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 **************************************************************************/
31 #include "program/program.h"
32 #include "program/prog_parameter.h"
33 #include "program/prog_cache.h"
34 #include "program/prog_instruction.h"
35 #include "program/prog_print.h"
36 #include "program/prog_statevars.h"
37 #include "program/programopt.h"
38 #include "texenvprogram.h"
42 * Note on texture units:
44 * The number of texture units supported by fixed-function fragment
45 * processing is MAX_TEXTURE_COORD_UNITS, not MAX_TEXTURE_IMAGE_UNITS.
46 * That's because there's a one-to-one correspondence between texture
47 * coordinates and samplers in fixed-function processing.
49 * Since fixed-function vertex processing is limited to MAX_TEXTURE_COORD_UNITS
50 * sets of texcoords, so is fixed-function fragment processing.
52 * We can safely use ctx->Const.MaxTextureUnits for loop bounds.
56 struct texenvprog_cache_item
60 struct gl_fragment_program
*data
;
61 struct texenvprog_cache_item
*next
;
65 texenv_doing_secondary_color(struct gl_context
*ctx
)
67 if (ctx
->Light
.Enabled
&&
68 (ctx
->Light
.Model
.ColorControl
== GL_SEPARATE_SPECULAR_COLOR
))
71 if (ctx
->Fog
.ColorSumEnabled
)
78 * Up to nine instructions per tex unit, plus fog, specular color.
80 #define MAX_INSTRUCTIONS ((MAX_TEXTURE_COORD_UNITS * 9) + 12)
82 #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
86 __extension__ GLubyte Source
:4; /**< SRC_x */
87 __extension__ GLubyte Operand
:3; /**< OPR_x */
89 GLubyte Source
; /**< SRC_x */
90 GLubyte Operand
; /**< OPR_x */
95 GLuint nr_enabled_units
:8;
96 GLuint enabled_units
:8;
97 GLuint separate_specular
:1;
99 GLuint fog_mode
:2; /**< FOG_x */
100 GLuint inputs_available
:12;
101 GLuint num_draw_buffers
:4;
103 /* NOTE: This array of structs must be last! (see "keySize" below) */
106 GLuint source_index
:3; /**< TEXTURE_x_INDEX */
108 GLuint ScaleShiftRGB
:2;
109 GLuint ScaleShiftA
:2;
111 GLuint NumArgsRGB
:3; /**< up to MAX_COMBINER_TERMS */
112 GLuint ModeRGB
:5; /**< MODE_x */
114 GLuint NumArgsA
:3; /**< up to MAX_COMBINER_TERMS */
115 GLuint ModeA
:5; /**< MODE_x */
117 GLuint texture_cyl_wrap
:1; /**< For gallium test/debug only */
119 struct mode_opt OptRGB
[MAX_COMBINER_TERMS
];
120 struct mode_opt OptA
[MAX_COMBINER_TERMS
];
121 } unit
[MAX_TEXTURE_UNITS
];
127 #define FOG_UNKNOWN 3
129 static GLuint
translate_fog_mode( GLenum mode
)
132 case GL_LINEAR
: return FOG_LINEAR
;
133 case GL_EXP
: return FOG_EXP
;
134 case GL_EXP2
: return FOG_EXP2
;
135 default: return FOG_UNKNOWN
;
139 #define OPR_SRC_COLOR 0
140 #define OPR_ONE_MINUS_SRC_COLOR 1
141 #define OPR_SRC_ALPHA 2
142 #define OPR_ONE_MINUS_SRC_ALPHA 3
145 #define OPR_UNKNOWN 7
147 static GLuint
translate_operand( GLenum operand
)
150 case GL_SRC_COLOR
: return OPR_SRC_COLOR
;
151 case GL_ONE_MINUS_SRC_COLOR
: return OPR_ONE_MINUS_SRC_COLOR
;
152 case GL_SRC_ALPHA
: return OPR_SRC_ALPHA
;
153 case GL_ONE_MINUS_SRC_ALPHA
: return OPR_ONE_MINUS_SRC_ALPHA
;
154 case GL_ZERO
: return OPR_ZERO
;
155 case GL_ONE
: return OPR_ONE
;
162 #define SRC_TEXTURE 0
163 #define SRC_TEXTURE0 1
164 #define SRC_TEXTURE1 2
165 #define SRC_TEXTURE2 3
166 #define SRC_TEXTURE3 4
167 #define SRC_TEXTURE4 5
168 #define SRC_TEXTURE5 6
169 #define SRC_TEXTURE6 7
170 #define SRC_TEXTURE7 8
171 #define SRC_CONSTANT 9
172 #define SRC_PRIMARY_COLOR 10
173 #define SRC_PREVIOUS 11
175 #define SRC_UNKNOWN 15
177 static GLuint
translate_source( GLenum src
)
180 case GL_TEXTURE
: return SRC_TEXTURE
;
188 case GL_TEXTURE7
: return SRC_TEXTURE0
+ (src
- GL_TEXTURE0
);
189 case GL_CONSTANT
: return SRC_CONSTANT
;
190 case GL_PRIMARY_COLOR
: return SRC_PRIMARY_COLOR
;
191 case GL_PREVIOUS
: return SRC_PREVIOUS
;
200 #define MODE_REPLACE 0 /* r = a0 */
201 #define MODE_MODULATE 1 /* r = a0 * a1 */
202 #define MODE_ADD 2 /* r = a0 + a1 */
203 #define MODE_ADD_SIGNED 3 /* r = a0 + a1 - 0.5 */
204 #define MODE_INTERPOLATE 4 /* r = a0 * a2 + a1 * (1 - a2) */
205 #define MODE_SUBTRACT 5 /* r = a0 - a1 */
206 #define MODE_DOT3_RGB 6 /* r = a0 . a1 */
207 #define MODE_DOT3_RGB_EXT 7 /* r = a0 . a1 */
208 #define MODE_DOT3_RGBA 8 /* r = a0 . a1 */
209 #define MODE_DOT3_RGBA_EXT 9 /* r = a0 . a1 */
210 #define MODE_MODULATE_ADD_ATI 10 /* r = a0 * a2 + a1 */
211 #define MODE_MODULATE_SIGNED_ADD_ATI 11 /* r = a0 * a2 + a1 - 0.5 */
212 #define MODE_MODULATE_SUBTRACT_ATI 12 /* r = a0 * a2 - a1 */
213 #define MODE_ADD_PRODUCTS 13 /* r = a0 * a1 + a2 * a3 */
214 #define MODE_ADD_PRODUCTS_SIGNED 14 /* r = a0 * a1 + a2 * a3 - 0.5 */
215 #define MODE_BUMP_ENVMAP_ATI 15 /* special */
216 #define MODE_UNKNOWN 16
