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 "ir_visitor.h"
32 #include "ir_hierarchical_visitor.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
, gl_shader
*sh
);
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
*ir
);
76 void create_overload(ir_function_signature
*ir
, nir_function
*function
);
77 void add_instr(nir_instr
*instr
, unsigned num_components
);
78 nir_ssa_def
*evaluate_rvalue(ir_rvalue
*ir
);
80 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
**srcs
);
81 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
);
82 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
,
84 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
,
85 nir_ssa_def
*src2
, nir_ssa_def
*src3
);
92 nir_function_impl
*impl
;
94 nir_ssa_def
*result
; /* result of the expression tree last visited */
96 nir_deref_var
*evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
);
98 /* the head of the dereference chain we're creating */
99 nir_deref_var
*deref_head
;
100 /* the tail of the dereference chain we're creating */
101 nir_deref
*deref_tail
;
103 nir_variable
*var
; /* variable created by ir_variable visitor */
105 /* whether the IR we're operating on is per-function or global */
108 /* map of ir_variable -> nir_variable */
109 struct hash_table
*var_table
;
111 /* map of ir_function_signature -> nir_function_overload */
112 struct hash_table
*overload_table
;
116 * This visitor runs before the main visitor, calling create_function() for
117 * each function so that the main visitor can resolve forward references in
121 class nir_function_visitor
: public ir_hierarchical_visitor
124 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
127 virtual ir_visitor_status
visit_enter(ir_function
*);
130 nir_visitor
*visitor
;
133 }; /* end of anonymous namespace */
136 glsl_to_nir(const struct gl_shader_program
*shader_prog
,
137 gl_shader_stage stage
,
138 const nir_shader_compiler_options
*options
)
140 struct gl_shader
*sh
= shader_prog
->_LinkedShaders
[stage
];
142 nir_shader
*shader
= nir_shader_create(NULL
, stage
, options
);
144 nir_visitor
v1(shader
, sh
);
145 nir_function_visitor
v2(&v1
);
147 visit_exec_list(sh
->ir
, &v1
);
149 nir_lower_outputs_to_temporaries(shader
);
151 shader
->info
.name
= ralloc_asprintf(shader
, "GLSL%d", shader_prog
->Name
);
152 if (shader_prog
->Label
)
153 shader
->info
.label
= ralloc_strdup(shader
, shader_prog
->Label
);
154 shader
->info
.num_textures
= _mesa_fls(sh
->Program
->SamplersUsed
);
155 shader
->info
.num_ubos
= sh
->NumUniformBlocks
;
156 shader
->info
.num_abos
= shader_prog
->NumAtomicBuffers
;
157 shader
->info
.num_ssbos
= sh
->NumShaderStorageBlocks
;
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
.patch_inputs_read
= sh
->Program
->PatchInputsRead
;
162 shader
->info
.patch_outputs_written
= sh
->Program
->PatchOutputsWritten
;
163 shader
->info
.system_values_read
= sh
->Program
->SystemValuesRead
;
164 shader
->info
.uses_texture_gather
= sh
->Program
->UsesGather
;
165 shader
->info
.uses_clip_distance_out
=
166 sh
->Program
->ClipDistanceArraySize
!= 0;
167 shader
->info
.separate_shader
= shader_prog
->SeparateShader
;
168 shader
->info
.has_transform_feedback_varyings
=
169 shader_prog
->TransformFeedback
.NumVarying
> 0;
172 case MESA_SHADER_TESS_CTRL
:
173 shader
->info
.tcs
.vertices_out
= shader_prog
->TessCtrl
.VerticesOut
;
176 case MESA_SHADER_GEOMETRY
:
177 shader
->info
.gs
.vertices_in
= shader_prog
->Geom
.VerticesIn
;
178 shader
->info
.gs
.output_primitive
= sh
->Geom
.OutputType
;
179 shader
->info
.gs
.vertices_out
= sh
->Geom
.VerticesOut
;
180 shader
->info
.gs
.invocations
= sh
->Geom
.Invocations
;
181 shader
->info
.gs
.uses_end_primitive
= shader_prog
->Geom
.UsesEndPrimitive
;
182 shader
->info
.gs
.uses_streams
= shader_prog
->Geom
.UsesStreams
;
185 case MESA_SHADER_FRAGMENT
: {
186 struct gl_fragment_program
*fp
=
187 (struct gl_fragment_program
*)sh
->Program
;
189 shader
->info
.fs
.uses_discard
= fp
->UsesKill
;
190 shader
->info
.fs
.early_fragment_tests
= sh
->EarlyFragmentTests
;
191 shader
->info
.fs
.depth_layout
= fp
->FragDepthLayout
;
195 case MESA_SHADER_COMPUTE
: {
196 struct gl_compute_program
*cp
= (struct gl_compute_program
*)sh
->Program
;
197 shader
->info
.cs
.local_size
[0] = cp
->LocalSize
[0];
198 shader
->info
.cs
.local_size
[1] = cp
->LocalSize
[1];
199 shader
->info
.cs
.local_size
[2] = cp
->LocalSize
[2];
204 break; /* No stage-specific info */
210 nir_visitor::nir_visitor(nir_shader
*shader
, gl_shader
*sh
)
212 this->supports_ints
= shader
->options
->native_integers
;
213 this->shader
= shader
;
215 this->is_global
= true;
216 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
217 _mesa_key_pointer_equal
);
218 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
219 _mesa_key_pointer_equal
);
222 nir_visitor::~nir_visitor()
224 _mesa_hash_table_destroy(this->var_table
, NULL
);
225 _mesa_hash_table_destroy(this->overload_table
, NULL
);
229 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
232 ralloc_steal(mem_ctx
, this->deref_head
);
233 return this->deref_head
;
236 static nir_constant
*
237 constant_copy(ir_constant
*ir
, void *mem_ctx
)
242 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
244 unsigned total_elems
= ir
->type
->components();
247 ret
->num_elements
= 0;
248 switch (ir
->type
->base_type
) {
250 for (i
= 0; i
< total_elems
; i
++)
251 ret
->value
.u
[i
] = ir
->value
.u
[i
];
255 for (i
= 0; i
< total_elems
; i
++)
256 ret
->value
.i
[i
] = ir
->value
.i
[i
];
259 case GLSL_TYPE_FLOAT
:
260 for (i
= 0; i
< total_elems
; i
++)
261 ret
->value
.f
[i
] = ir
->value
.f
[i
];
265 for (i
= 0; i
< total_elems
; i
++)
266 ret
->value
.b
[i
] = ir
->value
.b
[i
];
269 case GLSL_TYPE_STRUCT
:
270 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
272 ret
->num_elements
= ir
->type
->length
;
275 foreach_in_list(ir_constant
, field
, &ir
->components
) {
276 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
281 case GLSL_TYPE_ARRAY
:
282 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
284 ret
->num_elements
= ir
->type
->length
;
286 for (i
= 0; i
< ir
->type
->length
; i
++)
287 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
291 unreachable("not reached");
298 nir_visitor::visit(ir_variable
*ir
)
300 nir_variable
*var
= ralloc(shader
, nir_variable
);
301 var
->type
= ir
->type
;
302 var
->name
= ralloc_strdup(var
, ir
->name
);
304 var
->data
.read_only
= ir
->data
.read_only
;
305 var
->data
.centroid
= ir
->data
.centroid
;
306 var
->data
.sample
= ir
->data
.sample
;
307 var
->data
.patch
= ir
->data
.patch
;
308 var
->data
.invariant
= ir
->data
.invariant
;
309 var
->data
.location
= ir
->data
.location
;
311 switch(ir
->data
.mode
) {
313 case ir_var_temporary
:
315 var
->data
.mode
= nir_var_global
;
317 var
->data
.mode
= nir_var_local
;
320 case ir_var_function_in
:
321 case ir_var_function_out
:
322 case ir_var_function_inout
