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.
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
;
103 #define TXG_OBJ_LINEAR 1
104 #define TXG_EYE_LINEAR 2
105 #define TXG_SPHERE_MAP 3
106 #define TXG_REFLECTION_MAP 4
107 #define TXG_NORMAL_MAP 5
109 static GLuint
translate_texgen( GLboolean enabled
, GLenum mode
)
115 case GL_OBJECT_LINEAR
: return TXG_OBJ_LINEAR
;
116 case GL_EYE_LINEAR
: return TXG_EYE_LINEAR
;
117 case GL_SPHERE_MAP
: return TXG_SPHERE_MAP
;
118 case GL_REFLECTION_MAP_NV
: return TXG_REFLECTION_MAP
;
119 case GL_NORMAL_MAP_NV
: return TXG_NORMAL_MAP
;
120 default: return TXG_NONE
;
126 * Returns bitmask of flags indicating which materials are set per-vertex
128 * XXX get these from the VBO...
131 tnl_get_per_vertex_materials(GLcontext
*ctx
)
133 GLbitfield mask
= 0x0;
135 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
136 struct vertex_buffer
*VB
= &tnl
->vb
;
139 for (i
= _TNL_FIRST_MAT
; i
<= _TNL_LAST_MAT
; i
++)
140 if (VB
->AttribPtr
[i
] && VB
->AttribPtr
[i
]->stride
)
141 mask
|= 1 << (i
- _TNL_FIRST_MAT
);
147 * Should fog be computed per-vertex?
150 tnl_get_per_vertex_fog(GLcontext
*ctx
)
153 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
154 return tnl
->_DoVertexFog
;
160 static GLboolean
check_active_shininess( GLcontext
*ctx
,
161 const struct state_key
*key
,
164 GLuint bit
= 1 << (MAT_ATTRIB_FRONT_SHININESS
+ side
);
166 if (key
->light_color_material_mask
& bit
)
169 if (key
->light_material_mask
& bit
)
172 if (ctx
->Light
.Material
.Attrib
[MAT_ATTRIB_FRONT_SHININESS
+ side
][0] != 0.0F
)
181 static struct state_key
*make_state_key( GLcontext
*ctx
)
183 const struct gl_fragment_program
*fp
;
184 struct state_key
*key
= CALLOC_STRUCT(state_key
);
187 fp
= ctx
->FragmentProgram
._Current
;
189 /* This now relies on texenvprogram.c being active:
193 key
->need_eye_coords
= ctx
->_NeedEyeCoords
;
195 key
->fragprog_inputs_read
= fp
->Base
.InputsRead
;
197 if (ctx
->RenderMode
== GL_FEEDBACK
) {
198 /* make sure the vertprog emits color and tex0 */
199 key
->fragprog_inputs_read
|= (FRAG_BIT_COL0
| FRAG_BIT_TEX0
);
202 key
->separate_specular
= (ctx
->Light
.Model
.ColorControl
==
203 GL_SEPARATE_SPECULAR_COLOR
);
205 if (ctx
->Light
.Enabled
) {
206 key
->light_global_enabled
= 1;
208 if (ctx
->Light
.Model
.LocalViewer
)
209 key
->light_local_viewer
= 1;
211 if (ctx
->Light
.Model
.TwoSide
)
212 key
->light_twoside
= 1;
214 if (ctx
->Light
.ColorMaterialEnabled
) {
215 key
->light_color_material
= 1;
216 key
->light_color_material_mask
= ctx
->Light
.ColorMaterialBitmask
;
219 key
->light_material_mask
= tnl_get_per_vertex_materials(ctx
);
221 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
222 struct gl_light
*light
= &ctx
->Light
.Light
[i
];
224 if (light
->Enabled
) {
225 key
->unit
[i
].light_enabled
= 1;
227 if (light
->EyePosition
[3] == 0.0)
228 key
->unit
[i
].light_eyepos3_is_zero
= 1;
230 if (light
->SpotCutoff
== 180.0)
231 key
->unit
[i
].light_spotcutoff_is_180
= 1;
233 if (light
->ConstantAttenuation
!= 1.0 ||
234 light
->LinearAttenuation
!= 0.0 ||
235 light
->QuadraticAttenuation
!= 0.0)
236 key
->unit
[i
].light_attenuated
= 1;
240 if (check_active_shininess(ctx
, key
, 0)) {
241 key
->material_shininess_is_zero
= 0;
243 else if (key
->light_twoside
&&
244 check_active_shininess(ctx
, key
, 1)) {
245 key
->material_shininess_is_zero
= 0;
248 key
->material_shininess_is_zero
= 1;
252 if (ctx
->Transform
.Normalize
)
255 if (ctx
->Transform
.RescaleNormals
)
256 key
->rescale_normals
= 1;
258 key
->fog_mode
= translate_fog_mode(fp
->FogOption
);
260 if (ctx
->Fog
.FogCoordinateSource
== GL_FRAGMENT_DEPTH_EXT
)
261 key
->fog_source_is_depth
= 1;
263 key
->tnl_do_vertex_fog
= tnl_get_per_vertex_fog(ctx
);
265 if (ctx
->Point
._Attenuated
)
266 key
->point_attenuated
= 1;
268 #if FEATURE_point_size_array
269 if (ctx
->Array
.ArrayObj
->PointSize
.Enabled
)
270 key
->point_array
= 1;
273 if (ctx
->Texture
._TexGenEnabled
||
274 ctx
->Texture
._TexMatEnabled
||
275 ctx
->Texture
._EnabledUnits
)
276 key
->texture_enabled_global
= 1;
278 for (i
= 0; i
< MAX_TEXTURE_UNITS
; i
++) {
279 struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
281 if (texUnit
->_ReallyEnabled
)
282 key
->unit
[i
].texunit_really_enabled
= 1;
284 if (ctx
->Texture
._TexMatEnabled
& ENABLE_TEXMAT(i
))
285 key
->unit
[i
].texmat_enabled
= 1;
287 if (texUnit
->TexGenEnabled
) {
288 key
->unit
[i
].texgen_enabled
= 1;
290 key
->unit
[i
].texgen_mode0
=
291 translate_texgen( texUnit
->TexGenEnabled
& (1<<0),
293 key
->unit
[i
].texgen_mode1
=
294 translate_texgen( texUnit
->TexGenEnabled
& (1<<1),
296 key
->unit
[i
].texgen_mode2
=
297 translate_texgen( texUnit
->TexGenEnabled
& (1<<2),
299 key
->unit
[i
].texgen_mode3
=
300 translate_texgen( texUnit
->TexGenEnabled
& (1<<3),
310 /* Very useful debugging tool - produces annotated listing of
311 * generated program with line/function references for each
312 * instruction back into this file:
316 /* Should be tunable by the driver - do we want to do matrix
317 * multiplications with DP4's or with MUL/MAD's? SSE works better
318 * with the latter, drivers may differ.
