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
.is_unmatched_generic_inout
= ir
->data
.is_unmatched_generic_inout
;
371 var
->data
.location_frac
= ir
->data
.location_frac
;
372 var
->data
.from_named_ifc_block_array
= ir
->data
.from_named_ifc_block_array
;
373 var
->data
.from_named_ifc_block_nonarray
= ir
->data
.from_named_ifc_block_nonarray
;
375 switch (ir
->data
.depth_layout
) {
376 case ir_depth_layout_none
:
377 var
->data
.depth_layout
= nir_depth_layout_none
;
379 case ir_depth_layout_any
:
380 var
->data
.depth_layout
= nir_depth_layout_any
;
382 case ir_depth_layout_greater
:
383 var
->data
.depth_layout
= nir_depth_layout_greater
;
385 case ir_depth_layout_less
:
386 var
->data
.depth_layout
= nir_depth_layout_less
;
388 case ir_depth_layout_unchanged
:
389 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
392 unreachable("not reached");
395 var
->data
.index
= ir
->data
.index
;
396 var
->data
.descriptor_set
= 0;
397 var
->data
.binding
= ir
->data
.binding
;
398 var
->data
.atomic
.offset
= ir
->data
.atomic
.offset
;
399 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
400 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
401 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
402 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
403 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
404 var
->data
.image
.format
= ir
->data
.image_format
;
405 var
->data
.max_array_access
= ir
->data
.max_array_access
;
407 var
->num_state_slots
= ir
->get_num_state_slots();
408 if (var
->num_state_slots
> 0) {
409 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
410 var
->num_state_slots
);
412 ir_state_slot
*state_slots
= ir
->get_state_slots();
413 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
414 for (unsigned j
= 0; j
< 5; j
++)
415 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
416 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
419 var
->state_slots
= NULL
;
422 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
424 var
->interface_type
= ir
->get_interface_type();
426 if (var
->data
.mode
== nir_var_local
)
427 nir_function_impl_add_variable(impl
, var
);
429 nir_shader_add_variable(shader
, var
);
431 _mesa_hash_table_insert(var_table
, ir
, var
);
436 nir_function_visitor::visit_enter(ir_function
*ir
)
438 visitor
->create_function(ir
);
439 return visit_continue_with_parent
;
444 nir_visitor::create_function(ir_function
*ir
)
446 nir_function
*func
= nir_function_create(this->shader
, ir
->name
);
447 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
448 create_overload(sig
, func
);
455 nir_visitor::create_overload(ir_function_signature
*ir
, nir_function
*function
)
457 if (ir
->is_intrinsic
)
460 nir_function_overload
*overload
= nir_function_overload_create(function
);
462 unsigned num_params
= ir
->parameters
.length();
463 overload
->num_params
= num_params
;
464 overload
->params
= ralloc_array(shader
, nir_parameter
, num_params
);
467 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
468 switch (param
->data
.mode
) {
469 case ir_var_function_in
:
470 overload
->params
[i
].param_type
= nir_parameter_in
;
473 case ir_var_function_out
:
474 overload
->params
[i
].param_type
= nir_parameter_out
;
477 case ir_var_function_inout
:
478 overload
->params
[i
].param_type
= nir_parameter_inout
;
482 unreachable("not reached");
485 overload
->params
[i
].type
= param
->type
;
489 overload
->return_type
= ir
->return_type
;
491 _mesa_hash_table_insert(this->overload_table
, ir
, overload
);
495 nir_visitor::visit(ir_function
*ir
)
497 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
502 nir_visitor::visit(ir_function_signature
*ir
)
504 if (ir
->is_intrinsic
)
507 struct hash_entry
*entry
=
508 _mesa_hash_table_search(this->overload_table
, ir
);
511 nir_function_overload
*overload
= (nir_function_overload
*) entry
->data
;
513 if (ir
->is_defined
) {
514 nir_function_impl
*impl
= nir_function_impl_create(overload
);
517 unsigned num_params
= overload
->num_params
;
518 impl
->num_params
= num_params
;
519 impl
->params
= ralloc_array(this->shader
, nir_variable
*, num_params
);
521 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
523 impl
->params
[i
] = this->var
;
527 if (overload
->return_type
== glsl_type::void_type
) {
528 impl
->return_var
= NULL
;
530 impl
->return_var
= ralloc(this->shader
, nir_variable
);
531 impl
->return_var
->name
= ralloc_strdup(impl
->return_var
,
533 impl
->return_var
->type
= overload
->return_type
;
536 this->is_global
= false;
538 nir_builder_init(&b
, impl
);
539 b
.cursor
= nir_after_cf_list(&impl
->body
);
540 visit_exec_list(&ir
->body
, this);
542 this->is_global
= true;
544 overload
->impl
= NULL
;
549 nir_visitor::visit(ir_loop
*ir
)
551 nir_loop
*loop
= nir_loop_create(this->shader
);
552 nir_builder_cf_insert(&b
, &loop
->cf_node
);
554 b
.cursor
= nir_after_cf_list(&loop
->body
);
555 visit_exec_list(&ir
->body_instructions
, this);
556 b
.cursor
= nir_after_cf_node(&loop
->cf_node
);
560 nir_visitor::visit(ir_if
*ir
)
563 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
565 nir_if
*if_stmt
= nir_if_create(this->shader
);
566 if_stmt
->condition
= condition
;
567 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
569 b
.cursor
= nir_after_cf_list(&if_stmt
->then_list
);
570 visit_exec_list(&ir
->then_instructions
, this);
572 b
.cursor
= nir_after_cf_list(&if_stmt
->else_list
);
573 visit_exec_list(&ir
->else_instructions
, this);
575 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
579 nir_visitor::visit(ir_discard
*ir
)
582 * discards aren't treated as control flow, because before we lower them
583 * they can appear anywhere in the shader and the stuff after them may still
584 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
585 * discards will be immediately followed by a return.
