2 * Copyright © 2014 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Connor Abbott (cwabbott0@gmail.com)
28 #include "glsl_to_nir.h"
29 #include "nir_control_flow.h"
30 #include "nir_builder.h"
31 #include "compiler/glsl/ir_visitor.h"
32 #include "compiler/glsl/ir_hierarchical_visitor.h"
33 #include "compiler/glsl/ir.h"
34 #include "main/imports.h"
37 * pass to lower GLSL IR to NIR
39 * This will lower variable dereferences to loads/stores of corresponding
40 * variables in NIR - the variables will be converted to registers in a later
46 class nir_visitor
: public ir_visitor
49 nir_visitor(nir_shader
*shader
);
52 virtual void visit(ir_variable
*);
53 virtual void visit(ir_function
*);
54 virtual void visit(ir_function_signature
*);
55 virtual void visit(ir_loop
*);
56 virtual void visit(ir_if
*);
57 virtual void visit(ir_discard
*);
58 virtual void visit(ir_loop_jump
*);
59 virtual void visit(ir_return
*);
60 virtual void visit(ir_call
*);
61 virtual void visit(ir_assignment
*);
62 virtual void visit(ir_emit_vertex
*);
63 virtual void visit(ir_end_primitive
*);
64 virtual void visit(ir_expression
*);
65 virtual void visit(ir_swizzle
*);
66 virtual void visit(ir_texture
*);
67 virtual void visit(ir_constant
*);
68 virtual void visit(ir_dereference_variable
*);
69 virtual void visit(ir_dereference_record
*);
70 virtual void visit(ir_dereference_array
*);
71 virtual void visit(ir_barrier
*);
73 void create_function(ir_function_signature
*ir
);
76 void add_instr(nir_instr
*instr
, unsigned num_components
);
77 nir_ssa_def
*evaluate_rvalue(ir_rvalue
*ir
);
79 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
**srcs
);
80 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
);
81 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
,
83 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
,
84 nir_ssa_def
*src2
, nir_ssa_def
*src3
);
89 nir_function_impl
*impl
;
91 nir_ssa_def
*result
; /* result of the expression tree last visited */
93 nir_deref_var
*evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
);
95 /* the head of the dereference chain we're creating */
96 nir_deref_var
*deref_head
;
97 /* the tail of the dereference chain we're creating */
98 nir_deref
*deref_tail
;
100 nir_variable
*var
; /* variable created by ir_variable visitor */
102 /* whether the IR we're operating on is per-function or global */
105 /* map of ir_variable -> nir_variable */
106 struct hash_table
*var_table
;
108 /* map of ir_function_signature -> nir_function_overload */
109 struct hash_table
*overload_table
;
113 * This visitor runs before the main visitor, calling create_function() for
114 * each function so that the main visitor can resolve forward references in
118 class nir_function_visitor
: public ir_hierarchical_visitor
121 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
124 virtual ir_visitor_status
visit_enter(ir_function
*);
127 nir_visitor
*visitor
;
130 }; /* end of anonymous namespace */
133 glsl_to_nir(const struct gl_shader_program
*shader_prog
,
134 gl_shader_stage stage
,
135 const nir_shader_compiler_options
*options
)
137 struct gl_shader
*sh
= shader_prog
->_LinkedShaders
[stage
];
139 nir_shader
*shader
= nir_shader_create(NULL
, stage
, options
);
141 nir_visitor
v1(shader
);
142 nir_function_visitor
v2(&v1
);
144 visit_exec_list(sh
->ir
, &v1
);
146 nir_lower_outputs_to_temporaries(shader
);
148 shader
->info
.name
= ralloc_asprintf(shader
, "GLSL%d", shader_prog
->Name
);
149 if (shader_prog
->Label
)
150 shader
->info
.label
= ralloc_strdup(shader
, shader_prog
->Label
);
151 shader
->info
.num_textures
= _mesa_fls(sh
->Program
->SamplersUsed
);
152 shader
->info
.num_ubos
= sh
->NumUniformBlocks
;
153 shader
->info
.num_abos
= shader_prog
->NumAtomicBuffers
;
154 shader
->info
.num_ssbos
= sh
->NumShaderStorageBlocks
;
155 shader
->info
.num_images
= sh
->NumImages
;
156 shader
->info
.inputs_read
= sh
->Program
->InputsRead
;
157 shader
->info
.outputs_written
= sh
->Program
->OutputsWritten
;
158 shader
->info
.patch_inputs_read
= sh
->Program
->PatchInputsRead
;
159 shader
->info
.patch_outputs_written
= sh
->Program
->PatchOutputsWritten
;
160 shader
->info
.system_values_read
= sh
->Program
->SystemValuesRead
;
161 shader
->info
.uses_texture_gather
= sh
->Program
->UsesGather
;
162 shader
->info
.uses_clip_distance_out
=
163 sh
->Program
->ClipDistanceArraySize
!= 0;
164 shader
->info
.separate_shader
= shader_prog
->SeparateShader
;
165 shader
->info
.has_transform_feedback_varyings
=
166 shader_prog
->TransformFeedback
.NumVarying
> 0;
169 case MESA_SHADER_TESS_CTRL
:
170 shader
->info
.tcs
.vertices_out
= shader_prog
->TessCtrl
.VerticesOut
;
173 case MESA_SHADER_GEOMETRY
:
174 shader
->info
.gs
.vertices_in
= shader_prog
->Geom
.VerticesIn
;
175 shader
->info
.gs
.output_primitive
= sh
->Geom
.OutputType
;
176 shader
->info
.gs
.vertices_out
= sh
->Geom
.VerticesOut
;
177 shader
->info
.gs
.invocations
= sh
->Geom
.Invocations
;
178 shader
->info
.gs
.uses_end_primitive
= shader_prog
->Geom
.UsesEndPrimitive
;
179 shader
->info
.gs
.uses_streams
= shader_prog
->Geom
.UsesStreams
;
182 case MESA_SHADER_FRAGMENT
: {
183 struct gl_fragment_program
*fp
=
184 (struct gl_fragment_program
*)sh
->Program
;
186 shader
->info
.fs
.uses_discard
= fp
->UsesKill
;
187 shader
->info
.fs
.early_fragment_tests
= sh
->EarlyFragmentTests
;
188 shader
->info
.fs
.depth_layout
= fp
->FragDepthLayout
;
192 case MESA_SHADER_COMPUTE
: {
193 struct gl_compute_program
*cp
= (struct gl_compute_program
*)sh
->Program
;
194 shader
->info
.cs
.local_size
[0] = cp
->LocalSize
[0];
195 shader
->info
.cs
.local_size
[1] = cp
->LocalSize
[1];
196 shader
->info
.cs
.local_size
[2] = cp
->LocalSize
[2];
201 break; /* No stage-specific info */
207 nir_visitor::nir_visitor(nir_shader
*shader
)
209 this->supports_ints
= shader
->options
->native_integers
;
210 this->shader
= shader
;
211 this->is_global
= true;
212 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
213 _mesa_key_pointer_equal
);
214 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
215 _mesa_key_pointer_equal
);
218 nir_visitor::~nir_visitor()
220 _mesa_hash_table_destroy(this->var_table
, NULL
);
221 _mesa_hash_table_destroy(this->overload_table
, NULL
);
225 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
228 ralloc_steal(mem_ctx
, this->deref_head
);
229 return this->deref_head
;
232 static nir_constant
*
233 constant_copy(ir_constant
*ir
, void *mem_ctx
)
238 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
240 unsigned total_elems
= ir
->type
->components();
243 ret
->num_elements
= 0;
244 switch (ir
->type
->base_type
) {
246 for (i
= 0; i
< total_elems
; i
++)
247 ret
->value
.u
[i
] = ir
->value
.u
[i
];
251 for (i
= 0; i
< total_elems
; i
++)
252 ret
->value
.i
[i
] = ir
->value
.i
[i
];
255 case GLSL_TYPE_FLOAT
:
256 for (i
= 0; i
< total_elems
; i
++)
257 ret
->value
.f
[i
] = ir
->value
.f
[i
];
261 for (i
= 0; i
< total_elems
; i
++)
