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
);
141 nir_visitor
v1(shader
);
142 nir_function_visitor
v2(&v1
);
144 visit_exec_list(sh
->ir
, &v1
);
146 shader
->info
.name
= ralloc_asprintf(shader
, "GLSL%d", shader_prog
->Name
);
147 if (shader_prog
->Label
)
148 shader
->info
.label
= ralloc_strdup(shader
, shader_prog
->Label
);
149 shader
->info
.num_textures
= util_last_bit(sh
->Program
->SamplersUsed
);
150 shader
->info
.num_ubos
= sh
->NumUniformBlocks
;
151 shader
->info
.num_abos
= shader_prog
->NumAtomicBuffers
;
152 shader
->info
.num_ssbos
= sh
->NumShaderStorageBlocks
;
153 shader
->info
.num_images
= sh
->NumImages
;
154 shader
->info
.inputs_read
= sh
->Program
->InputsRead
;
155 shader
->info
.double_inputs_read
= sh
->Program
->DoubleInputsRead
;
156 shader
->info
.outputs_written
= sh
->Program
->OutputsWritten
;
157 shader
->info
.outputs_read
= sh
->Program
->OutputsRead
;
158 shader
->info
.patch_inputs_read
= sh
->Program
->PatchInputsRead
;
159 shader
->info
.patch_outputs_written
= sh
->Program
->PatchOutputsWritten
;
160 shader
->info
.system_values_read
= sh
->Program
->SystemValuesRead
;
161 shader
->info
.uses_texture_gather
= sh
->Program
->UsesGather
;
162 shader
->info
.uses_clip_distance_out
=
163 sh
->Program
->ClipDistanceArraySize
!= 0;
164 shader
->info
.separate_shader
= shader_prog
->SeparateShader
;
165 shader
->info
.has_transform_feedback_varyings
=
166 shader_prog
->TransformFeedback
.NumVarying
> 0;
169 case MESA_SHADER_TESS_CTRL
:
170 shader
->info
.tcs
.vertices_out
= sh
->info
.TessCtrl
.VerticesOut
;
173 case MESA_SHADER_GEOMETRY
:
174 shader
->info
.gs
.vertices_in
= shader_prog
->Geom
.VerticesIn
;
175 shader
->info
.gs
.output_primitive
= sh
->info
.Geom
.OutputType
;
176 shader
->info
.gs
.vertices_out
= sh
->info
.Geom
.VerticesOut
;
177 shader
->info
.gs
.invocations
= sh
->info
.Geom
.Invocations
;
178 shader
->info
.gs
.uses_end_primitive
= shader_prog
->Geom
.UsesEndPrimitive
;
179 shader
->info
.gs
.uses_streams
= shader_prog
->Geom
.UsesStreams
;
182 case MESA_SHADER_FRAGMENT
: {
183 struct gl_fragment_program
*fp
=
184 (struct gl_fragment_program
*)sh
->Program
;
186 shader
->info
.fs
.uses_discard
= fp
->UsesKill
;
187 shader
->info
.fs
.uses_sample_qualifier
= fp
->IsSample
!= 0;
188 shader
->info
.fs
.early_fragment_tests
= sh
->info
.EarlyFragmentTests
;
189 shader
->info
.fs
.depth_layout
= fp
->FragDepthLayout
;
193 case MESA_SHADER_COMPUTE
: {
194 struct gl_compute_program
*cp
= (struct gl_compute_program
*)sh
->Program
;
195 shader
->info
.cs
.local_size
[0] = cp
->LocalSize
[0];
196 shader
->info
.cs
.local_size
[1] = cp
->LocalSize
[1];
197 shader
->info
.cs
.local_size
[2] = cp
->LocalSize
[2];
202 break; /* No stage-specific info */
208 nir_visitor::nir_visitor(nir_shader
*shader
)
210 this->supports_ints
= shader
->options
->native_integers
;
211 this->shader
= shader
;
212 this->is_global
= true;
213 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
214 _mesa_key_pointer_equal
);
215 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
216 _mesa_key_pointer_equal
);
220 this->deref_head
= NULL
;
221 this->deref_tail
= NULL
;
222 memset(&this->b
, 0, sizeof(this->b
));
225 nir_visitor::~nir_visitor()
227 _mesa_hash_table_destroy(this->var_table
, NULL
);
228 _mesa_hash_table_destroy(this->overload_table
, NULL
);
232 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
235 ralloc_steal(mem_ctx
, this->deref_head
);
236 return this->deref_head
;
239 static nir_constant
*
240 constant_copy(ir_constant
*ir
, void *mem_ctx
)
245 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
247 unsigned total_elems
= ir
->type
->components();
250 ret
->num_elements
= 0;
251 switch (ir
->type
->base_type
) {
253 for (i
= 0; i
< total_elems
; i
++)
254 ret
->value
.u
[i
] = ir
->value
.u
[i
];
258 for (i
= 0; i
< total_elems
; i
++)
259 ret
->value
.i
[i
] = ir
->value
.i
[i
];
262 case GLSL_TYPE_FLOAT
:
263 for (i
= 0; i
< total_elems
; i
++)
264 ret
->value
.f
[i
] = ir
->value
.f
[i
];
267 case GLSL_TYPE_DOUBLE
:
268 for (i
= 0; i
< total_elems
; i
++)
269 ret
->value
.d
[i
] = ir
->value
.d
[i
];
273 for (i
= 0; i
< total_elems
; i
++)
274 ret
->value
.b
[i
] = ir
->value
.b
[i
];
277 case GLSL_TYPE_STRUCT
:
278 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
280 ret
->num_elements
= ir
->type
->length
;
283 foreach_in_list(ir_constant
, field
, &ir
->components
) {
284 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
289 case GLSL_TYPE_ARRAY
:
290 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
292 ret
->num_elements
= ir
->type
->length
;
294 for (i
= 0; i
< ir
->type
->length
; i
++)
295 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
299 unreachable("not reached");
306 nir_visitor::visit(ir_variable
*ir
)
308 nir_variable
*var
= ralloc(shader
, nir_variable
);
309 var
->type
= ir
->type
;
310 var
->name
= ralloc_strdup(var
, ir
->name
);
312 var
->data
.read_only
= ir
->data
.read_only
;
313 var
->data
.centroid
= ir
->data
.centroid
;
314 var
->data
.sample
= ir
->data
.sample
;
315 var
->data
.patch
= ir
->data
.patch
;
316 var
->data
.invariant
= ir
->data
.invariant
;
317 var
->data
.location
= ir
->data
.location
;
319 switch(ir
->data
.mode
) {
321 case ir_var_temporary
:
323 var
->data
.mode
= nir_var_global
;
325 var
->data
.mode
= nir_var_local
;
328 case ir_var_function_in
:
329 case ir_var_function_out
:
330 case ir_var_function_inout
:
331 case ir_var_const_in
:
332 var
->data
.mode
= nir_var_local
;
335 case ir_var_shader_in
:
336 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
337 ir
->data
.location
== VARYING_SLOT_FACE
) {
338 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
339 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
340 var
->data
.mode
= nir_var_system_value
;
341 } else if (shader
->stage
== MESA_SHADER_GEOMETRY
&&
342 ir
->data
.location
== VARYING_SLOT_PRIMITIVE_ID
) {
343 /* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
344 var
->data
.location
= SYSTEM_VALUE_PRIMITIVE_ID
;
345 var
->data
.mode
= nir_var_system_value
;
347 var
->data
.mode
= nir_var_shader_in
;
351 case ir_var_shader_out
:
352 var
->data
.mode
= nir_var_shader_out
;
356 var
->data
.mode
= nir_var_uniform
;
359 case ir_var_shader_storage
:
360 var
->data
.mode
= nir_var_shader_storage
;
363 case ir_var_system_value
:
364 var
->data
.mode
= nir_var_system_value
;
368 unreachable("not reached");
371 var
->data
.interpolation
= ir
->data
.interpolation
;
372 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
373 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
374 var
->data
.explicit_location
= ir
->data
.explicit_location
;
375 var
->data
.explicit_index
= ir
->data
.explicit_index
;
376 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
377 var
->data
.has_initializer
= ir
->data
.has_initializer
;
378 var
->data
.location_frac
= ir
->data
.location_frac
;
380 switch (ir
->data
.depth_layout
) {
381 case ir_depth_layout_none
:
382 var
->data
.depth_layout
= nir_depth_layout_none
;
384 case ir_depth_layout_any
:
385 var
->data
.depth_layout
= nir_depth_layout_any
;
387 case ir_depth_layout_greater
:
388 var
->data
.depth_layout
= nir_depth_layout_greater
;
390 case ir_depth_layout_less
:
391 var
->data
.depth_layout
= nir_depth_layout_less
;
393 case ir_depth_layout_unchanged
:
394 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
397 unreachable("not reached");
400 var
->data
.index
= ir
->data
.index
;
401 var
->data
.binding
= ir
->data
.binding
;
402 var
->data
.offset
= ir
->data
.offset
;
403 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
404 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
405 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
406 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
407 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
408 var
->data
.image
.format
= ir
->data
.image_format
;
409 var
->data
.max_array_access
= ir
->data
.max_array_access
;
410 var
->data
.fb_fetch_output
= ir
->data
.fb_fetch_output
;
412 var
->num_state_slots
= ir
->get_num_state_slots();
413 if (var
->num_state_slots
> 0) {
