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 * This MAX is probably a bit generous, but that's OK. There can be
41 * up to four instructions per texture unit (TEX + 3 for combine),
42 * then there's fog and specular add.
44 #define MAX_INSTRUCTIONS ((MAX_TEXTURE_UNITS * 4) + 12)
46 #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
54 GLbitfield enabled_units
;
55 GLuint separate_specular
:1;
61 GLuint source_index
:3; /* one of TEXTURE_1D/2D/3D/CUBE/RECT_INDEX */
62 GLuint ScaleShiftRGB
:2;
67 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
;
193 * Examine current texture environment state and generate a unique
194 * key to identify it.
196 static void make_state_key( GLcontext
*ctx
, struct state_key
*key
)
200 memset(key
, 0, sizeof(*key
));
202 for (i
=0;i
<MAX_TEXTURE_UNITS
;i
++) {
203 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
205 if (!texUnit
->_ReallyEnabled
)
208 key
->unit
[i
].enabled
= 1;
209 key
->enabled_units
|= (1<<i
);
211 key
->unit
[i
].source_index
=
212 translate_tex_src_bit(texUnit
->_ReallyEnabled
);
214 key
->unit
[i
].NumArgsRGB
= texUnit
->_CurrentCombine
->_NumArgsRGB
;
215 key
->unit
[i
].NumArgsA
= texUnit
->_CurrentCombine
->_NumArgsA
;
217 key
->unit
[i
].ModeRGB
=
218 translate_mode(texUnit
->_CurrentCombine
->ModeRGB
);
220 translate_mode(texUnit
->_CurrentCombine
->ModeA
);
222 key
->unit
[i
].ScaleShiftRGB
= texUnit
->_CurrentCombine
->ScaleShiftRGB
;
223 key
->unit
[i
].ScaleShiftA
= texUnit
->_CurrentCombine
->ScaleShiftA
;
226 key
->unit
[i
].OptRGB
[j
].Operand
=
227 translate_operand(texUnit
->_CurrentCombine
->OperandRGB
[j
]);
228 key
->unit
[i
].OptA
[j
].Operand
=
229 translate_operand(texUnit
->_CurrentCombine
->OperandA
[j
]);
230 key
->unit
[i
].OptRGB
[j
].Source
=
231 translate_source(texUnit
->_CurrentCombine
->SourceRGB
[j
]);
232 key
->unit
[i
].OptA
[j
].Source
=
233 translate_source(texUnit
->_CurrentCombine
->SourceA
[j
]);
237 if (ctx
->_TriangleCaps
& DD_SEPARATE_SPECULAR
)
238 key
->separate_specular
= 1;
240 if (ctx
->Fog
.Enabled
) {
241 key
->fog_enabled
= 1;
242 key
->fog_mode
= translate_fog_mode(ctx
->Fog
.Mode
);
246 /* Use uregs to represent registers internally, translate to Mesa's
247 * expected formats on emit.
249 * NOTE: These are passed by value extensively in this file rather
250 * than as usual by pointer reference. If this disturbs you, try
251 * remembering they are just 32bits in size.
253 * GCC is smart enough to deal with these dword-sized structures in
254 * much the same way as if I had defined them as dwords and was using
255 * macros to access and set the fields. This is much nicer and easier
268 static const struct ureg undef
= {
279 /* State used to build the fragment program:
281 struct texenv_fragment_program
{
282 struct gl_fragment_program
*program
;
284 struct state_key
*state
;
286 GLbitfield alu_temps
; /* Track texture indirections, see spec. */
287 GLbitfield temps_output
; /* Track texture indirections, see spec. */
288 GLbitfield temp_in_use
; /* Tracks temporary regs which are in use. */
291 struct ureg src_texture
[MAX_TEXTURE_UNITS
];
292 /* Reg containing each texture unit's sampled texture color,
296 struct ureg src_previous
; /* Reg containing color from previous
297 * stage. May need to be decl'd.
