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
;
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
;
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
,
277 int get_sampler_uniform_value(ir_dereference
*deref
);
282 ir_to_mesa_src_reg ir_to_mesa_undef
= ir_to_mesa_src_reg(PROGRAM_UNDEFINED
, 0, NULL
);
284 ir_to_mesa_dst_reg ir_to_mesa_undef_dst
= {
285 PROGRAM_UNDEFINED
, 0, SWIZZLE_NOOP
, COND_TR
, NULL
,
288 ir_to_mesa_dst_reg ir_to_mesa_address_reg
= {
289 PROGRAM_ADDRESS
, 0, WRITEMASK_X
, COND_TR
, NULL
292 static void fail_link(struct gl_shader_program
*prog
, const char *fmt
, ...) PRINTFLIKE(2, 3);
294 static void 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
;
304 static int swizzle_for_size(int size
)
306 int size_swizzles
[4] = {
307 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_X
),
308 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_Y
, SWIZZLE_Y
, SWIZZLE_Y
),
309 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_Y
, SWIZZLE_Z
, SWIZZLE_Z
),
310 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_Y
, SWIZZLE_Z
, SWIZZLE_W
),
313 return size_swizzles
[size
- 1];
316 ir_to_mesa_instruction
*
317 ir_to_mesa_visitor::ir_to_mesa_emit_op3(ir_instruction
*ir
,
319 ir_to_mesa_dst_reg dst
,
320 ir_to_mesa_src_reg src0
,
321 ir_to_mesa_src_reg src1
,
322 ir_to_mesa_src_reg src2
)
324 ir_to_mesa_instruction
*inst
= new(mem_ctx
) ir_to_mesa_instruction();
327 /* If we have to do relative addressing, we want to load the ARL
328 * reg directly for one of the regs, and preload the other reladdr
329 * sources into temps.
331 num_reladdr
+= dst
.reladdr
!= NULL
;
332 num_reladdr
+= src0
.reladdr
!= NULL
;
333 num_reladdr
+= src1
.reladdr
!= NULL
;
334 num_reladdr
+= src2
.reladdr
!= NULL
;
336 reladdr_to_temp(ir
, &src2
, &num_reladdr
);
337 reladdr_to_temp(ir
, &src1
, &num_reladdr
);
338 reladdr_to_temp(ir
, &src0
, &num_reladdr
);
341 ir_to_mesa_emit_op1(ir
, OPCODE_ARL
, ir_to_mesa_address_reg
,
346 assert(num_reladdr
== 0);
350 inst
->src_reg
[0] = src0
;
351 inst
->src_reg
[1] = src1
;
352 inst
->src_reg
[2] = src2
;
355 inst
->function
= NULL
;
357 this->instructions
.push_tail(inst
);
363 ir_to_mesa_instruction
*
364 ir_to_mesa_visitor::ir_to_mesa_emit_op2(ir_instruction
*ir
,
366 ir_to_mesa_dst_reg dst
,
367 ir_to_mesa_src_reg src0
,
368 ir_to_mesa_src_reg src1
)
370 return ir_to_mesa_emit_op3(ir
, op
, dst
, src0
, src1
, ir_to_mesa_undef
);
373 ir_to_mesa_instruction
*
374 ir_to_mesa_visitor::ir_to_mesa_emit_op1(ir_instruction
*ir
,
376 ir_to_mesa_dst_reg dst
,
377 ir_to_mesa_src_reg src0
)
379 assert(dst
.writemask
!= 0);
380 return ir_to_mesa_emit_op3(ir
, op
, dst
,
381 src0
, ir_to_mesa_undef
, ir_to_mesa_undef
);
384 ir_to_mesa_instruction
*
385 ir_to_mesa_visitor::ir_to_mesa_emit_op0(ir_instruction
*ir
,
388 return ir_to_mesa_emit_op3(ir
, op
, ir_to_mesa_undef_dst
,
394 inline ir_to_mesa_dst_reg
395 ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg reg
)
397 ir_to_mesa_dst_reg dst_reg
;
399 dst_reg
.file
= reg
.file
;
400 dst_reg
.index
= reg
.index
;
401 dst_reg
.writemask
= WRITEMASK_XYZW
;
402 dst_reg
.cond_mask
= COND_TR
;
403 dst_reg
.reladdr
= reg
.reladdr
;
408 inline ir_to_mesa_src_reg
409 ir_to_mesa_src_reg_from_dst(ir_to_mesa_dst_reg reg
)
411 return ir_to_mesa_src_reg(reg
.file
, reg
.index
, NULL
);
415 * Emits Mesa scalar opcodes to produce unique answers across channels.
417 * Some Mesa opcodes are scalar-only, like ARB_fp/vp. The src X
418 * channel determines the result across all channels. So to do a vec4
419 * of this operation, we want to emit a scalar per source channel used
420 * to produce dest channels.
423 ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op2(ir_instruction
*ir
,
425 ir_to_mesa_dst_reg dst
,
426 ir_to_mesa_src_reg orig_src0
,
427 ir_to_mesa_src_reg orig_src1
)
430 int done_mask
= ~dst
.writemask
;
432 /* Mesa RCP is a scalar operation splatting results to all channels,
433 * like ARB_fp/vp. So emit as many RCPs as necessary to cover our
436 for (i
= 0; i
< 4; i
++) {
437 GLuint this_mask
= (1 << i
);
438 ir_to_mesa_instruction
*inst
;
439 ir_to_mesa_src_reg src0
= orig_src0
;
440 ir_to_mesa_src_reg src1
= orig_src1
;
442 if (done_mask
& this_mask
)
445 GLuint src0_swiz
= GET_SWZ(src0
.swizzle
, i
);
446 GLuint src1_swiz
= GET_SWZ(src1
.swizzle
, i
);
447 for (j
= i
+ 1; j
< 4; j
++) {
448 if (!(done_mask
& (1 << j
)) &&
449 GET_SWZ(src0
.swizzle
, j
) == src0_swiz
&&
450 GET_SWZ(src1
.swizzle
, j
) == src1_swiz
) {
451 this_mask
|= (1 << j
);
454 src0
.swizzle
= MAKE_SWIZZLE4(src0_swiz
, src0_swiz
,
455 src0_swiz
, src0_swiz
);
456 src1
.swizzle
= MAKE_SWIZZLE4(src1_swiz
, src1_swiz
,
457 src1_swiz
, src1_swiz
);
459 inst
= ir_to_mesa_emit_op2(ir
, op
,
463 inst
->dst_reg
.writemask
= this_mask
;
464 done_mask
|= this_mask
;
469 ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op1(ir_instruction
*ir
,
471 ir_to_mesa_dst_reg dst
,
472 ir_to_mesa_src_reg src0
)
474 ir_to_mesa_src_reg undef
= ir_to_mesa_undef
;
476 undef
.swizzle
= SWIZZLE_XXXX
;
478 ir_to_mesa_emit_scalar_op2(ir
, op
, dst
, src0
, undef
);
481 struct ir_to_mesa_src_reg
482 ir_to_mesa_visitor::src_reg_for_float(float val
)
484 ir_to_mesa_src_reg
src_reg(PROGRAM_CONSTANT
, -1, NULL
);
486 src_reg
.index
= _mesa_add_unnamed_constant(this->prog
->Parameters
,
487 &val
, 1, &src_reg
.swizzle
);
493 type_size(const struct glsl_type
*type
)
498 switch (type
->base_type
) {
501 case GLSL_TYPE_FLOAT
:
503 if (type
->is_matrix()) {
504 return type
->matrix_columns
;
506 /* Regardless of size of vector, it gets a vec4. This is bad
507 * packing for things like floats, but otherwise arrays become a
508 * mess. Hopefully a later pass over the code can pack scalars
509 * down if appropriate.
513 case GLSL_TYPE_ARRAY
:
514 return type_size(type
->fields
.array
) * type
->length
;
515 case GLSL_TYPE_STRUCT
:
517 for (i
= 0; i
< type
->length
; i
++) {
518 size
+= type_size(type
->fields
.structure
[i
].type
);
521 case GLSL_TYPE_SAMPLER
:
522 /* Samplers take up one slot in UNIFORMS[], but they're baked in
533 * In the initial pass of codegen, we assign temporary numbers to
534 * intermediate results. (not SSA -- variable assignments will reuse
535 * storage). Actual register allocation for the Mesa VM occurs in a
536 * pass over the Mesa IR later.
