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 * Translates the IR to ARB_fragment_program text if possible,
34 #include "main/compiler.h"
36 #include "ir_visitor.h"
37 #include "ir_print_visitor.h"
38 #include "ir_expression_flattening.h"
39 #include "glsl_types.h"
40 #include "glsl_parser_extras.h"
41 #include "../glsl/program.h"
42 #include "ir_optimization.h"
46 #include "main/mtypes.h"
47 #include "main/shaderapi.h"
48 #include "main/shaderobj.h"
49 #include "main/uniforms.h"
50 #include "program/hash_table.h"
51 #include "program/prog_instruction.h"
52 #include "program/prog_optimize.h"
53 #include "program/prog_print.h"
54 #include "program/program.h"
55 #include "program/prog_uniform.h"
56 #include "program/prog_parameter.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
;
83 int file
; /**< PROGRAM_* from Mesa */
84 int index
; /**< temporary index, VERT_ATTRIB_*, FRAG_ATTRIB_*, etc. */
85 GLuint swizzle
; /**< SWIZZLE_XYZWONEZERO swizzles from Mesa. */
86 int negate
; /**< NEGATE_XYZW mask from mesa */
87 /** Register index should be offset by the integer in this reg. */
88 ir_to_mesa_src_reg
*reladdr
;
91 typedef struct ir_to_mesa_dst_reg
{
92 int file
; /**< PROGRAM_* from Mesa */
93 int index
; /**< temporary index, VERT_ATTRIB_*, FRAG_ATTRIB_*, etc. */
94 int writemask
; /**< Bitfield of WRITEMASK_[XYZW] */
96 /** Register index should be offset by the integer in this reg. */
97 ir_to_mesa_src_reg
*reladdr
;
100 extern ir_to_mesa_src_reg ir_to_mesa_undef
;
102 class ir_to_mesa_instruction
: public exec_node
{
105 ir_to_mesa_dst_reg dst_reg
;
106 ir_to_mesa_src_reg src_reg
[3];
107 /** Pointer to the ir source this tree came from for debugging */
109 GLboolean cond_update
;
110 int sampler
; /**< sampler index */
111 int tex_target
; /**< One of TEXTURE_*_INDEX */
112 GLboolean tex_shadow
;
114 class function_entry
*function
; /* Set on OPCODE_CAL or OPCODE_BGNSUB */
117 class variable_storage
: public exec_node
{
119 variable_storage(ir_variable
*var
, int file
, int index
)
120 : file(file
), index(index
), var(var
)
127 ir_variable
*var
; /* variable that maps to this, if any */
130 class function_entry
: public exec_node
{
132 ir_function_signature
*sig
;
135 * identifier of this function signature used by the program.
137 * At the point that Mesa instructions for function calls are
138 * generated, we don't know the address of the first instruction of
139 * the function body. So we make the BranchTarget that is called a
140 * small integer and rewrite them during set_branchtargets().
145 * Pointer to first instruction of the function body.
147 * Set during function body emits after main() is processed.
149 ir_to_mesa_instruction
*bgn_inst
;
152 * Index of the first instruction of the function body in actual
155 * Set after convertion from ir_to_mesa_instruction to prog_instruction.
159 /** Storage for the return value. */
160 ir_to_mesa_src_reg return_reg
;
163 class ir_to_mesa_visitor
: public ir_visitor
{
165 ir_to_mesa_visitor();
166 ~ir_to_mesa_visitor();
168 function_entry
*current_function
;
171 struct gl_program
*prog
;
175 variable_storage
*find_variable_storage(ir_variable
*var
);
177 function_entry
*get_function_signature(ir_function_signature
*sig
);
179 ir_to_mesa_src_reg
get_temp(const glsl_type
*type
);
180 void reladdr_to_temp(ir_instruction
*ir
,
181 ir_to_mesa_src_reg
*reg
, int *num_reladdr
);
183 struct ir_to_mesa_src_reg
src_reg_for_float(float val
);
186 * \name Visit methods
188 * As typical for the visitor pattern, there must be one \c visit method for
189 * each concrete subclass of \c ir_instruction. Virtual base classes within
190 * the hierarchy should not have \c visit methods.
193 virtual void visit(ir_variable
*);
194 virtual void visit(ir_loop
*);
195 virtual void visit(ir_loop_jump
*);
196 virtual void visit(ir_function_signature
*);
197 virtual void visit(ir_function
*);
198 virtual void visit(ir_expression
*);
199 virtual void visit(ir_swizzle
*);
200 virtual void visit(ir_dereference_variable
*);
201 virtual void visit(ir_dereference_array
*);
202 virtual void visit(ir_dereference_record
*);
203 virtual void visit(ir_assignment
*);
204 virtual void visit(ir_constant
*);
205 virtual void visit(ir_call
*);
206 virtual void visit(ir_return
*);
207 virtual void visit(ir_discard
*);
208 virtual void visit(ir_texture
*);
209 virtual void visit(ir_if
*);
212 struct ir_to_mesa_src_reg result
;
214 /** List of variable_storage */
217 /** List of function_entry */
218 exec_list function_signatures
;
219 int next_signature_id
;
221 /** List of ir_to_mesa_instruction */
222 exec_list instructions
;
224 ir_to_mesa_instruction
*ir_to_mesa_emit_op0(ir_instruction
*ir
,
225 enum prog_opcode op
);
227 ir_to_mesa_instruction
*ir_to_mesa_emit_op1(ir_instruction
*ir
,
229 ir_to_mesa_dst_reg dst
,
230 ir_to_mesa_src_reg src0
);
232 ir_to_mesa_instruction
*ir_to_mesa_emit_op2(ir_instruction
*ir
,
234 ir_to_mesa_dst_reg dst
,
235 ir_to_mesa_src_reg src0
,
236 ir_to_mesa_src_reg src1
);
238 ir_to_mesa_instruction
*ir_to_mesa_emit_op3(ir_instruction
*ir
,
240 ir_to_mesa_dst_reg dst
,
241 ir_to_mesa_src_reg src0
,
242 ir_to_mesa_src_reg src1
,
243 ir_to_mesa_src_reg src2
);
245 void ir_to_mesa_emit_scalar_op1(ir_instruction
*ir
,
247 ir_to_mesa_dst_reg dst
,
248 ir_to_mesa_src_reg src0
);
250 void ir_to_mesa_emit_scalar_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 GLboolean
try_emit_mad(ir_expression
*ir
,
259 int add_uniform(const char *name
,
260 const glsl_type
*type
,
261 ir_constant
*constant
);
262 void add_aggregate_uniform(ir_instruction
*ir
,
264 const struct glsl_type
*type
,
265 ir_constant
*constant
,
266 struct ir_to_mesa_dst_reg temp
);
268 struct hash_table
*sampler_map
;
270 void set_sampler_location(ir_variable
*sampler
, int location
);
271 int get_sampler_location(ir_variable
*sampler
);
276 ir_to_mesa_src_reg ir_to_mesa_undef
= ir_to_mesa_src_reg(PROGRAM_UNDEFINED
, 0, NULL
);
278 ir_to_mesa_dst_reg ir_to_mesa_undef_dst
= {
279 PROGRAM_UNDEFINED
, 0, SWIZZLE_NOOP
, COND_TR
, NULL
,
282 ir_to_mesa_dst_reg ir_to_mesa_address_reg
= {
283 PROGRAM_ADDRESS
, 0, WRITEMASK_X
, COND_TR
, NULL
286 static int swizzle_for_size(int size
)
288 int size_swizzles
[4] = {
289 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_X
),
290 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_Y
, SWIZZLE_Y
, SWIZZLE_Y
),
291 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_Y
, SWIZZLE_Z
, SWIZZLE_Z
),
292 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_Y
, SWIZZLE_Z
, SWIZZLE_W
),
295 return size_swizzles
[size
- 1];
298 ir_to_mesa_instruction
*
299 ir_to_mesa_visitor::ir_to_mesa_emit_op3(ir_instruction
*ir
,
301 ir_to_mesa_dst_reg dst
,
302 ir_to_mesa_src_reg src0
,
303 ir_to_mesa_src_reg src1
,
304 ir_to_mesa_src_reg src2
)
306 ir_to_mesa_instruction
*inst
= new(mem_ctx
) ir_to_mesa_instruction();
309 /* If we have to do relative addressing, we want to load the ARL
310 * reg directly for one of the regs, and preload the other reladdr
311 * sources into temps.
313 num_reladdr
+= dst
.reladdr
!= NULL
;
314 num_reladdr
+= src0
.reladdr
!= NULL
;
315 num_reladdr
+= src1
.reladdr
!= NULL
;
316 num_reladdr
+= src2
.reladdr
!= NULL
;
318 reladdr_to_temp(ir
, &src2
, &num_reladdr
);
319 reladdr_to_temp(ir
, &src1
, &num_reladdr
);
320 reladdr_to_temp(ir
, &src0
, &num_reladdr
);
323 ir_to_mesa_emit_op1(ir
, OPCODE_ARL
, ir_to_mesa_address_reg
,
328 assert(num_reladdr
== 0);
332 inst
->src_reg
[0] = src0
;
333 inst
->src_reg
[1] = src1
;
334 inst
->src_reg
[2] = src2
;
337 inst
->function
= NULL
;
339 this->instructions
.push_tail(inst
);
345 ir_to_mesa_instruction
*
346 ir_to_mesa_visitor::ir_to_mesa_emit_op2(ir_instruction
*ir
,
348 ir_to_mesa_dst_reg dst
,
349 ir_to_mesa_src_reg src0
,
350 ir_to_mesa_src_reg src1
)
352 return ir_to_mesa_emit_op3(ir
, op
, dst
, src0
, src1
, ir_to_mesa_undef
);
355 ir_to_mesa_instruction
*
356 ir_to_mesa_visitor::ir_to_mesa_emit_op1(ir_instruction
*ir
,
358 ir_to_mesa_dst_reg dst
,
359 ir_to_mesa_src_reg src0
)
361 assert(dst
.writemask
!= 0);
362 return ir_to_mesa_emit_op3(ir
, op
, dst
,
363 src0
, ir_to_mesa_undef
, ir_to_mesa_undef
);
366 ir_to_mesa_instruction
*
367 ir_to_mesa_visitor::ir_to_mesa_emit_op0(ir_instruction
*ir
,
370 return ir_to_mesa_emit_op3(ir
, op
, ir_to_mesa_undef_dst
,
377 ir_to_mesa_visitor::set_sampler_location(ir_variable
*sampler
, int location
)
379 if (this->sampler_map
== NULL
) {
380 this->sampler_map
= hash_table_ctor(0, hash_table_pointer_hash
,
381 hash_table_pointer_compare
);
384 hash_table_insert(this->sampler_map
, (void *)(uintptr_t)location
, sampler
);
388 ir_to_mesa_visitor::get_sampler_location(ir_variable
*sampler
)
390 void *result
= hash_table_find(this->sampler_map
, sampler
);
392 return (int)(uintptr_t)result
;
395 inline ir_to_mesa_dst_reg
396 ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg reg
)
398 ir_to_mesa_dst_reg dst_reg
;
400 dst_reg
.file
= reg
.file
;
401 dst_reg
.index
= reg
.index
;
402 dst_reg
.writemask
= WRITEMASK_XYZW
;
403 dst_reg
.cond_mask
= COND_TR
;
404 dst_reg
.reladdr
= reg
.reladdr
;
409 inline ir_to_mesa_src_reg
410 ir_to_mesa_src_reg_from_dst(ir_to_mesa_dst_reg reg
)
412 return ir_to_mesa_src_reg(reg
.file
, reg
.index
, NULL
);
416 * Emits Mesa scalar opcodes to produce unique answers across channels.