300 GLuint last_tex_stage
; /* Number of last enabled texture unit */
309 static struct ureg
make_ureg(GLuint file
, GLuint idx
)
317 reg
.swz
= SWIZZLE_NOOP
;
322 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
324 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
327 GET_SWZ(reg
.swz
, w
));
332 static struct ureg
swizzle1( struct ureg reg
, int x
)
334 return swizzle(reg
, x
, x
, x
, x
);
337 static struct ureg
negate( struct ureg reg
)
343 static GLboolean
is_undef( struct ureg reg
)
345 return reg
.file
== PROGRAM_UNDEFINED
;
349 static struct ureg
get_temp( struct texenv_fragment_program
*p
)
353 /* First try and reuse temps which have been used already:
355 bit
= _mesa_ffs( ~p
->temp_in_use
& p
->alu_temps
);
357 /* Then any unused temporary:
360 bit
= _mesa_ffs( ~p
->temp_in_use
);
363 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
367 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
368 p
->program
->Base
.NumTemporaries
= bit
;
370 p
->temp_in_use
|= 1<<(bit
-1);
371 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
374 static struct ureg
get_tex_temp( struct texenv_fragment_program
*p
)
378 /* First try to find availble temp not previously used (to avoid
379 * starting a new texture indirection). According to the spec, the
380 * ~p->temps_output isn't necessary, but will keep it there for
383 bit
= _mesa_ffs( ~p
->temp_in_use
& ~p
->alu_temps
& ~p
->temps_output
);
385 /* Then any unused temporary:
388 bit
= _mesa_ffs( ~p
->temp_in_use
);
391 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
395 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
396 p
->program
->Base
.NumTemporaries
= bit
;
398 p
->temp_in_use
|= 1<<(bit
-1);
399 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
403 static void release_temps(GLcontext
*ctx
, struct texenv_fragment_program
*p
)
405 GLuint max_temp
= ctx
->Const
.FragmentProgram
.MaxTemps
;
407 /* KW: To support tex_env_crossbar, don't release the registers in
410 if (max_temp
>= sizeof(int) * 8)
411 p
->temp_in_use
= p
->temps_output
;
413 p
->temp_in_use
= ~((1<<max_temp
)-1) | p
->temps_output
;
417 static struct ureg
register_param5( struct texenv_fragment_program
*p
,
424 gl_state_index tokens
[STATE_LENGTH
];
431 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
432 return make_ureg(PROGRAM_STATE_VAR
, idx
);
436 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
437 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
438 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
439 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
442 static struct ureg
register_input( struct texenv_fragment_program
*p
, GLuint input
)
444 p
->program
->Base
.InputsRead
|= (1 << input
);
445 return make_ureg(PROGRAM_INPUT
, input
);
449 static void emit_arg( struct prog_src_register
*reg
,
452 reg
->File
= ureg
.file
;
453 reg
->Index
= ureg
.idx
;
454 reg
->Swizzle
= ureg
.swz
;
455 reg
->NegateBase
= ureg
.negatebase
? 0xf : 0x0;
457 reg
->NegateAbs
= ureg
.negateabs
;
460 static void emit_dst( struct prog_dst_register
*dst
,
461 struct ureg ureg
, GLuint mask
)
463 dst
->File
= ureg
.file
;
464 dst
->Index
= ureg
.idx
;
465 dst
->WriteMask
= mask
;
466 dst
->CondMask
= COND_TR
; /* always pass cond test */
467 dst
->CondSwizzle
= SWIZZLE_NOOP
;
470 static struct prog_instruction
*
471 emit_op(struct texenv_fragment_program
*p
,
