1 /**************************************************************************
3 * Copyright 2003 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_instruction.h"
33 #include "shader/prog_print.h"
34 #include "shader/prog_statevars.h"
35 #include "texenvprogram.h"
38 * This MAX is probably a bit generous, but that's OK. There can be
39 * up to four instructions per texture unit (TEX + 3 for combine),
40 * then there's fog and specular add.
42 #define MAX_INSTRUCTIONS ((MAX_TEXTURE_UNITS * 4) + 12)
44 #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
52 GLbitfield enabled_units
;
53 GLuint separate_specular
:1;
59 GLuint source_index
:3; /* one of TEXTURE_1D/2D/3D/CUBE/RECT_INDEX */
60 GLuint ScaleShiftRGB
:2;
65 struct mode_opt OptRGB
[3];
69 struct mode_opt OptA
[3];
78 static GLuint
translate_fog_mode( GLenum mode
)
81 case GL_LINEAR
: return FOG_LINEAR
;
82 case GL_EXP
: return FOG_EXP
;
83 case GL_EXP2
: return FOG_EXP2
;
84 default: return FOG_UNKNOWN
;
88 #define OPR_SRC_COLOR 0
89 #define OPR_ONE_MINUS_SRC_COLOR 1
90 #define OPR_SRC_ALPHA 2
91 #define OPR_ONE_MINUS_SRC_ALPHA 3
96 static GLuint
translate_operand( GLenum operand
)
99 case GL_SRC_COLOR
: return OPR_SRC_COLOR
;
100 case GL_ONE_MINUS_SRC_COLOR
: return OPR_ONE_MINUS_SRC_COLOR
;
101 case GL_SRC_ALPHA
: return OPR_SRC_ALPHA
;
102 case GL_ONE_MINUS_SRC_ALPHA
: return OPR_ONE_MINUS_SRC_ALPHA
;
103 case GL_ZERO
: return OPR_ZERO
;
104 case GL_ONE
: return OPR_ONE
;
105 default: return OPR_UNKNOWN
;
109 #define SRC_TEXTURE 0
110 #define SRC_TEXTURE0 1
111 #define SRC_TEXTURE1 2
112 #define SRC_TEXTURE2 3
113 #define SRC_TEXTURE3 4
114 #define SRC_TEXTURE4 5
115 #define SRC_TEXTURE5 6
116 #define SRC_TEXTURE6 7
117 #define SRC_TEXTURE7 8
118 #define SRC_CONSTANT 9
119 #define SRC_PRIMARY_COLOR 10
120 #define SRC_PREVIOUS 11
121 #define SRC_UNKNOWN 15
123 static GLuint
translate_source( GLenum src
)
126 case GL_TEXTURE
: return SRC_TEXTURE
;
134 case GL_TEXTURE7
: return SRC_TEXTURE0
+ (src
- GL_TEXTURE0
);
135 case GL_CONSTANT
: return SRC_CONSTANT
;
136 case GL_PRIMARY_COLOR
: return SRC_PRIMARY_COLOR
;
137 case GL_PREVIOUS
: return SRC_PREVIOUS
;
138 default: return SRC_UNKNOWN
;
142 #define MODE_REPLACE 0
143 #define MODE_MODULATE 1
145 #define MODE_ADD_SIGNED 3
146 #define MODE_INTERPOLATE 4
147 #define MODE_SUBTRACT 5
148 #define MODE_DOT3_RGB 6
149 #define MODE_DOT3_RGB_EXT 7
150 #define MODE_DOT3_RGBA 8
151 #define MODE_DOT3_RGBA_EXT 9
152 #define MODE_MODULATE_ADD_ATI 10
153 #define MODE_MODULATE_SIGNED_ADD_ATI 11
154 #define MODE_MODULATE_SUBTRACT_ATI 12
155 #define MODE_UNKNOWN 15
157 static GLuint
translate_mode( GLenum mode
)
160 case GL_REPLACE
: return MODE_REPLACE
;
161 case GL_MODULATE
: return MODE_MODULATE
;
162 case GL_ADD
: return MODE_ADD
;
163 case GL_ADD_SIGNED
: return MODE_ADD_SIGNED
;
164 case GL_INTERPOLATE
: return MODE_INTERPOLATE
;
165 case GL_SUBTRACT
: return MODE_SUBTRACT
;
166 case GL_DOT3_RGB
: return MODE_DOT3_RGB
;
167 case GL_DOT3_RGB_EXT
: return MODE_DOT3_RGB_EXT
;
168 case GL_DOT3_RGBA
: return MODE_DOT3_RGBA
;
169 case GL_DOT3_RGBA_EXT
: return MODE_DOT3_RGBA_EXT
;
170 case GL_MODULATE_ADD_ATI
: return MODE_MODULATE_ADD_ATI
;
171 case GL_MODULATE_SIGNED_ADD_ATI
: return MODE_MODULATE_SIGNED_ADD_ATI
;
172 case GL_MODULATE_SUBTRACT_ATI
: return MODE_MODULATE_SUBTRACT_ATI
;
173 default: return MODE_UNKNOWN
;
177 #define TEXTURE_UNKNOWN_INDEX 7
178 static GLuint
translate_tex_src_bit( GLbitfield bit
)
181 case TEXTURE_1D_BIT
: return TEXTURE_1D_INDEX
;
182 case TEXTURE_2D_BIT
: return TEXTURE_2D_INDEX
;
183 case TEXTURE_RECT_BIT
: return TEXTURE_RECT_INDEX
;
184 case TEXTURE_3D_BIT
: return TEXTURE_3D_INDEX
;
185 case TEXTURE_CUBE_BIT
: return TEXTURE_CUBE_INDEX
;
186 default: return TEXTURE_UNKNOWN_INDEX
;
191 * Examine current texture environment state and generate a unique
192 * key to identify it.
