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 /* TODO: In future we should switch to using the NIR lowering pass but for
312 * now just ignore these variables as GLSL IR should have lowered them.
313 * Anything remaining are just dead vars that weren't cleaned up.
315 if (ir
->data
.mode
== ir_var_shader_shared
)
318 nir_variable
*var
= ralloc(shader
, nir_variable
);
319 var
->type
= ir
->type
;
320 var
->name
= ralloc_strdup(var
, ir
->name
);
322 var
->data
.read_only
= ir
->data
.read_only
;
323 var
->data
.centroid
= ir
->data
.centroid
;
324 var
->data
.sample
= ir
->data
.sample
;
325 var
->data
.patch
= ir
->data
.patch
;
326 var
->data
.invariant
= ir
->data
.invariant
;
327 var
->data
.location
= ir
->data
.location
;
328 var
->data
.compact
= false;
330 switch(ir
->data
.mode
) {
332 case ir_var_temporary
:
334 var
->data
.mode
= nir_var_global
;
336 var
->data
.mode
= nir_var_local
;
339 case ir_var_function_in
:
340 case ir_var_function_out
:
341 case ir_var_function_inout
:
342 case ir_var_const_in
:
343 var
->data
.mode
= nir_var_local
;
346 case ir_var_shader_in
:
347 if (shader
->stage
== MESA_SHADER_FRAGMENT
&&
348 ir
->data
.location
== VARYING_SLOT_FACE
) {
349 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
350 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
351 var
->data
.mode
= nir_var_system_value
;
352 } else if (shader
->stage
== MESA_SHADER_GEOMETRY
&&
353 ir
->data
.location
== VARYING_SLOT_PRIMITIVE_ID
) {
354 /* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
355 var
->data
.location
= SYSTEM_VALUE_PRIMITIVE_ID
;
356 var
->data
.mode
= nir_var_system_value
;
358 var
->data
.mode
= nir_var_shader_in
;
360 if (shader
->stage
== MESA_SHADER_TESS_EVAL
&&
361 (ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
||
362 ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
)) {
363 var
->data
.compact
= ir
->type
->without_array()->is_scalar();
367 /* Mark all the locations that require two slots */
368 if (glsl_type_is_dual_slot(glsl_without_array(var
->type
))) {
369 for (uint i
= 0; i
< glsl_count_attribute_slots(var
->type
, true); i
++) {
370 uint64_t bitfield
= BITFIELD64_BIT(var
->data
.location
+ i
);
371 shader
->info
.double_inputs_read
|= bitfield
;
376 case ir_var_shader_out
:
377 var
->data
.mode
= nir_var_shader_out
;
378 if (shader
->stage
== MESA_SHADER_TESS_CTRL
&&
379 (ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
||
380 ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
)) {
381 var
->data
.compact
= ir
->type
->without_array()->is_scalar();
386 var
->data
.mode
= nir_var_uniform
;
389 case ir_var_shader_storage
:
390 var
->data
.mode
= nir_var_shader_storage
;
393 case ir_var_system_value
:
394 var
->data
.mode
= nir_var_system_value
;
398 unreachable("not reached");
401 var
->data
.interpolation
= ir
->data
.interpolation
;
402 var
->data
.origin_upper_left
= ir
->data
.origin_upper_left
;
403 var
->data
.pixel_center_integer
= ir
->data
.pixel_center_integer
;
404 var
->data
.location_frac
= ir
->data
.location_frac
;
406 switch (ir
->data
.depth_layout
) {
407 case ir_depth_layout_none
:
408 var
->data
.depth_layout
= nir_depth_layout_none
;
410 case ir_depth_layout_any
:
411 var
->data
.depth_layout
= nir_depth_layout_any
;
413 case ir_depth_layout_greater
:
414 var
->data
.depth_layout
= nir_depth_layout_greater
;
416 case ir_depth_layout_less
:
417 var
->data
.depth_layout
= nir_depth_layout_less
;
419 case ir_depth_layout_unchanged
:
420 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
423 unreachable("not reached");
426 var
->data
.index
= ir
->data
.index
;
427 var
->data
.binding
= ir
->data
.binding
;
428 var
->data
.offset
= ir
->data
.offset
;
429 var
->data
.image
.read_only
= ir
->data
.memory_read_only
;
430 var
->data
.image
.write_only
= ir
->data
.memory_write_only
;
431 var
->data
.image
.coherent
= ir
->data
.memory_coherent
;
432 var
->data
.image
._volatile
= ir
->data
.memory_volatile
;
433 var
->data
.image
.restrict_flag
= ir
->data
.memory_restrict
;
434 var
->data
.image
.format
= ir
->data
.image_format
;
435 var
->data
.fb_fetch_output
= ir
->data
.fb_fetch_output
;
437 var
->num_state_slots
= ir
->get_num_state_slots();
438 if (var
->num_state_slots
> 0) {
439 var
->state_slots
= ralloc_array(var
, nir_state_slot
,
440 var
->num_state_slots
);
442 ir_state_slot
*state_slots
= ir
->get_state_slots();
443 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
444 for (unsigned j
= 0; j
< 5; j
++)
445 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
446 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
449 var
->state_slots
= NULL
;
452 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
454 var
->interface_type
= ir
->get_interface_type();
456 if (var
->data
.mode
== nir_var_local
)
457 nir_function_impl_add_variable(impl
, var
);
459 nir_shader_add_variable(shader
, var
);
461 _mesa_hash_table_insert(var_table
, ir
, var
);
466 nir_function_visitor::visit_enter(ir_function
*ir
)
468 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
469 visitor
->create_function(sig
);
471 return visit_continue_with_parent
;
475 nir_visitor::create_function(ir_function_signature
*ir
)
477 if (ir
->is_intrinsic())
480 nir_function
*func
= nir_function_create(shader
, ir
->function_name());
482 assert(ir
->parameters
.is_empty());
483 assert(ir
->return_type
== glsl_type::void_type
);
485 _mesa_hash_table_insert(this->overload_table
, ir
, func
);
489 nir_visitor::visit(ir_function
*ir
)
491 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
496 nir_visitor::visit(ir_function_signature
*ir
)
498 if (ir
->is_intrinsic())
501 struct hash_entry
*entry
=
502 _mesa_hash_table_search(this->overload_table
, ir
);
505 nir_function
*func
= (nir_function
*) entry
->data
;
507 if (ir
->is_defined
) {
508 nir_function_impl
*impl
= nir_function_impl_create(func
);
511 assert(strcmp(func
->name
, "main") == 0);
512 assert(ir
->parameters
.is_empty());
513 assert(func
->return_type
== glsl_type::void_type
);
515 this->is_global
= false;
517 nir_builder_init(&b
, impl
);
518 b
.cursor
= nir_after_cf_list(&impl
->body
);
519 visit_exec_list(&ir
->body
, this);
521 this->is_global
= true;
528 nir_visitor::visit(ir_loop
*ir
)
531 visit_exec_list(&ir
->body_instructions
, this);
532 nir_pop_loop(&b
, NULL
);
536 nir_visitor::visit(ir_if
*ir
)
538 nir_push_if(&b
, evaluate_rvalue(ir
->condition
));
539 visit_exec_list(&ir
->then_instructions
, this);
540 nir_push_else(&b
, NULL
);
541 visit_exec_list(&ir
->else_instructions
, this);
542 nir_pop_if(&b
, NULL
);
546 nir_visitor::visit(ir_discard
*ir
)
549 * discards aren't treated as control flow, because before we lower them
550 * they can appear anywhere in the shader and the stuff after them may still
551 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
552 * discards will be immediately followed by a return.
