2 * Copyright (C) 2005-2007 Brian Paul All Rights Reserved.
3 * Copyright (C) 2008 VMware, Inc. All Rights Reserved.
4 * Copyright © 2010 Intel Corporation
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
22 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
23 * DEALINGS IN THE SOFTWARE.
27 * \file ir_to_mesa.cpp
29 * Translate GLSL IR to Mesa's gl_program representation.
33 #include "main/compiler.h"
35 #include "ir_visitor.h"
36 #include "ir_print_visitor.h"
37 #include "ir_expression_flattening.h"
38 #include "glsl_types.h"
39 #include "glsl_parser_extras.h"
40 #include "../glsl/program.h"
41 #include "ir_optimization.h"
45 #include "main/mtypes.h"
46 #include "main/shaderapi.h"
47 #include "main/shaderobj.h"
48 #include "main/uniforms.h"
49 #include "program/hash_table.h"
50 #include "program/prog_instruction.h"
51 #include "program/prog_optimize.h"
52 #include "program/prog_print.h"
53 #include "program/program.h"
54 #include "program/prog_uniform.h"
55 #include "program/prog_parameter.h"
56 #include "program/sampler.h"
59 static int swizzle_for_size(int size
);
62 * This struct is a corresponding struct to Mesa prog_src_register, with
65 typedef struct ir_to_mesa_src_reg
{
66 ir_to_mesa_src_reg(int file
, int index
, const glsl_type
*type
)
70 if (type
&& (type
->is_scalar() || type
->is_vector() || type
->is_matrix()))
71 this->swizzle
= swizzle_for_size(type
->vector_elements
);
73 this->swizzle
= SWIZZLE_XYZW
;
80 this->file
= PROGRAM_UNDEFINED
;
87 int file
; /**< PROGRAM_* from Mesa */
88 int index
; /**< temporary index, VERT_ATTRIB_*, FRAG_ATTRIB_*, etc. */
89 GLuint swizzle
; /**< SWIZZLE_XYZWONEZERO swizzles from Mesa. */
90 int negate
; /**< NEGATE_XYZW mask from mesa */
91 /** Register index should be offset by the integer in this reg. */
92 ir_to_mesa_src_reg
*reladdr
;
95 typedef struct ir_to_mesa_dst_reg
{
96 int file
; /**< PROGRAM_* from Mesa */
97 int index
; /**< temporary index, VERT_ATTRIB_*, FRAG_ATTRIB_*, etc. */
98 int writemask
; /**< Bitfield of WRITEMASK_[XYZW] */
100 /** Register index should be offset by the integer in this reg. */
101 ir_to_mesa_src_reg
*reladdr
;
102 } ir_to_mesa_dst_reg
;
104 extern ir_to_mesa_src_reg ir_to_mesa_undef
;
106 class ir_to_mesa_instruction
: public exec_node
{
108 /* Callers of this talloc-based new need not call delete. It's
109 * easier to just talloc_free 'ctx' (or any of its ancestors). */
110 static void* operator new(size_t size
, void *ctx
)
114 node
= talloc_zero_size(ctx
, size
);
115 assert(node
!= NULL
);
121 ir_to_mesa_dst_reg dst_reg
;
122 ir_to_mesa_src_reg src_reg
[3];
123 /** Pointer to the ir source this tree came from for debugging */
125 GLboolean cond_update
;
126 int sampler
; /**< sampler index */
127 int tex_target
; /**< One of TEXTURE_*_INDEX */
128 GLboolean tex_shadow
;
130 class function_entry
*function
; /* Set on OPCODE_CAL or OPCODE_BGNSUB */
133 class variable_storage
: public exec_node
{
135 variable_storage(ir_variable
*var
, int file
, int index
)
136 : file(file
), index(index
), var(var
)
143 ir_variable
*var
; /* variable that maps to this, if any */
146 class function_entry
: public exec_node
{
148 ir_function_signature
*sig
;
151 * identifier of this function signature used by the program.
153 * At the point that Mesa instructions for function calls are
154 * generated, we don't know the address of the first instruction of
155 * the function body. So we make the BranchTarget that is called a
156 * small integer and rewrite them during set_branchtargets().
161 * Pointer to first instruction of the function body.
163 * Set during function body emits after main() is processed.
165 ir_to_mesa_instruction
*bgn_inst
;
168 * Index of the first instruction of the function body in actual
171 * Set after convertion from ir_to_mesa_instruction to prog_instruction.
175 /** Storage for the return value. */
176 ir_to_mesa_src_reg return_reg
;
179 class ir_to_mesa_visitor
: public ir_visitor
{
181 ir_to_mesa_visitor();
182 ~ir_to_mesa_visitor();
184 function_entry
*current_function
;
186 struct gl_context
*ctx
;
187 struct gl_program
*prog
;
188 struct gl_shader_program
*shader_program
;
189 struct gl_shader_compiler_options
*options
;
193 variable_storage
*find_variable_storage(ir_variable
*var
);
195 function_entry
*get_function_signature(ir_function_signature
*sig
);
197 ir_to_mesa_src_reg
get_temp(const glsl_type
*type
);
198 void reladdr_to_temp(ir_instruction
*ir
,
199 ir_to_mesa_src_reg
*reg
, int *num_reladdr
);
201 struct ir_to_mesa_src_reg
src_reg_for_float(float val
);
204 * \name Visit methods
206 * As typical for the visitor pattern, there must be one \c visit method for
207 * each concrete subclass of \c ir_instruction. Virtual base classes within
208 * the hierarchy should not have \c visit methods.
211 virtual void visit(ir_variable
*);
212 virtual void visit(ir_loop
*);
213 virtual void visit(ir_loop_jump
*);
214 virtual void visit(ir_function_signature
*);
215 virtual void visit(ir_function
*);
216 virtual void visit(ir_expression
*);
217 virtual void visit(ir_swizzle
*);
218 virtual void visit(ir_dereference_variable
*);
219 virtual void visit(ir_dereference_array
*);
220 virtual void visit(ir_dereference_record
*);
221 virtual void visit(ir_assignment
*);
222 virtual void visit(ir_constant
*);
223 virtual void visit(ir_call
*);
224 virtual void visit(ir_return
*);
225 virtual void visit(ir_discard
*);
226 virtual void visit(ir_texture
*);
227 virtual void visit(ir_if
*);
230 struct ir_to_mesa_src_reg result
;
232 /** List of variable_storage */
235 /** List of function_entry */
236 exec_list function_signatures
;
237 int next_signature_id
;
239 /** List of ir_to_mesa_instruction */
240 exec_list instructions
;
242 ir_to_mesa_instruction
*ir_to_mesa_emit_op0(ir_instruction
*ir
,
243 enum prog_opcode op
);
245 ir_to_mesa_instruction
*ir_to_mesa_emit_op1(ir_instruction
*ir
,
247 ir_to_mesa_dst_reg dst
,
248 ir_to_mesa_src_reg src0
);
250 ir_to_mesa_instruction
*ir_to_mesa_emit_op2(ir_instruction
*ir
,
252 ir_to_mesa_dst_reg dst
,
253 ir_to_mesa_src_reg src0
,
254 ir_to_mesa_src_reg src1
);
256 ir_to_mesa_instruction
*ir_to_mesa_emit_op3(ir_instruction
*ir
,
258 ir_to_mesa_dst_reg dst
,
259 ir_to_mesa_src_reg src0
,
260 ir_to_mesa_src_reg src1
,
261 ir_to_mesa_src_reg src2
);
263 void ir_to_mesa_emit_scalar_op1(ir_instruction
*ir
,
265 ir_to_mesa_dst_reg dst
,
266 ir_to_mesa_src_reg src0
);
268 void ir_to_mesa_emit_scalar_op2(ir_instruction
*ir
,
270 ir_to_mesa_dst_reg dst
,
271 ir_to_mesa_src_reg src0
,
272 ir_to_mesa_src_reg src1
);
274 GLboolean
try_emit_mad(ir_expression
*ir
,
280 ir_to_mesa_src_reg ir_to_mesa_undef
= ir_to_mesa_src_reg(PROGRAM_UNDEFINED
, 0, NULL
);
282 ir_to_mesa_dst_reg ir_to_mesa_undef_dst
= {
283 PROGRAM_UNDEFINED
, 0, SWIZZLE_NOOP
, COND_TR
, NULL
,
286 ir_to_mesa_dst_reg ir_to_mesa_address_reg
= {
287 PROGRAM_ADDRESS
, 0, WRITEMASK_X
, COND_TR
, NULL
291 fail_link(struct gl_shader_program
*prog
, const char *fmt
, ...) PRINTFLIKE(2, 3);
294 fail_link(struct gl_shader_program
*prog
, const char *fmt
, ...)
298 prog
->InfoLog
= talloc_vasprintf_append(prog
->InfoLog
, fmt
, args
);
301 prog
->LinkStatus
= GL_FALSE
;
305 swizzle_for_size(int size
)
307 int size_swizzles
[4] = {
308 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_X
),
309 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_Y
, SWIZZLE_Y
, SWIZZLE_Y
),
310 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_Y
, SWIZZLE_Z
, SWIZZLE_Z
),
311 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_Y
, SWIZZLE_Z
, SWIZZLE_W
),
314 assert((size
>= 1) && (size
<= 4));
315 return size_swizzles
[size
- 1];
318 ir_to_mesa_instruction
*
319 ir_to_mesa_visitor::ir_to_mesa_emit_op3(ir_instruction
*ir
,
321 ir_to_mesa_dst_reg dst
,
322 ir_to_mesa_src_reg src0
,
323 ir_to_mesa_src_reg src1
,
324 ir_to_mesa_src_reg src2
)
326 ir_to_mesa_instruction
*inst
= new(mem_ctx
) ir_to_mesa_instruction();
329 /* If we have to do relative addressing, we want to load the ARL
330 * reg directly for one of the regs, and preload the other reladdr
331 * sources into temps.
