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/program.h"
32 #include "shader/prog_parameter.h"
33 #include "shader/prog_cache.h"
34 #include "shader/prog_instruction.h"
35 #include "shader/prog_print.h"
36 #include "shader/prog_statevars.h"
37 #include "shader/programopt.h"
38 #include "texenvprogram.h"
41 struct texenvprog_cache_item
45 struct gl_fragment_program
*data
;
46 struct texenvprog_cache_item
*next
;
51 * Up to nine instructions per tex unit, plus fog, specular color.
53 #define MAX_INSTRUCTIONS ((MAX_TEXTURE_UNITS * 9) + 12)
55 #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
63 GLbitfield enabled_units
;
64 GLuint separate_specular
:1;
70 GLuint source_index
:3; /* one of TEXTURE_1D/2D/3D/CUBE/RECT_INDEX */
72 GLuint ScaleShiftRGB
:2;
77 struct mode_opt OptRGB
[3];
81 struct mode_opt OptA
[3];
90 static GLuint
translate_fog_mode( GLenum mode
)
93 case GL_LINEAR
: return FOG_LINEAR
;
94 case GL_EXP
: return FOG_EXP
;
95 case GL_EXP2
: return FOG_EXP2
;
96 default: return FOG_UNKNOWN
;
100 #define OPR_SRC_COLOR 0
101 #define OPR_ONE_MINUS_SRC_COLOR 1
102 #define OPR_SRC_ALPHA 2
103 #define OPR_ONE_MINUS_SRC_ALPHA 3
106 #define OPR_UNKNOWN 7
108 static GLuint
translate_operand( GLenum operand
)
111 case GL_SRC_COLOR
: return OPR_SRC_COLOR
;
112 case GL_ONE_MINUS_SRC_COLOR
: return OPR_ONE_MINUS_SRC_COLOR
;
113 case GL_SRC_ALPHA
: return OPR_SRC_ALPHA
;
114 case GL_ONE_MINUS_SRC_ALPHA
: return OPR_ONE_MINUS_SRC_ALPHA
;
115 case GL_ZERO
: return OPR_ZERO
;
116 case GL_ONE
: return OPR_ONE
;
117 default: return OPR_UNKNOWN
;
121 #define SRC_TEXTURE 0
122 #define SRC_TEXTURE0 1
123 #define SRC_TEXTURE1 2
124 #define SRC_TEXTURE2 3
125 #define SRC_TEXTURE3 4
126 #define SRC_TEXTURE4 5
127 #define SRC_TEXTURE5 6
128 #define SRC_TEXTURE6 7
129 #define SRC_TEXTURE7 8
130 #define SRC_CONSTANT 9
131 #define SRC_PRIMARY_COLOR 10
132 #define SRC_PREVIOUS 11
133 #define SRC_UNKNOWN 15
135 static GLuint
translate_source( GLenum src
)
138 case GL_TEXTURE
: return SRC_TEXTURE
;
146 case GL_TEXTURE7
: return SRC_TEXTURE0
+ (src
- GL_TEXTURE0
);
147 case GL_CONSTANT
: return SRC_CONSTANT
;
148 case GL_PRIMARY_COLOR
: return SRC_PRIMARY_COLOR
;
149 case GL_PREVIOUS
: return SRC_PREVIOUS
;
150 default: return SRC_UNKNOWN
;
154 #define MODE_REPLACE 0
155 #define MODE_MODULATE 1
157 #define MODE_ADD_SIGNED 3
158 #define MODE_INTERPOLATE 4
159 #define MODE_SUBTRACT 5
160 #define MODE_DOT3_RGB 6
161 #define MODE_DOT3_RGB_EXT 7
162 #define MODE_DOT3_RGBA 8
163 #define MODE_DOT3_RGBA_EXT 9
164 #define MODE_MODULATE_ADD_ATI 10
165 #define MODE_MODULATE_SIGNED_ADD_ATI 11
166 #define MODE_MODULATE_SUBTRACT_ATI 12
167 #define MODE_UNKNOWN 15
169 static GLuint
translate_mode( GLenum mode
)
172 case GL_REPLACE
: return MODE_REPLACE
;
173 case GL_MODULATE
: return MODE_MODULATE
;
174 case GL_ADD
: return MODE_ADD
;
175 case GL_ADD_SIGNED
: return MODE_ADD_SIGNED
;