418 * Some Mesa opcodes are scalar-only, like ARB_fp/vp. The src X
419 * channel determines the result across all channels. So to do a vec4
420 * of this operation, we want to emit a scalar per source channel used
421 * to produce dest channels.
424 ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op2(ir_instruction
*ir
,
426 ir_to_mesa_dst_reg dst
,
427 ir_to_mesa_src_reg orig_src0
,
428 ir_to_mesa_src_reg orig_src1
)
431 int done_mask
= ~dst
.writemask
;
433 /* Mesa RCP is a scalar operation splatting results to all channels,
434 * like ARB_fp/vp. So emit as many RCPs as necessary to cover our
437 for (i
= 0; i
< 4; i
++) {
438 GLuint this_mask
= (1 << i
);
439 ir_to_mesa_instruction
*inst
;
440 ir_to_mesa_src_reg src0
= orig_src0
;
441 ir_to_mesa_src_reg src1
= orig_src1
;
443 if (done_mask
& this_mask
)
446 GLuint src0_swiz
= GET_SWZ(src0
.swizzle
, i
);
447 GLuint src1_swiz
= GET_SWZ(src1
.swizzle
, i
);
448 for (j
= i
+ 1; j
< 4; j
++) {
449 if (!(done_mask
& (1 << j
)) &&
450 GET_SWZ(src0
.swizzle
, j
) == src0_swiz
&&
451 GET_SWZ(src1
.swizzle
, j
) == src1_swiz
) {
452 this_mask
|= (1 << j
);
455 src0
.swizzle
= MAKE_SWIZZLE4(src0_swiz
, src0_swiz
,
456 src0_swiz
, src0_swiz
);
457 src1
.swizzle
= MAKE_SWIZZLE4(src1_swiz
, src1_swiz
,
458 src1_swiz
, src1_swiz
);
460 inst
= ir_to_mesa_emit_op2(ir
, op
,
464 inst
->dst_reg
.writemask
= this_mask
;
465 done_mask
|= this_mask
;
470 ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op1(ir_instruction
*ir
,
472 ir_to_mesa_dst_reg dst
,
473 ir_to_mesa_src_reg src0
)
475 ir_to_mesa_src_reg undef
= ir_to_mesa_undef
;
477 undef
.swizzle
= SWIZZLE_XXXX
;
479 ir_to_mesa_emit_scalar_op2(ir
, op
, dst
, src0
, undef
);
482 struct ir_to_mesa_src_reg
483 ir_to_mesa_visitor::src_reg_for_float(float val
)
485 ir_to_mesa_src_reg
src_reg(PROGRAM_CONSTANT
, -1, NULL
);
487 src_reg
.index
= _mesa_add_unnamed_constant(this->prog
->Parameters
,
488 &val
, 1, &src_reg
.swizzle
);
494 type_size(const struct glsl_type
*type
)
499 switch (type
->base_type
) {
502 case GLSL_TYPE_FLOAT
:
504 if (type
->is_matrix()) {
505 return type
->matrix_columns
;
507 /* Regardless of size of vector, it gets a vec4. This is bad
508 * packing for things like floats, but otherwise arrays become a
509 * mess. Hopefully a later pass over the code can pack scalars
510 * down if appropriate.
514 case GLSL_TYPE_ARRAY
:
515 return type_size(type
->fields
.array
) * type
->length
;
516 case GLSL_TYPE_STRUCT
:
518 for (i
= 0; i
< type
->length
; i
++) {
519 size
+= type_size(type
->fields
.structure
[i
].type
);
522 case GLSL_TYPE_SAMPLER
:
523 /* Samplers take up no register space, since they're baked in at
534 * In the initial pass of codegen, we assign temporary numbers to
535 * intermediate results. (not SSA -- variable assignments will reuse
536 * storage). Actual register allocation for the Mesa VM occurs in a
537 * pass over the Mesa IR later.
540 ir_to_mesa_visitor::get_temp(const glsl_type
*type
)
542 ir_to_mesa_src_reg src_reg
;
546 src_reg
.file
= PROGRAM_TEMPORARY
;
547 src_reg
.index
= next_temp
;
548 src_reg
.reladdr
= NULL
;
549 next_temp
+= type_size(type
);
551 if (type
->is_array() || type
->is_record()) {
552 src_reg
.swizzle
= SWIZZLE_NOOP
;
554 for (i
= 0; i
< type
->vector_elements
; i
++)
557 swizzle
[i
] = type
->vector_elements
- 1;
558 src_reg
.swizzle
= MAKE_SWIZZLE4(swizzle
[0], swizzle
[1],
559 swizzle
[2], swizzle
[3]);
567 ir_to_mesa_visitor::find_variable_storage(ir_variable
*var
)
570 variable_storage
*entry
;
572 foreach_iter(exec_list_iterator
, iter
, this->variables
) {
573 entry
= (variable_storage
*)iter
.get();
575 if (entry
->var
== var
)
583 ir_to_mesa_visitor::visit(ir_variable
*ir
)
585 if (strcmp(ir
->name
, "gl_FragCoord") == 0) {
586 struct gl_fragment_program
*fp
= (struct gl_fragment_program
*)this->prog
;
588 fp
->OriginUpperLeft
= ir
->origin_upper_left
;
589 fp
->PixelCenterInteger
= ir
->pixel_center_integer
;
594 ir_to_mesa_visitor::visit(ir_loop
*ir
)
598 assert(!ir
->increment
);
599 assert(!ir
->counter
);
601 ir_to_mesa_emit_op0(NULL
, OPCODE_BGNLOOP
);
602 visit_exec_list(&ir
->body_instructions
, this);
603 ir_to_mesa_emit_op0(NULL
, OPCODE_ENDLOOP
);
607 ir_to_mesa_visitor::visit(ir_loop_jump
*ir
)
610 case ir_loop_jump::jump_break
:
611 ir_to_mesa_emit_op0(NULL
, OPCODE_BRK
);
613 case ir_loop_jump::jump_continue
:
614 ir_to_mesa_emit_op0(NULL
, OPCODE_CONT
);
621 ir_to_mesa_visitor::visit(ir_function_signature
*ir
)
628 ir_to_mesa_visitor::visit(ir_function
*ir
)
630 /* Ignore function bodies other than main() -- we shouldn't see calls to
631 * them since they should all be inlined before we get to ir_to_mesa.
633 if (strcmp(ir
->name
, "main") == 0) {
634 const ir_function_signature
*sig
;
637 sig
= ir
->matching_signature(&empty
);
641 foreach_iter(exec_list_iterator
, iter
, sig
->body
) {
642 ir_instruction
*ir
= (ir_instruction
*)iter
.get();
650 ir_to_mesa_visitor::try_emit_mad(ir_expression
*ir
, int mul_operand
)
652 int nonmul_operand
= 1 - mul_operand
;
653 ir_to_mesa_src_reg a
, b
, c
;
655 ir_expression
*expr
= ir
->operands
[mul_operand
]->as_expression();
656 if (!expr
|| expr
->operation
!= ir_binop_mul
)
659 expr
->operands
[0]->accept(this);
661 expr
->operands
[1]->accept(this);
663 ir
->operands
[nonmul_operand
]->accept(this);
666 this->result
= get_temp(ir
->type
);
667 ir_to_mesa_emit_op3(ir
, OPCODE_MAD
,
668 ir_to_mesa_dst_reg_from_src(this->result
), a
, b
, c
);
674 ir_to_mesa_visitor::reladdr_to_temp(ir_instruction
*ir
,
675 ir_to_mesa_src_reg
*reg
, int *num_reladdr
)
680 ir_to_mesa_emit_op1(ir
, OPCODE_ARL
, ir_to_mesa_address_reg
, *reg
->reladdr
);
682 if (*num_reladdr
!= 1) {
683 ir_to_mesa_src_reg temp
= get_temp(glsl_type::vec4_type
);
685 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
,
686 ir_to_mesa_dst_reg_from_src(temp
), *reg
);
694 ir_to_mesa_visitor::visit(ir_expression
*ir
)
696 unsigned int operand
;
697 struct ir_to_mesa_src_reg op
[2];
698 struct ir_to_mesa_src_reg result_src
;
699 struct ir_to_mesa_dst_reg result_dst
;
700 const glsl_type
*vec4_type
= glsl_type::get_instance(GLSL_TYPE_FLOAT
, 4, 1);
701 const glsl_type
*vec3_type
= glsl_type::get_instance(GLSL_TYPE_FLOAT
, 3, 1);
702 const glsl_type
*vec2_type
= glsl_type::get_instance(GLSL_TYPE_FLOAT
, 2, 1);
704 /* Quick peephole: Emit OPCODE_MAD(a, b, c) instead of ADD(MUL(a, b), c)
706 if (ir
->operation
== ir_binop_add
) {
707 if (try_emit_mad(ir
, 1))
709 if (try_emit_mad(ir
, 0))
713 for (operand
= 0; operand
< ir
->get_num_operands(); operand
++) {
714 this->result
.file
= PROGRAM_UNDEFINED
;
715 ir
->operands
[operand
]->accept(this);
716 if (this->result
.file
== PROGRAM_UNDEFINED
) {
718 printf("Failed to get tree for expression operand:\n");
719 ir
->operands
[operand
]->accept(&v
);
722 op
[operand
] = this->result
;
724 /* Matrix expression operands should have been broken down to vector
725 * operations already.
727 assert(!ir
->operands
[operand
]->type
->is_matrix());
730 this->result
.file
= PROGRAM_UNDEFINED
;
732 /* Storage for our result. Ideally for an assignment we'd be using
733 * the actual storage for the result here, instead.
735 result_src
= get_temp(ir
->type
);
736 /* convenience for the emit functions below. */
737 result_dst
= ir_to_mesa_dst_reg_from_src(result_src
);
738 /* Limit writes to the channels that will be used by result_src later.
