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", sh
->Name
);
154 shader
->info
.num_textures
= num_textures
;
155 shader
->info
.num_ubos
= sh
->NumBufferInterfaceBlocks
;
156 shader
->info
.num_abos
= shader_prog
->NumAtomicBuffers
;
157 shader
->info
.num_ssbos
= shader_prog
->NumBufferInterfaceBlocks
;
158 shader
->info
.num_images
= sh
->NumImages
;
159 shader
->info
.inputs_read
= sh
->Program
->InputsRead
;
160 shader
->info
.outputs_written
= sh
->Program
->OutputsWritten
;
161 shader
->info
.system_values_read
= sh
->Program
->SystemValuesRead
;
162 shader
->info
.uses_texture_gather
= sh
->Program
->UsesGather
;
163 shader
->info
.uses_clip_distance_out
= sh
->Program
->UsesClipDistanceOut
;
164 shader
->info
.separate_shader
= shader_prog
->SeparateShader
;
165 shader
->info
.gs
.vertices_out
= sh
->Geom
.VerticesOut
;
166 shader
->info
.gs
.invocations
= sh
->Geom
.Invocations
;
167 shader
->info
.has_transform_feedback_varyings
=
168 shader_prog
->TransformFeedback
.NumVarying
> 0;
173 nir_visitor::nir_visitor(nir_shader
*shader
)
175 this->supports_ints
= shader
->options
->native_integers
;
176 this->shader
= shader
;
177 this->is_global
= true;
178 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
179 _mesa_key_pointer_equal
);
180 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
181 _mesa_key_pointer_equal
);
184 nir_visitor::~nir_visitor()
186 _mesa_hash_table_destroy(this->var_table
, NULL
);
187 _mesa_hash_table_destroy(this->overload_table
, NULL
);
191 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
194 ralloc_steal(mem_ctx
, this->deref_head
);
195 return this->deref_head
;
198 static nir_constant
*
199 constant_copy(ir_constant
*ir
, void *mem_ctx
)
204 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
206 unsigned total_elems
= ir
->type
->components();
208 switch (ir
->type
->base_type
) {
210 for (i
= 0; i
< total_elems
; i
++)
211 ret
->value
.u
[i
] = ir
->value
.u
[i
];
215 for (i
= 0; i
< total_elems
; i
++)
216 ret
->value
.i
[i
] = ir
->value
.i
[i
];
219 case GLSL_TYPE_FLOAT
:
220 for (i
= 0; i
< total_elems
; i
++)
221 ret
->value
.f
[i
] = ir
->value
.f
[i
];
225 for (i
= 0; i
< total_elems
; i
++)
226 ret
->value
.b
[i
] = ir
->value
.b
[i
];
229 case GLSL_TYPE_STRUCT
:
230 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
233 foreach_in_list(ir_constant
, field
, &ir
->components
) {
234 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
239 case GLSL_TYPE_ARRAY
:
240 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
243 for (i
= 0; i
< ir
->type
->length
; i
++)
244 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
248 unreachable("not reached");
255 nir_visitor::visit(ir_variable
*ir
)
257 nir_variable
*var
= ralloc(shader
, nir_variable
);
258 var
->type
= ir
->type
;
259 var
->name
= ralloc_strdup(var
, ir
->name
);
261 if (ir
->is_interface_instance() && ir
->get_max_ifc_array_access() != NULL
) {
262 unsigned size
= ir
->get_interface_type()->length
;
263 var
->max_ifc_array_access
= ralloc_array(var
, unsigned, size
);
264 memcpy(var
->max_ifc_array_access
, ir
->get_max_ifc_array_access(),
265 size
* sizeof(unsigned));
267 var
->max_ifc_array_access
= NULL
;
270 var
->data
.read_only
= ir
->data
.read_only
;
271 var
->data
.centroid
= ir
->data
.centroid
;
272 var
->data
.sample
= ir
->data
.sample
;
273 var
->data
.invariant
= ir
->data
.invariant
;
274 var
->data
.location
= ir
->data
.location
;
276 switch(ir
->data
.mode
) {
278 case ir_var_temporary
:
280 var
->data
.mode
= nir_var_global
;
282 var
->data
.mode
= nir_var_local
;
285 case ir_var_function_in
:
286 case ir_var_function_out
:
287 case ir_var_function_inout
:
288 case ir_var_const_in
:
289 var
->data
.mode
= nir_var_local
;
292 case ir_var_shader_in
:
293 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
294 ir
->data
.location
== VARYING_SLOT_FACE
) {
295 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
296 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
297 var
->data
.mode
= nir_var_system_value
;
298 } else if (shader
->stage
== MESA_SHADER_GEOMETRY
&&
299 ir
->data
.location
== VARYING_SLOT_PRIMITIVE_ID
) {
300 /* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
301 var
->data
.location
= SYSTEM_VALUE_PRIMITIVE_ID
;
302 var
->data
.mode
= nir_var_system_value
;
304 var
->data
.mode
= nir_var_shader_in
;
308 case ir_var_shader_out
:
309 var
->data
.mode
= nir_var_shader_out
;
313 var
->data
.mode
= nir_var_uniform
;
316 case ir_var_shader_storage
:
317 var
->data
.mode
= nir_var_shader_storage
;
320 case ir_var_system_value
:
321 var
->data
.mode
= nir_var_system_value
;
325 unreachable("not reached");
328 var
->data
.interpolation
= ir
->data
.interpolation
;
329 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
330 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
331 var
->data
.explicit_location
= ir
->data
.explicit_location
;
332 var
->data
.explicit_index
= ir
->data
.explicit_index
;
333 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
334 var
->data
.has_initializer
= ir
->data
.has_initializer
;
335 var
->data
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
336 var
->data
.location_frac
= ir
->data
.location_frac
;
337 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
338 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
340 switch (ir
->data
.depth_layout
) {
341 case ir_depth_layout_none
:
342 var
->data
.depth_layout
= nir_depth_layout_none
;
344 case ir_depth_layout_any
:
345 var
->data
.depth_layout
= nir_depth_layout_any
;
347 case ir_depth_layout_greater
:
348 var
->data
.depth_layout
= nir_depth_layout_greater
;
350 case ir_depth_layout_less
:
351 var
->data
.depth_layout
= nir_depth_layout_less
;
353 case ir_depth_layout_unchanged
:
354 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
357 unreachable("not reached");
360 var
->data
.index
= ir
->data
.index
;
361 var
->data
.binding
= ir
->data
.binding
;
362 /* XXX Get rid of buffer_index */
363 var
->data
.atomic
.buffer_index
= ir
->data
.binding
;
364 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
365 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
366 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
367 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
368 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
369 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
370 var
->data
.image
.format
= ir
->data
.image_format
;
371 var
->data
.max_array_access
= ir
->data
.max_array_access
;
373 var
->num_state_slots
= ir
->get_num_state_slots();
374 if (var
->num_state_slots
> 0) {
375 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
376 var
->num_state_slots
);
378 ir_state_slot
*state_slots
= ir
->get_state_slots();
379 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
380 for (unsigned j
= 0; j
< 5; j
++)
381 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
382 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
385 var
->state_slots
= NULL
;
388 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
390 var
->interface_type
= ir
->get_interface_type();
392 switch (var
->data
.mode
) {
394 exec_list_push_tail(&impl
->locals
, &var
->node
);
398 exec_list_push_tail(&shader
->globals
, &var
->node
);
401 case nir_var_shader_in
:
402 exec_list_push_tail(&shader
->inputs
, &var
->node
);
405 case nir_var_shader_out
:
406 exec_list_push_tail(&shader
->outputs
, &var
->node
);
409 case nir_var_uniform
:
410 case nir_var_shader_storage
:
411 exec_list_push_tail(&shader
->uniforms
, &var
->node
);
414 case nir_var_system_value
:
415 exec_list_push_tail(&shader
->system_values
, &var
->node
);
419 unreachable("not reached");
422 _mesa_hash_table_insert(var_table
, ir
, var
);
427 nir_function_visitor::visit_enter(ir_function
*ir
)
429 visitor
->create_function(ir
);
430 return visit_continue_with_parent
;
435 nir_visitor::create_function(ir_function
*ir
)
437 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
438 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
439 create_overload(sig
, func
);
446 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
448 if (ir
->is_intrinsic
)
451 nir_function_overload
*overload
= nir_function_overload_create(function
);
453 unsigned num_params
= ir
->parameters
.length();
454 overload
->num_params
= num_params
;
455 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
458 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
459 switch (param
->data
.mode
) {
