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 GLuint nr_enabled_units
:8;
64 GLuint enabled_units
:8;
65 GLuint separate_specular
:1;
68 GLuint inputs_available
:12;
72 GLuint source_index
:3; /* one of TEXTURE_1D/2D/3D/CUBE/RECT_INDEX */
74 GLuint ScaleShiftRGB
:2;
82 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
;
204 #define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0)
205 #define VERT_RESULT_TEX_ANY (0xff << VERT_RESULT_TEX0)
208 * Identify all possible varying inputs. The fragment program will
209 * never reference non-varying inputs, but will track them via state
212 * This function figures out all the inputs that the fragment program
213 * has access to. The bitmask is later reduced to just those which
214 * are actually referenced.
216 static GLbitfield
get_fp_input_mask( GLcontext
*ctx
)
218 GLbitfield fp_inputs
= 0x0;
220 if (ctx
->VertexProgram
._Overriden
) {
221 /* Somebody's messing with the vertex program and we don't have
222 * a clue what's happening. Assume that it could be producing
223 * all possible outputs.
227 else if (ctx
->RenderMode
== GL_FEEDBACK
) {
228 fp_inputs
= (FRAG_BIT_COL0
| FRAG_BIT_TEX0
);
230 else if (!ctx
->VertexProgram
._Enabled
||
231 !ctx
->VertexProgram
._Current
) {
233 /* Fixed function logic */
234 GLbitfield varying_inputs
= ctx
->varying_vp_inputs
;
236 /* First look at what values may be computed by the generated
239 if (ctx
->Light
.Enabled
) {
240 fp_inputs
|= FRAG_BIT_COL0
;
242 if (ctx
->_TriangleCaps
& DD_SEPARATE_SPECULAR
)
243 fp_inputs
|= FRAG_BIT_COL1
;
246 fp_inputs
|= (ctx
->Texture
._TexGenEnabled
|
247 ctx
->Texture
._TexMatEnabled
) << FRAG_ATTRIB_TEX0
;
249 /* Then look at what might be varying as a result of enabled
252 if (varying_inputs
& VERT_BIT_COLOR0
) fp_inputs
|= FRAG_BIT_COL0
;
253 if (varying_inputs
& VERT_BIT_COLOR1
) fp_inputs
|= FRAG_BIT_COL1
;
255 fp_inputs
|= (((varying_inputs
& VERT_BIT_TEX_ANY
) >> VERT_ATTRIB_TEX0
)
256 << FRAG_ATTRIB_TEX0
);
260 /* calculate from vp->outputs */
261 GLbitfield vp_outputs
= ctx
->VertexProgram
._Current
->Base
.OutputsWritten
;
263 if (vp_outputs
& (1 << VERT_RESULT_COL0
)) fp_inputs
|= FRAG_BIT_COL0
;
264 if (vp_outputs
& (1 << VERT_RESULT_COL1
)) fp_inputs
|= FRAG_BIT_COL1
;
266 fp_inputs
|= (((vp_outputs
& VERT_RESULT_TEX_ANY
) >> VERT_RESULT_TEX0
)
267 << FRAG_ATTRIB_TEX0
);
275 * Examine current texture environment state and generate a unique
276 * key to identify it.