539 ir_to_mesa_visitor::get_temp(const glsl_type
*type
)
541 ir_to_mesa_src_reg src_reg
;
545 src_reg
.file
= PROGRAM_TEMPORARY
;
546 src_reg
.index
= next_temp
;
547 src_reg
.reladdr
= NULL
;
548 next_temp
+= type_size(type
);
550 if (type
->is_array() || type
->is_record()) {
551 src_reg
.swizzle
= SWIZZLE_NOOP
;
553 for (i
= 0; i
< type
->vector_elements
; i
++)
556 swizzle
[i
] = type
->vector_elements
- 1;
557 src_reg
.swizzle
= MAKE_SWIZZLE4(swizzle
[0], swizzle
[1],
558 swizzle
[2], swizzle
[3]);
566 ir_to_mesa_visitor::find_variable_storage(ir_variable
*var
)
569 variable_storage
*entry
;
571 foreach_iter(exec_list_iterator
, iter
, this->variables
) {
572 entry
= (variable_storage
*)iter
.get();
574 if (entry
->var
== var
)
581 struct statevar_element
{
583 int tokens
[STATE_LENGTH
];
588 static struct statevar_element gl_DepthRange_elements
[] = {
589 {"near", {STATE_DEPTH_RANGE
, 0, 0}, SWIZZLE_XXXX
},
590 {"far", {STATE_DEPTH_RANGE
, 0, 0}, SWIZZLE_YYYY
},
591 {"diff", {STATE_DEPTH_RANGE
, 0, 0}, SWIZZLE_ZZZZ
},
594 static struct statevar_element gl_ClipPlane_elements
[] = {
595 {NULL
, {STATE_CLIPPLANE
, 0, 0}, SWIZZLE_XYZW
}
598 static struct statevar_element gl_Point_elements
[] = {
599 {"size", {STATE_POINT_SIZE
}, SWIZZLE_XXXX
},
600 {"sizeMin", {STATE_POINT_SIZE
}, SWIZZLE_YYYY
},
601 {"sizeMax", {STATE_POINT_SIZE
}, SWIZZLE_ZZZZ
},
602 {"fadeThresholdSize", {STATE_POINT_SIZE
}, SWIZZLE_WWWW
},
603 {"distanceConstantAttenuation", {STATE_POINT_ATTENUATION
}, SWIZZLE_XXXX
},
604 {"distanceLinearAttenuation", {STATE_POINT_ATTENUATION
}, SWIZZLE_YYYY
},
605 {"distanceQuadraticAttenuation", {STATE_POINT_ATTENUATION
}, SWIZZLE_ZZZZ
},
608 static struct statevar_element gl_FrontMaterial_elements
[] = {
609 {"emission", {STATE_MATERIAL
, 0, STATE_EMISSION
}, SWIZZLE_XYZW
},
610 {"ambient", {STATE_MATERIAL
, 0, STATE_AMBIENT
}, SWIZZLE_XYZW
},
611 {"diffuse", {STATE_MATERIAL
, 0, STATE_DIFFUSE
}, SWIZZLE_XYZW
},
612 {"specular", {STATE_MATERIAL
, 0, STATE_SPECULAR
}, SWIZZLE_XYZW
},
613 {"shininess", {STATE_MATERIAL
, 0, STATE_SHININESS
}, SWIZZLE_XXXX
},
616 static struct statevar_element gl_BackMaterial_elements
[] = {
617 {"emission", {STATE_MATERIAL
, 1, STATE_EMISSION
}, SWIZZLE_XYZW
},
618 {"ambient", {STATE_MATERIAL
, 1, STATE_AMBIENT
}, SWIZZLE_XYZW
},
619 {"diffuse", {STATE_MATERIAL
, 1, STATE_DIFFUSE
}, SWIZZLE_XYZW
},
620 {"specular", {STATE_MATERIAL
, 1, STATE_SPECULAR
}, SWIZZLE_XYZW
},
621 {"shininess", {STATE_MATERIAL
, 1, STATE_SHININESS
}, SWIZZLE_XXXX
},
624 static struct statevar_element gl_LightSource_elements
[] = {
625 {"ambient", {STATE_LIGHT
, 0, STATE_AMBIENT
}, SWIZZLE_XYZW
},
626 {"diffuse", {STATE_LIGHT
, 0, STATE_DIFFUSE
}, SWIZZLE_XYZW
},
627 {"specular", {STATE_LIGHT
, 0, STATE_SPECULAR
}, SWIZZLE_XYZW
},
628 {"position", {STATE_LIGHT
, 0, STATE_POSITION
}, SWIZZLE_XYZW
},
629 {"halfVector", {STATE_LIGHT
, 0, STATE_HALF_VECTOR
}, SWIZZLE_XYZW
},
630 {"spotDirection", {STATE_LIGHT
, 0, STATE_SPOT_DIRECTION
}, SWIZZLE_XYZW
},
631 {"spotCosCutoff", {STATE_LIGHT
, 0, STATE_SPOT_DIRECTION
}, SWIZZLE_WWWW
},
632 {"spotCutoff", {STATE_LIGHT
, 0, STATE_SPOT_CUTOFF
}, SWIZZLE_XXXX
},
633 {"spotExponent", {STATE_LIGHT
, 0, STATE_ATTENUATION
}, SWIZZLE_WWWW
},
634 {"constantAttenuation", {STATE_LIGHT
, 0, STATE_ATTENUATION
}, SWIZZLE_XXXX
},
635 {"linearAttenuation", {STATE_LIGHT
, 0, STATE_ATTENUATION
}, SWIZZLE_YYYY
},
636 {"quadraticAttenuation", {STATE_LIGHT
, 0, STATE_ATTENUATION
}, SWIZZLE_ZZZZ
},
639 static struct statevar_element gl_LightModel_elements
[] = {
640 {"ambient", {STATE_LIGHTMODEL_AMBIENT
, 0}, SWIZZLE_XYZW
},
643 static struct statevar_element gl_FrontLightModelProduct_elements
[] = {
644 {"sceneColor", {STATE_LIGHTMODEL_SCENECOLOR
, 0}, SWIZZLE_XYZW
},
647 static struct statevar_element gl_BackLightModelProduct_elements
[] = {
648 {"sceneColor", {STATE_LIGHTMODEL_SCENECOLOR
, 1}, SWIZZLE_XYZW
},
651 static struct statevar_element gl_FrontLightProduct_elements
[] = {
652 {"ambient", {STATE_LIGHTPROD
, 0, 0, STATE_AMBIENT
}, SWIZZLE_XYZW
},
653 {"diffuse", {STATE_LIGHTPROD
, 0, 0, STATE_DIFFUSE
}, SWIZZLE_XYZW
},
654 {"specular", {STATE_LIGHTPROD
, 0, 0, STATE_SPECULAR
}, SWIZZLE_XYZW
},
657 static struct statevar_element gl_BackLightProduct_elements
[] = {
658 {"ambient", {STATE_LIGHTPROD
, 0, 1, STATE_AMBIENT
}, SWIZZLE_XYZW
},
659 {"diffuse", {STATE_LIGHTPROD
, 0, 1, STATE_DIFFUSE
}, SWIZZLE_XYZW
},
660 {"specular", {STATE_LIGHTPROD
, 0, 1, STATE_SPECULAR
}, SWIZZLE_XYZW
},
663 static struct statevar_element gl_TextureEnvColor_elements
[] = {
664 {NULL
, {STATE_TEXENV_COLOR
, 0}, SWIZZLE_XYZW
},
667 static struct statevar_element gl_EyePlaneS_elements
[] = {
668 {NULL
, {STATE_TEXGEN
, 0, STATE_TEXGEN_EYE_S
}, SWIZZLE_XYZW
},
671 static struct statevar_element gl_EyePlaneT_elements
[] = {
672 {NULL
, {STATE_TEXGEN
, 0, STATE_TEXGEN_EYE_T
}, SWIZZLE_XYZW
},
675 static struct statevar_element gl_EyePlaneR_elements
[] = {
676 {NULL
, {STATE_TEXGEN
, 0, STATE_TEXGEN_EYE_R
}, SWIZZLE_XYZW
},
679 static struct statevar_element gl_EyePlaneQ_elements
[] = {
680 {NULL
, {STATE_TEXGEN
, 0, STATE_TEXGEN_EYE_Q
}, SWIZZLE_XYZW
},
683 static struct statevar_element gl_ObjectPlaneS_elements
[] = {
684 {NULL
, {STATE_TEXGEN
, 0, STATE_TEXGEN_OBJECT_S
}, SWIZZLE_XYZW
},
687 static struct statevar_element gl_ObjectPlaneT_elements
[] = {
688 {NULL
, {STATE_TEXGEN
, 0, STATE_TEXGEN_OBJECT_T
}, SWIZZLE_XYZW
},
691 static struct statevar_element gl_ObjectPlaneR_elements
[] = {
692 {NULL
, {STATE_TEXGEN
, 0, STATE_TEXGEN_OBJECT_R
}, SWIZZLE_XYZW
},
695 static struct statevar_element gl_ObjectPlaneQ_elements
[] = {
696 {NULL
, {STATE_TEXGEN
, 0, STATE_TEXGEN_OBJECT_Q
}, SWIZZLE_XYZW
},
699 static struct statevar_element gl_Fog_elements
[] = {
700 {"color", {STATE_FOG_COLOR
}, SWIZZLE_XYZW
},
701 {"density", {STATE_FOG_PARAMS
}, SWIZZLE_XXXX
},
702 {"start", {STATE_FOG_PARAMS
}, SWIZZLE_YYYY
},
703 {"end", {STATE_FOG_PARAMS
}, SWIZZLE_ZZZZ
},
704 {"scale", {STATE_FOG_PARAMS
}, SWIZZLE_WWWW
},
707 static struct statevar_element gl_NormalScale_elements
[] = {
708 {NULL
, {STATE_NORMAL_SCALE
}, SWIZZLE_XXXX
},
711 #define MATRIX(name, statevar, modifier) \
712 static struct statevar_element name ## _elements[] = { \
713 { NULL, { statevar, 0, 0, 0, modifier}, SWIZZLE_XYZW }, \
714 { NULL, { statevar, 0, 1, 1, modifier}, SWIZZLE_XYZW }, \
715 { NULL, { statevar, 0, 2, 2, modifier}, SWIZZLE_XYZW }, \
716 { NULL, { statevar, 0, 3, 3, modifier}, SWIZZLE_XYZW }, \
719 MATRIX(gl_ModelViewMatrix
,
720 STATE_MODELVIEW_MATRIX
, STATE_MATRIX_TRANSPOSE
);
721 MATRIX(gl_ModelViewMatrixInverse
,
722 STATE_MODELVIEW_MATRIX
, STATE_MATRIX_INVTRANS
);
723 MATRIX(gl_ModelViewMatrixTranspose
,
724 STATE_MODELVIEW_MATRIX
, 0);
725 MATRIX(gl_ModelViewMatrixInverseTranspose
,
726 STATE_MODELVIEW_MATRIX
, STATE_MATRIX_INVERSE
);
728 MATRIX(gl_ProjectionMatrix
,
729 STATE_PROJECTION_MATRIX
, STATE_MATRIX_TRANSPOSE
);
730 MATRIX(gl_ProjectionMatrixInverse
,
731 STATE_PROJECTION_MATRIX
, STATE_MATRIX_INVTRANS
);
732 MATRIX(gl_ProjectionMatrixTranspose
,
733 STATE_PROJECTION_MATRIX
, 0);
734 MATRIX(gl_ProjectionMatrixInverseTranspose
,
735 STATE_PROJECTION_MATRIX
, STATE_MATRIX_INVERSE
);
737 MATRIX(gl_ModelViewProjectionMatrix
,
738 STATE_MVP_MATRIX
, STATE_MATRIX_TRANSPOSE
);
739 MATRIX(gl_ModelViewProjectionMatrixInverse
,
740 STATE_MVP_MATRIX
, STATE_MATRIX_INVTRANS
);
741 MATRIX(gl_ModelViewProjectionMatrixTranspose
,
742 STATE_MVP_MATRIX
, 0);
743 MATRIX(gl_ModelViewProjectionMatrixInverseTranspose
,
744 STATE_MVP_MATRIX
, STATE_MATRIX_INVERSE
);
746 MATRIX(gl_TextureMatrix
,
747 STATE_TEXTURE_MATRIX
, STATE_MATRIX_TRANSPOSE
);
748 MATRIX(gl_TextureMatrixInverse
,
749 STATE_TEXTURE_MATRIX
, STATE_MATRIX_INVTRANS
);
750 MATRIX(gl_TextureMatrixTranspose
,
751 STATE_TEXTURE_MATRIX
, 0);
752 MATRIX(gl_TextureMatrixInverseTranspose
,
753 STATE_TEXTURE_MATRIX
, STATE_MATRIX_INVERSE
);
755 static struct statevar_element gl_NormalMatrix_elements
[] = {
756 { NULL
, { STATE_MODELVIEW_MATRIX
, 0, 0, 0, STATE_MATRIX_INVERSE
},
758 { NULL
, { STATE_MODELVIEW_MATRIX
, 0, 1, 1, STATE_MATRIX_INVERSE
},
760 { NULL
, { STATE_MODELVIEW_MATRIX
, 0, 2, 2, STATE_MATRIX_INVERSE
},
766 #define STATEVAR(name) {#name, name ## _elements, Elements(name ## _elements)}
768 static const struct statevar
{
