2 * Copyright © 2014 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Connor Abbott (cwabbott0@gmail.com)
28 #include "glsl_to_nir.h"
29 #include "nir_control_flow.h"
30 #include "nir_builder.h"
31 #include "compiler/glsl/ir_visitor.h"
32 #include "compiler/glsl/ir_hierarchical_visitor.h"
33 #include "compiler/glsl/ir.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_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 nir_lower_outputs_to_temporaries(shader
, nir_shader_get_entrypoint(shader
));
148 shader
->info
.name
= ralloc_asprintf(shader
, "GLSL%d", shader_prog
->Name
);
149 if (shader_prog
->Label
)
150 shader
->info
.label
= ralloc_strdup(shader
, shader_prog
->Label
);
151 shader
->info
.num_textures
= _mesa_fls(sh
->Program
->SamplersUsed
);
152 shader
->info
.num_ubos
= sh
->NumUniformBlocks
;
153 shader
->info
.num_abos
= shader_prog
->NumAtomicBuffers
;
154 shader
->info
.num_ssbos
= sh
->NumShaderStorageBlocks
;
155 shader
->info
.num_images
= sh
->NumImages
;
156 shader
->info
.inputs_read
= sh
->Program
->InputsRead
;
157 shader
->info
.outputs_written
= sh
->Program
->OutputsWritten
;
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
= shader_prog
->TessCtrl
.VerticesOut
;
173 case MESA_SHADER_GEOMETRY
:
174 shader
->info
.gs
.vertices_in
= shader_prog
->Geom
.VerticesIn
;
175 shader
->info
.gs
.output_primitive
= sh
->Geom
.OutputType
;
176 shader
->info
.gs
.vertices_out
= sh
->Geom
.VerticesOut
;
177 shader
->info
.gs
.invocations
= sh
->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
.early_fragment_tests
= sh
->EarlyFragmentTests
;
188 shader
->info
.fs
.depth_layout
= fp
->FragDepthLayout
;
192 case MESA_SHADER_COMPUTE
: {
193 struct gl_compute_program
*cp
= (struct gl_compute_program
*)sh
->Program
;
194 shader
->info
.cs
.local_size
[0] = cp
->LocalSize
[0];
195 shader
->info
.cs
.local_size
[1] = cp
->LocalSize
[1];
196 shader
->info
.cs
.local_size
[2] = cp
->LocalSize
[2];
201 break; /* No stage-specific info */
207 nir_visitor::nir_visitor(nir_shader
*shader
)
209 this->supports_ints
= shader
->options
->native_integers
;
210 this->shader
= shader
;
211 this->is_global
= true;
212 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
213 _mesa_key_pointer_equal
);
214 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
215 _mesa_key_pointer_equal
);
218 nir_visitor::~nir_visitor()
220 _mesa_hash_table_destroy(this->var_table
, NULL
);
221 _mesa_hash_table_destroy(this->overload_table
, NULL
);
225 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
228 ralloc_steal(mem_ctx
, this->deref_head
);
229 return this->deref_head
;
232 static nir_constant
*
233 constant_copy(ir_constant
*ir
, void *mem_ctx
)
238 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
240 unsigned total_elems
= ir
->type
->components();
243 ret
->num_elements
= 0;
244 switch (ir
->type
->base_type
) {
246 for (i
= 0; i
< total_elems
; i
++)
247 ret
->value
.u
[i
] = ir
->value
.u
[i
];
251 for (i
= 0; i
< total_elems
; i
++)
252 ret
->value
.i
[i
] = ir
->value
.i
[i
];
255 case GLSL_TYPE_FLOAT
:
256 for (i
= 0; i
< total_elems
; i
++)
257 ret
->value
.f
[i
] = ir
->value
.f
[i
];
260 case GLSL_TYPE_DOUBLE
:
261 for (i
= 0; i
< total_elems
; i
++)
262 ret
->value
.d
[i
] = ir
->value
.d
[i
];
266 for (i
= 0; i
< total_elems
; i
++)
267 ret
->value
.b
[i
] = ir
->value
.b
[i
];
270 case GLSL_TYPE_STRUCT
:
271 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
273 ret
->num_elements
= ir
->type
->length
;
276 foreach_in_list(ir_constant
, field
, &ir
->components
) {
277 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
282 case GLSL_TYPE_ARRAY
:
283 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
285 ret
->num_elements
= ir
->type
->length
;
287 for (i
= 0; i
< ir
->type
->length
; i
++)
288 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
292 unreachable("not reached");
299 nir_visitor::visit(ir_variable
*ir
)
301 nir_variable
*var
= ralloc(shader
, nir_variable
);
302 var
->type
= ir
->type
;
303 var
->name
= ralloc_strdup(var
, ir
->name
);
305 var
->data
.read_only
= ir
->data
.read_only
;
306 var
->data
.centroid
= ir
->data
.centroid
;
307 var
->data
.sample
= ir
->data
.sample
;
308 var
->data
.patch
= ir
->data
.patch
;
309 var
->data
.invariant
= ir
->data
.invariant
;
310 var
->data
.location
= ir
->data
.location
;
312 switch(ir
->data
.mode
) {
314 case ir_var_temporary
:
316 var
->data
.mode
= nir_var_global
;
318 var
->data
.mode
= nir_var_local
;
321 case ir_var_function_in
:
322 case ir_var_function_out
:
323 case ir_var_function_inout
:
324 case ir_var_const_in
:
325 var
->data
.mode
= nir_var_local
;
328 case ir_var_shader_in
:
329 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
330 ir
->data
.location
== VARYING_SLOT_FACE
) {
331 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
332 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
333 var
->data
.mode
= nir_var_system_value
;
334 } else if (shader
->stage
== MESA_SHADER_GEOMETRY
&&
335 ir
->data
.location
== VARYING_SLOT_PRIMITIVE_ID
) {
336 /* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
337 var
->data
.location
= SYSTEM_VALUE_PRIMITIVE_ID
;
338 var
->data
.mode
= nir_var_system_value
;
340 var
->data
.mode
= nir_var_shader_in
;
344 case ir_var_shader_out
:
345 var
->data
.mode
= nir_var_shader_out
;
349 var
->data
.mode
= nir_var_uniform
;
352 case ir_var_shader_storage
:
353 var
->data
.mode
= nir_var_shader_storage
;
356 case ir_var_system_value
:
357 var
->data
.mode
= nir_var_system_value
;
361 unreachable("not reached");
364 var
->data
.interpolation
= ir
->data
.interpolation
;
365 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
366 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
367 var
->data
.explicit_location
= ir
->data
.explicit_location
;
368 var
->data
.explicit_index
= ir
->data
.explicit_index
;
369 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
370 var
->data
.has_initializer
= ir
->data
.has_initializer
;
371 var
->data
.location_frac
= ir
->data
.location_frac
;
373 switch (ir
->data
.depth_layout
) {
374 case ir_depth_layout_none
:
375 var
->data
.depth_layout
= nir_depth_layout_none
;
377 case ir_depth_layout_any
:
378 var
->data
.depth_layout
= nir_depth_layout_any
;
380 case ir_depth_layout_greater
:
381 var
->data
.depth_layout
= nir_depth_layout_greater
;
383 case ir_depth_layout_less
:
384 var
->data
.depth_layout
= nir_depth_layout_less
;
386 case ir_depth_layout_unchanged
:
387 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
390 unreachable("not reached");
393 var
->data
.index
= ir
->data
.index
;
394 var
->data
.descriptor_set
= 0;
395 var
->data
.binding
= ir
->data
.binding
;
396 var
->data
.offset
= ir
->data
.offset
;
397 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
398 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
399 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
400 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
401 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
402 var
->data
.image
.format
= ir
->data
.image_format
;
403 var
->data
.max_array_access
= ir
->data
.max_array_access
;
405 var
->num_state_slots
= ir
->get_num_state_slots();
406 if (var
->num_state_slots
> 0) {
407 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
408 var
->num_state_slots
);
410 ir_state_slot
*state_slots
= ir
->get_state_slots();
411 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
412 for (unsigned j
= 0; j
< 5; j
++)
413 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
414 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
417 var
->state_slots
= NULL
;
420 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
422 var
->interface_type
= ir
->get_interface_type();
424 if (var
->data
.mode
== nir_var_local
)
425 nir_function_impl_add_variable(impl
, var
);