278 static void make_state_key( GLcontext
*ctx
, struct state_key
*key
)
281 GLbitfield inputs_referenced
= FRAG_BIT_COL0
;
282 GLbitfield inputs_available
= get_fp_input_mask( ctx
);
284 memset(key
, 0, sizeof(*key
));
286 for (i
=0;i
<MAX_TEXTURE_UNITS
;i
++) {
287 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
289 if (!texUnit
->_ReallyEnabled
|| !texUnit
->Enabled
)
292 key
->unit
[i
].enabled
= 1;
293 key
->enabled_units
|= (1<<i
);
294 key
->nr_enabled_units
= i
+1;
295 inputs_referenced
|= FRAG_BIT_TEX(i
);
297 key
->unit
[i
].source_index
=
298 translate_tex_src_bit(texUnit
->_ReallyEnabled
);
299 key
->unit
[i
].shadow
= texUnit
->_Current
->CompareMode
== GL_COMPARE_R_TO_TEXTURE
;
301 key
->unit
[i
].NumArgsRGB
= texUnit
->_CurrentCombine
->_NumArgsRGB
;
302 key
->unit
[i
].NumArgsA
= texUnit
->_CurrentCombine
->_NumArgsA
;
304 key
->unit
[i
].ModeRGB
=
305 translate_mode(texUnit
->_CurrentCombine
->ModeRGB
);
307 translate_mode(texUnit
->_CurrentCombine
->ModeA
);
309 key
->unit
[i
].ScaleShiftRGB
= texUnit
->_CurrentCombine
->ScaleShiftRGB
;
310 key
->unit
[i
].ScaleShiftA
= texUnit
->_CurrentCombine
->ScaleShiftA
;
313 key
->unit
[i
].OptRGB
[j
].Operand
=
314 translate_operand(texUnit
->_CurrentCombine
->OperandRGB
[j
]);
315 key
->unit
[i
].OptA
[j
].Operand
=
316 translate_operand(texUnit
->_CurrentCombine
->OperandA
[j
]);
317 key
->unit
[i
].OptRGB
[j
].Source
=
318 translate_source(texUnit
->_CurrentCombine
->SourceRGB
[j
]);
319 key
->unit
[i
].OptA
[j
].Source
=
320 translate_source(texUnit
->_CurrentCombine
->SourceA
[j
]);
324 if (ctx
->_TriangleCaps
& DD_SEPARATE_SPECULAR
) {
325 key
->separate_specular
= 1;
326 inputs_referenced
|= FRAG_BIT_COL1
;
329 if (ctx
->Fog
.Enabled
) {
330 key
->fog_enabled
= 1;
331 key
->fog_mode
= translate_fog_mode(ctx
->Fog
.Mode
);
332 inputs_referenced
|= FRAG_BIT_FOGC
; /* maybe */
335 key
->inputs_available
= (inputs_available
& inputs_referenced
);
339 * Use uregs to represent registers internally, translate to Mesa's
340 * expected formats on emit.
342 * NOTE: These are passed by value extensively in this file rather
343 * than as usual by pointer reference. If this disturbs you, try
344 * remembering they are just 32bits in size.
346 * GCC is smart enough to deal with these dword-sized structures in
347 * much the same way as if I had defined them as dwords and was using
348 * macros to access and set the fields. This is much nicer and easier
361 static const struct ureg undef
= {
372 /** State used to build the fragment program:
374 struct texenv_fragment_program
{
375 struct gl_fragment_program
*program
;
377 struct state_key
*state
;
379 GLbitfield alu_temps
; /**< Track texture indirections, see spec. */
380 GLbitfield temps_output
; /**< Track texture indirections, see spec. */
381 GLbitfield temp_in_use
; /**< Tracks temporary regs which are in use. */
384 struct ureg src_texture
[MAX_TEXTURE_UNITS
];
385 /* Reg containing each texture unit's sampled texture color,
389 struct ureg src_previous
; /**< Reg containing color from previous
390 * stage. May need to be decl'd.
393 GLuint last_tex_stage
; /**< Number of last enabled texture unit */
402 static struct ureg
make_ureg(GLuint file
, GLuint idx
)
410 reg
.swz
= SWIZZLE_NOOP
;
415 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
417 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
420 GET_SWZ(reg
.swz
, w
));
425 static struct ureg
swizzle1( struct ureg reg
, int x
)
427 return swizzle(reg
, x
, x
, x
, x
);
430 static struct ureg
negate( struct ureg reg
)
436 static GLboolean
is_undef( struct ureg reg
)
438 return reg
.file
== PROGRAM_UNDEFINED
;
442 static struct ureg
get_temp( struct texenv_fragment_program
*p
)
446 /* First try and reuse temps which have been used already:
448 bit
= _mesa_ffs( ~p
->temp_in_use
& p
->alu_temps
);
450 /* Then any unused temporary:
453 bit
= _mesa_ffs( ~p
->temp_in_use
);
456 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
460 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
461 p
->program
->Base
.NumTemporaries
= bit
;
463 p
->temp_in_use
|= 1<<(bit
-1);
464 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
467 static struct ureg
get_tex_temp( struct texenv_fragment_program
*p
)
471 /* First try to find available temp not previously used (to avoid
472 * starting a new texture indirection). According to the spec, the
