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 "float64_glsl.h"
29 #include "glsl_to_nir.h"
30 #include "ir_visitor.h"
31 #include "ir_hierarchical_visitor.h"
33 #include "ir_optimization.h"
35 #include "compiler/nir/nir_control_flow.h"
36 #include "compiler/nir/nir_builder.h"
37 #include "main/errors.h"
38 #include "main/imports.h"
39 #include "main/mtypes.h"
40 #include "main/shaderobj.h"
41 #include "util/u_math.h"
44 * pass to lower GLSL IR to NIR
46 * This will lower variable dereferences to loads/stores of corresponding
47 * variables in NIR - the variables will be converted to registers in a later
53 class nir_visitor
: public ir_visitor
56 nir_visitor(gl_context
*ctx
, nir_shader
*shader
);
59 virtual void visit(ir_variable
*);
60 virtual void visit(ir_function
*);
61 virtual void visit(ir_function_signature
*);
62 virtual void visit(ir_loop
*);
63 virtual void visit(ir_if
*);
64 virtual void visit(ir_discard
*);
65 virtual void visit(ir_loop_jump
*);
66 virtual void visit(ir_return
*);
67 virtual void visit(ir_call
*);
68 virtual void visit(ir_assignment
*);
69 virtual void visit(ir_emit_vertex
*);
70 virtual void visit(ir_end_primitive
*);
71 virtual void visit(ir_expression
*);
72 virtual void visit(ir_swizzle
*);
73 virtual void visit(ir_texture
*);
74 virtual void visit(ir_constant
*);
75 virtual void visit(ir_dereference_variable
*);
76 virtual void visit(ir_dereference_record
*);
77 virtual void visit(ir_dereference_array
*);
78 virtual void visit(ir_barrier
*);
80 void create_function(ir_function_signature
*ir
);
83 void add_instr(nir_instr
*instr
, unsigned num_components
, unsigned bit_size
);
84 nir_ssa_def
*evaluate_rvalue(ir_rvalue
*ir
);
86 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
**srcs
);
87 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
);
88 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
,
90 nir_alu_instr
*emit(nir_op op
, unsigned dest_size
, nir_ssa_def
*src1
,
91 nir_ssa_def
*src2
, nir_ssa_def
*src3
);
97 nir_function_impl
*impl
;
99 nir_ssa_def
*result
; /* result of the expression tree last visited */
101 nir_deref_instr
*evaluate_deref(ir_instruction
*ir
);
103 nir_constant
*constant_copy(ir_constant
*ir
, void *mem_ctx
);
105 /* most recent deref instruction created */
106 nir_deref_instr
*deref
;
108 /* whether the IR we're operating on is per-function or global */
111 ir_function_signature
*sig
;
113 /* map of ir_variable -> nir_variable */
114 struct hash_table
*var_table
;
116 /* map of ir_function_signature -> nir_function_overload */
117 struct hash_table
*overload_table
;
121 * This visitor runs before the main visitor, calling create_function() for
122 * each function so that the main visitor can resolve forward references in
126 class nir_function_visitor
: public ir_hierarchical_visitor
129 nir_function_visitor(nir_visitor
*v
) : visitor(v
)
132 virtual ir_visitor_status
visit_enter(ir_function
*);
135 nir_visitor
*visitor
;
138 /* glsl_to_nir can only handle converting certain function paramaters
139 * to NIR. This visitor checks for parameters it can't currently handle.
141 class ir_function_param_visitor
: public ir_hierarchical_visitor
144 ir_function_param_visitor()
149 virtual ir_visitor_status
visit_enter(ir_function_signature
*ir
)
152 if (ir
->is_intrinsic())
153 return visit_continue
;
155 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
156 if (!param
->type
->is_vector() || !param
->type
->is_scalar()) {
161 if (param
->data
.mode
== ir_var_function_inout
) {
167 return visit_continue
;
173 } /* end of anonymous namespace */
177 has_unsupported_function_param(exec_list
*ir
)
179 ir_function_param_visitor visitor
;
180 visit_list_elements(&visitor
, ir
);
181 return visitor
.unsupported
;
185 glsl_to_nir(struct gl_context
*ctx
,
186 const struct gl_shader_program
*shader_prog
,
187 gl_shader_stage stage
,
188 const nir_shader_compiler_options
*options
)
190 struct gl_linked_shader
*sh
= shader_prog
->_LinkedShaders
[stage
];
192 const struct gl_shader_compiler_options
*gl_options
=
193 &ctx
->Const
.ShaderCompilerOptions
[stage
];
195 /* glsl_to_nir can only handle converting certain function paramaters
196 * to NIR. If we find something we can't handle then we get the GLSL IR
197 * opts to remove it before we continue on.
199 * TODO: add missing glsl ir to nir support and remove this loop.
201 while (has_unsupported_function_param(sh
->ir
)) {
202 do_common_optimization(sh
->ir
, true, true, gl_options
,
203 ctx
->Const
.NativeIntegers
);
206 nir_shader
*shader
= nir_shader_create(NULL
, stage
, options
,
209 nir_visitor
v1(ctx
, shader
);
210 nir_function_visitor
v2(&v1
);
212 visit_exec_list(sh
->ir
, &v1
);
214 nir_validate_shader(shader
, "after glsl to nir, before function inline");
216 /* We have to lower away local constant initializers right before we
217 * inline functions. That way they get properly initialized at the top
218 * of the function and not at the top of its caller.
220 nir_lower_constant_initializers(shader
, (nir_variable_mode
)~0);
221 nir_lower_returns(shader
);
222 nir_inline_functions(shader
);
223 nir_opt_deref(shader
);
225 nir_validate_shader(shader
, "after function inlining and return lowering");
227 /* Now that we have inlined everything remove all of the functions except
230 foreach_list_typed_safe(nir_function
, function
, node
, &(shader
)->functions
){
231 if (strcmp("main", function
->name
) != 0) {
232 exec_node_remove(&function
->node
);
236 /* Remap the locations to slots so those requiring two slots will occupy
237 * two locations. For instance, if we have in the IR code a dvec3 attr0 in
238 * location 0 and vec4 attr1 in location 1, in NIR attr0 will use
239 * locations/slots 0 and 1, and attr1 will use location/slot 2 */
240 if (shader
->info
.stage
== MESA_SHADER_VERTEX
)
241 nir_remap_dual_slot_attributes(shader
, &sh
->Program
->DualSlotInputs
);
243 shader
->info
.name
= ralloc_asprintf(shader
, "GLSL%d", shader_prog
->Name
);
244 if (shader_prog
->Label
)
245 shader
->info
.label
= ralloc_strdup(shader
, shader_prog
->Label
);
247 /* Check for transform feedback varyings specified via the API */
248 shader
->info
.has_transform_feedback_varyings
=
249 shader_prog
->TransformFeedback
.NumVarying
> 0;
251 /* Check for transform feedback varyings specified in the Shader */
252 if (shader_prog
->last_vert_prog
)
253 shader
->info
.has_transform_feedback_varyings
|=
254 shader_prog
->last_vert_prog
->sh
.LinkedTransformFeedback
->NumVarying
> 0;
256 if (shader
->info
.stage
== MESA_SHADER_FRAGMENT
) {
257 shader
->info
.fs
.pixel_center_integer
= sh
->Program
->info
.fs
.pixel_center_integer
;
258 shader
->info
.fs
.origin_upper_left
= sh
->Program
->info
.fs
.origin_upper_left
;
264 nir_visitor::nir_visitor(gl_context
*ctx
, nir_shader
*shader
)
266 this->supports_ints
= shader
->options
->native_integers
;
267 this->supports_std430
= ctx
->Const
.UseSTD430AsDefaultPacking
;
268 this->shader
= shader
;
269 this->is_global
= true;
270 this->var_table
= _mesa_pointer_hash_table_create(NULL
);
271 this->overload_table
= _mesa_pointer_hash_table_create(NULL
);
275 memset(&this->b
, 0, sizeof(this->b
));
278 nir_visitor::~nir_visitor()
280 _mesa_hash_table_destroy(this->var_table
, NULL
);
281 _mesa_hash_table_destroy(this->overload_table
, NULL
);
285 nir_visitor::evaluate_deref(ir_instruction
*ir
)
292 nir_visitor::constant_copy(ir_constant
*ir
, void *mem_ctx
)
297 nir_constant
*ret
= rzalloc(mem_ctx
, nir_constant
);
299 const unsigned rows
= ir
->type
->vector_elements
;
300 const unsigned cols
= ir
->type
->matrix_columns
;
303 ret
->num_elements
= 0;
304 switch (ir
->type
->base_type
) {
306 /* Only float base types can be matrices. */
309 for (unsigned r
= 0; r
< rows
; r
++)
311 ret
->values
[0].u32
[r
] = ir
->value
.u
[r
];
313 ret
->values
[0].f32
[r
] = ir
->value
.u
[r
];
318 /* Only float base types can be matrices. */
321 for (unsigned r
= 0; r
< rows
; r
++)
323 ret
->values
[0].i32
[r
] = ir
->value
.i
[r
];
325 ret
->values
[0].f32
[r
] = ir
->value
.i
[r
];
329 case GLSL_TYPE_FLOAT
:
330 for (unsigned c
= 0; c
< cols
; c
++) {
331 for (unsigned r
= 0; r
< rows
; r
++)
332 ret
->values
[c
].f32
[r
] = ir
->value
.f
[c
* rows
+ r
];
336 case GLSL_TYPE_DOUBLE
:
337 for (unsigned c
= 0; c
< cols
; c
++) {
338 for (unsigned r
= 0; r
< rows
; r
++)
339 ret
->values
[c
].f64
[r
] = ir
->value
.d
[c
* rows
+ r
];
343 case GLSL_TYPE_UINT64
:
344 /* Only float base types can be matrices. */
347 for (unsigned r
= 0; r
< rows
; r
++)
348 ret
->values
[0].u64
[r
] = ir
->value
.u64
[r
];
351 case GLSL_TYPE_INT64
:
352 /* Only float base types can be matrices. */
355 for (unsigned r
= 0; r
< rows
; r
++)
356 ret
->values
[0].i64
[r
] = ir
->value
.i64
[r
];
360 /* Only float base types can be matrices. */
363 for (unsigned r
= 0; r
< rows
; r
++)
364 ret
->values
[0].b
[r
] = ir
->value
.b
[r
];
368 case GLSL_TYPE_STRUCT
:
369 case GLSL_TYPE_ARRAY
:
370 ret
->elements
= ralloc_array(mem_ctx
, nir_constant
*,
372 ret
->num_elements
= ir
->type
->length
;
374 for (i
= 0; i
< ir
->type
->length
; i
++)
375 ret
->elements
[i
] = constant_copy(ir
->const_elements
[i
], mem_ctx
);
379 unreachable("not reached");
385 static const glsl_type
*
386 wrap_type_in_array(const glsl_type
*elem_type
, const glsl_type
*array_type
)
388 if (!array_type
->is_array())
391 elem_type
= wrap_type_in_array(elem_type
, array_type
->fields
.array
);
393 return glsl_type::get_array_instance(elem_type
, array_type
->length
);
397 nir_visitor::visit(ir_variable
*ir
)
399 /* TODO: In future we should switch to using the NIR lowering pass but for
400 * now just ignore these variables as GLSL IR should have lowered them.
401 * Anything remaining are just dead vars that weren't cleaned up.
