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 assert(!"GLSL 1.30 features unsupported");
1137 this->result
= result_src
;
1142 ir_to_mesa_visitor::visit(ir_swizzle
*ir
)
1144 ir_to_mesa_src_reg src_reg
;
1148 /* Note that this is only swizzles in expressions, not those on the left
1149 * hand side of an assignment, which do write masking. See ir_assignment
1153 ir
->val
->accept(this);
1154 src_reg
= this->result
;
1155 assert(src_reg
.file
!= PROGRAM_UNDEFINED
);
1157 for (i
= 0; i
< 4; i
++) {
1158 if (i
< ir
->type
->vector_elements
) {
1161 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.x
);
1164 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.y
);
1167 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.z
);
1170 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.w
);
1174 /* If the type is smaller than a vec4, replicate the last
1177 swizzle
[i
] = swizzle
[ir
->type
->vector_elements
- 1];
1181 src_reg
.swizzle
= MAKE_SWIZZLE4(swizzle
[0],
1186 this->result
= src_reg
;
1190 ir_to_mesa_visitor::visit(ir_dereference_variable
*ir
)
1192 variable_storage
*entry
= find_variable_storage(ir
->var
);
1195 switch (ir
->var
->mode
) {
1196 case ir_var_uniform
:
1197 entry
= new(mem_ctx
) variable_storage(ir
->var
, PROGRAM_UNIFORM
,
1199 this->variables
.push_tail(entry
);
1204 /* The linker assigns locations for varyings and attributes,
1205 * including deprecated builtins (like gl_Color), user-assign
1206 * generic attributes (glBindVertexLocation), and
1207 * user-defined varyings.
1209 * FINISHME: We would hit this path for function arguments. Fix!
1211 assert(ir
->var
->location
!= -1);
1212 if (ir
->var
->mode
== ir_var_in
||
1213 ir
->var
->mode
== ir_var_inout
) {
1214 entry
= new(mem_ctx
) variable_storage(ir
->var
,
1218 if (this->prog
->Target
== GL_VERTEX_PROGRAM_ARB
&&
1219 ir
->var
->location
>= VERT_ATTRIB_GENERIC0
) {
1220 _mesa_add_attribute(prog
->Attributes
,
1222 _mesa_sizeof_glsl_type(ir
->var
->type
->gl_type
),
1223 ir
->var
->type
->gl_type
,
1224 ir
->var
->location
- VERT_ATTRIB_GENERIC0
);
1227 entry
= new(mem_ctx
) variable_storage(ir
->var
,
1234 case ir_var_temporary
:
1235 entry
= new(mem_ctx
) variable_storage(ir
->var
, PROGRAM_TEMPORARY
,
1237 this->variables
.push_tail(entry
);
1239 next_temp
+= type_size(ir
->var
->type
);
1244 printf("Failed to make storage for %s\n", ir
->var
->name
);
1249 this->result
= ir_to_mesa_src_reg(entry
->file
, entry
->index
, ir
->var
->type
);
1253 ir_to_mesa_visitor::visit(ir_dereference_array
*ir
)
1256 ir_to_mesa_src_reg src_reg
;
1257 int element_size
= type_size(ir
->type
);
1259 index
= ir
->array_index
->constant_expression_value();
1261 ir
->array
->accept(this);
1262 src_reg
= this->result
;
1265 src_reg
.index
+= index
->value
.i
[0] * element_size
;
1267 ir_to_mesa_src_reg array_base
= this->result
;
1268 /* Variable index array dereference. It eats the "vec4" of the
1269 * base of the array and an index that offsets the Mesa register
1272 ir
->array_index
->accept(this);
1274 ir_to_mesa_src_reg index_reg
;
1276 if (element_size
== 1) {
1277 index_reg
= this->result
;
1279 index_reg
= get_temp(glsl_type::float_type
);
1281 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
,
1282 ir_to_mesa_dst_reg_from_src(index_reg
),
1283 this->result
, src_reg_for_float(element_size
));
1286 src_reg
.reladdr
= talloc(mem_ctx
, ir_to_mesa_src_reg
);
1287 memcpy(src_reg
.reladdr
, &index_reg
, sizeof(index_reg
));
1290 /* If the type is smaller than a vec4, replicate the last channel out. */
1291 if (ir
->type
->is_scalar() || ir
->type
->is_vector())
1292 src_reg
.swizzle
= swizzle_for_size(ir
->type
->vector_elements
);
1294 src_reg
.swizzle
= SWIZZLE_NOOP
;
1296 this->result
= src_reg
;
1300 ir_to_mesa_visitor::visit(ir_dereference_record
*ir
)
1303 const glsl_type
*struct_type
= ir
->record
->type
;
1306 ir
->record
->accept(this);
1308 for (i
= 0; i
< struct_type
->length
; i
++) {
1309 if (strcmp(struct_type
->fields
.structure
[i
].name
, ir
->field
) == 0)
1311 offset
+= type_size(struct_type
->fields
.structure
[i
].type
);
1313 this->result
.swizzle
= swizzle_for_size(ir
->type
->vector_elements
);
1314 this->result
.index
+= offset
;
1318 * We want to be careful in assignment setup to hit the actual storage
1319 * instead of potentially using a temporary like we might with the
1320 * ir_dereference handler.
1322 static struct ir_to_mesa_dst_reg
1323 get_assignment_lhs(ir_dereference
*ir
, ir_to_mesa_visitor
*v
)
1325 /* The LHS must be a dereference. If the LHS is a variable indexed array
1326 * access of a vector, it must be separated into a series conditional moves
1327 * before reaching this point (see ir_vec_index_to_cond_assign).
1329 assert(ir
->as_dereference());
1330 ir_dereference_array
*deref_array
= ir
->as_dereference_array();
1332 assert(!deref_array
->array
->type
->is_vector());
1335 /* Use the rvalue deref handler for the most part. We'll ignore
1336 * swizzles in it and write swizzles using writemask, though.
1339 return ir_to_mesa_dst_reg_from_src(v
->result
);
1343 ir_to_mesa_visitor::visit(ir_assignment
*ir
)
1345 struct ir_to_mesa_dst_reg l
;
1346 struct ir_to_mesa_src_reg r
;
1349 ir
->rhs
->accept(this);
1352 l
= get_assignment_lhs(ir
->lhs
, this);
1354 /* FINISHME: This should really set to the correct maximal writemask for each
1355 * FINISHME: component written (in the loops below). This case can only
1356 * FINISHME: occur for matrices, arrays, and structures.
