1 /**************************************************************************
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
31 #include "shader/prog_parameter.h"
32 #include "shader/prog_cache.h"
33 #include "shader/prog_instruction.h"
34 #include "shader/prog_print.h"
35 #include "shader/prog_statevars.h"
36 #include "shader/programopt.h"
37 #include "texenvprogram.h"
40 * Up to nine instructions per tex unit, plus fog, specular color.
42 #define MAX_INSTRUCTIONS ((MAX_TEXTURE_UNITS * 9) + 12)
44 #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
52 GLuint nr_enabled_units
:8;
53 GLuint enabled_units
:8;
54 GLuint separate_specular
:1;
57 GLuint inputs_available
:12;
61 GLuint source_index
:3; /* one of TEXTURE_1D/2D/3D/CUBE/RECT_INDEX */
62 GLuint ScaleShiftRGB
:2;
70 struct mode_opt OptRGB
[3];
71 struct mode_opt OptA
[3];
80 static GLuint
translate_fog_mode( GLenum mode
)
83 case GL_LINEAR
: return FOG_LINEAR
;
84 case GL_EXP
: return FOG_EXP
;
85 case GL_EXP2
: return FOG_EXP2
;
86 default: return FOG_UNKNOWN
;
90 #define OPR_SRC_COLOR 0
91 #define OPR_ONE_MINUS_SRC_COLOR 1
92 #define OPR_SRC_ALPHA 2
93 #define OPR_ONE_MINUS_SRC_ALPHA 3
98 static GLuint
translate_operand( GLenum operand
)
101 case GL_SRC_COLOR
: return OPR_SRC_COLOR
;
102 case GL_ONE_MINUS_SRC_COLOR
: return OPR_ONE_MINUS_SRC_COLOR
;
103 case GL_SRC_ALPHA
: return OPR_SRC_ALPHA
;
104 case GL_ONE_MINUS_SRC_ALPHA
: return OPR_ONE_MINUS_SRC_ALPHA
;
105 case GL_ZERO
: return OPR_ZERO
;
106 case GL_ONE
: return OPR_ONE
;
107 default: return OPR_UNKNOWN
;
111 #define SRC_TEXTURE 0
112 #define SRC_TEXTURE0 1
113 #define SRC_TEXTURE1 2
114 #define SRC_TEXTURE2 3
115 #define SRC_TEXTURE3 4
116 #define SRC_TEXTURE4 5
117 #define SRC_TEXTURE5 6
118 #define SRC_TEXTURE6 7
119 #define SRC_TEXTURE7 8
120 #define SRC_CONSTANT 9
121 #define SRC_PRIMARY_COLOR 10
122 #define SRC_PREVIOUS 11
123 #define SRC_UNKNOWN 15
125 static GLuint
translate_source( GLenum src
)
128 case GL_TEXTURE
: return SRC_TEXTURE
;
136 case GL_TEXTURE7
: return SRC_TEXTURE0
+ (src
- GL_TEXTURE0
);
137 case GL_CONSTANT
: return SRC_CONSTANT
;
138 case GL_PRIMARY_COLOR
: return SRC_PRIMARY_COLOR
;
139 case GL_PREVIOUS
: return SRC_PREVIOUS
;
140 default: return SRC_UNKNOWN
;
144 #define MODE_REPLACE 0
145 #define MODE_MODULATE 1
147 #define MODE_ADD_SIGNED 3
148 #define MODE_INTERPOLATE 4
149 #define MODE_SUBTRACT 5
150 #define MODE_DOT3_RGB 6
151 #define MODE_DOT3_RGB_EXT 7
152 #define MODE_DOT3_RGBA 8
153 #define MODE_DOT3_RGBA_EXT 9
154 #define MODE_MODULATE_ADD_ATI 10
155 #define MODE_MODULATE_SIGNED_ADD_ATI 11
156 #define MODE_MODULATE_SUBTRACT_ATI 12
157 #define MODE_UNKNOWN 15
159 static GLuint
translate_mode( GLenum mode
)
162 case GL_REPLACE
: return MODE_REPLACE
;
163 case GL_MODULATE
: return MODE_MODULATE
;
164 case GL_ADD
: return MODE_ADD
;
165 case GL_ADD_SIGNED
: return MODE_ADD_SIGNED
;
166 case GL_INTERPOLATE
: return MODE_INTERPOLATE
;
167 case GL_SUBTRACT
: return MODE_SUBTRACT
;
168 case GL_DOT3_RGB
: return MODE_DOT3_RGB
;
169 case GL_DOT3_RGB_EXT
: return MODE_DOT3_RGB_EXT
;
170 case GL_DOT3_RGBA
: return MODE_DOT3_RGBA
;
171 case GL_DOT3_RGBA_EXT
: return MODE_DOT3_RGBA_EXT
;
172 case GL_MODULATE_ADD_ATI
: return MODE_MODULATE_ADD_ATI
;
173 case GL_MODULATE_SIGNED_ADD_ATI
: return MODE_MODULATE_SIGNED_ADD_ATI
;
174 case GL_MODULATE_SUBTRACT_ATI
: return MODE_MODULATE_SUBTRACT_ATI
;
175 default: return MODE_UNKNOWN
;
179 #define TEXTURE_UNKNOWN_INDEX 7
180 static GLuint
translate_tex_src_bit( GLbitfield bit
)
183 case TEXTURE_1D_BIT
: return TEXTURE_1D_INDEX
;
184 case TEXTURE_2D_BIT
: return TEXTURE_2D_INDEX
;
185 case TEXTURE_RECT_BIT
: return TEXTURE_RECT_INDEX
;
186 case TEXTURE_3D_BIT
: return TEXTURE_3D_INDEX
;
187 case TEXTURE_CUBE_BIT
: return TEXTURE_CUBE_INDEX
;
188 default: return TEXTURE_UNKNOWN_INDEX
;
192 #define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0)
193 #define VERT_RESULT_TEX_ANY (0xff << VERT_RESULT_TEX0)
196 * Identify all possible varying inputs. The fragment program will
197 * never reference non-varying inputs, but will track them via state
200 * This function figures out all the inputs that the fragment program
201 * has access to. The bitmask is later reduced to just those which
202 * are actually referenced.
