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 "ir_visitor.h"
30 #include "ir_hierarchical_visitor.h"
32 #include "compiler/nir/nir_control_flow.h"
33 #include "compiler/nir/nir_builder.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 nir_remap_attributes(nir_shader
*shader
)
135 nir_foreach_variable(var
, &shader
->inputs
) {
136 var
->data
.location
+= _mesa_bitcount_64(shader
->info
->double_inputs_read
&
137 BITFIELD64_MASK(var
->data
.location
));
140 /* Once the remap is done, reset double_inputs_read, so later it will have
141 * which location/slots are doubles */
142 shader
->info
->double_inputs_read
= 0;
146 glsl_to_nir(const struct gl_shader_program
*shader_prog
,
147 gl_shader_stage stage
,
148 const nir_shader_compiler_options
*options
)
150 struct gl_linked_shader
*sh
= shader_prog
->_LinkedShaders
[stage
];
152 nir_shader
*shader
= nir_shader_create(NULL
, stage
, options
,
155 nir_visitor
v1(shader
);
156 nir_function_visitor
v2(&v1
);
158 visit_exec_list(sh
->ir
, &v1
);
160 nir_lower_constant_initializers(shader
, (nir_variable_mode
)~0);
162 /* Remap the locations to slots so those requiring two slots will occupy
163 * two locations. For instance, if we have in the IR code a dvec3 attr0 in
164 * location 0 and vec4 attr1 in location 1, in NIR attr0 will use
165 * locations/slots 0 and 1, and attr1 will use location/slot 2 */
166 if (shader
->stage
== MESA_SHADER_VERTEX
)
167 nir_remap_attributes(shader
);
169 shader
->info
->name
= ralloc_asprintf(shader
, "GLSL%d", shader_prog
->Name
);
170 if (shader_prog
->Label
)
171 shader
->info
->label
= ralloc_strdup(shader
, shader_prog
->Label
);
172 shader
->info
->has_transform_feedback_varyings
=
173 shader_prog
->TransformFeedback
.NumVarying
> 0;
178 nir_visitor::nir_visitor(nir_shader
*shader
)
180 this->supports_ints
= shader
->options
->native_integers
;
181 this->shader
= shader
;
182 this->is_global
= true;
183 this->var_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
184 _mesa_key_pointer_equal
);
185 this->overload_table
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
186 _mesa_key_pointer_equal
);
190 this->deref_head
= NULL
;
191 this->deref_tail
= NULL
;
192 memset(&this->b
, 0, sizeof(this->b
));
195 nir_visitor::~nir_visitor()
197 _mesa_hash_table_destroy(this->var_table
, NULL
);
198 _mesa_hash_table_destroy(this->overload_table
, NULL
);
202 nir_visitor::evaluate_deref(nir_instr
*mem_ctx
, ir_instruction
*ir
)
205 ralloc_steal(mem_ctx
, this->deref_head
);
206 return this->deref_head
;
209 static nir_constant
*
210 constant_copy(ir_constant
*ir
, void *mem_ctx
)
215 nir_constant
*ret
= ralloc(mem_ctx
, nir_constant
);
217 const unsigned rows
= ir
->type
->vector_elements
;
218 const unsigned cols
= ir
->type
->matrix_columns
;
221 ret
->num_elements
= 0;
222 switch (ir
->type
->base_type
) {
224 /* Only float base types can be matrices. */
227 for (unsigned r
= 0; r
< rows
; r
++)
228 ret
->values
[0].u32
[r
] = ir
->value
.u
[r
];
233 /* Only float base types can be matrices. */
236 for (unsigned r
= 0; r
< rows
; r
++)
237 ret
->values
[0].i32
[r
] = ir
->value
.i
[r
];
241 case GLSL_TYPE_FLOAT
:
242 for (unsigned c
= 0; c
< cols
; c
++) {
243 for (unsigned r
= 0; r
< rows
; r
++)
244 ret
->values
[c
].f32
[r
] = ir
->value
.f
[c
* rows
+ r
];
248 case GLSL_TYPE_DOUBLE
:
249 for (unsigned c
= 0; c
< cols
; c
++) {
250 for (unsigned r
= 0; r
< rows
; r
++)
251 ret
->values
[c
].f64
[r
] = ir
->value
.d
[c
* rows
+ r
];
255 case GLSL_TYPE_UINT64
:
256 /* Only float base types can be matrices. */
259 for (unsigned r
= 0; r
< rows
; r
++)
260 ret
->values
[0].u64
[r
] = ir
->value
.u64
[r
];
263 case GLSL_TYPE_INT64
:
264 /* Only float base types can be matrices. */
267 for (unsigned r
= 0; r
< rows
; r
++)
268 ret
->values
[0].i64
[r
] = ir
->value
.i64
[r
];
272 /* Only float base types can be matrices. */
275 for (unsigned r
= 0; r
< rows
; r
++)
276 ret
->values
[0].u32
[r
] = ir
->value
.b
[r
] ? NIR_TRUE
: NIR_FALSE
;
280 case GLSL_TYPE_STRUCT
:
281 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
283 ret
->num_elements
= ir
->type
->length
;
286 foreach_in_list(ir_constant
, field
, &ir
->components
) {
287 ret
->elements
[i
] = constant_copy(field
, mem_ctx
);
292 case GLSL_TYPE_ARRAY
:
293 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
295 ret
->num_elements
= ir
->type
->length
;
297 for (i
= 0; i
< ir
->type
->length
; i
++)
298 ret
->elements
[i
] = constant_copy(ir
->array_elements
[i
], mem_ctx
);
302 unreachable("not reached");
309 nir_visitor::visit(ir_variable
*ir
)
311 nir_variable
*var
= ralloc(shader
, nir_variable
);
312 var
->type
= ir
->type
;
313 var
->name
= ralloc_strdup(var
, ir
->name
);
315 var
->data
.read_only
= ir
->data
.read_only
;
316 var
->data
.centroid
= ir
->data
.centroid
;
317 var
->data
.sample
= ir
->data
.sample
;
318 var
->data
.patch
= ir
->data
.patch
;
319 var
->data
.invariant
= ir
->data
.invariant
;
320 var
->data
.location
= ir
->data
.location
;
321 var
->data
.compact
= false;
323 switch(ir
->data
.mode
) {
325 case ir_var_temporary
:
327 var
->data
.mode
= nir_var_global
;
329 var
->data
.mode
= nir_var_local
;
332 case ir_var_function_in
:
333 case ir_var_function_out
:
334 case ir_var_function_inout
:
335 case ir_var_const_in
:
336 var
->data
.mode
= nir_var_local
;
339 case ir_var_shader_in
:
340 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
341 ir
->data
.location
== VARYING_SLOT_FACE
) {
342 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
343 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
344 var
->data
.mode
= nir_var_system_value
;
345 } else if (shader
->stage
== MESA_SHADER_GEOMETRY
&&
346 ir
->data
.location
== VARYING_SLOT_PRIMITIVE_ID
) {
347 /* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
348 var
->data
.location
= SYSTEM_VALUE_PRIMITIVE_ID
;
349 var
->data
.mode
= nir_var_system_value
;
351 var
->data
.mode
= nir_var_shader_in
;
353 if (shader
->stage
== MESA_SHADER_TESS_EVAL
&&
354 (ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
||
355 ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
)) {
356 var
->data
.compact
= ir
->type
->without_array()->is_scalar();
360 /* Mark all the locations that require two slots */
361 if (glsl_type_is_dual_slot(glsl_without_array(var
->type
))) {
362 for (uint i
= 0; i
< glsl_count_attribute_slots(var
->type
, true); i
++) {
363 uint64_t bitfield
= BITFIELD64_BIT(var
->data
.location
+ i
);
364 shader
->info
->double_inputs_read
|= bitfield
;
369 case ir_var_shader_out
:
370 var
->data
.mode
= nir_var_shader_out
;
371 if (shader
->stage
== MESA_SHADER_TESS_CTRL
&&
372 (ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
||
373 ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
)) {
374 var
->data
.compact
= ir
->type
->without_array()->is_scalar();
379 var
->data
.mode
= nir_var_uniform
;
382 case ir_var_shader_storage
:
383 var
->data
.mode
= nir_var_shader_storage
;
386 case ir_var_system_value
:
387 var
->data
.mode
= nir_var_system_value
;
391 unreachable("not reached");
394 var
->data
.interpolation
= ir
->data
.interpolation
;
395 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
396 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
397 var
->data
.location_frac
= ir
->data
.location_frac
;
399 switch (ir
->data
.depth_layout
) {
400 case ir_depth_layout_none
:
401 var
->data
.depth_layout
= nir_depth_layout_none
;
403 case ir_depth_layout_any
:
404 var
->data
.depth_layout
= nir_depth_layout_any
;
406 case ir_depth_layout_greater
:
407 var
->data
.depth_layout
= nir_depth_layout_greater
;
409 case ir_depth_layout_less
:
410 var
->data
.depth_layout
= nir_depth_layout_less
;
412 case ir_depth_layout_unchanged
:
413 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
416 unreachable("not reached");
419 var
->data
.index
= ir
->data
.index
;
420 var
->data
.binding
= ir
->data
.binding
;
421 var
->data
.offset
= ir
->data
.offset
;
422 var
->data
.image
.read_only
= ir
->data
.image_read_only
;
423 var
->data
.image
.write_only
= ir
