2 * Copyright © 2014 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Connor Abbott (cwabbott0@gmail.com)
28 #include "glsl_to_nir.h"
29 #include "nir_control_flow.h"
30 #include "ir_visitor.h"
31 #include "ir_hierarchical_visitor.h"
35 * pass to lower GLSL IR to NIR
37 * This will lower variable dereferences to loads/stores of corresponding
38 * variables in NIR - the variables will be converted to registers in a later
44 class nir_visitor
: public ir_visitor
47 nir_visitor(nir_shader
*shader
);
50 virtual void visit(ir_variable
*);
51 virtual void visit(ir_function
*);
52 virtual void visit(ir_function_signature
*);
53 virtual void visit(ir_loop
*);
54 virtual void visit(ir_if
*);
55 virtual void visit(ir_discard
*);
56 virtual void visit(ir_loop_jump
*);
57 virtual void visit(ir_return
*);
58 virtual void visit(ir_call
*);
59 virtual void visit(ir_assignment
*);
60 virtual void visit(ir_emit_vertex
*);
61 virtual void visit(ir_end_primitive
*);
62 virtual void visit(ir_expression
*);
63 virtual void visit(ir_swizzle
*);
64 virtual void visit(ir_texture
*);
65 virtual void visit(ir_constant
*);
66 virtual void visit(ir_dereference_variable
*);
67 virtual void visit(ir_dereference_record
*);
68 virtual void visit(ir_dereference_array
*);
69 virtual void visit(ir_barrier
*);
71 void create_function(ir_function
*ir
);
74 void create_overload(ir_function_signature
*ir
, nir_function
*function
);
75 void add_instr(nir_instr
*instr
, unsigned num_components
);
76 nir_src
evaluate_rvalue(ir_rvalue
*ir
);
78 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
);
79 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
);
80 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
,
82 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_src src1
,
83 nir_src src2
, nir_src src3
);
88 nir_function_impl
*impl
;
89 exec_list
*cf_node_list
;
90 nir_instr
*result
; /* result of the expression tree last visited */
92 nir_deref_var
*evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
);
94 /* the head of the dereference chain we're creating */
95 nir_deref_var
*deref_head
;
96 /* the tail of the dereference chain we're creating */
97 nir_deref
*deref_tail
;
99 nir_variable
*var
; /* variable created by ir_variable visitor */
101 /* whether the IR we're operating on is per-function or global */
104 /* map of ir_variable -> nir_variable */
105 struct hash_table
*var_table
;
107 /* map of ir_function_signature -> nir_function_overload */
108 struct hash_table
*overload_table
;
112 * This visitor runs before the main visitor, calling create_function() for
113 * each function so that the main visitor can resolve forward references in
117 class nir_function_visitor
: public ir_hierarchical_visitor
120 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
123 virtual ir_visitor_status
visit_enter(ir_function
*);
126 nir_visitor
*visitor
;
129 }; /* end of anonymous namespace */
132 glsl_to_nir(const struct gl_shader_program
*shader_prog
,
133 gl_shader_stage stage
,
134 const nir_shader_compiler_options
*options
)
136 struct gl_shader
*sh
= shader_prog
->_LinkedShaders
[stage
];
138 nir_shader
*shader
= nir_shader_create(NULL
, stage
, options
);
140 nir_visitor
v1(shader
);
141 nir_function_visitor
v2(&v1
);
143 visit_exec_list(sh
->ir
, &v1
);
145 nir_lower_outputs_to_temporaries(shader
);
147 /* TODO: Use _mesa_fls instead */
148 unsigned num_textures
= 0;
149 for (unsigned i
= 0; i
< 8 * sizeof(sh
->Program
->SamplersUsed
); i
++)
150 if (sh
->Program
->SamplersUsed
& (1 << i
))
153 shader
->info
.name
= ralloc_asprintf(shader
, "GLSL%d", shader_prog
->Name
);
154 if (shader_prog
->Label
)
155 shader
->info
.label
= ralloc_strdup(shader
, shader_prog
->Label
);
156 shader
->info
.num_textures
= num_textures
;
157 shader
->info
.num_ubos
= sh
->NumUniformBlocks
;
158 shader
->info
.num_abos
= shader_prog
->NumAtomicBuffers
;
159 shader
->info
.num_ssbos
= sh
->NumShaderStorageBlocks
;
160 shader
->info
.num_images
= sh
->NumImages
;
161 shader
->info
.inputs_read
= sh
->Program
->InputsRead
;
162 shader
->info
.outputs_written
= sh
->Program
->OutputsWritten
;
163 shader
->info
.system_values_read
= sh
->Program
->SystemValuesRead
;
164 shader
->info
.uses_texture_gather
= sh
->Program
->UsesGather
;
165 shader
->info
.uses_clip_distance_out
= sh
->Program
->UsesClipDistanceOut
;
166 shader
->info
.separate_shader
= shader_prog
->SeparateShader
;
167 shader
->info
.gs
.vertices_out
= sh
->Geom
.VerticesOut
;
168 shader
->info
.gs
.invocations
= sh
->Geom
.Invocations
;
169 shader
->info
.has_transform_feedback_varyings
=
170 shader_prog
->TransformFeedback
.NumVarying
> 0;
175 nir_visitor::nir_visitor(nir_shader
*shader
)
177 this->supports_ints
= shader
->options
->native_integers
;
178 this->shader
= shader
;
179 this->is_global
= true;
180 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
181 _mesa_key_pointer_equal
);
182 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
183 _mesa_key_pointer_equal
);
186 nir_visitor::~nir_visitor()
188 _mesa_hash_table_destroy(this->var_table
, NULL
);
189 _mesa_hash_table_destroy(this->overload_table
, NULL
);
193 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
196 ralloc_steal(mem_ctx
, this->deref_head
);
197 return this->deref_head
;
200 static nir_constant
*
201 constant_copy(ir_constant
*ir
, void *mem_ctx
)
206 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
208 unsigned total_elems
= ir
->type
->components();
210 switch (ir
->type
->base_type
) {
212 for (i
= 0; i
< total_elems
; i
++)
213 ret
->value
.u
[i
] = ir
->value
.u
[i
];
217 for (i
= 0; i
< total_elems
; i
++)
218 ret
->value
.i
[i
] = ir
->value
.i
[i
];
221 case GLSL_TYPE_FLOAT
:
222 for (i
= 0; i
< total_elems
; i
++)
223 ret
->value
.f
[i
] = ir
->value
.f
[i
];
227 for (i
= 0; i
< total_elems
; i
++)
228 ret
->value
.b
[i
] = ir
->value
.b
[i
];
231 case GLSL_TYPE_STRUCT
:
232 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
235 foreach_in_list(ir_constant
, field
, &ir
->components
) {
236 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
241 case GLSL_TYPE_ARRAY
:
242 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
245 for (i
= 0; i
< ir
->type
->length
; i
++)
246 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
250 unreachable("not reached");
257 nir_visitor::visit(ir_variable
*ir
)
259 nir_variable
*var
= ralloc(shader
, nir_variable
);
260 var
->type
= ir
->type
;
261 var
->name
= ralloc_strdup(var
, ir
->name
);
263 if (ir
->is_interface_instance() && ir
->get_max_ifc_array_access() != NULL
) {
264 unsigned size
= ir
->get_interface_type()->length
;
265 var
->max_ifc_array_access
= ralloc_array(var
, unsigned, size
);
266 memcpy(var
->max_ifc_array_access
, ir
->get_max_ifc_array_access(),
267 size
* sizeof(unsigned));
269 var
->max_ifc_array_access
= NULL
;
272 var
->data
.read_only
= ir
->data
.read_only
;
273 var
->data
.centroid
= ir
->data
.centroid
;
274 var
->data
.sample
= ir
->data
.sample
;
275 var
->data
.invariant
= ir
->data
.invariant
;
276 var
->data
.location
= ir
->data
.location
;
278 switch(ir
->data
.mode
) {
280 case ir_var_temporary
:
282 var
->data
.mode
= nir_var_global
;
284 var