204 static GLbitfield
get_fp_input_mask( GLcontext
*ctx
)
206 GLbitfield fp_inputs
= 0x0;
208 if (ctx
->VertexProgram
._Overriden
) {
209 /* Somebody's messing with the vertex program and we don't have
210 * a clue what's happening. Assume that it could be producing
211 * all possible outputs.
215 else if (ctx
->RenderMode
== GL_FEEDBACK
) {
216 fp_inputs
= (FRAG_BIT_COL0
| FRAG_BIT_TEX0
);
218 else if (!ctx
->VertexProgram
._Enabled
||
219 !ctx
->VertexProgram
._Current
) {
221 /* Fixed function logic */
222 GLbitfield varying_inputs
= ctx
->varying_vp_inputs
;
224 /* First look at what values may be computed by the generated
227 if (ctx
->Light
.Enabled
) {
228 fp_inputs
|= FRAG_BIT_COL0
;
230 if (ctx
->_TriangleCaps
& DD_SEPARATE_SPECULAR
)
231 fp_inputs
|= FRAG_BIT_COL1
;
234 fp_inputs
|= (ctx
->Texture
._TexGenEnabled
|
235 ctx
->Texture
._TexMatEnabled
) << FRAG_ATTRIB_TEX0
;
237 /* Then look at what might be varying as a result of enabled
240 if (varying_inputs
& VERT_BIT_COLOR0
) fp_inputs
|= FRAG_BIT_COL0
;
241 if (varying_inputs
& VERT_BIT_COLOR1
) fp_inputs
|= FRAG_BIT_COL1
;
243 fp_inputs
|= (((varying_inputs
& VERT_BIT_TEX_ANY
) >> VERT_ATTRIB_TEX0
)
244 << FRAG_ATTRIB_TEX0
);
248 /* calculate from vp->outputs */
249 GLbitfield vp_outputs
= ctx
->VertexProgram
._Current
->Base
.OutputsWritten
;
251 if (vp_outputs
& (1 << VERT_RESULT_COL0
)) fp_inputs
|= FRAG_BIT_COL0
;
252 if (vp_outputs
& (1 << VERT_RESULT_COL1
)) fp_inputs
|= FRAG_BIT_COL1
;
254 fp_inputs
|= (((vp_outputs
& VERT_RESULT_TEX_ANY
) >> VERT_RESULT_TEX0
)
255 << FRAG_ATTRIB_TEX0
);
263 * Examine current texture environment state and generate a unique
264 * key to identify it.