480 GLuint nr
= p
->program
->Base
.NumInstructions
++;
481 struct prog_instruction
*inst
= &p
->program
->Base
.Instructions
[nr
];
483 assert(nr
< MAX_INSTRUCTIONS
);
485 _mesa_init_instructions(inst
, 1);
488 emit_arg( &inst
->SrcReg
[0], src0
);
489 emit_arg( &inst
->SrcReg
[1], src1
);
490 emit_arg( &inst
->SrcReg
[2], src2
);
492 inst
->SaturateMode
= saturate
? SATURATE_ZERO_ONE
: SATURATE_OFF
;
494 emit_dst( &inst
->DstReg
, dest
, mask
);
496 /* Accounting for indirection tracking:
498 if (dest
.file
== PROGRAM_TEMPORARY
)
499 p
->temps_output
|= 1 << dest
.idx
;
505 static struct ureg
emit_arith( struct texenv_fragment_program
*p
,
514 emit_op(p
, op
, dest
, mask
, saturate
, src0
, src1
, src2
);
516 /* Accounting for indirection tracking:
518 if (src0
.file
== PROGRAM_TEMPORARY
)
519 p
->alu_temps
|= 1 << src0
.idx
;
521 if (!is_undef(src1
) && src1
.file
== PROGRAM_TEMPORARY
)
522 p
->alu_temps
|= 1 << src1
.idx
;
524 if (!is_undef(src2
) && src2
.file
== PROGRAM_TEMPORARY
)
525 p
->alu_temps
|= 1 << src2
.idx
;
527 if (dest
.file
== PROGRAM_TEMPORARY
)
528 p
->alu_temps
|= 1 << dest
.idx
;
530 p
->program
->Base
.NumAluInstructions
++;
534 static struct ureg
emit_texld( struct texenv_fragment_program
*p
,
542 struct prog_instruction
*inst
= emit_op( p
, op
,
544 GL_FALSE
, /* don't saturate? */
549 inst
->TexSrcTarget
= tex_idx
;
550 inst
->TexSrcUnit
= tex_unit
;
552 p
->program
->Base
.NumTexInstructions
++;
554 /* Is this a texture indirection?
556 if ((coord
.file
== PROGRAM_TEMPORARY
&&
557 (p
->temps_output
& (1<<coord
.idx
))) ||
558 (dest
.file
== PROGRAM_TEMPORARY
&&
559 (p
->alu_temps
& (1<<dest
.idx
)))) {
560 p
->program
->Base
.NumTexIndirections
++;
561 p
->temps_output
= 1<<coord
.idx
;
563 assert(0); /* KW: texture env crossbar */
570 static struct ureg
register_const4f( struct texenv_fragment_program
*p
,
582 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
584 ASSERT(swizzle
== SWIZZLE_NOOP
);
585 return make_ureg(PROGRAM_CONSTANT
, idx
);
588 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
589 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
590 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
591 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
594 static struct ureg
get_one( struct texenv_fragment_program
*p
)
596 if (is_undef(p
->one
))
597 p
->one
= register_scalar_const(p
, 1.0);
601 static struct ureg
get_half( struct texenv_fragment_program
*p
)
603 if (is_undef(p
->half
))
604 p
->half
= register_scalar_const(p
, 0.5);
608 static struct ureg
get_zero( struct texenv_fragment_program
*p
)
610 if (is_undef(p
->zero
))
611 p
->zero
= register_scalar_const(p
, 0.0);
616 static void program_error( struct texenv_fragment_program
*p
, const char *msg
)
618 _mesa_problem(NULL
, msg
);
622 static struct ureg
get_source( struct texenv_fragment_program
*p
,
623 GLuint src
, GLuint unit
)
627 assert(!is_undef(p
->src_texture
[unit
]));
628 return p
->src_texture
[unit
];
638 assert(!is_undef(p
->src_texture
[src
- SRC_TEXTURE0
]));
639 return p
->src_texture
[src
- SRC_TEXTURE0
];
642 return register_param2(p
, STATE_TEXENV_COLOR
, unit
);
644 case SRC_PRIMARY_COLOR
:
645 return register_input(p
, FRAG_ATTRIB_COL0
);
649 if (is_undef(p
->src_previous
))
650 return register_input(p
, FRAG_ATTRIB_COL0
);
652 return p
->src_previous
;
656 static struct ureg