:
323 case ir_var_const_in
:
324 var
->data
.mode
= nir_var_local
;
327 case ir_var_shader_in
:
328 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
329 ir
->data
.location
== VARYING_SLOT_FACE
) {
330 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
331 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
332 var
->data
.mode
= nir_var_system_value
;
333 } else if (shader
->stage
== MESA_SHADER_GEOMETRY
&&
334 ir
->data
.location
== VARYING_SLOT_PRIMITIVE_ID
) {
335 /* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
336 var
->data
.location
= SYSTEM_VALUE_PRIMITIVE_ID
;
337 var
->data
.mode
= nir_var_system_value
;
339 var
->data
.mode
= nir_var_shader_in
;
343 case ir_var_shader_out
:
344 var
->data
.mode
= nir_var_shader_out
;
348 var
->data
.mode
= nir_var_uniform
;
351 case ir_var_shader_storage
:
352 var
->data
.mode
= nir_var_shader_storage
;
355 case ir_var_system_value
:
356 var
->data
.mode
= nir_var_system_value
;
360 unreachable("not reached");
363 var
->data
.interpolation
= ir
->data
.interpolation
;
364 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
365 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
366 var
->data
.explicit_location
= ir
->data
.explicit_location
;
367 var
->data
.explicit_index
= ir
->data
.explicit_index
;
368 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
369 var
->data
.has_initializer
= ir
->data
.has_initializer
;
370 var
->data
.location_frac
= ir
->data
.location_frac
;
371 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
372 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
374 switch (ir
->data
.depth_layout
) {
375 case ir_depth_layout_none
:
376 var
->data
.depth_layout
= nir_depth_layout_none
;
378 case ir_depth_layout_any
:
379 var
->data
.depth_layout
= nir_depth_layout_any
;
381 case ir_depth_layout_greater
:
382 var
->data
.depth_layout
= nir_depth_layout_greater
;
384 case ir_depth_layout_less
:
385 var
->data
.depth_layout
= nir_depth_layout_less
;
387 case ir_depth_layout_unchanged
:
388 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
391 unreachable("not reached");
394 var
->data
.index
= ir
->data
.index
;
395 var
->data
.descriptor_set
= 0;
396 var
->data
.binding
= ir
->data
.binding
;
397 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
398 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
399 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
400 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
401 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
402 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
403 var
->data
.image
.format
= ir
->data
.image_format
;
404 var
->data
.max_array_access
= ir
->data
.max_array_access
;
406 var
->num_state_slots
= ir
->get_num_state_slots();
407 if (var
->num_state_slots
> 0) {
408 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
409 var
->num_state_slots
);
411 ir_state_slot
*state_slots
= ir
->get_state_slots();
412 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
413 for (unsigned j
= 0; j
< 5; j
++)
414 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
415 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
418 var
->state_slots
= NULL
;
421 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
423 var
->interface_type
= ir
->get_interface_type();
425 if (var
->data
.mode
== nir_var_local
)
426 nir_function_impl_add_variable(impl
, var
);
428 nir_shader_add_variable(shader
, var
);
430 _mesa_hash_table_insert(var_table
, ir
, var
);
435 nir_function_visitor::visit_enter(ir_function
*ir
)
437 visitor
->create_function(ir
);
438 return visit_continue_with_parent
;
443 nir_visitor::create_function(ir_function
*ir
)
445 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
446 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
447 create_overload(sig
, func
);
454 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
456 if (ir
->is_intrinsic
)
459 nir_function_overload
*overload
= nir_function_overload_create(function
);
461 unsigned num_params
= ir
->parameters
.length();
462 overload
->num_params
= num_params
;
463 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
466 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
467 switch (param
->data
.mode
) {
468 case ir_var_function_in
:
469 overload
->params
[i
].param_type
= nir_parameter_in
;
472 case ir_var_function_out
:
473 overload
->params
[i
].param_type
= nir_parameter_out
;
476 case ir_var_function_inout
:
477 overload
->params
[i
].param_type
= nir_parameter_inout
;
481 unreachable("not reached");
484 overload
->params
[i
].type
= param
->type
;
488 overload
->return_type
= ir
->return_type
;
490 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
494 nir_visitor::visit(ir_function
*ir
)
496 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
501 nir_visitor::visit(ir_function_signature
*ir
)
503 if (ir
->is_intrinsic
)
506 struct hash_entry
*entry
=
507 _mesa_hash_table_search(this->overload_table
, ir
);
510 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
512 if (ir
->is_defined
) {
513 nir_function_impl
*impl
= nir_function_impl_create(overload
);
516 unsigned num_params
= overload
->num_params
;
517 impl
->num_params
= num_params
;
518 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
520 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
522 impl
->params
[i
] = this->var
;
526 if (overload
->return_type
== glsl_type::void_type
) {
527 impl
->return_var
= NULL
;
529 impl
->return_var
= ralloc(this->shader
, nir_variable
);
530 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
532 impl
->return_var
->type
= overload
->return_type
;
535 this->is_global
= false;
537 nir_builder_init(&b
, impl
);
538 b
.cursor
= nir_after_cf_list(&impl
->body
);
539 visit_exec_list(&ir
->body
, this);
541 this->is_global
= true;
543 overload
->impl
= NULL
;
548 nir_visitor::visit(ir_loop
*ir
)
550 nir_loop
*loop
= nir_loop_create(this->shader
);
551 nir_builder_cf_insert(&b
, &loop
->cf_node
);
553 b
.cursor
= nir_after_cf_list(&loop
->body
);
554 visit_exec_list(&ir
->body_instructions
, this);
555 b
.cursor
= nir_after_cf_node(&loop
->cf_node
);
559 nir_visitor::visit(ir_if
*ir
)
562 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
564 nir_if
*if_stmt
= nir_if_create(this->shader
);
565 if_stmt
->condition
= condition
;
566 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
568 b
.cursor
= nir_after_cf_list(&if_stmt
->then_list
);
569 visit_exec_list(&ir
->then_instructions
, this);
571 b
.cursor
= nir_after_cf_list(&if_stmt
->else_list
);
572 visit_exec_list(&ir
->else_instructions
, this);
574 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
578 nir_visitor::visit(ir_discard
*ir
)
581 * discards aren't treated as control flow, because before we lower them
582 * they can appear anywhere in the shader and the stuff after them may still
583 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
584 * discards will be immediately followed by a return.