473 * ~p->temps_output isn't necessary, but will keep it there for
476 bit
= _mesa_ffs( ~p
->temp_in_use
& ~p
->alu_temps
& ~p
->temps_output
);
478 /* Then any unused temporary:
481 bit
= _mesa_ffs( ~p
->temp_in_use
);
484 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
488 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
489 p
->program
->Base
.NumTemporaries
= bit
;
491 p
->temp_in_use
|= 1<<(bit
-1);
492 return make_ureg(PROGRAM_TEMPORARY
, (bit
-1));
496 /** Mark a temp reg as being no longer allocatable. */
497 static void reserve_temp( struct texenv_fragment_program
*p
, struct ureg r
)
499 if (r
.file
== PROGRAM_TEMPORARY
)
500 p
->temps_output
|= (1 << r
.idx
);
504 static void release_temps(GLcontext
*ctx
, struct texenv_fragment_program
*p
)
506 GLuint max_temp
= ctx
->Const
.FragmentProgram
.MaxTemps
;
508 /* KW: To support tex_env_crossbar, don't release the registers in
511 if (max_temp
>= sizeof(int) * 8)
512 p
->temp_in_use
= p
->temps_output
;
514 p
->temp_in_use
= ~((1<<max_temp
)-1) | p
->temps_output
;
518 static struct ureg
register_param5( struct texenv_fragment_program
*p
,
525 gl_state_index tokens
[STATE_LENGTH
];
532 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
533 return make_ureg(PROGRAM_STATE_VAR
, idx
);
537 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
538 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
539 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
540 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
542 static GLuint
frag_to_vert_attrib( GLuint attrib
)
545 case FRAG_ATTRIB_COL0
: return VERT_ATTRIB_COLOR0
;
546 case FRAG_ATTRIB_COL1
: return VERT_ATTRIB_COLOR1
;
548 assert(attrib
>= FRAG_ATTRIB_TEX0
);
549 assert(attrib
<= FRAG_ATTRIB_TEX7
);
550 return attrib
- FRAG_ATTRIB_TEX0
+ VERT_ATTRIB_TEX0
;
555 static struct ureg
register_input( struct texenv_fragment_program
*p
, GLuint input
)
557 if (p
->state
->inputs_available
& (1<<input
)) {
558 p
->program
->Base
.InputsRead
|= (1 << input
);
559 return make_ureg(PROGRAM_INPUT
, input
);
562 GLuint idx
= frag_to_vert_attrib( input
);
563 return register_param3( p
, STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, idx
);
568 static void emit_arg( struct prog_src_register
*reg
,
571 reg
->File
= ureg
.file
;
572 reg
->Index
= ureg
.idx
;
573 reg
->Swizzle
= ureg
.swz
;
574 reg
->NegateBase
= ureg
.negatebase
? 0xf : 0x0;
576 reg
->NegateAbs
= ureg
.negateabs
;
579 static void emit_dst( struct prog_dst_register
*dst
,
580 struct ureg ureg
, GLuint mask
)
582 dst
->File
= ureg
.file
;
583 dst
->Index
= ureg
.idx
;
584 dst
->WriteMask
= mask
;
585 dst
->CondMask
= COND_TR
; /* always pass cond test */
586 dst
->CondSwizzle
= SWIZZLE_NOOP
;
589 static struct prog_instruction
*
590 emit_op(struct texenv_fragment_program
*p
,
599 GLuint nr
= p
->program
->Base
.NumInstructions
++;
600 struct prog_instruction
*inst
= &p
->program
->Base
.Instructions
[nr
];
602 assert(nr
< MAX_INSTRUCTIONS
);
604 _mesa_init_instructions(inst
, 1);
607 emit_arg( &inst
->SrcReg
[0], src0
);
608 emit_arg( &inst
->SrcReg
[1], src1
);
609 emit_arg( &inst
->SrcReg
[2], src2
);
611 inst
->SaturateMode
= saturate
? SATURATE_ZERO_ONE
: SATURATE_OFF
;
613 emit_dst( &inst
->DstReg
, dest
, mask
);
616 /* Accounting for indirection tracking:
618 if (dest
.file
== PROGRAM_TEMPORARY
)
619 p
->temps_output
|= 1 << dest
.idx
;
626 static struct ureg
emit_arith( struct texenv_fragment_program
*p
,
635 emit_op(p
, op
, dest
, mask
, saturate
, src0
, src1
, src2
);
637 /* Accounting for indirection tracking:
639 if (src0
.file
== PROGRAM_TEMPORARY
)
640 p
->alu_temps
|= 1 << src0
.idx
;
642 if (!is_undef(src1
) && src1
.file
== PROGRAM_TEMPORARY
)
643 p
->alu_temps
|= 1 << src1
.idx
;
645 if (!is_undef(src2
) && src2
.file
== PROGRAM_TEMPORARY
)
646 p
->alu_temps
|= 1 << src2
.idx
;
648 if (dest
.file
== PROGRAM_TEMPORARY
)
649 p
->alu_temps
|= 1 << dest
.idx
;
651 p
->program
->Base
.NumAluInstructions
++;
655 static struct ureg
emit_texld( struct texenv_fragment_program
*p
,
663 struct prog_instruction
*inst
= emit_op( p
, op
,
665 GL_FALSE
, /* don't saturate? */
670 inst
->TexSrcTarget
= tex_idx
;
671 inst
->TexSrcUnit
= tex_unit
;
673 p
->program
->Base
.NumTexInstructions
++;
675 /* Accounting for indirection tracking:
677 reserve_temp(p
, dest
);
679 /* Is this a texture indirection?
