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 **************************************************************************/
29 * \file ffvertex_prog.c
31 * Create a vertex program to execute the current fixed function T&L pipeline.
32 * \author Keith Whitwell
36 #include "main/glheader.h"
37 #include "main/mtypes.h"
38 #include "main/macros.h"
39 #include "main/enums.h"
40 #include "main/ffvertex_prog.h"
41 #include "shader/program.h"
42 #include "shader/prog_cache.h"
43 #include "shader/prog_instruction.h"
44 #include "shader/prog_parameter.h"
45 #include "shader/prog_print.h"
46 #include "shader/prog_statevars.h"
50 unsigned light_global_enabled
:1;
51 unsigned light_local_viewer
:1;
52 unsigned light_twoside
:1;
53 unsigned light_color_material
:1;
54 unsigned light_color_material_mask
:12;
55 unsigned light_material_mask
:12;
56 unsigned material_shininess_is_zero
:1;
58 unsigned need_eye_coords
:1;
60 unsigned rescale_normals
:1;
61 unsigned fog_source_is_depth
:1;
62 unsigned tnl_do_vertex_fog
:1;
63 unsigned separate_specular
:1;
65 unsigned point_attenuated
:1;
66 unsigned point_array
:1;
67 unsigned texture_enabled_global
:1;
68 unsigned fragprog_inputs_read
:12;
71 unsigned light_enabled
:1;
72 unsigned light_eyepos3_is_zero
:1;
73 unsigned light_spotcutoff_is_180
:1;
74 unsigned light_attenuated
:1;
75 unsigned texunit_really_enabled
:1;
76 unsigned texmat_enabled
:1;
77 unsigned texgen_enabled
:4;
78 unsigned texgen_mode0
:4;
79 unsigned texgen_mode1
:4;
80 unsigned texgen_mode2
:4;
81 unsigned texgen_mode3
:4;
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_NONE
;
104 #define TXG_OBJ_LINEAR 1
105 #define TXG_EYE_LINEAR 2
106 #define TXG_SPHERE_MAP 3
107 #define TXG_REFLECTION_MAP 4
108 #define TXG_NORMAL_MAP 5
110 static GLuint
translate_texgen( GLboolean enabled
, GLenum mode
)
116 case GL_OBJECT_LINEAR
: return TXG_OBJ_LINEAR
;
117 case GL_EYE_LINEAR
: return TXG_EYE_LINEAR
;
118 case GL_SPHERE_MAP
: return TXG_SPHERE_MAP
;
119 case GL_REFLECTION_MAP_NV
: return TXG_REFLECTION_MAP
;
120 case GL_NORMAL_MAP_NV
: return TXG_NORMAL_MAP
;
121 default: return TXG_NONE
;
127 * Returns bitmask of flags indicating which materials are set per-vertex
129 * XXX get these from the VBO...
132 tnl_get_per_vertex_materials(GLcontext
*ctx
)
134 GLbitfield mask
= 0x0;
136 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
137 struct vertex_buffer
*VB
= &tnl
->vb
;
140 for (i
= _TNL_FIRST_MAT
; i
<= _TNL_LAST_MAT
; i
++)
141 if (VB
->AttribPtr
[i
] && VB
->AttribPtr
[i
]->stride
)
142 mask
|= 1 << (i
- _TNL_FIRST_MAT
);
149 * Should fog be computed per-vertex?
152 tnl_get_per_vertex_fog(GLcontext
*ctx
)
155 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
156 return tnl
->_DoVertexFog
;
163 static GLboolean
check_active_shininess( GLcontext
*ctx
,
164 const struct state_key
*key
,
167 GLuint bit
= 1 << (MAT_ATTRIB_FRONT_SHININESS
+ side
);
169 if (key
->light_color_material_mask
& bit
)
172 if (key
->light_material_mask
& bit
)
175 if (ctx
->Light
.Material
.Attrib
[MAT_ATTRIB_FRONT_SHININESS
+ side
][0] != 0.0F
)
182 static void make_state_key( GLcontext
*ctx
, struct state_key
*key
)
184 const struct gl_fragment_program
*fp
;
187 memset(key
, 0, sizeof(struct state_key
));
188 fp
= ctx
->FragmentProgram
._Current
;
190 /* This now relies on texenvprogram.c being active:
194 key
->need_eye_coords
= ctx
->_NeedEyeCoords
;
196 key
->fragprog_inputs_read
= fp
->Base
.InputsRead
;
198 if (ctx
->RenderMode
== GL_FEEDBACK
) {
199 /* make sure the vertprog emits color and tex0 */
200 key
->fragprog_inputs_read
|= (FRAG_BIT_COL0
| FRAG_BIT_TEX0
);
203 key
->separate_specular
= (ctx
->Light
.Model
.ColorControl
==
204 GL_SEPARATE_SPECULAR_COLOR
);
206 if (ctx
->Light
.Enabled
) {
207 key
->light_global_enabled
= 1;
209 if (ctx
->Light
.Model
.LocalViewer
)
210 key
->light_local_viewer
= 1;
212 if (ctx
->Light
.Model
.TwoSide
)
213 key
->light_twoside
= 1;
215 if (ctx
->Light
.ColorMaterialEnabled
) {
216 key
->light_color_material
= 1;
217 key
->light_color_material_mask
= ctx
->Light
.ColorMaterialBitmask
;
220 key
->light_material_mask
= tnl_get_per_vertex_materials(ctx
);
222 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
223 struct gl_light
*light
= &ctx
->Light
.Light
[i
];
225 if (light
->Enabled
) {
226 key
->unit
[i
].light_enabled
= 1;
228 if (light
->EyePosition
[3] == 0.0)
229 key
->unit
[i
].light_eyepos3_is_zero
= 1;
231 if (light
->SpotCutoff
== 180.0)
232 key
->unit
[i
].light_spotcutoff_is_180
= 1;
234 if (light
->ConstantAttenuation
!= 1.0 ||
235 light
->LinearAttenuation
!= 0.0 ||
