139 nir_shader
*shader
= nir_shader_create(NULL
, stage
, options
,
142 nir_visitor
v1(shader
);
143 nir_function_visitor
v2(&v1
);
145 visit_exec_list(sh
->ir
, &v1
);
147 shader
->info
->name
= ralloc_asprintf(shader
, "GLSL%d", shader_prog
->Name
);
148 if (shader_prog
->Label
)
149 shader
->info
->label
= ralloc_strdup(shader
, shader_prog
->Label
);
150 shader
->info
->num_textures
= util_last_bit(sh
->Program
->SamplersUsed
);
151 shader
->info
->num_ubos
= sh
->NumUniformBlocks
;
152 shader
->info
->num_abos
= shader_prog
->NumAtomicBuffers
;
153 shader
->info
->num_ssbos
= sh
->NumShaderStorageBlocks
;
154 shader
->info
->num_images
= sh
->NumImages
;
155 shader
->info
->uses_clip_distance_out
=
156 sh
->Program
->ClipDistanceArraySize
!= 0;
157 shader
->info
->separate_shader
= shader_prog
->SeparateShader
;
158 shader
->info
->has_transform_feedback_varyings
=
159 shader_prog
->TransformFeedback
.NumVarying
> 0;
164 nir_visitor::nir_visitor(nir_shader
*shader
)
166 this->supports_ints
= shader
->options
->native_integers
;
167 this->shader
= shader
;
168 this->is_global
= true;
169 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
170 _mesa_key_pointer_equal
);
171 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
172 _mesa_key_pointer_equal
);
176 this->deref_head
= NULL
;
177 this->deref_tail
= NULL
;
178 memset(&this->b
, 0, sizeof(this->b
));
181 nir_visitor::~nir_visitor()
183 _mesa_hash_table_destroy(this->var_table
, NULL
);
184 _mesa_hash_table_destroy(this->overload_table
, NULL
);
188 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
191 ralloc_steal(mem_ctx
, this->deref_head
);
192 return this->deref_head
;
195 static nir_constant
*
196 constant_copy(ir_constant
*ir
, void *mem_ctx
)
201 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
203 unsigned total_elems
= ir
->type
->components();
206 ret
->num_elements
= 0;
207 switch (ir
->type
->base_type
) {
209 for (i
= 0; i
< total_elems
; i
++)
210 ret
->value
.u
[i
] = ir
->value
.u
[i
];
214 for (i
= 0; i
< total_elems
; i
++)
215 ret
->value
.i
[i
] = ir
->value
.i
[i
];
218 case GLSL_TYPE_FLOAT
:
219 for (i
= 0; i
< total_elems
; i
++)
220 ret
->value
.f
[i
] = ir
->value
.f
[i
];
223 case GLSL_TYPE_DOUBLE
:
224 for (i
= 0; i
< total_elems
; i
++)
225 ret
->value
.d
[i
] = ir
->value
.d
[i
];
229 for (i
= 0; i
< total_elems
; i
++)
230 ret
->value
.b
[i
] = ir
->value
.b
[i
];
233 case GLSL_TYPE_STRUCT
:
234 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
236 ret
->num_elements
= ir
->type
->length
;
239 foreach_in_list(ir_constant
, field
, &ir
->components
) {
240 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
245 case GLSL_TYPE_ARRAY
:
246 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
248 ret
->num_elements
= ir
->type
->length
;
250 for (i
= 0; i
< ir
->type
->length
; i
++)
251 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
255 unreachable("not reached");
262 nir_visitor::visit(ir_variable
*ir
)
264 nir_variable
*var
= ralloc(shader
, nir_variable
);
265 var
->type
= ir
->type
;
266 var
->name
= ralloc_strdup(var
, ir
->name
);
268 var
->data
.read_only
= ir
->data
.read_only
;
269 var
->data
.centroid
= ir
->data
.centroid
;
270 var
->data
.sample
= ir
->data
.sample
;
271 var
->data
.patch
= ir
->data
.patch
;
272 var
->data
.invariant
= ir
->data
.invariant
;
273 var
->data
.location
= ir
->data
.location
;
275 switch(ir
->data
.mode
) {
277 case ir_var_temporary
:
279 var
->data
.mode
= nir_var_global
;
281 var
->data
.mode
= nir_var_local
;
284 case ir_var_function_in
:
285 case ir_var_function_out
:
286 case ir_var_function_inout
:
287 case ir_var_const_in
:
288 var
->data
.mode
= nir_var_local
;
291 case ir_var_shader_in
:
292 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
293 ir
->data
.location
== VARYING_SLOT_FACE
) {
294 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
295 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
296 var
->data
.mode
= nir_var_system_value
;
297 } else if (shader
->stage
== MESA_SHADER_GEOMETRY
&&
298 ir
->data
.location
== VARYING_SLOT_PRIMITIVE_ID
) {
299 /* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
300 var
->data
.location
= SYSTEM_VALUE_PRIMITIVE_ID
;
301 var
->data
.mode
= nir_var_system_value
;
303 var
->data
.mode
= nir_var_shader_in
;
307 case ir_var_shader_out
:
308 var
->data
.mode
= nir_var_shader_out
;
312 var
->data
.mode
= nir_var_uniform
;
315 case ir_var_shader_storage
:
316 var
->data
.mode
= nir_var_shader_storage
;
319 case ir_var_system_value
:
320 var
->data
.mode
= nir_var_system_value
;
324 unreachable("not reached");
327 var
->data
.interpolation
= ir
->data
.interpolation
;
328 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
329 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
330 var
->data
.explicit_location
= ir
->data
.explicit_location
;
331 var
->data
.explicit_index
= ir
->data
.explicit_index
;
332 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
333 var
->data
.has_initializer
= ir
->data
.has_initializer
;
334 var
->data
.location_frac
= ir
->data
.location_frac
;
336 switch (ir
->data
.depth_layout
) {
337 case ir_depth_layout_none
:
338 var
->data
.depth_layout
= nir_depth_layout_none
;
340 case ir_depth_layout_any
:
341 var
->data
.depth_layout
= nir_depth_layout_any
;
343 case ir_depth_layout_greater
:
344 var
->data
.depth_layout
= nir_depth_layout_greater
;
346 case ir_depth_layout_less
:
347 var
->data
.depth_layout
= nir_depth_layout_less
;
349 case ir_depth_layout_unchanged
:
350 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
353 unreachable("not reached");
356 var
->data
.index
= ir
->data
.index
;
357 var
->data
.binding
= ir
->data
.binding
;
358 var
->data
.offset
= ir
->data
.offset
;
359 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
360 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
361 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
362 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
363 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
364 var
->data
.image
.format
= ir
->data
.image_format
;
365 var
->data
.max_array_access
= ir
->data
.max_array_access
;
366 var
->data
.fb_fetch_output
= ir
->data
.fb_fetch_output
;
368 var
->num_state_slots
= ir
->get_num_state_slots();
369 if (var
->num_state_slots
> 0) {
370 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
371 var
->num_state_slots
);
373 ir_state_slot
*state_slots
= ir
->get_state_slots();
374 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
375 for (unsigned j
= 0; j
< 5; j
++)
376 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
377 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
380 var
->state_slots
= NULL
;
383 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
385 var
->interface_type
= ir
->get_interface_type();
387 if (var
->data
.mode
== nir_var_local
)
388 nir_function_impl_add_variable(impl
, var
);
390 nir_shader_add_variable(shader
, var
);
392 _mesa_hash_table_insert(var_table
, ir
, var
);
397 nir_function_visitor::visit_enter(ir_function
*ir
)
399 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
400 visitor
->create_function(sig
);
402 return visit_continue_with_parent
;
406 nir_visitor::create_function(ir_function_signature
*ir
)
408 if (ir
->is_intrinsic())
411 nir_function
*func
= nir_function_create(shader
, ir
->function_name());
413 assert(ir
->parameters
.is_empty());
414 assert(ir
->return_type
== glsl_type::void_type
);
416 _mesa_hash_table_insert(this->overload_table
, ir
, func
);
420 nir_visitor::visit(ir_function
*ir
)
422 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
427 nir_visitor::visit(ir_function_signature
*ir
)
429 if (ir
->is_intrinsic())
432 struct hash_entry
*entry
=
433 _mesa_hash_table_search(this->overload_table
, ir
);
436 nir_function
*func
= (nir_function
*) entry
->data
;
438 if (ir
->is_defined
) {
439 nir_function_impl
*impl
= nir_function_impl_create(func
);
442 assert(strcmp(func
->name
, "main") == 0);
443 assert(ir
->parameters
.is_empty());
444 assert(func
->return_type
== glsl_type::void_type
);
446 this->is_global
= false;
448 nir_builder_init(&b
, impl
);
449 b
.cursor
= nir_after_cf_list(&impl
->body
);
450 visit_exec_list(&ir
->body
, this);
452 this->is_global
= true;
459 nir_visitor::visit(ir_loop
*ir
)
461 nir_loop
*loop
= nir_loop_create(this->shader
);
462 nir_builder_cf_insert(&b
, &loop
->cf_node
);
464 b
.cursor
= nir_after_cf_list(&loop
->body
);
465 visit_exec_list(&ir
->body_instructions
, this);
466 b
.cursor
= nir_after_cf_node(&loop
->cf_node
);
470 nir_visitor::visit(ir_if
*ir
)
473 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
475 nir_if
*if_stmt
= nir_if_create(this->shader
);
476 if_stmt
->condition
= condition
;
477 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
479 b
.cursor
= nir_after_cf_list(&if_stmt
->then_list
);
480 visit_exec_list(&ir
->then_instructions
, this);
482 b
.cursor
= nir_after_cf_list(&if_stmt
->else_list
);
483 visit_exec_list(&ir
->else_instructions
, this);
485 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
489 nir_visitor::visit(ir_discard
*ir
)
492 * discards aren't treated as control flow, because before we lower them
493 * they can appear anywhere in the shader and the stuff after them may still
494 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
495 * discards will be immediately followed by a return.