->data
.mode
= nir_var_local
;
287 case ir_var_function_in
:
288 case ir_var_function_out
:
289 case ir_var_function_inout
:
290 case ir_var_const_in
:
291 var
->data
.mode
= nir_var_local
;
294 case ir_var_shader_in
:
295 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
296 ir
->data
.location
== VARYING_SLOT_FACE
) {
297 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
298 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
299 var
->data
.mode
= nir_var_system_value
;
300 } else if (shader
->stage
== MESA_SHADER_GEOMETRY
&&
301 ir
->data
.location
== VARYING_SLOT_PRIMITIVE_ID
) {
302 /* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
303 var
->data
.location
= SYSTEM_VALUE_PRIMITIVE_ID
;
304 var
->data
.mode
= nir_var_system_value
;
306 var
->data
.mode
= nir_var_shader_in
;
310 case ir_var_shader_out
:
311 var
->data
.mode
= nir_var_shader_out
;
315 var
->data
.mode
= nir_var_uniform
;
318 case ir_var_shader_storage
:
319 var
->data
.mode
= nir_var_shader_storage
;
322 case ir_var_system_value
:
323 var
->data
.mode
= nir_var_system_value
;
327 unreachable("not reached");
330 var
->data
.interpolation
= ir
->data
.interpolation
;
331 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
332 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
333 var
->data
.explicit_location
= ir
->data
.explicit_location
;
334 var
->data
.explicit_index
= ir
->data
.explicit_index
;
335 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
336 var
->data
.has_initializer
= ir
->data
.has_initializer
;
337 var
->data
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
338 var
->data
.location_frac
= ir
->data
.location_frac
;
339 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
340 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
342 switch (ir
->data
.depth_layout
) {
343 case ir_depth_layout_none
:
344 var
->data
.depth_layout
= nir_depth_layout_none
;
346 case ir_depth_layout_any
:
347 var
->data
.depth_layout
= nir_depth_layout_any
;
349 case ir_depth_layout_greater
:
350 var
->data
.depth_layout
= nir_depth_layout_greater
;
352 case ir_depth_layout_less
:
353 var
->data
.depth_layout
= nir_depth_layout_less
;
355 case ir_depth_layout_unchanged
:
356 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
359 unreachable("not reached");
362 var
->data
.index
= ir
->data
.index
;
363 var
->data
.binding
= ir
->data
.binding
;
364 /* XXX Get rid of buffer_index */
365 var
->data
.atomic
.buffer_index
= ir
->data
.binding
;
366 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
367 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
368 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
369 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
370 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
371 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
372 var
->data
.image
.format
= ir
->data
.image_format
;
373 var
->data
.max_array_access
= ir
->data
.max_array_access
;
375 var
->num_state_slots
= ir
->get_num_state_slots();
376 if (var
->num_state_slots
> 0) {
377 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
378 var
->num_state_slots
);
380 ir_state_slot
*state_slots
= ir
->get_state_slots();
381 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
382 for (unsigned j
= 0; j
< 5; j
++)
383 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
384 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
387 var
->state_slots
= NULL
;
390 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
392 var
->interface_type
= ir
->get_interface_type();
394 if (var
->data
.mode
== nir_var_local
)
395 nir_function_impl_add_variable(impl
, var
);
397 nir_shader_add_variable(shader
, var
);
399 _mesa_hash_table_insert(var_table
, ir
, var
);
404 nir_function_visitor::visit_enter(ir_function
*ir
)
406 visitor
->create_function(ir
);
407 return visit_continue_with_parent
;
412 nir_visitor::create_function(ir_function
*ir
)
414 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
415 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
416 create_overload(sig
, func
);
423 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
425 if (ir
->is_intrinsic
)
428 nir_function_overload
*overload
= nir_function_overload_create(function
);
430 unsigned num_params
= ir
->parameters
.length();
431 overload
->num_params
= num_params
;
432 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
435 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
436 switch (param
->data
.mode
) {
437 case ir_var_function_in
:
438 overload
->params
[i
].param_type
= nir_parameter_in
;
441 case ir_var_function_out
:
442 overload
->params
[i
].param_type
= nir_parameter_out
;
445 case ir_var_function_inout
:
446 overload
->params
[i
].param_type
= nir_parameter_inout
;
450 unreachable("not reached");
453 overload
->params
[i
].type
= param
->type
;
457 overload
->return_type
= ir
->return_type
;
459 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
463 nir_visitor::visit(ir_function
*ir
)
465 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
470 nir_visitor::visit(ir_function_signature
*ir
)
472 if (ir
->is_intrinsic
)
475 struct hash_entry
*entry
=
476 _mesa_hash_table_search(this->overload_table
, ir
);
479 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
481 if (ir
->is_defined
) {
482 nir_function_impl
*impl
= nir_function_impl_create(overload
);
485 unsigned num_params
= overload
->num_params
;
486 impl
->num_params
= num_params
;
487 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
489 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
491 impl
->params
[i
] = this->var
;
495 if (overload
->return_type
== glsl_type::void_type
) {
496 impl
->return_var
= NULL
;
498 impl
->return_var
= ralloc(this->shader
, nir_variable
);
499 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
501 impl
->return_var
->type
= overload
->return_type
;
504 this->is_global
= false;
506 this->cf_node_list
= &impl
->body
;
507 visit_exec_list(&ir
->body
, this);
509 this->is_global
= true;
511 overload
->impl
= NULL
;
516 nir_visitor::visit(ir_loop
*ir
)
518 exec_list
*old_list
= this->cf_node_list
;
520 nir_loop
*loop
= nir_loop_create(this->shader
);
521 nir_cf_node_insert_end(old_list
, &loop
->cf_node
);
522 this->cf_node_list
= &loop
->body
;
523 visit_exec_list(&ir
->body_instructions
, this);
525 this->cf_node_list
= old_list
;
529 nir_visitor::visit(ir_if
*ir
)
531 nir_src condition
= evaluate_rvalue(ir
->condition
);
533 exec_list
*old_list
= this->cf_node_list
;
535 nir_if
*if_stmt
= nir_if_create(this->shader
);
536 if_stmt
->condition
= condition
;
537 nir_cf_node_insert_end(old_list
, &if_stmt
->cf_node
);
539 this->cf_node_list
= &if_stmt
->then_list
;
540 visit_exec_list(&ir
->then_instructions
, this);
542 this->cf_node_list
= &if_stmt
->else_list
;
543 visit_exec_list(&ir
->else_instructions
, this);
545 this->cf_node_list
= old_list
;
549 nir_visitor::visit(ir_discard
*ir
)
552 * discards aren't treated as control flow, because before we lower them
553 * they can appear anywhere in the shader and the stuff after them may still
554 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
555 * discards will be immediately followed by a return.