555 nir_intrinsic_instr
*discard
;
557 discard
= nir_intrinsic_instr_create(this->shader
,
558 nir_intrinsic_discard_if
);
560 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
562 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
565 nir_builder_instr_insert(&b
, &discard
->instr
);
569 nir_visitor::visit(ir_emit_vertex
*ir
)
571 nir_intrinsic_instr
*instr
=
572 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
573 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
574 nir_builder_instr_insert(&b
, &instr
->instr
);
578 nir_visitor::visit(ir_end_primitive
*ir
)
580 nir_intrinsic_instr
*instr
=
581 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
582 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
583 nir_builder_instr_insert(&b
, &instr
->instr
);
587 nir_visitor::visit(ir_loop_jump
*ir
)
591 case ir_loop_jump::jump_break
:
592 type
= nir_jump_break
;
594 case ir_loop_jump::jump_continue
:
595 type
= nir_jump_continue
;
598 unreachable("not reached");
601 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
602 nir_builder_instr_insert(&b
, &instr
->instr
);
606 nir_visitor::visit(ir_return
*ir
)
608 if (ir
->value
!= NULL
) {
609 nir_intrinsic_instr
*copy
=
610 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
612 copy
->variables
[0] = nir_deref_var_create(copy
, this->impl
->return_var
);
613 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->value
);
616 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
617 nir_builder_instr_insert(&b
, &instr
->instr
);
621 nir_visitor::visit(ir_call
*ir
)
623 if (ir
->callee
->is_intrinsic()) {
626 switch (ir
->callee
->intrinsic_id
) {
627 case ir_intrinsic_atomic_counter_read
:
628 op
= nir_intrinsic_atomic_counter_read_var
;
630 case ir_intrinsic_atomic_counter_increment
:
631 op
= nir_intrinsic_atomic_counter_inc_var
;
633 case ir_intrinsic_atomic_counter_predecrement
:
634 op
= nir_intrinsic_atomic_counter_dec_var
;
636 case ir_intrinsic_atomic_counter_add
:
637 op
= nir_intrinsic_atomic_counter_add_var
;
639 case ir_intrinsic_atomic_counter_and
:
640 op
= nir_intrinsic_atomic_counter_and_var
;
642 case ir_intrinsic_atomic_counter_or
:
643 op
= nir_intrinsic_atomic_counter_or_var
;
645 case ir_intrinsic_atomic_counter_xor
:
646 op
= nir_intrinsic_atomic_counter_xor_var
;
648 case ir_intrinsic_atomic_counter_min
:
649 op
= nir_intrinsic_atomic_counter_min_var
;
651 case ir_intrinsic_atomic_counter_max
:
652 op
= nir_intrinsic_atomic_counter_max_var
;
654 case ir_intrinsic_atomic_counter_exchange
:
655 op
= nir_intrinsic_atomic_counter_exchange_var
;
657 case ir_intrinsic_atomic_counter_comp_swap
:
658 op
= nir_intrinsic_atomic_counter_comp_swap_var
;
660 case ir_intrinsic_image_load
:
661 op
= nir_intrinsic_image_load
;
663 case ir_intrinsic_image_store
:
664 op
= nir_intrinsic_image_store
;
666 case ir_intrinsic_image_atomic_add
:
667 op
= nir_intrinsic_image_atomic_add
;
669 case ir_intrinsic_image_atomic_min
:
670 op
= nir_intrinsic_image_atomic_min
;
672 case ir_intrinsic_image_atomic_max
:
673 op
= nir_intrinsic_image_atomic_max
;
675 case ir_intrinsic_image_atomic_and
:
676 op
= nir_intrinsic_image_atomic_and
;
678 case ir_intrinsic_image_atomic_or
:
679 op
= nir_intrinsic_image_atomic_or
;
681 case ir_intrinsic_image_atomic_xor
:
682 op
= nir_intrinsic_image_atomic_xor
;
684 case ir_intrinsic_image_atomic_exchange
:
685 op
= nir_intrinsic_image_atomic_exchange
;
687 case ir_intrinsic_image_atomic_comp_swap
:
688 op
= nir_intrinsic_image_atomic_comp_swap
;
690 case ir_intrinsic_memory_barrier
:
691 op
= nir_intrinsic_memory_barrier
;
693 case ir_intrinsic_image_size
:
694 op
= nir_intrinsic_image_size
;
696 case ir_intrinsic_image_samples
:
697 op
= nir_intrinsic_image_samples
;
699 case ir_intrinsic_ssbo_store
:
700 op
= nir_intrinsic_store_ssbo
;
702 case ir_intrinsic_ssbo_load
:
703 op
= nir_intrinsic_load_ssbo
;
705 case ir_intrinsic_ssbo_atomic_add
:
706 op
= nir_intrinsic_ssbo_atomic_add
;
708 case ir_intrinsic_ssbo_atomic_and
:
709 op
= nir_intrinsic_ssbo_atomic_and
;
711 case ir_intrinsic_ssbo_atomic_or
:
712 op
= nir_intrinsic_ssbo_atomic_or
;
714 case ir_intrinsic_ssbo_atomic_xor
:
715 op
= nir_intrinsic_ssbo_atomic_xor
;
717 case ir_intrinsic_ssbo_atomic_min
:
718 assert(ir
->return_deref
);
719 if (ir
->return_deref
->type
== glsl_type::int_type
)
720 op
= nir_intrinsic_ssbo_atomic_imin
;
721 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
722 op
= nir_intrinsic_ssbo_atomic_umin
;
724 unreachable("Invalid type");
726 case ir_intrinsic_ssbo_atomic_max
:
727 assert(ir
->return_deref
);
728 if (ir
->return_deref
->type
== glsl_type::int_type
)
729 op
= nir_intrinsic_ssbo_atomic_imax
;
730 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
731 op
= nir_intrinsic_ssbo_atomic_umax
;
733 unreachable("Invalid type");
735 case ir_intrinsic_ssbo_atomic_exchange
:
736 op
= nir_intrinsic_ssbo_atomic_exchange
;
738 case ir_intrinsic_ssbo_atomic_comp_swap
:
739 op
= nir_intrinsic_ssbo_atomic_comp_swap
;
741 case ir_intrinsic_shader_clock
:
742 op
= nir_intrinsic_shader_clock
;
744 case ir_intrinsic_group_memory_barrier
:
745 op
= nir_intrinsic_group_memory_barrier
;
747 case ir_intrinsic_memory_barrier_atomic_counter
:
748 op
= nir_intrinsic_memory_barrier_atomic_counter
;
750 case ir_intrinsic_memory_barrier_buffer
:
751 op
= nir_intrinsic_memory_barrier_buffer
;
753 case ir_intrinsic_memory_barrier_image
:
754 op
= nir_intrinsic_memory_barrier_image
;
756 case ir_intrinsic_memory_barrier_shared
:
757 op
= nir_intrinsic_memory_barrier_shared
;
759 case ir_intrinsic_shared_load
:
760 op
= nir_intrinsic_load_shared
;
762 case ir_intrinsic_shared_store
:
763 op
= nir_intrinsic_store_shared
;
765 case ir_intrinsic_shared_atomic_add
:
766 op
= nir_intrinsic_shared_atomic_add
;
768 case ir_intrinsic_shared_atomic_and
:
769 op
= nir_intrinsic_shared_atomic_and
;
771 case ir_intrinsic_shared_atomic_or
:
772 op
= nir_intrinsic_shared_atomic_or
;
774 case ir_intrinsic_shared_atomic_xor
:
775 op
= nir_intrinsic_shared_atomic_xor
;
777 case ir_intrinsic_shared_atomic_min
:
778 assert(ir
->return_deref
);
779 if (ir
->return_deref
->type
== glsl_type::int_type
)
780 op
= nir_intrinsic_shared_atomic_imin
;
781 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
782 op
= nir_intrinsic_shared_atomic_umin
;
784 unreachable("Invalid type");
786 case ir_intrinsic_shared_atomic_max
:
787 assert(ir
->return_deref
);
788 if (ir
->return_deref
->type
== glsl_type::int_type
)
789 op
= nir_intrinsic_shared_atomic_imax
;
790 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
791 op
= nir_intrinsic_shared_atomic_umax
;
793 unreachable("Invalid type");
795 case ir_intrinsic_shared_atomic_exchange
:
796 op
= nir_intrinsic_shared_atomic_exchange
;
798 case ir_intrinsic_shared_atomic_comp_swap
:
799 op
= nir_intrinsic_shared_atomic_comp_swap
;
802 unreachable("not reached");
805 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
806 nir_dest
*dest
= &instr
->dest
;
809 case nir_intrinsic_atomic_counter_read_var
:
810 case nir_intrinsic_atomic_counter_inc_var
:
811 case nir_intrinsic_atomic_counter_dec_var
:
812 case nir_intrinsic_atomic_counter_add_var
:
813 case nir_intrinsic_atomic_counter_min_var
:
814 case nir_intrinsic_atomic_counter_max_var
:
815 case nir_intrinsic_atomic_counter_and_var
:
816 case nir_intrinsic_atomic_counter_or_var
:
817 case nir_intrinsic_atomic_counter_xor_var
:
818 case nir_intrinsic_atomic_counter_exchange_var
:
819 case nir_intrinsic_atomic_counter_comp_swap_var
: {
820 /* Set the counter variable dereference. */
821 exec_node
*param
= ir
->actual_parameters
.get_head();
822 ir_dereference
*counter
= (ir_dereference
*)param
;
824 instr
->variables
[0] = evaluate_deref(&instr
->instr
, counter
);
825 param
= param
->get_next();
827 /* Set the intrinsic destination. */
828 if (ir
->return_deref
) {
829 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
832 /* Set the intrinsic parameters. */
833 if (!param
->is_tail_sentinel()) {
835 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
836 param
= param
->get_next();
839 if (!param
->is_tail_sentinel()) {
841 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
842 param
= param
->get_next();
845 nir_builder_instr_insert(&b
, &instr
->instr
);
848 case nir_intrinsic_image_load
:
849 case nir_intrinsic_image_store
:
850 case nir_intrinsic_image_atomic_add
:
851 case nir_intrinsic_image_atomic_min
:
852 case nir_intrinsic_image_atomic_max
:
853 case nir_intrinsic_image_atomic_and
:
854 case nir_intrinsic_image_atomic_or
:
855 case nir_intrinsic_image_atomic_xor
:
856 case nir_intrinsic_image_atomic_exchange
:
857 case nir_intrinsic_image_atomic_comp_swap
:
858 case nir_intrinsic_image_samples
:
859 case nir_intrinsic_image_size
: {
860 nir_ssa_undef_instr
*instr_undef
=
861 nir_ssa_undef_instr_create(shader
, 1, 32);
862 nir_builder_instr_insert(&b
, &instr_undef
->instr
);
864 /* Set the image variable dereference. */
865 exec_node
*param
= ir
->actual_parameters
.get_head();
866 ir_dereference
*image
= (ir_dereference
*)param
;
867 const glsl_type
*type
=
868 image
->variable_referenced()->type
->without_array();
870 instr
->variables
[0] = evaluate_deref(&instr
->instr
, image
);
871 param
= param
->get_next();
873 /* Set the intrinsic destination. */
874 if (ir
->return_deref
) {
875 unsigned num_components
= ir
->return_deref
->type
->vector_elements
;
876 if (instr
->intrinsic
== nir_intrinsic_image_size
)
877 instr
->num_components
= num_components
;
878 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
879 num_components
, 32, NULL
);
882 if (op
== nir_intrinsic_image_size
||
883 op
== nir_intrinsic_image_samples
) {
884 nir_builder_instr_insert(&b
, &instr
->instr
);
888 /* Set the address argument, extending the coordinate vector to four
891 nir_ssa_def
*src_addr
=
892 evaluate_rvalue((ir_dereference
*)param
);
893 nir_ssa_def
*srcs
[4];
895 for (int i
= 0; i
< 4; i
++) {
896 if (i
< type
->coordinate_components())
897 srcs
[i
] = nir_channel(&b
, src_addr
, i
);
899 srcs
[i
] = &instr_undef
->def
;
902 instr
->src
[0] = nir_src_for_ssa(nir_vec(&b
, srcs
, 4));
903 param
= param
->get_next();
905 /* Set the sample argument, which is undefined for single-sample
908 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
910 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
911 param
= param
->get_next();
913 instr
->src
[1] = nir_src_for_ssa(&instr_undef
->def
);
916 /* Set the intrinsic parameters. */
917 if (!param
->is_tail_sentinel()) {
919 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
920 param
= param
->get_next();
923 if (!param
->is_tail_sentinel()) {
925 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
926 param
= param
->get_next();
928 nir_builder_instr_insert(&b
, &instr
->instr
);
931 case nir_intrinsic_memory_barrier
:
932 case nir_intrinsic_group_memory_barrier
:
933 case nir_intrinsic_memory_barrier_atomic_counter
:
934 case nir_intrinsic_memory_barrier_buffer
:
935 case nir_intrinsic_memory_barrier_image
:
936 case nir_intrinsic_memory_barrier_shared
:
937 nir_builder_instr_insert(&b
, &instr
->instr
);
939 case nir_intrinsic_shader_clock
:
940 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 2, 32, NULL
);
941 instr
->num_components
= 2;
942 nir_builder_instr_insert(&b
, &instr
->instr
);
944 case nir_intrinsic_store_ssbo
: {
945 exec_node
*param
= ir
->actual_parameters
.get_head();
946 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
948 param
= param
->get_next();
949 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
951 param
= param
->get_next();
952 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
954 param
= param
->get_next();
955 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
958 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
959 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(block
));
960 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(offset
));
961 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
962 instr
->num_components
= val
->type
->vector_elements
;
964 nir_builder_instr_insert(&b
, &instr
->instr
);
967 case nir_intrinsic_load_ssbo
: {
968 exec_node
*param
= ir
->actual_parameters
.get_head();
969 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
971 param
= param
->get_next();
972 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
974 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(block
));
975 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
977 const glsl_type
*type
= ir
->return_deref
->var
->type
;
978 instr
->num_components
= type
->vector_elements
;
980 /* Setup destination register */
981 unsigned bit_size
= glsl_get_bit_size(type
);
982 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
983 type
->vector_elements
, bit_size
, NULL
);
985 /* Insert the created nir instruction now since in the case of boolean
986 * result we will need to emit another instruction after it
988 nir_builder_instr_insert(&b
, &instr
->instr
);
991 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
992 * consider a true boolean to be ~0. Fix this up with a != 0
995 if (type
->is_boolean()) {
996 nir_alu_instr
*load_ssbo_compare
=
997 nir_alu_instr_create(shader
, nir_op_ine
);
998 load_ssbo_compare
->src
[0].src
.is_ssa
= true;
999 load_ssbo_compare
->src
[0].src
.ssa
= &instr
->dest
.ssa
;
1000 load_ssbo_compare
->src
[1].src
=
1001 nir_src_for_ssa(nir_imm_int(&b
, 0));
1002 for (unsigned i
= 0; i
< type
->vector_elements
; i
++)
1003 load_ssbo_compare
->src
[1].swizzle
[i
] = 0;
1004 nir_ssa_dest_init(&load_ssbo_compare
->instr
,
1005 &load_ssbo_compare
->dest
.dest
,
1006 type
->vector_elements
, bit_size
, NULL
);
1007 load_ssbo_compare
->dest
.write_mask
= (1 << type
->vector_elements
) - 1;
1008 nir_builder_instr_insert(&b
, &load_ssbo_compare
->instr
);
1009 dest
= &load_ssbo_compare
->dest
.dest
;
1013 case nir_intrinsic_ssbo_atomic_add
:
1014 case nir_intrinsic_ssbo_atomic_imin
:
1015 case nir_intrinsic_ssbo_atomic_umin
:
1016 case nir_intrinsic_ssbo_atomic_imax
:
1017 case nir_intrinsic_ssbo_atomic_umax
:
1018 case nir_intrinsic_ssbo_atomic_and
:
1019 case nir_intrinsic_ssbo_atomic_or
:
1020 case nir_intrinsic_ssbo_atomic_xor
:
1021 case nir_intrinsic_ssbo_atomic_exchange
:
1022 case nir_intrinsic_ssbo_atomic_comp_swap
: {
1023 int param_count
= ir
->actual_parameters
.length();