414 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
415 var
->num_state_slots
);
417 ir_state_slot
*state_slots
= ir
->get_state_slots();
418 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
419 for (unsigned j
= 0; j
< 5; j
++)
420 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
421 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
424 var
->state_slots
= NULL
;
427 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
429 var
->interface_type
= ir
->get_interface_type();
431 if (var
->data
.mode
== nir_var_local
)
432 nir_function_impl_add_variable(impl
, var
);
434 nir_shader_add_variable(shader
, var
);
436 _mesa_hash_table_insert(var_table
, ir
, var
);
441 nir_function_visitor::visit_enter(ir_function
*ir
)
443 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
444 visitor
->create_function(sig
);
446 return visit_continue_with_parent
;
450 nir_visitor::create_function(ir_function_signature
*ir
)
452 if (ir
->is_intrinsic())
455 nir_function
*func
= nir_function_create(shader
, ir
->function_name());
457 assert(ir
->parameters
.is_empty());
458 assert(ir
->return_type
== glsl_type::void_type
);
460 _mesa_hash_table_insert(this->overload_table
, ir
, func
);
464 nir_visitor::visit(ir_function
*ir
)
466 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
471 nir_visitor::visit(ir_function_signature
*ir
)
473 if (ir
->is_intrinsic())
476 struct hash_entry
*entry
=
477 _mesa_hash_table_search(this->overload_table
, ir
);
480 nir_function
*func
= (nir_function
*) entry
->data
;
482 if (ir
->is_defined
) {
483 nir_function_impl
*impl
= nir_function_impl_create(func
);
486 assert(strcmp(func
->name
, "main") == 0);
487 assert(ir
->parameters
.is_empty());
488 assert(func
->return_type
== glsl_type::void_type
);
490 this->is_global
= false;
492 nir_builder_init(&b
, impl
);
493 b
.cursor
= nir_after_cf_list(&impl
->body
);
494 visit_exec_list(&ir
->body
, this);
496 this->is_global
= true;
503 nir_visitor::visit(ir_loop
*ir
)
505 nir_loop
*loop
= nir_loop_create(this->shader
);
506 nir_builder_cf_insert(&b
, &loop
->cf_node
);
508 b
.cursor
= nir_after_cf_list(&loop
->body
);
509 visit_exec_list(&ir
->body_instructions
, this);
510 b
.cursor
= nir_after_cf_node(&loop
->cf_node
);
514 nir_visitor::visit(ir_if
*ir
)
517 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
519 nir_if
*if_stmt
= nir_if_create(this->shader
);
520 if_stmt
->condition
= condition
;
521 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
523 b
.cursor
= nir_after_cf_list(&if_stmt
->then_list
);
524 visit_exec_list(&ir
->then_instructions
, this);
526 b
.cursor
= nir_after_cf_list(&if_stmt
->else_list
);
527 visit_exec_list(&ir
->else_instructions
, this);
529 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
533 nir_visitor::visit(ir_discard
*ir
)
536 * discards aren't treated as control flow, because before we lower them
537 * they can appear anywhere in the shader and the stuff after them may still
538 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
539 * discards will be immediately followed by a return.
542 nir_intrinsic_instr
*discard
;
544 discard
= nir_intrinsic_instr_create(this->shader
,
545 nir_intrinsic_discard_if
);
547 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
549 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
552 nir_builder_instr_insert(&b
, &discard
->instr
);
556 nir_visitor::visit(ir_emit_vertex
*ir
)
558 nir_intrinsic_instr
*instr
=
559 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
560 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
561 nir_builder_instr_insert(&b
, &instr
->instr
);
565 nir_visitor::visit(ir_end_primitive
*ir
)
567 nir_intrinsic_instr
*instr
=
568 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
569 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
570 nir_builder_instr_insert(&b
, &instr
->instr
);
574 nir_visitor::visit(ir_loop_jump
*ir
)
578 case ir_loop_jump::jump_break
:
579 type
= nir_jump_break
;
581 case ir_loop_jump::jump_continue
:
582 type
= nir_jump_continue
;
585 unreachable("not reached");
588 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
589 nir_builder_instr_insert(&b
, &instr
->instr
);
593 nir_visitor::visit(ir_return
*ir
)
595 if (ir
->value
!= NULL
) {
596 nir_intrinsic_instr
*copy
=
597 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
599 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
600 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
603 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
604 nir_builder_instr_insert(&b
, &instr
->instr
);
608 nir_visitor::visit(ir_call
*ir
)
610 if (ir
->callee
->is_intrinsic()) {
613 switch (ir
->callee
->intrinsic_id
) {
614 case ir_intrinsic_atomic_counter_read
:
615 op
= nir_intrinsic_atomic_counter_read_var
;
617 case ir_intrinsic_atomic_counter_increment
:
618 op
= nir_intrinsic_atomic_counter_inc_var
;
620 case ir_intrinsic_atomic_counter_predecrement
:
621 op
= nir_intrinsic_atomic_counter_dec_var
;
623 case ir_intrinsic_atomic_counter_add
:
624 op
= nir_intrinsic_atomic_counter_add_var
;
626 case ir_intrinsic_atomic_counter_and
:
627 op
= nir_intrinsic_atomic_counter_and_var
;
629 case ir_intrinsic_atomic_counter_or
:
630 op
= nir_intrinsic_atomic_counter_or_var
;
632 case ir_intrinsic_atomic_counter_xor
:
633 op
= nir_intrinsic_atomic_counter_xor_var
;
635 case ir_intrinsic_atomic_counter_min
:
636 op
= nir_intrinsic_atomic_counter_min_var
;
638 case ir_intrinsic_atomic_counter_max
:
639 op
= nir_intrinsic_atomic_counter_max_var
;
641 case ir_intrinsic_atomic_counter_exchange
:
642 op
= nir_intrinsic_atomic_counter_exchange_var
;
644 case ir_intrinsic_atomic_counter_comp_swap
:
645 op
= nir_intrinsic_atomic_counter_comp_swap_var
;
647 case ir_intrinsic_image_load
:
648 op
= nir_intrinsic_image_load
;
650 case ir_intrinsic_image_store
:
651 op
= nir_intrinsic_image_store
;
653 case ir_intrinsic_image_atomic_add
:
654 op
= nir_intrinsic_image_atomic_add
;
656 case ir_intrinsic_image_atomic_min
:
657 op
= nir_intrinsic_image_atomic_min
;
659 case ir_intrinsic_image_atomic_max
:
660 op
= nir_intrinsic_image_atomic_max
;
662 case ir_intrinsic_image_atomic_and
:
663 op
= nir_intrinsic_image_atomic_and
;
665 case ir_intrinsic_image_atomic_or
:
666 op
= nir_intrinsic_image_atomic_or
;
668 case ir_intrinsic_image_atomic_xor
:
669 op
= nir_intrinsic_image_atomic_xor
;
671 case ir_intrinsic_image_atomic_exchange
:
672 op
= nir_intrinsic_image_atomic_exchange
;
674 case ir_intrinsic_image_atomic_comp_swap
:
675 op
= nir_intrinsic_image_atomic_comp_swap
;
677 case ir_intrinsic_memory_barrier
:
678 op
= nir_intrinsic_memory_barrier
;
680 case ir_intrinsic_image_size
:
681 op
= nir_intrinsic_image_size
;
683 case ir_intrinsic_image_samples
:
684 op
= nir_intrinsic_image_samples
;
686 case ir_intrinsic_ssbo_store
:
687 op
= nir_intrinsic_store_ssbo
;
689 case ir_intrinsic_ssbo_load
:
690 op
= nir_intrinsic_load_ssbo
;
692 case ir_intrinsic_ssbo_atomic_add
:
693 op
= nir_intrinsic_ssbo_atomic_add
;
695 case ir_intrinsic_ssbo_atomic_and
:
696 op
= nir_intrinsic_ssbo_atomic_and
;
698 case ir_intrinsic_ssbo_atomic_or
:
699 op
= nir_intrinsic_ssbo_atomic_or
;
701 case ir_intrinsic_ssbo_atomic_xor
:
702 op
= nir_intrinsic_ssbo_atomic_xor
;
704 case ir_intrinsic_ssbo_atomic_min
:
705 assert(ir
->return_deref
);
706 if (ir
->return_deref
->type
== glsl_type::int_type
)
707 op
= nir_intrinsic_ssbo_atomic_imin
;
708 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
709 op
= nir_intrinsic_ssbo_atomic_umin
;
711 unreachable("Invalid type");
713 case ir_intrinsic_ssbo_atomic_max
:
714 assert(ir
->return_deref
);
715 if (ir
->return_deref
->type
== glsl_type::int_type
)
716 op
= nir_intrinsic_ssbo_atomic_imax
;
717 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
718 op
= nir_intrinsic_ssbo_atomic_umax
;
720 unreachable("Invalid type");
722 case ir_intrinsic_ssbo_atomic_exchange
:
723 op
= nir_intrinsic_ssbo_atomic_exchange
;
725 case ir_intrinsic_ssbo_atomic_comp_swap
:
726 op
= nir_intrinsic_ssbo_atomic_comp_swap
;
728 case ir_intrinsic_shader_clock
:
729 op
= nir_intrinsic_shader_clock
;
731 case ir_intrinsic_group_memory_barrier
:
732 op
= nir_intrinsic_group_memory_barrier