460 case ir_var_function_in
:
461 overload
->params
[i
].param_type
= nir_parameter_in
;
464 case ir_var_function_out
:
465 overload
->params
[i
].param_type
= nir_parameter_out
;
468 case ir_var_function_inout
:
469 overload
->params
[i
].param_type
= nir_parameter_inout
;
473 unreachable("not reached");
476 overload
->params
[i
].type
= param
->type
;
480 overload
->return_type
= ir
->return_type
;
482 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
486 nir_visitor::visit(ir_function
*ir
)
488 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
493 nir_visitor::visit(ir_function_signature
*ir
)
495 if (ir
->is_intrinsic
)
498 struct hash_entry
*entry
=
499 _mesa_hash_table_search(this->overload_table
, ir
);
502 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
504 if (ir
->is_defined
) {
505 nir_function_impl
*impl
= nir_function_impl_create(overload
);
508 unsigned num_params
= overload
->num_params
;
509 impl
->num_params
= num_params
;
510 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
512 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
514 impl
->params
[i
] = this->var
;
518 if (overload
->return_type
== glsl_type::void_type
) {
519 impl
->return_var
= NULL
;
521 impl
->return_var
= ralloc(this->shader
, nir_variable
);
522 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
524 impl
->return_var
->type
= overload
->return_type
;
527 this->is_global
= false;
529 this->cf_node_list
= &impl
->body
;
530 visit_exec_list(&ir
->body
, this);
532 this->is_global
= true;
534 overload
->impl
= NULL
;
539 nir_visitor::visit(ir_loop
*ir
)
541 exec_list
*old_list
= this->cf_node_list
;
543 nir_loop
*loop
= nir_loop_create(this->shader
);
544 nir_cf_node_insert_end(old_list
, &loop
->cf_node
);
545 this->cf_node_list
= &loop
->body
;
546 visit_exec_list(&ir
->body_instructions
, this);
548 this->cf_node_list
= old_list
;
552 nir_visitor::visit(ir_if
*ir
)
554 nir_src condition
= evaluate_rvalue(ir
->condition
);
556 exec_list
*old_list
= this->cf_node_list
;
558 nir_if
*if_stmt
= nir_if_create(this->shader
);
559 if_stmt
->condition
= condition
;
560 nir_cf_node_insert_end(old_list
, &if_stmt
->cf_node
);
562 this->cf_node_list
= &if_stmt
->then_list
;
563 visit_exec_list(&ir
->then_instructions
, this);
565 this->cf_node_list
= &if_stmt
->else_list
;
566 visit_exec_list(&ir
->else_instructions
, this);
568 this->cf_node_list
= old_list
;
572 nir_visitor::visit(ir_discard
*ir
)
575 * discards aren't treated as control flow, because before we lower them
576 * they can appear anywhere in the shader and the stuff after them may still
577 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
578 * discards will be immediately followed by a return.
581 nir_intrinsic_instr
*discard
;
583 discard
= nir_intrinsic_instr_create(this->shader
,
584 nir_intrinsic_discard_if
);
585 discard
->src
[0] = evaluate_rvalue(ir
->condition
);
587 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
589 nir_instr_insert_after_cf_list(this->cf_node_list
, &discard
->instr
);
593 nir_visitor::visit(ir_emit_vertex
*ir
)
595 nir_intrinsic_instr
*instr
=
596 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
597 instr
->const_index
[0] = ir
->stream_id();
598 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
602 nir_visitor::visit(ir_end_primitive
*ir
)
604 nir_intrinsic_instr
*instr
=
605 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
606 instr
->const_index
[0] = ir
->stream_id();
607 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
611 nir_visitor::visit(ir_loop_jump
*ir
)
615 case ir_loop_jump::jump_break
:
616 type
= nir_jump_break
;
618 case ir_loop_jump::jump_continue
:
619 type
= nir_jump_continue
;
622 unreachable("not reached");
625 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
626 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
630 nir_visitor::visit(ir_return
*ir
)
632 if (ir
->value
!= NULL
) {
633 nir_intrinsic_instr
*copy
=
634 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
636 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
637 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
640 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
641 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
645 nir_visitor::visit(ir_call
*ir
)
647 if (ir
->callee
->is_intrinsic
) {
649 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
650 op
= nir_intrinsic_atomic_counter_read_var
;
651 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
652 op
= nir_intrinsic_atomic_counter_inc_var
;
653 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
654 op
= nir_intrinsic_atomic_counter_dec_var
;
655 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
656 op
= nir_intrinsic_image_load
;
657 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
658 op
= nir_intrinsic_image_store
;
659 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
660 op
= nir_intrinsic_image_atomic_add
;
661 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
662 op
= nir_intrinsic_image_atomic_min
;
663 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
664 op
= nir_intrinsic_image_atomic_max
;
665 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
666 op
= nir_intrinsic_image_atomic_and
;
667 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
668 op
= nir_intrinsic_image_atomic_or
;
669 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
670 op
= nir_intrinsic_image_atomic_xor
;
671 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
672 op
= nir_intrinsic_image_atomic_exchange
;
673 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
674 op
= nir_intrinsic_image_atomic_comp_swap
;
675 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
676 op
= nir_intrinsic_memory_barrier
;
677 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_size") == 0) {
678 op
= nir_intrinsic_image_size
;
679 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_samples") == 0) {
680 op
= nir_intrinsic_image_samples
;
681 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_ssbo") == 0) {
682 op
= nir_intrinsic_store_ssbo
;
683 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_ssbo") == 0) {
684 op
= nir_intrinsic_load_ssbo
;
685 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_add_internal") == 0) {
686 op
= nir_intrinsic_ssbo_atomic_add
;
687 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_and_internal") == 0) {
688 op
= nir_intrinsic_ssbo_atomic_and
;
689 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_or_internal") == 0) {
690 op
= nir_intrinsic_ssbo_atomic_or
;
691 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_xor_internal") == 0) {
692 op
= nir_intrinsic_ssbo_atomic_xor
;
693 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_min_internal") == 0) {
694 assert(ir
->return_deref
);
695 if (ir
->return_deref
->type
== glsl_type::int_type
)
696 op
= nir_intrinsic_ssbo_atomic_imin
;
697 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
698 op
= nir_intrinsic_ssbo_atomic_umin
;
700 unreachable("Invalid type");
701 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_max_internal") == 0) {
702 assert(ir
->return_deref
);
703 if (ir
->return_deref
->type
== glsl_type::int_type
)
704 op
= nir_intrinsic_ssbo_atomic_imax
;
705 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
706 op
= nir_intrinsic_ssbo_atomic_umax
;
708 unreachable("Invalid type");
709 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_exchange_internal") == 0) {
710 op
= nir_intrinsic_ssbo_atomic_exchange
;
711 } else if (strcmp(ir
->callee_name(), "__intrinsic_ssbo_atomic_comp_swap_internal") == 0) {
712 op
= nir_intrinsic_ssbo_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, NULL
);
728 nir_instr_insert_after_cf_list(this->cf_node_list
, &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);
745 nir_instr_insert_after_cf_list(this->cf_node_list
,
746 &instr_undef
->instr
);
748 /* Set the image variable dereference. */
749 exec_node
*param
= ir
->actual_parameters
.get_head();
750 ir_dereference
*image
= (ir_dereference
*)param
;
751 const glsl_type
*type
=
752 image
->variable_referenced()->type
->without_array();
754 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
755 param
= param
->get_next();
757 /* Set the intrinsic destination. */
758 if (ir
->return_deref
) {
759 const nir_intrinsic_info
*info
=
760 &nir_intrinsic_infos
[instr
->intrinsic
];
761 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
762 info
->dest_components
, NULL
);
765 if (op