739 * This does limit this temp's use as a temporary for multi-instruction
742 result_dst
.writemask
= (1 << ir
->type
->vector_elements
) - 1;
744 switch (ir
->operation
) {
745 case ir_unop_logic_not
:
746 ir_to_mesa_emit_op2(ir
, OPCODE_SEQ
, result_dst
,
747 op
[0], src_reg_for_float(0.0));
750 op
[0].negate
= ~op
[0].negate
;
754 ir_to_mesa_emit_op1(ir
, OPCODE_ABS
, result_dst
, op
[0]);
757 ir_to_mesa_emit_op1(ir
, OPCODE_SSG
, result_dst
, op
[0]);
760 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_RCP
, result_dst
, op
[0]);
764 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_EX2
, result_dst
, op
[0]);
768 assert(!"not reached: should be handled by ir_explog_to_explog2");
771 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_LG2
, result_dst
, op
[0]);
774 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_SIN
, result_dst
, op
[0]);
777 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_COS
, result_dst
, op
[0]);
781 ir_to_mesa_emit_op1(ir
, OPCODE_DDX
, result_dst
, op
[0]);
784 ir_to_mesa_emit_op1(ir
, OPCODE_DDY
, result_dst
, op
[0]);
788 ir_to_mesa_emit_op2(ir
, OPCODE_ADD
, result_dst
, op
[0], op
[1]);
791 ir_to_mesa_emit_op2(ir
, OPCODE_SUB
, result_dst
, op
[0], op
[1]);
795 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
, result_dst
, op
[0], op
[1]);
798 assert(!"not reached: should be handled by ir_div_to_mul_rcp");
800 assert(!"ir_binop_mod should have been converted to b * fract(a/b)");
804 ir_to_mesa_emit_op2(ir
, OPCODE_SLT
, result_dst
, op
[0], op
[1]);
806 case ir_binop_greater
:
807 ir_to_mesa_emit_op2(ir
, OPCODE_SGT
, result_dst
, op
[0], op
[1]);
809 case ir_binop_lequal
:
810 ir_to_mesa_emit_op2(ir
, OPCODE_SLE
, result_dst
, op
[0], op
[1]);
812 case ir_binop_gequal
:
813 ir_to_mesa_emit_op2(ir
, OPCODE_SGE
, result_dst
, op
[0], op
[1]);
816 /* "==" operator producing a scalar boolean. */
817 if (ir
->operands
[0]->type
->is_vector() ||
818 ir
->operands
[1]->type
->is_vector()) {
819 ir_to_mesa_src_reg temp
= get_temp(glsl_type::vec4_type
);
820 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
821 ir_to_mesa_dst_reg_from_src(temp
), op
[0], op
[1]);
822 ir_to_mesa_emit_op2(ir
, OPCODE_DP4
, result_dst
, temp
, temp
);
823 ir_to_mesa_emit_op2(ir
, OPCODE_SEQ
,
824 result_dst
, result_src
, src_reg_for_float(0.0));
826 ir_to_mesa_emit_op2(ir
, OPCODE_SEQ
, result_dst
, op
[0], op
[1]);
829 case ir_binop_nequal
:
830 /* "!=" operator producing a scalar boolean. */
831 if (ir
->operands
[0]->type
->is_vector() ||
832 ir
->operands
[1]->type
->is_vector()) {
833 ir_to_mesa_src_reg temp
= get_temp(glsl_type::vec4_type
);
834 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
835 ir_to_mesa_dst_reg_from_src(temp
), op
[0], op
[1]);
836 ir_to_mesa_emit_op2(ir
, OPCODE_DP4
, result_dst
, temp
, temp
);
837 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
838 result_dst
, result_src
, src_reg_for_float(0.0));
840 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
, result_dst
, op
[0], op
[1]);
843 case ir_binop_logic_xor
:
844 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
, result_dst
, op
[0], op
[1]);
847 case ir_binop_logic_or
:
848 /* This could be a saturated add and skip the SNE. */
849 ir_to_mesa_emit_op2(ir
, OPCODE_ADD
,
853 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
855 result_src
, src_reg_for_float(0.0));
858 case ir_binop_logic_and
:
859 /* the bool args are stored as float 0.0 or 1.0, so "mul" gives us "and". */
860 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
,
866 if (ir
->operands
[0]->type
== vec4_type
) {
867 assert(ir
->operands
[1]->type
== vec4_type
);
868 ir_to_mesa_emit_op2(ir
, OPCODE_DP4
,
871 } else if (ir
->operands
[0]->type
== vec3_type
) {
872 assert(ir
->operands
[1]->type
== vec3_type
);
873 ir_to_mesa_emit_op2(ir
, OPCODE_DP3
,
876 } else if (ir
->operands
[0]->type
== vec2_type
) {
877 assert(ir
->operands
[1]->type
== vec2_type
);
878 ir_to_mesa_emit_op2(ir
, OPCODE_DP2
,
885 ir_to_mesa_emit_op2(ir
, OPCODE_XPD
, result_dst
, op
[0], op
[1]);
889 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_RSQ
, result_dst
, op
[0]);
890 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_RCP
, result_dst
, result_src
);
891 /* For incoming channels < 0, set the result to 0. */
892 ir_to_mesa_emit_op3(ir
, OPCODE_CMP
, result_dst
,
893 op
[0], src_reg_for_float(0.0), result_src
);
896 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_RSQ
, result_dst
, op
[0]);
901 /* Mesa IR lacks types, ints are stored as truncated floats. */
905 ir_to_mesa_emit_op1(ir
, OPCODE_TRUNC
, result_dst
, op
[0]);
909 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
, result_dst
,
910 result_src
, src_reg_for_float(0.0));
913 ir_to_mesa_emit_op1(ir
, OPCODE_TRUNC
, result_dst
, op
[0]);
916 op
[0].negate
= ~op
[0].negate
;
917 ir_to_mesa_emit_op1(ir
, OPCODE_FLR
, result_dst
, op
[0]);
918 result_src
.negate
= ~result_src
.negate
;
921 ir_to_mesa_emit_op1(ir
, OPCODE_FLR
, result_dst
, op
[0]);
924 ir_to_mesa_emit_op1(ir
, OPCODE_FRC
, result_dst
, op
[0]);
928 ir_to_mesa_emit_op2(ir
, OPCODE_MIN
, result_dst
, op
[0], op
[1]);
931 ir_to_mesa_emit_op2(ir
, OPCODE_MAX
, result_dst
, op
[0], op
[1]);
934 ir_to_mesa_emit_scalar_op2(ir
, OPCODE_POW
, result_dst
, op
[0], op
[1]);
937 case ir_unop_bit_not
:
939 case ir_binop_lshift
:
940 case ir_binop_rshift
:
941 case ir_binop_bit_and
:
942 case ir_binop_bit_xor
:
943 case ir_binop_bit_or
:
944 assert(!"GLSL 1.30 features unsupported");
948 this->result
= result_src
;
953 ir_to_mesa_visitor::visit(ir_swizzle
*ir
)
955 ir_to_mesa_src_reg src_reg
;
959 /* Note that this is only swizzles in expressions, not those on the left
960 * hand side of an assignment, which do write masking. See ir_assignment
964 ir
->val
->accept(this);
965 src_reg
= this->result
;
966 assert(src_reg
.file
!= PROGRAM_UNDEFINED
);
968 for (i
= 0; i
< 4; i
++) {
969 if (i
< ir
->type
->vector_elements
) {
972 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.x
);
975 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.y
);
978 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.z
);
981 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.w
);
985 /* If the type is smaller than a vec4, replicate the last
988 swizzle
[i
] = swizzle
[ir
->type
->vector_elements
- 1];
992 src_reg
.swizzle
= MAKE_SWIZZLE4(swizzle
[0],
997 this->result
= src_reg
;
1000 static const struct {
1003 int tokens
[STATE_LENGTH
];
1007 {"gl_DepthRange", "near",
1008 {STATE_DEPTH_RANGE
, 0, 0}, SWIZZLE_XXXX
, false},
1009 {"gl_DepthRange", "far",
1010 {STATE_DEPTH_RANGE
, 0, 0}, SWIZZLE_YYYY
, false},
1011 {"gl_DepthRange", "diff",
1012 {STATE_DEPTH_RANGE
, 0, 0}, SWIZZLE_ZZZZ
, false},
1014 {"gl_ClipPlane", NULL
,
1015 {STATE_CLIPPLANE
, 0, 0}, SWIZZLE_XYZW
, true}
1017 {"gl_Point", "size",
1018 {STATE_POINT_SIZE
}, SWIZZLE_XXXX
, false},
1019 {"gl_Point", "sizeMin",
1020 {STATE_POINT_SIZE
}, SWIZZLE_YYYY
, false},
1021 {"gl_Point", "sizeMax",
1022 {STATE_POINT_SIZE
}, SWIZZLE_ZZZZ
, false},
1023 {"gl_Point", "fadeThresholdSize",
1024 {STATE_POINT_SIZE
}, SWIZZLE_WWWW
, false},
1025 {"gl_Point", "distanceConstantAttenuation",
1026 {STATE_POINT_ATTENUATION
}, SWIZZLE_XXXX
, false},
1027 {"gl_Point", "distanceLinearAttenuation",
1028 {STATE_POINT_ATTENUATION
}, SWIZZLE_YYYY
, false},
1029 {"gl_Point", "distanceQuadraticAttenuation",
1030 {STATE_POINT_ATTENUATION
}, SWIZZLE_ZZZZ
, false},
1032 {"gl_FrontMaterial", "emission",
1033 {STATE_MATERIAL
, 0, STATE_EMISSION
}, SWIZZLE_XYZW
, false},
1034 {"gl_FrontMaterial", "ambient",
1035 {STATE_MATERIAL
, 0, STATE_AMBIENT
}, SWIZZLE_XYZW
, false},
1036 {"gl_FrontMaterial", "diffuse",
1037 {STATE_MATERIAL
, 0, STATE_DIFFUSE
}, SWIZZLE_XYZW
, false},
1038 {"gl_FrontMaterial", "specular",
1039 {STATE_MATERIAL
, 0, STATE_SPECULAR
}, SWIZZLE_XYZW
, false},
1040 {"gl_FrontMaterial", "shininess",
1041 {STATE_MATERIAL
, 0, STATE_SHININESS
}, SWIZZLE_XXXX
, false},
1043 {"gl_BackMaterial", "emission",
1044 {STATE_MATERIAL
, 1, STATE_EMISSION
}, SWIZZLE_XYZW
, false},
1045 {"gl_BackMaterial", "ambient",
1046 {STATE_MATERIAL
, 1, STATE_AMBIENT
}, SWIZZLE_XYZW
, false},
1047 {"gl_BackMaterial", "diffuse",
1048 {STATE_MATERIAL
, 1, STATE_DIFFUSE
}, SWIZZLE_XYZW
, false},
1049 {"gl_BackMaterial", "specular",
1050 {STATE_MATERIAL
, 1, STATE_SPECULAR
}, SWIZZLE_XYZW
, false},
1051 {"gl_BackMaterial", "shininess",
1052 {STATE_MATERIAL
, 1, STATE_SHININESS
}, SWIZZLE_XXXX
, false},
1054 {"gl_LightSource", "ambient",
1055 {STATE_LIGHT
, 0, STATE_AMBIENT
}, SWIZZLE_XYZW
, true},
1056 {"gl_LightSource", "diffuse",
1057 {STATE_LIGHT
, 0, STATE_DIFFUSE
}, SWIZZLE_XYZW
, true},
1058 {"gl_LightSource", "specular",
1059 {STATE_LIGHT
, 0, STATE_SPECULAR
}, SWIZZLE_XYZW
, true},
1060 {"gl_LightSource", "position",
1061 {STATE_LIGHT
, 0, STATE_POSITION
}, SWIZZLE_XYZW
, true},
1062 {"gl_LightSource", "halfVector",
1063 {STATE_LIGHT