403 if (ir
->data
.mode
== ir_var_shader_shared
)
406 /* FINISHME: inout parameters */
407 assert(ir
->data
.mode
!= ir_var_function_inout
);
409 if (ir
->data
.mode
== ir_var_function_out
)
412 nir_variable
*var
= rzalloc(shader
, nir_variable
);
413 var
->type
= ir
->type
;
414 var
->name
= ralloc_strdup(var
, ir
->name
);
416 var
->data
.always_active_io
= ir
->data
.always_active_io
;
417 var
->data
.read_only
= ir
->data
.read_only
;
418 var
->data
.centroid
= ir
->data
.centroid
;
419 var
->data
.sample
= ir
->data
.sample
;
420 var
->data
.patch
= ir
->data
.patch
;
421 var
->data
.invariant
= ir
->data
.invariant
;
422 var
->data
.location
= ir
->data
.location
;
423 var
->data
.stream
= ir
->data
.stream
;
424 var
->data
.compact
= false;
426 switch(ir
->data
.mode
) {
428 case ir_var_temporary
:
430 var
->data
.mode
= nir_var_shader_temp
;
432 var
->data
.mode
= nir_var_function_temp
;
435 case ir_var_function_in
:
436 case ir_var_const_in
:
437 var
->data
.mode
= nir_var_function_temp
;
440 case ir_var_shader_in
:
441 if (shader
->info
.stage
== MESA_SHADER_FRAGMENT
&&
442 ir
->data
.location
== VARYING_SLOT_FACE
) {
443 /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
444 var
->data
.location
= SYSTEM_VALUE_FRONT_FACE
;
445 var
->data
.mode
= nir_var_system_value
;
446 } else if (shader
->info
.stage
== MESA_SHADER_GEOMETRY
&&
447 ir
->data
.location
== VARYING_SLOT_PRIMITIVE_ID
) {
448 /* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
449 var
->data
.location
= SYSTEM_VALUE_PRIMITIVE_ID
;
450 var
->data
.mode
= nir_var_system_value
;
452 var
->data
.mode
= nir_var_shader_in
;
454 if (shader
->info
.stage
== MESA_SHADER_TESS_EVAL
&&
455 (ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
||
456 ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
)) {
457 var
->data
.compact
= ir
->type
->without_array()->is_scalar();
460 if (shader
->info
.stage
> MESA_SHADER_VERTEX
&&
461 ir
->data
.location
>= VARYING_SLOT_CLIP_DIST0
&&
462 ir
->data
.location
<= VARYING_SLOT_CULL_DIST1
) {
463 var
->data
.compact
= ir
->type
->without_array()->is_scalar();
468 case ir_var_shader_out
:
469 var
->data
.mode
= nir_var_shader_out
;
470 if (shader
->info
.stage
== MESA_SHADER_TESS_CTRL
&&
471 (ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
||
472 ir
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
)) {
473 var
->data
.compact
= ir
->type
->without_array()->is_scalar();
476 if (shader
->info
.stage
<= MESA_SHADER_GEOMETRY
&&
477 ir
->data
.location
>= VARYING_SLOT_CLIP_DIST0
&&
478 ir
->data
.location
<= VARYING_SLOT_CULL_DIST1
) {
479 var
->data
.compact
= ir
->type
->without_array()->is_scalar();
484 if (ir
->get_interface_type())
485 var
->data
.mode
= nir_var_mem_ubo
;
487 var
->data
.mode
= nir_var_uniform
;
490 case ir_var_shader_storage
:
491 var
->data
.mode
= nir_var_mem_ssbo
;
494 case ir_var_system_value
:
495 var
->data
.mode
= nir_var_system_value
;
499 unreachable("not reached");
502 unsigned image_access
= 0;
503 if (ir
->data
.memory_read_only
)
504 image_access
|= ACCESS_NON_WRITEABLE
;
505 if (ir
->data
.memory_write_only
)
506 image_access
|= ACCESS_NON_READABLE
;
507 if (ir
->data
.memory_coherent
)
508 image_access
|= ACCESS_COHERENT
;
509 if (ir
->data
.memory_volatile
)
510 image_access
|= ACCESS_VOLATILE
;
511 if (ir
->data
.memory_restrict
)
512 image_access
|= ACCESS_RESTRICT
;
514 /* For UBO and SSBO variables, we need explicit types */
515 if (var
->data
.mode
& (nir_var_mem_ubo
| nir_var_mem_ssbo
)) {
516 const glsl_type
*explicit_ifc_type
=
517 ir
->get_interface_type()->get_explicit_interface_type(supports_std430
);
519 if (ir
->type
->without_array()->is_interface()) {
520 /* If the type contains the interface, wrap the explicit type in the
521 * right number of arrays.
523 var
->type
= wrap_type_in_array(explicit_ifc_type
, ir
->type
);
525 /* Otherwise, this variable is one entry in the interface */
527 for (unsigned i
= 0; i
< explicit_ifc_type
->length
; i
++) {
528 const glsl_struct_field
*field
=
529 &explicit_ifc_type
->fields
.structure
[i
];
530 if (strcmp(ir
->name
, field
->name
) != 0)
533 var
->type
= field
->type
;
534 if (field
->memory_read_only
)
535 image_access
|= ACCESS_NON_WRITEABLE
;
536 if (field
->memory_write_only
)
537 image_access
|= ACCESS_NON_READABLE
;
538 if (field
->memory_coherent
)
539 image_access
|= ACCESS_COHERENT
;
540 if (field
->memory_volatile
)
541 image_access
|= ACCESS_VOLATILE
;
542 if (field
->memory_restrict
)
543 image_access
|= ACCESS_RESTRICT
;
552 var
->data
.interpolation
= ir
->data
.interpolation
;
553 var
->data
.location_frac
= ir
->data
.location_frac
;
555 switch (ir
->data
.depth_layout
) {
556 case ir_depth_layout_none
:
557 var
->data
.depth_layout
= nir_depth_layout_none
;
559 case ir_depth_layout_any
:
560 var
->data
.depth_layout
= nir_depth_layout_any
;
562 case ir_depth_layout_greater
:
563 var
->data
.depth_layout
= nir_depth_layout_greater
;
565 case ir_depth_layout_less
:
566 var
->data
.depth_layout
= nir_depth_layout_less
;
568 case ir_depth_layout_unchanged
:
569 var
->data
.depth_layout
= nir_depth_layout_unchanged
;
572 unreachable("not reached");
575 var
->data
.index
= ir
->data
.index
;
576 var
->data
.descriptor_set
= 0;
577 var
->data
.binding
= ir
->data
.binding
;
578 var
->data
.explicit_binding
= ir
->data
.explicit_binding
;
579 var
->data
.bindless
= ir
->data
.bindless
;
580 var
->data
.offset
= ir
->data
.offset
;
582 var
->data
.image
.access
= (gl_access_qualifier
)image_access
;
583 var
->data
.image
.format
= ir
->data
.image_format
;
585 var
->data
.fb_fetch_output
= ir
->data
.fb_fetch_output
;
586 var
->data
.explicit_xfb_buffer
= ir
->data
.explicit_xfb_buffer
;
587 var
->data
.explicit_xfb_stride
= ir
->data
.explicit_xfb_stride
;
588 var
->data
.xfb_buffer
= ir
->data
.xfb_buffer
;
589 var
->data
.xfb_stride
= ir
->data
.xfb_stride
;
591 var
->num_state_slots
= ir
->get_num_state_slots();
592 if (var
->num_state_slots
> 0) {
593 var
->state_slots
= rzalloc_array(var
, nir_state_slot
,
594 var
->num_state_slots
);
596 ir_state_slot
*state_slots
= ir
->get_state_slots();
597 for (unsigned i
= 0; i
< var
->num_state_slots
; i
++) {
598 for (unsigned j
= 0; j
< 5; j
++)
599 var
->state_slots
[i
].tokens
[j
] = state_slots
[i
].tokens
[j
];
600 var
->state_slots
[i
].swizzle
= state_slots
[i
].swizzle
;
603 var
->state_slots
= NULL
;
606 var
->constant_initializer
= constant_copy(ir
->constant_initializer
, var
);
608 var
->interface_type
= ir
->get_interface_type();
610 if (var
->data
.mode
== nir_var_function_temp
)
611 nir_function_impl_add_variable(impl
, var
);
613 nir_shader_add_variable(shader
, var
);
615 _mesa_hash_table_insert(var_table
, ir
, var
);
619 nir_function_visitor::visit_enter(ir_function
*ir
)
621 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
) {
622 visitor
->create_function(sig
);
624 return visit_continue_with_parent
;
628 nir_visitor::create_function(ir_function_signature
*ir
)
630 if (ir
->is_intrinsic())
633 nir_function
*func
= nir_function_create(shader
, ir
->function_name());
634 if (strcmp(ir
->function_name(), "main") == 0)
635 func
->is_entrypoint
= true;
637 func
->num_params
= ir
->parameters
.length() +
638 (ir
->return_type
!= glsl_type::void_type
);
639 func
->params
= ralloc_array(shader
, nir_parameter
, func
->num_params
);
643 if (ir
->return_type
!= glsl_type::void_type
) {
644 /* The return value is a variable deref (basically an out parameter) */
645 func
->params
[np
].num_components
= 1;
646 func
->params
[np
].bit_size
= 32;
650 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
651 /* FINISHME: pass arrays, structs, etc by reference? */
652 assert(param
->type
->is_vector() || param
->type
->is_scalar());
654 if (param
->data
.mode
== ir_var_function_in
) {
655 func
->params
[np
].num_components
= param
->type
->vector_elements
;
656 func
->params
[np
].bit_size
= glsl_get_bit_size(param
->type
);
658 func
->params
[np
].num_components
= 1;
659 func
->params
[np
].bit_size
= 32;
663 assert(np
== func
->num_params
);
665 _mesa_hash_table_insert(this->overload_table
, ir
, func
);
669 nir_visitor::visit(ir_function
*ir
)
671 foreach_in_list(ir_function_signature
, sig
, &ir
->signatures
)
676 nir_visitor::visit(ir_function_signature
*ir
)
678 if (ir
->is_intrinsic())
683 struct hash_entry
*entry
=
684 _mesa_hash_table_search(this->overload_table
, ir
);
687 nir_function
*func
= (nir_function
*) entry
->data
;
689 if (ir
->is_defined
) {
690 nir_function_impl
*impl
= nir_function_impl_create(func
);
693 this->is_global
= false;
695 nir_builder_init(&b
, impl
);
696 b
.cursor
= nir_after_cf_list(&impl
->body
);
698 unsigned i
= (ir
->return_type
!= glsl_type::void_type
) ? 1 : 0;
700 foreach_in_list(ir_variable
, param
, &ir
->parameters
) {
702 nir_local_variable_create(impl
, param
->type
, param
->name
);
704 if (param
->data
.mode
== ir_var_function_in
) {
705 nir_store_var(&b
, var
, nir_load_param(&b
, i
), ~0);
708 _mesa_hash_table_insert(var_table
, param
, var
);
712 visit_exec_list(&ir
->body
, this);
714 this->is_global
= true;
721 nir_visitor::visit(ir_loop
*ir
)
724 visit_exec_list(&ir
->body_instructions
, this);
725 nir_pop_loop(&b
, NULL
);
729 nir_visitor::visit(ir_if
*ir
)
731 nir_push_if(&b
, evaluate_rvalue(ir
->condition
));
732 visit_exec_list(&ir
->then_instructions
, this);
733 nir_push_else(&b
, NULL
);
734 visit_exec_list(&ir
->else_instructions
, this);
735 nir_pop_if(&b
, NULL
);
739 nir_visitor::visit(ir_discard
*ir
)
742 * discards aren't treated as control flow, because before we lower them
743 * they can appear anywhere in the shader and the stuff after them may still
744 * be executed (yay, crazy GLSL rules!). However, after lowering, all the
745 * discards will be immediately followed by a return.