219 * Translate GL combiner state into a MODE_x value
221 static GLuint
translate_mode( GLenum envMode
, GLenum mode
)
224 case GL_REPLACE
: return MODE_REPLACE
;
225 case GL_MODULATE
: return MODE_MODULATE
;
227 if (envMode
== GL_COMBINE4_NV
)
228 return MODE_ADD_PRODUCTS
;
232 if (envMode
== GL_COMBINE4_NV
)
233 return MODE_ADD_PRODUCTS_SIGNED
;
235 return MODE_ADD_SIGNED
;
236 case GL_INTERPOLATE
: return MODE_INTERPOLATE
;
237 case GL_SUBTRACT
: return MODE_SUBTRACT
;
238 case GL_DOT3_RGB
: return MODE_DOT3_RGB
;
239 case GL_DOT3_RGB_EXT
: return MODE_DOT3_RGB_EXT
;
240 case GL_DOT3_RGBA
: return MODE_DOT3_RGBA
;
241 case GL_DOT3_RGBA_EXT
: return MODE_DOT3_RGBA_EXT
;
242 case GL_MODULATE_ADD_ATI
: return MODE_MODULATE_ADD_ATI
;
243 case GL_MODULATE_SIGNED_ADD_ATI
: return MODE_MODULATE_SIGNED_ADD_ATI
;
244 case GL_MODULATE_SUBTRACT_ATI
: return MODE_MODULATE_SUBTRACT_ATI
;
245 case GL_BUMP_ENVMAP_ATI
: return MODE_BUMP_ENVMAP_ATI
;
254 * Do we need to clamp the results of the given texture env/combine mode?
255 * If the inputs to the mode are in [0,1] we don't always have to clamp
259 need_saturate( GLuint mode
)
264 case MODE_INTERPOLATE
:
267 case MODE_ADD_SIGNED
:
270 case MODE_DOT3_RGB_EXT
:
272 case MODE_DOT3_RGBA_EXT
:
273 case MODE_MODULATE_ADD_ATI
:
274 case MODE_MODULATE_SIGNED_ADD_ATI
:
275 case MODE_MODULATE_SUBTRACT_ATI
:
276 case MODE_ADD_PRODUCTS
:
277 case MODE_ADD_PRODUCTS_SIGNED
:
278 case MODE_BUMP_ENVMAP_ATI
:
289 * Translate TEXTURE_x_BIT to TEXTURE_x_INDEX.
291 static GLuint
translate_tex_src_bit( GLbitfield bit
)
294 return _mesa_ffs(bit
) - 1;
298 #define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0)
299 #define VERT_RESULT_TEX_ANY (0xff << VERT_RESULT_TEX0)
302 * Identify all possible varying inputs. The fragment program will
303 * never reference non-varying inputs, but will track them via state
306 * This function figures out all the inputs that the fragment program
307 * has access to. The bitmask is later reduced to just those which
308 * are actually referenced.
310 static GLbitfield
get_fp_input_mask( struct gl_context
*ctx
)
313 const GLboolean vertexShader
=
314 (ctx
->Shader
.CurrentVertexProgram
&&
315 ctx
->Shader
.CurrentVertexProgram
->LinkStatus
&&
316 ctx
->Shader
.CurrentVertexProgram
->VertexProgram
);
317 const GLboolean vertexProgram
= ctx
->VertexProgram
._Enabled
;
318 GLbitfield fp_inputs
= 0x0;
320 if (ctx
->VertexProgram
._Overriden
) {
321 /* Somebody's messing with the vertex program and we don't have
322 * a clue what's happening. Assume that it could be producing
323 * all possible outputs.
327 else if (ctx
->RenderMode
== GL_FEEDBACK
) {
328 /* _NEW_RENDERMODE */
329 fp_inputs
= (FRAG_BIT_COL0
| FRAG_BIT_TEX0
);
331 else if (!(vertexProgram
|| vertexShader
) ||
332 !ctx
->VertexProgram
._Current
) {
333 /* Fixed function vertex logic */
335 GLbitfield varying_inputs
= ctx
->varying_vp_inputs
;
337 /* These get generated in the setup routine regardless of the
341 if (ctx
->Point
.PointSprite
)
342 varying_inputs
|= FRAG_BITS_TEX_ANY
;
344 /* First look at what values may be computed by the generated
348 if (ctx
->Light
.Enabled
) {
349 fp_inputs
|= FRAG_BIT_COL0
;
351 if (texenv_doing_secondary_color(ctx
))
352 fp_inputs
|= FRAG_BIT_COL1
;
356 fp_inputs
|= (ctx
->Texture
._TexGenEnabled
|
357 ctx
->Texture
._TexMatEnabled
) << FRAG_ATTRIB_TEX0
;
359 /* Then look at what might be varying as a result of enabled
362 if (varying_inputs
& VERT_BIT_COLOR0
)
363 fp_inputs
|= FRAG_BIT_COL0
;
364 if (varying_inputs
& VERT_BIT_COLOR1
)
365 fp_inputs
|= FRAG_BIT_COL1
;
367 fp_inputs
|= (((varying_inputs
& VERT_BIT_TEX_ANY
) >> VERT_ATTRIB_TEX0
)
368 << FRAG_ATTRIB_TEX0
);
372 /* calculate from vp->outputs */
373 struct gl_vertex_program
*vprog
;
374 GLbitfield64 vp_outputs
;
376 /* Choose GLSL vertex shader over ARB vertex program. Need this
377 * since vertex shader state validation comes after fragment state
378 * validation (see additional comments in state.c).
381 vprog
= ctx
->Shader
.CurrentVertexProgram
->VertexProgram
;
383 vprog
= ctx
->VertexProgram
.Current
;
385 vp_outputs
= vprog
->Base
.OutputsWritten
;
387 /* These get generated in the setup routine regardless of the
391 if (ctx
->Point
.PointSprite
)
392 vp_outputs
|= FRAG_BITS_TEX_ANY
;
394 if (vp_outputs
& (1 << VERT_RESULT_COL0
))
395 fp_inputs
|= FRAG_BIT_COL0
;
396 if (vp_outputs
& (1 << VERT_RESULT_COL1
))
397 fp_inputs
|= FRAG_BIT_COL1
;
399 fp_inputs
|= (((vp_outputs
& VERT_RESULT_TEX_ANY
) >> VERT_RESULT_TEX0
)
400 << FRAG_ATTRIB_TEX0
);
408 * Examine current texture environment state and generate a unique
409 * key to identify it.
