1 /**************************************************************************
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
5 * Copyright 2009 VMware, Inc. All Rights Reserved.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **************************************************************************/
32 #include "program/program.h"
33 #include "program/prog_parameter.h"
34 #include "program/prog_cache.h"
35 #include "program/prog_instruction.h"
36 #include "program/prog_print.h"
37 #include "program/prog_statevars.h"
38 #include "program/programopt.h"
39 #include "texenvprogram.h"
43 * Note on texture units:
45 * The number of texture units supported by fixed-function fragment
46 * processing is MAX_TEXTURE_COORD_UNITS, not MAX_TEXTURE_IMAGE_UNITS.
47 * That's because there's a one-to-one correspondence between texture
48 * coordinates and samplers in fixed-function processing.
50 * Since fixed-function vertex processing is limited to MAX_TEXTURE_COORD_UNITS
51 * sets of texcoords, so is fixed-function fragment processing.
53 * We can safely use ctx->Const.MaxTextureUnits for loop bounds.
57 struct texenvprog_cache_item
61 struct gl_fragment_program
*data
;
62 struct texenvprog_cache_item
*next
;
66 texenv_doing_secondary_color(struct gl_context
*ctx
)
68 if (ctx
->Light
.Enabled
&&
69 (ctx
->Light
.Model
.ColorControl
== GL_SEPARATE_SPECULAR_COLOR
))
72 if (ctx
->Fog
.ColorSumEnabled
)
79 * Up to nine instructions per tex unit, plus fog, specular color.
81 #define MAX_INSTRUCTIONS ((MAX_TEXTURE_COORD_UNITS * 9) + 12)
83 #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
87 __extension__ GLubyte Source
:4; /**< SRC_x */
88 __extension__ GLubyte Operand
:3; /**< OPR_x */
90 GLubyte Source
; /**< SRC_x */
91 GLubyte Operand
; /**< OPR_x */
96 GLuint nr_enabled_units
:8;
97 GLuint enabled_units
:8;
98 GLuint separate_specular
:1;
100 GLuint fog_mode
:2; /**< FOG_x */
101 GLuint inputs_available
:12;
102 GLuint num_draw_buffers
:4;
104 /* NOTE: This array of structs must be last! (see "keySize" below) */
107 GLuint source_index
:3; /**< TEXTURE_x_INDEX */
109 GLuint ScaleShiftRGB
:2;
110 GLuint ScaleShiftA
:2;
112 GLuint NumArgsRGB
:3; /**< up to MAX_COMBINER_TERMS */
113 GLuint ModeRGB
:5; /**< MODE_x */
115 GLuint NumArgsA
:3; /**< up to MAX_COMBINER_TERMS */
116 GLuint ModeA
:5; /**< MODE_x */
118 GLuint texture_cyl_wrap
:1; /**< For gallium test/debug only */
120 struct mode_opt OptRGB
[MAX_COMBINER_TERMS
];
121 struct mode_opt OptA
[MAX_COMBINER_TERMS
];
122 } unit
[MAX_TEXTURE_UNITS
];
128 #define FOG_UNKNOWN 3
130 static GLuint
translate_fog_mode( GLenum mode
)
133 case GL_LINEAR
: return FOG_LINEAR
;
134 case GL_EXP
: return FOG_EXP
;
135 case GL_EXP2
: return FOG_EXP2
;
136 default: return FOG_UNKNOWN
;
140 #define OPR_SRC_COLOR 0
141 #define OPR_ONE_MINUS_SRC_COLOR 1
142 #define OPR_SRC_ALPHA 2
143 #define OPR_ONE_MINUS_SRC_ALPHA 3
146 #define OPR_UNKNOWN 7
148 static GLuint
translate_operand( GLenum operand
)
151 case GL_SRC_COLOR
: return OPR_SRC_COLOR
;
152 case GL_ONE_MINUS_SRC_COLOR
: return OPR_ONE_MINUS_SRC_COLOR
;
153 case GL_SRC_ALPHA
: return OPR_SRC_ALPHA
;
154 case GL_ONE_MINUS_SRC_ALPHA
: return OPR_ONE_MINUS_SRC_ALPHA
;
155 case GL_ZERO
: return OPR_ZERO
;
156 case GL_ONE
: return OPR_ONE
;
163 #define SRC_TEXTURE 0
164 #define SRC_TEXTURE0 1
165 #define SRC_TEXTURE1 2
166 #define SRC_TEXTURE2 3
167 #define SRC_TEXTURE3 4
168 #define SRC_TEXTURE4 5
169 #define SRC_TEXTURE5 6
170 #define SRC_TEXTURE6 7
171 #define SRC_TEXTURE7 8
172 #define SRC_CONSTANT 9
173 #define SRC_PRIMARY_COLOR 10
174 #define SRC_PREVIOUS 11
176 #define SRC_UNKNOWN 15
178 static GLuint
translate_source( GLenum src
)
181 case GL_TEXTURE
: return SRC_TEXTURE
;
189 case GL_TEXTURE7
: return SRC_TEXTURE0
+ (src
- GL_TEXTURE0
);
190 case GL_CONSTANT
: return SRC_CONSTANT
;
191 case GL_PRIMARY_COLOR
: return SRC_PRIMARY_COLOR
;
192 case GL_PREVIOUS
: return SRC_PREVIOUS
;
201 #define MODE_REPLACE 0 /* r = a0 */
202 #define MODE_MODULATE 1 /* r = a0 * a1 */
203 #define MODE_ADD 2 /* r = a0 + a1 */
204 #define MODE_ADD_SIGNED 3 /* r = a0 + a1 - 0.5 */
205 #define MODE_INTERPOLATE 4 /* r = a0 * a2 + a1 * (1 - a2) */
206 #define MODE_SUBTRACT 5 /* r = a0 - a1 */
207 #define MODE_DOT3_RGB 6 /* r = a0 . a1 */
208 #define MODE_DOT3_RGB_EXT 7 /* r = a0 . a1 */
209 #define MODE_DOT3_RGBA 8 /* r = a0 . a1 */
210 #define MODE_DOT3_RGBA_EXT 9 /* r = a0 . a1 */
211 #define MODE_MODULATE_ADD_ATI 10 /* r = a0 * a2 + a1 */
212 #define MODE_MODULATE_SIGNED_ADD_ATI 11 /* r = a0 * a2 + a1 - 0.5 */
213 #define MODE_MODULATE_SUBTRACT_ATI 12 /* r = a0 * a2 - a1 */
214 #define MODE_ADD_PRODUCTS 13 /* r = a0 * a1 + a2 * a3 */
215 #define MODE_ADD_PRODUCTS_SIGNED 14 /* r = a0 * a1 + a2 * a3 - 0.5 */