333 num_reladdr
+= dst
.reladdr
!= NULL
;
334 num_reladdr
+= src0
.reladdr
!= NULL
;
335 num_reladdr
+= src1
.reladdr
!= NULL
;
336 num_reladdr
+= src2
.reladdr
!= NULL
;
338 reladdr_to_temp(ir
, &src2
, &num_reladdr
);
339 reladdr_to_temp(ir
, &src1
, &num_reladdr
);
340 reladdr_to_temp(ir
, &src0
, &num_reladdr
);
343 ir_to_mesa_emit_op1(ir
, OPCODE_ARL
, ir_to_mesa_address_reg
,
348 assert(num_reladdr
== 0);
352 inst
->src_reg
[0] = src0
;
353 inst
->src_reg
[1] = src1
;
354 inst
->src_reg
[2] = src2
;
357 inst
->function
= NULL
;
359 this->instructions
.push_tail(inst
);
365 ir_to_mesa_instruction
*
366 ir_to_mesa_visitor::ir_to_mesa_emit_op2(ir_instruction
*ir
,
368 ir_to_mesa_dst_reg dst
,
369 ir_to_mesa_src_reg src0
,
370 ir_to_mesa_src_reg src1
)
372 return ir_to_mesa_emit_op3(ir
, op
, dst
, src0
, src1
, ir_to_mesa_undef
);
375 ir_to_mesa_instruction
*
376 ir_to_mesa_visitor::ir_to_mesa_emit_op1(ir_instruction
*ir
,
378 ir_to_mesa_dst_reg dst
,
379 ir_to_mesa_src_reg src0
)
381 assert(dst
.writemask
!= 0);
382 return ir_to_mesa_emit_op3(ir
, op
, dst
,
383 src0
, ir_to_mesa_undef
, ir_to_mesa_undef
);
386 ir_to_mesa_instruction
*
387 ir_to_mesa_visitor::ir_to_mesa_emit_op0(ir_instruction
*ir
,
390 return ir_to_mesa_emit_op3(ir
, op
, ir_to_mesa_undef_dst
,
396 inline ir_to_mesa_dst_reg
397 ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg reg
)
399 ir_to_mesa_dst_reg dst_reg
;
401 dst_reg
.file
= reg
.file
;
402 dst_reg
.index
= reg
.index
;
403 dst_reg
.writemask
= WRITEMASK_XYZW
;
404 dst_reg
.cond_mask
= COND_TR
;
405 dst_reg
.reladdr
= reg
.reladdr
;
410 inline ir_to_mesa_src_reg
411 ir_to_mesa_src_reg_from_dst(ir_to_mesa_dst_reg reg
)
413 return ir_to_mesa_src_reg(reg
.file
, reg
.index
, NULL
);
417 * Emits Mesa scalar opcodes to produce unique answers across channels.
419 * Some Mesa opcodes are scalar-only, like ARB_fp/vp. The src X
420 * channel determines the result across all channels. So to do a vec4
421 * of this operation, we want to emit a scalar per source channel used
422 * to produce dest channels.
425 ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op2(ir_instruction
*ir
,
427 ir_to_mesa_dst_reg dst
,
428 ir_to_mesa_src_reg orig_src0
,
429 ir_to_mesa_src_reg orig_src1
)
432 int done_mask
= ~dst
.writemask
;
434 /* Mesa RCP is a scalar operation splatting results to all channels,
435 * like ARB_fp/vp. So emit as many RCPs as necessary to cover our
438 for (i
= 0; i
< 4; i
++) {
439 GLuint this_mask
= (1 << i
);
440 ir_to_mesa_instruction
*inst
;
441 ir_to_mesa_src_reg src0
= orig_src0
;
442 ir_to_mesa_src_reg src1
= orig_src1
;
444 if (done_mask
& this_mask
)
447 GLuint src0_swiz
= GET_SWZ(src0
.swizzle
, i
);
448 GLuint src1_swiz
= GET_SWZ(src1
.swizzle
, i
);
449 for (j
= i
+ 1; j
< 4; j
++) {
450 if (!(done_mask
& (1 << j
)) &&
451 GET_SWZ(src0
.swizzle
, j
) == src0_swiz
&&
452 GET_SWZ(src1
.swizzle
, j
) == src1_swiz
) {
453 this_mask
|= (1 << j
);
456 src0
.swizzle
= MAKE_SWIZZLE4(src0_swiz
, src0_swiz
,
457 src0_swiz
, src0_swiz
);
458 src1
.swizzle
= MAKE_SWIZZLE4(src1_swiz
, src1_swiz
,
459 src1_swiz
, src1_swiz
);
461 inst
= ir_to_mesa_emit_op2(ir
, op
,
465 inst
->dst_reg
.writemask
= this_mask
;
466 done_mask
|= this_mask
;
471 ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op1(ir_instruction
*ir
,
473 ir_to_mesa_dst_reg dst
,
474 ir_to_mesa_src_reg src0
)
476 ir_to_mesa_src_reg undef
= ir_to_mesa_undef
;
478 undef
.swizzle
= SWIZZLE_XXXX
;
480 ir_to_mesa_emit_scalar_op2(ir
, op
, dst
, src0
, undef
);
483 struct ir_to_mesa_src_reg
484 ir_to_mesa_visitor::src_reg_for_float(float val
)
486 ir_to_mesa_src_reg
src_reg(PROGRAM_CONSTANT
, -1, NULL
);
488 src_reg
.index
= _mesa_add_unnamed_constant(this->prog
->Parameters
,
489 &val
, 1, &src_reg
.swizzle
);
495 type_size(const struct glsl_type
*type
)
500 switch (type
->base_type
) {
503 case GLSL_TYPE_FLOAT
:
505 if (type
->is_matrix()) {
506 return type
->matrix_columns
;
508 /* Regardless of size of vector, it gets a vec4. This is bad
509 * packing for things like floats, but otherwise arrays become a
510 * mess. Hopefully a later pass over the code can pack scalars
511 * down if appropriate.
515 case GLSL_TYPE_ARRAY
:
516 return type_size(type
->fields
.array
) * type
->length
;
517 case GLSL_TYPE_STRUCT
:
519 for (i
= 0; i
< type
->length
; i
++) {
520 size
+= type_size(type
->fields
.structure
[i
].type
);
523 case GLSL_TYPE_SAMPLER
:
524 /* Samplers take up one slot in UNIFORMS[], but they're baked in
535 * In the initial pass of codegen, we assign temporary numbers to
536 * intermediate results. (not SSA -- variable assignments will reuse
537 * storage). Actual register allocation for the Mesa VM occurs in a
538 * pass over the Mesa IR later.
541 ir_to_mesa_visitor::get_temp(const glsl_type
*type
)
543 ir_to_mesa_src_reg src_reg
;
547 src_reg
.file
= PROGRAM_TEMPORARY
;
548 src_reg
.index
= next_temp
;
549 src_reg
.reladdr
= NULL
;
550 next_temp
+= type_size(type
);
552 if (type
->is_array() || type
->is_record()) {
553 src_reg
.swizzle
= SWIZZLE_NOOP
;
555 for (i
= 0; i
< type
->vector_elements
; i
++)
558 swizzle
[i
] = type
->vector_elements
- 1;
559 src_reg
.swizzle
= MAKE_SWIZZLE4(swizzle
[0], swizzle
[1],
560 swizzle
[2], swizzle
[3]);
568 ir_to_mesa_visitor::find_variable_storage(ir_variable
*var
)
571 variable_storage
*entry
;
573 foreach_iter(exec_list_iterator
, iter
, this->variables
) {
574 entry
= (variable_storage
*)iter
.get();
576 if (entry
->var
== var
)
584 ir_to_mesa_visitor::visit(ir_variable
*ir
)
586 if (strcmp(ir
->name
, "gl_FragCoord") == 0) {
587 struct gl_fragment_program
*fp
= (struct gl_fragment_program
*)this->prog
;
589 fp
->OriginUpperLeft
= ir
->origin_upper_left
;
590 fp
->PixelCenterInteger
= ir
->pixel_center_integer
;
593 if (ir
->mode
== ir_var_uniform
&& strncmp(ir
->name
, "gl_", 3) == 0) {
595 const struct gl_builtin_uniform_desc
*statevar
;
597 for (i
= 0; _mesa_builtin_uniform_desc
[i
].name
; i
++) {
598 if (strcmp(ir
->name
, _mesa_builtin_uniform_desc
[i
].name
) == 0)
602 if (!_mesa_builtin_uniform_desc
[i
].name
) {
603 fail_link(this->shader_program
,
604 "Failed to find builtin uniform `%s'\n", ir
->name
);
608 statevar
= &_mesa_builtin_uniform_desc
[i
];
611 if (ir
->type
->is_array()) {
612 array_count
= ir
->type
->length
;
617 /* Check if this statevar's setup in the STATE file exactly
618 * matches how we'll want to reference it as a
619 * struct/array/whatever. If not, then we need to move it into
620 * temporary storage and hope that it'll get copy-propagated
623 for (i
= 0; i
< statevar
->num_elements
; i
++) {
624 if (statevar
->elements
[i
].swizzle
!= SWIZZLE_XYZW
) {
629 struct variable_storage
*storage
;
630 ir_to_mesa_dst_reg dst
;
631 if (i
== statevar
->num_elements
) {
632 /* We'll set the index later. */
633 storage
= new(mem_ctx
) variable_storage(ir
, PROGRAM_STATE_VAR
, -1);
634 this->variables
.push_tail(storage
);
636 dst
= ir_to_mesa_undef_dst
;
638 storage
= new(mem_ctx
) variable_storage(ir
, PROGRAM_TEMPORARY
,
640 this->variables
.push_tail(storage
);
641 this->next_temp
+= type_size(ir
->type
);
643 dst
= ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg(PROGRAM_TEMPORARY
,
649 for (int a
= 0; a
< array_count
; a
++) {
650 for (unsigned int i
= 0; i
< statevar
->num_elements
; i
++) {
651 struct gl_builtin_uniform_element
*element
= &statevar
->elements
[i
];
652 int tokens
[STATE_LENGTH
];
654 memcpy(tokens
, element
->tokens
, sizeof(element
->tokens
));
655 if (ir
->type
->is_array()) {
659 int index
= _mesa_add_state_reference(this->prog
->Parameters
,
660 (gl_state_index
*)tokens
);
662 if (storage
->file
== PROGRAM_STATE_VAR
) {
663 if (storage
->index
== -1) {
664 storage
->index
= index
;
667 (int)(storage
->index
+ a
* statevar
->num_elements
+ i
));
670 ir_to_mesa_src_reg
src(PROGRAM_STATE_VAR
, index
, NULL
);
671 src
.swizzle
= element
->swizzle
;
672 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, dst
, src
);
673 /* even a float takes up a whole vec4 reg in a struct/array. */
678 if (storage
->file
== PROGRAM_TEMPORARY
&&
679 dst
.index
!= storage
->index
+ type_size(ir
->type
)) {
680 fail_link(this->shader_program
,
681 "failed to load builtin uniform `%s' (%d/%d regs loaded)\n",
682 ir
->name
, dst
.index
- storage
->index
,
683 type_size(ir
->type
));
689 ir_to_mesa_visitor::visit(ir_loop
*ir
)
691 ir_dereference_variable
*counter
= NULL
;
693 if (ir
->counter
!= NULL
)
694 counter
= new(ir
) ir_dereference_variable(ir
->counter
);
696 if (ir
->from
!= NULL
) {
697 assert(ir
->counter
!= NULL
);
699 ir_assignment
*a
= new(ir
) ir_assignment(counter
, ir
->from
, NULL
);
705 ir_to_mesa_emit_op0(NULL
, OPCODE_BGNLOOP
);
709 new(ir
) ir_expression(ir
->cmp
, glsl_type::bool_type
,
711 ir_if
*if_stmt
= new(ir
) ir_if(e
);
713 ir_loop_jump
*brk
= new(ir
) ir_loop_jump(ir_loop_jump::jump_break
);
715 if_stmt
->then_instructions
.push_tail(brk
);
717 if_stmt
->accept(this);
724 visit_exec_list(&ir
->body_instructions
, this);
728 new(ir
) ir_expression(ir_binop_add
, counter
->type
,
729 counter
, ir
->increment
);
731 ir_assignment
*a
= new(ir
) ir_assignment(counter
, e
, NULL
);
738 ir_to_mesa_emit_op0(NULL
, OPCODE_ENDLOOP
);
742 ir_to_mesa_visitor::visit(ir_loop_jump
*ir
)
745 case ir_loop_jump::jump_break
:
746 ir_to_mesa_emit_op0(NULL
, OPCODE_BRK
);
748 case ir_loop_jump::jump_continue
:
749 ir_to_mesa_emit_op0(NULL
, OPCODE_CONT
);
756 ir_to_mesa_visitor::visit(ir_function_signature
*ir
)
763 ir_to_mesa_visitor::visit(ir_function
*ir
)
765 /* Ignore function bodies other than main() -- we shouldn't see calls to
766 * them since they should all be inlined before we get to ir_to_mesa.