587 nir_intrinsic_instr
*discard
;
589 discard
= nir_intrinsic_instr_create(this->shader
,
590 nir_intrinsic_discard_if
);
592 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
594 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
597 nir_builder_instr_insert(&b
, &discard
->instr
);
601 nir_visitor::visit(ir_emit_vertex
*ir
)
603 nir_intrinsic_instr
*instr
=
604 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
605 instr
->const_index
[0] = ir
->stream_id();
606 nir_builder_instr_insert(&b
, &instr
->instr
);
610 nir_visitor::visit(ir_end_primitive
*ir
)
612 nir_intrinsic_instr
*instr
=
613 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
614 instr
->const_index
[0] = ir
->stream_id();
615 nir_builder_instr_insert(&b
, &instr
->instr
);
619 nir_visitor::visit(ir_loop_jump
*ir
)
623 case ir_loop_jump::jump_break
:
624 type
= nir_jump_break
;
626 case ir_loop_jump::jump_continue
:
627 type
= nir_jump_continue
;
630 unreachable("not reached");
633 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
634 nir_builder_instr_insert(&b
, &instr
->instr
);
638 nir_visitor::visit(ir_return
*ir
)
640 if (ir
->value
!= NULL
) {
641 nir_intrinsic_instr
*copy
=
642 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
644 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
645 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
648 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
649 nir_builder_instr_insert(&b
, &instr
->instr
);
653 nir_visitor::visit(ir_call
*ir
)
655 if (ir
->callee
->is_intrinsic
) {
657 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
658 op
= nir_intrinsic_atomic_counter_read_var
;
659 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
660 op
= nir_intrinsic_atomic_counter_inc_var
;
661 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
662 op
= nir_intrinsic_atomic_counter_dec_var
;
663 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
664 op
= nir_intrinsic_image_load
;
665 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
666 op
= nir_intrinsic_image_store
;
667 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
668 op
= nir_intrinsic_image_atomic_add
;
669 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
670 op
= nir_intrinsic_image_atomic_min
;
671 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
672 op
= nir_intrinsic_image_atomic_max
;
673 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
674 op
= nir_intrinsic_image_atomic_and
;
675 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
676 op
= nir_intrinsic_image_atomic_or
;
677 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
678 op
= nir_intrinsic_image_atomic_xor
;
679 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
680 op
= nir_intrinsic_image_atomic_exchange
;
681 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
682 op
= nir_intrinsic_image_atomic_comp_swap
;
683 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
684 op
= nir_intrinsic_memory_barrier
;
685 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_size") == 0) {
686 op
= nir_intrinsic_image_size
;
687 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_samples") == 0) {
688 op
= nir_intrinsic_image_samples
;
689 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_ssbo") == 0) {
690 op
= nir_intrinsic_store_ssbo
;
691 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_ssbo") == 0) {
692 op
= nir_intrinsic_load_ssbo
;
693 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_add_ssbo") == 0) {
694 op
= nir_intrinsic_ssbo_atomic_add
;
695 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_and_ssbo") == 0) {
696 op
= nir_intrinsic_ssbo_atomic_and
;
697 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_or_ssbo") == 0) {
698 op
= nir_intrinsic_ssbo_atomic_or
;
699 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_xor_ssbo") == 0) {
700 op
= nir_intrinsic_ssbo_atomic_xor
;
701 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_min_ssbo") == 0) {
702 assert(ir
->return_deref
);
703 if (ir
->return_deref
->type
== glsl_type::int_type
)
704 op
= nir_intrinsic_ssbo_atomic_imin
;
705 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
706 op
= nir_intrinsic_ssbo_atomic_umin
;
708 unreachable("Invalid type");
709 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_max_ssbo") == 0) {
710 assert(ir
->return_deref
);
711 if (ir
->return_deref
->type
== glsl_type::int_type
)
712 op
= nir_intrinsic_ssbo_atomic_imax
;
713 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
714 op
= nir_intrinsic_ssbo_atomic_umax
;
716 unreachable("Invalid type");
717 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_exchange_ssbo") == 0) {
718 op
= nir_intrinsic_ssbo_atomic_exchange
;
719 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_comp_swap_ssbo") == 0) {
720 op
= nir_intrinsic_ssbo_atomic_comp_swap
;
721 } else if (strcmp(ir
->callee_name(), "__intrinsic_shader_clock") == 0) {
722 op
= nir_intrinsic_shader_clock
;
723 } else if (strcmp(ir
->callee_name(), "__intrinsic_group_memory_barrier") == 0) {
724 op
= nir_intrinsic_group_memory_barrier
;
725 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_atomic_counter") == 0) {
726 op
= nir_intrinsic_memory_barrier_atomic_counter
;
727 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_buffer") == 0) {
728 op
= nir_intrinsic_memory_barrier_buffer
;
729 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_image") == 0) {
730 op
= nir_intrinsic_memory_barrier_image
;
731 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_shared") == 0) {
732 op
= nir_intrinsic_memory_barrier_shared
;
733 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_shared") == 0) {
734 op
= nir_intrinsic_load_shared
;
735 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_shared") == 0) {
736 op
= nir_intrinsic_store_shared
;
737 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_add_shared") == 0) {
738 op
= nir_intrinsic_shared_atomic_add
;
739 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_and_shared") == 0) {
740 op
= nir_intrinsic_shared_atomic_and
;
741 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_or_shared") == 0) {
742 op
= nir_intrinsic_shared_atomic_or
;
743 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_xor_shared") == 0) {
744 op
= nir_intrinsic_shared_atomic_xor
;
745 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_min_shared") == 0) {
746 assert(ir
->return_deref
);
747 if (ir
->return_deref
->type
== glsl_type::int_type
)
748 op
= nir_intrinsic_shared_atomic_imin
;
749 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
750 op
= nir_intrinsic_shared_atomic_umin
;
752 unreachable("Invalid type");
753 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_max_shared") == 0) {
754 assert(ir
->return_deref
);
755 if (ir
->return_deref
->type
== glsl_type::int_type
)
756 op
= nir_intrinsic_shared_atomic_imax
;
757 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
758 op
= nir_intrinsic_shared_atomic_umax
;
760 unreachable("Invalid type");
761 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_exchange_shared") == 0) {
762 op
= nir_intrinsic_shared_atomic_exchange
;
763 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_comp_swap_shared") == 0) {
764 op
= nir_intrinsic_shared_atomic_comp_swap
;
766 unreachable("not reached");
769 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
770 nir_dest
*dest
= &instr
->dest
;
773 case nir_intrinsic_atomic_counter_read_var
:
774 case nir_intrinsic_atomic_counter_inc_var
:
775 case nir_intrinsic_atomic_counter_dec_var
: {
776 ir_dereference
*param
=
777 (ir_dereference
*) ir
->actual_parameters
.get_head();
778 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
779 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
780 nir_builder_instr_insert(&b
, &instr
->instr
);
783 case nir_intrinsic_image_load
:
784 case nir_intrinsic_image_store
:
785 case nir_intrinsic_image_atomic_add
:
786 case nir_intrinsic_image_atomic_min
:
787 case nir_intrinsic_image_atomic_max
:
788 case nir_intrinsic_image_atomic_and
:
789 case nir_intrinsic_image_atomic_or
:
790 case nir_intrinsic_image_atomic_xor
:
791 case nir_intrinsic_image_atomic_exchange
:
792 case nir_intrinsic_image_atomic_comp_swap
:
793 case nir_intrinsic_image_samples
:
794 case nir_intrinsic_image_size
: {
795 nir_ssa_undef_instr
*instr_undef
=
796 nir_ssa_undef_instr_create(shader
, 1);
797 nir_builder_instr_insert(&b
, &instr_undef
->instr
);
799 /* Set the image variable dereference. */
800 exec_node
*param
= ir
->actual_parameters
.get_head();
801 ir_dereference
*image
= (ir_dereference
*)param
;
802 const glsl_type
*type
=
803 image
->variable_referenced()->type
->without_array();
805 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
806 param
= param
->get_next();
808 /* Set the intrinsic destination. */
809 if (ir
->return_deref
) {
810 const nir_intrinsic_info
*info
=
811 &nir_intrinsic_infos
[instr
->intrinsic
];
812 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
813 info
->dest_components
, NULL
);
816 if (op
== nir_intrinsic_image_size
||
817 op
== nir_intrinsic_image_samples
) {
818 nir_builder_instr_insert(&b
, &instr
->instr
);
822 /* Set the address argument, extending the coordinate vector to four
825 nir_ssa_def
*src_addr
=
826 evaluate_rvalue((ir_dereference
*)param
);
827 nir_ssa_def
*srcs
[4];
829 for (int i
= 0; i
< 4; i
++) {
830 if (i
< type
->coordinate_components())
831 srcs
[i
] = nir_channel(&b
, src_addr
, i
);
833 srcs
[i
] = &instr_undef
->def
;
836 instr
->src
[0] = nir_src_for_ssa(nir_vec(&b
, srcs
, 4));
837 param
= param
->get_next();
839 /* Set the sample argument, which is undefined for single-sample
842 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
844 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
845 param
= param
->get_next();
847 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
850 /* Set the intrinsic parameters. */
851 if (!param
->is_tail_sentinel()) {
853 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
854 param
= param
->get_next();
857 if (!param
->is_tail_sentinel()) {
859 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
860 param
= param
->get_next();
862 nir_builder_instr_insert(&b
, &instr
->instr
);
865 case nir_intrinsic_memory_barrier
:
866 case nir_intrinsic_group_memory_barrier
:
867 case nir_intrinsic_memory_barrier_atomic_counter