216 #define MODE_BUMP_ENVMAP_ATI 15 /* special */
217 #define MODE_UNKNOWN 16
220 * Translate GL combiner state into a MODE_x value
222 static GLuint
translate_mode( GLenum envMode
, GLenum mode
)
225 case GL_REPLACE
: return MODE_REPLACE
;
226 case GL_MODULATE
: return MODE_MODULATE
;
228 if (envMode
== GL_COMBINE4_NV
)
229 return MODE_ADD_PRODUCTS
;
233 if (envMode
== GL_COMBINE4_NV
)
234 return MODE_ADD_PRODUCTS_SIGNED
;
236 return MODE_ADD_SIGNED
;
237 case GL_INTERPOLATE
: return MODE_INTERPOLATE
;
238 case GL_SUBTRACT
: return MODE_SUBTRACT
;
239 case GL_DOT3_RGB
: return MODE_DOT3_RGB
;
240 case GL_DOT3_RGB_EXT
: return MODE_DOT3_RGB_EXT
;
241 case GL_DOT3_RGBA
: return MODE_DOT3_RGBA
;
242 case GL_DOT3_RGBA_EXT
: return MODE_DOT3_RGBA_EXT
;
243 case GL_MODULATE_ADD_ATI
: return MODE_MODULATE_ADD_ATI
;
244 case GL_MODULATE_SIGNED_ADD_ATI
: return MODE_MODULATE_SIGNED_ADD_ATI
;
245 case GL_MODULATE_SUBTRACT_ATI
: return MODE_MODULATE_SUBTRACT_ATI
;
246 case GL_BUMP_ENVMAP_ATI
: return MODE_BUMP_ENVMAP_ATI
;
255 * Do we need to clamp the results of the given texture env/combine mode?
256 * If the inputs to the mode are in [0,1] we don't always have to clamp
260 need_saturate( GLuint mode
)
265 case MODE_INTERPOLATE
:
268 case MODE_ADD_SIGNED
:
271 case MODE_DOT3_RGB_EXT
:
273 case MODE_DOT3_RGBA_EXT
:
274 case MODE_MODULATE_ADD_ATI
:
275 case MODE_MODULATE_SIGNED_ADD_ATI
:
276 case MODE_MODULATE_SUBTRACT_ATI
:
277 case MODE_ADD_PRODUCTS
:
278 case MODE_ADD_PRODUCTS_SIGNED
:
279 case MODE_BUMP_ENVMAP_ATI
:
290 * Translate TEXTURE_x_BIT to TEXTURE_x_INDEX.
292 static GLuint
translate_tex_src_bit( GLbitfield bit
)
295 return _mesa_ffs(bit
) - 1;
299 #define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0)
300 #define VERT_RESULT_TEX_ANY (0xff << VERT_RESULT_TEX0)
303 * Identify all possible varying inputs. The fragment program will
304 * never reference non-varying inputs, but will track them via state
307 * This function figures out all the inputs that the fragment program
308 * has access to. The bitmask is later reduced to just those which
309 * are actually referenced.
311 static GLbitfield
get_fp_input_mask( struct gl_context
*ctx
)
314 const GLboolean vertexShader
=
315 (ctx
->Shader
.CurrentVertexProgram
&&
316 ctx
->Shader
.CurrentVertexProgram
->LinkStatus
&&
317 ctx
->Shader
.CurrentVertexProgram
->VertexProgram
);
318 const GLboolean vertexProgram
= ctx
->VertexProgram
._Enabled
;
319 GLbitfield fp_inputs
= 0x0;
321 if (ctx
->VertexProgram
._Overriden
) {
322 /* Somebody's messing with the vertex program and we don't have
323 * a clue what's happening. Assume that it could be producing
324 * all possible outputs.
328 else if (ctx
->RenderMode
== GL_FEEDBACK
) {
329 /* _NEW_RENDERMODE */
330 fp_inputs
= (FRAG_BIT_COL0
| FRAG_BIT_TEX0
);
332 else if (!(vertexProgram
|| vertexShader
) ||
333 !ctx
->VertexProgram
._Current
) {
334 /* Fixed function vertex logic */
336 GLbitfield varying_inputs
= ctx
->varying_vp_inputs
;
338 /* These get generated in the setup routine regardless of the
342 if (ctx
->Point
.PointSprite
)
343 varying_inputs
|= FRAG_BITS_TEX_ANY
;
345 /* First look at what values may be computed by the generated
349 if (ctx
->Light
.Enabled
) {
350 fp_inputs
|= FRAG_BIT_COL0
;
352 if (texenv_doing_secondary_color(ctx
))
353 fp_inputs
|= FRAG_BIT_COL1
;
357 fp_inputs
|= (ctx
->Texture
._TexGenEnabled
|
358 ctx
->Texture
._TexMatEnabled
) << FRAG_ATTRIB_TEX0
;
360 /* Then look at what might be varying as a result of enabled
363 if (varying_inputs
& VERT_BIT_COLOR0
)
364 fp_inputs
|= FRAG_BIT_COL0
;
365 if (varying_inputs
& VERT_BIT_COLOR1
)
366 fp_inputs
|= FRAG_BIT_COL1
;
368 fp_inputs
|= (((varying_inputs
& VERT_BIT_TEX_ANY
) >> VERT_ATTRIB_TEX0
)
369 << FRAG_ATTRIB_TEX0
);
373 /* calculate from vp->outputs */
374 struct gl_vertex_program
*vprog
;
375 GLbitfield64 vp_outputs
;
377 /* Choose GLSL vertex shader over ARB vertex program. Need this
378 * since vertex shader state validation comes after fragment state
379 * validation (see additional comments in state.c).
382 vprog
= ctx
->Shader
.CurrentVertexProgram
->VertexProgram
;
384 vprog
= ctx
->VertexProgram
.Current
;
386 vp_outputs
= vprog
->Base
.OutputsWritten
;
388 /* These get generated in the setup routine regardless of the
392 if (ctx
->Point
.PointSprite
)
393 vp_outputs
|= FRAG_BITS_TEX_ANY
;
395 if (vp_outputs
& (1 << VERT_RESULT_COL0
))
396 fp_inputs
|= FRAG_BIT_COL0
;
397 if (vp_outputs
& (1 << VERT_RESULT_COL1
))
398 fp_inputs
|= FRAG_BIT_COL1
;
400 fp_inputs
|= (((vp_outputs
& VERT_RESULT_TEX_ANY
) >> VERT_RESULT_TEX0
)
401 << FRAG_ATTRIB_TEX0
);
409 * Examine current texture environment state and generate a unique
410 * key to identify it.