266 static void make_state_key( GLcontext
*ctx
, struct state_key
*key
)
269 GLbitfield inputs_referenced
= FRAG_BIT_COL0
;
270 GLbitfield inputs_available
= get_fp_input_mask( ctx
);
272 memset(key
, 0, sizeof(*key
));
274 for (i
=0;i
<MAX_TEXTURE_UNITS
;i
++) {
275 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
277 if (!texUnit
->_ReallyEnabled
)
280 key
->unit
[i
].enabled
= 1;
281 key
->enabled_units
|= (1<<i
);
282 key
->nr_enabled_units
= i
+1;
283 inputs_referenced
|= FRAG_BIT_TEX(i
);
285 key
->unit
[i
].source_index
=
286 translate_tex_src_bit(texUnit
->_ReallyEnabled
);
288 key
->unit
[i
].NumArgsRGB
= texUnit
->_CurrentCombine
->_NumArgsRGB
;
289 key
->unit
[i
].NumArgsA
= texUnit
->_CurrentCombine
->_NumArgsA
;
291 key
->unit
[i
].ModeRGB
=
292 translate_mode(texUnit
->_CurrentCombine
->ModeRGB
);
294 translate_mode(texUnit
->_CurrentCombine
->ModeA
);
296 key
->unit
[i
].ScaleShiftRGB
= texUnit
->_CurrentCombine
->ScaleShiftRGB
;
297 key
->unit
[i
].ScaleShiftA
= texUnit
->_CurrentCombine
->ScaleShiftA
;
300 key
->unit
[i
].OptRGB
[j
].Operand
=
301 translate_operand(texUnit
->_CurrentCombine
->OperandRGB
[j
]);
302 key
->unit
[i
].OptA
[j
].Operand
=
303 translate_operand(texUnit
->_CurrentCombine
->OperandA
[j
]);
304 key
->unit
[i
].OptRGB
[j
].Source
=
305 translate_source(texUnit
->_CurrentCombine
->SourceRGB
[j
]);
306 key
->unit
[i
].OptA
[j
].Source
=
307 translate_source(texUnit
->_CurrentCombine
->SourceA
[j
]);
311 if (ctx
->_TriangleCaps
& DD_SEPARATE_SPECULAR
) {
312 key
->separate_specular
= 1;
313 inputs_referenced
|= FRAG_BIT_COL1
;
316 if (ctx
->Fog
.Enabled
) {
317 key
->fog_enabled
= 1;
318 key
->fog_mode
= translate_fog_mode(ctx
->Fog
.Mode
);
319 inputs_referenced
|= FRAG_BIT_FOGC
; /* maybe */
322 key
->inputs_available
= (inputs_available
& inputs_referenced
);
326 * Use uregs to represent registers internally, translate to Mesa's
327 * expected formats on emit.
329 * NOTE: These are passed by value extensively in this file rather
330 * than as usual by pointer reference. If this disturbs you, try
331 * remembering they are just 32bits in size.
333 * GCC is smart enough to deal with these dword-sized structures in
334 * much the same way as if I had defined them as dwords and was using
335 * macros to access and set the fields. This is much nicer and easier
348 static const struct ureg undef
= {
359 /** State used to build the fragment program:
361 struct texenv_fragment_program
{
362 struct gl_fragment_program
*program
;
364 struct state_key
*state
;
366 GLbitfield alu_temps
; /**< Track texture indirections, see spec. */
367 GLbitfield temps_output
; /**< Track texture indirections, see spec. */
368 GLbitfield temp_in_use
; /**< Tracks temporary regs which are in use. */
371 struct ureg src_texture
[MAX_TEXTURE_UNITS
];
372 /* Reg containing each texture unit's sampled texture color,
376 struct ureg src_previous
; /**< Reg containing color from previous
377 * stage. May need to be decl'd.
380 GLuint last_tex_stage
; /**< Number of last enabled texture unit */
389 static struct ureg
make_ureg(GLuint file
, GLuint idx
)
397 reg
.swz
= SWIZZLE_NOOP
;
402 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
404 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
407 GET_SWZ(reg
.swz
, w
));
412 static struct ureg
swizzle1( struct ureg reg
, int x
)
414 return swizzle(reg
, x
, x
, x
, x
);
417 static struct ureg
negate( struct ureg reg
)
423 static GLboolean
is_undef( struct ureg reg
)
425 return reg
.file
== PROGRAM_UNDEFINED
;
429 static struct ureg
get_temp( struct texenv_fragment_program
*p
)
433 /* First try and reuse temps which have been used already:
435 bit
= _mesa_ffs( ~p
->temp_in_use
& p
->alu_temps
);
437 /* Then any unused temporary:
440 bit
= _mesa_ffs( ~p
->temp_in_use
);
443 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
447 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
448 p
->program
->Base
.NumTemporaries
= bit
;
450 p
->temp_in_use
|= 1<<(bit
-1);
451 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
454 static struct ureg
get_tex_temp( struct texenv_fragment_program
*p
)
458 /* First try to find available temp not previously used (to avoid
459 * starting a new texture indirection). According to the spec, the
460 * ~p->temps_output isn't necessary, but will keep it there for