558 nir_intrinsic_instr
*discard
;
560 discard
= nir_intrinsic_instr_create(this->shader
,
561 nir_intrinsic_discard_if
);
562 discard
->src
[0] = evaluate_rvalue(ir
->condition
);
564 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
566 nir_instr_insert_after_cf_list(this->cf_node_list
, &discard
->instr
);
570 nir_visitor::visit(ir_emit_vertex
*ir
)
572 nir_intrinsic_instr
*instr
=
573 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
574 instr
->const_index
[0] = ir
->stream_id();
575 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
579 nir_visitor::visit(ir_end_primitive
*ir
)
581 nir_intrinsic_instr
*instr
=
582 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
583 instr
->const_index
[0] = ir
->stream_id();
584 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
588 nir_visitor::visit(ir_loop_jump
*ir
)
592 case ir_loop_jump::jump_break
:
593 type
= nir_jump_break
;
595 case ir_loop_jump::jump_continue
:
596 type
= nir_jump_continue
;
599 unreachable("not reached");
602 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
603 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
607 nir_visitor::visit(ir_return
*ir
)
609 if (ir
->value
!= NULL
) {
610 nir_intrinsic_instr
*copy
=
611 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
613 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
614 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
617 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
618 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
622 nir_visitor::visit(ir_call
*ir
)
624 if (ir
->callee
->is_intrinsic
) {
626 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
627 op
= nir_intrinsic_atomic_counter_read_var
;
628 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
629 op
= nir_intrinsic_atomic_counter_inc_var
;
630 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
631 op
= nir_intrinsic_atomic_counter_dec_var
;
632 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
633 op
= nir_intrinsic_image_load
;
634 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
635 op
= nir_intrinsic_image_store
;
636 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
637 op
= nir_intrinsic_image_atomic_add
;
638 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
639 op
= nir_intrinsic_image_atomic_min
;
640 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
641 op
= nir_intrinsic_image_atomic_max
;
642 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
643 op
= nir_intrinsic_image_atomic_and
;
644 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
645 op
= nir_intrinsic_image_atomic_or
;
646 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
647 op
= nir_intrinsic_image_atomic_xor
;
648 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
649 op
= nir_intrinsic_image_atomic_exchange
;
650 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
651 op
= nir_intrinsic_image_atomic_comp_swap
;
652 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
653 op
= nir_intrinsic_memory_barrier
;
654 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_size") == 0) {
655 op
= nir_intrinsic_image_size
;
656 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_samples") == 0) {
657 op
= nir_intrinsic_image_samples
;
658 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_ssbo") == 0) {
659 op
= nir_intrinsic_store_ssbo
;
660 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_ssbo") == 0) {
661 op
= nir_intrinsic_load_ssbo
;
662 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_add_internal") == 0) {
663 op
= nir_intrinsic_ssbo_atomic_add
;
664 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_and_internal") == 0) {
665 op
= nir_intrinsic_ssbo_atomic_and
;
666 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_or_internal") == 0) {
667 op
= nir_intrinsic_ssbo_atomic_or
;
668 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_xor_internal") == 0) {
669 op
= nir_intrinsic_ssbo_atomic_xor
;
670 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_min_internal") == 0) {
671 assert(ir
->return_deref
);
672 if (ir
->return_deref
->type
== glsl_type::int_type
)
673 op
= nir_intrinsic_ssbo_atomic_imin
;
674 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
675 op
= nir_intrinsic_ssbo_atomic_umin
;
677 unreachable("Invalid type");
678 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_max_internal") == 0) {
679 assert(ir
->return_deref
);
680 if (ir
->return_deref
->type
== glsl_type::int_type
)
681 op
= nir_intrinsic_ssbo_atomic_imax
;
682 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
683 op
= nir_intrinsic_ssbo_atomic_umax
;
685 unreachable("Invalid type");
686 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_exchange_internal") == 0) {
687 op
= nir_intrinsic_ssbo_atomic_exchange
;
688 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_comp_swap_internal") == 0) {
689 op
= nir_intrinsic_ssbo_atomic_comp_swap
;
691 unreachable("not reached");
694 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
695 nir_dest
*dest
= &instr
->dest
;
698 case nir_intrinsic_atomic_counter_read_var
:
699 case nir_intrinsic_atomic_counter_inc_var
:
700 case nir_intrinsic_atomic_counter_dec_var
: {
701 ir_dereference
*param
=
702 (ir_dereference
*) ir
->actual_parameters
.get_head();
703 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
704 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
705 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
708 case nir_intrinsic_image_load
:
709 case nir_intrinsic_image_store
:
710 case nir_intrinsic_image_atomic_add
:
711 case nir_intrinsic_image_atomic_min
:
712 case nir_intrinsic_image_atomic_max
:
713 case nir_intrinsic_image_atomic_and
:
714 case nir_intrinsic_image_atomic_or
:
715 case nir_intrinsic_image_atomic_xor
:
716 case nir_intrinsic_image_atomic_exchange
:
717 case nir_intrinsic_image_atomic_comp_swap
:
718 case nir_intrinsic_image_samples
:
719 case nir_intrinsic_image_size
: {
720 nir_ssa_undef_instr
*instr_undef
=
721 nir_ssa_undef_instr_create(shader
, 1);
722 nir_instr_insert_after_cf_list(this->cf_node_list
,
723 &instr_undef
->instr
);
725 /* Set the image variable dereference. */
726 exec_node
*param
= ir
->actual_parameters
.get_head();