323 /* Use uregs to represent registers internally, translate to Mesa's
324 * expected formats on emit.
326 * NOTE: These are passed by value extensively in this file rather
327 * than as usual by pointer reference. If this disturbs you, try
328 * remembering they are just 32bits in size.
330 * GCC is smart enough to deal with these dword-sized structures in
331 * much the same way as if I had defined them as dwords and was using
332 * macros to access and set the fields. This is much nicer and easier
337 GLint idx
:9; /* relative addressing may be negative */
338 /* sizeof(idx) should == sizeof(prog_src_reg::Index) */
346 const struct state_key
*state
;
347 struct gl_vertex_program
*program
;
348 GLint max_inst
; /** number of instructions allocated for program */
351 GLuint temp_reserved
;
353 struct ureg eye_position
;
354 struct ureg eye_position_z
;
355 struct ureg eye_position_normalized
;
356 struct ureg transformed_normal
;
357 struct ureg identity
;
360 GLuint color_materials
;
364 static const struct ureg undef
= {
382 static struct ureg
make_ureg(GLuint file
, GLint idx
)
388 reg
.swz
= SWIZZLE_NOOP
;
395 static struct ureg
negate( struct ureg reg
)
402 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
404 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
407 GET_SWZ(reg
.swz
, w
));
412 static struct ureg
swizzle1( struct ureg reg
, int x
)
414 return swizzle(reg
, x
, x
, x
, x
);
417 static struct ureg
get_temp( struct tnl_program
*p
)
419 int bit
= _mesa_ffs( ~p
->temp_in_use
);
421 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
425 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
426 p
->program
->Base
.NumTemporaries
= bit
;
428 p
->temp_in_use
|= 1<<(bit
-1);
429 return make_ureg(PROGRAM_TEMPORARY
, bit
-1);
432 static struct ureg
reserve_temp( struct tnl_program
*p
)
434 struct ureg temp
= get_temp( p
);
435 p
->temp_reserved
|= 1<<temp
.idx
;
439 static void release_temp( struct tnl_program
*p
, struct ureg reg
)
441 if (reg
.file
== PROGRAM_TEMPORARY
) {
442 p
->temp_in_use
&= ~(1<<reg
.idx
);
443 p
->temp_in_use
|= p
->temp_reserved
; /* can't release reserved temps */
447 static void release_temps( struct tnl_program
*p
)
449 p
->temp_in_use
= p
->temp_reserved
;
455 * \param input one of VERT_ATTRIB_x tokens.
457 static struct ureg
register_input( struct tnl_program
*p
, GLuint input
)
459 p
->program
->Base
.InputsRead
|= (1<<input
);
460 return make_ureg(PROGRAM_INPUT
, input
);
464 * \param input one of VERT_RESULT_x tokens.
466 static struct ureg
register_output( struct tnl_program
*p
, GLuint output
)
468 p
->program
->Base
.OutputsWritten
|= (1<<output
);
469 return make_ureg(PROGRAM_OUTPUT
, output
);
472 static struct ureg
register_const4f( struct tnl_program
*p
,
485 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
487 ASSERT(swizzle
== SWIZZLE_NOOP
);
488 return make_ureg(PROGRAM_CONSTANT
, idx
);
491 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
492 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
493 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
494 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
496 static GLboolean
is_undef( struct ureg reg
)
498 return reg
.file
== PROGRAM_UNDEFINED
;
501 static struct ureg
get_identity_param( struct tnl_program
*p
)
503 if (is_undef(p
->identity
))
504 p
->identity
= register_const4f(p
, 0,0,0,1);
509 static struct ureg
register_param5(struct tnl_program
*p
,
516 gl_state_index tokens
[STATE_LENGTH
];
523 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
524 return make_ureg(PROGRAM_STATE_VAR
, idx
);
528 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
529 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
530 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
531 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
534 static void register_matrix_param5( struct tnl_program
*p
,
535 GLint s0
, /* modelview, projection, etc */
536 GLint s1
, /* texture matrix number */
537 GLint s2
, /* first row */
538 GLint s3
, /* last row */
539 GLint s4
, /* inverse, transpose, etc */
540 struct ureg
*matrix
)
544 /* This is a bit sad as the support is there to pull the whole
545 * matrix out in one go:
547 for (i
= 0; i
<= s3
- s2
; i
++)
548 matrix
[i
] = register_param5( p
, s0
, s1
, i
, i
, s4
);
552 static void emit_arg( struct prog_src_register
*src
,
555 src
->File
= reg
.file
;
556 src
->Index
= reg
.idx
;
557 src
->Swizzle
= reg
.swz
;
558 src
->NegateBase
= reg
.negate
? NEGATE_XYZW
: 0;
562 /* Check that bitfield sizes aren't exceeded */
563 ASSERT(src
->Index
== reg
.idx
);
566 static void emit_dst( struct prog_dst_register
*dst
,
567 struct ureg reg
, GLuint mask
)
569 dst
->File
= reg
.file
;
570 dst
->Index
= reg
.idx
;
571 /* allow zero as a shorthand for xyzw */
572 dst
->WriteMask
= mask
? mask
: WRITEMASK_XYZW
;
573 dst
->CondMask
= COND_TR
; /* always pass cond test */
574 dst
->CondSwizzle
= SWIZZLE_NOOP
;
577 /* Check that bitfield sizes aren't exceeded */
578 ASSERT(dst
->Index
== reg
.idx
);
581 static void debug_insn( struct prog_instruction
*inst
, const char *fn
,
585 static const char *last_fn
;
589 _mesa_printf("%s:\n", fn
);
592 _mesa_printf("%d:\t", line
);
593 _mesa_print_instruction(inst
);
598 static void emit_op3fn(struct tnl_program
*p
,
609 struct prog_instruction
*inst
;
611 assert((GLint
) p
->program
->Base
.NumInstructions
<= p
->max_inst
);
613 if (p
->program
->Base
.NumInstructions
== p
->max_inst
) {
614 /* need to extend the program's instruction array */
615 struct prog_instruction
*newInst
;
617 /* double the size */
620 newInst
= _mesa_alloc_instructions(p
->max_inst
);
622 _mesa_error(NULL
, GL_OUT_OF_MEMORY
, "vertex program build");
626 _mesa_copy_instructions(newInst
,
627 p
->program
->Base
.Instructions
,
628 p
->program
->Base
.NumInstructions
);
630 _mesa_free_instructions(p
->program
->Base
.Instructions
,
631 p
->program
->Base
.NumInstructions
);
633 p
->program
->Base
.Instructions
= newInst
;
636 nr
= p
->program
->Base
.NumInstructions
++;
638 inst
= &p
->program
->Base
.Instructions
[nr
];
639 inst
->Opcode
= (enum prog_opcode
) op
;
643 emit_arg( &inst
->SrcReg
[0], src0
);
644 emit_arg( &inst
->SrcReg
[1], src1
);