463 bit
= _mesa_ffs( ~p
->temp_in_use
& ~p
->alu_temps
& ~p
->temps_output
);
465 /* Then any unused temporary:
468 bit
= _mesa_ffs( ~p
->temp_in_use
);
471 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
475 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
476 p
->program
->Base
.NumTemporaries
= bit
;
478 p
->temp_in_use
|= 1<<(bit
-1);
479 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
483 /** Mark a temp reg as being no longer allocatable. */
484 static void reserve_temp( struct texenv_fragment_program
*p
, struct ureg r
)
486 if (r
.file
== PROGRAM_TEMPORARY
)
487 p
->temps_output
|= (1 << r
.idx
);
491 static void release_temps(GLcontext
*ctx
, struct texenv_fragment_program
*p
)
493 GLuint max_temp
= ctx
->Const
.FragmentProgram
.MaxTemps
;
495 /* KW: To support tex_env_crossbar, don't release the registers in
498 if (max_temp
>= sizeof(int) * 8)
499 p
->temp_in_use
= p
->temps_output
;
501 p
->temp_in_use
= ~((1<<max_temp
)-1) | p
->temps_output
;
505 static struct ureg
register_param5( struct texenv_fragment_program
*p
,
512 gl_state_index tokens
[STATE_LENGTH
];
519 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
520 return make_ureg(PROGRAM_STATE_VAR
, idx
);
524 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
525 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
526 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
527 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
529 static GLuint
frag_to_vert_attrib( GLuint attrib
)
532 case FRAG_ATTRIB_COL0
: return VERT_ATTRIB_COLOR0
;
533 case FRAG_ATTRIB_COL1
: return VERT_ATTRIB_COLOR1
;
535 assert(attrib
>= FRAG_ATTRIB_TEX0
);
536 assert(attrib
<= FRAG_ATTRIB_TEX7
);
537 return attrib
- FRAG_ATTRIB_TEX0
+ VERT_ATTRIB_TEX0
;
542 static struct ureg
register_input( struct texenv_fragment_program
*p
, GLuint input
)
544 if (p
->state
->inputs_available
& (1<<input
)) {
545 p
->program
->Base
.InputsRead
|= (1 << input
);
546 return make_ureg(PROGRAM_INPUT
, input
);
549 GLuint idx
= frag_to_vert_attrib( input
);
550 return register_param3( p
, STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, idx
);
555 static void emit_arg( struct prog_src_register
*reg
,
558 reg
->File
= ureg
.file
;
559 reg
->Index
= ureg
.idx
;
560 reg
->Swizzle
= ureg
.swz
;
561 reg
->NegateBase
= ureg
.negatebase
? 0xf : 0x0;
563 reg
->NegateAbs
= ureg
.negateabs
;
566 static void emit_dst( struct prog_dst_register
*dst
,
567 struct ureg ureg
, GLuint mask
)
569 dst
->File
= ureg
.file
;
570 dst
->Index
= ureg
.idx
;
571 dst
->WriteMask
= mask
;
572 dst
->CondMask
= COND_TR
; /* always pass cond test */
573 dst
->CondSwizzle
= SWIZZLE_NOOP
;
576 static struct prog_instruction
*
577 emit_op(struct texenv_fragment_program
*p
,
586 GLuint nr
= p
->program
->Base
.NumInstructions
++;
587 struct prog_instruction
*inst
= &p
->program
->Base
.Instructions
[nr
];
589 assert(nr
< MAX_INSTRUCTIONS
);
591 _mesa_init_instructions(inst
, 1);
594 emit_arg( &inst
->SrcReg
[0], src0
);
595 emit_arg( &inst
->SrcReg
[1], src1
);
596 emit_arg( &inst
->SrcReg
[2], src2
);
598 inst
->SaturateMode
= saturate
? SATURATE_ZERO_ONE
: SATURATE_OFF
;
600 emit_dst( &inst
->DstReg
, dest
, mask
);
603 /* Accounting for indirection tracking:
605 if (dest
.file
== PROGRAM_TEMPORARY
)
606 p
->temps_output
|= 1 << dest
.idx
;
613 static struct ureg
emit_arith( struct texenv_fragment_program
*p
,
622 emit_op(p
, op
, dest
, mask
, saturate
, src0
, src1
, src2
);
624 /* Accounting for indirection tracking:
626 if (src0
.file
== PROGRAM_TEMPORARY
)
627 p
->alu_temps
|= 1 << src0
.idx
;
629 if (!is_undef(src1
) && src1
.file
== PROGRAM_TEMPORARY
)
630 p
->alu_temps
|= 1 << src1
.idx
;
632 if (!is_undef(src2
) && src2
.file
== PROGRAM_TEMPORARY
)
633 p
->alu_temps
|= 1 << src2
.idx
;
635 if (dest
.file
== PROGRAM_TEMPORARY
)
636 p
->alu_temps
|= 1 << dest
.idx
;
638 p
->program
->Base
.NumAluInstructions
++;
642 static struct ureg
emit_texld( struct texenv_fragment_program
*p
,
650 struct prog_instruction
*inst
= emit_op( p
, op
,
652 GL_FALSE
, /* don't saturate? */
657 inst
->TexSrcTarget
= tex_idx
;
658 inst
->TexSrcUnit
= tex_unit
;
660 p
->program
->Base
.NumTexInstructions
++;
662 /* Accounting for indirection tracking:
664 reserve_temp(p
, dest
);
666 /* Is this a texture indirection?