427 nir_shader_add_variable(shader
, var
);
429 _mesa_hash_table_insert(var_table
, ir
, var
);
434 nir_function_visitor::visit_enter(ir_function
*ir
)
436 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
437 visitor
->create_function(sig
);
439 return visit_continue_with_parent
;
443 nir_visitor::create_function(ir_function_signature
*ir
)
445 if (ir
->is_intrinsic
)
448 nir_function
*func
= nir_function_create(shader
, ir
->function_name());
450 assert(ir
->parameters
.is_empty());
451 assert(ir
->return_type
== glsl_type::void_type
);
453 _mesa_hash_table_insert(this->overload_table
, ir
, func
);
457 nir_visitor::visit(ir_function
*ir
)
459 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
464 nir_visitor::visit(ir_function_signature
*ir
)
466 if (ir
->is_intrinsic
)
469 struct hash_entry
*entry
=
470 _mesa_hash_table_search(this->overload_table
, ir
);
473 nir_function
*func
= (nir_function
*) entry
->data
;
475 if (ir
->is_defined
) {
476 nir_function_impl
*impl
= nir_function_impl_create(func
);
479 assert(strcmp(func
->name
, "main") == 0);
480 assert(ir
->parameters
.is_empty());
481 assert(func
->return_type
== glsl_type::void_type
);
483 this->is_global
= false;
485 nir_builder_init(&b
, impl
);
486 b
.cursor
= nir_after_cf_list(&impl
->body
);
487 visit_exec_list(&ir
->body
, this);
489 this->is_global
= true;
496 nir_visitor::visit(ir_loop
*ir
)
498 nir_loop
*loop
= nir_loop_create(this->shader
);
499 nir_builder_cf_insert(&b
, &loop
->cf_node
);
501 b
.cursor
= nir_after_cf_list(&loop
->body
);
502 visit_exec_list(&ir
->body_instructions
, this);
503 b
.cursor
= nir_after_cf_node(&loop
->cf_node
);
507 nir_visitor::visit(ir_if
*ir
)
510 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
512 nir_if
*if_stmt
= nir_if_create(this->shader
);
513 if_stmt
->condition
= condition
;
514 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
516 b
.cursor
= nir_after_cf_list(&if_stmt
->then_list
);
517 visit_exec_list(&ir
->then_instructions
, this);
519 b
.cursor
= nir_after_cf_list(&if_stmt
->else_list
);
520 visit_exec_list(&ir
->else_instructions
, this);
522 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
526 nir_visitor::visit(ir_discard
*ir
)
529 * discards aren't treated as control flow, because before we lower them
530 * they can appear anywhere in the shader and the stuff after them may still
531 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
532 * discards will be immediately followed by a return.
535 nir_intrinsic_instr
*discard
;
537 discard
= nir_intrinsic_instr_create(this->shader
,
538 nir_intrinsic_discard_if
);
540 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
542 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
545 nir_builder_instr_insert(&b
, &discard
->instr
);
549 nir_visitor::visit(ir_emit_vertex
*ir
)
551 nir_intrinsic_instr
*instr
=
552 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
553 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
554 nir_builder_instr_insert(&b
, &instr
->instr
);
558 nir_visitor::visit(ir_end_primitive
*ir
)
560 nir_intrinsic_instr
*instr
=
561 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
562 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
563 nir_builder_instr_insert(&b
, &instr
->instr
);
567 nir_visitor::visit(ir_loop_jump
*ir
)
571 case ir_loop_jump::jump_break
:
572 type
= nir_jump_break
;
574 case ir_loop_jump::jump_continue
:
575 type
= nir_jump_continue
;
578 unreachable("not reached");
581 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
582 nir_builder_instr_insert(&b
, &instr
->instr
);
586 nir_visitor::visit(ir_return
*ir
)
588 if (ir
->value
!= NULL
) {
589 nir_intrinsic_instr
*copy
=
590 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
592 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
593 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
596 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
597 nir_builder_instr_insert(&b
, &instr
->instr
);
601 nir_visitor::visit(ir_call
*ir
)
603 if (ir
->callee
->is_intrinsic
) {
605 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
606 op
= nir_intrinsic_atomic_counter_read_var
;
607 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
608 op
= nir_intrinsic_atomic_counter_inc_var
;
609 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
610 op
= nir_intrinsic_atomic_counter_dec_var
;
611 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
612 op
= nir_intrinsic_image_load
;
613 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
614 op
= nir_intrinsic_image_store
;
615 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
616 op
= nir_intrinsic_image_atomic_add
;
617 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
618 op
= nir_intrinsic_image_atomic_min
;
619 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
620 op
= nir_intrinsic_image_atomic_max
;
621 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
622 op
= nir_intrinsic_image_atomic_and
;
623 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
624 op
= nir_intrinsic_image_atomic_or
;
625 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
626 op
= nir_intrinsic_image_atomic_xor
;
627 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
628 op
= nir_intrinsic_image_atomic_exchange
;
629 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
630 op
= nir_intrinsic_image_atomic_comp_swap
;
631 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
632 op
= nir_intrinsic_memory_barrier
;
633 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_size") == 0) {
634 op
= nir_intrinsic_image_size
;
635 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_samples") == 0) {
636 op
= nir_intrinsic_image_samples
;
637 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_ssbo") == 0) {
638 op
= nir_intrinsic_store_ssbo
;
639 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_ssbo") == 0) {
640 op
= nir_intrinsic_load_ssbo
;
641 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_add_ssbo") == 0) {
642 op
= nir_intrinsic_ssbo_atomic_add
;
643 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_and_ssbo") == 0) {
644 op
= nir_intrinsic_ssbo_atomic_and
;
645 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_or_ssbo") == 0) {
646 op
= nir_intrinsic_ssbo_atomic_or
;
647 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_xor_ssbo") == 0) {
648 op
= nir_intrinsic_ssbo_atomic_xor
;
649 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_min_ssbo") == 0) {
650 assert(ir
->return_deref
);
651 if (ir
->return_deref
->type
== glsl_type::int_type
)
652 op
= nir_intrinsic_ssbo_atomic_imin
;
653 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
654 op
= nir_intrinsic_ssbo_atomic_umin
;
656 unreachable("Invalid type");
657 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_max_ssbo") == 0) {
658 assert(ir
->return_deref
);
659 if (ir
->return_deref
->type
== glsl_type::int_type
)
660 op
= nir_intrinsic_ssbo_atomic_imax
;
661 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
662 op
= nir_intrinsic_ssbo_atomic_umax
;
664 unreachable("Invalid type");
665 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_exchange_ssbo") == 0) {
666 op
= nir_intrinsic_ssbo_atomic_exchange
;
667 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_comp_swap_ssbo") == 0) {
668 op
= nir_intrinsic_ssbo_atomic_comp_swap
;
669 } else if (strcmp(ir
->callee_name(), "__intrinsic_shader_clock") == 0) {
670 op
= nir_intrinsic_shader_clock
;
671 } else if (strcmp(ir
->callee_name(), "__intrinsic_group_memory_barrier") == 0) {
672 op
= nir_intrinsic_group_memory_barrier
;
673 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_atomic_counter") == 0) {
674 op
= nir_intrinsic_memory_barrier_atomic_counter
;
675 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_buffer") == 0) {
676 op
= nir_intrinsic_memory_barrier_buffer
;