194 static void make_state_key( GLcontext
*ctx
, struct state_key
*key
)
198 memset(key
, 0, sizeof(*key
));
200 for (i
=0;i
<MAX_TEXTURE_UNITS
;i
++) {
201 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
203 if (!texUnit
->_ReallyEnabled
|| !texUnit
->Enabled
)
206 key
->unit
[i
].enabled
= 1;
207 key
->enabled_units
|= (1<<i
);
209 key
->unit
[i
].source_index
=
210 translate_tex_src_bit(texUnit
->_ReallyEnabled
);
212 key
->unit
[i
].NumArgsRGB
= texUnit
->_CurrentCombine
->_NumArgsRGB
;
213 key
->unit
[i
].NumArgsA
= texUnit
->_CurrentCombine
->_NumArgsA
;
215 key
->unit
[i
].ModeRGB
=
216 translate_mode(texUnit
->_CurrentCombine
->ModeRGB
);
218 translate_mode(texUnit
->_CurrentCombine
->ModeA
);
220 key
->unit
[i
].ScaleShiftRGB
= texUnit
->_CurrentCombine
->ScaleShiftRGB
;
221 key
->unit
[i
].ScaleShiftA
= texUnit
->_CurrentCombine
->ScaleShiftA
;
224 key
->unit
[i
].OptRGB
[j
].Operand
=
225 translate_operand(texUnit
->_CurrentCombine
->OperandRGB
[j
]);
226 key
->unit
[i
].OptA
[j
].Operand
=
227 translate_operand(texUnit
->_CurrentCombine
->OperandA
[j
]);
228 key
->unit
[i
].OptRGB
[j
].Source
=
229 translate_source(texUnit
->_CurrentCombine
->SourceRGB
[j
]);
230 key
->unit
[i
].OptA
[j
].Source
=
231 translate_source(texUnit
->_CurrentCombine
->SourceA
[j
]);
235 if (ctx
->_TriangleCaps
& DD_SEPARATE_SPECULAR
)
236 key
->separate_specular
= 1;
238 if (ctx
->Fog
.Enabled
) {
239 key
->fog_enabled
= 1;
240 key
->fog_mode
= translate_fog_mode(ctx
->Fog
.Mode
);
244 /* Use uregs to represent registers internally, translate to Mesa's
245 * expected formats on emit.
247 * NOTE: These are passed by value extensively in this file rather
248 * than as usual by pointer reference. If this disturbs you, try
249 * remembering they are just 32bits in size.
251 * GCC is smart enough to deal with these dword-sized structures in
252 * much the same way as if I had defined them as dwords and was using
253 * macros to access and set the fields. This is much nicer and easier
266 static const struct ureg undef
= {
277 /* State used to build the fragment program:
279 struct texenv_fragment_program
{
280 struct gl_fragment_program
*program
;
282 struct state_key
*state
;
284 GLbitfield alu_temps
; /* Track texture indirections, see spec. */
285 GLbitfield temps_output
; /* Track texture indirections, see spec. */
286 GLbitfield temp_in_use
; /* Tracks temporary regs which are in use. */
289 struct ureg src_texture
[MAX_TEXTURE_UNITS
];
290 /* Reg containing each texture unit's sampled texture color,
294 struct ureg src_previous
; /* Reg containing color from previous
295 * stage. May need to be decl'd.