1358 if (ir
->write_mask
== 0) {
1359 assert(!ir
->lhs
->type
->is_scalar() && !ir
->lhs
->type
->is_vector());
1360 l
.writemask
= WRITEMASK_XYZW
;
1361 } else if (ir
->lhs
->type
->is_scalar()) {
1362 /* FINISHME: This hack makes writing to gl_FragDepth, which lives in the
1363 * FINISHME: W component of fragment shader output zero, work correctly.
1365 l
.writemask
= WRITEMASK_XYZW
;
1368 int first_enabled_chan
= 0;
1371 assert(ir
->lhs
->type
->is_vector());
1372 l
.writemask
= ir
->write_mask
;
1374 for (int i
= 0; i
< 4; i
++) {
1375 if (l
.writemask
& (1 << i
)) {
1376 first_enabled_chan
= GET_SWZ(r
.swizzle
, i
);
1381 /* Swizzle a small RHS vector into the channels being written.
1383 * glsl ir treats write_mask as dictating how many channels are
1384 * present on the RHS while Mesa IR treats write_mask as just
1385 * showing which channels of the vec4 RHS get written.
1387 for (int i
= 0; i
< 4; i
++) {
1388 if (l
.writemask
& (1 << i
))
1389 swizzles
[i
] = GET_SWZ(r
.swizzle
, rhs_chan
++);
1391 swizzles
[i
] = first_enabled_chan
;
1393 r
.swizzle
= MAKE_SWIZZLE4(swizzles
[0], swizzles
[1],
1394 swizzles
[2], swizzles
[3]);
1397 assert(l
.file
!= PROGRAM_UNDEFINED
);
1398 assert(r
.file
!= PROGRAM_UNDEFINED
);
1400 if (ir
->condition
) {
1401 ir_to_mesa_src_reg condition
;
1403 ir
->condition
->accept(this);
1404 condition
= this->result
;
1406 /* We use the OPCODE_CMP (a < 0 ? b : c) for conditional moves,
1407 * and the condition we produced is 0.0 or 1.0. By flipping the
1408 * sign, we can choose which value OPCODE_CMP produces without
1409 * an extra computing the condition.
1411 condition
.negate
= ~condition
.negate
;
1412 for (i
= 0; i
< type_size(ir
->lhs
->type
); i
++) {
1413 ir_to_mesa_emit_op3(ir
, OPCODE_CMP
, l
,
1414 condition
, r
, ir_to_mesa_src_reg_from_dst(l
));
1419 for (i
= 0; i
< type_size(ir
->lhs
->type
); i
++) {
1420 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1429 ir_to_mesa_visitor::visit(ir_constant
*ir
)
1431 ir_to_mesa_src_reg src_reg
;
1432 GLfloat stack_vals
[4] = { 0 };
1433 GLfloat
*values
= stack_vals
;
1436 /* Unfortunately, 4 floats is all we can get into
1437 * _mesa_add_unnamed_constant. So, make a temp to store an
1438 * aggregate constant and move each constant value into it. If we
1439 * get lucky, copy propagation will eliminate the extra moves.
1442 if (ir
->type
->base_type
== GLSL_TYPE_STRUCT
) {
1443 ir_to_mesa_src_reg temp_base
= get_temp(ir
->type
);
1444 ir_to_mesa_dst_reg temp
= ir_to_mesa_dst_reg_from_src(temp_base
);
1446 foreach_iter(exec_list_iterator
, iter
, ir
->components
) {
1447 ir_constant
*field_value
= (ir_constant
*)iter
.get();
1448 int size
= type_size(field_value
->type
);
1452 field_value
->accept(this);
1453 src_reg
= this->result
;
1455 for (i
= 0; i
< (unsigned int)size
; i
++) {
1456 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, temp
, src_reg
);
1462 this->result
= temp_base
;
1466 if (ir
->type
->is_array()) {
1467 ir_to_mesa_src_reg temp_base
= get_temp(ir
->type
);
1468 ir_to_mesa_dst_reg temp
= ir_to_mesa_dst_reg_from_src(temp_base
);
1469 int size
= type_size(ir
->type
->fields
.array
);
1473 for (i
= 0; i
< ir
->type
->length
; i
++) {
1474 ir
->array_elements
[i
]->accept(this);
1475 src_reg
= this->result
;
1476 for (int j
= 0; j
< size
; j
++) {
1477 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, temp
, src_reg
);
1483 this->result
= temp_base
;
1487 if (ir
->type
->is_matrix()) {
1488 ir_to_mesa_src_reg mat
= get_temp(ir
->type
);
1489 ir_to_mesa_dst_reg mat_column
= ir_to_mesa_dst_reg_from_src(mat
);
1491 for (i
= 0; i
< ir
->type
->matrix_columns
; i
++) {
1492 assert(ir
->type
->base_type
== GLSL_TYPE_FLOAT
);
1493 values
= &ir
->value
.f
[i
* ir
->type
->vector_elements
];
1495 src_reg
= ir_to_mesa_src_reg(PROGRAM_CONSTANT
, -1, NULL
);
1496 src_reg
.index
= _mesa_add_unnamed_constant(this->prog
->Parameters
,
1498 ir
->type
->vector_elements
,
1500 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, mat_column
, src_reg
);
1509 src_reg
.file
= PROGRAM_CONSTANT
;
1510 switch (ir
->type
->base_type
) {
1511 case GLSL_TYPE_FLOAT
:
1512 values
= &ir
->value
.f
[0];
1514 case GLSL_TYPE_UINT
:
1515 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1516 values
[i
] = ir
->value
.u
[i
];
1520 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1521 values
[i
] = ir
->value
.i
[i
];
1524 case GLSL_TYPE_BOOL
:
1525 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1526 values
[i
] = ir
->value
.b
[i
];
1530 assert(!"Non-float/uint/int/bool constant");
1533 this->result
= ir_to_mesa_src_reg(PROGRAM_CONSTANT
, -1, ir
->type
);
1534 this->result
.index
= _mesa_add_unnamed_constant(this->prog
->Parameters
,
1536 ir
->type
->vector_elements
,
1537 &this->result
.swizzle
);
1541 ir_to_mesa_visitor::get_function_signature(ir_function_signature