176 case GL_INTERPOLATE
: return MODE_INTERPOLATE
;
177 case GL_SUBTRACT
: return MODE_SUBTRACT
;
178 case GL_DOT3_RGB
: return MODE_DOT3_RGB
;
179 case GL_DOT3_RGB_EXT
: return MODE_DOT3_RGB_EXT
;
180 case GL_DOT3_RGBA
: return MODE_DOT3_RGBA
;
181 case GL_DOT3_RGBA_EXT
: return MODE_DOT3_RGBA_EXT
;
182 case GL_MODULATE_ADD_ATI
: return MODE_MODULATE_ADD_ATI
;
183 case GL_MODULATE_SIGNED_ADD_ATI
: return MODE_MODULATE_SIGNED_ADD_ATI
;
184 case GL_MODULATE_SUBTRACT_ATI
: return MODE_MODULATE_SUBTRACT_ATI
;
185 default: return MODE_UNKNOWN
;
189 #define TEXTURE_UNKNOWN_INDEX 7
190 static GLuint
translate_tex_src_bit( GLbitfield bit
)
193 case TEXTURE_1D_BIT
: return TEXTURE_1D_INDEX
;
194 case TEXTURE_2D_BIT
: return TEXTURE_2D_INDEX
;
195 case TEXTURE_RECT_BIT
: return TEXTURE_RECT_INDEX
;
196 case TEXTURE_3D_BIT
: return TEXTURE_3D_INDEX
;
197 case TEXTURE_CUBE_BIT
: return TEXTURE_CUBE_INDEX
;
198 default: return TEXTURE_UNKNOWN_INDEX
;
203 * Examine current texture environment state and generate a unique
204 * key to identify it.
206 static void make_state_key( GLcontext
*ctx
, struct state_key
*key
)
210 memset(key
, 0, sizeof(*key
));
212 for (i
=0;i
<MAX_TEXTURE_UNITS
;i
++) {
213 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
215 if (!texUnit
->_ReallyEnabled
|| !texUnit
->Enabled
)
218 key
->unit
[i
].enabled
= 1;
219 key
->enabled_units
|= (1<<i
);
221 key
->unit
[i
].source_index
=
222 translate_tex_src_bit(texUnit
->_ReallyEnabled
);
223 key
->unit
[i
].shadow
= texUnit
->_Current
->CompareMode
== GL_COMPARE_R_TO_TEXTURE
;
225 key
->unit
[i
].NumArgsRGB
= texUnit
->_CurrentCombine
->_NumArgsRGB
;
226 key
->unit
[i
].NumArgsA
= texUnit
->_CurrentCombine
->_NumArgsA
;
228 key
->unit
[i
].ModeRGB
=
229 translate_mode(texUnit
->_CurrentCombine
->ModeRGB
);
231 translate_mode(texUnit
->_CurrentCombine
->ModeA
);
233 key
->unit
[i
].ScaleShiftRGB
= texUnit
->_CurrentCombine
->ScaleShiftRGB
;
234 key
->unit
[i
].ScaleShiftA
= texUnit
->_CurrentCombine
->ScaleShiftA
;
237 key
->unit
[i
].OptRGB
[j
].Operand
=
238 translate_operand(texUnit
->_CurrentCombine
->OperandRGB
[j
]);
239 key
->unit
[i
].OptA
[j
].Operand
=
240 translate_operand(texUnit
->_CurrentCombine
->OperandA
[j
]);
241 key
->unit
[i
].OptRGB
[j
].Source
=
242 translate_source(texUnit
->_CurrentCombine
->SourceRGB
[j
]);
243 key
->unit
[i
].OptA
[j
].Source
=
244 translate_source(texUnit
->_CurrentCombine
->SourceA
[j
]);
248 if (ctx
->_TriangleCaps
& DD_SEPARATE_SPECULAR
)
249 key
->separate_specular
= 1;
251 if (ctx
->Fog
.Enabled
) {
252 key
->fog_enabled
= 1;
253 key
->fog_mode
= translate_fog_mode(ctx
->Fog
.Mode
);
257 /* Use uregs to represent registers internally, translate to Mesa's
258 * expected formats on emit.
260 * NOTE: These are passed by value extensively in this file rather
261 * than as usual by pointer reference. If this disturbs you, try
262 * remembering they are just 32bits in size.