677 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_image") == 0) {
678 op
= nir_intrinsic_memory_barrier_image
;
679 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_shared") == 0) {
680 op
= nir_intrinsic_memory_barrier_shared
;
681 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_shared") == 0) {
682 op
= nir_intrinsic_load_shared
;
683 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_shared") == 0) {
684 op
= nir_intrinsic_store_shared
;
685 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_add_shared") == 0) {
686 op
= nir_intrinsic_shared_atomic_add
;
687 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_and_shared") == 0) {
688 op
= nir_intrinsic_shared_atomic_and
;
689 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_or_shared") == 0) {
690 op
= nir_intrinsic_shared_atomic_or
;
691 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_xor_shared") == 0) {
692 op
= nir_intrinsic_shared_atomic_xor
;
693 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_min_shared") == 0) {
694 assert(ir
->return_deref
);
695 if (ir
->return_deref
->type
== glsl_type::int_type
)
696 op
= nir_intrinsic_shared_atomic_imin
;
697 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
698 op
= nir_intrinsic_shared_atomic_umin
;
700 unreachable("Invalid type");
701 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_max_shared") == 0) {
702 assert(ir
->return_deref
);
703 if (ir
->return_deref
->type
== glsl_type::int_type
)
704 op
= nir_intrinsic_shared_atomic_imax
;
705 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
706 op
= nir_intrinsic_shared_atomic_umax
;
708 unreachable("Invalid type");
709 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_exchange_shared") == 0) {
710 op
= nir_intrinsic_shared_atomic_exchange
;
711 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_comp_swap_shared") == 0) {
712 op
= nir_intrinsic_shared_atomic_comp_swap
;
714 unreachable("not reached");
717 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
718 nir_dest
*dest
= &instr
->dest
;
721 case nir_intrinsic_atomic_counter_read_var
:
722 case nir_intrinsic_atomic_counter_inc_var
:
723 case nir_intrinsic_atomic_counter_dec_var
: {
724 ir_dereference
*param
=
725 (ir_dereference
*) ir
->actual_parameters
.get_head();
726 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
727 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
728 nir_builder_instr_insert(&b
, &instr
->instr
);
731 case nir_intrinsic_image_load
:
732 case nir_intrinsic_image_store
:
733 case nir_intrinsic_image_atomic_add
:
734 case nir_intrinsic_image_atomic_min
:
735 case nir_intrinsic_image_atomic_max
:
736 case nir_intrinsic_image_atomic_and
:
737 case nir_intrinsic_image_atomic_or
:
738 case nir_intrinsic_image_atomic_xor
:
739 case nir_intrinsic_image_atomic_exchange
:
740 case nir_intrinsic_image_atomic_comp_swap
:
741 case nir_intrinsic_image_samples
:
742 case nir_intrinsic_image_size
: {
743 nir_ssa_undef_instr
*instr_undef
=
744 nir_ssa_undef_instr_create(shader
, 1, 32);
745 nir_builder_instr_insert(&b
, &instr_undef
->instr
);
747 /* Set the image variable dereference. */
748 exec_node
*param
= ir
->actual_parameters
.get_head();
749 ir_dereference
*image
= (ir_dereference
*)param
;
750 const glsl_type
*type
=
751 image
->variable_referenced()->type
->without_array();
753 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
754 param
= param
->get_next();
756 /* Set the intrinsic destination. */
757 if (ir
->return_deref
) {
758 const nir_intrinsic_info
*info
=
759 &nir_intrinsic_infos
[instr
->intrinsic
];
760 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
761 info
->dest_components
, 32, NULL
);
764 if (op
== nir_intrinsic_image_size
||
765 op
== nir_intrinsic_image_samples
) {
766 nir_builder_instr_insert(&b
, &instr
->instr
);
770 /* Set the address argument, extending the coordinate vector to four
773 nir_ssa_def
*src_addr
=
774 evaluate_rvalue((ir_dereference
*)param
);
775 nir_ssa_def
*srcs
[4];
777 for (int i
= 0; i
< 4; i
++) {
778 if (i
< type
->coordinate_components())
779 srcs
[i
] = nir_channel(&b
, src_addr
, i
);
781 srcs
[i
] = &instr_undef
->def
;
784 instr
->src
[0] = nir_src_for_ssa(nir_vec(&b
, srcs
, 4));
785 param
= param
->get_next();
787 /* Set the sample argument, which is undefined for single-sample
790 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
792 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
793 param
= param
->get_next();
795 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
798 /* Set the intrinsic parameters. */
799 if (!param
->is_tail_sentinel()) {
801 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
802 param
= param
->get_next();
805 if (!param
->is_tail_sentinel()) {
807 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
808 param
= param
->get_next();
810 nir_builder_instr_insert(&b
, &instr
->instr
);
813 case nir_intrinsic_memory_barrier
:
814 case nir_intrinsic_group_memory_barrier
:
815 case nir_intrinsic_memory_barrier_atomic_counter
:
816 case nir_intrinsic_memory_barrier_buffer
:
817 case nir_intrinsic_memory_barrier_image
:
818 case nir_intrinsic_memory_barrier_shared
:
819 nir_builder_instr_insert(&b
, &instr
->instr
);
821 case nir_intrinsic_shader_clock
:
822 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
823 nir_builder_instr_insert(&b
, &instr
->instr
);
825 case nir_intrinsic_store_ssbo
: {
826 exec_node
*param
= ir
->actual_parameters
.get_head();
827 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
829 param
= param
->get_next();
830 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
832 param
= param
->get_next();
833 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
835 param
= param
->get_next();
836 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
839 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
840 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(block
));
841 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(offset
));
842 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
843 instr
->num_components
= val
->type
->vector_elements
;
845 nir_builder_instr_insert(&b
, &instr
->instr
);
848 case nir_intrinsic_load_ssbo
: {
849 exec_node
*param
= ir
->actual_parameters
.get_head();
850 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
852 param
= param
->get_next();
853 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
855 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(block
));
856 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
858 const glsl_type
*type
= ir
->return_deref
->var
->type
;
859 instr
->num_components
= type
->vector_elements
;
861 /* Setup destination register */
862 unsigned bit_size
= glsl_get_bit_size(type
->base_type
);
863 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
864 type
->vector_elements
, bit_size
, NULL
);
866 /* Insert the created nir instruction now since in the case of boolean
867 * result we will need to emit another instruction after it
869 nir_builder_instr_insert(&b
, &instr
->instr
);
872 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
873 * consider a true boolean to be ~0. Fix this up with a != 0
876 if (type
->base_type
== GLSL_TYPE_BOOL
) {
877 nir_alu_instr
*load_ssbo_compare
=
878 nir_alu_instr_create(shader
, nir_op_ine
);
879 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
880 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
881 load_ssbo_compare
->src
[1].src
=
882 nir_src_for_ssa(nir_imm_int(&b
, 0));
883 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
884 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
885 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
886 &load_ssbo_compare
->dest
.dest
,
887 type
->vector_elements
, bit_size
, NULL
);
888 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
889 nir_builder_instr_insert(&b
, &load_ssbo_compare
->instr
);
890 dest
= &load_ssbo_compare
