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_color_material_mask
:12;
51 unsigned light_material_mask
:12;
52 unsigned light_global_enabled
:1;
53 unsigned light_local_viewer
:1;
54 unsigned light_twoside
:1;
55 unsigned light_color_material
:1;
56 unsigned material_shininess_is_zero
:1;
57 unsigned need_eye_coords
:1;
59 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;
70 unsigned varying_vp_inputs
;
73 unsigned light_enabled
:1;
74 unsigned light_eyepos3_is_zero
:1;
75 unsigned light_spotcutoff_is_180
:1;
76 unsigned light_attenuated
:1;
77 unsigned texunit_really_enabled
:1;
78 unsigned texmat_enabled
:1;
79 unsigned texgen_enabled
:4;
80 unsigned texgen_mode0
:4;
81 unsigned texgen_mode1
:4;
82 unsigned texgen_mode2
:4;
83 unsigned texgen_mode3
:4;
94 static GLuint
translate_fog_mode( GLenum mode
)
97 case GL_LINEAR
: return FOG_LINEAR
;
98 case GL_EXP
: return FOG_EXP
;
99 case GL_EXP2
: return FOG_EXP2
;
100 default: return FOG_NONE
;
105 #define TXG_OBJ_LINEAR 1
106 #define TXG_EYE_LINEAR 2
107 #define TXG_SPHERE_MAP 3
108 #define TXG_REFLECTION_MAP 4
109 #define TXG_NORMAL_MAP 5
111 static GLuint
translate_texgen( GLboolean enabled
, GLenum mode
)
117 case GL_OBJECT_LINEAR
: return TXG_OBJ_LINEAR
;
118 case GL_EYE_LINEAR
: return TXG_EYE_LINEAR
;
119 case GL_SPHERE_MAP
: return TXG_SPHERE_MAP
;
120 case GL_REFLECTION_MAP_NV
: return TXG_REFLECTION_MAP
;
121 case GL_NORMAL_MAP_NV
: return TXG_NORMAL_MAP
;
122 default: return TXG_NONE
;
128 * Returns bitmask of flags indicating which materials are set per-vertex
130 * XXX get these from the VBO...
133 tnl_get_per_vertex_materials(GLcontext
*ctx
)
135 GLbitfield mask
= 0x0;
137 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
138 struct vertex_buffer
*VB
= &tnl
->vb
;
141 for (i
= _TNL_FIRST_MAT
; i
<= _TNL_LAST_MAT
; i
++)
142 if (VB
->AttribPtr
[i
] && VB
->AttribPtr
[i
]->stride
)
143 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
;
162 static GLboolean
check_active_shininess( GLcontext
*ctx
,
163 const struct state_key
*key
,
166 GLuint bit
= 1 << (MAT_ATTRIB_FRONT_SHININESS
+ side
);
168 if (key
->light_color_material_mask
& bit
)
171 if (key
->light_material_mask
& bit
)
174 if (ctx
->Light
.Material
.Attrib
[MAT_ATTRIB_FRONT_SHININESS
+ side
][0] != 0.0F
)
183 static void make_state_key( GLcontext
*ctx
, struct state_key
*key
)
185 const struct gl_fragment_program
*fp
;
188 memset(key
, 0, sizeof(struct state_key
));
189 fp
= ctx
->FragmentProgram
._Current
;
191 /* This now relies on texenvprogram.c being active:
195 key
->need_eye_coords
= ctx
->_NeedEyeCoords
;
197 key
->fragprog_inputs_read
= fp
->Base
.InputsRead
;
198 key
->varying_vp_inputs
= ctx
->varying_vp_inputs
;
200 if (ctx
->RenderMode
== GL_FEEDBACK
) {
201 /* make sure the vertprog emits color and tex0 */
202 key
->fragprog_inputs_read
|= (FRAG_BIT_COL0
| FRAG_BIT_TEX0
);
205 key
->separate_specular
= (ctx
->Light
.Model
.ColorControl
==
206 GL_SEPARATE_SPECULAR_COLOR
);
208 if (ctx
->Light
.Enabled
) {
209 key
->light_global_enabled
= 1;
211 if (ctx
->Light
.Model
.LocalViewer
)
212 key
->light_local_viewer
= 1;
214 if (ctx
->Light
.Model
.TwoSide
)
215 key
->light_twoside
= 1;
217 if (ctx
->Light
.ColorMaterialEnabled
) {
218 key
->light_color_material
= 1;
219 key
->light_color_material_mask
= ctx
->Light
.ColorMaterialBitmask
;
222 key
->light_material_mask
= tnl_get_per_vertex_materials(ctx
);
224 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
225 struct gl_light
*light
= &ctx
->Light
.Light
[i
];
227 if (light
->Enabled
) {
228 key
->unit
[i
].light_enabled
= 1;
230 if (light
->EyePosition
[3] == 0.0)
231 key
->unit
[i
].light_eyepos3_is_zero
= 1;
233 if (light
->SpotCutoff
== 180.0)
234 key
->unit
[i
].light_spotcutoff_is_180
= 1;
236 if (light
->ConstantAttenuation
!= 1.0 ||
237 light
->LinearAttenuation
!= 0.0 ||
238 light
->QuadraticAttenuation
!= 0.0)
239 key
->unit
[i
].light_attenuated
= 1;
243 if (check_active_shininess(ctx
, key
, 0)) {
244 key
->material_shininess_is_zero
= 0;
246 else if (key
->light_twoside
&&
247 check_active_shininess(ctx
, key
, 1)) {
248 key
->material_shininess_is_zero
= 0;
251 key