298 GLuint last_tex_stage
; /* Number of last enabled texture unit */
307 static struct ureg
make_ureg(GLuint file
, GLuint idx
)
315 reg
.swz
= SWIZZLE_NOOP
;
320 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
322 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
325 GET_SWZ(reg
.swz
, w
));
330 static struct ureg
swizzle1( struct ureg reg
, int x
)
332 return swizzle(reg
, x
, x
, x
, x
);
335 static struct ureg
negate( struct ureg reg
)
341 static GLboolean
is_undef( struct ureg reg
)
343 return reg
.file
== PROGRAM_UNDEFINED
;
347 static struct ureg
get_temp( struct texenv_fragment_program
*p
)
351 /* First try and reuse temps which have been used already:
353 bit
= _mesa_ffs( ~p
->temp_in_use
& p
->alu_temps
);
355 /* Then any unused temporary:
358 bit
= _mesa_ffs( ~p
->temp_in_use
);
361 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
365 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
366 p
->program
->Base
.NumTemporaries
= bit
;
368 p
->temp_in_use
|= 1<<(bit
-1);
369 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
372 static struct ureg
get_tex_temp( struct texenv_fragment_program
*p
)
376 /* First try to find availble temp not previously used (to avoid
377 * starting a new texture indirection). According to the spec, the
378 * ~p->temps_output isn't necessary, but will keep it there for
381 bit
= _mesa_ffs( ~p
->temp_in_use
& ~p
->alu_temps
& ~p
->temps_output
);
383 /* Then any unused temporary:
386 bit
= _mesa_ffs( ~p
->temp_in_use
);
389 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
393 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
394 p
->program
->Base
.NumTemporaries
= bit
;
396 p
->temp_in_use
|= 1<<(bit
-1);
397 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
401 static void release_temps( struct texenv_fragment_program
*p
)
403 GLuint max_temp
= p
->ctx
->Const
.FragmentProgram
.MaxTemps
;
405 /* KW: To support tex_env_crossbar, don't release the registers in
408 if (max_temp
>= sizeof(int) * 8)
409 p
->temp_in_use
= p
->temps_output
;
411 p
->temp_in_use
= ~((1<<max_temp
)-1) | p
->temps_output
;
415 static struct ureg
register_param5( struct texenv_fragment_program
*p
,
422 gl_state_index tokens
[STATE_LENGTH
];
429 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
430 return make_ureg(PROGRAM_STATE_VAR
, idx
);
434 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
435 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
436 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
437 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
440 static struct ureg
register_input( struct texenv_fragment_program
*p
, GLuint input
)
442 p
->program
->Base
.InputsRead
|= (1 << input
);
443 return make_ureg(PROGRAM_INPUT
, input
);
447 static void emit_arg( struct prog_src_register
*reg
,
450 reg
->File
= ureg
.file
;
451 reg
->Index
= ureg
.idx
;
452 reg
->Swizzle
= ureg
.swz
;
453 reg
->NegateBase
= ureg
.negatebase
? 0xf : 0x0;
455 reg
->NegateAbs
= ureg
.negateabs
;
458 static void emit_dst( struct prog_dst_register
*dst
,
459 struct ureg ureg
, GLuint mask
)
461 dst
->File
= ureg
.file
;
462 dst
->Index
= ureg
.idx
;
463 dst
->WriteMask
= mask
;
464 dst
->CondMask
= COND_TR
; /* always pass cond test */
465 dst
->CondSwizzle
= SWIZZLE_NOOP
;
468 static struct prog_instruction
*
469 emit_op(struct texenv_fragment_program
*p
,
478 GLuint nr
= p
->program
->Base
.NumInstructions
++;
479 struct prog_instruction
*inst
= &p
->program
->Base
.Instructions
[nr
];
481 assert(nr
< MAX_INSTRUCTIONS
);
483 _mesa_init_instructions(inst
, 1);
486 emit_arg( &inst
->SrcReg
[0], src0
);
487 emit_arg( &inst
->SrcReg
[1], src1
);
488 emit_arg( &inst
->SrcReg
[2], src2
);
490 inst
->SaturateMode
= saturate
? SATURATE_ZERO_ONE
: SATURATE_OFF
;
492 emit_dst( &inst
->DstReg
, dest
, mask
);
494 /* Accounting for indirection tracking:
496 if (dest
.file
== PROGRAM_TEMPORARY
)
497 p
->temps_output
|= 1 << dest
.idx
;
503 static struct ureg
emit_arith( struct texenv_fragment_program
*p
,
512 emit_op(p
, op
, dest
, mask
, saturate
, src0
, src1
, src2
);
514 /* Accounting for indirection tracking:
516 if (src0
.file
== PROGRAM_TEMPORARY
)
517 p
->alu_temps
|= 1 << src0
.idx
;
519 if (!is_undef(src1
) && src1
.file
== PROGRAM_TEMPORARY
)
520 p
->alu_temps
|= 1 << src1
.idx
;
522 if (!is_undef(src2
) && src2
.file
== PROGRAM_TEMPORARY
)
523 p
->alu_temps
|= 1 << src2
.idx
;
525 if (dest
.file
== PROGRAM_TEMPORARY
)
526 p
->alu_temps
|= 1 << dest
.idx
;
528 p
->program
->Base
.NumAluInstructions
++;
532 static struct ureg
emit_texld( struct texenv_fragment_program
*p
,
540 struct prog_instruction
*inst
= emit_op( p
, op
,
542 GL_FALSE
, /* don't saturate? */
547 inst
->TexSrcTarget
= tex_idx
;
548 inst
->TexSrcUnit
= tex_unit
;
550 p
->program
->Base
.NumTexInstructions
++;
552 /* Is this a texture indirection?