, 0, STATE_HALF_VECTOR
}, SWIZZLE_XYZW
, true},
1064 {"gl_LightSource", "spotDirection",
1065 {STATE_LIGHT
, 0, STATE_SPOT_DIRECTION
}, SWIZZLE_XYZW
, true},
1066 {"gl_LightSource", "spotCosCutoff",
1067 {STATE_LIGHT
, 0, STATE_SPOT_DIRECTION
}, SWIZZLE_WWWW
, true},
1068 {"gl_LightSource", "spotCutoff",
1069 {STATE_LIGHT
, 0, STATE_SPOT_CUTOFF
}, SWIZZLE_XXXX
, true},
1070 {"gl_LightSource", "spotExponent",
1071 {STATE_LIGHT
, 0, STATE_ATTENUATION
}, SWIZZLE_WWWW
, true},
1072 {"gl_LightSource", "constantAttenuation",
1073 {STATE_LIGHT
, 0, STATE_ATTENUATION
}, SWIZZLE_XXXX
, true},
1074 {"gl_LightSource", "linearAttenuation",
1075 {STATE_LIGHT
, 0, STATE_ATTENUATION
}, SWIZZLE_YYYY
, true},
1076 {"gl_LightSource", "quadraticAttenuation",
1077 {STATE_LIGHT
, 0, STATE_ATTENUATION
}, SWIZZLE_ZZZZ
, true},
1079 {"gl_LightModel", NULL
,
1080 {STATE_LIGHTMODEL_AMBIENT
, 0}, SWIZZLE_XYZW
, false},
1082 {"gl_FrontLightModelProduct", "sceneColor",
1083 {STATE_LIGHTMODEL_SCENECOLOR
, 0}, SWIZZLE_XYZW
, false},
1084 {"gl_BackLightModelProduct", "sceneColor",
1085 {STATE_LIGHTMODEL_SCENECOLOR
, 1}, SWIZZLE_XYZW
, false},
1087 {"gl_FrontLightProduct", "ambient",
1088 {STATE_LIGHTPROD
, 0, 0, STATE_AMBIENT
}, SWIZZLE_XYZW
, true},
1089 {"gl_FrontLightProduct", "diffuse",
1090 {STATE_LIGHTPROD
, 0, 0, STATE_DIFFUSE
}, SWIZZLE_XYZW
, true},
1091 {"gl_FrontLightProduct", "specular",
1092 {STATE_LIGHTPROD
, 0, 0, STATE_SPECULAR
}, SWIZZLE_XYZW
, true},
1094 {"gl_BackLightProduct", "ambient",
1095 {STATE_LIGHTPROD
, 0, 1, STATE_AMBIENT
}, SWIZZLE_XYZW
, true},
1096 {"gl_BackLightProduct", "diffuse",
1097 {STATE_LIGHTPROD
, 0, 1, STATE_DIFFUSE
}, SWIZZLE_XYZW
, true},
1098 {"gl_BackLightProduct", "specular",
1099 {STATE_LIGHTPROD
, 0, 1, STATE_SPECULAR
}, SWIZZLE_XYZW
, true},
1101 {"gl_TextureEnvColor", "ambient",
1102 {STATE_TEXENV_COLOR
, 0}, SWIZZLE_XYZW
, true},
1104 {"gl_EyePlaneS", NULL
,
1105 {STATE_TEXGEN
, 0, STATE_TEXGEN_EYE_S
}, SWIZZLE_XYZW
, true},
1106 {"gl_EyePlaneT", NULL
,
1107 {STATE_TEXGEN
, 0, STATE_TEXGEN_EYE_T
}, SWIZZLE_XYZW
, true},
1108 {"gl_EyePlaneR", NULL
,
1109 {STATE_TEXGEN
, 0, STATE_TEXGEN_EYE_R
}, SWIZZLE_XYZW
, true},
1110 {"gl_EyePlaneQ", NULL
,
1111 {STATE_TEXGEN
, 0, STATE_TEXGEN_EYE_Q
}, SWIZZLE_XYZW
, true},
1113 {"gl_ObjectPlaneS", NULL
,
1114 {STATE_TEXGEN
, 0, STATE_TEXGEN_OBJECT_S
}, SWIZZLE_XYZW
, true},
1115 {"gl_ObjectPlaneT", NULL
,
1116 {STATE_TEXGEN
, 0, STATE_TEXGEN_OBJECT_T
}, SWIZZLE_XYZW
, true},
1117 {"gl_ObjectPlaneR", NULL
,
1118 {STATE_TEXGEN
, 0, STATE_TEXGEN_OBJECT_R
}, SWIZZLE_XYZW
, true},
1119 {"gl_ObjectPlaneQ", NULL
,
1120 {STATE_TEXGEN
, 0, STATE_TEXGEN_OBJECT_Q
}, SWIZZLE_XYZW
, true},
1123 {STATE_FOG_COLOR
}, SWIZZLE_XYZW
, false},
1124 {"gl_Fog", "density",
1125 {STATE_FOG_PARAMS
}, SWIZZLE_XXXX
, false},
1127 {STATE_FOG_PARAMS
}, SWIZZLE_YYYY
, false},
1129 {STATE_FOG_PARAMS
}, SWIZZLE_ZZZZ
, false},
1131 {STATE_FOG_PARAMS
}, SWIZZLE_WWWW
, false},
1134 static ir_to_mesa_src_reg
1135 get_builtin_uniform_reg(struct gl_program
*prog
,
1136 const char *name
, int array_index
, const char *field
)
1139 ir_to_mesa_src_reg src_reg
;
1140 int tokens
[STATE_LENGTH
];
1142 for (i
= 0; i
< Elements(statevars
); i
++) {
1143 if (strcmp(statevars
[i
].name
, name
) != 0)
1145 if (!field
&& statevars
[i
].field
) {
1146 assert(!"FINISHME: whole-structure state var dereference");
1148 if (field
&& (!statevars
[i
].field
|| strcmp(statevars
[i
].field
, field
) != 0))
1153 if (i
== Elements(statevars
)) {
1154 printf("builtin uniform %s%s%s not found\n",
1157 field
? field
: "");
1161 memcpy(&tokens
, statevars
[i
].tokens
, sizeof(tokens
));
1162 if (statevars
[i
].array_indexed
)
1163 tokens
[1] = array_index
;
1165 src_reg
.file
= PROGRAM_STATE_VAR
;
1166 src_reg
.index
= _mesa_add_state_reference(prog
->Parameters
,
1167 (gl_state_index
*)tokens
);
1168 src_reg
.swizzle
= statevars
[i
].swizzle
;
1170 src_reg
.reladdr
= false;
1176 add_matrix_ref(struct gl_program
*prog
, int *tokens
)
1181 /* Add a ref for each column. It looks like the reason we do
1182 * it this way is that _mesa_add_state_reference doesn't work
1183 * for things that aren't vec4s, so the tokens[2]/tokens[3]
1184 * range has to be equal.
1186 for (i
= 0; i
< 4; i
++) {
1189 int pos
= _mesa_add_state_reference(prog
->Parameters
,
1190 (gl_state_index
*)tokens
);
1194 assert(base_pos
+ i
== pos
);
1200 static variable_storage
*
1201 get_builtin_matrix_ref(void *mem_ctx
, struct gl_program
*prog
, ir_variable
*var
,
1202 ir_rvalue
*array_index
)
1205 * NOTE: The ARB_vertex_program extension specified that matrices get
1206 * loaded in registers in row-major order. With GLSL, we want column-
1207 * major order. So, we need to transpose all matrices here...
1209 static const struct {
1214 { "gl_ModelViewMatrix", STATE_MODELVIEW_MATRIX
, STATE_MATRIX_TRANSPOSE
},
1215 { "gl_ModelViewMatrixInverse", STATE_MODELVIEW_MATRIX
, STATE_MATRIX_INVTRANS
},
1216 { "gl_ModelViewMatrixTranspose", STATE_MODELVIEW_MATRIX
, 0 },
1217 { "gl_ModelViewMatrixInverseTranspose", STATE_MODELVIEW_MATRIX
, STATE_MATRIX_INVERSE
},
1219 { "gl_ProjectionMatrix", STATE_PROJECTION_MATRIX
, STATE_MATRIX_TRANSPOSE
},
1220 { "gl_ProjectionMatrixInverse", STATE_PROJECTION_MATRIX
, STATE_MATRIX_INVTRANS
},
1221 { "gl_ProjectionMatrixTranspose", STATE_PROJECTION_MATRIX
, 0 },
1222 { "gl_ProjectionMatrixInverseTranspose", STATE_PROJECTION_MATRIX
, STATE_MATRIX_INVERSE
},
1224 { "gl_ModelViewProjectionMatrix", STATE_MVP_MATRIX
, STATE_MATRIX_TRANSPOSE
},
1225 { "gl_ModelViewProjectionMatrixInverse", STATE_MVP_MATRIX
, STATE_MATRIX_INVTRANS
},
1226 { "gl_ModelViewProjectionMatrixTranspose", STATE_MVP_MATRIX
, 0 },
1227 { "gl_ModelViewProjectionMatrixInverseTranspose", STATE_MVP_MATRIX
, STATE_MATRIX_INVERSE
},
1229 { "gl_TextureMatrix", STATE_TEXTURE_MATRIX
, STATE_MATRIX_TRANSPOSE
},
1230 { "gl_TextureMatrixInverse", STATE_TEXTURE_MATRIX
, STATE_MATRIX_INVTRANS
},
1231 { "gl_TextureMatrixTranspose", STATE_TEXTURE_MATRIX
, 0 },
1232 { "gl_TextureMatrixInverseTranspose", STATE_TEXTURE_MATRIX
, STATE_MATRIX_INVERSE
},
1234 { "gl_NormalMatrix", STATE_MODELVIEW_MATRIX
, STATE_MATRIX_INVERSE
},
1238 variable_storage
*entry
;
1240 /* C++ gets angry when we try to use an int as a gl_state_index, so we use
1241 * ints for gl_state_index. Make sure they're compatible.
1243 assert(sizeof(gl_state_index
) == sizeof(int));
1245 for (i
= 0; i
< Elements(matrices
); i
++) {
1246 if (strcmp(var
->name
, matrices
[i
].name
) == 0) {
1247 int tokens
[STATE_LENGTH
];
1250 tokens
[0] = matrices
[i
].matrix
;
1251 tokens
[4] = matrices
[i
].modifier
;
1252 if (matrices
[i
].matrix
== STATE_TEXTURE_MATRIX
) {
1253 ir_constant
*index
= array_index
->constant_expression_value();
1255 tokens
[1] = index
->value
.i
[0];
1256 base_pos
= add_matrix_ref(prog
, tokens
);
1258 for (i
= 0; i
< var
->type
->length
; i
++) {
1260 int pos
= add_matrix_ref(prog
, tokens
);
1264 assert(base_pos
+ (int)i
* 4 == pos
);
1268 tokens
[1] = 0; /* unused array index */
1269 base_pos
= add_matrix_ref(prog
, tokens
);
1271 tokens
[4] = matrices
[i
].modifier
;
1273 entry
= new(mem_ctx
) variable_storage(var
,
1285 ir_to_mesa_visitor::add_uniform(const char *name
,
1286 const glsl_type
*type
,
1287 ir_constant
*constant
)
1291 if (type
->is_vector() ||
1292 type
->is_scalar()) {
1293 len
= type
->vector_elements
;
1295 len
= type_size(type
) * 4;
1298 float *values
= NULL
;
1299 if (constant
&& type
->is_array()) {
1300 values
= (float *)malloc(type
->length
* 4 * sizeof(float));
1302 assert(type
->fields
.array
->is_scalar() ||
1303 type
->fields
.array
->is_vector() ||
1304 !"FINISHME: uniform array initializers for non-vector");
1306 for (unsigned int i
= 0; i
< type
->length
; i
++) {
1307 ir_constant
*element
= constant
->array_elements
[i
];
1310 for (c
= 0; c
< type
->fields
.array
->vector_elements
; c
++) {
1311 switch (type
->fields
.array
->base_type
) {
1312 case GLSL_TYPE_FLOAT
:
1313 values
[4 * i
+ c
] = element
->value
.f
[c
];
1316 values
[4 * i
+ c
] = element
->value
.i
[c
];
1318 case GLSL_TYPE_UINT
:
1319 values
[4 * i
+ c
] = element
->value
.u
[c
];
1321 case GLSL_TYPE_BOOL
:
1322 values
[4 * i
+ c
] = element
->value
.b
[c
];
1325 assert(!"not reached");
1329 } else if (constant
) {
1330 values
= (float *)malloc(16 * sizeof(float));
1331 for (unsigned int i
= 0; i
< type
->components(); i
++) {
1332 switch (type
->base_type
) {
1333 case GLSL_TYPE_FLOAT
:
1334 values
[i
] = constant
->value
.f
[i
];
1337 values
[i
] = constant
->value
.i
[i
];
1339 case GLSL_TYPE_UINT
:
1340 values
[i
] = constant
->value
.u
[i
];
1342 case GLSL_TYPE_BOOL
:
1343 values
[i
] = constant
->value
.b
[i
];
1346 assert(!"not reached");
1351 int loc
= _mesa_add_uniform(this->prog
->Parameters
,
1361 /* Recursively add all the members of the aggregate uniform as uniform names
1362 * to Mesa, moving those uniforms to our structured temporary.
