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 * This MAX is probably a bit generous, but that's OK. There can be
52 * up to four instructions per texture unit (TEX + 3 for combine),
53 * then there's fog and specular add.
55 #define MAX_INSTRUCTIONS ((MAX_TEXTURE_UNITS * 4) + 12)
57 #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
65 GLbitfield enabled_units
;
66 GLuint separate_specular
:1;
72 GLuint source_index
:3; /* one of TEXTURE_1D/2D/3D/CUBE/RECT_INDEX */
74 GLuint ScaleShiftRGB
:2;
79 struct mode_opt OptRGB
[3];
83 struct mode_opt OptA
[3];
92 static GLuint
translate_fog_mode( GLenum mode
)
95 case GL_LINEAR
: return FOG_LINEAR
;
96 case GL_EXP
: return FOG_EXP
;
97 case GL_EXP2
: return FOG_EXP2
;
98 default: return FOG_UNKNOWN
;
102 #define OPR_SRC_COLOR 0
103 #define OPR_ONE_MINUS_SRC_COLOR 1
104 #define OPR_SRC_ALPHA 2
105 #define OPR_ONE_MINUS_SRC_ALPHA 3
108 #define OPR_UNKNOWN 7
110 static GLuint
translate_operand( GLenum operand
)
113 case GL_SRC_COLOR
: return OPR_SRC_COLOR
;
114 case GL_ONE_MINUS_SRC_COLOR
: return OPR_ONE_MINUS_SRC_COLOR
;
115 case GL_SRC_ALPHA
: return OPR_SRC_ALPHA
;
116 case GL_ONE_MINUS_SRC_ALPHA
: return OPR_ONE_MINUS_SRC_ALPHA
;
117 case GL_ZERO
: return OPR_ZERO
;
118 case GL_ONE
: return OPR_ONE
;
119 default: return OPR_UNKNOWN
;
123 #define SRC_TEXTURE 0
124 #define SRC_TEXTURE0 1
125 #define SRC_TEXTURE1 2
126 #define SRC_TEXTURE2 3
127 #define SRC_TEXTURE3 4
128 #define SRC_TEXTURE4 5
129 #define SRC_TEXTURE5 6
130 #define SRC_TEXTURE6 7
131 #define SRC_TEXTURE7 8
132 #define SRC_CONSTANT 9
133 #define SRC_PRIMARY_COLOR 10
134 #define SRC_PREVIOUS 11
135 #define SRC_UNKNOWN 15
137 static GLuint
translate_source( GLenum src
)
140 case GL_TEXTURE
: return SRC_TEXTURE
;
148 case GL_TEXTURE7
: return SRC_TEXTURE0
+ (src
- GL_TEXTURE0
);
149 case GL_CONSTANT
: return SRC_CONSTANT
;
150 case GL_PRIMARY_COLOR
: return SRC_PRIMARY_COLOR
;
151 case GL_PREVIOUS
: return SRC_PREVIOUS
;
152 default: return SRC_UNKNOWN
;
156 #define MODE_REPLACE 0
157 #define MODE_MODULATE 1
159 #define MODE_ADD_SIGNED 3
160 #define MODE_INTERPOLATE 4
161 #define MODE_SUBTRACT 5
162 #define MODE_DOT3_RGB 6
163 #define MODE_DOT3_RGB_EXT 7
164 #define MODE_DOT3_RGBA 8
165 #define MODE_DOT3_RGBA_EXT 9
166 #define MODE_MODULATE_ADD_ATI 10
167 #define MODE_MODULATE_SIGNED_ADD_ATI 11
168 #define MODE_MODULATE_SUBTRACT_ATI 12
169 #define MODE_UNKNOWN 15
171 static GLuint
translate_mode( GLenum mode
)
174 case GL_REPLACE
: return MODE_REPLACE
;
175 case GL_MODULATE
: return MODE_MODULATE
;
176 case GL_ADD
: return MODE_ADD
;
177 case GL_ADD_SIGNED
: return MODE_ADD_SIGNED
;
178 case GL_INTERPOLATE
: return MODE_INTERPOLATE
;
179 case GL_SUBTRACT
: return MODE_SUBTRACT
;
180 case GL_DOT3_RGB
: return MODE_DOT3_RGB