236 light
->QuadraticAttenuation
!= 0.0)
237 key
->unit
[i
].light_attenuated
= 1;
241 if (check_active_shininess(ctx
, key
, 0)) {
242 key
->material_shininess_is_zero
= 0;
244 else if (key
->light_twoside
&&
245 check_active_shininess(ctx
, key
, 1)) {
246 key
->material_shininess_is_zero
= 0;
249 key
->material_shininess_is_zero
= 1;
253 if (ctx
->Transform
.Normalize
)
256 if (ctx
->Transform
.RescaleNormals
)
257 key
->rescale_normals
= 1;
259 key
->fog_mode
= translate_fog_mode(fp
->FogOption
);
261 if (ctx
->Fog
.FogCoordinateSource
== GL_FRAGMENT_DEPTH_EXT
)
262 key
->fog_source_is_depth
= 1;
264 key
->tnl_do_vertex_fog
= tnl_get_per_vertex_fog(ctx
);
266 if (ctx
->Point
._Attenuated
)
267 key
->point_attenuated
= 1;
269 #if FEATURE_point_size_array
270 if (ctx
->Array
.ArrayObj
->PointSize
.Enabled
)
271 key
->point_array
= 1;
274 if (ctx
->Texture
._TexGenEnabled
||
275 ctx
->Texture
._TexMatEnabled
||
276 ctx
->Texture
._EnabledUnits
)
277 key
->texture_enabled_global
= 1;
279 for (i
= 0; i
< MAX_TEXTURE_COORD_UNITS
; i
++) {
280 struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
282 if (texUnit
->_ReallyEnabled
)
283 key
->unit
[i
].texunit_really_enabled
= 1;
285 if (ctx
->Texture
._TexMatEnabled
& ENABLE_TEXMAT(i
))
286 key
->unit
[i
].texmat_enabled
= 1;
288 if (texUnit
->TexGenEnabled
) {
289 key
->unit
[i
].texgen_enabled
= 1;
291 key
->unit
[i
].texgen_mode0
=
292 translate_texgen( texUnit
->TexGenEnabled
& (1<<0),
294 key
->unit
[i
].texgen_mode1
=
295 translate_texgen( texUnit
->TexGenEnabled
& (1<<1),
297 key
->unit
[i
].texgen_mode2
=
298 translate_texgen( texUnit
->TexGenEnabled
& (1<<2),
300 key
->unit
[i
].texgen_mode3
=
301 translate_texgen( texUnit
->TexGenEnabled
& (1<<3),
309 /* Very useful debugging tool - produces annotated listing of
310 * generated program with line/function references for each
311 * instruction back into this file:
315 /* Should be tunable by the driver - do we want to do matrix
316 * multiplications with DP4's or with MUL/MAD's? SSE works better
317 * with the latter, drivers may differ.
322 /* Use uregs to represent registers internally, translate to Mesa's
323 * expected formats on emit.
325 * NOTE: These are passed by value extensively in this file rather
326 * than as usual by pointer reference. If this disturbs you, try
327 * remembering they are just 32bits in size.
329 * GCC is smart enough to deal with these dword-sized structures in
330 * much the same way as if I had defined them as dwords and was using
331 * macros to access and set the fields. This is much nicer and easier
336 GLint idx
:9; /* relative addressing may be negative */
337 /* sizeof(idx) should == sizeof(prog_src_reg::Index) */
345 const struct state_key
*state
;
346 struct gl_vertex_program
*program
;
347 GLint max_inst
; /** number of instructions allocated for program */
350 GLuint temp_reserved
;
352 struct ureg eye_position
;
353 struct ureg eye_position_z
;
354 struct ureg eye_position_normalized
;
355 struct ureg transformed_normal
;
356 struct ureg identity
;
359 GLuint color_materials
;
363 static const struct ureg undef
= {
381 static struct ureg
make_ureg(GLuint file
, GLint idx
)
387 reg
.swz
= SWIZZLE_NOOP
;
394 static struct ureg
negate( struct ureg reg
)
401 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
403 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
406 GET_SWZ(reg
.swz
, w
));
412 static struct ureg
swizzle1( struct ureg reg
, int x
)
414 return swizzle(reg
, x
, x
, x
, x
);
418 static struct ureg
get_temp( struct tnl_program
*p
)
420 int bit
= _mesa_ffs( ~p
->temp_in_use
);
422 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
426 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
427 p
->program
->Base
.NumTemporaries
= bit
;
429 p
->temp_in_use
|= 1<<(bit
-1);
430 return make_ureg(PROGRAM_TEMPORARY
, bit
-1);
434 static struct ureg
reserve_temp( struct tnl_program
*p
)
436 struct ureg temp
= get_temp( p
);
437 p
->temp_reserved
|= 1<<temp
.idx
;
442 static void release_temp( struct tnl_program
*p
, struct ureg reg
)
444 if (reg
.file
== PROGRAM_TEMPORARY
) {
445 p
->temp_in_use
&= ~(1<<reg
.idx
);
446 p
->temp_in_use
|= p
->temp_reserved
; /* can't release reserved temps */
451 static void release_temps( struct tnl_program
*p
)
453 p
->temp_in_use
= p
->temp_reserved
;
458 * \param input one of VERT_ATTRIB_x tokens.
460 static struct ureg
register_input( struct tnl_program
*p
, GLuint input
)
462 p
->program
->Base
.InputsRead
|= (1<<input
);
463 return make_ureg(PROGRAM_INPUT
, input
);
468 * \param input one of VERT_RESULT_x tokens.