411 static GLuint
make_state_key( struct gl_context
*ctx
, struct state_key
*key
)
414 GLbitfield inputs_referenced
= FRAG_BIT_COL0
;
415 const GLbitfield inputs_available
= get_fp_input_mask( ctx
);
418 memset(key
, 0, sizeof(*key
));
421 for (i
= 0; i
< ctx
->Const
.MaxTextureUnits
; i
++) {
422 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
423 const struct gl_texture_object
*texObj
= texUnit
->_Current
;
424 const struct gl_tex_env_combine_state
*comb
= texUnit
->_CurrentCombine
;
427 if (!texUnit
->_ReallyEnabled
|| !texUnit
->Enabled
)
430 format
= texObj
->Image
[0][texObj
->BaseLevel
]->_BaseFormat
;
432 key
->unit
[i
].enabled
= 1;
433 key
->enabled_units
|= (1<<i
);
434 key
->nr_enabled_units
= i
+ 1;
435 inputs_referenced
|= FRAG_BIT_TEX(i
);
437 key
->unit
[i
].source_index
=
438 translate_tex_src_bit(texUnit
->_ReallyEnabled
);
440 key
->unit
[i
].shadow
= ((texObj
->CompareMode
== GL_COMPARE_R_TO_TEXTURE
) &&
441 ((format
== GL_DEPTH_COMPONENT
) ||
442 (format
== GL_DEPTH_STENCIL_EXT
)));
444 key
->unit
[i
].NumArgsRGB
= comb
->_NumArgsRGB
;
445 key
->unit
[i
].NumArgsA
= comb
->_NumArgsA
;
447 key
->unit
[i
].ModeRGB
=
448 translate_mode(texUnit
->EnvMode
, comb
->ModeRGB
);
450 translate_mode(texUnit
->EnvMode
, comb
->ModeA
);
452 key
->unit
[i
].ScaleShiftRGB
= comb
->ScaleShiftRGB
;
453 key
->unit
[i
].ScaleShiftA
= comb
->ScaleShiftA
;
455 for (j
= 0; j
< MAX_COMBINER_TERMS
; j
++) {
456 key
->unit
[i
].OptRGB
[j
].Operand
= translate_operand(comb
->OperandRGB
[j
]);
457 key
->unit
[i
].OptA
[j
].Operand
= translate_operand(comb
->OperandA
[j
]);
458 key
->unit
[i
].OptRGB
[j
].Source
= translate_source(comb
->SourceRGB
[j
]);
459 key
->unit
[i
].OptA
[j
].Source
= translate_source(comb
->SourceA
[j
]);
462 if (key
->unit
[i
].ModeRGB
== MODE_BUMP_ENVMAP_ATI
) {
463 /* requires some special translation */
464 key
->unit
[i
].NumArgsRGB
= 2;
465 key
->unit
[i
].ScaleShiftRGB
= 0;
466 key
->unit
[i
].OptRGB
[0].Operand
= OPR_SRC_COLOR
;
467 key
->unit
[i
].OptRGB
[0].Source
= SRC_TEXTURE
;
468 key
->unit
[i
].OptRGB
[1].Operand
= OPR_SRC_COLOR
;
469 key
->unit
[i
].OptRGB
[1].Source
= texUnit
->BumpTarget
- GL_TEXTURE0
+ SRC_TEXTURE0
;
472 /* this is a back-door for enabling cylindrical texture wrap mode */
473 if (texObj
->Priority
== 0.125)
474 key
->unit
[i
].texture_cyl_wrap
= 1;
477 /* _NEW_LIGHT | _NEW_FOG */
478 if (texenv_doing_secondary_color(ctx
)) {
479 key
->separate_specular
= 1;
480 inputs_referenced
|= FRAG_BIT_COL1
;
484 if (ctx
->Fog
.Enabled
) {
485 key
->fog_enabled
= 1;
486 key
->fog_mode
= translate_fog_mode(ctx
->Fog
.Mode
);
487 inputs_referenced
|= FRAG_BIT_FOGC
; /* maybe */
491 key
->num_draw_buffers
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
493 key
->inputs_available
= (inputs_available
& inputs_referenced
);
495 /* compute size of state key, ignoring unused texture units */
496 keySize
= sizeof(*key
) - sizeof(key
->unit
)
497 + key
->nr_enabled_units
* sizeof(key
->unit
[0]);
504 * Use uregs to represent registers internally, translate to Mesa's
505 * expected formats on emit.
507 * NOTE: These are passed by value extensively in this file rather
508 * than as usual by pointer reference. If this disturbs you, try
509 * remembering they are just 32bits in size.
511 * GCC is smart enough to deal with these dword-sized structures in
512 * much the same way as if I had defined them as dwords and was using
513 * macros to access and set the fields. This is much nicer and easier
524 static const struct ureg undef
= {
533 /** State used to build the fragment program:
535 struct texenv_fragment_program
{
536 struct gl_fragment_program
*program
;
537 struct state_key
*state
;
539 GLbitfield alu_temps
; /**< Track texture indirections, see spec. */
540 GLbitfield temps_output
; /**< Track texture indirections, see spec. */
541 GLbitfield temp_in_use
; /**< Tracks temporary regs which are in use. */
544 struct ureg src_texture
[MAX_TEXTURE_COORD_UNITS
];
545 /* Reg containing each texture unit's sampled texture color,
549 struct ureg texcoord_tex
[MAX_TEXTURE_COORD_UNITS
];
550 /* Reg containing texcoord for a texture unit,
551 * needed for bump mapping, else undef.
554 struct ureg src_previous
; /**< Reg containing color from previous
555 * stage. May need to be decl'd.