== nir_intrinsic_image_size
||
766 op
== nir_intrinsic_image_samples
) {
767 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
771 /* Set the address argument, extending the coordinate vector to four
774 const nir_src src_addr
= evaluate_rvalue((ir_dereference
*)param
);
775 nir_alu_instr
*instr_addr
= nir_alu_instr_create(shader
, nir_op_vec4
);
776 nir_ssa_dest_init(&instr_addr
->instr
, &instr_addr
->dest
.dest
, 4, NULL
);
778 for (int i
= 0; i
< 4; i
++) {
779 if (i
< type
->coordinate_components()) {
780 instr_addr
->src
[i
].src
= src_addr
;
781 instr_addr
->src
[i
].swizzle
[0] = i
;
783 instr_addr
->src
[i
].src
= nir_src_for_ssa(&instr_undef
->def
);
787 nir_instr_insert_after_cf_list(cf_node_list
, &instr_addr
->instr
);
788 instr
->src
[0] = nir_src_for_ssa(&instr_addr
->dest
.dest
.ssa
);
789 param
= param
->get_next();
791 /* Set the sample argument, which is undefined for single-sample
794 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
795 instr
->src
[1] = evaluate_rvalue((ir_dereference
*)param
);
796 param
= param
->get_next();
798 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
801 /* Set the intrinsic parameters. */
802 if (!param
->is_tail_sentinel()) {
803 instr
->src
[2] = evaluate_rvalue((ir_dereference
*)param
);
804 param
= param
->get_next();
807 if (!param
->is_tail_sentinel()) {
808 instr
->src
[3] = evaluate_rvalue((ir_dereference
*)param
);
809 param
= param
->get_next();
811 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
814 case nir_intrinsic_memory_barrier
:
815 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
817 case nir_intrinsic_store_ssbo
: {
818 exec_node
*param
= ir
->actual_parameters
.get_head();
819 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
821 param
= param
->get_next();
822 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
824 param
= param
->get_next();
825 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
827 param
= param
->get_next();
828 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
831 /* Check if we need the indirect version */
832 ir_constant
*const_offset
= offset
->as_constant();
834 op
= nir_intrinsic_store_ssbo_indirect
;
836 instr
= nir_intrinsic_instr_create(shader
, op
);
837 instr
->src
[2] = evaluate_rvalue(offset
);
838 instr
->const_index
[0] = 0;
840 instr
->const_index
[0] = const_offset
->value
.u
[0];
843 instr
->const_index
[1] = write_mask
->value
.u
[0];
845 instr
->src
[0] = evaluate_rvalue(val
);
846 instr
->num_components
= val
->type
->vector_elements
;
848 instr
->src
[1] = evaluate_rvalue(block
);
849 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
852 case nir_intrinsic_load_ssbo
: {
853 exec_node
*param
= ir
->actual_parameters
.get_head();
854 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
856 param
= param
->get_next();
857 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
859 /* Check if we need the indirect version */
860 ir_constant
*const_offset
= offset
->as_constant();
862 op
= nir_intrinsic_load_ssbo_indirect
;
864 instr
= nir_intrinsic_instr_create(shader
, op
);
865 instr
->src
[1] = evaluate_rvalue(offset
);
866 instr
->const_index
[0] = 0;
869 instr
->const_index
[0] = const_offset
->value
.u
[0];
872 instr
->src
[0] = evaluate_rvalue(block
);
874 const glsl_type
*type
= ir
->return_deref
->var
->type
;
875 instr
->num_components
= type
->vector_elements
;
877 /* Setup destination register */
878 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
879 type
->vector_elements
, NULL
);
881 /* Insert the created nir instruction now since in the case of boolean
882 * result we will need to emit another instruction after it
884 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
887 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
888 * consider a true boolean to be ~0. Fix this up with a != 0
891 if (type
->base_type
== GLSL_TYPE_BOOL
) {
892 nir_load_const_instr
*const_zero
=
893 nir_load_const_instr_create(shader
, 1);
894 const_zero
->value
.u
[0] = 0;
895 nir_instr_insert_after_cf_list(this->cf_node_list
,
898 nir_alu_instr
*load_ssbo_compare
=
899 nir_alu_instr_create(shader
, nir_op_ine
);
900 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
901 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
902 load_ssbo_compare
->src
[1].src
.is_ssa
= true;
903 load_ssbo_compare
->src
[1].src
.ssa
= &const_zero
->def
;
904 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
905 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
906 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
907 &load_ssbo_compare
->dest
.dest
,
908 type
->vector_elements
, NULL
);
909 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
910 nir_instr_insert_after_cf_list(this->cf_node_list
,
911 &load_ssbo_compare
->instr
);
912 dest
= &load_ssbo_compare
->dest
.dest
;
916 case nir_intrinsic_ssbo_atomic_add
:
917 case nir_intrinsic_ssbo_atomic_imin
:
918 case nir_intrinsic_ssbo_atomic_umin
:
919 case nir_intrinsic_ssbo_atomic_imax
:
920 case nir_intrinsic_ssbo_atomic_umax
:
921 case nir_intrinsic_ssbo_atomic_and
:
922 case nir_intrinsic_ssbo_atomic_or
:
923 case nir_intrinsic_ssbo_atomic_xor
:
924 case nir_intrinsic_ssbo_atomic_exchange
:
925 case nir_intrinsic_ssbo_atomic_comp_swap
: {
926 int param_count
= ir
->actual_parameters
.length();
927 assert(param_count
== 3 || param_count
== 4);
930 exec_node
*param
= ir
->actual_parameters
.get_head();
931 ir_instruction
*inst
= (ir_instruction
*) param
;
932 instr
->src
[0] = evaluate_rvalue(inst
->as_rvalue());
935 param
= param
->get_next();
936 inst
= (ir_instruction
*) param
;
937 instr
->src
[1] = evaluate_rvalue(inst
->as_rvalue());
939 /* data1 parameter (this is always present) */
940 param
= param
->get_next();
941 inst
= (ir_instruction
*) param
;
942 instr
->src
[2] = evaluate_rvalue(inst
->as_rvalue());
944 /* data2 parameter (only with atomic_comp_swap) */
945 if (param_count
== 4) {
946 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
947 param
= param
->get_next();
948 inst
= (ir_instruction
*) param
;
949 instr
->src
[3] = evaluate_rvalue(inst
->as_rvalue());
953 assert(ir
->return_deref
);
954 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
955 ir
->return_deref
->type
->vector_elements
, NULL
);
956 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
960 unreachable("not reached");
963 if (ir
->return_deref
) {
964 nir_intrinsic_instr
*store_instr
=
965 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
966 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
968 store_instr
->variables
[0] =
969 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
970 store_instr
->src
[0] = nir_src_for_ssa(&dest
->ssa
);
972 nir_instr_insert_after_cf_list(this->cf_node_list
,
973 &store_instr
->instr
);
979 struct hash_entry
*entry
=
980 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
982 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
984 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
987 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
988 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
992 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
993 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);
997 nir_visitor::visit(ir_assignment
*ir
)
999 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
1001 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
1002 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
1003 /* We're doing a plain-as-can-be copy, so emit a copy_var */
1004 nir_intrinsic_instr
*copy
=
1005 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
1007 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
1008 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
1010 if (ir
->condition
) {
1011 nir_if
*if_stmt
= nir_if_create(this->shader
);
1012 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
1013 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
1014 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
1016 nir_instr_insert_after_cf_list(this->cf_node_list
, ©
->instr
);
1021 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
1023 ir
->lhs
->accept(this);
1024 nir_deref_var
*lhs_deref
= this->deref_head
;
1025 nir_src src
= evaluate_rvalue(ir
->rhs
);
1027 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
1029 * We have no good way to update only part of a variable, so just load
1030 * the LHS and do a vec operation to combine the old with the new, and
1032 * back into the LHS. Copy propagation should get rid of the mess.