:
868 case nir_intrinsic_memory_barrier_buffer
:
869 case nir_intrinsic_memory_barrier_image
:
870 case nir_intrinsic_memory_barrier_shared
:
871 nir_builder_instr_insert(&b
, &instr
->instr
);
873 case nir_intrinsic_shader_clock
:
874 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
875 nir_builder_instr_insert(&b
, &instr
->instr
);
877 case nir_intrinsic_store_ssbo
: {
878 exec_node
*param
= ir
->actual_parameters
.get_head();
879 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
881 param
= param
->get_next();
882 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
884 param
= param
->get_next();
885 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
887 param
= param
->get_next();
888 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
891 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
892 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(block
));
893 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(offset
));
894 instr
->const_index
[0] = write_mask
->value
.u
[0];
895 instr
->num_components
= val
->type
->vector_elements
;
897 nir_builder_instr_insert(&b
, &instr
->instr
);
900 case nir_intrinsic_load_ssbo
: {
901 exec_node
*param
= ir
->actual_parameters
.get_head();
902 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
904 param
= param
->get_next();
905 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
907 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(block
));
908 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
910 const glsl_type
*type
= ir
->return_deref
->var
->type
;
911 instr
->num_components
= type
->vector_elements
;
913 /* Setup destination register */
914 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
915 type
->vector_elements
, NULL
);
917 /* Insert the created nir instruction now since in the case of boolean
918 * result we will need to emit another instruction after it
920 nir_builder_instr_insert(&b
, &instr
->instr
);
923 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
924 * consider a true boolean to be ~0. Fix this up with a != 0
927 if (type
->base_type
== GLSL_TYPE_BOOL
) {
928 nir_alu_instr
*load_ssbo_compare
=
929 nir_alu_instr_create(shader
, nir_op_ine
);
930 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
931 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
932 load_ssbo_compare
->src
[1].src
=
933 nir_src_for_ssa(nir_imm_int(&b
, 0));
934 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
935 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
936 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
937 &load_ssbo_compare
->dest
.dest
,
938 type
->vector_elements
, NULL
);
939 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
940 nir_builder_instr_insert(&b
, &load_ssbo_compare
->instr
);
941 dest
= &load_ssbo_compare
->dest
.dest
;
945 case nir_intrinsic_ssbo_atomic_add
:
946 case nir_intrinsic_ssbo_atomic_imin
:
947 case nir_intrinsic_ssbo_atomic_umin
:
948 case nir_intrinsic_ssbo_atomic_imax
:
949 case nir_intrinsic_ssbo_atomic_umax
:
950 case nir_intrinsic_ssbo_atomic_and
:
951 case nir_intrinsic_ssbo_atomic_or
:
952 case nir_intrinsic_ssbo_atomic_xor
:
953 case nir_intrinsic_ssbo_atomic_exchange
:
954 case nir_intrinsic_ssbo_atomic_comp_swap
: {
955 int param_count
= ir
->actual_parameters
.length();
956 assert(param_count
== 3 || param_count
== 4);
959 exec_node
*param
= ir
->actual_parameters
.get_head();
960 ir_instruction
*inst
= (ir_instruction
*) param
;
961 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
964 param
= param
->get_next();
965 inst
= (ir_instruction
*) param
;
966 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
968 /* data1 parameter (this is always present) */
969 param
= param
->get_next();
970 inst
= (ir_instruction
*) param
;
971 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
973 /* data2 parameter (only with atomic_comp_swap) */
974 if (param_count
== 4) {
975 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
976 param
= param
->get_next();
977 inst
= (ir_instruction
*) param
;
978 instr
->src
[3] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
982 assert(ir
->return_deref
);
983 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
984 ir
->return_deref
->type
->vector_elements
, NULL
);
985 nir_builder_instr_insert(&b
, &instr
->instr
);
988 case nir_intrinsic_load_shared
: {
989 exec_node
*param
= ir
->actual_parameters
.get_head();
990 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
992 instr
->const_index
[0] = 0;
993 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(offset
));
995 const glsl_type
*type
= ir
->return_deref
->var
->type
;
996 instr
->num_components
= type
->vector_elements
;
998 /* Setup destination register */
999 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1000 type
->vector_elements
, NULL
);
1002 nir_builder_instr_insert(&b
, &instr
->instr
);
1005 case nir_intrinsic_store_shared
: {
1006 exec_node
*param
= ir
->actual_parameters
.get_head();
1007 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1009 param
= param
->get_next();
1010 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
1012 param
= param
->get_next();
1013 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
1016 instr
->const_index
[0] = 0;
1017 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
1019 instr
->const_index
[1] = write_mask
->value
.u
[0];
1021 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
1022 instr
->num_components
= val
->type
->vector_elements
;
1024 nir_builder_instr_insert(&b
, &instr
->instr
);
1027 case nir_intrinsic_shared_atomic_add
:
1028 case nir_intrinsic_shared_atomic_imin
:
1029 case nir_intrinsic_shared_atomic_umin
:
1030 case nir_intrinsic_shared_atomic_imax
:
1031 case nir_intrinsic_shared_atomic_umax
:
1032 case nir_intrinsic_shared_atomic_and
:
1033 case nir_intrinsic_shared_atomic_or
:
1034 case nir_intrinsic_shared_atomic_xor
:
1035 case nir_intrinsic_shared_atomic_exchange
:
1036 case nir_intrinsic_shared_atomic_comp_swap
: {
1037 int param_count
= ir
->actual_parameters
.length();
1038 assert(param_count
== 2 || param_count
== 3);
1041 exec_node
*param
= ir
->actual_parameters
.get_head();
1042 ir_instruction
*inst
= (ir_instruction
*) param
;
1043 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1045 /* data1 parameter (this is always present) */
1046 param
= param
->get_next();
1047 inst
= (ir_instruction
*) param
;
1048 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1050 /* data2 parameter (only with atomic_comp_swap) */
1051 if (param_count
== 3) {
1052 assert(op
== nir_intrinsic_shared_atomic_comp_swap
);
1053 param
= param
->get_next();
1054 inst
= (ir_instruction
*) param
;
1056 nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1060 assert(ir
->return_deref
);
1061 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1062 ir
->return_deref
->type
->vector_elements
, NULL
);
1063 nir_builder_instr_insert(&b
, &instr
->instr
);
1067 unreachable("not reached");
1070 if (ir
->return_deref
) {
1071 nir_intrinsic_instr
*store_instr
=
1072 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
1073 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
1074 store_instr
->const_index
[0] = (1 << store_instr
->num_components
) - 1;
1076 store_instr
->variables
[0] =
1077 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
1078 store_instr
->src
[0] = nir_src_for_ssa(&dest
->ssa
);
1080 nir_builder_instr_insert(&b
, &store_instr
->instr
);
1086 struct hash_entry
*entry
=
1087 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
1089 nir_function_overload
*callee
= (nir_function_overload
*) entry
->data
;
1091 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
1094 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
1095 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
1099 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
1100 nir_builder_instr_insert(&b
, &instr
->instr
);
1104 nir_visitor::visit(ir_assignment
*ir
)
1106 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
1108 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
1109 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
1110 /* We're doing a plain-as-can-be copy, so emit a copy_var */
1111 nir_intrinsic_instr
*copy
=
1112 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
1114 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
1115 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
1117 if (ir
->condition
) {
1118 nir_if
*if_stmt
= nir_if_create(this->shader
);
1119 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1120 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1121 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
1122 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1124 nir_builder_instr_insert(&b
, ©
->instr
);
1129 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
1131 ir
->lhs
->accept(this);
1132 nir_deref_var
*lhs_deref
= this->deref_head
;
1133 nir_ssa_def
*src
= evaluate_rvalue(ir
->rhs
);
1135 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
1136 /* GLSL IR will give us the input to the write-masked assignment in a
1137 * single packed vector. So, for example, if the writemask is xzw, then
1138 * we have to swizzle x -> x, y -> z, and z -> w and get the y component
1142 unsigned component
= 0;
1143 for (unsigned i
= 0; i
< 4; i
++) {
1144 swiz
[i
] = ir
->write_mask
& (1 << i
) ? component
++ : 0;
1146 src
= nir_swizzle(&b
, src
, swiz
, num_components
, !supports_ints
);
1149 nir_intrinsic_instr
*store
=
1150 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1151 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1152 store
->const_index
[0] = ir
->write_mask
;
1153 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
1154 store
->variables
[0] = nir_deref_as_var(store_deref
);
1155 store
->src
[0] = nir_src_for_ssa(src
);
1157 if (ir
->condition
) {
1158 nir_if
*if_stmt
= nir_if_create(this->shader
);
1159 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1160 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1161 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
1162 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1164 nir_builder_instr_insert(&b
, &store
->instr
);
1169 * Given an instruction, returns a pointer to its destination or NULL if there
1170 * is no destination.