264 * GCC is smart enough to deal with these dword-sized structures in
265 * much the same way as if I had defined them as dwords and was using
266 * macros to access and set the fields. This is much nicer and easier
279 static const struct ureg undef
= {
290 /* State used to build the fragment program:
292 struct texenv_fragment_program
{
293 struct gl_fragment_program
*program
;
295 struct state_key
*state
;
297 GLbitfield alu_temps
; /* Track texture indirections, see spec. */
298 GLbitfield temps_output
; /* Track texture indirections, see spec. */
299 GLbitfield temp_in_use
; /* Tracks temporary regs which are in use. */
302 struct ureg src_texture
[MAX_TEXTURE_UNITS
];
303 /* Reg containing each texture unit's sampled texture color,
307 struct ureg src_previous
; /* Reg containing color from previous
308 * stage. May need to be decl'd.
311 GLuint last_tex_stage
; /* Number of last enabled texture unit */
320 static struct ureg
make_ureg(GLuint file
, GLuint idx
)
328 reg
.swz
= SWIZZLE_NOOP
;
333 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
335 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
338 GET_SWZ(reg
.swz
, w
));
343 static struct ureg
swizzle1( struct ureg reg
, int x
)
345 return swizzle(reg
, x
, x
, x
, x
);
348 static struct ureg
negate( struct ureg reg
)
354 static GLboolean
is_undef( struct ureg reg
)
356 return reg
.file
== PROGRAM_UNDEFINED
;
360 static struct ureg
get_temp( struct texenv_fragment_program
*p
)
364 /* First try and reuse temps which have been used already:
366 bit
= _mesa_ffs( ~p
->temp_in_use
& p
->alu_temps
);
368 /* Then any unused temporary:
371 bit
= _mesa_ffs( ~p
->temp_in_use
);
374 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
378 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
379 p
->program
->Base
.NumTemporaries
= bit
;
381 p
->temp_in_use
|= 1<<(bit
-1);
382 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
385 static struct ureg
get_tex_temp( struct texenv_fragment_program
*p
)
389 /* First try to find available temp not previously used (to avoid
390 * starting a new texture indirection). According to the spec, the
391 * ~p->temps_output isn't necessary, but will keep it there for
394 bit
= _mesa_ffs( ~p
->temp_in_use
& ~p
->alu_temps
& ~p
->temps_output
);
396 /* Then any unused temporary:
399 bit
= _mesa_ffs( ~p
->temp_in_use
);
402 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
406 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
407 p
->program
->Base
.NumTemporaries
= bit
;
409 p
->temp_in_use
|= 1<<(bit
-1);
410 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
414 /** Mark a temp reg as being no longer allocatable. */
415 static void reserve_temp( struct texenv_fragment_program
*p
, struct ureg r
)
417 if (r
.file
== PROGRAM_TEMPORARY
)
418 p
->temps_output
|= (1 << r
.idx
);
422 static void release_temps(GLcontext
*ctx
, struct texenv_fragment_program
*p
)
424 GLuint max_temp
= ctx
->Const
.FragmentProgram
.MaxTemps
;
426 /* KW: To support tex_env_crossbar, don't release the registers in
429 if (max_temp
>= sizeof(int) * 8)
430 p
->temp_in_use
= p
->temps_output
;
432 p
->temp_in_use
= ~((1<<max_temp
)-1) | p
->temps_output
;
436 static struct ureg
register_param5( struct texenv_fragment_program
*p
,
443 gl_state_index tokens
[STATE_LENGTH
];
450 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
451 return make_ureg(PROGRAM_STATE_VAR
, idx
);
455 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
456 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
457 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
458 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
461 static struct ureg
register_input( struct texenv_fragment_program
*p
, GLuint input
)
463 p