681 if ((coord
.file
== PROGRAM_TEMPORARY
&&
682 (p
->temps_output
& (1<<coord
.idx
))) ||
683 (dest
.file
== PROGRAM_TEMPORARY
&&
684 (p
->alu_temps
& (1<<dest
.idx
)))) {
685 p
->program
->Base
.NumTexIndirections
++;
686 p
->temps_output
= 1<<coord
.idx
;
688 assert(0); /* KW: texture env crossbar */
695 static struct ureg
register_const4f( struct texenv_fragment_program
*p
,
708 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
710 r
= make_ureg(PROGRAM_CONSTANT
, idx
);
715 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
716 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
717 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
718 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
721 static struct ureg
get_one( struct texenv_fragment_program
*p
)
723 if (is_undef(p
->one
))
724 p
->one
= register_scalar_const(p
, 1.0);
728 static struct ureg
get_half( struct texenv_fragment_program
*p
)
730 if (is_undef(p
->half
))
731 p
->half
= register_scalar_const(p
, 0.5);
735 static struct ureg
get_zero( struct texenv_fragment_program
*p
)
737 if (is_undef(p
->zero
))
738 p
->zero
= register_scalar_const(p
, 0.0);
743 static void program_error( struct texenv_fragment_program
*p
, const char *msg
)
745 _mesa_problem(NULL
, msg
);
749 static struct ureg
get_source( struct texenv_fragment_program
*p
,
750 GLuint src
, GLuint unit
)
754 assert(!is_undef(p
->src_texture
[unit
]));
755 return p
->src_texture
[unit
];
765 assert(!is_undef(p
->src_texture
[src
- SRC_TEXTURE0
]));
766 return p
->src_texture
[src
- SRC_TEXTURE0
];
769 return register_param2(p
, STATE_TEXENV_COLOR
, unit
);
771 case SRC_PRIMARY_COLOR
:
772 return register_input(p
, FRAG_ATTRIB_COL0
);
776 if (is_undef(p
->src_previous
))
777 return register_input(p
, FRAG_ATTRIB_COL0
);
779 return p
->src_previous
;
783 static struct ureg
emit_combine_source( struct texenv_fragment_program
*p
,
789 struct ureg arg
, src
, one
;
791 src
= get_source(p
, source
, unit
);
794 case OPR_ONE_MINUS_SRC_COLOR
:
796 * Emit tmp = 1.0 - arg.xyzw
800 return emit_arith( p
, OPCODE_SUB
, arg
, mask
, 0, one
, src
, undef
);
803 if (mask
== WRITEMASK_W
)
806 return swizzle1( src
, SWIZZLE_W
);
807 case OPR_ONE_MINUS_SRC_ALPHA
:
809 * Emit tmp = 1.0 - arg.wwww
813 return emit_arith(p
, OPCODE_SUB
, arg
, mask
, 0,
814 one
, swizzle1(src
, SWIZZLE_W
), undef
);
825 static GLboolean
args_match( struct state_key
*key
, GLuint unit
)
827 GLuint i
, nr
= key
->unit
[unit
].NumArgsRGB
;
829 for (i
= 0 ; i
< nr
; i
++) {
830 if (key
->unit
[unit
].OptA
[i
].Source
!= key
->unit
[unit
].OptRGB
[i
].Source
)
833 switch(key