;
181 case GL_DOT3_RGB_EXT
: return MODE_DOT3_RGB_EXT
;
182 case GL_DOT3_RGBA
: return MODE_DOT3_RGBA
;
183 case GL_DOT3_RGBA_EXT
: return MODE_DOT3_RGBA_EXT
;
184 case GL_MODULATE_ADD_ATI
: return MODE_MODULATE_ADD_ATI
;
185 case GL_MODULATE_SIGNED_ADD_ATI
: return MODE_MODULATE_SIGNED_ADD_ATI
;
186 case GL_MODULATE_SUBTRACT_ATI
: return MODE_MODULATE_SUBTRACT_ATI
;
187 default: return MODE_UNKNOWN
;
191 #define TEXTURE_UNKNOWN_INDEX 7
192 static GLuint
translate_tex_src_bit( GLbitfield bit
)
195 case TEXTURE_1D_BIT
: return TEXTURE_1D_INDEX
;
196 case TEXTURE_2D_BIT
: return TEXTURE_2D_INDEX
;
197 case TEXTURE_RECT_BIT
: return TEXTURE_RECT_INDEX
;
198 case TEXTURE_3D_BIT
: return TEXTURE_3D_INDEX
;
199 case TEXTURE_CUBE_BIT
: return TEXTURE_CUBE_INDEX
;
200 default: return TEXTURE_UNKNOWN_INDEX
;
205 * Examine current texture environment state and generate a unique
206 * key to identify it.
208 static void make_state_key( GLcontext
*ctx
, struct state_key
*key
)
212 memset(key
, 0, sizeof(*key
));
214 for (i
=0;i
<MAX_TEXTURE_UNITS
;i
++) {
215 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
217 if (!texUnit
->_ReallyEnabled
|| !texUnit
->Enabled
)
220 key
->unit
[i
].enabled
= 1;
221 key
->enabled_units
|= (1<<i
);
223 key
->unit
[i
].source_index
=
224 translate_tex_src_bit(texUnit
->_ReallyEnabled
);
225 key
->unit
[i
].shadow
= texUnit
->_Current
->CompareMode
== GL_COMPARE_R_TO_TEXTURE
;
227 key
->unit
[i
].NumArgsRGB
= texUnit
->_CurrentCombine
->_NumArgsRGB
;
228 key
->unit
[i
].NumArgsA
= texUnit
->_CurrentCombine
->_NumArgsA
;
230 key
->unit
[i
].ModeRGB
=
231 translate_mode(texUnit
->_CurrentCombine
->ModeRGB
);
233 translate_mode(texUnit
->_CurrentCombine
->ModeA
);
235 key
->unit
[i
].ScaleShiftRGB
= texUnit
->_CurrentCombine
->ScaleShiftRGB
;
236 key
->unit
[i
].ScaleShiftA
= texUnit
->_CurrentCombine
->ScaleShiftA
;
239 key
->unit
[i
].OptRGB
[j
].Operand
=
240 translate_operand(texUnit
->_CurrentCombine
->OperandRGB
[j
]);
241 key
->unit
[i
].OptA
[j
].Operand
=
242 translate_operand(texUnit
->_CurrentCombine
->OperandA
[j
]);
243 key
->unit
[i
].OptRGB
[j
].Source
=
244 translate_source(texUnit
->_CurrentCombine
->SourceRGB
[j
]);
245 key
->unit
[i
].OptA
[j
].Source
=
246 translate_source(texUnit
->_CurrentCombine
->SourceA
[j
]);
250 if (ctx
->_TriangleCaps
& DD_SEPARATE_SPECULAR
)
251 key
->separate_specular
= 1;
253 if (ctx
->Fog
.Enabled
) {
254 key
->fog_enabled
= 1;
255 key
->fog_mode
= translate_fog_mode(ctx
->Fog
.Mode
);
259 /* Use uregs to represent registers internally, translate to Mesa's
260 * expected formats on emit.
262 * NOTE: These are passed by value extensively in this file rather
263 * than as usual by pointer reference. If this disturbs you, try
264 * remembering they are just 32bits in size.