1365 ir_to_mesa_visitor::add_aggregate_uniform(ir_instruction
*ir
,
1367 const struct glsl_type
*type
,
1368 ir_constant
*constant
,
1369 struct ir_to_mesa_dst_reg temp
)
1373 if (type
->is_record()) {
1374 void *mem_ctx
= talloc_new(NULL
);
1375 ir_constant
*field_constant
= NULL
;
1378 field_constant
= (ir_constant
*)constant
->components
.get_head();
1380 for (unsigned int i
= 0; i
< type
->length
; i
++) {
1381 const glsl_type
*field_type
= type
->fields
.structure
[i
].type
;
1383 add_aggregate_uniform(ir
,
1384 talloc_asprintf(mem_ctx
, "%s.%s", name
,
1385 type
->fields
.structure
[i
].name
),
1386 field_type
, field_constant
, temp
);
1387 temp
.index
+= type_size(field_type
);
1390 field_constant
= (ir_constant
*)field_constant
->next
;
1393 talloc_free(mem_ctx
);
1398 assert(type
->is_vector() || type
->is_scalar() || !"FINISHME: other types");
1400 loc
= add_uniform(name
, type
, constant
);
1402 ir_to_mesa_src_reg
uniform(PROGRAM_UNIFORM
, loc
, type
);
1404 for (int i
= 0; i
< type_size(type
); i
++) {
1405 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, temp
, uniform
);
1413 ir_to_mesa_visitor::visit(ir_dereference_variable
*ir
)
1415 variable_storage
*entry
= find_variable_storage(ir
->var
);
1419 switch (ir
->var
->mode
) {
1420 case ir_var_uniform
:
1421 entry
= get_builtin_matrix_ref(this->mem_ctx
, this->prog
, ir
->var
,
1426 /* FINISHME: Fix up uniform name for arrays and things */
1427 if (ir
->var
->type
->base_type
== GLSL_TYPE_SAMPLER
||
1428 (ir
->var
->type
->base_type
== GLSL_TYPE_ARRAY
&&
1429 ir
->var
->type
->fields
.array
->base_type
== GLSL_TYPE_SAMPLER
)) {
1432 if (ir
->var
->type
->base_type
== GLSL_TYPE_ARRAY
)
1433 array_length
= ir
->var
->type
->length
;
1436 int sampler
= _mesa_add_sampler(this->prog
->Parameters
,
1438 ir
->var
->type
->gl_type
,
1440 set_sampler_location(ir
->var
, sampler
);
1442 entry
= new(mem_ctx
) variable_storage(ir
->var
, PROGRAM_SAMPLER
,
1444 this->variables
.push_tail(entry
);
1448 assert(ir
->var
->type
->gl_type
!= 0 &&
1449 ir
->var
->type
->gl_type
!= GL_INVALID_ENUM
);
1451 /* Oh, the joy of aggregate types in Mesa. Like constants,
1452 * we can only really do vec4s. So, make a temp, chop the
1453 * aggregate up into vec4s, and move those vec4s to the temp.
1455 if (ir
->var
->type
->is_record()) {
1456 ir_to_mesa_src_reg temp
= get_temp(ir
->var
->type
);
1458 entry
= new(mem_ctx
) variable_storage(ir
->var
,
1461 this->variables
.push_tail(entry
);
1463 add_aggregate_uniform(ir
->var
, ir
->var
->name
, ir
->var
->type
,
1464 ir
->var
->constant_value
,
1465 ir_to_mesa_dst_reg_from_src(temp
));
1469 loc
= add_uniform(ir
->var
->name
,
1471 ir
->var
->constant_value
);
1473 /* Always mark the uniform used at this point. If it isn't
1474 * used, dead code elimination should have nuked the decl already.
1476 this->prog
->Parameters
->Parameters
[loc
].Used
= GL_TRUE
;
1478 entry
= new(mem_ctx
) variable_storage(ir
->var
, PROGRAM_UNIFORM
, loc
);
1479 this->variables
.push_tail(entry
);
1484 /* The linker assigns locations for varyings and attributes,
1485 * including deprecated builtins (like gl_Color), user-assign
1486 * generic attributes (glBindVertexLocation), and
1487 * user-defined varyings.
1489 * FINISHME: We would hit this path for function arguments. Fix!
1491 assert(ir
->var
->location
!= -1);
1492 if (ir
->var
->mode
== ir_var_in
||
1493 ir
->var
->mode
== ir_var_inout
) {
1494 entry
= new(mem_ctx
) variable_storage(ir
->var
,
1498 if (this->prog
->Target
== GL_VERTEX_PROGRAM_ARB
&&
1499 ir
->var
->location
>= VERT_ATTRIB_GENERIC0
) {
1500 _mesa_add_attribute(prog
->Attributes
,
1502 _mesa_sizeof_glsl_type(ir
->var
->type
->gl_type
),
1503 ir
->var
->type
->gl_type
,
1504 ir
->var
->location
- VERT_ATTRIB_GENERIC0
);
1507 entry
= new(mem_ctx
) variable_storage(ir
->var
,
1514 case ir_var_temporary
:
1515 entry
= new(mem_ctx
) variable_storage(ir
->var
, PROGRAM_TEMPORARY
,
1517 this->variables
.push_tail(entry
);
1519 next_temp
+= type_size(ir
->var
->type
);
1524 printf("Failed to make storage for %s\n", ir
->var
->name
);
1529 this->result
= ir_to_mesa_src_reg(entry
->file
, entry
->index
, ir
->var
->type
);
1533 ir_to_mesa_visitor::visit(ir_dereference_array
*ir
)
1535 ir_variable
*var
= ir
->variable_referenced();
1537 ir_to_mesa_src_reg src_reg
;
1538 ir_dereference_variable
*deref_var
= ir
->array
->as_dereference_variable();
1539 int element_size
= type_size(ir
->type
);
1541 index
= ir
->array_index
->constant_expression_value();
1543 if (deref_var
&& strncmp(deref_var
->var
->name
,
1545 strlen("gl_TextureMatrix")) == 0) {
1546 struct variable_storage
*entry
;
1548 entry
= get_builtin_matrix_ref(this->mem_ctx
, this->prog
, deref_var
->var
,
1552 ir_to_mesa_src_reg
src_reg(entry
->file
, entry
->index
, ir
->type
);
1555 src_reg
.reladdr
= NULL
;
1557 ir_to_mesa_src_reg index_reg
= get_temp(glsl_type::float_type
);
1559 ir
->array_index
->accept(this);
1560 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
,
1561 ir_to_mesa_dst_reg_from_src(index_reg
),
1562 this->result
, src_reg_for_float(element_size
));
1564 src_reg
.reladdr
= talloc(mem_ctx
, ir_to_mesa_src_reg
);
1565 memcpy(src_reg
.reladdr
, &index_reg
, sizeof(index_reg
));
1568 this->result
= src_reg
;
1572 if (strncmp(var
->name
, "gl_", 3) == 0 && var
->mode
== ir_var_uniform
&&
1573 !var
->type
->is_matrix()) {
1574 ir_dereference_record
*record
= NULL
;
1575 if (ir
->array
->ir_type
== ir_type_dereference_record
)
1576 record
= (ir_dereference_record
*)ir
->array
;
1578 assert(index
|| !"FINISHME: variable-indexed builtin uniform access");
1580 this->result
= get_builtin_uniform_reg(prog
,
1583 record
? record
->field
: NULL
);
1586 ir
->array
->accept(this);
1587 src_reg
= this->result
;
1590 src_reg
.index
+= index
->value
.i
[0] * element_size
;
1592 ir_to_mesa_src_reg array_base
= this->result
;
1593 /* Variable index array dereference. It eats the "vec4" of the
1594 * base of the array and an index that offsets the Mesa register
1597 ir
->array_index
->accept(this);
1599 ir_to_mesa_src_reg index_reg
;
1601 if (element_size
== 1) {
1602 index_reg
= this->result
;
1604 index_reg
= get_temp(glsl_type::float_type
);
1606 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
,
1607 ir_to_mesa_dst_reg_from_src(index_reg
),
1608 this->result
, src_reg_for_float(element_size
));
1611 src_reg
.reladdr
= talloc(mem_ctx
, ir_to_mesa_src_reg
);
1612 memcpy(src_reg
.reladdr
, &index_reg
, sizeof(index_reg
));
1615 /* If the type is smaller than a vec4, replicate the last channel out. */
1616 if (ir
->type
->is_scalar() || ir
->type
->is_vector())
1617 src_reg
.swizzle
= swizzle_for_size(ir
->type
->vector_elements
);
1619 src_reg
.swizzle
= SWIZZLE_NOOP
;
1621 this->result
= src_reg
;
1625 ir_to_mesa_visitor::visit(ir_dereference_record
*ir
)
1628 const glsl_type
*struct_type
= ir
->record
->type
;
1630 ir_variable
*var
= ir
->record
->variable_referenced();
1632 if (strncmp(var
->name
, "gl_", 3) == 0 && var
->mode
== ir_var_uniform
) {
1635 this->result
= get_builtin_uniform_reg(prog
,
1642 ir
->record
->accept(this);
1644 for (i
= 0; i
< struct_type
->length
; i
++) {
1645 if (strcmp(struct_type
->fields
.structure
[i
].name
, ir
->field
) == 0)
1647 offset
+= type_size(struct_type
->fields
.structure
[i
].type
);
1649 this->result
.swizzle
= swizzle_for_size(ir
->type
->vector_elements
);
1650 this->result
.index
+= offset
;
1654 * We want to be careful in assignment setup to hit the actual storage
1655 * instead of potentially using a temporary like we might with the
1656 * ir_dereference handler.
1658 static struct ir_to_mesa_dst_reg
1659 get_assignment_lhs(ir_dereference
*ir
, ir_to_mesa_visitor
*v
)
1661 /* The LHS must be a dereference. If the LHS is a variable indexed array
1662 * access of a vector, it must be separated into a series conditional moves
1663 * before reaching this point (see ir_vec_index_to_cond_assign).
1665 assert(ir
->as_dereference());
1666 ir_dereference_array
*deref_array
= ir
->as_dereference_array();
1668 assert(!deref_array
->array
->type
->is_vector());
1671 /* Use the rvalue deref handler for the most part. We'll ignore
1672 * swizzles in it and write swizzles using writemask, though.
1675 return ir_to_mesa_dst_reg_from_src(v
->result
);
1679 ir_to_mesa_visitor::visit(ir_assignment
*ir
)
1681 struct ir_to_mesa_dst_reg l
;
1682 struct ir_to_mesa_src_reg r
;
1685 ir
->rhs
->accept(this);
1688 l
= get_assignment_lhs(ir
->lhs
, this);
1690 /* FINISHME: This should really set to the correct maximal writemask for each
1691 * FINISHME: component written (in the loops below). This case can only
1692 * FINISHME: occur for matrices, arrays, and structures.
1694 if (ir
->write_mask
== 0) {
1695 assert(!ir
->lhs
->type
->is_scalar() && !ir
->lhs
->type
->is_vector());
1696 l
.writemask
= WRITEMASK_XYZW
;
1697 } else if (ir
->lhs
->type
->is_scalar()) {
1698 /* FINISHME: This hack makes writing to gl_FragData, which lives in the
1699 * FINISHME: W component of fragment shader output zero, work correctly.
1701 l
.writemask
= WRITEMASK_XYZW
;
1703 assert(ir
->lhs
->type
->is_vector());
1704 l
.writemask
= ir
->write_mask
;
1707 assert(l
.file
!= PROGRAM_UNDEFINED
);
1708 assert(r
.file
!= PROGRAM_UNDEFINED
);
1710 if (ir
->condition
) {
1711 ir_to_mesa_src_reg condition
;
1713 ir
->condition
->accept(this);
1714 condition
= this->result
;
1716 /* We use the OPCODE_CMP (a < 0 ? b : c) for conditional moves,
1717 * and the condition we produced is 0.0 or 1.0. By flipping the
1718 * sign, we can choose which value OPCODE_CMP produces without
1719 * an extra computing the condition.