768 if (strcmp(ir
->name
, "main") == 0) {
769 const ir_function_signature
*sig
;
772 sig
= ir
->matching_signature(&empty
);
776 foreach_iter(exec_list_iterator
, iter
, sig
->body
) {
777 ir_instruction
*ir
= (ir_instruction
*)iter
.get();
785 ir_to_mesa_visitor::try_emit_mad(ir_expression
*ir
, int mul_operand
)
787 int nonmul_operand
= 1 - mul_operand
;
788 ir_to_mesa_src_reg a
, b
, c
;
790 ir_expression
*expr
= ir
->operands
[mul_operand
]->as_expression();
791 if (!expr
|| expr
->operation
!= ir_binop_mul
)
794 expr
->operands
[0]->accept(this);
796 expr
->operands
[1]->accept(this);
798 ir
->operands
[nonmul_operand
]->accept(this);
801 this->result
= get_temp(ir
->type
);
802 ir_to_mesa_emit_op3(ir
, OPCODE_MAD
,
803 ir_to_mesa_dst_reg_from_src(this->result
), a
, b
, c
);
809 ir_to_mesa_visitor::reladdr_to_temp(ir_instruction
*ir
,
810 ir_to_mesa_src_reg
*reg
, int *num_reladdr
)
815 ir_to_mesa_emit_op1(ir
, OPCODE_ARL
, ir_to_mesa_address_reg
, *reg
->reladdr
);
817 if (*num_reladdr
!= 1) {
818 ir_to_mesa_src_reg temp
= get_temp(glsl_type::vec4_type
);
820 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
,
821 ir_to_mesa_dst_reg_from_src(temp
), *reg
);
829 ir_to_mesa_visitor::visit(ir_expression
*ir
)
831 unsigned int operand
;
832 struct ir_to_mesa_src_reg op
[2];
833 struct ir_to_mesa_src_reg result_src
;
834 struct ir_to_mesa_dst_reg result_dst
;
835 const glsl_type
*vec4_type
= glsl_type::get_instance(GLSL_TYPE_FLOAT
, 4, 1);
836 const glsl_type
*vec3_type
= glsl_type::get_instance(GLSL_TYPE_FLOAT
, 3, 1);
837 const glsl_type
*vec2_type
= glsl_type::get_instance(GLSL_TYPE_FLOAT
, 2, 1);
839 /* Quick peephole: Emit OPCODE_MAD(a, b, c) instead of ADD(MUL(a, b), c)
841 if (ir
->operation
== ir_binop_add
) {
842 if (try_emit_mad(ir
, 1))
844 if (try_emit_mad(ir
, 0))
848 for (operand
= 0; operand
< ir
->get_num_operands(); operand
++) {
849 this->result
.file
= PROGRAM_UNDEFINED
;
850 ir
->operands
[operand
]->accept(this);
851 if (this->result
.file
== PROGRAM_UNDEFINED
) {
853 printf("Failed to get tree for expression operand:\n");
854 ir
->operands
[operand
]->accept(&v
);
857 op
[operand
] = this->result
;
859 /* Matrix expression operands should have been broken down to vector
860 * operations already.
862 assert(!ir
->operands
[operand
]->type
->is_matrix());
865 int vector_elements
= ir
->operands
[0]->type
->vector_elements
;
866 if (ir
->operands
[1]) {
867 vector_elements
= MAX2(vector_elements
,
868 ir
->operands
[1]->type
->vector_elements
);
871 this->result
.file
= PROGRAM_UNDEFINED
;
873 /* Storage for our result. Ideally for an assignment we'd be using
874 * the actual storage for the result here, instead.
876 result_src
= get_temp(ir
->type
);
877 /* convenience for the emit functions below. */
878 result_dst
= ir_to_mesa_dst_reg_from_src(result_src
);
879 /* Limit writes to the channels that will be used by result_src later.
880 * This does limit this temp's use as a temporary for multi-instruction
883 result_dst
.writemask
= (1 << ir
->type
->vector_elements
) - 1;
885 switch (ir
->operation
) {
886 case ir_unop_logic_not
:
887 ir_to_mesa_emit_op2(ir
, OPCODE_SEQ
, result_dst
,
888 op
[0], src_reg_for_float(0.0));
891 op
[0].negate
= ~op
[0].negate
;
895 ir_to_mesa_emit_op1(ir
, OPCODE_ABS
, result_dst
, op
[0]);
898 ir_to_mesa_emit_op1(ir
, OPCODE_SSG
, result_dst
, op
[0]);
901 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_RCP
, result_dst
, op
[0]);
905 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_EX2
, result_dst
, op
[0]);
909 assert(!"not reached: should be handled by ir_explog_to_explog2");
912 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_LG2
, result_dst
, op
[0]);
915 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_SIN
, result_dst
, op
[0]);
918 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_COS
, result_dst
, op
[0]);
922 ir_to_mesa_emit_op1(ir
, OPCODE_DDX
, result_dst
, op
[0]);
925 ir_to_mesa_emit_op1(ir
, OPCODE_DDY
, result_dst
, op
[0]);
928 case ir_unop_noise
: {
929 const enum prog_opcode opcode
=
930 prog_opcode(OPCODE_NOISE1
931 + (ir
->operands
[0]->type
->vector_elements
) - 1);
932 assert((opcode
>= OPCODE_NOISE1
) && (opcode
<= OPCODE_NOISE4
));
934 ir_to_mesa_emit_op1(ir
, opcode
, result_dst
, op
[0]);
939 ir_to_mesa_emit_op2(ir
, OPCODE_ADD
, result_dst
, op
[0], op
[1]);
942 ir_to_mesa_emit_op2(ir
, OPCODE_SUB
, result_dst
, op
[0], op
[1]);
946 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
, result_dst
, op
[0], op
[1]);
949 assert(!"not reached: should be handled by ir_div_to_mul_rcp");
951 assert(!"ir_binop_mod should have been converted to b * fract(a/b)");
955 ir_to_mesa_emit_op2(ir
, OPCODE_SLT
, result_dst
, op
[0], op
[1]);
957 case ir_binop_greater
:
958 ir_to_mesa_emit_op2(ir
, OPCODE_SGT
, result_dst
, op
[0], op
[1]);
960 case ir_binop_lequal
:
961 ir_to_mesa_emit_op2(ir
, OPCODE_SLE
, result_dst
, op
[0], op
[1]);
963 case ir_binop_gequal
:
964 ir_to_mesa_emit_op2(ir
, OPCODE_SGE
, result_dst
, op
[0], op
[1]);
967 ir_to_mesa_emit_op2(ir
, OPCODE_SEQ
, result_dst
, op
[0], op
[1]);
969 case ir_binop_nequal
:
970 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
, result_dst
, op
[0], op
[1]);
972 case ir_binop_all_equal
:
973 /* "==" operator producing a scalar boolean. */
974 if (ir
->operands
[0]->type
->is_vector() ||
975 ir
->operands
[1]->type
->is_vector()) {
976 ir_to_mesa_src_reg temp
= get_temp(glsl_type::vec4_type
);
977 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
978 ir_to_mesa_dst_reg_from_src(temp
), op
[0], op
[1]);
979 if (vector_elements
== 4)
980 ir_to_mesa_emit_op2(ir
, OPCODE_DP4
, result_dst
, temp
, temp
);
981 else if (vector_elements
== 3)
982 ir_to_mesa_emit_op2(ir
, OPCODE_DP3
, result_dst
, temp
, temp
);
984 ir_to_mesa_emit_op2(ir
, OPCODE_DP2
, result_dst
, temp
, temp
);
985 ir_to_mesa_emit_op2(ir
, OPCODE_SEQ
,
986 result_dst
, result_src
, src_reg_for_float(0.0));
988 ir_to_mesa_emit_op2(ir
, OPCODE_SEQ
, result_dst
, op
[0], op
[1]);
991 case ir_binop_any_nequal
:
992 /* "!=" operator producing a scalar boolean. */
993 if (ir
->operands
[0]->type
->is_vector() ||
994 ir
->operands
[1]->type
->is_vector()) {
995 ir_to_mesa_src_reg temp
= get_temp(glsl_type::vec4_type
);
996 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
997 ir_to_mesa_dst_reg_from_src(temp
), op
[0], op
[1]);
998 if (vector_elements
== 4)
999 ir_to_mesa_emit_op2(ir
, OPCODE_DP4
, result_dst
, temp
, temp
);
1000 else if (vector_elements
== 3)
1001 ir_to_mesa_emit_op2(ir
, OPCODE_DP3
, result_dst
, temp
, temp
);
1003 ir_to_mesa_emit_op2(ir
, OPCODE_DP2
, result_dst
, temp
, temp
);
1004 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
1005 result_dst
, result_src
, src_reg_for_float(0.0));
1007 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
, result_dst
, op
[0], op
[1]);
1012 switch (ir
->operands
[0]->type
->vector_elements
) {
1014 ir_to_mesa_emit_op2(ir
, OPCODE_DP4
, result_dst
, op
[0], op
[0]);
1017 ir_to_mesa_emit_op2(ir
, OPCODE_DP3
, result_dst
, op
[0], op
[0]);
1020 ir_to_mesa_emit_op2(ir
, OPCODE_DP2
, result_dst
, op
[0], op
[0]);
1023 assert(!"unreached: ir_unop_any of non-bvec");
1026 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
1027 result_dst
, result_src
, src_reg_for_float(0.0));
1030 case ir_binop_logic_xor
:
1031 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
, result_dst
, op
[0], op
[1]);
1034 case ir_binop_logic_or
:
1035 /* This could be a saturated add and skip the SNE. */
1036 ir_to_mesa_emit_op2(ir
, OPCODE_ADD
,
1040 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
1042 result_src
, src_reg_for_float(0.0));
1045 case ir_binop_logic_and
:
1046 /* the bool args are stored as float 0.0 or 1.0, so "mul" gives us "and". */
1047 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
,
1053 if (ir
->operands
[0]->type
== vec4_type
) {
1054 assert(ir
->operands
[1]->type
== vec4_type
);
1055 ir_to_mesa_emit_op2(ir
, OPCODE_DP4
,
1058 } else if (ir
->operands
[0]->type
== vec3_type
) {
1059 assert(ir
->operands
[1]->type
== vec3_type
);
1060 ir_to_mesa_emit_op2(ir
, OPCODE_DP3