->unit
[unit
].OptA
[i
].Operand
) {
835 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
843 case OPR_ONE_MINUS_SRC_ALPHA
:
844 switch(key
->unit
[unit
].OptRGB
[i
].Operand
) {
845 case OPR_ONE_MINUS_SRC_COLOR
:
846 case OPR_ONE_MINUS_SRC_ALPHA
:
853 return GL_FALSE
; /* impossible */
860 static struct ureg
emit_combine( struct texenv_fragment_program
*p
,
867 const struct mode_opt
*opt
)
870 struct ureg tmp
, half
;
873 tmp
= undef
; /* silence warning (bug 5318) */
875 for (i
= 0; i
< nr
; i
++)
876 src
[i
] = emit_combine_source( p
, mask
, unit
, opt
[i
].Source
, opt
[i
].Operand
);
880 if (mask
== WRITEMASK_XYZW
&& !saturate
)
883 return emit_arith( p
, OPCODE_MOV
, dest
, mask
, saturate
, src
[0], undef
, undef
);
885 return emit_arith( p
, OPCODE_MUL
, dest
, mask
, saturate
,
886 src
[0], src
[1], undef
);
888 return emit_arith( p
, OPCODE_ADD
, dest
, mask
, saturate
,
889 src
[0], src
[1], undef
);
890 case MODE_ADD_SIGNED
:
896 emit_arith( p
, OPCODE_ADD
, tmp
, mask
, 0, src
[0], src
[1], undef
);
897 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp
, half
, undef
);
899 case MODE_INTERPOLATE
:
900 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
902 return emit_arith( p
, OPCODE_LRP
, dest
, mask
, saturate
, src
[2], src
[0], src
[1] );
905 return emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, src
[0], src
[1], undef
);
908 case MODE_DOT3_RGBA_EXT
:
909 case MODE_DOT3_RGB_EXT
:
910 case MODE_DOT3_RGB
: {
911 struct ureg tmp0
= get_temp( p
);
912 struct ureg tmp1
= get_temp( p
);
913 struct ureg neg1
= register_scalar_const(p
, -1);
914 struct ureg two
= register_scalar_const(p
, 2);
919 * dst = tmp0 dot3 tmp1
921 emit_arith( p
, OPCODE_MAD
, tmp0
, WRITEMASK_XYZW
, 0,
924 if (_mesa_memcmp(&src
[0], &src
[1], sizeof(struct ureg
)) == 0)
927 emit_arith( p
, OPCODE_MAD
, tmp1
, WRITEMASK_XYZW
, 0,
929 emit_arith( p
, OPCODE_DP3
, dest
, mask
, saturate
, tmp0
, tmp1
, undef
);
932 case MODE_MODULATE_ADD_ATI
:
933 /* Arg0 * Arg2 + Arg1 */
934 return emit_arith( p
, OPCODE_MAD
, dest
, mask
, saturate
,
935 src
[0], src
[2], src
[1] );
936 case MODE_MODULATE_SIGNED_ADD_ATI
: {
937 /* Arg0 * Arg2 + Arg1 - 0.5 */
938 struct ureg tmp0
= get_temp(p
);
940 emit_arith( p
, OPCODE_MAD
, tmp0
, mask
, 0, src
[0], src
[2], src
[1] );
941 emit_arith( p
, OPCODE_SUB
, dest
, mask
, saturate
, tmp0
, half
, undef
);
944 case MODE_MODULATE_SUBTRACT_ATI
:
945 /* Arg0 * Arg2 - Arg1 */
946 emit_arith( p
, OPCODE_MAD
, dest
, mask
, 0, src
[0], src
[2], negate(src
[1]) );
955 * Generate instructions for one texture unit's env/combiner mode.