727 ir_dereference
*image
= (ir_dereference
*)param
;
728 const glsl_type
*type
=
729 image
->variable_referenced()->type
->without_array();
731 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
732 param
= param
->get_next();
734 /* Set the intrinsic destination. */
735 if (ir
->return_deref
) {
736 const nir_intrinsic_info
*info
=
737 &nir_intrinsic_infos
[instr
->intrinsic
];
738 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
739 info
->dest_components
, NULL
);
742 if (op
== nir_intrinsic_image_size
||
743 op
== nir_intrinsic_image_samples
) {
744 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
748 /* Set the address argument, extending the coordinate vector to four
751 const nir_src src_addr
= evaluate_rvalue((ir_dereference
*)param
);
752 nir_alu_instr
*instr_addr
= nir_alu_instr_create(shader
, nir_op_vec4
);
753 nir_ssa_dest_init(&instr_addr
->instr
, &instr_addr
->dest
.dest
, 4, NULL
);
755 for (int i
= 0; i
< 4; i
++) {
756 if (i
< type
->coordinate_components()) {
757 instr_addr
->src
[i
].src
= src_addr
;
758 instr_addr
->src
[i
].swizzle
[0] = i
;
760 instr_addr
->src
[i
].src
= nir_src_for_ssa(&instr_undef
->def
);
764 nir_instr_insert_after_cf_list(cf_node_list
, &instr_addr
->instr
);
765 instr
->src
[0] = nir_src_for_ssa(&instr_addr
->dest
.dest
.ssa
);
766 param
= param
->get_next();
768 /* Set the sample argument, which is undefined for single-sample
771 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
772 instr
->src
[1] = evaluate_rvalue((ir_dereference
*)param
);
773 param
= param
->get_next();
775 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
778 /* Set the intrinsic parameters. */
779 if (!param
->is_tail_sentinel()) {
780 instr
->src
[2] = evaluate_rvalue((ir_dereference
*)param
);
781 param
= param
->get_next();
784 if (!param
->is_tail_sentinel()) {
785 instr
->src
[3] = evaluate_rvalue((ir_dereference
*)param
);
786 param
= param
->get_next();
788 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
791 case nir_intrinsic_memory_barrier
:
792 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
794 case nir_intrinsic_store_ssbo
: {
795 exec_node
*param
= ir
->actual_parameters
.get_head();
796 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
798 param
= param
->get_next();
799 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
801 param
= param
->get_next();
802 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
804 param
= param
->get_next();
805 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
808 /* Check if we need the indirect version */
809 ir_constant
*const_offset
= offset
->as_constant();
811 op
= nir_intrinsic_store_ssbo_indirect
;
813 instr
= nir_intrinsic_instr_create(shader
, op
);
814 instr
->src
[2] = evaluate_rvalue(offset
);
815 instr
->const_index
[0] = 0;
817 instr
->const_index
[0] = const_offset
->value
.u
[0];
820 instr
->const_index
[1] = write_mask
->value
.u
[0];
822 instr
->src
[0] = evaluate_rvalue(val
);
823 instr
->num_components
= val
->type
->vector_elements
;
825 instr
->src
[1] = evaluate_rvalue(block
);
826 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
829 case nir_intrinsic_load_ssbo
: {
830 exec_node
*param
= ir
->actual_parameters
.get_head();
831 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
833 param
= param
->get_next();
834 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
836 /* Check if we need the indirect version */
837 ir_constant
*const_offset
= offset
->as_constant();
839 op
= nir_intrinsic_load_ssbo_indirect
;
841 instr
= nir_intrinsic_instr_create(shader
, op
);
842 instr
->src
[1] = evaluate_rvalue(offset
);
843 instr
->const_index
[0] = 0;
846 instr
->const_index
[0] = const_offset
->value
.u
[0];
849 instr
->src
[0] = evaluate_rvalue(block
);
851 const glsl_type
*type
= ir
->return_deref
->var
->type
;
852 instr
->num_components
= type
->vector_elements
;
854 /* Setup destination register */
855 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
856 type
->vector_elements
, NULL
);
858 /* Insert the created nir instruction now since in the case of boolean
859 * result we will need to emit another instruction after it
861 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
864 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
865 * consider a true boolean to be ~0. Fix this up with a != 0
868 if (type
->base_type
== GLSL_TYPE_BOOL
) {
869 nir_load_const_instr
*const_zero
=
870 nir_load_const_instr_create(shader
, 1);
871 const_zero
->value
.u
[0] = 0;
872 nir_instr_insert_after_cf_list(this->cf_node_list
,
875 nir_alu_instr
*load_ssbo_compare
=
876 nir_alu_instr_create(shader
, nir_op_ine
);
877 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
878 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
879 load_ssbo_compare
->src
[1].src
.is_ssa
= true;
880 load_ssbo_compare
->src
[1].src
.ssa
= &const_zero
->def
;
881 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
882 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
883 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
884 &load_ssbo_compare
->dest
.dest
,
885 type
->vector_elements
, NULL
);
886 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
887 nir_instr_insert_after_cf_list(this->cf_node_list
,
888 &load_ssbo_compare
->instr
);
889 dest
= &load_ssbo_compare
->dest
.dest
;
893 case nir_intrinsic_ssbo_atomic_add
:
894 case nir_intrinsic_ssbo_atomic_imin
:
895 case nir_intrinsic_ssbo_atomic_umin
:
896 case nir_intrinsic_ssbo_atomic_imax
:
897 case nir_intrinsic_ssbo_atomic_umax
:
898 case nir_intrinsic_ssbo_atomic_and
:
899 case nir_intrinsic_ssbo_atomic_or
:
900 case nir_intrinsic_ssbo_atomic_xor
:
901 case nir_intrinsic_ssbo_atomic_exchange
:
902 case nir_intrinsic_ssbo_atomic_comp_swap
: {
903 int param_count
= ir
->actual_parameters
.length();
904 assert(param_count
== 3 || param_count
== 4);
907 exec_node
*param
= ir
->actual_parameters
.get_head();
908 ir_instruction
*inst
= (ir_instruction
*) param
;
909 instr
->src
[0] = evaluate_rvalue(inst
->as_rvalue());
912 param
= param
->get_next();
913 inst
= (ir_instruction
*) param
;
914 instr
->src
[1] = evaluate_rvalue(inst
->as_rvalue());
916 /* data1 parameter (this is always present) */
