2 * Copyright © 2014 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Connor Abbott (cwabbott0@gmail.com)
28 #include "glsl_to_nir.h"
29 #include "nir_control_flow.h"
30 #include "nir_builder.h"
31 #include "compiler/glsl/ir_visitor.h"
32 #include "compiler/glsl/ir_hierarchical_visitor.h"
33 #include "compiler/glsl/ir.h"
34 #include "main/imports.h"
37 * pass to lower GLSL IR to NIR
39 * This will lower variable dereferences to loads/stores of corresponding
40 * variables in NIR - the variables will be converted to registers in a later
46 class nir_visitor
: public ir_visitor
49 nir_visitor(nir_shader
*shader
);
52 virtual void visit(ir_variable
*);
53 virtual void visit(ir_function
*);
54 virtual void visit(ir_function_signature
*);
55 virtual void visit(ir_loop
*);
56 virtual void visit(ir_if
*);
57 virtual void visit(ir_discard
*);
58 virtual void visit(ir_loop_jump
*);
59 virtual void visit(ir_return
*);
60 virtual void visit(ir_call
*);
61 virtual void visit(ir_assignment
*);
62 virtual void visit(ir_emit_vertex
*);
63 virtual void visit(ir_end_primitive
*);
64 virtual void visit(ir_expression
*);
65 virtual void visit(ir_swizzle
*);
66 virtual void visit(ir_texture
*);
67 virtual void visit(ir_constant
*);
68 virtual void visit(ir_dereference_variable
*);
69 virtual void visit(ir_dereference_record
*);
70 virtual void visit(ir_dereference_array
*);
71 virtual void visit(ir_barrier
*);
73 void create_function(ir_function_signature
*ir
);
76 void add_instr(nir_instr
*instr
, unsigned num_components
, unsigned bit_size
);
77 nir_ssa_def
*evaluate_rvalue(ir_rvalue
*ir
);
79 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
**srcs
);
80 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
);
81 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
,
83 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
,
84 nir_ssa_def
*src2
, nir_ssa_def
*src3
);
89 nir_function_impl
*impl
;
91 nir_ssa_def
*result
; /* result of the expression tree last visited */
93 nir_deref_var
*evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
);
95 /* the head of the dereference chain we're creating */
96 nir_deref_var
*deref_head
;
97 /* the tail of the dereference chain we're creating */
98 nir_deref
*deref_tail
;
100 nir_variable
*var
; /* variable created by ir_variable visitor */
102 /* whether the IR we're operating on is per-function or global */
105 /* map of ir_variable -> nir_variable */
106 struct hash_table
*var_table
;
108 /* map of ir_function_signature -> nir_function_overload */
109 struct hash_table
*overload_table
;
113 * This visitor runs before the main visitor, calling create_function() for
114 * each function so that the main visitor can resolve forward references in
118 class nir_function_visitor
: public ir_hierarchical_visitor
121 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
124 virtual ir_visitor_status
visit_enter(ir_function
*);
127 nir_visitor
*visitor
;
130 }; /* end of anonymous namespace */
133 glsl_to_nir(const struct gl_shader_program
*shader_prog
,
134 gl_shader_stage stage
,
135 const nir_shader_compiler_options
*options
)
137 struct gl_shader
*sh
= shader_prog
->_LinkedShaders
[stage
];
139 nir_shader
*shader
= nir_shader_create(NULL
, stage
, options
);
141 nir_visitor
v1(shader
);
142 nir_function_visitor
v2(&v1
);
144 visit_exec_list(sh
->ir
, &v1
);
146 nir_lower_outputs_to_temporaries(shader
, nir_shader_get_entrypoint(shader
));
148 shader
->info
.name
= ralloc_asprintf(shader
, "GLSL%d", shader_prog
->Name
);
149 if (shader_prog
->Label
)
150 shader
->info
.label
= ralloc_strdup(shader
, shader_prog
->Label
);
151 shader
->info
.num_textures
= _mesa_fls(sh
->Program
->SamplersUsed
);
152 shader
->info
.num_ubos
= sh
->NumUniformBlocks
;
153 shader
->info
.num_abos
= shader_prog
->NumAtomicBuffers
;
154 shader
->info
.num_ssbos
= sh
->NumShaderStorageBlocks
;
155 shader
->info
.num_images
= sh
->NumImages
;
156 shader
->info
.inputs_read
= sh
->Program
->InputsRead
;
157 shader
->info
.outputs_written
= sh
->Program
->OutputsWritten
;
158 shader
->info
.patch_inputs_read
= sh
->Program
->PatchInputsRead
;
159 shader
->info
.patch_outputs_written
= sh
->Program
->PatchOutputsWritten
;
160 shader
->info
.system_values_read
= sh
->Program
->SystemValuesRead
;
161 shader
->info
.uses_texture_gather
= sh
->Program
->UsesGather
;
162 shader
->info
.uses_clip_distance_out
=
163 sh
->Program
->ClipDistanceArraySize
!= 0;
164 shader
->info
.separate_shader
= shader_prog
->SeparateShader
;
165 shader
->info
.has_transform_feedback_varyings
=
166 shader_prog
->TransformFeedback
.NumVarying
> 0;
169 case MESA_SHADER_TESS_CTRL
:
170 shader
->info
.tcs
.vertices_out
= shader_prog
->TessCtrl
.VerticesOut
;
173 case MESA_SHADER_GEOMETRY
:
174 shader
->info
.gs
.vertices_in
= shader_prog
->Geom
.VerticesIn
;
175 shader
->info
.gs
.output_primitive
= sh
->Geom
.OutputType
;
176 shader
->info
.gs
.vertices_out
= sh
->Geom
.VerticesOut
;
177 shader
->info
.gs
.invocations
= sh
->Geom
.Invocations
;
178 shader
->info
.gs
.uses_end_primitive
= shader_prog
->Geom
.UsesEndPrimitive
;
179 shader
->info
.gs
.uses_streams
= shader_prog
->Geom
.UsesStreams
;
182 case MESA_SHADER_FRAGMENT
: {
183 struct gl_fragment_program
*fp
=
184 (struct gl_fragment_program
*)sh
->Program
;
186 shader
->info
.fs
.uses_discard
= fp
->UsesKill
;
187 shader
->info
.fs
.early_fragment_tests
= sh
->EarlyFragmentTests
;
188 shader
->info
.fs
.depth_layout
= fp
->FragDepthLayout
;
192 case MESA_SHADER_COMPUTE
: {
193 struct gl_compute_program
*cp
= (struct gl_compute_program
*)sh
->Program
;
194 shader
->info
.cs
.local_size
[0] = cp
->LocalSize
[0];
195 shader
->info
.cs
.local_size
[1] = cp
->LocalSize
[1];
196 shader
->info
.cs
.local_size
[2] = cp
->LocalSize
[2];
201 break; /* No stage-specific info */
207 nir_visitor::nir_visitor(nir_shader
*shader
)
209 this->supports_ints
= shader
->options
->native_integers
;
210 this->shader
= shader
;
211 this->is_global
= true;
212 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
213 _mesa_key_pointer_equal
);
214 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
215 _mesa_key_pointer_equal
);
218 nir_visitor::~nir_visitor()
220 _mesa_hash_table_destroy(this->var_table
, NULL
);
221 _mesa_hash_table_destroy(this->overload_table
, NULL
);
225 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
228 ralloc_steal(mem_ctx
, this->deref_head
);
229 return this->deref_head
;
232 static nir_constant
*
233 constant_copy(ir_constant
*ir
, void *mem_ctx
)
238 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
240 unsigned total_elems
= ir
->type
->components();
243 ret
->num_elements
= 0;
244 switch (ir
->type
->base_type
) {
246 for (i
= 0; i
< total_elems
; i
++)
247 ret
->value
.u
[i
] = ir
->value
.u
[i
];
251 for (i
= 0; i
< total_elems
; i
++)
252 ret
->value
.i
[i
] = ir
->value
.i
[i
];
255 case GLSL_TYPE_FLOAT
:
256 for (i
= 0; i
< total_elems
; i
++)
257 ret
->value
.f
[i
] = ir
->value
.f
[i
];
260 case GLSL_TYPE_DOUBLE
:
261 for (i
= 0; i
< total_elems
; i
++)