,
1063 } else if (ir
->operands
[0]->type
== vec2_type
) {
1064 assert(ir
->operands
[1]->type
== vec2_type
);
1065 ir_to_mesa_emit_op2(ir
, OPCODE_DP2
,
1071 case ir_binop_cross
:
1072 ir_to_mesa_emit_op2(ir
, OPCODE_XPD
, result_dst
, op
[0], op
[1]);
1076 /* sqrt(x) = x * rsq(x). */
1077 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_RSQ
, result_dst
, op
[0]);
1078 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
, result_dst
, result_src
, op
[0]);
1079 /* For incoming channels <= 0, set the result to 0. */
1080 op
[0].negate
= ~op
[0].negate
;
1081 ir_to_mesa_emit_op3(ir
, OPCODE_CMP
, result_dst
,
1082 op
[0], result_src
, src_reg_for_float(0.0));
1085 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_RSQ
, result_dst
, op
[0]);
1090 /* Mesa IR lacks types, ints are stored as truncated floats. */
1094 ir_to_mesa_emit_op1(ir
, OPCODE_TRUNC
, result_dst
, op
[0]);
1098 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
, result_dst
,
1099 op
[0], src_reg_for_float(0.0));
1102 ir_to_mesa_emit_op1(ir
, OPCODE_TRUNC
, result_dst
, op
[0]);
1105 op
[0].negate
= ~op
[0].negate
;
1106 ir_to_mesa_emit_op1(ir
, OPCODE_FLR
, result_dst
, op
[0]);
1107 result_src
.negate
= ~result_src
.negate
;
1110 ir_to_mesa_emit_op1(ir
, OPCODE_FLR
, result_dst
, op
[0]);
1113 ir_to_mesa_emit_op1(ir
, OPCODE_FRC
, result_dst
, op
[0]);
1117 ir_to_mesa_emit_op2(ir
, OPCODE_MIN
, result_dst
, op
[0], op
[1]);
1120 ir_to_mesa_emit_op2(ir
, OPCODE_MAX
, result_dst
, op
[0], op
[1]);
1123 ir_to_mesa_emit_scalar_op2(ir
, OPCODE_POW
, result_dst
, op
[0], op
[1]);
1126 case ir_unop_bit_not
:
1128 case ir_binop_lshift
:
1129 case ir_binop_rshift
:
1130 case ir_binop_bit_and
:
1131 case ir_binop_bit_xor
:
1132 case ir_binop_bit_or
:
1133 case ir_unop_round_even
:
1134 assert(!"GLSL 1.30 features unsupported");
1138 this->result
= result_src
;
1143 ir_to_mesa_visitor::visit(ir_swizzle
*ir
)
1145 ir_to_mesa_src_reg src_reg
;
1149 /* Note that this is only swizzles in expressions, not those on the left
1150 * hand side of an assignment, which do write masking. See ir_assignment
1154 ir
->val
->accept(this);
1155 src_reg
= this->result
;
1156 assert(src_reg
.file
!= PROGRAM_UNDEFINED
);
1158 for (i
= 0; i
< 4; i
++) {
1159 if (i
< ir
->type
->vector_elements
) {
1162 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.x
);
1165 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.y
);
1168 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.z
);
1171 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.w
);
1175 /* If the type is smaller than a vec4, replicate the last
1178 swizzle
[i
] = swizzle
[ir
->type
->vector_elements
- 1];
1182 src_reg
.swizzle
= MAKE_SWIZZLE4(swizzle
[0],
1187 this->result
= src_reg
;
1191 ir_to_mesa_visitor::visit(ir_dereference_variable
*ir
)
1193 variable_storage
*entry
= find_variable_storage(ir
->var
);
1196 switch (ir
->var
->mode
) {
1197 case ir_var_uniform
:
1198 entry
= new(mem_ctx
) variable_storage(ir
->var
, PROGRAM_UNIFORM
,
1200 this->variables
.push_tail(entry
);
1205 /* The linker assigns locations for varyings and attributes,
1206 * including deprecated builtins (like gl_Color), user-assign
1207 * generic attributes (glBindVertexLocation), and
1208 * user-defined varyings.
1210 * FINISHME: We would hit this path for function arguments. Fix!
1212 assert(ir
->var
->location
!= -1);
1213 if (ir
->var
->mode
== ir_var_in
||
1214 ir
->var
->mode
== ir_var_inout
) {
1215 entry
= new(mem_ctx
) variable_storage(ir
->var
,
1219 if (this->prog
->Target
== GL_VERTEX_PROGRAM_ARB
&&
1220 ir
->var
->location
>= VERT_ATTRIB_GENERIC0
) {
1221 _mesa_add_attribute(prog
->Attributes
,
1223 _mesa_sizeof_glsl_type(ir
->var
->type
->gl_type
),
1224 ir
->var
->type
->gl_type
,
1225 ir
->var
->location
- VERT_ATTRIB_GENERIC0
);
1228 entry
= new(mem_ctx
) variable_storage(ir
->var
,
1235 case ir_var_temporary
:
1236 entry
= new(mem_ctx
) variable_storage(ir
->var
, PROGRAM_TEMPORARY
,
1238 this->variables
.push_tail(entry
);
1240 next_temp
+= type_size(ir
->var
->type
);
1245 printf("Failed to make storage for %s\n", ir
->var
->name
);
1250 this->result
= ir_to_mesa_src_reg(entry
->file
, entry
->index
, ir
->var
->type
);
1254 ir_to_mesa_visitor::visit(ir_dereference_array
*ir
)
1257 ir_to_mesa_src_reg src_reg
;
1258 int element_size
= type_size(ir
->type
);
1260 index
= ir
->array_index
->constant_expression_value();
1262 ir
->array
->accept(this);
1263 src_reg
= this->result
;
1266 src_reg
.index
+= index
->value
.i
[0] * element_size
;
1268 ir_to_mesa_src_reg array_base
= this->result
;
1269 /* Variable index array dereference. It eats the "vec4" of the
1270 * base of the array and an index that offsets the Mesa register
1273 ir
->array_index
->accept(this);
1275 ir_to_mesa_src_reg index_reg
;
1277 if (element_size
== 1) {
1278 index_reg
= this->result
;
1280 index_reg
= get_temp(glsl_type::float_type
);
1282 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
,
1283 ir_to_mesa_dst_reg_from_src(index_reg
),
1284 this->result
, src_reg_for_float(element_size
));
1287 src_reg
.reladdr
= talloc(mem_ctx
, ir_to_mesa_src_reg
);
1288 memcpy(src_reg
.reladdr
, &index_reg
, sizeof(index_reg
));
1291 /* If the type is smaller than a vec4, replicate the last channel out. */
1292 if (ir
->type
->is_scalar() || ir
->type
->is_vector())
1293 src_reg
.swizzle
= swizzle_for_size(ir
->type
->vector_elements
);
1295 src_reg
.swizzle
= SWIZZLE_NOOP
;
1297 this->result
= src_reg
;
1301 ir_to_mesa_visitor::visit(ir_dereference_record
*ir
)
1304 const glsl_type
*struct_type
= ir
->record
->type
;
1307 ir
->record
->accept(this);
1309 for (i
= 0; i
< struct_type
->length
; i
++) {
1310 if (strcmp(struct_type
->fields
.structure
[i
].name
, ir
->field
) == 0)
1312 offset
+= type_size(struct_type
->fields
.structure
[i
].type
);
1314 this->result
.swizzle
= swizzle_for_size(ir
->type
->vector_elements
);
1315 this->result
.index
+= offset
;
1319 * We want to be careful in assignment setup to hit the actual storage
1320 * instead of potentially using a temporary like we might with the
1321 * ir_dereference handler.
1323 static struct ir_to_mesa_dst_reg
1324 get_assignment_lhs(ir_dereference
*ir
, ir_to_mesa_visitor
*v
)
1326 /* The LHS must be a dereference. If the LHS is a variable indexed array
1327 * access of a vector, it must be separated into a series conditional moves
1328 * before reaching this point (see ir_vec_index_to_cond_assign).
1330 assert(ir
->as_dereference());
1331 ir_dereference_array
*deref_array
= ir
->as_dereference_array();
1333 assert(!deref_array
->array
->type
->is_vector());
1336 /* Use the rvalue deref handler for the most part. We'll ignore
1337 * swizzles in it and write swizzles using writemask, though.
1340 return ir_to_mesa_dst_reg_from_src(v
->result
);
1344 ir_to_mesa_visitor::visit(ir_assignment
*ir
)
1346 struct ir_to_mesa_dst_reg l
;
1347 struct ir_to_mesa_src_reg r
;
1350 ir
->rhs
->accept(this);
1353 l
= get_assignment_lhs(ir
->lhs
, this);
1355 /* FINISHME: This should really set to the correct maximal writemask for each
1356 * FINISHME: component written (in the loops below). This case can only
1357 * FINISHME: occur for matrices, arrays, and structures.
1359 if (ir
->write_mask
== 0) {
1360 assert(!ir
->lhs
->type
->is_scalar() && !ir
->lhs
->type
->is_vector());
1361 l
.writemask
= WRITEMASK_XYZW
;
1362 } else if (ir
->lhs
->type
->is_scalar()) {
1363 /* FINISHME: This hack makes writing to gl_FragDepth, which lives in the
1364 * FINISHME: W component of fragment shader output zero, work correctly.
1366 l
.writemask
= WRITEMASK_XYZW
;
1369 int first_enabled_chan
= 0;
1372 assert(ir
->lhs
->type
->is_vector());
1373 l
.writemask
= ir
->write_mask
;
1375 for (int i
= 0; i
< 4; i
++) {
1376 if (l
.writemask
& (1 << i
)) {
1377 first_enabled_chan
= GET_SWZ(r
.swizzle
, i
);
1382 /* Swizzle a small RHS vector into the channels being written.
1384 * glsl ir treats write_mask as dictating how many channels are
1385 * present on the RHS while Mesa IR treats write_mask as just
1386 * showing which channels of the vec4 RHS get written.