->material_shininess_is_zero
= 1;
255 if (ctx
->Transform
.Normalize
)
258 if (ctx
->Transform
.RescaleNormals
)
259 key
->rescale_normals
= 1;
261 key
->fog_mode
= translate_fog_mode(fp
->FogOption
);
263 if (ctx
->Fog
.FogCoordinateSource
== GL_FRAGMENT_DEPTH_EXT
)
264 key
->fog_source_is_depth
= 1;
266 key
->tnl_do_vertex_fog
= tnl_get_per_vertex_fog(ctx
);
268 if (ctx
->Point
._Attenuated
)
269 key
->point_attenuated
= 1;
271 #if FEATURE_point_size_array
272 if (ctx
->Array
.ArrayObj
->PointSize
.Enabled
)
273 key
->point_array
= 1;
276 if (ctx
->Texture
._TexGenEnabled
||
277 ctx
->Texture
._TexMatEnabled
||
278 ctx
->Texture
._EnabledUnits
)
279 key
->texture_enabled_global
= 1;
281 for (i
= 0; i
< MAX_TEXTURE_UNITS
; i
++) {
282 struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
284 if (texUnit
->_ReallyEnabled
)
285 key
->unit
[i
].texunit_really_enabled
= 1;
287 if (ctx
->Texture
._TexMatEnabled
& ENABLE_TEXMAT(i
))
288 key
->unit
[i
].texmat_enabled
= 1;
290 if (texUnit
->TexGenEnabled
) {
291 key
->unit
[i
].texgen_enabled
= 1;
293 key
->unit
[i
].texgen_mode0
=
294 translate_texgen( texUnit
->TexGenEnabled
& (1<<0),
296 key
->unit
[i
].texgen_mode1
=
297 translate_texgen( texUnit
->TexGenEnabled
& (1<<1),
299 key
->unit
[i
].texgen_mode2
=
300 translate_texgen( texUnit
->TexGenEnabled
& (1<<2),
302 key
->unit
[i
].texgen_mode3
=
303 translate_texgen( texUnit
->TexGenEnabled
& (1<<3),
311 /* Very useful debugging tool - produces annotated listing of
312 * generated program with line/function references for each
313 * instruction back into this file:
317 /* Should be tunable by the driver - do we want to do matrix
318 * multiplications with DP4's or with MUL/MAD's? SSE works better
319 * with the latter, drivers may differ.
324 /* Use uregs to represent registers internally, translate to Mesa's
325 * expected formats on emit.
327 * NOTE: These are passed by value extensively in this file rather
328 * than as usual by pointer reference. If this disturbs you, try
329 * remembering they are just 32bits in size.
331 * GCC is smart enough to deal with these dword-sized structures in
332 * much the same way as if I had defined them as dwords and was using
333 * macros to access and set the fields. This is much nicer and easier
338 GLint idx
:9; /* relative addressing may be negative */
339 /* sizeof(idx) should == sizeof(prog_src_reg::Index) */
347 const struct state_key
*state
;
348 struct gl_vertex_program
*program
;
349 GLint max_inst
; /** number of instructions allocated for program */
352 GLuint temp_reserved
;
354 struct ureg eye_position
;
355 struct ureg eye_position_z
;
356 struct ureg eye_position_normalized
;
357 struct ureg transformed_normal
;
358 struct ureg identity
;
361 GLuint color_materials
;
365 static const struct ureg undef
= {
383 static struct ureg
make_ureg(GLuint file
, GLint idx
)
389 reg
.swz
= SWIZZLE_NOOP
;
396 static struct ureg
negate( struct ureg reg
)
403 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
405 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
408 GET_SWZ(reg
.swz
, w
));
413 static struct ureg
swizzle1( struct ureg reg
, int x
)
415 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);
433 static struct ureg
reserve_temp( struct tnl_program
*p
)
435 struct ureg temp
= get_temp( p
);
436 p
->temp_reserved
|= 1<<temp
.idx
;
440 static void release_temp( struct tnl_program
*p
, struct ureg reg
)
442 if (reg
.file
== PROGRAM_TEMPORARY
) {
443 p
->temp_in_use
&= ~(1<<reg
.idx
);
444 p
->temp_in_use
|= p
->temp_reserved
; /* can't release reserved temps */
448 static void release_temps( struct tnl_program
*p
)
450 p
->temp_in_use
= p
->temp_reserved
;
454 static struct ureg
register_param5(struct tnl_program
*p
,
461 gl_state_index tokens
[STATE_LENGTH
];
468 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
469 return make_ureg(PROGRAM_STATE_VAR
, idx
);
473 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
474 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
475 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
476 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
481 * \param input one of VERT_ATTRIB_x tokens.