emit_combine_source( struct texenv_fragment_program
*p
,
662 struct ureg arg
, src
, one
;
664 src
= get_source(p
, source
, unit
);
667 case OPR_ONE_MINUS_SRC_COLOR
:
669 * Emit tmp = 1.0 - arg.xyzw
673 return emit_arith( p
, OPCODE_SUB
, arg
, mask
, 0, one
, src
, undef
);
676 if (mask
== WRITEMASK_W
)
679 return swizzle1( src
, SWIZZLE_W
);
680 case OPR_ONE_MINUS_SRC_ALPHA
:
682 * Emit tmp = 1.0 - arg.wwww
686 return emit_arith(p
, OPCODE_SUB
, arg
, mask
, 0,
687 one
, swizzle1(src
, SWIZZLE_W
), undef
);
698 static GLboolean
args_match( struct state_key
*key
, GLuint unit
)
700 GLuint i
, nr
= key
->unit
[unit
].NumArgsRGB
;
702 for (i
= 0 ; i
< nr
; i
++) {
703 if (key
->unit
[unit
].OptA
[i
].Source
!= key
->unit
[unit
].OptRGB
[i
].Source
)
706 switch(key
->unit
[unit
].OptA
[i
].Operand
) {
708 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
716 case OPR_ONE_MINUS_SRC_ALPHA
:
717 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
718 case OPR_ONE_MINUS_SRC_COLOR
:
719 case OPR_ONE_MINUS_SRC_ALPHA
:
726 return GL_FALSE
; /* impossible */
733 static struct ureg
emit_combine( struct texenv_fragment_program
*p
,
740 const struct mode_opt
*opt
)
743 struct ureg tmp
, half
;
746 tmp
= undef
; /* silence warning (bug 5318) */
748 for (i
= 0; i
< nr
; i
++)
749 src
[i
] = emit_combine_source( p
, mask
, unit
, opt
[i
].Source
, opt
[i
].Operand
);
753 if (mask
== WRITEMASK_XYZW
&& !saturate
)
756 return emit_arith( p
, OPCODE_MOV
, dest
, mask
, saturate
, src
[0], undef
, undef
);
758 return emit_arith( p
, OPCODE_MUL
, dest
, mask
, saturate
,
759 src
[0], src
[1], undef
);
761 return emit_arith( p
, OPCODE_ADD
, dest
, mask
, saturate
,
762 src
[0], src
[1], undef
);
763 case MODE_ADD_SIGNED
:
769 emit_arith( p
, OPCODE_ADD
, tmp
, mask
, 0, src
[0], src
[1], undef
);
770 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp
, half
, undef
);
772 case MODE_INTERPOLATE
:
773 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
775 return emit_arith( p
, OPCODE_LRP
, dest
, mask
, saturate
, src
[2], src
[0], src
[1] );
778 return emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, src
[0], src
[1], undef
);
781 case MODE_DOT3_RGBA_EXT
:
782 case MODE_DOT3_RGB_EXT
:
783 case MODE_DOT3_RGB
: {
784 struct ureg tmp0
= get_temp( p
);
785 struct ureg tmp1
= get_temp( p
);
786 struct ureg neg1
= register_scalar_const(p
, -1);
787 struct ureg two
= register_scalar_const(p
, 2);
792 * dst = tmp0 dot3 tmp1
794 emit_arith( p
, OPCODE_MAD
, tmp0
, WRITEMASK_XYZW
, 0,
797 if (_mesa_memcmp(&src
[0], &src
[1], sizeof(struct ureg
)) == 0)
800 emit_arith( p
, OPCODE_MAD
, tmp1
, WRITEMASK_XYZW
, 0,
802 emit_arith( p
, OPCODE_DP3
, dest
, mask
, saturate
, tmp0
, tmp1
, undef
);
805 case MODE_MODULATE_ADD_ATI
:
806 /* Arg0 * Arg2 + Arg1 */
807 return emit_arith( p
, OPCODE_MAD
, dest
, mask
, saturate
,
808 src
[0], src
[2], src
[1] );
809 case MODE_MODULATE_SIGNED_ADD_ATI
: {
810 /* Arg0 * Arg2 + Arg1 - 0.5 */
811 struct ureg tmp0
= get_temp(p
);
813 emit_arith( p
, OPCODE_MAD
, tmp0
, mask
, 0, src
[0], src
[2], src
[1] );
814 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp0
, half
, undef
);
817 case MODE_MODULATE_SUBTRACT_ATI
:
818 /* Arg0 * Arg2 - Arg1 */
819 emit_arith( p
, OPCODE_MAD
, dest
, mask
, 0, src
[0], src
[2], negate(src
[1]) );
828 * Generate instructions for one texture unit's env/combiner mode.