1388 for (int i
= 0; i
< 4; i
++) {
1389 if (l
.writemask
& (1 << i
))
1390 swizzles
[i
] = GET_SWZ(r
.swizzle
, rhs_chan
++);
1392 swizzles
[i
] = first_enabled_chan
;
1394 r
.swizzle
= MAKE_SWIZZLE4(swizzles
[0], swizzles
[1],
1395 swizzles
[2], swizzles
[3]);
1398 assert(l
.file
!= PROGRAM_UNDEFINED
);
1399 assert(r
.file
!= PROGRAM_UNDEFINED
);
1401 if (ir
->condition
) {
1402 ir_to_mesa_src_reg condition
;
1404 ir
->condition
->accept(this);
1405 condition
= this->result
;
1407 /* We use the OPCODE_CMP (a < 0 ? b : c) for conditional moves,
1408 * and the condition we produced is 0.0 or 1.0. By flipping the
1409 * sign, we can choose which value OPCODE_CMP produces without
1410 * an extra computing the condition.
1412 condition
.negate
= ~condition
.negate
;
1413 for (i
= 0; i
< type_size(ir
->lhs
->type
); i
++) {
1414 ir_to_mesa_emit_op3(ir
, OPCODE_CMP
, l
,
1415 condition
, r
, ir_to_mesa_src_reg_from_dst(l
));
1420 for (i
= 0; i
< type_size(ir
->lhs
->type
); i
++) {
1421 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1430 ir_to_mesa_visitor::visit(ir_constant
*ir
)
1432 ir_to_mesa_src_reg src_reg
;
1433 GLfloat stack_vals
[4] = { 0 };
1434 GLfloat
*values
= stack_vals
;
1437 /* Unfortunately, 4 floats is all we can get into
1438 * _mesa_add_unnamed_constant. So, make a temp to store an
1439 * aggregate constant and move each constant value into it. If we
1440 * get lucky, copy propagation will eliminate the extra moves.
1443 if (ir
->type
->base_type
== GLSL_TYPE_STRUCT
) {
1444 ir_to_mesa_src_reg temp_base
= get_temp(ir
->type
);
1445 ir_to_mesa_dst_reg temp
= ir_to_mesa_dst_reg_from_src(temp_base
);
1447 foreach_iter(exec_list_iterator
, iter
, ir
->components
) {
1448 ir_constant
*field_value
= (ir_constant
*)iter
.get();
1449 int size
= type_size(field_value
->type
);
1453 field_value
->accept(this);
1454 src_reg
= this->result
;
1456 for (i
= 0; i
< (unsigned int)size
; i
++) {
1457 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, temp
, src_reg
);
1463 this->result
= temp_base
;
1467 if (ir
->type
->is_array()) {
1468 ir_to_mesa_src_reg temp_base
= get_temp(ir
->type
);
1469 ir_to_mesa_dst_reg temp
= ir_to_mesa_dst_reg_from_src(temp_base
);
1470 int size
= type_size(ir
->type
->fields
.array
);
1474 for (i
= 0; i
< ir
->type
->length
; i
++) {
1475 ir
->array_elements
[i
]->accept(this);
1476 src_reg
= this->result
;
1477 for (int j
= 0; j
< size
; j
++) {
1478 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, temp
, src_reg
);
1484 this->result
= temp_base
;
1488 if (ir
->type
->is_matrix()) {
1489 ir_to_mesa_src_reg mat
= get_temp(ir
->type
);
1490 ir_to_mesa_dst_reg mat_column
= ir_to_mesa_dst_reg_from_src(mat
);
1492 for (i
= 0; i
< ir
->type
->matrix_columns
; i
++) {
1493 assert(ir
->type
->base_type
== GLSL_TYPE_FLOAT
);
1494 values
= &ir
->value
.f
[i
* ir
->type
->vector_elements
];
1496 src_reg
= ir_to_mesa_src_reg(PROGRAM_CONSTANT
, -1, NULL
);
1497 src_reg
.index
= _mesa_add_unnamed_constant(this->prog
->Parameters
,
1499 ir
->type
->vector_elements
,
1501 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, mat_column
, src_reg
);
1510 src_reg
.file
= PROGRAM_CONSTANT
;
1511 switch (ir
->type
->base_type
) {
1512 case GLSL_TYPE_FLOAT
:
1513 values
= &ir
->value
.f
[0];
1515 case GLSL_TYPE_UINT
:
1516 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1517 values
[i
] = ir
->value
.u
[i
];
1521 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1522 values
[i
] = ir
->value
.i
[i
];
1525 case GLSL_TYPE_BOOL
:
1526 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1527 values
[i
] = ir
->value
.b
[i
];
1531 assert(!"Non-float/uint/int/bool constant");
1534 this->result
= ir_to_mesa_src_reg(PROGRAM_CONSTANT
, -1, ir
->type
);
1535 this->result
.index
= _mesa_add_unnamed_constant(this->prog
->Parameters
,
1537 ir
->type
->vector_elements
,
1538 &this->result
.swizzle
);
1542 ir_to_mesa_visitor::get_function_signature(ir_function_signature
*sig
)
1544 function_entry
*entry
;
1546 foreach_iter(exec_list_iterator
, iter
, this->function_signatures
) {
1547 entry
= (function_entry
*)iter
.get();
1549 if (entry
->sig
== sig
)
1553 entry
= talloc(mem_ctx
, function_entry
);
1555 entry
->sig_id
= this->next_signature_id
++;
1556 entry
->bgn_inst
= NULL
;
1558 /* Allocate storage for all the parameters. */
1559 foreach_iter(exec_list_iterator
, iter
, sig
->parameters
) {
1560 ir_variable
*param
= (ir_variable
*)iter
.get();
1561 variable_storage
*storage
;
1563 storage
= find_variable_storage(param
);
1566 storage
= new(mem_ctx
) variable_storage(param
, PROGRAM_TEMPORARY
,
1568 this->variables
.push_tail(storage
);
1570 this->next_temp
+= type_size(param
->type
);
1573 if (!sig
->return_type
->is_void()) {
1574 entry
->return_reg
= get_temp(sig
->return_type
);
1576 entry
->return_reg
= ir_to_mesa_undef
;
1579 this->function_signatures
.push_tail(entry
);
1584 ir_to_mesa_visitor::visit(ir_call
*ir
)
1586 ir_to_mesa_instruction
*call_inst
;
1587 ir_function_signature
*sig
= ir
->get_callee();
1588 function_entry
*entry
= get_function_signature(sig
);
1591 /* Process in parameters. */
1592 exec_list_iterator sig_iter
= sig
->parameters
.iterator();
1593 foreach_iter(exec_list_iterator
, iter
, *ir
) {
1594 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
1595 ir_variable
*param
= (ir_variable
*)sig_iter
.get();
1597 if (param
->mode
== ir_var_in
||
1598 param
->mode
== ir_var_inout
) {
1599 variable_storage
*storage
= find_variable_storage(param
);
1602 param_rval
->accept(this);
1603 ir_to_mesa_src_reg r
= this->result
;
1605 ir_to_mesa_dst_reg l
;
1606 l
.file
= storage
->file
;
1607 l
.index
= storage
->index
;
1609 l
.writemask
= WRITEMASK_XYZW
;
1610 l
.cond_mask
= COND_TR
;
1612 for (i
= 0; i
< type_size(param
->type
); i
++) {
1613 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1621 assert(!sig_iter
.has_next());
1623 /* Emit call instruction */
1624 call_inst
= ir_to_mesa_emit_op1(ir
, OPCODE_CAL
,
1625 ir_to_mesa_undef_dst
, ir_to_mesa_undef
);
1626 call_inst
->function
= entry
;
1628 /* Process out parameters. */
1629 sig_iter
= sig
->parameters
.iterator();
1630 foreach_iter(exec_list_iterator
, iter
, *ir
) {
1631 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
1632 ir_variable
*param
= (ir_variable
*)sig_iter
.get();
1634 if (param
->mode
== ir_var_out
||
1635 param
->mode
== ir_var_inout
) {
1636 variable_storage
*storage
= find_variable_storage(param
);
1639 ir_to_mesa_src_reg r
;
1640 r
.file
= storage
->file
;
1641 r
.index
= storage
->index
;
1643 r
.swizzle
= SWIZZLE_NOOP
;
1646 param_rval
->accept(this);
1647 ir_to_mesa_dst_reg l
= ir_to_mesa_dst_reg_from_src(this->result
);
1649 for (i
= 0; i
< type_size(param
->type
); i
++) {
1650 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1658 assert(!sig_iter
.has_next());
1660 /* Process return value. */
1661 this->result
= entry
->return_reg
;
1665 ir_to_mesa_visitor::visit(ir_texture
*ir
)
1667 ir_to_mesa_src_reg result_src
, coord
, lod_info
, projector
;
1668 ir_to_mesa_dst_reg result_dst
, coord_dst
;
1669 ir_to_mesa_instruction
*inst
= NULL
;
1670 prog_opcode opcode
= OPCODE_NOP
;
1672 ir
->coordinate
->accept(this);
1674 /* Put our coords in a temp. We'll need to modify them for shadow,
1675 * projection, or LOD, so the only case we'd use it as is is if
1676 * we're doing plain old texturing. Mesa IR optimization should
1677 * handle cleaning up our mess in that case.
1679 coord
= get_temp(glsl_type::vec4_type
);
1680 coord_dst
= ir_to_mesa_dst_reg_from_src(coord
);
1681 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
,
1684 if (ir
->projector
) {
1685 ir
->projector
->accept(this);
1686 projector
= this->result
;
1689 /* Storage for our result. Ideally for an assignment we'd be using
1690 * the actual storage for the result here, instead.
1692 result_src
= get_temp(glsl_type::vec4_type
);
1693 result_dst
= ir_to_mesa_dst_reg_from_src(result_src
);
1697 opcode
= OPCODE_TEX
;
1700 opcode
= OPCODE_TXB
;
1701 ir
->lod_info
.bias
->accept(this);
1702 lod_info
= this->result
;
1705 opcode
= OPCODE_TXL
;
1706 ir
->lod_info
.lod
->accept(this);
1707 lod_info
= this->result
;
1711 assert(!"GLSL 1.30 features unsupported");
1715 if (ir
->projector
) {
1716 if (opcode
== OPCODE_TEX
) {
1717 /* Slot the projector in as the last component of the coord. */
1718 coord_dst
.writemask
= WRITEMASK_W
;
1719 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, projector
);
1720 coord_dst
.writemask
= WRITEMASK_XYZW
;
1721 opcode
= OPCODE_TXP
;
1723 ir_to_mesa_src_reg coord_w
= coord
;
1724 coord_w
.swizzle
= SWIZZLE_WWWW
;
1726 /* For the other TEX opcodes there's no projective version
1727 * since the last slot is taken up by lod info. Do the
1728 * projective divide now.