->data
.image_write_only
;
424 var
->data
.image
.coherent
= ir
->data
.image_coherent
;
425 var
->data
.image
._volatile
= ir
->data
.image_volatile
;
426 var
->data
.image
.restrict_flag
= ir
->data
.image_restrict
;
427 var
->data
.image
.format
= ir
->data
.image_format
;
428 var
->data
.fb_fetch_output
= ir
->data
.fb_fetch_output
;
430 var
->num_state_slots
= ir
->get_num_state_slots();
431 if (var
->num_state_slots
> 0) {
432 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
433 var
->num_state_slots
);
435 ir_state_slot
*state_slots
= ir
->get_state_slots();
436 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
437 for (unsigned j
= 0; j
< 5; j
++)
438 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
439 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
442 var
->state_slots
= NULL
;
445 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
447 var
->interface_type
= ir
->get_interface_type();
449 if (var
->data
.mode
== nir_var_local
)
450 nir_function_impl_add_variable(impl
, var
);
452 nir_shader_add_variable(shader
, var
);
454 _mesa_hash_table_insert(var_table
, ir
, var
);
459 nir_function_visitor::visit_enter(ir_function
*ir
)
461 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
462 visitor
->create_function(sig
);
464 return visit_continue_with_parent
;
468 nir_visitor::create_function(ir_function_signature
*ir
)
470 if (ir
->is_intrinsic())
473 nir_function
*func
= nir_function_create(shader
, ir
->function_name());
475 assert(ir
->parameters
.is_empty());
476 assert(ir
->return_type
== glsl_type::void_type
);
478 _mesa_hash_table_insert(this->overload_table
, ir
, func
);
482 nir_visitor::visit(ir_function
*ir
)
484 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
489 nir_visitor::visit(ir_function_signature
*ir
)
491 if (ir
->is_intrinsic())
494 struct hash_entry
*entry
=
495 _mesa_hash_table_search(this->overload_table
, ir
);
498 nir_function
*func
= (nir_function
*) entry
->data
;
500 if (ir
->is_defined
) {
501 nir_function_impl
*impl
= nir_function_impl_create(func
);
504 assert(strcmp(func
->name
, "main") == 0);
505 assert(ir
->parameters
.is_empty());
506 assert(func
->return_type
== glsl_type::void_type
);
508 this->is_global
= false;
510 nir_builder_init(&b
, impl
);
511 b
.cursor
= nir_after_cf_list(&impl
->body
);
512 visit_exec_list(&ir
->body
, this);
514 this->is_global
= true;
521 nir_visitor::visit(ir_loop
*ir
)
524 visit_exec_list(&ir
->body_instructions
, this);
525 nir_pop_loop(&b
, NULL
);
529 nir_visitor::visit(ir_if
*ir
)
531 nir_push_if(&b
, evaluate_rvalue(ir
->condition
));
532 visit_exec_list(&ir
->then_instructions
, this);
533 nir_push_else(&b
, NULL
);
534 visit_exec_list(&ir
->else_instructions
, this);
535 nir_pop_if(&b
, NULL
);
539 nir_visitor::visit(ir_discard
*ir
)
542 * discards aren't treated as control flow, because before we lower them
543 * they can appear anywhere in the shader and the stuff after them may still
544 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
545 * discards will be immediately followed by a return.
548 nir_intrinsic_instr
*discard
;
550 discard
= nir_intrinsic_instr_create(this->shader
,
551 nir_intrinsic_discard_if
);
553 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
555 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
558 nir_builder_instr_insert(&b
, &discard
->instr
);
562 nir_visitor::visit(ir_emit_vertex
*ir
)
564 nir_intrinsic_instr
*instr
=
565 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
566 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
567 nir_builder_instr_insert(&b
, &instr
->instr
);
571 nir_visitor::visit(ir_end_primitive
*ir
)
573 nir_intrinsic_instr
*instr
=
574 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
575 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
576 nir_builder_instr_insert(&b
, &instr
->instr
);
580 nir_visitor::visit(ir_loop_jump
*ir
)
584 case ir_loop_jump::jump_break
:
585 type
= nir_jump_break
;
587 case ir_loop_jump::jump_continue
:
588 type
= nir_jump_continue
;
591 unreachable("not reached");
594 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
595 nir_builder_instr_insert(&b
, &instr
->instr
);
599 nir_visitor::visit(ir_return
*ir
)
601 if (ir
->value
!= NULL
) {
602 nir_intrinsic_instr
*copy
=
603 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
605 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
606 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
609 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
610 nir_builder_instr_insert(&b
, &instr
->instr
);
614 nir_visitor::visit(ir_call
*ir
)
616 if (ir
->callee
->is_intrinsic()) {
619 switch (ir
->callee
->intrinsic_id
) {
620 case ir_intrinsic_atomic_counter_read
:
621 op
= nir_intrinsic_atomic_counter_read_var
;
623 case ir_intrinsic_atomic_counter_increment
:
624 op
= nir_intrinsic_atomic_counter_inc_var
;
626 case ir_intrinsic_atomic_counter_predecrement
:
627 op
= nir_intrinsic_atomic_counter_dec_var
;
629 case ir_intrinsic_atomic_counter_add
:
630 op
= nir_intrinsic_atomic_counter_add_var
;
632 case ir_intrinsic_atomic_counter_and
:
633 op
= nir_intrinsic_atomic_counter_and_var
;
635 case ir_intrinsic_atomic_counter_or
:
636 op
= nir_intrinsic_atomic_counter_or_var
;
638 case ir_intrinsic_atomic_counter_xor
:
639 op
= nir_intrinsic_atomic_counter_xor_var
;
641 case ir_intrinsic_atomic_counter_min
:
642 op
= nir_intrinsic_atomic_counter_min_var
;
644 case ir_intrinsic_atomic_counter_max
:
645 op
= nir_intrinsic_atomic_counter_max_var
;
647 case ir_intrinsic_atomic_counter_exchange
:
648 op
= nir_intrinsic_atomic_counter_exchange_var
;
650 case ir_intrinsic_atomic_counter_comp_swap
:
651 op
= nir_intrinsic_atomic_counter_comp_swap_var
;
653 case ir_intrinsic_image_load
:
654 op
= nir_intrinsic_image_load
;
656 case ir_intrinsic_image_store
:
657 op
= nir_intrinsic_image_store
;
659 case ir_intrinsic_image_atomic_add
:
660 op
= nir_intrinsic_image_atomic_add
;
662 case ir_intrinsic_image_atomic_min
:
663 op
= nir_intrinsic_image_atomic_min
;
665 case ir_intrinsic_image_atomic_max
:
666 op
= nir_intrinsic_image_atomic_max
;
668 case ir_intrinsic_image_atomic_and
:
669 op
= nir_intrinsic_image_atomic_and
;
671 case ir_intrinsic_image_atomic_or
:
672 op
= nir_intrinsic_image_atomic_or
;
674 case ir_intrinsic_image_atomic_xor
:
675 op
= nir_intrinsic_image_atomic_xor
;
677 case ir_intrinsic_image_atomic_exchange
:
678 op
= nir_intrinsic_image_atomic_exchange
;
680 case ir_intrinsic_image_atomic_comp_swap
:
681 op
= nir_intrinsic_image_atomic_comp_swap
;
683 case ir_intrinsic_memory_barrier
:
684 op
= nir_intrinsic_memory_barrier
;
686 case ir_intrinsic_image_size
:
687 op
= nir_intrinsic_image_size
;
689 case ir_intrinsic_image_samples
:
690 op
= nir_intrinsic_image_samples
;
692 case ir_intrinsic_ssbo_store
:
693 op
= nir_intrinsic_store_ssbo
;
695 case ir_intrinsic_ssbo_load
:
696 op
= nir_intrinsic_load_ssbo
;
698 case ir_intrinsic_ssbo_atomic_add
:
699 op
= nir_intrinsic_ssbo_atomic_add
;
701 case ir_intrinsic_ssbo_atomic_and
:
702 op
= nir_intrinsic_ssbo_atomic_and
;
704 case ir_intrinsic_ssbo_atomic_or
:
705 op
= nir_intrinsic_ssbo_atomic_or
;
707 case ir_intrinsic_ssbo_atomic_xor
:
708 op
= nir_intrinsic_ssbo_atomic_xor
;
710 case ir_intrinsic_ssbo_atomic_min
:
711 assert(ir
->return_deref
);
712 if (ir
->return_deref
->type
== glsl_type::int_type
)
713 op
= nir_intrinsic_ssbo_atomic_imin
;
714 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
715 op
= nir_intrinsic_ssbo_atomic_umin
;
717 unreachable("Invalid type");
719 case ir_intrinsic_ssbo_atomic_max
:
720 assert(ir
->return_deref
);
721 if (ir
->return_deref
->type
== glsl_type::int_type
)
722 op
= nir_intrinsic_ssbo_atomic_imax
;
723 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
724 op
= nir_intrinsic_ssbo_atomic_umax
;
726 unreachable("Invalid type");
728 case ir_intrinsic_ssbo_atomic_exchange
:
729 op
= nir_intrinsic_ssbo_atomic_exchange
;
731 case ir_intrinsic_ssbo_atomic_comp_swap
:
732 op
= nir_intrinsic_ssbo_atomic_comp_swap
;
734 case ir_intrinsic_shader_clock