->dest
.dest
;
894 case nir_intrinsic_ssbo_atomic_add
:
895 case nir_intrinsic_ssbo_atomic_imin
:
896 case nir_intrinsic_ssbo_atomic_umin
:
897 case nir_intrinsic_ssbo_atomic_imax
:
898 case nir_intrinsic_ssbo_atomic_umax
:
899 case nir_intrinsic_ssbo_atomic_and
:
900 case nir_intrinsic_ssbo_atomic_or
:
901 case nir_intrinsic_ssbo_atomic_xor
:
902 case nir_intrinsic_ssbo_atomic_exchange
:
903 case nir_intrinsic_ssbo_atomic_comp_swap
: {
904 int param_count
= ir
->actual_parameters
.length();
905 assert(param_count
== 3 || param_count
== 4);
908 exec_node
*param
= ir
->actual_parameters
.get_head();
909 ir_instruction
*inst
= (ir_instruction
*) param
;
910 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
913 param
= param
->get_next();
914 inst
= (ir_instruction
*) param
;
915 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
917 /* data1 parameter (this is always present) */
918 param
= param
->get_next();
919 inst
= (ir_instruction
*) param
;
920 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
922 /* data2 parameter (only with atomic_comp_swap) */
923 if (param_count
== 4) {
924 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
925 param
= param
->get_next();
926 inst
= (ir_instruction
*) param
;
927 instr
->src
[3] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
931 assert(ir
->return_deref
);
932 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
933 ir
->return_deref
->type
->vector_elements
, 32, NULL
);
934 nir_builder_instr_insert(&b
, &instr
->instr
);
937 case nir_intrinsic_load_shared
: {
938 exec_node
*param
= ir
->actual_parameters
.get_head();
939 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
941 nir_intrinsic_set_base(instr
, 0);
942 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(offset
));
944 const glsl_type
*type
= ir
->return_deref
->var
->type
;
945 instr
->num_components
= type
->vector_elements
;
947 /* Setup destination register */
948 unsigned bit_size
= glsl_get_bit_size(type
->base_type
);
949 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
950 type
->vector_elements
, bit_size
, NULL
);
952 nir_builder_instr_insert(&b
, &instr
->instr
);
955 case nir_intrinsic_store_shared
: {
956 exec_node
*param
= ir
->actual_parameters
.get_head();
957 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
959 param
= param
->get_next();
960 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
962 param
= param
->get_next();
963 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
966 nir_intrinsic_set_base(instr
, 0);
967 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
969 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
971 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
972 instr
->num_components
= val
->type
->vector_elements
;
974 nir_builder_instr_insert(&b
, &instr
->instr
);
977 case nir_intrinsic_shared_atomic_add
:
978 case nir_intrinsic_shared_atomic_imin
:
979 case nir_intrinsic_shared_atomic_umin
:
980 case nir_intrinsic_shared_atomic_imax
:
981 case nir_intrinsic_shared_atomic_umax
:
982 case nir_intrinsic_shared_atomic_and
:
983 case nir_intrinsic_shared_atomic_or
:
984 case nir_intrinsic_shared_atomic_xor
:
985 case nir_intrinsic_shared_atomic_exchange
:
986 case nir_intrinsic_shared_atomic_comp_swap
: {
987 int param_count
= ir
->actual_parameters
.length();
988 assert(param_count
== 2 || param_count
== 3);
991 exec_node
*param
= ir
->actual_parameters
.get_head();
992 ir_instruction
*inst
= (ir_instruction
*) param
;
993 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
995 /* data1 parameter (this is always present) */
996 param
= param
->get_next();
997 inst
= (ir_instruction
*) param
;
998 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1000 /* data2 parameter (only with atomic_comp_swap) */
1001 if (param_count
== 3) {
1002 assert(op
== nir_intrinsic_shared_atomic_comp_swap
);
1003 param
= param
->get_next();
1004 inst
= (ir_instruction
*) param
;
1006 nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1010 assert(ir
->return_deref
);
1011 unsigned bit_size
= glsl_get_bit_size(ir
->return_deref
->type
->base_type
);
1012 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1013 ir
->return_deref
->type
->vector_elements
,
1015 nir_builder_instr_insert(&b
, &instr
->instr
);
1019 unreachable("not reached");
1022 if (ir
->return_deref
) {
1023 nir_intrinsic_instr
*store_instr
=
1024 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
1025 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
1026 nir_intrinsic_set_write_mask(store_instr
,
1027 (1 << store_instr
->num_components
) - 1);
1029 store_instr
->variables
[0] =
1030 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
1031 store_instr
->src
[0] = nir_src_for_ssa(&dest
->ssa
);
1033 nir_builder_instr_insert(&b
, &store_instr
->instr
);
1039 struct hash_entry
*entry
=
1040 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
1042 nir_function
*callee
= (nir_function
*) entry
->data
;
1044 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
1047 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
1048 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
1052 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
1053 nir_builder_instr_insert(&b
, &instr
->instr
);
1057 nir_visitor::visit(ir_assignment
*ir
)
1059 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
1061 b
.exact
= ir
->lhs
->variable_referenced()->data
.invariant
||
1062 ir
->lhs
->variable_referenced()->data
.precise
;
1064 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
1065 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
1066 /* We're doing a plain-as-can-be copy, so emit a copy_var */
1067 nir_intrinsic_instr
*copy
=
1068 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
1070 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
1071 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
1073 if (ir
->condition
) {
1074 nir_if
*if_stmt
= nir_if_create(this->shader
);
1075 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1076 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1077 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
1078 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1080 nir_builder_instr_insert(&b
, ©
->instr
);
1085 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
1087 ir
->lhs
->accept(this);
1088 nir_deref_var
*lhs_deref
= this->deref_head
;
1089 nir_ssa_def
*src
= evaluate_rvalue(ir
->rhs
);
1091 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
1092 /* GLSL IR will give us the input to the write-masked assignment in a
1093 * single packed vector. So, for example, if the writemask is xzw, then
1094 * we have to swizzle x -> x, y -> z, and z -> w and get the y component
1098 unsigned component
= 0;
1099 for (unsigned i
= 0; i
< 4; i
++) {
1100 swiz
[i
] = ir
->write_mask
& (1 << i
) ? component
++ : 0;
1102 src
= nir_swizzle(&b
, src
, swiz
, num_components
, !supports_ints
);
1105 nir_intrinsic_instr
*store
=
1106 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1107 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1108 nir_intrinsic_set_write_mask(store
, ir
->write_mask
);
1109 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
1110 store
->variables
[0] = nir_deref_as_var(store_deref
);
1111 store
->src
[0] = nir_src_for_ssa(src
);
1113 if (ir
->condition
) {
1114 nir_if
*if_stmt
= nir_if_create(this->shader
);
1115 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1116 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1117 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
1118 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1120 nir_builder_instr_insert(&b
, &store
->instr
);
1125 * Given an instruction, returns a pointer to its destination or NULL if there
1126 * is no destination.
1128 * Note that this only handles instructions we generate at this level.