1730 coord_dst
.writemask
= WRITEMASK_W
;
1731 ir_to_mesa_emit_op1(ir
, OPCODE_RCP
, coord_dst
, projector
);
1733 coord_dst
.writemask
= WRITEMASK_XYZ
;
1734 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
, coord_dst
, coord
, coord_w
);
1736 coord_dst
.writemask
= WRITEMASK_XYZW
;
1737 coord
.swizzle
= SWIZZLE_XYZW
;
1741 if (ir
->shadow_comparitor
) {
1742 /* Slot the shadow value in as the second to last component of the
1745 ir
->shadow_comparitor
->accept(this);
1746 coord_dst
.writemask
= WRITEMASK_Z
;
1747 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, this->result
);
1748 coord_dst
.writemask
= WRITEMASK_XYZW
;
1751 if (opcode
== OPCODE_TXL
|| opcode
== OPCODE_TXB
) {
1752 /* Mesa IR stores lod or lod bias in the last channel of the coords. */
1753 coord_dst
.writemask
= WRITEMASK_W
;
1754 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, lod_info
);
1755 coord_dst
.writemask
= WRITEMASK_XYZW
;
1758 inst
= ir_to_mesa_emit_op1(ir
, opcode
, result_dst
, coord
);
1760 if (ir
->shadow_comparitor
)
1761 inst
->tex_shadow
= GL_TRUE
;
1763 inst
->sampler
= _mesa_get_sampler_uniform_value(ir
->sampler
,
1764 this->shader_program
,
1767 const glsl_type
*sampler_type
= ir
->sampler
->type
;
1769 switch (sampler_type
->sampler_dimensionality
) {
1770 case GLSL_SAMPLER_DIM_1D
:
1771 inst
->tex_target
= (sampler_type
->sampler_array
)
1772 ? TEXTURE_1D_ARRAY_INDEX
: TEXTURE_1D_INDEX
;
1774 case GLSL_SAMPLER_DIM_2D
:
1775 inst
->tex_target
= (sampler_type
->sampler_array
)
1776 ? TEXTURE_2D_ARRAY_INDEX
: TEXTURE_2D_INDEX
;
1778 case GLSL_SAMPLER_DIM_3D
:
1779 inst
->tex_target
= TEXTURE_3D_INDEX
;
1781 case GLSL_SAMPLER_DIM_CUBE
:
1782 inst
->tex_target
= TEXTURE_CUBE_INDEX
;
1784 case GLSL_SAMPLER_DIM_RECT
:
1785 inst
->tex_target
= TEXTURE_RECT_INDEX
;
1787 case GLSL_SAMPLER_DIM_BUF
:
1788 assert(!"FINISHME: Implement ARB_texture_buffer_object");
1791 assert(!"Should not get here.");
1794 this->result
= result_src
;
1798 ir_to_mesa_visitor::visit(ir_return
*ir
)
1800 if (ir
->get_value()) {
1801 ir_to_mesa_dst_reg l
;
1804 assert(current_function
);
1806 ir
->get_value()->accept(this);
1807 ir_to_mesa_src_reg r
= this->result
;
1809 l
= ir_to_mesa_dst_reg_from_src(current_function
->return_reg
);
1811 for (i
= 0; i
< type_size(current_function
->sig
->return_type
); i
++) {
1812 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1818 ir_to_mesa_emit_op0(ir
, OPCODE_RET
);
1822 ir_to_mesa_visitor::visit(ir_discard
*ir
)
1824 struct gl_fragment_program
*fp
= (struct gl_fragment_program
*)this->prog
;
1826 assert(ir
->condition
== NULL
); /* FINISHME */
1828 ir_to_mesa_emit_op0(ir
, OPCODE_KIL_NV
);
1829 fp
->UsesKill
= GL_TRUE
;
1833 ir_to_mesa_visitor::visit(ir_if
*ir
)
1835 ir_to_mesa_instruction
*cond_inst
, *if_inst
, *else_inst
= NULL
;
1836 ir_to_mesa_instruction
*prev_inst
;
1838 prev_inst
= (ir_to_mesa_instruction
*)this->instructions
.get_tail();
1840 ir
->condition
->accept(this);
1841 assert(this->result
.file
!= PROGRAM_UNDEFINED
);
1843 if (this->options
->EmitCondCodes
) {
1844 cond_inst
= (ir_to_mesa_instruction
*)this->instructions
.get_tail();
1846 /* See if we actually generated any instruction for generating
1847 * the condition. If not, then cook up a move to a temp so we
1848 * have something to set cond_update on.
1850 if (cond_inst
== prev_inst
) {
1851 ir_to_mesa_src_reg temp
= get_temp(glsl_type::bool_type
);
1852 cond_inst
= ir_to_mesa_emit_op1(ir
->condition
, OPCODE_MOV
,
1853 ir_to_mesa_dst_reg_from_src(temp
),
1856 cond_inst
->cond_update
= GL_TRUE
;
1858 if_inst
= ir_to_mesa_emit_op0(ir
->condition
, OPCODE_IF
);
1859 if_inst
->dst_reg
.cond_mask
= COND_NE
;
1861 if_inst
= ir_to_mesa_emit_op1(ir
->condition
,
1862 OPCODE_IF
, ir_to_mesa_undef_dst
,
1866 this->instructions
.push_tail(if_inst
);
1868 visit_exec_list(&ir
->then_instructions
, this);
1870 if (!ir
->else_instructions
.is_empty()) {
1871 else_inst
= ir_to_mesa_emit_op0(ir
->condition
, OPCODE_ELSE
);
1872 visit_exec_list(&ir
->else_instructions
, this);
1875 if_inst
= ir_to_mesa_emit_op1(ir
->condition
, OPCODE_ENDIF
,
1876 ir_to_mesa_undef_dst
, ir_to_mesa_undef
);
1879 ir_to_mesa_visitor::ir_to_mesa_visitor()
1881 result
.file
= PROGRAM_UNDEFINED
;
1883 next_signature_id
= 1;
1884 current_function
= NULL
;
1885 mem_ctx
= talloc_new(NULL
);
1888 ir_to_mesa_visitor::~ir_to_mesa_visitor()
1890 talloc_free(mem_ctx
);
1893 static struct prog_src_register
1894 mesa_src_reg_from_ir_src_reg(ir_to_mesa_src_reg reg
)
1896 struct prog_src_register mesa_reg
;
1898 mesa_reg
.File
= reg
.file
;
1899 assert(reg
.index
< (1 << INST_INDEX_BITS
) - 1);
1900 mesa_reg
.Index
= reg
.index
;
1901 mesa_reg
.Swizzle
= reg
.swizzle
;
1902 mesa_reg
.RelAddr
= reg
.reladdr
!= NULL
;
1903 mesa_reg
.Negate
= reg
.negate
;
1905 mesa_reg
.HasIndex2
= GL_FALSE
;
1906 mesa_reg
.RelAddr2
= 0;
1907 mesa_reg
.Index2
= 0;
1913 set_branchtargets(ir_to_mesa_visitor
*v
,
1914 struct prog_instruction
*mesa_instructions
,
1915 int num_instructions
)
1917 int if_count
= 0, loop_count
= 0;
1918 int *if_stack
, *loop_stack
;
1919 int if_stack_pos
= 0, loop_stack_pos
= 0;
1922 for (i
= 0; i
< num_instructions
; i
++) {
1923 switch (mesa_instructions
[i
].Opcode
) {
1927 case OPCODE_BGNLOOP
:
1932 mesa_instructions
[i
].BranchTarget
= -1;
1939 if_stack
= talloc_zero_array(v
->mem_ctx
, int, if_count
);
1940 loop_stack
= talloc_zero_array(v
->mem_ctx
, int, loop_count
);
1942 for (i
= 0; i
< num_instructions
; i
++) {
1943 switch (mesa_instructions
[i
].Opcode
) {
1945 if_stack
[if_stack_pos
] = i
;
1949 mesa_instructions
[if_stack
[if_stack_pos
- 1]].BranchTarget
= i
;
1950 if_stack
[if_stack_pos
- 1] = i
;
1953 mesa_instructions
[if_stack
[if_stack_pos
- 1]].BranchTarget
= i
;
1956 case OPCODE_BGNLOOP
:
1957 loop_stack
[loop_stack_pos
] = i
;
1960 case OPCODE_ENDLOOP
:
1962 /* Rewrite any breaks/conts at this nesting level (haven't
1963 * already had a BranchTarget assigned) to point to the end
1966 for (j
= loop_stack
[loop_stack_pos
]; j
< i
; j
++) {
1967 if (mesa_instructions
[j
].Opcode
== OPCODE_BRK
||
1968 mesa_instructions
[j
].Opcode
== OPCODE_CONT
) {
1969 if (mesa_instructions
[j
].BranchTarget
== -1) {
1970 mesa_instructions
[j
].BranchTarget
= i
;
1974 /* The loop ends point at each other. */
1975 mesa_instructions
[i
].BranchTarget
= loop_stack
[loop_stack_pos
];
1976 mesa_instructions
[loop_stack
[loop_stack_pos
]].BranchTarget
= i
;
1979 foreach_iter(exec_list_iterator
, iter
, v
->function_signatures
) {
1980 function_entry
*entry
= (function_entry
*)iter
.get();
1982 if (entry
->sig_id
== mesa_instructions
[i
].BranchTarget
) {
1983 mesa_instructions
[i
].BranchTarget
= entry
->inst
;
1995 print_program(struct prog_instruction
*mesa_instructions
,
1996 ir_instruction
**mesa_instruction_annotation
,
1997 int num_instructions
)
1999 ir_instruction
*last_ir
= NULL
;
2003 for (i
= 0; i
< num_instructions
; i
++) {
2004 struct prog_instruction
*mesa_inst
= mesa_instructions
+ i
;
2005 ir_instruction
*ir
= mesa_instruction_annotation
[i
];
2007 fprintf(stdout
, "%3d: ", i
);
2009 if (last_ir
!= ir
&& ir
) {
2012 for (j
= 0; j
< indent
; j
++) {
2013 fprintf(stdout
, " ");
2019 fprintf(stdout
, " "); /* line number spacing. */
2022 indent
= _mesa_fprint_instruction_opt(stdout
, mesa_inst
, indent
,
2023 PROG_PRINT_DEBUG
, NULL
);
2028 count_resources(struct gl_program
*prog
)
2032 prog
->SamplersUsed
= 0;
2034 for (i
= 0; i
< prog
->NumInstructions
; i
++) {
2035 struct prog_instruction
*inst
= &prog
->Instructions
[i
];
2037 if (_mesa_is_tex_instruction(inst
->Opcode
)) {
2038 prog
->SamplerTargets
[inst
->TexSrcUnit
] =
2039 (gl_texture_index
)inst
->TexSrcTarget
;
2040 prog
->SamplersUsed
|= 1 << inst
->TexSrcUnit
;
2041 if (inst
->TexShadow
) {
2042 prog
->ShadowSamplers
|= 1 << inst
->TexSrcUnit
;
2047 _mesa_update_shader_textures_used(prog
);
2050 struct uniform_sort
{
2051 struct gl_uniform
*u
;
2055 /* The shader_program->Uniforms list is almost sorted in increasing
2056 * uniform->{Frag,Vert}Pos locations, but not quite when there are
2057 * uniforms shared between targets. We need to add parameters in
2058 * increasing order for the targets.