645 emit_arg( &inst
->SrcReg
[2], src2
);
647 emit_dst( &inst
->DstReg
, dest
, mask
);
649 debug_insn(inst
, fn
, line
);
653 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
654 emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
656 #define emit_op2(p, op, dst, mask, src0, src1) \
657 emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
659 #define emit_op1(p, op, dst, mask, src0) \
660 emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
663 static struct ureg
make_temp( struct tnl_program
*p
, struct ureg reg
)
665 if (reg
.file
== PROGRAM_TEMPORARY
&&
666 !(p
->temp_reserved
& (1<<reg
.idx
)))
669 struct ureg temp
= get_temp(p
);
670 emit_op1(p
, OPCODE_MOV
, temp
, 0, reg
);
676 /* Currently no tracking performed of input/output/register size or
677 * active elements. Could be used to reduce these operations, as
678 * could the matrix type.
680 static void emit_matrix_transform_vec4( struct tnl_program
*p
,
682 const struct ureg
*mat
,
685 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_X
, src
, mat
[0]);
686 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Y
, src
, mat
[1]);
687 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Z
, src
, mat
[2]);
688 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_W
, src
, mat
[3]);
691 /* This version is much easier to implement if writemasks are not
692 * supported natively on the target or (like SSE), the target doesn't
693 * have a clean/obvious dotproduct implementation.
695 static void emit_transpose_matrix_transform_vec4( struct tnl_program
*p
,
697 const struct ureg
*mat
,
702 if (dest
.file
!= PROGRAM_TEMPORARY
)
707 emit_op2(p
, OPCODE_MUL
, tmp
, 0, swizzle1(src
,X
), mat
[0]);
708 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Y
), mat
[1], tmp
);
709 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Z
), mat
[2], tmp
);
710 emit_op3(p
, OPCODE_MAD
, dest
, 0, swizzle1(src
,W
), mat
[3], tmp
);
712 if (dest
.file
!= PROGRAM_TEMPORARY
)
713 release_temp(p
, tmp
);
716 static void emit_matrix_transform_vec3( struct tnl_program
*p
,
718 const struct ureg
*mat
,
721 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_X
, src
, mat
[0]);
722 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Y
, src
, mat
[1]);
723 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Z
, src
, mat
[2]);
727 static void emit_normalize_vec3( struct tnl_program
*p
,
731 struct ureg tmp
= get_temp(p
);
732 emit_op2(p
, OPCODE_DP3
, tmp
, WRITEMASK_X
, src
, src
);
733 emit_op1(p
, OPCODE_RSQ
, tmp
, WRITEMASK_X
, tmp
);
734 emit_op2(p
, OPCODE_MUL
, dest
, 0, src
, swizzle1(tmp
, X
));
735 release_temp(p
, tmp
);
738 static void emit_passthrough( struct tnl_program
*p
,
742 struct ureg out
= register_output(p
, output
);
743 emit_op1(p
, OPCODE_MOV
, out
, 0, register_input(p
, input
));
746 static struct ureg
get_eye_position( struct tnl_program
*p
)
748 if (is_undef(p
->eye_position
)) {
749 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
750 struct ureg modelview
[4];
752 p
->eye_position
= reserve_temp(p
);
755 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
758 emit_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
761 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
762 STATE_MATRIX_TRANSPOSE
, modelview
);
764 emit_transpose_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
768 return p
->eye_position
;
772 static struct ureg
get_eye_position_z( struct tnl_program
*p
)
774 if (!is_undef(p
->eye_position
))
775 return swizzle1(p
->eye_position
, Z
);
777 if (is_undef(p
->eye_position_z
)) {
778 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
779 struct ureg modelview
[4];
781 p
->eye_position_z
= reserve_temp(p
);
783 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
786 emit_op2(p
, OPCODE_DP4
, p
->eye_position_z
, 0, pos
, modelview
[2]);
789 return p
->eye_position_z
;
794 static struct ureg
get_eye_position_normalized( struct tnl_program
*p
)
796 if (is_undef(p
->eye_position_normalized
)) {
797 struct ureg eye
= get_eye_position(p
);
798 p
->eye_position_normalized
= reserve_temp(p
);
799 emit_normalize_vec3(p
, p
->eye_position_normalized
, eye
);
802 return p
->eye_position_normalized
;
806 static struct ureg
get_transformed_normal( struct tnl_program
*p
)
808 if (is_undef(p
->transformed_normal
) &&
809 !p
->state
->need_eye_coords
&&
810 !p
->state
->normalize
&&
811 !(p
->state
->need_eye_coords
== p
->state
->rescale_normals
))
813 p
->transformed_normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
815 else if (is_undef(p
->transformed_normal
))
817 struct ureg normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
818 struct ureg mvinv
[3];
819 struct ureg transformed_normal
= reserve_temp(p
);
821 if (p
->state
->need_eye_coords
) {
822 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 2,
823 STATE_MATRIX_INVTRANS
, mvinv
);
825 /* Transform to eye space:
827 emit_matrix_transform_vec3( p
, transformed_normal
, mvinv
, normal
);
828 normal
= transformed_normal
;
831 /* Normalize/Rescale:
833 if (p
->state
->normalize
) {
834 emit_normalize_vec3( p
, transformed_normal
, normal
);
835 normal
= transformed_normal
;
837 else if (p
->state
->need_eye_coords
== p
->state
->rescale_normals
) {
838 /* This is already adjusted for eye/non-eye rendering:
840 struct ureg rescale
= register_param2(p
, STATE_INTERNAL
,
843 emit_op2( p
, OPCODE_MUL
, transformed_normal
, 0, normal
, rescale
);
844 normal
= transformed_normal
;
847 assert(normal
.file
== PROGRAM_TEMPORARY
);
848 p
->transformed_normal
= normal
;
851 return p
->transformed_normal
;
856 static void build_hpos( struct tnl_program
*p
)
858 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
859 struct ureg hpos
= register_output( p
, VERT_RESULT_HPOS
);
863 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
865 emit_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
868 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
869 STATE_MATRIX_TRANSPOSE
, mvp
);
870 emit_transpose_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
875 static GLuint
material_attrib( GLuint side
, GLuint property
)
877 return ((property
- STATE_AMBIENT
) * 2 +
881 /* Get a bitmask of which material values vary on a per-vertex basis.