1035 nir_intrinsic_instr
*load
=
1036 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1037 load
->num_components
= ir
->lhs
->type
->vector_elements
;
1038 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
1039 load
->variables
[0] = lhs_deref
;
1040 ralloc_steal(load
, load
->variables
[0]);
1041 nir_instr_insert_after_cf_list(this->cf_node_list
, &load
->instr
);
1044 switch (ir
->lhs
->type
->vector_elements
) {
1045 case 1: vec_op
= nir_op_imov
; break;
1046 case 2: vec_op
= nir_op_vec2
; break;
1047 case 3: vec_op
= nir_op_vec3
; break;
1048 case 4: vec_op
= nir_op_vec4
; break;
1049 default: unreachable("Invalid number of components"); break;
1051 nir_alu_instr
*vec
= nir_alu_instr_create(this->shader
, vec_op
);
1052 nir_ssa_dest_init(&vec
->instr
, &vec
->dest
.dest
, num_components
, NULL
);
1053 vec
->dest
.write_mask
= (1 << num_components
) - 1;
1055 unsigned component
= 0;
1056 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
1057 if (ir
->write_mask
& (1 << i
)) {
1058 vec
->src
[i
].src
= src
;
1060 /* GLSL IR will give us the input to the write-masked assignment
1061 * in a single packed vector. So, for example, if the
1062 * writemask is xzw, then we have to swizzle x -> x, y -> z,
1063 * and z -> w and get the y component from the load.
1065 vec
->src
[i
].swizzle
[0] = component
++;
1067 vec
->src
[i
].src
.is_ssa
= true;
1068 vec
->src
[i
].src
.ssa
= &load
->dest
.ssa
;
1069 vec
->src
[i
].swizzle
[0] = i
;
1073 nir_instr_insert_after_cf_list(this->cf_node_list
, &vec
->instr
);
1076 src
.ssa
= &vec
->dest
.dest
.ssa
;
1079 nir_intrinsic_instr
*store
=
1080 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1081 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1082 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
1083 store
->variables
[0] = nir_deref_as_var(store_deref
);
1084 store
->src
[0] = src
;
1086 if (ir
->condition
) {
1087 nir_if
*if_stmt
= nir_if_create(this->shader
);
1088 if_stmt
->condition
= evaluate_rvalue(ir
->condition
);
1089 nir_cf_node_insert_end(this->cf_node_list
, &if_stmt
->cf_node
);
1090 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
1092 nir_instr_insert_after_cf_list(this->cf_node_list
, &store
->instr
);
1097 * Given an instruction, returns a pointer to its destination or NULL if there
1098 * is no destination.
1100 * Note that this only handles instructions we generate at this level.
1103 get_instr_dest(nir_instr
*instr
)
1105 nir_alu_instr
*alu_instr
;
1106 nir_intrinsic_instr
*intrinsic_instr
;
1107 nir_tex_instr
*tex_instr
;
1109 switch (instr
->type
) {
1110 case nir_instr_type_alu
:
1111 alu_instr
= nir_instr_as_alu(instr
);
1112 return &alu_instr
->dest
.dest
;
1114 case nir_instr_type_intrinsic
:
1115 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1116 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1117 return &intrinsic_instr
->dest
;
1121 case nir_instr_type_tex
:
1122 tex_instr
= nir_instr_as_tex(instr
);
1123 return &tex_instr
->dest
;
1126 unreachable("not reached");
1133 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
1135 nir_dest
*dest
= get_instr_dest(instr
);
1138 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
1140 nir_instr_insert_after_cf_list(this->cf_node_list
, instr
);
1141 this->result
= instr
;
1145 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1148 if (ir
->as_dereference() || ir
->as_constant()) {
1150 * A dereference is being used on the right hand side, which means we
1151 * must emit a variable load.
1154 nir_intrinsic_instr
*load_instr
=
1155 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1156 load_instr
->num_components
= ir
->type
->vector_elements
;
1157 load_instr
->variables
[0] = this->deref_head
;
1158 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1159 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
1162 nir_dest
*dest
= get_instr_dest(this->result
);
1163 assert(dest
->is_ssa
);
1165 return nir_src_for_ssa(&dest
->ssa
);
1169 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src
*srcs
)
1171 nir_alu_instr
*instr
= nir_alu_instr_create(this->shader
, op
);
1172 for (unsigned i
= 0; i
< nir_op_infos
[op
].num_inputs
; i
++)
1173 instr
->src
[i
].src
= srcs
[i
];
1174 instr
->dest
.write_mask
= (1 << dest_size
) - 1;
1175 add_instr(&instr
->instr
, dest_size
);
1180 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
)
1182 assert(nir_op_infos
[op
].num_inputs
== 1);
1183 return emit(op
, dest_size
, &src1
);
1187 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
1190 assert(nir_op_infos
[op
].num_inputs
== 2);
1191 nir_src srcs
[] = { src1
, src2
};
1192 return emit(op
, dest_size
, srcs
);
1196 nir_visitor::emit(nir_op op
, unsigned dest_size
, nir_src src1
,
1197 nir_src src2
, nir_src src3
)
1199 assert(nir_op_infos
[op
].num_inputs
== 3);
1200 nir_src srcs
[] = { src1
, src2
, src3
};
1201 return emit(op
, dest_size
, srcs
);
1205 nir_visitor::visit(ir_expression
*ir
)
1207 /* Some special cases */
1208 switch (ir
->operation
) {
1209 case ir_binop_ubo_load
: {
1210 ir_constant
*const_index
= ir
->operands
[1]->as_constant();
1212 nir_intrinsic_op op
;
1214 op
= nir_intrinsic_load_ubo
;
1216 op
= nir_intrinsic_load_ubo_indirect
;
1218 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(this->shader
, op
);
1219 load
->num_components
= ir
->type
->vector_elements
;
1220 load
->const_index
[0] = const_index
? const_index
->value
.u
[0] : 0; /* base offset */
1221 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
1223 load
->src
[1] = evaluate_rvalue(ir
->operands
[1]);
1224 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1227 * In UBO's, a true boolean value is any non-zero value, but we consider
1228 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1231 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
) {
1232 nir_load_const_instr
*const_zero
= nir_load_const_instr_create(shader
, 1);
1233 const_zero
->value
.u
[0] = 0;
1234 nir_instr_insert_after_cf_list(this->cf_node_list
, &const_zero
->instr
);
1236 nir_alu_instr
*compare
= nir_alu_instr_create(shader
, nir_op_ine
);
1237 compare
->src
[0].src
.is_ssa
= true;
1238 compare
->src
[0].src
.ssa
= &load
->dest
.ssa
;
1239 compare
->src
[1].src
.is_ssa
= true;
1240 compare
->src
[1].src
.ssa
= &const_zero
->def
;
1241 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1242 compare
->src
[1].swizzle
[i
] = 0;
1243 compare
->dest
.write_mask
= (1 << ir
->type
->vector_elements
) - 1;
1245 add_instr(&compare
->instr
, ir
->type
->vector_elements
);
1251 case ir_unop_interpolate_at_centroid
:
1252 case ir_binop_interpolate_at_offset
:
1253 case ir_binop_interpolate_at_sample
: {
1254 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1255 ir_swizzle
*swizzle
= NULL
;
1257 /* the api does not allow a swizzle here, but the varying packing code
1258 * may have pushed one into here.