262 ret
->value
.d
[i
] = ir
->value
.d
[i
];
266 for (i
= 0; i
< total_elems
; i
++)
267 ret
->value
.b
[i
] = ir
->value
.b
[i
];
270 case GLSL_TYPE_STRUCT
:
271 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
273 ret
->num_elements
= ir
->type
->length
;
276 foreach_in_list(ir_constant
, field
, &ir
->components
) {
277 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
282 case GLSL_TYPE_ARRAY
:
283 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
285 ret
->num_elements
= ir
->type
->length
;
287 for (i
= 0; i
< ir
->type
->length
; i
++)
288 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
292 unreachable("not reached");
299 nir_visitor::visit(ir_variable
*ir
)
301 nir_variable
*var
= ralloc(shader
, nir_variable
);
302 var
->type
= ir
->type
;
303 var
->name
= ralloc_strdup(var
, ir
->name
);
305 var
->data
.read_only
= ir
->data
.read_only
;
306 var
->data
.centroid
= ir
->data
.centroid
;
307 var
->data
.sample
= ir
->data
.sample
;
308 var
->data
.patch
= ir
->data
.patch
;
309 var
->data
.invariant
= ir
->data
.invariant
;
310 var
->data
.location
= ir
->data
.location
;
312 switch(ir
->data
.mode
) {
314 case ir_var_temporary
:
316 var
->data
.mode
= nir_var_global
;
318 var
->data
.mode
= nir_var_local
;
321 case ir_var_function_in
:
322 case ir_var_function_out
:
323 case ir_var_function_inout
:
324 case ir_var_const_in
:
325 var
->data
.mode
= nir_var_local
;
328 case ir_var_shader_in
:
329 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
330 ir
->data
.location
== VARYING_SLOT_FACE
) {
331 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
332 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
333 var
->data
.mode
= nir_var_system_value
;
334 } else if (shader
->stage
== MESA_SHADER_GEOMETRY
&&
335 ir
->data
.location
== VARYING_SLOT_PRIMITIVE_ID
) {
336 /* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
337 var
->data
.location
= SYSTEM_VALUE_PRIMITIVE_ID
;
338 var
->data
.mode
= nir_var_system_value
;
340 var
->data
.mode
= nir_var_shader_in
;
344 case ir_var_shader_out
:
345 var
->data
.mode
= nir_var_shader_out
;
349 var
->data
.mode
= nir_var_uniform
;
352 case ir_var_shader_storage
:
353 var
->data
.mode
= nir_var_shader_storage
;
356 case ir_var_system_value
:
357 var
->data
.mode
= nir_var_system_value
;
361 unreachable("not reached");
364 var
->data
.interpolation
= ir
->data
.interpolation
;
365 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
366 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
367 var
->data
.explicit_location
= ir
->data
.explicit_location
;
368 var
->data
.explicit_index
= ir
->data
.explicit_index
;
369 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
370 var
->data
.has_initializer
= ir
->data
.has_initializer
;
371 var
->data
.location_frac
= ir
->data
.location_frac
;
373 switch (ir
->data
.depth_layout
) {
374 case ir_depth_layout_none
:
375 var
->data
.depth_layout
= nir_depth_layout_none
;
377 case ir_depth_layout_any
:
378 var
->data
.depth_layout
= nir_depth_layout_any
;
380 case ir_depth_layout_greater
:
381 var
->data
.depth_layout
= nir_depth_layout_greater
;
383 case ir_depth_layout_less
:
384 var
->data
.depth_layout
= nir_depth_layout_less
;
386 case ir_depth_layout_unchanged
:
387 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
390 unreachable("not reached");
393 var
->data
.index
= ir
->data
.index
;
394 var
->data
.binding
= ir
->data
.binding
;
395 var
->data
.offset
= ir
->data
.offset
;
396 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
397 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
398 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
399 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
400 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
401 var
->data
.image
.format
= ir
->data
.image_format
;
402 var
->data
.max_array_access
= ir
->data
.max_array_access
;
404 var
->num_state_slots
= ir
->get_num_state_slots();
405 if (var
->num_state_slots
> 0) {
406 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
407 var
->num_state_slots
);
409 ir_state_slot
*state_slots
= ir
->get_state_slots();
410 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
411 for (unsigned j
= 0; j
< 5; j
++)
412 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
413 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
416 var
->state_slots
= NULL
;
419 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
421 var
->interface_type
= ir
->get_interface_type();
423 if (var
->data
.mode
== nir_var_local
)
424 nir_function_impl_add_variable(impl
, var
);
426 nir_shader_add_variable(shader
, var
);
428 _mesa_hash_table_insert(var_table
, ir
, var
);
433 nir_function_visitor::visit_enter(ir_function
*ir
)
435 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
436 visitor
->create_function(sig
);
438 return visit_continue_with_parent
;
442 nir_visitor::create_function(ir_function_signature
*ir
)
444 if (ir
->is_intrinsic
)
447 nir_function
*func
= nir_function_create(shader
, ir
->function_name());
449 assert(ir
->parameters
.is_empty());
450 assert(ir
->return_type
== glsl_type::void_type
);
452 _mesa_hash_table_insert(this->overload_table
, ir
, func
);
456 nir_visitor::visit(ir_function
*ir
)
458 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
463 nir_visitor::visit(ir_function_signature
*ir
)
465 if (ir
->is_intrinsic
)
468 struct hash_entry
*entry
=
469 _mesa_hash_table_search(this->overload_table
, ir
);
472 nir_function
*func
= (nir_function
*) entry
->data
;
474 if (ir
->is_defined
) {
475 nir_function_impl
*impl
= nir_function_impl_create(func
);
478 assert(strcmp(func
->name
, "main") == 0);
479 assert(ir
->parameters
.is_empty());
480 assert(func
->return_type
== glsl_type::void_type
);
482 this->is_global
= false;
484 nir_builder_init(&b
, impl
);
485 b
.cursor
= nir_after_cf_list(&impl
->body
);
486 visit_exec_list(&ir
->body
, this);
488 this->is_global
= true;
495 nir_visitor::visit(ir_loop
*ir
)
497 nir_loop
*loop
= nir_loop_create(this->shader
);
498 nir_builder_cf_insert(&b
, &loop
->cf_node
);
500 b
.cursor
= nir_after_cf_list(&loop
->body
);
501 visit_exec_list(&ir
->body_instructions
, this);
502 b
.cursor
= nir_after_cf_node(&loop
->cf_node
);
506 nir_visitor::visit(ir_if
*ir
)
509 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
511 nir_if
*if_stmt
= nir_if_create(this->shader
);
512 if_stmt
->condition
= condition
;
513 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
515 b
.cursor
= nir_after_cf_list(&if_stmt
->then_list
);
516 visit_exec_list(&ir
->then_instructions
, this);
518 b
.cursor
= nir_after_cf_list(&if_stmt
->else_list
);
519 visit_exec_list(&ir
->else_instructions
, this);
521 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
525 nir_visitor::visit(ir_discard
*ir
)
528 * discards aren't treated as control flow, because before we lower them
529 * they can appear anywhere in the shader and the stuff after them may still
530 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
531 * discards will be immediately followed by a return.