883 static void set_material_flags( struct tnl_program
*p
)
885 p
->color_materials
= 0;
888 if (p
->state
->light_color_material
) {
890 p
->color_materials
= p
->state
->light_color_material_mask
;
893 p
->materials
|= p
->state
->light_material_mask
;
897 /* XXX temporary!!! */
898 #define _TNL_ATTRIB_MAT_FRONT_AMBIENT 32
900 static struct ureg
get_material( struct tnl_program
*p
, GLuint side
,
903 GLuint attrib
= material_attrib(side
, property
);
905 if (p
->color_materials
& (1<<attrib
))
906 return register_input(p
, VERT_ATTRIB_COLOR0
);
907 else if (p
->materials
& (1<<attrib
))
908 return register_input( p
, attrib
+ _TNL_ATTRIB_MAT_FRONT_AMBIENT
);
910 return register_param3( p
, STATE_MATERIAL
, side
, property
);
913 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
914 MAT_BIT_FRONT_AMBIENT | \
915 MAT_BIT_FRONT_DIFFUSE) << (side))
917 /* Either return a precalculated constant value or emit code to
918 * calculate these values dynamically in the case where material calls
919 * are present between begin/end pairs.
921 * Probably want to shift this to the program compilation phase - if
922 * we always emitted the calculation here, a smart compiler could
923 * detect that it was constant (given a certain set of inputs), and
924 * lift it out of the main loop. That way the programs created here
925 * would be independent of the vertex_buffer details.
927 static struct ureg
get_scenecolor( struct tnl_program
*p
, GLuint side
)
929 if (p
->materials
& SCENE_COLOR_BITS(side
)) {
930 struct ureg lm_ambient
= register_param1(p
, STATE_LIGHTMODEL_AMBIENT
);
931 struct ureg material_emission
= get_material(p
, side
, STATE_EMISSION
);
932 struct ureg material_ambient
= get_material(p
, side
, STATE_AMBIENT
);
933 struct ureg material_diffuse
= get_material(p
, side
, STATE_DIFFUSE
);
934 struct ureg tmp
= make_temp(p
, material_diffuse
);
935 emit_op3(p
, OPCODE_MAD
, tmp
, WRITEMASK_XYZ
, lm_ambient
,
936 material_ambient
, material_emission
);
940 return register_param2( p
, STATE_LIGHTMODEL_SCENECOLOR
, side
);
944 static struct ureg
get_lightprod( struct tnl_program
*p
, GLuint light
,
945 GLuint side
, GLuint property
)
947 GLuint attrib
= material_attrib(side
, property
);
948 if (p
->materials
& (1<<attrib
)) {
949 struct ureg light_value
=
950 register_param3(p
, STATE_LIGHT
, light
, property
);
951 struct ureg material_value
= get_material(p
, side
, property
);
952 struct ureg tmp
= get_temp(p
);
953 emit_op2(p
, OPCODE_MUL
, tmp
, 0, light_value
, material_value
);
957 return register_param4(p
, STATE_LIGHTPROD
, light
, side
, property
);
960 static struct ureg
calculate_light_attenuation( struct tnl_program
*p
,
965 struct ureg attenuation
= register_param3(p
, STATE_LIGHT
, i
,
967 struct ureg att
= get_temp(p
);
969 /* Calculate spot attenuation:
971 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
972 struct ureg spot_dir_norm
= register_param3(p
, STATE_INTERNAL
,
973 STATE_LIGHT_SPOT_DIR_NORMALIZED
, i
);
974 struct ureg spot
= get_temp(p
);
975 struct ureg slt
= get_temp(p
);
977 emit_op2(p
, OPCODE_DP3
, spot
, 0, negate(VPpli
), spot_dir_norm
);
978 emit_op2(p
, OPCODE_SLT
, slt
, 0, swizzle1(spot_dir_norm
,W
), spot
);
979 emit_op2(p
, OPCODE_POW
, spot
, 0, spot
, swizzle1(attenuation
, W
));
980 emit_op2(p
, OPCODE_MUL
, att
, 0, slt
, spot
);
982 release_temp(p
, spot
);
983 release_temp(p
, slt
);
986 /* Calculate distance attenuation:
988 if (p
->state
->unit
[i
].light_attenuated
) {
991 emit_op1(p
, OPCODE_RCP
, dist
, WRITEMASK_YZ
, dist
);
993 emit_op2(p
, OPCODE_MUL
, dist
, WRITEMASK_XZ
, dist
, swizzle1(dist
,Y
));
995 emit_op2(p
, OPCODE_DP3
, dist
, 0, attenuation
, dist
);
997 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
999 emit_op1(p
, OPCODE_RCP
, dist
, 0, dist
);
1000 /* spot-atten * dist-atten */
1001 emit_op2(p
, OPCODE_MUL
, att
, 0, dist
, att
);
1004 emit_op1(p
, OPCODE_RCP
, att
, 0, dist
);
1014 * lit.y = MAX(0, dots.x)
1015 * lit.z = SLT(0, dots.x)
1017 static void emit_degenerate_lit( struct tnl_program
*p
,
1021 struct ureg id
= get_identity_param(p
); /* id = {0,0,0,1} */
1023 /* Note that lit.x & lit.w will not be examined. Note also that
1024 * dots.xyzw == dots.xxxx.