470 static struct ureg
register_output( struct tnl_program
*p
, GLuint output
)
472 p
->program
->Base
.OutputsWritten
|= (1<<output
);
473 return make_ureg(PROGRAM_OUTPUT
, output
);
477 static struct ureg
register_const4f( struct tnl_program
*p
,
490 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
492 ASSERT(swizzle
== SWIZZLE_NOOP
);
493 return make_ureg(PROGRAM_CONSTANT
, idx
);
497 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
498 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
499 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
500 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
502 static GLboolean
is_undef( struct ureg reg
)
504 return reg
.file
== PROGRAM_UNDEFINED
;
508 static struct ureg
get_identity_param( struct tnl_program
*p
)
510 if (is_undef(p
->identity
))
511 p
->identity
= register_const4f(p
, 0,0,0,1);
517 static struct ureg
register_param5(struct tnl_program
*p
,
524 gl_state_index tokens
[STATE_LENGTH
];
531 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
532 return make_ureg(PROGRAM_STATE_VAR
, idx
);
536 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
537 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
538 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
539 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
542 static void register_matrix_param5( struct tnl_program
*p
,
543 GLint s0
, /* modelview, projection, etc */
544 GLint s1
, /* texture matrix number */
545 GLint s2
, /* first row */
546 GLint s3
, /* last row */
547 GLint s4
, /* inverse, transpose, etc */
548 struct ureg
*matrix
)
552 /* This is a bit sad as the support is there to pull the whole
553 * matrix out in one go:
555 for (i
= 0; i
<= s3
- s2
; i
++)
556 matrix
[i
] = register_param5( p
, s0
, s1
, i
, i
, s4
);
560 static void emit_arg( struct prog_src_register
*src
,
563 src
->File
= reg
.file
;
564 src
->Index
= reg
.idx
;
565 src
->Swizzle
= reg
.swz
;
566 src
->NegateBase
= reg
.negate
? NEGATE_XYZW
: 0;
570 /* Check that bitfield sizes aren't exceeded */
571 ASSERT(src
->Index
== reg
.idx
);
575 static void emit_dst( struct prog_dst_register
*dst
,
576 struct ureg reg
, GLuint mask
)
578 dst
->File
= reg
.file
;
579 dst
->Index
= reg
.idx
;
580 /* allow zero as a shorthand for xyzw */
581 dst
->WriteMask
= mask
? mask
: WRITEMASK_XYZW
;
582 dst
->CondMask
= COND_TR
; /* always pass cond test */
583 dst
->CondSwizzle
= SWIZZLE_NOOP
;
586 /* Check that bitfield sizes aren't exceeded */
587 ASSERT(dst
->Index
== reg
.idx
);
591 static void debug_insn( struct prog_instruction
*inst
, const char *fn
,
595 static const char *last_fn
;
599 _mesa_printf("%s:\n", fn
);
602 _mesa_printf("%d:\t", line
);
603 _mesa_print_instruction(inst
);
608 static void emit_op3fn(struct tnl_program
*p
,
619 struct prog_instruction
*inst
;
621 assert((GLint
) p
->program
->Base
.NumInstructions
<= p
->max_inst
);
623 if (p
->program
->Base
.NumInstructions
== p
->max_inst
) {
624 /* need to extend the program's instruction array */
625 struct prog_instruction
*newInst
;
627 /* double the size */
630 newInst
= _mesa_alloc_instructions(p
->max_inst
);
632 _mesa_error(NULL
, GL_OUT_OF_MEMORY
, "vertex program build");
636 _mesa_copy_instructions(newInst
,
637 p
->program
->Base
.Instructions
,
638 p
->program
->Base
.NumInstructions
);
640 _mesa_free_instructions(p
->program
->Base
.Instructions
,
641 p
->program
->Base
.NumInstructions
);
643 p
->program
->Base
.Instructions
= newInst
;
646 nr
= p
->program
->Base
.NumInstructions
++;
648 inst
= &p
->program
->Base
.Instructions
[nr
];
649 inst
->Opcode
= (enum prog_opcode
) op
;
653 emit_arg( &inst
->SrcReg
[0], src0
);
654 emit_arg( &inst
->SrcReg
[1], src1
);
655 emit_arg( &inst
->SrcReg
[2], src2
);
657 emit_dst( &inst
->DstReg
, dest
, mask
);
659 debug_insn(inst
, fn
, line
);
663 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
664 emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
666 #define emit_op2(p, op, dst, mask, src0, src1) \
667 emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
669 #define emit_op1(p, op, dst, mask, src0) \
670 emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
673 static struct ureg
make_temp( struct tnl_program
*p
, struct ureg reg
)
675 if (reg
.file
== PROGRAM_TEMPORARY
&&
676 !(p
->temp_reserved
& (1<<reg
.idx
)))
679 struct ureg temp
= get_temp(p
);
680 emit_op1(p
, OPCODE_MOV
, temp
, 0, reg
);
686 /* Currently no tracking performed of input/output/register size or
687 * active elements. Could be used to reduce these operations, as
688 * could the matrix type.
690 static void emit_matrix_transform_vec4( struct tnl_program
*p
,
692 const struct ureg
*mat
,
695 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_X
, src
, mat
[0]);
696 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Y
, src
, mat
[1]);
697 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Z
, src
, mat
[2]);
698 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_W
, src
, mat
[3]);
702 /* This version is much easier to implement if writemasks are not
703 * supported natively on the target or (like SSE), the target doesn't
704 * have a clean/obvious dotproduct implementation.
706 static void emit_transpose_matrix_transform_vec4( struct tnl_program
*p
,
708 const struct ureg
*mat
,
713 if (dest
.file
!= PROGRAM_TEMPORARY
)
718 emit_op2(p
, OPCODE_MUL
, tmp
, 0, swizzle1(src
,X
), mat
[0]);
719 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Y
), mat
[1], tmp
);
720 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Z
), mat
[2], tmp
);
721 emit_op3(p
, OPCODE_MAD
, dest
, 0, swizzle1(src
,W
), mat
[3], tmp
);
723 if (dest
.file
!= PROGRAM_TEMPORARY
)
724 release_temp(p
, tmp
);
728 static void emit_matrix_transform_vec3( struct tnl_program
*p
,
730 const struct ureg
*mat
,
733 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_X
, src
, mat
[0]);
734 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Y
, src
, mat
[1]);
735 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Z
, src
, mat
[2]);
739 static void emit_normalize_vec3( struct tnl_program
*p
,
744 /* XXX use this when drivers are ready for NRM3 */
745 emit_op1(p
, OPCODE_NRM3
, dest
, WRITEMASK_XYZ
, src
);
747 struct ureg tmp
= get_temp(p
);
748 emit_op2(p
, OPCODE_DP3
, tmp
, WRITEMASK_X
, src
, src
);
749 emit_op1(p
, OPCODE_RSQ
, tmp
, WRITEMASK_X
, tmp
);
750 emit_op2(p
, OPCODE_MUL
, dest
, 0, src
, swizzle1(tmp
, X
));
751 release_temp(p
, tmp
);
756 static void emit_passthrough( struct tnl_program
*p
,
760 struct ureg out
= register_output(p
, output
);
761 emit_op1(p
, OPCODE_MOV
, out
, 0, register_input(p
, input
));
765 static struct ureg
get_eye_position( struct tnl_program
*p
)
767 if (is_undef(p
->eye_position
)) {
768 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
769 struct ureg modelview
[4];
771 p
->eye_position
= reserve_temp(p
);
774 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
777 emit_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
780 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
781 STATE_MATRIX_TRANSPOSE
, modelview
);
783 emit_transpose_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