:
735 op
= nir_intrinsic_shader_clock
;
737 case ir_intrinsic_group_memory_barrier
:
738 op
= nir_intrinsic_group_memory_barrier
;
740 case ir_intrinsic_memory_barrier_atomic_counter
:
741 op
= nir_intrinsic_memory_barrier_atomic_counter
;
743 case ir_intrinsic_memory_barrier_buffer
:
744 op
= nir_intrinsic_memory_barrier_buffer
;
746 case ir_intrinsic_memory_barrier_image
:
747 op
= nir_intrinsic_memory_barrier_image
;
749 case ir_intrinsic_memory_barrier_shared
:
750 op
= nir_intrinsic_memory_barrier_shared
;
752 case ir_intrinsic_shared_load
:
753 op
= nir_intrinsic_load_shared
;
755 case ir_intrinsic_shared_store
:
756 op
= nir_intrinsic_store_shared
;
758 case ir_intrinsic_shared_atomic_add
:
759 op
= nir_intrinsic_shared_atomic_add
;
761 case ir_intrinsic_shared_atomic_and
:
762 op
= nir_intrinsic_shared_atomic_and
;
764 case ir_intrinsic_shared_atomic_or
:
765 op
= nir_intrinsic_shared_atomic_or
;
767 case ir_intrinsic_shared_atomic_xor
:
768 op
= nir_intrinsic_shared_atomic_xor
;
770 case ir_intrinsic_shared_atomic_min
:
771 assert(ir
->return_deref
);
772 if (ir
->return_deref
->type
== glsl_type::int_type
)
773 op
= nir_intrinsic_shared_atomic_imin
;
774 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
775 op
= nir_intrinsic_shared_atomic_umin
;
777 unreachable("Invalid type");
779 case ir_intrinsic_shared_atomic_max
:
780 assert(ir
->return_deref
);
781 if (ir
->return_deref
->type
== glsl_type::int_type
)
782 op
= nir_intrinsic_shared_atomic_imax
;
783 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
784 op
= nir_intrinsic_shared_atomic_umax
;
786 unreachable("Invalid type");
788 case ir_intrinsic_shared_atomic_exchange
:
789 op
= nir_intrinsic_shared_atomic_exchange
;
791 case ir_intrinsic_shared_atomic_comp_swap
:
792 op
= nir_intrinsic_shared_atomic_comp_swap
;
795 unreachable("not reached");
798 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
799 nir_dest
*dest
= &instr
->dest
;
802 case nir_intrinsic_atomic_counter_read_var
:
803 case nir_intrinsic_atomic_counter_inc_var
:
804 case nir_intrinsic_atomic_counter_dec_var
:
805 case nir_intrinsic_atomic_counter_add_var
:
806 case nir_intrinsic_atomic_counter_min_var
:
807 case nir_intrinsic_atomic_counter_max_var
:
808 case nir_intrinsic_atomic_counter_and_var
:
809 case nir_intrinsic_atomic_counter_or_var
:
810 case nir_intrinsic_atomic_counter_xor_var
:
811 case nir_intrinsic_atomic_counter_exchange_var
:
812 case nir_intrinsic_atomic_counter_comp_swap_var
: {
813 /* Set the counter variable dereference. */
814 exec_node
*param
= ir
->actual_parameters
.get_head();
815 ir_dereference
*counter
= (ir_dereference
*)param
;
817 instr
->variables
[0] = evaluate_deref(&instr
->instr
, counter
);
818 param
= param
->get_next();
820 /* Set the intrinsic destination. */
821 if (ir
->return_deref
) {
822 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
825 /* Set the intrinsic parameters. */
826 if (!param
->is_tail_sentinel()) {
828 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
829 param
= param
->get_next();
832 if (!param
->is_tail_sentinel()) {
834 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
835 param
= param
->get_next();
838 nir_builder_instr_insert(&b
, &instr
->instr
);
841 case nir_intrinsic_image_load
:
842 case nir_intrinsic_image_store
:
843 case nir_intrinsic_image_atomic_add
:
844 case nir_intrinsic_image_atomic_min
:
845 case nir_intrinsic_image_atomic_max
:
846 case nir_intrinsic_image_atomic_and
:
847 case nir_intrinsic_image_atomic_or
:
848 case nir_intrinsic_image_atomic_xor
:
849 case nir_intrinsic_image_atomic_exchange
:
850 case nir_intrinsic_image_atomic_comp_swap
:
851 case nir_intrinsic_image_samples
:
852 case nir_intrinsic_image_size
: {
853 nir_ssa_undef_instr
*instr_undef
=
854 nir_ssa_undef_instr_create(shader
, 1, 32);
855 nir_builder_instr_insert(&b
, &instr_undef
->instr
);
857 /* Set the image variable dereference. */
858 exec_node
*param
= ir
->actual_parameters
.get_head();
859 ir_dereference
*image
= (ir_dereference
*)param
;
860 const glsl_type
*type
=
861 image
->variable_referenced()->type
->without_array();
863 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
864 param
= param
->get_next();
866 /* Set the intrinsic destination. */
867 if (ir
->return_deref
) {
868 const nir_intrinsic_info
*info
=
869 &nir_intrinsic_infos
[instr
->intrinsic
];
870 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
871 info
->dest_components
, 32, NULL
);
874 if (op
== nir_intrinsic_image_size
||
875 op
== nir_intrinsic_image_samples
) {
876 nir_builder_instr_insert(&b
, &instr
->instr
);
880 /* Set the address argument, extending the coordinate vector to four
883 nir_ssa_def
*src_addr
=
884 evaluate_rvalue((ir_dereference
*)param
);
885 nir_ssa_def
*srcs
[4];
887 for (int i
= 0; i
< 4; i
++) {
888 if (i
< type
->coordinate_components())
889 srcs
[i
] = nir_channel(&b
, src_addr
, i
);
891 srcs
[i
] = &instr_undef
->def
;
894 instr
->src
[0] = nir_src_for_ssa(nir_vec(&b
, srcs
, 4));
895 param
= param
->get_next();
897 /* Set the sample argument, which is undefined for single-sample
900 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
902 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
903 param
= param
->get_next();
905 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
908 /* Set the intrinsic parameters. */
909 if (!param
->is_tail_sentinel()) {
911 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
912 param
= param
->get_next();
915 if (!param
->is_tail_sentinel()) {
917 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
918 param
= param
->get_next();
920 nir_builder_instr_insert(&b
, &instr
->instr
);
923 case nir_intrinsic_memory_barrier
:
924 case nir_intrinsic_group_memory_barrier
:
925 case nir_intrinsic_memory_barrier_atomic_counter
:
926 case nir_intrinsic_memory_barrier_buffer
:
927 case nir_intrinsic_memory_barrier_image
:
928 case nir_intrinsic_memory_barrier_shared
:
929 nir_builder_instr_insert(&b
, &instr
->instr
);
931 case nir_intrinsic_shader_clock
:
932 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
933 nir_builder_instr_insert(&b
, &instr
->instr
);
935 case nir_intrinsic_store_ssbo
: {
936 exec_node
*param
= ir
->actual_parameters
.get_head();
937 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
939 param
= param
->get_next();
940 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
942 param
= param
->get_next();
943 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
945 param
= param
->get_next();
946 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
949 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
950 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(block
));
951 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(offset
));
952 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
953 instr
->num_components
= val
->type
->vector_elements
;
955 nir_builder_instr_insert(&b
, &instr
->instr
);
958 case nir_intrinsic_load_ssbo
: {
959 exec_node
*param
= ir
->actual_parameters
.get_head();
960 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
962 param
= param
->get_next();
963 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
965 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(block
));
966 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
968 const glsl_type
*type
= ir
->return_deref
->var
->type
;
969 instr
->num_components
= type
->vector_elements
;
971 /* Setup destination register */
972 unsigned bit_size
= glsl_get_bit_size(type
);
973 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
974 type
->vector_elements
, bit_size
, NULL
);
976 /* Insert the created nir instruction now since in the case of boolean
977 * result we will need to emit another instruction after it
979 nir_builder_instr_insert(&b
, &instr
->instr
);
982 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
983 * consider a true boolean to be ~0. Fix this up with a != 0
986 if (type
->base_type
== GLSL_TYPE_BOOL
) {
987 nir_alu_instr
*load_ssbo_compare
=
988 nir_alu_instr_create(shader
, nir_op_ine
);