*sig
)
1543 function_entry
*entry
;
1545 foreach_iter(exec_list_iterator
, iter
, this->function_signatures
) {
1546 entry
= (function_entry
*)iter
.get();
1548 if (entry
->sig
== sig
)
1552 entry
= talloc(mem_ctx
, function_entry
);
1554 entry
->sig_id
= this->next_signature_id
++;
1555 entry
->bgn_inst
= NULL
;
1557 /* Allocate storage for all the parameters. */
1558 foreach_iter(exec_list_iterator
, iter
, sig
->parameters
) {
1559 ir_variable
*param
= (ir_variable
*)iter
.get();
1560 variable_storage
*storage
;
1562 storage
= find_variable_storage(param
);
1565 storage
= new(mem_ctx
) variable_storage(param
, PROGRAM_TEMPORARY
,
1567 this->variables
.push_tail(storage
);
1569 this->next_temp
+= type_size(param
->type
);
1572 if (!sig
->return_type
->is_void()) {
1573 entry
->return_reg
= get_temp(sig
->return_type
);
1575 entry
->return_reg
= ir_to_mesa_undef
;
1578 this->function_signatures
.push_tail(entry
);
1583 ir_to_mesa_visitor::visit(ir_call
*ir
)
1585 ir_to_mesa_instruction
*call_inst
;
1586 ir_function_signature
*sig
= ir
->get_callee();
1587 function_entry
*entry
= get_function_signature(sig
);
1590 /* Process in parameters. */
1591 exec_list_iterator sig_iter
= sig
->parameters
.iterator();
1592 foreach_iter(exec_list_iterator
, iter
, *ir
) {
1593 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
1594 ir_variable
*param
= (ir_variable
*)sig_iter
.get();
1596 if (param
->mode
== ir_var_in
||
1597 param
->mode
== ir_var_inout
) {
1598 variable_storage
*storage
= find_variable_storage(param
);
1601 param_rval
->accept(this);
1602 ir_to_mesa_src_reg r
= this->result
;
1604 ir_to_mesa_dst_reg l
;
1605 l
.file
= storage
->file
;
1606 l
.index
= storage
->index
;
1608 l
.writemask
= WRITEMASK_XYZW
;
1609 l
.cond_mask
= COND_TR
;
1611 for (i
= 0; i
< type_size(param
->type
); i
++) {
1612 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1620 assert(!sig_iter
.has_next());
1622 /* Emit call instruction */
1623 call_inst
= ir_to_mesa_emit_op1(ir
, OPCODE_CAL
,
1624 ir_to_mesa_undef_dst
, ir_to_mesa_undef
);
1625 call_inst
->function
= entry
;
1627 /* Process out parameters. */
1628 sig_iter
= sig
->parameters
.iterator();
1629 foreach_iter(exec_list_iterator
, iter
, *ir
) {
1630 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
1631 ir_variable
*param
= (ir_variable
*)sig_iter
.get();
1633 if (param
->mode
== ir_var_out
||
1634 param
->mode
== ir_var_inout
) {
1635 variable_storage
*storage
= find_variable_storage(param
);
1638 ir_to_mesa_src_reg r
;
1639 r
.file
= storage
->file
;
1640 r
.index
= storage
->index
;
1642 r
.swizzle
= SWIZZLE_NOOP
;
1645 param_rval
->accept(this);
1646 ir_to_mesa_dst_reg l
= ir_to_mesa_dst_reg_from_src(this->result
);
1648 for (i
= 0; i
< type_size(param
->type
); i
++) {
1649 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1657 assert(!sig_iter
.has_next());
1659 /* Process return value. */
1660 this->result
= entry
->return_reg
;
1664 ir_to_mesa_visitor::visit(ir_texture
*ir
)
1666 ir_to_mesa_src_reg result_src
, coord
, lod_info
, projector
;
1667 ir_to_mesa_dst_reg result_dst
, coord_dst
;
1668 ir_to_mesa_instruction
*inst
= NULL
;
1669 prog_opcode opcode
= OPCODE_NOP
;
1671 ir
->coordinate
->accept(this);
1673 /* Put our coords in a temp. We'll need to modify them for shadow,
1674 * projection, or LOD, so the only case we'd use it as is is if
1675 * we're doing plain old texturing. Mesa IR optimization should
1676 * handle cleaning up our mess in that case.
1678 coord
= get_temp(glsl_type::vec4_type
);
1679 coord_dst
= ir_to_mesa_dst_reg_from_src(coord
);
1680 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
,
1683 if (ir
->projector
) {
1684 ir
->projector
->accept(this);
1685 projector
= this->result
;
1688 /* Storage for our result. Ideally for an assignment we'd be using
1689 * the actual storage for the result here, instead.
1691 result_src
= get_temp(glsl_type::vec4_type
);
1692 result_dst
= ir_to_mesa_dst_reg_from_src(result_src
);
1696 opcode
= OPCODE_TEX
;
1699 opcode
= OPCODE_TXB
;
1700 ir
->lod_info
.bias
->accept(this);
1701 lod_info
= this->result
;
1704 opcode
= OPCODE_TXL
;
1705 ir
->lod_info
.lod
->accept(this);
1706 lod_info
= this->result
;
1710 assert(!"GLSL 1.30 features unsupported");
1714 if (ir
->projector
) {
1715 if (opcode
== OPCODE_TEX
) {
1716 /* Slot the projector in as the last component of the coord. */
1717 coord_dst
.writemask
= WRITEMASK_W
;
1718 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, projector
);
1719 coord_dst
.writemask
= WRITEMASK_XYZW
;
1720 opcode
= OPCODE_TXP
;
1722 ir_to_mesa_src_reg coord_w
= coord
;
1723 coord_w
.swizzle
= SWIZZLE_WWWW
;
1725 /* For the other TEX opcodes there's no projective version
1726 * since the last slot is taken up by lod info. Do the
1727 * projective divide now.