1721 condition
.negate
= ~condition
.negate
;
1722 for (i
= 0; i
< type_size(ir
->lhs
->type
); i
++) {
1723 ir_to_mesa_emit_op3(ir
, OPCODE_CMP
, l
,
1724 condition
, r
, ir_to_mesa_src_reg_from_dst(l
));
1729 for (i
= 0; i
< type_size(ir
->lhs
->type
); i
++) {
1730 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1739 ir_to_mesa_visitor::visit(ir_constant
*ir
)
1741 ir_to_mesa_src_reg src_reg
;
1742 GLfloat stack_vals
[4];
1743 GLfloat
*values
= stack_vals
;
1746 /* Unfortunately, 4 floats is all we can get into
1747 * _mesa_add_unnamed_constant. So, make a temp to store an
1748 * aggregate constant and move each constant value into it. If we
1749 * get lucky, copy propagation will eliminate the extra moves.
1752 if (ir
->type
->base_type
== GLSL_TYPE_STRUCT
) {
1753 ir_to_mesa_src_reg temp_base
= get_temp(ir
->type
);
1754 ir_to_mesa_dst_reg temp
= ir_to_mesa_dst_reg_from_src(temp_base
);
1756 foreach_iter(exec_list_iterator
, iter
, ir
->components
) {
1757 ir_constant
*field_value
= (ir_constant
*)iter
.get();
1758 int size
= type_size(field_value
->type
);
1762 field_value
->accept(this);
1763 src_reg
= this->result
;
1765 for (i
= 0; i
< (unsigned int)size
; i
++) {
1766 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, temp
, src_reg
);
1772 this->result
= temp_base
;
1776 if (ir
->type
->is_array()) {
1777 ir_to_mesa_src_reg temp_base
= get_temp(ir
->type
);
1778 ir_to_mesa_dst_reg temp
= ir_to_mesa_dst_reg_from_src(temp_base
);
1779 int size
= type_size(ir
->type
->fields
.array
);
1783 for (i
= 0; i
< ir
->type
->length
; i
++) {
1784 ir
->array_elements
[i
]->accept(this);
1785 src_reg
= this->result
;
1786 for (int j
= 0; j
< size
; j
++) {
1787 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, temp
, src_reg
);
1793 this->result
= temp_base
;
1797 if (ir
->type
->is_matrix()) {
1798 ir_to_mesa_src_reg mat
= get_temp(ir
->type
);
1799 ir_to_mesa_dst_reg mat_column
= ir_to_mesa_dst_reg_from_src(mat
);
1801 for (i
= 0; i
< ir
->type
->matrix_columns
; i
++) {
1802 assert(ir
->type
->base_type
== GLSL_TYPE_FLOAT
);
1803 values
= &ir
->value
.f
[i
* ir
->type
->vector_elements
];
1805 src_reg
= ir_to_mesa_src_reg(PROGRAM_CONSTANT
, -1, NULL
);
1806 src_reg
.index
= _mesa_add_unnamed_constant(this->prog
->Parameters
,
1808 ir
->type
->vector_elements
,
1810 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, mat_column
, src_reg
);
1818 src_reg
.file
= PROGRAM_CONSTANT
;
1819 switch (ir
->type
->base_type
) {
1820 case GLSL_TYPE_FLOAT
:
1821 values
= &ir
->value
.f
[0];
1823 case GLSL_TYPE_UINT
:
1824 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1825 values
[i
] = ir
->value
.u
[i
];
1829 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1830 values
[i
] = ir
->value
.i
[i
];
1833 case GLSL_TYPE_BOOL
:
1834 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1835 values
[i
] = ir
->value
.b
[i
];
1839 assert(!"Non-float/uint/int/bool constant");
1842 this->result
= ir_to_mesa_src_reg(PROGRAM_CONSTANT
, -1, ir
->type
);
1843 this->result
.index
= _mesa_add_unnamed_constant(this->prog
->Parameters
,
1845 ir
->type
->vector_elements
,
1846 &this->result
.swizzle
);
1850 ir_to_mesa_visitor::get_function_signature(ir_function_signature
*sig
)
1852 function_entry
*entry
;
1854 foreach_iter(exec_list_iterator
, iter
, this->function_signatures
) {
1855 entry
= (function_entry
*)iter
.get();
1857 if (entry
->sig
== sig
)
1861 entry
= talloc(mem_ctx
, function_entry
);
1863 entry
->sig_id
= this->next_signature_id
++;
1864 entry
->bgn_inst
= NULL
;
1866 /* Allocate storage for all the parameters. */
1867 foreach_iter(exec_list_iterator
, iter
, sig
->parameters
) {
1868 ir_variable
*param
= (ir_variable
*)iter
.get();
1869 variable_storage
*storage
;
1871 storage
= find_variable_storage(param
);
1874 storage
= new(mem_ctx
) variable_storage(param
, PROGRAM_TEMPORARY
,
1876 this->variables
.push_tail(storage
);
1878 this->next_temp
+= type_size(param
->type
);
1881 if (!sig
->return_type
->is_void()) {
1882 entry
->return_reg
= get_temp(sig
->return_type
);
1884 entry
->return_reg
= ir_to_mesa_undef
;
1887 this->function_signatures
.push_tail(entry
);
1892 ir_to_mesa_visitor::visit(ir_call
*ir
)
1894 ir_to_mesa_instruction
*call_inst
;
1895 ir_function_signature
*sig
= ir
->get_callee();
1896 function_entry
*entry
= get_function_signature(sig
);
1899 /* Process in parameters. */
1900 exec_list_iterator sig_iter
= sig
->parameters
.iterator();
1901 foreach_iter(exec_list_iterator
, iter
, *ir
) {
1902 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
1903 ir_variable
*param
= (ir_variable
*)sig_iter
.get();
1905 if (param
->mode
== ir_var_in
||
1906 param
->mode
== ir_var_inout
) {
1907 variable_storage
*storage
= find_variable_storage(param
);
1910 param_rval
->accept(this);
1911 ir_to_mesa_src_reg r
= this->result
;
1913 ir_to_mesa_dst_reg l
;
1914 l
.file
= storage
->file
;
1915 l
.index
= storage
->index
;
1917 l
.writemask
= WRITEMASK_XYZW
;
1918 l
.cond_mask
= COND_TR
;
1920 for (i
= 0; i
< type_size(param
->type
); i
++) {
1921 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1929 assert(!sig_iter
.has_next());
1931 /* Emit call instruction */
1932 call_inst
= ir_to_mesa_emit_op1(ir
, OPCODE_CAL
,
1933 ir_to_mesa_undef_dst
, ir_to_mesa_undef
);
1934 call_inst
->function
= entry
;
1936 /* Process out parameters. */
1937 sig_iter
= sig
->parameters
.iterator();
1938 foreach_iter(exec_list_iterator
, iter
, *ir
) {
1939 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
1940 ir_variable
*param
= (ir_variable
*)sig_iter
.get();
1942 if (param
->mode
== ir_var_out
||
1943 param
->mode
== ir_var_inout
) {
1944 variable_storage
*storage
= find_variable_storage(param
);
1947 ir_to_mesa_src_reg r
;
1948 r
.file
= storage
->file
;
1949 r
.index
= storage
->index
;
1951 r
.swizzle
= SWIZZLE_NOOP
;
1954 param_rval
->accept(this);
1955 ir_to_mesa_dst_reg l
= ir_to_mesa_dst_reg_from_src(this->result
);
1957 for (i
= 0; i
< type_size(param
->type
); i
++) {
1958 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1966 assert(!sig_iter
.has_next());
1968 /* Process return value. */
1969 this->result
= entry
->return_reg
;
1974 ir_to_mesa_visitor::visit(ir_texture
*ir
)
1976 ir_to_mesa_src_reg result_src
, coord
, lod_info
, projector
;
1977 ir_to_mesa_dst_reg result_dst
, coord_dst
;
1978 ir_to_mesa_instruction
*inst
= NULL
;
1979 prog_opcode opcode
= OPCODE_NOP
;
1981 ir
->coordinate
->accept(this);
1983 /* Put our coords in a temp. We'll need to modify them for shadow,
1984 * projection, or LOD, so the only case we'd use it as is is if
1985 * we're doing plain old texturing. Mesa IR optimization should
1986 * handle cleaning up our mess in that case.
1988 coord
= get_temp(glsl_type::vec4_type
);
1989 coord_dst
= ir_to_mesa_dst_reg_from_src(coord
);
1990 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
,
1993 if (ir
->projector
) {
1994 ir
->projector
->accept(this);
1995 projector
= this->result
;
1998 /* Storage for our result. Ideally for an assignment we'd be using
1999 * the actual storage for the result here, instead.
2001 result_src
= get_temp(glsl_type::vec4_type
);
2002 result_dst
= ir_to_mesa_dst_reg_from_src(result_src
);
2006 opcode
= OPCODE_TEX
;
2009 opcode
= OPCODE_TXB
;
2010 ir
->lod_info
.bias
->accept(this);
2011 lod_info
= this->result
;
2014 opcode
= OPCODE_TXL
;
2015 ir
->lod_info
.lod
->accept(this);
2016 lod_info
= this->result
;
2020 assert(!"GLSL 1.30 features unsupported");
2024 if (ir
->projector
) {
2025 if (opcode
== OPCODE_TEX
) {
2026 /* Slot the projector in as the last component of the coord. */
2027 coord_dst
.writemask
= WRITEMASK_W
;
2028 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, projector
);
2029 coord_dst
.writemask
= WRITEMASK_XYZW
;
2030 opcode
= OPCODE_TXP
;
2032 ir_to_mesa_src_reg coord_w
= coord
;
2033 coord_w
.swizzle
= SWIZZLE_WWWW
;
2035 /* For the other TEX opcodes there's no projective version
2036 * since the last slot is taken up by lod info. Do the
2037 * projective divide now.
2039 coord_dst
.writemask
= WRITEMASK_W
;
2040 ir_to_mesa_emit_op1(ir
, OPCODE_RCP
, coord_dst
, projector
);
2042 coord_dst
.writemask
= WRITEMASK_XYZ
;
2043 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
, coord_dst
, coord
, coord_w
);
2045 coord_dst
.writemask
= WRITEMASK_XYZW
;
2046 coord
.swizzle
= SWIZZLE_XYZW
;
2050 if (ir
->shadow_comparitor
) {
2051 /* Slot the shadow value in as the second to last component of the
2054 ir
->shadow_comparitor
->accept(this);
2055 coord_dst
.writemask
= WRITEMASK_Z
;
2056 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, this->result
);
2057 coord_dst
.writemask
= WRITEMASK_XYZW
;
2060 if (opcode
== OPCODE_TXL
|| opcode
== OPCODE_TXB
) {
2061 /* Mesa IR stores lod or lod bias in the last channel of the coords. */
2062 coord_dst
.writemask
= WRITEMASK_W
;
2063 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, lod_info
);
2064 coord_dst
.writemask
= WRITEMASK_XYZW
;
2067 inst
= ir_to_mesa_emit_op1(ir
, opcode
, result_dst
, coord
);
2069 if (ir
->shadow_comparitor
)
2070 inst
->tex_shadow
= GL_TRUE
;
2072 ir_variable
*sampler
= ir
->sampler
->variable_referenced();
2074 /* generate the mapping, remove when we generate storage at
2077 ir
->sampler
->accept(this);
2079 inst
->sampler
= get_sampler_location(sampler
);
2081 ir_dereference_array
*sampler_array
= ir
->sampler
->as_dereference_array();
2082 if (sampler_array
) {
2083 ir_constant
*array_index
=
2084 sampler_array
->array_index
->constant_expression_value();
2086 /* GLSL 1.10 and 1.20 allowed variable sampler array indices,
2087 * while GLSL 1.30 requires that the array indices be constant
2088 * integer expressions. We don't expect any driver to actually
2089 * work with a really variable array index, and in 1.20 all that
2090 * would work would be an unrolled loop counter, so assert that
2091 * we ended up with a constant at least..