1027 /* MAX lit, id, dots;
1029 emit_op2(p
, OPCODE_MAX
, lit
, WRITEMASK_XYZW
, id
, dots
);
1031 /* result[2] = (in > 0 ? 1 : 0)
1032 * SLT lit.z, id.z, dots; # lit.z = (0 < dots.z) ? 1 : 0
1034 emit_op2(p
, OPCODE_SLT
, lit
, WRITEMASK_Z
, swizzle1(id
,Z
), dots
);
1038 /* Need to add some addtional parameters to allow lighting in object
1039 * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
1042 static void build_lighting( struct tnl_program
*p
)
1044 const GLboolean twoside
= p
->state
->light_twoside
;
1045 const GLboolean separate
= p
->state
->separate_specular
;
1046 GLuint nr_lights
= 0, count
= 0;
1047 struct ureg normal
= get_transformed_normal(p
);
1048 struct ureg lit
= get_temp(p
);
1049 struct ureg dots
= get_temp(p
);
1050 struct ureg _col0
= undef
, _col1
= undef
;
1051 struct ureg _bfc0
= undef
, _bfc1
= undef
;
1056 * dot.x = dot(normal, VPpli)
1057 * dot.y = dot(normal, halfAngle)
1058 * dot.z = back.shininess
1059 * dot.w = front.shininess
1062 for (i
= 0; i
< MAX_LIGHTS
; i
++)
1063 if (p
->state
->unit
[i
].light_enabled
)
1066 set_material_flags(p
);
1069 if (!p
->state
->material_shininess_is_zero
) {
1070 struct ureg shininess
= get_material(p
, 0, STATE_SHININESS
);
1071 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_W
, swizzle1(shininess
,X
));
1072 release_temp(p
, shininess
);
1075 _col0
= make_temp(p
, get_scenecolor(p
, 0));
1077 _col1
= make_temp(p
, get_identity_param(p
));
1084 if (!p
->state
->material_shininess_is_zero
) {
1085 struct ureg shininess
= get_material(p
, 1, STATE_SHININESS
);
1086 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_Z
,
1087 negate(swizzle1(shininess
,X
)));
1088 release_temp(p
, shininess
);
1091 _bfc0
= make_temp(p
, get_scenecolor(p
, 1));
1093 _bfc1
= make_temp(p
, get_identity_param(p
));
1098 /* If no lights, still need to emit the scenecolor.
1101 struct ureg res0
= register_output( p
, VERT_RESULT_COL0
);
1102 emit_op1(p
, OPCODE_MOV
, res0
, 0, _col0
);
1106 struct ureg res1
= register_output( p
, VERT_RESULT_COL1
);
1107 emit_op1(p
, OPCODE_MOV
, res1
, 0, _col1
);
1111 struct ureg res0
= register_output( p
, VERT_RESULT_BFC0
);
1112 emit_op1(p
, OPCODE_MOV
, res0
, 0, _bfc0
);
1115 if (twoside
&& separate
) {
1116 struct ureg res1
= register_output( p
, VERT_RESULT_BFC1
);
1117 emit_op1(p
, OPCODE_MOV
, res1
, 0, _bfc1
);
1120 if (nr_lights
== 0) {
1125 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
1126 if (p
->state
->unit
[i
].light_enabled
) {
1127 struct ureg half
= undef
;
1128 struct ureg att
= undef
, VPpli
= undef
;
1132 if (p
->state
->unit
[i
].light_eyepos3_is_zero
) {
1133 /* Can used precomputed constants in this case.
1134 * Attenuation never applies to infinite lights.
1136 VPpli
= register_param3(p
, STATE_INTERNAL
,
1137 STATE_LIGHT_POSITION_NORMALIZED
, i
);
1139 if (!p
->state
->material_shininess_is_zero
) {
1140 if (p
->state
->light_local_viewer
) {
1141 struct ureg eye_hat
= get_eye_position_normalized(p
);
1143 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1144 emit_normalize_vec3(p
, half
, half
);
1146 half
= register_param3(p
, STATE_INTERNAL
,
1147 STATE_LIGHT_HALF_VECTOR
, i
);
1152 struct ureg Ppli
= register_param3(p
, STATE_INTERNAL
,
1153 STATE_LIGHT_POSITION
, i
);
1154 struct ureg V
= get_eye_position(p
);
1155 struct ureg dist
= get_temp(p
);
1157 VPpli
= get_temp(p
);
1159 /* Calculate VPpli vector
1161 emit_op2(p
, OPCODE_SUB
, VPpli
, 0, Ppli
, V
);
1163 /* Normalize VPpli. The dist value also used in
1164 * attenuation below.