917 param
= param
->get_next();
918 inst
= (ir_instruction
*) param
;
919 instr
->src
[2] = evaluate_rvalue(inst
->as_rvalue());
921 /* data2 parameter (only with atomic_comp_swap) */
922 if (param_count
== 4) {
923 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
924 param
= param
->get_next();
925 inst
= (ir_instruction
*) param
;
926 instr
->src
[3] = evaluate_rvalue(inst
->as_rvalue());
930 assert(ir
->return_deref
);
931 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
932 ir
->return_deref
->type
->vector_elements
, NULL
);
933 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
937 unreachable("not reached");
940 if (ir
->return_deref
) {
941 nir_intrinsic_instr
*store_instr
=
942 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
943 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
945 store_instr
->variables
[0] =
946 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
947 store_instr
->src
[0] = nir_src_for_ssa(&dest
->ssa
);
949 nir_instr_insert_after_cf_list(this->cf_node_list
,
950 &store_instr
->instr
);
956 struct hash_entry
*entry
=
957 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
959 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
961 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
964 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
965 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
969 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
970 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
974 nir_visitor::visit(ir_assignment
*ir
)
976 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
978 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
979 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
980 /* We're doing a plain-as-can-be copy, so emit a copy_var */
981 nir_intrinsic_instr
*copy
=
982 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
984 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
985 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
988 nir_if
*if_stmt
= nir_if_create(this->shader
);
989 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
990 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
991 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
993 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
998 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
1000 ir
->lhs
->accept(this);
1001 nir_deref_var
*lhs_deref
= this->deref_head
;
1002 nir_src src
= evaluate_rvalue(ir
->rhs
);
1004 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
1006 * We have no good way to update only part of a variable, so just load
1007 * the LHS and do a vec operation to combine the old with the new, and
1009 * back into the LHS. Copy propagation should get rid of the mess.
1012 nir_intrinsic_instr
*load
=
1013 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1014 load
->num_components
= ir
->lhs
->type
->vector_elements
;
1015 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
1016 load
->variables
[0] = lhs_deref
;
1017 ralloc_steal(load
, load
->variables
[0]);
1018 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
1021 switch (ir
->lhs
->type
->vector_elements
) {
1022 case 1: vec_op
= nir_op_imov
; break;
1023 case 2: vec_op
= nir_op_vec2
; break;
1024 case 3: vec_op
= nir_op_vec3
; break;
1025 case 4: vec_op
= nir_op_vec4
; break;
1026 default: unreachable("Invalid number of components"); break;
1028 nir_alu_instr
*vec
= nir_alu_instr_create(this->shader
, vec_op
);
1029 nir_ssa_dest_init(&vec
->instr
, &vec
->dest
.dest
, num_components
, NULL
);
1030 vec
->dest
.write_mask
= (1 << num_components
) - 1;
1032 unsigned component
= 0;
1033 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
1034 if (ir
->write_mask
& (1 << i
)) {
1035 vec
->src
[i
].src
= src
;
1037 /* GLSL IR will give us the input to the write-masked assignment
1038 * in a single packed vector. So, for example, if the
1039 * writemask is xzw, then we have to swizzle x -> x, y -> z,
1040 * and z -> w and get the y component from the load.
1042 vec
->src
[i
].swizzle
[0] = component
++;
1044 vec
->src
[i
].src
.is_ssa
= true;
1045 vec
->src
[i
].src
.ssa
= &load
->dest
.ssa
;
1046 vec
->src
[i
].swizzle
[0] = i
;
1050 nir_instr_insert_after_cf_list(this->cf_node_list
, &vec
->instr
);
1053 src
.ssa
= &vec
->dest
.dest
.ssa
;
1056 nir_intrinsic_instr
*store
=
1057 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1058 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1059 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
1060 store
->variables
[0] = nir_deref_as_var(store_deref
);
1061 store
->src
[0] = src
;
1063 if (ir
->condition
) {
1064 nir_if
*if_stmt
= nir_if_create(this->shader
);
1065 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
1066 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
1067 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
1069 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
1074 * Given an instruction, returns a pointer to its destination or NULL if there
1075 * is no destination.
1077 * Note that this only handles instructions we generate at this level.
1080 get_instr_dest(nir_instr
*instr
)
1082 nir_alu_instr
*alu_instr
;
1083 nir_intrinsic_instr
*intrinsic_instr
;
1084 nir_tex_instr
*tex_instr
;
1086 switch (instr
->type
) {
1087 case nir_instr_type_alu
:
1088 alu_instr
= nir_instr_as_alu(instr
);
1089 return &alu_instr
->dest
.dest
;
1091 case nir_instr_type_intrinsic
:
1092 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1093 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1094 return &intrinsic_instr
->dest
;
1098 case nir_instr_type_tex
:
1099 tex_instr
= nir_instr_as_tex(instr
);
1100 return &tex_instr
->dest
;
1103 unreachable("not reached");
1110 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
1112 nir_dest
*dest
= get_instr_dest(instr
);
1115 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
1117 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
1118 this->result
= instr
;
1122 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1125 if (ir
->as_dereference() || ir
->as_constant()) {
1127 * A dereference is being used on the right hand side, which means we
1128 * must emit a variable load.