668 if ((coord
.file
== PROGRAM_TEMPORARY
&&
669 (p
->temps_output
& (1<<coord
.idx
))) ||
670 (dest
.file
== PROGRAM_TEMPORARY
&&
671 (p
->alu_temps
& (1<<dest
.idx
)))) {
672 p
->program
->Base
.NumTexIndirections
++;
673 p
->temps_output
= 1<<coord
.idx
;
675 assert(0); /* KW: texture env crossbar */
682 static struct ureg
register_const4f( struct texenv_fragment_program
*p
,
695 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
697 r
= make_ureg(PROGRAM_CONSTANT
, idx
);
702 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
703 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
704 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
705 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
708 static struct ureg
get_one( struct texenv_fragment_program
*p
)
710 if (is_undef(p
->one
))
711 p
->one
= register_scalar_const(p
, 1.0);
715 static struct ureg
get_half( struct texenv_fragment_program
*p
)
717 if (is_undef(p
->half
))
718 p
->half
= register_scalar_const(p
, 0.5);
722 static struct ureg
get_zero( struct texenv_fragment_program
*p
)
724 if (is_undef(p
->zero
))
725 p
->zero
= register_scalar_const(p
, 0.0);
730 static void program_error( struct texenv_fragment_program
*p
, const char *msg
)
732 _mesa_problem(NULL
, msg
);
736 static struct ureg
get_source( struct texenv_fragment_program
*p
,
737 GLuint src
, GLuint unit
)
741 assert(!is_undef(p
->src_texture
[unit
]));
742 return p
->src_texture
[unit
];
752 assert(!is_undef(p
->src_texture
[src
- SRC_TEXTURE0
]));
753 return p
->src_texture
[src
- SRC_TEXTURE0
];
756 return register_param2(p
, STATE_TEXENV_COLOR
, unit
);
758 case SRC_PRIMARY_COLOR
:
759 return register_input(p
, FRAG_ATTRIB_COL0
);
763 if (is_undef(p
->src_previous
))
764 return register_input(p
, FRAG_ATTRIB_COL0
);
766 return p
->src_previous
;
770 static struct ureg
emit_combine_source( struct texenv_fragment_program
*p
,
776 struct ureg arg
, src
, one
;
778 src
= get_source(p
, source
, unit
);
781 case OPR_ONE_MINUS_SRC_COLOR
:
783 * Emit tmp = 1.0 - arg.xyzw
787 return emit_arith( p
, OPCODE_SUB
, arg
, mask
, 0, one
, src
, undef
);
790 if (mask
== WRITEMASK_W
)
793 return swizzle1( src
, SWIZZLE_W
);
794 case OPR_ONE_MINUS_SRC_ALPHA
:
796 * Emit tmp = 1.0 - arg.wwww
800 return emit_arith(p
, OPCODE_SUB
, arg
, mask
, 0,
801 one
, swizzle1(src
, SWIZZLE_W
), undef
);
812 static GLboolean
args_match( struct state_key
*key
, GLuint unit
)
814 GLuint i
, nr
= key
->unit
[unit
].NumArgsRGB
;
816 for (i
= 0 ; i
< nr
; i
++) {
817 if (key
->unit
[unit
].OptA
[i
].Source
!= key
->unit
[unit
].OptRGB
[i
].Source
)
820 switch(key
->unit
[unit
].OptA
[i
].Operand
) {
822 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
830 case OPR_ONE_MINUS_SRC_ALPHA
:
831 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
832 case OPR_ONE_MINUS_SRC_COLOR
:
833 case OPR_ONE_MINUS_SRC_ALPHA
:
840 return GL_FALSE
; /* impossible */
847 static struct ureg
emit_combine( struct texenv_fragment_program
*p
,
854 const struct mode_opt
*opt
)
857 struct ureg tmp
, half
;
860 tmp
= undef
; /* silence warning (bug 5318) */
862 for (i
= 0; i
< nr
; i
++)
863 src
[i
] = emit_combine_source( p
, mask
, unit
, opt
[i
].Source
, opt
[i
].Operand
);
867 if (mask
== WRITEMASK_XYZW
&& !saturate
)
870 return emit_arith( p
, OPCODE_MOV
, dest
, mask
, saturate
, src
[0], undef
, undef
);
872 return emit_arith( p
, OPCODE_MUL
, dest
, mask
, saturate
,
873 src
[0], src
[1], undef
);
875 return emit_arith( p
, OPCODE_ADD
, dest
, mask
, saturate
,
876 src
[0], src
[1], undef
);
877 case MODE_ADD_SIGNED
:
883 emit_arith( p
, OPCODE_ADD
, tmp
, mask
, 0, src
[0], src
[1], undef
);
884 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp
, half
, undef
);
886 case MODE_INTERPOLATE
:
887 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