770 struct statevar_element
*elements
;
771 unsigned int num_elements
;
773 STATEVAR(gl_DepthRange
),
774 STATEVAR(gl_ClipPlane
),
776 STATEVAR(gl_FrontMaterial
),
777 STATEVAR(gl_BackMaterial
),
778 STATEVAR(gl_LightSource
),
779 STATEVAR(gl_LightModel
),
780 STATEVAR(gl_FrontLightModelProduct
),
781 STATEVAR(gl_BackLightModelProduct
),
782 STATEVAR(gl_FrontLightProduct
),
783 STATEVAR(gl_BackLightProduct
),
784 STATEVAR(gl_TextureEnvColor
),
785 STATEVAR(gl_EyePlaneS
),
786 STATEVAR(gl_EyePlaneT
),
787 STATEVAR(gl_EyePlaneR
),
788 STATEVAR(gl_EyePlaneQ
),
789 STATEVAR(gl_ObjectPlaneS
),
790 STATEVAR(gl_ObjectPlaneT
),
791 STATEVAR(gl_ObjectPlaneR
),
792 STATEVAR(gl_ObjectPlaneQ
),
795 STATEVAR(gl_ModelViewMatrix
),
796 STATEVAR(gl_ModelViewMatrixInverse
),
797 STATEVAR(gl_ModelViewMatrixTranspose
),
798 STATEVAR(gl_ModelViewMatrixInverseTranspose
),
800 STATEVAR(gl_ProjectionMatrix
),
801 STATEVAR(gl_ProjectionMatrixInverse
),
802 STATEVAR(gl_ProjectionMatrixTranspose
),
803 STATEVAR(gl_ProjectionMatrixInverseTranspose
),
805 STATEVAR(gl_ModelViewProjectionMatrix
),
806 STATEVAR(gl_ModelViewProjectionMatrixInverse
),
807 STATEVAR(gl_ModelViewProjectionMatrixTranspose
),
808 STATEVAR(gl_ModelViewProjectionMatrixInverseTranspose
),
810 STATEVAR(gl_TextureMatrix
),
811 STATEVAR(gl_TextureMatrixInverse
),
812 STATEVAR(gl_TextureMatrixTranspose
),
813 STATEVAR(gl_TextureMatrixInverseTranspose
),
815 STATEVAR(gl_NormalMatrix
),
816 STATEVAR(gl_NormalScale
),
820 ir_to_mesa_visitor::visit(ir_variable
*ir
)
822 if (strcmp(ir
->name
, "gl_FragCoord") == 0) {
823 struct gl_fragment_program
*fp
= (struct gl_fragment_program
*)this->prog
;
825 fp
->OriginUpperLeft
= ir
->origin_upper_left
;
826 fp
->PixelCenterInteger
= ir
->pixel_center_integer
;
829 if (ir
->mode
== ir_var_uniform
&& strncmp(ir
->name
, "gl_", 3) == 0) {
832 for (i
= 0; i
< Elements(statevars
); i
++) {
833 if (strcmp(ir
->name
, statevars
[i
].name
) == 0)
837 if (i
== Elements(statevars
)) {
838 fail_link(this->shader_program
,
839 "Failed to find builtin uniform `%s'\n", ir
->name
);
843 const struct statevar
*statevar
= &statevars
[i
];
846 if (ir
->type
->is_array()) {
847 array_count
= ir
->type
->length
;
852 /* Check if this statevar's setup in the STATE file exactly
853 * matches how we'll want to reference it as a
854 * struct/array/whatever. If not, then we need to move it into
855 * temporary storage and hope that it'll get copy-propagated
858 for (i
= 0; i
< statevar
->num_elements
; i
++) {
859 if (statevar
->elements
[i
].swizzle
!= SWIZZLE_XYZW
) {
864 struct variable_storage
*storage
;
865 ir_to_mesa_dst_reg dst
;
866 if (i
== statevar
->num_elements
) {
867 /* We'll set the index later. */
868 storage
= new(mem_ctx
) variable_storage(ir
, PROGRAM_STATE_VAR
, -1);
869 this->variables
.push_tail(storage
);
871 dst
= ir_to_mesa_undef_dst
;
873 storage
= new(mem_ctx
) variable_storage(ir
, PROGRAM_TEMPORARY
,
875 this->variables
.push_tail(storage
);
876 this->next_temp
+= type_size(ir
->type
);
878 dst
= ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg(PROGRAM_TEMPORARY
,
884 for (int a
= 0; a
< array_count
; a
++) {
885 for (unsigned int i
= 0; i
< statevar
->num_elements
; i
++) {
886 struct statevar_element
*element
= &statevar
->elements
[i
];
887 int tokens
[STATE_LENGTH
];
889 memcpy(tokens
, element
->tokens
, sizeof(element
->tokens
));
890 if (ir
->type
->is_array()) {
894 int index
= _mesa_add_state_reference(this->prog
->Parameters
,
895 (gl_state_index
*)tokens
);
897 if (storage
->file
== PROGRAM_STATE_VAR
) {
898 if (storage
->index
== -1) {
899 storage
->index
= index
;
902 (int)(storage
->index
+ a
* statevar
->num_elements
+ i
));
905 ir_to_mesa_src_reg
src(PROGRAM_STATE_VAR
, index
, NULL
);
906 src
.swizzle
= element
->swizzle
;
907 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, dst
, src
);
908 /* even a float takes up a whole vec4 reg in a struct/array. */
913 if (storage
->file
== PROGRAM_TEMPORARY
&&
914 dst
.index
!= storage
->index
+ type_size(ir
->type
)) {
915 fail_link(this->shader_program
,
916 "failed to load builtin uniform `%s' (%d/%d regs loaded)\n",
917 ir
->name
, dst
.index
- storage
->index
,
918 type_size(ir
->type
));
924 ir_to_mesa_visitor::visit(ir_loop
*ir
)
926 ir_dereference_variable
*counter
= NULL
;
928 if (ir
->counter
!= NULL
)
929 counter
= new(ir
) ir_dereference_variable(ir
->counter
);
931 if (ir
->from
!= NULL
) {
932 assert(ir
->counter
!= NULL
);
934 ir_assignment
*a
= new(ir
) ir_assignment(counter
, ir
->from
, NULL
);
940 ir_to_mesa_emit_op0(NULL
, OPCODE_BGNLOOP
);
944 new(ir
) ir_expression(ir
->cmp
, glsl_type::bool_type
,
946 ir_if
*if_stmt
= new(ir
) ir_if(e
);
948 ir_loop_jump
*brk
= new(ir
) ir_loop_jump(ir_loop_jump::jump_break
);
950 if_stmt
->then_instructions
.push_tail(brk
);
952 if_stmt
->accept(this);
959 visit_exec_list(&ir
->body_instructions
, this);
963 new(ir
) ir_expression(ir_binop_add
, counter
->type
,
964 counter
, ir
->increment
);
966 ir_assignment
*a
= new(ir
) ir_assignment(counter
, e
, NULL
);
973 ir_to_mesa_emit_op0(NULL
, OPCODE_ENDLOOP
);
977 ir_to_mesa_visitor::visit(ir_loop_jump
*ir
)
980 case ir_loop_jump::jump_break
:
981 ir_to_mesa_emit_op0(NULL
, OPCODE_BRK
);
983 case ir_loop_jump::jump_continue
:
984 ir_to_mesa_emit_op0(NULL
, OPCODE_CONT
);
991 ir_to_mesa_visitor::visit(ir_function_signature
*ir
)
998 ir_to_mesa_visitor::visit(ir_function
*ir
)
1000 /* Ignore function bodies other than main() -- we shouldn't see calls to
1001 * them since they should all be inlined before we get to ir_to_mesa.
1003 if (strcmp(ir
->name
, "main") == 0) {
1004 const ir_function_signature
*sig
;
1007 sig
= ir
->matching_signature(&empty
);
1011 foreach_iter(exec_list_iterator
, iter
, sig
->body
) {
1012 ir_instruction
*ir
= (ir_instruction
*)iter
.get();
1020 ir_to_mesa_visitor::try_emit_mad(ir_expression
*ir
, int mul_operand
)
1022 int nonmul_operand
= 1 - mul_operand
;
1023 ir_to_mesa_src_reg a
, b
, c
;
1025 ir_expression
*expr
= ir
->operands
[mul_operand
]->as_expression();
1026 if (!expr
|| expr
->operation
!= ir_binop_mul
)
1029 expr
->operands
[0]->accept(this);
1031 expr
->operands
[1]->accept(this);
1033 ir
->operands
[nonmul_operand
]->accept(this);
1036 this->result
= get_temp(ir
->type
);
1037 ir_to_mesa_emit_op3(ir
, OPCODE_MAD
,
1038 ir_to_mesa_dst_reg_from_src(this->result
), a
, b
, c
);
1044 ir_to_mesa_visitor::reladdr_to_temp(ir_instruction
*ir
,
1045 ir_to_mesa_src_reg
*reg
, int *num_reladdr
)
1050 ir_to_mesa_emit_op1(ir
, OPCODE_ARL
, ir_to_mesa_address_reg
, *reg
->reladdr
);
1052 if (*num_reladdr
!= 1) {
1053 ir_to_mesa_src_reg temp
= get_temp(glsl_type::vec4_type
);
1055 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
,
1056 ir_to_mesa_dst_reg_from_src(temp
), *reg
);
1064 ir_to_mesa_visitor::visit(ir_expression
*ir
)
1066 unsigned int operand
;
1067 struct ir_to_mesa_src_reg op
[2];
1068 struct ir_to_mesa_src_reg result_src
;
1069 struct ir_to_mesa_dst_reg result_dst
;
1070 const glsl_type
*vec4_type
= glsl_type::get_instance(GLSL_TYPE_FLOAT
, 4, 1);
1071 const glsl_type
*vec3_type
= glsl_type::get_instance(GLSL_TYPE_FLOAT
, 3, 1);
1072 const glsl_type
*vec2_type
= glsl_type::get_instance(GLSL_TYPE_FLOAT
, 2, 1);
1074 /* Quick peephole: Emit OPCODE_MAD(a, b, c) instead of ADD(MUL(a, b), c)
1076 if (ir
->operation
== ir_binop_add
) {
1077 if (try_emit_mad(ir
, 1))
1079 if (try_emit_mad(ir
, 0))
1083 for (operand
= 0; operand
< ir
->get_num_operands(); operand
++) {
1084 this->result
.file
= PROGRAM_UNDEFINED
;
1085 ir
->operands
[operand
]->accept(this);
1086 if (this->result
.file
== PROGRAM_UNDEFINED
) {
1088 printf("Failed to get tree for expression operand:\n");
1089 ir
->operands
[operand
]->accept(&v
);
1092 op
[operand
] = this->result
;
1094 /* Matrix expression operands should have been broken down to vector
1095 * operations already.
1097 assert(!ir
->operands
[operand
]->type
->is_matrix());
1100 this->result
.file
= PROGRAM_UNDEFINED
;
1102 /* Storage for our result. Ideally for an assignment we'd be using
1103 * the actual storage for the result here, instead.
1105 result_src
= get_temp(ir
->type
);
1106 /* convenience for the emit functions below. */
1107 result_dst
= ir_to_mesa_dst_reg_from_src(result_src
);
1108 /* Limit writes to the channels that will be used by result_src later.