534 nir_intrinsic_instr
*discard
;
536 discard
= nir_intrinsic_instr_create(this->shader
,
537 nir_intrinsic_discard_if
);
539 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
541 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
544 nir_builder_instr_insert(&b
, &discard
->instr
);
548 nir_visitor::visit(ir_emit_vertex
*ir
)
550 nir_intrinsic_instr
*instr
=
551 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
552 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
553 nir_builder_instr_insert(&b
, &instr
->instr
);
557 nir_visitor::visit(ir_end_primitive
*ir
)
559 nir_intrinsic_instr
*instr
=
560 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
561 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
562 nir_builder_instr_insert(&b
, &instr
->instr
);
566 nir_visitor::visit(ir_loop_jump
*ir
)
570 case ir_loop_jump::jump_break
:
571 type
= nir_jump_break
;
573 case ir_loop_jump::jump_continue
:
574 type
= nir_jump_continue
;
577 unreachable("not reached");
580 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
581 nir_builder_instr_insert(&b
, &instr
->instr
);
585 nir_visitor::visit(ir_return
*ir
)
587 if (ir
->value
!= NULL
) {
588 nir_intrinsic_instr
*copy
=
589 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
591 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
592 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
595 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
596 nir_builder_instr_insert(&b
, &instr
->instr
);
600 nir_visitor::visit(ir_call
*ir
)
602 if (ir
->callee
->is_intrinsic
) {
604 if (strcmp(ir
->callee_name(), "__intrinsic_atomic_read") == 0) {
605 op
= nir_intrinsic_atomic_counter_read_var
;
606 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_increment") == 0) {
607 op
= nir_intrinsic_atomic_counter_inc_var
;
608 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_predecrement") == 0) {
609 op
= nir_intrinsic_atomic_counter_dec_var
;
610 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_load") == 0) {
611 op
= nir_intrinsic_image_load
;
612 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_store") == 0) {
613 op
= nir_intrinsic_image_store
;
614 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_add") == 0) {
615 op
= nir_intrinsic_image_atomic_add
;
616 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_min") == 0) {
617 op
= nir_intrinsic_image_atomic_min
;
618 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_max") == 0) {
619 op
= nir_intrinsic_image_atomic_max
;
620 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_and") == 0) {
621 op
= nir_intrinsic_image_atomic_and
;
622 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_or") == 0) {
623 op
= nir_intrinsic_image_atomic_or
;
624 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_xor") == 0) {
625 op
= nir_intrinsic_image_atomic_xor
;
626 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_exchange") == 0) {
627 op
= nir_intrinsic_image_atomic_exchange
;
628 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_atomic_comp_swap") == 0) {
629 op
= nir_intrinsic_image_atomic_comp_swap
;
630 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier") == 0) {
631 op
= nir_intrinsic_memory_barrier
;
632 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_size") == 0) {
633 op
= nir_intrinsic_image_size
;
634 } else if (strcmp(ir
->callee_name(), "__intrinsic_image_samples") == 0) {
635 op
= nir_intrinsic_image_samples
;
636 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_ssbo") == 0) {
637 op
= nir_intrinsic_store_ssbo
;
638 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_ssbo") == 0) {
639 op
= nir_intrinsic_load_ssbo
;
640 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_add_ssbo") == 0) {
641 op
= nir_intrinsic_ssbo_atomic_add
;
642 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_and_ssbo") == 0) {
643 op
= nir_intrinsic_ssbo_atomic_and
;
644 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_or_ssbo") == 0) {
645 op
= nir_intrinsic_ssbo_atomic_or
;
646 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_xor_ssbo") == 0) {
647 op
= nir_intrinsic_ssbo_atomic_xor
;
648 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_min_ssbo") == 0) {
649 assert(ir
->return_deref
);
650 if (ir
->return_deref
->type
== glsl_type::int_type
)
651 op
= nir_intrinsic_ssbo_atomic_imin
;
652 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
653 op
= nir_intrinsic_ssbo_atomic_umin
;
655 unreachable("Invalid type");
656 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_max_ssbo") == 0) {
657 assert(ir
->return_deref
);
658 if (ir
->return_deref
->type
== glsl_type::int_type
)
659 op
= nir_intrinsic_ssbo_atomic_imax
;
660 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
661 op
= nir_intrinsic_ssbo_atomic_umax
;
663 unreachable("Invalid type");
664 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_exchange_ssbo") == 0) {
665 op
= nir_intrinsic_ssbo_atomic_exchange
;
666 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_comp_swap_ssbo") == 0) {
667 op
= nir_intrinsic_ssbo_atomic_comp_swap
;
668 } else if (strcmp(ir
->callee_name(), "__intrinsic_shader_clock") == 0) {
669 op
= nir_intrinsic_shader_clock
;
670 } else if (strcmp(ir
->callee_name(), "__intrinsic_group_memory_barrier") == 0) {
671 op
= nir_intrinsic_group_memory_barrier
;
672 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_atomic_counter") == 0) {
673 op
= nir_intrinsic_memory_barrier_atomic_counter
;
674 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_buffer") == 0) {
675 op
= nir_intrinsic_memory_barrier_buffer
;
676 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_image") == 0) {
677 op
= nir_intrinsic_memory_barrier_image
;
678 } else if (strcmp(ir
->callee_name(), "__intrinsic_memory_barrier_shared") == 0) {
679 op
= nir_intrinsic_memory_barrier_shared
;
680 } else if (strcmp(ir
->callee_name(), "__intrinsic_load_shared") == 0) {
681 op
= nir_intrinsic_load_shared
;
682 } else if (strcmp(ir
->callee_name(), "__intrinsic_store_shared") == 0) {
683 op
= nir_intrinsic_store_shared
;
684 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_add_shared") == 0) {
685 op
= nir_intrinsic_shared_atomic_add
;
686 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_and_shared") == 0) {
687 op
= nir_intrinsic_shared_atomic_and
;
688 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_or_shared") == 0) {
689 op
= nir_intrinsic_shared_atomic_or
;
690 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_xor_shared") == 0) {
691 op
= nir_intrinsic_shared_atomic_xor
;
692 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_min_shared") == 0) {
693 assert(ir
->return_deref
);
694 if (ir
->return_deref
->type
== glsl_type::int_type
)
695 op
= nir_intrinsic_shared_atomic_imin
;
696 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
697 op
= nir_intrinsic_shared_atomic_umin
;
699 unreachable("Invalid type");
700 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_max_shared") == 0) {
701 assert(ir
->return_deref
);
702 if (ir
->return_deref
->type
== glsl_type::int_type
)
703 op
= nir_intrinsic_shared_atomic_imax
;
704 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
705 op
= nir_intrinsic_shared_atomic_umax
;
707 unreachable("Invalid type");
708 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_exchange_shared") == 0) {
709 op
= nir_intrinsic_shared_atomic_exchange
;
710 } else if (strcmp(ir
->callee_name(), "__intrinsic_atomic_comp_swap_shared") == 0) {
711 op
= nir_intrinsic_shared_atomic_comp_swap
;
713 unreachable("not reached");
716 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
717 nir_dest
*dest
= &instr
->dest
;
720 case nir_intrinsic_atomic_counter_read_var
:
721 case nir_intrinsic_atomic_counter_inc_var
:
722 case nir_intrinsic_atomic_counter_dec_var
: {
723 ir_dereference
*param
=
724 (ir_dereference
*) ir
->actual_parameters
.get_head();
725 instr
->variables
[0] = evaluate_deref(&instr
->instr
, param
);
726 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
727 nir_builder_instr_insert(&b
, &instr
->instr
);
730 case nir_intrinsic_image_load
:
731 case nir_intrinsic_image_store
:
732 case nir_intrinsic_image_atomic_add
:
733 case nir_intrinsic_image_atomic_min
:
734 case nir_intrinsic_image_atomic_max
:
735 case nir_intrinsic_image_atomic_and
:
736 case nir_intrinsic_image_atomic_or
:
737 case nir_intrinsic_image_atomic_xor