1131 get_instr_dest(nir_instr
*instr
)
1133 nir_alu_instr
*alu_instr
;
1134 nir_intrinsic_instr
*intrinsic_instr
;
1135 nir_tex_instr
*tex_instr
;
1137 switch (instr
->type
) {
1138 case nir_instr_type_alu
:
1139 alu_instr
= nir_instr_as_alu(instr
);
1140 return &alu_instr
->dest
.dest
;
1142 case nir_instr_type_intrinsic
:
1143 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1144 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1145 return &intrinsic_instr
->dest
;
1149 case nir_instr_type_tex
:
1150 tex_instr
= nir_instr_as_tex(instr
);
1151 return &tex_instr
->dest
;
1154 unreachable("not reached");
1161 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
,
1164 nir_dest
*dest
= get_instr_dest(instr
);
1167 nir_ssa_dest_init(instr
, dest
, num_components
, bit_size
, NULL
);
1169 nir_builder_instr_insert(&b
, instr
);
1172 assert(dest
->is_ssa
);
1173 this->result
= &dest
->ssa
;
1178 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1181 if (ir
->as_dereference() || ir
->as_constant()) {
1183 * A dereference is being used on the right hand side, which means we
1184 * must emit a variable load.
1187 nir_intrinsic_instr
*load_instr
=
1188 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1189 load_instr
->num_components
= ir
->type
->vector_elements
;
1190 load_instr
->variables
[0] = this->deref_head
;
1191 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1192 unsigned bit_size
= glsl_get_bit_size(ir
->type
->base_type
);
1193 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
, bit_size
);
1196 return this->result
;
1200 type_is_float(glsl_base_type type
)
1202 return type
== GLSL_TYPE_FLOAT
|| type
== GLSL_TYPE_DOUBLE
;
1206 nir_visitor::visit(ir_expression
*ir
)
1208 /* Some special cases */
1209 switch (ir
->operation
) {
1210 case ir_binop_ubo_load
: {
1211 nir_intrinsic_instr
*load
=
1212 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_ubo
);
1213 unsigned bit_size
= glsl_get_bit_size(ir
->type
->base_type
);
1214 load
->num_components
= ir
->type
->vector_elements
;
1215 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1216 load
->src
[1] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1217 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1220 * In UBO's, a true boolean value is any non-zero value, but we consider
1221 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1224 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
)
1225 this->result
= nir_ine(&b
, &load
->dest
.ssa
, nir_imm_int(&b
, 0));
1230 case ir_unop_interpolate_at_centroid
:
1231 case ir_binop_interpolate_at_offset
:
1232 case ir_binop_interpolate_at_sample
: {
1233 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1234 ir_swizzle
*swizzle
= NULL
;
1236 /* the api does not allow a swizzle here, but the varying packing code
1237 * may have pushed one into here.
1239 swizzle
= ir
->operands
[0]->as_swizzle();
1241 deref
= swizzle
->val
->as_dereference();
1245 deref
->accept(this);
1247 nir_intrinsic_op op
;
1248 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1249 switch (ir
->operation
) {
1250 case ir_unop_interpolate_at_centroid
:
1251 op
= nir_intrinsic_interp_var_at_centroid
;
1253 case ir_binop_interpolate_at_offset
:
1254 op
= nir_intrinsic_interp_var_at_offset
;
1256 case ir_binop_interpolate_at_sample
:
1257 op
= nir_intrinsic_interp_var_at_sample
;
1260 unreachable("Invalid interpolation intrinsic");
1263 /* This case can happen if the vertex shader does not write the
1264 * given varying. In this case, the linker will lower it to a
1265 * global variable. Since interpolating a variable makes no
1266 * sense, we'll just turn it into a load which will probably
1267 * eventually end up as an SSA definition.
1269 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1270 op
= nir_intrinsic_load_var
;
1273 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1274 intrin
->num_components
= deref
->type
->vector_elements
;
1275 intrin
->variables
[0] = this->deref_head
;
1276 ralloc_steal(intrin
, intrin
->variables
[0]);
1278 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1279 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1280 intrin
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1282 unsigned bit_size
= glsl_get_bit_size(deref
->type
->base_type
);
1283 add_instr(&intrin
->instr
, deref
->type
->vector_elements
, bit_size
);
1286 unsigned swiz
[4] = {
1287 swizzle
->mask
.x
, swizzle
->mask
.y
, swizzle
->mask
.z
, swizzle
->mask
.w
1290 result
= nir_swizzle(&b
, result
, swiz
,
1291 swizzle
->type
->vector_elements
, false);
1301 nir_ssa_def
*srcs
[4];
1302 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1303 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1305 glsl_base_type types
[4];
1306 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1308 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1310 types
[i
] = GLSL_TYPE_FLOAT
;
1312 glsl_base_type out_type
;
1314 out_type
= ir
->type
->base_type
;
1316 out_type
= GLSL_TYPE_FLOAT
;
1318 switch (ir
->operation
) {
1319 case ir_unop_bit_not
: result
= nir_inot(&b
, srcs
[0]); break;
1320 case ir_unop_logic_not
:
1321 result
= supports_ints
? nir_inot(&b
, srcs
[0]) : nir_fnot(&b
, srcs
[0]);
1324 result
= type_is_float(types
[0]) ? nir_fneg(&b
, srcs
[0])
1325 : nir_ineg(&b
, srcs
[0]);
1328 result
= type_is_float(types
[0]) ? nir_fabs(&b
, srcs
[0])
1329 : nir_iabs(&b
, srcs
[0]);
1331 case ir_unop_saturate
:
1332 assert(type_is_float(types
[0]));
1333 result
= nir_fsat(&b
, srcs
[0]);
1336 result
= type_is_float(types
[0]) ? nir_fsign(&b
, srcs
[0])
1337 : nir_isign(&b
, srcs
[0]);
1339 case ir_unop_rcp
: result
= nir_frcp(&b
, srcs
[0]); break;
1340 case ir_unop_rsq
: result
= nir_frsq(&b
, srcs
[0]); break;
1341 case ir_unop_sqrt
: result
= nir_fsqrt(&b
, srcs
[0]); break;
1342 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1343 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1344 case ir_unop_exp2
: result
= nir_fexp2(&b
, srcs
[0]); break;
1345 case ir_unop_log2
: result
= nir_flog2(&b
, srcs
[0]); break;
1347 result
= supports_ints
? nir_i2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1350 result
= supports_ints
? nir_u2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1353 result
= supports_ints