;
734 case ir_intrinsic_memory_barrier_atomic_counter
:
735 op
= nir_intrinsic_memory_barrier_atomic_counter
;
737 case ir_intrinsic_memory_barrier_buffer
:
738 op
= nir_intrinsic_memory_barrier_buffer
;
740 case ir_intrinsic_memory_barrier_image
:
741 op
= nir_intrinsic_memory_barrier_image
;
743 case ir_intrinsic_memory_barrier_shared
:
744 op
= nir_intrinsic_memory_barrier_shared
;
746 case ir_intrinsic_shared_load
:
747 op
= nir_intrinsic_load_shared
;
749 case ir_intrinsic_shared_store
:
750 op
= nir_intrinsic_store_shared
;
752 case ir_intrinsic_shared_atomic_add
:
753 op
= nir_intrinsic_shared_atomic_add
;
755 case ir_intrinsic_shared_atomic_and
:
756 op
= nir_intrinsic_shared_atomic_and
;
758 case ir_intrinsic_shared_atomic_or
:
759 op
= nir_intrinsic_shared_atomic_or
;
761 case ir_intrinsic_shared_atomic_xor
:
762 op
= nir_intrinsic_shared_atomic_xor
;
764 case ir_intrinsic_shared_atomic_min
:
765 assert(ir
->return_deref
);
766 if (ir
->return_deref
->type
== glsl_type::int_type
)
767 op
= nir_intrinsic_shared_atomic_imin
;
768 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
769 op
= nir_intrinsic_shared_atomic_umin
;
771 unreachable("Invalid type");
773 case ir_intrinsic_shared_atomic_max
:
774 assert(ir
->return_deref
);
775 if (ir
->return_deref
->type
== glsl_type::int_type
)
776 op
= nir_intrinsic_shared_atomic_imax
;
777 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
778 op
= nir_intrinsic_shared_atomic_umax
;
780 unreachable("Invalid type");
782 case ir_intrinsic_shared_atomic_exchange
:
783 op
= nir_intrinsic_shared_atomic_exchange
;
785 case ir_intrinsic_shared_atomic_comp_swap
:
786 op
= nir_intrinsic_shared_atomic_comp_swap
;
789 unreachable("not reached");
792 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
793 nir_dest
*dest
= &instr
->dest
;
796 case nir_intrinsic_atomic_counter_read_var
:
797 case nir_intrinsic_atomic_counter_inc_var
:
798 case nir_intrinsic_atomic_counter_dec_var
:
799 case nir_intrinsic_atomic_counter_add_var
:
800 case nir_intrinsic_atomic_counter_min_var
:
801 case nir_intrinsic_atomic_counter_max_var
:
802 case nir_intrinsic_atomic_counter_and_var
:
803 case nir_intrinsic_atomic_counter_or_var
:
804 case nir_intrinsic_atomic_counter_xor_var
:
805 case nir_intrinsic_atomic_counter_exchange_var
:
806 case nir_intrinsic_atomic_counter_comp_swap_var
: {
807 /* Set the counter variable dereference. */
808 exec_node
*param
= ir
->actual_parameters
.get_head();
809 ir_dereference
*counter
= (ir_dereference
*)param
;
811 instr
->variables
[0] = evaluate_deref(&instr
->instr
, counter
);
812 param
= param
->get_next();
814 /* Set the intrinsic destination. */
815 if (ir
->return_deref
) {
816 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
819 /* Set the intrinsic parameters. */
820 if (!param
->is_tail_sentinel()) {
822 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
823 param
= param
->get_next();
826 if (!param
->is_tail_sentinel()) {
828 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
829 param
= param
->get_next();
832 nir_builder_instr_insert(&b
, &instr
->instr
);
835 case nir_intrinsic_image_load
:
836 case nir_intrinsic_image_store
:
837 case nir_intrinsic_image_atomic_add
:
838 case nir_intrinsic_image_atomic_min
:
839 case nir_intrinsic_image_atomic_max
:
840 case nir_intrinsic_image_atomic_and
:
841 case nir_intrinsic_image_atomic_or
:
842 case nir_intrinsic_image_atomic_xor
:
843 case nir_intrinsic_image_atomic_exchange
:
844 case nir_intrinsic_image_atomic_comp_swap
:
845 case nir_intrinsic_image_samples
:
846 case nir_intrinsic_image_size
: {
847 nir_ssa_undef_instr
*instr_undef
=
848 nir_ssa_undef_instr_create(shader
, 1, 32);
849 nir_builder_instr_insert(&b
, &instr_undef
->instr
);
851 /* Set the image variable dereference. */
852 exec_node
*param
= ir
->actual_parameters
.get_head();
853 ir_dereference
*image
= (ir_dereference
*)param
;
854 const glsl_type
*type
=
855 image
->variable_referenced()->type
->without_array();
857 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
858 param
= param
->get_next();
860 /* Set the intrinsic destination. */
861 if (ir
->return_deref
) {
862 const nir_intrinsic_info
*info
=
863 &nir_intrinsic_infos
[instr
->intrinsic
];
864 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
865 info
->dest_components
, 32, NULL
);
868 if (op
== nir_intrinsic_image_size
||
869 op
== nir_intrinsic_image_samples
) {
870 nir_builder_instr_insert(&b
, &instr
->instr
);
874 /* Set the address argument, extending the coordinate vector to four
877 nir_ssa_def
*src_addr
=
878 evaluate_rvalue((ir_dereference
*)param
);
879 nir_ssa_def
*srcs
[4];
881 for (int i
= 0; i
< 4; i
++) {
882 if (i
< type
->coordinate_components())
883 srcs
[i
] = nir_channel(&b
, src_addr
, i
);
885 srcs
[i
] = &instr_undef
->def
;
888 instr
->src
[0] = nir_src_for_ssa(nir_vec(&b
, srcs
, 4));
889 param
= param
->get_next();
891 /* Set the sample argument, which is undefined for single-sample
894 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
896 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
897 param
= param
->get_next();
899 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
902 /* Set the intrinsic parameters. */
903 if (!param
->is_tail_sentinel()) {
905 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
906 param
= param
->get_next();
909 if (!param
->is_tail_sentinel()) {
911 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
912 param
= param
->get_next();
914 nir_builder_instr_insert(&b
, &instr
->instr
);
917 case nir_intrinsic_memory_barrier
:
918 case nir_intrinsic_group_memory_barrier
:
919 case nir_intrinsic_memory_barrier_atomic_counter
:
920 case nir_intrinsic_memory_barrier_buffer
:
921 case nir_intrinsic_memory_barrier_image
:
922 case nir_intrinsic_memory_barrier_shared
:
923 nir_builder_instr_insert(&b
, &instr
->instr
);
925 case nir_intrinsic_shader_clock
:
926 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
927 nir_builder_instr_insert(&b
, &instr
->instr
);
929 case nir_intrinsic_store_ssbo
: {
930 exec_node
*param
= ir
->actual_parameters
.get_head();
931 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
933 param
= param
->get_next();
934 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
936 param
= param
->get_next();
937 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
939 param
= param
->get_next();
940 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
943 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
944 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(block
));
945 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(offset
));
946 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
947 instr
->num_components
= val
->type
->vector_elements
;
949 nir_builder_instr_insert(&b
, &instr
->instr
);
952 case nir_intrinsic_load_ssbo
: {
953 exec_node
*param
= ir
->actual_parameters
.get_head();
954 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
956 param
= param
->get_next();
957 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
959 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(block
));
960 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
962 const glsl_type
*type
= ir
->return_deref
->var
->type
;
963 instr
->num_components
= type
->vector_elements
;
965 /* Setup destination register */
966 unsigned bit_size
= glsl_get_bit_size(type
);
967 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
968 type
->vector_elements
, bit_size
, NULL
);
970 /* Insert the created nir instruction now since in the case of boolean
971 * result we will need to emit another instruction after it
973 nir_builder_instr_insert(&b
, &instr
->instr
);
976 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
977 * consider a true boolean to be ~0. Fix this up with a != 0
980 if (type
->base_type
== GLSL_TYPE_BOOL
) {
981 nir_alu_instr
*load_ssbo_compare
=
982 nir_alu_instr_create(shader
, nir_op_ine
);
983 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
984 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
985 load_ssbo_compare
->src
[1].src
=
986 nir_src_for_ssa(nir_imm_int(&b
, 0));
987 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