989 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
990 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
991 load_ssbo_compare
->src
[1].src
=
992 nir_src_for_ssa(nir_imm_int(&b
, 0));
993 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
994 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
995 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
996 &load_ssbo_compare
->dest
.dest
,
997 type
->vector_elements
, bit_size
, NULL
);
998 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
999 nir_builder_instr_insert(&b
, &load_ssbo_compare
->instr
);
1000 dest
= &load_ssbo_compare
->dest
.dest
;
1004 case nir_intrinsic_ssbo_atomic_add
:
1005 case nir_intrinsic_ssbo_atomic_imin
:
1006 case nir_intrinsic_ssbo_atomic_umin
:
1007 case nir_intrinsic_ssbo_atomic_imax
:
1008 case nir_intrinsic_ssbo_atomic_umax
:
1009 case nir_intrinsic_ssbo_atomic_and
:
1010 case nir_intrinsic_ssbo_atomic_or
:
1011 case nir_intrinsic_ssbo_atomic_xor
:
1012 case nir_intrinsic_ssbo_atomic_exchange
:
1013 case nir_intrinsic_ssbo_atomic_comp_swap
: {
1014 int param_count
= ir
->actual_parameters
.length();
1015 assert(param_count
== 3 || param_count
== 4);
1018 exec_node
*param
= ir
->actual_parameters
.get_head();
1019 ir_instruction
*inst
= (ir_instruction
*) param
;
1020 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1023 param
= param
->get_next();
1024 inst
= (ir_instruction
*) param
;
1025 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1027 /* data1 parameter (this is always present) */
1028 param
= param
->get_next();
1029 inst
= (ir_instruction
*) param
;
1030 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1032 /* data2 parameter (only with atomic_comp_swap) */
1033 if (param_count
== 4) {
1034 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
1035 param
= param
->get_next();
1036 inst
= (ir_instruction
*) param
;
1037 instr
->src
[3] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1041 assert(ir
->return_deref
);
1042 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1043 ir
->return_deref
->type
->vector_elements
, 32, NULL
);
1044 nir_builder_instr_insert(&b
, &instr
->instr
);
1047 case nir_intrinsic_load_shared
: {
1048 exec_node
*param
= ir
->actual_parameters
.get_head();
1049 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1051 nir_intrinsic_set_base(instr
, 0);
1052 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(offset
));
1054 const glsl_type
*type
= ir
->return_deref
->var
->type
;
1055 instr
->num_components
= type
->vector_elements
;
1057 /* Setup destination register */
1058 unsigned bit_size
= glsl_get_bit_size(type
);
1059 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1060 type
->vector_elements
, bit_size
, NULL
);
1062 nir_builder_instr_insert(&b
, &instr
->instr
);
1065 case nir_intrinsic_store_shared
: {
1066 exec_node
*param
= ir
->actual_parameters
.get_head();
1067 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1069 param
= param
->get_next();
1070 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
1072 param
= param
->get_next();
1073 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
1076 nir_intrinsic_set_base(instr
, 0);
1077 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
1079 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
1081 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
1082 instr
->num_components
= val
->type
->vector_elements
;
1084 nir_builder_instr_insert(&b
, &instr
->instr
);
1087 case nir_intrinsic_shared_atomic_add
:
1088 case nir_intrinsic_shared_atomic_imin
:
1089 case nir_intrinsic_shared_atomic_umin
:
1090 case nir_intrinsic_shared_atomic_imax
:
1091 case nir_intrinsic_shared_atomic_umax
:
1092 case nir_intrinsic_shared_atomic_and
:
1093 case nir_intrinsic_shared_atomic_or
:
1094 case nir_intrinsic_shared_atomic_xor
:
1095 case nir_intrinsic_shared_atomic_exchange
:
1096 case nir_intrinsic_shared_atomic_comp_swap
: {
1097 int param_count
= ir
->actual_parameters
.length();
1098 assert(param_count
== 2 || param_count
== 3);
1101 exec_node
*param
= ir
->actual_parameters
.get_head();
1102 ir_instruction
*inst
= (ir_instruction
*) param
;
1103 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1105 /* data1 parameter (this is always present) */
1106 param
= param
->get_next();
1107 inst
= (ir_instruction
*) param
;
1108 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1110 /* data2 parameter (only with atomic_comp_swap) */
1111 if (param_count
== 3) {
1112 assert(op
== nir_intrinsic_shared_atomic_comp_swap
);
1113 param
= param
->get_next();
1114 inst
= (ir_instruction
*) param
;
1116 nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1120 assert(ir
->return_deref
);
1121 unsigned bit_size
= glsl_get_bit_size(ir
->return_deref
->type
);
1122 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1123 ir
->return_deref
->type
->vector_elements
,
1125 nir_builder_instr_insert(&b
, &instr
->instr
);
1129 unreachable("not reached");
1132 if (ir
->return_deref
) {
1133 nir_intrinsic_instr
*store_instr
=
1134 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
1135 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
1136 nir_intrinsic_set_write_mask(store_instr
,
1137 (1 << store_instr
->num_components
) - 1);
1139 store_instr
->variables
[0] =
1140 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
1141 store_instr
->src
[0] = nir_src_for_ssa(&dest
->ssa
);
1143 nir_builder_instr_insert(&b
, &store_instr
->instr
);
1149 struct hash_entry
*entry
=
1150 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
1152 nir_function
*callee
= (nir_function
*) entry
->data
;
1154 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
1157 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
1158 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
1162 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
1163 nir_builder_instr_insert(&b
, &instr
->instr
);
1167 nir_visitor::visit(ir_assignment
*ir
)
1169 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
1171 b
.exact
= ir
->lhs
->variable_referenced()->data
.invariant
||
1172 ir
->lhs
->variable_referenced()->data
.precise
;
1174 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
1175 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
1176 /* We're doing a plain-as-can-be copy, so emit a copy_var */
1177 nir_intrinsic_instr
*copy
=
1178 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
1180 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
1181 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
1183 if (ir
->condition
) {
1184 nir_push_if(&b
, evaluate_rvalue(ir
->condition
));
1185 nir_builder_instr_insert(&b
, ©
->instr
);
1186 nir_pop_if(&b
, NULL
);
1188 nir_builder_instr_insert(&b
, ©
->instr
);
1193 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
1195 ir
->lhs
->accept(this);
1196 nir_deref_var
*lhs_deref
= this->deref_head
;
1197 nir_ssa_def
*src
= evaluate_rvalue(ir
->rhs
);
1199 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
1200 /* GLSL IR will give us the input to the write-masked assignment in a
1201 * single packed vector. So, for example, if the writemask is xzw, then
1202 * we have to swizzle x -> x, y -> z, and z -> w and get the y component
1206 unsigned component
= 0;
1207 for (unsigned i
= 0; i
< 4; i
++) {
1208 swiz
[i
] = ir
->write_mask
& (1 << i
) ? component
++ : 0;
1210 src
= nir_swizzle(&b
, src
, swiz
, num_components
, !supports_ints
);
1213 nir_intrinsic_instr
*store
=
1214 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1215 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1216 nir_intrinsic_set_write_mask(store
, ir
->write_mask
);
1217 store
->variables
[0] = nir_deref_var_clone(lhs_deref
, store
);
1218 store
->src
[0] = nir_src_for_ssa(src
);
1220 if (ir
->condition
) {
1221 nir_push_if(&b
, evaluate_rvalue(ir
->condition
));
1222 nir_builder_instr_insert(&b
, &store
->instr
);
1223 nir_pop_if(&b
, NULL
);
1225 nir_builder_instr_insert(&b
, &store
->instr
);
1230 * Given an instruction, returns a pointer to its destination or NULL if there
1231 * is no destination.