588 nir_intrinsic_instr
*discard
;
590 discard
= nir_intrinsic_instr_create(this->shader
,
591 nir_intrinsic_discard_if
);
593 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
595 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
598 nir_builder_instr_insert(&b
, &discard
->instr
);
602 nir_visitor::visit(ir_emit_vertex
*ir
)
604 nir_intrinsic_instr
*instr
=
605 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
606 instr
->const_index
[0] = ir
->stream_id();
607 nir_builder_instr_insert(&b
, &instr
->instr
);
611 nir_visitor::visit(ir_end_primitive
*ir
)
613 nir_intrinsic_instr
*instr
=
614 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
615 instr
->const_index
[0] = ir
->stream_id();
616 nir_builder_instr_insert(&b
, &instr
->instr
);
620 nir_visitor::visit(ir_loop_jump
*ir
)
624 case ir_loop_jump::jump_break
:
625 type
= nir_jump_break
;
627 case ir_loop_jump::jump_continue
:
628 type
= nir_jump_continue
;
631 unreachable("not reached");
634 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
635 nir_builder_instr_insert(&b
, &instr
->instr
);
639 nir_visitor::visit(ir_return
*ir
)
641 if (ir
->value
!= NULL
) {
642 nir_intrinsic_instr
*copy
=
643 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
645 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
646 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
649 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
650 nir_builder_instr_insert(&b
, &instr
->instr
);
654 nir_visitor::visit(ir_call
*ir
)
656 if (ir
->callee
->is_intrinsic
) {
658 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
659 op
= nir_intrinsic_atomic_counter_read_var
;
660 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
661 op
= nir_intrinsic_atomic_counter_inc_var
;
662 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
663 op
= nir_intrinsic_atomic_counter_dec_var
;
664 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
665 op
= nir_intrinsic_image_load
;
666 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
667 op
= nir_intrinsic_image_store
;
668 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
669 op
= nir_intrinsic_image_atomic_add
;
670 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
671 op
= nir_intrinsic_image_atomic_min
;
672 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
673 op
= nir_intrinsic_image_atomic_max
;
674 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
675 op
= nir_intrinsic_image_atomic_and
;
676 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
677 op
= nir_intrinsic_image_atomic_or
;
678 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
679 op
= nir_intrinsic_image_atomic_xor
;
680 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
681 op
= nir_intrinsic_image_atomic_exchange
;
682 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
683 op
= nir_intrinsic_image_atomic_comp_swap
;
684 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
685 op
= nir_intrinsic_memory_barrier
;
686 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_size") == 0) {
687 op
= nir_intrinsic_image_size
;
688 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_samples") == 0) {
689 op
= nir_intrinsic_image_samples
;
690 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_ssbo") == 0) {
691 op
= nir_intrinsic_store_ssbo
;
692 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_ssbo") == 0) {
693 op
= nir_intrinsic_load_ssbo
;
694 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_add_ssbo") == 0) {
695 op
= nir_intrinsic_ssbo_atomic_add
;
696 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_and_ssbo") == 0) {
697 op
= nir_intrinsic_ssbo_atomic_and
;
698 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_or_ssbo") == 0) {
699 op
= nir_intrinsic_ssbo_atomic_or
;
700 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_xor_ssbo") == 0) {
701 op
= nir_intrinsic_ssbo_atomic_xor
;
702 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_min_ssbo") == 0) {
703 assert(ir
->return_deref
);
704 if (ir
->return_deref
->type
== glsl_type::int_type
)
705 op
= nir_intrinsic_ssbo_atomic_imin
;
706 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
707 op
= nir_intrinsic_ssbo_atomic_umin
;
709 unreachable("Invalid type");
710 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_max_ssbo") == 0) {
711 assert(ir
->return_deref
);
712 if (ir
->return_deref
->type
== glsl_type::int_type
)
713 op
= nir_intrinsic_ssbo_atomic_imax
;
714 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
715 op
= nir_intrinsic_ssbo_atomic_umax
;
717 unreachable("Invalid type");
718 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_exchange_ssbo") == 0) {
719 op
= nir_intrinsic_ssbo_atomic_exchange
;
720 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_comp_swap_ssbo") == 0) {
721 op
= nir_intrinsic_ssbo_atomic_comp_swap
;
722 } else if (strcmp(ir
->callee_name(), "__intrinsic_shader_clock") == 0) {
723 op
= nir_intrinsic_shader_clock
;
724 } else if (strcmp(ir
->callee_name(), "__intrinsic_group_memory_barrier") == 0) {
725 op
= nir_intrinsic_group_memory_barrier
;
726 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_atomic_counter") == 0) {
727 op
= nir_intrinsic_memory_barrier_atomic_counter
;
728 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_buffer") == 0) {
729 op
= nir_intrinsic_memory_barrier_buffer
;
730 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_image") == 0) {
731 op
= nir_intrinsic_memory_barrier_image
;
732 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_shared") == 0) {
733 op
= nir_intrinsic_memory_barrier_shared
;
734 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_shared") == 0) {
735 op
= nir_intrinsic_load_shared
;
736 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_shared") == 0) {
737 op
= nir_intrinsic_store_shared
;
738 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_add_shared") == 0) {
739 op
= nir_intrinsic_shared_atomic_add
;
740 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_and_shared") == 0) {
741 op
= nir_intrinsic_shared_atomic_and
;
742 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_or_shared") == 0) {
743 op
= nir_intrinsic_shared_atomic_or
;
744 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_xor_shared") == 0) {
745 op
= nir_intrinsic_shared_atomic_xor
;
746 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_min_shared") == 0) {
747 assert(ir
->return_deref
);
748 if (ir
->return_deref
->type
== glsl_type::int_type
)
749 op
= nir_intrinsic_shared_atomic_imin
;
750 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
751 op
= nir_intrinsic_shared_atomic_umin
;
753 unreachable("Invalid type");
754 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_max_shared") == 0) {
755 assert(ir
->return_deref
);
756 if (ir
->return_deref
->type
== glsl_type::int_type
)
757 op
= nir_intrinsic_shared_atomic_imax
;
758 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
759 op
= nir_intrinsic_shared_atomic_umax
;
761 unreachable("Invalid type");
762 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_exchange_shared") == 0) {
763 op
= nir_intrinsic_shared_atomic_exchange
;
764 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_comp_swap_shared") == 0) {
765 op
= nir_intrinsic_shared_atomic_comp_swap
;
767 unreachable("not reached");
770 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
771 nir_dest
*dest
= &instr
->dest
;
774 case nir_intrinsic_atomic_counter_read_var
:
775 case nir_intrinsic_atomic_counter_inc_var
:
776 case nir_intrinsic_atomic_counter_dec_var
: {
777 ir_dereference
*param
=
778 (ir_dereference
*) ir
->actual_parameters
.get_head();
779 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
780 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
781 nir_builder_instr_insert(&b
, &instr
->instr
);
784 case nir_intrinsic_image_load
:
785 case nir_intrinsic_image_store
:
786 case nir_intrinsic_image_atomic_add
:
787 case nir_intrinsic_image_atomic_min
:
788 case nir_intrinsic_image_atomic_max
:
789 case nir_intrinsic_image_atomic_and
:
790 case nir_intrinsic_image_atomic_or
:
791 case nir_intrinsic_image_atomic_xor
:
792 case nir_intrinsic_image_atomic_exchange
:
793 case nir_intrinsic_image_atomic_comp_swap
:
794 case nir_intrinsic_image_samples
:
795 case nir_intrinsic_image_size
: {
796 nir_ssa_undef_instr
*instr_undef
=
797 nir_ssa_undef_instr_create(shader
, 1);
798 nir_builder_instr_insert(&b
, &instr_undef
->instr
);
800 /* Set the image variable dereference. */
801 exec_node
*param
= ir
->actual_parameters
.get_head();
802 ir_dereference
*image
= (ir_dereference
*)param
;
803 const glsl_type
*type
=
804 image
->variable_referenced()->type
->without_array();
806 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
807 param
= param
->get_next();
809 /* Set the intrinsic destination. */
810 if (ir
->return_deref
) {
811 const nir_intrinsic_info
*info
=
812 &nir_intrinsic_infos
[instr