412 static GLuint
make_state_key( struct gl_context
*ctx
, struct state_key
*key
)
415 GLbitfield inputs_referenced
= FRAG_BIT_COL0
;
416 const GLbitfield inputs_available
= get_fp_input_mask( ctx
);
419 memset(key
, 0, sizeof(*key
));
422 for (i
= 0; i
< ctx
->Const
.MaxTextureUnits
; i
++) {
423 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
424 const struct gl_texture_object
*texObj
= texUnit
->_Current
;
425 const struct gl_tex_env_combine_state
*comb
= texUnit
->_CurrentCombine
;
428 if (!texUnit
->_ReallyEnabled
|| !texUnit
->Enabled
)
431 format
= texObj
->Image
[0][texObj
->BaseLevel
]->_BaseFormat
;
433 key
->unit
[i
].enabled
= 1;
434 key
->enabled_units
|= (1<<i
);
435 key
->nr_enabled_units
= i
+ 1;
436 inputs_referenced
|= FRAG_BIT_TEX(i
);
438 key
->unit
[i
].source_index
=
439 translate_tex_src_bit(texUnit
->_ReallyEnabled
);
441 key
->unit
[i
].shadow
= ((texObj
->CompareMode
== GL_COMPARE_R_TO_TEXTURE
) &&
442 ((format
== GL_DEPTH_COMPONENT
) ||
443 (format
== GL_DEPTH_STENCIL_EXT
)));
445 key
->unit
[i
].NumArgsRGB
= comb
->_NumArgsRGB
;
446 key
->unit
[i
].NumArgsA
= comb
->_NumArgsA
;
448 key
->unit
[i
].ModeRGB
=
449 translate_mode(texUnit
->EnvMode
, comb
->ModeRGB
);
451 translate_mode(texUnit
->EnvMode
, comb
->ModeA
);
453 key
->unit
[i
].ScaleShiftRGB
= comb
->ScaleShiftRGB
;
454 key
->unit
[i
].ScaleShiftA
= comb
->ScaleShiftA
;
456 for (j
= 0; j
< MAX_COMBINER_TERMS
; j
++) {
457 key
->unit
[i
].OptRGB
[j
].Operand
= translate_operand(comb
->OperandRGB
[j
]);
458 key
->unit
[i
].OptA
[j
].Operand
= translate_operand(comb
->OperandA
[j
]);
459 key
->unit
[i
].OptRGB
[j
].Source
= translate_source(comb
->SourceRGB
[j
]);
460 key
->unit
[i
].OptA
[j
].Source
= translate_source(comb
->SourceA
[j
]);
463 if (key
->unit
[i
].ModeRGB
== MODE_BUMP_ENVMAP_ATI
) {
464 /* requires some special translation */
465 key
->unit
[i
].NumArgsRGB
= 2;
466 key
->unit
[i
].ScaleShiftRGB
= 0;
467 key
->unit
[i
].OptRGB
[0].Operand
= OPR_SRC_COLOR
;
468 key
->unit
[i
].OptRGB
[0].Source
= SRC_TEXTURE
;
469 key
->unit
[i
].OptRGB
[1].Operand
= OPR_SRC_COLOR
;
470 key
->unit
[i
].OptRGB
[1].Source
= texUnit
->BumpTarget
- GL_TEXTURE0
+ SRC_TEXTURE0
;
473 /* this is a back-door for enabling cylindrical texture wrap mode */
474 if (texObj
->Priority
== 0.125)
475 key
->unit
[i
].texture_cyl_wrap
= 1;
478 /* _NEW_LIGHT | _NEW_FOG */
479 if (texenv_doing_secondary_color(ctx
)) {
480 key
->separate_specular
= 1;
481 inputs_referenced
|= FRAG_BIT_COL1
;
485 if (ctx
->Fog
.Enabled
) {
486 key
->fog_enabled
= 1;
487 key
->fog_mode
= translate_fog_mode(ctx
->Fog
.Mode
);
488 inputs_referenced
|= FRAG_BIT_FOGC
; /* maybe */
492 key
->num_draw_buffers
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
494 key
->inputs_available
= (inputs_available
& inputs_referenced
);
496 /* compute size of state key, ignoring unused texture units */
497 keySize
= sizeof(*key
) - sizeof(key
->unit
)
498 + key
->nr_enabled_units
* sizeof(key
->unit
[0]);
505 * Use uregs to represent registers internally, translate to Mesa's
506 * expected formats on emit.
508 * NOTE: These are passed by value extensively in this file rather
509 * than as usual by pointer reference. If this disturbs you, try
510 * remembering they are just 32bits in size.
512 * GCC is smart enough to deal with these dword-sized structures in
513 * much the same way as if I had defined them as dwords and was using
514 * macros to access and set the fields. This is much nicer and easier
525 static const struct ureg undef
= {
534 /** State used to build the fragment program:
536 struct texenv_fragment_program
{
537 struct gl_fragment_program
*program
;
538 struct state_key
*state
;
540 GLbitfield alu_temps
; /**< Track texture indirections, see spec. */
541 GLbitfield temps_output
; /**< Track texture indirections, see spec. */
542 GLbitfield temp_in_use
; /**< Tracks temporary regs which are in use. */
545 struct ureg src_texture
[MAX_TEXTURE_COORD_UNITS
];
546 /* Reg containing each texture unit's sampled texture color,
550 struct ureg texcoord_tex
[MAX_TEXTURE_COORD_UNITS
];
551 /* Reg containing texcoord for a texture unit,
552 * needed for bump mapping, else undef.
555 struct ureg src_previous
; /**< Reg containing color from previous
556 * stage. May need to be decl'd.