748 nir_intrinsic_instr
*discard
;
750 discard
= nir_intrinsic_instr_create(this->shader
,
751 nir_intrinsic_discard_if
);
753 nir_src_for_ssa(evaluate_rvalue(ir
->condition
));
755 discard
= nir_intrinsic_instr_create(this->shader
, nir_intrinsic_discard
);
758 nir_builder_instr_insert(&b
, &discard
->instr
);
762 nir_visitor::visit(ir_emit_vertex
*ir
)
764 nir_intrinsic_instr
*instr
=
765 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_emit_vertex
);
766 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
767 nir_builder_instr_insert(&b
, &instr
->instr
);
771 nir_visitor::visit(ir_end_primitive
*ir
)
773 nir_intrinsic_instr
*instr
=
774 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_end_primitive
);
775 nir_intrinsic_set_stream_id(instr
, ir
->stream_id());
776 nir_builder_instr_insert(&b
, &instr
->instr
);
780 nir_visitor::visit(ir_loop_jump
*ir
)
784 case ir_loop_jump::jump_break
:
785 type
= nir_jump_break
;
787 case ir_loop_jump::jump_continue
:
788 type
= nir_jump_continue
;
791 unreachable("not reached");
794 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, type
);
795 nir_builder_instr_insert(&b
, &instr
->instr
);
799 nir_visitor::visit(ir_return
*ir
)
801 if (ir
->value
!= NULL
) {
802 nir_deref_instr
*ret_deref
=
803 nir_build_deref_cast(&b
, nir_load_param(&b
, 0),
804 nir_var_function_temp
, ir
->value
->type
, 0);
806 nir_ssa_def
*val
= evaluate_rvalue(ir
->value
);
807 nir_store_deref(&b
, ret_deref
, val
, ~0);
810 nir_jump_instr
*instr
= nir_jump_instr_create(this->shader
, nir_jump_return
);
811 nir_builder_instr_insert(&b
, &instr
->instr
);
815 intrinsic_set_std430_align(nir_intrinsic_instr
*intrin
, const glsl_type
*type
)
817 unsigned bit_size
= type
->is_boolean() ? 32 : glsl_get_bit_size(type
);
818 unsigned pow2_components
= util_next_power_of_two(type
->vector_elements
);
819 nir_intrinsic_set_align(intrin
, (bit_size
/ 8) * pow2_components
, 0);
823 nir_visitor::visit(ir_call
*ir
)
825 if (ir
->callee
->is_intrinsic()) {
828 switch (ir
->callee
->intrinsic_id
) {
829 case ir_intrinsic_generic_atomic_add
:
830 op
= ir
->return_deref
->type
->is_integer_32_64()
831 ? nir_intrinsic_deref_atomic_add
: nir_intrinsic_deref_atomic_fadd
;
833 case ir_intrinsic_generic_atomic_and
:
834 op
= nir_intrinsic_deref_atomic_and
;
836 case ir_intrinsic_generic_atomic_or
:
837 op
= nir_intrinsic_deref_atomic_or
;
839 case ir_intrinsic_generic_atomic_xor
:
840 op
= nir_intrinsic_deref_atomic_xor
;
842 case ir_intrinsic_generic_atomic_min
:
843 assert(ir
->return_deref
);
844 if (ir
->return_deref
->type
== glsl_type::int_type
)
845 op
= nir_intrinsic_deref_atomic_imin
;
846 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
847 op
= nir_intrinsic_deref_atomic_umin
;
848 else if (ir
->return_deref
->type
== glsl_type::float_type
)
849 op
= nir_intrinsic_deref_atomic_fmin
;
851 unreachable("Invalid type");
853 case ir_intrinsic_generic_atomic_max
:
854 assert(ir
->return_deref
);
855 if (ir
->return_deref
->type
== glsl_type::int_type
)
856 op
= nir_intrinsic_deref_atomic_imax
;
857 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
858 op
= nir_intrinsic_deref_atomic_umax
;
859 else if (ir
->return_deref
->type
== glsl_type::float_type
)
860 op
= nir_intrinsic_deref_atomic_fmax
;
862 unreachable("Invalid type");
864 case ir_intrinsic_generic_atomic_exchange
:
865 op
= nir_intrinsic_deref_atomic_exchange
;
867 case ir_intrinsic_generic_atomic_comp_swap
:
868 op
= ir
->return_deref
->type
->is_integer_32_64()
869 ? nir_intrinsic_deref_atomic_comp_swap
870 : nir_intrinsic_deref_atomic_fcomp_swap
;
872 case ir_intrinsic_atomic_counter_read
:
873 op
= nir_intrinsic_atomic_counter_read_deref
;
875 case ir_intrinsic_atomic_counter_increment
:
876 op
= nir_intrinsic_atomic_counter_inc_deref
;
878 case ir_intrinsic_atomic_counter_predecrement
:
879 op
= nir_intrinsic_atomic_counter_pre_dec_deref
;
881 case ir_intrinsic_atomic_counter_add
:
882 op
= nir_intrinsic_atomic_counter_add_deref
;
884 case ir_intrinsic_atomic_counter_and
:
885 op
= nir_intrinsic_atomic_counter_and_deref
;
887 case ir_intrinsic_atomic_counter_or
:
888 op
= nir_intrinsic_atomic_counter_or_deref
;
890 case ir_intrinsic_atomic_counter_xor
:
891 op
= nir_intrinsic_atomic_counter_xor_deref
;
893 case ir_intrinsic_atomic_counter_min
:
894 op
= nir_intrinsic_atomic_counter_min_deref
;
896 case ir_intrinsic_atomic_counter_max
:
897 op
= nir_intrinsic_atomic_counter_max_deref
;
899 case ir_intrinsic_atomic_counter_exchange
:
900 op
= nir_intrinsic_atomic_counter_exchange_deref
;
902 case ir_intrinsic_atomic_counter_comp_swap
:
903 op
= nir_intrinsic_atomic_counter_comp_swap_deref
;
905 case ir_intrinsic_image_load
:
906 op
= nir_intrinsic_image_deref_load
;
908 case ir_intrinsic_image_store
:
909 op
= nir_intrinsic_image_deref_store
;
911 case ir_intrinsic_image_atomic_add
:
912 op
= ir
->return_deref
->type
->is_integer_32_64()
913 ? nir_intrinsic_image_deref_atomic_add
914 : nir_intrinsic_image_deref_atomic_fadd
;
916 case ir_intrinsic_image_atomic_min
:
917 op
= nir_intrinsic_image_deref_atomic_min
;
919 case ir_intrinsic_image_atomic_max
:
920 op
= nir_intrinsic_image_deref_atomic_max
;
922 case ir_intrinsic_image_atomic_and
:
923 op
= nir_intrinsic_image_deref_atomic_and
;
925 case ir_intrinsic_image_atomic_or
:
926 op
= nir_intrinsic_image_deref_atomic_or
;
928 case ir_intrinsic_image_atomic_xor
:
929 op
= nir_intrinsic_image_deref_atomic_xor
;
931 case ir_intrinsic_image_atomic_exchange
:
932 op
= nir_intrinsic_image_deref_atomic_exchange
;
934 case ir_intrinsic_image_atomic_comp_swap
:
935 op
= nir_intrinsic_image_deref_atomic_comp_swap
;
937 case ir_intrinsic_memory_barrier
:
938 op
= nir_intrinsic_memory_barrier
;
940 case ir_intrinsic_image_size
:
941 op
= nir_intrinsic_image_deref_size
;
943 case ir_intrinsic_image_samples
:
944 op
= nir_intrinsic_image_deref_samples
;
946 case ir_intrinsic_ssbo_store
:
947 op
= nir_intrinsic_store_ssbo
;
949 case ir_intrinsic_ssbo_load
:
950 op
= nir_intrinsic_load_ssbo
;
952 case ir_intrinsic_ssbo_atomic_add
:
953 op
= ir
->return_deref
->type
->is_integer_32_64()
954 ? nir_intrinsic_ssbo_atomic_add
: nir_intrinsic_ssbo_atomic_fadd
;
956 case ir_intrinsic_ssbo_atomic_and
:
957 op
= nir_intrinsic_ssbo_atomic_and
;
959 case ir_intrinsic_ssbo_atomic_or
:
960 op
= nir_intrinsic_ssbo_atomic_or
;
962 case ir_intrinsic_ssbo_atomic_xor
:
963 op
= nir_intrinsic_ssbo_atomic_xor
;
965 case ir_intrinsic_ssbo_atomic_min
:
966 assert(ir
->return_deref
);
967 if (ir
->return_deref
->type
== glsl_type::int_type
)
968 op
= nir_intrinsic_ssbo_atomic_imin
;
969 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
970 op
= nir_intrinsic_ssbo_atomic_umin
;
971 else if (ir
->return_deref
->type
== glsl_type::float_type
)
972 op
= nir_intrinsic_ssbo_atomic_fmin
;
974 unreachable("Invalid type");
976 case ir_intrinsic_ssbo_atomic_max
:
977 assert(ir
->return_deref
);
978 if (ir
->return_deref
->type
== glsl_type::int_type
)
979 op
= nir_intrinsic_ssbo_atomic_imax
;
980 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
981 op
= nir_intrinsic_ssbo_atomic_umax
;
982 else if (ir
->return_deref
->type
== glsl_type::float_type
)
983 op
= nir_intrinsic_ssbo_atomic_fmax
;
985 unreachable("Invalid type");
987 case ir_intrinsic_ssbo_atomic_exchange
:
988 op
= nir_intrinsic_ssbo_atomic_exchange
;
990 case ir_intrinsic_ssbo_atomic_comp_swap
:
991 op
= ir
->return_deref
->type
->is_integer_32_64()
992 ? nir_intrinsic_ssbo_atomic_comp_swap
993 : nir_intrinsic_ssbo_atomic_fcomp_swap
;
995 case ir_intrinsic_shader_clock
:
996 op
= nir_intrinsic_shader_clock
;
998 case ir_intrinsic_begin_invocation_interlock
:
999 op
= nir_intrinsic_begin_invocation_interlock
;
1001 case ir_intrinsic_end_invocation_interlock
:
1002 op
= nir_intrinsic_end_invocation_interlock
;
1004 case ir_intrinsic_group_memory_barrier
:
1005 op
= nir_intrinsic_group_memory_barrier
;
1007 case ir_intrinsic_memory_barrier_atomic_counter
:
1008 op
= nir_intrinsic_memory_barrier_atomic_counter
;
1010 case ir_intrinsic_memory_barrier_buffer
:
1011 op
= nir_intrinsic_memory_barrier_buffer
;
1013 case ir_intrinsic_memory_barrier_image
:
1014 op
= nir_intrinsic_memory_barrier_image
;
1016 case ir_intrinsic_memory_barrier_shared
:
1017 op
= nir_intrinsic_memory_barrier_shared
;
1019 case ir_intrinsic_shared_load
:
1020 op
= nir_intrinsic_load_shared
;
1022 case ir_intrinsic_shared_store
:
1023 op
= nir_intrinsic_store_shared
;
1025 case ir_intrinsic_shared_atomic_add
:
1026 op
= ir
->return_deref
->type
->is_integer_32_64()
1027 ? nir_intrinsic_shared_atomic_add
1028 : nir_intrinsic_shared_atomic_fadd
;
1030 case ir_intrinsic_shared_atomic_and
:
1031 op
= nir_intrinsic_shared_atomic_and
;
1033 case ir_intrinsic_shared_atomic_or
:
1034 op
= nir_intrinsic_shared_atomic_or
;
1036 case ir_intrinsic_shared_atomic_xor
:
1037 op
= nir_intrinsic_shared_atomic_xor
;
1039 case ir_intrinsic_shared_atomic_min
:
1040 assert(ir
->return_deref
);
1041 if (ir
->return_deref
->type
== glsl_type::int_type
)
1042 op
= nir_intrinsic_shared_atomic_imin
;
1043 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
1044 op
= nir_intrinsic_shared_atomic_umin
;
1045 else if (ir
->return_deref
->type
== glsl_type::float_type
)
1046 op
= nir_intrinsic_shared_atomic_fmin
;
1048 unreachable("Invalid type");
1050 case ir_intrinsic_shared_atomic_max
:
1051 assert(ir
->return_deref
);
1052 if (ir
->return_deref
->type
== glsl_type::int_type
)
1053 op
= nir_intrinsic_shared_atomic_imax
;
1054 else if (ir
->return_deref
->type
== glsl_type::uint_type
)
1055 op
= nir_intrinsic_shared_atomic_umax
;
1056 else if (ir
->return_deref
->type
== glsl_type::float_type
)
1057 op
= nir_intrinsic_shared_atomic_fmax
;
1059 unreachable("Invalid type");
1061 case ir_intrinsic_shared_atomic_exchange
:
1062 op
= nir_intrinsic_shared_atomic_exchange
;
1064 case ir_intrinsic_shared_atomic_comp_swap
:
1065 op
= ir
->return_deref
->type
->is_integer_32_64()
1066 ? nir_intrinsic_shared_atomic_comp_swap
1067 : nir_intrinsic_shared_atomic_fcomp_swap
;
1069 case ir_intrinsic_vote_any
:
1070 op
= nir_intrinsic_vote_any