1024 assert(param_count
== 3 || param_count
== 4);
1027 exec_node
*param
= ir
->actual_parameters
.get_head();
1028 ir_instruction
*inst
= (ir_instruction
*) param
;
1029 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1032 param
= param
->get_next();
1033 inst
= (ir_instruction
*) param
;
1034 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1036 /* data1 parameter (this is always present) */
1037 param
= param
->get_next();
1038 inst
= (ir_instruction
*) param
;
1039 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1041 /* data2 parameter (only with atomic_comp_swap) */
1042 if (param_count
== 4) {
1043 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
);
1044 param
= param
->get_next();
1045 inst
= (ir_instruction
*) param
;
1046 instr
->src
[3] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1050 assert(ir
->return_deref
);
1051 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1052 ir
->return_deref
->type
->vector_elements
, 32, NULL
);
1053 nir_builder_instr_insert(&b
, &instr
->instr
);
1056 case nir_intrinsic_load_shared
: {
1057 exec_node
*param
= ir
->actual_parameters
.get_head();
1058 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1060 nir_intrinsic_set_base(instr
, 0);
1061 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(offset
));
1063 const glsl_type
*type
= ir
->return_deref
->var
->type
;
1064 instr
->num_components
= type
->vector_elements
;
1066 /* Setup destination register */
1067 unsigned bit_size
= glsl_get_bit_size(type
);
1068 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1069 type
->vector_elements
, bit_size
, NULL
);
1071 nir_builder_instr_insert(&b
, &instr
->instr
);
1074 case nir_intrinsic_store_shared
: {
1075 exec_node
*param
= ir
->actual_parameters
.get_head();
1076 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1078 param
= param
->get_next();
1079 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
1081 param
= param
->get_next();
1082 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
1085 nir_intrinsic_set_base(instr
, 0);
1086 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
1088 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
1090 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(val
));
1091 instr
->num_components
= val
->type
->vector_elements
;
1093 nir_builder_instr_insert(&b
, &instr
->instr
);
1096 case nir_intrinsic_shared_atomic_add
:
1097 case nir_intrinsic_shared_atomic_imin
:
1098 case nir_intrinsic_shared_atomic_umin
:
1099 case nir_intrinsic_shared_atomic_imax
:
1100 case nir_intrinsic_shared_atomic_umax
:
1101 case nir_intrinsic_shared_atomic_and
:
1102 case nir_intrinsic_shared_atomic_or
:
1103 case nir_intrinsic_shared_atomic_xor
:
1104 case nir_intrinsic_shared_atomic_exchange
:
1105 case nir_intrinsic_shared_atomic_comp_swap
: {
1106 int param_count
= ir
->actual_parameters
.length();
1107 assert(param_count
== 2 || param_count
== 3);
1110 exec_node
*param
= ir
->actual_parameters
.get_head();
1111 ir_instruction
*inst
= (ir_instruction
*) param
;
1112 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1114 /* data1 parameter (this is always present) */
1115 param
= param
->get_next();
1116 inst
= (ir_instruction
*) param
;
1117 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1119 /* data2 parameter (only with atomic_comp_swap) */
1120 if (param_count
== 3) {
1121 assert(op
== nir_intrinsic_shared_atomic_comp_swap
);
1122 param
= param
->get_next();
1123 inst
= (ir_instruction
*) param
;
1125 nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1129 assert(ir
->return_deref
);
1130 unsigned bit_size
= glsl_get_bit_size(ir
->return_deref
->type
);
1131 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1132 ir
->return_deref
->type
->vector_elements
,
1134 nir_builder_instr_insert(&b
, &instr
->instr
);
1138 unreachable("not reached");
1141 if (ir
->return_deref
) {
1142 nir_intrinsic_instr
*store_instr
=
1143 nir_intrinsic_instr_create(shader
, nir_intrinsic_store_var
);
1144 store_instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
1145 nir_intrinsic_set_write_mask(store_instr
,
1146 (1 << store_instr
->num_components
) - 1);
1148 store_instr
->variables
[0] =
1149 evaluate_deref(&store_instr
->instr
, ir
->return_deref
);
1150 store_instr
->src
[0] = nir_src_for_ssa(&dest
->ssa
);
1152 nir_builder_instr_insert(&b
, &store_instr
->instr
);
1158 struct hash_entry
*entry
=
1159 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
1161 nir_function
*callee
= (nir_function
*) entry
->data
;
1163 nir_call_instr
*instr
= nir_call_instr_create(this->shader
, callee
);
1166 foreach_in_list(ir_dereference
, param
, &ir
->actual_parameters
) {
1167 instr
->params
[i
] = evaluate_deref(&instr
->instr
, param
);
1171 instr
->return_deref
= evaluate_deref(&instr
->instr
, ir
->return_deref
);
1172 nir_builder_instr_insert(&b
, &instr
->instr
);
1176 nir_visitor::visit(ir_assignment
*ir
)
1178 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
1180 b
.exact
= ir
->lhs
->variable_referenced()->data
.invariant
||
1181 ir
->lhs
->variable_referenced()->data
.precise
;
1183 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
1184 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
1185 /* We're doing a plain-as-can-be copy, so emit a copy_var */
1186 nir_intrinsic_instr
*copy
=
1187 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_copy_var
);
1189 copy
->variables
[0] = evaluate_deref(©
->instr
, ir
->lhs
);
1190 copy
->variables
[1] = evaluate_deref(©
->instr
, ir
->rhs
);
1192 if (ir
->condition
) {
1193 nir_push_if(&b
, evaluate_rvalue(ir
->condition
));
1194 nir_builder_instr_insert(&b
, ©
->instr
);
1195 nir_pop_if(&b
, NULL
);
1197 nir_builder_instr_insert(&b
, ©
->instr
);
1202 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
1204 ir
->lhs
->accept(this);
1205 nir_deref_var
*lhs_deref
= this->deref_head
;
1206 nir_ssa_def
*src
= evaluate_rvalue(ir
->rhs
);
1208 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
1209 /* GLSL IR will give us the input to the write-masked assignment in a
1210 * single packed vector. So, for example, if the writemask is xzw, then
1211 * we have to swizzle x -> x, y -> z, and z -> w and get the y component
1215 unsigned component
= 0;
1216 for (unsigned i
= 0; i
< 4; i
++) {
1217 swiz
[i
] = ir
->write_mask
& (1 << i
) ? component
++ : 0;
1219 src
= nir_swizzle(&b
, src
, swiz
, num_components
, !supports_ints
);
1222 nir_intrinsic_instr
*store
=
1223 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_store_var
);
1224 store
->num_components
= ir
->lhs
->type
->vector_elements
;
1225 nir_intrinsic_set_write_mask(store
, ir
->write_mask
);
1226 store
->variables
[0] = nir_deref_var_clone(lhs_deref
, store
);
1227 store
->src
[0] = nir_src_for_ssa(src
);
1229 if (ir
->condition
) {
1230 nir_push_if(&b
, evaluate_rvalue(ir
->condition
));
1231 nir_builder_instr_insert(&b
, &store
->instr
);
1232 nir_pop_if(&b
, NULL
);
1234 nir_builder_instr_insert(&b
, &store
->instr
);
1239 * Given an instruction, returns a pointer to its destination or NULL if there
1240 * is no destination.
1242 * Note that this only handles instructions we generate at this level.