554 if ((coord
.file
== PROGRAM_TEMPORARY
&&
555 (p
->temps_output
& (1<<coord
.idx
))) ||
556 (dest
.file
== PROGRAM_TEMPORARY
&&
557 (p
->alu_temps
& (1<<dest
.idx
)))) {
558 p
->program
->Base
.NumTexIndirections
++;
559 p
->temps_output
= 1<<coord
.idx
;
561 assert(0); /* KW: texture env crossbar */
568 static struct ureg
register_const4f( struct texenv_fragment_program
*p
,
580 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
582 ASSERT(swizzle
== SWIZZLE_NOOP
);
583 return make_ureg(PROGRAM_STATE_VAR
, idx
);
586 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
587 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
588 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
589 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
592 static struct ureg
get_one( struct texenv_fragment_program
*p
)
594 if (is_undef(p
->one
))
595 p
->one
= register_scalar_const(p
, 1.0);
599 static struct ureg
get_half( struct texenv_fragment_program
*p
)
601 if (is_undef(p
->half
))
602 p
->half
= register_scalar_const(p
, 0.5);
606 static struct ureg
get_zero( struct texenv_fragment_program
*p
)
608 if (is_undef(p
->zero
))
609 p
->zero
= register_scalar_const(p
, 0.0);
614 static void program_error( struct texenv_fragment_program
*p
, const char *msg
)
616 _mesa_problem(NULL
, msg
);
620 static struct ureg
get_source( struct texenv_fragment_program
*p
,
621 GLuint src
, GLuint unit
)
625 assert(!is_undef(p
->src_texture
[unit
]));
626 return p
->src_texture
[unit
];
636 assert(!is_undef(p
->src_texture
[src
- SRC_TEXTURE0
]));
637 return p
->src_texture
[src
- SRC_TEXTURE0
];
640 return register_param2(p
, STATE_TEXENV_COLOR
, unit
);
642 case SRC_PRIMARY_COLOR
:
643 return register_input(p
, FRAG_ATTRIB_COL0
);
647 if (is_undef(p
->src_previous
))
648 return register_input(p
, FRAG_ATTRIB_COL0
);
650 return p
->src_previous
;
654 static struct ureg
emit_combine_source( struct texenv_fragment_program
*p
,
660 struct ureg arg
, src
, one
;
662 src
= get_source(p
, source
, unit
);
665 case OPR_ONE_MINUS_SRC_COLOR
:
667 * Emit tmp = 1.0 - arg.xyzw
671 return emit_arith( p
, OPCODE_SUB
, arg
, mask
, 0, one
, src
, undef
);
674 if (mask
== WRITEMASK_W
)
677 return swizzle1( src
, SWIZZLE_W
);
678 case OPR_ONE_MINUS_SRC_ALPHA
:
680 * Emit tmp = 1.0 - arg.wwww
684 return emit_arith(p
, OPCODE_SUB
, arg
, mask
, 0,
685 one
, swizzle1(src
, SWIZZLE_W
), undef
);
696 static GLboolean
args_match( struct state_key
*key
, GLuint unit
)
698 GLuint i
, nr
= key
->unit
[unit
].NumArgsRGB
;
700 for (i
= 0 ; i
< nr
; i
++) {
701 if (key
->unit
[unit
].OptA
[i
].Source
!= key
->unit
[unit
].OptRGB
[i