1729 coord_dst
.writemask
= WRITEMASK_W
;
1730 ir_to_mesa_emit_op1(ir
, OPCODE_RCP
, coord_dst
, projector
);
1732 coord_dst
.writemask
= WRITEMASK_XYZ
;
1733 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
, coord_dst
, coord
, coord_w
);
1735 coord_dst
.writemask
= WRITEMASK_XYZW
;
1736 coord
.swizzle
= SWIZZLE_XYZW
;
1740 if (ir
->shadow_comparitor
) {
1741 /* Slot the shadow value in as the second to last component of the
1744 ir
->shadow_comparitor
->accept(this);
1745 coord_dst
.writemask
= WRITEMASK_Z
;
1746 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, this->result
);
1747 coord_dst
.writemask
= WRITEMASK_XYZW
;
1750 if (opcode
== OPCODE_TXL
|| opcode
== OPCODE_TXB
) {
1751 /* Mesa IR stores lod or lod bias in the last channel of the coords. */
1752 coord_dst
.writemask
= WRITEMASK_W
;
1753 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, lod_info
);
1754 coord_dst
.writemask
= WRITEMASK_XYZW
;
1757 inst
= ir_to_mesa_emit_op1(ir
, opcode
, result_dst
, coord
);
1759 if (ir
->shadow_comparitor
)
1760 inst
->tex_shadow
= GL_TRUE
;
1762 inst
->sampler
= _mesa_get_sampler_uniform_value(ir
->sampler
,
1763 this->shader_program
,
1766 const glsl_type
*sampler_type
= ir
->sampler
->type
;
1768 switch (sampler_type
->sampler_dimensionality
) {
1769 case GLSL_SAMPLER_DIM_1D
:
1770 inst
->tex_target
= (sampler_type
->sampler_array
)
1771 ? TEXTURE_1D_ARRAY_INDEX
: TEXTURE_1D_INDEX
;
1773 case GLSL_SAMPLER_DIM_2D
:
1774 inst
->tex_target
= (sampler_type
->sampler_array
)
1775 ? TEXTURE_2D_ARRAY_INDEX
: TEXTURE_2D_INDEX
;
1777 case GLSL_SAMPLER_DIM_3D
:
1778 inst
->tex_target
= TEXTURE_3D_INDEX
;
1780 case GLSL_SAMPLER_DIM_CUBE
:
1781 inst
->tex_target
= TEXTURE_CUBE_INDEX
;
1783 case GLSL_SAMPLER_DIM_RECT
:
1784 inst
->tex_target
= TEXTURE_RECT_INDEX
;
1786 case GLSL_SAMPLER_DIM_BUF
:
1787 assert(!"FINISHME: Implement ARB_texture_buffer_object");
1790 assert(!"Should not get here.");
1793 this->result
= result_src
;
1797 ir_to_mesa_visitor::visit(ir_return
*ir
)
1799 if (ir
->get_value()) {
1800 ir_to_mesa_dst_reg l
;
1803 assert(current_function
);
1805 ir
->get_value()->accept(this);
1806 ir_to_mesa_src_reg r
= this->result
;
1808 l
= ir_to_mesa_dst_reg_from_src(current_function
->return_reg
);
1810 for (i
= 0; i
< type_size(current_function
->sig
->return_type
); i
++) {
1811 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1817 ir_to_mesa_emit_op0(ir
, OPCODE_RET
);
1821 ir_to_mesa_visitor::visit(ir_discard
*ir
)
1823 struct gl_fragment_program
*fp
= (struct gl_fragment_program
*)this->prog
;
1825 assert(ir
->condition
== NULL
); /* FINISHME */
1827 ir_to_mesa_emit_op0(ir
, OPCODE_KIL_NV
);
1828 fp
->UsesKill
= GL_TRUE
;
1832 ir_to_mesa_visitor::visit(ir_if
*ir
)
1834 ir_to_mesa_instruction
*cond_inst
, *if_inst
, *else_inst
= NULL
;
1835 ir_to_mesa_instruction
*prev_inst
;
1837 prev_inst
= (ir_to_mesa_instruction
*)this->instructions
.get_tail();
1839 ir
->condition
->accept(this);
1840 assert(this->result
.file
!= PROGRAM_UNDEFINED
);
1842 if (this->options
->EmitCondCodes
) {
1843 cond_inst
= (ir_to_mesa_instruction
*)this->instructions
.get_tail();
1845 /* See if we actually generated any instruction for generating
1846 * the condition. If not, then cook up a move to a temp so we
1847 * have something to set cond_update on.
1849 if (cond_inst
== prev_inst
) {
1850 ir_to_mesa_src_reg temp
= get_temp(glsl_type::bool_type
);
1851 cond_inst
= ir_to_mesa_emit_op1(ir
->condition
, OPCODE_MOV
,
1852 ir_to_mesa_dst_reg_from_src(temp
),
1855 cond_inst
->cond_update
= GL_TRUE
;
1857 if_inst
= ir_to_mesa_emit_op0(ir
->condition
, OPCODE_IF
);
1858 if_inst
->dst_reg
.cond_mask
= COND_NE
;
1860 if_inst
= ir_to_mesa_emit_op1(ir
->condition
,
1861 OPCODE_IF
, ir_to_mesa_undef_dst
,
1865 this->instructions
.push_tail(if_inst
);
1867 visit_exec_list(&ir
->then_instructions
, this);
1869 if (!ir
->else_instructions
.is_empty()) {
1870 else_inst
= ir_to_mesa_emit_op0(ir
->condition
, OPCODE_ELSE
);
1871 visit_exec_list(&ir
->else_instructions
, this);
1874 if_inst
= ir_to_mesa_emit_op1(ir
->condition
, OPCODE_ENDIF
,
1875 ir_to_mesa_undef_dst
, ir_to_mesa_undef
);
1878 ir_to_mesa_visitor::ir_to_mesa_visitor()
1880 result
.file
= PROGRAM_UNDEFINED
;
1882 next_signature_id
= 1;
1883 current_function
= NULL
;
1884 mem_ctx
= talloc_new(NULL
);
1887 ir_to_mesa_visitor::~ir_to_mesa_visitor()
1889 talloc_free(mem_ctx
);
1892 static struct prog_src_register
1893 mesa_src_reg_from_ir_src_reg(ir_to_mesa_src_reg reg
)
1895 struct prog_src_register mesa_reg
;
1897 mesa_reg
.File
= reg
.file
;
1898 assert(reg
.index
< (1 << INST_INDEX_BITS
) - 1);
1899 mesa_reg
.Index
= reg
.index
;
1900 mesa_reg
.Swizzle
= reg
.swizzle
;
1901 mesa_reg
.RelAddr
= reg
.reladdr
!= NULL
;
1902 mesa_reg
.Negate
= reg
.negate
;
1904 mesa_reg
.HasIndex2
= GL_FALSE
;
1905 mesa_reg
.RelAddr2
= 0;
1906 mesa_reg
.Index2
= 0;
1912 set_branchtargets(ir_to_mesa_visitor
*v
,
1913 struct prog_instruction
*mesa_instructions
,
1914 int num_instructions
)
1916 int if_count
= 0, loop_count
= 0;
1917 int *if_stack
, *loop_stack
;
1918 int if_stack_pos
= 0, loop_stack_pos
= 0;
1921 for (i
= 0; i
< num_instructions
; i
++) {
1922 switch (mesa_instructions
[i
].Opcode
) {
1926 case OPCODE_BGNLOOP
:
1931 mesa_instructions