483 static struct ureg
register_input( struct tnl_program
*p
, GLuint input
)
485 /* Material attribs are passed here as inputs >= 32
487 if (input
>= 32 || (p
->state
->varying_vp_inputs
& (1<<input
))) {
488 p
->program
->Base
.InputsRead
|= (1<<input
);
489 return make_ureg(PROGRAM_INPUT
, input
);
492 return register_param3( p
, STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, input
);
497 * \param input one of VERT_RESULT_x tokens.
499 static struct ureg
register_output( struct tnl_program
*p
, GLuint output
)
501 p
->program
->Base
.OutputsWritten
|= (1<<output
);
502 return make_ureg(PROGRAM_OUTPUT
, output
);
505 static struct ureg
register_const4f( struct tnl_program
*p
,
518 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
520 ASSERT(swizzle
== SWIZZLE_NOOP
);
521 return make_ureg(PROGRAM_CONSTANT
, idx
);
524 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
525 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
526 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
527 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
529 static GLboolean
is_undef( struct ureg reg
)
531 return reg
.file
== PROGRAM_UNDEFINED
;
534 static struct ureg
get_identity_param( struct tnl_program
*p
)
536 if (is_undef(p
->identity
))
537 p
->identity
= register_const4f(p
, 0,0,0,1);
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
);
574 static void emit_dst( struct prog_dst_register
*dst
,
575 struct ureg reg
, GLuint mask
)
577 dst
->File
= reg
.file
;
578 dst
->Index
= reg
.idx
;
579 /* allow zero as a shorthand for xyzw */
580 dst
->WriteMask
= mask
? mask
: WRITEMASK_XYZW
;
581 dst
->CondMask
= COND_TR
; /* always pass cond test */
582 dst
->CondSwizzle
= SWIZZLE_NOOP
;
585 /* Check that bitfield sizes aren't exceeded */
586 ASSERT(dst
->Index
== reg
.idx
);
589 static void debug_insn( struct prog_instruction
*inst
, const char *fn
,
593 static const char *last_fn
;
597 _mesa_printf("%s:\n", fn
);
600 _mesa_printf("%d:\t", line
);
601 _mesa_print_instruction(inst
);
606 static void emit_op3fn(struct tnl_program
*p
,
617 struct prog_instruction
*inst
;
619 assert((GLint
) p
->program
->Base
.NumInstructions
<= p
->max_inst
);
621 if (p
->program
->Base
.NumInstructions
== p
->max_inst
) {
622 /* need to extend the program's instruction array */
623 struct prog_instruction
*newInst
;
625 /* double the size */
628 newInst
= _mesa_alloc_instructions(p
->max_inst
);
630 _mesa_error(NULL
, GL_OUT_OF_MEMORY
, "vertex program build");
634 _mesa_copy_instructions(newInst
,
635 p
->program
->Base
.Instructions
,
636 p
->program
->Base
.NumInstructions
);
638 _mesa_free_instructions(p
->program
->Base
.Instructions
,
639 p
->program
->Base
.NumInstructions
);
641 p
->program
->Base
.Instructions
= newInst
;
644 nr
= p
->program
->Base
.NumInstructions
++;
646 inst
= &p
->program
->Base
.Instructions
[nr
];
647 inst
->Opcode
= (enum prog_opcode
) op
;
651 emit_arg( &inst
->SrcReg
[0], src0
);
652 emit_arg( &inst
->SrcReg
[1], src1
);
653 emit_arg( &inst
->SrcReg
[2], src2
);
655 emit_dst( &inst
->DstReg
, dest
, mask
);
657 debug_insn(inst
, fn
, line
);
661 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
662 emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
664 #define emit_op2(p, op, dst, mask, src0, src1) \
665 emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
667 #define emit_op1(p, op, dst, mask, src0) \
668 emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
671 static struct ureg
make_temp( struct tnl_program
*p
, struct ureg reg
)
673 if (reg
.file
== PROGRAM_TEMPORARY
&&
674 !(p
->temp_reserved
& (1<<reg
.idx
)))
677 struct ureg temp
= get_temp(p
);
678 emit_op1(p
, OPCODE_MOV
, temp
, 0, reg
);
684 /* Currently no tracking performed of input/output/register size or
685 * active elements. Could be used to reduce these operations, as
686 * could the matrix type.