1109 * This does limit this temp's use as a temporary for multi-instruction
1112 result_dst
.writemask
= (1 << ir
->type
->vector_elements
) - 1;
1114 switch (ir
->operation
) {
1115 case ir_unop_logic_not
:
1116 ir_to_mesa_emit_op2(ir
, OPCODE_SEQ
, result_dst
,
1117 op
[0], src_reg_for_float(0.0));
1120 op
[0].negate
= ~op
[0].negate
;
1124 ir_to_mesa_emit_op1(ir
, OPCODE_ABS
, result_dst
, op
[0]);
1127 ir_to_mesa_emit_op1(ir
, OPCODE_SSG
, result_dst
, op
[0]);
1130 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_RCP
, result_dst
, op
[0]);
1134 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_EX2
, result_dst
, op
[0]);
1138 assert(!"not reached: should be handled by ir_explog_to_explog2");
1141 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_LG2
, result_dst
, op
[0]);
1144 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_SIN
, result_dst
, op
[0]);
1147 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_COS
, result_dst
, op
[0]);
1151 ir_to_mesa_emit_op1(ir
, OPCODE_DDX
, result_dst
, op
[0]);
1154 ir_to_mesa_emit_op1(ir
, OPCODE_DDY
, result_dst
, op
[0]);
1157 case ir_unop_noise
: {
1158 const enum prog_opcode opcode
=
1159 prog_opcode(OPCODE_NOISE1
1160 + (ir
->operands
[0]->type
->vector_elements
) - 1);
1161 assert((opcode
>= OPCODE_NOISE1
) && (opcode
<= OPCODE_NOISE4
));
1163 ir_to_mesa_emit_op1(ir
, opcode
, result_dst
, op
[0]);
1168 ir_to_mesa_emit_op2(ir
, OPCODE_ADD
, result_dst
, op
[0], op
[1]);
1171 ir_to_mesa_emit_op2(ir
, OPCODE_SUB
, result_dst
, op
[0], op
[1]);
1175 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
, result_dst
, op
[0], op
[1]);
1178 assert(!"not reached: should be handled by ir_div_to_mul_rcp");
1180 assert(!"ir_binop_mod should have been converted to b * fract(a/b)");
1184 ir_to_mesa_emit_op2(ir
, OPCODE_SLT
, result_dst
, op
[0], op
[1]);
1186 case ir_binop_greater
:
1187 ir_to_mesa_emit_op2(ir
, OPCODE_SGT
, result_dst
, op
[0], op
[1]);
1189 case ir_binop_lequal
:
1190 ir_to_mesa_emit_op2(ir
, OPCODE_SLE
, result_dst
, op
[0], op
[1]);
1192 case ir_binop_gequal
:
1193 ir_to_mesa_emit_op2(ir
, OPCODE_SGE
, result_dst
, op
[0], op
[1]);
1195 case ir_binop_equal
:
1196 ir_to_mesa_emit_op2(ir
, OPCODE_SEQ
, result_dst
, op
[0], op
[1]);
1198 case ir_binop_nequal
:
1199 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
, result_dst
, op
[0], op
[1]);
1201 case ir_binop_all_equal
:
1202 /* "==" operator producing a scalar boolean. */
1203 if (ir
->operands
[0]->type
->is_vector() ||
1204 ir
->operands
[1]->type
->is_vector()) {
1205 ir_to_mesa_src_reg temp
= get_temp(glsl_type::vec4_type
);
1206 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
1207 ir_to_mesa_dst_reg_from_src(temp
), op
[0], op
[1]);
1208 ir_to_mesa_emit_op2(ir
, OPCODE_DP4
, result_dst
, temp
, temp
);
1209 ir_to_mesa_emit_op2(ir
, OPCODE_SEQ
,
1210 result_dst
, result_src
, src_reg_for_float(0.0));
1212 ir_to_mesa_emit_op2(ir
, OPCODE_SEQ
, result_dst
, op
[0], op
[1]);
1215 case ir_binop_any_nequal
:
1216 /* "!=" operator producing a scalar boolean. */
1217 if (ir
->operands
[0]->type
->is_vector() ||
1218 ir
->operands
[1]->type
->is_vector()) {
1219 ir_to_mesa_src_reg temp
= get_temp(glsl_type::vec4_type
);
1220 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
1221 ir_to_mesa_dst_reg_from_src(temp
), op
[0], op
[1]);
1222 ir_to_mesa_emit_op2(ir
, OPCODE_DP4
, result_dst
, temp
, temp
);
1223 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
1224 result_dst
, result_src
, src_reg_for_float(0.0));
1226 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
, result_dst
, op
[0], op
[1]);
1231 switch (ir
->operands
[0]->type
->vector_elements
) {
1233 ir_to_mesa_emit_op2(ir
, OPCODE_DP4
, result_dst
, op
[0], op
[0]);
1236 ir_to_mesa_emit_op2(ir
, OPCODE_DP3
, result_dst
, op
[0], op
[0]);
1239 ir_to_mesa_emit_op2(ir
, OPCODE_DP2
, result_dst
, op
[0], op
[0]);
1242 assert(!"unreached: ir_unop_any of non-bvec");
1245 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
1246 result_dst
, result_src
, src_reg_for_float(0.0));
1249 case ir_binop_logic_xor
:
1250 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
, result_dst
, op
[0], op
[1]);
1253 case ir_binop_logic_or
:
1254 /* This could be a saturated add and skip the SNE. */
1255 ir_to_mesa_emit_op2(ir
, OPCODE_ADD
,
1259 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
,
1261 result_src
, src_reg_for_float(0.0));
1264 case ir_binop_logic_and
:
1265 /* the bool args are stored as float 0.0 or 1.0, so "mul" gives us "and". */
1266 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
,
1272 if (ir
->operands
[0]->type
== vec4_type
) {
1273 assert(ir
->operands
[1]->type
== vec4_type
);
1274 ir_to_mesa_emit_op2(ir
, OPCODE_DP4
,
1277 } else if (ir
->operands
[0]->type
== vec3_type
) {
1278 assert(ir
->operands
[1]->type
== vec3_type
);
1279 ir_to_mesa_emit_op2(ir
, OPCODE_DP3
,
1282 } else if (ir
->operands
[0]->type
== vec2_type
) {
1283 assert(ir
->operands
[1]->type
== vec2_type
);
1284 ir_to_mesa_emit_op2(ir
, OPCODE_DP2
,
1290 case ir_binop_cross
:
1291 ir_to_mesa_emit_op2(ir
, OPCODE_XPD
, result_dst
, op
[0], op
[1]);
1295 /* sqrt(x) = x * rsq(x). */
1296 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_RSQ
, result_dst
, op
[0]);
1297 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
, result_dst
, result_src
, op
[0]);
1298 /* For incoming channels <= 0, set the result to 0. */
1299 op
[0].negate
= ~op
[0].negate
;
1300 ir_to_mesa_emit_op3(ir
, OPCODE_CMP
, result_dst
,
1301 op
[0], result_src
, src_reg_for_float(0.0));
1304 ir_to_mesa_emit_scalar_op1(ir
, OPCODE_RSQ
, result_dst
, op
[0]);
1309 /* Mesa IR lacks types, ints are stored as truncated floats. */
1313 ir_to_mesa_emit_op1(ir
, OPCODE_TRUNC
, result_dst
, op
[0]);
1317 ir_to_mesa_emit_op2(ir
, OPCODE_SNE
, result_dst
,
1318 op
[0], src_reg_for_float(0.0));
1321 ir_to_mesa_emit_op1(ir
, OPCODE_TRUNC
, result_dst
, op
[0]);
1324 op
[0].negate
= ~op
[0].negate
;
1325 ir_to_mesa_emit_op1(ir
, OPCODE_FLR
, result_dst
, op
[0]);
1326 result_src
.negate
= ~result_src
.negate
;
1329 ir_to_mesa_emit_op1(ir
, OPCODE_FLR
, result_dst
, op
[0]);
1332 ir_to_mesa_emit_op1(ir
, OPCODE_FRC
, result_dst
, op
[0]);
1336 ir_to_mesa_emit_op2(ir
, OPCODE_MIN
, result_dst
, op
[0], op
[1]);
1339 ir_to_mesa_emit_op2(ir
, OPCODE_MAX
, result_dst
, op
[0], op
[1]);
1342 ir_to_mesa_emit_scalar_op2(ir
, OPCODE_POW
, result_dst
, op
[0], op
[1]);
1345 case ir_unop_bit_not
:
1347 case ir_binop_lshift
:
1348 case ir_binop_rshift
:
1349 case ir_binop_bit_and
:
1350 case ir_binop_bit_xor
:
1351 case ir_binop_bit_or
:
1352 assert(!"GLSL 1.30 features unsupported");
1356 this->result
= result_src
;
1361 ir_to_mesa_visitor::visit(ir_swizzle
*ir
)
1363 ir_to_mesa_src_reg src_reg
;
1367 /* Note that this is only swizzles in expressions, not those on the left
1368 * hand side of an assignment, which do write masking. See ir_assignment
1372 ir
->val
->accept(this);
1373 src_reg
= this->result
;
1374 assert(src_reg
.file
!= PROGRAM_UNDEFINED
);
1376 for (i
= 0; i
< 4; i
++) {
1377 if (i
< ir
->type
->vector_elements
) {
1380 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.x
);
1383 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.y
);
1386 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.z
);
1389 swizzle
[i
] = GET_SWZ(src_reg
.swizzle
, ir
->mask
.w
);
1393 /* If the type is smaller than a vec4, replicate the last
1396 swizzle
[i
] = swizzle
[ir
->type
->vector_elements
- 1];
1400 src_reg
.swizzle
= MAKE_SWIZZLE4(swizzle
[0],
1405 this->result
= src_reg
;
1409 ir_to_mesa_visitor::visit(ir_dereference_variable
*ir
)
1411 variable_storage
*entry
= find_variable_storage(ir
->var
);
1414 switch (ir
->var
->mode
) {
1415 case ir_var_uniform
:
1416 entry
= new(mem_ctx
) variable_storage(ir
->var
, PROGRAM_UNIFORM
,
1418 this->variables
.push_tail(entry
);
1423 /* The linker assigns locations for varyings and attributes,
1424 * including deprecated builtins (like gl_Color), user-assign
1425 * generic attributes (glBindVertexLocation), and
1426 * user-defined varyings.
1428 * FINISHME: We would hit this path for function arguments. Fix!
1430 assert(ir
->var
->location
!= -1);
1431 if (ir
->var
->mode
== ir_var_in
||
1432 ir
->var
->mode
== ir_var_inout
) {
1433 entry
= new(mem_ctx
) variable_storage(ir
->var
,
1437 if (this->prog
->Target
== GL_VERTEX_PROGRAM_ARB
&&
1438 ir
->var
->location
>= VERT_ATTRIB_GENERIC0
) {
1439 _mesa_add_attribute(prog
->Attributes
,
1441 _mesa_sizeof_glsl_type(ir
->var
->type
->gl_type
),
1442 ir
->var
->type
->gl_type
,
1443 ir
->var
->location
- VERT_ATTRIB_GENERIC0
);
1446 entry
= new(mem_ctx
) variable_storage(ir
->var
,
1453 case ir_var_temporary
:
1454 entry
= new(mem_ctx
) variable_storage(ir
->var
, PROGRAM_TEMPORARY
,
1456 this->variables
.push_tail(entry
);
1458 next_temp
+= type_size(ir
->var
->type
);
1463 printf("Failed to make storage for %s\n", ir
->var
->name
);
1468 this->result
= ir_to_mesa_src_reg(entry
->file
, entry
->index
, ir
->var
->type
);
1472 ir_to_mesa_visitor::visit(ir_dereference_array
*ir
)
1475 ir_to_mesa_src_reg src_reg
;
1476 int element_size
= type_size(ir
->type
);
1478 index
= ir
->array_index
->constant_expression_value();
1480 ir
->array
->accept(this);
1481 src_reg
= this->result
;
1484 src_reg
.index
+= index
->value
.i
[0] * element_size
;
1486 ir_to_mesa_src_reg array_base
= this->result
;
1487 /* Variable index array dereference. It eats the "vec4" of the
1488 * base of the array and an index that offsets the Mesa register
1491 ir
->array_index
->accept(this);
1493 ir_to_mesa_src_reg index_reg
;
1495 if (element_size
== 1) {
1496 index_reg
= this->result
;
1498 index_reg
= get_temp(glsl_type::float_type
);
1500 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
,
1501 ir_to_mesa_dst_reg_from_src(index_reg
),
1502 this->result
, src_reg_for_float(element_size
));
1505 src_reg
.reladdr
= talloc(mem_ctx
, ir_to_mesa_src_reg
);
1506 memcpy(src_reg
.reladdr
, &index_reg
, sizeof(index_reg
));
1509 /* If the type is smaller than a vec4, replicate the last channel out. */
1510 if (ir
->type
->is_scalar() || ir
->type
->is_vector())
1511 src_reg
.swizzle
= swizzle_for_size(ir
->type
->vector_elements
);
1513 src_reg
.swizzle
= SWIZZLE_NOOP
;
1515 this->result
= src_reg
;
1519 ir_to_mesa_visitor::visit(ir_dereference_record
*ir
)
1522 const glsl_type
*struct_type
= ir
->record
->type
;
1525 ir
->record
->accept(this);
1527 for (i
= 0; i
< struct_type
->length
; i
++) {
1528 if (strcmp(struct_type
->fields
.structure
[i
].name
, ir
->field
) == 0)
1530 offset
+= type_size(struct_type
->fields
.structure
[i
].type
);
1532 this->result
.swizzle
= swizzle_for_size(ir
->type
->vector_elements
);
1533 this->result
.index
+= offset
;
1537 * We want to be careful in assignment setup to hit the actual storage
1538 * instead of potentially using a temporary like we might with the
1539 * ir_dereference handler.