1233 * Note that this only handles instructions we generate at this level.
1236 get_instr_dest(nir_instr
*instr
)
1238 nir_alu_instr
*alu_instr
;
1239 nir_intrinsic_instr
*intrinsic_instr
;
1240 nir_tex_instr
*tex_instr
;
1242 switch (instr
->type
) {
1243 case nir_instr_type_alu
:
1244 alu_instr
= nir_instr_as_alu(instr
);
1245 return &alu_instr
->dest
.dest
;
1247 case nir_instr_type_intrinsic
:
1248 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1249 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1250 return &intrinsic_instr
->dest
;
1254 case nir_instr_type_tex
:
1255 tex_instr
= nir_instr_as_tex(instr
);
1256 return &tex_instr
->dest
;
1259 unreachable("not reached");
1266 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
,
1269 nir_dest
*dest
= get_instr_dest(instr
);
1272 nir_ssa_dest_init(instr
, dest
, num_components
, bit_size
, NULL
);
1274 nir_builder_instr_insert(&b
, instr
);
1277 assert(dest
->is_ssa
);
1278 this->result
= &dest
->ssa
;
1283 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1286 if (ir
->as_dereference() || ir
->as_constant()) {
1288 * A dereference is being used on the right hand side, which means we
1289 * must emit a variable load.
1292 nir_intrinsic_instr
*load_instr
=
1293 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1294 load_instr
->num_components
= ir
->type
->vector_elements
;
1295 load_instr
->variables
[0] = this->deref_head
;
1296 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1297 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1298 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
, bit_size
);
1301 return this->result
;
1305 type_is_float(glsl_base_type type
)
1307 return type
== GLSL_TYPE_FLOAT
|| type
== GLSL_TYPE_DOUBLE
;
1311 type_is_signed(glsl_base_type type
)
1313 return type
== GLSL_TYPE_INT
|| type
== GLSL_TYPE_INT64
;
1317 nir_visitor::visit(ir_expression
*ir
)
1319 /* Some special cases */
1320 switch (ir
->operation
) {
1321 case ir_binop_ubo_load
: {
1322 nir_intrinsic_instr
*load
=
1323 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_ubo
);
1324 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1325 load
->num_components
= ir
->type
->vector_elements
;
1326 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1327 load
->src
[1] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1328 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1331 * In UBO's, a true boolean value is any non-zero value, but we consider
1332 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1335 if (ir
->type
->base_type
== GLSL_TYPE_BOOL
)
1336 this->result
= nir_ine(&b
, &load
->dest
.ssa
, nir_imm_int(&b
, 0));
1341 case ir_unop_interpolate_at_centroid
:
1342 case ir_binop_interpolate_at_offset
:
1343 case ir_binop_interpolate_at_sample
: {
1344 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1345 ir_swizzle
*swizzle
= NULL
;
1347 /* the api does not allow a swizzle here, but the varying packing code
1348 * may have pushed one into here.
1350 swizzle
= ir
->operands
[0]->as_swizzle();
1352 deref
= swizzle
->val
->as_dereference();
1356 deref
->accept(this);
1358 nir_intrinsic_op op
;
1359 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1360 switch (ir
->operation
) {
1361 case ir_unop_interpolate_at_centroid
:
1362 op
= nir_intrinsic_interp_var_at_centroid
;
1364 case ir_binop_interpolate_at_offset
:
1365 op
= nir_intrinsic_interp_var_at_offset
;
1367 case ir_binop_interpolate_at_sample
:
1368 op
= nir_intrinsic_interp_var_at_sample
;
1371 unreachable("Invalid interpolation intrinsic");
1374 /* This case can happen if the vertex shader does not write the
1375 * given varying. In this case, the linker will lower it to a
1376 * global variable. Since interpolating a variable makes no
1377 * sense, we'll just turn it into a load which will probably
1378 * eventually end up as an SSA definition.
1380 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1381 op
= nir_intrinsic_load_var
;
1384 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1385 intrin
->num_components
= deref
->type
->vector_elements
;
1386 intrin
->variables
[0] = this->deref_head
;
1387 ralloc_steal(intrin
, intrin
->variables
[0]);
1389 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1390 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1391 intrin
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1393 unsigned bit_size
= glsl_get_bit_size(deref
->type
);
1394 add_instr(&intrin
->instr
, deref
->type
->vector_elements
, bit_size
);
1397 unsigned swiz
[4] = {
1398 swizzle
->mask
.x
, swizzle
->mask
.y
, swizzle
->mask
.z
, swizzle
->mask
.w
1401 result
= nir_swizzle(&b
, result
, swiz
,
1402 swizzle
->type
->vector_elements
, false);
1412 nir_ssa_def
*srcs
[4];
1413 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1414 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1416 glsl_base_type types
[4];
1417 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1419 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1421 types
[i
] = GLSL_TYPE_FLOAT
;
1423 glsl_base_type out_type
;
1425 out_type
= ir
->type
->base_type
;
1427 out_type
= GLSL_TYPE_FLOAT
;
1429 switch (ir
->operation
) {
1430 case ir_unop_bit_not
: result
= nir_inot(&b
, srcs
[0]); break;
1431 case ir_unop_logic_not
:
1432 result
= supports_ints
? nir_inot(&b
, srcs
[0]) : nir_fnot(&b
, srcs
[0]);
1435 result
= type_is_float(types
[0]) ? nir_fneg(&b
, srcs
[0])
1436 : nir_ineg(&b
, srcs
[0]);
1439 result
= type_is_float(types
[0]) ? nir_fabs(&b
, srcs
[0])
1440 : nir_iabs(&b
, srcs
[0]);
1442 case ir_unop_saturate
:
1443 assert(type_is_float(types
[0]));
1444 result
= nir_fsat(&b
, srcs
[0]);
1447 result
= type_is_float(types
[0]) ? nir_fsign(&b
, srcs
[0])
1448 : nir_isign(&b
, srcs
[0]);
1450 case ir_unop_rcp
: result
= nir_frcp(&b
, srcs
[0]); break;
1451 case ir_unop_rsq
: result
= nir_frsq(&b
, srcs
[0]); break;
1452 case ir_unop_sqrt
: result
= nir_fsqrt(&b
, srcs
[0]); break;
1453 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1454 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1455 case ir_unop_exp2
: result
= nir_fexp2(&b
, srcs
[0]); break;
1456 case ir_unop_log2
: result
= nir_flog2(&b
, srcs
[0]); break;
1458 result
= supports_ints
? nir_i2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1461 result
= supports_ints
? nir_u2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1464 result
= supports_ints
? nir_b2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1466 case ir_unop_f2i
: result
= nir_f2i(&b
, srcs
[0]); break;
1467 case ir_unop_f2u
: result
= nir_f2u(&b
, srcs
[0]); break;
1468 case ir_unop_f2b
: result
= nir_f2b(&b
, srcs
[0]); break;
1469 case ir_unop_i2b
: result
= nir_i2b(&b
, srcs
[0]); break;
1470 case ir_unop_b2i
: result
= nir_b2i(&b
, srcs