262 ret
->value
.b
[i
] = ir
->value
.b
[i
];
265 case GLSL_TYPE_STRUCT
:
266 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
268 ret
->num_elements
= ir
->type
->length
;
271 foreach_in_list(ir_constant
, field
, &ir
->components
) {
272 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
277 case GLSL_TYPE_ARRAY
:
278 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
280 ret
->num_elements
= ir
->type
->length
;
282 for (i
= 0; i
< ir
->type
->length
; i
++)
283 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
287 unreachable("not reached");
294 nir_visitor::visit(ir_variable
*ir
)
296 nir_variable
*var
= ralloc(shader
, nir_variable
);
297 var
->type
= ir
->type
;
298 var
->name
= ralloc_strdup(var
, ir
->name
);
300 var
->data
.read_only
= ir
->data
.read_only
;
301 var
->data
.centroid
= ir
->data
.centroid
;
302 var
->data
.sample
= ir
->data
.sample
;
303 var
->data
.patch
= ir
->data
.patch
;
304 var
->data
.invariant
= ir
->data
.invariant
;
305 var
->data
.location
= ir
->data
.location
;
307 switch(ir
->data
.mode
) {
309 case ir_var_temporary
:
311 var
->data
.mode
= nir_var_global
;
313 var
->data
.mode
= nir_var_local
;
316 case ir_var_function_in
:
317 case ir_var_function_out
:
318 case ir_var_function_inout
:
319 case ir_var_const_in
:
320 var
->data
.mode
= nir_var_local
;
323 case ir_var_shader_in
:
324 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
325 ir
->data
.location
== VARYING_SLOT_FACE
) {
326 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
327 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
328 var
->data
.mode
= nir_var_system_value
;
329 } else if (shader
->stage
== MESA_SHADER_GEOMETRY
&&
330 ir
->data
.location
== VARYING_SLOT_PRIMITIVE_ID
) {
331 /* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
332 var
->data
.location
= SYSTEM_VALUE_PRIMITIVE_ID
;
333 var
->data
.mode
= nir_var_system_value
;
335 var
->data
.mode
= nir_var_shader_in
;
339 case ir_var_shader_out
:
340 var
->data
.mode
= nir_var_shader_out
;
344 var
->data
.mode
= nir_var_uniform
;
347 case ir_var_shader_storage
:
348 var
->data
.mode
= nir_var_shader_storage
;
351 case ir_var_system_value
:
352 var
->data
.mode
= nir_var_system_value
;
356 unreachable("not reached");
359 var
->data
.interpolation
= ir
->data
.interpolation
;
360 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
361 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
362 var
->data
.explicit_location
= ir
->data
.explicit_location
;
363 var
->data
.explicit_index
= ir
->data
.explicit_index
;
364 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
365 var
->data
.has_initializer
= ir
->data
.has_initializer
;
366 var
->data
.location_frac
= ir
->data
.location_frac
;
368 switch (ir
->data
.depth_layout
) {
369 case ir_depth_layout_none
:
370 var
->data
.depth_layout
= nir_depth_layout_none
;
372 case ir_depth_layout_any
:
373 var
->data
.depth_layout
= nir_depth_layout_any
;
375 case ir_depth_layout_greater
:
376 var
->data
.depth_layout
= nir_depth_layout_greater
;
378 case ir_depth_layout_less
:
379 var
->data
.depth_layout
= nir_depth_layout_less
;
381 case ir_depth_layout_unchanged
:
382 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
385 unreachable("not reached");
388 var
->data
.index
= ir
->data
.index
;
389 var
->data
.binding
= ir
->data
.binding
;
390 var
->data
.offset
= ir
->data
.offset
;
391 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
392 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
393 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
394 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
395 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
396 var
->data
.image
.format
= ir
->data
.image_format
;
397 var
->data
.max_array_access
= ir
->data
.max_array_access
;
399 var
->num_state_slots
= ir
->get_num_state_slots();
400 if (var
->num_state_slots
> 0) {
401 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
402 var
->num_state_slots
);
404 ir_state_slot
*state_slots
= ir
->get_state_slots();
405 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
406 for (unsigned j
= 0; j
< 5; j
++)
407 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
408 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
411 var
->state_slots
= NULL
;
414 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
416 var
->interface_type
= ir
->get_interface_type();
418 if (var
->data
.mode
== nir_var_local
)
419 nir_function_impl_add_variable(impl
, var
);
421 nir_shader_add_variable(shader
, var
);
423 _mesa_hash_table_insert(var_table
, ir
, var
);
428 nir_function_visitor::visit_enter(ir_function
*ir
)
430 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
431 visitor
->create_function(sig
);
433 return visit_continue_with_parent
;
437 nir_visitor::create_function(ir_function_signature
*ir
)
439 if (ir
->is_intrinsic
)
442 nir_function
*func
= nir_function_create(shader
, ir
->function_name());
444 assert(ir
->parameters
.is_empty());
445 assert(ir
->return_type
== glsl_type::void_type
);
447 _mesa_hash_table_insert(this->overload_table
, ir
, func
);
451 nir_visitor::visit(ir_function
*ir
)
453 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
458 nir_visitor::visit(ir_function_signature
*ir
)
460 if (ir
->is_intrinsic
)
463 struct hash_entry
*entry
=
464 _mesa_hash_table_search(this->overload_table
, ir
);
467 nir_function
*func
= (nir_function
*) entry
->data
;
469 if (ir
->is_defined
) {
470 nir_function_impl
*impl
= nir_function_impl_create(func
);
473 assert(strcmp(func
->name
, "main") == 0);
474 assert(ir
->parameters
.is_empty());
475 assert(func
->return_type
== glsl_type::void_type
);
477 this->is_global
= false;
479 nir_builder_init(&b
, impl
);
480 b
.cursor
= nir_after_cf_list(&impl
->body
);
481 visit_exec_list(&ir
->body
, this);
483 this->is_global
= true;
490 nir_visitor::visit(ir_loop
*ir
)
492 nir_loop
*loop
= nir_loop_create(this->shader
);
493 nir_builder_cf_insert(&b
, &loop
->cf_node
);
495 b
.cursor
= nir_after_cf_list(&loop
->body
);
496 visit_exec_list(&ir
->body_instructions
, this);
497 b
.cursor
= nir_after_cf_node(&loop
->cf_node
);
501 nir_visitor::visit(ir_if
*ir
)
504 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
506 nir_if
*if_stmt
= nir_if_create(this->shader
);
507 if_stmt
->condition
= condition
;
508 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
510 b
.cursor
= nir_after_cf_list(&if_stmt
->then_list
);
511 visit_exec_list(&ir
->then_instructions
, this);
513 b
.cursor
= nir_after_cf_list(&if_stmt
->else_list
);
514 visit_exec_list(&ir
->else_instructions
, this);
516 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
520 nir_visitor::visit(ir_discard
*ir
)
523 * discards aren't treated as control flow, because before we lower them
524 * they can appear anywhere in the shader and the stuff after them may still
525 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
526 * discards will be immediately followed by a return.