1166 emit_op2(p
, OPCODE_DP3
, dist
, 0, VPpli
, VPpli
);
1167 emit_op1(p
, OPCODE_RSQ
, dist
, 0, dist
);
1168 emit_op2(p
, OPCODE_MUL
, VPpli
, 0, VPpli
, dist
);
1170 /* Calculate attenuation:
1172 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
||
1173 p
->state
->unit
[i
].light_attenuated
) {
1174 att
= calculate_light_attenuation(p
, i
, VPpli
, dist
);
1177 /* Calculate viewer direction, or use infinite viewer:
1179 if (!p
->state
->material_shininess_is_zero
) {
1182 if (p
->state
->light_local_viewer
) {
1183 struct ureg eye_hat
= get_eye_position_normalized(p
);
1184 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1187 struct ureg z_dir
= swizzle(get_identity_param(p
),X
,Y
,W
,Z
);
1188 emit_op2(p
, OPCODE_ADD
, half
, 0, VPpli
, z_dir
);
1191 emit_normalize_vec3(p
, half
, half
);
1194 release_temp(p
, dist
);
1197 /* Calculate dot products:
1199 if (p
->state
->material_shininess_is_zero
) {
1200 emit_op2(p
, OPCODE_DP3
, dots
, 0, normal
, VPpli
);
1203 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_X
, normal
, VPpli
);
1204 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_Y
, normal
, half
);
1207 /* Front face lighting:
1210 struct ureg ambient
= get_lightprod(p
, i
, 0, STATE_AMBIENT
);
1211 struct ureg diffuse
= get_lightprod(p
, i
, 0, STATE_DIFFUSE
);
1212 struct ureg specular
= get_lightprod(p
, i
, 0, STATE_SPECULAR
);
1213 struct ureg res0
, res1
;
1214 GLuint mask0
, mask1
;
1217 if (count
== nr_lights
) {
1219 mask0
= WRITEMASK_XYZ
;
1220 mask1
= WRITEMASK_XYZ
;
1221 res0
= register_output( p
, VERT_RESULT_COL0
);
1222 res1
= register_output( p
, VERT_RESULT_COL1
);
1226 mask1
= WRITEMASK_XYZ
;
1228 res1
= register_output( p
, VERT_RESULT_COL0
);
1238 if (!is_undef(att
)) {
1239 /* light is attenuated by distance */
1240 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1241 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1242 emit_op3(p
, OPCODE_MAD
, _col0
, 0, swizzle1(lit
,X
), ambient
, _col0
);
1244 else if (!p
->state
->material_shininess_is_zero
) {
1245 /* there's a non-zero specular term */
1246 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1247 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1250 /* no attenutation, no specular */
1251 emit_degenerate_lit(p
, lit
, dots
);
1252 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1255 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _col0
);
1256 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _col1
);
1258 release_temp(p
, ambient
);
1259 release_temp(p
, diffuse
);
1260 release_temp(p
, specular
);
1263 /* Back face lighting:
1266 struct ureg ambient
= get_lightprod(p
, i
, 1, STATE_AMBIENT
);
1267 struct ureg diffuse
= get_lightprod(p
, i
, 1, STATE_DIFFUSE
);
1268 struct ureg specular
= get_lightprod(p
, i
, 1, STATE_SPECULAR
);
1269 struct ureg res0
, res1
;
1270 GLuint mask0
, mask1
;
1272 if (count
== nr_lights
) {
1274 mask0
= WRITEMASK_XYZ
;
1275 mask1
= WRITEMASK_XYZ
;
1276 res0
= register_output( p
, VERT_RESULT_BFC0
);
1277 res1
= register_output( p
, VERT_RESULT_BFC1
);
1281 mask1
= WRITEMASK_XYZ
;
1283 res1
= register_output( p
, VERT_RESULT_BFC0
);
1292 dots
= negate(swizzle(dots
,X
,Y
,W
,Z
));
1294 if (!is_undef(att
)) {
1295 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1296 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1297 emit_op3(p
, OPCODE_MAD
, _bfc0
, 0, swizzle1(lit
,X
), ambient
, _bfc0
);
1299 else if (!p
->state
->material_shininess_is_zero
) {
1300 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1301 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1304 emit_degenerate_lit(p
, lit
, dots
);
1305 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1308 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
);
1309 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _bfc0
);
1310 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _bfc1
);
1312 release_temp(p
, ambient
);
1313 release_temp(p
, diffuse
);
1314 release_temp(p
, specular
);
1317 release_temp(p
, half
);
1318 release_temp(p
, VPpli
);
1319 release_temp(p
, att
);
1327 static void build_fog( struct tnl_program
*p
)
1329 struct ureg fog
= register_output(p
, VERT_RESULT_FOGC
);
1332 if (p
->state
->fog_source_is_depth
) {
1333 input
= get_eye_position_z(p
);
1336 input
= swizzle1(register_input(p
, VERT_ATTRIB_FOG
), X
);
1339 if (p
->state
->fog_mode
&& p
->state
->tnl_do_vertex_fog
) {
1340 struct ureg params
= register_param2(p
, STATE_INTERNAL
,
1341 STATE_FOG_PARAMS_OPTIMIZED
);
1342 struct ureg tmp
= get_temp(p
);
1343 GLboolean useabs
= (p
->state
->fog_mode
!= FOG_EXP2
);
1346 emit_op1(p
, OPCODE_ABS
, tmp
, 0, input
);
1349 switch (p
->state
->fog_mode
) {
1351 struct ureg id
= get_identity_param(p
);
1352 emit_op3(p
, OPCODE_MAD
, tmp
, 0, useabs
? tmp
: input
,
1353 swizzle1(params
,X
), swizzle1(params
,Y
));
1354 emit_op2(p
, OPCODE_MAX
, tmp
, 0, tmp
, swizzle1(id
,X
)); /* saturate */
1355 emit_op2(p
, OPCODE_MIN
, fog
, WRITEMASK_X
, tmp
, swizzle1(id
,W
));
1359 emit_op2(p
, OPCODE_MUL
, tmp
, 0, useabs
? tmp
: input
,
1360 swizzle1(params
,Z
));
1361 emit_op1(p
, OPCODE_EX2
, fog
, WRITEMASK_X
, negate(tmp
));
1364 emit_op2(p
, OPCODE_MUL
, tmp
, 0, input
, swizzle1(params
,W
));
1365 emit_op2(p
, OPCODE_MUL
, tmp
, 0, tmp
, tmp
);
1366 emit_op1(p
, OPCODE_EX2
, fog
, WRITEMASK_X
, negate(tmp
));
1370 release_temp(p
, tmp
);
1373 /* results = incoming fog coords (compute fog per-fragment later)
1375 * KW: Is it really necessary to do anything in this case?