558 GLuint last_tex_stage
; /**< Number of last enabled texture unit */
567 static struct ureg
make_ureg(GLuint file
, GLuint idx
)
573 reg
.swz
= SWIZZLE_NOOP
;
578 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
580 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
583 GET_SWZ(reg
.swz
, w
));
588 static struct ureg
swizzle1( struct ureg reg
, int x
)
590 return swizzle(reg
, x
, x
, x
, x
);
593 static struct ureg
negate( struct ureg reg
)
599 static GLboolean
is_undef( struct ureg reg
)
601 return reg
.file
== PROGRAM_UNDEFINED
;
605 static struct ureg
get_temp( struct texenv_fragment_program
*p
)
609 /* First try and reuse temps which have been used already:
611 bit
= _mesa_ffs( ~p
->temp_in_use
& p
->alu_temps
);
613 /* Then any unused temporary:
616 bit
= _mesa_ffs( ~p
->temp_in_use
);
619 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
623 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
624 p
->program
->Base
.NumTemporaries
= bit
;
626 p
->temp_in_use
|= 1<<(bit
-1);
627 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
630 static struct ureg
get_tex_temp( struct texenv_fragment_program
*p
)
634 /* First try to find available temp not previously used (to avoid
635 * starting a new texture indirection). According to the spec, the
636 * ~p->temps_output isn't necessary, but will keep it there for
639 bit
= _mesa_ffs( ~p
->temp_in_use
& ~p
->alu_temps
& ~p
->temps_output
);
641 /* Then any unused temporary:
644 bit
= _mesa_ffs( ~p
->temp_in_use
);
647 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
651 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
652 p
->program
->Base
.NumTemporaries
= bit
;
654 p
->temp_in_use
|= 1<<(bit
-1);
655 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
659 /** Mark a temp reg as being no longer allocatable. */
660 static void reserve_temp( struct texenv_fragment_program
*p
, struct ureg r
)
662 if (r
.file
== PROGRAM_TEMPORARY
)
663 p
->temps_output
|= (1 << r
.idx
);
667 static void release_temps(struct gl_context
*ctx
, struct texenv_fragment_program
*p
)
669 GLuint max_temp
= ctx
->Const
.FragmentProgram
.MaxTemps
;
671 /* KW: To support tex_env_crossbar, don't release the registers in
674 if (max_temp
>= sizeof(int) * 8)
675 p
->temp_in_use
= p
->temps_output
;
677 p
->temp_in_use
= ~((1<<max_temp
)-1) | p
->temps_output
;
681 static struct ureg
register_param5( struct texenv_fragment_program
*p
,
688 gl_state_index tokens
[STATE_LENGTH
];
695 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
696 return make_ureg(PROGRAM_STATE_VAR
, idx
);
700 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
701 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
702 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
703 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
705 static GLuint
frag_to_vert_attrib( GLuint attrib
)
708 case FRAG_ATTRIB_COL0
: return VERT_ATTRIB_COLOR0
;
709 case FRAG_ATTRIB_COL1
: return VERT_ATTRIB_COLOR1
;
711 assert(attrib
>= FRAG_ATTRIB_TEX0
);
712 assert(attrib
<= FRAG_ATTRIB_TEX7
);
713 return attrib
- FRAG_ATTRIB_TEX0
+ VERT_ATTRIB_TEX0
;
718 static struct ureg
register_input( struct texenv_fragment_program
*p
, GLuint input
)
720 if (p
->state
->inputs_available
& (1<<input
)) {
721 p
->program
->Base
.InputsRead
|= (1 << input
);
722 return make_ureg(PROGRAM_INPUT
, input
);
725 GLuint idx
= frag_to_vert_attrib( input
);
726 return register_param3( p
, STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, idx
);
731 static void emit_arg( struct prog_src_register
*reg
,
734 reg
->File
= ureg
.file
;
735 reg
->Index
= ureg
.idx
;
736 reg
->Swizzle
= ureg
.swz
;
737 reg
->Negate
= ureg
.negatebase
? NEGATE_XYZW
: NEGATE_NONE
;
741 static void emit_dst( struct prog_dst_register
*dst
,
742 struct ureg ureg
, GLuint mask
)
744 dst
->File
= ureg
.file
;
745 dst
->Index
= ureg
.idx
;
746 dst
->WriteMask
= mask
;
747 dst
->CondMask
= COND_TR
; /* always pass cond test */
748 dst
->CondSwizzle
= SWIZZLE_NOOP
;
751 static struct prog_instruction
*
752 emit_op(struct texenv_fragment_program
*p
,
761 const GLuint nr
= p
->program
->Base
.NumInstructions
++;
762 struct prog_instruction
*inst
= &p
->program
->Base
.Instructions
[nr
];
764 assert(nr
< MAX_INSTRUCTIONS
);
766 _mesa_init_instructions(inst
, 1);
769 emit_arg( &inst
->SrcReg
[0], src0
);
770 emit_arg( &inst
->SrcReg
[1], src1
);
771 emit_arg( &inst
->SrcReg
[2], src2
);
773 inst
->SaturateMode
= saturate
? SATURATE_ZERO_ONE
: SATURATE_OFF
;
775 emit_dst( &inst
->DstReg
, dest
, mask
);
778 /* Accounting for indirection tracking:
780 if (dest
.file
== PROGRAM_TEMPORARY
)
781 p
->temps_output
|= 1 << dest
.idx
;
788 static struct ureg
emit_arith( struct texenv_fragment_program
*p
,
797 emit_op(p
, op
, dest
, mask
, saturate
, src0
, src1
, src2
);
799 /* Accounting for indirection tracking:
801 if (src0
.file
== PROGRAM_TEMPORARY
)
802 p
->alu_temps
|= 1 << src0
.idx
;
804 if (!is_undef(src1
) && src1
.file
== PROGRAM_TEMPORARY
)
805 p
->alu_temps
|= 1 << src1
.idx
;
807 if (!is_undef(src2
) && src2
.file
== PROGRAM_TEMPORARY
)
808 p
->alu_temps
|= 1 << src2
.idx
;
810 if (dest
.file
== PROGRAM_TEMPORARY
)
811 p
->alu_temps
|= 1 << dest
.idx
;
813 p
->program
->Base
.NumAluInstructions
++;
817 static struct ureg
emit_texld( struct texenv_fragment_program
*p
,
826 struct prog_instruction
*inst
= emit_op( p
, op
,
828 GL_FALSE
, /* don't saturate? */
833 inst
->TexSrcTarget
= tex_idx
;
834 inst
->TexSrcUnit
= tex_unit
;
835 inst
->TexShadow
= tex_shadow
;
837 p
->program
->Base
.NumTexInstructions
++;
839 /* Accounting for indirection tracking:
841 reserve_temp(p
, dest
);
844 /* Is this a texture indirection?