1260 swizzle
= ir
->operands
[0]->as_swizzle();
1262 deref
= swizzle
->val
->as_dereference();
1266 deref
->accept(this);
1268 nir_intrinsic_op op
;
1269 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1270 switch (ir
->operation
) {
1271 case ir_unop_interpolate_at_centroid
:
1272 op
= nir_intrinsic_interp_var_at_centroid
;
1274 case ir_binop_interpolate_at_offset
:
1275 op
= nir_intrinsic_interp_var_at_offset
;
1277 case ir_binop_interpolate_at_sample
:
1278 op
= nir_intrinsic_interp_var_at_sample
;
1281 unreachable("Invalid interpolation intrinsic");
1284 /* This case can happen if the vertex shader does not write the
1285 * given varying. In this case, the linker will lower it to a
1286 * global variable. Since interpolating a variable makes no
1287 * sense, we'll just turn it into a load which will probably
1288 * eventually end up as an SSA definition.
1290 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1291 op
= nir_intrinsic_load_var
;
1294 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1295 intrin
->num_components
= deref
->type
->vector_elements
;
1296 intrin
->variables
[0] = this->deref_head
;
1297 ralloc_steal(intrin
, intrin
->variables
[0]);
1299 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1300 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1301 intrin
->src
[0] = evaluate_rvalue(ir
->operands
[1]);
1303 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
1306 nir_alu_instr
*mov
= nir_alu_instr_create(shader
, nir_op_imov
);
1307 mov
->dest
.write_mask
= (1 << swizzle
->type
->vector_elements
) - 1;
1308 mov
->src
[0].src
.is_ssa
= true;
1309 mov
->src
[0].src
.ssa
= &intrin
->dest
.ssa
;
1311 mov
->src
[0].swizzle
[0] = swizzle
->mask
.x
;
1312 mov
->src
[0].swizzle
[1] = swizzle
->mask
.y
;
1313 mov
->src
[0].swizzle
[2] = swizzle
->mask
.z
;
1314 mov
->src
[0].swizzle
[3] = swizzle
->mask
.w
;
1315 for (unsigned i
= deref
->type
->vector_elements
; i
< 4; i
++)
1316 mov
->src
[0].swizzle
[i
] = 0;
1318 add_instr(&mov
->instr
, swizzle
->type
->vector_elements
);
1329 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1330 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1332 glsl_base_type types
[4];
1333 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1335 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1337 types
[i
] = GLSL_TYPE_FLOAT
;
1339 glsl_base_type out_type
;
1341 out_type
= ir
->type
->base_type
;
1343 out_type
= GLSL_TYPE_FLOAT
;
1345 unsigned dest_size
= ir
->type
->vector_elements
;
1347 nir_alu_instr
*instr
;
1350 switch (ir
->operation
) {
1351 case ir_unop_bit_not
: emit(nir_op_inot
, dest_size
, srcs
); break;
1352 case ir_unop_logic_not
:
1353 emit(supports_ints
? nir_op_inot
: nir_op_fnot
, dest_size
, srcs
);
1356 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fneg
: nir_op_ineg
,
1360 instr
= emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fabs
: nir_op_iabs
,
1363 case ir_unop_saturate
:
1364 assert(types
[0] == GLSL_TYPE_FLOAT
);
1365 instr
= emit(nir_op_fsat
, dest_size
, srcs
);
1368 emit(types
[0] == GLSL_TYPE_FLOAT
? nir_op_fsign
: nir_op_isign
,
1371 case ir_unop_rcp
: emit(nir_op_frcp
, dest_size
, srcs
); break;
1372 case ir_unop_rsq
: emit(nir_op_frsq
, dest_size
, srcs
); break;
1373 case ir_unop_sqrt
: emit(nir_op_fsqrt
, dest_size
, srcs
); break;
1374 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1375 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1376 case ir_unop_exp2
: emit(nir_op_fexp2
, dest_size
, srcs
); break;
1377 case ir_unop_log2
: emit(nir_op_flog2
, dest_size
, srcs
); break;
1379 emit(supports_ints
? nir_op_i2f
: nir_op_fmov
, dest_size
, srcs
);
1382 emit(supports_ints
? nir_op_u2f
: nir_op_fmov
, dest_size
, srcs
);
1385 emit(supports_ints
? nir_op_b2f
: nir_op_fmov
, dest_size
, srcs
);
1387 case ir_unop_f2i
: emit(nir_op_f2i
, dest_size
, srcs
); break;
1388 case ir_unop_f2u
: emit(nir_op_f2u
, dest_size
, srcs
); break;
1389 case ir_unop_f2b
: emit(nir_op_f2b
, dest_size
, srcs
); break;
1390 case ir_unop_i2b
: emit(nir_op_i2b
, dest_size
, srcs
); break;
1391 case ir_unop_b2i
: emit(nir_op_b2i
, dest_size
, srcs
); break;
1394 case ir_unop_bitcast_i2f
:
1395 case ir_unop_bitcast_f2i
:
1396 case ir_unop_bitcast_u2f
:
1397 case ir_unop_bitcast_f2u
:
1398 case ir_unop_subroutine_to_int
:
1400 emit(nir_op_imov
, dest_size
, srcs
);
1403 switch (ir
->operands
[0]->type
->vector_elements
) {
1405 emit(supports_ints
? nir_op_bany2
: nir_op_fany2
,
1409 emit(supports_ints
? nir_op_bany3
: nir_op_fany3
,
1413 emit(supports_ints
? nir_op_bany4
: nir_op_fany4
,
1417 unreachable("not reached");
1420 case ir_unop_trunc
: emit(nir_op_ftrunc
, dest_size
, srcs
); break;
1421 case ir_unop_ceil
: emit(nir_op_fceil
, dest_size
, srcs
); break;
1422 case ir_unop_floor
: emit(nir_op_ffloor
, dest_size
, srcs
); break;
1423 case ir_unop_fract
: emit(nir_op_ffract
, dest_size
, srcs
); break;
1424 case ir_unop_round_even
: emit(nir_op_fround_even
, dest_size
, srcs
); break;
1425 case ir_unop_sin
: emit(nir_op_fsin
, dest_size
, srcs
); break;
1426 case ir_unop_cos
: emit(nir_op_fcos
, dest_size
, srcs
); break;
1427 case ir_unop_dFdx
: emit(nir_op_fddx
, dest_size
, srcs
); break;
1428 case ir_unop_dFdy
: emit(nir_op_fddy
, dest_size
, srcs
); break;
1429 case ir_unop_dFdx_fine
: emit(nir_op_fddx_fine
, dest_size
, srcs
); break;
1430 case ir_unop_dFdy_fine
: emit(nir_op_fddy_fine
, dest_size
, srcs
); break;
1431 case ir_unop_dFdx_coarse
: emit(nir_op_fddx_coarse
, dest_size
, srcs
); break;
1432 case ir_unop_dFdy_coarse
: emit(nir_op_fddy_coarse
, dest_size
, srcs
); break;
1433 case ir_unop_pack_snorm_2x16
:
1434 emit(nir_op_pack_snorm_2x16
, dest_size
, srcs
);
1436 case ir_unop_pack_snorm_4x8