266 * GCC is smart enough to deal with these dword-sized structures in
267 * much the same way as if I had defined them as dwords and was using
268 * macros to access and set the fields. This is much nicer and easier
281 static const struct ureg undef
= {
292 /* State used to build the fragment program:
294 struct texenv_fragment_program
{
295 struct gl_fragment_program
*program
;
297 struct state_key
*state
;
299 GLbitfield alu_temps
; /* Track texture indirections, see spec. */
300 GLbitfield temps_output
; /* Track texture indirections, see spec. */
301 GLbitfield temp_in_use
; /* Tracks temporary regs which are in use. */
304 struct ureg src_texture
[MAX_TEXTURE_UNITS
];
305 /* Reg containing each texture unit's sampled texture color,
309 struct ureg src_previous
; /* Reg containing color from previous
310 * stage. May need to be decl'd.
313 GLuint last_tex_stage
; /* Number of last enabled texture unit */
322 static struct ureg
make_ureg(GLuint file
, GLuint idx
)
330 reg
.swz
= SWIZZLE_NOOP
;
335 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
337 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
340 GET_SWZ(reg
.swz
, w
));
345 static struct ureg
swizzle1( struct ureg reg
, int x
)
347 return swizzle(reg
, x
, x
, x
, x
);
350 static struct ureg
negate( struct ureg reg
)
356 static GLboolean
is_undef( struct ureg reg
)
358 return reg
.file
== PROGRAM_UNDEFINED
;
362 static struct ureg
get_temp( struct texenv_fragment_program
*p
)
366 /* First try and reuse temps which have been used already:
368 bit
= _mesa_ffs( ~p
->temp_in_use
& p
->alu_temps
);
370 /* Then any unused temporary:
373 bit
= _mesa_ffs( ~p
->temp_in_use
);
376 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
380 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
381 p
->program
->Base
.NumTemporaries
= bit
;
383 p
->temp_in_use
|= 1<<(bit
-1);
384 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
387 static struct ureg
get_tex_temp( struct texenv_fragment_program
*p
)
391 /* First try to find availble temp not previously used (to avoid
392 * starting a new texture indirection). According to the spec, the
393 * ~p->temps_output isn't necessary, but will keep it there for
396 bit
= _mesa_ffs( ~p
->temp_in_use
& ~p
->alu_temps
& ~p
->temps_output
);
398 /* Then any unused temporary:
401 bit
= _mesa_ffs( ~p
->temp_in_use
);
404 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
408 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
409 p
->program
->Base
.NumTemporaries
= bit
;
411 p
->temp_in_use
|= 1<<(bit
-1);
412 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
416 static void release_temps(GLcontext
*ctx
, struct texenv_fragment_program
*p
)
418 GLuint max_temp
= ctx
->Const
.FragmentProgram
.MaxTemps
;
420 /* KW: To support tex_env_crossbar, don't release the registers in
423 if (max_temp
>= sizeof(int) * 8)
424 p
->temp_in_use
= p
->temps_output
;
426 p
->temp_in_use
= ~((1<<max_temp
)-1) | p
->temps_output
;
430 static struct ureg
register_param5( struct texenv_fragment_program
*p
,
437 gl_state_index tokens
[STATE_LENGTH
];
444 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
445 return make_ureg(PROGRAM_STATE_VAR
, idx
);
449 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
450 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
451 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
452 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
455 static struct ureg
register_input( struct texenv_fragment_program
*p
, GLuint input
)
457 p
->program
->Base
.InputsRead
|= (1 << input
);
458 return make_ureg(PROGRAM_INPUT
, input
);
462 static void emit_arg( struct prog_src_register