559 GLuint last_tex_stage
; /**< Number of last enabled texture unit */
568 static struct ureg
make_ureg(GLuint file
, GLuint idx
)
574 reg
.swz
= SWIZZLE_NOOP
;
579 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
581 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
584 GET_SWZ(reg
.swz
, w
));
589 static struct ureg
swizzle1( struct ureg reg
, int x
)
591 return swizzle(reg
, x
, x
, x
, x
);
594 static struct ureg
negate( struct ureg reg
)
600 static GLboolean
is_undef( struct ureg reg
)
602 return reg
.file
== PROGRAM_UNDEFINED
;
606 static struct ureg
get_temp( struct texenv_fragment_program
*p
)
610 /* First try and reuse temps which have been used already:
612 bit
= _mesa_ffs( ~p
->temp_in_use
& p
->alu_temps
);
614 /* Then any unused temporary:
617 bit
= _mesa_ffs( ~p
->temp_in_use
);
620 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
624 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
625 p
->program
->Base
.NumTemporaries
= bit
;
627 p
->temp_in_use
|= 1<<(bit
-1);
628 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
631 static struct ureg
get_tex_temp( struct texenv_fragment_program
*p
)
635 /* First try to find available temp not previously used (to avoid
636 * starting a new texture indirection). According to the spec, the
637 * ~p->temps_output isn't necessary, but will keep it there for
640 bit
= _mesa_ffs( ~p
->temp_in_use
& ~p
->alu_temps
& ~p
->temps_output
);
642 /* Then any unused temporary:
645 bit
= _mesa_ffs( ~p
->temp_in_use
);
648 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
652 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
653 p
->program
->Base
.NumTemporaries
= bit
;
655 p
->temp_in_use
|= 1<<(bit
-1);
656 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
660 /** Mark a temp reg as being no longer allocatable. */
661 static void reserve_temp( struct texenv_fragment_program
*p
, struct ureg r
)
663 if (r
.file
== PROGRAM_TEMPORARY
)
664 p
->temps_output
|= (1 << r
.idx
);
668 static void release_temps(struct gl_context
*ctx
, struct texenv_fragment_program
*p
)
670 GLuint max_temp
= ctx
->Const
.FragmentProgram
.MaxTemps
;
672 /* KW: To support tex_env_crossbar, don't release the registers in
675 if (max_temp
>= sizeof(int) * 8)
676 p
->temp_in_use
= p
->temps_output
;
678 p
->temp_in_use
= ~((1<<max_temp
)-1) | p
->temps_output
;
682 static struct ureg
register_param5( struct texenv_fragment_program
*p
,
689 gl_state_index tokens
[STATE_LENGTH
];
696 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
697 return make_ureg(PROGRAM_STATE_VAR
, idx
);
701 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
702 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
703 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
704 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
706 static GLuint
frag_to_vert_attrib( GLuint attrib
)
709 case FRAG_ATTRIB_COL0
: return VERT_ATTRIB_COLOR0
;
710 case FRAG_ATTRIB_COL1
: return VERT_ATTRIB_COLOR1
;
712 assert(attrib
>= FRAG_ATTRIB_TEX0
);
713 assert(attrib
<= FRAG_ATTRIB_TEX7
);
714 return attrib
- FRAG_ATTRIB_TEX0
+ VERT_ATTRIB_TEX0
;
719 static struct ureg
register_input( struct texenv_fragment_program
*p
, GLuint input
)
721 if (p
->state
->inputs_available
& (1<<input
)) {
722 p
->program
->Base
.InputsRead
|= (1 << input
);
723 return make_ureg(PROGRAM_INPUT
, input
);
726 GLuint idx
= frag_to_vert_attrib( input
);
727 return register_param3( p
, STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, idx
);
732 static void emit_arg( struct prog_src_register
*reg
,
735 reg
->File
= ureg
.file
;
736 reg
->Index
= ureg
.idx
;
737 reg
->Swizzle
= ureg
.swz
;
738 reg
->Negate
= ureg
.negatebase
? NEGATE_XYZW
: NEGATE_NONE
;
742 static void emit_dst( struct prog_dst_register
*dst
,
743 struct ureg ureg
, GLuint mask
)
745 dst
->File
= ureg
.file
;
746 dst
->Index
= ureg
.idx
;
747 dst
->WriteMask
= mask
;
748 dst
->CondMask
= COND_TR
; /* always pass cond test */
749 dst
->CondSwizzle
= SWIZZLE_NOOP
;
752 static struct prog_instruction
*
753 emit_op(struct texenv_fragment_program
*p
,
762 const GLuint nr
= p
->program
->Base
.NumInstructions
++;
763 struct prog_instruction
*inst
= &p
->program
->Base
.Instructions
[nr
];
765 assert(nr
< MAX_INSTRUCTIONS
);
767 _mesa_init_instructions(inst
, 1);
770 emit_arg( &inst
->SrcReg
[0], src0
);
771 emit_arg( &inst
->SrcReg
[1], src1
);
772 emit_arg( &inst
->SrcReg
[2], src2
);
774 inst
->SaturateMode
= saturate
? SATURATE_ZERO_ONE
: SATURATE_OFF
;
776 emit_dst( &inst
->DstReg
, dest
, mask
);
779 /* Accounting for indirection tracking:
781 if (dest
.file
== PROGRAM_TEMPORARY
)
782 p
->temps_output
|= 1 << dest
.idx
;
789 static struct ureg
emit_arith( struct texenv_fragment_program
*p
,
798 emit_op(p
, op
, dest
, mask
, saturate
, src0
, src1
, src2
);
800 /* Accounting for indirection tracking:
802 if (src0
.file
== PROGRAM_TEMPORARY
)
803 p
->alu_temps
|= 1 << src0
.idx
;
805 if (!is_undef(src1
) && src1
.file
== PROGRAM_TEMPORARY
)
806 p
->alu_temps
|= 1 << src1
.idx
;
808 if (!is_undef(src2
) && src2
.file
== PROGRAM_TEMPORARY
)
809 p
->alu_temps
|= 1 << src2
.idx
;
811 if (dest
.file
== PROGRAM_TEMPORARY
)
812 p
->alu_temps
|= 1 << dest
.idx
;
814 p
->program
->Base
.NumAluInstructions
++;
818 static struct ureg
emit_texld( struct texenv_fragment_program
*p
,
827 struct prog_instruction
*inst
= emit_op( p
, op
,
829 GL_FALSE
, /* don't saturate? */
834 inst
->TexSrcTarget
= tex_idx
;
835 inst
->TexSrcUnit
= tex_unit
;
836 inst
->TexShadow
= tex_shadow
;
838 p
->program
->Base
.NumTexInstructions
++;
840 /* Accounting for indirection tracking:
842 reserve_temp(p
, dest
);
845 /* Is this a texture indirection?