].Source
)
704 switch(key
->unit
[unit
].OptA
[i
].Operand
) {
706 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
714 case OPR_ONE_MINUS_SRC_ALPHA
:
715 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
716 case OPR_ONE_MINUS_SRC_COLOR
:
717 case OPR_ONE_MINUS_SRC_ALPHA
:
724 return GL_FALSE
; /* impossible */
731 static struct ureg
emit_combine( struct texenv_fragment_program
*p
,
738 const struct mode_opt
*opt
)
741 struct ureg tmp
, half
;
744 tmp
= undef
; /* silence warning (bug 5318) */
746 for (i
= 0; i
< nr
; i
++)
747 src
[i
] = emit_combine_source( p
, mask
, unit
, opt
[i
].Source
, opt
[i
].Operand
);
751 if (mask
== WRITEMASK_XYZW
&& !saturate
)
754 return emit_arith( p
, OPCODE_MOV
, dest
, mask
, saturate
, src
[0], undef
, undef
);
756 return emit_arith( p
, OPCODE_MUL
, dest
, mask
, saturate
,
757 src
[0], src
[1], undef
);
759 return emit_arith( p
, OPCODE_ADD
, dest
, mask
, saturate
,
760 src
[0], src
[1], undef
);
761 case MODE_ADD_SIGNED
:
767 emit_arith( p
, OPCODE_ADD
, tmp
, mask
, 0, src
[0], src
[1], undef
);
768 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp
, half
, undef
);
770 case MODE_INTERPOLATE
:
771 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
773 return emit_arith( p
, OPCODE_LRP
, dest
, mask
, saturate
, src
[2], src
[0], src
[1] );
776 return emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, src
[0], src
[1], undef
);
779 case MODE_DOT3_RGBA_EXT
:
780 case MODE_DOT3_RGB_EXT
:
781 case MODE_DOT3_RGB
: {
782 struct ureg tmp0
= get_temp( p
);
783 struct ureg tmp1
= get_temp( p
);
784 struct ureg neg1
= register_scalar_const(p
, -1);
785 struct ureg two
= register_scalar_const(p
, 2);
790 * dst = tmp0 dot3 tmp1
792 emit_arith( p
, OPCODE_MAD
, tmp0
, WRITEMASK_XYZW
, 0,
795 if (_mesa_memcmp(&src
[0], &src
[1], sizeof(struct ureg
)) == 0)
798 emit_arith( p
, OPCODE_MAD
, tmp1
, WRITEMASK_XYZW
, 0,
800 emit_arith( p
, OPCODE_DP3
, dest
, mask
, saturate
, tmp0
, tmp1
, undef
);
803 case MODE_MODULATE_ADD_ATI
:
804 /* Arg0 * Arg2 + Arg1 */
805 return emit_arith( p
, OPCODE_MAD
, dest
, mask
, saturate
,
806 src
[0], src
[2], src
[1] );
807 case MODE_MODULATE_SIGNED_ADD_ATI
: {
808 /* Arg0 * Arg2 + Arg1 - 0.5 */
809 struct ureg tmp0
= get_temp(p
);
811 emit_arith( p
, OPCODE_MAD
, tmp0
, mask
, 0, src
[0], src
[2], src
[1] );
812 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp0
, half
, undef
);
815 case MODE_MODULATE_SUBTRACT_ATI
:
816 /* Arg0 * Arg2 - Arg1 */
817 emit_arith( p
, OPCODE_MAD
, dest
, mask
, 0, src
[0], src
[2], negate(src
[1]) );
826 * Generate instructions for one texture unit's env/combiner mode.