688 static void emit_matrix_transform_vec4( struct tnl_program
*p
,
690 const struct ureg
*mat
,
693 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_X
, src
, mat
[0]);
694 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Y
, src
, mat
[1]);
695 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Z
, src
, mat
[2]);
696 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_W
, src
, mat
[3]);
699 /* This version is much easier to implement if writemasks are not
700 * supported natively on the target or (like SSE), the target doesn't
701 * have a clean/obvious dotproduct implementation.
703 static void emit_transpose_matrix_transform_vec4( struct tnl_program
*p
,
705 const struct ureg
*mat
,
710 if (dest
.file
!= PROGRAM_TEMPORARY
)
715 emit_op2(p
, OPCODE_MUL
, tmp
, 0, swizzle1(src
,X
), mat
[0]);
716 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Y
), mat
[1], tmp
);
717 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Z
), mat
[2], tmp
);
718 emit_op3(p
, OPCODE_MAD
, dest
, 0, swizzle1(src
,W
), mat
[3], tmp
);
720 if (dest
.file
!= PROGRAM_TEMPORARY
)
721 release_temp(p
, tmp
);
724 static void emit_matrix_transform_vec3( struct tnl_program
*p
,
726 const struct ureg
*mat
,
729 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_X
, src
, mat
[0]);
730 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Y
, src
, mat
[1]);
731 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Z
, src
, mat
[2]);
735 static void emit_normalize_vec3( struct tnl_program
*p
,
740 /* XXX use this when drivers are ready for NRM3 */
741 emit_op1(p
, OPCODE_NRM3
, dest
, WRITEMASK_XYZ
, src
);
743 struct ureg tmp
= get_temp(p
);
744 emit_op2(p
, OPCODE_DP3
, tmp
, WRITEMASK_X
, src
, src
);
745 emit_op1(p
, OPCODE_RSQ
, tmp
, WRITEMASK_X
, tmp
);
746 emit_op2(p
, OPCODE_MUL
, dest
, 0, src
, swizzle1(tmp
, X
));
747 release_temp(p
, tmp
);
751 static void emit_passthrough( struct tnl_program
*p
,
755 struct ureg out
= register_output(p
, output
);
756 emit_op1(p
, OPCODE_MOV
, out
, 0, register_input(p
, input
));
759 static struct ureg
get_eye_position( struct tnl_program
*p
)
761 if (is_undef(p
->eye_position
)) {
762 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
763 struct ureg modelview
[4];
765 p
->eye_position
= reserve_temp(p
);
768 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
771 emit_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
774 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
775 STATE_MATRIX_TRANSPOSE
, modelview
);
777 emit_transpose_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
781 return p
->eye_position
;
785 static struct ureg
get_eye_position_z( struct tnl_program
*p
)
787 if (!is_undef(p
->eye_position
))
788 return swizzle1(p
->eye_position
, Z
);
790 if (is_undef(p
->eye_position_z
)) {
791 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
792 struct ureg modelview
[4];
794 p
->eye_position_z
= reserve_temp(p
);
796 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
799 emit_op2(p
, OPCODE_DP4
, p
->eye_position_z
, 0, pos
, modelview
[2]);
802 return p
->eye_position_z
;
807 static struct ureg
get_eye_position_normalized( struct tnl_program
*p
)
809 if (is_undef(p
->eye_position_normalized
)) {
810 struct ureg eye
= get_eye_position(p
);
811 p
->eye_position_normalized
= reserve_temp(p
);
812 emit_normalize_vec3(p
, p
->eye_position_normalized
, eye
);
815 return p
->eye_position_normalized
;
819 static struct ureg
get_transformed_normal( struct tnl_program
*p
)
821 if (is_undef(p
->transformed_normal
) &&
822 !p
->state
->need_eye_coords
&&
823 !p
->state
->normalize
&&
824 !(p
->state
->need_eye_coords
== p
->state
->rescale_normals
))
826 p
->transformed_normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
828 else if (is_undef(p
->transformed_normal
))
830 struct ureg normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
831 struct ureg mvinv
[3];
832 struct ureg transformed_normal
= reserve_temp(p
);
834 if (p
->state
->need_eye_coords
) {
835 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 2,
836 STATE_MATRIX_INVTRANS
, mvinv
);
838 /* Transform to eye space:
840 emit_matrix_transform_vec3( p
, transformed_normal
, mvinv
, normal
);
841 normal
= transformed_normal
;
844 /* Normalize/Rescale:
846 if (p
->state
->normalize
) {
847 emit_normalize_vec3( p
, transformed_normal
, normal
);
848 normal
= transformed_normal
;
850 else if (p
->state
->need_eye_coords
== p
->state
->rescale_normals
) {
851 /* This is already adjusted for eye/non-eye rendering:
853 struct ureg rescale
= register_param2(p
, STATE_INTERNAL
,
856 emit_op2( p
, OPCODE_MUL
, transformed_normal
, 0, normal
, rescale
);
857 normal
= transformed_normal
;
860 assert(normal
.file
== PROGRAM_TEMPORARY
);
861 p
->transformed_normal
= normal
;
864 return p
->transformed_normal
;
869 static void build_hpos( struct tnl_program
*p
)
871 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
872 struct ureg hpos
= register_output( p
, VERT_RESULT_HPOS
);
876 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
878 emit_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
881 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
882 STATE_MATRIX_TRANSPOSE
, mvp
);
883 emit_transpose_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
888 static GLuint
material_attrib( GLuint side
, GLuint property
)
890 return ((property
- STATE_AMBIENT
) * 2 +
894 /* Get a bitmask of which material values vary on a per-vertex basis.