988 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
989 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
990 &load_ssbo_compare
->dest
.dest
,
991 type
->vector_elements
, bit_size
, NULL
);
992 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
993 nir_builder_instr_insert(&b
, &load_ssbo_compare
->instr
);
994 dest
= &load_ssbo_compare
->dest
.dest
;
998 case nir_intrinsic_ssbo_atomic_add
:
999 case nir_intrinsic_ssbo_atomic_imin
:
1000 case nir_intrinsic_ssbo_atomic_umin
:
1001 case nir_intrinsic_ssbo_atomic_imax
:
1002 case nir_intrinsic_ssbo_atomic_umax
:
1003 case nir_intrinsic_ssbo_atomic_and
:
1004 case nir_intrinsic_ssbo_atomic_or
:
1005 case nir_intrinsic_ssbo_atomic_xor
:
1006 case nir_intrinsic_ssbo_atomic_exchange
:
1007 case nir_intrinsic_ssbo_atomic_comp_swap
: {
1008 int param_count
= ir
->actual_parameters
.length();
1009 assert(param_count
== 3 || param_count
== 4);
1012 exec_node
*param
= ir
->actual_parameters
.get_head();
1013 ir_instruction
*inst
= (ir_instruction
*) param
;
1014 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1017 param
= param
->get_next();
1018 inst
= (ir_instruction
*) param
;
1019 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1021 /* data1 parameter (this is always present) */
1022 param
= param
->get_next();
1023 inst
= (ir_instruction
*) param
;
1024 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1026 /* data2 parameter (only with atomic_comp_swap) */
1027 if (param_count
== 4) {
1028 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
1029 param
= param
->get_next();
1030 inst
= (ir_instruction
*) param
;
1031 instr
->src
[3] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1035 assert(ir
->return_deref
);
1036 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1037 ir
->return_deref
->type
->vector_elements
, 32, NULL
);
1038 nir_builder_instr_insert(&b
, &instr
->instr
);
1041 case nir_intrinsic_load_shared
: {
1042 exec_node
*param
= ir
->actual_parameters
.get_head();
1043 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1045 nir_intrinsic_set_base(instr
, 0);
1046 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(offset
));
1048 const glsl_type
*type
= ir
->return_deref
->var
->type
;
1049 instr
->num_components
= type
->vector_elements
;
1051 /* Setup destination register */
1052 unsigned bit_size
= glsl_get_bit_size(type
);
1053 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1054 type
->vector_elements
, bit_size
, NULL
);
1056 nir_builder_instr_insert(&b
, &instr
->instr
);
1059 case nir_intrinsic_store_shared
: {
1060 exec_node
*param
= ir
->actual_parameters
.get_head();
1061 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1063 param
= param
->get_next();
1064 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
1066 param
= param
->get_next();
1067 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
1070 nir_intrinsic_set_base(instr
, 0);
1071 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
1073 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
1075 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
1076 instr
->num_components
= val
->type
->vector_elements
;
1078 nir_builder_instr_insert(&b
, &instr
->instr
);
1081 case nir_intrinsic_shared_atomic_add
:
1082 case nir_intrinsic_shared_atomic_imin
:
1083 case nir_intrinsic_shared_atomic_umin
:
1084 case nir_intrinsic_shared_atomic_imax
:
1085 case nir_intrinsic_shared_atomic_umax
:
1086 case nir_intrinsic_shared_atomic_and
:
1087 case nir_intrinsic_shared_atomic_or
:
1088 case nir_intrinsic_shared_atomic_xor
:
1089 case nir_intrinsic_shared_atomic_exchange
:
1090 case nir_intrinsic_shared_atomic_comp_swap
: {
1091 int param_count
= ir
->actual_parameters
.length();
1092 assert(param_count
== 2 || param_count
== 3);
1095 exec_node
*param
= ir
->actual_parameters
.get_head();
1096 ir_instruction
*inst
= (ir_instruction
*) param
;
1097 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1099 /* data1 parameter (this is always present) */
1100 param
= param
->get_next();
1101 inst
= (ir_instruction
*) param
;
1102 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1104 /* data2 parameter (only with atomic_comp_swap) */
1105 if (param_count
== 3) {
1106 assert(op
== nir_intrinsic_shared_atomic_comp_swap
);
1107 param
= param
->get_next();
1108 inst
= (ir_instruction
*) param
;
1110 nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1114 assert(ir
->return_deref
);
1115 unsigned bit_size
= glsl_get_bit_size(ir
->return_deref
->type
);
1116 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1117 ir
->return_deref
->type
->vector_elements
,
1119 nir_builder_instr_insert(&b
, &instr
->instr
);
1123 unreachable("not reached");
1126 if (ir
->return_deref
) {
1127 nir_intrinsic_instr
*store_instr
=
1128 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
1129 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
1130 nir_intrinsic_set_write_mask(store_instr
,
1131 (1 << store_instr
->num_components
) - 1);
1133 store_instr
->variables
[0] =
1134 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
1135 store_instr
->src
[0] = nir_src_for_ssa(&dest
->ssa
);
1137 nir_builder_instr_insert(&b
, &store_instr
->instr
);
1143 struct hash_entry
*entry
=
1144 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
1146 nir_function
*callee
= (nir_function
*) entry
->data
;
1148 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
1151 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
1152 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
1156 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
1157 nir_builder_instr_insert(&b
, &instr
->instr
);
1161 nir_visitor::visit(ir_assignment
*ir
)
1163 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
1165 b
.exact
= ir
->lhs
->variable_referenced()->data
.invariant
||
1166 ir
->lhs
->variable_referenced()->data
.precise
;
1168 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
1169 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
1170 /* We're doing a plain-as-can-be copy, so emit a copy_var */
1171 nir_intrinsic_instr
*copy
=
1172 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
1174 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
1175 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
1177 if (ir
->condition
) {
1178 nir_if
*if_stmt
= nir_if_create(this->shader
);
1179 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1180 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1181 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
1182 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1184 nir_builder_instr_insert(&b
, ©
->instr
);
1189 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
1191 ir
->lhs
->accept(this);
1192 nir_deref_var
*lhs_deref
= this->deref_head
;
1193 nir_ssa_def
*src
= evaluate_rvalue(ir
->rhs
);
1195 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
1196 /* GLSL IR will give us the input to the write-masked assignment in a
1197 * single packed vector. So, for example, if the writemask is xzw, then
1198 * we have to swizzle x -> x, y -> z, and z -> w and get the y component
1202 unsigned component
= 0;
1203 for (unsigned i
= 0; i
< 4; i
++) {
1204 swiz
[i
] = ir
->write_mask
& (1 << i
) ? component
++ : 0;
1206 src
= nir_swizzle(&b
, src
, swiz
, num_components
, !supports_ints
);
1209 nir_intrinsic_instr
*store
=
1210 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1211 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1212 nir_intrinsic_set_write_mask(store
, ir
->write_mask
);
1213 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
1214 store
->variables
[0] = nir_deref_as_var(store_deref
);
1215 store
->src
[0] = nir_src_for_ssa(src
);
1217 if (ir
->condition
) {
1218 nir_if
*if_stmt
= nir_if_create(this->shader
);
1219 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1220 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1221 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
1222 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1224 nir_builder_instr_insert(&b
, &store
->instr
);
1229 * Given an instruction, returns a pointer to its destination or NULL if there
1230 * is no destination.