1245 get_instr_dest(nir_instr
*instr
)
1247 nir_alu_instr
*alu_instr
;
1248 nir_intrinsic_instr
*intrinsic_instr
;
1249 nir_tex_instr
*tex_instr
;
1251 switch (instr
->type
) {
1252 case nir_instr_type_alu
:
1253 alu_instr
= nir_instr_as_alu(instr
);
1254 return &alu_instr
->dest
.dest
;
1256 case nir_instr_type_intrinsic
:
1257 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1258 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1259 return &intrinsic_instr
->dest
;
1263 case nir_instr_type_tex
:
1264 tex_instr
= nir_instr_as_tex(instr
);
1265 return &tex_instr
->dest
;
1268 unreachable("not reached");
1275 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
,
1278 nir_dest
*dest
= get_instr_dest(instr
);
1281 nir_ssa_dest_init(instr
, dest
, num_components
, bit_size
, NULL
);
1283 nir_builder_instr_insert(&b
, instr
);
1286 assert(dest
->is_ssa
);
1287 this->result
= &dest
->ssa
;
1292 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1295 if (ir
->as_dereference() || ir
->as_constant()) {
1297 * A dereference is being used on the right hand side, which means we
1298 * must emit a variable load.
1301 nir_intrinsic_instr
*load_instr
=
1302 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_var
);
1303 load_instr
->num_components
= ir
->type
->vector_elements
;
1304 load_instr
->variables
[0] = this->deref_head
;
1305 ralloc_steal(load_instr
, load_instr
->variables
[0]);
1306 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1307 add_instr(&load_instr
->instr
, ir
->type
->vector_elements
, bit_size
);
1310 return this->result
;
1314 type_is_float(glsl_base_type type
)
1316 return type
== GLSL_TYPE_FLOAT
|| type
== GLSL_TYPE_DOUBLE
;
1320 type_is_signed(glsl_base_type type
)
1322 return type
== GLSL_TYPE_INT
|| type
== GLSL_TYPE_INT64
;
1326 nir_visitor::visit(ir_expression
*ir
)
1328 /* Some special cases */
1329 switch (ir
->operation
) {
1330 case ir_binop_ubo_load
: {
1331 nir_intrinsic_instr
*load
=
1332 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_ubo
);
1333 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1334 load
->num_components
= ir
->type
->vector_elements
;
1335 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1336 load
->src
[1] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1337 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1340 * In UBO's, a true boolean value is any non-zero value, but we consider
1341 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1344 if (ir
->type
->is_boolean())
1345 this->result
= nir_ine(&b
, &load
->dest
.ssa
, nir_imm_int(&b
, 0));
1350 case ir_unop_interpolate_at_centroid
:
1351 case ir_binop_interpolate_at_offset
:
1352 case ir_binop_interpolate_at_sample
: {
1353 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1354 ir_swizzle
*swizzle
= NULL
;
1356 /* the api does not allow a swizzle here, but the varying packing code
1357 * may have pushed one into here.
1359 swizzle
= ir
->operands
[0]->as_swizzle();
1361 deref
= swizzle
->val
->as_dereference();
1365 deref
->accept(this);
1367 nir_intrinsic_op op
;
1368 if (this->deref_head
->var
->data
.mode
== nir_var_shader_in
) {
1369 switch (ir
->operation
) {
1370 case ir_unop_interpolate_at_centroid
:
1371 op
= nir_intrinsic_interp_var_at_centroid
;
1373 case ir_binop_interpolate_at_offset
:
1374 op
= nir_intrinsic_interp_var_at_offset
;
1376 case ir_binop_interpolate_at_sample
:
1377 op
= nir_intrinsic_interp_var_at_sample
;
1380 unreachable("Invalid interpolation intrinsic");
1383 /* This case can happen if the vertex shader does not write the
1384 * given varying. In this case, the linker will lower it to a
1385 * global variable. Since interpolating a variable makes no
1386 * sense, we'll just turn it into a load which will probably
1387 * eventually end up as an SSA definition.
1389 assert(this->deref_head
->var
->data
.mode
== nir_var_global
);
1390 op
= nir_intrinsic_load_var
;
1393 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1394 intrin
->num_components
= deref
->type
->vector_elements
;
1395 intrin
->variables
[0] = this->deref_head
;
1396 ralloc_steal(intrin
, intrin
->variables
[0]);
1398 if (intrin
->intrinsic
== nir_intrinsic_interp_var_at_offset
||
1399 intrin
->intrinsic
== nir_intrinsic_interp_var_at_sample
)
1400 intrin
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1402 unsigned bit_size
= glsl_get_bit_size(deref
->type
);
1403 add_instr(&intrin
->instr
, deref
->type
->vector_elements
, bit_size
);
1406 unsigned swiz
[4] = {
1407 swizzle
->mask
.x
, swizzle
->mask
.y
, swizzle
->mask
.z
, swizzle
->mask
.w
1410 result
= nir_swizzle(&b
, result
, swiz
,
1411 swizzle
->type
->vector_elements
, false);
1421 nir_ssa_def
*srcs
[4];
1422 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1423 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1425 glsl_base_type types
[4];
1426 for (unsigned i
= 0; i
< ir
->get_num_operands(); i
++)
1428 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1430 types
[i
] = GLSL_TYPE_FLOAT
;
1432 glsl_base_type out_type
;
1434 out_type
= ir
->type
->base_type
;
1436 out_type
= GLSL_TYPE_FLOAT
;
1438 switch (ir
->operation
) {
1439 case ir_unop_bit_not
: result
= nir_inot(&b
, srcs
[0]); break;
1440 case ir_unop_logic_not
:
1441 result
= supports_ints
? nir_inot(&b
, srcs
[0]) : nir_fnot(&b
, srcs
[0]);
1444 result
= type_is_float(types
[0]) ? nir_fneg(&b
, srcs
[0])
1445 : nir_ineg(&b
, srcs
[0]);
1448 result
= type_is_float(types
[0]) ? nir_fabs(&b
, srcs
[0])
1449 : nir_iabs(&b
, srcs
[0]);
1451 case ir_unop_saturate
:
1452 assert(type_is_float(types
[0]));
1453 result
= nir_fsat(&b
, srcs
[0]);
1456 result
= type_is_float(types
[0]) ? nir_fsign(&b
, srcs
[0])
1457 : nir_isign(&b
, srcs
[0]);
1459 case ir_unop_rcp
: result
= nir_frcp(&b
, srcs
[0]); break;
1460 case ir_unop_rsq
: result
= nir_frsq(&b
, srcs
[0]); break;
1461 case ir_unop_sqrt
: result
= nir_fsqrt(&b
, srcs
[0]); break;
1462 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1463 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1464 case ir_unop_exp2
: result
= nir_fexp2(&b
, srcs
[0]); break;
1465 case ir_unop_log2
: result
= nir_flog2(&b
, srcs
[0]); break;
1467 result
= supports_ints
? nir_i2f32(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1470 result
= supports_ints
? nir_u2f32(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1473 result
= supports_ints
? nir_b2f(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1507 case ir_unop_i642u64
:
1508 case ir_unop_u642i64
: {
1509 nir_alu_type src_type
= nir_get_nir_type_for_glsl_base_type(types
[0]);
1510 nir_alu_type dst_type
= nir_get_nir_type_for_glsl_base_type(out_type
);
1511 result
= nir_build_alu(&b
, nir_type_conversion_op(src_type
, dst_type
),
1512 srcs
[0], NULL
, NULL
, NULL
);
1513 /* b2i and b2f don't have fixed bit-size versions so the builder will
1514 * just assume 32 and we have to fix it up here.