->program
->Base
.InputsRead
|= (1 << input
);
464 return make_ureg(PROGRAM_INPUT
, input
);
468 static void emit_arg( struct prog_src_register
*reg
,
471 reg
->File
= ureg
.file
;
472 reg
->Index
= ureg
.idx
;
473 reg
->Swizzle
= ureg
.swz
;
474 reg
->NegateBase
= ureg
.negatebase
? 0xf : 0x0;
476 reg
->NegateAbs
= ureg
.negateabs
;
479 static void emit_dst( struct prog_dst_register
*dst
,
480 struct ureg ureg
, GLuint mask
)
482 dst
->File
= ureg
.file
;
483 dst
->Index
= ureg
.idx
;
484 dst
->WriteMask
= mask
;
485 dst
->CondMask
= COND_TR
; /* always pass cond test */
486 dst
->CondSwizzle
= SWIZZLE_NOOP
;
489 static struct prog_instruction
*
490 emit_op(struct texenv_fragment_program
*p
,
499 GLuint nr
= p
->program
->Base
.NumInstructions
++;
500 struct prog_instruction
*inst
= &p
->program
->Base
.Instructions
[nr
];
502 assert(nr
< MAX_INSTRUCTIONS
);
504 _mesa_init_instructions(inst
, 1);
507 emit_arg( &inst
->SrcReg
[0], src0
);
508 emit_arg( &inst
->SrcReg
[1], src1
);
509 emit_arg( &inst
->SrcReg
[2], src2
);
511 inst
->SaturateMode
= saturate
? SATURATE_ZERO_ONE
: SATURATE_OFF
;
513 emit_dst( &inst
->DstReg
, dest
, mask
);
516 /* Accounting for indirection tracking:
518 if (dest
.file
== PROGRAM_TEMPORARY
)
519 p
->temps_output
|= 1 << dest
.idx
;
526 static struct ureg
emit_arith( struct texenv_fragment_program
*p
,
535 emit_op(p
, op
, dest
, mask
, saturate
, src0
, src1
, src2
);
537 /* Accounting for indirection tracking:
539 if (src0
.file
== PROGRAM_TEMPORARY
)
540 p
->alu_temps
|= 1 << src0
.idx
;
542 if (!is_undef(src1
) && src1
.file
== PROGRAM_TEMPORARY
)
543 p
->alu_temps
|= 1 << src1
.idx
;
545 if (!is_undef(src2
) && src2
.file
== PROGRAM_TEMPORARY
)
546 p
->alu_temps
|= 1 << src2
.idx
;
548 if (dest
.file
== PROGRAM_TEMPORARY
)
549 p
->alu_temps
|= 1 << dest
.idx
;
551 p
->program
->Base
.NumAluInstructions
++;
555 static struct ureg
emit_texld( struct texenv_fragment_program
*p
,
563 struct prog_instruction
*inst
= emit_op( p
, op
,
565 GL_FALSE
, /* don't saturate? */
570 inst
->TexSrcTarget
= tex_idx
;
571 inst
->TexSrcUnit
= tex_unit
;
573 p
->program
->Base
.NumTexInstructions
++;
575 /* Accounting for indirection tracking:
577 reserve_temp(p
, dest
);
579 /* Is this a texture indirection?
581 if ((coord
.file
== PROGRAM_TEMPORARY
&&
582 (p
->temps_output
& (1<<coord
.idx
))) ||
583 (dest
.file
== PROGRAM_TEMPORARY
&&
584 (p
->alu_temps
& (1<<dest
.idx
)))) {
585 p
->program
->Base
.NumTexIndirections
++;
586 p
->temps_output
= 1<<coord
.idx
;
588 assert(0); /* KW: texture env crossbar */
595 static struct ureg
register_const4f( struct texenv_fragment_program
*p
,
608 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
610 r
= make_ureg(PROGRAM_CONSTANT
, idx
);
615 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
616 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
617 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
618 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
621 static struct ureg
get_one( struct texenv_fragment_program
*p
)
623 if (is_undef(p
->one
))
624 p
->one
= register_scalar_const(p
, 1.0);
628 static struct ureg
get_half( struct texenv_fragment_program
*p
)
630 if (is_undef(p
->half
))
631 p
->half
= register_scalar_const(p
, 0.5);
635 static struct ureg
get_zero( struct texenv_fragment_program
*p
)
637 if (is_undef(p
->zero
))
638 p
->zero
= register_scalar_const(p
, 0.0);
643 static void program_error( struct texenv_fragment_program
*p
, const char *msg
)
645 _mesa_problem(NULL
, msg
);
649 static struct ureg
get_source( struct texenv_fragment_program
*p
,
650 GLuint src
, GLuint unit
)
654 assert(!is_undef(p
->src_texture
[unit
]));
655 return p
->src_texture
[unit
];
665 assert(!is_undef(p
->src_texture