896 static void set_material_flags( struct tnl_program
*p
)
898 p
->color_materials
= 0;
901 if (p
->state
->light_color_material
) {
903 p
->color_materials
= p
->state
->light_color_material_mask
;
906 p
->materials
|= p
->state
->light_material_mask
;
910 /* XXX temporary!!! */
911 #define _TNL_ATTRIB_MAT_FRONT_AMBIENT 32
913 static struct ureg
get_material( struct tnl_program
*p
, GLuint side
,
916 GLuint attrib
= material_attrib(side
, property
);
918 if (p
->color_materials
& (1<<attrib
))
919 return register_input(p
, VERT_ATTRIB_COLOR0
);
920 else if (p
->materials
& (1<<attrib
))
921 return register_input( p
, attrib
+ _TNL_ATTRIB_MAT_FRONT_AMBIENT
);
923 return register_param3( p
, STATE_MATERIAL
, side
, property
);
926 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
927 MAT_BIT_FRONT_AMBIENT | \
928 MAT_BIT_FRONT_DIFFUSE) << (side))
930 /* Either return a precalculated constant value or emit code to
931 * calculate these values dynamically in the case where material calls
932 * are present between begin/end pairs.
934 * Probably want to shift this to the program compilation phase - if
935 * we always emitted the calculation here, a smart compiler could
936 * detect that it was constant (given a certain set of inputs), and
937 * lift it out of the main loop. That way the programs created here
938 * would be independent of the vertex_buffer details.
940 static struct ureg
get_scenecolor( struct tnl_program
*p
, GLuint side
)
942 if (p
->materials
& SCENE_COLOR_BITS(side
)) {
943 struct ureg lm_ambient
= register_param1(p
, STATE_LIGHTMODEL_AMBIENT
);
944 struct ureg material_emission
= get_material(p
, side
, STATE_EMISSION
);
945 struct ureg material_ambient
= get_material(p
, side
, STATE_AMBIENT
);
946 struct ureg material_diffuse
= get_material(p
, side
, STATE_DIFFUSE
);
947 struct ureg tmp
= make_temp(p
, material_diffuse
);
948 emit_op3(p
, OPCODE_MAD
, tmp
, WRITEMASK_XYZ
, lm_ambient
,
949 material_ambient
, material_emission
);
953 return register_param2( p
, STATE_LIGHTMODEL_SCENECOLOR
, side
);
957 static struct ureg
get_lightprod( struct tnl_program
*p
, GLuint light
,
958 GLuint side
, GLuint property
)
960 GLuint attrib
= material_attrib(side
, property
);
961 if (p
->materials
& (1<<attrib
)) {
962 struct ureg light_value
=
963 register_param3(p
, STATE_LIGHT
, light
, property
);
964 struct ureg material_value
= get_material(p
, side
, property
);
965 struct ureg tmp
= get_temp(p
);
966 emit_op2(p
, OPCODE_MUL
, tmp
, 0, light_value
, material_value
);
970 return register_param4(p
, STATE_LIGHTPROD
, light
, side
, property
);
973 static struct ureg
calculate_light_attenuation( struct tnl_program
*p
,
978 struct ureg attenuation
= register_param3(p
, STATE_LIGHT
, i
,
980 struct ureg att
= get_temp(p
);
982 /* Calculate spot attenuation:
984 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
985 struct ureg spot_dir_norm
= register_param3(p
, STATE_INTERNAL
,
986 STATE_LIGHT_SPOT_DIR_NORMALIZED
, i
);
987 struct ureg spot
= get_temp(p
);
988 struct ureg slt
= get_temp(p
);
990 emit_op2(p
, OPCODE_DP3
, spot
, 0, negate(VPpli
), spot_dir_norm
);
991 emit_op2(p
, OPCODE_SLT
, slt
, 0, swizzle1(spot_dir_norm
,W
), spot
);
992 emit_op2(p
, OPCODE_POW
, spot
, 0, spot
, swizzle1(attenuation
, W
));
993 emit_op2(p
, OPCODE_MUL
, att
, 0, slt
, spot
);
995 release_temp(p
, spot
);
996 release_temp(p
, slt
);
999 /* Calculate distance attenuation:
1001 if (p
->state
->unit
[i
].light_attenuated
) {
1004 emit_op1(p
, OPCODE_RCP
, dist
, WRITEMASK_YZ
, dist
);
1006 emit_op2(p
, OPCODE_MUL
, dist
, WRITEMASK_XZ
, dist
, swizzle1(dist
,Y
));
1008 emit_op2(p
, OPCODE_DP3
, dist
, 0, attenuation
, dist
);
1010 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
1012 emit_op1(p
, OPCODE_RCP
, dist
, 0, dist
);
1013 /* spot-atten * dist-atten */
1014 emit_op2(p
, OPCODE_MUL
, att
, 0, dist
, att
);
1017 emit_op1(p
, OPCODE_RCP
, att
, 0, dist
);
1027 * lit.y = MAX(0, dots.x)
1028 * lit.z = SLT(0, dots.x)
1030 static void emit_degenerate_lit( struct tnl_program
*p
,
1034 struct ureg id
= get_identity_param(p
); /* id = {0,0,0,1} */
1036 /* Note that lit.x & lit.w will not be examined. Note also that
1037 * dots.xyzw == dots.xxxx.