1232 * Note that this only handles instructions we generate at this level.
1235 get_instr_dest(nir_instr
*instr
)
1237 nir_alu_instr
*alu_instr
;
1238 nir_intrinsic_instr
*intrinsic_instr
;
1239 nir_tex_instr
*tex_instr
;
1241 switch (instr
->type
) {
1242 case nir_instr_type_alu
:
1243 alu_instr
= nir_instr_as_alu(instr
);
1244 return &alu_instr
->dest
.dest
;
1246 case nir_instr_type_intrinsic
:
1247 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1248 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1249 return &intrinsic_instr
->dest
;
1253 case nir_instr_type_tex
:
1254 tex_instr
= nir_instr_as_tex(instr
);
1255 return &tex_instr
->dest
;
1258 unreachable("not reached");
1265 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
,
1268 nir_dest
*dest
= get_instr_dest(instr
);
1271 nir_ssa_dest_init(instr
, dest
, num_components
, bit_size
, NULL
);
1273 nir_builder_instr_insert(&b
, instr
);
1276 assert(dest
->is_ssa
);
1277 this->result
= &dest
->ssa
;
1282 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1285 if (ir
->as_dereference() || ir
->as_constant()) {
1287 * A dereference is being used on the right hand side, which means we
1288 * must emit a variable load.
1291 nir_intrinsic_instr
*load_instr
=
1292 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1293 load_instr
->num_components
= ir
->type
->vector_elements
;
1294 load_instr
->variables
[0] = this->deref_head
;
1295 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1296 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1297 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
, bit_size
);
1300 return this->result
;
1304 type_is_float(glsl_base_type type
)
1306 return type
== GLSL_TYPE_FLOAT
|| type
== GLSL_TYPE_DOUBLE
;
1310 nir_visitor::visit(ir_expression
*ir
)
1312 /* Some special cases */
1313 switch (ir
->operation
) {
1314 case ir_binop_ubo_load
: {
1315 nir_intrinsic_instr
*load
=
1316 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_ubo
);
1317 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1318 load
->num_components
= ir
->type
->vector_elements
;
1319 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1320 load
->src
[1] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1321 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1324 * In UBO's, a true boolean value is any non-zero value, but we consider
1325 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1328 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
)
1329 this->result
= nir_ine(&b
, &load
->dest
.ssa
, nir_imm_int(&b
, 0));
1334 case ir_unop_interpolate_at_centroid
:
1335 case ir_binop_interpolate_at_offset
:
1336 case ir_binop_interpolate_at_sample
: {
1337 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1338 ir_swizzle
*swizzle
= NULL
;
1340 /* the api does not allow a swizzle here, but the varying packing code
1341 * may have pushed one into here.
1343 swizzle
= ir
->operands
[0]->as_swizzle();
1345 deref
= swizzle
->val
->as_dereference();
1349 deref
->accept(this);
1351 nir_intrinsic_op op
;
1352 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1353 switch (ir
->operation
) {
1354 case ir_unop_interpolate_at_centroid
:
1355 op
= nir_intrinsic_interp_var_at_centroid
;
1357 case ir_binop_interpolate_at_offset
:
1358 op
= nir_intrinsic_interp_var_at_offset
;
1360 case ir_binop_interpolate_at_sample
:
1361 op
= nir_intrinsic_interp_var_at_sample
;
1364 unreachable("Invalid interpolation intrinsic");
1367 /* This case can happen if the vertex shader does not write the
1368 * given varying. In this case, the linker will lower it to a
1369 * global variable. Since interpolating a variable makes no
1370 * sense, we'll just turn it into a load which will probably
1371 * eventually end up as an SSA definition.