958 emit_texenv(struct texenv_fragment_program
*p
, GLuint unit
)
960 struct state_key
*key
= p
->state
;
961 GLboolean saturate
= (unit
< p
->last_tex_stage
);
962 GLuint rgb_shift
, alpha_shift
;
963 struct ureg out
, shift
;
966 if (!key
->unit
[unit
].enabled
) {
967 return get_source(p
, SRC_PREVIOUS
, 0);
970 switch (key
->unit
[unit
].ModeRGB
) {
971 case MODE_DOT3_RGB_EXT
:
972 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
975 case MODE_DOT3_RGBA_EXT
:
980 rgb_shift
= key
->unit
[unit
].ScaleShiftRGB
;
981 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
985 /* If this is the very last calculation, emit direct to output reg:
987 if (key
->separate_specular
||
988 unit
!= p
->last_tex_stage
||
991 dest
= get_temp( p
);
993 dest
= make_ureg(PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
995 /* Emit the RGB and A combine ops
997 if (key
->unit
[unit
].ModeRGB
== key
->unit
[unit
].ModeA
&&
998 args_match(key
, unit
)) {
999 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
1001 key
->unit
[unit
].NumArgsRGB
,
1002 key
->unit
[unit
].ModeRGB
,
1003 key
->unit
[unit
].OptRGB
);
1005 else if (key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA_EXT
||
1006 key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA
) {
1008 out
= emit_combine( p
, dest
, WRITEMASK_XYZW
, saturate
,
1010 key
->unit
[unit
].NumArgsRGB
,
1011 key
->unit
[unit
].ModeRGB
,
1012 key
->unit
[unit
].OptRGB
);
1015 /* Need to do something to stop from re-emitting identical
1016 * argument calculations here:
1018 out
= emit_combine( p
, dest
, WRITEMASK_XYZ
, saturate
,
1020 key
->unit
[unit
].NumArgsRGB
,
1021 key
->unit
[unit
].ModeRGB
,
1022 key
->unit
[unit
].OptRGB
);
1023 out
= emit_combine( p
, dest
, WRITEMASK_W
, saturate
,
1025 key
->unit
[unit
].NumArgsA
,
1026 key
->unit
[unit
].ModeA
,
1027 key
->unit
[unit
].OptA
);
1030 /* Deal with the final shift:
1032 if (alpha_shift
|| rgb_shift
) {
1033 if (rgb_shift
== alpha_shift
) {
1034 shift
= register_scalar_const(p
, (GLfloat
)(1<<rgb_shift
));
1037 shift
= register_const4f(p
,
1038 (GLfloat
)(1<<rgb_shift
),
1039 (GLfloat
)(1<<rgb_shift
),
1040 (GLfloat
)(1<<rgb_shift
),
1041 (GLfloat
)(1<<alpha_shift
));
1043 return emit_arith( p
, OPCODE_MUL
, dest
, WRITEMASK_XYZW
,
1044 saturate
, out
, shift
, undef
);
1052 * Generate instruction for getting a texture source term.
1054 static void load_texture( struct texenv_fragment_program
*p
, GLuint unit
)
1056 if (is_undef(p
->src_texture
[unit
])) {
1057 GLuint dim
= p
->state
->unit
[unit
].source_index
;
1058 struct ureg texcoord
= register_input(p
, FRAG_ATTRIB_TEX0
+unit
);
1059 struct ureg tmp
= get_tex_temp( p
);
1061 if (dim
== TEXTURE_UNKNOWN_INDEX
)
1062 program_error(p
, "TexSrcBit");
1064 /* TODO: Use D0_MASK_XY where possible.
1066 if (p
->state
->unit
[unit
].enabled
) {
1067 p
->src_texture
[unit
] = emit_texld( p
, OPCODE_TXP
,
1068 tmp
, WRITEMASK_XYZW
,
1069 unit
, dim
, texcoord
);
1071 if (p
->state
->unit
[unit
].shadow
)
1072 p
->program
->Base
.ShadowSamplers
|= 1 << unit
;
1074 p
->program
->Base
.SamplersUsed
|= (1 << unit
);
1075 /* This identity mapping should already be in place
1076 * (see _mesa_init_program_struct()) but let's be safe.
1078 p
->program
->Base
.SamplerUnits
[unit
] = unit
;
1081 p
->src_texture
[unit
] = get_zero(p
);
1085 static GLboolean
load_texenv_source( struct texenv_fragment_program
*p
,
1086 GLuint src
, GLuint unit
)
1090 load_texture(p
, unit
);
1101 load_texture(p
, src
- SRC_TEXTURE0
);
1113 * Generate instructions for loading all texture source terms.