787 return p
->eye_position
;
791 static struct ureg
get_eye_position_z( struct tnl_program
*p
)
793 if (!is_undef(p
->eye_position
))
794 return swizzle1(p
->eye_position
, Z
);
796 if (is_undef(p
->eye_position_z
)) {
797 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
798 struct ureg modelview
[4];
800 p
->eye_position_z
= reserve_temp(p
);
802 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
805 emit_op2(p
, OPCODE_DP4
, p
->eye_position_z
, 0, pos
, modelview
[2]);
808 return p
->eye_position_z
;
812 static struct ureg
get_eye_position_normalized( struct tnl_program
*p
)
814 if (is_undef(p
->eye_position_normalized
)) {
815 struct ureg eye
= get_eye_position(p
);
816 p
->eye_position_normalized
= reserve_temp(p
);
817 emit_normalize_vec3(p
, p
->eye_position_normalized
, eye
);
820 return p
->eye_position_normalized
;
824 static struct ureg
get_transformed_normal( struct tnl_program
*p
)
826 if (is_undef(p
->transformed_normal
) &&
827 !p
->state
->need_eye_coords
&&
828 !p
->state
->normalize
&&
829 !(p
->state
->need_eye_coords
== p
->state
->rescale_normals
))
831 p
->transformed_normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
833 else if (is_undef(p
->transformed_normal
))
835 struct ureg normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
836 struct ureg mvinv
[3];
837 struct ureg transformed_normal
= reserve_temp(p
);
839 if (p
->state
->need_eye_coords
) {
840 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 2,
841 STATE_MATRIX_INVTRANS
, mvinv
);
843 /* Transform to eye space:
845 emit_matrix_transform_vec3( p
, transformed_normal
, mvinv
, normal
);
846 normal
= transformed_normal
;
849 /* Normalize/Rescale:
851 if (p
->state
->normalize
) {
852 emit_normalize_vec3( p
, transformed_normal
, normal
);
853 normal
= transformed_normal
;
855 else if (p
->state
->need_eye_coords
== p
->state
->rescale_normals
) {
856 /* This is already adjusted for eye/non-eye rendering:
858 struct ureg rescale
= register_param2(p
, STATE_INTERNAL
,
861 emit_op2( p
, OPCODE_MUL
, transformed_normal
, 0, normal
, rescale
);
862 normal
= transformed_normal
;
865 assert(normal
.file
== PROGRAM_TEMPORARY
);
866 p
->transformed_normal
= normal
;
869 return p
->transformed_normal
;
873 static void build_hpos( struct tnl_program
*p
)
875 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
876 struct ureg hpos
= register_output( p
, VERT_RESULT_HPOS
);
880 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
882 emit_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
885 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
886 STATE_MATRIX_TRANSPOSE
, mvp
);
887 emit_transpose_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
892 static GLuint
material_attrib( GLuint side
, GLuint property
)
894 return ((property
- STATE_AMBIENT
) * 2 +
900 * Get a bitmask of which material values vary on a per-vertex basis.
902 static void set_material_flags( struct tnl_program
*p
)
904 p
->color_materials
= 0;
907 if (p
->state
->light_color_material
) {
909 p
->color_materials
= p
->state
->light_color_material_mask
;
912 p
->materials
|= p
->state
->light_material_mask
;
916 /* XXX temporary!!! */
917 #define _TNL_ATTRIB_MAT_FRONT_AMBIENT 32
919 static struct ureg
get_material( struct tnl_program
*p
, GLuint side
,
922 GLuint attrib
= material_attrib(side
, property
);
924 if (p
->color_materials
& (1<<attrib
))
925 return register_input(p
, VERT_ATTRIB_COLOR0
);
926 else if (p
->materials
& (1<<attrib
))
927 return register_input( p
, attrib
+ _TNL_ATTRIB_MAT_FRONT_AMBIENT
);
929 return register_param3( p
, STATE_MATERIAL
, side
, property
);
932 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
933 MAT_BIT_FRONT_AMBIENT | \
934 MAT_BIT_FRONT_DIFFUSE) << (side))
938 * Either return a precalculated constant value or emit code to
939 * calculate these values dynamically in the case where material calls
940 * are present between begin/end pairs.
942 * Probably want to shift this to the program compilation phase - if
943 * we always emitted the calculation here, a smart compiler could
944 * detect that it was constant (given a certain set of inputs), and
945 * lift it out of the main loop. That way the programs created here
946 * would be independent of the vertex_buffer details.
948 static struct ureg
get_scenecolor( struct tnl_program
*p
, GLuint side
)
950 if (p
->materials
& SCENE_COLOR_BITS(side
)) {
951 struct ureg lm_ambient
= register_param1(p
, STATE_LIGHTMODEL_AMBIENT
);
952 struct ureg material_emission
= get_material(p
, side
, STATE_EMISSION
);
953 struct ureg material_ambient
= get_material(p
, side
, STATE_AMBIENT
);
954 struct ureg material_diffuse
= get_material(p
, side
, STATE_DIFFUSE
);
955 struct ureg tmp
= make_temp(p
, material_diffuse
);
956 emit_op3(p
, OPCODE_MAD
, tmp
, WRITEMASK_XYZ
, lm_ambient
,
957 material_ambient
, material_emission
);
961 return register_param2( p
, STATE_LIGHTMODEL_SCENECOLOR
, side
);
965 static struct ureg
get_lightprod( struct tnl_program
*p
, GLuint light
,
966 GLuint side
, GLuint property
)
968 GLuint attrib
= material_attrib(side
, property
);
969 if (p
->materials
& (1<<attrib
)) {
970 struct ureg light_value
=
971 register_param3(p
, STATE_LIGHT
, light
, property
);
972 struct ureg material_value
= get_material(p
, side
, property
);
973 struct ureg tmp
= get_temp(p
);
974 emit_op2(p
, OPCODE_MUL
, tmp
, 0, light_value
, material_value
);
978 return register_param4(p
, STATE_LIGHTPROD
, light
, side
, property
);
982 static struct ureg
calculate_light_attenuation( struct tnl_program
*p
,
987 struct ureg attenuation
= register_param3(p
, STATE_LIGHT
, i
,
989 struct ureg att
= get_temp(p
);
991 /* Calculate spot attenuation:
993 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
994 struct ureg spot_dir_norm
= register_param3(p
, STATE_INTERNAL
,
995 STATE_LIGHT_SPOT_DIR_NORMALIZED
, i
);
996 struct ureg spot
= get_temp(p
);
997 struct ureg slt
= get_temp(p
);
999 emit_op2(p
, OPCODE_DP3
, spot
, 0, negate(VPpli
), spot_dir_norm
);
1000 emit_op2(p
, OPCODE_SLT
, slt
, 0, swizzle1(spot_dir_norm
,W
), spot
);
1001 emit_op2(p
, OPCODE_POW
, spot
, 0, spot
, swizzle1(attenuation
, W
));
1002 emit_op2(p
, OPCODE_MUL
, att
, 0, slt
, spot
);
1004 release_temp(p
, spot
);
1005 release_temp(p
, slt
);
1008 /* Calculate distance attenuation:
1010 if (p
->state
->unit
[i
].light_attenuated
) {
1013 emit_op1(p
, OPCODE_RCP
, dist
, WRITEMASK_YZ
, dist
);
1015 emit_op2(p
, OPCODE_MUL
, dist
, WRITEMASK_XZ
, dist
, swizzle1(dist
,Y
));
1017 emit_op2(p
, OPCODE_DP3
, dist
, 0, attenuation
, dist
);
1019 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
1021 emit_op1(p
, OPCODE_RCP
, dist
, 0, dist
);
1022 /* spot-atten * dist-atten */
1023 emit_op2(p
, OPCODE_MUL
, att
, 0, dist
, att
);
1026 emit_op1(p
, OPCODE_RCP
, att
, 0, dist
);
1036 * lit.y = MAX(0, dots.x)
1037 * lit.z = SLT(0, dots.x)
1039 static void emit_degenerate_lit( struct tnl_program
*p
,
1043 struct ureg id
= get_identity_param(p
); /* id = {0,0,0,1} */
1045 /* Note that lit.x & lit.w will not be examined. Note also that
1046 * dots.xyzw == dots.xxxx.