1131 nir_intrinsic_instr
*load_instr
=
1132 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1133 load_instr
->num_components
= ir
->type
->vector_elements
;
1134 load_instr
->variables
[0] = this->deref_head
;
1135 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1136 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
1139 nir_dest
*dest
= get_instr_dest(this->result
);
1140 assert(dest
->is_ssa
);
1142 return nir_src_for_ssa(&dest
->ssa
);
1146 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
1148 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
1149 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
1150 instr
->src
[i
].src
= srcs
[i
];
1151 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
1152 add_instr(&instr
->instr
, dest_size
);
1157 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
1159 assert(nir_op_infos
[op
].num_inputs
== 1);
1160 return emit(op
, dest_size
, &src1
);
1164 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
1167 assert(nir_op_infos
[op
].num_inputs
== 2);
1168 nir_src srcs
[] = { src1
, src2
};
1169 return emit(op
, dest_size
, srcs
);
1173 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
1174 nir_src src2
, nir_src src3
)
1176 assert(nir_op_infos
[op
].num_inputs
== 3);
1177 nir_src srcs
[] = { src1
, src2
, src3
};
1178 return emit(op
, dest_size
, srcs
);
1182 nir_visitor::visit(ir_expression
*ir
)
1184 /* Some special cases */
1185 switch (ir
->operation
) {
1186 case ir_binop_ubo_load
: {
1187 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
1189 nir_intrinsic_op op
;
1191 op
= nir_intrinsic_load_ubo
;
1193 op
= nir_intrinsic_load_ubo_indirect
;
1195 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
1196 load
->num_components
= ir
->type
->vector_elements
;
1197 load
->const_index
[0] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
1198 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
1200 load
->src
[1] = evaluate_rvalue(ir
->operands
[1]);
1201 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1204 * In UBO's, a true boolean value is any non-zero value, but we consider
1205 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1208 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
1209 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
, 1);
1210 const_zero
->value
.u
[0] = 0;
1211 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
1213 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
1214 compare
->src
[0].src
.is_ssa
= true;
1215 compare
->src
[0].src
.ssa
= &load
->dest
.ssa
;
1216 compare
->src
[1].src
.is_ssa
= true;
1217 compare
->src
[1].src
.ssa
= &const_zero
->def
;
1218 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1219 compare
->src
[1].swizzle
[i
] = 0;
1220 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
1222 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
1228 case ir_unop_interpolate_at_centroid
:
1229 case ir_binop_interpolate_at_offset
:
1230 case ir_binop_interpolate_at_sample
: {
1231 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1232 ir_swizzle
*swizzle
= NULL
;
1234 /* the api does not allow a swizzle here, but the varying packing code
1235 * may have pushed one into here.
1237 swizzle
= ir
->operands
[0]->as_swizzle();
1239 deref
= swizzle
->val
->as_dereference();
1243 deref
->accept(this);
1245 nir_intrinsic_op op
;
1246 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1247 switch (ir
->operation
) {
1248 case ir_unop_interpolate_at_centroid
:
1249 op
= nir_intrinsic_interp_var_at_centroid
;
1251 case ir_binop_interpolate_at_offset
:
1252 op
= nir_intrinsic_interp_var_at_offset
;
1254 case ir_binop_interpolate_at_sample
:
1255 op
= nir_intrinsic_interp_var_at_sample
;
1258 unreachable("Invalid interpolation intrinsic");
1261 /* This case can happen if the vertex shader does not write the
1262 * given varying. In this case, the linker will lower it to a
1263 * global variable. Since interpolating a variable makes no
1264 * sense, we'll just turn it into a load which will probably
1265 * eventually end up as an SSA definition.
1267 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1268 op
= nir_intrinsic_load_var
;
1271 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1272 intrin
->num_components
= deref
->type
->vector_elements
;
1273 intrin
->variables
[0] = this->deref_head
;
1274 ralloc_steal(intrin
, intrin
->variables
[0]);
1276 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1277 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1278 intrin
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
1280 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
1283 nir_alu_instr
*mov
= nir_alu_instr_create(shader
, nir_op_imov
);
1284 mov
->dest
.write_mask
= (1 << swizzle
->type
->vector_elements
) - 1;
1285 mov
->src
[0].src
.is_ssa
= true;
1286 mov
->src
[0].src
.ssa
= &intrin
->dest
.ssa
;
1288 mov
->src
[0].swizzle
[0] = swizzle
->mask
.x
;
1289 mov
->src
[0].swizzle
[1] = swizzle
->mask
.y
;
1290 mov
->src
[0].swizzle
[2] = swizzle
->mask
.z
;
1291 mov
->src
[0].swizzle
[3] = swizzle
->mask
.w
;
1292 for (unsigned i
= deref
->type
->vector_elements
; i
< 4; i
++)
1293 mov
->src
[0].swizzle
[i
] = 0;
1295 add_instr(&mov
->instr
, swizzle
->type
->vector_elements
);
1306 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1307 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1309 glsl_base_type types
[4];
1310 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1312 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1314 types
[i
] = GLSL_TYPE_FLOAT
;
1316 glsl_base_type out_type
;
1318 out_type
= ir
->type
->base_type
;
1320 out_type
= GLSL_TYPE_FLOAT
;
1322 unsigned dest_size
= ir
->type
->vector_elements
;
1324 nir_alu_instr
*instr
;
1327 switch (ir
->operation
) {
1328 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
1329 case ir_unop_logic_not
:
1330 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1333 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fneg
: nir_op_ineg
,
1337 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fabs
: nir_op_iabs
,
1340 case ir_unop_saturate
:
1341 assert(types
[0] == GLSL_TYPE_FLOAT
);
1342 instr
= emit(nir_op_fsat
, dest_size
, srcs
);
1345 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1348 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1349 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1350 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1351 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1352 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1353 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1354 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1356 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1359 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1362 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1364 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1365 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1366 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1367 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1368 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1371 case ir_unop_bitcast_i2f
:
1372 case ir_unop_bitcast_f2i
:
1373 case ir_unop_bitcast_u2f
:
1374 case ir_unop_bitcast_f2u
:
1375 case ir_unop_subroutine_to_int
:
1377 emit(nir_op_imov
, dest_size
, srcs
);
1380 switch (ir
->operands
[0]->type
->vector_elements
) {
1382 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1386 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1390 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1394 unreachable("not reached");
1397 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1398 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1399 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1400 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1401 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1402 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1403 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1404 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1405 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1406 case ir_unop_dFdx_fine
: emit(nir_op_fddx_fine
, dest_size
, srcs
); break;
1407 case ir_unop_dFdy_fine
: emit(nir_op_fddy_fine
, dest_size
, srcs
); break;
1408 case ir_unop_dFdx_coarse
: emit(nir_op_fddx_coarse