[src
- SRC_TEXTURE0
]));
666 return p
->src_texture
[src
- SRC_TEXTURE0
];
669 return register_param2(p
, STATE_TEXENV_COLOR
, unit
);
671 case SRC_PRIMARY_COLOR
:
672 return register_input(p
, FRAG_ATTRIB_COL0
);
676 if (is_undef(p
->src_previous
))
677 return register_input(p
, FRAG_ATTRIB_COL0
);
679 return p
->src_previous
;
683 static struct ureg
emit_combine_source( struct texenv_fragment_program
*p
,
689 struct ureg arg
, src
, one
;
691 src
= get_source(p
, source
, unit
);
694 case OPR_ONE_MINUS_SRC_COLOR
:
696 * Emit tmp = 1.0 - arg.xyzw
700 return emit_arith( p
, OPCODE_SUB
, arg
, mask
, 0, one
, src
, undef
);
703 if (mask
== WRITEMASK_W
)
706 return swizzle1( src
, SWIZZLE_W
);
707 case OPR_ONE_MINUS_SRC_ALPHA
:
709 * Emit tmp = 1.0 - arg.wwww
713 return emit_arith(p
, OPCODE_SUB
, arg
, mask
, 0,
714 one
, swizzle1(src
, SWIZZLE_W
), undef
);
725 static GLboolean
args_match( struct state_key
*key
, GLuint unit
)
727 GLuint i
, nr
= key
->unit
[unit
].NumArgsRGB
;
729 for (i
= 0 ; i
< nr
; i
++) {
730 if (key
->unit
[unit
].OptA
[i
].Source
!= key
->unit
[unit
].OptRGB
[i
].Source
)
733 switch(key
->unit
[unit
].OptA
[i
].Operand
) {
735 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
743 case OPR_ONE_MINUS_SRC_ALPHA
:
744 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
745 case OPR_ONE_MINUS_SRC_COLOR
:
746 case OPR_ONE_MINUS_SRC_ALPHA
:
753 return GL_FALSE
; /* impossible */
760 static struct ureg
emit_combine( struct texenv_fragment_program
*p
,
767 const struct mode_opt
*opt
)
770 struct ureg tmp
, half
;
773 tmp
= undef
; /* silence warning (bug 5318) */
775 for (i
= 0; i
< nr
; i
++)
776 src
[i
] = emit_combine_source( p
, mask
, unit
, opt
[i
].Source
, opt
[i
].Operand
);
780 if (mask
== WRITEMASK_XYZW
&& !saturate
)
783 return emit_arith( p
, OPCODE_MOV
, dest
, mask
, saturate
, src
[0], undef
, undef
);
785 return emit_arith( p
, OPCODE_MUL
, dest
, mask
, saturate
,
786 src
[0], src
[1], undef
);
788 return emit_arith( p
, OPCODE_ADD
, dest
, mask
, saturate
,
789 src
[0], src
[1], undef
);
790 case MODE_ADD_SIGNED
:
796 emit_arith( p
, OPCODE_ADD
, tmp
, mask
, 0, src
[0], src
[1], undef
);
797 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp
, half
, undef
);
799 case MODE_INTERPOLATE
:
800 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
802 return emit_arith( p
, OPCODE_LRP
, dest
, mask
, saturate
, src
[2], src
[0], src
[1] );
805 return emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, src
[0], src
[1], undef
);
808 case MODE_DOT3_RGBA_EXT
:
809 case MODE_DOT3_RGB_EXT
:
810 case MODE_DOT3_RGB
: {
811 struct ureg tmp0
= get_temp( p
);
812 struct ureg tmp1
= get_temp( p
);
813 struct ureg neg1
= register_scalar_const(p
, -1);
814 struct ureg two
= register_scalar_const(p
, 2);
819 * dst = tmp0 dot3 tmp1
821 emit_arith( p
, OPCODE_MAD
, tmp0
, WRITEMASK_XYZW
, 0,
824 if (_mesa_memcmp(&src
[0], &src
[1], sizeof(struct ureg
)) == 0)
827 emit_arith( p
, OPCODE_MAD
, tmp1
, WRITEMASK_XYZW
, 0,
829 emit_arith( p
, OPCODE_DP3
, dest
, mask
, saturate
, tmp0
, tmp1
, undef
);
832 case MODE_MODULATE_ADD_ATI
:
833 /* Arg0 * Arg2 + Arg1 */
834 return emit_arith( p
, OPCODE_MAD
, dest
, mask
, saturate
,
835 src
[0], src
[2], src
[1] );
836 case MODE_MODULATE_SIGNED_ADD_ATI
: {
837 /* Arg0 * Arg2 + Arg1 - 0.5 */
838 struct ureg tmp0
= get_temp(p
);
840 emit_arith( p
, OPCODE_MAD
, tmp0
, mask
, 0, src
[0], src
[2], src
[1] );
841 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp0
, half
, undef
);
844 case MODE_MODULATE_SUBTRACT_ATI
:
845 /* Arg0 * Arg2 - Arg1 */
846 emit_arith( p
, OPCODE_MAD
, dest
, mask
, 0, src
[0], src
[2], negate(src
[1]) );
855 * Generate instructions for one texture unit's env/combiner mode.