:
1437 emit(nir_op_pack_snorm_4x8
, dest_size
, srcs
);
1439 case ir_unop_pack_unorm_2x16
:
1440 emit(nir_op_pack_unorm_2x16
, dest_size
, srcs
);
1442 case ir_unop_pack_unorm_4x8
:
1443 emit(nir_op_pack_unorm_4x8
, dest_size
, srcs
);
1445 case ir_unop_pack_half_2x16
:
1446 emit(nir_op_pack_half_2x16
, dest_size
, srcs
);
1448 case ir_unop_unpack_snorm_2x16
:
1449 emit(nir_op_unpack_snorm_2x16
, dest_size
, srcs
);
1451 case ir_unop_unpack_snorm_4x8
:
1452 emit(nir_op_unpack_snorm_4x8
, dest_size
, srcs
);
1454 case ir_unop_unpack_unorm_2x16
:
1455 emit(nir_op_unpack_unorm_2x16
, dest_size
, srcs
);
1457 case ir_unop_unpack_unorm_4x8
:
1458 emit(nir_op_unpack_unorm_4x8
, dest_size
, srcs
);
1460 case ir_unop_unpack_half_2x16
:
1461 emit(nir_op_unpack_half_2x16
, dest_size
, srcs
);
1463 case ir_unop_unpack_half_2x16_split_x
:
1464 emit(nir_op_unpack_half_2x16_split_x
, dest_size
, srcs
);
1466 case ir_unop_unpack_half_2x16_split_y
:
1467 emit(nir_op_unpack_half_2x16_split_y
, dest_size
, srcs
);
1469 case ir_unop_bitfield_reverse
:
1470 emit(nir_op_bitfield_reverse
, dest_size
, srcs
);
1472 case ir_unop_bit_count
:
1473 emit(nir_op_bit_count
, dest_size
, srcs
);
1475 case ir_unop_find_msb
:
1477 case GLSL_TYPE_UINT
:
1478 emit(nir_op_ufind_msb
, dest_size
, srcs
);
1481 emit(nir_op_ifind_msb
, dest_size
, srcs
);
1484 unreachable("Invalid type for findMSB()");
1487 case ir_unop_find_lsb
:
1488 emit(nir_op_find_lsb
, dest_size
, srcs
);
1492 switch (ir
->type
->vector_elements
) {
1494 switch (ir
->operands
[0]->type
->vector_elements
) {
1495 case 1: emit(nir_op_fnoise1_1
, dest_size
, srcs
); break;
1496 case 2: emit(nir_op_fnoise1_2
, dest_size
, srcs
); break;
1497 case 3: emit(nir_op_fnoise1_3
, dest_size
, srcs
); break;
1498 case 4: emit(nir_op_fnoise1_4
, dest_size
, srcs
); break;
1499 default: unreachable("not reached");
1503 switch (ir
->operands
[0]->type
->vector_elements
) {
1504 case 1: emit(nir_op_fnoise2_1
, dest_size
, srcs
); break;
1505 case 2: emit(nir_op_fnoise2_2
, dest_size
, srcs
); break;
1506 case 3: emit(nir_op_fnoise2_3
, dest_size
, srcs
); break;
1507 case 4: emit(nir_op_fnoise2_4
, dest_size
, srcs
); break;
1508 default: unreachable("not reached");
1512 switch (ir
->operands
[0]->type
->vector_elements
) {
1513 case 1: emit(nir_op_fnoise3_1
, dest_size
, srcs
); break;
1514 case 2: emit(nir_op_fnoise3_2
, dest_size
, srcs
); break;
1515 case 3: emit(nir_op_fnoise3_3
, dest_size
, srcs
); break;
1516 case 4: emit(nir_op_fnoise3_4
, dest_size
, srcs
); break;
1517 default: unreachable("not reached");
1521 switch (ir
->operands
[0]->type
->vector_elements
) {
1522 case 1: emit(nir_op_fnoise4_1
, dest_size
, srcs
); break;
1523 case 2: emit(nir_op_fnoise4_2
, dest_size
, srcs
); break;
1524 case 3: emit(nir_op_fnoise4_3
, dest_size
, srcs
); break;
1525 case 4: emit(nir_op_fnoise4_4
, dest_size
, srcs
); break;
1526 default: unreachable("not reached");
1530 unreachable("not reached");
1533 case ir_unop_get_buffer_size
: {
1534 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1536 nir_intrinsic_get_buffer_size
);
1537 load
->num_components
= ir
->type
->vector_elements
;
1538 load
->src
[0] = evaluate_rvalue(ir
->operands
[0]);
1539 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1551 case ir_binop_bit_and
:
1552 case ir_binop_bit_or
:
1553 case ir_binop_bit_xor
:
1554 case ir_binop_logic_and
:
1555 case ir_binop_logic_or
:
1556 case ir_binop_logic_xor
:
1557 case ir_binop_lshift
:
1558 case ir_binop_rshift
:
1559 switch (ir
->operation
) {
1561 if (out_type
== GLSL_TYPE_FLOAT
)
1567 if (out_type
== GLSL_TYPE_FLOAT
)
1573 if (out_type
== GLSL_TYPE_FLOAT
)
1579 if (out_type
== GLSL_TYPE_FLOAT
)
1581 else if (out_type
== GLSL_TYPE_INT
)
1587 if (out_type
== GLSL_TYPE_FLOAT
)
1593 if (out_type
== GLSL_TYPE_FLOAT
)
1595 else if (out_type
== GLSL_TYPE_INT
)
1601 if (out_type
== GLSL_TYPE_FLOAT
)
1603 else if (out_type
== GLSL_TYPE_INT
)
1608 case ir_binop_bit_and
:
1611 case ir_binop_bit_or
:
1614 case ir_binop_bit_xor
:
1617 case ir_binop_logic_and
:
1623 case ir_binop_logic_or
:
1629 case ir_binop_logic_xor
:
1635 case ir_binop_lshift
:
1638 case ir_binop_rshift
:
1639 if (out_type
== GLSL_TYPE_INT
)
1649 unreachable("not reached");
1652 instr
= emit(op
, dest_size
, srcs
);
1654 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1655 ir
->operands
[1]->type
->vector_elements
== 1) {
1656 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1658 instr
->src
[1].swizzle
[i
] = 0;
1662 if (ir
->operands
[1]->type
->vector_elements
!= 1 &&
1663 ir
->operands
[0]->type
->vector_elements
== 1) {
1664 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
;
1666 instr
->src
[0].swizzle
[i
] = 0;
1671 case ir_binop_imul_high
:
1672 emit(out_type
== GLSL_TYPE_UINT
? nir_op_umul_high
: nir_op_imul_high
,
1675 case ir_binop_carry
: emit(nir_op_uadd_carry
, dest_size
, srcs
); break;
1676 case ir_binop_borrow
: emit(nir_op_usub_borrow
, dest_size
, srcs
); break;
1678 if (supports_ints
) {
1679 if (types
[0] == GLSL_TYPE_FLOAT
)
1680 emit(nir_op_flt
, dest_size
, srcs
);
1681 else if (types
[0] == GLSL_TYPE_INT
)
1682 emit(nir_op_ilt
, dest_size
, srcs
);
1684 emit(nir_op_ult
, dest_size
, srcs
);
1686 emit(nir_op_slt
, dest_size
, srcs
);
1689 case ir_binop_greater
:
1690 if (supports_ints
) {
1691 if (types
[0] == GLSL_TYPE_FLOAT
)
1692 emit(nir_op_flt
, dest_size
, srcs
[1], srcs
[0]);
1693 else if (types
[0] == GLSL_TYPE_INT
)
1694 emit(nir_op_ilt
, dest_size
, srcs
[1], srcs
[0]);
1696 emit(nir_op_ult
, dest_size
, srcs
[1], srcs
[0]);
1698 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1701 case ir_binop_lequal
:
1702 if (supports_ints
) {
1703 if (types
[0] == GLSL_TYPE_FLOAT