[i
].BranchTarget
= -1;
1938 if_stack
= talloc_zero_array(v
->mem_ctx
, int, if_count
);
1939 loop_stack
= talloc_zero_array(v
->mem_ctx
, int, loop_count
);
1941 for (i
= 0; i
< num_instructions
; i
++) {
1942 switch (mesa_instructions
[i
].Opcode
) {
1944 if_stack
[if_stack_pos
] = i
;
1948 mesa_instructions
[if_stack
[if_stack_pos
- 1]].BranchTarget
= i
;
1949 if_stack
[if_stack_pos
- 1] = i
;
1952 mesa_instructions
[if_stack
[if_stack_pos
- 1]].BranchTarget
= i
;
1955 case OPCODE_BGNLOOP
:
1956 loop_stack
[loop_stack_pos
] = i
;
1959 case OPCODE_ENDLOOP
:
1961 /* Rewrite any breaks/conts at this nesting level (haven't
1962 * already had a BranchTarget assigned) to point to the end
1965 for (j
= loop_stack
[loop_stack_pos
]; j
< i
; j
++) {
1966 if (mesa_instructions
[j
].Opcode
== OPCODE_BRK
||
1967 mesa_instructions
[j
].Opcode
== OPCODE_CONT
) {
1968 if (mesa_instructions
[j
].BranchTarget
== -1) {
1969 mesa_instructions
[j
].BranchTarget
= i
;
1973 /* The loop ends point at each other. */
1974 mesa_instructions
[i
].BranchTarget
= loop_stack
[loop_stack_pos
];
1975 mesa_instructions
[loop_stack
[loop_stack_pos
]].BranchTarget
= i
;
1978 foreach_iter(exec_list_iterator
, iter
, v
->function_signatures
) {
1979 function_entry
*entry
= (function_entry
*)iter
.get();
1981 if (entry
->sig_id
== mesa_instructions
[i
].BranchTarget
) {
1982 mesa_instructions
[i
].BranchTarget
= entry
->inst
;
1994 print_program(struct prog_instruction
*mesa_instructions
,
1995 ir_instruction
**mesa_instruction_annotation
,
1996 int num_instructions
)
1998 ir_instruction
*last_ir
= NULL
;
2002 for (i
= 0; i
< num_instructions
; i
++) {
2003 struct prog_instruction
*mesa_inst
= mesa_instructions
+ i
;
2004 ir_instruction
*ir
= mesa_instruction_annotation
[i
];
2006 fprintf(stdout
, "%3d: ", i
);
2008 if (last_ir
!= ir
&& ir
) {
2011 for (j
= 0; j
< indent
; j
++) {
2012 fprintf(stdout
, " ");
2018 fprintf(stdout
, " "); /* line number spacing. */
2021 indent
= _mesa_fprint_instruction_opt(stdout
, mesa_inst
, indent
,
2022 PROG_PRINT_DEBUG
, NULL
);
2027 count_resources(struct gl_program
*prog
)
2031 prog
->SamplersUsed
= 0;
2033 for (i
= 0; i
< prog
->NumInstructions
; i
++) {
2034 struct prog_instruction
*inst
= &prog
->Instructions
[i
];
2036 if (_mesa_is_tex_instruction(inst
->Opcode
)) {
2037 prog
->SamplerTargets
[inst
->TexSrcUnit
] =
2038 (gl_texture_index
)inst
->TexSrcTarget
;
2039 prog
->SamplersUsed
|= 1 << inst
->TexSrcUnit
;
2040 if (inst
->TexShadow
) {
2041 prog
->ShadowSamplers
|= 1 << inst
->TexSrcUnit
;
2046 _mesa_update_shader_textures_used(prog
);
2049 struct uniform_sort
{
2050 struct gl_uniform
*u
;
2054 /* The shader_program->Uniforms list is almost sorted in increasing
2055 * uniform->{Frag,Vert}Pos locations, but not quite when there are
2056 * uniforms shared between targets. We need to add parameters in
2057 * increasing order for the targets.
2060 sort_uniforms(const void *a
, const void *b
)
2062 struct uniform_sort
*u1
= (struct uniform_sort
*)a
;
2063 struct uniform_sort
*u2
= (struct uniform_sort
*)b
;
2065 return u1
->pos
- u2
->pos
;
2068 /* Add the uniforms to the parameters. The linker chose locations
2069 * in our parameters lists (which weren't created yet), which the
2070 * uniforms code will use to poke values into our parameters list
2071 * when uniforms are updated.
2074 add_uniforms_to_parameters_list(struct gl_shader_program
*shader_program
,
2075 struct gl_shader
*shader
,
2076 struct gl_program
*prog
)
2079 unsigned int next_sampler
= 0, num_uniforms
= 0;
2080 struct uniform_sort
*sorted_uniforms
;
2082 sorted_uniforms
= talloc_array(NULL
, struct uniform_sort
,
2083 shader_program
->Uniforms
->NumUniforms
);
2085 for (i
= 0; i
< shader_program
->Uniforms
->NumUniforms
; i
++) {
2086 struct gl_uniform
*uniform
= shader_program
->Uniforms
->Uniforms
+ i
;
2087 int parameter_index
= -1;
2089 switch (shader
->Type
) {
2090 case GL_VERTEX_SHADER
:
2091 parameter_index
= uniform
->VertPos
;
2093 case GL_FRAGMENT_SHADER
:
2094 parameter_index
= uniform
->FragPos
;
2096 case GL_GEOMETRY_SHADER
:
2097 parameter_index
= uniform
->GeomPos
;
2101 /* Only add uniforms used in our target. */
2102 if (parameter_index
!= -1) {
2103 sorted_uniforms
[num_uniforms
].pos
= parameter_index
;
2104 sorted_uniforms
[num_uniforms
].u
= uniform
;
2109 qsort(sorted_uniforms
, num_uniforms
, sizeof(struct uniform_sort
),
2112 for (i
= 0; i
< num_uniforms
; i
++) {
2113 struct gl_uniform
*uniform
= sorted_uniforms
[i
].u
;
2114 int parameter_index
= sorted_uniforms
[i
].pos
;
2115 const glsl_type
*type
= uniform
->Type
;
2118 if (type
->is_vector() ||
2119 type
->is_scalar()) {
2120 size
= type
->vector_elements
;
2122 size
= type_size(type
) * 4;
2125 gl_register_file file
;
2126 if (type
->is_sampler() ||
2127 (type
->is_array() && type
->fields
.array
->is_sampler())) {
2128 file
= PROGRAM_SAMPLER
;
2130 file
= PROGRAM_UNIFORM
;
2133 GLint index
= _mesa_lookup_parameter_index(prog
->Parameters
, -1,
2137 index
= _mesa_add_parameter(prog
->Parameters
, file
,
2138 uniform
->Name
, size
, type
->gl_type
,
2141 /* Sampler uniform values are stored in prog->SamplerUnits,
2142 * and the entry in that array is selected by this index we
2143 * store in ParameterValues[].