1516 result
->bit_size
= nir_alu_type_get_type_size(dst_type
);
1520 case ir_unop_bitcast_i2f
:
1521 case ir_unop_bitcast_f2i
:
1522 case ir_unop_bitcast_u2f
:
1523 case ir_unop_bitcast_f2u
:
1524 case ir_unop_bitcast_i642d
:
1525 case ir_unop_bitcast_d2i64
:
1526 case ir_unop_bitcast_u642d
:
1527 case ir_unop_bitcast_d2u64
:
1528 case ir_unop_subroutine_to_int
:
1530 result
= nir_imov(&b
, srcs
[0]);
1532 case ir_unop_trunc
: result
= nir_ftrunc(&b
, srcs
[0]); break;
1533 case ir_unop_ceil
: result
= nir_fceil(&b
, srcs
[0]); break;
1534 case ir_unop_floor
: result
= nir_ffloor(&b
, srcs
[0]); break;
1535 case ir_unop_fract
: result
= nir_ffract(&b
, srcs
[0]); break;
1536 case ir_unop_round_even
: result
= nir_fround_even(&b
, srcs
[0]); break;
1537 case ir_unop_sin
: result
= nir_fsin(&b
, srcs
[0]); break;
1538 case ir_unop_cos
: result
= nir_fcos(&b
, srcs
[0]); break;
1539 case ir_unop_dFdx
: result
= nir_fddx(&b
, srcs
[0]); break;
1540 case ir_unop_dFdy
: result
= nir_fddy(&b
, srcs
[0]); break;
1541 case ir_unop_dFdx_fine
: result
= nir_fddx_fine(&b
, srcs
[0]); break;
1542 case ir_unop_dFdy_fine
: result
= nir_fddy_fine(&b
, srcs
[0]); break;
1543 case ir_unop_dFdx_coarse
: result
= nir_fddx_coarse(&b
, srcs
[0]); break;
1544 case ir_unop_dFdy_coarse
: result
= nir_fddy_coarse(&b
, srcs
[0]); break;
1545 case ir_unop_pack_snorm_2x16
:
1546 result
= nir_pack_snorm_2x16(&b
, srcs
[0]);
1548 case ir_unop_pack_snorm_4x8
:
1549 result
= nir_pack_snorm_4x8(&b
, srcs
[0]);
1551 case ir_unop_pack_unorm_2x16
:
1552 result
= nir_pack_unorm_2x16(&b
, srcs
[0]);
1554 case ir_unop_pack_unorm_4x8
:
1555 result
= nir_pack_unorm_4x8(&b
, srcs
[0]);
1557 case ir_unop_pack_half_2x16
:
1558 result
= nir_pack_half_2x16(&b
, srcs
[0]);
1560 case ir_unop_unpack_snorm_2x16
:
1561 result
= nir_unpack_snorm_2x16(&b
, srcs
[0]);
1563 case ir_unop_unpack_snorm_4x8
:
1564 result
= nir_unpack_snorm_4x8(&b
, srcs
[0]);
1566 case ir_unop_unpack_unorm_2x16
:
1567 result
= nir_unpack_unorm_2x16(&b
, srcs
[0]);
1569 case ir_unop_unpack_unorm_4x8
:
1570 result
= nir_unpack_unorm_4x8(&b
, srcs
[0]);
1572 case ir_unop_unpack_half_2x16
:
1573 result
= nir_unpack_half_2x16(&b
, srcs
[0]);
1575 case ir_unop_pack_double_2x32
:
1576 case ir_unop_pack_int_2x32
:
1577 case ir_unop_pack_uint_2x32
:
1578 result
= nir_pack_64_2x32(&b
, srcs
[0]);
1580 case ir_unop_unpack_double_2x32
:
1581 case ir_unop_unpack_int_2x32
:
1582 case ir_unop_unpack_uint_2x32
:
1583 result
= nir_unpack_64_2x32(&b
, srcs
[0]);
1585 case ir_unop_bitfield_reverse
:
1586 result
= nir_bitfield_reverse(&b
, srcs
[0]);
1588 case ir_unop_bit_count
:
1589 result
= nir_bit_count(&b
, srcs
[0]);
1591 case ir_unop_find_msb
:
1593 case GLSL_TYPE_UINT
:
1594 result
= nir_ufind_msb(&b
, srcs
[0]);
1597 result
= nir_ifind_msb(&b
, srcs
[0]);
1600 unreachable("Invalid type for findMSB()");
1603 case ir_unop_find_lsb
:
1604 result
= nir_find_lsb(&b
, srcs
[0]);
1608 switch (ir
->type
->vector_elements
) {
1610 switch (ir
->operands
[0]->type
->vector_elements
) {
1611 case 1: result
= nir_fnoise1_1(&b
, srcs
[0]); break;
1612 case 2: result
= nir_fnoise1_2(&b
, srcs
[0]); break;
1613 case 3: result
= nir_fnoise1_3(&b
, srcs
[0]); break;
1614 case 4: result
= nir_fnoise1_4(&b
, srcs
[0]); break;
1615 default: unreachable("not reached");
1619 switch (ir
->operands
[0]->type
->vector_elements
) {
1620 case 1: result
= nir_fnoise2_1(&b
, srcs
[0]); break;
1621 case 2: result
= nir_fnoise2_2(&b
, srcs
[0]); break;
1622 case 3: result
= nir_fnoise2_3(&b
, srcs
[0]); break;
1623 case 4: result
= nir_fnoise2_4(&b
, srcs
[0]); break;
1624 default: unreachable("not reached");
1628 switch (ir
->operands
[0]->type
->vector_elements
) {
1629 case 1: result
= nir_fnoise3_1(&b
, srcs
[0]); break;
1630 case 2: result
= nir_fnoise3_2(&b
, srcs
[0]); break;
1631 case 3: result
= nir_fnoise3_3(&b
, srcs
[0]); break;
1632 case 4: result
= nir_fnoise3_4(&b
, srcs
[0]); break;
1633 default: unreachable("not reached");
1637 switch (ir
->operands
[0]->type
->vector_elements
) {
1638 case 1: result
= nir_fnoise4_1(&b
, srcs
[0]); break;
1639 case 2: result
= nir_fnoise4_2(&b
, srcs
[0]); break;
1640 case 3: result
= nir_fnoise4_3(&b
, srcs
[0]); break;
1641 case 4: result
= nir_fnoise4_4(&b
, srcs
[0]); break;
1642 default: unreachable("not reached");
1646 unreachable("not reached");
1649 case ir_unop_get_buffer_size
: {
1650 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
1652 nir_intrinsic_get_buffer_size
);
1653 load
->num_components
= ir
->type
->vector_elements
;
1654 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1655 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
1656 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1661 result
= type_is_float(out_type
) ? nir_fadd(&b
, srcs
[0], srcs
[1])
1662 : nir_iadd(&b
, srcs
[0], srcs
[1]);
1665 result
= type_is_float(out_type
) ? nir_fsub(&b
, srcs
[0], srcs
[1])
1666 : nir_isub(&b
, srcs
[0], srcs
[1]);
1669 result
= type_is_float(out_type
) ? nir_fmul(&b
, srcs
[0], srcs
[1])
1670 : nir_imul(&b
, srcs
[0], srcs
[1]);
1673 if (type_is_float(out_type
))
1674 result
= nir_fdiv(&b
, srcs
[0], srcs
[1]);
1675 else if (type_is_signed(out_type
))
1676 result
= nir_idiv(&b
, srcs
[0], srcs
[1]);
1678 result
= nir_udiv(&b
, srcs
[0], srcs
[1]);
1681 result
= type_is_float(out_type
) ? nir_fmod(&b
, srcs
[0], srcs
[1])
1682 : nir_umod(&b
, srcs
[0], srcs
[1]);
1685 if (type_is_float(out_type
))
1686 result
= nir_fmin(&b
, srcs
[0], srcs
[1]);
1687 else if (type_is_signed(out_type
))
1688 result
= nir_imin(&b
, srcs
[0], srcs
[1]);
1690 result
= nir_umin(&b
, srcs
[0], srcs
[1]);
1693 if (type_is_float(out_type
))
1694 result
= nir_fmax(&b
, srcs
[0], srcs
[1]);
1695 else if (type_is_signed(out_type
))
1696 result
= nir_imax(&b
, srcs
[0], srcs
[1]);
1698 result
= nir_umax(&b
, srcs
[0], srcs
[1]);
1700 case ir_binop_pow
: result
= nir_fpow(&b
, srcs
[0], srcs
[1]); break;
1701 case ir_binop_bit_and
: result
= nir_iand(&b
, srcs
[0], srcs