1373 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1374 op
= nir_intrinsic_load_var
;
1377 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1378 intrin
->num_components
= deref
->type
->vector_elements
;
1379 intrin
->variables
[0] = this->deref_head
;
1380 ralloc_steal(intrin
, intrin
->variables
[0]);
1382 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1383 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1384 intrin
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1386 unsigned bit_size
= glsl_get_bit_size(deref
->type
);
1387 add_instr(&intrin
->instr
, deref
->type
->vector_elements
, bit_size
);
1390 unsigned swiz
[4] = {
1391 swizzle
->mask
.x
, swizzle
->mask
.y
, swizzle
->mask
.z
, swizzle
->mask
.w
1394 result
= nir_swizzle(&b
, result
, swiz
,
1395 swizzle
->type
->vector_elements
, false);
1405 nir_ssa_def
*srcs
[4];
1406 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1407 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1409 glsl_base_type types
[4];
1410 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1412 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1414 types
[i
] = GLSL_TYPE_FLOAT
;
1416 glsl_base_type out_type
;
1418 out_type
= ir
->type
->base_type
;
1420 out_type
= GLSL_TYPE_FLOAT
;
1422 switch (ir
->operation
) {
1423 case ir_unop_bit_not
: result
= nir_inot(&b
, srcs
[0]); break;
1424 case ir_unop_logic_not
:
1425 result
= supports_ints
? nir_inot(&b
, srcs
[0]) : nir_fnot(&b
, srcs
[0]);
1428 result
= type_is_float(types
[0]) ? nir_fneg(&b
, srcs
[0])
1429 : nir_ineg(&b
, srcs
[0]);
1432 result
= type_is_float(types
[0]) ? nir_fabs(&b
, srcs
[0])
1433 : nir_iabs(&b
, srcs
[0]);
1435 case ir_unop_saturate
:
1436 assert(type_is_float(types
[0]));
1437 result
= nir_fsat(&b
, srcs
[0]);
1440 result
= type_is_float(types
[0]) ? nir_fsign(&b
, srcs
[0])
1441 : nir_isign(&b
, srcs
[0]);
1443 case ir_unop_rcp
: result
= nir_frcp(&b
, srcs
[0]); break;
1444 case ir_unop_rsq
: result
= nir_frsq(&b
, srcs
[0]); break;
1445 case ir_unop_sqrt
: result
= nir_fsqrt(&b
, srcs
[0]); break;
1446 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1447 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1448 case ir_unop_exp2
: result
= nir_fexp2(&b
, srcs
[0]); break;
1449 case ir_unop_log2
: result
= nir_flog2(&b
, srcs
[0]); break;
1451 result
= supports_ints
? nir_i2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1454 result
= supports_ints
? nir_u2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1457 result
= supports_ints
? nir_b2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1459 case ir_unop_f2i
: result
= nir_f2i(&b
, srcs
[0]); break;
1460 case ir_unop_f2u
: result
= nir_f2u(&b
, srcs
[0]); break;
1461 case ir_unop_f2b
: result
= nir_f2b(&b
, srcs
[0]); break;
1462 case ir_unop_i2b
: result
= nir_i2b(&b
, srcs
[0]); break;
1463 case ir_unop_b2i
: result
= nir_b2i(&b
, srcs
[0]); break;
1464 case ir_unop_d2f
: result
= nir_d2f(&b
, srcs
[0]); break;
1465 case ir_unop_f2d
: result
= nir_f2d(&b
, srcs
[0]); break;
1466 case ir_unop_d2i
: result
= nir_d2i(&b
, srcs
[0]); break;
1467 case ir_unop_d2u
: result
= nir_d2u(&b
, srcs
[0]); break;
1468 case ir_unop_d2b
: result
= nir_d2b(&b
, srcs
[0]); break;
1470 assert(supports_ints
);
1471 result
= nir_i2d(&b
, srcs
[0]);
1474 assert(supports_ints
);
1475 result
= nir_u2d(&b
, srcs
[0]);
1479 case ir_unop_bitcast_i2f
:
1480 case ir_unop_bitcast_f2i
:
1481 case ir_unop_bitcast_u2f
:
1482 case ir_unop_bitcast_f2u
:
1483 case ir_unop_subroutine_to_int
:
1485 result
= nir_imov(&b
, srcs
[0]);
1487 case ir_unop_trunc
: result
= nir_ftrunc(&b
, srcs
[0]); break;
1488 case ir_unop_ceil
: result
= nir_fceil(&b
, srcs
[0]); break;
1489 case ir_unop_floor
: result
= nir_ffloor(&b
, srcs
[0]); break;
1490 case ir_unop_fract
: result
= nir_ffract(&b
, srcs
[0]); break;
1491 case ir_unop_round_even
: result
= nir_fround_even(&b
, srcs
[0]); break;
1492 case ir_unop_sin
: result
= nir_fsin(&b
, srcs
[0]); break;
1493 case ir_unop_cos
: result
= nir_fcos(&b
, srcs
[0]); break;
1494 case ir_unop_dFdx
: result
= nir_fddx(&b
, srcs
[0]); break;
1495 case ir_unop_dFdy
: result
= nir_fddy(&b
, srcs
[0]); break;
1496 case ir_unop_dFdx_fine
: result
= nir_fddx_fine(&b
, srcs
[0]); break;
1497 case ir_unop_dFdy_fine
: result
= nir_fddy_fine(&b
, srcs
[0]); break;
1498 case ir_unop_dFdx_coarse
: result
= nir_fddx_coarse(&b
, srcs
[0]); break;
1499 case ir_unop_dFdy_coarse
: result
= nir_fddy_coarse(&b
, srcs
[0]); break;
1500 case ir_unop_pack_snorm_2x16
:
1501 result
= nir_pack_snorm_2x16(&b
, srcs
[0]);
1503 case ir_unop_pack_snorm_4x8
:
1504 result
= nir_pack_snorm_4x8(&b
, srcs
[0]);
1506 case ir_unop_pack_unorm_2x16
:
1507 result
= nir_pack_unorm_2x16(&b
, srcs
[0]);
1509 case ir_unop_pack_unorm_4x8
:
1510 result
= nir_pack_unorm_4x8(&b
, srcs
[0]);
1512 case ir_unop_pack_half_2x16
:
1513 result
= nir_pack_half_2x16(&b
, srcs
[0]);
1515 case ir_unop_unpack_snorm_2x16
:
1516 result
= nir_unpack_snorm_2x16(&b
, srcs
[0]);
1518 case ir_unop_unpack_snorm_4x8
:
1519 result
= nir_unpack_snorm_4x8(&b
, srcs
[0]);
1521 case ir_unop_unpack_unorm_2x16
:
1522 result
= nir_unpack_unorm_2x16(&b
, srcs
[0]);
1524 case ir_unop_unpack_unorm_4x8
:
1525 result
= nir_unpack_unorm_4x8(&b
, srcs
[0]);
1527 case ir_unop_unpack_half_2x16
:
1528 result
= nir_unpack_half_2x16(&b
, srcs
[0]);
1530 case ir_unop_pack_double_2x32
:
1531 result
= nir_pack_double_2x32(&b
, srcs
[0]);
1533 case ir_unop_unpack_double_2x32
:
1534 result
= nir_unpack_double_2x32(&b
, srcs
[0]);
1536 case ir_unop_bitfield_reverse
:
1537 result
= nir_bitfield_reverse(&b
, srcs
[0]);
1539 case ir_unop_bit_count
:
1540 result
= nir_bit_count(&b
, srcs
[0]);
1542 case ir_unop_find_msb
:
1544 case GLSL_TYPE_UINT
:
1545 result
= nir_ufind_msb(&b
, srcs
[0]);
1548 result
= nir_ifind_msb(&b
, srcs
[0]);
1551 unreachable("Invalid type for findMSB()");
1554 case ir_unop_find_lsb
:
1555 result
= nir_find_lsb(&b
, srcs
[0]);
1559 switch (ir
->type
->vector_elements
) {
1561 switch (ir
->operands
[0]->type
->vector_elements
) {
1562 case 1: result
= nir_fnoise1_1(&b
, srcs
[0]); break;
1563 case 2: result
= nir_fnoise1_2(&b
, srcs
[0]); break;
1564 case 3: result
= nir_fnoise1_3(&b
, srcs
[0]); break;
1565 case 4: result
= nir_fnoise1_4(&b
, srcs
[0]); break;
1566 default: unreachable("not reached");
1570 switch (ir
->operands
[0]->type
->vector_elements
) {
1571 case 1: result
= nir_fnoise2_1(&b
, srcs
[0]); break;
1572 case 2: result
= nir_fnoise2_2(&b
, srcs
[0]); break;
1573 case 3: result
= nir_fnoise2_3(&b
, srcs
[0]); break;
1574 case 4: result
= nir_fnoise2_4(&b
, srcs
[0]); break;
1575 default: unreachable("not reached");
1579 switch (ir
->operands
[0]->type
->vector_elements
) {
1580 case 1: result
= nir_fnoise3_1(&b
, srcs
[0]); break;
1581 case 2: result
= nir_fnoise3_2(&b
, srcs
[0]); break;
1582 case 3: result
= nir_fnoise3_3(&b
, srcs
[0]); break;
1583 case 4: result
= nir_fnoise3_4(&b
, srcs
[0]); break;
1584 default: unreachable("not reached");
1588 switch (ir
->operands
[0]->type
->vector_elements
) {
1589 case 1: result
= nir_fnoise4_1(&b
, srcs
[0]); break;
1590 case 2: result
= nir_fnoise4_2(&b
, srcs
[0]); break;
1591 case 3: result
= nir_fnoise4_3(&b
, srcs
[0]); break;
1592 case 4: result
= nir_fnoise4_4(&b
, srcs
[0]); break;
1593 default: unreachable("not reached");
1597 unreachable("not reached");
1600 case ir_unop_get_buffer_size
: {
1601 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1603 nir_intrinsic_get_buffer_size
);
1604 load