*reg
,
465 reg
->File
= ureg
.file
;
466 reg
->Index
= ureg
.idx
;
467 reg
->Swizzle
= ureg
.swz
;
468 reg
->NegateBase
= ureg
.negatebase
? 0xf : 0x0;
470 reg
->NegateAbs
= ureg
.negateabs
;
473 static void emit_dst( struct prog_dst_register
*dst
,
474 struct ureg ureg
, GLuint mask
)
476 dst
->File
= ureg
.file
;
477 dst
->Index
= ureg
.idx
;
478 dst
->WriteMask
= mask
;
479 dst
->CondMask
= COND_TR
; /* always pass cond test */
480 dst
->CondSwizzle
= SWIZZLE_NOOP
;
483 static struct prog_instruction
*
484 emit_op(struct texenv_fragment_program
*p
,
493 GLuint nr
= p
->program
->Base
.NumInstructions
++;
494 struct prog_instruction
*inst
= &p
->program
->Base
.Instructions
[nr
];
496 assert(nr
< MAX_INSTRUCTIONS
);
498 _mesa_init_instructions(inst
, 1);
501 emit_arg( &inst
->SrcReg
[0], src0
);
502 emit_arg( &inst
->SrcReg
[1], src1
);
503 emit_arg( &inst
->SrcReg
[2], src2
);
505 inst
->SaturateMode
= saturate
? SATURATE_ZERO_ONE
: SATURATE_OFF
;
507 emit_dst( &inst
->DstReg
, dest
, mask
);
509 /* Accounting for indirection tracking:
511 if (dest
.file
== PROGRAM_TEMPORARY
)
512 p
->temps_output
|= 1 << dest
.idx
;
518 static struct ureg
emit_arith( struct texenv_fragment_program
*p
,
527 emit_op(p
, op
, dest
, mask
, saturate
, src0
, src1
, src2
);
529 /* Accounting for indirection tracking:
531 if (src0
.file
== PROGRAM_TEMPORARY
)
532 p
->alu_temps
|= 1 << src0
.idx
;
534 if (!is_undef(src1
) && src1
.file
== PROGRAM_TEMPORARY
)
535 p
->alu_temps
|= 1 << src1
.idx
;
537 if (!is_undef(src2
) && src2
.file
== PROGRAM_TEMPORARY
)
538 p
->alu_temps
|= 1 << src2
.idx
;
540 if (dest
.file
== PROGRAM_TEMPORARY
)
541 p
->alu_temps
|= 1 << dest
.idx
;
543 p
->program
->Base
.NumAluInstructions
++;
547 static struct ureg
emit_texld( struct texenv_fragment_program
*p
,
555 struct prog_instruction
*inst
= emit_op( p
, op
,
557 GL_FALSE
, /* don't saturate? */
562 inst
->TexSrcTarget
= tex_idx
;
563 inst
->TexSrcUnit
= tex_unit
;
565 p
->program
->Base
.NumTexInstructions
++;
567 /* Is this a texture indirection?
569 if ((coord
.file
== PROGRAM_TEMPORARY
&&
570 (p
->temps_output
& (1<<coord
.idx
))) ||
571 (dest
.file
== PROGRAM_TEMPORARY
&&
572 (p
->alu_temps
& (1<<dest
.idx
)))) {
573 p
->program
->Base
.NumTexIndirections
++;
574 p
->temps_output
= 1<<coord
.idx
;
576 assert(0); /* KW: texture env crossbar */
583 static struct ureg
register_const4f( struct texenv_fragment_program
*p
,
595 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
597 ASSERT(swizzle
== SWIZZLE_NOOP
);
598 return make_ureg(PROGRAM_CONSTANT
, idx
);
601 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
602 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
603 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
604 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
607 static struct ureg
get_one( struct texenv_fragment_program
*p
)
609 if (is_undef(p
->one
))
610 p
->one
= register_scalar_const(p
, 1.0);
614 static struct ureg
get_half( struct texenv_fragment_program
*p
)
616 if (is_undef(p
->half
))
617 p
->half
= register_scalar_const(p
, 0.5);
621 static struct ureg
get_zero( struct texenv_fragment_program
*p
)
623 if (is_undef(p
->zero
))
624 p
->zero
= register_scalar_const(p
, 0.0);
629 static void program_error( struct texenv_fragment_program
*p
, const char *msg
)
631 _mesa_problem(NULL
, msg
);
635 static struct ureg
get_source( struct texenv_fragment_program
*p
,
636 GLuint src
, GLuint unit
)
640 assert(!is_undef(p
->src_texture
[unit
]));
641 return p
->src_texture
[unit
];
651 assert(!is_undef(p
->src_texture
[src
- SRC_TEXTURE0
]));
652 return p
->src_texture
[src
- SRC_TEXTURE0
];