831 emit_texenv(struct texenv_fragment_program
*p
, GLuint unit
)
833 struct state_key
*key
= p
->state
;
834 GLboolean saturate
= (unit
< p
->last_tex_stage
);
835 GLuint rgb_shift
, alpha_shift
;
836 struct ureg out
, shift
;
839 if (!key
->unit
[unit
].enabled
) {
840 return get_source(p
, SRC_PREVIOUS
, 0);
843 switch (key
->unit
[unit
].ModeRGB
) {
844 case MODE_DOT3_RGB_EXT
:
845 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
848 case MODE_DOT3_RGBA_EXT
:
853 rgb_shift
= key
->unit
[unit
].ScaleShiftRGB
;
854 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
858 /* If this is the very last calculation, emit direct to output reg:
860 if (key
->separate_specular
||
861 unit
!= p
->last_tex_stage
||
864 dest
= get_temp( p
);
866 dest
= make_ureg(PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
868 /* Emit the RGB and A combine ops
870 if (key
->unit
[unit
].ModeRGB
== key
->unit
[unit
].ModeA
&&
871 args_match(key
, unit
)) {
872 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
874 key
->unit
[unit
].NumArgsRGB
,
875 key
->unit
[unit
].ModeRGB
,
876 key
->unit
[unit
].OptRGB
);
878 else if (key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA_EXT
||
879 key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA
) {
881 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
883 key
->unit
[unit
].NumArgsRGB
,
884 key
->unit
[unit
].ModeRGB
,
885 key
->unit
[unit
].OptRGB
);
888 /* Need to do something to stop from re-emitting identical
889 * argument calculations here:
891 out
= emit_combine( p
, dest
, WRITEMASK_XYZ
, saturate
,
893 key
->unit
[unit
].NumArgsRGB
,
894 key
->unit
[unit
].ModeRGB
,
895 key
->unit
[unit
].OptRGB
);
896 out
= emit_combine( p
, dest
, WRITEMASK_W
, saturate
,
898 key
->unit
[unit
].NumArgsA
,
899 key
->unit
[unit
].ModeA
,
900 key
->unit
[unit
].OptA
);
903 /* Deal with the final shift:
905 if (alpha_shift
|| rgb_shift
) {
906 if (rgb_shift
== alpha_shift
) {
907 shift
= register_scalar_const(p
, (GLfloat
)(1<<rgb_shift
));
910 shift
= register_const4f(p
,
911 (GLfloat
)(1<<rgb_shift
),
912 (GLfloat
)(1<<rgb_shift
),
913 (GLfloat
)(1<<rgb_shift
),
914 (GLfloat
)(1<<alpha_shift
));
916 return emit_arith( p
, OPCODE_MUL
, dest
, WRITEMASK_XYZW
,
917 saturate
, out
, shift
, undef
);
925 * Generate instruction for getting a texture source term.
927 static void load_texture( struct texenv_fragment_program
*p
, GLuint unit
)
929 if (is_undef(p
->src_texture
[unit
])) {
930 GLuint dim
= p
->state
->unit
[unit
].source_index
;
931 struct ureg texcoord
= register_input(p
, FRAG_ATTRIB_TEX0
+unit
);
932 struct ureg tmp
= get_tex_temp( p
);
934 if (dim
== TEXTURE_UNKNOWN_INDEX
)
935 program_error(p
, "TexSrcBit");
937 /* TODO: Use D0_MASK_XY where possible.
939 if (p
->state
->unit
[unit
].enabled
) {
940 p
->src_texture
[unit
] = emit_texld( p
, OPCODE_TXP
,
942 unit
, dim
, texcoord
);
943 p
->program
->Base
.SamplersUsed
|= (1 << unit
);
944 /* This identity mapping should already be in place
945 * (see _mesa_init_program_struct()) but let's be safe.