[0]); break;
1471 case ir_unop_b2i64
:result
= nir_b2i64(&b
, srcs
[0]); break;
1472 case ir_unop_d2f
: result
= nir_d2f(&b
, srcs
[0]); break;
1473 case ir_unop_f2d
: result
= nir_f2d(&b
, srcs
[0]); break;
1474 case ir_unop_d2i
: result
= nir_d2i(&b
, srcs
[0]); break;
1475 case ir_unop_d2u
: result
= nir_d2u(&b
, srcs
[0]); break;
1476 case ir_unop_d2b
: result
= nir_d2b(&b
, srcs
[0]); break;
1478 assert(supports_ints
);
1479 result
= nir_i2d(&b
, srcs
[0]);
1482 assert(supports_ints
);
1483 result
= nir_u2d(&b
, srcs
[0]);
1485 case ir_unop_i642i
: result
= nir_i2i32(&b
, srcs
[0]); break;
1486 case ir_unop_i642u
: result
= nir_i2u32(&b
, srcs
[0]); break;
1487 case ir_unop_i642f
: result
= nir_i642f(&b
, srcs
[0]); break;
1488 case ir_unop_i642b
: result
= nir_i642b(&b
, srcs
[0]); break;
1489 case ir_unop_i642d
: result
= nir_i642d(&b
, srcs
[0]); break;
1491 case ir_unop_u642i
: result
= nir_u2i32(&b
, srcs
[0]); break;
1492 case ir_unop_u642u
: result
= nir_u2u32(&b
, srcs
[0]); break;
1493 case ir_unop_u642f
: result
= nir_u642f(&b
, srcs
[0]); break;
1494 case ir_unop_u642d
: result
= nir_u642d(&b
, srcs
[0]); break;
1496 case ir_unop_i2i64
: result
= nir_i2i64(&b
, srcs
[0]); break;
1497 case ir_unop_u2i64
: result
= nir_u2i64(&b
, srcs
[0]); break;
1500 result
= nir_f2i64(&b
, srcs
[0]);
1502 case ir_unop_i2u64
: result
= nir_i2u64(&b
, srcs
[0]); break;
1503 case ir_unop_u2u64
: result
= nir_u2u64(&b
, srcs
[0]); break;
1506 result
= nir_f2u64(&b
, srcs
[0]);
1510 case ir_unop_i642u64
:
1511 case ir_unop_u642i64
:
1512 case ir_unop_bitcast_i2f
:
1513 case ir_unop_bitcast_f2i
:
1514 case ir_unop_bitcast_u2f
:
1515 case ir_unop_bitcast_f2u
:
1516 case ir_unop_bitcast_i642d
:
1517 case ir_unop_bitcast_d2i64
:
1518 case ir_unop_bitcast_u642d
:
1519 case ir_unop_bitcast_d2u64
:
1520 case ir_unop_subroutine_to_int
:
1522 result
= nir_imov(&b
, srcs
[0]);
1524 case ir_unop_trunc
: result
= nir_ftrunc(&b
, srcs
[0]); break;
1525 case ir_unop_ceil
: result
= nir_fceil(&b
, srcs
[0]); break;
1526 case ir_unop_floor
: result
= nir_ffloor(&b
, srcs
[0]); break;
1527 case ir_unop_fract
: result
= nir_ffract(&b
, srcs
[0]); break;
1528 case ir_unop_round_even
: result
= nir_fround_even(&b
, srcs
[0]); break;
1529 case ir_unop_sin
: result
= nir_fsin(&b
, srcs
[0]); break;
1530 case ir_unop_cos
: result
= nir_fcos(&b
, srcs
[0]); break;
1531 case ir_unop_dFdx
: result
= nir_fddx(&b
, srcs
[0]); break;
1532 case ir_unop_dFdy
: result
= nir_fddy(&b
, srcs
[0]); break;
1533 case ir_unop_dFdx_fine
: result
= nir_fddx_fine(&b
, srcs
[0]); break;
1534 case ir_unop_dFdy_fine
: result
= nir_fddy_fine(&b
, srcs
[0]); break;
1535 case ir_unop_dFdx_coarse
: result
= nir_fddx_coarse(&b
, srcs
[0]); break;
1536 case ir_unop_dFdy_coarse
: result
= nir_fddy_coarse(&b
, srcs
[0]); break;
1537 case ir_unop_pack_snorm_2x16
:
1538 result
= nir_pack_snorm_2x16(&b
, srcs
[0]);
1540 case ir_unop_pack_snorm_4x8
:
1541 result
= nir_pack_snorm_4x8(&b
, srcs
[0]);
1543 case ir_unop_pack_unorm_2x16
:
1544 result
= nir_pack_unorm_2x16(&b
, srcs
[0]);
1546 case ir_unop_pack_unorm_4x8
:
1547 result
= nir_pack_unorm_4x8(&b
, srcs
[0]);
1549 case ir_unop_pack_half_2x16
:
1550 result
= nir_pack_half_2x16(&b
, srcs
[0]);
1552 case ir_unop_unpack_snorm_2x16
:
1553 result
= nir_unpack_snorm_2x16(&b
, srcs
[0]);
1555 case ir_unop_unpack_snorm_4x8
:
1556 result
= nir_unpack_snorm_4x8(&b
, srcs
[0]);
1558 case ir_unop_unpack_unorm_2x16
:
1559 result
= nir_unpack_unorm_2x16(&b
, srcs
[0]);
1561 case ir_unop_unpack_unorm_4x8
:
1562 result
= nir_unpack_unorm_4x8(&b
, srcs
[0]);
1564 case ir_unop_unpack_half_2x16
:
1565 result
= nir_unpack_half_2x16(&b
, srcs
[0]);
1567 case ir_unop_pack_double_2x32
:
1568 case ir_unop_pack_int_2x32
:
1569 case ir_unop_pack_uint_2x32
:
1570 result
= nir_pack_64_2x32(&b
, srcs
[0]);
1572 case ir_unop_unpack_double_2x32
:
1573 case ir_unop_unpack_int_2x32
:
1574 case ir_unop_unpack_uint_2x32
:
1575 result
= nir_unpack_64_2x32(&b
, srcs
[0]);
1577 case ir_unop_bitfield_reverse
:
1578 result
= nir_bitfield_reverse(&b
, srcs
[0]);
1580 case ir_unop_bit_count
:
1581 result
= nir_bit_count(&b
, srcs
[0]);
1583 case ir_unop_find_msb
:
1585 case GLSL_TYPE_UINT
:
1586 result
= nir_ufind_msb(&b
, srcs
[0]);
1589 result
= nir_ifind_msb(&b
, srcs
[0]);
1592 unreachable("Invalid type for findMSB()");
1595 case ir_unop_find_lsb
:
1596 result
= nir_find_lsb(&b
, srcs
[0]);
1600 switch (ir
->type
->vector_elements
) {
1602 switch (ir
->operands
[0]->type
->vector_elements
) {
1603 case 1: result
= nir_fnoise1_1(&b
, srcs
[0]); break;
1604 case 2: result
= nir_fnoise1_2(&b
, srcs
[0]); break;
1605 case 3: result
= nir_fnoise1_3(&b
, srcs
[0]); break;
1606 case 4: result
= nir_fnoise1_4(&b
, srcs
[0]); break;
1607 default: unreachable("not reached");
1611 switch (ir
->operands
[0]->type
->vector_elements
) {
1612 case 1: result
= nir_fnoise2_1(&b
, srcs
[0]); break;
1613 case 2: result
= nir_fnoise2_2(&b
, srcs
[0]); break;
1614 case 3: result
= nir_fnoise2_3(&b
, srcs
[0]); break;
1615 case 4: result
= nir_fnoise2_4(&b
, srcs
[0]); break;
1616 default: unreachable("not reached");
1620 switch (ir
->operands
[0]->type
->vector_elements
) {
1621 case 1: result
= nir_fnoise3_1(&b
, srcs
[0]); break;
1622 case 2: result
= nir_fnoise3_2(&b
, srcs
[0]); break;
1623 case 3: result
= nir_fnoise3_3(&b
, srcs
[0]); break;
1624 case 4: result
= nir_fnoise3_4(&b
, srcs
[0]); break;
1625 default: unreachable("not reached");
1629 switch (ir
->operands
[0]->type
->vector_elements
) {
1630 case 1: result
= nir_fnoise4_1(&b
, srcs
[0]); break;
1631 case 2: result
= nir_fnoise4_2(&b
, srcs
[0]); break;
1632 case 3: result
= nir_fnoise4_3(&b
, srcs
[0]); break;
1633 case 4: result
= nir_fnoise4_4(&b
, srcs
[0]); break;
1634 default: unreachable("not reached");
1638 unreachable("not reached");
1641 case ir_unop_get_buffer_size
: {
1642 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1644 nir_intrinsic_get_buffer_size
);
1645 load
->num_components
= ir
->type
->vector_elements
;
1646 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1647 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1648 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1653 result
= type_is_float(out_type
) ? nir_fadd(&b
, srcs
[0], srcs
[1])
1654 : nir_iadd(&b
, srcs
[0], srcs
[1]);
1657 result
= type_is_float(out_type
) ? nir_fsub(&b
, srcs
[0], srcs
[1])
1658 : nir_isub(&b
, srcs
[0], srcs
[1]);
1661 result
= type_is_float(out_type
) ? nir_fmul(&b
, srcs
[0], srcs
[1])
1662 : nir_imul(&b
, srcs
[0], srcs
[1]);
1665 if (type_is_float(out_type
))
1666 result
= nir_fdiv(&b
, srcs
[0], srcs
[1]);
1667 else if (type_is_signed(out_type
))
1668 result
= nir_idiv(&b
, srcs
[0], srcs
[1]);
1670 result
= nir_udiv(&b