->intrinsic
];
813 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
814 info
->dest_components
, NULL
);
817 if (op
== nir_intrinsic_image_size
||
818 op
== nir_intrinsic_image_samples
) {
819 nir_builder_instr_insert(&b
, &instr
->instr
);
823 /* Set the address argument, extending the coordinate vector to four
826 nir_ssa_def
*src_addr
=
827 evaluate_rvalue((ir_dereference
*)param
);
828 nir_ssa_def
*srcs
[4];
830 for (int i
= 0; i
< 4; i
++) {
831 if (i
< type
->coordinate_components())
832 srcs
[i
] = nir_channel(&b
, src_addr
, i
);
834 srcs
[i
] = &instr_undef
->def
;
837 instr
->src
[0] = nir_src_for_ssa(nir_vec(&b
, srcs
, 4));
838 param
= param
->get_next();
840 /* Set the sample argument, which is undefined for single-sample
843 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
845 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
846 param
= param
->get_next();
848 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
851 /* Set the intrinsic parameters. */
852 if (!param
->is_tail_sentinel()) {
854 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
855 param
= param
->get_next();
858 if (!param
->is_tail_sentinel()) {
860 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
861 param
= param
->get_next();
863 nir_builder_instr_insert(&b
, &instr
->instr
);
866 case nir_intrinsic_memory_barrier
:
867 case nir_intrinsic_group_memory_barrier
:
868 case nir_intrinsic_memory_barrier_atomic_counter
:
869 case nir_intrinsic_memory_barrier_buffer
:
870 case nir_intrinsic_memory_barrier_image
:
871 case nir_intrinsic_memory_barrier_shared
:
872 nir_builder_instr_insert(&b
, &instr
->instr
);
874 case nir_intrinsic_shader_clock
:
875 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, NULL
);
876 nir_builder_instr_insert(&b
, &instr
->instr
);
878 case nir_intrinsic_store_ssbo
: {
879 exec_node
*param
= ir
->actual_parameters
.get_head();
880 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
882 param
= param
->get_next();
883 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
885 param
= param
->get_next();
886 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
888 param
= param
->get_next();
889 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
892 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
893 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(block
));
894 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(offset
));
895 instr
->const_index
[0] = write_mask
->value
.u
[0];
896 instr
->num_components
= val
->type
->vector_elements
;
898 nir_builder_instr_insert(&b
, &instr
->instr
);
901 case nir_intrinsic_load_ssbo
: {
902 exec_node
*param
= ir
->actual_parameters
.get_head();
903 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
905 param
= param
->get_next();
906 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
908 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(block
));
909 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
911 const glsl_type
*type
= ir
->return_deref
->var
->type
;
912 instr
->num_components
= type
->vector_elements
;
914 /* Setup destination register */
915 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
916 type
->vector_elements
, NULL
);
918 /* Insert the created nir instruction now since in the case of boolean
919 * result we will need to emit another instruction after it
921 nir_builder_instr_insert(&b
, &instr
->instr
);
924 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
925 * consider a true boolean to be ~0. Fix this up with a != 0
928 if (type
->base_type
== GLSL_TYPE_BOOL
) {
929 nir_alu_instr
*load_ssbo_compare
=
930 nir_alu_instr_create(shader
, nir_op_ine
);
931 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
932 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
933 load_ssbo_compare
->src
[1].src
=
934 nir_src_for_ssa(nir_imm_int(&b
, 0));
935 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
936 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
937 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
938 &load_ssbo_compare
->dest
.dest
,
939 type
->vector_elements
, NULL
);
940 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
941 nir_builder_instr_insert(&b
, &load_ssbo_compare
->instr
);
942 dest
= &load_ssbo_compare
->dest
.dest
;
946 case nir_intrinsic_ssbo_atomic_add
:
947 case nir_intrinsic_ssbo_atomic_imin
:
948 case nir_intrinsic_ssbo_atomic_umin
:
949 case nir_intrinsic_ssbo_atomic_imax
:
950 case nir_intrinsic_ssbo_atomic_umax
:
951 case nir_intrinsic_ssbo_atomic_and
:
952 case nir_intrinsic_ssbo_atomic_or
:
953 case nir_intrinsic_ssbo_atomic_xor
:
954 case nir_intrinsic_ssbo_atomic_exchange
:
955 case nir_intrinsic_ssbo_atomic_comp_swap
: {
956 int param_count
= ir
->actual_parameters
.length();
957 assert(param_count
== 3 || param_count
== 4);
960 exec_node
*param
= ir
->actual_parameters
.get_head();
961 ir_instruction
*inst
= (ir_instruction
*) param
;
962 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
965 param
= param
->get_next();
966 inst
= (ir_instruction
*) param
;
967 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
969 /* data1 parameter (this is always present) */
970 param
= param
->get_next();
971 inst
= (ir_instruction
*) param
;
972 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
974 /* data2 parameter (only with atomic_comp_swap) */
975 if (param_count
== 4) {
976 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
977 param
= param
->get_next();
978 inst
= (ir_instruction
*) param
;
979 instr
->src
[3] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
983 assert(ir
->return_deref
);
984 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
985 ir
->return_deref
->type
->vector_elements
, NULL
);
986 nir_builder_instr_insert(&b
, &instr
->instr
);
989 case nir_intrinsic_load_shared
: {
990 exec_node
*param
= ir
->actual_parameters
.get_head();
991 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
993 instr
->const_index
[0] = 0;
994 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(offset
));
996 const glsl_type
*type
= ir
->return_deref
->var
->type
;
997 instr
->num_components
= type
->vector_elements
;
999 /* Setup destination register */
1000 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1001 type
->vector_elements
, NULL
);
1003 nir_builder_instr_insert(&b
, &instr
->instr
);
1006 case nir_intrinsic_store_shared
: {
1007 exec_node
*param
= ir
->actual_parameters
.get_head();
1008 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1010 param
= param
->get_next();
1011 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
1013 param
= param
->get_next();
1014 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
1017 instr
->const_index
[0] = 0;
1018 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
1020 instr
->const_index
[1] = write_mask
->value
.u
[0];
1022 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
1023 instr
->num_components
= val
->type
->vector_elements
;
1025 nir_builder_instr_insert(&b
, &instr
->instr
);
1028 case nir_intrinsic_shared_atomic_add
:
1029 case nir_intrinsic_shared_atomic_imin
:
1030 case nir_intrinsic_shared_atomic_umin
:
1031 case nir_intrinsic_shared_atomic_imax
:
1032 case nir_intrinsic_shared_atomic_umax
:
1033 case nir_intrinsic_shared_atomic_and
:
1034 case nir_intrinsic_shared_atomic_or
:
1035 case nir_intrinsic_shared_atomic_xor
:
1036 case nir_intrinsic_shared_atomic_exchange
:
1037 case nir_intrinsic_shared_atomic_comp_swap
: {
1038 int param_count
= ir
->actual_parameters
.length();
1039 assert(param_count
== 2 || param_count
== 3);
1042 exec_node
*param
= ir
->actual_parameters
.get_head();
1043 ir_instruction
*inst
= (ir_instruction
*) param
;
1044 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1046 /* data1 parameter (this is always present) */
1047 param
= param
->get_next();
1048 inst
= (ir_instruction
*) param
;
1049 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1051 /* data2 parameter (only with atomic_comp_swap) */
1052 if (param_count
== 3) {
1053 assert(op
== nir_intrinsic_shared_atomic_comp_swap
);
1054 param
= param
->get_next();
1055 inst
= (ir_instruction
*) param
;
1057 nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1061 assert(ir
->return_deref
);
1062 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1063 ir
->return_deref
->type
->vector_elements
, NULL
);
1064 nir_builder_instr_insert(&b
, &instr
->instr
);
1068 unreachable("not reached");
1071 if (ir
->return_deref
) {
1072 nir_intrinsic_instr
*store_instr
=
1073 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
1074 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
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) {
1137 * We have no good way to update only part of a variable, so just load
1138 * the LHS and do a vec operation to combine the old with the new, and
1140 * back into the LHS. Copy propagation should get rid of the mess.