1116 load_texunit_sources( struct texenv_fragment_program
*p
, int unit
)
1118 struct state_key
*key
= p
->state
;
1121 for (i
= 0; i
< key
->unit
[unit
].NumArgsRGB
; i
++) {
1122 load_texenv_source( p
, key
->unit
[unit
].OptRGB
[i
].Source
, unit
);
1125 for (i
= 0; i
< key
->unit
[unit
].NumArgsA
; i
++) {
1126 load_texenv_source( p
, key
->unit
[unit
].OptA
[i
].Source
, unit
);
1134 * Generate a new fragment program which implements the context's
1135 * current texture env/combine mode.
1138 create_new_program(GLcontext
*ctx
, struct state_key
*key
,
1139 struct gl_fragment_program
*program
)
1141 struct prog_instruction instBuffer
[MAX_INSTRUCTIONS
];
1142 struct texenv_fragment_program p
;
1144 struct ureg cf
, out
;
1146 _mesa_memset(&p
, 0, sizeof(p
));
1149 p
.program
= program
;
1151 /* During code generation, use locally-allocated instruction buffer,
1152 * then alloc dynamic storage below.
1154 p
.program
->Base
.Instructions
= instBuffer
;
1155 p
.program
->Base
.Target
= GL_FRAGMENT_PROGRAM_ARB
;
1156 p
.program
->Base
.NumTexIndirections
= 1; /* correct? */
1157 p
.program
->Base
.NumTexInstructions
= 0;
1158 p
.program
->Base
.NumAluInstructions
= 0;
1159 p
.program
->Base
.String
= NULL
;
1160 p
.program
->Base
.NumInstructions
=
1161 p
.program
->Base
.NumTemporaries
=
1162 p
.program
->Base
.NumParameters
=
1163 p
.program
->Base
.NumAttributes
= p
.program
->Base
.NumAddressRegs
= 0;
1164 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1166 p
.program
->Base
.InputsRead
= 0;
1167 p
.program
->Base
.OutputsWritten
= 1 << FRAG_RESULT_COLR
;
1169 for (unit
= 0; unit
< MAX_TEXTURE_UNITS
; unit
++)
1170 p
.src_texture
[unit
] = undef
;
1172 p
.src_previous
= undef
;
1177 p
.last_tex_stage
= 0;
1178 release_temps(ctx
, &p
);
1180 if (key
->enabled_units
) {
1181 /* First pass - to support texture_env_crossbar, first identify
1182 * all referenced texture sources and emit texld instructions
1185 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1186 if (key
->unit
[unit
].enabled
) {
1187 load_texunit_sources( &p
, unit
);
1188 p
.last_tex_stage
= unit
;
1191 /* Second pass - emit combine instructions to build final color:
1193 for (unit
= 0 ; unit
< ctx
->Const
.MaxTextureUnits
; unit
++)
1194 if (key
->enabled_units
& (1<<unit
)) {
1195 p
.src_previous
= emit_texenv( &p
, unit
);
1196 reserve_temp(&p
, p
.src_previous
); /* don't re-use this temp reg */
1197 release_temps(ctx
, &p
); /* release all temps */
1201 cf
= get_source( &p
, SRC_PREVIOUS
, 0 );
1202 out
= make_ureg( PROGRAM_OUTPUT
, FRAG_RESULT_COLR
);
1204 if (key
->separate_specular
) {
1205 /* Emit specular add.
1207 struct ureg s
= register_input(&p
, FRAG_ATTRIB_COL1
);
1208 emit_arith( &p
, OPCODE_ADD
, out
, WRITEMASK_XYZ
, 0, cf
, s
, undef
);
1209 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_W
, 0, cf
, undef
, undef
);
1211 else if (_mesa_memcmp(&cf
, &out
, sizeof(cf
)) != 0) {
1212 /* Will wind up in here if no texture enabled or a couple of
1213 * other scenarios (GL_REPLACE for instance).