;
1072 case ir_intrinsic_vote_all
:
1073 op
= nir_intrinsic_vote_all
;
1075 case ir_intrinsic_vote_eq
:
1076 op
= nir_intrinsic_vote_ieq
;
1078 case ir_intrinsic_ballot
:
1079 op
= nir_intrinsic_ballot
;
1081 case ir_intrinsic_read_invocation
:
1082 op
= nir_intrinsic_read_invocation
;
1084 case ir_intrinsic_read_first_invocation
:
1085 op
= nir_intrinsic_read_first_invocation
;
1088 unreachable("not reached");
1091 nir_intrinsic_instr
*instr
= nir_intrinsic_instr_create(shader
, op
);
1092 nir_ssa_def
*ret
= &instr
->dest
.ssa
;
1095 case nir_intrinsic_deref_atomic_add
:
1096 case nir_intrinsic_deref_atomic_imin
:
1097 case nir_intrinsic_deref_atomic_umin
:
1098 case nir_intrinsic_deref_atomic_imax
:
1099 case nir_intrinsic_deref_atomic_umax
:
1100 case nir_intrinsic_deref_atomic_and
:
1101 case nir_intrinsic_deref_atomic_or
:
1102 case nir_intrinsic_deref_atomic_xor
:
1103 case nir_intrinsic_deref_atomic_exchange
:
1104 case nir_intrinsic_deref_atomic_comp_swap
:
1105 case nir_intrinsic_deref_atomic_fadd
:
1106 case nir_intrinsic_deref_atomic_fmin
:
1107 case nir_intrinsic_deref_atomic_fmax
:
1108 case nir_intrinsic_deref_atomic_fcomp_swap
: {
1109 int param_count
= ir
->actual_parameters
.length();
1110 assert(param_count
== 2 || param_count
== 3);
1113 exec_node
*param
= ir
->actual_parameters
.get_head();
1114 ir_rvalue
*rvalue
= (ir_rvalue
*) param
;
1115 ir_dereference
*deref
= rvalue
->as_dereference();
1116 ir_swizzle
*swizzle
= NULL
;
1118 /* We may have a swizzle to pick off a single vec4 component */
1119 swizzle
= rvalue
->as_swizzle();
1120 assert(swizzle
&& swizzle
->type
->vector_elements
== 1);
1121 deref
= swizzle
->val
->as_dereference();
1124 nir_deref_instr
*nir_deref
= evaluate_deref(deref
);
1126 nir_deref
= nir_build_deref_array_imm(&b
, nir_deref
,
1129 instr
->src
[0] = nir_src_for_ssa(&nir_deref
->dest
.ssa
);
1131 /* data1 parameter (this is always present) */
1132 param
= param
->get_next();
1133 ir_instruction
*inst
= (ir_instruction
*) param
;
1134 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1136 /* data2 parameter (only with atomic_comp_swap) */
1137 if (param_count
== 3) {
1138 assert(op
== nir_intrinsic_deref_atomic_comp_swap
||
1139 op
== nir_intrinsic_deref_atomic_fcomp_swap
);
1140 param
= param
->get_next();
1141 inst
= (ir_instruction
*) param
;
1142 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1146 assert(ir
->return_deref
);
1147 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1148 ir
->return_deref
->type
->vector_elements
, 32, NULL
);
1149 nir_builder_instr_insert(&b
, &instr
->instr
);
1152 case nir_intrinsic_atomic_counter_read_deref
:
1153 case nir_intrinsic_atomic_counter_inc_deref
:
1154 case nir_intrinsic_atomic_counter_pre_dec_deref
:
1155 case nir_intrinsic_atomic_counter_add_deref
:
1156 case nir_intrinsic_atomic_counter_min_deref
:
1157 case nir_intrinsic_atomic_counter_max_deref
:
1158 case nir_intrinsic_atomic_counter_and_deref
:
1159 case nir_intrinsic_atomic_counter_or_deref
:
1160 case nir_intrinsic_atomic_counter_xor_deref
:
1161 case nir_intrinsic_atomic_counter_exchange_deref
:
1162 case nir_intrinsic_atomic_counter_comp_swap_deref
: {
1163 /* Set the counter variable dereference. */
1164 exec_node
*param
= ir
->actual_parameters
.get_head();
1165 ir_dereference
*counter
= (ir_dereference
*)param
;
1167 instr
->src
[0] = nir_src_for_ssa(&evaluate_deref(counter
)->dest
.ssa
);
1168 param
= param
->get_next();
1170 /* Set the intrinsic destination. */
1171 if (ir
->return_deref
) {
1172 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 32, NULL
);
1175 /* Set the intrinsic parameters. */
1176 if (!param
->is_tail_sentinel()) {
1178 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
1179 param
= param
->get_next();
1182 if (!param
->is_tail_sentinel()) {
1184 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
1185 param
= param
->get_next();
1188 nir_builder_instr_insert(&b
, &instr
->instr
);
1191 case nir_intrinsic_image_deref_load
:
1192 case nir_intrinsic_image_deref_store
:
1193 case nir_intrinsic_image_deref_atomic_add
:
1194 case nir_intrinsic_image_deref_atomic_min
:
1195 case nir_intrinsic_image_deref_atomic_max
:
1196 case nir_intrinsic_image_deref_atomic_and
:
1197 case nir_intrinsic_image_deref_atomic_or
:
1198 case nir_intrinsic_image_deref_atomic_xor
:
1199 case nir_intrinsic_image_deref_atomic_exchange
:
1200 case nir_intrinsic_image_deref_atomic_comp_swap
:
1201 case nir_intrinsic_image_deref_atomic_fadd
:
1202 case nir_intrinsic_image_deref_samples
:
1203 case nir_intrinsic_image_deref_size
: {
1204 nir_ssa_undef_instr
*instr_undef
=
1205 nir_ssa_undef_instr_create(shader
, 1, 32);
1206 nir_builder_instr_insert(&b
, &instr_undef
->instr
);
1208 /* Set the image variable dereference. */
1209 exec_node
*param
= ir
->actual_parameters
.get_head();
1210 ir_dereference
*image
= (ir_dereference
*)param
;
1211 const glsl_type
*type
=
1212 image
->variable_referenced()->type
->without_array();
1214 instr
->src
[0] = nir_src_for_ssa(&evaluate_deref(image
)->dest
.ssa
);
1215 param
= param
->get_next();
1217 /* Set the intrinsic destination. */
1218 if (ir
->return_deref
) {
1219 unsigned num_components
= ir
->return_deref
->type
->vector_elements
;
1220 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1221 num_components
, 32, NULL
);
1224 if (op
== nir_intrinsic_image_deref_size
) {
1225 instr
->num_components
= instr
->dest
.ssa
.num_components
;
1226 } else if (op
== nir_intrinsic_image_deref_load
||
1227 op
== nir_intrinsic_image_deref_store
) {
1228 instr
->num_components
= 4;
1231 if (op
== nir_intrinsic_image_deref_size
||
1232 op
== nir_intrinsic_image_deref_samples
) {
1233 nir_builder_instr_insert(&b
, &instr
->instr
);
1237 /* Set the address argument, extending the coordinate vector to four
1240 nir_ssa_def
*src_addr
=
1241 evaluate_rvalue((ir_dereference
*)param
);
1242 nir_ssa_def
*srcs
[4];
1244 for (int i
= 0; i
< 4; i
++) {
1245 if (i
< type
->coordinate_components())
1246 srcs
[i
] = nir_channel(&b
, src_addr
, i
);
1248 srcs
[i
] = &instr_undef
->def
;
1251 instr
->src
[1] = nir_src_for_ssa(nir_vec(&b
, srcs
, 4));
1252 param
= param
->get_next();
1254 /* Set the sample argument, which is undefined for single-sample
1257 if (type
->sampler_dimensionality
== GLSL_SAMPLER_DIM_MS
) {
1259 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
1260 param
= param
->get_next();
1262 instr
->src
[2] = nir_src_for_ssa(&instr_undef
->def
);
1265 /* Set the intrinsic parameters. */
1266 if (!param
->is_tail_sentinel()) {
1268 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
1269 param
= param
->get_next();
1272 if (!param
->is_tail_sentinel()) {
1274 nir_src_for_ssa(evaluate_rvalue((ir_dereference
*)param
));
1275 param
= param
->get_next();
1277 nir_builder_instr_insert(&b
, &instr
->instr
);
1280 case nir_intrinsic_memory_barrier
:
1281 case nir_intrinsic_group_memory_barrier
:
1282 case nir_intrinsic_memory_barrier_atomic_counter
:
1283 case nir_intrinsic_memory_barrier_buffer
:
1284 case nir_intrinsic_memory_barrier_image
:
1285 case nir_intrinsic_memory_barrier_shared
:
1286 nir_builder_instr_insert(&b
, &instr
->instr
);
1288 case nir_intrinsic_shader_clock
:
1289 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 2, 32, NULL
);
1290 instr
->num_components
= 2;
1291 nir_builder_instr_insert(&b
, &instr
->instr
);
1293 case nir_intrinsic_begin_invocation_interlock
:
1294 nir_builder_instr_insert(&b
, &instr
->instr
);
1296 case nir_intrinsic_end_invocation_interlock
:
1297 nir_builder_instr_insert(&b
, &instr
->instr
);
1299 case nir_intrinsic_store_ssbo
: {
1300 exec_node
*param
= ir
->actual_parameters
.get_head();
1301 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
1303 param
= param
->get_next();
1304 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1306 param
= param
->get_next();
1307 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
1309 param
= param
->get_next();
1310 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
1313 nir_ssa_def
*nir_val
= evaluate_rvalue(val
);
1314 if (val
->type
->is_boolean())
1315 nir_val
= nir_b2i32(&b
, nir_val
);
1317 instr
->src
[0] = nir_src_for_ssa(nir_val
);
1318 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(block
));
1319 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(offset
));
1320 intrinsic_set_std430_align(instr
, val
->type
);
1321 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
1322 instr
->num_components
= val
->type
->vector_elements
;
1324 nir_builder_instr_insert(&b
, &instr
->instr
);
1327 case nir_intrinsic_load_ssbo
: {
1328 exec_node
*param
= ir
->actual_parameters
.get_head();
1329 ir_rvalue
*block
= ((ir_instruction
*)param
)->as_rvalue();
1331 param
= param
->get_next();
1332 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1334 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(block
));
1335 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
1337 const glsl_type
*type
= ir
->return_deref
->var
->type
;
1338 instr
->num_components
= type
->vector_elements
;
1339 intrinsic_set_std430_align(instr
, type
);
1341 /* Setup destination register */
1342 unsigned bit_size
= type
->is_boolean() ? 32 : glsl_get_bit_size(type
);
1343 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1344 type
->vector_elements
, bit_size
, NULL
);
1346 /* Insert the created nir instruction now since in the case of boolean
1347 * result we will need to emit another instruction after it
1349 nir_builder_instr_insert(&b
, &instr
->instr
);
1352 * In SSBO/UBO's, a true boolean value is any non-zero value, but we
1353 * consider a true boolean to be ~0. Fix this up with a != 0
1356 if (type
->is_boolean())
1357 ret
= nir_i2b(&b
, &instr
->dest
.ssa
);
1360 case nir_intrinsic_ssbo_atomic_add
:
1361 case nir_intrinsic_ssbo_atomic_imin
:
1362 case nir_intrinsic_ssbo_atomic_umin
:
1363 case nir_intrinsic_ssbo_atomic_imax
:
1364 case nir_intrinsic_ssbo_atomic_umax
:
1365 case nir_intrinsic_ssbo_atomic_and
:
1366 case nir_intrinsic_ssbo_atomic_or
:
1367 case nir_intrinsic_ssbo_atomic_xor
:
1368 case nir_intrinsic_ssbo_atomic_exchange