? nir_b2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1355 case ir_unop_f2i
: result
= nir_f2i(&b
, srcs
[0]); break;
1356 case ir_unop_f2u
: result
= nir_f2u(&b
, srcs
[0]); break;
1357 case ir_unop_f2b
: result
= nir_f2b(&b
, srcs
[0]); break;
1358 case ir_unop_i2b
: result
= nir_i2b(&b
, srcs
[0]); break;
1359 case ir_unop_b2i
: result
= nir_b2i(&b
, srcs
[0]); break;
1360 case ir_unop_d2f
: result
= nir_d2f(&b
, srcs
[0]); break;
1361 case ir_unop_f2d
: result
= nir_f2d(&b
, srcs
[0]); break;
1362 case ir_unop_d2i
: result
= nir_d2i(&b
, srcs
[0]); break;
1363 case ir_unop_d2u
: result
= nir_d2u(&b
, srcs
[0]); break;
1364 case ir_unop_d2b
: result
= nir_d2b(&b
, srcs
[0]); break;
1366 assert(supports_ints
);
1367 result
= nir_i2d(&b
, srcs
[0]);
1370 assert(supports_ints
);
1371 result
= nir_u2d(&b
, srcs
[0]);
1375 case ir_unop_bitcast_i2f
:
1376 case ir_unop_bitcast_f2i
:
1377 case ir_unop_bitcast_u2f
:
1378 case ir_unop_bitcast_f2u
:
1379 case ir_unop_subroutine_to_int
:
1381 result
= nir_imov(&b
, srcs
[0]);
1383 case ir_unop_trunc
: result
= nir_ftrunc(&b
, srcs
[0]); break;
1384 case ir_unop_ceil
: result
= nir_fceil(&b
, srcs
[0]); break;
1385 case ir_unop_floor
: result
= nir_ffloor(&b
, srcs
[0]); break;
1386 case ir_unop_fract
: result
= nir_ffract(&b
, srcs
[0]); break;
1387 case ir_unop_round_even
: result
= nir_fround_even(&b
, srcs
[0]); break;
1388 case ir_unop_sin
: result
= nir_fsin(&b
, srcs
[0]); break;
1389 case ir_unop_cos
: result
= nir_fcos(&b
, srcs
[0]); break;
1390 case ir_unop_dFdx
: result
= nir_fddx(&b
, srcs
[0]); break;
1391 case ir_unop_dFdy
: result
= nir_fddy(&b
, srcs
[0]); break;
1392 case ir_unop_dFdx_fine
: result
= nir_fddx_fine(&b
, srcs
[0]); break;
1393 case ir_unop_dFdy_fine
: result
= nir_fddy_fine(&b
, srcs
[0]); break;
1394 case ir_unop_dFdx_coarse
: result
= nir_fddx_coarse(&b
, srcs
[0]); break;
1395 case ir_unop_dFdy_coarse
: result
= nir_fddy_coarse(&b
, srcs
[0]); break;
1396 case ir_unop_pack_snorm_2x16
:
1397 result
= nir_pack_snorm_2x16(&b
, srcs
[0]);
1399 case ir_unop_pack_snorm_4x8
:
1400 result
= nir_pack_snorm_4x8(&b
, srcs
[0]);
1402 case ir_unop_pack_unorm_2x16
:
1403 result
= nir_pack_unorm_2x16(&b
, srcs
[0]);
1405 case ir_unop_pack_unorm_4x8
:
1406 result
= nir_pack_unorm_4x8(&b
, srcs
[0]);
1408 case ir_unop_pack_half_2x16
:
1409 result
= nir_pack_half_2x16(&b
, srcs
[0]);
1411 case ir_unop_unpack_snorm_2x16
:
1412 result
= nir_unpack_snorm_2x16(&b
, srcs
[0]);
1414 case ir_unop_unpack_snorm_4x8
:
1415 result
= nir_unpack_snorm_4x8(&b
, srcs
[0]);
1417 case ir_unop_unpack_unorm_2x16
:
1418 result
= nir_unpack_unorm_2x16(&b
, srcs
[0]);
1420 case ir_unop_unpack_unorm_4x8
:
1421 result
= nir_unpack_unorm_4x8(&b
, srcs
[0]);
1423 case ir_unop_unpack_half_2x16
:
1424 result
= nir_unpack_half_2x16(&b
, srcs
[0]);
1426 case ir_unop_pack_double_2x32
:
1427 result
= nir_pack_double_2x32(&b
, srcs
[0]);
1429 case ir_unop_unpack_double_2x32
:
1430 result
= nir_unpack_double_2x32(&b
, srcs
[0]);
1432 case ir_unop_bitfield_reverse
:
1433 result
= nir_bitfield_reverse(&b
, srcs
[0]);
1435 case ir_unop_bit_count
:
1436 result
= nir_bit_count(&b
, srcs
[0]);
1438 case ir_unop_find_msb
:
1440 case GLSL_TYPE_UINT
:
1441 result
= nir_ufind_msb(&b
, srcs
[0]);
1444 result
= nir_ifind_msb(&b
, srcs
[0]);
1447 unreachable("Invalid type for findMSB()");
1450 case ir_unop_find_lsb
:
1451 result
= nir_find_lsb(&b
, srcs
[0]);
1455 switch (ir
->type
->vector_elements
) {
1457 switch (ir
->operands
[0]->type
->vector_elements
) {
1458 case 1: result
= nir_fnoise1_1(&b
, srcs
[0]); break;
1459 case 2: result
= nir_fnoise1_2(&b
, srcs
[0]); break;
1460 case 3: result
= nir_fnoise1_3(&b
, srcs
[0]); break;
1461 case 4: result
= nir_fnoise1_4(&b
, srcs
[0]); break;
1462 default: unreachable("not reached");
1466 switch (ir
->operands
[0]->type
->vector_elements
) {
1467 case 1: result
= nir_fnoise2_1(&b
, srcs
[0]); break;
1468 case 2: result
= nir_fnoise2_2(&b
, srcs
[0]); break;
1469 case 3: result
= nir_fnoise2_3(&b
, srcs
[0]); break;
1470 case 4: result
= nir_fnoise2_4(&b
, srcs
[0]); break;
1471 default: unreachable("not reached");
1475 switch (ir
->operands
[0]->type
->vector_elements
) {
1476 case 1: result
= nir_fnoise3_1(&b
, srcs
[0]); break;
1477 case 2: result
= nir_fnoise3_2(&b
, srcs
[0]); break;
1478 case 3: result
= nir_fnoise3_3(&b
, srcs
[0]); break;
1479 case 4: result
= nir_fnoise3_4(&b
, srcs
[0]); break;
1480 default: unreachable("not reached");
1484 switch (ir
->operands
[0]->type
->vector_elements
) {
1485 case 1: result
= nir_fnoise4_1(&b
, srcs
[0]); break;
1486 case 2: result
= nir_fnoise4_2(&b
, srcs
[0]); break;
1487 case 3: result
= nir_fnoise4_3(&b
, srcs
[0]); break;
1488 case 4: result
= nir_fnoise4_4(&b
, srcs
[0]); break;
1489 default: unreachable("not reached");
1493 unreachable("not reached");
1496 case ir_unop_get_buffer_size
: {
1497 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1499 nir_intrinsic_get_buffer_size
);
1500 load
->num_components
= ir
->type
->vector_elements
;
1501 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1502 unsigned bit_size
= glsl_get_bit_size(ir
->type
->base_type
);
1503 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1508 result
= type_is_float(out_type
) ? nir_fadd(&b
, srcs
[0], srcs
[1])
1509 : nir_iadd(&b
, srcs
[0], srcs
[1]);
1512 result
= type_is_float(out_type
) ? nir_fsub(&b
, srcs
[0], srcs
[1])
1513 : nir_isub(&b
, srcs
[0], srcs
[1]);
1516 result
= type_is_float(out_type
) ? nir_fmul(&b
, srcs
[0], srcs
[1])
1517 : nir_imul(&b
, srcs
[0], srcs
[1]);
1520 if (type_is_float(out_type
))
1521 result
= nir_fdiv(&b
, srcs
[0], srcs
[1]);
1522 else if (out_type
== GLSL_TYPE_INT
)
1523 result
= nir_idiv(&b
, srcs
[0], srcs
[1]);
1525 result
= nir_udiv(&b
, srcs
[0], srcs
[1]);
1528 result
= type_is_float(out_type
) ? nir_fmod(&b
, srcs
[0], srcs
[1])
1529 : nir_umod(&b
, srcs
[0], srcs
[1]);
1532 if (type_is_float(out_type
))
1533 result
= nir_fmin(&b
, srcs
[0], srcs
[1]);
1534 else if (out_type
== GLSL_TYPE_INT
)
1535 result
= nir_imin(&b
, srcs
[0], srcs
[1]);
1537 result
= nir_umin(&b
, srcs
[0], srcs
[1]);
1540 if (type_is_float(out_type
))
1541 result
= nir_fmax(&b
, srcs
[0], srcs
[1]);
1542 else if (out_type
== GLSL_TYPE_INT
)
1543 result