498 nir_intrinsic_instr
*discard
;
500 discard
= nir_intrinsic_instr_create(this->shader
,
501 nir_intrinsic_discard_if
);
503 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
505 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
508 nir_builder_instr_insert(&b
, &discard
->instr
);
512 nir_visitor::visit(ir_emit_vertex
*ir
)
514 nir_intrinsic_instr
*instr
=
515 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
516 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
517 nir_builder_instr_insert(&b
, &instr
->instr
);
521 nir_visitor::visit(ir_end_primitive
*ir
)
523 nir_intrinsic_instr
*instr
=
524 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
525 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
526 nir_builder_instr_insert(&b
, &instr
->instr
);
530 nir_visitor::visit(ir_loop_jump
*ir
)
534 case ir_loop_jump::jump_break
:
535 type
= nir_jump_break
;
537 case ir_loop_jump::jump_continue
:
538 type
= nir_jump_continue
;
541 unreachable("not reached");
544 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
545 nir_builder_instr_insert(&b
, &instr
->instr
);
549 nir_visitor::visit(ir_return
*ir
)
551 if (ir
->value
!= NULL
) {
552 nir_intrinsic_instr
*copy
=
553 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
555 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
556 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
559 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
560 nir_builder_instr_insert(&b
, &instr
->instr
);
564 nir_visitor::visit(ir_call
*ir
)
566 if (ir
->callee
->is_intrinsic()) {
569 switch (ir
->callee
->intrinsic_id
) {
570 case ir_intrinsic_atomic_counter_read
:
571 op
= nir_intrinsic_atomic_counter_read_var
;
573 case ir_intrinsic_atomic_counter_increment
:
574 op
= nir_intrinsic_atomic_counter_inc_var
;
576 case ir_intrinsic_atomic_counter_predecrement
:
577 op
= nir_intrinsic_atomic_counter_dec_var
;
579 case ir_intrinsic_atomic_counter_add
:
580 op
= nir_intrinsic_atomic_counter_add_var
;
582 case ir_intrinsic_atomic_counter_and
:
583 op
= nir_intrinsic_atomic_counter_and_var
;
585 case ir_intrinsic_atomic_counter_or
:
586 op
= nir_intrinsic_atomic_counter_or_var
;
588 case ir_intrinsic_atomic_counter_xor
:
589 op
= nir_intrinsic_atomic_counter_xor_var
;
591 case ir_intrinsic_atomic_counter_min
:
592 op
= nir_intrinsic_atomic_counter_min_var
;
594 case ir_intrinsic_atomic_counter_max
:
595 op
= nir_intrinsic_atomic_counter_max_var
;
597 case ir_intrinsic_atomic_counter_exchange
:
598 op
= nir_intrinsic_atomic_counter_exchange_var
;
600 case ir_intrinsic_atomic_counter_comp_swap
:
601 op
= nir_intrinsic_atomic_counter_comp_swap_var
;
603 case ir_intrinsic_image_load
:
604 op
= nir_intrinsic_image_load
;
606 case ir_intrinsic_image_store
:
607 op
= nir_intrinsic_image_store
;
609 case ir_intrinsic_image_atomic_add
:
610 op
= nir_intrinsic_image_atomic_add
;
612 case ir_intrinsic_image_atomic_min
:
613 op
= nir_intrinsic_image_atomic_min
;
615 case ir_intrinsic_image_atomic_max
:
616 op
= nir_intrinsic_image_atomic_max
;
618 case ir_intrinsic_image_atomic_and
:
619 op
= nir_intrinsic_image_atomic_and
;
621 case ir_intrinsic_image_atomic_or
:
622 op
= nir_intrinsic_image_atomic_or
;
624 case ir_intrinsic_image_atomic_xor
:
625 op
= nir_intrinsic_image_atomic_xor
;
627 case ir_intrinsic_image_atomic_exchange
:
628 op
= nir_intrinsic_image_atomic_exchange
;
630 case ir_intrinsic_image_atomic_comp_swap
:
631 op
= nir_intrinsic_image_atomic_comp_swap
;
633 case ir_intrinsic_memory_barrier
:
634 op
= nir_intrinsic_memory_barrier
;
636 case ir_intrinsic_image_size
:
637 op
= nir_intrinsic_image_size
;
639 case ir_intrinsic_image_samples
:
640 op
= nir_intrinsic_image_samples
;
642 case ir_intrinsic_ssbo_store
:
643 op
= nir_intrinsic_store_ssbo
;
645 case ir_intrinsic_ssbo_load
:
646 op
= nir_intrinsic_load_ssbo
;
648 case ir_intrinsic_ssbo_atomic_add
:
649 op
= nir_intrinsic_ssbo_atomic_add
;
651 case ir_intrinsic_ssbo_atomic_and
:
652 op
= nir_intrinsic_ssbo_atomic_and
;
654 case ir_intrinsic_ssbo_atomic_or
:
655 op
= nir_intrinsic_ssbo_atomic_or
;
657 case ir_intrinsic_ssbo_atomic_xor
:
658 op
= nir_intrinsic_ssbo_atomic_xor
;
660 case ir_intrinsic_ssbo_atomic_min
:
661 assert(ir
->return_deref
);
662 if (ir
->return_deref
->type
== glsl_type::int_type
)
663 op
= nir_intrinsic_ssbo_atomic_imin
;
664 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
665 op
= nir_intrinsic_ssbo_atomic_umin
;
667 unreachable("Invalid type");
669 case ir_intrinsic_ssbo_atomic_max
:
670 assert(ir
->return_deref
);
671 if (ir
->return_deref
->type
== glsl_type::int_type
)
672 op
= nir_intrinsic_ssbo_atomic_imax
;
673 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
674 op
= nir_intrinsic_ssbo_atomic_umax
;
676 unreachable("Invalid type");
678 case ir_intrinsic_ssbo_atomic_exchange
:
679 op
= nir_intrinsic_ssbo_atomic_exchange
;
681 case ir_intrinsic_ssbo_atomic_comp_swap
:
682 op
= nir_intrinsic_ssbo_atomic_comp_swap
;
684 case ir_intrinsic_shader_clock
:
685 op
= nir_intrinsic_shader_clock
;
687 case ir_intrinsic_group_memory_barrier
:
688 op
= nir_intrinsic_group_memory_barrier
;
690 case ir_intrinsic_memory_barrier_atomic_counter
:
691 op
= nir_intrinsic_memory_barrier_atomic_counter
;
693 case ir_intrinsic_memory_barrier_buffer
:
694 op
= nir_intrinsic_memory_barrier_buffer
;
696 case ir_intrinsic_memory_barrier_image
:
697 op
= nir_intrinsic_memory_barrier_image
;
699 case ir_intrinsic_memory_barrier_shared
:
700 op
= nir_intrinsic_memory_barrier_shared
;
702 case ir_intrinsic_shared_load
:
703 op
= nir_intrinsic_load_shared
;
705 case ir_intrinsic_shared_store
:
706 op
= nir_intrinsic_store_shared
;
708 case ir_intrinsic_shared_atomic_add
:
709 op
= nir_intrinsic_shared_atomic_add
;
711 case ir_intrinsic_shared_atomic_and
:
712 op
= nir_intrinsic_shared_atomic_and
;
714 case ir_intrinsic_shared_atomic_or
:
715 op
= nir_intrinsic_shared_atomic_or
;
717 case ir_intrinsic_shared_atomic_xor
:
718 op
= nir_intrinsic_shared_atomic_xor
;
720 case ir_intrinsic_shared_atomic_min
:
721 assert(ir
->return_deref
);
722 if (ir
->return_deref
->type
== glsl_type::int_type
)
723 op
= nir_intrinsic_shared_atomic_imin
;
724 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
725 op
= nir_intrinsic_shared_atomic_umin
;
727 unreachable("Invalid type");
729 case ir_intrinsic_shared_atomic_max
:
730 assert(ir
->return_deref
);
731 if (ir
->return_deref
->type
== glsl_type::int_type
)
732 op
= nir_intrinsic_shared_atomic_imax
;
733 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
734 op
= nir_intrinsic_shared_atomic_umax
;
736 unreachable("Invalid type");
738 case ir_intrinsic_shared_atomic_exchange
:
739 op
= nir_intrinsic_shared_atomic_exchange
;
741 case ir_intrinsic_shared_atomic_comp_swap
:
742 op
= nir_intrinsic_shared_atomic_comp_swap
;
745 unreachable("not reached");
748 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
749 nir_dest
*dest
= &instr
->dest
;
752 case nir_intrinsic_atomic_counter_read_var
:
753 case nir_intrinsic_atomic_counter_inc_var
:
754 case nir_intrinsic_atomic_counter_dec_var
:
755 case nir_intrinsic_atomic_counter_add_var
:
756 case nir_intrinsic_atomic_counter_min_var
:
757 case nir_intrinsic_atomic_counter_max_var
:
758 case nir_intrinsic_atomic_counter_and_var
:
759 case nir_intrinsic_atomic_counter_or_var
:
760 case nir_intrinsic_atomic_counter_xor_var
:
761 case nir_intrinsic_atomic_counter_exchange_var
:
762 case nir_intrinsic_atomic_counter_comp_swap_var
: {
763 /* Set the counter variable dereference. */
764 exec_node
*param
= ir
->actual_parameters
.get_head();
765 ir_dereference
*counter
= (ir_dereference
*)param
;
767 instr
->variables
[0] = evaluate_deref(&instr
->instr
, counter
);
768 param
= param
->get_next();
770 /* Set the intrinsic destination. */
771 if (ir
->return_deref
) {
772 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
775 /* Set the intrinsic parameters. */
776 if (!param
->is_tail_sentinel()) {
778 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
779 param
= param
->get_next();
782 if (!param
->is_tail_sentinel()) {
784 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
785 param
= param
->get_next();
788 nir_builder_instr_insert(&b
, &instr
->instr
);
791 case nir_intrinsic_image_load
:
792 case nir_intrinsic_image_store
:
793 case nir_intrinsic_image_atomic_add
:
794 case nir_intrinsic_image_atomic_min
:
795 case nir_intrinsic_image_atomic_max
:
796 case nir_intrinsic_image_atomic_and
:
797 case nir_intrinsic_image_atomic_or
:
798 case nir_intrinsic_image_atomic_xor
:
799 case nir_intrinsic_image_atomic_exchange
:
800 case nir_intrinsic_image_atomic_comp_swap
:
801 case nir_intrinsic_image_samples
:
802 case nir_intrinsic_image_size
: {
803 nir_ssa_undef_instr
*instr_undef
=
804 nir_ssa_undef_instr_create(shader
, 1, 32);
805 nir_builder_instr_insert(&b
, &instr_undef
->instr
);
807 /* Set the image variable dereference. */
808 exec_node
*param
= ir
->actual_parameters
.get_head();
809 ir_dereference
*image
= (ir_dereference
*)param
;
810 const glsl_type
*type
=
811 image
->variable_referenced()->type
->without_array();
813 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
814 param
= param
->get_next();
816 /* Set the intrinsic destination. */
817 if (ir
->return_deref
) {
818 const nir_intrinsic_info
*info
=
819 &nir_intrinsic_infos
[instr
->intrinsic
];
820 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
821 info
->dest_components
, 32, NULL
);
824 if (op
== nir_intrinsic_image_size
||
825 op
== nir_intrinsic_image_samples
) {
826 nir_builder_instr_insert(&b
, &instr
->instr
);
830 /* Set the address argument, extending the coordinate vector to four
833 nir_ssa_def
*src_addr
=
834 evaluate_rvalue((ir_dereference
*)param
);
835 nir_ssa_def
*srcs
[4];
837 for (int i
= 0; i
< 4; i
++) {
838 if (i
< type
->coordinate_components())
839 srcs
[i
] = nir_channel(&b
, src_addr
, i
);
841 srcs
[i
] = &instr_undef
->def
;
844 instr
->src
[0] = nir_src_for_ssa(nir_vec(&b
, srcs
, 4));
845 param
= param
->get_next();
847 /* Set the sample argument, which is undefined for single-sample
850 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
852 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
853 param
= param
->get_next();
855 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
858 /* Set the intrinsic parameters. */
859 if (!param
->is_tail_sentinel()) {
861 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
862 param
= param
->get_next();
865 if (!param
->is_tail_sentinel()) {
867 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
868 param
= param
->get_next();
870 nir_builder_instr_insert(&b
, &instr
->instr
);
873 case nir_intrinsic_memory_barrier
:
874 case nir_intrinsic_group_memory_barrier
:
875 case nir_intrinsic_memory_barrier_atomic_counter
:
876 case nir_intrinsic_memory_barrier_buffer
:
877 case nir_intrinsic_memory_barrier_image
:
878 case nir_intrinsic_memory_barrier_shared
:
879 nir_builder_instr_insert(&b
, &instr
->instr
);
881 case nir_intrinsic_shader_clock
:
882 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
883 nir_builder_instr_insert(&b
, &instr
->instr
);
885 case nir_intrinsic_store_ssbo
: {
886 exec_node
*param
= ir
->actual_parameters
.get_head();
887 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
889 param
= param
->get_next();
890 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
892 param
= param
->get_next();
893 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
895 param
= param
->get_next();
896 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
899 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
900 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(block
));
901 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(offset
));
902 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
903 instr
->num_components
= val
->type
->vector_elements
;
905 nir_builder_instr_insert(&b
, &instr
->instr
);
908 case nir_intrinsic_load_ssbo
: {
909 exec_node
*param
= ir
->actual_parameters
.get_head();
910 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
912 param
= param
->get_next();
913 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
915 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(block
));
916 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
918 const glsl_type
*type
= ir
->return_deref
->var
->type
;
919 instr
->num_components
= type
->vector_elements
;
921 /* Setup destination register */
922 unsigned bit_size
= glsl_get_bit_size(type
);
923 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
924 type
->vector_elements
, bit_size
, NULL
);
926 /* Insert the created nir instruction now since in the case of boolean
927 * result we will need to emit another instruction after it
929 nir_builder_instr_insert(&b
, &instr
->instr
);
932 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
933 * consider a true boolean to be ~0. Fix this up with a != 0
936 if (type
->base_type
== GLSL_TYPE_BOOL
) {
937 nir_alu_instr
*load_ssbo_compare
=
938 nir_alu_instr_create(shader
, nir_op_ine
);
939 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
940 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
941 load_ssbo_compare
->src
[1].src
=
942 nir_src_for_ssa(nir_imm_int(&b
, 0));
943 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
944 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
945 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
946 &load_ssbo_compare
->dest
.dest
,
947 type
->vector_elements
, bit_size
, NULL
);
948 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
949 nir_builder_instr_insert(&b
, &load_ssbo_compare
->instr
);
950 dest
= &load_ssbo_compare
->dest
.dest
;
954 case nir_intrinsic_ssbo_atomic_add
:
955 case nir_intrinsic_ssbo_atomic_imin
:
956 case nir_intrinsic_ssbo_atomic_umin
:
957 case nir_intrinsic_ssbo_atomic_imax
:
958 case nir_intrinsic_ssbo_atomic_umax
:
959 case nir_intrinsic_ssbo_atomic_and
:
960 case nir_intrinsic_ssbo_atomic_or
:
961 case nir_intrinsic_ssbo_atomic_xor
:
962 case nir_intrinsic_ssbo_atomic_exchange
:
963 case nir_intrinsic_ssbo_atomic_comp_swap
: {
964 int param_count
= ir
->actual_parameters
.length();
965 assert(param_count
== 3 || param_count
== 4);
968 exec_node
*param
= ir
->actual_parameters
.get_head();
969 ir_instruction
*inst