829 emit_texenv(struct texenv_fragment_program
*p
, GLuint unit
)
831 struct state_key
*key
= p
->state
;
832 GLboolean saturate
= (unit
< p
->last_tex_stage
);
833 GLuint rgb_shift
, alpha_shift
;
834 struct ureg out
, shift
;
837 if (!key
->unit
[unit
].enabled
) {
838 return get_source(p
, SRC_PREVIOUS
, 0);
841 switch (key
->unit
[unit
].ModeRGB
) {
842 case MODE_DOT3_RGB_EXT
:
843 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
846 case MODE_DOT3_RGBA_EXT
:
851 rgb_shift
= key
->unit
[unit
].ScaleShiftRGB
;
852 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
856 /* If this is the very last calculation, emit direct to output reg:
858 if (key
->separate_specular
||
859 unit
!= p
->last_tex_stage
||
862 dest
= get_temp( p
);
864 dest
= make_ureg(PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
866 /* Emit the RGB and A combine ops
868 if (key
->unit
[unit
].ModeRGB
== key
->unit
[unit
].ModeA
&&
869 args_match(key
, unit
)) {
870 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
872 key
->unit
[unit
].NumArgsRGB
,
873 key
->unit
[unit
].ModeRGB
,
874 key
->unit
[unit
].OptRGB
);
876 else if (key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA_EXT
||
877 key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA
) {
879 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
881 key
->unit
[unit
].NumArgsRGB
,
882 key
->unit
[unit
].ModeRGB
,
883 key
->unit
[unit
].OptRGB
);
886 /* Need to do something to stop from re-emitting identical
887 * argument calculations here:
889 out
= emit_combine( p
, dest
, WRITEMASK_XYZ
, saturate
,
891 key
->unit
[unit
].NumArgsRGB
,
892 key
->unit
[unit
].ModeRGB
,
893 key
->unit
[unit
].OptRGB
);
894 out
= emit_combine( p
, dest
, WRITEMASK_W
, saturate
,
896 key
->unit
[unit
].NumArgsA
,
897 key
->unit
[unit
].ModeA
,
898 key
->unit
[unit
].OptA
);
901 /* Deal with the final shift:
903 if (alpha_shift
|| rgb_shift
) {
904 if (rgb_shift
== alpha_shift
) {
905 shift
= register_scalar_const(p
, 1<<rgb_shift
);
908 shift
= register_const4f(p
,
914 return emit_arith( p
, OPCODE_MUL
, dest
, WRITEMASK_XYZW
,
915 saturate
, out
, shift
, undef
);
923 * Generate instruction for getting a texture source term.
925 static void load_texture( struct texenv_fragment_program
*p
, GLuint unit
)
927 if (is_undef(p
->src_texture
[unit
])) {
928 GLuint dim
= p
->state
->unit
[unit
].source_index
;
929 struct ureg texcoord
= register_input(p
, FRAG_ATTRIB_TEX0
+unit
);
930 struct ureg tmp
= get_tex_temp( p
);
932 if (dim
== TEXTURE_UNKNOWN_INDEX
)
933 program_error(p
, "TexSrcBit");
935 /* TODO: Use D0_MASK_XY where possible.
937 if (p
->state
->unit
[unit
].enabled
)
938 p
->src_texture
[unit
] = emit_texld( p
, OPCODE_TXP
,
940 unit
, dim
, texcoord
);
942 p
->src_texture
[unit
] = get_zero(p
);
946 static GLboolean
load_texenv_source( struct texenv_fragment_program
*p
,
947 GLuint src
, GLuint unit
)
951 load_texture(p
, unit
);
962 load_texture(p
, src
- SRC_TEXTURE0
);
974 * Generate instructions for loading all texture source terms.
977 load_texunit_sources( struct texenv_fragment_program
*p
, int unit
)
979 struct state_key
*key
= p
->state
;
982 for (i
= 0; i
< key
->unit
[unit
].NumArgsRGB
; i
++) {
983 load_texenv_source( p
, key
->unit
[unit
].OptRGB
[i
].Source
, unit
);
986 for (i
= 0; i
< key
->unit
[unit
].NumArgsA
; i
++) {
987 load_texenv_source( p
, key
->unit
[unit
].OptA
[i
].Source
, unit
);
995 * Generate a new fragment program which implements the context's
996 * current texture env/combine mode.
999 create_new_program(GLcontext
*ctx
, struct state_key
*key
,
1000 struct gl_fragment_program
*program
)
1002 struct prog_instruction instBuffer
[MAX_INSTRUCTIONS
];
1003 struct texenv_fragment_program p
;
1005 struct ureg cf
, out
;
1007 _mesa_memset(&p
, 0, sizeof(p
));
1010 p
.program
= program
;
1012 /* During code generation, use locally-allocated instruction buffer,
1013 * then alloc dynamic storage below.
1015 p
.program
->Base
.Instructions
= instBuffer
;
1016 p
.program
->Base
.Target
= GL_FRAGMENT_PROGRAM_ARB
;
1017 p
.program
->Base
.NumTexIndirections
= 1; /* correct? */
1018 p
.program
->Base
.NumTexInstructions
= 0;
1019 p
.program
->Base
.NumAluInstructions
= 0;
1020 p
.program
->Base
.String
= NULL
;
1021 p
.program
->Base
.NumInstructions
=
1022 p
.program
->Base
.NumTemporaries
=
1023 p
.program
->Base
.NumParameters
=
1024 p
.program
->Base
.NumAttributes
= p
.program
->Base
.NumAddressRegs
= 0;
1025 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1027 p
.program
->Base
.InputsRead
= 0;
1028 p
.program
->Base
.OutputsWritten
= 1 << FRAG_RESULT_COLR
;
1030 for (unit
= 0; unit
< MAX_TEXTURE_UNITS
; unit
++)
1031 p
.src_texture
[unit
] = undef
;
1033 p
.src_previous
= undef
;
1038 p
.last_tex_stage
= 0;
1041 if (key
->enabled_units
) {
1042 /* First pass - to support texture_env_crossbar, first identify
1043 * all referenced texture sources and emit texld instructions
1046 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1047 if (key
->unit
[unit
].enabled
) {
1048 load_texunit_sources( &p
, unit
);
1049 p
.last_tex_stage
= unit
;
1052 /* Second pass - emit combine instructions to build final color:
1054 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1055 if (key
->enabled_units
& (1<<unit
)) {
1056 p
.src_previous
= emit_texenv( &p
, unit
);
1057 release_temps(&p
); /* release all temps */
1061 cf
= get_source( &p
, SRC_PREVIOUS
, 0 );
1062 out
= make_ureg( PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
1064 if (key
->separate_specular
) {
1065 /* Emit specular add.