= nir_imax(&b
, srcs
[0], srcs
[1]);
1545 result
= nir_umax(&b
, srcs
[0], srcs
[1]);
1547 case ir_binop_pow
: result
= nir_fpow(&b
, srcs
[0], srcs
[1]); break;
1548 case ir_binop_bit_and
: result
= nir_iand(&b
, srcs
[0], srcs
[1]); break;
1549 case ir_binop_bit_or
: result
= nir_ior(&b
, srcs
[0], srcs
[1]); break;
1550 case ir_binop_bit_xor
: result
= nir_ixor(&b
, srcs
[0], srcs
[1]); break;
1551 case ir_binop_logic_and
:
1552 result
= supports_ints
? nir_iand(&b
, srcs
[0], srcs
[1])
1553 : nir_fand(&b
, srcs
[0], srcs
[1]);
1555 case ir_binop_logic_or
:
1556 result
= supports_ints
? nir_ior(&b
, srcs
[0], srcs
[1])
1557 : nir_for(&b
, srcs
[0], srcs
[1]);
1559 case ir_binop_logic_xor
:
1560 result
= supports_ints
? nir_ixor(&b
, srcs
[0], srcs
[1])
1561 : nir_fxor(&b
, srcs
[0], srcs
[1]);
1563 case ir_binop_lshift
: result
= nir_ishl(&b
, srcs
[0], srcs
[1]); break;
1564 case ir_binop_rshift
:
1565 result
= (out_type
== GLSL_TYPE_INT
) ? nir_ishr(&b
, srcs
[0], srcs
[1])
1566 : nir_ushr(&b
, srcs
[0], srcs
[1]);
1568 case ir_binop_imul_high
:
1569 result
= (out_type
== GLSL_TYPE_INT
) ? nir_imul_high(&b
, srcs
[0], srcs
[1])
1570 : nir_umul_high(&b
, srcs
[0], srcs
[1]);
1572 case ir_binop_carry
: result
= nir_uadd_carry(&b
, srcs
[0], srcs
[1]); break;
1573 case ir_binop_borrow
: result
= nir_usub_borrow(&b
, srcs
[0], srcs
[1]); break;
1575 if (supports_ints
) {
1576 if (type_is_float(types
[0]))
1577 result
= nir_flt(&b
, srcs
[0], srcs
[1]);
1578 else if (types
[0] == GLSL_TYPE_INT
)
1579 result
= nir_ilt(&b
, srcs
[0], srcs
[1]);
1581 result
= nir_ult(&b
, srcs
[0], srcs
[1]);
1583 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1586 case ir_binop_greater
:
1587 if (supports_ints
) {
1588 if (type_is_float(types
[0]))
1589 result
= nir_flt(&b
, srcs
[1], srcs
[0]);
1590 else if (types
[0] == GLSL_TYPE_INT
)
1591 result
= nir_ilt(&b
, srcs
[1], srcs
[0]);
1593 result
= nir_ult(&b
, srcs
[1], srcs
[0]);
1595 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1598 case ir_binop_lequal
:
1599 if (supports_ints
) {
1600 if (type_is_float(types
[0]))
1601 result
= nir_fge(&b
, srcs
[1], srcs
[0]);
1602 else if (types
[0] == GLSL_TYPE_INT
)
1603 result
= nir_ige(&b
, srcs
[1], srcs
[0]);
1605 result
= nir_uge(&b
, srcs
[1], srcs
[0]);
1607 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1610 case ir_binop_gequal
:
1611 if (supports_ints
) {
1612 if (type_is_float(types
[0]))
1613 result
= nir_fge(&b
, srcs
[0], srcs
[1]);
1614 else if (types
[0] == GLSL_TYPE_INT
)
1615 result
= nir_ige(&b
, srcs
[0], srcs
[1]);
1617 result
= nir_uge(&b
, srcs
[0], srcs
[1]);
1619 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1622 case ir_binop_equal
:
1623 if (supports_ints
) {
1624 if (type_is_float(types
[0]))
1625 result
= nir_feq(&b
, srcs
[0], srcs
[1]);
1627 result
= nir_ieq(&b
, srcs
[0], srcs
[1]);
1629 result
= nir_seq(&b
, srcs
[0], srcs
[1]);
1632 case ir_binop_nequal
:
1633 if (supports_ints
) {
1634 if (type_is_float(types
[0]))
1635 result
= nir_fne(&b
, srcs
[0], srcs
[1]);
1637 result
= nir_ine(&b
, srcs
[0], srcs
[1]);
1639 result
= nir_sne(&b
, srcs
[0], srcs
[1]);
1642 case ir_binop_all_equal
:
1643 if (supports_ints
) {
1644 if (type_is_float(types
[0])) {
1645 switch (ir
->operands
[0]->type
->vector_elements
) {
1646 case 1: result
= nir_feq(&b
, srcs
[0], srcs
[1]); break;
1647 case 2: result
= nir_ball_fequal2(&b
, srcs
[0], srcs
[1]); break;
1648 case 3: result
= nir_ball_fequal3(&b
, srcs
[0], srcs
[1]); break;
1649 case 4: result
= nir_ball_fequal4(&b
, srcs
[0], srcs
[1]); break;
1651 unreachable("not reached");
1654 switch (ir
->operands
[0]->type
->vector_elements
) {
1655 case 1: result
= nir_ieq(&b
, srcs
[0], srcs
[1]); break;
1656 case 2: result
= nir_ball_iequal2(&b
, srcs
[0], srcs
[1]); break;
1657 case 3: result
= nir_ball_iequal3(&b
, srcs
[0], srcs
[1]); break;
1658 case 4: result
= nir_ball_iequal4(&b
, srcs
[0], srcs
[1]); break;
1660 unreachable("not reached");
1664 switch (ir
->operands
[0]->type
->vector_elements
) {
1665 case 1: result
= nir_seq(&b
, srcs
[0], srcs
[1]); break;
1666 case 2: result
= nir_fall_equal2(&b
, srcs
[0], srcs
[1]); break;
1667 case 3: result
= nir_fall_equal3(&b
, srcs
[0], srcs
[1]); break;
1668 case 4: result
= nir_fall_equal4(&b
, srcs
[0], srcs
[1]); break;
1670 unreachable("not reached");
1674 case ir_binop_any_nequal
:
1675 if (supports_ints
) {
1676 if (type_is_float(types
[0])) {
1677 switch (ir
->operands
[0]->type
->vector_elements
) {
1678 case 1: result
= nir_fne(&b
, srcs
[0], srcs
[1]); break;
1679 case 2: result
= nir_bany_fnequal2(&b
, srcs
[0], srcs
[1]); break;
1680 case 3: result
= nir_bany_fnequal3(&b
, srcs
[0], srcs
[1]); break;
1681 case 4: result
= nir_bany_fnequal4(&b
, srcs
[0], srcs
[1]); break;
1683 unreachable("not reached");
1686 switch (ir
->operands
[0]->type
->vector_elements
) {
1687 case 1: result
= nir_ine(&b
, srcs
[0], srcs
[1]); break;
1688 case 2: result
= nir_bany_inequal2(&b
, srcs
[0], srcs
[1]); break;
1689 case 3: result
= nir_bany_inequal3(&b
, srcs
[0], srcs
[1]); break;
1690 case 4: result
= nir_bany_inequal4(&b
, srcs
[0], srcs
[1]); break;
1692 unreachable("not reached");
1696 switch (ir
->operands
[0]->type
->vector_elements
) {
1697 case 1: result
= nir_sne(&b
, srcs
[0], srcs
[1]); break;
1698 case 2: result
= nir_fany_nequal2(&b
, srcs
[0], srcs
[1]); break;
1699 case 3: result
= nir_fany_nequal3(&b
, srcs
[0], srcs
[1]); break;
1700 case 4: result
= nir_fany_nequal4(&b
, srcs
[0], srcs
[1]); break;
1702 unreachable("not reached");
1707 switch (ir
->operands
[0]->type
->vector_elements
) {
1708 case 2: result
= nir_fdot2(&b
, srcs
[0], srcs
[1]); break;
1709 case 3: result
= nir_fdot3(&b
, srcs
[0], srcs
[1]); break;
1710 case 4: result
= nir_fdot4(&b
, srcs
[0], srcs
[1]); break;
1712 unreachable("not reached");
1716 case ir_binop_ldexp
: result
= nir_ldexp(&b
, srcs
[0], srcs
[1]); break;
1718 result
= nir_ffma(&b
, srcs
[0], srcs
[1], srcs
[2]);
1721 result
= nir_flrp(&b
, srcs
[0], srcs
[1], srcs
[2]);
1725 result
= nir_bcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1727 result
= nir_fcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1729 case ir_triop_bitfield_extract
:
1730 result
= (out_type
== GLSL_TYPE_INT
) ?