1172 * Note that this only handles instructions we generate at this level.
1175 get_instr_dest(nir_instr
*instr
)
1177 nir_alu_instr
*alu_instr
;
1178 nir_intrinsic_instr
*intrinsic_instr
;
1179 nir_tex_instr
*tex_instr
;
1181 switch (instr
->type
) {
1182 case nir_instr_type_alu
:
1183 alu_instr
= nir_instr_as_alu(instr
);
1184 return &alu_instr
->dest
.dest
;
1186 case nir_instr_type_intrinsic
:
1187 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1188 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1189 return &intrinsic_instr
->dest
;
1193 case nir_instr_type_tex
:
1194 tex_instr
= nir_instr_as_tex(instr
);
1195 return &tex_instr
->dest
;
1198 unreachable("not reached");
1205 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
1207 nir_dest
*dest
= get_instr_dest(instr
);
1210 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
1212 nir_builder_instr_insert(&b
, instr
);
1215 assert(dest
->is_ssa
);
1216 this->result
= &dest
->ssa
;
1221 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1224 if (ir
->as_dereference() || ir
->as_constant()) {
1226 * A dereference is being used on the right hand side, which means we
1227 * must emit a variable load.
1230 nir_intrinsic_instr
*load_instr
=
1231 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1232 load_instr
->num_components
= ir
->type
->vector_elements
;
1233 load_instr
->variables
[0] = this->deref_head
;
1234 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1235 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
1238 return this->result
;
1242 nir_visitor::visit(ir_expression
*ir
)
1244 /* Some special cases */
1245 switch (ir
->operation
) {
1246 case ir_binop_ubo_load
: {
1247 nir_intrinsic_instr
*load
=
1248 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_ubo
);
1249 load
->num_components
= ir
->type
->vector_elements
;
1250 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1251 load
->src
[1] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1252 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1255 * In UBO's, a true boolean value is any non-zero value, but we consider
1256 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1259 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
)
1260 this->result
= nir_ine(&b
, &load
->dest
.ssa
, nir_imm_int(&b
, 0));
1265 case ir_unop_interpolate_at_centroid
:
1266 case ir_binop_interpolate_at_offset
:
1267 case ir_binop_interpolate_at_sample
: {
1268 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1269 ir_swizzle
*swizzle
= NULL
;
1271 /* the api does not allow a swizzle here, but the varying packing code
1272 * may have pushed one into here.
1274 swizzle
= ir
->operands
[0]->as_swizzle();
1276 deref
= swizzle
->val
->as_dereference();
1280 deref
->accept(this);
1282 nir_intrinsic_op op
;
1283 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1284 switch (ir
->operation
) {
1285 case ir_unop_interpolate_at_centroid
:
1286 op
= nir_intrinsic_interp_var_at_centroid
;
1288 case ir_binop_interpolate_at_offset
:
1289 op
= nir_intrinsic_interp_var_at_offset
;
1291 case ir_binop_interpolate_at_sample
:
1292 op
= nir_intrinsic_interp_var_at_sample
;
1295 unreachable("Invalid interpolation intrinsic");
1298 /* This case can happen if the vertex shader does not write the
1299 * given varying. In this case, the linker will lower it to a
1300 * global variable. Since interpolating a variable makes no
1301 * sense, we'll just turn it into a load which will probably
1302 * eventually end up as an SSA definition.
1304 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1305 op
= nir_intrinsic_load_var
;
1308 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1309 intrin
->num_components
= deref
->type
->vector_elements
;
1310 intrin
->variables
[0] = this->deref_head
;
1311 ralloc_steal(intrin
, intrin
->variables
[0]);
1313 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1314 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1315 intrin
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1317 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
1320 unsigned swiz
[4] = {
1321 swizzle
->mask
.x
, swizzle
->mask
.y
, swizzle
->mask
.z
, swizzle
->mask
.w
1324 result
= nir_swizzle(&b
, result
, swiz
,
1325 swizzle
->type
->vector_elements
, false);
1335 nir_ssa_def
*srcs
[4];
1336 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1337 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1339 glsl_base_type types
[4];
1340 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1342 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1344 types
[i
] = GLSL_TYPE_FLOAT
;
1346 glsl_base_type out_type
;
1348 out_type
= ir
->type
->base_type
;
1350 out_type
= GLSL_TYPE_FLOAT
;
1352 switch (ir
->operation
) {
1353 case ir_unop_bit_not
: result
= nir_inot(&b
, srcs
[0]); break;
1354 case ir_unop_logic_not
:
1355 result
= supports_ints
? nir_inot(&b
, srcs
[0]) : nir_fnot(&b
, srcs
[0]);
1358 result
= (types
[0] == GLSL_TYPE_FLOAT
) ? nir_fneg(&b
, srcs
[0])
1359 : nir_ineg(&b
, srcs
[0]);
1362 result
= (types
[0] == GLSL_TYPE_FLOAT
) ? nir_fabs(&b
, srcs
[0])
1363 : nir_iabs(&b
, srcs
[0]);
1365 case ir_unop_saturate
:
1366 assert(types
[0] == GLSL_TYPE_FLOAT
);
1367 result
= nir_fsat(&b
, srcs
[0]);
1370 result
= (types
[0] == GLSL_TYPE_FLOAT
) ? nir_fsign(&b
, srcs
[0])
1371 : nir_isign(&b
, srcs
[0]);
1373 case ir_unop_rcp
: result
= nir_frcp(&b
, srcs
[0]); break;
1374 case ir_unop_rsq
: result
= nir_frsq(&b
, srcs
[0]); break;
1375 case ir_unop_sqrt
: result
= nir_fsqrt(&b
, srcs
[0]); break;
1376 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1377 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1378 case ir_unop_exp2
: result
= nir_fexp2(&b
, srcs
[0]); break;
1379 case ir_unop_log2
: result
= nir_flog2(&b
, srcs
[0]); break;
1381 result
= supports_ints
? nir_i2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1384 result
= supports_ints
? nir_u2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1387 result
= supports_ints
? nir_b2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1389 case ir_unop_f2i
: result
= nir_f2i(&b
, srcs
[0]); break;
1390 case ir_unop_f2u