529 nir_intrinsic_instr
*discard
;
531 discard
= nir_intrinsic_instr_create(this->shader
,
532 nir_intrinsic_discard_if
);
534 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
536 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
539 nir_builder_instr_insert(&b
, &discard
->instr
);
543 nir_visitor::visit(ir_emit_vertex
*ir
)
545 nir_intrinsic_instr
*instr
=
546 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
547 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
548 nir_builder_instr_insert(&b
, &instr
->instr
);
552 nir_visitor::visit(ir_end_primitive
*ir
)
554 nir_intrinsic_instr
*instr
=
555 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
556 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
557 nir_builder_instr_insert(&b
, &instr
->instr
);
561 nir_visitor::visit(ir_loop_jump
*ir
)
565 case ir_loop_jump::jump_break
:
566 type
= nir_jump_break
;
568 case ir_loop_jump::jump_continue
:
569 type
= nir_jump_continue
;
572 unreachable("not reached");
575 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
576 nir_builder_instr_insert(&b
, &instr
->instr
);
580 nir_visitor::visit(ir_return
*ir
)
582 if (ir
->value
!= NULL
) {
583 nir_intrinsic_instr
*copy
=
584 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
586 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
587 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
590 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
591 nir_builder_instr_insert(&b
, &instr
->instr
);
595 nir_visitor::visit(ir_call
*ir
)
597 if (ir
->callee
->is_intrinsic
) {
599 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
600 op
= nir_intrinsic_atomic_counter_read_var
;
601 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
602 op
= nir_intrinsic_atomic_counter_inc_var
;
603 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
604 op
= nir_intrinsic_atomic_counter_dec_var
;
605 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
606 op
= nir_intrinsic_image_load
;
607 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
608 op
= nir_intrinsic_image_store
;
609 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
610 op
= nir_intrinsic_image_atomic_add
;
611 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
612 op
= nir_intrinsic_image_atomic_min
;
613 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
614 op
= nir_intrinsic_image_atomic_max
;
615 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
616 op
= nir_intrinsic_image_atomic_and
;
617 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
618 op
= nir_intrinsic_image_atomic_or
;
619 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
620 op
= nir_intrinsic_image_atomic_xor
;
621 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
622 op
= nir_intrinsic_image_atomic_exchange
;
623 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
624 op
= nir_intrinsic_image_atomic_comp_swap
;
625 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
626 op
= nir_intrinsic_memory_barrier
;
627 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_size") == 0) {
628 op
= nir_intrinsic_image_size
;
629 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_samples") == 0) {
630 op
= nir_intrinsic_image_samples
;
631 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_ssbo") == 0) {
632 op
= nir_intrinsic_store_ssbo
;
633 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_ssbo") == 0) {
634 op
= nir_intrinsic_load_ssbo
;
635 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_add_ssbo") == 0) {
636 op
= nir_intrinsic_ssbo_atomic_add
;
637 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_and_ssbo") == 0) {
638 op
= nir_intrinsic_ssbo_atomic_and
;
639 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_or_ssbo") == 0) {
640 op
= nir_intrinsic_ssbo_atomic_or
;
641 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_xor_ssbo") == 0) {
642 op
= nir_intrinsic_ssbo_atomic_xor
;
643 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_min_ssbo") == 0) {
644 assert(ir
->return_deref
);
645 if (ir
->return_deref
->type
== glsl_type::int_type
)
646 op
= nir_intrinsic_ssbo_atomic_imin
;
647 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
648 op
= nir_intrinsic_ssbo_atomic_umin
;
650 unreachable("Invalid type");
651 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_max_ssbo") == 0) {
652 assert(ir
->return_deref
);
653 if (ir
->return_deref
->type
== glsl_type::int_type
)
654 op
= nir_intrinsic_ssbo_atomic_imax
;
655 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
656 op
= nir_intrinsic_ssbo_atomic_umax
;
658 unreachable("Invalid type");
659 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_exchange_ssbo") == 0) {
660 op
= nir_intrinsic_ssbo_atomic_exchange
;
661 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_comp_swap_ssbo") == 0) {
662 op
= nir_intrinsic_ssbo_atomic_comp_swap
;
663 } else if (strcmp(ir
->callee_name(), "__intrinsic_shader_clock") == 0) {
664 op
= nir_intrinsic_shader_clock
;
665 } else if (strcmp(ir
->callee_name(), "__intrinsic_group_memory_barrier") == 0) {
666 op
= nir_intrinsic_group_memory_barrier
;
667 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_atomic_counter") == 0) {
668 op
= nir_intrinsic_memory_barrier_atomic_counter
;
669 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_buffer") == 0) {
670 op
= nir_intrinsic_memory_barrier_buffer
;
671 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_image") == 0) {
672 op
= nir_intrinsic_memory_barrier_image
;
673 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_shared") == 0) {
674 op
= nir_intrinsic_memory_barrier_shared
;
675 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_shared") == 0) {
676 op
= nir_intrinsic_load_shared
;
677 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_shared") == 0) {
678 op
= nir_intrinsic_store_shared
;
679 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_add_shared") == 0) {
680 op
= nir_intrinsic_shared_atomic_add
;
681 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_and_shared") == 0) {
682 op
= nir_intrinsic_shared_atomic_and
;
683 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_or_shared") == 0) {
684 op
= nir_intrinsic_shared_atomic_or
;
685 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_xor_shared") == 0) {
686 op
= nir_intrinsic_shared_atomic_xor
;
687 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_min_shared") == 0) {
688 assert(ir
->return_deref
);
689 if (ir
->return_deref
->type
== glsl_type::int_type
)
690 op
= nir_intrinsic_shared_atomic_imin
;
691 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
692 op
= nir_intrinsic_shared_atomic_umin
;
694 unreachable("Invalid type");
695 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_max_shared") == 0) {
696 assert(ir
->return_deref
);
697 if (ir
->return_deref
->type
== glsl_type::int_type
)
698 op
= nir_intrinsic_shared_atomic_imax
;
699 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
700 op
= nir_intrinsic_shared_atomic_umax
;
702 unreachable("Invalid type");
703 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_exchange_shared") == 0) {
704 op
= nir_intrinsic_shared_atomic_exchange
;
705 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_comp_swap_shared") == 0) {
706 op
= nir_intrinsic_shared_atomic_comp_swap
;
708 unreachable("not reached");
711 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
712 nir_dest
*dest
= &instr
->dest
;
715 case nir_intrinsic_atomic_counter_read_var
:
716 case nir_intrinsic_atomic_counter_inc_var
:
717 case nir_intrinsic_atomic_counter_dec_var
: {
718 ir_dereference
*param
=
719 (ir_dereference
*) ir
->actual_parameters
.get_head();
720 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
721 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
722 nir_builder_instr_insert(&b
, &instr
->instr
);
725 case nir_intrinsic_image_load
:
726 case nir_intrinsic_image_store
:
727 case nir_intrinsic_image_atomic_add
:
728 case nir_intrinsic_image_atomic_min
:
729 case nir_intrinsic_image_atomic_max
:
730 case nir_intrinsic_image_atomic_and
:
731 case nir_intrinsic_image_atomic_or
:
732 case nir_intrinsic_image_atomic_xor
:
733 case nir_intrinsic_image_atomic_exchange
:
734 case nir_intrinsic_image_atomic_comp_swap
:
735 case nir_intrinsic_image_samples
:
736 case nir_intrinsic_image_size
: {
737 nir_ssa_undef_instr
*instr_undef
=
738 nir_ssa_undef_instr_create(shader
, 1);
739 nir_builder_instr_insert(&b
, &instr_undef
->instr
);
741 /* Set the image variable dereference. */
742 exec_node
*param
= ir
->actual_parameters
.get_head();
743 ir_dereference
*image
= (ir_dereference
*)param
;
744 const glsl_type
*type
=
745 image
->variable_referenced()->type
->without_array();
747 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
748 param
= param
->get_next();
750 /* Set the intrinsic destination. */
751 if (ir
->return_deref
) {
752 const nir_intrinsic_info
*info
=
753 &nir_intrinsic_infos
[instr
->intrinsic
];
754 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
755 info
->dest_components
, 32, NULL
);
758 if (op
== nir_intrinsic_image_size
||
759 op
== nir_intrinsic_image_samples
) {
760 nir_builder_instr_insert(&b
, &instr
->instr
);
764 /* Set the address argument, extending the coordinate vector to four
767 nir_ssa_def
*src_addr
=
768 evaluate_rvalue((ir_dereference
*)param
);
769 nir_ssa_def
*srcs
[4];
771 for (int i
= 0; i
< 4; i
++) {
772 if (i
< type
->coordinate_components())
773 srcs
[i
] = nir_channel(&b
, src_addr
, i
);
775 srcs
[i
] = &instr_undef
->def
;
778 instr
->src
[0] = nir_src_for_ssa(nir_vec(&b
, srcs
, 4));
779 param
= param
->get_next();
781 /* Set the sample argument, which is undefined for single-sample
784 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
786 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
787 param
= param
->get_next();
789 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
792 /* Set the intrinsic parameters. */
793 if (!param
->is_tail_sentinel()) {
795 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
796 param
= param
->get_next();
799 if (!param
->is_tail_sentinel()) {
801 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
802 param
= param
->get_next();
804 nir_builder_instr_insert(&b
, &instr
->instr
);
807 case nir_intrinsic_memory_barrier
:
808 case nir_intrinsic_group_memory_barrier
:
809 case nir_intrinsic_memory_barrier_atomic_counter
:
810 case nir_intrinsic_memory_barrier_buffer
:
811 case nir_intrinsic_memory_barrier_image
:
812 case nir_intrinsic_memory_barrier_shared
:
813 nir_builder_instr_insert(&b
, &instr
->instr
);
815 case nir_intrinsic_shader_clock
:
816 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
817 nir_builder_instr_insert(&b
, &instr
->instr
);
819 case nir_intrinsic_store_ssbo
: {
820 exec_node
*param
= ir
->actual_parameters
.get_head();
821 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
823 param
= param
->get_next();
824 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
826 param
= param
->get_next();
827 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
829 param
= param
->get_next();
830 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
833 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
834 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(block
));
835 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(offset
));
836 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
837 instr
->num_components
= val
->type
->vector_elements
;
839 nir_builder_instr_insert(&b
, &instr
->instr
);
842 case nir_intrinsic_load_ssbo
: {
843 exec_node
*param
= ir
->actual_parameters
.get_head();
844 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
846 param
= param
->get_next();
847 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
849 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(block
));
850 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
852 const glsl_type
*type
= ir
->return_deref
->var
->type
;
853 instr
->num_components
= type
->vector_elements
;
855 /* Setup destination register */
856 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
857 type
->vector_elements
, 32, NULL
);
859 /* Insert the created nir instruction now since in the case of boolean
860 * result we will need to emit another instruction after it
862 nir_builder_instr_insert(&b
, &instr
->instr
);
865 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
866 * consider a true boolean to be ~0. Fix this up with a != 0
869 if (type
->base_type
== GLSL_TYPE_BOOL
) {
870 nir_alu_instr
*load_ssbo_compare
=
871 nir_alu_instr_create(shader
, nir_op_ine
);
872 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
873 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
874 load_ssbo_compare
->src
[1].src
=
875 nir_src_for_ssa(nir_imm_int(&b
, 0));
876 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
877 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
878 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
879 &load_ssbo_compare
->dest
.dest
,
880 type
->vector_elements
, 32, NULL
);
881 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
882 nir_builder_instr_insert(&b
, &load_ssbo_compare
->instr
);
883 dest
= &load_ssbo_compare
->dest
.dest
;
887 case nir_intrinsic_ssbo_atomic_add
:
888 case nir_intrinsic_ssbo_atomic_imin
:
889 case nir_intrinsic_ssbo_atomic_umin
:
890 case nir_intrinsic_ssbo_atomic_imax
:
891 case nir_intrinsic_ssbo_atomic_umax
:
892 case nir_intrinsic_ssbo_atomic_and
:
893 case nir_intrinsic_ssbo_atomic_or
:
894 case nir_intrinsic_ssbo_atomic_xor
:
895 case nir_intrinsic_ssbo_atomic_exchange
:
896 case nir_intrinsic_ssbo_atomic_comp_swap
: {
897 int param_count
= ir
->actual_parameters
.length();
898 assert(param_count
== 3 || param_count
== 4);
901 exec_node
*param
= ir
->actual_parameters
.get_head();
902 ir_instruction
*inst
= (ir_instruction
*) param
;
903 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
906 param
= param
->get_next();
907 inst
= (ir_instruction
*) param
;
908 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
910 /* data1 parameter (this is always present) */
911 param
= param
->get_next();
912 inst
= (ir_instruction
*) param
;
913 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
915 /* data2 parameter (only with atomic_comp_swap) */
916 if (param_count
== 4) {
917 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
918 param
= param
->get_next();
919 inst
= (ir_instruction
*) param
;
920 instr
->src
[3] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
924 assert(ir
->return_deref
);
925 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
926 ir
->return_deref
->type
->vector_elements
, 32, NULL
);
927 nir_builder_instr_insert(&b
, &instr
->instr
);
930 case nir_intrinsic_load_shared
: {
931 exec_node
*param
= ir
->actual_parameters
.get_head();
932 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
934 nir_intrinsic_set_base(instr
, 0);
935 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(offset
));
937 const glsl_type
*type
= ir
->return_deref
->var
->type
;
938 instr
->num_components
= type
->vector_elements
;
940 /* Setup destination register */
941 unsigned bit_size
= glsl_get_bit_size(type
->base_type
);
942 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
943 type
->vector_elements
, bit_size
, NULL
);
945 nir_builder_instr_insert(&b
, &instr
->instr
);
948 case nir_intrinsic_store_shared
: {
949 exec_node
*param
= ir
->actual_parameters
.get_head();
950 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
952 param
= param
->get_next();
953 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
955 param
= param
->get_next();
956 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
959 nir_intrinsic_set_base(instr
, 0);
960 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
962 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
964 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
965 instr
->num_components
= val
->type
->vector_elements
;
967 nir_builder_instr_insert(&b
, &instr
->instr
);
970 case nir_intrinsic_shared_atomic_add
:
971 case nir_intrinsic_shared_atomic_imin
:
972 case nir_intrinsic_shared_atomic_umin
:
973 case nir_intrinsic_shared_atomic_imax
:
974 case nir_intrinsic_shared_atomic_umax
:
975 case nir_intrinsic_shared_atomic_and
:
976 case nir_intrinsic_shared_atomic_or
:
977 case nir_intrinsic_shared_atomic_xor
:
978 case nir_intrinsic_shared_atomic_exchange
:
979 case nir_intrinsic_shared_atomic_comp_swap
: {
980 int param_count
= ir
->actual_parameters
.length();
981 assert(param_count
== 2 || param_count
== 3);
984 exec_node
*param
= ir
->actual_parameters
.get_head();
985 ir_instruction
*inst
= (ir_instruction
*) param
;
986 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
988 /* data1 parameter (this is always present) */
989 param
= param
->get_next();
990 inst
= (ir_instruction
*) param
;
991 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
993 /* data2 parameter (only with atomic_comp_swap) */
994 if (param_count
== 3) {
995 assert(op
== nir_intrinsic_shared_atomic_comp_swap
);
996 param
= param
->get_next();
997 inst
= (ir_instruction
*) param
;
999 nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1003 assert(ir
->return_deref
);
1004 unsigned bit_size
= glsl_get_bit_size(ir
->return_deref
->type
->base_type
);
1005 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1006 ir
->return_deref
->type
->vector_elements
,
1008 nir_builder_instr_insert(&b
, &instr
->instr
);
1012 unreachable("not reached");
1015 if (ir
->return_deref
) {
1016 nir_intrinsic_instr
*store_instr
=
1017 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
1018 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
1019 nir_intrinsic_set_write_mask(store_instr
,
1020 (1 << store_instr
->num_components
) - 1);
1022 store_instr
->variables
[0] =
1023 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
1024 store_instr
->src
[0] = nir_src_for_ssa(&dest
->ssa
);
1026 nir_builder_instr_insert(&b
, &store_instr
->instr
);
1032 struct hash_entry
*entry
=
1033 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
1035 nir_function
*callee
= (nir_function
*) entry
->data
;
1037 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
1040 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
1041 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
1045 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
1046 nir_builder_instr_insert(&b
, &instr
->instr
);
1050 nir_visitor::visit(ir_assignment
*ir
)
1052 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
1054 b
.exact
= ir
->lhs
->variable_referenced()->data
.invariant
||
1055 ir
->lhs
->variable_referenced()->data
.precise
;
1057 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
1058 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
1059 /* We're doing a plain-as-can-be copy, so emit a copy_var */
1060 nir_intrinsic_instr
*copy
=
1061 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
1063 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
1064 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
1066 if (ir
->condition
) {
1067 nir_if
*if_stmt
= nir_if_create(this->shader
);
1068 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1069 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1070 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
1071 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1073 nir_builder_instr_insert(&b
, ©
->instr
);
1078 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
1080 ir
->lhs
->accept(this);
1081 nir_deref_var
*lhs_deref
= this->deref_head
;
1082 nir_ssa_def
*src
= evaluate_rvalue(ir
->rhs
);
1084 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
1085 /* GLSL IR will give us the input to the write-masked assignment in a
1086 * single packed vector. So, for example, if the writemask is xzw, then
1087 * we have to swizzle x -> x, y -> z, and z -> w and get the y component
1091 unsigned component
= 0;
1092 for (unsigned i
= 0; i
< 4; i
++) {
1093 swiz
[i
] = ir
->write_mask
& (1 << i
) ? component
++ : 0;
1095 src
= nir_swizzle(&b
, src
, swiz
, num_components
, !supports_ints
);
1098 nir_intrinsic_instr
*store
=
1099 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1100 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1101 nir_intrinsic_set_write_mask(store
, ir
->write_mask
);
1102 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
1103 store
->variables
[0] = nir_deref_as_var(store_deref
);
1104 store
->src
[0] = nir_src_for_ssa(src
);
1106 if (ir
->condition
) {
1107 nir_if
*if_stmt
= nir_if_create(this->shader
);
1108 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1109 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1110 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
1111 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1113 nir_builder_instr_insert(&b
, &store
->instr
);
1118 * Given an instruction, returns a pointer to its destination or NULL if there
1119 * is no destination.