2061 sort_uniforms(const void *a
, const void *b
)
2063 struct uniform_sort
*u1
= (struct uniform_sort
*)a
;
2064 struct uniform_sort
*u2
= (struct uniform_sort
*)b
;
2066 return u1
->pos
- u2
->pos
;
2069 /* Add the uniforms to the parameters. The linker chose locations
2070 * in our parameters lists (which weren't created yet), which the
2071 * uniforms code will use to poke values into our parameters list
2072 * when uniforms are updated.
2075 add_uniforms_to_parameters_list(struct gl_shader_program
*shader_program
,
2076 struct gl_shader
*shader
,
2077 struct gl_program
*prog
)
2080 unsigned int next_sampler
= 0, num_uniforms
= 0;
2081 struct uniform_sort
*sorted_uniforms
;
2083 sorted_uniforms
= talloc_array(NULL
, struct uniform_sort
,
2084 shader_program
->Uniforms
->NumUniforms
);
2086 for (i
= 0; i
< shader_program
->Uniforms
->NumUniforms
; i
++) {
2087 struct gl_uniform
*uniform
= shader_program
->Uniforms
->Uniforms
+ i
;
2088 int parameter_index
= -1;
2090 switch (shader
->Type
) {
2091 case GL_VERTEX_SHADER
:
2092 parameter_index
= uniform
->VertPos
;
2094 case GL_FRAGMENT_SHADER
:
2095 parameter_index
= uniform
->FragPos
;
2097 case GL_GEOMETRY_SHADER
:
2098 parameter_index
= uniform
->GeomPos
;
2102 /* Only add uniforms used in our target. */
2103 if (parameter_index
!= -1) {
2104 sorted_uniforms
[num_uniforms
].pos
= parameter_index
;
2105 sorted_uniforms
[num_uniforms
].u
= uniform
;
2110 qsort(sorted_uniforms
, num_uniforms
, sizeof(struct uniform_sort
),
2113 for (i
= 0; i
< num_uniforms
; i
++) {
2114 struct gl_uniform
*uniform
= sorted_uniforms
[i
].u
;
2115 int parameter_index
= sorted_uniforms
[i
].pos
;
2116 const glsl_type
*type
= uniform
->Type
;
2119 if (type
->is_vector() ||
2120 type
->is_scalar()) {
2121 size
= type
->vector_elements
;
2123 size
= type_size(type
) * 4;
2126 gl_register_file file
;
2127 if (type
->is_sampler() ||
2128 (type
->is_array() && type
->fields
.array
->is_sampler())) {
2129 file
= PROGRAM_SAMPLER
;
2131 file
= PROGRAM_UNIFORM
;
2134 GLint index
= _mesa_lookup_parameter_index(prog
->Parameters
, -1,
2138 index
= _mesa_add_parameter(prog
->Parameters
, file
,
2139 uniform
->Name
, size
, type
->gl_type
,
2142 /* Sampler uniform values are stored in prog->SamplerUnits,
2143 * and the entry in that array is selected by this index we
2144 * store in ParameterValues[].
2146 if (file
== PROGRAM_SAMPLER
) {
2147 for (unsigned int j
= 0; j
< size
/ 4; j
++)
2148 prog
->Parameters
->ParameterValues
[index
+ j
][0] = next_sampler
++;
2151 /* The location chosen in the Parameters list here (returned
2152 * from _mesa_add_uniform) has to match what the linker chose.
2154 if (index
!= parameter_index
) {
2155 fail_link(shader_program
, "Allocation of uniform `%s' to target "
2156 "failed (%d vs %d)\n",
2157 uniform
->Name
, index
, parameter_index
);
2162 talloc_free(sorted_uniforms
);
2166 set_uniform_initializer(struct gl_context
*ctx
, void *mem_ctx
,
2167 struct gl_shader_program
*shader_program
,
2168 const char *name
, const glsl_type
*type
,
2171 if (type
->is_record()) {
2172 ir_constant
*field_constant
;
2174 field_constant
= (ir_constant
*)val
->components
.get_head();
2176 for (unsigned int i
= 0; i
< type
->length
; i
++) {
2177 const glsl_type
*field_type
= type
->fields
.structure
[i
].type
;
2178 const char *field_name
= talloc_asprintf(mem_ctx
, "%s.%s", name
,
2179 type
->fields
.structure
[i
].name
);
2180 set_uniform_initializer(ctx
, mem_ctx
, shader_program
, field_name
,
2181 field_type
, field_constant
);
2182 field_constant
= (ir_constant
*)field_constant
->next
;
2187 int loc
= _mesa_get_uniform_location(ctx
, shader_program
, name
);
2190 fail_link(shader_program
,
2191 "Couldn't find uniform for initializer %s\n", name
);
2195 for (unsigned int i
= 0; i
< (type
->is_array() ? type
->length
: 1); i
++) {
2196 ir_constant
*element
;
2197 const glsl_type
*element_type
;
2198 if (type
->is_array()) {
2199 element
= val
->array_elements
[i
];
2200 element_type
= type
->fields
.array
;
2203 element_type
= type
;
2208 if (element_type
->base_type
== GLSL_TYPE_BOOL
) {
2209 int *conv
= talloc_array(mem_ctx
, int, element_type
->components());
2210 for (unsigned int j
= 0; j
< element_type
->components(); j
++) {
2211 conv
[j
] = element
->value
.b
[j
];
2213 values
= (void *)conv
;
2214 element_type
= glsl_type::get_instance(GLSL_TYPE_INT
,
2215 element_type
->vector_elements
,
2218 values
= &element
->value
;
2221 if (element_type
->is_matrix()) {
2222 _mesa_uniform_matrix(ctx
, shader_program
,
2223 element_type
->matrix_columns
,
2224 element_type
->vector_elements
,
2225 loc
, 1, GL_FALSE
, (GLfloat
*)values
);
2226 loc
+= element_type
->matrix_columns
;
2228 _mesa_uniform(ctx
, shader_program
, loc
, element_type
->matrix_columns
,
2229 values
, element_type
->gl_type
);
2230 loc
+= type_size(element_type
);
2236 set_uniform_initializers(struct gl_context
*ctx
,
2237 struct gl_shader_program
*shader_program
)
2239 void *mem_ctx
= NULL
;
2241 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2242 struct gl_shader
*shader
= shader_program
->_LinkedShaders
[i
];
2247 foreach_iter(exec_list_iterator
, iter
, *shader
->ir
) {
2248 ir_instruction
*ir
= (ir_instruction
*)iter
.get();
2249 ir_variable
*var
= ir
->as_variable();
2251 if (!var
|| var
->mode
!= ir_var_uniform
|| !var
->constant_value
)
2255 mem_ctx
= talloc_new(NULL
);
2257 set_uniform_initializer(ctx
, mem_ctx
, shader_program
, var
->name
,
2258 var
->type
, var
->constant_value
);
2262 talloc_free(mem_ctx
);
2267 * Convert a shader's GLSL IR into a Mesa gl_program.
2270 get_mesa_program(struct gl_context
*ctx
, struct gl_shader_program
*shader_program
,
2271 struct gl_shader
*shader
)
2273 ir_to_mesa_visitor v
;
2274 struct prog_instruction
*mesa_instructions
, *mesa_inst
;
2275 ir_instruction
**mesa_instruction_annotation
;
2277 struct gl_program
*prog
;
2279 const char *target_string
;
2281 struct gl_shader_compiler_options
*options
=
2282 &ctx
->ShaderCompilerOptions
[_mesa_shader_type_to_index(shader
->Type
)];
2284 switch (shader
->Type
) {
2285 case GL_VERTEX_SHADER
:
2286 target
= GL_VERTEX_PROGRAM_ARB
;
2287 target_string
= "vertex";
2289 case GL_FRAGMENT_SHADER
:
2290 target
= GL_FRAGMENT_PROGRAM_ARB
;
2291 target_string
= "fragment";
2294 assert(!"should not be reached");
2298 validate_ir_tree(shader
->ir
);
2300 prog
= ctx
->Driver
.NewProgram(ctx
, target
, shader_program
->Name
);
2303 prog
->Parameters
= _mesa_new_parameter_list();
2304 prog
->Varying
= _mesa_new_parameter_list();
2305 prog
->Attributes
= _mesa_new_parameter_list();
2308 v
.shader_program
= shader_program
;
2309 v
.options
= options
;
2311 add_uniforms_to_parameters_list(shader_program
, shader
, prog
);
2313 /* Emit Mesa IR for main(). */
2314 visit_exec_list(shader
->ir
, &v
);
2315 v
.ir_to_mesa_emit_op0(NULL
, OPCODE_END
);
2317 /* Now emit bodies for any functions that were used. */
2319 progress
= GL_FALSE
;
2321 foreach_iter(exec_list_iterator
, iter
, v
.function_signatures
) {
2322 function_entry
*entry
= (function_entry
*)iter
.get();
2324 if (!entry
->bgn_inst
) {
2325 v
.current_function
= entry
;
2327 entry
->bgn_inst
= v
.ir_to_mesa_emit_op0(NULL
, OPCODE_BGNSUB
);
2328 entry
->bgn_inst
->function
= entry
;
2330 visit_exec_list(&entry
->sig
->body
, &v
);
2332 ir_to_mesa_instruction
*last
;
2333 last
= (ir_to_mesa_instruction
*)v
.instructions
.get_tail();
2334 if (last
->op
!= OPCODE_RET
)
2335 v
.ir_to_mesa_emit_op0(NULL
, OPCODE_RET
);
2337 ir_to_mesa_instruction
*end
;
2338 end
= v
.ir_to_mesa_emit_op0(NULL
, OPCODE_ENDSUB
);
2339 end
->function
= entry
;
2346 prog
->NumTemporaries
= v
.next_temp
;
2348 int num_instructions
= 0;
2349 foreach_iter(exec_list_iterator
, iter
, v
.instructions
) {
2354 (struct prog_instruction
*)calloc(num_instructions
,
2355 sizeof(*mesa_instructions
));
2356 mesa_instruction_annotation
= talloc_array(v
.mem_ctx
, ir_instruction
*,
2359 /* Convert ir_mesa_instructions into prog_instructions.