655 return register_param2(p
, STATE_TEXENV_COLOR
, unit
);
657 case SRC_PRIMARY_COLOR
:
658 return register_input(p
, FRAG_ATTRIB_COL0
);
662 if (is_undef(p
->src_previous
))
663 return register_input(p
, FRAG_ATTRIB_COL0
);
665 return p
->src_previous
;
669 static struct ureg
emit_combine_source( struct texenv_fragment_program
*p
,
675 struct ureg arg
, src
, one
;
677 src
= get_source(p
, source
, unit
);
680 case OPR_ONE_MINUS_SRC_COLOR
:
682 * Emit tmp = 1.0 - arg.xyzw
686 return emit_arith( p
, OPCODE_SUB
, arg
, mask
, 0, one
, src
, undef
);
689 if (mask
== WRITEMASK_W
)
692 return swizzle1( src
, SWIZZLE_W
);
693 case OPR_ONE_MINUS_SRC_ALPHA
:
695 * Emit tmp = 1.0 - arg.wwww
699 return emit_arith(p
, OPCODE_SUB
, arg
, mask
, 0,
700 one
, swizzle1(src
, SWIZZLE_W
), undef
);
711 static GLboolean
args_match( struct state_key
*key
, GLuint unit
)
713 GLuint i
, nr
= key
->unit
[unit
].NumArgsRGB
;
715 for (i
= 0 ; i
< nr
; i
++) {
716 if (key
->unit
[unit
].OptA
[i
].Source
!= key
->unit
[unit
].OptRGB
[i
].Source
)
719 switch(key
->unit
[unit
].OptA
[i
].Operand
) {
721 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
729 case OPR_ONE_MINUS_SRC_ALPHA
:
730 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
731 case OPR_ONE_MINUS_SRC_COLOR
:
732 case OPR_ONE_MINUS_SRC_ALPHA
:
739 return GL_FALSE
; /* impossible */
746 static struct ureg
emit_combine( struct texenv_fragment_program
*p
,
753 const struct mode_opt
*opt
)
756 struct ureg tmp
, half
;
759 tmp
= undef
; /* silence warning (bug 5318) */
761 for (i
= 0; i
< nr
; i
++)
762 src
[i
] = emit_combine_source( p
, mask
, unit
, opt
[i
].Source
, opt
[i
].Operand
);
766 if (mask
== WRITEMASK_XYZW
&& !saturate
)
769 return emit_arith( p
, OPCODE_MOV
, dest
, mask
, saturate
, src
[0], undef
, undef
);
771 return emit_arith( p
, OPCODE_MUL
, dest
, mask
, saturate
,
772 src
[0], src
[1], undef
);
774 return emit_arith( p
, OPCODE_ADD
, dest
, mask
, saturate
,
775 src
[0], src
[1], undef
);
776 case MODE_ADD_SIGNED
:
782 emit_arith( p
, OPCODE_ADD
, tmp
, mask
, 0, src
[0], src
[1], undef
);
783 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp
, half
, undef
);
785 case MODE_INTERPOLATE
:
786 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
788 return emit_arith( p
, OPCODE_LRP
, dest
, mask
, saturate
, src
[2], src
[0], src
[1] );
791 return emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, src
[0], src
[1], undef
);
794 case MODE_DOT3_RGBA_EXT
:
795 case MODE_DOT3_RGB_EXT
:
796 case MODE_DOT3_RGB
: {
797 struct ureg tmp0
= get_temp( p
);
798 struct ureg tmp1
= get_temp( p
);
799 struct ureg neg1
= register_scalar_const(p
, -1);
800 struct ureg two
= register_scalar_const(p
, 2);
805 * dst = tmp0 dot3 tmp1
807 emit_arith( p
, OPCODE_MAD
, tmp0
, WRITEMASK_XYZW
, 0,
810 if (_mesa_memcmp(&src
[0], &src
[1], sizeof(struct ureg
)) == 0)
813 emit_arith( p
, OPCODE_MAD
, tmp1
, WRITEMASK_XYZW
, 0,
815 emit_arith( p
, OPCODE_DP3
, dest
, mask
, saturate
, tmp0
, tmp1
, undef
);
818 case MODE_MODULATE_ADD_ATI
:
819 /* Arg0 * Arg2 + Arg1 */
820 return emit_arith( p
, OPCODE_MAD
, dest
, mask
, saturate
,
821 src
[0], src
[2], src
[1] );
822 case MODE_MODULATE_SIGNED_ADD_ATI
: {
823 /* Arg0 * Arg2 + Arg1 - 0.5 */
824 struct ureg tmp0
= get_temp(p
);
826 emit_arith( p
, OPCODE_MAD
, tmp0
, mask
, 0, src
[0], src
[2], src
[1] );
827 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp0
, half
, undef
);
830 case MODE_MODULATE_SUBTRACT_ATI
:
831 /* Arg0 * Arg2 - Arg1 */
832 emit_arith( p
, OPCODE_MAD
, dest
, mask
, 0, src
[0], src
[2], negate(src
[1]) );
841 * Generate instructions for one texture unit's env/combiner mode.