889 return emit_arith( p
, OPCODE_LRP
, dest
, mask
, saturate
, src
[2], src
[0], src
[1] );
892 return emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, src
[0], src
[1], undef
);
895 case MODE_DOT3_RGBA_EXT
:
896 case MODE_DOT3_RGB_EXT
:
897 case MODE_DOT3_RGB
: {
898 struct ureg tmp0
= get_temp( p
);
899 struct ureg tmp1
= get_temp( p
);
900 struct ureg neg1
= register_scalar_const(p
, -1);
901 struct ureg two
= register_scalar_const(p
, 2);
906 * dst = tmp0 dot3 tmp1
908 emit_arith( p
, OPCODE_MAD
, tmp0
, WRITEMASK_XYZW
, 0,
911 if (_mesa_memcmp(&src
[0], &src
[1], sizeof(struct ureg
)) == 0)
914 emit_arith( p
, OPCODE_MAD
, tmp1
, WRITEMASK_XYZW
, 0,
916 emit_arith( p
, OPCODE_DP3
, dest
, mask
, saturate
, tmp0
, tmp1
, undef
);
919 case MODE_MODULATE_ADD_ATI
:
920 /* Arg0 * Arg2 + Arg1 */
921 return emit_arith( p
, OPCODE_MAD
, dest
, mask
, saturate
,
922 src
[0], src
[2], src
[1] );
923 case MODE_MODULATE_SIGNED_ADD_ATI
: {
924 /* Arg0 * Arg2 + Arg1 - 0.5 */
925 struct ureg tmp0
= get_temp(p
);
927 emit_arith( p
, OPCODE_MAD
, tmp0
, mask
, 0, src
[0], src
[2], src
[1] );
928 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp0
, half
, undef
);
931 case MODE_MODULATE_SUBTRACT_ATI
:
932 /* Arg0 * Arg2 - Arg1 */
933 emit_arith( p
, OPCODE_MAD
, dest
, mask
, 0, src
[0], src
[2], negate(src
[1]) );
942 * Generate instructions for one texture unit's env/combiner mode.
945 emit_texenv(struct texenv_fragment_program
*p
, GLuint unit
)
947 struct state_key
*key
= p
->state
;
948 GLboolean saturate
= (unit
< p
->last_tex_stage
);
949 GLuint rgb_shift
, alpha_shift
;
950 struct ureg out
, shift
;
953 if (!key
->unit
[unit
].enabled
) {
954 return get_source(p
, SRC_PREVIOUS
, 0);
957 switch (key
->unit
[unit
].ModeRGB
) {
958 case MODE_DOT3_RGB_EXT
:
959 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
962 case MODE_DOT3_RGBA_EXT
:
967 rgb_shift
= key
->unit
[unit
].ScaleShiftRGB
;
968 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
972 /* If this is the very last calculation, emit direct to output reg:
974 if (key
->separate_specular
||
975 unit
!= p
->last_tex_stage
||
978 dest
= get_temp( p
);
980 dest
= make_ureg(PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
982 /* Emit the RGB and A combine ops
984 if (key
->unit
[unit
].ModeRGB
== key
->unit
[unit
].ModeA
&&
985 args_match(key
, unit
)) {
986 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
988 key
->unit
[unit
].NumArgsRGB
,
989 key
->unit
[unit
].ModeRGB
,
990 key
->unit
[unit
].OptRGB
);
992 else if (key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA_EXT
||
993 key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA
) {
995 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
997 key
->unit
[unit
].NumArgsRGB
,
998 key
->unit
[unit
].ModeRGB
,
999 key
->unit
[unit
].OptRGB
);
1002 /* Need to do something to stop from re-emitting identical
1003 * argument calculations here:
1005 out
= emit_combine( p
, dest
, WRITEMASK_XYZ
, saturate
,
1007 key
->unit
[unit
].NumArgsRGB
,
1008 key
->unit
[unit
].ModeRGB
,
1009 key
->unit
[unit
].OptRGB
);
1010 out
= emit_combine( p
, dest
, WRITEMASK_W
, saturate
,
1012 key
->unit
[unit
].NumArgsA
,
1013 key
->unit
[unit
].ModeA
,
1014 key
->unit
[unit
].OptA
);
1017 /* Deal with the final shift:
1019 if (alpha_shift
|| rgb_shift
) {
1020 if (rgb_shift
== alpha_shift
) {
1021 shift
= register_scalar_const(p
, (GLfloat
)(1<<rgb_shift
));
1024 shift
= register_const4f(p
,
1025 (GLfloat
)(1<<rgb_shift
),
1026 (GLfloat
)(1<<rgb_shift
),
1027 (GLfloat
)(1<<rgb_shift
),
1028 (GLfloat
)(1<<alpha_shift
));
1030 return emit_arith( p
, OPCODE_MUL
, dest
, WRITEMASK_XYZW
,
1031 saturate
, out
, shift
, undef
);
1039 * Generate instruction for getting a texture source term.