1541 static struct ir_to_mesa_dst_reg
1542 get_assignment_lhs(ir_dereference
*ir
, ir_to_mesa_visitor
*v
)
1544 /* The LHS must be a dereference. If the LHS is a variable indexed array
1545 * access of a vector, it must be separated into a series conditional moves
1546 * before reaching this point (see ir_vec_index_to_cond_assign).
1548 assert(ir
->as_dereference());
1549 ir_dereference_array
*deref_array
= ir
->as_dereference_array();
1551 assert(!deref_array
->array
->type
->is_vector());
1554 /* Use the rvalue deref handler for the most part. We'll ignore
1555 * swizzles in it and write swizzles using writemask, though.
1558 return ir_to_mesa_dst_reg_from_src(v
->result
);
1562 ir_to_mesa_visitor::visit(ir_assignment
*ir
)
1564 struct ir_to_mesa_dst_reg l
;
1565 struct ir_to_mesa_src_reg r
;
1568 ir
->rhs
->accept(this);
1571 l
= get_assignment_lhs(ir
->lhs
, this);
1573 /* FINISHME: This should really set to the correct maximal writemask for each
1574 * FINISHME: component written (in the loops below). This case can only
1575 * FINISHME: occur for matrices, arrays, and structures.
1577 if (ir
->write_mask
== 0) {
1578 assert(!ir
->lhs
->type
->is_scalar() && !ir
->lhs
->type
->is_vector());
1579 l
.writemask
= WRITEMASK_XYZW
;
1580 } else if (ir
->lhs
->type
->is_scalar()) {
1581 /* FINISHME: This hack makes writing to gl_FragData, which lives in the
1582 * FINISHME: W component of fragment shader output zero, work correctly.
1584 l
.writemask
= WRITEMASK_XYZW
;
1586 assert(ir
->lhs
->type
->is_vector());
1587 l
.writemask
= ir
->write_mask
;
1590 assert(l
.file
!= PROGRAM_UNDEFINED
);
1591 assert(r
.file
!= PROGRAM_UNDEFINED
);
1593 if (ir
->condition
) {
1594 ir_to_mesa_src_reg condition
;
1596 ir
->condition
->accept(this);
1597 condition
= this->result
;
1599 /* We use the OPCODE_CMP (a < 0 ? b : c) for conditional moves,
1600 * and the condition we produced is 0.0 or 1.0. By flipping the
1601 * sign, we can choose which value OPCODE_CMP produces without
1602 * an extra computing the condition.
1604 condition
.negate
= ~condition
.negate
;
1605 for (i
= 0; i
< type_size(ir
->lhs
->type
); i
++) {
1606 ir_to_mesa_emit_op3(ir
, OPCODE_CMP
, l
,
1607 condition
, r
, ir_to_mesa_src_reg_from_dst(l
));
1612 for (i
= 0; i
< type_size(ir
->lhs
->type
); i
++) {
1613 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1622 ir_to_mesa_visitor::visit(ir_constant
*ir
)
1624 ir_to_mesa_src_reg src_reg
;
1625 GLfloat stack_vals
[4] = { 0 };
1626 GLfloat
*values
= stack_vals
;
1629 /* Unfortunately, 4 floats is all we can get into
1630 * _mesa_add_unnamed_constant. So, make a temp to store an
1631 * aggregate constant and move each constant value into it. If we
1632 * get lucky, copy propagation will eliminate the extra moves.
1635 if (ir
->type
->base_type
== GLSL_TYPE_STRUCT
) {
1636 ir_to_mesa_src_reg temp_base
= get_temp(ir
->type
);
1637 ir_to_mesa_dst_reg temp
= ir_to_mesa_dst_reg_from_src(temp_base
);
1639 foreach_iter(exec_list_iterator
, iter
, ir
->components
) {
1640 ir_constant
*field_value
= (ir_constant
*)iter
.get();
1641 int size
= type_size(field_value
->type
);
1645 field_value
->accept(this);
1646 src_reg
= this->result
;
1648 for (i
= 0; i
< (unsigned int)size
; i
++) {
1649 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, temp
, src_reg
);
1655 this->result
= temp_base
;
1659 if (ir
->type
->is_array()) {
1660 ir_to_mesa_src_reg temp_base
= get_temp(ir
->type
);
1661 ir_to_mesa_dst_reg temp
= ir_to_mesa_dst_reg_from_src(temp_base
);
1662 int size
= type_size(ir
->type
->fields
.array
);
1666 for (i
= 0; i
< ir
->type
->length
; i
++) {
1667 ir
->array_elements
[i
]->accept(this);
1668 src_reg
= this->result
;
1669 for (int j
= 0; j
< size
; j
++) {
1670 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, temp
, src_reg
);
1676 this->result
= temp_base
;
1680 if (ir
->type
->is_matrix()) {
1681 ir_to_mesa_src_reg mat
= get_temp(ir
->type
);
1682 ir_to_mesa_dst_reg mat_column
= ir_to_mesa_dst_reg_from_src(mat
);
1684 for (i
= 0; i
< ir
->type
->matrix_columns
; i
++) {
1685 assert(ir
->type
->base_type
== GLSL_TYPE_FLOAT
);
1686 values
= &ir
->value
.f
[i
* ir
->type
->vector_elements
];
1688 src_reg
= ir_to_mesa_src_reg(PROGRAM_CONSTANT
, -1, NULL
);
1689 src_reg
.index
= _mesa_add_unnamed_constant(this->prog
->Parameters
,
1691 ir
->type
->vector_elements
,
1693 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, mat_column
, src_reg
);
1702 src_reg
.file
= PROGRAM_CONSTANT
;
1703 switch (ir
->type
->base_type
) {
1704 case GLSL_TYPE_FLOAT
:
1705 values
= &ir
->value
.f
[0];
1707 case GLSL_TYPE_UINT
:
1708 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1709 values
[i
] = ir
->value
.u
[i
];
1713 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1714 values
[i
] = ir
->value
.i
[i
];
1717 case GLSL_TYPE_BOOL
:
1718 for (i
= 0; i
< ir
->type
->vector_elements
; i
++) {
1719 values
[i
] = ir
->value
.b
[i
];
1723 assert(!"Non-float/uint/int/bool constant");
1726 this->result
= ir_to_mesa_src_reg(PROGRAM_CONSTANT
, -1, ir
->type
);
1727 this->result
.index
= _mesa_add_unnamed_constant(this->prog
->Parameters
,
1729 ir
->type
->vector_elements
,
1730 &this->result
.swizzle
);
1734 ir_to_mesa_visitor::get_function_signature(ir_function_signature
*sig
)
1736 function_entry
*entry
;
1738 foreach_iter(exec_list_iterator
, iter
, this->function_signatures
) {
1739 entry
= (function_entry
*)iter
.get();
1741 if (entry
->sig
== sig
)
1745 entry
= talloc(mem_ctx
, function_entry
);
1747 entry
->sig_id
= this->next_signature_id
++;
1748 entry
->bgn_inst
= NULL
;
1750 /* Allocate storage for all the parameters. */
1751 foreach_iter(exec_list_iterator
, iter
, sig
->parameters
) {
1752 ir_variable
*param
= (ir_variable
*)iter
.get();
1753 variable_storage
*storage
;
1755 storage
= find_variable_storage(param
);
1758 storage
= new(mem_ctx
) variable_storage(param
, PROGRAM_TEMPORARY
,
1760 this->variables
.push_tail(storage
);
1762 this->next_temp
+= type_size(param
->type
);
1765 if (!sig
->return_type
->is_void()) {
1766 entry
->return_reg
= get_temp(sig
->return_type
);
1768 entry
->return_reg
= ir_to_mesa_undef
;
1771 this->function_signatures
.push_tail(entry
);
1776 ir_to_mesa_visitor::visit(ir_call
*ir
)
1778 ir_to_mesa_instruction
*call_inst
;
1779 ir_function_signature
*sig
= ir
->get_callee();
1780 function_entry
*entry
= get_function_signature(sig
);
1783 /* Process in parameters. */
1784 exec_list_iterator sig_iter
= sig
->parameters
.iterator();
1785 foreach_iter(exec_list_iterator
, iter
, *ir
) {
1786 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
1787 ir_variable
*param
= (ir_variable
*)sig_iter
.get();
1789 if (param
->mode
== ir_var_in
||
1790 param
->mode
== ir_var_inout
) {
1791 variable_storage
*storage
= find_variable_storage(param
);
1794 param_rval
->accept(this);
1795 ir_to_mesa_src_reg r
= this->result
;
1797 ir_to_mesa_dst_reg l
;
1798 l
.file
= storage
->file
;
1799 l
.index
= storage
->index
;
1801 l
.writemask
= WRITEMASK_XYZW
;
1802 l
.cond_mask
= COND_TR
;
1804 for (i
= 0; i
< type_size(param
->type
); i
++) {
1805 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1813 assert(!sig_iter
.has_next());
1815 /* Emit call instruction */
1816 call_inst
= ir_to_mesa_emit_op1(ir
, OPCODE_CAL
,
1817 ir_to_mesa_undef_dst
, ir_to_mesa_undef
);
1818 call_inst
->function
= entry
;
1820 /* Process out parameters. */
1821 sig_iter
= sig
->parameters
.iterator();
1822 foreach_iter(exec_list_iterator
, iter
, *ir
) {
1823 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
1824 ir_variable
*param
= (ir_variable
*)sig_iter
.get();
1826 if (param
->mode
== ir_var_out
||
1827 param
->mode
== ir_var_inout
) {
1828 variable_storage
*storage
= find_variable_storage(param
);
1831 ir_to_mesa_src_reg r
;
1832 r
.file
= storage
->file
;
1833 r
.index
= storage
->index
;
1835 r
.swizzle
= SWIZZLE_NOOP
;
1838 param_rval
->accept(this);
1839 ir_to_mesa_dst_reg l
= ir_to_mesa_dst_reg_from_src(this->result
);
1841 for (i
= 0; i
< type_size(param
->type
); i
++) {
1842 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
1850 assert(!sig_iter
.has_next());
1852 /* Process return value. */
1853 this->result
= entry
->return_reg
;
1856 class get_sampler_name
: public ir_hierarchical_visitor
1859 get_sampler_name(ir_to_mesa_visitor
*mesa
, ir_dereference
*last
)
1861 this->mem_ctx
= mesa
->mem_ctx
;
1868 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1870 this->name
= ir
->var
->name
;
1871 return visit_continue
;
1874 virtual ir_visitor_status
visit_leave(ir_dereference_record
*ir
)
1876 this->name
= talloc_asprintf(mem_ctx
, "%s.%s", name
, ir
->field
);
1877 return visit_continue
;
1880 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
1882 ir_constant
*index
= ir
->array_index
->as_constant();
1886 i
= index
->value
.i
[0];
1888 /* GLSL 1.10 and 1.20 allowed variable sampler array indices,
1889 * while GLSL 1.30 requires that the array indices be
1890 * constant integer expressions. We don't expect any driver
1891 * to actually work with a really variable array index, so
1892 * all that would work would be an unrolled loop counter that ends
1893 * up being constant above.