1143 nir_intrinsic_instr
*load
=
1144 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1145 load
->num_components
= ir
->lhs
->type
->vector_elements
;
1146 nir_ssa_dest_init(&load
->instr
, &load
->dest
, num_components
, NULL
);
1147 load
->variables
[0] = lhs_deref
;
1148 ralloc_steal(load
, load
->variables
[0]);
1149 nir_builder_instr_insert(&b
, &load
->instr
);
1151 nir_ssa_def
*srcs
[4];
1153 unsigned component
= 0;
1154 for (unsigned i
= 0; i
< ir
->lhs
->type
->vector_elements
; i
++) {
1155 if (ir
->write_mask
& (1 << i
)) {
1156 /* GLSL IR will give us the input to the write-masked assignment
1157 * in a single packed vector. So, for example, if the
1158 * writemask is xzw, then we have to swizzle x -> x, y -> z,
1159 * and z -> w and get the y component from the load.
1161 srcs
[i
] = nir_channel(&b
, src
, component
++);
1163 srcs
[i
] = nir_channel(&b
, &load
->dest
.ssa
, i
);
1167 src
= nir_vec(&b
, srcs
, ir
->lhs
->type
->vector_elements
);
1170 nir_intrinsic_instr
*store
=
1171 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1172 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1173 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
1174 store
->variables
[0] = nir_deref_as_var(store_deref
);
1175 store
->src
[0] = nir_src_for_ssa(src
);
1177 if (ir
->condition
) {
1178 nir_if
*if_stmt
= nir_if_create(this->shader
);
1179 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1180 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1181 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
1182 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1184 nir_builder_instr_insert(&b
, &store
->instr
);
1189 * Given an instruction, returns a pointer to its destination or NULL if there
1190 * is no destination.
1192 * Note that this only handles instructions we generate at this level.
1195 get_instr_dest(nir_instr
*instr
)
1197 nir_alu_instr
*alu_instr
;
1198 nir_intrinsic_instr
*intrinsic_instr
;
1199 nir_tex_instr
*tex_instr
;
1201 switch (instr
->type
) {
1202 case nir_instr_type_alu
:
1203 alu_instr
= nir_instr_as_alu(instr
);
1204 return &alu_instr
->dest
.dest
;
1206 case nir_instr_type_intrinsic
:
1207 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1208 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1209 return &intrinsic_instr
->dest
;
1213 case nir_instr_type_tex
:
1214 tex_instr
= nir_instr_as_tex(instr
);
1215 return &tex_instr
->dest
;
1218 unreachable("not reached");
1225 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
)
1227 nir_dest
*dest
= get_instr_dest(instr
);
1230 nir_ssa_dest_init(instr
, dest
, num_components
, NULL
);
1232 nir_builder_instr_insert(&b
, instr
);
1235 assert(dest
->is_ssa
);
1236 this->result
= &dest
->ssa
;
1241 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1244 if (ir
->as_dereference() || ir
->as_constant()) {
1246 * A dereference is being used on the right hand side, which means we
1247 * must emit a variable load.
1250 nir_intrinsic_instr
*load_instr
=
1251 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1252 load_instr
->num_components
= ir
->type
->vector_elements
;
1253 load_instr
->variables
[0] = this->deref_head
;
1254 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1255 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
);
1258 return this->result
;
1262 nir_visitor::visit(ir_expression
*ir
)
1264 /* Some special cases */
1265 switch (ir
->operation
) {
1266 case ir_binop_ubo_load
: {
1267 nir_intrinsic_instr
*load
=
1268 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_ubo
);
1269 load
->num_components
= ir
->type
->vector_elements
;
1270 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1271 load
->src
[1] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1272 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1275 * In UBO's, a true boolean value is any non-zero value, but we consider
1276 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1279 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
)
1280 this->result
= nir_ine(&b
, &load
->dest
.ssa
, nir_imm_int(&b
, 0));
1285 case ir_unop_interpolate_at_centroid
:
1286 case ir_binop_interpolate_at_offset
:
1287 case ir_binop_interpolate_at_sample
: {
1288 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1289 ir_swizzle
*swizzle
= NULL
;
1291 /* the api does not allow a swizzle here, but the varying packing code
1292 * may have pushed one into here.
1294 swizzle
= ir
->operands
[0]->as_swizzle();
1296 deref
= swizzle
->val
->as_dereference();
1300 deref
->accept(this);
1302 nir_intrinsic_op op
;
1303 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1304 switch (ir
->operation
) {
1305 case ir_unop_interpolate_at_centroid
:
1306 op
= nir_intrinsic_interp_var_at_centroid
;
1308 case ir_binop_interpolate_at_offset
:
1309 op
= nir_intrinsic_interp_var_at_offset
;
1311 case ir_binop_interpolate_at_sample
:
1312 op
= nir_intrinsic_interp_var_at_sample
;
1315 unreachable("Invalid interpolation intrinsic");
1318 /* This case can happen if the vertex shader does not write the
1319 * given varying. In this case, the linker will lower it to a
1320 * global variable. Since interpolating a variable makes no
1321 * sense, we'll just turn it into a load which will probably
1322 * eventually end up as an SSA definition.