846 if ((coord
.file
== PROGRAM_TEMPORARY
&&
847 (p
->temps_output
& (1<<coord
.idx
))) ||
848 (dest
.file
== PROGRAM_TEMPORARY
&&
849 (p
->alu_temps
& (1<<dest
.idx
)))) {
850 p
->program
->Base
.NumTexIndirections
++;
851 p
->temps_output
= 1<<coord
.idx
;
853 assert(0); /* KW: texture env crossbar */
861 static struct ureg
register_const4f( struct texenv_fragment_program
*p
,
874 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
876 r
= make_ureg(PROGRAM_CONSTANT
, idx
);
881 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
882 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
883 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
884 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
887 static struct ureg
get_one( struct texenv_fragment_program
*p
)
889 if (is_undef(p
->one
))
890 p
->one
= register_scalar_const(p
, 1.0);
894 static struct ureg
get_half( struct texenv_fragment_program
*p
)
896 if (is_undef(p
->half
))
897 p
->half
= register_scalar_const(p
, 0.5);
901 static struct ureg
get_zero( struct texenv_fragment_program
*p
)
903 if (is_undef(p
->zero
))
904 p
->zero
= register_scalar_const(p
, 0.0);
909 static void program_error( struct texenv_fragment_program
*p
, const char *msg
)
911 _mesa_problem(NULL
, "%s", msg
);
915 static struct ureg
get_source( struct texenv_fragment_program
*p
,
916 GLuint src
, GLuint unit
)
920 assert(!is_undef(p
->src_texture
[unit
]));
921 return p
->src_texture
[unit
];
931 assert(!is_undef(p
->src_texture
[src
- SRC_TEXTURE0
]));
932 return p
->src_texture
[src
- SRC_TEXTURE0
];
935 return register_param2(p
, STATE_TEXENV_COLOR
, unit
);
937 case SRC_PRIMARY_COLOR
:
938 return register_input(p
, FRAG_ATTRIB_COL0
);
944 if (is_undef(p
->src_previous
))
945 return register_input(p
, FRAG_ATTRIB_COL0
);
947 return p
->src_previous
;
955 static struct ureg
emit_combine_source( struct texenv_fragment_program
*p
,
961 struct ureg arg
, src
, one
;
963 src
= get_source(p
, source
, unit
);
966 case OPR_ONE_MINUS_SRC_COLOR
:
968 * Emit tmp = 1.0 - arg.xyzw
972 return emit_arith( p
, OPCODE_SUB
, arg
, mask
, 0, one
, src
, undef
);
975 if (mask
== WRITEMASK_W
)
978 return swizzle1( src
, SWIZZLE_W
);
979 case OPR_ONE_MINUS_SRC_ALPHA
:
981 * Emit tmp = 1.0 - arg.wwww
985 return emit_arith(p
, OPCODE_SUB
, arg
, mask
, 0,
986 one
, swizzle1(src
, SWIZZLE_W
), undef
);
1000 * Check if the RGB and Alpha sources and operands match for the given
1001 * texture unit's combinder state. When the RGB and A sources and
1002 * operands match, we can emit fewer instructions.
1004 static GLboolean
args_match( const struct state_key
*key
, GLuint unit
)
1006 GLuint i
, numArgs
= key
->unit
[unit
].NumArgsRGB
;
1008 for (i
= 0; i
< numArgs
; i
++) {
1009 if (key
->unit
[unit
].OptA
[i
].Source
!= key
->unit
[unit
].OptRGB
[i
].Source
)
1012 switch (key
->unit
[unit
].OptA
[i
].Operand
) {
1014 switch (key
->unit
[unit
].OptRGB
[i
].Operand
) {
1022 case OPR_ONE_MINUS_SRC_ALPHA
:
1023 switch (key
->unit
[unit
].OptRGB
[i
].Operand
) {
1024 case OPR_ONE_MINUS_SRC_COLOR
:
1025 case OPR_ONE_MINUS_SRC_ALPHA
:
1032 return GL_FALSE
; /* impossible */
1039 static struct ureg
emit_combine( struct texenv_fragment_program
*p
,
1046 const struct mode_opt
*opt
)
1048 struct ureg src
[MAX_COMBINER_TERMS
];
1049 struct ureg tmp
, half
;
1052 assert(nr
<= MAX_COMBINER_TERMS
);
1054 for (i
= 0; i
< nr
; i
++)
1055 src
[i
] = emit_combine_source( p
, mask
, unit
, opt
[i
].Source
, opt
[i
].Operand
);
1059 if (mask
== WRITEMASK_XYZW
&& !saturate
)
1062 return emit_arith( p
, OPCODE_MOV
, dest
, mask
, saturate
, src
[0], undef
, undef
);
1064 return emit_arith( p
, OPCODE_MUL
, dest
, mask
, saturate
,
1065 src
[0], src
[1], undef
);
1067 return emit_arith( p
, OPCODE_ADD
, dest
, mask
, saturate
,
1068 src
[0], src
[1], undef
);
1069 case MODE_ADD_SIGNED
:
1070 /* tmp = arg0 + arg1
1074 tmp
= get_temp( p
);
1075 emit_arith( p
, OPCODE_ADD
, tmp
, mask
, 0, src
[0], src
[1], undef
);
1076 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp
, half
, undef
);
1078 case MODE_INTERPOLATE
:
1079 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
1081 return emit_arith( p
, OPCODE_LRP
, dest
, mask
, saturate
, src
[2], src
[0], src
[1] );
1084 return emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, src
[0], src
[1], undef
);
1086 case MODE_DOT3_RGBA
:
1087 case MODE_DOT3_RGBA_EXT
:
1088 case MODE_DOT3_RGB_EXT
:
1089 case MODE_DOT3_RGB
: {
1090 struct ureg tmp0
= get_temp( p
);
1091 struct ureg tmp1
= get_temp( p
);
1092 struct ureg neg1
= register_scalar_const(p
, -1);
1093 struct ureg two
= register_scalar_const(p
, 2);
1095 /* tmp0 = 2*src0 - 1
1098 * dst = tmp0 dot3 tmp1
1100 emit_arith( p
, OPCODE_MAD
, tmp0
, WRITEMASK_XYZW
, 0,
1103 if (memcmp(&src
[0], &src
[1], sizeof(struct ureg