= (ir_instruction
*) param
;
970 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
973 param
= param
->get_next();
974 inst
= (ir_instruction
*) param
;
975 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
977 /* data1 parameter (this is always present) */
978 param
= param
->get_next();
979 inst
= (ir_instruction
*) param
;
980 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
982 /* data2 parameter (only with atomic_comp_swap) */
983 if (param_count
== 4) {
984 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
985 param
= param
->get_next();
986 inst
= (ir_instruction
*) param
;
987 instr
->src
[3] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
991 assert(ir
->return_deref
);
992 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
993 ir
->return_deref
->type
->vector_elements
, 32, NULL
);
994 nir_builder_instr_insert(&b
, &instr
->instr
);
997 case nir_intrinsic_load_shared
: {
998 exec_node
*param
= ir
->actual_parameters
.get_head();
999 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1001 nir_intrinsic_set_base(instr
, 0);
1002 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(offset
));
1004 const glsl_type
*type
= ir
->return_deref
->var
->type
;
1005 instr
->num_components
= type
->vector_elements
;
1007 /* Setup destination register */
1008 unsigned bit_size
= glsl_get_bit_size(type
);
1009 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1010 type
->vector_elements
, bit_size
, NULL
);
1012 nir_builder_instr_insert(&b
, &instr
->instr
);
1015 case nir_intrinsic_store_shared
: {
1016 exec_node
*param
= ir
->actual_parameters
.get_head();
1017 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1019 param
= param
->get_next();
1020 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
1022 param
= param
->get_next();
1023 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
1026 nir_intrinsic_set_base(instr
, 0);
1027 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
1029 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
1031 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
1032 instr
->num_components
= val
->type
->vector_elements
;
1034 nir_builder_instr_insert(&b
, &instr
->instr
);
1037 case nir_intrinsic_shared_atomic_add
:
1038 case nir_intrinsic_shared_atomic_imin
:
1039 case nir_intrinsic_shared_atomic_umin
:
1040 case nir_intrinsic_shared_atomic_imax
:
1041 case nir_intrinsic_shared_atomic_umax
:
1042 case nir_intrinsic_shared_atomic_and
:
1043 case nir_intrinsic_shared_atomic_or
:
1044 case nir_intrinsic_shared_atomic_xor
:
1045 case nir_intrinsic_shared_atomic_exchange
:
1046 case nir_intrinsic_shared_atomic_comp_swap
: {
1047 int param_count
= ir
->actual_parameters
.length();
1048 assert(param_count
== 2 || param_count
== 3);
1051 exec_node
*param
= ir
->actual_parameters
.get_head();
1052 ir_instruction
*inst
= (ir_instruction
*) param
;
1053 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1055 /* data1 parameter (this is always present) */
1056 param
= param
->get_next();
1057 inst
= (ir_instruction
*) param
;
1058 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1060 /* data2 parameter (only with atomic_comp_swap) */
1061 if (param_count
== 3) {
1062 assert(op
== nir_intrinsic_shared_atomic_comp_swap
);
1063 param
= param
->get_next();
1064 inst
= (ir_instruction
*) param
;
1066 nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1070 assert(ir
->return_deref
);
1071 unsigned bit_size
= glsl_get_bit_size(ir
->return_deref
->type
);
1072 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1073 ir
->return_deref
->type
->vector_elements
,
1075 nir_builder_instr_insert(&b
, &instr
->instr
);
1079 unreachable("not reached");
1082 if (ir
->return_deref
) {
1083 nir_intrinsic_instr
*store_instr
=
1084 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
1085 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
1086 nir_intrinsic_set_write_mask(store_instr
,
1087 (1 << store_instr
->num_components
) - 1);
1089 store_instr
->variables
[0] =
1090 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
1091 store_instr
->src
[0] = nir_src_for_ssa(&dest
->ssa
);
1093 nir_builder_instr_insert(&b
, &store_instr
->instr
);
1099 struct hash_entry
*entry
=
1100 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
1102 nir_function
*callee
= (nir_function
*) entry
->data
;
1104 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
1107 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
1108 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
1112 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
1113 nir_builder_instr_insert(&b
, &instr
->instr
);
1117 nir_visitor::visit(ir_assignment
*ir
)
1119 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
1121 b
.exact
= ir
->lhs
->variable_referenced()->data
.invariant
||
1122 ir
->lhs
->variable_referenced()->data
.precise
;
1124 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
1125 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
1126 /* We're doing a plain-as-can-be copy, so emit a copy_var */
1127 nir_intrinsic_instr
*copy
=
1128 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
1130 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
1131 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
1133 if (ir
->condition
) {
1134 nir_if
*if_stmt
= nir_if_create(this->shader
);
1135 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1136 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1137 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
1138 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1140 nir_builder_instr_insert(&b
, ©
->instr
);
1145 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
1147 ir
->lhs
->accept(this);
1148 nir_deref_var
*lhs_deref
= this->deref_head
;
1149 nir_ssa_def
*src
= evaluate_rvalue(ir
->rhs
);
1151 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
1152 /* GLSL IR will give us the input to the write-masked assignment in a
1153 * single packed vector. So, for example, if the writemask is xzw, then
1154 * we have to swizzle x -> x, y -> z, and z -> w and get the y component
1158 unsigned component
= 0;
1159 for (unsigned i
= 0; i
< 4; i
++) {
1160 swiz
[i
] = ir
->write_mask
& (1 << i
) ? component
++ : 0;
1162 src
= nir_swizzle(&b
, src
, swiz
, num_components
, !supports_ints
);
1165 nir_intrinsic_instr
*store
=
1166 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1167 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1168 nir_intrinsic_set_write_mask(store
, ir
->write_mask
);
1169 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
1170 store
->variables
[0] = nir_deref_as_var(store_deref
);
1171 store
->src
[0] = nir_src_for_ssa(src
);
1173 if (ir
->condition
) {
1174 nir_if
*if_stmt
= nir_if_create(this->shader
);
1175 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1176 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1177 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
1178 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1180 nir_builder_instr_insert(&b
, &store
->instr
);
1185 * Given an instruction, returns a pointer to its destination or NULL if there
1186 * is no destination.