2361 mesa_inst
= mesa_instructions
;
2363 foreach_iter(exec_list_iterator
, iter
, v
.instructions
) {
2364 const ir_to_mesa_instruction
*inst
= (ir_to_mesa_instruction
*)iter
.get();
2366 mesa_inst
->Opcode
= inst
->op
;
2367 mesa_inst
->CondUpdate
= inst
->cond_update
;
2368 mesa_inst
->DstReg
.File
= inst
->dst_reg
.file
;
2369 mesa_inst
->DstReg
.Index
= inst
->dst_reg
.index
;
2370 mesa_inst
->DstReg
.CondMask
= inst
->dst_reg
.cond_mask
;
2371 mesa_inst
->DstReg
.WriteMask
= inst
->dst_reg
.writemask
;
2372 mesa_inst
->DstReg
.RelAddr
= inst
->dst_reg
.reladdr
!= NULL
;
2373 mesa_inst
->SrcReg
[0] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[0]);
2374 mesa_inst
->SrcReg
[1] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[1]);
2375 mesa_inst
->SrcReg
[2] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[2]);
2376 mesa_inst
->TexSrcUnit
= inst
->sampler
;
2377 mesa_inst
->TexSrcTarget
= inst
->tex_target
;
2378 mesa_inst
->TexShadow
= inst
->tex_shadow
;
2379 mesa_instruction_annotation
[i
] = inst
->ir
;
2381 /* Set IndirectRegisterFiles. */
2382 if (mesa_inst
->DstReg
.RelAddr
)
2383 prog
->IndirectRegisterFiles
|= 1 << mesa_inst
->DstReg
.File
;
2385 /* Update program's bitmask of indirectly accessed register files */
2386 for (unsigned src
= 0; src
< 3; src
++)
2387 if (mesa_inst
->SrcReg
[src
].RelAddr
)
2388 prog
->IndirectRegisterFiles
|= 1 << mesa_inst
->SrcReg
[src
].File
;
2390 if (options
->EmitNoIfs
&& mesa_inst
->Opcode
== OPCODE_IF
) {
2391 fail_link(shader_program
, "Couldn't flatten if statement\n");
2394 switch (mesa_inst
->Opcode
) {
2396 inst
->function
->inst
= i
;
2397 mesa_inst
->Comment
= strdup(inst
->function
->sig
->function_name());
2400 mesa_inst
->Comment
= strdup(inst
->function
->sig
->function_name());
2403 mesa_inst
->BranchTarget
= inst
->function
->sig_id
; /* rewritten later */
2406 prog
->NumAddressRegs
= 1;
2416 set_branchtargets(&v
, mesa_instructions
, num_instructions
);
2418 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2420 printf("GLSL IR for linked %s program %d:\n", target_string
,
2421 shader_program
->Name
);
2422 _mesa_print_ir(shader
->ir
, NULL
);
2425 printf("Mesa IR for linked %s program %d:\n", target_string
,
2426 shader_program
->Name
);
2427 print_program(mesa_instructions
, mesa_instruction_annotation
,
2431 prog
->Instructions
= mesa_instructions
;
2432 prog
->NumInstructions
= num_instructions
;
2434 do_set_program_inouts(shader
->ir
, prog
);
2435 count_resources(prog
);
2437 _mesa_reference_program(ctx
, &shader
->Program
, prog
);
2439 if ((ctx
->Shader
.Flags
& GLSL_NO_OPT
) == 0) {
2440 _mesa_optimize_program(ctx
, prog
);
2449 * Called via ctx->Driver.CompilerShader().
2451 * XXX can we remove the ctx->Driver.CompileShader() hook?
2454 _mesa_ir_compile_shader(struct gl_context
*ctx
, struct gl_shader
*shader
)
2456 assert(shader
->CompileStatus
);
2465 * Called via ctx->Driver.LinkShader()
2466 * This actually involves converting GLSL IR into Mesa gl_programs with
2467 * code lowering and other optimizations.
2470 _mesa_ir_link_shader(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2472 assert(prog
->LinkStatus
);
2474 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2475 if (prog
->_LinkedShaders
[i
] == NULL
)
2479 exec_list
*ir
= prog
->_LinkedShaders
[i
]->ir
;
2480 const struct gl_shader_compiler_options
*options
=
2481 &ctx
->ShaderCompilerOptions
[_mesa_shader_type_to_index(prog
->_LinkedShaders
[i
]->Type
)];
2487 do_mat_op_to_vec(ir
);
2488 do_mod_to_fract(ir
);
2489 do_div_to_mul_rcp(ir
);
2490 do_explog_to_explog2(ir
);
2492 progress
= do_lower_jumps(ir
, true, true, options
->EmitNoMainReturn
, options
->EmitNoCont
, options
->EmitNoLoops
) || progress
;
2494 progress
= do_common_optimization(ir
, true, options
->MaxUnrollIterations
) || progress
;
2496 if (options
->EmitNoIfs
)
2497 progress
= do_if_to_cond_assign(ir
) || progress
;
2499 if (options
->EmitNoNoise
)
2500 progress
= lower_noise(ir
) || progress
;
2502 /* If there are forms of indirect addressing that the driver
2503 * cannot handle, perform the lowering pass.
2505 if (options
->EmitNoIndirectInput
|| options
->EmitNoIndirectOutput
2506 || options
->EmitNoIndirectTemp
|| options
->EmitNoIndirectUniform
)
2508 lower_variable_index_to_cond_assign(ir
,
2509 options
->EmitNoIndirectInput
,
2510 options
->EmitNoIndirectOutput
,
2511 options
->EmitNoIndirectTemp
,
2512 options
->EmitNoIndirectUniform
)
2515 progress
= do_vec_index_to_cond_assign(ir
) || progress
;
2518 validate_ir_tree(ir
);
2521 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2522 struct gl_program
*linked_prog
;
2525 if (prog
->_LinkedShaders
[i
] == NULL
)
2528 linked_prog
= get_mesa_program(ctx
, prog
, prog
->_LinkedShaders
[i
]);
2530 switch (prog
->_LinkedShaders
[i
]->Type
) {
2531 case GL_VERTEX_SHADER
:
2532 _mesa_reference_vertprog(ctx
, &prog
->VertexProgram
,
2533 (struct gl_vertex_program
*)linked_prog
);
2534 ok
= ctx
->Driver
.ProgramStringNotify(ctx
, GL_VERTEX_PROGRAM_ARB
,
2537 case GL_FRAGMENT_SHADER
:
2538 _mesa_reference_fragprog(ctx
, &prog
->FragmentProgram
,
2539 (struct gl_fragment_program
*)linked_prog
);
2540 ok
= ctx
->Driver
.ProgramStringNotify(ctx
, GL_FRAGMENT_PROGRAM_ARB
,
2547 _mesa_reference_program(ctx
, &linked_prog
, NULL
);
2555 * Compile a GLSL shader. Called via glCompileShader().
2558 _mesa_glsl_compile_shader(struct gl_context
*ctx
, struct gl_shader
*shader
)
2560 struct _mesa_glsl_parse_state
*state
=
2561 new(shader
) _mesa_glsl_parse_state(ctx
, shader
->Type
, shader
);
2563 const char *source
= shader
->Source
;
2564 /* Check if the user called glCompileShader without first calling
2565 * glShaderSource. This should fail to compile, but not raise a GL_ERROR.
2567 if (source
== NULL
) {
2568 shader
->CompileStatus
= GL_FALSE
;
2572 state
->error
= preprocess(state
, &source
, &state
->info_log
,
2573 &ctx
->Extensions
, ctx
->API
);
2575 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2576 printf("GLSL source for shader %d:\n", shader
->Name
);
2577 printf("%s\n", shader
->Source
);
2580 if (!state
->error
) {
2581 _mesa_glsl_lexer_ctor(state
, source
);
2582 _mesa_glsl_parse(state
);
2583 _mesa_glsl_lexer_dtor(state
);
2586 talloc_free(shader
->ir
);
2587 shader
->ir
= new(shader
) exec_list
;
2588 if (!state
->error
&& !state
->translation_unit
.is_empty())
2589 _mesa_ast_to_hir(shader
->ir
, state
);
2591 if (!state
->error
&& !shader
->ir
->is_empty()) {
2592 validate_ir_tree(shader
->ir
);
2594 /* Do some optimization at compile time to reduce shader IR size
2595 * and reduce later work if the same shader is linked multiple times
2597 while (do_common_optimization(shader
->ir
, false, 32))
2600 validate_ir_tree(shader
->ir
);
2603 shader
->symbols
= state
->symbols
;
2605 shader
->CompileStatus
= !state
->error
;
2606 shader
->InfoLog
= state
->info_log
;
2607 shader
->Version
= state
->language_version
;
2608 memcpy(shader
->builtins_to_link
, state
->builtins_to_link
,
2609 sizeof(shader
->builtins_to_link
[0]) * state
->num_builtins_to_link
);
2610 shader
->num_builtins_to_link
= state
->num_builtins_to_link
;
2612 if (ctx
->Shader
.Flags
& GLSL_LOG
) {
2613 _mesa_write_shader_to_file(shader
);
2616 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2617 if (shader
->CompileStatus
) {
2618 printf("GLSL IR for shader %d:\n", shader
->Name
);
2619 _mesa_print_ir(shader
->ir
, NULL
);
2622 printf("GLSL shader %d failed to compile.\n", shader
->Name
);
2624 if (shader
->InfoLog
&& shader
->InfoLog
[0] != 0) {
2625 printf("GLSL shader %d info log:\n", shader
->Name
);
2626 printf("%s\n", shader
->InfoLog
);
2630 /* Retain any live IR, but trash the rest. */
2631 reparent_ir(shader
->ir
, shader
->ir
);
2635 if (shader
->CompileStatus
) {
2636 if (!ctx
->Driver
.CompileShader(ctx
, shader
))
2637 shader
->CompileStatus
= GL_FALSE
;
2643 * Link a GLSL shader program. Called via glLinkProgram().
2646 _mesa_glsl_link_shader(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2650 _mesa_clear_shader_program_data(ctx
, prog
);
2652 prog
->LinkStatus
= GL_TRUE
;
2654 for (i
= 0; i
< prog
->NumShaders
; i
++) {
2655 if (!prog
->Shaders
[i
]->CompileStatus
) {
2656 fail_link(prog
, "linking with uncompiled shader");
2657 prog
->LinkStatus
= GL_FALSE
;
2661 prog
->Varying
= _mesa_new_parameter_list();
2662 _mesa_reference_vertprog(ctx
, &prog
->VertexProgram
, NULL
);
2663 _mesa_reference_fragprog(ctx
, &prog
->FragmentProgram
, NULL
);
2665 if (prog
->LinkStatus
) {
2666 link_shaders(ctx
, prog
);
2669 if (prog
->LinkStatus
) {
2670 if (!ctx
->Driver
.LinkShader(ctx
, prog
)) {
2671 prog
->LinkStatus
= GL_FALSE
;
2675 set_uniform_initializers(ctx
, prog
);
2677 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2678 if (!prog
->LinkStatus
) {
2679 printf("GLSL shader program %d failed to link\n", prog
->Name
);
2682 if (prog
->InfoLog
&& prog
->InfoLog
[0] != 0) {
2683 printf("GLSL shader program %d info log:\n", prog
->Name
);
2684 printf("%s\n", prog
->InfoLog
);