: result
= nir_f2u(&b
, srcs
[0]); break;
1391 case ir_unop_f2b
: result
= nir_f2b(&b
, srcs
[0]); break;
1392 case ir_unop_i2b
: result
= nir_i2b(&b
, srcs
[0]); break;
1393 case ir_unop_b2i
: result
= nir_b2i(&b
, srcs
[0]); break;
1396 case ir_unop_bitcast_i2f
:
1397 case ir_unop_bitcast_f2i
:
1398 case ir_unop_bitcast_u2f
:
1399 case ir_unop_bitcast_f2u
:
1400 case ir_unop_subroutine_to_int
:
1402 result
= nir_imov(&b
, srcs
[0]);
1404 case ir_unop_trunc
: result
= nir_ftrunc(&b
, srcs
[0]); break;
1405 case ir_unop_ceil
: result
= nir_fceil(&b
, srcs
[0]); break;
1406 case ir_unop_floor
: result
= nir_ffloor(&b
, srcs
[0]); break;
1407 case ir_unop_fract
: result
= nir_ffract(&b
, srcs
[0]); break;
1408 case ir_unop_round_even
: result
= nir_fround_even(&b
, srcs
[0]); break;
1409 case ir_unop_sin
: result
= nir_fsin(&b
, srcs
[0]); break;
1410 case ir_unop_cos
: result
= nir_fcos(&b
, srcs
[0]); break;
1411 case ir_unop_dFdx
: result
= nir_fddx(&b
, srcs
[0]); break;
1412 case ir_unop_dFdy
: result
= nir_fddy(&b
, srcs
[0]); break;
1413 case ir_unop_dFdx_fine
: result
= nir_fddx_fine(&b
, srcs
[0]); break;
1414 case ir_unop_dFdy_fine
: result
= nir_fddy_fine(&b
, srcs
[0]); break;
1415 case ir_unop_dFdx_coarse
: result
= nir_fddx_coarse(&b
, srcs
[0]); break;
1416 case ir_unop_dFdy_coarse
: result
= nir_fddy_coarse(&b
, srcs
[0]); break;
1417 case ir_unop_pack_snorm_2x16
:
1418 result
= nir_pack_snorm_2x16(&b
, srcs
[0]);
1420 case ir_unop_pack_snorm_4x8
:
1421 result
= nir_pack_snorm_4x8(&b
, srcs
[0]);
1423 case ir_unop_pack_unorm_2x16
:
1424 result
= nir_pack_unorm_2x16(&b
, srcs
[0]);
1426 case ir_unop_pack_unorm_4x8
:
1427 result
= nir_pack_unorm_4x8(&b
, srcs
[0]);
1429 case ir_unop_pack_half_2x16
:
1430 result
= nir_pack_half_2x16(&b
, srcs
[0]);
1432 case ir_unop_unpack_snorm_2x16
:
1433 result
= nir_unpack_snorm_2x16(&b
, srcs
[0]);
1435 case ir_unop_unpack_snorm_4x8
:
1436 result
= nir_unpack_snorm_4x8(&b
, srcs
[0]);
1438 case ir_unop_unpack_unorm_2x16
:
1439 result
= nir_unpack_unorm_2x16(&b
, srcs
[0]);
1441 case ir_unop_unpack_unorm_4x8
:
1442 result
= nir_unpack_unorm_4x8(&b
, srcs
[0]);
1444 case ir_unop_unpack_half_2x16
:
1445 result
= nir_unpack_half_2x16(&b
, srcs
[0]);
1447 case ir_unop_unpack_half_2x16_split_x
:
1448 result
= nir_unpack_half_2x16_split_x(&b
, srcs
[0]);
1450 case ir_unop_unpack_half_2x16_split_y
:
1451 result
= nir_unpack_half_2x16_split_y(&b
, srcs
[0]);
1453 case ir_unop_bitfield_reverse
:
1454 result
= nir_bitfield_reverse(&b
, srcs
[0]);
1456 case ir_unop_bit_count
:
1457 result
= nir_bit_count(&b
, srcs
[0]);
1459 case ir_unop_find_msb
:
1461 case GLSL_TYPE_UINT
:
1462 result
= nir_ufind_msb(&b
, srcs
[0]);
1465 result
= nir_ifind_msb(&b
, srcs
[0]);
1468 unreachable("Invalid type for findMSB()");
1471 case ir_unop_find_lsb
:
1472 result
= nir_find_lsb(&b
, srcs
[0]);
1476 switch (ir
->type
->vector_elements
) {
1478 switch (ir
->operands
[0]->type
->vector_elements
) {
1479 case 1: result
= nir_fnoise1_1(&b
, srcs
[0]); break;
1480 case 2: result
= nir_fnoise1_2(&b
, srcs
[0]); break;
1481 case 3: result
= nir_fnoise1_3(&b
, srcs
[0]); break;
1482 case 4: result
= nir_fnoise1_4(&b
, srcs
[0]); break;
1483 default: unreachable("not reached");
1487 switch (ir
->operands
[0]->type
->vector_elements
) {
1488 case 1: result
= nir_fnoise2_1(&b
, srcs
[0]); break;
1489 case 2: result
= nir_fnoise2_2(&b
, srcs
[0]); break;
1490 case 3: result
= nir_fnoise2_3(&b
, srcs
[0]); break;
1491 case 4: result
= nir_fnoise2_4(&b
, srcs
[0]); break;
1492 default: unreachable("not reached");
1496 switch (ir
->operands
[0]->type
->vector_elements
) {
1497 case 1: result
= nir_fnoise3_1(&b
, srcs
[0]); break;
1498 case 2: result
= nir_fnoise3_2(&b
, srcs
[0]); break;
1499 case 3: result
= nir_fnoise3_3(&b
, srcs
[0]); break;
1500 case 4: result
= nir_fnoise3_4(&b
, srcs
[0]); break;
1501 default: unreachable("not reached");
1505 switch (ir
->operands
[0]->type
->vector_elements
) {
1506 case 1: result
= nir_fnoise4_1(&b
, srcs
[0]); break;
1507 case 2: result
= nir_fnoise4_2(&b
, srcs
[0]); break;
1508 case 3: result
= nir_fnoise4_3(&b
, srcs
[0]); break;
1509 case 4: result
= nir_fnoise4_4(&b
, srcs
[0]); break;
1510 default: unreachable("not reached");
1514 unreachable("not reached");
1517 case ir_unop_get_buffer_size
: {
1518 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1520 nir_intrinsic_get_buffer_size
);
1521 load
->num_components
= ir
->type
->vector_elements
;
1522 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1523 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1528 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fadd(&b
, srcs
[0], srcs
[1])
1529 : nir_iadd(&b
, srcs
[0], srcs
[1]);
1532 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fsub(&b
, srcs
[0], srcs
[1])
1533 : nir_isub(&b
, srcs
[0], srcs
[1]);
1536 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fmul(&b
, srcs
[0], srcs
[1])
1537 : nir_imul(&b
, srcs
[0], srcs
[1]);
1540 if (out_type
== GLSL_TYPE_FLOAT
)
1541 result
= nir_fdiv(&b
, srcs
[0], srcs
[1]);
1542 else if (out_type
== GLSL_TYPE_INT
)
1543 result
= nir_idiv(&b
, srcs
[0], srcs
[1]);
1545 result
= nir_udiv(&b
, srcs
[0], srcs
[1]);
1548 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fmod(&b
, srcs
[0], srcs
[1])
1549 : nir_umod(&b
, srcs
[0], srcs
[1]);
1552 if (out_type
== GLSL_TYPE_FLOAT
)
1553 result
= nir_fmin(&b
, srcs
[0], srcs
[1]);
1554 else if (out_type
== GLSL_TYPE_INT
)
1555 result
= nir_imin(&b
, srcs
[0], srcs
[1]);
1557 result
= nir_umin(&b
, srcs
[0], srcs
[1]);
1560 if (out_type
== GLSL_TYPE_FLOAT
)
1561 result
= nir_fmax(&b
, srcs
[0], srcs
[1]);
1562 else if (out_type
== GLSL_TYPE_INT
)
1563 result
= nir_imax(&b
, srcs
[0], srcs
[1]);
1565 result
= nir_umax(&b
, srcs
[0], srcs
[1]);
1567 case ir_binop_pow
: result
= nir_fpow(&b
, srcs
[0], srcs
[1]); break;
1568 case ir_binop_bit_and
: result
= nir_iand(&b
, srcs
[0], srcs
[1]); break;
1569 case ir_binop_bit_or
: result
= nir_ior(&b
, srcs
[0], srcs
[1]); break;
1570 case ir_binop_bit_xor
: result
= nir_ixor(&b
, srcs
[0], srcs
[1]); break;
1571 case ir_binop_logic_and