1731 nir_ibitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]) :
1732 nir_ubitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]);
1734 case ir_quadop_bitfield_insert
:
1735 result
= nir_bitfield_insert(&b
, srcs
[0], srcs
[1], srcs
[2], srcs
[3]);
1737 case ir_quadop_vector
:
1738 result
= nir_vec(&b
, srcs
, ir
->type
->vector_elements
);
1742 unreachable("not reached");
1747 nir_visitor::visit(ir_swizzle
*ir
)
1749 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1750 result
= nir_swizzle(&b
, evaluate_rvalue(ir
->val
), swizzle
,
1751 ir
->type
->vector_elements
, !supports_ints
);
1755 nir_visitor::visit(ir_texture
*ir
)
1762 num_srcs
= 1; /* coordinate */
1767 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1768 num_srcs
= 2; /* coordinate, bias/lod */
1772 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1778 if (ir
->lod_info
.lod
!= NULL
)
1779 num_srcs
= 2; /* coordinate, lod */
1781 num_srcs
= 1; /* coordinate */
1785 op
= nir_texop_txf_ms
;
1786 num_srcs
= 2; /* coordinate, sample_index */
1791 if (ir
->lod_info
.lod
!= NULL
)
1792 num_srcs
= 1; /* lod */
1799 num_srcs
= 1; /* coordinate */
1804 num_srcs
= 1; /* coordinate */
1807 case ir_query_levels
:
1808 op
= nir_texop_query_levels
;
1812 case ir_texture_samples
:
1813 op
= nir_texop_texture_samples
;
1817 case ir_samples_identical
:
1818 op
= nir_texop_samples_identical
;
1819 num_srcs
= 1; /* coordinate */
1823 unreachable("not reached");
1826 if (ir
->projector
!= NULL
)
1828 if (ir
->shadow_comparitor
!= NULL
)
1830 if (ir
->offset
!= NULL
)
1833 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1836 instr
->sampler_dim
=
1837 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1838 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1839 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1840 if (instr
->is_shadow
)
1841 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1842 switch (ir
->type
->base_type
) {
1843 case GLSL_TYPE_FLOAT
:
1844 instr
->dest_type
= nir_type_float
;
1847 instr
->dest_type
= nir_type_int
;
1849 case GLSL_TYPE_BOOL
:
1850 case GLSL_TYPE_UINT
:
1851 instr
->dest_type
= nir_type_uint
;
1854 unreachable("not reached");
1857 instr
->texture
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1859 unsigned src_number
= 0;
1861 if (ir
->coordinate
!= NULL
) {
1862 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1863 instr
->src
[src_number
].src
=
1864 nir_src_for_ssa(evaluate_rvalue(ir
->coordinate
));
1865 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1869 if (ir
->projector
!= NULL
) {
1870 instr
->src
[src_number
].src
=
1871 nir_src_for_ssa(evaluate_rvalue(ir
->projector
));
1872 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1876 if (ir
->shadow_comparitor
!= NULL
) {
1877 instr
->src
[src_number
].src
=
1878 nir_src_for_ssa(evaluate_rvalue(ir
->shadow_comparitor
));
1879 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1883 if (ir
->offset
!= NULL
) {
1884 /* we don't support multiple offsets yet */
1885 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1887 instr
->src
[src_number
].src
=
1888 nir_src_for_ssa(evaluate_rvalue(ir
->offset
));
1889 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1895 instr
->src
[src_number
].src
=
1896 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.bias
));
1897 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1904 if (ir
->lod_info
.lod
!= NULL
) {
1905 instr
->src
[src_number
].src
=
1906 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.lod
));
1907 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1913 instr
->src
[src_number
].src
=
1914 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdx
));
1915 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1917 instr
->src
[src_number
].src
=
1918 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdy
));
1919 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1924 instr
->src
[src_number
].src
=
1925 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.sample_index
));
1926 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1931 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1938 assert(src_number
== num_srcs
);
1940 unsigned bit_size
= glsl_get_bit_size(ir
->type
->base_type
);
1941 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
), bit_size
);
1945 nir_visitor::visit(ir_constant
*ir
)
1948 * We don't know if this variable is an an array or struct that gets
1949 * dereferenced, so do the safe thing an make it a variable with a
1950 * constant initializer and return a dereference.
1954 nir_local_variable_create(this->impl
, ir
->type
, "const_temp");
1955 var
->data
.read_only
= true;
1956 var
->constant_initializer
= constant_copy(ir
, var
);
1958 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1959 this->deref_tail
= &this->deref_head
->deref
;
1963 nir_visitor::visit(ir_dereference_variable
*ir
)
1965 struct hash_entry
*entry
=
1966 _mesa_hash_table_search(this->var_table
, ir
->var
);
1968 nir_variable
*var
= (nir_variable
*) entry
->data
;
1970 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
1971 this->deref_head
= deref
;
1972 this->deref_tail
= &deref
->deref
;
1976 nir_visitor::visit(ir_dereference_record
*ir
)
1978 ir
->record
->accept(this);
1980 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
1981 assert(field_index
>= 0);
1983 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
1984 deref
->deref
.type
= ir
->type
;
1985 this->deref_tail
->child
= &deref
->deref
;
1986 this->deref_tail
= &deref
->deref
;
1990 nir_visitor::visit(ir_dereference_array
*ir
)
1992 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
1993 deref
->deref
.type
= ir
->type
;
1995 ir_constant
*const_index
= ir
->array_index
->as_constant();
1996 if (const_index
!= NULL
) {
1997 deref
->deref_array_type
= nir_deref_array_type_direct
;
1998 deref
->base_offset
= const_index
->value
.u
[0];
2000 deref
->deref_array_type
= nir_deref_array_type_indirect
;
2002 nir_src_for_ssa(evaluate_rvalue(ir
->array_index
));
2005 ir
->array
->accept(this);
2007 this->deref_tail
->child
= &deref
->deref
;
2008 ralloc_steal(this->deref_tail
, deref
);
2009 this->deref_tail
= &deref
->deref
;
2013 nir_visitor::visit(ir_barrier
*ir
)
2015 nir_intrinsic_instr
*instr
=
2016 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
2017 nir_builder_instr_insert(&b
, &instr
->instr
);