1040 /* MAX lit, id, dots;
1042 emit_op2(p
, OPCODE_MAX
, lit
, WRITEMASK_XYZW
, id
, dots
);
1044 /* result[2] = (in > 0 ? 1 : 0)
1045 * SLT lit.z, id.z, dots; # lit.z = (0 < dots.z) ? 1 : 0
1047 emit_op2(p
, OPCODE_SLT
, lit
, WRITEMASK_Z
, swizzle1(id
,Z
), dots
);
1051 /* Need to add some addtional parameters to allow lighting in object
1052 * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
1055 static void build_lighting( struct tnl_program
*p
)
1057 const GLboolean twoside
= p
->state
->light_twoside
;
1058 const GLboolean separate
= p
->state
->separate_specular
;
1059 GLuint nr_lights
= 0, count
= 0;
1060 struct ureg normal
= get_transformed_normal(p
);
1061 struct ureg lit
= get_temp(p
);
1062 struct ureg dots
= get_temp(p
);
1063 struct ureg _col0
= undef
, _col1
= undef
;
1064 struct ureg _bfc0
= undef
, _bfc1
= undef
;
1069 * dot.x = dot(normal, VPpli)
1070 * dot.y = dot(normal, halfAngle)
1071 * dot.z = back.shininess
1072 * dot.w = front.shininess
1075 for (i
= 0; i
< MAX_LIGHTS
; i
++)
1076 if (p
->state
->unit
[i
].light_enabled
)
1079 set_material_flags(p
);
1082 if (!p
->state
->material_shininess_is_zero
) {
1083 struct ureg shininess
= get_material(p
, 0, STATE_SHININESS
);
1084 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_W
, swizzle1(shininess
,X
));
1085 release_temp(p
, shininess
);
1088 _col0
= make_temp(p
, get_scenecolor(p
, 0));
1090 _col1
= make_temp(p
, get_identity_param(p
));
1097 if (!p
->state
->material_shininess_is_zero
) {
1098 struct ureg shininess
= get_material(p
, 1, STATE_SHININESS
);
1099 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_Z
,
1100 negate(swizzle1(shininess
,X
)));
1101 release_temp(p
, shininess
);
1104 _bfc0
= make_temp(p
, get_scenecolor(p
, 1));
1106 _bfc1
= make_temp(p
, get_identity_param(p
));
1111 /* If no lights, still need to emit the scenecolor.
1114 struct ureg res0
= register_output( p
, VERT_RESULT_COL0
);
1115 emit_op1(p
, OPCODE_MOV
, res0
, 0, _col0
);
1119 struct ureg res1
= register_output( p
, VERT_RESULT_COL1
);
1120 emit_op1(p
, OPCODE_MOV
, res1
, 0, _col1
);
1124 struct ureg res0
= register_output( p
, VERT_RESULT_BFC0
);
1125 emit_op1(p
, OPCODE_MOV
, res0
, 0, _bfc0
);
1128 if (twoside
&& separate
) {
1129 struct ureg res1
= register_output( p
, VERT_RESULT_BFC1
);
1130 emit_op1(p
, OPCODE_MOV
, res1
, 0, _bfc1
);
1133 if (nr_lights
== 0) {
1138 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
1139 if (p
->state
->unit
[i
].light_enabled
) {
1140 struct ureg half
= undef
;
1141 struct ureg att
= undef
, VPpli
= undef
;
1145 if (p
->state
->unit
[i
].light_eyepos3_is_zero
) {
1146 /* Can used precomputed constants in this case.
1147 * Attenuation never applies to infinite lights.