2093 assert(array_index
);
2094 inst
->sampler
+= array_index
->value
.i
[0];
2097 const glsl_type
*sampler_type
= sampler
->type
;
2098 while (sampler_type
->base_type
== GLSL_TYPE_ARRAY
)
2099 sampler_type
= sampler_type
->fields
.array
;
2101 switch (sampler_type
->sampler_dimensionality
) {
2102 case GLSL_SAMPLER_DIM_1D
:
2103 inst
->tex_target
= (sampler_type
->sampler_array
)
2104 ? TEXTURE_1D_ARRAY_INDEX
: TEXTURE_1D_INDEX
;
2106 case GLSL_SAMPLER_DIM_2D
:
2107 inst
->tex_target
= (sampler_type
->sampler_array
)
2108 ? TEXTURE_2D_ARRAY_INDEX
: TEXTURE_2D_INDEX
;
2110 case GLSL_SAMPLER_DIM_3D
:
2111 inst
->tex_target
= TEXTURE_3D_INDEX
;
2113 case GLSL_SAMPLER_DIM_CUBE
:
2114 inst
->tex_target
= TEXTURE_CUBE_INDEX
;
2116 case GLSL_SAMPLER_DIM_RECT
:
2117 inst
->tex_target
= TEXTURE_RECT_INDEX
;
2119 case GLSL_SAMPLER_DIM_BUF
:
2120 assert(!"FINISHME: Implement ARB_texture_buffer_object");
2123 assert(!"Should not get here.");
2126 this->result
= result_src
;
2130 ir_to_mesa_visitor::visit(ir_return
*ir
)
2132 if (ir
->get_value()) {
2133 ir_to_mesa_dst_reg l
;
2136 assert(current_function
);
2138 ir
->get_value()->accept(this);
2139 ir_to_mesa_src_reg r
= this->result
;
2141 l
= ir_to_mesa_dst_reg_from_src(current_function
->return_reg
);
2143 for (i
= 0; i
< type_size(current_function
->sig
->return_type
); i
++) {
2144 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
2150 ir_to_mesa_emit_op0(ir
, OPCODE_RET
);
2154 ir_to_mesa_visitor::visit(ir_discard
*ir
)
2156 assert(ir
->condition
== NULL
); /* FINISHME */
2158 ir_to_mesa_emit_op0(ir
, OPCODE_KIL_NV
);
2162 ir_to_mesa_visitor::visit(ir_if
*ir
)
2164 ir_to_mesa_instruction
*cond_inst
, *if_inst
, *else_inst
= NULL
;
2165 ir_to_mesa_instruction
*prev_inst
;
2167 prev_inst
= (ir_to_mesa_instruction
*)this->instructions
.get_tail();
2169 ir
->condition
->accept(this);
2170 assert(this->result
.file
!= PROGRAM_UNDEFINED
);
2172 if (ctx
->Shader
.EmitCondCodes
) {
2173 cond_inst
= (ir_to_mesa_instruction
*)this->instructions
.get_tail();
2175 /* See if we actually generated any instruction for generating
2176 * the condition. If not, then cook up a move to a temp so we
2177 * have something to set cond_update on.
2179 if (cond_inst
== prev_inst
) {
2180 ir_to_mesa_src_reg temp
= get_temp(glsl_type::bool_type
);
2181 cond_inst
= ir_to_mesa_emit_op1(ir
->condition
, OPCODE_MOV
,
2182 ir_to_mesa_dst_reg_from_src(temp
),
2185 cond_inst
->cond_update
= GL_TRUE
;
2187 if_inst
= ir_to_mesa_emit_op0(ir
->condition
, OPCODE_IF
);
2188 if_inst
->dst_reg
.cond_mask
= COND_NE
;
2190 if_inst
= ir_to_mesa_emit_op1(ir
->condition
,
2191 OPCODE_IF
, ir_to_mesa_undef_dst
,
2195 this->instructions
.push_tail(if_inst
);
2197 visit_exec_list(&ir
->then_instructions
, this);
2199 if (!ir
->else_instructions
.is_empty()) {
2200 else_inst
= ir_to_mesa_emit_op0(ir
->condition
, OPCODE_ELSE
);
2201 visit_exec_list(&ir
->else_instructions
, this);
2204 if_inst
= ir_to_mesa_emit_op1(ir
->condition
, OPCODE_ENDIF
,
2205 ir_to_mesa_undef_dst
, ir_to_mesa_undef
);
2208 ir_to_mesa_visitor::ir_to_mesa_visitor()
2210 result
.file
= PROGRAM_UNDEFINED
;
2212 next_signature_id
= 1;
2214 current_function
= NULL
;
2215 mem_ctx
= talloc_new(NULL
);
2218 ir_to_mesa_visitor::~ir_to_mesa_visitor()
2220 talloc_free(mem_ctx
);
2221 if (this->sampler_map
)
2222 hash_table_dtor(this->sampler_map
);
2225 static struct prog_src_register
2226 mesa_src_reg_from_ir_src_reg(ir_to_mesa_src_reg reg
)
2228 struct prog_src_register mesa_reg
;
2230 mesa_reg
.File
= reg
.file
;
2231 assert(reg
.index
< (1 << INST_INDEX_BITS
) - 1);
2232 mesa_reg
.Index
= reg
.index
;
2233 mesa_reg
.Swizzle
= reg
.swizzle
;
2234 mesa_reg
.RelAddr
= reg
.reladdr
!= NULL
;
2235 mesa_reg
.Negate
= reg
.negate
;
2237 mesa_reg
.HasIndex2
= GL_FALSE
;
2243 set_branchtargets(ir_to_mesa_visitor
*v
,
2244 struct prog_instruction
*mesa_instructions
,
2245 int num_instructions
)
2247 int if_count
= 0, loop_count
= 0;
2248 int *if_stack
, *loop_stack
;
2249 int if_stack_pos
= 0, loop_stack_pos
= 0;
2252 for (i
= 0; i
< num_instructions
; i
++) {
2253 switch (mesa_instructions
[i
].Opcode
) {
2257 case OPCODE_BGNLOOP
:
2262 mesa_instructions
[i
].BranchTarget
= -1;
2269 if_stack
= talloc_zero_array(v
->mem_ctx
, int, if_count
);
2270 loop_stack
= talloc_zero_array(v
->mem_ctx
, int, loop_count
);
2272 for (i
= 0; i
< num_instructions
; i
++) {
2273 switch (mesa_instructions
[i
].Opcode
) {
2275 if_stack
[if_stack_pos
] = i
;
2279 mesa_instructions
[if_stack
[if_stack_pos
- 1]].BranchTarget
= i
;
2280 if_stack
[if_stack_pos
- 1] = i
;
2283 mesa_instructions
[if_stack
[if_stack_pos
- 1]].BranchTarget
= i
;
2286 case OPCODE_BGNLOOP
:
2287 loop_stack
[loop_stack_pos
] = i
;
2290 case OPCODE_ENDLOOP
:
2292 /* Rewrite any breaks/conts at this nesting level (haven't
2293 * already had a BranchTarget assigned) to point to the end
2296 for (j
= loop_stack
[loop_stack_pos
]; j
< i
; j
++) {
2297 if (mesa_instructions
[j
].Opcode
== OPCODE_BRK
||
2298 mesa_instructions
[j
].Opcode
== OPCODE_CONT
) {
2299 if (mesa_instructions
[j
].BranchTarget
== -1) {
2300 mesa_instructions
[j
].BranchTarget
= i
;
2304 /* The loop ends point at each other. */
2305 mesa_instructions
[i
].BranchTarget
= loop_stack
[loop_stack_pos
];
2306 mesa_instructions
[loop_stack
[loop_stack_pos
]].BranchTarget
= i
;
2309 foreach_iter(exec_list_iterator
, iter
, v
->function_signatures
) {
2310 function_entry
*entry
= (function_entry
*)iter
.get();
2312 if (entry
->sig_id
== mesa_instructions
[i
].BranchTarget
) {
2313 mesa_instructions
[i
].BranchTarget
= entry
->inst
;
2325 print_program(struct prog_instruction
*mesa_instructions
,
2326 ir_instruction
**mesa_instruction_annotation
,
2327 int num_instructions
)
2329 ir_instruction
*last_ir
= NULL
;
2333 for (i
= 0; i
< num_instructions
; i
++) {
2334 struct prog_instruction
*mesa_inst
= mesa_instructions
+ i
;
2335 ir_instruction
*ir
= mesa_instruction_annotation
[i
];
2337 fprintf(stdout
, "%3d: ", i
);
2339 if (last_ir
!= ir
&& ir
) {
2342 for (j
= 0; j
< indent
; j
++) {
2343 fprintf(stdout
, " ");
2349 fprintf(stdout
, " "); /* line number spacing. */
2352 indent
= _mesa_fprint_instruction_opt(stdout
, mesa_inst
, indent
,
2353 PROG_PRINT_DEBUG
, NULL
);
2358 count_resources(struct gl_program
*prog
)
2362 prog
->SamplersUsed
= 0;
2364 for (i
= 0; i
< prog
->NumInstructions
; i
++) {
2365 struct prog_instruction
*inst
= &prog
->Instructions
[i
];
2367 /* Instead of just using the uniform's value to map to a
2368 * sampler, Mesa first allocates a separate number for the
2369 * sampler (_mesa_add_sampler), then we reindex it down to a
2370 * small integer (sampler_map[], SamplersUsed), then that gets
2371 * mapped to the uniform's value, and we get an actual sampler.
2373 if (_mesa_is_tex_instruction(inst
->Opcode
)) {
2374 prog
->SamplerTargets
[inst
->TexSrcUnit
] =
2375 (gl_texture_index
)inst
->TexSrcTarget
;
2376 prog
->SamplersUsed
|= 1 << inst
->TexSrcUnit
;
2377 if (inst
->TexShadow
) {
2378 prog
->ShadowSamplers
|= 1 << inst
->TexSrcUnit
;
2383 _mesa_update_shader_textures_used(prog
);
2386 /* Each stage has some uniforms in its Parameters list. The Uniforms
2387 * list for the linked shader program has a pointer to these uniforms
2388 * in each of the stage's Parameters list, so that their values can be
2389 * updated when a uniform is set.