1215 emit_arith( &p
, OPCODE_MOV
, out
, WRITEMASK_XYZW
, 0, cf
, undef
, undef
);
1220 emit_arith( &p
, OPCODE_END
, undef
, WRITEMASK_XYZW
, 0, undef
, undef
, undef
);
1222 if (key
->fog_enabled
) {
1223 /* Pull fog mode from GLcontext, the value in the state key is
1224 * a reduced value and not what is expected in FogOption
1226 p
.program
->FogOption
= ctx
->Fog
.Mode
;
1227 p
.program
->Base
.InputsRead
|= FRAG_BIT_FOGC
; /* XXX new */
1229 p
.program
->FogOption
= GL_NONE
;
1231 if (p
.program
->Base
.NumTexIndirections
> ctx
->Const
.FragmentProgram
.MaxTexIndirections
)
1232 program_error(&p
, "Exceeded max nr indirect texture lookups");
1234 if (p
.program
->Base
.NumTexInstructions
> ctx
->Const
.FragmentProgram
.MaxTexInstructions
)
1235 program_error(&p
, "Exceeded max TEX instructions");
1237 if (p
.program
->Base
.NumAluInstructions
> ctx
->Const
.FragmentProgram
.MaxAluInstructions
)
1238 program_error(&p
, "Exceeded max ALU instructions");
1240 ASSERT(p
.program
->Base
.NumInstructions
<= MAX_INSTRUCTIONS
);
1242 /* Allocate final instruction array */
1243 p
.program
->Base
.Instructions
1244 = _mesa_alloc_instructions(p
.program
->Base
.NumInstructions
);
1245 if (!p
.program
->Base
.Instructions
) {
1246 _mesa_error(ctx
, GL_OUT_OF_MEMORY
,
1247 "generating tex env program");
1250 _mesa_copy_instructions(p
.program
->Base
.Instructions
, instBuffer
,
1251 p
.program
->Base
.NumInstructions
);
1253 if (p
.program
->FogOption
) {
1254 _mesa_append_fog_code(ctx
, p
.program
);
1255 p
.program
->FogOption
= GL_NONE
;
1259 /* Notify driver the fragment program has (actually) changed.
1261 if (ctx
->Driver
.ProgramStringNotify
) {
1262 ctx
->Driver
.ProgramStringNotify( ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1267 _mesa_print_program(&p
.program
->Base
);
1274 * Return a fragment program which implements the current
1275 * fixed-function texture, fog and color-sum operations.
1277 struct gl_fragment_program
*
1278 _mesa_get_fixed_func_fragment_program(GLcontext
*ctx
)
1280 struct gl_fragment_program
*prog
;
1281 struct state_key key
;
1283 make_state_key(ctx
, &key
);
1285 prog
= (struct gl_fragment_program
*)
1286 _mesa_search_program_cache(ctx
->FragmentProgram
.Cache
,
1290 prog
= (struct gl_fragment_program
*)
1291 ctx
->Driver
.NewProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
, 0);
1293 create_new_program(ctx
, &key
, prog
);
1295 _mesa_program_cache_insert(ctx
, ctx
->FragmentProgram
.Cache
,
1296 &key
, sizeof(key
), &prog
->Base
);
1305 * If _MaintainTexEnvProgram is set we'll generate a fragment program that
1306 * implements the current texture env/combine mode.
1307 * This function generates that program and puts it into effect.
1309 * XXX: remove this function. currently only called by some drivers,
1310 * not by mesa core. We now handle this properly from inside mesa.
1313 _mesa_UpdateTexEnvProgram( GLcontext
*ctx
)
1315 const struct gl_fragment_program
*prev
= ctx
->FragmentProgram
._Current
;
1317 ASSERT(ctx
->FragmentProgram
._MaintainTexEnvProgram
);
1319 /* If a conventional fragment program/shader isn't in effect... */
1320 if (!ctx
->FragmentProgram
._Enabled
&&
1321 (!ctx
->Shader
.CurrentProgram
||
1322 !ctx
->Shader
.CurrentProgram
->FragmentProgram
) )
1324 struct gl_fragment_program
*newProg
;
1326 newProg
= _mesa_get_fixed_func_fragment_program(ctx
);
1328 _mesa_reference_fragprog(ctx
, &ctx
->FragmentProgram
._Current
, newProg
);
1329 _mesa_reference_fragprog(ctx
, &ctx
->FragmentProgram
._TexEnvProgram
, newProg
);
1332 /* Tell the driver about the change. Could define a new target for
1335 if (ctx
->FragmentProgram
._Current
!= prev
&& ctx
->Driver
.BindProgram
) {
1336 ctx
->Driver
.BindProgram(ctx
, GL_FRAGMENT_PROGRAM_ARB
,
1337 (struct gl_program
*) ctx
->FragmentProgram
._Current
);