1188 * Note that this only handles instructions we generate at this level.
1191 get_instr_dest(nir_instr
*instr
)
1193 nir_alu_instr
*alu_instr
;
1194 nir_intrinsic_instr
*intrinsic_instr
;
1195 nir_tex_instr
*tex_instr
;
1197 switch (instr
->type
) {
1198 case nir_instr_type_alu
:
1199 alu_instr
= nir_instr_as_alu(instr
);
1200 return &alu_instr
->dest
.dest
;
1202 case nir_instr_type_intrinsic
:
1203 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1204 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1205 return &intrinsic_instr
->dest
;
1209 case nir_instr_type_tex
:
1210 tex_instr
= nir_instr_as_tex(instr
);
1211 return &tex_instr
->dest
;
1214 unreachable("not reached");
1221 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
,
1224 nir_dest
*dest
= get_instr_dest(instr
);
1227 nir_ssa_dest_init(instr
, dest
, num_components
, bit_size
, NULL
);
1229 nir_builder_instr_insert(&b
, instr
);
1232 assert(dest
->is_ssa
);
1233 this->result
= &dest
->ssa
;
1238 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1241 if (ir
->as_dereference() || ir
->as_constant()) {
1243 * A dereference is being used on the right hand side, which means we
1244 * must emit a variable load.
1247 nir_intrinsic_instr
*load_instr
=
1248 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1249 load_instr
->num_components
= ir
->type
->vector_elements
;
1250 load_instr
->variables
[0] = this->deref_head
;
1251 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1252 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1253 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
, bit_size
);
1256 return this->result
;
1260 type_is_float(glsl_base_type type
)
1262 return type
== GLSL_TYPE_FLOAT
|| type
== GLSL_TYPE_DOUBLE
;
1266 nir_visitor::visit(ir_expression
*ir
)
1268 /* Some special cases */
1269 switch (ir
->operation
) {
1270 case ir_binop_ubo_load
: {
1271 nir_intrinsic_instr
*load
=
1272 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_ubo
);
1273 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1274 load
->num_components
= ir
->type
->vector_elements
;
1275 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1276 load
->src
[1] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1277 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1280 * In UBO's, a true boolean value is any non-zero value, but we consider
1281 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1284 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
)
1285 this->result
= nir_ine(&b
, &load
->dest
.ssa
, nir_imm_int(&b
, 0));
1290 case ir_unop_interpolate_at_centroid
:
1291 case ir_binop_interpolate_at_offset
:
1292 case ir_binop_interpolate_at_sample
: {
1293 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1294 ir_swizzle
*swizzle
= NULL
;
1296 /* the api does not allow a swizzle here, but the varying packing code
1297 * may have pushed one into here.
1299 swizzle
= ir
->operands
[0]->as_swizzle();
1301 deref
= swizzle
->val
->as_dereference();
1305 deref
->accept(this);
1307 nir_intrinsic_op op
;
1308 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1309 switch (ir
->operation
) {
1310 case ir_unop_interpolate_at_centroid
:
1311 op
= nir_intrinsic_interp_var_at_centroid
;
1313 case ir_binop_interpolate_at_offset
:
1314 op
= nir_intrinsic_interp_var_at_offset
;
1316 case ir_binop_interpolate_at_sample
:
1317 op
= nir_intrinsic_interp_var_at_sample
;
1320 unreachable("Invalid interpolation intrinsic");
1323 /* This case can happen if the vertex shader does not write the
1324 * given varying. In this case, the linker will lower it to a
1325 * global variable. Since interpolating a variable makes no
1326 * sense, we'll just turn it into a load which will probably
1327 * eventually end up as an SSA definition.
1329 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1330 op
= nir_intrinsic_load_var
;
1333 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1334 intrin
->num_components
= deref
->type
->vector_elements
;
1335 intrin
->variables
[0] = this->deref_head
;
1336 ralloc_steal(intrin
, intrin
->variables
[0]);
1338 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1339 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1340 intrin
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1342 unsigned bit_size
= glsl_get_bit_size(deref
->type
);
1343 add_instr(&intrin
->instr
, deref
->type
->vector_elements
, bit_size
);
1346 unsigned swiz
[4] = {
1347 swizzle
->mask
.x
, swizzle
->mask
.y
, swizzle
->mask
.z
, swizzle
->mask
.w
1350 result
= nir_swizzle(&b
, result
, swiz
,
1351 swizzle
->type
->vector_elements
, false);
1361 nir_ssa_def
*srcs
[4];
1362 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1363 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1365 glsl_base_type types
[4];
1366 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1368 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1370 types
[i
] = GLSL_TYPE_FLOAT
;
1372 glsl_base_type out_type
;
1374 out_type
= ir
->type
->base_type
;
1376 out_type
= GLSL_TYPE_FLOAT
;
1378 switch (ir
->operation
) {
1379 case ir_unop_bit_not
: result
= nir_inot(&b
, srcs
[0]); break;
1380 case ir_unop_logic_not
:
1381 result
= supports_ints
? nir_inot(&b
, srcs
[0]) : nir_fnot(&b
, srcs
[0]);
1384 result
= type_is_float(types
[0]) ? nir_fneg(&b
, srcs
[0])
1385 : nir_ineg(&b
, srcs
[0]);
1388 result
= type_is_float(types
[0]) ? nir_fabs(&b
, srcs
[0])
1389 : nir_iabs(&b
, srcs
[0]);
1391 case ir_unop_saturate
:
1392 assert(type_is_float(types
[0]));
1393 result
= nir_fsat(&b
, srcs
[0]);
1396 result
= type_is_float(types
[0]) ? nir_fsign(&b
, srcs
[0])
1397 : nir_isign(&b
, srcs
[0]);
1399 case ir_unop_rcp
: result
= nir_frcp(&b
, srcs
[0]); break;
1400 case ir_unop_rsq
: result
= nir_frsq(&b
, srcs
[0]); break;
1401 case ir_unop_sqrt
: result
= nir_fsqrt(&b
, srcs
[0]); break;
1402 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1403 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1404 case ir_unop_exp2
: result
= nir_fexp2(&b
, srcs
[0]); break;
1405 case ir_unop_log2
: result
= nir_flog2(&b
, srcs
[0]); break;
1407 result
= supports_ints
? nir_i2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1410 result
= supports_ints
? nir_u2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1413 result