1041 static void load_texture( struct texenv_fragment_program
*p
, GLuint unit
)
1043 if (is_undef(p
->src_texture
[unit
])) {
1044 GLuint dim
= p
->state
->unit
[unit
].source_index
;
1045 struct ureg texcoord
= register_input(p
, FRAG_ATTRIB_TEX0
+unit
);
1046 struct ureg tmp
= get_tex_temp( p
);
1048 if (dim
== TEXTURE_UNKNOWN_INDEX
)
1049 program_error(p
, "TexSrcBit");
1051 /* TODO: Use D0_MASK_XY where possible.
1053 if (p
->state
->unit
[unit
].enabled
) {
1054 p
->src_texture
[unit
] = emit_texld( p
, OPCODE_TXP
,
1055 tmp
, WRITEMASK_XYZW
,
1056 unit
, dim
, texcoord
);
1057 p
->program
->Base
.SamplersUsed
|= (1 << unit
);
1058 /* This identity mapping should already be in place
1059 * (see _mesa_init_program_struct()) but let's be safe.
1061 p
->program
->Base
.SamplerUnits
[unit
] = unit
;
1064 p
->src_texture
[unit
] = get_zero(p
);
1068 static GLboolean
load_texenv_source( struct texenv_fragment_program
*p
,
1069 GLuint src
, GLuint unit
)
1073 load_texture(p
, unit
);
1084 load_texture(p
, src
- SRC_TEXTURE0
);
1096 * Generate instructions for loading all texture source terms.
1099 load_texunit_sources( struct texenv_fragment_program
*p
, int unit
)
1101 struct state_key
*key
= p
->state
;
1104 for (i
= 0; i
< key
->unit
[unit
].NumArgsRGB
; i
++) {
1105 load_texenv_source( p
, key
->unit
[unit
].OptRGB
[i
].Source
, unit
);
1108 for (i
= 0; i
< key
->unit
[unit
].NumArgsA
; i
++) {
1109 load_texenv_source( p
, key
->unit
[unit
].OptA
[i
].Source
, unit
);
1117 * Generate a new fragment program which implements the context's
1118 * current texture env/combine mode.
1121 create_new_program(GLcontext
*ctx
, struct state_key
*key
,
1122 struct gl_fragment_program
*program
)
1124 struct prog_instruction instBuffer
[MAX_INSTRUCTIONS
];
1125 struct texenv_fragment_program p
;
1127 struct ureg cf
, out
;
1129 _mesa_memset(&p
, 0, sizeof(p
));
1132 p
.program
= program
;
1134 /* During code generation, use locally-allocated instruction buffer,
1135 * then alloc dynamic storage below.
1137 p
.program
->Base
.Instructions
= instBuffer
;
1138 p
.program
->Base
.Target
= GL_FRAGMENT_PROGRAM_ARB
;
1139 p
.program
->Base
.NumTexIndirections
= 1; /* correct? */
1140 p
.program
->Base
.NumTexInstructions
= 0;
1141 p
.program
->Base
.NumAluInstructions
= 0;
1142 p
.program
->Base
.String
= NULL
;
1143 p
.program
->Base
.NumInstructions
=
1144 p
.program
->Base
.NumTemporaries
=
1145 p
.program
->Base
.NumParameters
=
1146 p
.program
->Base
.NumAttributes
= p
.program
->Base
.NumAddressRegs
= 0;
1147 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1149 p
.program
->Base
.InputsRead
= 0;
1150 p
.program
->Base
.OutputsWritten
= 1 << FRAG_RESULT_COLR
;
1152 for (unit
= 0; unit
< MAX_TEXTURE_UNITS
; unit
++)
1153 p
.src_texture
[unit
] = undef
;
1155 p
.src_previous
= undef
;
1160 p
.last_tex_stage
= 0;
1161 release_temps(ctx
, &p
);
1163 if (key
->enabled_units
) {
1164 /* First pass - to support texture_env_crossbar, first identify
1165 * all referenced texture sources and emit texld instructions
1168 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1169 if (key
->unit
[unit
].enabled
) {
1170 load_texunit_sources( &p
, unit
);
1171 p
.last_tex_stage
= unit
;
1174 /* Second pass - emit combine instructions to build final color:
1176 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1177 if (key
->enabled_units
& (1<<unit
)) {
1178 p
.src_previous
= emit_texenv( &p
, unit
);
1179 reserve_temp(&p
, p
.src_previous
); /* don't re-use this temp reg */
1180 release_temps(ctx
, &p
); /* release all temps */
1184 cf
= get_source( &p
, SRC_PREVIOUS
, 0 );
1185 out
= make_ureg( PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
1187 if (key
->separate_specular
) {
1188 /* Emit specular add.