1324 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1325 op
= nir_intrinsic_load_var
;
1328 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1329 intrin
->num_components
= deref
->type
->vector_elements
;
1330 intrin
->variables
[0] = this->deref_head
;
1331 ralloc_steal(intrin
, intrin
->variables
[0]);
1333 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1334 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1335 intrin
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1337 add_instr(&intrin
->instr
, deref
->type
->vector_elements
);
1340 unsigned swiz
[4] = {
1341 swizzle
->mask
.x
, swizzle
->mask
.y
, swizzle
->mask
.z
, swizzle
->mask
.w
1344 result
= nir_swizzle(&b
, result
, swiz
,
1345 swizzle
->type
->vector_elements
, false);
1355 nir_ssa_def
*srcs
[4];
1356 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1357 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1359 glsl_base_type types
[4];
1360 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1362 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1364 types
[i
] = GLSL_TYPE_FLOAT
;
1366 glsl_base_type out_type
;
1368 out_type
= ir
->type
->base_type
;
1370 out_type
= GLSL_TYPE_FLOAT
;
1372 switch (ir
->operation
) {
1373 case ir_unop_bit_not
: result
= nir_inot(&b
, srcs
[0]); break;
1374 case ir_unop_logic_not
:
1375 result
= supports_ints
? nir_inot(&b
, srcs
[0]) : nir_fnot(&b
, srcs
[0]);
1378 result
= (types
[0] == GLSL_TYPE_FLOAT
) ? nir_fneg(&b
, srcs
[0])
1379 : nir_ineg(&b
, srcs
[0]);
1382 result
= (types
[0] == GLSL_TYPE_FLOAT
) ? nir_fabs(&b
, srcs
[0])
1383 : nir_iabs(&b
, srcs
[0]);
1385 case ir_unop_saturate
:
1386 assert(types
[0] == GLSL_TYPE_FLOAT
);
1387 result
= nir_fsat(&b
, srcs
[0]);
1390 result
= (types
[0] == GLSL_TYPE_FLOAT
) ? nir_fsign(&b
, srcs
[0])
1391 : nir_isign(&b
, srcs
[0]);
1393 case ir_unop_rcp
: result
= nir_frcp(&b
, srcs
[0]); break;
1394 case ir_unop_rsq
: result
= nir_frsq(&b
, srcs
[0]); break;
1395 case ir_unop_sqrt
: result
= nir_fsqrt(&b
, srcs
[0]); break;
1396 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1397 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1398 case ir_unop_exp2
: result
= nir_fexp2(&b
, srcs
[0]); break;
1399 case ir_unop_log2
: result
= nir_flog2(&b
, srcs
[0]); break;
1401 result
= supports_ints
? nir_i2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1404 result
= supports_ints
? nir_u2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1407 result
= supports_ints
? nir_b2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1409 case ir_unop_f2i
: result
= nir_f2i(&b
, srcs
[0]); break;
1410 case ir_unop_f2u
: result
= nir_f2u(&b
, srcs
[0]); break;
1411 case ir_unop_f2b
: result
= nir_f2b(&b
, srcs
[0]); break;
1412 case ir_unop_i2b
: result
= nir_i2b(&b
, srcs
[0]); break;
1413 case ir_unop_b2i
: result
= nir_b2i(&b
, srcs
[0]); break;
1416 case ir_unop_bitcast_i2f
:
1417 case ir_unop_bitcast_f2i
:
1418 case ir_unop_bitcast_u2f
:
1419 case ir_unop_bitcast_f2u
:
1420 case ir_unop_subroutine_to_int
:
1422 result
= nir_imov(&b
, srcs
[0]);
1425 switch (ir
->operands
[0]->type
->vector_elements
) {
1427 result
= supports_ints
? nir_bany2(&b
, srcs
[0])
1428 : nir_fany2(&b
, srcs
[0]);
1431 result
= supports_ints
? nir_bany3(&b
, srcs
[0])
1432 : nir_fany3(&b
, srcs
[0]);
1435 result
= supports_ints
? nir_bany4(&b
, srcs
[0])
1436 : nir_fany4(&b
, srcs
[0]);
1439 unreachable("not reached");
1442 case ir_unop_trunc
: result
= nir_ftrunc(&b
, srcs
[0]); break;
1443 case ir_unop_ceil
: result
= nir_fceil(&b
, srcs
[0]); break;
1444 case ir_unop_floor
: result
= nir_ffloor(&b
, srcs
[0]); break;
1445 case ir_unop_fract
: result
= nir_ffract(&b
, srcs
[0]); break;
1446 case ir_unop_round_even
: result
= nir_fround_even(&b
, srcs
[0]); break;
1447 case ir_unop_sin
: result
= nir_fsin(&b
, srcs
[0]); break;
1448 case ir_unop_cos
: result
= nir_fcos(&b
, srcs
[0]); break;
1449 case ir_unop_dFdx
: result
= nir_fddx(&b
, srcs
[0]); break;
1450 case ir_unop_dFdy
: result
= nir_fddy(&b
, srcs
[0]); break;
1451 case ir_unop_dFdx_fine
: result
= nir_fddx_fine(&b
, srcs
[0]); break;
1452 case ir_unop_dFdy_fine
: result
= nir_fddy_fine(&b
, srcs
[0]); break;
1453 case ir_unop_dFdx_coarse
: result
= nir_fddx_coarse(&b
, srcs
[0]); break;
1454 case ir_unop_dFdy_coarse
: result
= nir_fddy_coarse(&b
, srcs
[0]); break;
1455 case ir_unop_pack_snorm_2x16
:
1456 result
= nir_pack_snorm_2x16(&b
, srcs
[0]);
1458 case ir_unop_pack_snorm_4x8
:
1459 result
= nir_pack_snorm_4x8(&b
, srcs
[0]);
1461 case ir_unop_pack_unorm_2x16
:
1462 result
= nir_pack_unorm_2x16(&b
, srcs
[0]);
1464 case ir_unop_pack_unorm_4x8
:
1465 result
= nir_pack_unorm_4x8(&b
, srcs
[0]);
1467 case ir_unop_pack_half_2x16
:
1468 result
= nir_pack_half_2x16(&b
, srcs
[0]);
1470 case ir_unop_unpack_snorm_2x16
:
1471 result
= nir_unpack_snorm_2x16(&b
, srcs
[0]);
1473 case ir_unop_unpack_snorm_4x8
:
1474 result
= nir_unpack_snorm_4x8(&b
, srcs
[0]);
1476 case ir_unop_unpack_unorm_2x16
:
1477 result
= nir_unpack_unorm_2x16(&b
, srcs
[0]);
1479 case ir_unop_unpack_unorm_4x8
:
1480 result
= nir_unpack_unorm_4x8(&b
, srcs
[0]);
1482 case ir_unop_unpack_half_2x16
:
1483 result
= nir_unpack_half_2x16(&b
, srcs
[0]);
1485 case ir_unop_unpack_half_2x16_split_x
:
1486 result
= nir_unpack_half_2x16_split_x(&b
, srcs
[0]);
1488 case ir_unop_unpack_half_2x16_split_y
:
1489 result
= nir_unpack_half_2x16_split_y(&b
, srcs
[0]);
1491 case ir_unop_bitfield_reverse
:
1492 result
= nir_bitfield_reverse(&b
, srcs
[0]);
1494 case ir_unop_bit_count
:
1495 result
= nir_bit_count(&b
, srcs
[0]);
1497 case ir_unop_find_msb
:
1499 case GLSL_TYPE_UINT
:
1500 result
= nir_ufind_msb(&b
, srcs
[0]);
1503 result
= nir_ifind_msb(&b
, srcs
[0]);
1506 unreachable("Invalid type for findMSB()");
1509 case ir_unop_find_lsb
:
1510 result
= nir_find_lsb(&b
, srcs
[0]);
1514 switch (ir
->type
->vector_elements
) {
1516 switch (ir
->operands
[0]->type
->vector_elements
) {
1517 case 1: result
= nir_fnoise1_1(&b
, srcs
[0]); break;
1518 case 2: result
= nir_fnoise1_2(&b
, srcs
[0]); break;
1519 case 3: result
= nir_fnoise1_3(&b
, srcs
[0]); break;
1520 case 4: result
= nir_fnoise1_4(&b
, srcs
[0]); break;
1521 default: unreachable("not reached");
1525 switch (ir
->operands
[0]->type
->vector_elements
) {
1526 case 1: result
= nir_fnoise2_1(&b
, srcs
[0]); break;
1527 case 2: result
= nir_fnoise2_2(&b
, srcs
[0]); break;
1528 case 3: result
= nir_fnoise2_3(&b
, srcs
[0]); break;
1529 case 4: result
= nir_fnoise2_4(&b
, srcs
[0]); break;
1530 default: unreachable("not reached");
1534 switch (ir
->operands
[0]->type
->vector_elements
) {
1535 case 1: result
= nir_fnoise3_1(&b
, srcs
[0]); break;
1536 case 2: result
= nir_fnoise3_2(&b
, srcs
[0]); break;
1537 case 3: result
= nir_fnoise3_3(&b
, srcs
[0]); break;
1538 case 4: result
= nir_fnoise3_4(&b
, srcs
[0]); break;
1539 default: unreachable("not reached");
1543 switch (ir
->operands
[0]->type
->vector_elements
) {
1544 case 1: result
= nir_fnoise4_1(&b
, srcs
[0]); break;
1545 case 2: result
= nir_fnoise4_2(&b
, srcs
[0]); break;
1546 case 3: result
= nir_fnoise4_3(&b
, srcs
[0]); break;
1547 case 4: result
= nir_fnoise4_4(&b
, srcs
[0]); break;
1548 default: unreachable("not reached");
1552 unreachable("not reached");
1555 case ir_unop_get_buffer_size
: {
1556 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1558 nir_intrinsic_get_buffer_size
);
1559 load
->num_components
= ir
->type
->vector_elements
;
1560 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1561 add_instr(&load
->instr
, ir
->type
->vector_elements
);
1566 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fadd(&b
, srcs
[0], srcs
[1])
1567 : nir_iadd(&b
, srcs
[0], srcs
[1]);
1570 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fsub(&b
, srcs
[0], srcs
[1])
1571 : nir_isub(&b
, srcs
[0], srcs
[1]);
1574 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fmul(&b
, srcs
[0], srcs
[1])
1575 : nir_imul(&b
, srcs
[0], srcs
[1]);
1578 if (out_type
== GLSL_TYPE_FLOAT
)
1579 result
= nir_fdiv(&b
, srcs
[0], srcs
[1]);
1580 else if (out_type
== GLSL_TYPE_INT
)
1581 result
= nir_idiv(&b
, srcs
[0], srcs
[1]);
1583 result
= nir_udiv(&b
, srcs
[0], srcs
[1]);
1586 result
= (out_type
== GLSL_TYPE_FLOAT
) ? nir_fmod(&b
, srcs
[0], srcs
[1])
1587 : nir_umod(&b
, srcs
[0], srcs
[1]);
1590 if (out_type
== GLSL_TYPE_FLOAT
)
1591 result
= nir_fmin(&b
, srcs
[0], srcs
[1]);
1592 else if (out_type
== GLSL_TYPE_INT
)
1593 result
= nir_imin(&b
, srcs
[0], srcs
[1]);
1595 result
= nir_umin(&b
, srcs
[0], srcs
[1]);
1598 if (out_type
== GLSL_TYPE_FLOAT
)
1599 result
= nir_fmax(&b
, srcs
[0], srcs
[1]);
1600 else if (out_type
== GLSL_TYPE_INT
)
1601 result
= nir_imax(&b
, srcs
[0], srcs
[1]);
1603 result
= nir_umax(&b
, srcs
[0], srcs
[1]);
1605 case ir_binop_pow
: result
= nir_fpow(&b
, srcs
[0], srcs
[1]); break;
1606 case ir_binop_bit_and
: result
= nir_iand(&b
, srcs
[0], srcs
[1]); break;
1607 case ir_binop_bit_or
: result
= nir_ior(&b
, srcs
[0], srcs
[1]); break;
1608 case ir_binop_bit_xor
: result
= nir_ixor(&b
, srcs
[0], srcs
[1]); break;
1609 case ir_binop_logic_and
:
1610 result
= supports_ints
? nir_iand(&b
, srcs
[0], srcs
[1])
1611 : nir_fand(&b
, srcs
[0], srcs
[1]);
1613 case ir_binop_logic_or
:
1614 result
= supports_ints
? nir_ior(&b
, srcs
[0], srcs
[1])
1615 : nir_for(&b
, srcs
[0], srcs
[1]);
1617 case ir_binop_logic_xor
:
1618 result
= supports_ints
? nir_ixor(&b
, srcs
[0], srcs
[1])
1619 : nir_fxor(&b
, srcs
[0], srcs
[1]);
1621 case ir_binop_lshift
: result
= nir_ishl(&b
, srcs
[0], srcs
[1]); break;
1622 case ir_binop_rshift
:
1623 result
= (out_type
== GLSL_TYPE_INT
) ? nir_ishr(&b
, srcs
[0], srcs
[1])
1624 : nir_ushr(&b
, srcs
[0], srcs
[1]);
1626 case ir_binop_imul_high
:
1627 result
= (out_type
== GLSL_TYPE_INT
) ? nir_imul_high(&b
, srcs
[0], srcs
[1])
1628 : nir_umul_high(&b
, srcs
[0], srcs
[1]);
1630 case ir_binop_carry
: result
= nir_uadd_carry(&b
, srcs
[0], srcs
[1]); break;
1631 case ir_binop_borrow
: result
= nir_usub_borrow(&b
, srcs
[0], srcs
[1]); break;
1633 if (supports_ints
) {
1634 if (types
[0] == GLSL_TYPE_FLOAT
)
1635 result
= nir_flt(&b
, srcs