847 if ((coord
.file
== PROGRAM_TEMPORARY
&&
848 (p
->temps_output
& (1<<coord
.idx
))) ||
849 (dest
.file
== PROGRAM_TEMPORARY
&&
850 (p
->alu_temps
& (1<<dest
.idx
)))) {
851 p
->program
->Base
.NumTexIndirections
++;
852 p
->temps_output
= 1<<coord
.idx
;
854 assert(0); /* KW: texture env crossbar */
862 static struct ureg
register_const4f( struct texenv_fragment_program
*p
,
875 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
877 r
= make_ureg(PROGRAM_CONSTANT
, idx
);
882 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
883 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
884 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
885 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
888 static struct ureg
get_one( struct texenv_fragment_program
*p
)
890 if (is_undef(p
->one
))
891 p
->one
= register_scalar_const(p
, 1.0);
895 static struct ureg
get_half( struct texenv_fragment_program
*p
)
897 if (is_undef(p
->half
))
898 p
->half
= register_scalar_const(p
, 0.5);
902 static struct ureg
get_zero( struct texenv_fragment_program
*p
)
904 if (is_undef(p
->zero
))
905 p
->zero
= register_scalar_const(p
, 0.0);
910 static void program_error( struct texenv_fragment_program
*p
, const char *msg
)
912 _mesa_problem(NULL
, "%s", msg
);
916 static struct ureg
get_source( struct texenv_fragment_program
*p
,
917 GLuint src
, GLuint unit
)
921 assert(!is_undef(p
->src_texture
[unit
]));
922 return p
->src_texture
[unit
];
932 assert(!is_undef(p
->src_texture
[src
- SRC_TEXTURE0
]));
933 return p
->src_texture
[src
- SRC_TEXTURE0
];
936 return register_param2(p
, STATE_TEXENV_COLOR
, unit
);
938 case SRC_PRIMARY_COLOR
:
939 return register_input(p
, FRAG_ATTRIB_COL0
);
945 if (is_undef(p
->src_previous
))
946 return register_input(p
, FRAG_ATTRIB_COL0
);
948 return p
->src_previous
;
956 static struct ureg
emit_combine_source( struct texenv_fragment_program
*p
,
962 struct ureg arg
, src
, one
;
964 src
= get_source(p
, source
, unit
);
967 case OPR_ONE_MINUS_SRC_COLOR
:
969 * Emit tmp = 1.0 - arg.xyzw
973 return emit_arith( p
, OPCODE_SUB
, arg
, mask
, 0, one
, src
, undef
);
976 if (mask
== WRITEMASK_W
)
979 return swizzle1( src
, SWIZZLE_W
);
980 case OPR_ONE_MINUS_SRC_ALPHA
:
982 * Emit tmp = 1.0 - arg.wwww
986 return emit_arith(p
, OPCODE_SUB
, arg
, mask
, 0,
987 one
, swizzle1(src
, SWIZZLE_W
), undef
);
1001 * Check if the RGB and Alpha sources and operands match for the given
1002 * texture unit's combinder state. When the RGB and A sources and
1003 * operands match, we can emit fewer instructions.
1005 static GLboolean
args_match( const struct state_key
*key
, GLuint unit
)
1007 GLuint i
, numArgs
= key
->unit
[unit
].NumArgsRGB
;
1009 for (i
= 0; i
< numArgs
; i
++) {
1010 if (key
->unit
[unit
].OptA
[i
].Source
!= key
->unit
[unit
].OptRGB
[i
].Source
)
1013 switch (key
->unit
[unit
].OptA
[i
].Operand
) {
1015 switch (key
->unit
[unit
].OptRGB
[i
].Operand
) {
1023 case OPR_ONE_MINUS_SRC_ALPHA
:
1024 switch (key
->unit
[unit
].OptRGB
[i
].Operand
) {
1025 case OPR_ONE_MINUS_SRC_COLOR
:
1026 case OPR_ONE_MINUS_SRC_ALPHA
:
1033 return GL_FALSE
; /* impossible */
1040 static struct ureg
emit_combine( struct texenv_fragment_program
*p
,
1047 const struct mode_opt
*opt
)
1049 struct ureg src
[MAX_COMBINER_TERMS
];
1050 struct ureg tmp
, half
;
1053 assert(nr
<= MAX_COMBINER_TERMS
);
1055 for (i
= 0; i
< nr
; i
++)
1056 src
[i
] = emit_combine_source( p
, mask
, unit
, opt
[i
].Source
, opt
[i
].Operand
);
1060 if (mask
== WRITEMASK_XYZW
&& !saturate
)
1063 return emit_arith( p
, OPCODE_MOV
, dest
, mask
, saturate
, src
[0], undef
, undef
);
1065 return emit_arith( p
, OPCODE_MUL
, dest
, mask
, saturate
,
1066 src
[0], src
[1], undef
);
1068 return emit_arith( p
, OPCODE_ADD
, dest
, mask
, saturate
,
1069 src
[0], src
[1], undef
);
1070 case MODE_ADD_SIGNED
:
1071 /* tmp = arg0 + arg1
1075 tmp
= get_temp( p
);
1076 emit_arith( p
, OPCODE_ADD
, tmp
, mask
, 0, src
[0], src
[1], undef
);
1077 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp
, half
, undef
);
1079 case MODE_INTERPOLATE
:
1080 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
1082 return emit_arith( p
, OPCODE_LRP
, dest
, mask
, saturate
, src
[2], src
[0], src
[1] );
1085 return emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, src
[0], src
[1], undef
);
1087 case MODE_DOT3_RGBA
:
1088 case MODE_DOT3_RGBA_EXT
:
1089 case MODE_DOT3_RGB_EXT
:
1090 case MODE_DOT3_RGB
: {
1091 struct ureg tmp0
= get_temp( p
);
1092 struct ureg tmp1
= get_temp( p
);
1093 struct ureg neg1
= register_scalar_const(p
, -1);
1094 struct ureg two
= register_scalar_const(p
, 2);
1096 /* tmp0 = 2*src0 - 1
1099 * dst = tmp0 dot3 tmp1
1101 emit_arith( p
, OPCODE_MAD
, tmp0
, WRITEMASK_XYZW
, 0,
1104 if (memcmp(&src
[0], &src
[1], sizeof(struct ureg