1049 /* MAX lit, id, dots;
1051 emit_op2(p
, OPCODE_MAX
, lit
, WRITEMASK_XYZW
, id
, dots
);
1053 /* result[2] = (in > 0 ? 1 : 0)
1054 * SLT lit.z, id.z, dots; # lit.z = (0 < dots.z) ? 1 : 0
1056 emit_op2(p
, OPCODE_SLT
, lit
, WRITEMASK_Z
, swizzle1(id
,Z
), dots
);
1060 /* Need to add some addtional parameters to allow lighting in object
1061 * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
1064 static void build_lighting( struct tnl_program
*p
)
1066 const GLboolean twoside
= p
->state
->light_twoside
;
1067 const GLboolean separate
= p
->state
->separate_specular
;
1068 GLuint nr_lights
= 0, count
= 0;
1069 struct ureg normal
= get_transformed_normal(p
);
1070 struct ureg lit
= get_temp(p
);
1071 struct ureg dots
= get_temp(p
);
1072 struct ureg _col0
= undef
, _col1
= undef
;
1073 struct ureg _bfc0
= undef
, _bfc1
= undef
;
1078 * dot.x = dot(normal, VPpli)
1079 * dot.y = dot(normal, halfAngle)
1080 * dot.z = back.shininess
1081 * dot.w = front.shininess
1084 for (i
= 0; i
< MAX_LIGHTS
; i
++)
1085 if (p
->state
->unit
[i
].light_enabled
)
1088 set_material_flags(p
);
1091 if (!p
->state
->material_shininess_is_zero
) {
1092 struct ureg shininess
= get_material(p
, 0, STATE_SHININESS
);
1093 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_W
, swizzle1(shininess
,X
));
1094 release_temp(p
, shininess
);
1097 _col0
= make_temp(p
, get_scenecolor(p
, 0));
1099 _col1
= make_temp(p
, get_identity_param(p
));
1106 if (!p
->state
->material_shininess_is_zero
) {
1107 struct ureg shininess
= get_material(p
, 1, STATE_SHININESS
);
1108 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_Z
,
1109 negate(swizzle1(shininess
,X
)));
1110 release_temp(p
, shininess
);
1113 _bfc0
= make_temp(p
, get_scenecolor(p
, 1));
1115 _bfc1
= make_temp(p
, get_identity_param(p
));
1120 /* If no lights, still need to emit the scenecolor.
1123 struct ureg res0
= register_output( p
, VERT_RESULT_COL0
);
1124 emit_op1(p
, OPCODE_MOV
, res0
, 0, _col0
);
1128 struct ureg res1
= register_output( p
, VERT_RESULT_COL1
);
1129 emit_op1(p
, OPCODE_MOV
, res1
, 0, _col1
);
1133 struct ureg res0
= register_output( p
, VERT_RESULT_BFC0
);
1134 emit_op1(p
, OPCODE_MOV
, res0
, 0, _bfc0
);
1137 if (twoside
&& separate
) {
1138 struct ureg res1
= register_output( p
, VERT_RESULT_BFC1
);
1139 emit_op1(p
, OPCODE_MOV
, res1
, 0, _bfc1
);
1142 if (nr_lights
== 0) {
1147 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
1148 if (p
->state
->unit
[i
].light_enabled
) {
1149 struct ureg half
= undef
;
1150 struct ureg att
= undef
, VPpli
= undef
;
1154 if (p
->state
->unit
[i
].light_eyepos3_is_zero
) {
1155 /* Can used precomputed constants in this case.
1156 * Attenuation never applies to infinite lights.
1158 VPpli
= register_param3(p
, STATE_INTERNAL
,
1159 STATE_LIGHT_POSITION_NORMALIZED
, i
);
1161 if (!p
->state
->material_shininess_is_zero
) {
1162 if (p
->state
->light_local_viewer
) {
1163 struct ureg eye_hat
= get_eye_position_normalized(p
);
1165 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1166 emit_normalize_vec3(p
, half
, half
);
1168 half
= register_param3(p
, STATE_INTERNAL
,
1169 STATE_LIGHT_HALF_VECTOR
, i
);
1174 struct ureg Ppli
= register_param3(p
, STATE_INTERNAL
,
1175 STATE_LIGHT_POSITION
, i
);
1176 struct ureg V
= get_eye_position(p
);
1177 struct ureg dist
= get_temp(p
);
1179 VPpli
= get_temp(p
);
1181 /* Calculate VPpli vector
1183 emit_op2(p
, OPCODE_SUB
, VPpli
, 0, Ppli
, V
);
1185 /* Normalize VPpli. The dist value also used in
1186 * attenuation below.