1895 mesa
->shader_program
->InfoLog
=
1896 talloc_asprintf_append(mesa
->shader_program
->InfoLog
,
1897 "warning: Variable sampler array index "
1898 "unsupported.\nThis feature of the language "
1899 "was removed in GLSL 1.20 and is unlikely "
1900 "to be supported for 1.10 in Mesa.\n");
1904 this->name
= talloc_asprintf(mem_ctx
, "%s[%d]", name
, i
);
1908 return visit_continue
;
1911 ir_to_mesa_visitor
*mesa
;
1915 ir_dereference
*last
;
1919 ir_to_mesa_visitor::get_sampler_uniform_value(ir_dereference
*sampler
)
1921 get_sampler_name
getname(this, sampler
);
1923 sampler
->accept(&getname
);
1925 GLint index
= _mesa_lookup_parameter_index(prog
->Parameters
, -1,
1929 fail_link(this->shader_program
,
1930 "failed to find sampler named %s.\n", getname
.name
);
1934 index
+= getname
.offset
;
1936 return this->prog
->Parameters
->ParameterValues
[index
][0];
1940 ir_to_mesa_visitor::visit(ir_texture
*ir
)
1942 ir_to_mesa_src_reg result_src
, coord
, lod_info
, projector
;
1943 ir_to_mesa_dst_reg result_dst
, coord_dst
;
1944 ir_to_mesa_instruction
*inst
= NULL
;
1945 prog_opcode opcode
= OPCODE_NOP
;
1947 ir
->coordinate
->accept(this);
1949 /* Put our coords in a temp. We'll need to modify them for shadow,
1950 * projection, or LOD, so the only case we'd use it as is is if
1951 * we're doing plain old texturing. Mesa IR optimization should
1952 * handle cleaning up our mess in that case.
1954 coord
= get_temp(glsl_type::vec4_type
);
1955 coord_dst
= ir_to_mesa_dst_reg_from_src(coord
);
1956 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
,
1959 if (ir
->projector
) {
1960 ir
->projector
->accept(this);
1961 projector
= this->result
;
1964 /* Storage for our result. Ideally for an assignment we'd be using
1965 * the actual storage for the result here, instead.
1967 result_src
= get_temp(glsl_type::vec4_type
);
1968 result_dst
= ir_to_mesa_dst_reg_from_src(result_src
);
1972 opcode
= OPCODE_TEX
;
1975 opcode
= OPCODE_TXB
;
1976 ir
->lod_info
.bias
->accept(this);
1977 lod_info
= this->result
;
1980 opcode
= OPCODE_TXL
;
1981 ir
->lod_info
.lod
->accept(this);
1982 lod_info
= this->result
;
1986 assert(!"GLSL 1.30 features unsupported");
1990 if (ir
->projector
) {
1991 if (opcode
== OPCODE_TEX
) {
1992 /* Slot the projector in as the last component of the coord. */
1993 coord_dst
.writemask
= WRITEMASK_W
;
1994 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, projector
);
1995 coord_dst
.writemask
= WRITEMASK_XYZW
;
1996 opcode
= OPCODE_TXP
;
1998 ir_to_mesa_src_reg coord_w
= coord
;
1999 coord_w
.swizzle
= SWIZZLE_WWWW
;
2001 /* For the other TEX opcodes there's no projective version
2002 * since the last slot is taken up by lod info. Do the
2003 * projective divide now.
2005 coord_dst
.writemask
= WRITEMASK_W
;
2006 ir_to_mesa_emit_op1(ir
, OPCODE_RCP
, coord_dst
, projector
);
2008 coord_dst
.writemask
= WRITEMASK_XYZ
;
2009 ir_to_mesa_emit_op2(ir
, OPCODE_MUL
, coord_dst
, coord
, coord_w
);
2011 coord_dst
.writemask
= WRITEMASK_XYZW
;
2012 coord
.swizzle
= SWIZZLE_XYZW
;
2016 if (ir
->shadow_comparitor
) {
2017 /* Slot the shadow value in as the second to last component of the
2020 ir
->shadow_comparitor
->accept(this);
2021 coord_dst
.writemask
= WRITEMASK_Z
;
2022 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, this->result
);
2023 coord_dst
.writemask
= WRITEMASK_XYZW
;
2026 if (opcode
== OPCODE_TXL
|| opcode
== OPCODE_TXB
) {
2027 /* Mesa IR stores lod or lod bias in the last channel of the coords. */
2028 coord_dst
.writemask
= WRITEMASK_W
;
2029 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, coord_dst
, lod_info
);
2030 coord_dst
.writemask
= WRITEMASK_XYZW
;
2033 inst
= ir_to_mesa_emit_op1(ir
, opcode
, result_dst
, coord
);
2035 if (ir
->shadow_comparitor
)
2036 inst
->tex_shadow
= GL_TRUE
;
2038 inst
->sampler
= get_sampler_uniform_value(ir
->sampler
);
2040 const glsl_type
*sampler_type
= ir
->sampler
->type
;
2042 switch (sampler_type
->sampler_dimensionality
) {
2043 case GLSL_SAMPLER_DIM_1D
:
2044 inst
->tex_target
= (sampler_type
->sampler_array
)
2045 ? TEXTURE_1D_ARRAY_INDEX
: TEXTURE_1D_INDEX
;
2047 case GLSL_SAMPLER_DIM_2D
:
2048 inst
->tex_target
= (sampler_type
->sampler_array
)
2049 ? TEXTURE_2D_ARRAY_INDEX
: TEXTURE_2D_INDEX
;
2051 case GLSL_SAMPLER_DIM_3D
:
2052 inst
->tex_target
= TEXTURE_3D_INDEX
;
2054 case GLSL_SAMPLER_DIM_CUBE
:
2055 inst
->tex_target
= TEXTURE_CUBE_INDEX
;
2057 case GLSL_SAMPLER_DIM_RECT
:
2058 inst
->tex_target
= TEXTURE_RECT_INDEX
;
2060 case GLSL_SAMPLER_DIM_BUF
:
2061 assert(!"FINISHME: Implement ARB_texture_buffer_object");
2064 assert(!"Should not get here.");
2067 this->result
= result_src
;
2071 ir_to_mesa_visitor::visit(ir_return
*ir
)
2073 if (ir
->get_value()) {
2074 ir_to_mesa_dst_reg l
;
2077 assert(current_function
);
2079 ir
->get_value()->accept(this);
2080 ir_to_mesa_src_reg r
= this->result
;
2082 l
= ir_to_mesa_dst_reg_from_src(current_function
->return_reg
);
2084 for (i
= 0; i
< type_size(current_function
->sig
->return_type
); i
++) {
2085 ir_to_mesa_emit_op1(ir
, OPCODE_MOV
, l
, r
);
2091 ir_to_mesa_emit_op0(ir
, OPCODE_RET
);
2095 ir_to_mesa_visitor::visit(ir_discard
*ir
)
2097 struct gl_fragment_program
*fp
= (struct gl_fragment_program
*)this->prog
;
2099 assert(ir
->condition
== NULL
); /* FINISHME */
2101 ir_to_mesa_emit_op0(ir
, OPCODE_KIL_NV
);
2102 fp
->UsesKill
= GL_TRUE
;
2106 ir_to_mesa_visitor::visit(ir_if
*ir
)
2108 ir_to_mesa_instruction
*cond_inst
, *if_inst
, *else_inst
= NULL
;
2109 ir_to_mesa_instruction
*prev_inst
;
2111 prev_inst
= (ir_to_mesa_instruction
*)this->instructions
.get_tail();
2113 ir
->condition
->accept(this);
2114 assert(this->result
.file
!= PROGRAM_UNDEFINED
);
2116 if (this->options
->EmitCondCodes
) {
2117 cond_inst
= (ir_to_mesa_instruction
*)this->instructions
.get_tail();
2119 /* See if we actually generated any instruction for generating
2120 * the condition. If not, then cook up a move to a temp so we
2121 * have something to set cond_update on.
2123 if (cond_inst
== prev_inst
) {
2124 ir_to_mesa_src_reg temp
= get_temp(glsl_type::bool_type
);
2125 cond_inst
= ir_to_mesa_emit_op1(ir
->condition
, OPCODE_MOV
,
2126 ir_to_mesa_dst_reg_from_src(temp
),
2129 cond_inst
->cond_update
= GL_TRUE
;
2131 if_inst
= ir_to_mesa_emit_op0(ir
->condition
, OPCODE_IF
);
2132 if_inst
->dst_reg
.cond_mask
= COND_NE
;
2134 if_inst
= ir_to_mesa_emit_op1(ir
->condition
,
2135 OPCODE_IF
, ir_to_mesa_undef_dst
,
2139 this->instructions
.push_tail(if_inst
);
2141 visit_exec_list(&ir
->then_instructions
, this);
2143 if (!ir
->else_instructions
.is_empty()) {
2144 else_inst
= ir_to_mesa_emit_op0(ir
->condition
, OPCODE_ELSE
);
2145 visit_exec_list(&ir
->else_instructions
, this);
2148 if_inst
= ir_to_mesa_emit_op1(ir
->condition
, OPCODE_ENDIF
,
2149 ir_to_mesa_undef_dst
, ir_to_mesa_undef
);
2152 ir_to_mesa_visitor::ir_to_mesa_visitor()
2154 result
.file
= PROGRAM_UNDEFINED
;
2156 next_signature_id
= 1;
2157 current_function
= NULL
;
2158 mem_ctx
= talloc_new(NULL
);
2161 ir_to_mesa_visitor::~ir_to_mesa_visitor()
2163 talloc_free(mem_ctx
);
2166 static struct prog_src_register
2167 mesa_src_reg_from_ir_src_reg(ir_to_mesa_src_reg reg
)
2169 struct prog_src_register mesa_reg
;
2171 mesa_reg
.File
= reg
.file
;
2172 assert(reg
.index
< (1 << INST_INDEX_BITS
) - 1);
2173 mesa_reg
.Index
= reg
.index
;
2174 mesa_reg
.Swizzle
= reg
.swizzle
;
2175 mesa_reg
.RelAddr
= reg
.reladdr
!= NULL
;
2176 mesa_reg
.Negate
= reg
.negate
;
2178 mesa_reg
.HasIndex2
= GL_FALSE
;
2179 mesa_reg
.RelAddr2
= 0;
2180 mesa_reg
.Index2
= 0;
2186 set_branchtargets(ir_to_mesa_visitor
*v
,
2187 struct prog_instruction
*mesa_instructions
,
2188 int num_instructions
)
2190 int if_count
= 0, loop_count
= 0;
2191 int *if_stack
, *loop_stack
;
2192 int if_stack_pos
= 0, loop_stack_pos
= 0;
2195 for (i
= 0; i
< num_instructions
; i
++) {
2196 switch (mesa_instructions
[i
].Opcode
) {
2200 case OPCODE_BGNLOOP
:
2205 mesa_instructions
[i
].BranchTarget
= -1;
2212 if_stack
= talloc_zero_array(v
->mem_ctx
, int, if_count
);
2213 loop_stack
= talloc_zero_array(v
->mem_ctx
, int, loop_count
);
2215 for (i
= 0; i
< num_instructions
; i
++) {
2216 switch (mesa_instructions
[i
].Opcode
) {
2218 if_stack
[if_stack_pos
] = i
;
2222 mesa_instructions
[if_stack
[if_stack_pos
- 1]].BranchTarget
= i
;
2223 if_stack
[if_stack_pos
- 1] = i
;
2226 mesa_instructions
[if_stack
[if_stack_pos
- 1]].BranchTarget
= i
;
2229 case OPCODE_BGNLOOP
:
2230 loop_stack
[loop_stack_pos
] = i
;
2233 case OPCODE_ENDLOOP
:
2235 /* Rewrite any breaks/conts at this nesting level (haven't
2236 * already had a BranchTarget assigned) to point to the end
2239 for (j
= loop_stack
[loop_stack_pos
]; j
< i
; j
++) {
2240 if (mesa_instructions
[j
].Opcode
== OPCODE_BRK
||
2241 mesa_instructions
[j
].Opcode
== OPCODE_CONT
) {
2242 if (mesa_instructions
[j
].BranchTarget
== -1) {
2243 mesa_instructions
[j
].BranchTarget
= i
;
2247 /* The loop ends point at each other. */
2248 mesa_instructions
[i
].BranchTarget
= loop_stack
[loop_stack_pos
];
2249 mesa_instructions
[loop_stack
[loop_stack_pos
]].BranchTarget
= i
;
2252 foreach_iter(exec_list_iterator
, iter
, v
->function_signatures
) {
2253 function_entry
*entry
= (function_entry
*)iter
.get();
2255 if (entry
->sig_id
== mesa_instructions
[i
].BranchTarget
) {
2256 mesa_instructions
[i
].BranchTarget
= entry
->inst
;
2268 print_program(struct prog_instruction
*mesa_instructions
,
2269 ir_instruction
**mesa_instruction_annotation
,
2270 int num_instructions
)
2272 ir_instruction
*last_ir
= NULL
;
2276 for (i
= 0; i
< num_instructions
; i
++) {
2277 struct prog_instruction
*mesa_inst
= mesa_instructions
+ i
;
2278 ir_instruction
*ir
= mesa_instruction_annotation
[i
];
2280 fprintf(stdout
, "%3d: ", i
);
2282 if (last_ir
!= ir
&& ir
) {
2285 for (j
= 0; j
< indent
; j
++) {
2286 fprintf(stdout
, " ");
2292 fprintf(stdout
, " "); /* line number spacing. */
2295 indent
= _mesa_fprint_instruction_opt(stdout
, mesa_inst
, indent
,
2296 PROG_PRINT_DEBUG
, NULL
);
2301 count_resources(struct gl_program
*prog
)
2305 prog
->SamplersUsed
= 0;
2307 for (i
= 0; i
< prog
->NumInstructions
; i
++) {
2308 struct prog_instruction
*inst
= &prog
->Instructions
[i
];
2310 if (_mesa_is_tex_instruction(inst
->Opcode
)) {
2311 prog
->SamplerTargets
[inst
->TexSrcUnit
] =
2312 (gl_texture_index
)inst
->TexSrcTarget
;
2313 prog
->SamplersUsed
|= 1 << inst
->TexSrcUnit
;
2314 if (inst
->TexShadow
) {
2315 prog
->ShadowSamplers
|= 1 << inst
->TexSrcUnit
;
2320 _mesa_update_shader_textures_used(prog
);
2323 struct uniform_sort
{
2324 struct gl_uniform
*u
;
2328 /* The shader_program->Uniforms list is almost sorted in increasing
2329 * uniform->{Frag,Vert}Pos locations, but not quite when there are
2330 * uniforms shared between targets. We need to add parameters in
2331 * increasing order for the targets.