:
738 case nir_intrinsic_image_atomic_exchange
:
739 case nir_intrinsic_image_atomic_comp_swap
:
740 case nir_intrinsic_image_samples
:
741 case nir_intrinsic_image_size
: {
742 nir_ssa_undef_instr
*instr_undef
=
743 nir_ssa_undef_instr_create(shader
, 1, 32);
744 nir_builder_instr_insert(&b
, &instr_undef
->instr
);
746 /* Set the image variable dereference. */
747 exec_node
*param
= ir
->actual_parameters
.get_head();
748 ir_dereference
*image
= (ir_dereference
*)param
;
749 const glsl_type
*type
=
750 image
->variable_referenced()->type
->without_array();
752 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
753 param
= param
->get_next();
755 /* Set the intrinsic destination. */
756 if (ir
->return_deref
) {
757 const nir_intrinsic_info
*info
=
758 &nir_intrinsic_infos
[instr
->intrinsic
];
759 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
760 info
->dest_components
, 32, NULL
);
763 if (op
== nir_intrinsic_image_size
||
764 op
== nir_intrinsic_image_samples
) {
765 nir_builder_instr_insert(&b
, &instr
->instr
);
769 /* Set the address argument, extending the coordinate vector to four
772 nir_ssa_def
*src_addr
=
773 evaluate_rvalue((ir_dereference
*)param
);
774 nir_ssa_def
*srcs
[4];
776 for (int i
= 0; i
< 4; i
++) {
777 if (i
< type
->coordinate_components())
778 srcs
[i
] = nir_channel(&b
, src_addr
, i
);
780 srcs
[i
] = &instr_undef
->def
;
783 instr
->src
[0] = nir_src_for_ssa(nir_vec(&b
, srcs
, 4));
784 param
= param
->get_next();
786 /* Set the sample argument, which is undefined for single-sample
789 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
791 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
792 param
= param
->get_next();
794 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
797 /* Set the intrinsic parameters. */
798 if (!param
->is_tail_sentinel()) {
800 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
801 param
= param
->get_next();
804 if (!param
->is_tail_sentinel()) {
806 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
807 param
= param
->get_next();
809 nir_builder_instr_insert(&b
, &instr
->instr
);
812 case nir_intrinsic_memory_barrier
:
813 case nir_intrinsic_group_memory_barrier
:
814 case nir_intrinsic_memory_barrier_atomic_counter
:
815 case nir_intrinsic_memory_barrier_buffer
:
816 case nir_intrinsic_memory_barrier_image
:
817 case nir_intrinsic_memory_barrier_shared
:
818 nir_builder_instr_insert(&b
, &instr
->instr
);
820 case nir_intrinsic_shader_clock
:
821 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
822 nir_builder_instr_insert(&b
, &instr
->instr
);
824 case nir_intrinsic_store_ssbo
: {
825 exec_node
*param
= ir
->actual_parameters
.get_head();
826 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
828 param
= param
->get_next();
829 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
831 param
= param
->get_next();
832 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
834 param
= param
->get_next();
835 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
838 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
839 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(block
));
840 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(offset
));
841 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
842 instr
->num_components
= val
->type
->vector_elements
;
844 nir_builder_instr_insert(&b
, &instr
->instr
);
847 case nir_intrinsic_load_ssbo
: {
848 exec_node
*param
= ir
->actual_parameters
.get_head();
849 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
851 param
= param
->get_next();
852 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
854 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(block
));
855 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
857 const glsl_type
*type
= ir
->return_deref
->var
->type
;
858 instr
->num_components
= type
->vector_elements
;
860 /* Setup destination register */
861 unsigned bit_size
= glsl_get_bit_size(type
->base_type
);
862 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
863 type
->vector_elements
, bit_size
, NULL
);
865 /* Insert the created nir instruction now since in the case of boolean
866 * result we will need to emit another instruction after it
868 nir_builder_instr_insert(&b
, &instr
->instr
);
871 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
872 * consider a true boolean to be ~0. Fix this up with a != 0
875 if (type
->base_type
== GLSL_TYPE_BOOL
) {
876 nir_alu_instr
*load_ssbo_compare
=
877 nir_alu_instr_create(shader
, nir_op_ine
);
878 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
879 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
880 load_ssbo_compare
->src
[1].src
=
881 nir_src_for_ssa(nir_imm_int(&b
, 0));
882 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
883 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
884 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
885 &load_ssbo_compare
->dest
.dest
,
886 type
->vector_elements
, bit_size
, NULL
);
887 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
888 nir_builder_instr_insert(&b
, &load_ssbo_compare
->instr
);
889 dest
= &load_ssbo_compare
->dest
.dest
;
893 case nir_intrinsic_ssbo_atomic_add
:
894 case nir_intrinsic_ssbo_atomic_imin
:
895 case nir_intrinsic_ssbo_atomic_umin
:
896 case nir_intrinsic_ssbo_atomic_imax
:
897 case nir_intrinsic_ssbo_atomic_umax
:
898 case nir_intrinsic_ssbo_atomic_and
:
899 case nir_intrinsic_ssbo_atomic_or
:
900 case nir_intrinsic_ssbo_atomic_xor
:
901 case nir_intrinsic_ssbo_atomic_exchange
:
902 case nir_intrinsic_ssbo_atomic_comp_swap
: {
903 int param_count
= ir
->actual_parameters
.length();
904 assert(param_count
== 3 || param_count
== 4);
907 exec_node
*param
= ir
->actual_parameters
.get_head();
908 ir_instruction
*inst
= (ir_instruction
*) param
;
909 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
912 param
= param
->get_next();
913 inst
= (ir_instruction
*) param
;
914 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
916 /* data1 parameter (this is always present) */
917 param
= param
->get_next();
918 inst
= (ir_instruction
*) param
;
919 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
921 /* data2 parameter (only with atomic_comp_swap) */
922 if (param_count
== 4) {
923 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
924 param
= param
->get_next();
925 inst
= (ir_instruction
*) param
;
926 instr
->src
[3] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
930 assert(ir
->return_deref
);
931 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
932 ir
->return_deref
->type
->vector_elements
, 32, NULL
);
933 nir_builder_instr_insert(&b
, &instr
->instr
);
936 case nir_intrinsic_load_shared
: {
937 exec_node
*param
= ir
->actual_parameters
.get_head();
938 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
940 nir_intrinsic_set_base(instr
, 0);
941 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(offset
));
943 const glsl_type
*type
= ir
->return_deref
->var
->type
;
944 instr
->num_components
= type
->vector_elements
;
946 /* Setup destination register */
947 unsigned bit_size
= glsl_get_bit_size(type
->base_type
);
948 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
949 type
->vector_elements
, bit_size
, NULL
);
951 nir_builder_instr_insert(&b
, &instr
->instr
);
954 case nir_intrinsic_store_shared
: {
955 exec_node
*param
= ir
->actual_parameters
.get_head();
956 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
958 param
= param
->get_next();
959 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
961 param
= param
->get_next();
962 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
965 nir_intrinsic_set_base(instr
, 0);
966 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
968 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
970 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
971 instr
->num_components
= val
->type
->vector_elements
;
973 nir_builder_instr_insert(&b
, &instr
->instr
);
976 case nir_intrinsic_shared_atomic_add
:
977 case nir_intrinsic_shared_atomic_imin
:
978 case nir_intrinsic_shared_atomic_umin
:
979 case nir_intrinsic_shared_atomic_imax
:
980 case nir_intrinsic_shared_atomic_umax
:
981 case nir_intrinsic_shared_atomic_and
:
982 case nir_intrinsic_shared_atomic_or
:
983 case nir_intrinsic_shared_atomic_xor
:
984 case nir_intrinsic_shared_atomic_exchange
:
985 case nir_intrinsic_shared_atomic_comp_swap