1067 struct ureg s
= register_input(&p
, FRAG_ATTRIB_COL1
);
1068 emit_arith( &p
, OPCODE_ADD
, out
, WRITEMASK_XYZ
, 0, cf
, s
, undef
);
1069 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_W
, 0, cf
, undef
, undef
);
1071 else if (_mesa_memcmp(&cf
, &out
, sizeof(cf
)) != 0) {
1072 /* Will wind up in here if no texture enabled or a couple of
1073 * other scenarios (GL_REPLACE for instance).
1075 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_XYZW
, 0, cf
, undef
, undef
);
1080 emit_arith( &p
, OPCODE_END
, undef
, WRITEMASK_XYZW
, 0, undef
, undef
, undef
);
1082 if (key
->fog_enabled
) {
1083 /* Pull fog mode from GLcontext, the value in the state key is
1084 * a reduced value and not what is expected in FogOption
1086 p
.program
->FogOption
= ctx
->Fog
.Mode
;
1088 p
.program
->FogOption
= GL_NONE
;
1090 if (p
.program
->Base
.NumTexIndirections
> ctx
->Const
.FragmentProgram
.MaxTexIndirections
)
1091 program_error(&p
, "Exceeded max nr indirect texture lookups");
1093 if (p
.program
->Base
.NumTexInstructions
> ctx
->Const
.FragmentProgram
.MaxTexInstructions
)
1094 program_error(&p
, "Exceeded max TEX instructions");
1096 if (p
.program
->Base
.NumAluInstructions
> ctx
->Const
.FragmentProgram
.MaxAluInstructions
)
1097 program_error(&p
, "Exceeded max ALU instructions");
1099 ASSERT(p
.program
->Base
.NumInstructions
<= MAX_INSTRUCTIONS
);
1101 /* Allocate final instruction array */
1102 program
->Base
.Instructions
1103 = _mesa_alloc_instructions(program
->Base
.NumInstructions
);
1104 if (!program
->Base
.Instructions
) {
1105 _mesa_error(ctx
, GL_OUT_OF_MEMORY
,
1106 "generating tex env program");
1109 _mesa_copy_instructions(program
->Base
.Instructions
, instBuffer
,
1110 program
->Base
.NumInstructions
);
1112 /* Notify driver the fragment program has (actually) changed.
1114 if (ctx
->Driver
.ProgramStringNotify
) {
1115 ctx
->Driver
.ProgramStringNotify( ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1120 _mesa_print_program(&p
.program
->Base
);
1126 static struct gl_fragment_program
*
1127 search_cache(const struct texenvprog_cache
*cache
,
1132 struct texenvprog_cache_item
*c
;
1134 for (c
= cache
->items
[hash
% cache
->size
]; c
; c
= c
->next
) {
1135 if (c
->hash
== hash
&& memcmp(c
->key
, key
, keysize
) == 0)
1136 return (struct gl_fragment_program
*) c
->data
;
1142 static void rehash( struct texenvprog_cache
*cache
)
1144 struct texenvprog_cache_item
**items
;
1145 struct texenvprog_cache_item
*c
, *next
;
1148 size
= cache
->size
* 3;
1149 items
= (struct texenvprog_cache_item
**) _mesa_malloc(size
* sizeof(*items
));
1150 _mesa_memset(items
, 0, size
* sizeof(*items
));
1152 for (i
= 0; i
< cache
->size
; i
++)
1153 for (c
= cache
->items
[i
]; c
; c
= next
) {
1155 c
->next
= items
[c
->hash
% size
];
1156 items
[c
->hash
% size
] = c
;
1159 _mesa_free(cache
->items
);
1160 cache
->items
= items
;
1164 static void clear_cache( struct texenvprog_cache
*cache
)
1166 struct texenvprog_cache_item
*c
, *next
;
1169 for (i
= 0; i
< cache
->size
; i
++) {
1170 for (c
= cache
->items
[i
]; c
; c
= next
) {
1173 cache
->ctx
->Driver
.DeleteProgram(cache
->ctx
,
1174 (struct gl_program
*) c
->data