:
1369 case nir_intrinsic_ssbo_atomic_comp_swap
:
1370 case nir_intrinsic_ssbo_atomic_fadd
:
1371 case nir_intrinsic_ssbo_atomic_fmin
:
1372 case nir_intrinsic_ssbo_atomic_fmax
:
1373 case nir_intrinsic_ssbo_atomic_fcomp_swap
: {
1374 int param_count
= ir
->actual_parameters
.length();
1375 assert(param_count
== 3 || param_count
== 4);
1378 exec_node
*param
= ir
->actual_parameters
.get_head();
1379 ir_instruction
*inst
= (ir_instruction
*) param
;
1380 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1383 param
= param
->get_next();
1384 inst
= (ir_instruction
*) param
;
1385 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1387 /* data1 parameter (this is always present) */
1388 param
= param
->get_next();
1389 inst
= (ir_instruction
*) param
;
1390 instr
->src
[2] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1392 /* data2 parameter (only with atomic_comp_swap) */
1393 if (param_count
== 4) {
1394 assert(op
== nir_intrinsic_ssbo_atomic_comp_swap
||
1395 op
== nir_intrinsic_ssbo_atomic_fcomp_swap
);
1396 param
= param
->get_next();
1397 inst
= (ir_instruction
*) param
;
1398 instr
->src
[3] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1402 assert(ir
->return_deref
);
1403 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1404 ir
->return_deref
->type
->vector_elements
, 32, NULL
);
1405 nir_builder_instr_insert(&b
, &instr
->instr
);
1408 case nir_intrinsic_load_shared
: {
1409 exec_node
*param
= ir
->actual_parameters
.get_head();
1410 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1412 nir_intrinsic_set_base(instr
, 0);
1413 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(offset
));
1415 const glsl_type
*type
= ir
->return_deref
->var
->type
;
1416 instr
->num_components
= type
->vector_elements
;
1417 intrinsic_set_std430_align(instr
, type
);
1419 /* Setup destination register */
1420 unsigned bit_size
= type
->is_boolean() ? 32 : glsl_get_bit_size(type
);
1421 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1422 type
->vector_elements
, bit_size
, NULL
);
1424 nir_builder_instr_insert(&b
, &instr
->instr
);
1426 /* The value in shared memory is a 32-bit value */
1427 if (type
->is_boolean())
1428 ret
= nir_i2b(&b
, &instr
->dest
.ssa
);
1431 case nir_intrinsic_store_shared
: {
1432 exec_node
*param
= ir
->actual_parameters
.get_head();
1433 ir_rvalue
*offset
= ((ir_instruction
*)param
)->as_rvalue();
1435 param
= param
->get_next();
1436 ir_rvalue
*val
= ((ir_instruction
*)param
)->as_rvalue();
1438 param
= param
->get_next();
1439 ir_constant
*write_mask
= ((ir_instruction
*)param
)->as_constant();
1442 nir_intrinsic_set_base(instr
, 0);
1443 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(offset
));
1445 nir_intrinsic_set_write_mask(instr
, write_mask
->value
.u
[0]);
1447 nir_ssa_def
*nir_val
= evaluate_rvalue(val
);
1448 /* The value in shared memory is a 32-bit value */
1449 if (val
->type
->is_boolean())
1450 nir_val
= nir_b2i32(&b
, nir_val
);
1452 instr
->src
[0] = nir_src_for_ssa(nir_val
);
1453 instr
->num_components
= val
->type
->vector_elements
;
1454 intrinsic_set_std430_align(instr
, val
->type
);
1456 nir_builder_instr_insert(&b
, &instr
->instr
);
1459 case nir_intrinsic_shared_atomic_add
:
1460 case nir_intrinsic_shared_atomic_imin
:
1461 case nir_intrinsic_shared_atomic_umin
:
1462 case nir_intrinsic_shared_atomic_imax
:
1463 case nir_intrinsic_shared_atomic_umax
:
1464 case nir_intrinsic_shared_atomic_and
:
1465 case nir_intrinsic_shared_atomic_or
:
1466 case nir_intrinsic_shared_atomic_xor
:
1467 case nir_intrinsic_shared_atomic_exchange
:
1468 case nir_intrinsic_shared_atomic_comp_swap
:
1469 case nir_intrinsic_shared_atomic_fadd
:
1470 case nir_intrinsic_shared_atomic_fmin
:
1471 case nir_intrinsic_shared_atomic_fmax
:
1472 case nir_intrinsic_shared_atomic_fcomp_swap
: {
1473 int param_count
= ir
->actual_parameters
.length();
1474 assert(param_count
== 2 || param_count
== 3);
1477 exec_node
*param
= ir
->actual_parameters
.get_head();
1478 ir_instruction
*inst
= (ir_instruction
*) param
;
1479 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1481 /* data1 parameter (this is always present) */
1482 param
= param
->get_next();
1483 inst
= (ir_instruction
*) param
;
1484 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1486 /* data2 parameter (only with atomic_comp_swap) */
1487 if (param_count
== 3) {
1488 assert(op
== nir_intrinsic_shared_atomic_comp_swap
||
1489 op
== nir_intrinsic_shared_atomic_fcomp_swap
);
1490 param
= param
->get_next();
1491 inst
= (ir_instruction
*) param
;
1493 nir_src_for_ssa(evaluate_rvalue(inst
->as_rvalue()));
1497 assert(ir
->return_deref
);
1498 unsigned bit_size
= glsl_get_bit_size(ir
->return_deref
->type
);
1499 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1500 ir
->return_deref
->type
->vector_elements
,
1502 nir_builder_instr_insert(&b
, &instr
->instr
);
1505 case nir_intrinsic_vote_any
:
1506 case nir_intrinsic_vote_all
:
1507 case nir_intrinsic_vote_ieq
: {
1508 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
, 1, 1, NULL
);
1509 instr
->num_components
= 1;
1511 ir_rvalue
*value
= (ir_rvalue
*) ir
->actual_parameters
.get_head();
1512 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(value
));
1514 nir_builder_instr_insert(&b
, &instr
->instr
);
1518 case nir_intrinsic_ballot
: {
1519 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1520 ir
->return_deref
->type
->vector_elements
, 64, NULL
);
1521 instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
1523 ir_rvalue
*value
= (ir_rvalue
*) ir
->actual_parameters
.get_head();
1524 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(value
));
1526 nir_builder_instr_insert(&b
, &instr
->instr
);
1529 case nir_intrinsic_read_invocation
: {
1530 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1531 ir
->return_deref
->type
->vector_elements
, 32, NULL
);
1532 instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
1534 ir_rvalue
*value
= (ir_rvalue
*) ir
->actual_parameters
.get_head();
1535 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(value
));
1537 ir_rvalue
*invocation
= (ir_rvalue
*) ir
->actual_parameters
.get_head()->next
;
1538 instr
->src
[1] = nir_src_for_ssa(evaluate_rvalue(invocation
));
1540 nir_builder_instr_insert(&b
, &instr
->instr
);
1543 case nir_intrinsic_read_first_invocation
: {
1544 nir_ssa_dest_init(&instr
->instr
, &instr
->dest
,
1545 ir
->return_deref
->type
->vector_elements
, 32, NULL
);
1546 instr
->num_components
= ir
->return_deref
->type
->vector_elements
;
1548 ir_rvalue
*value
= (ir_rvalue
*) ir
->actual_parameters
.get_head();
1549 instr
->src
[0] = nir_src_for_ssa(evaluate_rvalue(value
));
1551 nir_builder_instr_insert(&b
, &instr
->instr
);
1555 unreachable("not reached");
1558 if (ir
->return_deref
)
1559 nir_store_deref(&b
, evaluate_deref(ir
->return_deref
), ret
, ~0);
1564 struct hash_entry
*entry
=
1565 _mesa_hash_table_search(this->overload_table
, ir
->callee
);
1567 nir_function
*callee
= (nir_function
*) entry
->data
;
1569 nir_call_instr
*call
= nir_call_instr_create(this->shader
, callee
);
1572 nir_deref_instr
*ret_deref
= NULL
;
1573 if (ir
->return_deref
) {
1574 nir_variable
*ret_tmp
=
1575 nir_local_variable_create(this->impl
, ir
->return_deref
->type
,
1577 ret_deref
= nir_build_deref_var(&b
, ret_tmp
);
1578 call
->params
[i
++] = nir_src_for_ssa(&ret_deref
->dest
.ssa
);
1581 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
1582 actual_node
, &ir
->actual_parameters
) {
1583 ir_rvalue
*param_rvalue
= (ir_rvalue
*) actual_node
;
1584 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
1586 if (sig_param
->data
.mode
== ir_var_function_out
) {
1587 nir_deref_instr
*out_deref
= evaluate_deref(param_rvalue
);
1588 call
->params
[i
] = nir_src_for_ssa(&out_deref
->dest
.ssa
);
1589 } else if (sig_param
->data
.mode
== ir_var_function_in
) {
1590 nir_ssa_def
*val
= evaluate_rvalue(param_rvalue
);
1591 nir_src src
= nir_src_for_ssa(val
);
1593 nir_src_copy(&call
->params
[i
], &src
, call
);
1594 } else if (sig_param
->data
.mode
== ir_var_function_inout
) {
1595 unreachable("unimplemented: inout parameters");
1601 nir_builder_instr_insert(&b
, &call
->instr
);
1603 if (ir
->return_deref
)
1604 nir_store_deref(&b
, evaluate_deref(ir
->return_deref
), nir_load_deref(&b
, ret_deref
), ~0);
1608 nir_visitor::visit(ir_assignment
*ir
)
1610 unsigned num_components
= ir
->lhs
->type
->vector_elements
;
1612 b
.exact
= ir
->lhs
->variable_referenced()->data
.invariant
||
1613 ir
->lhs
->variable_referenced()->data
.precise
;
1615 if ((ir
->rhs
->as_dereference() || ir
->rhs
->as_constant()) &&
1616 (ir
->write_mask
== (1 << num_components
) - 1 || ir
->write_mask
== 0)) {
1617 if (ir
->condition
) {
1618 nir_push_if(&b
, evaluate_rvalue(ir
->condition
));
1619 nir_copy_deref(&b
, evaluate_deref(ir
->lhs
), evaluate_deref(ir
->rhs
));
1620 nir_pop_if(&b
, NULL
);
1622 nir_copy_deref(&b
, evaluate_deref(ir
->lhs
), evaluate_deref(ir
->rhs
));
1627 assert(ir
->rhs
->type
->is_scalar() || ir
->rhs
->type
->is_vector());
1629 ir
->lhs
->accept(this);
1630 nir_deref_instr
*lhs_deref
= this->deref
;
1631 nir_ssa_def
*src
= evaluate_rvalue(ir
->rhs
);
1633 if (ir
->write_mask
!= (1 << num_components
) - 1 && ir
->write_mask
!= 0) {
1634 /* GLSL IR will give us the input to the write-masked assignment in a
1635 * single packed vector. So, for example, if the writemask is xzw, then
1636 * we have to swizzle x -> x, y -> z, and z -> w and get the y component
1640 unsigned component
= 0;
1641 for (unsigned i
= 0; i
< 4; i
++) {
1642 swiz
[i
] = ir
->write_mask
& (1 << i
) ? component
++ : 0;
1644 src
= nir_swizzle(&b
, src
, swiz
, num_components
, false);
1647 if (ir
->condition
) {
1648 nir_push_if(&b
, evaluate_rvalue(ir
->condition
));
1649 nir_store_deref(&b
, lhs_deref
, src
, ir
->write_mask
);
1650 nir_pop_if(&b
, NULL
);
1652 nir_store_deref(&b
, lhs_deref
, src
, ir
->write_mask
);
1657 * Given an instruction, returns a pointer to its destination or NULL if there
1658 * is no destination.
1660 * Note that this only handles instructions we generate at this level.