: {
986 int param_count
= ir
->actual_parameters
.length();
987 assert(param_count
== 2 || param_count
== 3);
990 exec_node
*param
= ir
->actual_parameters
.get_head();
991 ir_instruction
*inst
= (ir_instruction
*) param
;
992 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
994 /* data1 parameter (this is always present) */
995 param
= param
->get_next();
996 inst
= (ir_instruction
*) param
;
997 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
999 /* data2 parameter (only with atomic_comp_swap) */
1000 if (param_count
== 3) {
1001 assert(op
== nir_intrinsic_shared_atomic_comp_swap
);
1002 param
= param
->get_next();
1003 inst
= (ir_instruction
*) param
;
1005 nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1009 assert(ir
->return_deref
);
1010 unsigned bit_size
= glsl_get_bit_size(ir
->return_deref
->type
->base_type
);
1011 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1012 ir
->return_deref
->type
->vector_elements
,
1014 nir_builder_instr_insert(&b
, &instr
->instr
);
1018 unreachable("not reached");
1021 if (ir
->return_deref
) {
1022 nir_intrinsic_instr
*store_instr
=
1023 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
1024 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
1025 nir_intrinsic_set_write_mask(store_instr
,
1026 (1 << store_instr
->num_components
) - 1);
1028 store_instr
->variables
[0] =
1029 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
1030 store_instr
->src
[0] = nir_src_for_ssa(&dest
->ssa
);
1032 nir_builder_instr_insert(&b
, &store_instr
->instr
);
1038 struct hash_entry
*entry
=
1039 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
1041 nir_function
*callee
= (nir_function
*) entry
->data
;
1043 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
1046 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
1047 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
1051 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
1052 nir_builder_instr_insert(&b
, &instr
->instr
);
1056 nir_visitor::visit(ir_assignment
*ir
)
1058 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
1060 b
.exact
= ir
->lhs
->variable_referenced()->data
.invariant
||
1061 ir
->lhs
->variable_referenced()->data
.precise
;
1063 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
1064 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
1065 /* We're doing a plain-as-can-be copy, so emit a copy_var */
1066 nir_intrinsic_instr
*copy
=
1067 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
1069 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
1070 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
1072 if (ir
->condition
) {
1073 nir_if
*if_stmt
= nir_if_create(this->shader
);
1074 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1075 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1076 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, ©
->instr
);
1077 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1079 nir_builder_instr_insert(&b
, ©
->instr
);
1084 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
1086 ir
->lhs
->accept(this);
1087 nir_deref_var
*lhs_deref
= this->deref_head
;
1088 nir_ssa_def
*src
= evaluate_rvalue(ir
->rhs
);
1090 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
1091 /* GLSL IR will give us the input to the write-masked assignment in a
1092 * single packed vector. So, for example, if the writemask is xzw, then
1093 * we have to swizzle x -> x, y -> z, and z -> w and get the y component
1097 unsigned component
= 0;
1098 for (unsigned i
= 0; i
< 4; i
++) {
1099 swiz
[i
] = ir
->write_mask
& (1 << i
) ? component
++ : 0;
1101 src
= nir_swizzle(&b
, src
, swiz
, num_components
, !supports_ints
);
1104 nir_intrinsic_instr
*store
=
1105 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1106 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1107 nir_intrinsic_set_write_mask(store
, ir
->write_mask
);
1108 nir_deref
*store_deref
= nir_copy_deref(store
, &lhs_deref
->deref
);
1109 store
->variables
[0] = nir_deref_as_var(store_deref
);
1110 store
->src
[0] = nir_src_for_ssa(src
);
1112 if (ir
->condition
) {
1113 nir_if
*if_stmt
= nir_if_create(this->shader
);
1114 if_stmt
->condition
= nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
1115 nir_builder_cf_insert(&b
, &if_stmt
->cf_node
);
1116 nir_instr_insert_after_cf_list(&if_stmt
->then_list
, &store
->instr
);
1117 b
.cursor
= nir_after_cf_node(&if_stmt
->cf_node
);
1119 nir_builder_instr_insert(&b
, &store
->instr
);
1124 * Given an instruction, returns a pointer to its destination or NULL if there
1125 * is no destination.
1127 * Note that this only handles instructions we generate at this level.
1130 get_instr_dest(nir_instr
*instr
)
1132 nir_alu_instr
*alu_instr
;
1133 nir_intrinsic_instr
*intrinsic_instr
;
1134 nir_tex_instr
*tex_instr
;
1136 switch (instr
->type
) {
1137 case nir_instr_type_alu
:
1138 alu_instr
= nir_instr_as_alu(instr
);
1139 return &alu_instr
->dest
.dest
;
1141 case nir_instr_type_intrinsic
:
1142 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1143 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1144 return &intrinsic_instr
->dest
;
1148 case nir_instr_type_tex
:
1149 tex_instr
= nir_instr_as_tex(instr
);
1150 return &tex_instr
->dest
;
1153 unreachable("not reached");
1160 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
,
1163 nir_dest
*dest
= get_instr_dest(instr
);
1166 nir_ssa_dest_init(instr
, dest
, num_components
, bit_size
, NULL
);
1168 nir_builder_instr_insert(&b
, instr
);
1171 assert(dest
->is_ssa
);
1172 this->result
= &dest
->ssa
;
1177 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1180 if (ir
->as_dereference() || ir
->as_constant()) {
1182 * A dereference is being used on the right hand side, which means we
1183 * must emit a variable load.
1186 nir_intrinsic_instr
*load_instr
=
1187 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1188 load_instr
->num_components
= ir
->type
->vector_elements
;
1189 load_instr
->variables
[0] = this->deref_head
;
1190 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1191 unsigned bit_size
= glsl_get_bit_size(ir
->type
->base_type
);
1192 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
, bit_size
);
1195 return this->result
;
1199 type_is_float(glsl_base_type type
)
1201 return type
== GLSL_TYPE_FLOAT
|| type
== GLSL_TYPE_DOUBLE
;
1205 nir_visitor::visit(ir_expression
*ir
)
1207 /* Some special cases */
1208 switch (ir
->operation
) {
1209 case ir_binop_ubo_load
: {
1210 nir_intrinsic_instr
*load
=
1211 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_ubo
);
1212 unsigned bit_size
= glsl_get_bit_size(ir
->type
->base_type
);
1213 load
->num_components
= ir
->type
->vector_elements
;
1214 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1215 load
->src
[1] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1216 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1219 * In UBO's, a true boolean value is any non-zero value, but we consider
1220 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1223 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
)
1224 this->result
= nir_ine(&b
, &load
->dest
.ssa
, nir_imm_int(&b
, 0));
1229 case ir_unop_interpolate_at_centroid
:
1230 case ir_binop_interpolate_at_offset
:
1231 case ir_binop_interpolate_at_sample
: {
1232 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1233 ir_swizzle
*swizzle
= NULL
;
1235 /* the api does not allow a swizzle here, but the varying packing code
1236 * may have pushed one into here.
1238 swizzle
= ir
->operands
[0]->as_swizzle();
1240 deref
= swizzle
->val
->as_dereference();
1244 deref
->accept(this);
1246 nir_intrinsic_op op
;
1247 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1248 switch (ir
->operation
) {
1249 case ir_unop_interpolate_at_centroid
:
1250 op
= nir_intrinsic_interp_var_at_centroid
;
1252 case ir_binop_interpolate_at_offset
:
1253 op
= nir_intrinsic_interp_var_at_offset
;
1255 case ir_binop_interpolate_at_sample
:
1256 op
= nir_intrinsic_interp_var_at_sample
;
1259 unreachable("Invalid interpolation intrinsic");
1262 /* This case can happen if the vertex shader does not write the
1263 * given varying. In this case, the linker will lower it to a
1264 * global variable. Since interpolating a variable makes no
1265 * sense, we'll just turn it into a load which will probably
1266 * eventually end up as an SSA definition.