1376 * BP: Yes, we always need to compute the absolute value, unless
1377 * we want to push that down into the fragment program...
1379 GLboolean useabs
= GL_TRUE
;
1380 emit_op1(p
, useabs
? OPCODE_ABS
: OPCODE_MOV
, fog
, WRITEMASK_X
, input
);
1384 static void build_reflect_texgen( struct tnl_program
*p
,
1388 struct ureg normal
= get_transformed_normal(p
);
1389 struct ureg eye_hat
= get_eye_position_normalized(p
);
1390 struct ureg tmp
= get_temp(p
);
1393 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1395 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1397 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, negate(tmp
), normal
, eye_hat
);
1399 release_temp(p
, tmp
);
1402 static void build_sphere_texgen( struct tnl_program
*p
,
1406 struct ureg normal
= get_transformed_normal(p
);
1407 struct ureg eye_hat
= get_eye_position_normalized(p
);
1408 struct ureg tmp
= get_temp(p
);
1409 struct ureg half
= register_scalar_const(p
, .5);
1410 struct ureg r
= get_temp(p
);
1411 struct ureg inv_m
= get_temp(p
);
1412 struct ureg id
= get_identity_param(p
);
1414 /* Could share the above calculations, but it would be
1415 * a fairly odd state for someone to set (both sphere and
1416 * reflection active for different texture coordinate
1417 * components. Of course - if two texture units enable
1418 * reflect and/or sphere, things start to tilt in favour
1419 * of seperating this out:
1423 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1425 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1427 emit_op3(p
, OPCODE_MAD
, r
, 0, negate(tmp
), normal
, eye_hat
);
1429 emit_op2(p
, OPCODE_ADD
, tmp
, 0, r
, swizzle(id
,X
,Y
,W
,Z
));
1430 /* rx^2 + ry^2 + (rz+1)^2 */
1431 emit_op2(p
, OPCODE_DP3
, tmp
, 0, tmp
, tmp
);
1433 emit_op1(p
, OPCODE_RSQ
, tmp
, 0, tmp
);
1435 emit_op2(p
, OPCODE_MUL
, inv_m
, 0, tmp
, half
);
1437 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, r
, inv_m
, half
);
1439 release_temp(p
, tmp
);
1441 release_temp(p
, inv_m
);
1445 static void build_texture_transform( struct tnl_program
*p
)
1449 for (i
= 0; i
< MAX_TEXTURE_UNITS
; i
++) {
1451 if (!(p
->state
->fragprog_inputs_read
& FRAG_BIT_TEX(i
)))
1454 if (p
->state
->unit
[i
].texgen_enabled
||
1455 p
->state
->unit
[i
].texmat_enabled
) {
1457 GLuint texmat_enabled
= p
->state
->unit
[i
].texmat_enabled
;
1458 struct ureg out
= register_output(p
, VERT_RESULT_TEX0
+ i
);
1459 struct ureg out_texgen
= undef
;
1461 if (p
->state
->unit
[i
].texgen_enabled
) {
1462 GLuint copy_mask
= 0;
1463 GLuint sphere_mask
= 0;
1464 GLuint reflect_mask
= 0;
1465 GLuint normal_mask
= 0;
1469 out_texgen
= get_temp(p
);
1473 modes
[0] = p
->state
->unit
[i
].texgen_mode0
;
1474 modes
[1] = p
->state
->unit
[i
].texgen_mode1
;
1475 modes
[2] = p
->state
->unit
[i
].texgen_mode2
;
1476 modes
[3] = p
->state
->unit
[i
].texgen_mode3
;
1478 for (j
= 0; j
< 4; j
++) {
1480 case TXG_OBJ_LINEAR
: {
1481 struct ureg obj
= register_input(p
, VERT_ATTRIB_POS
);
1483 register_param3(p
, STATE_TEXGEN
, i
,
1484 STATE_TEXGEN_OBJECT_S
+ j
);
1486 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1490 case TXG_EYE_LINEAR
: {
1491 struct ureg eye
= get_eye_position(p
);
1493 register_param3(p
, STATE_TEXGEN
, i
,
1494 STATE_TEXGEN_EYE_S
+ j
);
1496 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1500 case TXG_SPHERE_MAP
:
1501 sphere_mask
|= WRITEMASK_X
<< j
;
1503 case TXG_REFLECTION_MAP
:
1504 reflect_mask
|= WRITEMASK_X
<< j
;
1506 case TXG_NORMAL_MAP
:
1507 normal_mask
|= WRITEMASK_X
<< j
;
1510 copy_mask
|= WRITEMASK_X
<< j
;
1517 build_sphere_texgen(p
, out_texgen
, sphere_mask
);
1521 build_reflect_texgen(p
, out_texgen
, reflect_mask
);
1525 struct ureg normal
= get_transformed_normal(p
);
1526 emit_op1(p
, OPCODE_MOV
, out_texgen
, normal_mask
, normal
);
1530 struct ureg in
= register_input(p
, VERT_ATTRIB_TEX0
+i
);
1531 emit_op1(p
, OPCODE_MOV
, out_texgen
, copy_mask
, in
);
1535 if (texmat_enabled
) {
1536 struct ureg texmat
[4];
1537 struct ureg in
= (!is_undef(out_texgen
) ?
1539 register_input(p
, VERT_ATTRIB_TEX0
+i
));
1541 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1543 emit_matrix_transform_vec4( p
, out
, texmat
, in
);
1546 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1547 STATE_MATRIX_TRANSPOSE
, texmat
);
1548 emit_transpose_matrix_transform_vec4( p
, out
, texmat
, in
);
1555 emit_passthrough(p
, VERT_ATTRIB_TEX0
+i
, VERT_RESULT_TEX0
+i
);
1562 * Point size attenuation computation.