, dest_size
, srcs
); break;
1409 case ir_unop_dFdy_coarse
: emit(nir_op_fddy_coarse
, dest_size
, srcs
); break;
1410 case ir_unop_pack_snorm_2x16
:
1411 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1413 case ir_unop_pack_snorm_4x8
:
1414 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1416 case ir_unop_pack_unorm_2x16
:
1417 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1419 case ir_unop_pack_unorm_4x8
:
1420 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1422 case ir_unop_pack_half_2x16
:
1423 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1425 case ir_unop_unpack_snorm_2x16
:
1426 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1428 case ir_unop_unpack_snorm_4x8
:
1429 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1431 case ir_unop_unpack_unorm_2x16
:
1432 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1434 case ir_unop_unpack_unorm_4x8
:
1435 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1437 case ir_unop_unpack_half_2x16
:
1438 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1440 case ir_unop_unpack_half_2x16_split_x
:
1441 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1443 case ir_unop_unpack_half_2x16_split_y
:
1444 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1446 case ir_unop_bitfield_reverse
:
1447 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1449 case ir_unop_bit_count
:
1450 emit(nir_op_bit_count
, dest_size
, srcs
);
1452 case ir_unop_find_msb
:
1454 case GLSL_TYPE_UINT
:
1455 emit(nir_op_ufind_msb
, dest_size
, srcs
);
1458 emit(nir_op_ifind_msb
, dest_size
, srcs
);
1461 unreachable("Invalid type for findMSB()");
1464 case ir_unop_find_lsb
:
1465 emit(nir_op_find_lsb
, dest_size
, srcs
);
1469 switch (ir
->type
->vector_elements
) {
1471 switch (ir
->operands
[0]->type
->vector_elements
) {
1472 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1473 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1474 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1475 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1476 default: unreachable("not reached");
1480 switch (ir
->operands
[0]->type
->vector_elements
) {
1481 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1482 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1483 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1484 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1485 default: unreachable("not reached");
1489 switch (ir
->operands
[0]->type
->vector_elements
) {
1490 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1491 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1492 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1493 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1494 default: unreachable("not reached");
1498 switch (ir
->operands
[0]->type
->vector_elements
) {
1499 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1500 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1501 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1502 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1503 default: unreachable("not reached");
1507 unreachable("not reached");
1510 case ir_unop_get_buffer_size
: {
1511 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1513 nir_intrinsic_get_buffer_size
);
1514 load
->num_components
= ir
->type
->vector_elements
;
1515 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
1516 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1528 case ir_binop_bit_and
:
1529 case ir_binop_bit_or
:
1530 case ir_binop_bit_xor
:
1531 case ir_binop_logic_and
:
1532 case ir_binop_logic_or
:
1533 case ir_binop_logic_xor
:
1534 case ir_binop_lshift
:
1535 case ir_binop_rshift
:
1536 switch (ir
->operation
) {
1538 if (out_type
== GLSL_TYPE_FLOAT
)
1544 if (out_type
== GLSL_TYPE_FLOAT
)
1550 if (out_type
== GLSL_TYPE_FLOAT
)
1556 if (out_type
== GLSL_TYPE_FLOAT
)
1558 else if (out_type
== GLSL_TYPE_INT
)
1564 if (out_type
== GLSL_TYPE_FLOAT
)
1570 if (out_type
== GLSL_TYPE_FLOAT
)
1572 else if (out_type
== GLSL_TYPE_INT
)
1578 if (out_type
== GLSL_TYPE_FLOAT
)
1580 else if (out_type
== GLSL_TYPE_INT
)
1585 case ir_binop_bit_and
:
1588 case ir_binop_bit_or
:
1591 case ir_binop_bit_xor
:
1594 case ir_binop_logic_and
:
1600 case ir_binop_logic_or
:
1606 case ir_binop_logic_xor
:
1612 case ir_binop_lshift
:
1615 case ir_binop_rshift
:
1616 if (out_type
== GLSL_TYPE_INT
)
1626 unreachable("not reached");
1629 instr
= emit(op
, dest_size
, srcs
);
1631 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1632 ir
->operands
[1]->type
->vector_elements
== 1) {
1633 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1635 instr
->src
[1].swizzle
[i
] = 0;
1639 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1640 ir
->operands
[0]->type
->vector_elements
== 1) {
1641 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1643 instr
->src
[0].swizzle
[i
] = 0;
1648 case ir_binop_imul_high
:
1649 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1652 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1653 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1655 if (supports_ints
) {
1656 if (types
[0] == GLSL_TYPE_FLOAT
)
1657 emit(nir_op_flt
, dest_size
, srcs
);
1658 else if (types
[0] == GLSL_TYPE_INT
)
1659 emit(nir_op_ilt
, dest_size
, srcs
);
1661 emit(nir_op_ult
, dest_size
, srcs
);
1663 emit(nir_op_slt
, dest_size
, srcs
);
1666 case ir_binop_greater
:
1667 if (supports_ints
) {
1668 if (types
[0] == GLSL_TYPE_FLOAT
)
1669 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1670 else if (types
[0] == GLSL_TYPE_INT
)
1671 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1673 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1675 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1678 case ir_binop_lequal
:
1679 if (supports_ints
) {
1680 if (types
[0] == GLSL_TYPE_FLOAT
)
1681 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1682 else if (types
[0] == GLSL_TYPE_INT
)
1683 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1685 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1687 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1690 case ir_binop_gequal
:
1691 if (supports_ints
) {
1692 if (types
[0] == GLSL_TYPE_FLOAT
)
1693 emit(nir_op_fge
, dest_size
, srcs
);
1694 else if (types
[0] == GLSL_TYPE_INT
)
1695 emit(nir_op_ige
, dest_size
, srcs
);
1697 emit(nir_op_uge
, dest_size
, srcs
);
1699 emit(nir_op_slt
, dest_size
, srcs
);
1702 case ir_binop_equal
:
1703 if (supports_ints
) {
1704 if (types
[0] == GLSL_TYPE_FLOAT
)
1705 emit(nir_op_feq
, dest_size
, srcs
);
1707 emit(nir_op_ieq
, dest_size
, srcs
);
1709 emit(nir_op_seq
, dest_size
, srcs
);
1712 case ir_binop_nequal
:
1713 if (supports_ints
) {
1714 if (types
[0] == GLSL_TYPE_FLOAT
)
1715 emit(nir_op_fne
, dest_size
, srcs
);
1717 emit(nir_op_ine
, dest_size
, srcs
);
1719 emit(nir_op_sne
, dest_size
, srcs
);
1722 case ir_binop_all_equal
:
1723 if (supports_ints
) {
1724 if (types
[0] == GLSL_TYPE_FLOAT
) {
1725 switch (ir
->operands
[0]->type
->vector_elements
) {
1726 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1727 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1728 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1729 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1731 unreachable("not reached");
1734 switch (ir
->operands
[0]->type
->vector_elements
) {
1735 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1736 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1737 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1738 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1740 unreachable("not reached");
1744 switch (ir
->operands
[0]->type
->vector_elements
) {
1745 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1746 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1747 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1748 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1750 unreachable("not reached");
1754 case ir_binop_any_nequal
:
1755 if (supports_ints
) {
1756 if (types
[0] == GLSL_TYPE_FLOAT
) {
1757 switch (ir
->operands
[0]->type
->vector_elements
) {
1758 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1759 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1760 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1761 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1763 unreachable("not reached");