2145 if (file
== PROGRAM_SAMPLER
) {
2146 for (unsigned int j
= 0; j
< size
/ 4; j
++)
2147 prog
->Parameters
->ParameterValues
[index
+ j
][0] = next_sampler
++;
2150 /* The location chosen in the Parameters list here (returned
2151 * from _mesa_add_uniform) has to match what the linker chose.
2153 if (index
!= parameter_index
) {
2154 fail_link(shader_program
, "Allocation of uniform `%s' to target "
2155 "failed (%d vs %d)\n",
2156 uniform
->Name
, index
, parameter_index
);
2161 talloc_free(sorted_uniforms
);
2165 set_uniform_initializer(struct gl_context
*ctx
, void *mem_ctx
,
2166 struct gl_shader_program
*shader_program
,
2167 const char *name
, const glsl_type
*type
,
2170 if (type
->is_record()) {
2171 ir_constant
*field_constant
;
2173 field_constant
= (ir_constant
*)val
->components
.get_head();
2175 for (unsigned int i
= 0; i
< type
->length
; i
++) {
2176 const glsl_type
*field_type
= type
->fields
.structure
[i
].type
;
2177 const char *field_name
= talloc_asprintf(mem_ctx
, "%s.%s", name
,
2178 type
->fields
.structure
[i
].name
);
2179 set_uniform_initializer(ctx
, mem_ctx
, shader_program
, field_name
,
2180 field_type
, field_constant
);
2181 field_constant
= (ir_constant
*)field_constant
->next
;
2186 int loc
= _mesa_get_uniform_location(ctx
, shader_program
, name
);
2189 fail_link(shader_program
,
2190 "Couldn't find uniform for initializer %s\n", name
);
2194 for (unsigned int i
= 0; i
< (type
->is_array() ? type
->length
: 1); i
++) {
2195 ir_constant
*element
;
2196 const glsl_type
*element_type
;
2197 if (type
->is_array()) {
2198 element
= val
->array_elements
[i
];
2199 element_type
= type
->fields
.array
;
2202 element_type
= type
;
2207 if (element_type
->base_type
== GLSL_TYPE_BOOL
) {
2208 int *conv
= talloc_array(mem_ctx
, int, element_type
->components());
2209 for (unsigned int j
= 0; j
< element_type
->components(); j
++) {
2210 conv
[j
] = element
->value
.b
[j
];
2212 values
= (void *)conv
;
2213 element_type
= glsl_type::get_instance(GLSL_TYPE_INT
,
2214 element_type
->vector_elements
,
2217 values
= &element
->value
;
2220 if (element_type
->is_matrix()) {
2221 _mesa_uniform_matrix(ctx
, shader_program
,
2222 element_type
->matrix_columns
,
2223 element_type
->vector_elements
,
2224 loc
, 1, GL_FALSE
, (GLfloat
*)values
);
2225 loc
+= element_type
->matrix_columns
;
2227 _mesa_uniform(ctx
, shader_program
, loc
, element_type
->matrix_columns
,
2228 values
, element_type
->gl_type
);
2229 loc
+= type_size(element_type
);
2235 set_uniform_initializers(struct gl_context
*ctx
,
2236 struct gl_shader_program
*shader_program
)
2238 void *mem_ctx
= NULL
;
2240 for (unsigned int i
= 0; i
< shader_program
->_NumLinkedShaders
; i
++) {
2241 struct gl_shader
*shader
= shader_program
->_LinkedShaders
[i
];
2242 foreach_iter(exec_list_iterator
, iter
, *shader
->ir
) {
2243 ir_instruction
*ir
= (ir_instruction
*)iter
.get();
2244 ir_variable
*var
= ir
->as_variable();
2246 if (!var
|| var
->mode
!= ir_var_uniform
|| !var
->constant_value
)
2250 mem_ctx
= talloc_new(NULL
);
2252 set_uniform_initializer(ctx
, mem_ctx
, shader_program
, var
->name
,
2253 var
->type
, var
->constant_value
);
2257 talloc_free(mem_ctx
);
2262 * Convert a shader's GLSL IR into a Mesa gl_program.
2265 get_mesa_program(struct gl_context
*ctx
, struct gl_shader_program
*shader_program
,
2266 struct gl_shader
*shader
)
2268 ir_to_mesa_visitor v
;
2269 struct prog_instruction
*mesa_instructions
, *mesa_inst
;
2270 ir_instruction
**mesa_instruction_annotation
;
2272 struct gl_program
*prog
;
2274 const char *target_string
;
2276 struct gl_shader_compiler_options
*options
=
2277 &ctx
->ShaderCompilerOptions
[_mesa_shader_type_to_index(shader
->Type
)];
2279 switch (shader
->Type
) {
2280 case GL_VERTEX_SHADER
:
2281 target
= GL_VERTEX_PROGRAM_ARB
;
2282 target_string
= "vertex";
2284 case GL_FRAGMENT_SHADER
:
2285 target
= GL_FRAGMENT_PROGRAM_ARB
;
2286 target_string
= "fragment";
2289 assert(!"should not be reached");
2293 validate_ir_tree(shader
->ir
);
2295 prog
= ctx
->Driver
.NewProgram(ctx
, target
, shader_program
->Name
);
2298 prog
->Parameters
= _mesa_new_parameter_list();
2299 prog
->Varying
= _mesa_new_parameter_list();
2300 prog
->Attributes
= _mesa_new_parameter_list();
2303 v
.shader_program
= shader_program
;
2304 v
.options
= options
;
2306 add_uniforms_to_parameters_list(shader_program
, shader
, prog
);
2308 /* Emit Mesa IR for main(). */
2309 visit_exec_list(shader
->ir
, &v
);
2310 v
.ir_to_mesa_emit_op0(NULL
, OPCODE_END
);
2312 /* Now emit bodies for any functions that were used. */
2314 progress
= GL_FALSE
;
2316 foreach_iter(exec_list_iterator
, iter
, v
.function_signatures
) {
2317 function_entry
*entry
= (function_entry
*)iter
.get();
2319 if (!entry
->bgn_inst
) {
2320 v
.current_function
= entry
;
2322 entry
->bgn_inst
= v
.ir_to_mesa_emit_op0(NULL
, OPCODE_BGNSUB
);
2323 entry
->bgn_inst
->function
= entry
;
2325 visit_exec_list(&entry
->sig
->body
, &v
);
2327 ir_to_mesa_instruction
*last
;
2328 last
= (ir_to_mesa_instruction
*)v
.instructions
.get_tail();
2329 if (last
->op
!= OPCODE_RET
)
2330 v
.ir_to_mesa_emit_op0(NULL
, OPCODE_RET
);
2332 ir_to_mesa_instruction
*end
;
2333 end
= v
.ir_to_mesa_emit_op0(NULL
, OPCODE_ENDSUB
);
2334 end
->function
= entry
;
2341 prog
->NumTemporaries
= v
.next_temp
;
2343 int num_instructions
= 0;
2344 foreach_iter(exec_list_iterator
, iter
, v
.instructions
) {
2349 (struct prog_instruction
*)calloc(num_instructions
,
2350 sizeof(*mesa_instructions
));
2351 mesa_instruction_annotation
= talloc_array(v
.mem_ctx
, ir_instruction
*,
2354 /* Convert ir_mesa_instructions into prog_instructions.