, srcs
[0], srcs
[1]);
1673 result
= type_is_float(out_type
) ? nir_fmod(&b
, srcs
[0], srcs
[1])
1674 : nir_umod(&b
, srcs
[0], srcs
[1]);
1677 if (type_is_float(out_type
))
1678 result
= nir_fmin(&b
, srcs
[0], srcs
[1]);
1679 else if (type_is_signed(out_type
))
1680 result
= nir_imin(&b
, srcs
[0], srcs
[1]);
1682 result
= nir_umin(&b
, srcs
[0], srcs
[1]);
1685 if (type_is_float(out_type
))
1686 result
= nir_fmax(&b
, srcs
[0], srcs
[1]);
1687 else if (type_is_signed(out_type
))
1688 result
= nir_imax(&b
, srcs
[0], srcs
[1]);
1690 result
= nir_umax(&b
, srcs
[0], srcs
[1]);
1692 case ir_binop_pow
: result
= nir_fpow(&b
, srcs
[0], srcs
[1]); break;
1693 case ir_binop_bit_and
: result
= nir_iand(&b
, srcs
[0], srcs
[1]); break;
1694 case ir_binop_bit_or
: result
= nir_ior(&b
, srcs
[0], srcs
[1]); break;
1695 case ir_binop_bit_xor
: result
= nir_ixor(&b
, srcs
[0], srcs
[1]); break;
1696 case ir_binop_logic_and
:
1697 result
= supports_ints
? nir_iand(&b
, srcs
[0], srcs
[1])
1698 : nir_fand(&b
, srcs
[0], srcs
[1]);
1700 case ir_binop_logic_or
:
1701 result
= supports_ints
? nir_ior(&b
, srcs
[0], srcs
[1])
1702 : nir_for(&b
, srcs
[0], srcs
[1]);
1704 case ir_binop_logic_xor
:
1705 result
= supports_ints
? nir_ixor(&b
, srcs
[0], srcs
[1])
1706 : nir_fxor(&b
, srcs
[0], srcs
[1]);
1708 case ir_binop_lshift
: result
= nir_ishl(&b
, srcs
[0], srcs
[1]); break;
1709 case ir_binop_rshift
:
1710 result
= (type_is_signed(out_type
)) ? nir_ishr(&b
, srcs
[0], srcs
[1])
1711 : nir_ushr(&b
, srcs
[0], srcs
[1]);
1713 case ir_binop_imul_high
:
1714 result
= (out_type
== GLSL_TYPE_INT
) ? nir_imul_high(&b
, srcs
[0], srcs
[1])
1715 : nir_umul_high(&b
, srcs
[0], srcs
[1]);
1717 case ir_binop_carry
: result
= nir_uadd_carry(&b
, srcs
[0], srcs
[1]); break;
1718 case ir_binop_borrow
: result
= nir_usub_borrow(&b
, srcs
[0], srcs
[1]); break;
1720 if (supports_ints
) {
1721 if (type_is_float(types
[0]))
1722 result
= nir_flt(&b
, srcs
[0], srcs
[1]);
1723 else if (type_is_signed(types
[0]))
1724 result
= nir_ilt(&b
, srcs
[0], srcs
[1]);
1726 result
= nir_ult(&b
, srcs
[0], srcs
[1]);
1728 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1731 case ir_binop_greater
:
1732 if (supports_ints
) {
1733 if (type_is_float(types
[0]))
1734 result
= nir_flt(&b
, srcs
[1], srcs
[0]);
1735 else if (type_is_signed(types
[0]))
1736 result
= nir_ilt(&b
, srcs
[1], srcs
[0]);
1738 result
= nir_ult(&b
, srcs
[1], srcs
[0]);
1740 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1743 case ir_binop_lequal
:
1744 if (supports_ints
) {
1745 if (type_is_float(types
[0]))
1746 result
= nir_fge(&b
, srcs
[1], srcs
[0]);
1747 else if (type_is_signed(types
[0]))
1748 result
= nir_ige(&b
, srcs
[1], srcs
[0]);
1750 result
= nir_uge(&b
, srcs
[1], srcs
[0]);
1752 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1755 case ir_binop_gequal
:
1756 if (supports_ints
) {
1757 if (type_is_float(types
[0]))
1758 result
= nir_fge(&b
, srcs
[0], srcs
[1]);
1759 else if (type_is_signed(types
[0]))
1760 result
= nir_ige(&b
, srcs
[0], srcs
[1]);
1762 result
= nir_uge(&b
, srcs
[0], srcs
[1]);
1764 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1767 case ir_binop_equal
:
1768 if (supports_ints
) {
1769 if (type_is_float(types
[0]))
1770 result
= nir_feq(&b
, srcs
[0], srcs
[1]);
1772 result
= nir_ieq(&b
, srcs
[0], srcs
[1]);
1774 result
= nir_seq(&b
, srcs
[0], srcs
[1]);
1777 case ir_binop_nequal
:
1778 if (supports_ints
) {
1779 if (type_is_float(types
[0]))
1780 result
= nir_fne(&b
, srcs
[0], srcs
[1]);
1782 result
= nir_ine(&b
, srcs
[0], srcs
[1]);
1784 result
= nir_sne(&b
, srcs
[0], srcs
[1]);
1787 case ir_binop_all_equal
:
1788 if (supports_ints
) {
1789 if (type_is_float(types
[0])) {
1790 switch (ir
->operands
[0]->type
->vector_elements
) {
1791 case 1: result
= nir_feq(&b
, srcs
[0], srcs
[1]); break;
1792 case 2: result
= nir_ball_fequal2(&b
, srcs
[0], srcs
[1]); break;
1793 case 3: result
= nir_ball_fequal3(&b
, srcs
[0], srcs
[1]); break;
1794 case 4: result
= nir_ball_fequal4(&b
, srcs
[0], srcs
[1]); break;
1796 unreachable("not reached");
1799 switch (ir
->operands
[0]->type
->vector_elements
) {
1800 case 1: result
= nir_ieq(&b
, srcs
[0], srcs
[1]); break;
1801 case 2: result
= nir_ball_iequal2(&b
, srcs
[0], srcs
[1]); break;
1802 case 3: result
= nir_ball_iequal3(&b
, srcs
[0], srcs
[1]); break;
1803 case 4: result
= nir_ball_iequal4(&b
, srcs
[0], srcs
[1]); break;
1805 unreachable("not reached");
1809 switch (ir
->operands
[0]->type
->vector_elements
) {
1810 case 1: result
= nir_seq(&b
, srcs
[0], srcs
[1]); break;
1811 case 2: result
= nir_fall_equal2(&b
, srcs
[0], srcs
[1]); break;
1812 case 3: result
= nir_fall_equal3(&b
, srcs
[0], srcs
[1]); break;
1813 case 4: result
= nir_fall_equal4(&b
, srcs
[0], srcs
[1]); break;
1815 unreachable("not reached");
1819 case ir_binop_any_nequal
:
1820 if (supports_ints
) {
1821 if (type_is_float(types
[0])) {
1822 switch (ir
->operands
[0]->type
->vector_elements
) {
1823 case 1: result
= nir_fne(&b
, srcs
[0], srcs
[1]); break;
1824 case 2: result
= nir_bany_fnequal2(&b
, srcs
[0], srcs
[1]); break;
1825 case 3: result
= nir_bany_fnequal3(&b
, srcs
[0], srcs
[1]); break;
1826 case 4: result
= nir_bany_fnequal4(&b
, srcs
[0], srcs
[1]); break;
1828 unreachable("not reached");
1831 switch (ir
->operands
[0]->type
->vector_elements
) {
1832 case 1: result
= nir_ine(&b
, srcs
[0], srcs
[1]); break;
1833 case 2: result
= nir_bany_inequal2(&b
, srcs
[0], srcs
[1]); break;
1834 case 3: result
= nir_bany_inequal3(&b
, srcs
[0], srcs
[1]); break;
1835 case 4: result
= nir_bany_inequal4(&b
, srcs
[0], srcs
[1]); break;
1837 unreachable("not reached");
1841 switch (ir
->operands
[0]->type
->vector_elements
) {
1842 case 1: result
= nir_sne(&b
, srcs
[0], srcs
[1]); break;
1843 case 2: result
= nir_fany_nequal2(&b
, srcs
[0], srcs
[1]); break;
1844 case 3: result
= nir_fany_nequal3(&b
, srcs
[0], srcs
[1]); break;
1845 case 4: result
= nir_fany_nequal4(&b
, srcs
[0], srcs
[1]); break;
1847 unreachable("not reached");
1852 switch (ir
->operands
[0]->type
->vector_elements
) {
1853 case 2: result
= nir_fdot2(&b
, srcs
[0], srcs
[1]); break;
1854 case 3: result
= nir_fdot3(&b
, srcs
[0], srcs
[1]); break;
1855 case 4: result
= nir_fdot4(&b
, srcs
[0], srcs
[1]); break;
1857 unreachable("not reached");
1861 case ir_binop_ldexp
: result
= nir_ldexp(&b
, srcs
[0], srcs
[1]); break;
1863 result
= nir_ffma(&b
, srcs
[0], srcs
[1], srcs
[2]);
1866 result
= nir_flrp(&b
, srcs
[0], srcs
[1], srcs
[2]);
1870 result
= nir_bcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1872 result
= nir_fcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1874 case ir_triop_bitfield_extract
:
1875 result
= (out_type
== GLSL_TYPE_INT
) ?