->num_components
= ir
->type
->vector_elements
;
1605 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1606 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1607 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1612 result
= type_is_float(out_type
) ? nir_fadd(&b
, srcs
[0], srcs
[1])
1613 : nir_iadd(&b
, srcs
[0], srcs
[1]);
1616 result
= type_is_float(out_type
) ? nir_fsub(&b
, srcs
[0], srcs
[1])
1617 : nir_isub(&b
, srcs
[0], srcs
[1]);
1620 result
= type_is_float(out_type
) ? nir_fmul(&b
, srcs
[0], srcs
[1])
1621 : nir_imul(&b
, srcs
[0], srcs
[1]);
1624 if (type_is_float(out_type
))
1625 result
= nir_fdiv(&b
, srcs
[0], srcs
[1]);
1626 else if (out_type
== GLSL_TYPE_INT
)
1627 result
= nir_idiv(&b
, srcs
[0], srcs
[1]);
1629 result
= nir_udiv(&b
, srcs
[0], srcs
[1]);
1632 result
= type_is_float(out_type
) ? nir_fmod(&b
, srcs
[0], srcs
[1])
1633 : nir_umod(&b
, srcs
[0], srcs
[1]);
1636 if (type_is_float(out_type
))
1637 result
= nir_fmin(&b
, srcs
[0], srcs
[1]);
1638 else if (out_type
== GLSL_TYPE_INT
)
1639 result
= nir_imin(&b
, srcs
[0], srcs
[1]);
1641 result
= nir_umin(&b
, srcs
[0], srcs
[1]);
1644 if (type_is_float(out_type
))
1645 result
= nir_fmax(&b
, srcs
[0], srcs
[1]);
1646 else if (out_type
== GLSL_TYPE_INT
)
1647 result
= nir_imax(&b
, srcs
[0], srcs
[1]);
1649 result
= nir_umax(&b
, srcs
[0], srcs
[1]);
1651 case ir_binop_pow
: result
= nir_fpow(&b
, srcs
[0], srcs
[1]); break;
1652 case ir_binop_bit_and
: result
= nir_iand(&b
, srcs
[0], srcs
[1]); break;
1653 case ir_binop_bit_or
: result
= nir_ior(&b
, srcs
[0], srcs
[1]); break;
1654 case ir_binop_bit_xor
: result
= nir_ixor(&b
, srcs
[0], srcs
[1]); break;
1655 case ir_binop_logic_and
:
1656 result
= supports_ints
? nir_iand(&b
, srcs
[0], srcs
[1])
1657 : nir_fand(&b
, srcs
[0], srcs
[1]);
1659 case ir_binop_logic_or
:
1660 result
= supports_ints
? nir_ior(&b
, srcs
[0], srcs
[1])
1661 : nir_for(&b
, srcs
[0], srcs
[1]);
1663 case ir_binop_logic_xor
:
1664 result
= supports_ints
? nir_ixor(&b
, srcs
[0], srcs
[1])
1665 : nir_fxor(&b
, srcs
[0], srcs
[1]);
1667 case ir_binop_lshift
: result
= nir_ishl(&b
, srcs
[0], srcs
[1]); break;
1668 case ir_binop_rshift
:
1669 result
= (out_type
== GLSL_TYPE_INT
) ? nir_ishr(&b
, srcs
[0], srcs
[1])
1670 : nir_ushr(&b
, srcs
[0], srcs
[1]);
1672 case ir_binop_imul_high
:
1673 result
= (out_type
== GLSL_TYPE_INT
) ? nir_imul_high(&b
, srcs
[0], srcs
[1])
1674 : nir_umul_high(&b
, srcs
[0], srcs
[1]);
1676 case ir_binop_carry
: result
= nir_uadd_carry(&b
, srcs
[0], srcs
[1]); break;
1677 case ir_binop_borrow
: result
= nir_usub_borrow(&b
, srcs
[0], srcs
[1]); break;
1679 if (supports_ints
) {
1680 if (type_is_float(types
[0]))
1681 result
= nir_flt(&b
, srcs
[0], srcs
[1]);
1682 else if (types
[0] == GLSL_TYPE_INT
)
1683 result
= nir_ilt(&b
, srcs
[0], srcs
[1]);
1685 result
= nir_ult(&b
, srcs
[0], srcs
[1]);
1687 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1690 case ir_binop_greater
:
1691 if (supports_ints
) {
1692 if (type_is_float(types
[0]))
1693 result
= nir_flt(&b
, srcs
[1], srcs
[0]);
1694 else if (types
[0] == GLSL_TYPE_INT
)
1695 result
= nir_ilt(&b
, srcs
[1], srcs
[0]);
1697 result
= nir_ult(&b
, srcs
[1], srcs
[0]);
1699 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1702 case ir_binop_lequal
:
1703 if (supports_ints
) {
1704 if (type_is_float(types
[0]))
1705 result
= nir_fge(&b
, srcs
[1], srcs
[0]);
1706 else if (types
[0] == GLSL_TYPE_INT
)
1707 result
= nir_ige(&b
, srcs
[1], srcs
[0]);
1709 result
= nir_uge(&b
, srcs
[1], srcs
[0]);
1711 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1714 case ir_binop_gequal
:
1715 if (supports_ints
) {
1716 if (type_is_float(types
[0]))
1717 result
= nir_fge(&b
, srcs
[0], srcs
[1]);
1718 else if (types
[0] == GLSL_TYPE_INT
)
1719 result
= nir_ige(&b
, srcs
[0], srcs
[1]);
1721 result
= nir_uge(&b
, srcs
[0], srcs
[1]);
1723 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1726 case ir_binop_equal
:
1727 if (supports_ints
) {
1728 if (type_is_float(types
[0]))
1729 result
= nir_feq(&b
, srcs
[0], srcs
[1]);
1731 result
= nir_ieq(&b
, srcs
[0], srcs
[1]);
1733 result
= nir_seq(&b
, srcs
[0], srcs
[1]);
1736 case ir_binop_nequal
:
1737 if (supports_ints
) {
1738 if (type_is_float(types
[0]))
1739 result
= nir_fne(&b
, srcs
[0], srcs
[1]);
1741 result
= nir_ine(&b
, srcs
[0], srcs
[1]);
1743 result
= nir_sne(&b
, srcs
[0], srcs
[1]);
1746 case ir_binop_all_equal
:
1747 if (supports_ints
) {
1748 if (type_is_float(types
[0])) {
1749 switch (ir
->operands
[0]->type
->vector_elements
) {
1750 case 1: result
= nir_feq(&b
, srcs
[0], srcs
[1]); break;
1751 case 2: result
= nir_ball_fequal2(&b
, srcs
[0], srcs
[1]); break;
1752 case 3: result
= nir_ball_fequal3(&b
, srcs
[0], srcs
[1]); break;
1753 case 4: result
= nir_ball_fequal4(&b
, srcs
[0], srcs
[1]); break;
1755 unreachable("not reached");
1758 switch (ir
->operands
[0]->type
->vector_elements
) {
1759 case 1: result
= nir_ieq(&b
, srcs
[0], srcs
[1]); break;
1760 case 2: result
= nir_ball_iequal2(&b
, srcs
[0], srcs
[1]); break;
1761 case 3: result
= nir_ball_iequal3(&b
, srcs
[0], srcs
[1]); break;
1762 case 4: result
= nir_ball_iequal4(&b
, srcs
[0], srcs
[1]); break;
1764 unreachable("not reached");
1768 switch (ir
->operands
[0]->type
->vector_elements
) {
1769 case 1: result
= nir_seq(&b
, srcs
[0], srcs
[1]); break;
1770 case 2: result
= nir_fall_equal2(&b
, srcs
[0], srcs
[1]); break;
1771 case 3: result
= nir_fall_equal3(&b
, srcs
[0], srcs
[1]); break;
1772 case 4: result
= nir_fall_equal4(&b
, srcs
[0], srcs
[1]); break;
1774 unreachable("not reached");
1778 case ir_binop_any_nequal
:
1779 if (supports_ints
) {
1780 if (type_is_float(types
[0])) {
1781 switch (ir
->operands
[0]->type
->vector_elements
) {
1782 case 1: result
= nir_fne(&b
, srcs
[0], srcs
[1]); break;
1783 case 2: result
= nir_bany_fnequal2(&b
, srcs
[0], srcs
[1]); break;
1784 case 3: result
= nir_bany_fnequal3(&b
, srcs
[0], srcs
[1]); break;
1785 case 4: result
= nir_bany_fnequal4(&b
, srcs
[0], srcs
[1]); break;
1787 unreachable("not reached");
1790 switch (ir
->operands
[0]->type
->vector_elements
) {
1791 case 1: result
= nir_ine(&b
, srcs
[0], srcs
[1]); break;
1792 case 2: result
= nir_bany_inequal2(&b
, srcs
[0], srcs
[1]); break;
1793 case 3: result
= nir_bany_inequal3(&b
, srcs
[0], srcs
[1]); break;
1794 case 4: result
= nir_bany_inequal4(&b
, srcs
[0], srcs
[1]); break;
1796 unreachable("not reached");
1800 switch (ir
->operands
[0]->type
->vector_elements
) {
1801 case 1: result
= nir_sne(&b
, srcs
[0], srcs
[1]); break;
1802 case 2: result
= nir_fany_nequal2(&b
, srcs
[0], srcs
[1]); break;
1803 case 3: result
= nir_fany_nequal3(&b
, srcs
[0], srcs
[1]); break;
1804 case 4: result
= nir_fany_nequal4(&b
, srcs
[0], srcs
[1]); break;
1806 unreachable("not reached");
1811 switch (ir
->operands
[0]->type
->vector_elements
) {
1812 case 2: result
= nir_fdot2(&b
, srcs
[0], srcs
[1]); break;
1813 case 3: result
= nir_fdot3(&b
, srcs
[0], srcs
[1]); break;
1814 case 4: result
= nir_fdot4(&b
, srcs
[0], srcs
[1]); break;
1816 unreachable("not reached");
1820 case ir_binop_ldexp
: result
= nir_ldexp(&b
, srcs
[0], srcs
[1]); break;
1822 result
= nir_ffma(&b
, srcs
[0], srcs
[1], srcs
[2]);
1825 result
= nir_flrp(&b
, srcs
[0], srcs
[1], srcs
[2]);
1829 result
= nir_bcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1831 result
= nir_fcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1833 case ir_triop_bitfield_extract
:
1834 result
= (out_type
== GLSL_TYPE_INT
) ?