858 emit_texenv(struct texenv_fragment_program
*p
, GLuint unit
)
860 struct state_key
*key
= p
->state
;
861 GLboolean saturate
= (unit
< p
->last_tex_stage
);
862 GLuint rgb_shift
, alpha_shift
;
863 struct ureg out
, shift
;
866 if (!key
->unit
[unit
].enabled
) {
867 return get_source(p
, SRC_PREVIOUS
, 0);
870 switch (key
->unit
[unit
].ModeRGB
) {
871 case MODE_DOT3_RGB_EXT
:
872 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
875 case MODE_DOT3_RGBA_EXT
:
880 rgb_shift
= key
->unit
[unit
].ScaleShiftRGB
;
881 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
885 /* If this is the very last calculation, emit direct to output reg:
887 if (key
->separate_specular
||
888 unit
!= p
->last_tex_stage
||
891 dest
= get_temp( p
);
893 dest
= make_ureg(PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
895 /* Emit the RGB and A combine ops
897 if (key
->unit
[unit
].ModeRGB
== key
->unit
[unit
].ModeA
&&
898 args_match(key
, unit
)) {
899 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
901 key
->unit
[unit
].NumArgsRGB
,
902 key
->unit
[unit
].ModeRGB
,
903 key
->unit
[unit
].OptRGB
);
905 else if (key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA_EXT
||
906 key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA
) {
908 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
910 key
->unit
[unit
].NumArgsRGB
,
911 key
->unit
[unit
].ModeRGB
,
912 key
->unit
[unit
].OptRGB
);
915 /* Need to do something to stop from re-emitting identical
916 * argument calculations here:
918 out
= emit_combine( p
, dest
, WRITEMASK_XYZ
, saturate
,
920 key
->unit
[unit
].NumArgsRGB
,
921 key
->unit
[unit
].ModeRGB
,
922 key
->unit
[unit
].OptRGB
);
923 out
= emit_combine( p
, dest
, WRITEMASK_W
, saturate
,
925 key
->unit
[unit
].NumArgsA
,
926 key
->unit
[unit
].ModeA
,
927 key
->unit
[unit
].OptA
);
930 /* Deal with the final shift:
932 if (alpha_shift
|| rgb_shift
) {
933 if (rgb_shift
== alpha_shift
) {
934 shift
= register_scalar_const(p
, (GLfloat
)(1<<rgb_shift
));
937 shift
= register_const4f(p
,
938 (GLfloat
)(1<<rgb_shift
),
939 (GLfloat
)(1<<rgb_shift
),
940 (GLfloat
)(1<<rgb_shift
),
941 (GLfloat
)(1<<alpha_shift
));
943 return emit_arith( p
, OPCODE_MUL
, dest
, WRITEMASK_XYZW
,
944 saturate
, out
, shift
, undef
);
952 * Generate instruction for getting a texture source term.
954 static void load_texture( struct texenv_fragment_program
*p
, GLuint unit
)
956 if (is_undef(p
->src_texture
[unit
])) {
957 GLuint dim
= p
->state
->unit
[unit
].source_index
;
958 struct ureg texcoord
= register_input(p
, FRAG_ATTRIB_TEX0
+unit
);
959 struct ureg tmp
= get_tex_temp( p
);
961 if (dim
== TEXTURE_UNKNOWN_INDEX
)
962 program_error(p
, "TexSrcBit");
964 /* TODO: Use D0_MASK_XY where possible.
966 if (p
->state
->unit
[unit
].enabled
) {
967 p
->src_texture
[unit
] = emit_texld( p
, OPCODE_TXP
,
969 unit
, dim
, texcoord
);
971 if (p
->state
->unit
[unit
].shadow
)
972 p
->program
->Base
.ShadowSamplers
|= 1 << unit
;
974 p
->program
->Base
.SamplersUsed
|= (1 << unit
);
975 /* This identity mapping should already be in place
976 * (see _mesa_init_program_struct()) but let's be safe.