2356 mesa_inst
= mesa_instructions
;
2358 foreach_iter(exec_list_iterator
, iter
, v
.instructions
) {
2359 const ir_to_mesa_instruction
*inst
= (ir_to_mesa_instruction
*)iter
.get();
2361 mesa_inst
->Opcode
= inst
->op
;
2362 mesa_inst
->CondUpdate
= inst
->cond_update
;
2363 mesa_inst
->DstReg
.File
= inst
->dst_reg
.file
;
2364 mesa_inst
->DstReg
.Index
= inst
->dst_reg
.index
;
2365 mesa_inst
->DstReg
.CondMask
= inst
->dst_reg
.cond_mask
;
2366 mesa_inst
->DstReg
.WriteMask
= inst
->dst_reg
.writemask
;
2367 mesa_inst
->DstReg
.RelAddr
= inst
->dst_reg
.reladdr
!= NULL
;
2368 mesa_inst
->SrcReg
[0] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[0]);
2369 mesa_inst
->SrcReg
[1] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[1]);
2370 mesa_inst
->SrcReg
[2] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[2]);
2371 mesa_inst
->TexSrcUnit
= inst
->sampler
;
2372 mesa_inst
->TexSrcTarget
= inst
->tex_target
;
2373 mesa_inst
->TexShadow
= inst
->tex_shadow
;
2374 mesa_instruction_annotation
[i
] = inst
->ir
;
2376 /* Set IndirectRegisterFiles. */
2377 if (mesa_inst
->DstReg
.RelAddr
)
2378 prog
->IndirectRegisterFiles
|= 1 << mesa_inst
->DstReg
.File
;
2380 /* Update program's bitmask of indirectly accessed register files */
2381 for (unsigned src
= 0; src
< 3; src
++)
2382 if (mesa_inst
->SrcReg
[src
].RelAddr
)
2383 prog
->IndirectRegisterFiles
|= 1 << mesa_inst
->SrcReg
[src
].File
;
2385 if (options
->EmitNoIfs
&& mesa_inst
->Opcode
== OPCODE_IF
) {
2386 fail_link(shader_program
, "Couldn't flatten if statement\n");
2389 switch (mesa_inst
->Opcode
) {
2391 inst
->function
->inst
= i
;
2392 mesa_inst
->Comment
= strdup(inst
->function
->sig
->function_name());
2395 mesa_inst
->Comment
= strdup(inst
->function
->sig
->function_name());
2398 mesa_inst
->BranchTarget
= inst
->function
->sig_id
; /* rewritten later */
2401 prog
->NumAddressRegs
= 1;
2411 set_branchtargets(&v
, mesa_instructions
, num_instructions
);
2413 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2415 printf("GLSL IR for linked %s program %d:\n", target_string
,
2416 shader_program
->Name
);
2417 _mesa_print_ir(shader
->ir
, NULL
);
2420 printf("Mesa IR for linked %s program %d:\n", target_string
,
2421 shader_program
->Name
);
2422 print_program(mesa_instructions
, mesa_instruction_annotation
,
2426 prog
->Instructions
= mesa_instructions
;
2427 prog
->NumInstructions
= num_instructions
;
2429 do_set_program_inouts(shader
->ir
, prog
);
2430 count_resources(prog
);
2432 _mesa_reference_program(ctx
, &shader
->Program
, prog
);
2434 if ((ctx
->Shader
.Flags
& GLSL_NO_OPT
) == 0) {
2435 _mesa_optimize_program(ctx
, prog
);
2444 * Called via ctx->Driver.CompilerShader().
2446 * XXX can we remove the ctx->Driver.CompileShader() hook?
2449 _mesa_ir_compile_shader(struct gl_context
*ctx
, struct gl_shader
*shader
)
2451 assert(shader
->CompileStatus
);
2460 * Called via ctx->Driver.LinkShader()
2461 * This actually involves converting GLSL IR into Mesa gl_programs with
2462 * code lowering and other optimizations.
2465 _mesa_ir_link_shader(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2467 assert(prog
->LinkStatus
);
2469 for (unsigned i
= 0; i
< prog
->_NumLinkedShaders
; i
++) {
2471 exec_list
*ir
= prog
->_LinkedShaders
[i
]->ir
;
2472 const struct gl_shader_compiler_options
*options
=
2473 &ctx
->ShaderCompilerOptions
[_mesa_shader_type_to_index(prog
->_LinkedShaders
[i
]->Type
)];
2479 do_mat_op_to_vec(ir
);
2480 do_mod_to_fract(ir
);
2481 do_div_to_mul_rcp(ir
);
2482 do_explog_to_explog2(ir
);
2484 progress
= do_lower_jumps(ir
, true, true, options
->EmitNoMainReturn
, options
->EmitNoCont
, options
->EmitNoLoops
) || progress
;
2486 progress
= do_common_optimization(ir
, true, options
->MaxUnrollIterations
) || progress
;
2488 if (options
->EmitNoIfs
)
2489 progress
= do_if_to_cond_assign(ir
) || progress
;
2491 if (options
->EmitNoNoise
)
2492 progress
= lower_noise(ir
) || progress
;
2494 /* If there are forms of indirect addressing that the driver
2495 * cannot handle, perform the lowering pass.