844 emit_texenv(struct texenv_fragment_program
*p
, GLuint unit
)
846 struct state_key
*key
= p
->state
;
847 GLboolean saturate
= (unit
< p
->last_tex_stage
);
848 GLuint rgb_shift
, alpha_shift
;
849 struct ureg out
, shift
;
852 if (!key
->unit
[unit
].enabled
) {
853 return get_source(p
, SRC_PREVIOUS
, 0);
856 switch (key
->unit
[unit
].ModeRGB
) {
857 case MODE_DOT3_RGB_EXT
:
858 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
861 case MODE_DOT3_RGBA_EXT
:
866 rgb_shift
= key
->unit
[unit
].ScaleShiftRGB
;
867 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
871 /* If this is the very last calculation, emit direct to output reg:
873 if (key
->separate_specular
||
874 unit
!= p
->last_tex_stage
||
877 dest
= get_temp( p
);
879 dest
= make_ureg(PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
881 /* Emit the RGB and A combine ops
883 if (key
->unit
[unit
].ModeRGB
== key
->unit
[unit
].ModeA
&&
884 args_match(key
, unit
)) {
885 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
887 key
->unit
[unit
].NumArgsRGB
,
888 key
->unit
[unit
].ModeRGB
,
889 key
->unit
[unit
].OptRGB
);
891 else if (key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA_EXT
||
892 key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA
) {
894 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
896 key
->unit
[unit
].NumArgsRGB
,
897 key
->unit
[unit
].ModeRGB
,
898 key
->unit
[unit
].OptRGB
);
901 /* Need to do something to stop from re-emitting identical
902 * argument calculations here:
904 out
= emit_combine( p
, dest
, WRITEMASK_XYZ
, saturate
,
906 key
->unit
[unit
].NumArgsRGB
,
907 key
->unit
[unit
].ModeRGB
,
908 key
->unit
[unit
].OptRGB
);
909 out
= emit_combine( p
, dest
, WRITEMASK_W
, saturate
,
911 key
->unit
[unit
].NumArgsA
,
912 key
->unit
[unit
].ModeA
,
913 key
->unit
[unit
].OptA
);
916 /* Deal with the final shift:
918 if (alpha_shift
|| rgb_shift
) {
919 if (rgb_shift
== alpha_shift
) {
920 shift
= register_scalar_const(p
, (GLfloat
)(1<<rgb_shift
));
923 shift
= register_const4f(p
,
924 (GLfloat
)(1<<rgb_shift
),
925 (GLfloat
)(1<<rgb_shift
),
926 (GLfloat
)(1<<rgb_shift
),
927 (GLfloat
)(1<<alpha_shift
));
929 return emit_arith( p
, OPCODE_MUL
, dest
, WRITEMASK_XYZW
,
930 saturate
, out
, shift
, undef
);
938 * Generate instruction for getting a texture source term.
940 static void load_texture( struct texenv_fragment_program
*p
, GLuint unit
)
942 if (is_undef(p
->src_texture
[unit
])) {
943 GLuint dim
= p
->state
->unit
[unit
].source_index
;
944 struct ureg texcoord
= register_input(p
, FRAG_ATTRIB_TEX0
+unit
);
945 struct ureg tmp
= get_tex_temp( p
);
947 if (dim
== TEXTURE_UNKNOWN_INDEX
)
948 program_error(p
, "TexSrcBit");
950 /* TODO: Use D0_MASK_XY where possible.
952 if (p
->state
->unit
[unit
].enabled
) {
953 p
->src_texture
[unit
] = emit_texld( p
, OPCODE_TXP
,
955 unit
, dim
, texcoord
);
956 if (p
->state
->unit
[unit
].shadow
)
957 p
->program
->Base
.ShadowSamplers
|= 1 << unit
;
959 p
->program
->Base
.SamplersUsed
|= (1 << unit
);
960 /* This identity mapping should already be in place
961 * (see _mesa_init_program_struct()) but let's be safe.