947 p
->program
->Base
.SamplerUnits
[unit
] = unit
;
950 p
->src_texture
[unit
] = get_zero(p
);
954 static GLboolean
load_texenv_source( struct texenv_fragment_program
*p
,
955 GLuint src
, GLuint unit
)
959 load_texture(p
, unit
);
970 load_texture(p
, src
- SRC_TEXTURE0
);
982 * Generate instructions for loading all texture source terms.
985 load_texunit_sources( struct texenv_fragment_program
*p
, int unit
)
987 struct state_key
*key
= p
->state
;
990 for (i
= 0; i
< key
->unit
[unit
].NumArgsRGB
; i
++) {
991 load_texenv_source( p
, key
->unit
[unit
].OptRGB
[i
].Source
, unit
);
994 for (i
= 0; i
< key
->unit
[unit
].NumArgsA
; i
++) {
995 load_texenv_source( p
, key
->unit
[unit
].OptA
[i
].Source
, unit
);
1003 * Generate a new fragment program which implements the context's
1004 * current texture env/combine mode.
1007 create_new_program(GLcontext
*ctx
, struct state_key
*key
,
1008 struct gl_fragment_program
*program
)
1010 struct prog_instruction instBuffer
[MAX_INSTRUCTIONS
];
1011 struct texenv_fragment_program p
;
1013 struct ureg cf
, out
;
1015 _mesa_memset(&p
, 0, sizeof(p
));
1018 p
.program
= program
;
1020 /* During code generation, use locally-allocated instruction buffer,
1021 * then alloc dynamic storage below.
1023 p
.program
->Base
.Instructions
= instBuffer
;
1024 p
.program
->Base
.Target
= GL_FRAGMENT_PROGRAM_ARB
;
1025 p
.program
->Base
.NumTexIndirections
= 1; /* correct? */
1026 p
.program
->Base
.NumTexInstructions
= 0;
1027 p
.program
->Base
.NumAluInstructions
= 0;
1028 p
.program
->Base
.String
= NULL
;
1029 p
.program
->Base
.NumInstructions
=
1030 p
.program
->Base
.NumTemporaries
=
1031 p
.program
->Base
.NumParameters
=
1032 p
.program
->Base
.NumAttributes
= p
.program
->Base
.NumAddressRegs
= 0;
1033 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1035 p
.program
->Base
.InputsRead
= 0;
1036 p
.program
->Base
.OutputsWritten
= 1 << FRAG_RESULT_COLR
;
1038 for (unit
= 0; unit
< MAX_TEXTURE_UNITS
; unit
++)
1039 p
.src_texture
[unit
] = undef
;
1041 p
.src_previous
= undef
;
1046 p
.last_tex_stage
= 0;
1047 release_temps(ctx
, &p
);
1049 if (key
->enabled_units
) {
1050 /* First pass - to support texture_env_crossbar, first identify
1051 * all referenced texture sources and emit texld instructions
1054 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1055 if (key
->unit
[unit
].enabled
) {
1056 load_texunit_sources( &p
, unit
);
1057 p
.last_tex_stage
= unit
;
1060 /* Second pass - emit combine instructions to build final color:
1062 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1063 if (key
->enabled_units
& (1<<unit
)) {
1064 p
.src_previous
= emit_texenv( &p
, unit
);
1065 release_temps(ctx
, &p
); /* release all temps */
1069 cf
= get_source( &p
, SRC_PREVIOUS
, 0 );
1070 out
= make_ureg( PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
1072 if (key
->separate_specular
) {
1073 /* Emit specular add.
1075 struct ureg s
= register_input(&p
, FRAG_ATTRIB_COL1
);
1076 emit_arith( &p
, OPCODE_ADD
, out
, WRITEMASK_XYZ
, 0, cf
, s
, undef
);
1077 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_W
, 0, cf
, undef
, undef
);
1079 else if (_mesa_memcmp(&cf
, &out
, sizeof(cf
)) != 0) {
1080 /* Will wind up in here if no texture enabled or a couple of
1081 * other scenarios (GL_REPLACE for instance).