1188 emit_op2(p
, OPCODE_DP3
, dist
, 0, VPpli
, VPpli
);
1189 emit_op1(p
, OPCODE_RSQ
, dist
, 0, dist
);
1190 emit_op2(p
, OPCODE_MUL
, VPpli
, 0, VPpli
, dist
);
1192 /* Calculate attenuation:
1194 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
||
1195 p
->state
->unit
[i
].light_attenuated
) {
1196 att
= calculate_light_attenuation(p
, i
, VPpli
, dist
);
1199 /* Calculate viewer direction, or use infinite viewer:
1201 if (!p
->state
->material_shininess_is_zero
) {
1204 if (p
->state
->light_local_viewer
) {
1205 struct ureg eye_hat
= get_eye_position_normalized(p
);
1206 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1209 struct ureg z_dir
= swizzle(get_identity_param(p
),X
,Y
,W
,Z
);
1210 emit_op2(p
, OPCODE_ADD
, half
, 0, VPpli
, z_dir
);
1213 emit_normalize_vec3(p
, half
, half
);
1216 release_temp(p
, dist
);
1219 /* Calculate dot products:
1221 if (p
->state
->material_shininess_is_zero
) {
1222 emit_op2(p
, OPCODE_DP3
, dots
, 0, normal
, VPpli
);
1225 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_X
, normal
, VPpli
);
1226 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_Y
, normal
, half
);
1229 /* Front face lighting:
1232 struct ureg ambient
= get_lightprod(p
, i
, 0, STATE_AMBIENT
);
1233 struct ureg diffuse
= get_lightprod(p
, i
, 0, STATE_DIFFUSE
);
1234 struct ureg specular
= get_lightprod(p
, i
, 0, STATE_SPECULAR
);
1235 struct ureg res0
, res1
;
1236 GLuint mask0
, mask1
;
1238 if (count
== nr_lights
) {
1240 mask0
= WRITEMASK_XYZ
;
1241 mask1
= WRITEMASK_XYZ
;
1242 res0
= register_output( p
, VERT_RESULT_COL0
);
1243 res1
= register_output( p
, VERT_RESULT_COL1
);
1247 mask1
= WRITEMASK_XYZ
;
1249 res1
= register_output( p
, VERT_RESULT_COL0
);
1258 if (!is_undef(att
)) {
1259 /* light is attenuated by distance */
1260 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1261 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1262 emit_op3(p
, OPCODE_MAD
, _col0
, 0, swizzle1(lit
,X
), ambient
, _col0
);
1264 else if (!p
->state
->material_shininess_is_zero
) {
1265 /* there's a non-zero specular term */
1266 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1267 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1270 /* no attenutation, no specular */
1271 emit_degenerate_lit(p
, lit
, dots
);
1272 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1275 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _col0
);
1276 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _col1
);
1278 release_temp(p
, ambient
);
1279 release_temp(p
, diffuse
);
1280 release_temp(p
, specular
);
1283 /* Back face lighting:
1286 struct ureg ambient
= get_lightprod(p
, i
, 1, STATE_AMBIENT
);
1287 struct ureg diffuse
= get_lightprod(p
, i
, 1, STATE_DIFFUSE
);
1288 struct ureg specular
= get_lightprod(p
, i
, 1, STATE_SPECULAR
);
1289 struct ureg res0
, res1
;
1290 GLuint mask0
, mask1
;
1292 if (count
== nr_lights
) {
1294 mask0
= WRITEMASK_XYZ
;
1295 mask1
= WRITEMASK_XYZ
;
1296 res0
= register_output( p
, VERT_RESULT_BFC0
);
1297 res1
= register_output( p
, VERT_RESULT_BFC1
);
1301 mask1
= WRITEMASK_XYZ
;
1303 res1
= register_output( p
, VERT_RESULT_BFC0
);
1312 dots
= negate(swizzle(dots
,X
,Y
,W
,Z
));
1314 if (!is_undef(att
)) {
1315 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1316 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1317 emit_op3(p
, OPCODE_MAD
, _bfc0
, 0, swizzle1(lit
,X
), ambient
, _bfc0
);
1319 else if (!p
->state
->material_shininess_is_zero
) {
1320 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1321 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
);
1324 emit_degenerate_lit(p
, lit
, dots
);
1325 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
);
1328 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _bfc0
);
1329 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _bfc1
);
1330 /* restore negate flag for next lighting */
1331 dots
= negate(dots
);
1333 release_temp(p
, ambient
);
1334 release_temp(p
, diffuse
);
1335 release_temp(p
, specular
);
1338 release_temp(p
, half
);
1339 release_temp(p
, VPpli
);
1340 release_temp(p
, att
);
1348 static void build_fog( struct tnl_program
*p
)
1350 struct ureg fog
= register_output(p
, VERT_RESULT_FOGC
);
1353 if (p
->state
->fog_source_is_depth
) {
1354 input
= get_eye_position_z(p
);
1357 input
= swizzle1(register_input(p
, VERT_ATTRIB_FOG
), X
);
1360 if (p
->state
->fog_mode
&& p
->state
->tnl_do_vertex_fog
) {
1361 struct ureg params
= register_param2(p
, STATE_INTERNAL
,
1362 STATE_FOG_PARAMS_OPTIMIZED
);
1363 struct ureg tmp
= get_temp(p
);
1364 GLboolean useabs
= (p
->state
->fog_mode
!= FOG_EXP2
);
1367 emit_op1(p
, OPCODE_ABS
, tmp
, 0, input
);
1370 switch (p
->state
->fog_mode
) {
1372 struct ureg id
= get_identity_param(p
);
1373 emit_op3(p
, OPCODE_MAD
, tmp
, 0, useabs
? tmp
: input
,
1374 swizzle1(params
,X
), swizzle1(params
,Y
));
1375 emit_op2(p
, OPCODE_MAX
, tmp
, 0, tmp
, swizzle1(id
,X
)); /* saturate */
1376 emit_op2(p
, OPCODE_MIN
, fog
, WRITEMASK_X
, tmp
, swizzle1(id
,W
));
1380 emit_op2(p
, OPCODE_MUL
, tmp
, 0, useabs
? tmp
: input
,
1381 swizzle1(params
,Z
));
1382 emit_op1(p
, OPCODE_EX2
, fog
, WRITEMASK_X
, negate(tmp
));
1385 emit_op2(p
, OPCODE_MUL
, tmp
, 0, input
, swizzle1(params
,W
));
1386 emit_op2(p
, OPCODE_MUL
, tmp
, 0, tmp
, tmp
);
1387 emit_op1(p
, OPCODE_EX2
, fog
, WRITEMASK_X
, negate(tmp
));
1391 release_temp(p
, tmp
);
1394 /* results = incoming fog coords (compute fog per-fragment later)
1396 * KW: Is it really necessary to do anything in this case?