[1]); break;
1702 case ir_binop_bit_or
: result
= nir_ior(&b
, srcs
[0], srcs
[1]); break;
1703 case ir_binop_bit_xor
: result
= nir_ixor(&b
, srcs
[0], srcs
[1]); break;
1704 case ir_binop_logic_and
:
1705 result
= supports_ints
? nir_iand(&b
, srcs
[0], srcs
[1])
1706 : nir_fand(&b
, srcs
[0], srcs
[1]);
1708 case ir_binop_logic_or
:
1709 result
= supports_ints
? nir_ior(&b
, srcs
[0], srcs
[1])
1710 : nir_for(&b
, srcs
[0], srcs
[1]);
1712 case ir_binop_logic_xor
:
1713 result
= supports_ints
? nir_ixor(&b
, srcs
[0], srcs
[1])
1714 : nir_fxor(&b
, srcs
[0], srcs
[1]);
1716 case ir_binop_lshift
: result
= nir_ishl(&b
, srcs
[0], srcs
[1]); break;
1717 case ir_binop_rshift
:
1718 result
= (type_is_signed(out_type
)) ? nir_ishr(&b
, srcs
[0], srcs
[1])
1719 : nir_ushr(&b
, srcs
[0], srcs
[1]);
1721 case ir_binop_imul_high
:
1722 result
= (out_type
== GLSL_TYPE_INT
) ? nir_imul_high(&b
, srcs
[0], srcs
[1])
1723 : nir_umul_high(&b
, srcs
[0], srcs
[1]);
1725 case ir_binop_carry
: result
= nir_uadd_carry(&b
, srcs
[0], srcs
[1]); break;
1726 case ir_binop_borrow
: result
= nir_usub_borrow(&b
, srcs
[0], srcs
[1]); break;
1728 if (supports_ints
) {
1729 if (type_is_float(types
[0]))
1730 result
= nir_flt(&b
, srcs
[0], srcs
[1]);
1731 else if (type_is_signed(types
[0]))
1732 result
= nir_ilt(&b
, srcs
[0], srcs
[1]);
1734 result
= nir_ult(&b
, srcs
[0], srcs
[1]);
1736 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1739 case ir_binop_greater
:
1740 if (supports_ints
) {
1741 if (type_is_float(types
[0]))
1742 result
= nir_flt(&b
, srcs
[1], srcs
[0]);
1743 else if (type_is_signed(types
[0]))
1744 result
= nir_ilt(&b
, srcs
[1], srcs
[0]);
1746 result
= nir_ult(&b
, srcs
[1], srcs
[0]);
1748 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1751 case ir_binop_lequal
:
1752 if (supports_ints
) {
1753 if (type_is_float(types
[0]))
1754 result
= nir_fge(&b
, srcs
[1], srcs
[0]);
1755 else if (type_is_signed(types
[0]))
1756 result
= nir_ige(&b
, srcs
[1], srcs
[0]);
1758 result
= nir_uge(&b
, srcs
[1], srcs
[0]);
1760 result
= nir_slt(&b
, srcs
[1], srcs
[0]);
1763 case ir_binop_gequal
:
1764 if (supports_ints
) {
1765 if (type_is_float(types
[0]))
1766 result
= nir_fge(&b
, srcs
[0], srcs
[1]);
1767 else if (type_is_signed(types
[0]))
1768 result
= nir_ige(&b
, srcs
[0], srcs
[1]);
1770 result
= nir_uge(&b
, srcs
[0], srcs
[1]);
1772 result
= nir_slt(&b
, srcs
[0], srcs
[1]);
1775 case ir_binop_equal
:
1776 if (supports_ints
) {
1777 if (type_is_float(types
[0]))
1778 result
= nir_feq(&b
, srcs
[0], srcs
[1]);
1780 result
= nir_ieq(&b
, srcs
[0], srcs
[1]);
1782 result
= nir_seq(&b
, srcs
[0], srcs
[1]);
1785 case ir_binop_nequal
:
1786 if (supports_ints
) {
1787 if (type_is_float(types
[0]))
1788 result
= nir_fne(&b
, srcs
[0], srcs
[1]);
1790 result
= nir_ine(&b
, srcs
[0], srcs
[1]);
1792 result
= nir_sne(&b
, srcs
[0], srcs
[1]);
1795 case ir_binop_all_equal
:
1796 if (supports_ints
) {
1797 if (type_is_float(types
[0])) {
1798 switch (ir
->operands
[0]->type
->vector_elements
) {
1799 case 1: result
= nir_feq(&b
, srcs
[0], srcs
[1]); break;
1800 case 2: result
= nir_ball_fequal2(&b
, srcs
[0], srcs
[1]); break;
1801 case 3: result
= nir_ball_fequal3(&b
, srcs
[0], srcs
[1]); break;
1802 case 4: result
= nir_ball_fequal4(&b
, srcs
[0], srcs
[1]); break;
1804 unreachable("not reached");
1807 switch (ir
->operands
[0]->type
->vector_elements
) {
1808 case 1: result
= nir_ieq(&b
, srcs
[0], srcs
[1]); break;
1809 case 2: result
= nir_ball_iequal2(&b
, srcs
[0], srcs
[1]); break;
1810 case 3: result
= nir_ball_iequal3(&b
, srcs
[0], srcs
[1]); break;
1811 case 4: result
= nir_ball_iequal4(&b
, srcs
[0], srcs
[1]); break;
1813 unreachable("not reached");
1817 switch (ir
->operands
[0]->type
->vector_elements
) {
1818 case 1: result
= nir_seq(&b
, srcs
[0], srcs
[1]); break;
1819 case 2: result
= nir_fall_equal2(&b
, srcs
[0], srcs
[1]); break;
1820 case 3: result
= nir_fall_equal3(&b
, srcs
[0], srcs
[1]); break;
1821 case 4: result
= nir_fall_equal4(&b
, srcs
[0], srcs
[1]); break;
1823 unreachable("not reached");
1827 case ir_binop_any_nequal
:
1828 if (supports_ints
) {
1829 if (type_is_float(types
[0])) {
1830 switch (ir
->operands
[0]->type
->vector_elements
) {
1831 case 1: result
= nir_fne(&b
, srcs
[0], srcs
[1]); break;
1832 case 2: result
= nir_bany_fnequal2(&b
, srcs
[0], srcs
[1]); break;
1833 case 3: result
= nir_bany_fnequal3(&b
, srcs
[0], srcs
[1]); break;
1834 case 4: result
= nir_bany_fnequal4(&b
, srcs
[0], srcs
[1]); break;
1836 unreachable("not reached");
1839 switch (ir
->operands
[0]->type
->vector_elements
) {
1840 case 1: result
= nir_ine(&b
, srcs
[0], srcs
[1]); break;
1841 case 2: result
= nir_bany_inequal2(&b
, srcs
[0], srcs
[1]); break;
1842 case 3: result
= nir_bany_inequal3(&b
, srcs
[0], srcs
[1]); break;
1843 case 4: result
= nir_bany_inequal4(&b
, srcs
[0], srcs
[1]); break;
1845 unreachable("not reached");
1849 switch (ir
->operands
[0]->type
->vector_elements
) {
1850 case 1: result
= nir_sne(&b
, srcs
[0], srcs
[1]); break;
1851 case 2: result
= nir_fany_nequal2(&b
, srcs
[0], srcs
[1]); break;
1852 case 3: result
= nir_fany_nequal3(&b
, srcs
[0], srcs
[1]); break;
1853 case 4: result
= nir_fany_nequal4(&b
, srcs
[0], srcs
[1]); break;
1855 unreachable("not reached");
1860 switch (ir
->operands
[0]->type
->vector_elements
) {
1861 case 2: result
= nir_fdot2(&b
, srcs
[0], srcs
[1]); break;
1862 case 3: result
= nir_fdot3(&b
, srcs
[0], srcs
[1]); break;
1863 case 4: result
= nir_fdot4(&b
, srcs
[0], srcs
[1]); break;
1865 unreachable("not reached");
1869 case ir_binop_ldexp
: result
= nir_ldexp(&b
, srcs
[0], srcs
[1]); break;
1871 result
= nir_ffma(&b
, srcs
[0], srcs
[1], srcs
[2]);
1874 result
= nir_flrp(&b
, srcs
[0], srcs
[1], srcs
[2]);
1878 result
= nir_bcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1880 result
= nir_fcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
1882 case ir_triop_bitfield_extract
:
1883 result
= (out_type
== GLSL_TYPE_INT
) ?