1268 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1269 op
= nir_intrinsic_load_var
;
1272 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1273 intrin
->num_components
= deref
->type
->vector_elements
;
1274 intrin
->variables
[0] = this->deref_head
;
1275 ralloc_steal(intrin
, intrin
->variables
[0]);
1277 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1278 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1279 intrin
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1281 unsigned bit_size
= glsl_get_bit_size(deref
->type
->base_type
);
1282 add_instr(&intrin
->instr
, deref
->type
->vector_elements
, bit_size
);
1285 unsigned swiz
[4] = {
1286 swizzle
->mask
.x
, swizzle
->mask
.y
, swizzle
->mask
.z
, swizzle
->mask
.w
1289 result
= nir_swizzle(&b
, result
, swiz
,
1290 swizzle
->type
->vector_elements
, false);
1300 nir_ssa_def
*srcs
[4];
1301 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1302 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1304 glsl_base_type types
[4];
1305 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1307 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1309 types
[i
] = GLSL_TYPE_FLOAT
;
1311 glsl_base_type out_type
;
1313 out_type
= ir
->type
->base_type
;
1315 out_type
= GLSL_TYPE_FLOAT
;
1317 switch (ir
->operation
) {
1318 case ir_unop_bit_not
: result
= nir_inot(&b
, srcs
[0]); break;
1319 case ir_unop_logic_not
:
1320 result
= supports_ints
? nir_inot(&b
, srcs
[0]) : nir_fnot(&b
, srcs
[0]);
1323 result
= type_is_float(types
[0]) ? nir_fneg(&b
, srcs
[0])
1324 : nir_ineg(&b
, srcs
[0]);
1327 result
= type_is_float(types
[0]) ? nir_fabs(&b
, srcs
[0])
1328 : nir_iabs(&b
, srcs
[0]);
1330 case ir_unop_saturate
:
1331 assert(type_is_float(types
[0]));
1332 result
= nir_fsat(&b
, srcs
[0]);
1335 result
= type_is_float(types
[0]) ? nir_fsign(&b
, srcs
[0])
1336 : nir_isign(&b
, srcs
[0]);
1338 case ir_unop_rcp
: result
= nir_frcp(&b
, srcs
[0]); break;
1339 case ir_unop_rsq
: result
= nir_frsq(&b
, srcs
[0]); break;
1340 case ir_unop_sqrt
: result
= nir_fsqrt(&b
, srcs
[0]); break;
1341 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1342 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1343 case ir_unop_exp2
: result
= nir_fexp2(&b
, srcs
[0]); break;
1344 case ir_unop_log2
: result
= nir_flog2(&b
, srcs
[0]); break;
1346 result
= supports_ints
? nir_i2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1349 result
= supports_ints
? nir_u2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1352 result
= supports_ints
? nir_b2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1354 case ir_unop_f2i
: result
= nir_f2i(&b
, srcs
[0]); break;
1355 case ir_unop_f2u
: result
= nir_f2u(&b
, srcs
[0]); break;
1356 case ir_unop_f2b
: result
= nir_f2b(&b
, srcs
[0]); break;
1357 case ir_unop_i2b
: result
= nir_i2b(&b
, srcs
[0]); break;
1358 case ir_unop_b2i
: result
= nir_b2i(&b
, srcs
[0]); break;
1359 case ir_unop_d2f
: result
= nir_d2f(&b
, srcs
[0]); break;
1360 case ir_unop_f2d
: result
= nir_f2d(&b
, srcs
[0]); break;
1363 case ir_unop_bitcast_i2f
:
1364 case ir_unop_bitcast_f2i
:
1365 case ir_unop_bitcast_u2f
:
1366 case ir_unop_bitcast_f2u
:
1367 case ir_unop_subroutine_to_int
:
1369 result
= nir_imov(&b
, srcs
[0]);
1371 case ir_unop_trunc
: result
= nir_ftrunc(&b
, srcs
[0]); break;
1372 case ir_unop_ceil
: result
= nir_fceil(&b
, srcs
[0]); break;
1373 case ir_unop_floor
: result
= nir_ffloor(&b
, srcs
[0]); break;
1374 case ir_unop_fract
: result
= nir_ffract(&b
, srcs
[0]); break;
1375 case ir_unop_round_even
: result
= nir_fround_even(&b
, srcs
[0]); break;
1376 case ir_unop_sin
: result
= nir_fsin(&b
, srcs
[0]); break;
1377 case ir_unop_cos
: result
= nir_fcos(&b
, srcs
[0]); break;
1378 case ir_unop_dFdx
: result
= nir_fddx(&b
, srcs
[0]); break;
1379 case ir_unop_dFdy
: result
= nir_fddy(&b
, srcs
[0]); break;
1380 case ir_unop_dFdx_fine
: result
= nir_fddx_fine(&b
, srcs
[0]); break;
1381 case ir_unop_dFdy_fine
: result
= nir_fddy_fine(&b
, srcs
[0]); break;
1382 case ir_unop_dFdx_coarse
: result
= nir_fddx_coarse(&b
, srcs
[0]); break;
1383 case ir_unop_dFdy_coarse
: result
= nir_fddy_coarse(&b
, srcs
[0]); break;
1384 case ir_unop_pack_snorm_2x16
:
1385 result
= nir_pack_snorm_2x16(&b
, srcs
[0]);
1387 case ir_unop_pack_snorm_4x8
:
1388 result
= nir_pack_snorm_4x8(&b
, srcs
[0]);
1390 case ir_unop_pack_unorm_2x16
:
1391 result
= nir_pack_unorm_2x16(&b
, srcs
[0]);
1393 case ir_unop_pack_unorm_4x8
:
1394 result
= nir_pack_unorm_4x8(&b
, srcs
[0]);
1396 case ir_unop_pack_half_2x16
:
1397 result
= nir_pack_half_2x16(&b
, srcs
[0]);
1399 case ir_unop_unpack_snorm_2x16
:
1400 result
= nir_unpack_snorm_2x16(&b
, srcs
[0]);
1402 case ir_unop_unpack_snorm_4x8
:
1403 result
= nir_unpack_snorm_4x8(&b
, srcs
[0]);
1405 case ir_unop_unpack_unorm_2x16
:
1406 result
= nir_unpack_unorm_2x16(&b
, srcs
[0]);
1408 case ir_unop_unpack_unorm_4x8
:
1409 result
= nir_unpack_unorm_4x8(&b
, srcs
[0]);
1411 case ir_unop_unpack_half_2x16
:
1412 result
= nir_unpack_half_2x16(&b
, srcs
[0]);
1414 case ir_unop_bitfield_reverse
:
1415 result
= nir_bitfield_reverse(&b
, srcs
[0]);
1417 case ir_unop_bit_count
:
1418 result
= nir_bit_count(&b
, srcs
[0]);
1420 case ir_unop_find_msb
:
1422 case GLSL_TYPE_UINT
:
1423 result
= nir_ufind_msb(&b
, srcs
[0]);
1426 result
= nir_ifind_msb(&b
, srcs
[0]);
1429 unreachable("Invalid type for findMSB()");
1432 case ir_unop_find_lsb
:
1433 result
= nir_find_lsb(&b
, srcs
[0]);
1437 switch (ir
->type
->vector_elements
) {
1439 switch (ir
->operands
[0]->type
->vector_elements
) {
1440 case 1: result
= nir_fnoise1_1(&b
, srcs
[0]); break;
1441 case 2: result
= nir_fnoise1_2(&b
, srcs
[0]); break;
1442 case 3: result
= nir_fnoise1_3(&b
, srcs
[0]); break;
1443 case 4: result
= nir_fnoise1_4(&b
, srcs
[0]); break;
1444 default: unreachable("not reached");
1448 switch (ir
->operands
[0]->type
->vector_elements
) {
1449 case 1: result
= nir_fnoise2_1(&b
, srcs
[0]); break;
1450 case 2: result
= nir_fnoise2_2(&b
, srcs
[0]); break;
1451 case 3: result
= nir_fnoise2_3(&b
, srcs
[0]); break;
1452 case 4: result
= nir_fnoise2_4(&b
, srcs
[0]); break;
1453 default: unreachable("not reached");
1457 switch (ir
->operands
[0]->type
->vector_elements
) {
1458 case 1: result
= nir_fnoise3_1(&b
, srcs
[0]); break;
1459 case 2: result
= nir_fnoise3_2(&b
, srcs
[0]); break;
1460 case 3: result
= nir_fnoise3_3(&b
, srcs
[0]); break;
1461 case 4: result
= nir_fnoise3_4(&b
, srcs
[0]); break;
1462 default: unreachable("not reached");
1466 switch (ir
->operands
[0]->type
->vector_elements
) {
1467 case 1: result
= nir_fnoise4_1(&b
, srcs
[0]); break;
1468 case 2: result
= nir_fnoise4_2(&b
, srcs
[0]); break;
1469 case 3: result
= nir_fnoise4_3(&b
, srcs
[0]); break;
1470 case 4: result
= nir_fnoise4_4(&b
, srcs
[0]); break;
1471 default: unreachable("not reached");
1475 unreachable("not reached");
1478 case ir_unop_get_buffer_size
: {
1479 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1481 nir_intrinsic_get_buffer_size
);
1482 load
->num_components
= ir
->type
->vector_elements
;
1483 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1484 unsigned bit_size
= glsl_get_bit_size(ir
->type
->base_type
);
1485 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1490 result