:
1572 result
= supports_ints
? nir_iand(&b
, srcs
[0], srcs
[1])
1573 : nir_fand(&b
, srcs
[0], srcs
[1]);
1575 case ir_binop_logic_or
:
1576 result
= supports_ints
? nir_ior(&b
, srcs
[0], srcs
[1])
1577 : nir_for(&b
, srcs
[0], srcs
[1]);
1579 case ir_binop_logic_xor
:
1580 result
= supports_ints
? nir_ixor(&b
, srcs
[0], srcs
[1])
1581 : nir_fxor(&b
, srcs
[0], srcs
[1]);
1583 case ir_binop_lshift
: result
= nir_ishl(&b
, srcs
[0], srcs
[1]); break;
1584 case ir_binop_rshift
:
1585 result
= (out_type
== GLSL_TYPE_INT
) ? nir_ishr(&b
, srcs
[0], srcs
[1])
1586 : nir_ushr(&b
, srcs
[0], srcs
[1]);
1588 case ir_binop_imul_high
:
1589 result
= (out_type
== GLSL_TYPE_INT
) ? nir_imul_high(&b
, srcs
[0], srcs
[1])
1590 : nir_umul_high(&b
, srcs
[0], srcs
[1]);
1592 case ir_binop_carry
: result
= nir_uadd_carry(&b
, srcs
[0], srcs
[1]); break;
1593 case ir_binop_borrow
: result
= nir_usub_borrow(&b
, srcs
[0], srcs
[1]); break;
1595 if (supports_ints
) {
1596 if (types
[0] == GLSL_TYPE_FLOAT
)
1597 result
= nir_flt(&b
, srcs
[0], srcs
[1]);
1598 else if (types
[0] == GLSL_TYPE_INT
)
1599 result
= nir_ilt(&b
, srcs
[0], srcs
[1]);
1601 result
= nir_ult(&b
, srcs
[0], srcs
[1]);
1603 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1606 case ir_binop_greater
:
1607 if (supports_ints
) {
1608 if (types
[0] == GLSL_TYPE_FLOAT
)
1609 result
= nir_flt(&b
, srcs
[1], srcs
[0]);
1610 else if (types
[0] == GLSL_TYPE_INT
)
1611 result
= nir_ilt(&b
, srcs
[1], srcs
[0]);
1613 result
= nir_ult(&b
, srcs
[1], srcs
[0]);
1615 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1618 case ir_binop_lequal
:
1619 if (supports_ints
) {
1620 if (types
[0] == GLSL_TYPE_FLOAT
)
1621 result
= nir_fge(&b
, srcs
[1], srcs
[0]);
1622 else if (types
[0] == GLSL_TYPE_INT
)
1623 result
= nir_ige(&b
, srcs
[1], srcs
[0]);
1625 result
= nir_uge(&b
, srcs
[1], srcs
[0]);
1627 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1630 case ir_binop_gequal
:
1631 if (supports_ints
) {
1632 if (types
[0] == GLSL_TYPE_FLOAT
)
1633 result
= nir_fge(&b
, srcs
[0], srcs
[1]);
1634 else if (types
[0] == GLSL_TYPE_INT
)
1635 result
= nir_ige(&b
, srcs
[0], srcs
[1]);
1637 result
= nir_uge(&b
, srcs
[0], srcs
[1]);
1639 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1642 case ir_binop_equal
:
1643 if (supports_ints
) {
1644 if (types
[0] == GLSL_TYPE_FLOAT
)
1645 result
= nir_feq(&b
, srcs
[0], srcs
[1]);
1647 result
= nir_ieq(&b
, srcs
[0], srcs
[1]);
1649 result
= nir_seq(&b
, srcs
[0], srcs
[1]);
1652 case ir_binop_nequal
:
1653 if (supports_ints
) {
1654 if (types
[0] == GLSL_TYPE_FLOAT
)
1655 result
= nir_fne(&b
, srcs
[0], srcs
[1]);
1657 result
= nir_ine(&b
, srcs
[0], srcs
[1]);
1659 result
= nir_sne(&b
, srcs
[0], srcs
[1]);
1662 case ir_binop_all_equal
:
1663 if (supports_ints
) {
1664 if (types
[0] == GLSL_TYPE_FLOAT
) {
1665 switch (ir
->operands
[0]->type
->vector_elements
) {
1666 case 1: result
= nir_feq(&b
, srcs
[0], srcs
[1]); break;
1667 case 2: result
= nir_ball_fequal2(&b
, srcs
[0], srcs
[1]); break;
1668 case 3: result
= nir_ball_fequal3(&b
, srcs
[0], srcs
[1]); break;
1669 case 4: result
= nir_ball_fequal4(&b
, srcs
[0], srcs
[1]); break;
1671 unreachable("not reached");
1674 switch (ir
->operands
[0]->type
->vector_elements
) {
1675 case 1: result
= nir_ieq(&b
, srcs
[0], srcs
[1]); break;
1676 case 2: result
= nir_ball_iequal2(&b
, srcs
[0], srcs
[1]); break;
1677 case 3: result
= nir_ball_iequal3(&b
, srcs
[0], srcs
[1]); break;
1678 case 4: result
= nir_ball_iequal4(&b
, srcs
[0], srcs
[1]); break;
1680 unreachable("not reached");
1684 switch (ir
->operands
[0]->type
->vector_elements
) {
1685 case 1: result
= nir_seq(&b
, srcs
[0], srcs
[1]); break;
1686 case 2: result
= nir_fall_equal2(&b
, srcs
[0], srcs
[1]); break;
1687 case 3: result
= nir_fall_equal3(&b
, srcs
[0], srcs
[1]); break;
1688 case 4: result
= nir_fall_equal4(&b
, srcs
[0], srcs
[1]); break;
1690 unreachable("not reached");
1694 case ir_binop_any_nequal
:
1695 if (supports_ints
) {
1696 if (types
[0] == GLSL_TYPE_FLOAT
) {
1697 switch (ir
->operands
[0]->type
->vector_elements
) {
1698 case 1: result
= nir_fne(&b
, srcs
[0], srcs
[1]); break;
1699 case 2: result
= nir_bany_fnequal2(&b
, srcs
[0], srcs
[1]); break;
1700 case 3: result
= nir_bany_fnequal3(&b
, srcs
[0], srcs
[1]); break;
1701 case 4: result
= nir_bany_fnequal4(&b
, srcs
[0], srcs
[1]); break;
1703 unreachable("not reached");
1706 switch (ir
->operands
[0]->type
->vector_elements
) {
1707 case 1: result
= nir_ine(&b
, srcs
[0], srcs
[1]); break;
1708 case 2: result
= nir_bany_inequal2(&b
, srcs
[0], srcs
[1]); break;
1709 case 3: result
= nir_bany_inequal3(&b
, srcs
[0], srcs
[1]); break;
1710 case 4: result
= nir_bany_inequal4(&b
, srcs
[0], srcs
[1]); break;
1712 unreachable("not reached");
1716 switch (ir
->operands
[0]->type
->vector_elements
) {
1717 case 1: result
= nir_sne(&b
, srcs
[0], srcs
[1]); break;
1718 case 2: result
= nir_fany_nequal2(&b
, srcs
[0], srcs
[1]); break;
1719 case 3: result
= nir_fany_nequal3(&b
, srcs
[0], srcs
[1]); break;
1720 case 4: result
= nir_fany_nequal4(&b
, srcs
[0], srcs
[1]); break;
1722 unreachable("not reached");
1727 switch (ir
->operands
[0]->type
->vector_elements
) {
1728 case 2: result
= nir_fdot2(&b
, srcs
[0], srcs
[1]); break;
1729 case 3: result
= nir_fdot3(&b
, srcs
[0], srcs
[1]); break;
1730 case 4: result
= nir_fdot4(&b
, srcs
[0], srcs
[1]); break;
1732 unreachable("not reached");
1736 case ir_binop_pack_half_2x16_split
:
1737 result
= nir_pack_half_2x16_split(&b
, srcs
[0], srcs
[1]);
1739 case ir_binop_bfm
: result
= nir_bfm(&b
, srcs
[0], srcs
[1]); break;
1740 case ir_binop_ldexp
: result
= nir_ldexp(&b
, srcs
[0], srcs
[1]); break;
1742 result
= nir_ffma(&b
, srcs
[0], srcs
[1], srcs
[2]);
1745 result
= nir_flrp(&b
, srcs
[0], srcs
[1], srcs
[2]);
1749 result
= nir_bcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1751 result
= nir_fcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1754 result
= nir_bfi(&b
, srcs
[0], srcs
[1], srcs
[2]);
1756 case ir_triop_bitfield_extract
:
1757 result
= (out_type
== GLSL_TYPE_INT
) ?