2334 sort_uniforms(const void *a
, const void *b
)
2336 struct uniform_sort
*u1
= (struct uniform_sort
*)a
;
2337 struct uniform_sort
*u2
= (struct uniform_sort
*)b
;
2339 return u1
->pos
- u2
->pos
;
2342 /* Add the uniforms to the parameters. The linker chose locations
2343 * in our parameters lists (which weren't created yet), which the
2344 * uniforms code will use to poke values into our parameters list
2345 * when uniforms are updated.
2348 add_uniforms_to_parameters_list(struct gl_shader_program
*shader_program
,
2349 struct gl_shader
*shader
,
2350 struct gl_program
*prog
)
2353 unsigned int next_sampler
= 0, num_uniforms
= 0;
2354 struct uniform_sort
*sorted_uniforms
;
2356 sorted_uniforms
= talloc_array(NULL
, struct uniform_sort
,
2357 shader_program
->Uniforms
->NumUniforms
);
2359 for (i
= 0; i
< shader_program
->Uniforms
->NumUniforms
; i
++) {
2360 struct gl_uniform
*uniform
= shader_program
->Uniforms
->Uniforms
+ i
;
2361 int parameter_index
= -1;
2363 switch (shader
->Type
) {
2364 case GL_VERTEX_SHADER
:
2365 parameter_index
= uniform
->VertPos
;
2367 case GL_FRAGMENT_SHADER
:
2368 parameter_index
= uniform
->FragPos
;
2370 case GL_GEOMETRY_SHADER
:
2371 parameter_index
= uniform
->GeomPos
;
2375 /* Only add uniforms used in our target. */
2376 if (parameter_index
!= -1) {
2377 sorted_uniforms
[num_uniforms
].pos
= parameter_index
;
2378 sorted_uniforms
[num_uniforms
].u
= uniform
;
2383 qsort(sorted_uniforms
, num_uniforms
, sizeof(struct uniform_sort
),
2386 for (i
= 0; i
< num_uniforms
; i
++) {
2387 struct gl_uniform
*uniform
= sorted_uniforms
[i
].u
;
2388 int parameter_index
= sorted_uniforms
[i
].pos
;
2389 const glsl_type
*type
= uniform
->Type
;
2392 if (type
->is_vector() ||
2393 type
->is_scalar()) {
2394 size
= type
->vector_elements
;
2396 size
= type_size(type
) * 4;
2399 gl_register_file file
;
2400 if (type
->is_sampler() ||
2401 (type
->is_array() && type
->fields
.array
->is_sampler())) {
2402 file
= PROGRAM_SAMPLER
;
2404 file
= PROGRAM_UNIFORM
;
2407 GLint index
= _mesa_lookup_parameter_index(prog
->Parameters
, -1,
2411 index
= _mesa_add_parameter(prog
->Parameters
, file
,
2412 uniform
->Name
, size
, type
->gl_type
,
2415 /* Sampler uniform values are stored in prog->SamplerUnits,
2416 * and the entry in that array is selected by this index we
2417 * store in ParameterValues[].
2419 if (file
== PROGRAM_SAMPLER
) {
2420 for (unsigned int j
= 0; j
< size
/ 4; j
++)
2421 prog
->Parameters
->ParameterValues
[index
+ j
][0] = next_sampler
++;
2424 /* The location chosen in the Parameters list here (returned
2425 * from _mesa_add_uniform) has to match what the linker chose.
2427 if (index
!= parameter_index
) {
2428 fail_link(shader_program
, "Allocation of uniform `%s' to target "
2429 "failed (%d vs %d)\n",
2430 uniform
->Name
, index
, parameter_index
);
2435 talloc_free(sorted_uniforms
);
2439 set_uniform_initializer(GLcontext
*ctx
, void *mem_ctx
,
2440 struct gl_shader_program
*shader_program
,
2441 const char *name
, const glsl_type
*type
,
2444 if (type
->is_record()) {
2445 ir_constant
*field_constant
;
2447 field_constant
= (ir_constant
*)val
->components
.get_head();
2449 for (unsigned int i
= 0; i
< type
->length
; i
++) {
2450 const glsl_type
*field_type
= type
->fields
.structure
[i
].type
;
2451 const char *field_name
= talloc_asprintf(mem_ctx
, "%s.%s", name
,
2452 type
->fields
.structure
[i
].name
);
2453 set_uniform_initializer(ctx
, mem_ctx
, shader_program
, field_name
,
2454 field_type
, field_constant
);
2455 field_constant
= (ir_constant
*)field_constant
->next
;
2460 int loc
= _mesa_get_uniform_location(ctx
, shader_program
, name
);
2463 fail_link(shader_program
,
2464 "Couldn't find uniform for initializer %s\n", name
);
2468 for (unsigned int i
= 0; i
< (type
->is_array() ? type
->length
: 1); i
++) {
2469 ir_constant
*element
;
2470 const glsl_type
*element_type
;
2471 if (type
->is_array()) {
2472 element
= val
->array_elements
[i
];
2473 element_type
= type
->fields
.array
;
2476 element_type
= type
;
2481 if (element_type
->base_type
== GLSL_TYPE_BOOL
) {
2482 int *conv
= talloc_array(mem_ctx
, int, element_type
->components());
2483 for (unsigned int j
= 0; j
< element_type
->components(); j
++) {
2484 conv
[j
] = element
->value
.b
[j
];
2486 values
= (void *)conv
;
2487 element_type
= glsl_type::get_instance(GLSL_TYPE_INT
,
2488 element_type
->vector_elements
,
2491 values
= &element
->value
;
2494 if (element_type
->is_matrix()) {
2495 _mesa_uniform_matrix(ctx
, shader_program
,
2496 element_type
->matrix_columns
,
2497 element_type
->vector_elements
,
2498 loc
, 1, GL_FALSE
, (GLfloat
*)values
);
2499 loc
+= element_type
->matrix_columns
;
2501 _mesa_uniform(ctx
, shader_program
, loc
, element_type
->matrix_columns
,
2502 values
, element_type
->gl_type
);
2503 loc
+= type_size(element_type
);
2509 set_uniform_initializers(GLcontext
*ctx
,
2510 struct gl_shader_program
*shader_program
)
2512 void *mem_ctx
= NULL
;
2514 for (unsigned int i
= 0; i
< shader_program
->_NumLinkedShaders
; i
++) {
2515 struct gl_shader
*shader
= shader_program
->_LinkedShaders
[i
];
2516 foreach_iter(exec_list_iterator
, iter
, *shader
->ir
) {
2517 ir_instruction
*ir
= (ir_instruction
*)iter
.get();
2518 ir_variable
*var
= ir
->as_variable();
2520 if (!var
|| var
->mode
!= ir_var_uniform
|| !var
->constant_value
)
2524 mem_ctx
= talloc_new(NULL
);
2526 set_uniform_initializer(ctx
, mem_ctx
, shader_program
, var
->name
,
2527 var
->type
, var
->constant_value
);
2531 talloc_free(mem_ctx
);
2535 get_mesa_program(GLcontext
*ctx
, struct gl_shader_program
*shader_program
,
2536 struct gl_shader
*shader
)
2538 ir_to_mesa_visitor v
;
2539 struct prog_instruction
*mesa_instructions
, *mesa_inst
;
2540 ir_instruction
**mesa_instruction_annotation
;
2542 struct gl_program
*prog
;
2544 const char *target_string
;
2546 struct gl_shader_compiler_options
*options
=
2547 &ctx
->ShaderCompilerOptions
[_mesa_shader_type_to_index(shader
->Type
)];
2549 switch (shader
->Type
) {
2550 case GL_VERTEX_SHADER
:
2551 target
= GL_VERTEX_PROGRAM_ARB
;
2552 target_string
= "vertex";
2554 case GL_FRAGMENT_SHADER
:
2555 target
= GL_FRAGMENT_PROGRAM_ARB
;
2556 target_string
= "fragment";
2559 assert(!"should not be reached");
2563 validate_ir_tree(shader
->ir
);
2565 prog
= ctx
->Driver
.NewProgram(ctx
, target
, shader_program
->Name
);
2568 prog
->Parameters
= _mesa_new_parameter_list();
2569 prog
->Varying
= _mesa_new_parameter_list();
2570 prog
->Attributes
= _mesa_new_parameter_list();
2573 v
.shader_program
= shader_program
;
2574 v
.options
= options
;
2576 add_uniforms_to_parameters_list(shader_program
, shader
, prog
);
2578 /* Emit Mesa IR for main(). */
2579 visit_exec_list(shader
->ir
, &v
);
2580 v
.ir_to_mesa_emit_op0(NULL
, OPCODE_END
);
2582 /* Now emit bodies for any functions that were used. */
2584 progress
= GL_FALSE
;
2586 foreach_iter(exec_list_iterator
, iter
, v
.function_signatures
) {
2587 function_entry
*entry
= (function_entry
*)iter
.get();
2589 if (!entry
->bgn_inst
) {
2590 v
.current_function
= entry
;
2592 entry
->bgn_inst
= v
.ir_to_mesa_emit_op0(NULL
, OPCODE_BGNSUB
);
2593 entry
->bgn_inst
->function
= entry
;
2595 visit_exec_list(&entry
->sig
->body
, &v
);
2597 ir_to_mesa_instruction
*last
;
2598 last
= (ir_to_mesa_instruction
*)v
.instructions
.get_tail();
2599 if (last
->op
!= OPCODE_RET
)
2600 v
.ir_to_mesa_emit_op0(NULL
, OPCODE_RET
);
2602 ir_to_mesa_instruction
*end
;
2603 end
= v
.ir_to_mesa_emit_op0(NULL
, OPCODE_ENDSUB
);
2604 end
->function
= entry
;
2611 prog
->NumTemporaries
= v
.next_temp
;
2613 int num_instructions
= 0;
2614 foreach_iter(exec_list_iterator
, iter
, v
.instructions
) {
2619 (struct prog_instruction
*)calloc(num_instructions
,
2620 sizeof(*mesa_instructions
));
2621 mesa_instruction_annotation
= talloc_array(v
.mem_ctx
, ir_instruction
*,
2624 mesa_inst
= mesa_instructions
;