1121 * Note that this only handles instructions we generate at this level.
1124 get_instr_dest(nir_instr
*instr
)
1126 nir_alu_instr
*alu_instr
;
1127 nir_intrinsic_instr
*intrinsic_instr
;
1128 nir_tex_instr
*tex_instr
;
1130 switch (instr
->type
) {
1131 case nir_instr_type_alu
:
1132 alu_instr
= nir_instr_as_alu(instr
);
1133 return &alu_instr
->dest
.dest
;
1135 case nir_instr_type_intrinsic
:
1136 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1137 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1138 return &intrinsic_instr
->dest
;
1142 case nir_instr_type_tex
:
1143 tex_instr
= nir_instr_as_tex(instr
);
1144 return &tex_instr
->dest
;
1147 unreachable("not reached");
1154 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
1156 nir_dest
*dest
= get_instr_dest(instr
);
1159 nir_ssa_dest_init(instr
, dest
, num_components
, 32, NULL
);
1161 nir_builder_instr_insert(&b
, instr
);
1164 assert(dest
->is_ssa
);
1165 this->result
= &dest
->ssa
;
1170 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1173 if (ir
->as_dereference() || ir
->as_constant()) {
1175 * A dereference is being used on the right hand side, which means we
1176 * must emit a variable load.
1179 nir_intrinsic_instr
*load_instr
=
1180 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1181 load_instr
->num_components
= ir
->type
->vector_elements
;
1182 load_instr
->variables
[0] = this->deref_head
;
1183 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1184 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
1187 return this->result
;
1191 nir_visitor::visit(ir_expression
*ir
)
1193 /* Some special cases */
1194 switch (ir
->operation
) {
1195 case ir_binop_ubo_load
: {
1196 nir_intrinsic_instr
*load
=
1197 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_ubo
);
1198 load
->num_components
= ir
->type
->vector_elements
;
1199 load
->dest
.ssa
.bit_size
= glsl_get_bit_size(ir
->type
->base_type
);
1200 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1201 load
->src
[1] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1202 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1205 * In UBO's, a true boolean value is any non-zero value, but we consider
1206 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1209 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
)
1210 this->result
= nir_ine(&b
, &load
->dest
.ssa
, nir_imm_int(&b
, 0));
1215 case ir_unop_interpolate_at_centroid
:
1216 case ir_binop_interpolate_at_offset
:
1217 case ir_binop_interpolate_at_sample
: {
1218 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1219 ir_swizzle
*swizzle
= NULL
;
1221 /* the api does not allow a swizzle here, but the varying packing code
1222 * may have pushed one into here.
1224 swizzle
= ir
->operands
[0]->as_swizzle();
1226 deref
= swizzle
->val
->as_dereference();
1230 deref
->accept(this);
1232 nir_intrinsic_op op
;
1233 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1234 switch (ir
->operation
) {
1235 case ir_unop_interpolate_at_centroid
:
1236 op
= nir_intrinsic_interp_var_at_centroid
;
1238 case ir_binop_interpolate_at_offset
:
1239 op
= nir_intrinsic_interp_var_at_offset
;
1241 case ir_binop_interpolate_at_sample
:
1242 op
= nir_intrinsic_interp_var_at_sample
;
1245 unreachable("Invalid interpolation intrinsic");
1248 /* This case can happen if the vertex shader does not write the
1249 * given varying. In this case, the linker will lower it to a
1250 * global variable. Since interpolating a variable makes no
1251 * sense, we'll just turn it into a load which will probably
1252 * eventually end up as an SSA definition.