= type_is_float(out_type
) ? nir_fadd(&b
, srcs
[0], srcs
[1])
1491 : nir_iadd(&b
, srcs
[0], srcs
[1]);
1494 result
= type_is_float(out_type
) ? nir_fsub(&b
, srcs
[0], srcs
[1])
1495 : nir_isub(&b
, srcs
[0], srcs
[1]);
1498 result
= type_is_float(out_type
) ? nir_fmul(&b
, srcs
[0], srcs
[1])
1499 : nir_imul(&b
, srcs
[0], srcs
[1]);
1502 if (type_is_float(out_type
))
1503 result
= nir_fdiv(&b
, srcs
[0], srcs
[1]);
1504 else if (out_type
== GLSL_TYPE_INT
)
1505 result
= nir_idiv(&b
, srcs
[0], srcs
[1]);
1507 result
= nir_udiv(&b
, srcs
[0], srcs
[1]);
1510 result
= type_is_float(out_type
) ? nir_fmod(&b
, srcs
[0], srcs
[1])
1511 : nir_umod(&b
, srcs
[0], srcs
[1]);
1514 if (type_is_float(out_type
))
1515 result
= nir_fmin(&b
, srcs
[0], srcs
[1]);
1516 else if (out_type
== GLSL_TYPE_INT
)
1517 result
= nir_imin(&b
, srcs
[0], srcs
[1]);
1519 result
= nir_umin(&b
, srcs
[0], srcs
[1]);
1522 if (type_is_float(out_type
))
1523 result
= nir_fmax(&b
, srcs
[0], srcs
[1]);
1524 else if (out_type
== GLSL_TYPE_INT
)
1525 result
= nir_imax(&b
, srcs
[0], srcs
[1]);
1527 result
= nir_umax(&b
, srcs
[0], srcs
[1]);
1529 case ir_binop_pow
: result
= nir_fpow(&b
, srcs
[0], srcs
[1]); break;
1530 case ir_binop_bit_and
: result
= nir_iand(&b
, srcs
[0], srcs
[1]); break;
1531 case ir_binop_bit_or
: result
= nir_ior(&b
, srcs
[0], srcs
[1]); break;
1532 case ir_binop_bit_xor
: result
= nir_ixor(&b
, srcs
[0], srcs
[1]); break;
1533 case ir_binop_logic_and
:
1534 result
= supports_ints
? nir_iand(&b
, srcs
[0], srcs
[1])
1535 : nir_fand(&b
, srcs
[0], srcs
[1]);
1537 case ir_binop_logic_or
:
1538 result
= supports_ints
? nir_ior(&b
, srcs
[0], srcs
[1])
1539 : nir_for(&b
, srcs
[0], srcs
[1]);
1541 case ir_binop_logic_xor
:
1542 result
= supports_ints
? nir_ixor(&b
, srcs
[0], srcs
[1])
1543 : nir_fxor(&b
, srcs
[0], srcs
[1]);
1545 case ir_binop_lshift
: result
= nir_ishl(&b
, srcs
[0], srcs
[1]); break;
1546 case ir_binop_rshift
:
1547 result
= (out_type
== GLSL_TYPE_INT
) ? nir_ishr(&b
, srcs
[0], srcs
[1])
1548 : nir_ushr(&b
, srcs
[0], srcs
[1]);
1550 case ir_binop_imul_high
:
1551 result
= (out_type
== GLSL_TYPE_INT
) ? nir_imul_high(&b
, srcs
[0], srcs
[1])
1552 : nir_umul_high(&b
, srcs
[0], srcs
[1]);
1554 case ir_binop_carry
: result
= nir_uadd_carry(&b
, srcs
[0], srcs
[1]); break;
1555 case ir_binop_borrow
: result
= nir_usub_borrow(&b
, srcs
[0], srcs
[1]); break;
1557 if (supports_ints
) {
1558 if (type_is_float(types
[0]))
1559 result
= nir_flt(&b
, srcs
[0], srcs
[1]);
1560 else if (types
[0] == GLSL_TYPE_INT
)
1561 result
= nir_ilt(&b
, srcs
[0], srcs
[1]);
1563 result
= nir_ult(&b
, srcs
[0], srcs
[1]);
1565 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1568 case ir_binop_greater
:
1569 if (supports_ints
) {
1570 if (type_is_float(types
[0]))
1571 result
= nir_flt(&b
, srcs
[1], srcs
[0]);
1572 else if (types
[0] == GLSL_TYPE_INT
)
1573 result
= nir_ilt(&b
, srcs
[1], srcs
[0]);
1575 result
= nir_ult(&b
, srcs
[1], srcs
[0]);
1577 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1580 case ir_binop_lequal
:
1581 if (supports_ints
) {
1582 if (type_is_float(types
[0]))
1583 result
= nir_fge(&b
, srcs
[1], srcs
[0]);
1584 else if (types
[0] == GLSL_TYPE_INT
)
1585 result
= nir_ige(&b
, srcs
[1], srcs
[0]);
1587 result
= nir_uge(&b
, srcs
[1], srcs
[0]);
1589 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1592 case ir_binop_gequal
:
1593 if (supports_ints
) {
1594 if (type_is_float(types
[0]))
1595 result
= nir_fge(&b
, srcs
[0], srcs
[1]);
1596 else if (types
[0] == GLSL_TYPE_INT
)
1597 result
= nir_ige(&b
, srcs
[0], srcs
[1]);
1599 result
= nir_uge(&b
, srcs
[0], srcs
[1]);
1601 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1604 case ir_binop_equal
:
1605 if (supports_ints
) {
1606 if (type_is_float(types
[0]))
1607 result
= nir_feq(&b
, srcs
[0], srcs
[1]);
1609 result
= nir_ieq(&b
, srcs
[0], srcs
[1]);
1611 result
= nir_seq(&b
, srcs
[0], srcs
[1]);
1614 case ir_binop_nequal
:
1615 if (supports_ints
) {
1616 if (type_is_float(types
[0]))
1617 result
= nir_fne(&b
, srcs
[0], srcs
[1]);
1619 result
= nir_ine(&b
, srcs
[0], srcs
[1]);
1621 result
= nir_sne(&b
, srcs
[0], srcs
[1]);
1624 case ir_binop_all_equal
:
1625 if (supports_ints
) {
1626 if (type_is_float(types
[0])) {
1627 switch (ir
->operands
[0]->type
->vector_elements
) {
1628 case 1: result
= nir_feq(&b
, srcs
[0], srcs
[1]); break;
1629 case 2: result
= nir_ball_fequal2(&b
, srcs
[0], srcs
[1]); break;
1630 case 3: result
= nir_ball_fequal3(&b
, srcs
[0], srcs
[1]); break;
1631 case 4: result
= nir_ball_fequal4(&b
, srcs
[0], srcs
[1]); break;
1633 unreachable("not reached");
1636 switch (ir
->operands
[0]->type
->vector_elements
) {
1637 case 1: result
= nir_ieq(&b
, srcs
[0], srcs
[1]); break;
1638 case 2: result
= nir_ball_iequal2(&b
, srcs
[0], srcs
[1]); break;
1639 case 3: result
= nir_ball_iequal3(&b
, srcs
[0], srcs
[1]); break;
1640 case 4: result
= nir_ball_iequal4(&b
, srcs
[0], srcs
[1]); break;
1642 unreachable("not reached");
1646 switch (ir
->operands
[0]->type
->vector_elements
) {
1647 case 1: result
= nir_seq(&b
, srcs
[0], srcs
[1]); break;
1648 case 2: result
= nir_fall_equal2(&b
, srcs
[0], srcs
[1]); break;
1649 case 3: result
= nir_fall_equal3(&b
, srcs
[0], srcs
[1]); break;
1650 case 4: result
= nir_fall_equal4(&b
, srcs
[0], srcs
[1]); break;
1652 unreachable("not reached");
1656 case ir_binop_any_nequal
:
1657 if (supports_ints
) {
1658 if (type_is_float(types
[0])) {
1659 switch (ir
->operands
[0]->type
->vector_elements
) {
1660 case 1: result
= nir_fne(&b
, srcs
[0], srcs
[1]); break;
1661 case 2: result
= nir_bany_fnequal2(&b
, srcs
[0], srcs
[1]); break;
1662 case 3: result
= nir_bany_fnequal3(&b
, srcs
[0], srcs
[1]); break;
1663 case 4: result
= nir_bany_fnequal4(&b
, srcs
[0], srcs
[1]); break;
1665 unreachable("not reached");
1668 switch (ir
->operands
[0]->type
->vector_elements
) {
1669 case 1: result
= nir_ine(&b
, srcs
[0], srcs
[1]); break;
1670 case 2: result
= nir_bany_inequal2(&b
, srcs
[0], srcs
[1]); break;
1671 case 3: result
= nir_bany_inequal3(&b
, srcs
[0], srcs
[1]); break;
1672 case 4: result
= nir_bany_inequal4(&b
, srcs
[0], srcs
[1]); break;
1674 unreachable("not reached");
1678 switch (ir
->operands
[0]->type
->vector_elements
) {
1679 case 1: result
= nir_sne(&b
, srcs
[0], srcs
[1]); break;
1680 case 2: result
= nir_fany_nequal2(&b
, srcs
[0], srcs
[1]); break;
1681 case 3: result
= nir_fany_nequal3(&b
, srcs
[0], srcs
[1]); break;
1682 case 4: result
= nir_fany_nequal4(&b
, srcs
[0], srcs
[1]); break;
1684 unreachable("not reached");
1689 switch (ir
->operands
[0]->type
->vector_elements
) {
1690 case 2: result
= nir_fdot2(&b
, srcs
[0], srcs
[1]); break;
1691 case 3: result
= nir_fdot3(&b
, srcs
[0], srcs
[1]); break;
1692 case 4: result
= nir_fdot4(&b
, srcs
[0], srcs
[1]); break;
1694 unreachable("not reached");
1698 case ir_binop_ldexp
: result
= nir_ldexp(&b
, srcs
[0], srcs
[1]); break;
1700 result
= nir_ffma(&b
, srcs
[0], srcs
[1], srcs
[2]);
1703 result
= nir_flrp(&b
, srcs
[0], srcs
[1], srcs
[2]);
1707 result
= nir_bcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1709 result
= nir_fcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1711 case ir_triop_bitfield_extract
:
1712 result
= (out_type
== GLSL_TYPE_INT
) ?