1663 get_instr_dest(nir_instr
*instr
)
1665 nir_alu_instr
*alu_instr
;
1666 nir_intrinsic_instr
*intrinsic_instr
;
1667 nir_tex_instr
*tex_instr
;
1669 switch (instr
->type
) {
1670 case nir_instr_type_alu
:
1671 alu_instr
= nir_instr_as_alu(instr
);
1672 return &alu_instr
->dest
.dest
;
1674 case nir_instr_type_intrinsic
:
1675 intrinsic_instr
= nir_instr_as_intrinsic(instr
);
1676 if (nir_intrinsic_infos
[intrinsic_instr
->intrinsic
].has_dest
)
1677 return &intrinsic_instr
->dest
;
1681 case nir_instr_type_tex
:
1682 tex_instr
= nir_instr_as_tex(instr
);
1683 return &tex_instr
->dest
;
1686 unreachable("not reached");
1693 nir_visitor::add_instr(nir_instr
*instr
, unsigned num_components
,
1696 nir_dest
*dest
= get_instr_dest(instr
);
1699 nir_ssa_dest_init(instr
, dest
, num_components
, bit_size
, NULL
);
1701 nir_builder_instr_insert(&b
, instr
);
1704 assert(dest
->is_ssa
);
1705 this->result
= &dest
->ssa
;
1710 nir_visitor::evaluate_rvalue(ir_rvalue
* ir
)
1713 if (ir
->as_dereference() || ir
->as_constant()) {
1715 * A dereference is being used on the right hand side, which means we
1716 * must emit a variable load.
1719 this->result
= nir_load_deref(&b
, this->deref
);
1722 return this->result
;
1726 type_is_float(glsl_base_type type
)
1728 return type
== GLSL_TYPE_FLOAT
|| type
== GLSL_TYPE_DOUBLE
||
1729 type
== GLSL_TYPE_FLOAT16
;
1733 type_is_signed(glsl_base_type type
)
1735 return type
== GLSL_TYPE_INT
|| type
== GLSL_TYPE_INT64
||
1736 type
== GLSL_TYPE_INT16
;
1740 type_is_int(glsl_base_type type
)
1742 return type
== GLSL_TYPE_UINT
|| type
== GLSL_TYPE_INT
||
1743 type
== GLSL_TYPE_UINT8
|| type
== GLSL_TYPE_INT8
||
1744 type
== GLSL_TYPE_UINT16
|| type
== GLSL_TYPE_INT16
||
1745 type
== GLSL_TYPE_UINT64
|| type
== GLSL_TYPE_INT64
;
1749 nir_visitor::visit(ir_expression
*ir
)
1751 /* Some special cases */
1752 switch (ir
->operation
) {
1753 case ir_binop_ubo_load
: {
1754 nir_intrinsic_instr
*load
=
1755 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_load_ubo
);
1756 unsigned bit_size
= ir
->type
->is_boolean() ? 32 :
1757 glsl_get_bit_size(ir
->type
);
1758 load
->num_components
= ir
->type
->vector_elements
;
1759 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
1760 load
->src
[1] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1761 intrinsic_set_std430_align(load
, ir
->type
);
1762 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
1765 * In UBO's, a true boolean value is any non-zero value, but we consider
1766 * a true boolean to be ~0. Fix this up with a != 0 comparison.
1769 if (ir
->type
->is_boolean())
1770 this->result
= nir_i2b(&b
, &load
->dest
.ssa
);
1775 case ir_unop_interpolate_at_centroid
:
1776 case ir_binop_interpolate_at_offset
:
1777 case ir_binop_interpolate_at_sample
: {
1778 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1779 ir_swizzle
*swizzle
= NULL
;
1781 /* the api does not allow a swizzle here, but the varying packing code
1782 * may have pushed one into here.
1784 swizzle
= ir
->operands
[0]->as_swizzle();
1786 deref
= swizzle
->val
->as_dereference();
1790 deref
->accept(this);
1792 nir_intrinsic_op op
;
1793 if (this->deref
->mode
== nir_var_shader_in
) {
1794 switch (ir
->operation
) {
1795 case ir_unop_interpolate_at_centroid
:
1796 op
= nir_intrinsic_interp_deref_at_centroid
;
1798 case ir_binop_interpolate_at_offset
:
1799 op
= nir_intrinsic_interp_deref_at_offset
;
1801 case ir_binop_interpolate_at_sample
:
1802 op
= nir_intrinsic_interp_deref_at_sample
;
1805 unreachable("Invalid interpolation intrinsic");
1808 /* This case can happen if the vertex shader does not write the
1809 * given varying. In this case, the linker will lower it to a
1810 * global variable. Since interpolating a variable makes no
1811 * sense, we'll just turn it into a load which will probably
1812 * eventually end up as an SSA definition.
1814 assert(this->deref
->mode
== nir_var_shader_temp
);
1815 op
= nir_intrinsic_load_deref
;
1818 nir_intrinsic_instr
*intrin
= nir_intrinsic_instr_create(shader
, op
);
1819 intrin
->num_components
= deref
->type
->vector_elements
;
1820 intrin
->src
[0] = nir_src_for_ssa(&this->deref
->dest
.ssa
);
1822 if (intrin
->intrinsic
== nir_intrinsic_interp_deref_at_offset
||
1823 intrin
->intrinsic
== nir_intrinsic_interp_deref_at_sample
)
1824 intrin
->src
[1] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[1]));
1826 unsigned bit_size
= glsl_get_bit_size(deref
->type
);
1827 add_instr(&intrin
->instr
, deref
->type
->vector_elements
, bit_size
);
1830 unsigned swiz
[4] = {
1831 swizzle
->mask
.x
, swizzle
->mask
.y
, swizzle
->mask
.z
, swizzle
->mask
.w
1834 result
= nir_swizzle(&b
, result
, swiz
,
1835 swizzle
->type
->vector_elements
, false);
1841 case ir_unop_ssbo_unsized_array_length
: {
1842 nir_intrinsic_instr
*intrin
=
1843 nir_intrinsic_instr_create(b
.shader
,
1844 nir_intrinsic_deref_buffer_array_length
);
1846 ir_dereference
*deref
= ir
->operands
[0]->as_dereference();
1847 intrin
->src
[0] = nir_src_for_ssa(&evaluate_deref(deref
)->dest
.ssa
);
1849 add_instr(&intrin
->instr
, 1, 32);
1857 nir_ssa_def
*srcs
[4];
1858 for (unsigned i
= 0; i
< ir
->num_operands
; i
++)
1859 srcs
[i
] = evaluate_rvalue(ir
->operands
[i
]);
1861 glsl_base_type types
[4];
1862 for (unsigned i
= 0; i
< ir
->num_operands
; i
++)
1863 if (supports_ints
|| !type_is_int(ir
->operands
[i
]->type
->base_type
))
1864 types
[i
] = ir
->operands
[i
]->type
->base_type
;
1866 types
[i
] = GLSL_TYPE_FLOAT
;
1868 glsl_base_type out_type
;
1869 if (supports_ints
|| !type_is_int(ir
->type
->base_type
))
1870 out_type
= ir
->type
->base_type
;
1872 out_type
= GLSL_TYPE_FLOAT
;
1874 switch (ir
->operation
) {
1875 case ir_unop_bit_not
: result
= nir_inot(&b
, srcs
[0]); break;
1876 case ir_unop_logic_not
:
1877 result
= nir_inot(&b
, srcs
[0]);
1880 result
= type_is_float(types
[0]) ? nir_fneg(&b
, srcs
[0])
1881 : nir_ineg(&b
, srcs
[0]);
1884 result
= type_is_float(types
[0]) ? nir_fabs(&b
, srcs
[0])
1885 : nir_iabs(&b
, srcs
[0]);
1887 case ir_unop_saturate
:
1888 assert(type_is_float(types
[0]));
1889 result
= nir_fsat(&b
, srcs
[0]);
1892 result
= type_is_float(types
[0]) ? nir_fsign(&b
, srcs
[0])
1893 : nir_isign(&b
, srcs
[0]);
1895 case ir_unop_rcp
: result
= nir_frcp(&b
, srcs
[0]); break;
1896 case ir_unop_rsq
: result
= nir_frsq(&b
, srcs
[0]); break;
1897 case ir_unop_sqrt
: result
= nir_fsqrt(&b
, srcs
[0]); break;
1898 case ir_unop_exp
: unreachable("ir_unop_exp should have been lowered");
1899 case ir_unop_log
: unreachable("ir_unop_log should have been lowered");
1900 case ir_unop_exp2
: result
= nir_fexp2(&b
, srcs
[0]); break;
1901 case ir_unop_log2
: result
= nir_flog2(&b
, srcs
[0]); break;
1903 result
= supports_ints
? nir_i2f32(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1906 result
= supports_ints
? nir_u2f32(&b
, srcs
[0]) : nir_fmov(&b
, srcs
[0]);
1909 result
= nir_b2f32(&b
, srcs
[0]);
1912 result
= supports_ints
? nir_f2i32(&b
, srcs
[0]) : nir_ftrunc(&b
, srcs
[0]);
1915 result
= supports_ints
? nir_f2u32(&b
, srcs
[0]) : nir_ftrunc(&b
, srcs
[0]);
1947 case ir_unop_i642u64
:
1948 case ir_unop_u642i64
: {
1949 nir_alu_type src_type
= nir_get_nir_type_for_glsl_base_type(types
[0]);
1950 nir_alu_type dst_type
= nir_get_nir_type_for_glsl_base_type(out_type
);
1951 result
= nir_build_alu(&b
, nir_type_conversion_op(src_type
, dst_type
,
1952 nir_rounding_mode_undef
),
1953 srcs
[0], NULL
, NULL
, NULL
);
1954 /* b2i and b2f don't have fixed bit-size versions so the builder will
1955 * just assume 32 and we have to fix it up here.