2497 if (options
->EmitNoIndirectInput
|| options
->EmitNoIndirectOutput
2498 || options
->EmitNoIndirectTemp
|| options
->EmitNoIndirectUniform
)
2500 lower_variable_index_to_cond_assign(ir
,
2501 options
->EmitNoIndirectInput
,
2502 options
->EmitNoIndirectOutput
,
2503 options
->EmitNoIndirectTemp
,
2504 options
->EmitNoIndirectUniform
)
2507 progress
= do_vec_index_to_cond_assign(ir
) || progress
;
2510 validate_ir_tree(ir
);
2513 for (unsigned i
= 0; i
< prog
->_NumLinkedShaders
; i
++) {
2514 struct gl_program
*linked_prog
;
2517 linked_prog
= get_mesa_program(ctx
, prog
, prog
->_LinkedShaders
[i
]);
2519 switch (prog
->_LinkedShaders
[i
]->Type
) {
2520 case GL_VERTEX_SHADER
:
2521 _mesa_reference_vertprog(ctx
, &prog
->VertexProgram
,
2522 (struct gl_vertex_program
*)linked_prog
);
2523 ok
= ctx
->Driver
.ProgramStringNotify(ctx
, GL_VERTEX_PROGRAM_ARB
,
2526 case GL_FRAGMENT_SHADER
:
2527 _mesa_reference_fragprog(ctx
, &prog
->FragmentProgram
,
2528 (struct gl_fragment_program
*)linked_prog
);
2529 ok
= ctx
->Driver
.ProgramStringNotify(ctx
, GL_FRAGMENT_PROGRAM_ARB
,
2536 _mesa_reference_program(ctx
, &linked_prog
, NULL
);
2544 * Compile a GLSL shader. Called via glCompileShader().
2547 _mesa_glsl_compile_shader(struct gl_context
*ctx
, struct gl_shader
*shader
)
2549 struct _mesa_glsl_parse_state
*state
=
2550 new(shader
) _mesa_glsl_parse_state(ctx
, shader
->Type
, shader
);
2552 const char *source
= shader
->Source
;
2553 /* Check if the user called glCompileShader without first calling
2554 * glShaderSource. This should fail to compile, but not raise a GL_ERROR.
2556 if (source
== NULL
) {
2557 shader
->CompileStatus
= GL_FALSE
;
2561 state
->error
= preprocess(state
, &source
, &state
->info_log
,
2562 &ctx
->Extensions
, ctx
->API
);
2564 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2565 printf("GLSL source for shader %d:\n", shader
->Name
);
2566 printf("%s\n", shader
->Source
);
2569 if (!state
->error
) {
2570 _mesa_glsl_lexer_ctor(state
, source
);
2571 _mesa_glsl_parse(state
);
2572 _mesa_glsl_lexer_dtor(state
);
2575 talloc_free(shader
->ir
);
2576 shader
->ir
= new(shader
) exec_list
;
2577 if (!state
->error
&& !state
->translation_unit
.is_empty())
2578 _mesa_ast_to_hir(shader
->ir
, state
);
2580 if (!state
->error
&& !shader
->ir
->is_empty()) {
2581 validate_ir_tree(shader
->ir
);
2583 /* Do some optimization at compile time to reduce shader IR size
2584 * and reduce later work if the same shader is linked multiple times
2586 while (do_common_optimization(shader
->ir
, false, 32))
2589 validate_ir_tree(shader
->ir
);
2592 shader
->symbols
= state
->symbols
;
2594 shader
->CompileStatus
= !state
->error
;
2595 shader
->InfoLog
= state
->info_log
;
2596 shader
->Version
= state
->language_version
;
2597 memcpy(shader
->builtins_to_link
, state
->builtins_to_link
,
2598 sizeof(shader
->builtins_to_link
[0]) * state
->num_builtins_to_link
);
2599 shader
->num_builtins_to_link
= state
->num_builtins_to_link
;
2601 if (ctx
->Shader
.Flags
& GLSL_LOG
) {
2602 _mesa_write_shader_to_file(shader
);
2605 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2606 if (shader
->CompileStatus
) {
2607 printf("GLSL IR for shader %d:\n", shader
->Name
);
2608 _mesa_print_ir(shader
->ir
, NULL
);
2611 printf("GLSL shader %d failed to compile.\n", shader
->Name
);
2613 if (shader
->InfoLog
&& shader
->InfoLog
[0] != 0) {
2614 printf("GLSL shader %d info log:\n", shader
->Name
);
2615 printf("%s\n", shader
->InfoLog
);
2619 /* Retain any live IR, but trash the rest. */
2620 reparent_ir(shader
->ir
, shader
->ir
);
2624 if (shader
->CompileStatus
) {
2625 if (!ctx
->Driver
.CompileShader(ctx
, shader
))
2626 shader
->CompileStatus
= GL_FALSE
;
2632 * Link a GLSL shader program. Called via glLinkProgram().
2635 _mesa_glsl_link_shader(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2639 _mesa_clear_shader_program_data(ctx
, prog
);
2641 prog
->LinkStatus
= GL_TRUE
;
2643 for (i
= 0; i
< prog
->NumShaders
; i
++) {
2644 if (!prog
->Shaders
[i
]->CompileStatus
) {
2645 fail_link(prog
, "linking with uncompiled shader");
2646 prog
->LinkStatus
= GL_FALSE
;
2650 prog
->Varying
= _mesa_new_parameter_list();
2651 _mesa_reference_vertprog(ctx
, &prog
->VertexProgram
, NULL
);
2652 _mesa_reference_fragprog(ctx
, &prog
->FragmentProgram
, NULL
);
2654 if (prog
->LinkStatus
) {
2655 link_shaders(ctx
, prog
);
2658 if (prog
->LinkStatus
) {
2659 if (!ctx
->Driver
.LinkShader(ctx
, prog
)) {
2660 prog
->LinkStatus
= GL_FALSE
;
2664 set_uniform_initializers(ctx
, prog
);
2666 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2667 if (!prog
->LinkStatus
) {
2668 printf("GLSL shader program %d failed to link\n", prog
->Name
);
2671 if (prog
->InfoLog
&& prog
->InfoLog
[0] != 0) {
2672 printf("GLSL shader program %d info log:\n", prog
->Name
);
2673 printf("%s\n", prog
->InfoLog
);