963 p
->program
->Base
.SamplerUnits
[unit
] = unit
;
966 p
->src_texture
[unit
] = get_zero(p
);
970 static GLboolean
load_texenv_source( struct texenv_fragment_program
*p
,
971 GLuint src
, GLuint unit
)
975 load_texture(p
, unit
);
986 load_texture(p
, src
- SRC_TEXTURE0
);
998 * Generate instructions for loading all texture source terms.
1001 load_texunit_sources( struct texenv_fragment_program
*p
, int unit
)
1003 struct state_key
*key
= p
->state
;
1006 for (i
= 0; i
< key
->unit
[unit
].NumArgsRGB
; i
++) {
1007 load_texenv_source( p
, key
->unit
[unit
].OptRGB
[i
].Source
, unit
);
1010 for (i
= 0; i
< key
->unit
[unit
].NumArgsA
; i
++) {
1011 load_texenv_source( p
, key
->unit
[unit
].OptA
[i
].Source
, unit
);
1019 * Generate a new fragment program which implements the context's
1020 * current texture env/combine mode.
1023 create_new_program(GLcontext
*ctx
, struct state_key
*key
,
1024 struct gl_fragment_program
*program
)
1026 struct prog_instruction instBuffer
[MAX_INSTRUCTIONS
];
1027 struct texenv_fragment_program p
;
1029 struct ureg cf
, out
;
1031 _mesa_memset(&p
, 0, sizeof(p
));
1034 p
.program
= program
;
1036 /* During code generation, use locally-allocated instruction buffer,
1037 * then alloc dynamic storage below.
1039 p
.program
->Base
.Instructions
= instBuffer
;
1040 p
.program
->Base
.Target
= GL_FRAGMENT_PROGRAM_ARB
;
1041 p
.program
->Base
.NumTexIndirections
= 1; /* correct? */
1042 p
.program
->Base
.NumTexInstructions
= 0;
1043 p
.program
->Base
.NumAluInstructions
= 0;
1044 p
.program
->Base
.String
= NULL
;
1045 p
.program
->Base
.NumInstructions
=
1046 p
.program
->Base
.NumTemporaries
=
1047 p
.program
->Base
.NumParameters
=
1048 p
.program
->Base
.NumAttributes
= p
.program
->Base
.NumAddressRegs
= 0;
1049 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1051 p
.program
->Base
.InputsRead
= 0;
1052 p
.program
->Base
.OutputsWritten
= 1 << FRAG_RESULT_COLR
;
1054 for (unit
= 0; unit
< MAX_TEXTURE_UNITS
; unit
++)
1055 p
.src_texture
[unit
] = undef
;
1057 p
.src_previous
= undef
;
1062 p
.last_tex_stage
= 0;
1063 release_temps(ctx
, &p
);
1065 if (key
->enabled_units
) {
1066 /* First pass - to support texture_env_crossbar, first identify
1067 * all referenced texture sources and emit texld instructions
1070 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1071 if (key
->unit
[unit
].enabled
) {
1072 load_texunit_sources( &p
, unit
);
1073 p
.last_tex_stage
= unit
;
1076 /* Second pass - emit combine instructions to build final color:
1078 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1079 if (key
->enabled_units
& (1<<unit
)) {
1080 p
.src_previous
= emit_texenv( &p
, unit
);
1081 release_temps(ctx
, &p
); /* release all temps */
1085 cf
= get_source( &p
, SRC_PREVIOUS
, 0 );
1086 out
= make_ureg( PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
1088 if (key
->separate_specular
) {
1089 /* Emit specular add.
1091 struct ureg s
= register_input(&p
, FRAG_ATTRIB_COL1
);
1092 emit_arith( &p
, OPCODE_ADD
, out
, WRITEMASK_XYZ
, 0, cf
, s
, undef
);
1093 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_W
, 0, cf
, undef
, undef
);
1095 else if (_mesa_memcmp(&cf
, &out
, sizeof(cf
)) != 0) {
1096 /* Will wind up in here if no texture enabled or a couple of
1097 * other scenarios (GL_REPLACE for instance).