[0], srcs
[1]);
1636 else if (types
[0] == GLSL_TYPE_INT
)
1637 result
= nir_ilt(&b
, srcs
[0], srcs
[1]);
1639 result
= nir_ult(&b
, srcs
[0], srcs
[1]);
1641 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1644 case ir_binop_greater
:
1645 if (supports_ints
) {
1646 if (types
[0] == GLSL_TYPE_FLOAT
)
1647 result
= nir_flt(&b
, srcs
[1], srcs
[0]);
1648 else if (types
[0] == GLSL_TYPE_INT
)
1649 result
= nir_ilt(&b
, srcs
[1], srcs
[0]);
1651 result
= nir_ult(&b
, srcs
[1], srcs
[0]);
1653 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1656 case ir_binop_lequal
:
1657 if (supports_ints
) {
1658 if (types
[0] == GLSL_TYPE_FLOAT
)
1659 result
= nir_fge(&b
, srcs
[1], srcs
[0]);
1660 else if (types
[0] == GLSL_TYPE_INT
)
1661 result
= nir_ige(&b
, srcs
[1], srcs
[0]);
1663 result
= nir_uge(&b
, srcs
[1], srcs
[0]);
1665 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1668 case ir_binop_gequal
:
1669 if (supports_ints
) {
1670 if (types
[0] == GLSL_TYPE_FLOAT
)
1671 result
= nir_fge(&b
, srcs
[0], srcs
[1]);
1672 else if (types
[0] == GLSL_TYPE_INT
)
1673 result
= nir_ige(&b
, srcs
[0], srcs
[1]);
1675 result
= nir_uge(&b
, srcs
[0], srcs
[1]);
1677 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1680 case ir_binop_equal
:
1681 if (supports_ints
) {
1682 if (types
[0] == GLSL_TYPE_FLOAT
)
1683 result
= nir_feq(&b
, srcs
[0], srcs
[1]);
1685 result
= nir_ieq(&b
, srcs
[0], srcs
[1]);
1687 result
= nir_seq(&b
, srcs
[0], srcs
[1]);
1690 case ir_binop_nequal
:
1691 if (supports_ints
) {
1692 if (types
[0] == GLSL_TYPE_FLOAT
)
1693 result
= nir_fne(&b
, srcs
[0], srcs
[1]);
1695 result
= nir_ine(&b
, srcs
[0], srcs
[1]);
1697 result
= nir_sne(&b
, srcs
[0], srcs
[1]);
1700 case ir_binop_all_equal
:
1701 if (supports_ints
) {
1702 if (types
[0] == GLSL_TYPE_FLOAT
) {
1703 switch (ir
->operands
[0]->type
->vector_elements
) {
1704 case 1: result
= nir_feq(&b
, srcs
[0], srcs
[1]); break;
1705 case 2: result
= nir_ball_fequal2(&b
, srcs
[0], srcs
[1]); break;
1706 case 3: result
= nir_ball_fequal3(&b
, srcs
[0], srcs
[1]); break;
1707 case 4: result
= nir_ball_fequal4(&b
, srcs
[0], srcs
[1]); break;
1709 unreachable("not reached");
1712 switch (ir
->operands
[0]->type
->vector_elements
) {
1713 case 1: result
= nir_ieq(&b
, srcs
[0], srcs
[1]); break;
1714 case 2: result
= nir_ball_iequal2(&b
, srcs
[0], srcs
[1]); break;
1715 case 3: result
= nir_ball_iequal3(&b
, srcs
[0], srcs
[1]); break;
1716 case 4: result
= nir_ball_iequal4(&b
, srcs
[0], srcs
[1]); break;
1718 unreachable("not reached");
1722 switch (ir
->operands
[0]->type
->vector_elements
) {
1723 case 1: result
= nir_seq(&b
, srcs
[0], srcs
[1]); break;
1724 case 2: result
= nir_fall_equal2(&b
, srcs
[0], srcs
[1]); break;
1725 case 3: result
= nir_fall_equal3(&b
, srcs
[0], srcs
[1]); break;
1726 case 4: result
= nir_fall_equal4(&b
, srcs
[0], srcs
[1]); break;
1728 unreachable("not reached");
1732 case ir_binop_any_nequal
:
1733 if (supports_ints
) {
1734 if (types
[0] == GLSL_TYPE_FLOAT
) {
1735 switch (ir
->operands
[0]->type
->vector_elements
) {
1736 case 1: result
= nir_fne(&b
, srcs
[0], srcs
[1]); break;
1737 case 2: result
= nir_bany_fnequal2(&b
, srcs
[0], srcs
[1]); break;
1738 case 3: result
= nir_bany_fnequal3(&b
, srcs
[0], srcs
[1]); break;
1739 case 4: result
= nir_bany_fnequal4(&b
, srcs
[0], srcs
[1]); break;
1741 unreachable("not reached");
1744 switch (ir
->operands
[0]->type
->vector_elements
) {
1745 case 1: result
= nir_ine(&b
, srcs
[0], srcs
[1]); break;
1746 case 2: result
= nir_bany_inequal2(&b
, srcs
[0], srcs
[1]); break;
1747 case 3: result
= nir_bany_inequal3(&b
, srcs
[0], srcs
[1]); break;
1748 case 4: result
= nir_bany_inequal4(&b
, srcs
[0], srcs
[1]); break;
1750 unreachable("not reached");
1754 switch (ir
->operands
[0]->type
->vector_elements
) {
1755 case 1: result
= nir_sne(&b
, srcs
[0], srcs
[1]); break;
1756 case 2: result
= nir_fany_nequal2(&b
, srcs
[0], srcs
[1]); break;
1757 case 3: result
= nir_fany_nequal3(&b
, srcs
[0], srcs
[1]); break;
1758 case 4: result
= nir_fany_nequal4(&b
, srcs
[0], srcs
[1]); break;
1760 unreachable("not reached");
1765 switch (ir
->operands
[0]->type
->vector_elements
) {
1766 case 2: result
= nir_fdot2(&b
, srcs
[0], srcs
[1]); break;
1767 case 3: result
= nir_fdot3(&b
, srcs
[0], srcs
[1]); break;
1768 case 4: result
= nir_fdot4(&b
, srcs
[0], srcs
[1]); break;
1770 unreachable("not reached");
1774 case ir_binop_pack_half_2x16_split
:
1775 result
= nir_pack_half_2x16_split(&b
, srcs
[0], srcs
[1]);
1777 case ir_binop_bfm
: result
= nir_bfm(&b
, srcs
[0], srcs
[1]); break;
1778 case ir_binop_ldexp
: result
= nir_ldexp(&b
, srcs
[0], srcs
[1]); break;
1780 result
= nir_ffma(&b
, srcs
[0], srcs
[1], srcs
[2]);
1783 result
= nir_flrp(&b
, srcs
[0], srcs
[1], srcs
[2]);
1787 result
= nir_bcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1789 result
= nir_fcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1792 result
= nir_bfi(&b
, srcs
[0], srcs
[1], srcs
[2]);
1794 case ir_triop_bitfield_extract
:
1795 result
= (out_type
== GLSL_TYPE_INT
) ?