1397 * BP: Yes, we always need to compute the absolute value, unless
1398 * we want to push that down into the fragment program...
1400 GLboolean useabs
= GL_TRUE
;
1401 emit_op1(p
, useabs
? OPCODE_ABS
: OPCODE_MOV
, fog
, WRITEMASK_X
, input
);
1406 static void build_reflect_texgen( struct tnl_program
*p
,
1410 struct ureg normal
= get_transformed_normal(p
);
1411 struct ureg eye_hat
= get_eye_position_normalized(p
);
1412 struct ureg tmp
= get_temp(p
);
1415 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1417 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1419 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, negate(tmp
), normal
, eye_hat
);
1421 release_temp(p
, tmp
);
1425 static void build_sphere_texgen( struct tnl_program
*p
,
1429 struct ureg normal
= get_transformed_normal(p
);
1430 struct ureg eye_hat
= get_eye_position_normalized(p
);
1431 struct ureg tmp
= get_temp(p
);
1432 struct ureg half
= register_scalar_const(p
, .5);
1433 struct ureg r
= get_temp(p
);
1434 struct ureg inv_m
= get_temp(p
);
1435 struct ureg id
= get_identity_param(p
);
1437 /* Could share the above calculations, but it would be
1438 * a fairly odd state for someone to set (both sphere and
1439 * reflection active for different texture coordinate
1440 * components. Of course - if two texture units enable
1441 * reflect and/or sphere, things start to tilt in favour
1442 * of seperating this out:
1446 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1448 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1450 emit_op3(p
, OPCODE_MAD
, r
, 0, negate(tmp
), normal
, eye_hat
);
1452 emit_op2(p
, OPCODE_ADD
, tmp
, 0, r
, swizzle(id
,X
,Y
,W
,Z
));
1453 /* rx^2 + ry^2 + (rz+1)^2 */
1454 emit_op2(p
, OPCODE_DP3
, tmp
, 0, tmp
, tmp
);
1456 emit_op1(p
, OPCODE_RSQ
, tmp
, 0, tmp
);
1458 emit_op2(p
, OPCODE_MUL
, inv_m
, 0, tmp
, half
);
1460 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, r
, inv_m
, half
);
1462 release_temp(p
, tmp
);
1464 release_temp(p
, inv_m
);
1468 static void build_texture_transform( struct tnl_program
*p
)
1472 for (i
= 0; i
< MAX_TEXTURE_COORD_UNITS
; i
++) {
1474 if (!(p
->state
->fragprog_inputs_read
& FRAG_BIT_TEX(i
)))
1477 if (p
->state
->unit
[i
].texgen_enabled
||
1478 p
->state
->unit
[i
].texmat_enabled
) {
1480 GLuint texmat_enabled
= p
->state
->unit
[i
].texmat_enabled
;
1481 struct ureg out
= register_output(p
, VERT_RESULT_TEX0
+ i
);
1482 struct ureg out_texgen
= undef
;
1484 if (p
->state
->unit
[i
].texgen_enabled
) {
1485 GLuint copy_mask
= 0;
1486 GLuint sphere_mask
= 0;
1487 GLuint reflect_mask
= 0;
1488 GLuint normal_mask
= 0;
1492 out_texgen
= get_temp(p
);
1496 modes
[0] = p
->state
->unit
[i
].texgen_mode0
;
1497 modes
[1] = p
->state
->unit
[i
].texgen_mode1
;
1498 modes
[2] = p
->state
->unit
[i
].texgen_mode2
;
1499 modes
[3] = p
->state
->unit
[i
].texgen_mode3
;
1501 for (j
= 0; j
< 4; j
++) {
1503 case TXG_OBJ_LINEAR
: {
1504 struct ureg obj
= register_input(p
, VERT_ATTRIB_POS
);
1506 register_param3(p
, STATE_TEXGEN
, i
,
1507 STATE_TEXGEN_OBJECT_S
+ j
);
1509 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1513 case TXG_EYE_LINEAR
: {
1514 struct ureg eye
= get_eye_position(p
);
1516 register_param3(p
, STATE_TEXGEN
, i
,
1517 STATE_TEXGEN_EYE_S
+ j
);
1519 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1523 case TXG_SPHERE_MAP
:
1524 sphere_mask
|= WRITEMASK_X
<< j
;
1526 case TXG_REFLECTION_MAP
:
1527 reflect_mask
|= WRITEMASK_X
<< j
;
1529 case TXG_NORMAL_MAP
:
1530 normal_mask
|= WRITEMASK_X
<< j
;
1533 copy_mask
|= WRITEMASK_X
<< j
;
1538 build_sphere_texgen(p
, out_texgen
, sphere_mask
);
1542 build_reflect_texgen(p
, out_texgen
, reflect_mask
);
1546 struct ureg normal
= get_transformed_normal(p
);
1547 emit_op1(p
, OPCODE_MOV
, out_texgen
, normal_mask
, normal
);
1551 struct ureg in
= register_input(p
, VERT_ATTRIB_TEX0
+i
);
1552 emit_op1(p
, OPCODE_MOV
, out_texgen
, copy_mask
, in
);
1556 if (texmat_enabled
) {
1557 struct ureg texmat
[4];
1558 struct ureg in
= (!is_undef(out_texgen
) ?
1560 register_input(p
, VERT_ATTRIB_TEX0
+i
));
1562 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1564 emit_matrix_transform_vec4( p
, out
, texmat
, in
);
1567 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1568 STATE_MATRIX_TRANSPOSE
, texmat
);
1569 emit_transpose_matrix_transform_vec4( p
, out
, texmat
, in
);
1576 emit_passthrough(p
, VERT_ATTRIB_TEX0
+i
, VERT_RESULT_TEX0
+i
);
1583 * Point size attenuation computation.