2626 foreach_iter(exec_list_iterator
, iter
, v
.instructions
) {
2627 ir_to_mesa_instruction
*inst
= (ir_to_mesa_instruction
*)iter
.get();
2629 mesa_inst
->Opcode
= inst
->op
;
2630 mesa_inst
->CondUpdate
= inst
->cond_update
;
2631 mesa_inst
->DstReg
.File
= inst
->dst_reg
.file
;
2632 mesa_inst
->DstReg
.Index
= inst
->dst_reg
.index
;
2633 mesa_inst
->DstReg
.CondMask
= inst
->dst_reg
.cond_mask
;
2634 mesa_inst
->DstReg
.WriteMask
= inst
->dst_reg
.writemask
;
2635 mesa_inst
->DstReg
.RelAddr
= inst
->dst_reg
.reladdr
!= NULL
;
2636 mesa_inst
->SrcReg
[0] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[0]);
2637 mesa_inst
->SrcReg
[1] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[1]);
2638 mesa_inst
->SrcReg
[2] = mesa_src_reg_from_ir_src_reg(inst
->src_reg
[2]);
2639 mesa_inst
->TexSrcUnit
= inst
->sampler
;
2640 mesa_inst
->TexSrcTarget
= inst
->tex_target
;
2641 mesa_inst
->TexShadow
= inst
->tex_shadow
;
2642 mesa_instruction_annotation
[i
] = inst
->ir
;
2644 /* Set IndirectRegisterFiles. */
2645 if (mesa_inst
->DstReg
.RelAddr
)
2646 prog
->IndirectRegisterFiles
|= 1 << mesa_inst
->DstReg
.File
;
2648 for (unsigned src
= 0; src
< 3; src
++)
2649 if (mesa_inst
->SrcReg
[src
].RelAddr
)
2650 prog
->IndirectRegisterFiles
|= 1 << mesa_inst
->SrcReg
[src
].File
;
2652 if (options
->EmitNoIfs
&& mesa_inst
->Opcode
== OPCODE_IF
) {
2653 fail_link(shader_program
, "Couldn't flatten if statement\n");
2656 switch (mesa_inst
->Opcode
) {
2658 inst
->function
->inst
= i
;
2659 mesa_inst
->Comment
= strdup(inst
->function
->sig
->function_name());
2662 mesa_inst
->Comment
= strdup(inst
->function
->sig
->function_name());
2665 mesa_inst
->BranchTarget
= inst
->function
->sig_id
; /* rewritten later */
2668 prog
->NumAddressRegs
= 1;
2678 set_branchtargets(&v
, mesa_instructions
, num_instructions
);
2680 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2682 printf("GLSL IR for linked %s program %d:\n", target_string
,
2683 shader_program
->Name
);
2684 _mesa_print_ir(shader
->ir
, NULL
);
2687 printf("Mesa IR for linked %s program %d:\n", target_string
,
2688 shader_program
->Name
);
2689 print_program(mesa_instructions
, mesa_instruction_annotation
,
2693 prog
->Instructions
= mesa_instructions
;
2694 prog
->NumInstructions
= num_instructions
;
2696 do_set_program_inouts(shader
->ir
, prog
);
2697 count_resources(prog
);
2699 _mesa_reference_program(ctx
, &shader
->Program
, prog
);
2701 if ((ctx
->Shader
.Flags
& GLSL_NO_OPT
) == 0) {
2702 _mesa_optimize_program(ctx
, prog
);
2710 _mesa_ir_compile_shader(GLcontext
*ctx
, struct gl_shader
*shader
)
2712 assert(shader
->CompileStatus
);
2719 _mesa_ir_link_shader(GLcontext
*ctx
, struct gl_shader_program
*prog
)
2721 assert(prog
->LinkStatus
);
2723 for (unsigned i
= 0; i
< prog
->_NumLinkedShaders
; i
++) {
2725 exec_list
*ir
= prog
->_LinkedShaders
[i
]->ir
;
2726 struct gl_shader_compiler_options
*options
=
2727 &ctx
->ShaderCompilerOptions
[_mesa_shader_type_to_index(prog
->_LinkedShaders
[i
]->Type
)];
2733 do_mat_op_to_vec(ir
);
2734 do_mod_to_fract(ir
);
2735 do_div_to_mul_rcp(ir
);
2736 do_explog_to_explog2(ir
);
2738 progress
= do_lower_jumps(ir
, true, true, options
->EmitNoMainReturn
, options
->EmitNoCont
, options
->EmitNoLoops
) || progress
;
2740 progress
= do_common_optimization(ir
, true, options
->MaxUnrollIterations
) || progress
;
2742 if (options
->EmitNoIfs
)
2743 progress
= do_if_to_cond_assign(ir
) || progress
;
2745 if (options
->EmitNoNoise
)
2746 progress
= lower_noise(ir
) || progress
;
2748 progress
= do_vec_index_to_cond_assign(ir
) || progress
;
2751 validate_ir_tree(ir
);
2754 for (unsigned i
= 0; i
< prog
->_NumLinkedShaders
; i
++) {
2755 struct gl_program
*linked_prog
;
2758 linked_prog
= get_mesa_program(ctx
, prog
, prog
->_LinkedShaders
[i
]);
2760 switch (prog
->_LinkedShaders
[i
]->Type
) {
2761 case GL_VERTEX_SHADER
:
2762 _mesa_reference_vertprog(ctx
, &prog
->VertexProgram
,
2763 (struct gl_vertex_program
*)linked_prog
);
2764 ok
= ctx
->Driver
.ProgramStringNotify(ctx
, GL_VERTEX_PROGRAM_ARB
,
2767 case GL_FRAGMENT_SHADER
:
2768 _mesa_reference_fragprog(ctx
, &prog
->FragmentProgram
,
2769 (struct gl_fragment_program
*)linked_prog
);
2770 ok
= ctx
->Driver
.ProgramStringNotify(ctx
, GL_FRAGMENT_PROGRAM_ARB
,
2777 _mesa_reference_program(ctx
, &linked_prog
, NULL
);
2784 _mesa_glsl_compile_shader(GLcontext
*ctx
, struct gl_shader
*shader
)
2786 struct _mesa_glsl_parse_state
*state
=
2787 new(shader
) _mesa_glsl_parse_state(ctx
, shader
->Type
, shader
);
2789 const char *source
= shader
->Source
;
2790 /* Check if the user called glCompileShader without first calling
2791 * glShaderSource. This should fail to compile, but not raise a GL_ERROR.
2793 if (source
== NULL
) {
2794 shader
->CompileStatus
= GL_FALSE
;
2798 state
->error
= preprocess(state
, &source
, &state
->info_log
,
2799 &ctx
->Extensions
, ctx
->API
);
2801 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2802 printf("GLSL source for shader %d:\n", shader
->Name
);
2803 printf("%s\n", shader
->Source
);
2806 if (!state
->error
) {
2807 _mesa_glsl_lexer_ctor(state
, source
);
2808 _mesa_glsl_parse(state
);
2809 _mesa_glsl_lexer_dtor(state
);
2812 talloc_free(shader
->ir
);
2813 shader
->ir
= new(shader
) exec_list
;
2814 if (!state
->error
&& !state
->translation_unit
.is_empty())
2815 _mesa_ast_to_hir(shader
->ir
, state
);
2817 if (!state
->error
&& !shader
->ir
->is_empty()) {
2818 validate_ir_tree(shader
->ir
);
2820 /* Do some optimization at compile time to reduce shader IR size
2821 * and reduce later work if the same shader is linked multiple times
2823 while (do_common_optimization(shader
->ir
, false, 32))
2826 validate_ir_tree(shader
->ir
);
2829 shader
->symbols
= state
->symbols
;
2831 shader
->CompileStatus
= !state
->error
;
2832 shader
->InfoLog
= state
->info_log
;
2833 shader
->Version
= state
->language_version
;
2834 memcpy(shader
->builtins_to_link
, state
->builtins_to_link
,
2835 sizeof(shader
->builtins_to_link
[0]) * state
->num_builtins_to_link
);
2836 shader
->num_builtins_to_link
= state
->num_builtins_to_link
;
2838 if (ctx
->Shader
.Flags
& GLSL_LOG
) {
2839 _mesa_write_shader_to_file(shader
);
2842 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2843 if (shader
->CompileStatus
) {
2844 printf("GLSL IR for shader %d:\n", shader
->Name
);
2845 _mesa_print_ir(shader
->ir
, NULL
);
2848 printf("GLSL shader %d failed to compile.\n", shader
->Name
);
2850 if (shader
->InfoLog
&& shader
->InfoLog
[0] != 0) {
2851 printf("GLSL shader %d info log:\n", shader
->Name
);
2852 printf("%s\n", shader
->InfoLog
);
2856 /* Retain any live IR, but trash the rest. */
2857 reparent_ir(shader
->ir
, shader
->ir
);
2861 if (shader
->CompileStatus
) {
2862 if (!ctx
->Driver
.CompileShader(ctx
, shader
))
2863 shader
->CompileStatus
= GL_FALSE
;
2868 _mesa_glsl_link_shader(GLcontext
*ctx
, struct gl_shader_program
*prog
)
2872 _mesa_clear_shader_program_data(ctx
, prog
);
2874 prog
->LinkStatus
= GL_TRUE
;
2876 for (i
= 0; i
< prog
->NumShaders
; i
++) {
2877 if (!prog
->Shaders
[i
]->CompileStatus
) {
2878 fail_link(prog
, "linking with uncompiled shader");
2879 prog
->LinkStatus
= GL_FALSE
;
2883 prog
->Varying
= _mesa_new_parameter_list();
2884 _mesa_reference_vertprog(ctx
, &prog
->VertexProgram
, NULL
);
2885 _mesa_reference_fragprog(ctx
, &prog
->FragmentProgram
, NULL
);
2887 if (prog
->LinkStatus
) {
2888 link_shaders(ctx
, prog
);
2891 if (prog
->LinkStatus
) {
2892 if (!ctx
->Driver
.LinkShader(ctx
, prog
)) {
2893 prog
->LinkStatus
= GL_FALSE
;
2897 set_uniform_initializers(ctx
, prog
);
2899 if (ctx
->Shader
.Flags
& GLSL_DUMP
) {
2900 if (!prog
->LinkStatus
) {
2901 printf("GLSL shader program %d failed to link\n", prog
->Name
);
2904 if (prog
->InfoLog
&& prog
->InfoLog
[0] != 0) {
2905 printf("GLSL shader program %d info log:\n", prog
->Name
);
2906 printf("%s\n", prog
->InfoLog
);