1564 static void build_atten_pointsize( struct tnl_program
*p
)
1566 struct ureg eye
= get_eye_position_z(p
);
1567 struct ureg state_size
= register_param1(p
, STATE_POINT_SIZE
);
1568 struct ureg state_attenuation
= register_param1(p
, STATE_POINT_ATTENUATION
);
1569 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1570 struct ureg ut
= get_temp(p
);
1573 emit_op1(p
, OPCODE_ABS
, ut
, WRITEMASK_Y
, swizzle1(eye
, Z
));
1574 /* p1 + dist * (p2 + dist * p3); */
1575 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1576 swizzle1(state_attenuation
, Z
), swizzle1(state_attenuation
, Y
));
1577 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1578 ut
, swizzle1(state_attenuation
, X
));
1580 /* 1 / sqrt(factor) */
1581 emit_op1(p
, OPCODE_RSQ
, ut
, WRITEMASK_X
, ut
);
1584 /* out = pointSize / sqrt(factor) */
1585 emit_op2(p
, OPCODE_MUL
, out
, WRITEMASK_X
, ut
, state_size
);
1587 /* this is a good place to clamp the point size since there's likely
1588 * no hardware registers to clamp point size at rasterization time.
1590 emit_op2(p
, OPCODE_MUL
, ut
, WRITEMASK_X
, ut
, state_size
);
1591 emit_op2(p
, OPCODE_MAX
, ut
, WRITEMASK_X
, ut
, swizzle1(state_size
, Y
));
1592 emit_op2(p
, OPCODE_MIN
, out
, WRITEMASK_X
, ut
, swizzle1(state_size
, Z
));
1595 release_temp(p
, ut
);
1599 * Emit constant point size.
1601 static void build_constant_pointsize( struct tnl_program
*p
)
1603 struct ureg state_size
= register_param1(p
, STATE_POINT_SIZE
);
1604 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1605 emit_op1(p
, OPCODE_MOV
, out
, WRITEMASK_X
, state_size
);
1609 * Pass-though per-vertex point size, from user's point size array.
1611 static void build_array_pointsize( struct tnl_program
*p
)
1613 struct ureg in
= register_input(p
, VERT_ATTRIB_POINT_SIZE
);
1614 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1615 emit_op1(p
, OPCODE_MOV
, out
, WRITEMASK_X
, in
);
1619 static void build_tnl_program( struct tnl_program
*p
)
1620 { /* Emit the program, starting with modelviewproject:
1624 /* Lighting calculations:
1626 if (p
->state
->fragprog_inputs_read
& (FRAG_BIT_COL0
|FRAG_BIT_COL1
)) {
1627 if (p
->state
->light_global_enabled
)
1630 if (p
->state
->fragprog_inputs_read
& FRAG_BIT_COL0
)
1631 emit_passthrough(p
, VERT_ATTRIB_COLOR0
, VERT_RESULT_COL0
);
1633 if (p
->state
->fragprog_inputs_read
& FRAG_BIT_COL1
)
1634 emit_passthrough(p
, VERT_ATTRIB_COLOR1
, VERT_RESULT_COL1
);
1638 if ((p
->state
->fragprog_inputs_read
& FRAG_BIT_FOGC
) ||
1639 p
->state
->fog_mode
!= FOG_NONE
)
1642 if (p
->state
->fragprog_inputs_read
& FRAG_BITS_TEX_ANY
)
1643 build_texture_transform(p
);
1645 if (p
->state
->point_attenuated
)
1646 build_atten_pointsize(p
);
1647 else if (p
->state
->point_array
)
1648 build_array_pointsize(p
);
1651 build_constant_pointsize(p
);
1653 (void) build_constant_pointsize
;
1658 emit_op1(p
, OPCODE_END
, undef
, 0, undef
);
1663 _mesa_printf ("\n");
1669 create_new_program( const struct state_key
*key
,
1670 struct gl_vertex_program
*program
,
1673 struct tnl_program p
;
1675 _mesa_memset(&p
, 0, sizeof(p
));
1677 p
.program
= program
;
1678 p
.eye_position
= undef
;
1679 p
.eye_position_z
= undef
;
1680 p
.eye_position_normalized
= undef
;
1681 p
.transformed_normal
= undef
;
1685 if (max_temps
>= sizeof(int) * 8)
1686 p
.temp_reserved
= 0;
1688 p
.temp_reserved
= ~((1<<max_temps
)-1);
1690 /* Start by allocating 32 instructions.
1691 * If we need more, we'll grow the instruction array as needed.
1694 p
.program
->Base
.Instructions
= _mesa_alloc_instructions(p
.max_inst
);
1695 p
.program
->Base
.String
= NULL
;
1696 p
.program
->Base
.NumInstructions
=
1697 p
.program
->Base
.NumTemporaries
=
1698 p
.program
->Base
.NumParameters
=
1699 p
.program
->Base
.NumAttributes
= p
.program
->Base
.NumAddressRegs
= 0;
1700 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1701 p
.program
->Base
.InputsRead
= 0;
1702 p
.program
->Base
.OutputsWritten
= 0;
1704 build_tnl_program( &p
);
1709 * Return a vertex program which implements the current fixed-function
1710 * transform/lighting/texgen operations.
1711 * XXX move this into core mesa (main/)
1713 struct gl_vertex_program
*
1714 _mesa_get_fixed_func_vertex_program(GLcontext
*ctx
)
1716 struct gl_vertex_program
*prog
;
1717 struct state_key
*key
;
1719 /* Grab all the relevent state and put it in a single structure:
1721 key
= make_state_key(ctx
);
1723 /* Look for an already-prepared program for this state:
1725 prog
= (struct gl_vertex_program
*)
1726 _mesa_search_program_cache(ctx
->VertexProgram
.Cache
, key
, sizeof(*key
));
1729 /* OK, we'll have to build a new one */
1731 _mesa_printf("Build new TNL program\n");
1733 prog
= (struct gl_vertex_program
*)
1734 ctx
->Driver
.NewProgram(ctx
, GL_VERTEX_PROGRAM_ARB
, 0);
1738 create_new_program( key
, prog
,
1739 ctx
->Const
.VertexProgram
.MaxTemps
);
1742 if (ctx
->Driver
.ProgramStringNotify
)
1743 ctx
->Driver
.ProgramStringNotify( ctx
, GL_VERTEX_PROGRAM_ARB
,
1746 _mesa_program_cache_insert(ctx
, ctx
->VertexProgram
.Cache
,
1747 key
, sizeof(*key
), &prog
->Base
);