1876 nir_ibitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]) :
1877 nir_ubitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]);
1879 case ir_quadop_bitfield_insert
:
1880 result
= nir_bitfield_insert(&b
, srcs
[0], srcs
[1], srcs
[2], srcs
[3]);
1882 case ir_quadop_vector
:
1883 result
= nir_vec(&b
, srcs
, ir
->type
->vector_elements
);
1887 unreachable("not reached");
1892 nir_visitor::visit(ir_swizzle
*ir
)
1894 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1895 result
= nir_swizzle(&b
, evaluate_rvalue(ir
->val
), swizzle
,
1896 ir
->type
->vector_elements
, !supports_ints
);
1900 nir_visitor::visit(ir_texture
*ir
)
1907 num_srcs
= 1; /* coordinate */
1912 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1913 num_srcs
= 2; /* coordinate, bias/lod */
1917 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1923 if (ir
->lod_info
.lod
!= NULL
)
1924 num_srcs
= 2; /* coordinate, lod */
1926 num_srcs
= 1; /* coordinate */
1930 op
= nir_texop_txf_ms
;
1931 num_srcs
= 2; /* coordinate, sample_index */
1936 if (ir
->lod_info
.lod
!= NULL
)
1937 num_srcs
= 1; /* lod */
1944 num_srcs
= 1; /* coordinate */
1949 num_srcs
= 1; /* coordinate */
1952 case ir_query_levels
:
1953 op
= nir_texop_query_levels
;
1957 case ir_texture_samples
:
1958 op
= nir_texop_texture_samples
;
1962 case ir_samples_identical
:
1963 op
= nir_texop_samples_identical
;
1964 num_srcs
= 1; /* coordinate */
1968 unreachable("not reached");
1971 if (ir
->projector
!= NULL
)
1973 if (ir
->shadow_comparator
!= NULL
)
1975 if (ir
->offset
!= NULL
)
1978 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1981 instr
->sampler_dim
=
1982 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1983 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1984 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1985 if (instr
->is_shadow
)
1986 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1987 switch (ir
->type
->base_type
) {
1988 case GLSL_TYPE_FLOAT
:
1989 instr
->dest_type
= nir_type_float
;
1992 instr
->dest_type
= nir_type_int
;
1994 case GLSL_TYPE_BOOL
:
1995 case GLSL_TYPE_UINT
:
1996 instr
->dest_type
= nir_type_uint
;
1999 unreachable("not reached");
2002 instr
->texture
= evaluate_deref(&instr
->instr
, ir
->sampler
);
2004 unsigned src_number
= 0;
2006 if (ir
->coordinate
!= NULL
) {
2007 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
2008 instr
->src
[src_number
].src
=
2009 nir_src_for_ssa(evaluate_rvalue(ir
->coordinate
));
2010 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
2014 if (ir
->projector
!= NULL
) {
2015 instr
->src
[src_number
].src
=
2016 nir_src_for_ssa(evaluate_rvalue(ir
->projector
));
2017 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
2021 if (ir
->shadow_comparator
!= NULL
) {
2022 instr
->src
[src_number
].src
=
2023 nir_src_for_ssa(evaluate_rvalue(ir
->shadow_comparator
));
2024 instr
->src
[src_number
].src_type
= nir_tex_src_comparator
;
2028 if (ir
->offset
!= NULL
) {
2029 /* we don't support multiple offsets yet */
2030 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
2032 instr
->src
[src_number
].src
=
2033 nir_src_for_ssa(evaluate_rvalue(ir
->offset
));
2034 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
2040 instr
->src
[src_number
].src
=
2041 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.bias
));
2042 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
2049 if (ir
->lod_info
.lod
!= NULL
) {
2050 instr
->src
[src_number
].src
=
2051 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.lod
));
2052 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
2058 instr
->src
[src_number
].src
=
2059 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdx
));
2060 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
2062 instr
->src
[src_number
].src
=
2063 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdy
));
2064 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
2069 instr
->src
[src_number
].src
=
2070 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.sample_index
));
2071 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
2076 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
2083 assert(src_number
== num_srcs
);
2085 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
2086 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
), bit_size
);
2090 nir_visitor::visit(ir_constant
*ir
)
2093 * We don't know if this variable is an array or struct that gets
2094 * dereferenced, so do the safe thing an make it a variable with a
2095 * constant initializer and return a dereference.
2099 nir_local_variable_create(this->impl
, ir
->type
, "const_temp");
2100 var
->data
.read_only
= true;
2101 var
->constant_initializer
= constant_copy(ir
, var
);
2103 this->deref_head
= nir_deref_var_create(this->shader
, var
);
2104 this->deref_tail
= &this->deref_head
->deref
;
2108 nir_visitor::visit(ir_dereference_variable
*ir
)
2110 struct hash_entry
*entry
=
2111 _mesa_hash_table_search(this->var_table
, ir
->var
);
2113 nir_variable
*var
= (nir_variable
*) entry
->data
;
2115 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
2116 this->deref_head
= deref
;
2117 this->deref_tail
= &deref
->deref
;
2121 nir_visitor::visit(ir_dereference_record
*ir
)
2123 ir
->record
->accept(this);
2125 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
2126 assert(field_index
>= 0);
2128 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
2129 deref
->deref
.type
= ir
->type
;
2130 this->deref_tail
->child
= &deref
->deref
;
2131 this->deref_tail
= &deref
->deref
;
2135 nir_visitor::visit(ir_dereference_array
*ir
)
2137 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
2138 deref
->deref
.type
= ir
->type
;
2140 ir_constant
*const_index
= ir
->array_index
->as_constant();
2141 if (const_index
!= NULL
) {
2142 deref
->deref_array_type
= nir_deref_array_type_direct
;
2143 deref
->base_offset
= const_index
->value
.u
[0];
2145 deref
->deref_array_type
= nir_deref_array_type_indirect
;
2147 nir_src_for_ssa(evaluate_rvalue(ir
->array_index
));
2150 ir
->array
->accept(this);
2152 this->deref_tail
->child
= &deref
->deref
;
2153 ralloc_steal(this->deref_tail
, deref
);
2154 this->deref_tail
= &deref
->deref
;
2158 nir_visitor::visit(ir_barrier
*)
2160 nir_intrinsic_instr
*instr
=
2161 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
2162 nir_builder_instr_insert(&b
, &instr
->instr
);