1254 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1255 op
= nir_intrinsic_load_var
;
1258 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1259 intrin
->num_components
= deref
->type
->vector_elements
;
1260 intrin
->variables
[0] = this->deref_head
;
1261 ralloc_steal(intrin
, intrin
->variables
[0]);
1263 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1264 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1265 intrin
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1267 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
1270 unsigned swiz
[4] = {
1271 swizzle
->mask
.x
, swizzle
->mask
.y
, swizzle
->mask
.z
, swizzle
->mask
.w
1274 result
= nir_swizzle(&b
, result
, swiz
,
1275 swizzle
->type
->vector_elements
, false);
1285 nir_ssa_def
*srcs
[4];
1286 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1287 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1289 glsl_base_type types
[4];
1290 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1292 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1294 types
[i
] = GLSL_TYPE_FLOAT
;
1296 glsl_base_type out_type
;
1298 out_type
= ir
->type
->base_type
;
1300 out_type
= GLSL_TYPE_FLOAT
;
1302 switch (ir
->operation
) {
1303 case ir_unop_bit_not
: result
= nir_inot(&b
, srcs
[0]); break;
1304 case ir_unop_logic_not
:
1305 result
= supports_ints
? nir_inot(&b
, srcs
[0]) : nir_fnot(&b
, srcs
[0]);
1308 result
= (types
[0] == GLSL_TYPE_FLOAT
) ? nir_fneg(&b
, srcs
[0])
1309 : nir_ineg(&b
, srcs
[0]);
1312 result
= (types
[0] == GLSL_TYPE_FLOAT
) ? nir_fabs(&b
, srcs
[0])
1313 : nir_iabs(&b
, srcs
[0]);
1315 case ir_unop_saturate
:
1316 assert(types
[0] == GLSL_TYPE_FLOAT
);
1317 result
= nir_fsat(&b
, srcs
[0]);
1320 result
= (types
[0] == GLSL_TYPE_FLOAT
) ? nir_fsign(&b
, srcs
[0])
1321 : nir_isign(&b
, srcs
[0]);
1323 case ir_unop_rcp
: result
= nir_frcp(&b
, srcs
[0]); break;
1324 case ir_unop_rsq
: result
= nir_frsq(&b
, srcs
[0]); break;
1325 case ir_unop_sqrt
: result
= nir_fsqrt(&b
, srcs
[0]); break;
1326 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1327 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1328 case ir_unop_exp2
: result
= nir_fexp2(&b
, srcs
[0]); break;
1329 case ir_unop_log2
: result
= nir_flog2(&b
, srcs
[0]); break;
1331 result
= supports_ints
? nir_i2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1334 result
= supports_ints
? nir_u2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1337 result
= supports_ints
? nir_b2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1339 case ir_unop_f2i
: result
= nir_f2i(&b
, srcs
[0]); break;
1340 case ir_unop_f2u
: result
= nir_f2u(&b
, srcs
[0]); break;
1341 case ir_unop_f2b
: result
= nir_f2b(&b
, srcs
[0]); break;
1342 case ir_unop_i2b
: result
= nir_i2b(&b
, srcs
[0]); break;
1343 case ir_unop_b2i
: result
= nir_b2i(&b
, srcs
[0]); break;
1346 case ir_unop_bitcast_i2f
:
1347 case ir_unop_bitcast_f2i
:
1348 case ir_unop_bitcast_u2f
:
1349 case ir_unop_bitcast_f2u
:
1350 case ir_unop_subroutine_to_int
:
1352 result
= nir_imov(&b
, srcs
[0]);
1354 case ir_unop_trunc
: result
= nir_ftrunc(&b
, srcs
[0]); break;
1355 case ir_unop_ceil
: result
= nir_fceil(&b
, srcs
[0]); break;
1356 case ir_unop_floor
: result
= nir_ffloor(&b
, srcs
[0]); break;
1357 case ir_unop_fract
: result
= nir_ffract(&b
, srcs
[0]); break;
1358 case ir_unop_round_even
: result
= nir_fround_even(&b
, srcs
[0]); break;
1359 case ir_unop_sin
: result
= nir_fsin(&b
, srcs
[0]); break;
1360 case ir_unop_cos
: result
= nir_fcos(&b
, srcs
[0]); break;
1361 case ir_unop_dFdx
: result
= nir_fddx(&b
, srcs
[0]); break;
1362 case ir_unop_dFdy
: result
= nir_fddy(&b
, srcs
[0]); break;
1363 case ir_unop_dFdx_fine
: result
= nir_fddx_fine(&b
, srcs
[0]); break;
1364 case ir_unop_dFdy_fine
: result
= nir_fddy_fine(&b
, srcs
[0]); break;
1365 case ir_unop_dFdx_coarse
: result
= nir_fddx_coarse(&b
, srcs
[0]); break;
1366 case ir_unop_dFdy_coarse
: result
= nir_fddy_coarse(&b
, srcs
[0]); break;
1367 case ir_unop_pack_snorm_2x16
:
1368 result
= nir_pack_snorm_2x16(&b
, srcs
[0]);
1370 case ir_unop_pack_snorm_4x8
:
1371 result
= nir_pack_snorm_4x8(&b
, srcs
[0]);
1373 case ir_unop_pack_unorm_2x16
:
1374 result
= nir_pack_unorm_2x16(&b
, srcs
[0]);
1376 case ir_unop_pack_unorm_4x8
:
1377 result
= nir_pack_unorm_4x8(&b
, srcs
[0]);
1379 case ir_unop_pack_half_2x16
:
1380 result
= nir_pack_half_2x16(&b
, srcs
[0]);
1382 case ir_unop_unpack_snorm_2x16
:
1383 result
= nir_unpack_snorm_2x16(&b
, srcs
[0]);
1385 case ir_unop_unpack_snorm_4x8
:
1386 result
= nir_unpack_snorm_4x8(&b
, srcs
[0]);
1388 case ir_unop_unpack_unorm_2x16
:
1389 result
= nir_unpack_unorm_2x16(&b
, srcs
[0]);
1391 case ir_unop_unpack_unorm_4x8
:
1392 result
= nir_unpack_unorm_4x8(&b
, srcs
[0]);
1394 case ir_unop_unpack_half_2x16
:
1395 result
= nir_unpack_half_2x16(&b
, srcs
[0]);
1397 case ir_unop_bitfield_reverse
:
1398 result
= nir_bitfield_reverse(&b
, srcs
[0]);
1400 case ir_unop_bit_count
:
1401 result
= nir_bit_count(&b
, srcs
[0]);
1403 case ir_unop_find_msb
:
1405 case GLSL_TYPE_UINT
:
1406 result
= nir_ufind_msb(&b
, srcs
[0]);
1409 result
= nir_ifind_msb(&b
, srcs
[0]);
1412 unreachable("Invalid type for findMSB()");
1415 case ir_unop_find_lsb
:
1416 result
= nir_find_lsb(&b
, srcs
[0]);
1420 switch (ir
->type
->vector_elements
) {
1422 switch (ir
->operands
[0]->type
->vector_elements
) {
1423 case 1: result
= nir_fnoise1_1(&b
, srcs
[0]); break;
1424 case 2: result
= nir_fnoise1_2(&b
, srcs
[0]); break;
1425 case 3: result
= nir_fnoise1_3(&b
, srcs
[0]); break;
1426 case 4: result
= nir_fnoise1_4(&b
, srcs
[0]); break;
1427 default: unreachable("not reached");
1431 switch (ir
->operands
[0]->type
->vector_elements
) {
1432 case 1: result
= nir_fnoise2_1(&b
, srcs
[0]); break;
1433 case 2: result
= nir_fnoise2_2(&b
, srcs
[0]); break;
1434 case 3: result
= nir_fnoise2_3(&b
, srcs
[0]); break;
1435 case 4: result
= nir_fnoise2_4(&b
, srcs
[0]); break;
1436 default: unreachable("not reached");
1440 switch (ir
->operands
[0]->type
->vector_elements
) {
1441 case 1: result
= nir_fnoise3_1(&b
, srcs
[0]); break;
1442 case 2: result
= nir_fnoise3_2(&b
, srcs
[0]); break;
1443 case 3: result
= nir_fnoise3_3(&b
, srcs
[0]); break;
1444 case 4: result
= nir_fnoise3_4(&b
, srcs
[0]); break;
1445 default: unreachable("not reached");
1449 switch (ir
->operands
[0]->type
->vector_elements
) {
1450 case 1: result
= nir_fnoise4_1(&b
, srcs
[0]); break;
1451 case 2: result
= nir_fnoise4_2(&b
, srcs
[0]); break;
1452 case 3: result
= nir_fnoise4_3(&b
, srcs
[0]); break;
1453 case 4: result
= nir_fnoise4_4(&b
, srcs
[0]); break;
1454 default: unreachable("not reached");
1458 unreachable("not reached");
1461 case ir_unop_get_buffer_size
: {
1462 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1464 nir_intrinsic_get_buffer_size
);
1465 load
->num_components
= ir
->type
->vector_elements
;
1466 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1467 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1472 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fadd(&b
, srcs
[0], srcs
[1])
1473 : nir_iadd(&b
, srcs
[0], srcs
[1]);
1476 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fsub(&b
, srcs