)) == 0)
1106 emit_arith( p
, OPCODE_MAD
, tmp1
, WRITEMASK_XYZW
, 0,
1108 emit_arith( p
, OPCODE_DP3
, dest
, mask
, saturate
, tmp0
, tmp1
, undef
);
1111 case MODE_MODULATE_ADD_ATI
:
1112 /* Arg0 * Arg2 + Arg1 */
1113 return emit_arith( p
, OPCODE_MAD
, dest
, mask
, saturate
,
1114 src
[0], src
[2], src
[1] );
1115 case MODE_MODULATE_SIGNED_ADD_ATI
: {
1116 /* Arg0 * Arg2 + Arg1 - 0.5 */
1117 struct ureg tmp0
= get_temp(p
);
1119 emit_arith( p
, OPCODE_MAD
, tmp0
, mask
, 0, src
[0], src
[2], src
[1] );
1120 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp0
, half
, undef
);
1123 case MODE_MODULATE_SUBTRACT_ATI
:
1124 /* Arg0 * Arg2 - Arg1 */
1125 emit_arith( p
, OPCODE_MAD
, dest
, mask
, 0, src
[0], src
[2], negate(src
[1]) );
1127 case MODE_ADD_PRODUCTS
:
1128 /* Arg0 * Arg1 + Arg2 * Arg3 */
1130 struct ureg tmp0
= get_temp(p
);
1131 emit_arith( p
, OPCODE_MUL
, tmp0
, mask
, 0, src
[0], src
[1], undef
);
1132 emit_arith( p
, OPCODE_MAD
, dest
, mask
, saturate
, src
[2], src
[3], tmp0
);
1135 case MODE_ADD_PRODUCTS_SIGNED
:
1136 /* Arg0 * Arg1 + Arg2 * Arg3 - 0.5 */
1138 struct ureg tmp0
= get_temp(p
);
1140 emit_arith( p
, OPCODE_MUL
, tmp0
, mask
, 0, src
[0], src
[1], undef
);
1141 emit_arith( p
, OPCODE_MAD
, tmp0
, mask
, 0, src
[2], src
[3], tmp0
);
1142 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp0
, half
, undef
);
1145 case MODE_BUMP_ENVMAP_ATI
:
1146 /* special - not handled here */
1157 * Generate instructions for one texture unit's env/combiner mode.
1160 emit_texenv(struct texenv_fragment_program
*p
, GLuint unit
)
1162 const struct state_key
*key
= p
->state
;
1163 GLboolean rgb_saturate
, alpha_saturate
;
1164 GLuint rgb_shift
, alpha_shift
;
1165 struct ureg out
, dest
;
1167 if (!key
->unit
[unit
].enabled
) {
1168 return get_source(p
, SRC_PREVIOUS
, 0);
1170 if (key
->unit
[unit
].ModeRGB
== MODE_BUMP_ENVMAP_ATI
) {
1171 /* this isn't really a env stage delivering a color and handled elsewhere */
1172 return get_source(p
, SRC_PREVIOUS
, 0);
1175 switch (key
->unit
[unit
].ModeRGB
) {
1176 case MODE_DOT3_RGB_EXT
:
1177 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
1180 case MODE_DOT3_RGBA_EXT
:
1185 rgb_shift
= key
->unit
[unit
].ScaleShiftRGB
;
1186 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
1190 /* If we'll do rgb/alpha shifting don't saturate in emit_combine().
1191 * We don't want to clamp twice.
1194 rgb_saturate
= GL_FALSE
; /* saturate after rgb shift */
1195 else if (need_saturate(key
->unit
[unit
].ModeRGB
))
1196 rgb_saturate
= GL_TRUE
;
1198 rgb_saturate
= GL_FALSE
;
1201 alpha_saturate
= GL_FALSE
; /* saturate after alpha shift */
1202 else if (need_saturate(key
->unit
[unit
].ModeA
))
1203 alpha_saturate
= GL_TRUE
;
1205 alpha_saturate
= GL_FALSE
;
1207 /* If this is the very last calculation (and various other conditions
1208 * are met), emit directly to the color output register. Otherwise,
1209 * emit to a temporary register.
1211 if (key
->separate_specular
||
1212 unit
!= p
->last_tex_stage
||
1214 key
->num_draw_buffers
!= 1 ||
1216 dest
= get_temp( p
);
1218 dest
= make_ureg(PROGRAM_OUTPUT
, FRAG_RESULT_COLOR
);
1220 /* Emit the RGB and A combine ops
1222 if (key
->unit
[unit
].ModeRGB
== key
->unit
[unit
].ModeA
&&
1223 args_match(key
, unit
)) {
1224 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, rgb_saturate
,
1226 key
->unit
[unit
].NumArgsRGB
,
1227 key
->unit
[unit
].ModeRGB
,
1228 key
->unit
[unit
].OptRGB
);
1230 else if (key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA_EXT
||
1231 key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA
) {
1232 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, rgb_saturate
,
1234 key
->unit
[unit
].NumArgsRGB
,
1235 key
->unit
[unit
].ModeRGB
,
1236 key
->unit
[unit
].OptRGB
);
1239 /* Need to do something to stop from re-emitting identical
1240 * argument calculations here:
1242 out
= emit_combine( p
, dest
, WRITEMASK_XYZ
, rgb_saturate
,
1244 key
->unit
[unit
].NumArgsRGB
,
1245 key
->unit
[unit
].ModeRGB
,
1246 key
->unit
[unit
].OptRGB
);
1247 out
= emit_combine( p
, dest
, WRITEMASK_W
, alpha_saturate
,
1249 key
->unit
[unit
].NumArgsA
,
1250 key
->unit
[unit
].ModeA
,
1251 key
->unit
[unit
].OptA
);
1254 /* Deal with the final shift:
1256 if (alpha_shift
|| rgb_shift
) {
1258 GLboolean saturate
= GL_TRUE
; /* always saturate at this point */
1260 if (rgb_shift
== alpha_shift
) {
1261 shift
= register_scalar_const(p
, (GLfloat
)(1<<rgb_shift
));
1264 shift
= register_const4f(p
,
1265 (GLfloat
)(1<<rgb_shift
),
1266 (GLfloat
)(1<<rgb_shift
),
1267 (GLfloat
)(1<<rgb_shift
),
1268 (GLfloat
)(1<<alpha_shift
));
1270 return emit_arith( p
, OPCODE_MUL
, dest
, WRITEMASK_XYZW
,
1271 saturate
, out
, shift
, undef
);
1279 * Generate instruction for getting a texture source term.