2392 link_uniforms_to_shared_uniform_list(struct gl_uniform_list
*uniforms
,
2393 struct gl_program
*prog
)
2397 for (i
= 0; i
< prog
->Parameters
->NumParameters
; i
++) {
2398 const struct gl_program_parameter
*p
= prog
->Parameters
->Parameters
+ i
;
2400 if (p
->Type
== PROGRAM_UNIFORM
|| p
->Type
== PROGRAM_SAMPLER
) {
2401 struct gl_uniform
*uniform
=
2402 _mesa_append_uniform(uniforms
, p
->Name
, prog
->Target
, i
);
2404 uniform
->Initialized
= p
->Initialized
;
2410 get_mesa_program(GLcontext
*ctx
, struct gl_shader_program
*shader_program
,
2411 struct gl_shader
*shader
)
2413 ir_to_mesa_visitor v
;
2414 struct prog_instruction
*mesa_instructions
, *mesa_inst
;
2415 ir_instruction
**mesa_instruction_annotation
;
2417 struct gl_program
*prog
;
2419 const char *target_string
;
2422 switch (shader
->Type
) {
2423 case GL_VERTEX_SHADER
:
2424 target
= GL_VERTEX_PROGRAM_ARB
;
2425 target_string
= "vertex";
2427 case GL_FRAGMENT_SHADER
:
2428 target
= GL_FRAGMENT_PROGRAM_ARB
;
2429 target_string
= "fragment";
2432 assert(!"should not be reached");
2436 validate_ir_tree(shader
->ir
);
2438 prog
= ctx
->Driver
.NewProgram(ctx
, target
, shader_program
->Name
);
2441 prog
->Parameters
= _mesa_new_parameter_list();
2442 prog
->Varying
= _mesa_new_parameter_list();
2443 prog
->Attributes
= _mesa_new_parameter_list();
2447 /* Emit Mesa IR for main(). */
2448 visit_exec_list(shader
->ir
, &v
);
2449 v
.ir_to_mesa_emit_op0(NULL
, OPCODE_END
);
2451 /* Now emit bodies for any functions that were used. */
2453 progress
= GL_FALSE
;
2455 foreach_iter(exec_list_iterator
, iter
, v
.function_signatures
) {
2456 function_entry
*entry
= (function_entry
*)iter
.get();
2458 if (!entry
->bgn_inst
) {
2459 v
.current_function
= entry
;
2461 entry
->bgn_inst
= v
.ir_to_mesa_emit_op0(NULL
, OPCODE_BGNSUB
);
2462 entry
->bgn_inst
->function
= entry
;
2464 visit_exec_list(&entry
->sig
->body
, &v
);
2466 ir_to_mesa_instruction
*last
;
2467 last
= (ir_to_mesa_instruction
*)v
.instructions
.get_tail();
2468 if (last
->op
!= OPCODE_RET
)
2469 v
.ir_to_mesa_emit_op0(NULL
, OPCODE_RET
);
2471 ir_to_mesa_instruction
*end
;
2472 end
= v
.ir_to_mesa_emit_op0(NULL
, OPCODE_ENDSUB
);
2473 end
->function
= entry
;
2480 prog
->NumTemporaries
= v
.next_temp
;
2482 int num_instructions
= 0;
2483 foreach_iter(exec_list_iterator
, iter
, v
.instructions
) {
2488 (struct prog_instruction
*)calloc(num_instructions
,
2489 sizeof(*mesa_instructions
));
2490 mesa_instruction_annotation
= talloc_array(v
.mem_ctx
, ir_instruction
*,
2493 mesa_inst
= mesa_instructions
;
2495 foreach_iter(exec_list_iterator
, iter
, v
.instructions
) {
2496 ir_to_mesa_instruction
*inst
= (ir_to_mesa_instruction
*)iter
.get();
2498 mesa_inst
->Opcode
= inst
->op
;
2499 mesa_inst
->CondUpdate
= inst
->cond_update
;
2500 mesa_inst
->DstReg
.File
= inst
->dst_reg
.file
;
2501 mesa_inst
->DstReg
.Index
= inst
->dst_reg
.index
;
2502 mesa_inst
->DstReg
.CondMask
= inst
->dst_reg
.cond_mask
;
2503 mesa_inst
->DstReg
.WriteMask
= inst
->dst_reg
.writemask
;
2504 mesa_inst
->DstReg
.RelAddr
= inst
->dst_reg
.reladdr
!= NULL
;
2505 mesa_inst
->SrcReg
[0] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[0]);
2506 mesa_inst
->SrcReg
[1] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[1]);
2507 mesa_inst
->SrcReg
[2] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[2]);
2508 mesa_inst
->TexSrcUnit
= inst
->sampler
;
2509 mesa_inst
->TexSrcTarget
= inst
->tex_target
;
2510 mesa_inst
->TexShadow
= inst
->tex_shadow
;
2511 mesa_instruction_annotation
[i
] = inst
->ir
;
2513 if (ctx
->Shader
.EmitNoIfs
&& mesa_inst
->Opcode
== OPCODE_IF
) {
2514 shader_program
->InfoLog
=
2515 talloc_asprintf_append(shader_program
->InfoLog
,
2516 "Couldn't flatten if statement\n");
2517 shader_program
->LinkStatus
= false;
2520 switch (mesa_inst
->Opcode
) {
2522 inst
->function
->inst
= i
;
2523 mesa_inst
->Comment
= strdup(inst
->function
->sig
->function_name());
2526 mesa_inst
->Comment
= strdup(inst
->function
->sig
->function_name());
2529 mesa_inst
->BranchTarget
= inst
->function
->sig_id
; /* rewritten later */
2532 prog
->NumAddressRegs
= 1;
2542 set_branchtargets(&v
, mesa_instructions
, num_instructions
);
2544 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2546 printf("GLSL IR for linked %s program %d:\n", target_string
,
2547 shader_program
->Name
);
2548 _mesa_print_ir(shader
->ir
, NULL
);
2551 printf("Mesa IR for linked %s program %d:\n", target_string
,
2552 shader_program
->Name
);
2553 print_program(mesa_instructions
, mesa_instruction_annotation
,
2557 prog
->Instructions
= mesa_instructions
;
2558 prog
->NumInstructions
= num_instructions
;
2560 do_set_program_inouts(shader
->ir
, prog
);
2561 count_resources(prog
);
2563 _mesa_reference_program(ctx
, &shader
->Program
, prog
);
2565 if ((ctx
->Shader
.Flags
& GLSL_NO_OPT
) == 0) {
2566 _mesa_optimize_program(ctx
, prog
);
2574 _mesa_ir_compile_shader(GLcontext
*ctx
, struct gl_shader
*shader
)
2576 assert(shader
->CompileStatus
);
2583 _mesa_ir_link_shader(GLcontext
*ctx
, struct gl_shader_program
*prog
)
2585 assert(prog
->LinkStatus
);
2587 for (unsigned i
= 0; i
< prog
->_NumLinkedShaders
; i
++) {
2589 exec_list
*ir
= prog
->_LinkedShaders
[i
]->ir
;
2595 do_mat_op_to_vec(ir
);
2596 do_mod_to_fract(ir
);
2597 do_div_to_mul_rcp(ir
);
2598 do_explog_to_explog2(ir
);
2600 progress
= do_common_optimization(ir
, true) || progress
;
2602 if (ctx
->Shader
.EmitNoIfs
)
2603 progress
= do_if_to_cond_assign(ir
) || progress
;
2605 progress
= do_vec_index_to_cond_assign(ir
) || progress
;
2608 validate_ir_tree(ir
);
2611 for (unsigned i
= 0; i
< prog
->_NumLinkedShaders
; i
++) {
2612 struct gl_program
*linked_prog
;
2615 linked_prog
= get_mesa_program(ctx
, prog
, prog
->_LinkedShaders
[i
]);
2617 link_uniforms_to_shared_uniform_list(prog
->Uniforms
, linked_prog
);
2619 switch (prog
->_LinkedShaders
[i
]->Type
) {
2620 case GL_VERTEX_SHADER
:
2621 _mesa_reference_vertprog(ctx
, &prog
->VertexProgram
,
2622 (struct gl_vertex_program
*)linked_prog
);
2623 ok
= ctx
->Driver
.ProgramStringNotify(ctx
, GL_VERTEX_PROGRAM_ARB
,
2626 case GL_FRAGMENT_SHADER
:
2627 _mesa_reference_fragprog(ctx
, &prog
->FragmentProgram
,
2628 (struct gl_fragment_program
*)linked_prog
);
2629 ok
= ctx
->Driver
.ProgramStringNotify(ctx
, GL_FRAGMENT_PROGRAM_ARB
,
2636 _mesa_reference_program(ctx
, &linked_prog
, NULL
);
2643 _mesa_glsl_compile_shader(GLcontext
*ctx
, struct gl_shader
*shader
)
2645 struct _mesa_glsl_parse_state
*state
=
2646 new(shader
) _mesa_glsl_parse_state(ctx
, shader
->Type
, shader
);
2648 const char *source
= shader
->Source
;
2649 state
->error
= preprocess(state
, &source
, &state
->info_log
,
2652 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2653 printf("GLSL source for shader %d:\n", shader
->Name
);
2654 printf("%s\n", shader
->Source
);
2657 if (!state
->error
) {
2658 _mesa_glsl_lexer_ctor(state
, source
);
2659 _mesa_glsl_parse(state
);
2660 _mesa_glsl_lexer_dtor(state
);
2663 talloc_free(shader
->ir
);
2664 shader
->ir
= new(shader
) exec_list
;
2665 if (!state
->error
&& !state
->translation_unit
.is_empty())
2666 _mesa_ast_to_hir(shader
->ir
, state
);
2668 if (!state
->error
&& !shader
->ir
->is_empty()) {
2669 validate_ir_tree(shader
->ir
);
2671 /* Do some optimization at compile time to reduce shader IR size
2672 * and reduce later work if the same shader is linked multiple times
2674 while (do_common_optimization(shader
->ir
, false))
2677 validate_ir_tree(shader
->ir
);
2680 shader
->symbols
= state
->symbols
;
2682 shader
->CompileStatus
= !state
->error
;
2683 shader
->InfoLog
= state
->info_log
;
2684 shader
->Version
= state
->language_version
;
2685 memcpy(shader
->builtins_to_link
, state
->builtins_to_link
,
2686 sizeof(shader
->builtins_to_link
[0]) * state
->num_builtins_to_link
);
2687 shader
->num_builtins_to_link
= state
->num_builtins_to_link
;
2689 if (ctx
->Shader
.Flags
& GLSL_LOG
) {
2690 _mesa_write_shader_to_file(shader
);
2693 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2694 if (shader
->CompileStatus
) {
2695 printf("GLSL IR for shader %d:\n", shader
->Name
);
2696 _mesa_print_ir(shader
->ir
, NULL
);
2699 printf("GLSL shader %d failed to compile.\n", shader
->Name
);
2701 if (shader
->InfoLog
&& shader
->InfoLog
[0] != 0) {
2702 printf("GLSL shader %d info log:\n", shader
->Name
);
2703 printf("%s\n", shader
->InfoLog
);
2707 /* Retain any live IR, but trash the rest. */
2708 reparent_ir(shader
->ir
, shader
->ir
);
2712 if (shader
->CompileStatus
) {
2713 if (!ctx
->Driver
.CompileShader(ctx
, shader
))
2714 shader
->CompileStatus
= GL_FALSE
;
2719 _mesa_glsl_link_shader(GLcontext
*ctx
, struct gl_shader_program
*prog
)
2723 _mesa_clear_shader_program_data(ctx
, prog
);
2725 prog
->LinkStatus
= GL_TRUE
;
2727 for (i
= 0; i
< prog
->NumShaders
; i
++) {
2728 if (!prog
->Shaders
[i
]->CompileStatus
) {
2730 talloc_asprintf_append(prog
->InfoLog
,
2731 "linking with uncompiled shader");
2732 prog
->LinkStatus
= GL_FALSE
;
2736 prog
->Varying
= _mesa_new_parameter_list();
2737 _mesa_reference_vertprog(ctx
, &prog
->VertexProgram
, NULL
);
2738 _mesa_reference_fragprog(ctx
, &prog
->FragmentProgram
, NULL
);
2740 if (prog
->LinkStatus
) {
2741 link_shaders(ctx
, prog
);
2743 /* We don't use the linker's uniforms list, and cook up our own at
2746 _mesa_free_uniform_list(prog
->Uniforms
);
2747 prog
->Uniforms
= _mesa_new_uniform_list();
2750 if (prog
->LinkStatus
) {
2751 if (!ctx
->Driver
.LinkShader(ctx
, prog
)) {
2752 prog
->LinkStatus
= GL_FALSE
;
2756 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2757 if (!prog
->LinkStatus
) {
2758 printf("GLSL shader program %d failed to link\n", prog
->Name
);
2761 if (prog
->InfoLog
&& prog
->InfoLog
[0] != 0) {
2762 printf("GLSL shader program %d info log:\n", prog
->Name
);
2763 printf("%s\n", prog
->InfoLog
);