);
1177 cache
->items
[i
] = NULL
;
1185 static void cache_item( struct texenvprog_cache
*cache
,
1187 const struct state_key
*key
,
1190 struct texenvprog_cache_item
*c
1191 = (struct texenvprog_cache_item
*) MALLOC(sizeof(*c
));
1194 c
->key
= _mesa_malloc(sizeof(*key
));
1195 memcpy(c
->key
, key
, sizeof(*key
));
1197 c
->data
= (struct gl_fragment_program
*) data
;
1199 if (cache
->n_items
> cache
->size
* 1.5) {
1200 if (cache
->size
< 1000)
1207 c
->next
= cache
->items
[hash
% cache
->size
];
1208 cache
->items
[hash
% cache
->size
] = c
;
1211 static GLuint
hash_key( const struct state_key
*key
)
1213 GLuint
*ikey
= (GLuint
*)key
;
1216 /* Make a slightly better attempt at a hash function:
1218 for (i
= 0; i
< sizeof(*key
)/sizeof(*ikey
); i
++)
1221 hash
+= (hash
<< 10);
1222 hash
^= (hash
>> 6);
1230 * If _MaintainTexEnvProgram is set we'll generate a fragment program that
1231 * implements the current texture env/combine mode.
1232 * This function generates that program and puts it into effect.
1235 _mesa_UpdateTexEnvProgram( GLcontext
*ctx
)
1237 struct state_key key
;
1239 const struct gl_fragment_program
*prev
= ctx
->FragmentProgram
._Current
;
1241 ASSERT(ctx
->FragmentProgram
._MaintainTexEnvProgram
);
1243 /* If a conventional fragment program/shader isn't in effect... */
1244 if (!ctx
->FragmentProgram
._Enabled
&&
1245 (!ctx
->Shader
.CurrentProgram
|| !ctx
->Shader
.CurrentProgram
->FragmentProgram
)) {
1246 make_state_key(ctx
, &key
);
1247 hash
= hash_key(&key
);
1249 ctx
->FragmentProgram
._Current
=
1250 ctx
->FragmentProgram
._TexEnvProgram
=
1251 search_cache(&ctx
->Texture
.env_fp_cache
, hash
, &key
, sizeof(key
));
1253 if (!ctx
->FragmentProgram
._TexEnvProgram
) {
1255 _mesa_printf("Building new texenv proggy for key %x\n", hash
);
1257 /* create new tex env program */
1258 ctx
->FragmentProgram
._Current
=
1259 ctx
->FragmentProgram
._TexEnvProgram
=
1260 (struct gl_fragment_program
*)
1261 ctx
->Driver
.NewProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
, 0);
1263 create_new_program(ctx
, &key
, ctx
->FragmentProgram
._TexEnvProgram
);
1265 cache_item(&ctx
->Texture
.env_fp_cache
, hash
, &key
,
1266 ctx
->FragmentProgram
._TexEnvProgram
);
1270 _mesa_printf("Found existing texenv program for key %x\n", hash
);
1274 ctx
->FragmentProgram
._Current
= ctx
->FragmentProgram
.Current
;
1277 /* Tell the driver about the change. Could define a new target for
1280 if (ctx
->FragmentProgram
._Current
!= prev
&& ctx
->Driver
.BindProgram
) {
1281 ctx
->Driver
.BindProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1282 (struct gl_program
*) ctx
->FragmentProgram
._Current
);
1287 void _mesa_TexEnvProgramCacheInit( GLcontext
*ctx
)
1289 ctx
->Texture
.env_fp_cache
.ctx
= ctx
;
1290 ctx
->Texture
.env_fp_cache
.size
= 17;
1291 ctx
->Texture
.env_fp_cache
.n_items
= 0;
1292 ctx
->Texture
.env_fp_cache
.items
= (struct texenvprog_cache_item
**)
1293 _mesa_calloc(ctx
->Texture
.env_fp_cache
.size
*
1294 sizeof(struct texenvprog_cache_item
));
1298 void _mesa_TexEnvProgramCacheDestroy( GLcontext
*ctx
)
1300 clear_cache(&ctx
->Texture
.env_fp_cache
);
1301 _mesa_free(ctx
->Texture
.env_fp_cache
.items
);