= supports_ints
? nir_b2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1415 case ir_unop_f2i
: result
= nir_f2i(&b
, srcs
[0]); break;
1416 case ir_unop_f2u
: result
= nir_f2u(&b
, srcs
[0]); break;
1417 case ir_unop_f2b
: result
= nir_f2b(&b
, srcs
[0]); break;
1418 case ir_unop_i2b
: result
= nir_i2b(&b
, srcs
[0]); break;
1419 case ir_unop_b2i
: result
= nir_b2i(&b
, srcs
[0]); break;
1420 case ir_unop_d2f
: result
= nir_d2f(&b
, srcs
[0]); break;
1421 case ir_unop_f2d
: result
= nir_f2d(&b
, srcs
[0]); break;
1422 case ir_unop_d2i
: result
= nir_d2i(&b
, srcs
[0]); break;
1423 case ir_unop_d2u
: result
= nir_d2u(&b
, srcs
[0]); break;
1424 case ir_unop_d2b
: result
= nir_d2b(&b
, srcs
[0]); break;
1426 assert(supports_ints
);
1427 result
= nir_i2d(&b
, srcs
[0]);
1430 assert(supports_ints
);
1431 result
= nir_u2d(&b
, srcs
[0]);
1435 case ir_unop_bitcast_i2f
:
1436 case ir_unop_bitcast_f2i
:
1437 case ir_unop_bitcast_u2f
:
1438 case ir_unop_bitcast_f2u
:
1439 case ir_unop_subroutine_to_int
:
1441 result
= nir_imov(&b
, srcs
[0]);
1443 case ir_unop_trunc
: result
= nir_ftrunc(&b
, srcs
[0]); break;
1444 case ir_unop_ceil
: result
= nir_fceil(&b
, srcs
[0]); break;
1445 case ir_unop_floor
: result
= nir_ffloor(&b
, srcs
[0]); break;
1446 case ir_unop_fract
: result
= nir_ffract(&b
, srcs
[0]); break;
1447 case ir_unop_round_even
: result
= nir_fround_even(&b
, srcs
[0]); break;
1448 case ir_unop_sin
: result
= nir_fsin(&b
, srcs
[0]); break;
1449 case ir_unop_cos
: result
= nir_fcos(&b
, srcs
[0]); break;
1450 case ir_unop_dFdx
: result
= nir_fddx(&b
, srcs
[0]); break;
1451 case ir_unop_dFdy
: result
= nir_fddy(&b
, srcs
[0]); break;
1452 case ir_unop_dFdx_fine
: result
= nir_fddx_fine(&b
, srcs
[0]); break;
1453 case ir_unop_dFdy_fine
: result
= nir_fddy_fine(&b
, srcs
[0]); break;
1454 case ir_unop_dFdx_coarse
: result
= nir_fddx_coarse(&b
, srcs
[0]); break;
1455 case ir_unop_dFdy_coarse
: result
= nir_fddy_coarse(&b
, srcs
[0]); break;
1456 case ir_unop_pack_snorm_2x16
:
1457 result
= nir_pack_snorm_2x16(&b
, srcs
[0]);
1459 case ir_unop_pack_snorm_4x8
:
1460 result
= nir_pack_snorm_4x8(&b
, srcs
[0]);
1462 case ir_unop_pack_unorm_2x16
:
1463 result
= nir_pack_unorm_2x16(&b
, srcs
[0]);
1465 case ir_unop_pack_unorm_4x8
:
1466 result
= nir_pack_unorm_4x8(&b
, srcs
[0]);
1468 case ir_unop_pack_half_2x16
:
1469 result
= nir_pack_half_2x16(&b
, srcs
[0]);
1471 case ir_unop_unpack_snorm_2x16
:
1472 result
= nir_unpack_snorm_2x16(&b
, srcs
[0]);
1474 case ir_unop_unpack_snorm_4x8
:
1475 result
= nir_unpack_snorm_4x8(&b
, srcs
[0]);
1477 case ir_unop_unpack_unorm_2x16
:
1478 result
= nir_unpack_unorm_2x16(&b
, srcs
[0]);
1480 case ir_unop_unpack_unorm_4x8
:
1481 result
= nir_unpack_unorm_4x8(&b
, srcs
[0]);
1483 case ir_unop_unpack_half_2x16
:
1484 result
= nir_unpack_half_2x16(&b
, srcs
[0]);
1486 case ir_unop_pack_double_2x32
:
1487 result
= nir_pack_double_2x32(&b
, srcs
[0]);
1489 case ir_unop_unpack_double_2x32
:
1490 result
= nir_unpack_double_2x32(&b
, srcs
[0]);
1492 case ir_unop_bitfield_reverse
:
1493 result
= nir_bitfield_reverse(&b
, srcs
[0]);
1495 case ir_unop_bit_count
:
1496 result
= nir_bit_count(&b
, srcs
[0]);
1498 case ir_unop_find_msb
:
1500 case GLSL_TYPE_UINT
:
1501 result
= nir_ufind_msb(&b
, srcs
[0]);
1504 result
= nir_ifind_msb(&b
, srcs
[0]);
1507 unreachable("Invalid type for findMSB()");
1510 case ir_unop_find_lsb
:
1511 result
= nir_find_lsb(&b
, srcs
[0]);
1515 switch (ir
->type
->vector_elements
) {
1517 switch (ir
->operands
[0]->type
->vector_elements
) {
1518 case 1: result
= nir_fnoise1_1(&b
, srcs
[0]); break;
1519 case 2: result
= nir_fnoise1_2(&b
, srcs
[0]); break;
1520 case 3: result
= nir_fnoise1_3(&b
, srcs
[0]); break;
1521 case 4: result
= nir_fnoise1_4(&b
, srcs
[0]); break;
1522 default: unreachable("not reached");
1526 switch (ir
->operands
[0]->type
->vector_elements
) {
1527 case 1: result
= nir_fnoise2_1(&b
, srcs
[0]); break;
1528 case 2: result
= nir_fnoise2_2(&b
, srcs
[0]); break;
1529 case 3: result
= nir_fnoise2_3(&b
, srcs
[0]); break;
1530 case 4: result
= nir_fnoise2_4(&b
, srcs
[0]); break;
1531 default: unreachable("not reached");
1535 switch (ir
->operands
[0]->type
->vector_elements
) {
1536 case 1: result
= nir_fnoise3_1(&b
, srcs
[0]); break;
1537 case 2: result
= nir_fnoise3_2(&b
, srcs
[0]); break;
1538 case 3: result
= nir_fnoise3_3(&b
, srcs
[0]); break;
1539 case 4: result
= nir_fnoise3_4(&b
, srcs
[0]); break;
1540 default: unreachable("not reached");
1544 switch (ir
->operands
[0]->type
->vector_elements
) {
1545 case 1: result
= nir_fnoise4_1(&b
, srcs
[0]); break;
1546 case 2: result
= nir_fnoise4_2(&b
, srcs
[0]); break;
1547 case 3: result
= nir_fnoise4_3(&b
, srcs
[0]); break;
1548 case 4: result
= nir_fnoise4_4(&b
, srcs
[0]); break;
1549 default: unreachable("not reached");
1553 unreachable("not reached");
1556 case ir_unop_get_buffer_size
: {
1557 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1559 nir_intrinsic_get_buffer_size
);
1560 load
->num_components
= ir
->type
->vector_elements
;
1561 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1562 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1563 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1568 result
= type_is_float(out_type
) ? nir_fadd(&b
, srcs
[0], srcs
[1])
1569 : nir_iadd(&b
, srcs
[0], srcs
[1]);
1572 result
= type_is_float(out_type
) ? nir_fsub(&b
, srcs
[0], srcs
[1])
1573 : nir_isub(&b
, srcs
[0], srcs
[1]);
1576 result
= type_is_float(out_type
) ? nir_fmul(&b
, srcs
[0], srcs
[1])
1577 : nir_imul(&b
, srcs
[0], srcs
[1]);
1580 if (type_is_float(out_type
))
1581 result
= nir_fdiv(&b
, srcs
[0], srcs
[1]);
1582 else if (out_type
== GLSL_TYPE_INT
)
1583 result
= nir_idiv(&b
, srcs
[0], srcs
[1]);
1585 result
= nir_udiv(&b
, srcs
[0], srcs
[1]);
1588 result
= type_is_float(out_type
) ? nir_fmod(&b
, srcs
[0], srcs
[1])
1589 : nir_umod(&b
, srcs
[0], srcs
[1]);
1592 if (type_is_float(out_type
))
1593 result
= nir_fmin(&b
, srcs
[0], srcs
[1]);
1594 else if (out_type
== GLSL_TYPE_INT
)
1595 result
= nir_imin(&b
, srcs
[0], srcs
[1]);
1597 result
= nir_umin(&b
, srcs
[0], srcs
[1]);
1600 if (type_is_float(out_type
))
1601 result
= nir_fmax(&b
, srcs
[0], srcs
[1]);
1602 else if (out_type
== GLSL_TYPE_INT
)