1957 result
->bit_size
= nir_alu_type_get_type_size(dst_type
);
1961 case ir_unop_bitcast_i2f
:
1962 case ir_unop_bitcast_f2i
:
1963 case ir_unop_bitcast_u2f
:
1964 case ir_unop_bitcast_f2u
:
1965 case ir_unop_bitcast_i642d
:
1966 case ir_unop_bitcast_d2i64
:
1967 case ir_unop_bitcast_u642d
:
1968 case ir_unop_bitcast_d2u64
:
1969 case ir_unop_subroutine_to_int
:
1971 result
= nir_imov(&b
, srcs
[0]);
1973 case ir_unop_trunc
: result
= nir_ftrunc(&b
, srcs
[0]); break;
1974 case ir_unop_ceil
: result
= nir_fceil(&b
, srcs
[0]); break;
1975 case ir_unop_floor
: result
= nir_ffloor(&b
, srcs
[0]); break;
1976 case ir_unop_fract
: result
= nir_ffract(&b
, srcs
[0]); break;
1977 case ir_unop_frexp_exp
: result
= nir_frexp_exp(&b
, srcs
[0]); break;
1978 case ir_unop_frexp_sig
: result
= nir_frexp_sig(&b
, srcs
[0]); break;
1979 case ir_unop_round_even
: result
= nir_fround_even(&b
, srcs
[0]); break;
1980 case ir_unop_sin
: result
= nir_fsin(&b
, srcs
[0]); break;
1981 case ir_unop_cos
: result
= nir_fcos(&b
, srcs
[0]); break;
1982 case ir_unop_dFdx
: result
= nir_fddx(&b
, srcs
[0]); break;
1983 case ir_unop_dFdy
: result
= nir_fddy(&b
, srcs
[0]); break;
1984 case ir_unop_dFdx_fine
: result
= nir_fddx_fine(&b
, srcs
[0]); break;
1985 case ir_unop_dFdy_fine
: result
= nir_fddy_fine(&b
, srcs
[0]); break;
1986 case ir_unop_dFdx_coarse
: result
= nir_fddx_coarse(&b
, srcs
[0]); break;
1987 case ir_unop_dFdy_coarse
: result
= nir_fddy_coarse(&b
, srcs
[0]); break;
1988 case ir_unop_pack_snorm_2x16
:
1989 result
= nir_pack_snorm_2x16(&b
, srcs
[0]);
1991 case ir_unop_pack_snorm_4x8
:
1992 result
= nir_pack_snorm_4x8(&b
, srcs
[0]);
1994 case ir_unop_pack_unorm_2x16
:
1995 result
= nir_pack_unorm_2x16(&b
, srcs
[0]);
1997 case ir_unop_pack_unorm_4x8
:
1998 result
= nir_pack_unorm_4x8(&b
, srcs
[0]);
2000 case ir_unop_pack_half_2x16
:
2001 result
= nir_pack_half_2x16(&b
, srcs
[0]);
2003 case ir_unop_unpack_snorm_2x16
:
2004 result
= nir_unpack_snorm_2x16(&b
, srcs
[0]);
2006 case ir_unop_unpack_snorm_4x8
:
2007 result
= nir_unpack_snorm_4x8(&b
, srcs
[0]);
2009 case ir_unop_unpack_unorm_2x16
:
2010 result
= nir_unpack_unorm_2x16(&b
, srcs
[0]);
2012 case ir_unop_unpack_unorm_4x8
:
2013 result
= nir_unpack_unorm_4x8(&b
, srcs
[0]);
2015 case ir_unop_unpack_half_2x16
:
2016 result
= nir_unpack_half_2x16(&b
, srcs
[0]);
2018 case ir_unop_pack_sampler_2x32
:
2019 case ir_unop_pack_image_2x32
:
2020 case ir_unop_pack_double_2x32
:
2021 case ir_unop_pack_int_2x32
:
2022 case ir_unop_pack_uint_2x32
:
2023 result
= nir_pack_64_2x32(&b
, srcs
[0]);
2025 case ir_unop_unpack_sampler_2x32
:
2026 case ir_unop_unpack_image_2x32
:
2027 case ir_unop_unpack_double_2x32
:
2028 case ir_unop_unpack_int_2x32
:
2029 case ir_unop_unpack_uint_2x32
:
2030 result
= nir_unpack_64_2x32(&b
, srcs
[0]);
2032 case ir_unop_bitfield_reverse
:
2033 result
= nir_bitfield_reverse(&b
, srcs
[0]);
2035 case ir_unop_bit_count
:
2036 result
= nir_bit_count(&b
, srcs
[0]);
2038 case ir_unop_find_msb
:
2040 case GLSL_TYPE_UINT
:
2041 result
= nir_ufind_msb(&b
, srcs
[0]);
2044 result
= nir_ifind_msb(&b
, srcs
[0]);
2047 unreachable("Invalid type for findMSB()");
2050 case ir_unop_find_lsb
:
2051 result
= nir_find_lsb(&b
, srcs
[0]);
2055 switch (ir
->type
->vector_elements
) {
2057 switch (ir
->operands
[0]->type
->vector_elements
) {
2058 case 1: result
= nir_fnoise1_1(&b
, srcs
[0]); break;
2059 case 2: result
= nir_fnoise1_2(&b
, srcs
[0]); break;
2060 case 3: result
= nir_fnoise1_3(&b
, srcs
[0]); break;
2061 case 4: result
= nir_fnoise1_4(&b
, srcs
[0]); break;
2062 default: unreachable("not reached");
2066 switch (ir
->operands
[0]->type
->vector_elements
) {
2067 case 1: result
= nir_fnoise2_1(&b
, srcs
[0]); break;
2068 case 2: result
= nir_fnoise2_2(&b
, srcs
[0]); break;
2069 case 3: result
= nir_fnoise2_3(&b
, srcs
[0]); break;
2070 case 4: result
= nir_fnoise2_4(&b
, srcs
[0]); break;
2071 default: unreachable("not reached");
2075 switch (ir
->operands
[0]->type
->vector_elements
) {
2076 case 1: result
= nir_fnoise3_1(&b
, srcs
[0]); break;
2077 case 2: result
= nir_fnoise3_2(&b
, srcs
[0]); break;
2078 case 3: result
= nir_fnoise3_3(&b
, srcs
[0]); break;
2079 case 4: result
= nir_fnoise3_4(&b
, srcs
[0]); break;
2080 default: unreachable("not reached");
2084 switch (ir
->operands
[0]->type
->vector_elements
) {
2085 case 1: result
= nir_fnoise4_1(&b
, srcs
[0]); break;
2086 case 2: result
= nir_fnoise4_2(&b
, srcs
[0]); break;
2087 case 3: result
= nir_fnoise4_3(&b
, srcs
[0]); break;
2088 case 4: result
= nir_fnoise4_4(&b
, srcs
[0]); break;
2089 default: unreachable("not reached");
2093 unreachable("not reached");
2096 case ir_unop_get_buffer_size
: {
2097 nir_intrinsic_instr
*load
= nir_intrinsic_instr_create(
2099 nir_intrinsic_get_buffer_size
);
2100 load
->num_components
= ir
->type
->vector_elements
;
2101 load
->src
[0] = nir_src_for_ssa(evaluate_rvalue(ir
->operands
[0]));
2102 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
2103 add_instr(&load
->instr
, ir
->type
->vector_elements
, bit_size
);
2108 result
= type_is_float(out_type
) ? nir_fadd(&b
, srcs
[0], srcs
[1])
2109 : nir_iadd(&b
, srcs
[0], srcs
[1]);
2112 result
= type_is_float(out_type
) ? nir_fsub(&b
, srcs
[0], srcs
[1])
2113 : nir_isub(&b
, srcs
[0], srcs
[1]);
2116 if (type_is_float(out_type
))
2117 result
= nir_fmul(&b
, srcs
[0], srcs
[1]);
2118 else if (out_type
== GLSL_TYPE_INT64
&&
2119 (ir
->operands
[0]->type
->base_type
== GLSL_TYPE_INT
||
2120 ir
->operands
[1]->type
->base_type
== GLSL_TYPE_INT
))
2121 result
= nir_imul_2x32_64(&b
, srcs
[0], srcs
[1]);
2122 else if (out_type
== GLSL_TYPE_UINT64
&&
2123 (ir
->operands
[0]->type
->base_type
== GLSL_TYPE_UINT
||
2124 ir
->operands
[1]->type
->base_type
== GLSL_TYPE_UINT
))
2125 result
= nir_umul_2x32_64(&b
, srcs
[0], srcs
[1]);
2127 result
= nir_imul(&b
, srcs
[0], srcs
[1]);
2130 if (type_is_float(out_type
))
2131 result
= nir_fdiv(&b
, srcs
[0], srcs
[1]);
2132 else if (type_is_signed(out_type
))
2133 result
= nir_idiv(&b
, srcs
[0], srcs
[1]);
2135 result
= nir_udiv(&b
, srcs
[0], srcs
[1]);
2138 result
= type_is_float(out_type
) ? nir_fmod(&b
, srcs
[0], srcs
[1])
2139 : nir_umod(&b
, srcs
[0], srcs
[1]);
2142 if (type_is_float(out_type
))
2143 result
= nir_fmin(&b
, srcs
[0], srcs
[1]);
2144 else if (type_is_signed(out_type
))
2145 result
= nir_imin(&b
, srcs
[0], srcs
[1]);
2147 result
= nir_umin(&b
, srcs
[0], srcs
[1]);
2150 if (type_is_float(out_type
))
2151 result
= nir_fmax(&b
, srcs
[0], srcs
[1]);
2152 else if (type_is_signed(out_type
))
2153 result
= nir_imax(&b
, srcs
[0], srcs
[1]);
2155 result
= nir_umax(&b
, srcs
[0], srcs
[1]);
2157 case ir_binop_pow
: result
= nir_fpow(&b
, srcs
[0], srcs
[1]); break;
2158 case ir_binop_bit_and
: result
= nir_iand(&b
, srcs
[0], srcs
[1]); break;
2159 case ir_binop_bit_or
: result
= nir_ior(&b
, srcs
[0], srcs
[1]); break;
2160 case ir_binop_bit_xor
: result
= nir_ixor(&b
, srcs
[0], srcs
[1]); break;
2161 case ir_binop_logic_and
:
2162 result
= nir_iand(&b
, srcs
[0], srcs
[1]);
2164 case ir_binop_logic_or
:
2165 result
= nir_ior(&b
, srcs
[0], srcs
[1]);
2167 case ir_binop_logic_xor
:
2168 result
= nir_ixor(&b
, srcs
[0], srcs
[1]);
2170 case ir_binop_lshift
: result
= nir_ishl(&b
, srcs
[0], srcs
[1]); break;
2171 case ir_binop_rshift
:
2172 result
= (type_is_signed(out_type
)) ? nir_ishr(&b
, srcs
[0], srcs
[1])
2173 : nir_ushr(&b
, srcs
[0], srcs
[1]);
2175 case ir_binop_imul_high
:
2176 result
= (out_type
== GLSL_TYPE_INT
) ? nir_imul_high(&b
, srcs
[0], srcs
[1])
2177 : nir_umul_high(&b
, srcs
[0], srcs
[1]);
2179 case ir_binop_carry
: result
= nir_uadd_carry(&b
, srcs
[0], srcs
[1]); break;
2180 case ir_binop_borrow
: result
= nir_usub_borrow(&b
, srcs
[0], srcs
[1]); break;
2182 if (type_is_float(types
[0]))
2183 result
= nir_flt(&b
, srcs
[0], srcs
[1]);
2184 else if (type_is_signed(types
[0]))
2185 result
= nir_ilt(&b
, srcs
[0], srcs
[1]);
2187 result
= nir_ult(&b
, srcs
[0], srcs
[1]);
2189 case ir_binop_gequal
:
2190 if (type_is_float(types
[0]))
2191 result
= nir_fge(&b
, srcs
[0], srcs
[1]);
2192 else if (type_is_signed(types
[0]))
2193 result
= nir_ige(&b
, srcs
[0], srcs
[1]);
2195 result
= nir_uge(&b
, srcs
[0], srcs
[1]);
2197 case ir_binop_equal
:
2198 if (type_is_float(types
[0]))
2199 result
= nir_feq(&b
, srcs
[0], srcs
[1]);
2201 result
= nir_ieq(&b
, srcs
[0], srcs
[1]);
2203 case ir_binop_nequal
:
2204 if (type_is_float(types
[0]))
2205 result
= nir_fne(&b
, srcs
[0], srcs
[1]);
2207 result
= nir_ine(&b
, srcs
[0], srcs
[1]);
2209 case ir_binop_all_equal
:
2210 if (type_is_float(types
[0])) {
2211 switch (ir
->operands
[0]->type
->vector_elements
) {
2212 case 1: result
= nir_feq(&b
, srcs
[0], srcs
[1]); break;
2213 case 2: result
= nir_ball_fequal2(&b
, srcs
[0], srcs
[1]); break;
2214 case 3: result
= nir_ball_fequal3(&b
, srcs
[0], srcs
[1]); break;
2215 case 4: result
= nir_ball_fequal4(&b
, srcs
[0], srcs
[1]); break;
2217 unreachable("not reached");
2220 switch (ir
->operands
[0]->type
->vector_elements
) {
2221 case 1: result
= nir_ieq(&b
, srcs
[0], srcs
[1]); break;
2222 case 2: result
= nir_ball_iequal2(&b
, srcs
[0], srcs
[1]); break;
2223 case 3: result
= nir_ball_iequal3(&b
, srcs
[0], srcs
[1]); break;
2224 case 4: result
= nir_ball_iequal4(&b
, srcs
[0], srcs
[1]); break;
2226 unreachable("not reached");
2230 case ir_binop_any_nequal
:
2231 if (type_is_float(types
[0])) {
2232 switch (ir
->operands
[0]->type
->vector_elements
) {
2233 case 1: result
= nir_fne(&b
, srcs
[0], srcs
[1]); break;
2234 case 2: result
= nir_bany_fnequal2(&b
, srcs
[0], srcs
[1]); break;
2235 case 3: result
= nir_bany_fnequal3(&b
, srcs
[0], srcs
[1]); break;
2236 case 4: result
= nir_bany_fnequal4(&b
, srcs
[0], srcs
[1]); break;
2238 unreachable("not reached");
2241 switch (ir
->operands
[0]->type
->vector_elements
) {
2242 case 1: result
= nir_ine(&b
, srcs
[0], srcs
[1]); break;
2243 case 2: result
= nir_bany_inequal2(&b
, srcs
[0], srcs
[1]); break;
2244 case 3: result
= nir_bany_inequal3(&b
, srcs
[0], srcs
[1]); break;
2245 case 4: result
= nir_bany_inequal4(&b
, srcs
[0], srcs
[1]); break;
2247 unreachable("not reached");
2252 switch (ir
->operands
[0]->type
->vector_elements
) {
2253 case 2: result
= nir_fdot2(&b
, srcs
[0], srcs
[1]); break;
2254 case 3: result
= nir_fdot3(&b
, srcs
[0], srcs
[1]); break;
2255 case 4: result
= nir_fdot4(&b
, srcs
[0], srcs
[1]); break;
2257 unreachable("not reached");
2260 case ir_binop_vector_extract
: {
2261 result
= nir_channel(&b
, srcs
[0], 0);
2262 for (unsigned i
= 1; i
< ir
->operands
[0]->type
->vector_elements
; i
++) {
2263 nir_ssa_def
*swizzled
= nir_channel(&b
, srcs
[0], i
);
2264 result
= nir_bcsel(&b
, nir_ieq(&b
, srcs
[1], nir_imm_int(&b
, i
)),
2270 case ir_binop_ldexp
: result
= nir_ldexp(&b
, srcs
[0], srcs
[1]); break;
2272 result
= nir_ffma(&b
, srcs
[0], srcs
[1], srcs
[2]);
2275 result
= nir_flrp(&b
, srcs
[0], srcs
[1], srcs
[2]);
2278 result
= nir_bcsel(&b
, srcs
[0], srcs
[1], srcs
[2]);
2280 case ir_triop_bitfield_extract
:
2281 result
= (out_type
== GLSL_TYPE_INT
) ?