1281 static void load_texture( struct texenv_fragment_program
*p
, GLuint unit
)
1283 if (is_undef(p
->src_texture
[unit
])) {
1284 const GLuint texTarget
= p
->state
->unit
[unit
].source_index
;
1285 struct ureg texcoord
;
1286 struct ureg tmp
= get_tex_temp( p
);
1288 if (is_undef(p
->texcoord_tex
[unit
])) {
1289 texcoord
= register_input(p
, FRAG_ATTRIB_TEX0
+unit
);
1292 /* might want to reuse this reg for tex output actually */
1293 texcoord
= p
->texcoord_tex
[unit
];
1296 /* TODO: Use D0_MASK_XY where possible.
1298 if (p
->state
->unit
[unit
].enabled
) {
1299 GLboolean shadow
= GL_FALSE
;
1301 if (p
->state
->unit
[unit
].shadow
) {
1302 p
->program
->Base
.ShadowSamplers
|= 1 << unit
;
1306 p
->src_texture
[unit
] = emit_texld( p
, OPCODE_TXP
,
1307 tmp
, WRITEMASK_XYZW
,
1308 unit
, texTarget
, shadow
,
1311 p
->program
->Base
.SamplersUsed
|= (1 << unit
);
1312 /* This identity mapping should already be in place
1313 * (see _mesa_init_program_struct()) but let's be safe.
1315 p
->program
->Base
.SamplerUnits
[unit
] = unit
;
1318 p
->src_texture
[unit
] = get_zero(p
);
1320 if (p
->state
->unit
[unit
].texture_cyl_wrap
) {
1321 /* set flag which is checked by Mesa->Gallium program translation */
1322 p
->program
->Base
.InputFlags
[0] |= PROG_PARAM_BIT_CYL_WRAP
;
1328 static GLboolean
load_texenv_source( struct texenv_fragment_program
*p
,
1329 GLuint src
, GLuint unit
)
1333 load_texture(p
, unit
);
1344 load_texture(p
, src
- SRC_TEXTURE0
);
1348 /* not a texture src - do nothing */
1357 * Generate instructions for loading all texture source terms.
1360 load_texunit_sources( struct texenv_fragment_program
*p
, GLuint unit
)
1362 const struct state_key
*key
= p
->state
;
1365 for (i
= 0; i
< key
->unit
[unit
].NumArgsRGB
; i
++) {
1366 load_texenv_source( p
, key
->unit
[unit
].OptRGB
[i
].Source
, unit
);
1369 for (i
= 0; i
< key
->unit
[unit
].NumArgsA
; i
++) {
1370 load_texenv_source( p
, key
->unit
[unit
].OptA
[i
].Source
, unit
);
1377 * Generate instructions for loading bump map textures.
1380 load_texunit_bumpmap( struct texenv_fragment_program
*p
, GLuint unit
)
1382 const struct state_key
*key
= p
->state
;
1383 GLuint bumpedUnitNr
= key
->unit
[unit
].OptRGB
[1].Source
- SRC_TEXTURE0
;
1384 struct ureg texcDst
, bumpMapRes
;
1385 struct ureg constdudvcolor
= register_const4f(p
, 0.0, 0.0, 0.0, 1.0);
1386 struct ureg texcSrc
= register_input(p
, FRAG_ATTRIB_TEX0
+ bumpedUnitNr
);
1387 struct ureg rotMat0
= register_param3( p
, STATE_INTERNAL
, STATE_ROT_MATRIX_0
, unit
);
1388 struct ureg rotMat1
= register_param3( p
, STATE_INTERNAL
, STATE_ROT_MATRIX_1
, unit
);
1390 load_texenv_source( p
, unit
+ SRC_TEXTURE0
, unit
);
1392 bumpMapRes
= get_source(p
, key
->unit
[unit
].OptRGB
[0].Source
, unit
);
1393 texcDst
= get_tex_temp( p
);
1394 p
->texcoord_tex
[bumpedUnitNr
] = texcDst
;
1396 /* Apply rot matrix and add coords to be available in next phase.
1397 * dest = (Arg0.xxxx * rotMat0 + Arg1) + (Arg0.yyyy * rotMat1)
1398 * note only 2 coords are affected the rest are left unchanged (mul by 0)
1400 emit_arith( p
, OPCODE_MAD
, texcDst
, WRITEMASK_XYZW
, 0,
1401 swizzle1(bumpMapRes
, SWIZZLE_X
), rotMat0
, texcSrc
);
1402 emit_arith( p
, OPCODE_MAD
, texcDst
, WRITEMASK_XYZW
, 0,
1403 swizzle1(bumpMapRes
, SWIZZLE_Y
), rotMat1
, texcDst
);
1405 /* Move 0,0,0,1 into bumpmap src if someone (crossbar) is foolish
1406 * enough to access this later, should optimize away.
1408 emit_arith( p
, OPCODE_MOV
, bumpMapRes
, WRITEMASK_XYZW
, 0,
1409 constdudvcolor
, undef
, undef
);
1415 * Generate a new fragment program which implements the context's
1416 * current texture env/combine mode.
1419 create_new_program(struct gl_context
*ctx
, struct state_key
*key
,
1420 struct gl_fragment_program
*program
)
1422 struct prog_instruction instBuffer
[MAX_INSTRUCTIONS
];
1423 struct texenv_fragment_program p
;
1425 struct ureg cf
, out
;
1428 memset(&p
, 0, sizeof(p
));
1430 p
.program
= program
;
1432 /* During code generation, use locally-allocated instruction buffer,
1433 * then alloc dynamic storage below.