1758 nir_ibitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]) :
1759 nir_ubitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]);
1761 case ir_quadop_bitfield_insert
:
1762 result
= nir_bitfield_insert(&b
, srcs
[0], srcs
[1], srcs
[2], srcs
[3]);
1764 case ir_quadop_vector
:
1765 result
= nir_vec(&b
, srcs
, ir
->type
->vector_elements
);
1769 unreachable("not reached");
1774 nir_visitor::visit(ir_swizzle
*ir
)
1776 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1777 result
= nir_swizzle(&b
, evaluate_rvalue(ir
->val
), swizzle
,
1778 ir
->type
->vector_elements
, !supports_ints
);
1782 nir_visitor::visit(ir_texture
*ir
)
1789 num_srcs
= 1; /* coordinate */
1794 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1795 num_srcs
= 2; /* coordinate, bias/lod */
1799 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1805 if (ir
->lod_info
.lod
!= NULL
)
1806 num_srcs
= 2; /* coordinate, lod */
1808 num_srcs
= 1; /* coordinate */
1812 op
= nir_texop_txf_ms
;
1813 num_srcs
= 2; /* coordinate, sample_index */
1818 if (ir
->lod_info
.lod
!= NULL
)
1819 num_srcs
= 1; /* lod */
1826 num_srcs
= 1; /* coordinate */
1831 num_srcs
= 1; /* coordinate */
1834 case ir_query_levels
:
1835 op
= nir_texop_query_levels
;
1839 case ir_texture_samples
:
1840 op
= nir_texop_texture_samples
;
1844 case ir_samples_identical
:
1845 op
= nir_texop_samples_identical
;
1846 num_srcs
= 1; /* coordinate */
1850 unreachable("not reached");
1853 if (ir
->projector
!= NULL
)
1855 if (ir
->shadow_comparitor
!= NULL
)
1857 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1860 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1863 instr
->sampler_dim
=
1864 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1865 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1866 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1867 if (instr
->is_shadow
)
1868 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1869 switch (ir
->type
->base_type
) {
1870 case GLSL_TYPE_FLOAT
:
1871 instr
->dest_type
= nir_type_float
;
1874 instr
->dest_type
= nir_type_int
;
1876 case GLSL_TYPE_BOOL
:
1877 case GLSL_TYPE_UINT
:
1878 instr
->dest_type
= nir_type_uint
;
1881 unreachable("not reached");
1884 instr
->sampler
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1886 unsigned src_number
= 0;
1888 if (ir
->coordinate
!= NULL
) {
1889 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1890 instr
->src
[src_number
].src
=
1891 nir_src_for_ssa(evaluate_rvalue(ir
->coordinate
));
1892 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1896 if (ir
->projector
!= NULL
) {
1897 instr
->src
[src_number
].src
=
1898 nir_src_for_ssa(evaluate_rvalue(ir
->projector
));
1899 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1903 if (ir
->shadow_comparitor
!= NULL
) {
1904 instr
->src
[src_number
].src
=
1905 nir_src_for_ssa(evaluate_rvalue(ir
->shadow_comparitor
));
1906 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1910 if (ir
->offset
!= NULL
) {
1911 /* we don't support multiple offsets yet */
1912 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1914 ir_constant
*const_offset
= ir
->offset
->as_constant();
1915 if (const_offset
!= NULL
) {
1916 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1917 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1919 instr
->src
[src_number
].src
=
1920 nir_src_for_ssa(evaluate_rvalue(ir
->offset
));
1921 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1928 instr
->src
[src_number
].src
=
1929 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.bias
));
1930 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1937 if (ir
->lod_info
.lod
!= NULL
) {
1938 instr
->src
[src_number
].src
=
1939 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.lod
));
1940 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1946 instr
->src
[src_number
].src
=
1947 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdx
));
1948 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1950 instr
->src
[src_number
].src
=
1951 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdy
));
1952 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1957 instr
->src
[src_number
].src
=
1958 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.sample_index
));
1959 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1964 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1971 assert(src_number
== num_srcs
);
1973 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
1977 nir_visitor::visit(ir_constant
*ir
)
1980 * We don't know if this variable is an an array or struct that gets
1981 * dereferenced, so do the safe thing an make it a variable with a
1982 * constant initializer and return a dereference.
1986 nir_local_variable_create(this->impl
, ir
->type
, "const_temp");
1987 var
->data
.read_only
= true;
1988 var
->constant_initializer
= constant_copy(ir
, var
);
1990 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1991 this->deref_tail
= &this->deref_head
->deref
;
1995 nir_visitor::visit(ir_dereference_variable
*ir
)
1997 struct hash_entry
*entry
=
1998 _mesa_hash_table_search(this->var_table
, ir
->var
);
2000 nir_variable
*var
= (nir_variable
*) entry
->data
;
2002 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
2003 this->deref_head
= deref
;
2004 this->deref_tail
= &deref
->deref
;
2008 nir_visitor::visit(ir_dereference_record
*ir
)
2010 ir
->record
->accept(this);
2012 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
2013 assert(field_index
>= 0);
2015 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
2016 deref
->deref
.type
= ir
->type
;
2017 this->deref_tail
->child
= &deref
->deref
;
2018 this->deref_tail
= &deref
->deref
;
2022 nir_visitor::visit(ir_dereference_array
*ir
)
2024 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
2025 deref
->deref
.type
= ir
->type
;
2027 ir_constant
*const_index
= ir
->array_index
->as_constant();
2028 if (const_index
!= NULL
) {
2029 deref
->deref_array_type
= nir_deref_array_type_direct
;
2030 deref
->base_offset
= const_index
->value
.u
[0];
2032 deref
->deref_array_type
= nir_deref_array_type_indirect
;
2034 nir_src_for_ssa(evaluate_rvalue(ir
->array_index
));
2037 ir
->array
->accept(this);
2039 this->deref_tail
->child
= &deref
->deref
;
2040 ralloc_steal(this->deref_tail
, deref
);
2041 this->deref_tail
= &deref
->deref
;
2045 nir_visitor::visit(ir_barrier
*ir
)
2047 nir_intrinsic_instr
*instr
=
2048 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
2049 nir_builder_instr_insert(&b
, &instr
->instr
);