1099 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_XYZW
, 0, cf
, undef
, undef
);
1104 emit_arith( &p
, OPCODE_END
, undef
, WRITEMASK_XYZW
, 0, undef
, undef
, undef
);
1106 if (key
->fog_enabled
) {
1107 /* Pull fog mode from GLcontext, the value in the state key is
1108 * a reduced value and not what is expected in FogOption
1110 p
.program
->FogOption
= ctx
->Fog
.Mode
;
1111 p
.program
->Base
.InputsRead
|= FRAG_BIT_FOGC
; /* XXX new */
1113 p
.program
->FogOption
= GL_NONE
;
1115 if (p
.program
->Base
.NumTexIndirections
> ctx
->Const
.FragmentProgram
.MaxTexIndirections
)
1116 program_error(&p
, "Exceeded max nr indirect texture lookups");
1118 if (p
.program
->Base
.NumTexInstructions
> ctx
->Const
.FragmentProgram
.MaxTexInstructions
)
1119 program_error(&p
, "Exceeded max TEX instructions");
1121 if (p
.program
->Base
.NumAluInstructions
> ctx
->Const
.FragmentProgram
.MaxAluInstructions
)
1122 program_error(&p
, "Exceeded max ALU instructions");
1124 ASSERT(p
.program
->Base
.NumInstructions
<= MAX_INSTRUCTIONS
);
1126 /* Allocate final instruction array */
1127 p
.program
->Base
.Instructions
1128 = _mesa_alloc_instructions(p
.program
->Base
.NumInstructions
);
1129 if (!p
.program
->Base
.Instructions
) {
1130 _mesa_error(ctx
, GL_OUT_OF_MEMORY
,
1131 "generating tex env program");
1134 _mesa_copy_instructions(p
.program
->Base
.Instructions
, instBuffer
,
1135 p
.program
->Base
.NumInstructions
);
1137 if (p
.program
->FogOption
) {
1138 _mesa_append_fog_code(ctx
, p
.program
);
1139 p
.program
->FogOption
= GL_NONE
;
1143 /* Notify driver the fragment program has (actually) changed.
1145 if (ctx
->Driver
.ProgramStringNotify
) {
1146 ctx
->Driver
.ProgramStringNotify( ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1151 _mesa_print_program(&p
.program
->Base
);
1158 * Return a fragment program which implements the current
1159 * fixed-function texture, fog and color-sum operations.
1161 struct gl_fragment_program
*
1162 _mesa_get_fixed_func_fragment_program(GLcontext
*ctx
)
1164 struct gl_fragment_program
*prog
;
1165 struct state_key key
;
1167 make_state_key(ctx
, &key
);
1169 prog
= (struct gl_fragment_program
*)
1170 _mesa_search_program_cache(ctx
->FragmentProgram
.Cache
,
1174 prog
= (struct gl_fragment_program
*)
1175 ctx
->Driver
.NewProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
, 0);
1177 create_new_program(ctx
, &key
, prog
);
1179 _mesa_program_cache_insert(ctx
, ctx
->FragmentProgram
.Cache
,
1180 &key
, sizeof(key
), &prog
->Base
);
1189 * If _MaintainTexEnvProgram is set we'll generate a fragment program that
1190 * implements the current texture env/combine mode.
1191 * This function generates that program and puts it into effect.
1194 _mesa_UpdateTexEnvProgram( GLcontext
*ctx
)
1196 const struct gl_fragment_program
*prev
= ctx
->FragmentProgram
._Current
;
1198 ASSERT(ctx
->FragmentProgram
._MaintainTexEnvProgram
);
1200 /* If a conventional fragment program/shader isn't in effect... */
1201 if (!ctx
->FragmentProgram
._Enabled
&&
1202 (!ctx
->Shader
.CurrentProgram
||
1203 !ctx
->Shader
.CurrentProgram
->FragmentProgram
) )
1205 struct gl_fragment_program
*newProg
;
1207 newProg
= _mesa_get_fixed_func_fragment_program(ctx
);
1209 _mesa_reference_fragprog(ctx
, &ctx
->FragmentProgram
._Current
, newProg
);
1210 _mesa_reference_fragprog(ctx
, &ctx
->FragmentProgram
._TexEnvProgram
, newProg
);
1213 /* Tell the driver about the change. Could define a new target for
1216 if (ctx
->FragmentProgram
._Current
!= prev
&& ctx
->Driver
.BindProgram
) {
1217 ctx
->Driver
.BindProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1218 (struct gl_program
*) ctx
->FragmentProgram
._Current
);