1835 nir_ibitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]) :
1836 nir_ubitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]);
1838 case ir_quadop_bitfield_insert
:
1839 result
= nir_bitfield_insert(&b
, srcs
[0], srcs
[1], srcs
[2], srcs
[3]);
1841 case ir_quadop_vector
:
1842 result
= nir_vec(&b
, srcs
, ir
->type
->vector_elements
);
1846 unreachable("not reached");
1851 nir_visitor::visit(ir_swizzle
*ir
)
1853 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1854 result
= nir_swizzle(&b
, evaluate_rvalue(ir
->val
), swizzle
,
1855 ir
->type
->vector_elements
, !supports_ints
);
1859 nir_visitor::visit(ir_texture
*ir
)
1866 num_srcs
= 1; /* coordinate */
1871 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1872 num_srcs
= 2; /* coordinate, bias/lod */
1876 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1882 if (ir
->lod_info
.lod
!= NULL
)
1883 num_srcs
= 2; /* coordinate, lod */
1885 num_srcs
= 1; /* coordinate */
1889 op
= nir_texop_txf_ms
;
1890 num_srcs
= 2; /* coordinate, sample_index */
1895 if (ir
->lod_info
.lod
!= NULL
)
1896 num_srcs
= 1; /* lod */
1903 num_srcs
= 1; /* coordinate */
1908 num_srcs
= 1; /* coordinate */
1911 case ir_query_levels
:
1912 op
= nir_texop_query_levels
;
1916 case ir_texture_samples
:
1917 op
= nir_texop_texture_samples
;
1921 case ir_samples_identical
:
1922 op
= nir_texop_samples_identical
;
1923 num_srcs
= 1; /* coordinate */
1927 unreachable("not reached");
1930 if (ir
->projector
!= NULL
)
1932 if (ir
->shadow_comparitor
!= NULL
)
1934 if (ir
->offset
!= NULL
)
1937 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1940 instr
->sampler_dim
=
1941 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1942 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1943 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1944 if (instr
->is_shadow
)
1945 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1946 switch (ir
->type
->base_type
) {
1947 case GLSL_TYPE_FLOAT
:
1948 instr
->dest_type
= nir_type_float
;
1951 instr
->dest_type
= nir_type_int
;
1953 case GLSL_TYPE_BOOL
:
1954 case GLSL_TYPE_UINT
:
1955 instr
->dest_type
= nir_type_uint
;
1958 unreachable("not reached");
1961 instr
->texture
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1963 unsigned src_number
= 0;
1965 if (ir
->coordinate
!= NULL
) {
1966 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1967 instr
->src
[src_number
].src
=
1968 nir_src_for_ssa(evaluate_rvalue(ir
->coordinate
));
1969 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1973 if (ir
->projector
!= NULL
) {
1974 instr
->src
[src_number
].src
=
1975 nir_src_for_ssa(evaluate_rvalue(ir
->projector
));
1976 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1980 if (ir
->shadow_comparitor
!= NULL
) {
1981 instr
->src
[src_number
].src
=
1982 nir_src_for_ssa(evaluate_rvalue(ir
->shadow_comparitor
));
1983 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1987 if (ir
->offset
!= NULL
) {
1988 /* we don't support multiple offsets yet */
1989 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1991 instr
->src
[src_number
].src
=
1992 nir_src_for_ssa(evaluate_rvalue(ir
->offset
));
1993 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1999 instr
->src
[src_number
].src
=
2000 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.bias
));
2001 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
2008 if (ir
->lod_info
.lod
!= NULL
) {
2009 instr
->src
[src_number
].src
=
2010 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.lod
));
2011 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
2017 instr
->src
[src_number
].src
=
2018 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdx
));
2019 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
2021 instr
->src
[src_number
].src
=
2022 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdy
));
2023 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
2028 instr
->src
[src_number
].src
=
2029 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.sample_index
));
2030 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
2035 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
2042 assert(src_number
== num_srcs
);
2044 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
2045 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
), bit_size
);
2049 nir_visitor::visit(ir_constant
*ir
)
2052 * We don't know if this variable is an array or struct that gets
2053 * dereferenced, so do the safe thing an make it a variable with a
2054 * constant initializer and return a dereference.
2058 nir_local_variable_create(this->impl
, ir
->type
, "const_temp");
2059 var
->data
.read_only
= true;
2060 var
->constant_initializer
= constant_copy(ir
, var
);
2062 this->deref_head
= nir_deref_var_create(this->shader
, var
);
2063 this->deref_tail
= &this->deref_head
->deref
;
2067 nir_visitor::visit(ir_dereference_variable
*ir
)
2069 struct hash_entry
*entry
=
2070 _mesa_hash_table_search(this->var_table
, ir
->var
);
2072 nir_variable
*var
= (nir_variable
*) entry
->data
;
2074 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
2075 this->deref_head
= deref
;
2076 this->deref_tail
= &deref
->deref
;
2080 nir_visitor::visit(ir_dereference_record
*ir
)
2082 ir
->record
->accept(this);
2084 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
2085 assert(field_index
>= 0);
2087 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
2088 deref
->deref
.type
= ir
->type
;
2089 this->deref_tail
->child
= &deref
->deref
;
2090 this->deref_tail
= &deref
->deref
;
2094 nir_visitor::visit(ir_dereference_array
*ir
)
2096 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
2097 deref
->deref
.type
= ir
->type
;
2099 ir_constant
*const_index
= ir
->array_index
->as_constant();
2100 if (const_index
!= NULL
) {
2101 deref
->deref_array_type
= nir_deref_array_type_direct
;
2102 deref
->base_offset
= const_index
->value
.u
[0];
2104 deref
->deref_array_type
= nir_deref_array_type_indirect
;
2106 nir_src_for_ssa(evaluate_rvalue(ir
->array_index
));
2109 ir
->array
->accept(this);
2111 this->deref_tail
->child
= &deref
->deref
;
2112 ralloc_steal(this->deref_tail
, deref
);
2113 this->deref_tail
= &deref
->deref
;
2117 nir_visitor::visit(ir_barrier
*)
2119 nir_intrinsic_instr
*instr
=
2120 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
2121 nir_builder_instr_insert(&b
, &instr
->instr
);