1149 VPpli
= register_param3(p
, STATE_INTERNAL
,
1150 STATE_LIGHT_POSITION_NORMALIZED
, i
);
1152 if (!p
->state
->material_shininess_is_zero
) {
1153 if (p
->state
->light_local_viewer
) {
1154 struct ureg eye_hat
= get_eye_position_normalized(p
);
1156 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1157 emit_normalize_vec3(p
, half
, half
);
1159 half
= register_param3(p
, STATE_INTERNAL
,
1160 STATE_LIGHT_HALF_VECTOR
, i
);
1165 struct ureg Ppli
= register_param3(p
, STATE_INTERNAL
,
1166 STATE_LIGHT_POSITION
, i
);
1167 struct ureg V
= get_eye_position(p
);
1168 struct ureg dist
= get_temp(p
);
1170 VPpli
= get_temp(p
);
1172 /* Calculate VPpli vector
1174 emit_op2(p
, OPCODE_SUB
, VPpli
, 0, Ppli
, V
);
1176 /* Normalize VPpli. The dist value also used in
1177 * attenuation below.
1179 emit_op2(p
, OPCODE_DP3
, dist
, 0, VPpli
, VPpli
);
1180 emit_op1(p
, OPCODE_RSQ
, dist
, 0, dist
);
1181 emit_op2(p
, OPCODE_MUL
, VPpli
, 0, VPpli
, dist
);
1183 /* Calculate attenuation:
1185 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
||
1186 p
->state
->unit
[i
].light_attenuated
) {
1187 att
= calculate_light_attenuation(p
, i
, VPpli
, dist
);
1190 /* Calculate viewer direction, or use infinite viewer:
1192 if (!p
->state
->material_shininess_is_zero
) {
1195 if (p
->state
->light_local_viewer
) {
1196 struct ureg eye_hat
= get_eye_position_normalized(p
);
1197 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1200 struct ureg z_dir
= swizzle(get_identity_param(p
),X
,Y
,W
,Z
);
1201 emit_op2(p
, OPCODE_ADD
, half
, 0, VPpli
, z_dir
);
1204 emit_normalize_vec3(p
, half
, half
);
1207 release_temp(p
, dist
);
1210 /* Calculate dot products:
1212 if (p
->state
->material_shininess_is_zero
) {
1213 emit_op2(p
, OPCODE_DP3
, dots
, 0, normal
, VPpli
);
1216 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_X
, normal
, VPpli
);
1217 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_Y
, normal
, half
);
1220 /* Front face lighting:
1223 struct ureg ambient
= get_lightprod(p
, i
, 0, STATE_AMBIENT
);
1224 struct ureg diffuse
= get_lightprod(p
, i
, 0, STATE_DIFFUSE
);
1225 struct ureg specular
= get_lightprod(p
, i
, 0, STATE_SPECULAR
);
1226 struct ureg res0
, res1
;
1227 GLuint mask0
, mask1
;
1230 if (count
== nr_lights
) {
1232 mask0
= WRITEMASK_XYZ
;
1233 mask1
= WRITEMASK_XYZ
;
1234 res0
= register_output( p
, VERT_RESULT_COL0
);
1235 res1
= register_output( p
, VERT_RESULT_COL1
);
1239 mask1
= WRITEMASK_XYZ
;
1241 res1
= register_output( p
, VERT_RESULT_COL0
);
1251 if (!is_undef(att
)) {
1252 /* light is attenuated by distance */
1253 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1254 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1255 emit_op3(p
, OPCODE_MAD
, _col0
, 0, swizzle1(lit
,X
), ambient
, _col0
);
1257 else if (!p
->state
->material_shininess_is_zero
) {
1258 /* there's a non-zero specular term */
1259 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1260 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1263 /* no attenutation, no specular */
1264 emit_degenerate_lit(p
, lit
, dots
);
1265 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1268 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _col0
);
1269 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _col1
);
1271 release_temp(p
, ambient
);
1272 release_temp(p
, diffuse
);
1273 release_temp(p
, specular
);
1276 /* Back face lighting:
1279 struct ureg ambient
= get_lightprod(p
, i
, 1, STATE_AMBIENT
);
1280 struct ureg diffuse
= get_lightprod(p
, i
, 1, STATE_DIFFUSE
);
1281 struct ureg specular
= get_lightprod(p
, i
, 1, STATE_SPECULAR
);
1282 struct ureg res0
, res1
;
1283 GLuint mask0
, mask1
;
1285 if (count
== nr_lights
) {
1287 mask0
= WRITEMASK_XYZ
;
1288 mask1
= WRITEMASK_XYZ
;
1289 res0
= register_output( p
, VERT_RESULT_BFC0
);
1290 res1
= register_output( p
, VERT_RESULT_BFC1
);
1294 mask1
= WRITEMASK_XYZ
;
1296 res1
= register_output( p
, VERT_RESULT_BFC0
);
1305 dots
= negate(swizzle(dots
,X
,Y
,W
,Z
));
1307 if (!is_undef(att
)) {