)) == 0)
1107 emit_arith( p
, OPCODE_MAD
, tmp1
, WRITEMASK_XYZW
, 0,
1109 emit_arith( p
, OPCODE_DP3
, dest
, mask
, saturate
, tmp0
, tmp1
, undef
);
1112 case MODE_MODULATE_ADD_ATI
:
1113 /* Arg0 * Arg2 + Arg1 */
1114 return emit_arith( p
, OPCODE_MAD
, dest
, mask
, saturate
,
1115 src
[0], src
[2], src
[1] );
1116 case MODE_MODULATE_SIGNED_ADD_ATI
: {
1117 /* Arg0 * Arg2 + Arg1 - 0.5 */
1118 struct ureg tmp0
= get_temp(p
);
1120 emit_arith( p
, OPCODE_MAD
, tmp0
, mask
, 0, src
[0], src
[2], src
[1] );
1121 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp0
, half
, undef
);
1124 case MODE_MODULATE_SUBTRACT_ATI
:
1125 /* Arg0 * Arg2 - Arg1 */
1126 emit_arith( p
, OPCODE_MAD
, dest
, mask
, 0, src
[0], src
[2], negate(src
[1]) );
1128 case MODE_ADD_PRODUCTS
:
1129 /* Arg0 * Arg1 + Arg2 * Arg3 */
1131 struct ureg tmp0
= get_temp(p
);
1132 emit_arith( p
, OPCODE_MUL
, tmp0
, mask
, 0, src
[0], src
[1], undef
);
1133 emit_arith( p
, OPCODE_MAD
, dest
, mask
, saturate
, src
[2], src
[3], tmp0
);
1136 case MODE_ADD_PRODUCTS_SIGNED
:
1137 /* Arg0 * Arg1 + Arg2 * Arg3 - 0.5 */
1139 struct ureg tmp0
= get_temp(p
);
1141 emit_arith( p
, OPCODE_MUL
, tmp0
, mask
, 0, src
[0], src
[1], undef
);
1142 emit_arith( p
, OPCODE_MAD
, tmp0
, mask
, 0, src
[2], src
[3], tmp0
);
1143 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp0
, half
, undef
);
1146 case MODE_BUMP_ENVMAP_ATI
:
1147 /* special - not handled here */
1158 * Generate instructions for one texture unit's env/combiner mode.
1161 emit_texenv(struct texenv_fragment_program
*p
, GLuint unit
)
1163 const struct state_key
*key
= p
->state
;
1164 GLboolean rgb_saturate
, alpha_saturate
;
1165 GLuint rgb_shift
, alpha_shift
;
1166 struct ureg out
, dest
;
1168 if (!key
->unit
[unit
].enabled
) {
1169 return get_source(p
, SRC_PREVIOUS
, 0);
1171 if (key
->unit
[unit
].ModeRGB
== MODE_BUMP_ENVMAP_ATI
) {
1172 /* this isn't really a env stage delivering a color and handled elsewhere */
1173 return get_source(p
, SRC_PREVIOUS
, 0);
1176 switch (key
->unit
[unit
].ModeRGB
) {
1177 case MODE_DOT3_RGB_EXT
:
1178 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
1181 case MODE_DOT3_RGBA_EXT
:
1186 rgb_shift
= key
->unit
[unit
].ScaleShiftRGB
;
1187 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
1191 /* If we'll do rgb/alpha shifting don't saturate in emit_combine().
1192 * We don't want to clamp twice.
1195 rgb_saturate
= GL_FALSE
; /* saturate after rgb shift */
1196 else if (need_saturate(key
->unit
[unit
].ModeRGB
))
1197 rgb_saturate
= GL_TRUE
;
1199 rgb_saturate
= GL_FALSE
;
1202 alpha_saturate
= GL_FALSE
; /* saturate after alpha shift */
1203 else if (need_saturate(key
->unit
[unit
].ModeA
))
1204 alpha_saturate
= GL_TRUE
;
1206 alpha_saturate
= GL_FALSE
;
1208 /* If this is the very last calculation (and various other conditions
1209 * are met), emit directly to the color output register. Otherwise,
1210 * emit to a temporary register.
1212 if (key
->separate_specular
||
1213 unit
!= p
->last_tex_stage
||
1215 key
->num_draw_buffers
!= 1 ||
1217 dest
= get_temp( p
);
1219 dest
= make_ureg(PROGRAM_OUTPUT
, FRAG_RESULT_COLOR
);
1221 /* Emit the RGB and A combine ops
1223 if (key
->unit
[unit
].ModeRGB
== key
->unit
[unit
].ModeA
&&
1224 args_match(key
, unit
)) {
1225 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, rgb_saturate
,
1227 key
->unit
[unit
].NumArgsRGB
,
1228 key
->unit
[unit
].ModeRGB
,
1229 key
->unit
[unit
].OptRGB
);
1231 else if (key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA_EXT
||
1232 key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA
) {
1233 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, rgb_saturate
,
1235 key
->unit
[unit
].NumArgsRGB
,
1236 key
->unit
[unit
].ModeRGB
,
1237 key
->unit
[unit
].OptRGB
);
1240 /* Need to do something to stop from re-emitting identical
1241 * argument calculations here:
1243 out
= emit_combine( p
, dest
, WRITEMASK_XYZ
, rgb_saturate
,
1245 key
->unit
[unit
].NumArgsRGB
,
1246 key
->unit
[unit
].ModeRGB
,
1247 key
->unit
[unit
].OptRGB
);
1248 out
= emit_combine( p
, dest
, WRITEMASK_W
, alpha_saturate
,
1250 key
->unit
[unit
].NumArgsA
,
1251 key
->unit
[unit
].ModeA
,
1252 key
->unit
[unit
].OptA
);
1255 /* Deal with the final shift:
1257 if (alpha_shift
|| rgb_shift
) {
1259 GLboolean saturate
= GL_TRUE
; /* always saturate at this point */
1261 if (rgb_shift
== alpha_shift
) {
1262 shift
= register_scalar_const(p
, (GLfloat
)(1<<rgb_shift
));
1265 shift
= register_const4f(p
,
1266 (GLfloat
)(1<<rgb_shift
),
1267 (GLfloat
)(1<<rgb_shift
),
1268 (GLfloat
)(1<<rgb_shift
),
1269 (GLfloat
)(1<<alpha_shift
));
1271 return emit_arith( p
, OPCODE_MUL
, dest
, WRITEMASK_XYZW
,
1272 saturate
, out
, shift
, undef
);
1280 * Generate instruction for getting a texture source term.