1083 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_XYZW
, 0, cf
, undef
, undef
);
1088 emit_arith( &p
, OPCODE_END
, undef
, WRITEMASK_XYZW
, 0, undef
, undef
, undef
);
1090 if (key
->fog_enabled
) {
1091 /* Pull fog mode from GLcontext, the value in the state key is
1092 * a reduced value and not what is expected in FogOption
1094 p
.program
->FogOption
= ctx
->Fog
.Mode
;
1095 p
.program
->Base
.InputsRead
|= FRAG_BIT_FOGC
; /* XXX new */
1097 p
.program
->FogOption
= GL_NONE
;
1099 if (p
.program
->Base
.NumTexIndirections
> ctx
->Const
.FragmentProgram
.MaxTexIndirections
)
1100 program_error(&p
, "Exceeded max nr indirect texture lookups");
1102 if (p
.program
->Base
.NumTexInstructions
> ctx
->Const
.FragmentProgram
.MaxTexInstructions
)
1103 program_error(&p
, "Exceeded max TEX instructions");
1105 if (p
.program
->Base
.NumAluInstructions
> ctx
->Const
.FragmentProgram
.MaxAluInstructions
)
1106 program_error(&p
, "Exceeded max ALU instructions");
1108 ASSERT(p
.program
->Base
.NumInstructions
<= MAX_INSTRUCTIONS
);
1110 /* Allocate final instruction array */
1111 p
.program
->Base
.Instructions
1112 = _mesa_alloc_instructions(p
.program
->Base
.NumInstructions
);
1113 if (!p
.program
->Base
.Instructions
) {
1114 _mesa_error(ctx
, GL_OUT_OF_MEMORY
,
1115 "generating tex env program");
1118 _mesa_copy_instructions(p
.program
->Base
.Instructions
, instBuffer
,
1119 p
.program
->Base
.NumInstructions
);
1121 if (p
.program
->FogOption
) {
1122 _mesa_append_fog_code(ctx
, p
.program
);
1123 p
.program
->FogOption
= GL_NONE
;
1127 /* Notify driver the fragment program has (actually) changed.
1129 if (ctx
->Driver
.ProgramStringNotify
) {
1130 ctx
->Driver
.ProgramStringNotify( ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1135 _mesa_print_program(&p
.program
->Base
);
1142 * Return a fragment program which implements the current
1143 * fixed-function texture, fog and color-sum operations.
1145 struct gl_fragment_program
*
1146 _mesa_get_fixed_func_fragment_program(GLcontext
*ctx
)
1148 struct gl_fragment_program
*prog
;
1149 struct state_key key
;
1151 make_state_key(ctx
, &key
);
1153 prog
= (struct gl_fragment_program
*)
1154 _mesa_search_program_cache(ctx
->FragmentProgram
.Cache
,
1158 prog
= (struct gl_fragment_program
*)
1159 ctx
->Driver
.NewProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
, 0);
1161 create_new_program(ctx
, &key
, prog
);
1163 _mesa_program_cache_insert(ctx
, ctx
->FragmentProgram
.Cache
,
1164 &key
, sizeof(key
), &prog
->Base
);
1173 * If _MaintainTexEnvProgram is set we'll generate a fragment program that
1174 * implements the current texture env/combine mode.
1175 * This function generates that program and puts it into effect.
1178 _mesa_UpdateTexEnvProgram( GLcontext
*ctx
)
1180 const struct gl_fragment_program
*prev
= ctx
->FragmentProgram
._Current
;
1182 ASSERT(ctx
->FragmentProgram
._MaintainTexEnvProgram
);
1184 /* If a conventional fragment program/shader isn't in effect... */
1185 if (!ctx
->FragmentProgram
._Enabled
&&
1186 (!ctx
->Shader
.CurrentProgram
||
1187 !ctx
->Shader
.CurrentProgram
->FragmentProgram
) ) {
1189 ctx
->FragmentProgram
._Current
1190 = ctx
->FragmentProgram
._TexEnvProgram
1191 = _mesa_get_fixed_func_fragment_program(ctx
);
1194 /* Tell the driver about the change. Could define a new target for
1197 if (ctx
->FragmentProgram
._Current
!= prev
&& ctx
->Driver
.BindProgram
) {
1198 ctx
->Driver
.BindProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1199 (struct gl_program
*) ctx
->FragmentProgram
._Current
);