1796 nir_ibitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]) :
1797 nir_ubitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]);
1799 case ir_quadop_bitfield_insert
:
1800 result
= nir_bitfield_insert(&b
, srcs
[0], srcs
[1], srcs
[2], srcs
[3]);
1802 case ir_quadop_vector
:
1803 result
= nir_vec(&b
, srcs
, ir
->type
->vector_elements
);
1807 unreachable("not reached");
1812 nir_visitor::visit(ir_swizzle
*ir
)
1814 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1815 result
= nir_swizzle(&b
, evaluate_rvalue(ir
->val
), swizzle
,
1816 ir
->type
->vector_elements
, !supports_ints
);
1820 nir_visitor::visit(ir_texture
*ir
)
1827 num_srcs
= 1; /* coordinate */
1832 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1833 num_srcs
= 2; /* coordinate, bias/lod */
1837 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1843 if (ir
->lod_info
.lod
!= NULL
)
1844 num_srcs
= 2; /* coordinate, lod */
1846 num_srcs
= 1; /* coordinate */
1850 op
= nir_texop_txf_ms
;
1851 num_srcs
= 2; /* coordinate, sample_index */
1856 if (ir
->lod_info
.lod
!= NULL
)
1857 num_srcs
= 1; /* lod */
1864 num_srcs
= 1; /* coordinate */
1869 num_srcs
= 1; /* coordinate */
1872 case ir_query_levels
:
1873 op
= nir_texop_query_levels
;
1877 case ir_texture_samples
:
1878 op
= nir_texop_texture_samples
;
1882 case ir_samples_identical
:
1883 op
= nir_texop_samples_identical
;
1884 num_srcs
= 1; /* coordinate */
1888 unreachable("not reached");
1891 if (ir
->projector
!= NULL
)
1893 if (ir
->shadow_comparitor
!= NULL
)
1895 if (ir
->offset
!= NULL
&& ir
->offset
->as_constant() == NULL
)
1898 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1901 instr
->sampler_dim
=
1902 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1903 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1904 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1905 if (instr
->is_shadow
)
1906 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1907 switch (ir
->type
->base_type
) {
1908 case GLSL_TYPE_FLOAT
:
1909 instr
->dest_type
= nir_type_float
;
1912 instr
->dest_type
= nir_type_int
;
1914 case GLSL_TYPE_BOOL
:
1915 case GLSL_TYPE_UINT
:
1916 instr
->dest_type
= nir_type_uint
;
1919 unreachable("not reached");
1922 instr
->sampler
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1924 unsigned src_number
= 0;
1926 if (ir
->coordinate
!= NULL
) {
1927 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1928 instr
->src
[src_number
].src
=
1929 nir_src_for_ssa(evaluate_rvalue(ir
->coordinate
));
1930 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1934 if (ir
->projector
!= NULL
) {
1935 instr
->src
[src_number
].src
=
1936 nir_src_for_ssa(evaluate_rvalue(ir
->projector
));
1937 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1941 if (ir
->shadow_comparitor
!= NULL
) {
1942 instr
->src
[src_number
].src
=
1943 nir_src_for_ssa(evaluate_rvalue(ir
->shadow_comparitor
));
1944 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1948 if (ir
->offset
!= NULL
) {
1949 /* we don't support multiple offsets yet */
1950 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1952 ir_constant
*const_offset
= ir
->offset
->as_constant();
1953 if (const_offset
!= NULL
) {
1954 for (unsigned i
= 0; i
< const_offset
->type
->vector_elements
; i
++)
1955 instr
->const_offset
[i
] = const_offset
->value
.i
[i
];
1957 instr
->src
[src_number
].src
=
1958 nir_src_for_ssa(evaluate_rvalue(ir
->offset
));
1959 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1966 instr
->src
[src_number
].src
=
1967 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.bias
));
1968 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1975 if (ir
->lod_info
.lod
!= NULL
) {
1976 instr
->src
[src_number
].src
=
1977 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.lod
));
1978 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1984 instr
->src
[src_number
].src
=
1985 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdx
));
1986 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1988 instr
->src
[src_number
].src
=
1989 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdy
));
1990 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1995 instr
->src
[src_number
].src
=
1996 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.sample_index
));
1997 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
2002 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
2009 assert(src_number
== num_srcs
);
2011 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
));
2015 nir_visitor::visit(ir_constant
*ir
)
2018 * We don't know if this variable is an an array or struct that gets
2019 * dereferenced, so do the safe thing an make it a variable with a
2020 * constant initializer and return a dereference.
2024 nir_local_variable_create(this->impl
, ir
->type
, "const_temp");
2025 var
->data
.read_only
= true;
2026 var
->constant_initializer
= constant_copy(ir
, var
);
2028 this->deref_head
= nir_deref_var_create(this->shader
, var
);
2029 this->deref_tail
= &this->deref_head
->deref
;
2033 nir_visitor::visit(ir_dereference_variable
*ir
)
2035 struct hash_entry
*entry
=
2036 _mesa_hash_table_search(this->var_table
, ir
->var
);
2038 nir_variable
*var
= (nir_variable
*) entry
->data
;
2040 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
2041 this->deref_head
= deref
;
2042 this->deref_tail
= &deref
->deref
;
2046 nir_visitor::visit(ir_dereference_record
*ir
)
2048 ir
->record
->accept(this);
2050 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
2051 assert(field_index
>= 0);
2053 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
2054 deref
->deref
.type
= ir
->type
;
2055 this->deref_tail
->child
= &deref
->deref
;
2056 this->deref_tail
= &deref
->deref
;
2060 nir_visitor::visit(ir_dereference_array
*ir
)
2062 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
2063 deref
->deref
.type
= ir
->type
;
2065 ir_constant
*const_index
= ir
->array_index
->as_constant();
2066 if (const_index
!= NULL
) {
2067 deref
->deref_array_type
= nir_deref_array_type_direct
;
2068 deref
->base_offset
= const_index
->value
.u
[0];
2070 deref
->deref_array_type
= nir_deref_array_type_indirect
;
2072 nir_src_for_ssa(evaluate_rvalue(ir
->array_index
));
2075 ir
->array
->accept(this);
2077 this->deref_tail
->child
= &deref
->deref
;
2078 ralloc_steal(this->deref_tail
, deref
);
2079 this->deref_tail
= &deref
->deref
;
2083 nir_visitor::visit(ir_barrier
*ir
)
2085 nir_intrinsic_instr
*instr
=
2086 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
2087 nir_builder_instr_insert(&b
, &instr
->instr
);