2282 nir_ibitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]) :
2283 nir_ubitfield_extract(&b
, srcs
[0], srcs
[1], srcs
[2]);
2285 case ir_quadop_bitfield_insert
:
2286 result
= nir_bitfield_insert(&b
, srcs
[0], srcs
[1], srcs
[2], srcs
[3]);
2288 case ir_quadop_vector
:
2289 result
= nir_vec(&b
, srcs
, ir
->type
->vector_elements
);
2293 unreachable("not reached");
2298 nir_visitor::visit(ir_swizzle
*ir
)
2300 unsigned swizzle
[4] = { ir
->mask
.x
, ir
->mask
.y
, ir
->mask
.z
, ir
->mask
.w
};
2301 result
= nir_swizzle(&b
, evaluate_rvalue(ir
->val
), swizzle
,
2302 ir
->type
->vector_elements
, false);
2306 nir_visitor::visit(ir_texture
*ir
)
2313 num_srcs
= 1; /* coordinate */
2318 op
= (ir
->op
== ir_txb
) ? nir_texop_txb
: nir_texop_txl
;
2319 num_srcs
= 2; /* coordinate, bias/lod */
2323 op
= nir_texop_txd
; /* coordinate, dPdx, dPdy */
2329 if (ir
->lod_info
.lod
!= NULL
)
2330 num_srcs
= 2; /* coordinate, lod */
2332 num_srcs
= 1; /* coordinate */
2336 op
= nir_texop_txf_ms
;
2337 num_srcs
= 2; /* coordinate, sample_index */
2342 if (ir
->lod_info
.lod
!= NULL
)
2343 num_srcs
= 1; /* lod */
2350 num_srcs
= 1; /* coordinate */
2355 num_srcs
= 1; /* coordinate */
2358 case ir_query_levels
:
2359 op
= nir_texop_query_levels
;
2363 case ir_texture_samples
:
2364 op
= nir_texop_texture_samples
;
2368 case ir_samples_identical
:
2369 op
= nir_texop_samples_identical
;
2370 num_srcs
= 1; /* coordinate */
2374 unreachable("not reached");
2377 if (ir
->projector
!= NULL
)
2379 if (ir
->shadow_comparator
!= NULL
)
2381 /* offsets are constants we store inside nir_tex_intrs.offsets */
2382 if (ir
->offset
!= NULL
&& !ir
->offset
->type
->is_array())
2385 /* Add one for the texture deref */
2388 nir_tex_instr
*instr
= nir_tex_instr_create(this->shader
, num_srcs
);
2391 instr
->sampler_dim
=
2392 (glsl_sampler_dim
) ir
->sampler
->type
->sampler_dimensionality
;
2393 instr
->is_array
= ir
->sampler
->type
->sampler_array
;
2394 instr
->is_shadow
= ir
->sampler
->type
->sampler_shadow
;
2395 if (instr
->is_shadow
)
2396 instr
->is_new_style_shadow
= (ir
->type
->vector_elements
== 1);
2397 switch (ir
->type
->base_type
) {
2398 case GLSL_TYPE_FLOAT
:
2399 instr
->dest_type
= nir_type_float
;
2402 instr
->dest_type
= nir_type_int
;
2404 case GLSL_TYPE_BOOL
:
2405 case GLSL_TYPE_UINT
:
2406 instr
->dest_type
= nir_type_uint
;
2409 unreachable("not reached");
2412 nir_deref_instr
*sampler_deref
= evaluate_deref(ir
->sampler
);
2413 instr
->src
[0].src
= nir_src_for_ssa(&sampler_deref
->dest
.ssa
);
2414 instr
->src
[0].src_type
= nir_tex_src_texture_deref
;
2415 instr
->src
[1].src
= nir_src_for_ssa(&sampler_deref
->dest
.ssa
);
2416 instr
->src
[1].src_type
= nir_tex_src_sampler_deref
;
2418 unsigned src_number
= 2;
2420 if (ir
->coordinate
!= NULL
) {
2421 instr
->coord_components
= ir
->coordinate
->type
->vector_elements
;
2422 instr
->src
[src_number
].src
=
2423 nir_src_for_ssa(evaluate_rvalue(ir
->coordinate
));
2424 instr
->src
[src_number
].src_type
= nir_tex_src_coord
;
2428 if (ir
->projector
!= NULL
) {
2429 instr
->src
[src_number
].src
=
2430 nir_src_for_ssa(evaluate_rvalue(ir
->projector
));
2431 instr
->src
[src_number
].src_type
= nir_tex_src_projector
;
2435 if (ir
->shadow_comparator
!= NULL
) {
2436 instr
->src
[src_number
].src
=
2437 nir_src_for_ssa(evaluate_rvalue(ir
->shadow_comparator
));
2438 instr
->src
[src_number
].src_type
= nir_tex_src_comparator
;
2442 if (ir
->offset
!= NULL
) {
2443 if (ir
->offset
->type
->is_array()) {
2444 for (int i
= 0; i
< ir
->offset
->type
->array_size(); i
++) {
2445 const ir_constant
*c
=
2446 ir
->offset
->as_constant()->get_array_element(i
);
2448 for (unsigned j
= 0; j
< 2; ++j
) {
2449 int val
= c
->get_int_component(j
);
2450 assert(val
<= 31 && val
>= -32);
2451 instr
->tg4_offsets
[i
][j
] = val
;
2455 assert(ir
->offset
->type
->is_vector() || ir
->offset
->type
->is_scalar());
2457 instr
->src
[src_number
].src
=
2458 nir_src_for_ssa(evaluate_rvalue(ir
->offset
));
2459 instr
->src
[src_number
].src_type
= nir_tex_src_offset
;
2466 instr
->src
[src_number
].src
=
2467 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.bias
));
2468 instr
->src
[src_number
].src_type
= nir_tex_src_bias
;
2475 if (ir
->lod_info
.lod
!= NULL
) {
2476 instr
->src
[src_number
].src
=
2477 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.lod
));
2478 instr
->src
[src_number
].src_type
= nir_tex_src_lod
;
2484 instr
->src
[src_number
].src
=
2485 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdx
));
2486 instr
->src
[src_number
].src_type
= nir_tex_src_ddx
;
2488 instr
->src
[src_number
].src
=
2489 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.grad
.dPdy
));
2490 instr
->src
[src_number
].src_type
= nir_tex_src_ddy
;
2495 instr
->src
[src_number
].src
=
2496 nir_src_for_ssa(evaluate_rvalue(ir
->lod_info
.sample_index
));
2497 instr
->src
[src_number
].src_type
= nir_tex_src_ms_index
;
2502 instr
->component
= ir
->lod_info
.component
->as_constant()->value
.u
[0];
2509 assert(src_number
== num_srcs
);
2511 unsigned bit_size
= glsl_get_bit_size(ir
->type
);
2512 add_instr(&instr
->instr
, nir_tex_instr_dest_size(instr
), bit_size
);
2516 nir_visitor::visit(ir_constant
*ir
)
2519 * We don't know if this variable is an array or struct that gets
2520 * dereferenced, so do the safe thing an make it a variable with a
2521 * constant initializer and return a dereference.
2525 nir_local_variable_create(this->impl
, ir
->type
, "const_temp");
2526 var
->data
.read_only
= true;
2527 var
->constant_initializer
= constant_copy(ir
, var
);
2529 this->deref
= nir_build_deref_var(&b
, var
);
2533 nir_visitor::visit(ir_dereference_variable
*ir
)
2535 if (ir
->variable_referenced()->data
.mode
== ir_var_function_out
) {
2536 unsigned i
= (sig
->return_type
!= glsl_type::void_type
) ? 1 : 0;
2538 foreach_in_list(ir_variable
, param
, &sig
->parameters
) {
2539 if (param
== ir
->variable_referenced()) {
2545 this->deref
= nir_build_deref_cast(&b
, nir_load_param(&b
, i
),
2546 nir_var_function_temp
, ir
->type
, 0);
2550 assert(ir
->variable_referenced()->data
.mode
!= ir_var_function_inout
);
2552 struct hash_entry
*entry
=
2553 _mesa_hash_table_search(this->var_table
, ir
->var
);
2555 nir_variable
*var
= (nir_variable
*) entry
->data
;
2557 this->deref
= nir_build_deref_var(&b
, var
);
2561 nir_visitor::visit(ir_dereference_record
*ir
)
2563 ir
->record
->accept(this);
2565 int field_index
= ir
->field_idx
;
2566 assert(field_index
>= 0);
2568 this->deref
= nir_build_deref_struct(&b
, this->deref
, field_index
);
2572 nir_visitor::visit(ir_dereference_array
*ir
)
2574 nir_ssa_def
*index
= evaluate_rvalue(ir
->array_index
);
2576 ir
->array
->accept(this);
2578 this->deref
= nir_build_deref_array(&b
, this->deref
, index
);
2582 nir_visitor::visit(ir_barrier
*)
2584 nir_intrinsic_instr
*instr
=
2585 nir_intrinsic_instr_create(this->shader
, nir_intrinsic_barrier
);
2586 nir_builder_instr_insert(&b
, &instr
->instr
);
2590 glsl_float64_funcs_to_nir(struct gl_context
*ctx
,
2591 const nir_shader_compiler_options
*options
)
2593 /* We pretend it's a vertex shader. Ultimately, the stage shouldn't
2594 * matter because we're not optimizing anything here.
2596 struct gl_shader
*sh
= _mesa_new_shader(-1, MESA_SHADER_VERTEX
);
2597 sh
->Source
= float64_source
;
2598 sh
->CompileStatus
= COMPILE_FAILURE
;
2599 _mesa_glsl_compile_shader(ctx
, sh
, false, false, true);
2601 if (!sh
->CompileStatus
) {
2604 "fp64 software impl compile failed:\n%s\nsource:\n%s\n",
2605 sh
->InfoLog
, float64_source
);
2610 nir_shader
*nir
= nir_shader_create(NULL
, MESA_SHADER_VERTEX
, options
, NULL
);
2612 nir_visitor
v1(ctx
, nir
);
2613 nir_function_visitor
v2(&v1
);
2615 visit_exec_list(sh
->ir
, &v1
);
2617 /* _mesa_delete_shader will try to free sh->Source but it's static const */
2619 _mesa_delete_shader(ctx
, sh
);
2621 nir_validate_shader(nir
, "float64_funcs_to_nir");
2623 NIR_PASS_V(nir
, nir_lower_constant_initializers
, nir_var_function_temp
);
2624 NIR_PASS_V(nir
, nir_lower_returns
);
2625 NIR_PASS_V(nir
, nir_inline_functions
);
2626 NIR_PASS_V(nir
, nir_opt_deref
);