)
1704 emit(nir_op_fge
, dest_size
, srcs
[1], srcs
[0]);
1705 else if (types
[0] == GLSL_TYPE_INT
)
1706 emit(nir_op_ige
, dest_size
, srcs
[1], srcs
[0]);
1708 emit(nir_op_uge
, dest_size
, srcs
[1], srcs
[0]);
1710 emit(nir_op_slt
, dest_size
, srcs
[1], srcs
[0]);
1713 case ir_binop_gequal
:
1714 if (supports_ints
) {
1715 if (types
[0] == GLSL_TYPE_FLOAT
)
1716 emit(nir_op_fge
, dest_size
, srcs
);
1717 else if (types
[0] == GLSL_TYPE_INT
)
1718 emit(nir_op_ige
, dest_size
, srcs
);
1720 emit(nir_op_uge
, dest_size
, srcs
);
1722 emit(nir_op_slt
, dest_size
, srcs
);
1725 case ir_binop_equal
:
1726 if (supports_ints
) {
1727 if (types
[0] == GLSL_TYPE_FLOAT
)
1728 emit(nir_op_feq
, dest_size
, srcs
);
1730 emit(nir_op_ieq
, dest_size
, srcs
);
1732 emit(nir_op_seq
, dest_size
, srcs
);
1735 case ir_binop_nequal
:
1736 if (supports_ints
) {
1737 if (types
[0] == GLSL_TYPE_FLOAT
)
1738 emit(nir_op_fne
, dest_size
, srcs
);
1740 emit(nir_op_ine
, dest_size
, srcs
);
1742 emit(nir_op_sne
, dest_size
, srcs
);
1745 case ir_binop_all_equal
:
1746 if (supports_ints
) {
1747 if (types
[0] == GLSL_TYPE_FLOAT
) {
1748 switch (ir
->operands
[0]->type
->vector_elements
) {
1749 case 1: emit(nir_op_feq
, dest_size
, srcs
); break;
1750 case 2: emit(nir_op_ball_fequal2
, dest_size
, srcs
); break;
1751 case 3: emit(nir_op_ball_fequal3
, dest_size
, srcs
); break;
1752 case 4: emit(nir_op_ball_fequal4
, dest_size
, srcs
); break;
1754 unreachable("not reached");
1757 switch (ir
->operands
[0]->type
->vector_elements
) {
1758 case 1: emit(nir_op_ieq
, dest_size
, srcs
); break;
1759 case 2: emit(nir_op_ball_iequal2
, dest_size
, srcs
); break;
1760 case 3: emit(nir_op_ball_iequal3
, dest_size
, srcs
); break;
1761 case 4: emit(nir_op_ball_iequal4
, dest_size
, srcs
); break;
1763 unreachable("not reached");
1767 switch (ir
->operands
[0]->type
->vector_elements
) {
1768 case 1: emit(nir_op_seq
, dest_size
, srcs
); break;
1769 case 2: emit(nir_op_fall_equal2
, dest_size
, srcs
); break;
1770 case 3: emit(nir_op_fall_equal3
, dest_size
, srcs
); break;
1771 case 4: emit(nir_op_fall_equal4
, dest_size
, srcs
); break;
1773 unreachable("not reached");
1777 case ir_binop_any_nequal
:
1778 if (supports_ints
) {
1779 if (types
[0] == GLSL_TYPE_FLOAT
) {
1780 switch (ir
->operands
[0]->type
->vector_elements
) {
1781 case 1: emit(nir_op_fne
, dest_size
, srcs
); break;
1782 case 2: emit(nir_op_bany_fnequal2
, dest_size
, srcs
); break;
1783 case 3: emit(nir_op_bany_fnequal3
, dest_size
, srcs
); break;
1784 case 4: emit(nir_op_bany_fnequal4
, dest_size
, srcs
); break;
1786 unreachable("not reached");
1789 switch (ir
->operands
[0]->type
->vector_elements
) {
1790 case 1: emit(nir_op_ine
, dest_size
, srcs
); break;
1791 case 2: emit(nir_op_bany_inequal2
, dest_size
, srcs
); break;
1792 case 3: emit(nir_op_bany_inequal3
, dest_size
, srcs
); break;
1793 case 4: emit(nir_op_bany_inequal4
, dest_size
, srcs
); break;
1795 unreachable("not reached");
1799 switch (ir
->operands
[0]->type
->vector_elements
) {
1800 case 1: emit(nir_op_sne
, dest_size
, srcs
); break;
1801 case 2: emit(nir_op_fany_nequal2
, dest_size
, srcs
); break;
1802 case 3: emit(nir_op_fany_nequal3
, dest_size
, srcs
); break;
1803 case 4: emit(nir_op_fany_nequal4
, dest_size
, srcs
); break;
1805 unreachable("not reached");
1810 switch (ir
->operands
[0]->type
->vector_elements
) {
1811 case 2: emit(nir_op_fdot2
, dest_size
, srcs
); break;
1812 case 3: emit(nir_op_fdot3
, dest_size
, srcs
); break;
1813 case 4: emit(nir_op_fdot4
, dest_size
, srcs
); break;
1815 unreachable("not reached");
1819 case ir_binop_pack_half_2x16_split
:
1820 emit(nir_op_pack_half_2x16_split
, dest_size
, srcs
);
1822 case ir_binop_bfm
: emit(nir_op_bfm
, dest_size
, srcs
); break;
1823 case ir_binop_ldexp
: emit(nir_op_ldexp
, dest_size
, srcs
); break;
1824 case ir_triop_fma
: emit(nir_op_ffma
, dest_size
, srcs
); break;
1826 instr
= emit(nir_op_flrp
, dest_size
, srcs
);
1827 if (ir
->operands
[0]->type
->vector_elements
!= 1 &&
1828 ir
->operands
[2]->type
->vector_elements
== 1) {
1829 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
;
1831 instr
->src
[2].swizzle
[i
] = 0;
1837 emit(nir_op_bcsel
, dest_size
, srcs
);
1839 emit(nir_op_fcsel
, dest_size
, srcs
);
1842 instr
= emit(nir_op_bfi
, dest_size
, srcs
);
1843 for (unsigned i
= 0; i
< ir
->operands
[1]->type
->vector_elements
; i
++) {
1844 instr
->src
[0].swizzle
[i
] = 0;
1847 case ir_triop_bitfield_extract
:
1848 instr
= emit(out_type
== GLSL_TYPE_INT
? nir_op_ibitfield_extract
:
1849 nir_op_ubitfield_extract
, dest_size
, srcs
);
1850 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1851 instr
->src
[1].swizzle
[i
] = 0;
1852 instr
->src
[2].swizzle
[i
] = 0;
1855 case ir_quadop_bitfield_insert
:
1856 instr
= emit(nir_op_bitfield_insert
, dest_size
, srcs
);
1857 for (unsigned i
= 0; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
1858 instr
->src
[2].swizzle
[i
] = 0;
1859 instr
->src
[3].swizzle
[i
] = 0;
1862 case ir_quadop_vector
:
1863 switch (ir
->type
->vector_elements
) {
1864 case 2: emit(nir_op_vec2
, dest_size
, srcs
); break;
1865 case 3: emit(nir_op_vec3
, dest_size
, srcs
); break;
1866 case 4: emit(nir_op_vec4
, dest_size
, srcs
); break;
1867 default: unreachable("not reached");
1872 unreachable("not reached");
1877 nir_visitor::visit(ir_swizzle
*ir
)
1879 nir_alu_instr
*instr
= emit(supports_ints
? nir_op_imov
: nir_op_fmov
,
1880 ir
->type
->vector_elements
,
1881 evaluate_rvalue(ir
->val
));
1883 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1884 for (unsigned i
= 0; i
< ir
->type
->vector_elements
; i
++)
1885 instr
->src
[0].swizzle
[i
] = swizzle
[i
];
1889 nir_visitor::visit(ir_texture