1766 switch (ir
->operands
[0]->type
->vector_elements
) {
1767 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1768 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1769 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1770 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1772 unreachable("not reached");
1776 switch (ir
->operands
[0]->type
->vector_elements
) {
1777 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1778 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1779 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1780 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1782 unreachable("not reached");
1787 switch (ir
->operands
[0]->type
->vector_elements
) {
1788 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1789 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1790 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1792 unreachable("not reached");
1796 case ir_binop_pack_half_2x16_split
:
1797 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1799 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1800 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1801 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1803 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1804 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1805 ir
->operands
[2]->type
->vector_elements
== 1) {
1806 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1808 instr
->src
[2].swizzle
[i
] = 0;
1814 emit(nir_op_bcsel
, dest_size
, srcs
);
1816 emit(nir_op_fcsel
, dest_size
, srcs
);
1819 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1820 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1821 instr
->src
[0].swizzle
[i
] = 0;
1824 case ir_triop_bitfield_extract
:
1825 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1826 nir_op_ubitfield_extract
, dest_size
, srcs
);
1827 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1828 instr
->src
[1].swizzle
[i
] = 0;
1829 instr
->src
[2].swizzle
[i
] = 0;
1832 case ir_quadop_bitfield_insert
:
1833 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1834 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1835 instr
->src
[2].swizzle
[i
] = 0;
1836 instr
->src
[3].swizzle
[i
] = 0;
1839 case ir_quadop_vector
:
1840 switch (ir
->type
->vector_elements
) {
1841 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1842 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1843 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1844 default: unreachable("not reached");
1849 unreachable("not reached");
1854 nir_visitor::visit(ir_swizzle
*ir
)
1856 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1857 ir
->type
->vector_elements
,
1858 evaluate_rvalue(ir
->val
));
1860 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1861 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1862 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1866 nir_visitor::visit(ir_texture
*ir
)
1873 num_srcs
= 1; /* coordinate */
1878 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1879 num_srcs
= 2; /* coordinate, bias/lod */
1883 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1889 if (ir
->lod_info
.lod
!= NULL
)
1890 num_srcs
= 2; /* coordinate, lod */
1892 num_srcs
= 1; /* coordinate */
1896 op
= nir_texop_txf_ms
;
1897 num_srcs
= 2; /* coordinate, sample_index */
1902 if (ir
->lod_info
.lod
!= NULL
)
1903 num_srcs
= 1; /* lod */
1910 num_srcs
= 1; /* coordinate */
1915 num_srcs
= 1; /* coordinate */
1918 case ir_query_levels
:
1919 op
= nir_texop_query_levels
;
1923 case ir_texture_samples
:
1924 op
= nir_texop_texture_samples
;
1929 unreachable("not reached");
1932 if (ir
->projector
!= NULL
)
1934 if (ir
->shadow_comparitor
!= NULL
)
1936 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1939 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1942 instr
->sampler_dim
=
1943 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1944 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1945 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1946 if (instr
->is_shadow
)
1947 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1948 switch (ir
->type
->base_type
) {
1949 case GLSL_TYPE_FLOAT
:
1950 instr
->dest_type
= nir_type_float
;
1953 instr
->dest_type
= nir_type_int
;
1955 case GLSL_TYPE_UINT
:
1956 instr
->dest_type
= nir_type_unsigned
;
1959 unreachable("not reached");
1962 instr
->sampler
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1964 unsigned src_number
= 0;
1966 if (ir
->coordinate
!= NULL
) {
1967 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1968 instr
->src
[src_number
].src
= 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
= evaluate_rvalue(ir
->projector
);
1975 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1979 if (ir
->shadow_comparitor
!= NULL
) {
1980 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->shadow_comparitor
);
1981 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1985 if (ir
->offset
!= NULL
) {
1986 /* we don't support multiple offsets yet */
1987 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1989 ir_constant
*const_offset
= ir
->offset
->as_constant();
1990 if (const_offset
!= NULL
) {
1991 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1992 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1994 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->offset
);
1995 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
2002 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.bias
);
2003 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
2010 if (ir
->lod_info
.lod
!= NULL
) {
2011 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.lod
);
2012 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
2018 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdx
);
2019 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
2021 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdy
);
2022 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
2027 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.sample_index
);
2028 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
2033 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
2040 assert(src_number
== num_srcs
);
2042 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
2046 nir_visitor::visit(ir_constant
*ir
)
2049 * We don't know if this variable is an an array or struct that gets
2050 * dereferenced, so do the safe thing an make it a variable with a
2051 * constant initializer and return a dereference.
2055 nir_local_variable_create(this->impl
, ir
->type
, "const_temp");
2056 var
->data
.read_only
= true;
2057 var
->constant_initializer
= constant_copy(ir
, var
);
2059 this->deref_head
= nir_deref_var_create(this->shader
, var
);
2060 this->deref_tail
= &this->deref_head
->deref
;
2064 nir_visitor::visit(ir_dereference_variable
*ir
)
2066 struct hash_entry
*entry
=
2067 _mesa_hash_table_search(this->var_table
, ir
->var
);
2069 nir_variable
*var
= (nir_variable
*) entry
->data
;
2071 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
2072 this->deref_head
= deref
;
2073 this->deref_tail
= &deref
->deref
;
2077 nir_visitor::visit(ir_dereference_record
*ir
)
2079 ir
->record
->accept(this);
2081 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
2082 assert(field_index
>= 0);
2084 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
2085 deref
->deref
.type
= ir
->type
;
2086 this->deref_tail
->child
= &deref
->deref
;
2087 this->deref_tail
= &deref
->deref
;
2091 nir_visitor::visit(ir_dereference_array
*ir
)
2093 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
2094 deref
->deref
.type
= ir
->type
;
2096 ir_constant
*const_index
= ir
->array_index
->as_constant();
2097 if (const_index
!= NULL
) {
2098 deref
->deref_array_type
= nir_deref_array_type_direct
;
2099 deref
->base_offset
= const_index
->value
.u
[0];
2101 deref
->deref_array_type
= nir_deref_array_type_indirect
;
2102 deref
->indirect
= evaluate_rvalue(ir
->array_index
);
2105 ir
->array
->accept(this);
2107 this->deref_tail
->child
= &deref
->deref
;
2108 ralloc_steal(this->deref_tail
, deref
);
2109 this->deref_tail
= &deref
->deref
;
2113 nir_visitor::visit(ir_barrier
*ir
)
2115 nir_intrinsic_instr
*instr
=
2116 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
2117 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);