978 p
->program
->Base
.SamplerUnits
[unit
] = unit
;
981 p
->src_texture
[unit
] = get_zero(p
);
985 static GLboolean
load_texenv_source( struct texenv_fragment_program
*p
,
986 GLuint src
, GLuint unit
)
990 load_texture(p
, unit
);
1001 load_texture(p
, src
- SRC_TEXTURE0
);
1013 * Generate instructions for loading all texture source terms.
1016 load_texunit_sources( struct texenv_fragment_program
*p
, int unit
)
1018 struct state_key
*key
= p
->state
;
1021 for (i
= 0; i
< key
->unit
[unit
].NumArgsRGB
; i
++) {
1022 load_texenv_source( p
, key
->unit
[unit
].OptRGB
[i
].Source
, unit
);
1025 for (i
= 0; i
< key
->unit
[unit
].NumArgsA
; i
++) {
1026 load_texenv_source( p
, key
->unit
[unit
].OptA
[i
].Source
, unit
);
1034 * Generate a new fragment program which implements the context's
1035 * current texture env/combine mode.
1038 create_new_program(GLcontext
*ctx
, struct state_key
*key
,
1039 struct gl_fragment_program
*program
)
1041 struct prog_instruction instBuffer
[MAX_INSTRUCTIONS
];
1042 struct texenv_fragment_program p
;
1044 struct ureg cf
, out
;
1046 _mesa_memset(&p
, 0, sizeof(p
));
1049 p
.program
= program
;
1051 /* During code generation, use locally-allocated instruction buffer,
1052 * then alloc dynamic storage below.
1054 p
.program
->Base
.Instructions
= instBuffer
;
1055 p
.program
->Base
.Target
= GL_FRAGMENT_PROGRAM_ARB
;
1056 p
.program
->Base
.NumTexIndirections
= 1; /* correct? */
1057 p
.program
->Base
.NumTexInstructions
= 0;
1058 p
.program
->Base
.NumAluInstructions
= 0;
1059 p
.program
->Base
.String
= NULL
;
1060 p
.program
->Base
.NumInstructions
=
1061 p
.program
->Base
.NumTemporaries
=
1062 p
.program
->Base
.NumParameters
=
1063 p
.program
->Base
.NumAttributes
= p
.program
->Base
.NumAddressRegs
= 0;
1064 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1066 p
.program
->Base
.InputsRead
= 0;
1067 p
.program
->Base
.OutputsWritten
= 1 << FRAG_RESULT_COLR
;
1069 for (unit
= 0; unit
< MAX_TEXTURE_UNITS
; unit
++)
1070 p
.src_texture
[unit
] = undef
;
1072 p
.src_previous
= undef
;
1077 p
.last_tex_stage
= 0;
1078 release_temps(ctx
, &p
);
1080 if (key
->enabled_units
) {
1081 /* First pass - to support texture_env_crossbar, first identify
1082 * all referenced texture sources and emit texld instructions
1085 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1086 if (key
->unit
[unit
].enabled
) {
1087 load_texunit_sources( &p
, unit
);
1088 p
.last_tex_stage
= unit
;
1091 /* Second pass - emit combine instructions to build final color:
1093 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1094 if (key
->enabled_units
& (1<<unit
)) {
1095 p
.src_previous
= emit_texenv( &p
, unit
);
1096 reserve_temp(&p
, p
.src_previous
); /* don't re-use this temp reg */
1097 release_temps(ctx
, &p
); /* release all temps */
1101 cf
= get_source( &p
, SRC_PREVIOUS
, 0 );
1102 out
= make_ureg( PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
1104 if (key
->separate_specular
) {
1105 /* Emit specular add.
1107 struct ureg s
= register_input(&p
, FRAG_ATTRIB_COL1
);
1108 emit_arith( &p
, OPCODE_ADD
, out
, WRITEMASK_XYZ
, 0, cf
, s
, undef
);
1109 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_W
, 0, cf
, undef
, undef
);
1111 else if (_mesa_memcmp(&cf
, &out
, sizeof(cf
)) != 0) {
1112 /* Will wind up in here if no texture enabled or a couple of
1113 * other scenarios (GL_REPLACE for instance).