[0], srcs
[1])
1477 : nir_isub(&b
, srcs
[0], srcs
[1]);
1480 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fmul(&b
, srcs
[0], srcs
[1])
1481 : nir_imul(&b
, srcs
[0], srcs
[1]);
1484 if (out_type
== GLSL_TYPE_FLOAT
)
1485 result
= nir_fdiv(&b
, srcs
[0], srcs
[1]);
1486 else if (out_type
== GLSL_TYPE_INT
)
1487 result
= nir_idiv(&b
, srcs
[0], srcs
[1]);
1489 result
= nir_udiv(&b
, srcs
[0], srcs
[1]);
1492 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fmod(&b
, srcs
[0], srcs
[1])
1493 : nir_umod(&b
, srcs
[0], srcs
[1]);
1496 if (out_type
== GLSL_TYPE_FLOAT
)
1497 result
= nir_fmin(&b
, srcs
[0], srcs
[1]);
1498 else if (out_type
== GLSL_TYPE_INT
)
1499 result
= nir_imin(&b
, srcs
[0], srcs
[1]);
1501 result
= nir_umin(&b
, srcs
[0], srcs
[1]);
1504 if (out_type
== GLSL_TYPE_FLOAT
)
1505 result
= nir_fmax(&b
, srcs
[0], srcs
[1]);
1506 else if (out_type
== GLSL_TYPE_INT
)
1507 result
= nir_imax(&b
, srcs
[0], srcs
[1]);
1509 result
= nir_umax(&b
, srcs
[0], srcs
[1]);
1511 case ir_binop_pow
: result
= nir_fpow(&b
, srcs
[0], srcs
[1]); break;
1512 case ir_binop_bit_and
: result
= nir_iand(&b
, srcs
[0], srcs
[1]); break;
1513 case ir_binop_bit_or
: result
= nir_ior(&b
, srcs
[0], srcs
[1]); break;
1514 case ir_binop_bit_xor
: result
= nir_ixor(&b
, srcs
[0], srcs
[1]); break;
1515 case ir_binop_logic_and
:
1516 result
= supports_ints
? nir_iand(&b
, srcs
[0], srcs
[1])
1517 : nir_fand(&b
, srcs
[0], srcs
[1]);
1519 case ir_binop_logic_or
:
1520 result
= supports_ints
? nir_ior(&b
, srcs
[0], srcs
[1])
1521 : nir_for(&b
, srcs
[0], srcs
[1]);
1523 case ir_binop_logic_xor
:
1524 result
= supports_ints
? nir_ixor(&b
, srcs
[0], srcs
[1])
1525 : nir_fxor(&b
, srcs
[0], srcs
[1]);
1527 case ir_binop_lshift
: result
= nir_ishl(&b
, srcs
[0], srcs
[1]); break;
1528 case ir_binop_rshift
:
1529 result
= (out_type
== GLSL_TYPE_INT
) ? nir_ishr(&b
, srcs
[0], srcs
[1])
1530 : nir_ushr(&b
, srcs
[0], srcs
[1]);
1532 case ir_binop_imul_high
:
1533 result
= (out_type
== GLSL_TYPE_INT
) ? nir_imul_high(&b
, srcs
[0], srcs
[1])
1534 : nir_umul_high(&b
, srcs
[0], srcs
[1]);
1536 case ir_binop_carry
: result
= nir_uadd_carry(&b
, srcs
[0], srcs
[1]); break;
1537 case ir_binop_borrow
: result
= nir_usub_borrow(&b
, srcs
[0], srcs
[1]); break;
1539 if (supports_ints
) {
1540 if (types
[0] == GLSL_TYPE_FLOAT
)
1541 result
= nir_flt(&b
, srcs
[0], srcs
[1]);
1542 else if (types
[0] == GLSL_TYPE_INT
)
1543 result
= nir_ilt(&b
, srcs
[0], srcs
[1]);
1545 result
= nir_ult(&b
, srcs
[0], srcs
[1]);
1547 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1550 case ir_binop_greater
:
1551 if (supports_ints
) {
1552 if (types
[0] == GLSL_TYPE_FLOAT
)
1553 result
= nir_flt(&b
, srcs
[1], srcs
[0]);
1554 else if (types
[0] == GLSL_TYPE_INT
)
1555 result
= nir_ilt(&b
, srcs
[1], srcs
[0]);
1557 result
= nir_ult(&b
, srcs
[1], srcs
[0]);
1559 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1562 case ir_binop_lequal
:
1563 if (supports_ints
) {
1564 if (types
[0] == GLSL_TYPE_FLOAT
)
1565 result
= nir_fge(&b
, srcs
[1], srcs
[0]);
1566 else if (types
[0] == GLSL_TYPE_INT
)
1567 result
= nir_ige(&b
, srcs
[1], srcs
[0]);
1569 result
= nir_uge(&b
, srcs
[1], srcs
[0]);
1571 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1574 case ir_binop_gequal
:
1575 if (supports_ints
) {
1576 if (types
[0] == GLSL_TYPE_FLOAT
)
1577 result
= nir_fge(&b
, srcs
[0], srcs
[1]);
1578 else if (types
[0] == GLSL_TYPE_INT
)
1579 result
= nir_ige(&b
, srcs
[0], srcs
[1]);
1581 result
= nir_uge(&b
, srcs
[0], srcs
[1]);
1583 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1586 case ir_binop_equal
:
1587 if (supports_ints
) {
1588 if (types
[0] == GLSL_TYPE_FLOAT
)
1589 result
= nir_feq(&b
, srcs
[0], srcs
[1]);
1591 result
= nir_ieq(&b
, srcs
[0], srcs
[1]);
1593 result
= nir_seq(&b
, srcs
[0], srcs
[1]);
1596 case ir_binop_nequal
:
1597 if (supports_ints
) {
1598 if (types
[0] == GLSL_TYPE_FLOAT
)
1599 result
= nir_fne(&b
, srcs
[0], srcs
[1]);
1601 result
= nir_ine(&b
, srcs
[0], srcs
[1]);
1603 result
= nir_sne(&b
, srcs
[0], srcs
[1]);
1606 case ir_binop_all_equal
:
1607 if (supports_ints
) {
1608 if (types
[0] == GLSL_TYPE_FLOAT
) {
1609 switch (ir
->operands
[0]->type
->vector_elements
) {
1610 case 1: result
= nir_feq(&b
, srcs
[0], srcs
[1]); break;
1611 case 2: result
= nir_ball_fequal2(&b
, srcs
[0], srcs
[1]); break;
1612 case 3: result
= nir_ball_fequal3(&b
, srcs
[0], srcs
[1]); break;
1613 case 4: result
= nir_ball_fequal4(&b
, srcs
[0], srcs
[1]); break;
1615 unreachable("not reached");
1618 switch (ir
->operands
[0]->type
->vector_elements
) {
1619 case 1: result
= nir_ieq(&b
, srcs
[0], srcs
[1]); break;
1620 case 2: result
= nir_ball_iequal2(&b
, srcs
[0], srcs
[1]); break;
1621 case 3: result
= nir_ball_iequal3(&b
, srcs
[0], srcs
[1]); break;
1622 case 4: result
= nir_ball_iequal4(&b
, srcs
[0], srcs
[1]); break;
1624 unreachable("not reached");
1628 switch (ir
->operands
[0]->type
->vector_elements
) {
1629 case 1: result
= nir_seq(&b
, srcs
[0], srcs
[1]); break;
1630 case 2: result
= nir_fall_equal2(&b
, srcs
[0], srcs
[1]); break;
1631 case 3: result
= nir_fall_equal3(&b
, srcs
[0], srcs
[1]); break;
1632 case 4: result
= nir_fall_equal4(&b
, srcs
[0], srcs
[1]); break;
1634 unreachable("not reached");
1638 case ir_binop_any_nequal
:
1639 if (supports_ints
) {
1640 if (types
[0] == GLSL_TYPE_FLOAT
) {
1641 switch (ir
->operands
[0]->type
->vector_elements
) {
1642 case 1: result
= nir_fne(&b
, srcs
[0], srcs
[1]); break;
1643 case 2: result
= nir_bany_fnequal2(&b
, srcs
[0], srcs
[1]); break;
1644 case 3: result
= nir_bany_fnequal3(&b
, srcs
[0], srcs
[1]); break;
1645 case 4: result
= nir_bany_fnequal4(&b
, srcs
[0], srcs
[1]); break;
1647 unreachable("not reached");
1650 switch (ir
->operands
[0]->type
->vector_elements
) {
1651 case 1: result
= nir_ine(&b
, srcs
[0], srcs
[1]); break;
1652 case 2: result
= nir_bany_inequal2(&b
, srcs
[0], srcs
[1]); break;
1653 case 3: result
= nir_bany_inequal3(&b
, srcs
[0], srcs
[1]); break;
1654 case 4: result
= nir_bany_inequal4(&b
, srcs
[0], srcs
[1]); break;
1656 unreachable("not reached");
1660 switch (ir
->operands
[0]->type
->vector_elements
) {
1661 case 1: result
= nir_sne(&b
, srcs
[0], srcs
[1]); break;
1662 case 2: result
= nir_fany_nequal2(&b
, srcs
[0], srcs
[1]); break;
1663 case 3: result
= nir_fany_nequal3(&b
, srcs
[0], srcs
[1]); break;
1664 case 4: result
= nir_fany_nequal4(&b
, srcs
[0], srcs
[1]); break;
1666 unreachable("not reached");
1671 switch (ir
->operands
[0]->type
->vector_elements
) {
1672 case 2: result
= nir_fdot2(&b
, srcs
[0], srcs
[1]); break;
1673 case 3: result
= nir_fdot3(&b
, srcs
[0], srcs
[1]); break;
1674 case 4: result
= nir_fdot4(&b
, srcs
[0], srcs
[1]); break;
1676 unreachable("not reached");
1680 case ir_binop_ldexp
: result
= nir_ldexp(&b
, srcs
[0], srcs
[1]); break;
1682 result
= nir_ffma(&b
, srcs
[0], srcs
[1], srcs
[2]);
1685 result
= nir_flrp(&b
, srcs
[0], srcs
[1], srcs
[2]);
1689 result
= nir_bcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1691 result
= nir_fcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1693 case ir_triop_bitfield_extract
:
1694 result
= (out_type
== GLSL_TYPE_INT
) ?