1190 struct ureg s
= register_input(&p
, FRAG_ATTRIB_COL1
);
1191 emit_arith( &p
, OPCODE_ADD
, out
, WRITEMASK_XYZ
, 0, cf
, s
, undef
);
1192 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_W
, 0, cf
, undef
, undef
);
1194 else if (_mesa_memcmp(&cf
, &out
, sizeof(cf
)) != 0) {
1195 /* Will wind up in here if no texture enabled or a couple of
1196 * other scenarios (GL_REPLACE for instance).
1198 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_XYZW
, 0, cf
, undef
, undef
);
1203 emit_arith( &p
, OPCODE_END
, undef
, WRITEMASK_XYZW
, 0, undef
, undef
, undef
);
1205 if (key
->fog_enabled
) {
1206 /* Pull fog mode from GLcontext, the value in the state key is
1207 * a reduced value and not what is expected in FogOption
1209 p
.program
->FogOption
= ctx
->Fog
.Mode
;
1210 p
.program
->Base
.InputsRead
|= FRAG_BIT_FOGC
; /* XXX new */
1212 p
.program
->FogOption
= GL_NONE
;
1214 if (p
.program
->Base
.NumTexIndirections
> ctx
->Const
.FragmentProgram
.MaxTexIndirections
)
1215 program_error(&p
, "Exceeded max nr indirect texture lookups");
1217 if (p
.program
->Base
.NumTexInstructions
> ctx
->Const
.FragmentProgram
.MaxTexInstructions
)
1218 program_error(&p
, "Exceeded max TEX instructions");
1220 if (p
.program
->Base
.NumAluInstructions
> ctx
->Const
.FragmentProgram
.MaxAluInstructions
)
1221 program_error(&p
, "Exceeded max ALU instructions");
1223 ASSERT(p
.program
->Base
.NumInstructions
<= MAX_INSTRUCTIONS
);
1225 /* Allocate final instruction array */
1226 p
.program
->Base
.Instructions
1227 = _mesa_alloc_instructions(p
.program
->Base
.NumInstructions
);
1228 if (!p
.program
->Base
.Instructions
) {
1229 _mesa_error(ctx
, GL_OUT_OF_MEMORY
,
1230 "generating tex env program");
1233 _mesa_copy_instructions(p
.program
->Base
.Instructions
, instBuffer
,
1234 p
.program
->Base
.NumInstructions
);
1236 if (p
.program
->FogOption
) {
1237 _mesa_append_fog_code(ctx
, p
.program
);
1238 p
.program
->FogOption
= GL_NONE
;
1242 /* Notify driver the fragment program has (actually) changed.
1244 if (ctx
->Driver
.ProgramStringNotify
) {
1245 ctx
->Driver
.ProgramStringNotify( ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1250 _mesa_print_program(&p
.program
->Base
);
1257 * Return a fragment program which implements the current
1258 * fixed-function texture, fog and color-sum operations.
1260 struct gl_fragment_program
*
1261 _mesa_get_fixed_func_fragment_program(GLcontext
*ctx
)
1263 struct gl_fragment_program
*prog
;
1264 struct state_key key
;
1266 make_state_key(ctx
, &key
);
1268 prog
= (struct gl_fragment_program
*)
1269 _mesa_search_program_cache(ctx
->FragmentProgram
.Cache
,
1273 prog
= (struct gl_fragment_program
*)
1274 ctx
->Driver
.NewProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
, 0);
1276 create_new_program(ctx
, &key
, prog
);
1278 _mesa_program_cache_insert(ctx
, ctx
->FragmentProgram
.Cache
,
1279 &key
, sizeof(key
), &prog
->Base
);
1288 * If _MaintainTexEnvProgram is set we'll generate a fragment program that
1289 * implements the current texture env/combine mode.
1290 * This function generates that program and puts it into effect.
1293 _mesa_UpdateTexEnvProgram( GLcontext
*ctx
)
1295 const struct gl_fragment_program
*prev
= ctx
->FragmentProgram
._Current
;
1297 ASSERT(ctx
->FragmentProgram
._MaintainTexEnvProgram
);
1299 /* If a conventional fragment program/shader isn't in effect... */
1300 if (!ctx
->FragmentProgram
._Enabled
&&
1301 (!ctx
->Shader
.CurrentProgram
||
1302 !ctx
->Shader
.CurrentProgram
->FragmentProgram
) ) {
1304 ctx
->FragmentProgram
._Current
1305 = ctx
->FragmentProgram
._TexEnvProgram
1306 = _mesa_get_fixed_func_fragment_program(ctx
);
1309 /* Tell the driver about the change. Could define a new target for
1312 if (ctx
->FragmentProgram
._Current
!= prev
&& ctx
->Driver
.BindProgram
) {
1313 ctx
->Driver
.BindProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1314 (struct gl_program
*) ctx
->FragmentProgram
._Current
);