1603 result
= nir_imax(&b
, srcs
[0], srcs
[1]);
1605 result
= nir_umax(&b
, srcs
[0], srcs
[1]);
1607 case ir_binop_pow
: result
= nir_fpow(&b
, srcs
[0], srcs
[1]); break;
1608 case ir_binop_bit_and
: result
= nir_iand(&b
, srcs
[0], srcs
[1]); break;
1609 case ir_binop_bit_or
: result
= nir_ior(&b
, srcs
[0], srcs
[1]); break;
1610 case ir_binop_bit_xor
: result
= nir_ixor(&b
, srcs
[0], srcs
[1]); break;
1611 case ir_binop_logic_and
:
1612 result
= supports_ints
? nir_iand(&b
, srcs
[0], srcs
[1])
1613 : nir_fand(&b
, srcs
[0], srcs
[1]);
1615 case ir_binop_logic_or
:
1616 result
= supports_ints
? nir_ior(&b
, srcs
[0], srcs
[1])
1617 : nir_for(&b
, srcs
[0], srcs
[1]);
1619 case ir_binop_logic_xor
:
1620 result
= supports_ints
? nir_ixor(&b
, srcs
[0], srcs
[1])
1621 : nir_fxor(&b
, srcs
[0], srcs
[1]);
1623 case ir_binop_lshift
: result
= nir_ishl(&b
, srcs
[0], srcs
[1]); break;
1624 case ir_binop_rshift
:
1625 result
= (out_type
== GLSL_TYPE_INT
) ? nir_ishr(&b
, srcs
[0], srcs
[1])
1626 : nir_ushr(&b
, srcs
[0], srcs
[1]);
1628 case ir_binop_imul_high
:
1629 result
= (out_type
== GLSL_TYPE_INT
) ? nir_imul_high(&b
, srcs
[0], srcs
[1])
1630 : nir_umul_high(&b
, srcs
[0], srcs
[1]);
1632 case ir_binop_carry
: result
= nir_uadd_carry(&b
, srcs
[0], srcs
[1]); break;
1633 case ir_binop_borrow
: result
= nir_usub_borrow(&b
, srcs
[0], srcs
[1]); break;
1635 if (supports_ints
) {
1636 if (type_is_float(types
[0]))
1637 result
= nir_flt(&b
, srcs
[0], srcs
[1]);
1638 else if (types
[0] == GLSL_TYPE_INT
)
1639 result
= nir_ilt(&b
, srcs
[0], srcs
[1]);
1641 result
= nir_ult(&b
, srcs
[0], srcs
[1]);
1643 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1646 case ir_binop_greater
:
1647 if (supports_ints
) {
1648 if (type_is_float(types
[0]))
1649 result
= nir_flt(&b
, srcs
[1], srcs
[0]);
1650 else if (types
[0] == GLSL_TYPE_INT
)
1651 result
= nir_ilt(&b
, srcs
[1], srcs
[0]);
1653 result
= nir_ult(&b
, srcs
[1], srcs
[0]);
1655 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1658 case ir_binop_lequal
:
1659 if (supports_ints
) {
1660 if (type_is_float(types
[0]))
1661 result
= nir_fge(&b
, srcs
[1], srcs
[0]);
1662 else if (types
[0] == GLSL_TYPE_INT
)
1663 result
= nir_ige(&b
, srcs
[1], srcs
[0]);
1665 result
= nir_uge(&b
, srcs
[1], srcs
[0]);
1667 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1670 case ir_binop_gequal
:
1671 if (supports_ints
) {
1672 if (type_is_float(types
[0]))
1673 result
= nir_fge(&b
, srcs
[0], srcs
[1]);
1674 else if (types
[0] == GLSL_TYPE_INT
)
1675 result
= nir_ige(&b
, srcs
[0], srcs
[1]);
1677 result
= nir_uge(&b
, srcs
[0], srcs
[1]);
1679 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1682 case ir_binop_equal
:
1683 if (supports_ints
) {
1684 if (type_is_float(types
[0]))
1685 result
= nir_feq(&b
, srcs
[0], srcs
[1]);
1687 result
= nir_ieq(&b
, srcs
[0], srcs
[1]);
1689 result
= nir_seq(&b
, srcs
[0], srcs
[1]);
1692 case ir_binop_nequal
:
1693 if (supports_ints
) {
1694 if (type_is_float(types
[0]))
1695 result
= nir_fne(&b
, srcs
[0], srcs
[1]);
1697 result
= nir_ine(&b
, srcs
[0], srcs
[1]);
1699 result
= nir_sne(&b
, srcs
[0], srcs
[1]);
1702 case ir_binop_all_equal
:
1703 if (supports_ints
) {
1704 if (type_is_float(types
[0])) {
1705 switch (ir
->operands
[0]->type
->vector_elements
) {
1706 case 1: result
= nir_feq(&b
, srcs
[0], srcs
[1]); break;
1707 case 2: result
= nir_ball_fequal2(&b
, srcs
[0], srcs
[1]); break;
1708 case 3: result
= nir_ball_fequal3(&b
, srcs
[0], srcs
[1]); break;
1709 case 4: result
= nir_ball_fequal4(&b
, srcs
[0], srcs
[1]); break;
1711 unreachable("not reached");
1714 switch (ir
->operands
[0]->type
->vector_elements
) {
1715 case 1: result
= nir_ieq(&b
, srcs
[0], srcs
[1]); break;
1716 case 2: result
= nir_ball_iequal2(&b
, srcs
[0], srcs
[1]); break;
1717 case 3: result
= nir_ball_iequal3(&b
, srcs
[0], srcs
[1]); break;
1718 case 4: result
= nir_ball_iequal4(&b
, srcs
[0], srcs
[1]); break;
1720 unreachable("not reached");
1724 switch (ir
->operands
[0]->type
->vector_elements
) {
1725 case 1: result
= nir_seq(&b
, srcs
[0], srcs
[1]); break;
1726 case 2: result
= nir_fall_equal2(&b
, srcs
[0], srcs
[1]); break;
1727 case 3: result
= nir_fall_equal3(&b
, srcs
[0], srcs
[1]); break;
1728 case 4: result
= nir_fall_equal4(&b
, srcs
[0], srcs
[1]); break;
1730 unreachable("not reached");
1734 case ir_binop_any_nequal
:
1735 if (supports_ints
) {
1736 if (type_is_float(types
[0])) {
1737 switch (ir
->operands
[0]->type
->vector_elements
) {
1738 case 1: result
= nir_fne(&b
, srcs
[0], srcs
[1]); break;
1739 case 2: result
= nir_bany_fnequal2(&b
, srcs
[0], srcs
[1]); break;
1740 case 3: result
= nir_bany_fnequal3(&b
, srcs
[0], srcs
[1]); break;
1741 case 4: result
= nir_bany_fnequal4(&b
, srcs
[0], srcs
[1]); break;
1743 unreachable("not reached");
1746 switch (ir
->operands
[0]->type
->vector_elements
) {
1747 case 1: result
= nir_ine(&b
, srcs
[0], srcs
[1]); break;
1748 case 2: result
= nir_bany_inequal2(&b
, srcs
[0], srcs
[1]); break;
1749 case 3: result
= nir_bany_inequal3(&b
, srcs
[0], srcs
[1]); break;
1750 case 4: result
= nir_bany_inequal4(&b
, srcs
[0], srcs
[1]); break;
1752 unreachable("not reached");
1756 switch (ir
->operands
[0]->type
->vector_elements
) {
1757 case 1: result
= nir_sne(&b
, srcs
[0], srcs
[1]); break;
1758 case 2: result
= nir_fany_nequal2(&b
, srcs
[0], srcs
[1]); break;
1759 case 3: result
= nir_fany_nequal3(&b
, srcs
[0], srcs
[1]); break;
1760 case 4: result
= nir_fany_nequal4(&b
, srcs
[0], srcs
[1]); break;
1762 unreachable("not reached");
1767 switch (ir
->operands
[0]->type
->vector_elements
) {
1768 case 2: result
= nir_fdot2(&b
, srcs
[0], srcs
[1]); break;
1769 case 3: result
= nir_fdot3(&b
, srcs
[0], srcs
[1]); break;
1770 case 4: result
= nir_fdot4(&b
, srcs
[0], srcs
[1]); break;
1772 unreachable("not reached");
1776 case ir_binop_ldexp
: result
= nir_ldexp(&b
, srcs
[0], srcs
[1]); break;
1778 result
= nir_ffma(&b
, srcs
[0], srcs
[1], srcs
[2]);
1781 result
= nir_flrp(&b
, srcs
[0], srcs
[1], srcs
[2]);
1785 result
= nir_bcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1787 result
= nir_fcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1789 case ir_triop_bitfield_extract
:
1790 result
= (out_type
== GLSL_TYPE_INT
) ?