1115 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_XYZW
, 0, cf
, undef
, undef
);
1120 emit_arith( &p
, OPCODE_END
, undef
, WRITEMASK_XYZW
, 0, undef
, undef
, undef
);
1122 if (key
->fog_enabled
) {
1123 /* Pull fog mode from GLcontext, the value in the state key is
1124 * a reduced value and not what is expected in FogOption
1126 p
.program
->FogOption
= ctx
->Fog
.Mode
;
1127 p
.program
->Base
.InputsRead
|= FRAG_BIT_FOGC
; /* XXX new */
1129 p
.program
->FogOption
= GL_NONE
;
1131 if (p
.program
->Base
.NumTexIndirections
> ctx
->Const
.FragmentProgram
.MaxTexIndirections
)
1132 program_error(&p
, "Exceeded max nr indirect texture lookups");
1134 if (p
.program
->Base
.NumTexInstructions
> ctx
->Const
.FragmentProgram
.MaxTexInstructions
)
1135 program_error(&p
, "Exceeded max TEX instructions");
1137 if (p
.program
->Base
.NumAluInstructions
> ctx
->Const
.FragmentProgram
.MaxAluInstructions
)
1138 program_error(&p
, "Exceeded max ALU instructions");
1140 ASSERT(p
.program
->Base
.NumInstructions
<= MAX_INSTRUCTIONS
);
1142 /* Allocate final instruction array */
1143 p
.program
->Base
.Instructions
1144 = _mesa_alloc_instructions(p
.program
->Base
.NumInstructions
);
1145 if (!p
.program
->Base
.Instructions
) {
1146 _mesa_error(ctx
, GL_OUT_OF_MEMORY
,
1147 "generating tex env program");
1150 _mesa_copy_instructions(p
.program
->Base
.Instructions
, instBuffer
,
1151 p
.program
->Base
.NumInstructions
);
1153 if (p
.program
->FogOption
) {
1154 _mesa_append_fog_code(ctx
, p
.program
);
1155 p
.program
->FogOption
= GL_NONE
;
1159 /* Notify driver the fragment program has (actually) changed.
1161 if (ctx
->Driver
.ProgramStringNotify
) {
1162 ctx
->Driver
.ProgramStringNotify( ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1167 _mesa_print_program(&p
.program
->Base
);
1174 * Return a fragment program which implements the current
1175 * fixed-function texture, fog and color-sum operations.
1177 struct gl_fragment_program
*
1178 _mesa_get_fixed_func_fragment_program(GLcontext
*ctx
)
1180 struct gl_fragment_program
*prog
;
1181 struct state_key key
;
1183 make_state_key(ctx
, &key
);
1185 prog
= (struct gl_fragment_program
*)
1186 _mesa_search_program_cache(ctx
->FragmentProgram
.Cache
,
1190 prog
= (struct gl_fragment_program
*)
1191 ctx
->Driver
.NewProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
, 0);
1193 create_new_program(ctx
, &key
, prog
);
1195 _mesa_program_cache_insert(ctx
, ctx
->FragmentProgram
.Cache
,
1196 &key
, sizeof(key
), &prog
->Base
);
1205 * If _MaintainTexEnvProgram is set we'll generate a fragment program that
1206 * implements the current texture env/combine mode.
1207 * This function generates that program and puts it into effect.
1209 * XXX: remove this function. currently only called by some drivers,
1210 * not by mesa core. We now handle this properly from inside mesa.
1213 _mesa_UpdateTexEnvProgram( GLcontext
*ctx
)
1215 const struct gl_fragment_program
*prev
= ctx
->FragmentProgram
._Current
;
1217 ASSERT(ctx
->FragmentProgram
._MaintainTexEnvProgram
);
1219 /* If a conventional fragment program/shader isn't in effect... */
1220 if (!ctx
->FragmentProgram
._Enabled
&&
1221 (!ctx
->Shader
.CurrentProgram
||
1222 !ctx
->Shader
.CurrentProgram
->FragmentProgram
) )
1224 struct gl_fragment_program
*newProg
;
1226 newProg
= _mesa_get_fixed_func_fragment_program(ctx
);
1228 _mesa_reference_fragprog(ctx
, &ctx
->FragmentProgram
._Current
, newProg
);
1229 _mesa_reference_fragprog(ctx
, &ctx
->FragmentProgram
._TexEnvProgram
, newProg
);
1232 /* Tell the driver about the change. Could define a new target for
1235 if (ctx
->FragmentProgram
._Current
!= prev
&& ctx
->Driver
.BindProgram
) {
1236 ctx
->Driver
.BindProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1237 (struct gl_program
*) ctx
->FragmentProgram
._Current
);