1884 nir_ibitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]) :
1885 nir_ubitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]);
1887 case ir_quadop_bitfield_insert
:
1888 result
= nir_bitfield_insert(&b
, srcs
[0], srcs
[1], srcs
[2], srcs
[3]);
1890 case ir_quadop_vector
:
1891 result
= nir_vec(&b
, srcs
, ir
->type
->vector_elements
);
1895 unreachable("not reached");
1900 nir_visitor::visit(ir_swizzle
*ir
)
1902 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
1903 result
= nir_swizzle(&b
, evaluate_rvalue(ir
->val
), swizzle
,
1904 ir
->type
->vector_elements
, !supports_ints
);
1908 nir_visitor::visit(ir_texture
*ir
)
1915 num_srcs
= 1; /* coordinate */
1920 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
1921 num_srcs
= 2; /* coordinate, bias/lod */
1925 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
1931 if (ir
->lod_info
.lod
!= NULL
)
1932 num_srcs
= 2; /* coordinate, lod */
1934 num_srcs
= 1; /* coordinate */
1938 op
= nir_texop_txf_ms
;
1939 num_srcs
= 2; /* coordinate, sample_index */
1944 if (ir
->lod_info
.lod
!= NULL
)
1945 num_srcs
= 1; /* lod */
1952 num_srcs
= 1; /* coordinate */
1957 num_srcs
= 1; /* coordinate */
1960 case ir_query_levels
:
1961 op
= nir_texop_query_levels
;
1965 case ir_texture_samples
:
1966 op
= nir_texop_texture_samples
;
1970 case ir_samples_identical
:
1971 op
= nir_texop_samples_identical
;
1972 num_srcs
= 1; /* coordinate */
1976 unreachable("not reached");
1979 if (ir
->projector
!= NULL
)
1981 if (ir
->shadow_comparator
!= NULL
)
1983 if (ir
->offset
!= NULL
)
1986 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
1989 instr
->sampler_dim
=
1990 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
1991 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
1992 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
1993 if (instr
->is_shadow
)
1994 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
1995 switch (ir
->type
->base_type
) {
1996 case GLSL_TYPE_FLOAT
:
1997 instr
->dest_type
= nir_type_float
;
2000 instr
->dest_type
= nir_type_int
;
2002 case GLSL_TYPE_BOOL
:
2003 case GLSL_TYPE_UINT
:
2004 instr
->dest_type
= nir_type_uint
;
2007 unreachable("not reached");
2010 instr
->texture
= evaluate_deref(&instr
->instr
, ir
->sampler
);
2012 unsigned src_number
= 0;
2014 if (ir
->coordinate
!= NULL
) {
2015 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
2016 instr
->src
[src_number
].src
=
2017 nir_src_for_ssa(evaluate_rvalue(ir
->coordinate
));
2018 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
2022 if (ir
->projector
!= NULL
) {
2023 instr
->src
[src_number
].src
=
2024 nir_src_for_ssa(evaluate_rvalue(ir
->projector
));
2025 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
2029 if (ir
->shadow_comparator
!= NULL
) {
2030 instr
->src
[src_number
].src
=
2031 nir_src_for_ssa(evaluate_rvalue(ir
->shadow_comparator
));
2032 instr
->src
[src_number
].src_type
= nir_tex_src_comparator
;
2036 if (ir
->offset
!= NULL
) {
2037 /* we don't support multiple offsets yet */
2038 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
2040 instr
->src
[src_number
].src
=
2041 nir_src_for_ssa(evaluate_rvalue(ir
->offset
));
2042 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
2048 instr
->src
[src_number
].src
=
2049 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.bias
));
2050 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
2057 if (ir
->lod_info
.lod
!= NULL
) {
2058 instr
->src
[src_number
].src
=
2059 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.lod
));
2060 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
2066 instr
->src
[src_number
].src
=
2067 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdx
));
2068 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
2070 instr
->src
[src_number
].src
=
2071 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdy
));
2072 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
2077 instr
->src
[src_number
].src
=
2078 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.sample_index
));
2079 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
2084 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
2091 assert(src_number
== num_srcs
);
2093 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
2094 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
), bit_size
);
2098 nir_visitor::visit(ir_constant
*ir
)
2101 * We don't know if this variable is an array or struct that gets
2102 * dereferenced, so do the safe thing an make it a variable with a
2103 * constant initializer and return a dereference.
2107 nir_local_variable_create(this->impl
, ir
->type
, "const_temp");
2108 var
->data
.read_only
= true;
2109 var
->constant_initializer
= constant_copy(ir
, var
);
2111 this->deref_head
= nir_deref_var_create(this->shader
, var
);
2112 this->deref_tail
= &this->deref_head
->deref
;
2116 nir_visitor::visit(ir_dereference_variable
*ir
)
2118 struct hash_entry
*entry
=
2119 _mesa_hash_table_search(this->var_table
, ir
->var
);
2121 nir_variable
*var
= (nir_variable
*) entry
->data
;
2123 nir_deref_var
*deref
= nir_deref_var_create(this->shader
, var
);
2124 this->deref_head
= deref
;
2125 this->deref_tail
= &deref
->deref
;
2129 nir_visitor::visit(ir_dereference_record
*ir
)
2131 ir
->record
->accept(this);
2133 int field_index
= this->deref_tail
->type
->field_index(ir
->field
);
2134 assert(field_index
>= 0);
2136 nir_deref_struct
*deref
= nir_deref_struct_create(this->deref_tail
, field_index
);
2137 deref
->deref
.type
= ir
->type
;
2138 this->deref_tail
->child
= &deref
->deref
;
2139 this->deref_tail
= &deref
->deref
;
2143 nir_visitor::visit(ir_dereference_array
*ir
)
2145 nir_deref_array
*deref
= nir_deref_array_create(this->shader
);
2146 deref
->deref
.type
= ir
->type
;
2148 ir_constant
*const_index
= ir
->array_index
->as_constant();
2149 if (const_index
!= NULL
) {
2150 deref
->deref_array_type
= nir_deref_array_type_direct
;
2151 deref
->base_offset
= const_index
->value
.u
[0];
2153 deref
->deref_array_type
= nir_deref_array_type_indirect
;
2155 nir_src_for_ssa(evaluate_rvalue(ir
->array_index
));
2158 ir
->array
->accept(this);
2160 this->deref_tail
->child
= &deref
->deref
;
2161 ralloc_steal(this->deref_tail
, deref
);
2162 this->deref_tail
= &deref
->deref
;
2166 nir_visitor::visit(ir_barrier
*)
2168 nir_intrinsic_instr
*instr
=
2169 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
2170 nir_builder_instr_insert(&b
, &instr
->instr
);