1435 p
.program
->Base
.Instructions
= instBuffer
;
1436 p
.program
->Base
.Target
= GL_FRAGMENT_PROGRAM_ARB
;
1437 p
.program
->Base
.String
= NULL
;
1438 p
.program
->Base
.NumTexIndirections
= 1; /* is this right? */
1439 p
.program
->Base
.NumTexInstructions
= 0;
1440 p
.program
->Base
.NumAluInstructions
= 0;
1441 p
.program
->Base
.NumInstructions
= 0;
1442 p
.program
->Base
.NumTemporaries
= 0;
1443 p
.program
->Base
.NumParameters
= 0;
1444 p
.program
->Base
.NumAttributes
= 0;
1445 p
.program
->Base
.NumAddressRegs
= 0;
1446 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1447 p
.program
->Base
.InputsRead
= 0x0;
1449 if (key
->num_draw_buffers
== 1)
1450 p
.program
->Base
.OutputsWritten
= 1 << FRAG_RESULT_COLOR
;
1452 for (i
= 0; i
< key
->num_draw_buffers
; i
++)
1453 p
.program
->Base
.OutputsWritten
|= (1 << (FRAG_RESULT_DATA0
+ i
));
1456 for (unit
= 0; unit
< ctx
->Const
.MaxTextureUnits
; unit
++) {
1457 p
.src_texture
[unit
] = undef
;
1458 p
.texcoord_tex
[unit
] = undef
;
1461 p
.src_previous
= undef
;
1466 p
.last_tex_stage
= 0;
1467 release_temps(ctx
, &p
);
1469 if (key
->enabled_units
) {
1470 GLboolean needbumpstage
= GL_FALSE
;
1472 /* Zeroth pass - bump map textures first */
1473 for (unit
= 0; unit
< key
->nr_enabled_units
; unit
++)
1474 if (key
->unit
[unit
].enabled
&&
1475 key
->unit
[unit
].ModeRGB
== MODE_BUMP_ENVMAP_ATI
) {
1476 needbumpstage
= GL_TRUE
;
1477 load_texunit_bumpmap( &p
, unit
);
1480 p
.program
->Base
.NumTexIndirections
++;
1482 /* First pass - to support texture_env_crossbar, first identify
1483 * all referenced texture sources and emit texld instructions
1486 for (unit
= 0; unit
< key
->nr_enabled_units
; unit
++)
1487 if (key
->unit
[unit
].enabled
) {
1488 load_texunit_sources( &p
, unit
);
1489 p
.last_tex_stage
= unit
;
1492 /* Second pass - emit combine instructions to build final color:
1494 for (unit
= 0; unit
< key
->nr_enabled_units
; unit
++)
1495 if (key
->unit
[unit
].enabled
) {
1496 p
.src_previous
= emit_texenv( &p
, unit
);
1497 reserve_temp(&p
, p
.src_previous
); /* don't re-use this temp reg */
1498 release_temps(ctx
, &p
); /* release all temps */
1502 cf
= get_source( &p
, SRC_PREVIOUS
, 0 );
1504 for (i
= 0; i
< key
->num_draw_buffers
; i
++) {
1505 if (key
->num_draw_buffers
== 1)
1506 out
= make_ureg( PROGRAM_OUTPUT
, FRAG_RESULT_COLOR
);
1508 out
= make_ureg( PROGRAM_OUTPUT
, FRAG_RESULT_DATA0
+ i
);
1511 if (key
->separate_specular
) {
1512 /* Emit specular add.
1514 struct ureg s
= register_input(&p
, FRAG_ATTRIB_COL1
);
1515 emit_arith( &p
, OPCODE_ADD
, out
, WRITEMASK_XYZ
, 0, cf
, s
, undef
);
1516 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_W
, 0, cf
, undef
, undef
);
1518 else if (memcmp(&cf
, &out
, sizeof(cf
)) != 0) {
1519 /* Will wind up in here if no texture enabled or a couple of
1520 * other scenarios (GL_REPLACE for instance).
1522 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_XYZW
, 0, cf
, undef
, undef
);
1527 emit_arith( &p
, OPCODE_END
, undef
, WRITEMASK_XYZW
, 0, undef
, undef
, undef
);
1529 if (key
->fog_enabled
) {
1530 /* Pull fog mode from struct gl_context, the value in the state key is
1531 * a reduced value and not what is expected in FogOption
1533 p
.program
->FogOption
= ctx
->Fog
.Mode
;
1534 p
.program
->Base
.InputsRead
|= FRAG_BIT_FOGC
;
1537 p
.program
->FogOption
= GL_NONE
;
1540 if (p
.program
->Base
.NumTexIndirections
> ctx
->Const
.FragmentProgram
.MaxTexIndirections
)
1541 program_error(&p
, "Exceeded max nr indirect texture lookups");
1543 if (p
.program
->Base
.NumTexInstructions
> ctx
->Const
.FragmentProgram
.MaxTexInstructions
)
1544 program_error(&p
, "Exceeded max TEX instructions");
1546 if (p
.program
->Base
.NumAluInstructions
> ctx
->Const
.FragmentProgram
.MaxAluInstructions
)
1547 program_error(&p
, "Exceeded max ALU instructions");
1549 ASSERT(p
.program
->Base
.NumInstructions
<= MAX_INSTRUCTIONS
);
1551 /* Allocate final instruction array */
1552 p
.program
->Base
.Instructions
1553 = _mesa_alloc_instructions(p
.program
->Base
.NumInstructions
);
1554 if (!p
.program
->Base
.Instructions
) {
1555 _mesa_error(ctx
, GL_OUT_OF_MEMORY
,
1556 "generating tex env program");
1559 _mesa_copy_instructions(p
.program
->Base
.Instructions
, instBuffer
,
1560 p
.program
->Base
.NumInstructions
);
1562 if (p
.program
->FogOption
) {
1563 _mesa_append_fog_code(ctx
, p
.program
);
1564 p
.program
->FogOption
= GL_NONE
;
1568 /* Notify driver the fragment program has (actually) changed.
1570 if (ctx
->Driver
.ProgramStringNotify
) {
1571 GLboolean ok
= ctx
->Driver
.ProgramStringNotify(ctx
,
1572 GL_FRAGMENT_PROGRAM_ARB
,
1574 /* Driver should be able to handle any texenv programs as long as
1575 * the driver correctly reported max number of texture units correctly,
1579 (void) ok
; /* silence unused var warning */
1583 _mesa_print_program(&p
.program
->Base
);
1590 * Return a fragment program which implements the current
1591 * fixed-function texture, fog and color-sum operations.
1593 struct gl_fragment_program
*
1594 _mesa_get_fixed_func_fragment_program(struct gl_context
*ctx
)
1596 struct gl_fragment_program
*prog
;
1597 struct state_key key
;
1600 keySize
= make_state_key(ctx
, &key
);
1602 prog
= (struct gl_fragment_program
*)
1603 _mesa_search_program_cache(ctx
->FragmentProgram
.Cache
,
1607 prog
= (struct gl_fragment_program
*)
1608 ctx
->Driver
.NewProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
, 0);
1610 create_new_program(ctx
, &key
, prog
);
1612 _mesa_program_cache_insert(ctx
, ctx
->FragmentProgram
.Cache
,
1613 &key
, keySize
, &prog
->Base
);