1791 nir_ibitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]) :
1792 nir_ubitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]);
1794 case ir_quadop_bitfield_insert
:
1795 result
= nir_bitfield_insert(&b
, srcs
[0], srcs
[1], srcs
[2], srcs
[3]);
1797 case ir_quadop_vector
:
1798 result
= nir_vec(&b
, srcs
, ir
->type
->vector_elements
);
1802 unreachable("not reached");
1807 nir_visitor::visit(ir_swizzle
*ir
)
1809 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1810 result
= nir_swizzle(&b
, evaluate_rvalue(ir
->val
), swizzle
,
1811 ir
->type
->vector_elements
, !supports_ints
);
1815 nir_visitor::visit(ir_texture
*ir
)
1822 num_srcs
= 1; /* coordinate */
1827 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1828 num_srcs
= 2; /* coordinate, bias/lod */
1832 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1838 if (ir
->lod_info
.lod
!= NULL
)
1839 num_srcs
= 2; /* coordinate, lod */
1841 num_srcs
= 1; /* coordinate */
1845 op
= nir_texop_txf_ms
;
1846 num_srcs
= 2; /* coordinate, sample_index */
1851 if (ir
->lod_info
.lod
!= NULL
)
1852 num_srcs
= 1; /* lod */
1859 num_srcs
= 1; /* coordinate */
1864 num_srcs
= 1; /* coordinate */
1867 case ir_query_levels
:
1868 op
= nir_texop_query_levels
;
1872 case ir_texture_samples
:
1873 op
= nir_texop_texture_samples
;
1877 case ir_samples_identical
:
1878 op
= nir_texop_samples_identical
;
1879 num_srcs
= 1; /* coordinate */
1883 unreachable("not reached");
1886 if (ir
->projector
!= NULL
)
1888 if (ir
->shadow_comparitor
!= NULL
)
1890 if (ir
->offset
!= NULL
)
1893 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1896 instr
->sampler_dim
=
1897 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1898 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1899 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1900 if (instr
->is_shadow
)
1901 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1902 switch (ir
->type
->base_type
) {
1903 case GLSL_TYPE_FLOAT
:
1904 instr
->dest_type
= nir_type_float
;
1907 instr
->dest_type
= nir_type_int
;
1909 case GLSL_TYPE_BOOL
:
1910 case GLSL_TYPE_UINT
:
1911 instr
->dest_type
= nir_type_uint
;
1914 unreachable("not reached");
1917 instr
->texture
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1919 unsigned src_number
= 0;
1921 if (ir
->coordinate
!= NULL
) {
1922 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1923 instr
->src
[src_number
].src
=
1924 nir_src_for_ssa(evaluate_rvalue(ir
->coordinate
));
1925 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1929 if (ir
->projector
!= NULL
) {
1930 instr
->src
[src_number
].src
=
1931 nir_src_for_ssa(evaluate_rvalue(ir
->projector
));
1932 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1936 if (ir
->shadow_comparitor
!= NULL
) {
1937 instr
->src
[src_number
].src
=
1938 nir_src_for_ssa(evaluate_rvalue(ir
->shadow_comparitor
));
1939 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1943 if (ir
->offset
!= NULL
) {
1944 /* we don't support multiple offsets yet */
1945 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1947 instr
->src
[src_number
].src
=
1948 nir_src_for_ssa(evaluate_rvalue(ir
->offset
));
1949 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1955 instr
->src
[src_number
].src
=
1956 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.bias
));
1957 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1964 if (ir
->lod_info
.lod
!= NULL
) {
1965 instr
->src
[src_number
].src
=
1966 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.lod
));
1967 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1973 instr
->src
[src_number
].src
=
1974 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdx
));
1975 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1977 instr
->src
[src_number
].src
=
1978 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdy
));
1979 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1984 instr
->src
[src_number
].src
=
1985 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.sample_index
));
1986 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1991 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1998 assert(src_number
== num_srcs
);
2000 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
2001 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
), bit_size
);
2005 nir_visitor::visit(ir_constant
*ir
)
2008 * We don't know if this variable is an array or struct that gets
2009 * dereferenced, so do the safe thing an make it a variable with a
2010 * constant initializer and return a dereference.
2014 nir_local_variable_create(this->impl
, ir
->type
, "const_temp");
2015 var
->data
.read_only
= true;
2016 var
->constant_initializer
= constant_copy(ir
, var
);
2018 this->deref_head
= nir_deref_var_create(this->shader
, var
);
2019 this->deref_tail
= &this->deref_head
->deref
;
2023 nir_visitor::visit(ir_dereference_variable
*ir
)
2025 struct hash_entry
*entry
=
2026 _mesa_hash_table_search(this->var_table
, ir
->var
);
2028 nir_variable
*var
= (nir_variable
*) entry
->data
;
2030 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
2031 this->deref_head
= deref
;
2032 this->deref_tail
= &deref
->deref
;
2036 nir_visitor::visit(ir_dereference_record
*ir
)
2038 ir
->record
->accept(this);
2040 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
2041 assert(field_index
>= 0);
2043 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
2044 deref
->deref
.type
= ir
->type
;
2045 this->deref_tail
->child
= &deref
->deref
;
2046 this->deref_tail
= &deref
->deref
;
2050 nir_visitor::visit(ir_dereference_array
*ir
)
2052 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
2053 deref
->deref
.type
= ir
->type
;
2055 ir_constant
*const_index
= ir
->array_index
->as_constant();
2056 if (const_index
!= NULL
) {
2057 deref
->deref_array_type
= nir_deref_array_type_direct
;
2058 deref
->base_offset
= const_index
->value
.u
[0];
2060 deref
->deref_array_type
= nir_deref_array_type_indirect
;
2062 nir_src_for_ssa(evaluate_rvalue(ir
->array_index
));
2065 ir
->array
->accept(this);
2067 this->deref_tail
->child
= &deref
->deref
;
2068 ralloc_steal(this->deref_tail
, deref
);
2069 this->deref_tail
= &deref
->deref
;
2073 nir_visitor::visit(ir_barrier
*)
2075 nir_intrinsic_instr
*instr
=
2076 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
2077 nir_builder_instr_insert(&b
, &instr
->instr
);