1282 static void load_texture( struct texenv_fragment_program
*p
, GLuint unit
)
1284 if (is_undef(p
->src_texture
[unit
])) {
1285 const GLuint texTarget
= p
->state
->unit
[unit
].source_index
;
1286 struct ureg texcoord
;
1287 struct ureg tmp
= get_tex_temp( p
);
1289 if (is_undef(p
->texcoord_tex
[unit
])) {
1290 texcoord
= register_input(p
, FRAG_ATTRIB_TEX0
+unit
);
1293 /* might want to reuse this reg for tex output actually */
1294 texcoord
= p
->texcoord_tex
[unit
];
1297 /* TODO: Use D0_MASK_XY where possible.
1299 if (p
->state
->unit
[unit
].enabled
) {
1300 GLboolean shadow
= GL_FALSE
;
1302 if (p
->state
->unit
[unit
].shadow
) {
1303 p
->program
->Base
.ShadowSamplers
|= 1 << unit
;
1307 p
->src_texture
[unit
] = emit_texld( p
, OPCODE_TXP
,
1308 tmp
, WRITEMASK_XYZW
,
1309 unit
, texTarget
, shadow
,
1312 p
->program
->Base
.SamplersUsed
|= (1 << unit
);
1313 /* This identity mapping should already be in place
1314 * (see _mesa_init_program_struct()) but let's be safe.
1316 p
->program
->Base
.SamplerUnits
[unit
] = unit
;
1319 p
->src_texture
[unit
] = get_zero(p
);
1321 if (p
->state
->unit
[unit
].texture_cyl_wrap
) {
1322 /* set flag which is checked by Mesa->Gallium program translation */
1323 p
->program
->Base
.InputFlags
[0] |= PROG_PARAM_BIT_CYL_WRAP
;
1329 static GLboolean
load_texenv_source( struct texenv_fragment_program
*p
,
1330 GLuint src
, GLuint unit
)
1334 load_texture(p
, unit
);
1345 load_texture(p
, src
- SRC_TEXTURE0
);
1349 /* not a texture src - do nothing */
1358 * Generate instructions for loading all texture source terms.
1361 load_texunit_sources( struct texenv_fragment_program
*p
, GLuint unit
)
1363 const struct state_key
*key
= p
->state
;
1366 for (i
= 0; i
< key
->unit
[unit
].NumArgsRGB
; i
++) {
1367 load_texenv_source( p
, key
->unit
[unit
].OptRGB
[i
].Source
, unit
);
1370 for (i
= 0; i
< key
->unit
[unit
].NumArgsA
; i
++) {
1371 load_texenv_source( p
, key
->unit
[unit
].OptA
[i
].Source
, unit
);
1378 * Generate instructions for loading bump map textures.
1381 load_texunit_bumpmap( struct texenv_fragment_program
*p
, GLuint unit
)
1383 const struct state_key
*key
= p
->state
;
1384 GLuint bumpedUnitNr
= key
->unit
[unit
].OptRGB
[1].Source
- SRC_TEXTURE0
;
1385 struct ureg texcDst
, bumpMapRes
;
1386 struct ureg constdudvcolor
= register_const4f(p
, 0.0, 0.0, 0.0, 1.0);
1387 struct ureg texcSrc
= register_input(p
, FRAG_ATTRIB_TEX0
+ bumpedUnitNr
);
1388 struct ureg rotMat0
= register_param3( p
, STATE_INTERNAL
, STATE_ROT_MATRIX_0
, unit
);
1389 struct ureg rotMat1
= register_param3( p
, STATE_INTERNAL
, STATE_ROT_MATRIX_1
, unit
);
1391 load_texenv_source( p
, unit
+ SRC_TEXTURE0
, unit
);
1393 bumpMapRes
= get_source(p
, key
->unit
[unit
].OptRGB
[0].Source
, unit
);
1394 texcDst
= get_tex_temp( p
);
1395 p
->texcoord_tex
[bumpedUnitNr
] = texcDst
;
1397 /* Apply rot matrix and add coords to be available in next phase.
1398 * dest = (Arg0.xxxx * rotMat0 + Arg1) + (Arg0.yyyy * rotMat1)
1399 * note only 2 coords are affected the rest are left unchanged (mul by 0)
1401 emit_arith( p
, OPCODE_MAD
, texcDst
, WRITEMASK_XYZW
, 0,
1402 swizzle1(bumpMapRes
, SWIZZLE_X
), rotMat0
, texcSrc
);
1403 emit_arith( p
, OPCODE_MAD
, texcDst
, WRITEMASK_XYZW
, 0,
1404 swizzle1(bumpMapRes
, SWIZZLE_Y
), rotMat1
, texcDst
);
1406 /* Move 0,0,0,1 into bumpmap src if someone (crossbar) is foolish
1407 * enough to access this later, should optimize away.
1409 emit_arith( p
, OPCODE_MOV
, bumpMapRes
, WRITEMASK_XYZW
, 0,
1410 constdudvcolor
, undef
, undef
);
1416 * Generate a new fragment program which implements the context's
1417 * current texture env/combine mode.
1420 create_new_program(struct gl_context
*ctx
, struct state_key
*key
,
1421 struct gl_fragment_program
*program
)
1423 struct prog_instruction instBuffer
[MAX_INSTRUCTIONS
];
1424 struct texenv_fragment_program p
;
1426 struct ureg cf
, out
;
1429 memset(&p
, 0, sizeof(p
));
1431 p
.program
= program
;
1433 /* During code generation, use locally-allocated instruction buffer,
1434 * then alloc dynamic storage below.