1585 static void build_atten_pointsize( struct tnl_program
*p
)
1587 struct ureg eye
= get_eye_position_z(p
);
1588 struct ureg state_size
= register_param1(p
, STATE_POINT_SIZE
);
1589 struct ureg state_attenuation
= register_param1(p
, STATE_POINT_ATTENUATION
);
1590 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1591 struct ureg ut
= get_temp(p
);
1594 emit_op1(p
, OPCODE_ABS
, ut
, WRITEMASK_Y
, swizzle1(eye
, Z
));
1595 /* p1 + dist * (p2 + dist * p3); */
1596 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1597 swizzle1(state_attenuation
, Z
), swizzle1(state_attenuation
, Y
));
1598 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1599 ut
, swizzle1(state_attenuation
, X
));
1601 /* 1 / sqrt(factor) */
1602 emit_op1(p
, OPCODE_RSQ
, ut
, WRITEMASK_X
, ut
);
1605 /* out = pointSize / sqrt(factor) */
1606 emit_op2(p
, OPCODE_MUL
, out
, WRITEMASK_X
, ut
, state_size
);
1608 /* this is a good place to clamp the point size since there's likely
1609 * no hardware registers to clamp point size at rasterization time.
1611 emit_op2(p
, OPCODE_MUL
, ut
, WRITEMASK_X
, ut
, state_size
);
1612 emit_op2(p
, OPCODE_MAX
, ut
, WRITEMASK_X
, ut
, swizzle1(state_size
, Y
));
1613 emit_op2(p
, OPCODE_MIN
, out
, WRITEMASK_X
, ut
, swizzle1(state_size
, Z
));
1616 release_temp(p
, ut
);
1621 * Emit constant point size.
1623 static void build_constant_pointsize( struct tnl_program
*p
)
1625 struct ureg state_size
= register_param1(p
, STATE_POINT_SIZE
);
1626 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1627 emit_op1(p
, OPCODE_MOV
, out
, WRITEMASK_X
, state_size
);
1632 * Pass-though per-vertex point size, from user's point size array.
1634 static void build_array_pointsize( struct tnl_program
*p
)
1636 struct ureg in
= register_input(p
, VERT_ATTRIB_POINT_SIZE
);
1637 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1638 emit_op1(p
, OPCODE_MOV
, out
, WRITEMASK_X
, in
);
1642 static void build_tnl_program( struct tnl_program
*p
)
1643 { /* Emit the program, starting with modelviewproject:
1647 /* Lighting calculations:
1649 if (p
->state
->fragprog_inputs_read
& (FRAG_BIT_COL0
|FRAG_BIT_COL1
)) {
1650 if (p
->state
->light_global_enabled
)
1653 if (p
->state
->fragprog_inputs_read
& FRAG_BIT_COL0
)
1654 emit_passthrough(p
, VERT_ATTRIB_COLOR0
, VERT_RESULT_COL0
);
1656 if (p
->state
->fragprog_inputs_read
& FRAG_BIT_COL1
)
1657 emit_passthrough(p
, VERT_ATTRIB_COLOR1
, VERT_RESULT_COL1
);
1661 if ((p
->state
->fragprog_inputs_read
& FRAG_BIT_FOGC
) ||
1662 p
->state
->fog_mode
!= FOG_NONE
)
1665 if (p
->state
->fragprog_inputs_read
& FRAG_BITS_TEX_ANY
)
1666 build_texture_transform(p
);
1668 if (p
->state
->point_attenuated
)
1669 build_atten_pointsize(p
);
1670 else if (p
->state
->point_array
)
1671 build_array_pointsize(p
);
1674 build_constant_pointsize(p
);
1676 (void) build_constant_pointsize
;
1681 emit_op1(p
, OPCODE_END
, undef
, 0, undef
);
1686 _mesa_printf ("\n");
1692 create_new_program( const struct state_key
*key
,
1693 struct gl_vertex_program
*program
,
1696 struct tnl_program p
;
1698 _mesa_memset(&p
, 0, sizeof(p
));
1700 p
.program
= program
;
1701 p
.eye_position
= undef
;
1702 p
.eye_position_z
= undef
;
1703 p
.eye_position_normalized
= undef
;
1704 p
.transformed_normal
= undef
;
1708 if (max_temps
>= sizeof(int) * 8)
1709 p
.temp_reserved
= 0;
1711 p
.temp_reserved
= ~((1<<max_temps
)-1);
1713 /* Start by allocating 32 instructions.
1714 * If we need more, we'll grow the instruction array as needed.
1717 p
.program
->Base
.Instructions
= _mesa_alloc_instructions(p
.max_inst
);
1718 p
.program
->Base
.String
= NULL
;
1719 p
.program
->Base
.NumInstructions
=
1720 p
.program
->Base
.NumTemporaries
=
1721 p
.program
->Base
.NumParameters
=
1722 p
.program
->Base
.NumAttributes
= p
.program
->Base
.NumAddressRegs
= 0;
1723 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1724 p
.program
->Base
.InputsRead
= 0;
1725 p
.program
->Base
.OutputsWritten
= 0;
1727 build_tnl_program( &p
);
1732 * Return a vertex program which implements the current fixed-function
1733 * transform/lighting/texgen operations.
1734 * XXX move this into core mesa (main/)
1736 struct gl_vertex_program
*
1737 _mesa_get_fixed_func_vertex_program(GLcontext
*ctx
)
1739 struct gl_vertex_program
*prog
;
1740 struct state_key key
;
1742 /* Grab all the relevent state and put it in a single structure:
1744 make_state_key(ctx
, &key
);
1746 /* Look for an already-prepared program for this state:
1748 prog
= (struct gl_vertex_program
*)
1749 _mesa_search_program_cache(ctx
->VertexProgram
.Cache
, &key
, sizeof(key
));
1752 /* OK, we'll have to build a new one */
1754 _mesa_printf("Build new TNL program\n");
1756 prog
= (struct gl_vertex_program
*)
1757 ctx
->Driver
.NewProgram(ctx
, GL_VERTEX_PROGRAM_ARB
, 0);
1761 create_new_program( &key
, prog
,
1762 ctx
->Const
.VertexProgram
.MaxTemps
);
1765 if (ctx
->Driver
.ProgramStringNotify
)
1766 ctx
->Driver
.ProgramStringNotify( ctx
, GL_VERTEX_PROGRAM_ARB
,
1769 _mesa_program_cache_insert(ctx
, ctx
->VertexProgram
.Cache
,
1770 &key
, sizeof(key
), &prog
->Base
);