1713 nir_ibitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]) :
1714 nir_ubitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]);
1716 case ir_quadop_bitfield_insert
:
1717 result
= nir_bitfield_insert(&b
, srcs
[0], srcs
[1], srcs
[2], srcs
[3]);
1719 case ir_quadop_vector
:
1720 result
= nir_vec(&b
, srcs
, ir
->type
->vector_elements
);
1724 unreachable("not reached");
1729 nir_visitor::visit(ir_swizzle
*ir
)
1731 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1732 result
= nir_swizzle(&b
, evaluate_rvalue(ir
->val
), swizzle
,
1733 ir
->type
->vector_elements
, !supports_ints
);
1737 nir_visitor::visit(ir_texture
*ir
)
1744 num_srcs
= 1; /* coordinate */
1749 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1750 num_srcs
= 2; /* coordinate, bias/lod */
1754 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1760 if (ir
->lod_info
.lod
!= NULL
)
1761 num_srcs
= 2; /* coordinate, lod */
1763 num_srcs
= 1; /* coordinate */
1767 op
= nir_texop_txf_ms
;
1768 num_srcs
= 2; /* coordinate, sample_index */
1773 if (ir
->lod_info
.lod
!= NULL
)
1774 num_srcs
= 1; /* lod */
1781 num_srcs
= 1; /* coordinate */
1786 num_srcs
= 1; /* coordinate */
1789 case ir_query_levels
:
1790 op
= nir_texop_query_levels
;
1794 case ir_texture_samples
:
1795 op
= nir_texop_texture_samples
;
1799 case ir_samples_identical
:
1800 op
= nir_texop_samples_identical
;
1801 num_srcs
= 1; /* coordinate */
1805 unreachable("not reached");
1808 if (ir
->projector
!= NULL
)
1810 if (ir
->shadow_comparitor
!= NULL
)
1812 if (ir
->offset
!= NULL
)
1815 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1818 instr
->sampler_dim
=
1819 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1820 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1821 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1822 if (instr
->is_shadow
)
1823 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1824 switch (ir
->type
->base_type
) {
1825 case GLSL_TYPE_FLOAT
:
1826 instr
->dest_type
= nir_type_float
;
1829 instr
->dest_type
= nir_type_int
;
1831 case GLSL_TYPE_BOOL
:
1832 case GLSL_TYPE_UINT
:
1833 instr
->dest_type
= nir_type_uint
;
1836 unreachable("not reached");
1839 instr
->texture
= evaluate_deref(&instr
->instr
, ir
->sampler
);
1841 unsigned src_number
= 0;
1843 if (ir
->coordinate
!= NULL
) {
1844 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
1845 instr
->src
[src_number
].src
=
1846 nir_src_for_ssa(evaluate_rvalue(ir
->coordinate
));
1847 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
1851 if (ir
->projector
!= NULL
) {
1852 instr
->src
[src_number
].src
=
1853 nir_src_for_ssa(evaluate_rvalue(ir
->projector
));
1854 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
1858 if (ir
->shadow_comparitor
!= NULL
) {
1859 instr
->src
[src_number
].src
=
1860 nir_src_for_ssa(evaluate_rvalue(ir
->shadow_comparitor
));
1861 instr
->src
[src_number
].src_type
= nir_tex_src_comparitor
;
1865 if (ir
->offset
!= NULL
) {
1866 /* we don't support multiple offsets yet */
1867 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
1869 instr
->src
[src_number
].src
=
1870 nir_src_for_ssa(evaluate_rvalue(ir
->offset
));
1871 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
1877 instr
->src
[src_number
].src
=
1878 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.bias
));
1879 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
1886 if (ir
->lod_info
.lod
!= NULL
) {
1887 instr
->src
[src_number
].src
=
1888 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.lod
));
1889 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
1895 instr
->src
[src_number
].src
=
1896 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdx
));
1897 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
1899 instr
->src
[src_number
].src
=
1900 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdy
));
1901 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
1906 instr
->src
[src_number
].src
=
1907 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.sample_index
));
1908 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
1913 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
1920 assert(src_number
== num_srcs
);
1922 unsigned bit_size
= glsl_get_bit_size(ir
->type
->base_type
);
1923 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
), bit_size
);
1927 nir_visitor::visit(ir_constant
*ir
)
1930 * We don't know if this variable is an an array or struct that gets
1931 * dereferenced, so do the safe thing an make it a variable with a
1932 * constant initializer and return a dereference.
1936 nir_local_variable_create(this->impl
, ir
->type
, "const_temp");
1937 var
->data
.read_only
= true;
1938 var
->constant_initializer
= constant_copy(ir
, var
);
1940 this->deref_head
= nir_deref_var_create(this->shader
, var
);
1941 this->deref_tail
= &this->deref_head
->deref
;
1945 nir_visitor::visit(ir_dereference_variable
*ir
)
1947 struct hash_entry
*entry
=
1948 _mesa_hash_table_search(this->var_table
, ir
->var
);
1950 nir_variable
*var
= (nir_variable
*) entry
->data
;
1952 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
1953 this->deref_head
= deref
;
1954 this->deref_tail
= &deref
->deref
;
1958 nir_visitor::visit(ir_dereference_record
*ir
)
1960 ir
->record
->accept(this);
1962 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
1963 assert(field_index
>= 0);
1965 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
1966 deref
->deref
.type
= ir
->type
;
1967 this->deref_tail
->child
= &deref
->deref
;
1968 this->deref_tail
= &deref
->deref
;
1972 nir_visitor::visit(ir_dereference_array
*ir
)
1974 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
1975 deref
->deref
.type
= ir
->type
;
1977 ir_constant
*const_index
= ir
->array_index
->as_constant();
1978 if (const_index
!= NULL
) {
1979 deref
->deref_array_type
= nir_deref_array_type_direct
;
1980 deref
->base_offset
= const_index
->value
.u
[0];
1982 deref
->deref_array_type
= nir_deref_array_type_indirect
;
1984 nir_src_for_ssa(evaluate_rvalue(ir
->array_index
));
1987 ir
->array
->accept(this);
1989 this->deref_tail
->child
= &deref
->deref
;
1990 ralloc_steal(this->deref_tail
, deref
);
1991 this->deref_tail
= &deref
->deref
;
1995 nir_visitor::visit(ir_barrier
*ir
)
1997 nir_intrinsic_instr
*instr
=
1998 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
1999 nir_builder_instr_insert(&b
, &instr
->instr
);