*ir
)
1896 num_srcs
= 1; /* coordinate */
1901 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1902 num_srcs
= 2; /* coordinate, bias/lod */
1906 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1912 if (ir
->lod_info
.lod
!= NULL
)
1913 num_srcs
= 2; /* coordinate, lod */
1915 num_srcs
= 1; /* coordinate */
1919 op
= nir_texop_txf_ms
;
1920 num_srcs
= 2; /* coordinate, sample_index */
1925 if (ir
->lod_info
.lod
!= NULL
)
1926 num_srcs
= 1; /* lod */
1933 num_srcs
= 1; /* coordinate */
1938 num_srcs
= 1; /* coordinate */
1941 case ir_query_levels
:
1942 op
= nir_texop_query_levels
;
1946 case ir_texture_samples
:
1947 op
= nir_texop_texture_samples
;
1952 unreachable("not reached");
1955 if (ir
->projector
!= NULL
)
1957 if (ir
->shadow_comparitor
!= NULL
)
1959 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1962 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1965 instr
->sampler_dim
=
1966 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1967 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1968 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1969 if (instr
->is_shadow
)
1970 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1971 switch (ir
->type
->base_type
) {
1972 case GLSL_TYPE_FLOAT
:
1973 instr
->dest_type
= nir_type_float
;
1976 instr
->dest_type
= nir_type_int
;
1978 case GLSL_TYPE_UINT
:
1979 instr
->dest_type
= nir_type_unsigned
;
1982 unreachable("not reached");
1985 instr
->sampler
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1987 unsigned src_number
= 0;
1989 if (ir
->coordinate
!= NULL
) {
1990 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1991 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->coordinate
);
1992 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1996 if (ir
->projector
!= NULL
) {
1997 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->projector
);
1998 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
2002 if (ir
->shadow_comparitor
!= NULL
) {
2003 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->shadow_comparitor
);
2004 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
2008 if (ir
->offset
!= NULL
) {
2009 /* we don't support multiple offsets yet */
2010 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
2012 ir_constant
*const_offset
= ir
->offset
->as_constant();
2013 if (const_offset
!= NULL
) {
2014 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
2015 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
2017 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->offset
);
2018 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
2025 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.bias
);
2026 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
2033 if (ir
->lod_info
.lod
!= NULL
) {
2034 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.lod
);
2035 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
2041 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdx
);
2042 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
2044 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.grad
.dPdy
);
2045 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
2050 instr
->src
[src_number
].src
= evaluate_rvalue(ir
->lod_info
.sample_index
);
2051 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
2056 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
2063 assert(src_number
== num_srcs
);
2065 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
2069 nir_visitor::visit(ir_constant
*ir
)
2072 * We don't know if this variable is an an array or struct that gets
2073 * dereferenced, so do the safe thing an make it a variable with a
2074 * constant initializer and return a dereference.
2077 nir_variable
*var
= ralloc(this->shader
, nir_variable
);
2078 var
->name
= ralloc_strdup(var
, "const_temp");
2079 var
->type
= ir
->type
;
2080 var
->data
.mode
= nir_var_local
;
2081 var
->data
.read_only
= true;
2082 var
->constant_initializer
= constant_copy(ir
, var
);
2083 exec_list_push_tail(&this->impl
->locals
, &var
->node
);
2085 this->deref_head
= nir_deref_var_create(this->shader
, var
);
2086 this->deref_tail
= &this->deref_head
->deref
;
2090 nir_visitor::visit(ir_dereference_variable
*ir
)
2092 struct hash_entry
*entry
=
2093 _mesa_hash_table_search(this->var_table
, ir
->var
);
2095 nir_variable
*var
= (nir_variable
*) entry
->data
;
2097 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
2098 this->deref_head
= deref
;
2099 this->deref_tail
= &deref
->deref
;
2103 nir_visitor::visit(ir_dereference_record
*ir
)
2105 ir
->record
->accept(this);
2107 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
2108 assert(field_index
>= 0);
2110 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
2111 deref
->deref
.type
= ir
->type
;
2112 this->deref_tail
->child
= &deref
->deref
;
2113 this->deref_tail
= &deref
->deref
;
2117 nir_visitor::visit(ir_dereference_array
*ir
)
2119 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
2120 deref
->deref
.type
= ir
->type
;
2122 ir_constant
*const_index
= ir
->array_index
->as_constant();
2123 if (const_index
!= NULL
) {
2124 deref
->deref_array_type
= nir_deref_array_type_direct
;
2125 deref
->base_offset
= const_index
->value
.u
[0];
2127 deref
->deref_array_type
= nir_deref_array_type_indirect
;
2128 deref
->indirect
= evaluate_rvalue(ir
->array_index
);
2131 ir
->array
->accept(this);
2133 this->deref_tail
->child
= &deref
->deref
;
2134 ralloc_steal(this->deref_tail
, deref
);
2135 this->deref_tail
= &deref
->deref
;
2139 nir_visitor::visit(ir_barrier
*ir
)
2141 nir_intrinsic_instr
*instr
=
2142 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
2143 nir_instr_insert_after_cf_list(this->cf_node_list
, &instr
->instr
);