1695 nir_ibitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]) :
1696 nir_ubitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]);
1698 case ir_quadop_bitfield_insert
:
1699 result
= nir_bitfield_insert(&b
, srcs
[0], srcs
[1], srcs
[2], srcs
[3]);
1701 case ir_quadop_vector
:
1702 result
= nir_vec(&b
, srcs
, ir
->type
->vector_elements
);
1706 unreachable("not reached");
1711 nir_visitor::visit(ir_swizzle
*ir
)
1713 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1714 result
= nir_swizzle(&b
, evaluate_rvalue(ir
->val
), swizzle
,
1715 ir
->type
->vector_elements
, !supports_ints
);
1719 nir_visitor::visit(ir_texture
*ir
)
1726 num_srcs
= 1; /* coordinate */
1731 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1732 num_srcs
= 2; /* coordinate, bias/lod */
1736 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1742 if (ir
->lod_info
.lod
!= NULL
)
1743 num_srcs
= 2; /* coordinate, lod */
1745 num_srcs
= 1; /* coordinate */
1749 op
= nir_texop_txf_ms
;
1750 num_srcs
= 2; /* coordinate, sample_index */
1755 if (ir
->lod_info
.lod
!= NULL
)
1756 num_srcs
= 1; /* lod */
1763 num_srcs
= 1; /* coordinate */
1768 num_srcs
= 1; /* coordinate */
1771 case ir_query_levels
:
1772 op
= nir_texop_query_levels
;
1776 case ir_texture_samples
:
1777 op
= nir_texop_texture_samples
;
1781 case ir_samples_identical
:
1782 op
= nir_texop_samples_identical
;
1783 num_srcs
= 1; /* coordinate */
1787 unreachable("not reached");
1790 if (ir
->projector
!= NULL
)
1792 if (ir
->shadow_comparitor
!= NULL
)
1794 if (ir
->offset
!= NULL
)
1797 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1800 instr
->sampler_dim
=
1801 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1802 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1803 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1804 if (instr
->is_shadow
)
1805 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1806 switch (ir
->type
->base_type
) {
1807 case GLSL_TYPE_FLOAT
:
1808 instr
->dest_type
= nir_type_float
;
1811 instr
->dest_type
= nir_type_int
;
1813 case GLSL_TYPE_BOOL
:
1814 case GLSL_TYPE_UINT
:
1815 instr
->dest_type
= nir_type_uint
;
1818 unreachable("not reached");
1821 instr
->texture
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1823 unsigned src_number
= 0;
1825 if (ir
->coordinate
!= NULL
) {
1826 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1827 instr
->src
[src_number
].src
=
1828 nir_src_for_ssa(evaluate_rvalue(ir
->coordinate
));
1829 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1833 if (ir
->projector
!= NULL
) {
1834 instr
->src
[src_number
].src
=
1835 nir_src_for_ssa(evaluate_rvalue(ir
->projector
));
1836 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1840 if (ir
->shadow_comparitor
!= NULL
) {
1841 instr
->src
[src_number
].src
=
1842 nir_src_for_ssa(evaluate_rvalue(ir
->shadow_comparitor
));
1843 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1847 if (ir
->offset
!= NULL
) {
1848 /* we don't support multiple offsets yet */
1849 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1851 instr
->src
[src_number
].src
=
1852 nir_src_for_ssa(evaluate_rvalue(ir
->offset
));
1853 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1859 instr
->src
[src_number
].src
=
1860 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.bias
));
1861 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1868 if (ir
->lod_info
.lod
!= NULL
) {
1869 instr
->src
[src_number
].src
=
1870 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.lod
));
1871 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1877 instr
->src
[src_number
].src
=
1878 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdx
));
1879 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1881 instr
->src
[src_number
].src
=
1882 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdy
));
1883 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1888 instr
->src
[src_number
].src
=
1889 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.sample_index
));
1890 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1895 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1902 assert(src_number
== num_srcs
);
1904 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
1908 nir_visitor::visit(ir_constant
*ir
)
1911 * We don't know if this variable is an an array or struct that gets
1912 * dereferenced, so do the safe thing an make it a variable with a
1913 * constant initializer and return a dereference.
1917 nir_local_variable_create(this->impl
, ir
->type
, "const_temp");
1918 var
->data
.read_only
= true;
1919 var
->constant_initializer
= constant_copy(ir
, var
);
1921 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1922 this->deref_tail
= &this->deref_head
->deref
;
1926 nir_visitor::visit(ir_dereference_variable
*ir
)
1928 struct hash_entry
*entry
=
1929 _mesa_hash_table_search(this->var_table
, ir
->var
);
1931 nir_variable
*var
= (nir_variable
*) entry
->data
;
1933 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
1934 this->deref_head
= deref
;
1935 this->deref_tail
= &deref
->deref
;
1939 nir_visitor::visit(ir_dereference_record
*ir
)
1941 ir
->record
->accept(this);
1943 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
1944 assert(field_index
>= 0);
1946 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
1947 deref
->deref
.type
= ir
->type
;
1948 this->deref_tail
->child
= &deref
->deref
;
1949 this->deref_tail
= &deref
->deref
;
1953 nir_visitor::visit(ir_dereference_array
*ir
)
1955 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
1956 deref
->deref
.type
= ir
->type
;
1958 ir_constant
*const_index
= ir
->array_index
->as_constant();
1959 if (const_index
!= NULL
) {
1960 deref
->deref_array_type
= nir_deref_array_type_direct
;
1961 deref
->base_offset
= const_index
->value
.u
[0];
1963 deref
->deref_array_type
= nir_deref_array_type_indirect
;
1965 nir_src_for_ssa(evaluate_rvalue(ir
->array_index
));
1968 ir
->array
->accept(this);
1970 this->deref_tail
->child
= &deref
->deref
;
1971 ralloc_steal(this->deref_tail
, deref
);
1972 this->deref_tail
= &deref
->deref
;
1976 nir_visitor::visit(ir_barrier
*ir
)
1978 nir_intrinsic_instr
*instr
=
1979 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
1980 nir_builder_instr_insert(&b
, &instr
->instr
);