1436 p
.program
->Base
.Instructions
= instBuffer
;
1437 p
.program
->Base
.Target
= GL_FRAGMENT_PROGRAM_ARB
;
1438 p
.program
->Base
.String
= NULL
;
1439 p
.program
->Base
.NumTexIndirections
= 1; /* is this right? */
1440 p
.program
->Base
.NumTexInstructions
= 0;
1441 p
.program
->Base
.NumAluInstructions
= 0;
1442 p
.program
->Base
.NumInstructions
= 0;
1443 p
.program
->Base
.NumTemporaries
= 0;
1444 p
.program
->Base
.NumParameters
= 0;
1445 p
.program
->Base
.NumAttributes
= 0;
1446 p
.program
->Base
.NumAddressRegs
= 0;
1447 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1448 p
.program
->Base
.InputsRead
= 0x0;
1450 if (key
->num_draw_buffers
== 1)
1451 p
.program
->Base
.OutputsWritten
= 1 << FRAG_RESULT_COLOR
;
1453 for (i
= 0; i
< key
->num_draw_buffers
; i
++)
1454 p
.program
->Base
.OutputsWritten
|= (1 << (FRAG_RESULT_DATA0
+ i
));
1457 for (unit
= 0; unit
< ctx
->Const
.MaxTextureUnits
; unit
++) {
1458 p
.src_texture
[unit
] = undef
;
1459 p
.texcoord_tex
[unit
] = undef
;
1462 p
.src_previous
= undef
;
1467 p
.last_tex_stage
= 0;
1468 release_temps(ctx
, &p
);
1470 if (key
->enabled_units
) {
1471 GLboolean needbumpstage
= GL_FALSE
;
1473 /* Zeroth pass - bump map textures first */
1474 for (unit
= 0; unit
< key
->nr_enabled_units
; unit
++)
1475 if (key
->unit
[unit
].enabled
&&
1476 key
->unit
[unit
].ModeRGB
== MODE_BUMP_ENVMAP_ATI
) {
1477 needbumpstage
= GL_TRUE
;
1478 load_texunit_bumpmap( &p
, unit
);
1481 p
.program
->Base
.NumTexIndirections
++;
1483 /* First pass - to support texture_env_crossbar, first identify
1484 * all referenced texture sources and emit texld instructions
1487 for (unit
= 0; unit
< key
->nr_enabled_units
; unit
++)
1488 if (key
->unit
[unit
].enabled
) {
1489 load_texunit_sources( &p
, unit
);
1490 p
.last_tex_stage
= unit
;
1493 /* Second pass - emit combine instructions to build final color:
1495 for (unit
= 0; unit
< key
->nr_enabled_units
; unit
++)
1496 if (key
->unit
[unit
].enabled
) {
1497 p
.src_previous
= emit_texenv( &p
, unit
);
1498 reserve_temp(&p
, p
.src_previous
); /* don't re-use this temp reg */
1499 release_temps(ctx
, &p
); /* release all temps */
1503 cf
= get_source( &p
, SRC_PREVIOUS
, 0 );
1505 for (i
= 0; i
< key
->num_draw_buffers
; i
++) {
1506 if (key
->num_draw_buffers
== 1)
1507 out
= make_ureg( PROGRAM_OUTPUT
, FRAG_RESULT_COLOR
);
1509 out
= make_ureg( PROGRAM_OUTPUT
, FRAG_RESULT_DATA0
+ i
);
1512 if (key
->separate_specular
) {
1513 /* Emit specular add.
1515 struct ureg s
= register_input(&p
, FRAG_ATTRIB_COL1
);
1516 emit_arith( &p
, OPCODE_ADD
, out
, WRITEMASK_XYZ
, 0, cf
, s
, undef
);
1517 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_W
, 0, cf
, undef
, undef
);
1519 else if (memcmp(&cf
, &out
, sizeof(cf
)) != 0) {
1520 /* Will wind up in here if no texture enabled or a couple of
1521 * other scenarios (GL_REPLACE for instance).
1523 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_XYZW
, 0, cf
, undef
, undef
);
1528 emit_arith( &p
, OPCODE_END
, undef
, WRITEMASK_XYZW
, 0, undef
, undef
, undef
);
1530 if (key
->fog_enabled
) {
1531 /* Pull fog mode from struct gl_context, the value in the state key is
1532 * a reduced value and not what is expected in FogOption
1534 p
.program
->FogOption
= ctx
->Fog
.Mode
;
1535 p
.program
->Base
.InputsRead
|= FRAG_BIT_FOGC
;
1538 p
.program
->FogOption
= GL_NONE
;
1541 if (p
.program
->Base
.NumTexIndirections
> ctx
->Const
.FragmentProgram
.MaxTexIndirections
)
1542 program_error(&p
, "Exceeded max nr indirect texture lookups");
1544 if (p
.program
->Base
.NumTexInstructions
> ctx
->Const
.FragmentProgram
.MaxTexInstructions
)
1545 program_error(&p
, "Exceeded max TEX instructions");
1547 if (p
.program
->Base
.NumAluInstructions
> ctx
->Const
.FragmentProgram
.MaxAluInstructions
)
1548 program_error(&p
, "Exceeded max ALU instructions");
1550 ASSERT(p
.program
->Base
.NumInstructions
<= MAX_INSTRUCTIONS
);
1552 /* Allocate final instruction array */
1553 p
.program
->Base
.Instructions
1554 = _mesa_alloc_instructions(p
.program
->Base
.NumInstructions
);
1555 if (!p
.program
->Base
.Instructions
) {
1556 _mesa_error(ctx
, GL_OUT_OF_MEMORY
,
1557 "generating tex env program");
1560 _mesa_copy_instructions(p
.program
->Base
.Instructions
, instBuffer
,
1561 p
.program
->Base
.NumInstructions
);
1563 if (p
.program
->FogOption
) {
1564 _mesa_append_fog_code(ctx
, p
.program
);
1565 p
.program
->FogOption
= GL_NONE
;
1569 /* Notify driver the fragment program has (actually) changed.
1571 if (ctx
->Driver
.ProgramStringNotify
) {
1572 GLboolean ok
= ctx
->Driver
.ProgramStringNotify(ctx
,
1573 GL_FRAGMENT_PROGRAM_ARB
,
1575 /* Driver should be able to handle any texenv programs as long as
1576 * the driver correctly reported max number of texture units correctly,
1580 (void) ok
; /* silence unused var warning */
1584 _mesa_print_program(&p
.program
->Base
);
1591 * Return a fragment program which implements the current
1592 * fixed-function texture, fog and color-sum operations.
1594 struct gl_fragment_program
*
1595 _mesa_get_fixed_func_fragment_program(struct gl_context
*ctx
)
1597 struct gl_fragment_program
*prog
;
1598 struct state_key key
;
1601 keySize
= make_state_key(ctx
, &key
);
1603 prog
= (struct gl_fragment_program
*)
1604 _mesa_search_program_cache(ctx
->FragmentProgram
.Cache
,
1608 prog
= (struct gl_fragment_program
*)
1609 ctx
->Driver
.NewProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
, 0);
1611 create_new_program(ctx
, &key
, prog
);
1613 _mesa_program_cache_insert(ctx
, ctx
->FragmentProgram
.Cache
,
1614 &key
, keySize
, &prog
->Base
);