1308 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1309 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1310 emit_op3(p
, OPCODE_MAD
, _bfc0
, 0, swizzle1(lit
,X
), ambient
, _bfc0
);
1312 else if (!p
->state
->material_shininess_is_zero
) {
1313 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1314 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
);
1317 emit_degenerate_lit(p
, lit
, dots
);
1318 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
);
1321 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _bfc0
);
1322 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _bfc1
);
1324 /* restore negate flag for next lighting */
1325 dots
= negate(dots
);
1327 release_temp(p
, ambient
);
1328 release_temp(p
, diffuse
);
1329 release_temp(p
, specular
);
1332 release_temp(p
, half
);
1333 release_temp(p
, VPpli
);
1334 release_temp(p
, att
);
1342 static void build_fog( struct tnl_program
*p
)
1344 struct ureg fog
= register_output(p
, VERT_RESULT_FOGC
);
1347 if (p
->state
->fog_source_is_depth
) {
1348 input
= get_eye_position_z(p
);
1351 input
= swizzle1(register_input(p
, VERT_ATTRIB_FOG
), X
);
1354 if (p
->state
->fog_mode
&& p
->state
->tnl_do_vertex_fog
) {
1355 struct ureg params
= register_param2(p
, STATE_INTERNAL
,
1356 STATE_FOG_PARAMS_OPTIMIZED
);
1357 struct ureg tmp
= get_temp(p
);
1358 GLboolean useabs
= (p
->state
->fog_mode
!= FOG_EXP2
);
1361 emit_op1(p
, OPCODE_ABS
, tmp
, 0, input
);
1364 switch (p
->state
->fog_mode
) {
1366 struct ureg id
= get_identity_param(p
);
1367 emit_op3(p
, OPCODE_MAD
, tmp
, 0, useabs
? tmp
: input
,
1368 swizzle1(params
,X
), swizzle1(params
,Y
));
1369 emit_op2(p
, OPCODE_MAX
, tmp
, 0, tmp
, swizzle1(id
,X
)); /* saturate */
1370 emit_op2(p
, OPCODE_MIN
, fog
, WRITEMASK_X
, tmp
, swizzle1(id
,W
));
1374 emit_op2(p
, OPCODE_MUL
, tmp
, 0, useabs
? tmp
: input
,
1375 swizzle1(params
,Z
));
1376 emit_op1(p
, OPCODE_EX2
, fog
, WRITEMASK_X
, negate(tmp
));
1379 emit_op2(p
, OPCODE_MUL
, tmp
, 0, input
, swizzle1(params
,W
));
1380 emit_op2(p
, OPCODE_MUL
, tmp
, 0, tmp
, tmp
);
1381 emit_op1(p
, OPCODE_EX2
, fog
, WRITEMASK_X
, negate(tmp
));
1385 release_temp(p
, tmp
);
1388 /* results = incoming fog coords (compute fog per-fragment later)
1390 * KW: Is it really necessary to do anything in this case?
1391 * BP: Yes, we always need to compute the absolute value, unless
1392 * we want to push that down into the fragment program...
1394 GLboolean useabs
= GL_TRUE
;
1395 emit_op1(p
, useabs
? OPCODE_ABS
: OPCODE_MOV
, fog
, WRITEMASK_X
, input
);
1399 static void build_reflect_texgen( struct tnl_program
*p
,
1403 struct ureg normal
= get_transformed_normal(p
);
1404 struct ureg eye_hat
= get_eye_position_normalized(p
);
1405 struct ureg tmp
= get_temp(p
);
1408 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1410 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1412 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, negate(tmp
), normal
, eye_hat
);
1414 release_temp(p
, tmp
);
1417 static void build_sphere_texgen( struct tnl_program
*p
,
1421 struct ureg normal
= get_transformed_normal(p
);
1422 struct ureg eye_hat
= get_eye_position_normalized(p
);
1423 struct ureg tmp
= get_temp(p
);
1424 struct ureg half
= register_scalar_const(p
, .5);
1425 struct ureg r
= get_temp(p
);
1426 struct ureg inv_m
= get_temp(p
);
1427 struct ureg id
= get_identity_param(p
);
1429 /* Could share the above calculations, but it would be
1430 * a fairly odd state for someone to set (both sphere and
1431 * reflection active for different texture coordinate
1432 * components. Of course - if two texture units enable
1433 * reflect and/or sphere, things start to tilt in favour
1434 * of seperating this out:
1438 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1440 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1442 emit_op3(p
, OPCODE_MAD
, r
, 0, negate(tmp
), normal
, eye_hat
);
1444 emit_op2(p
, OPCODE_ADD
, tmp
, 0, r
, swizzle(id
,X
,Y
,W
,Z
));
1445 /* rx^2 + ry^2 + (rz+1)^2 */
1446 emit_op2(p
, OPCODE_DP3
, tmp
, 0, tmp
, tmp
);
1448 emit_op1(p
, OPCODE_RSQ
, tmp
, 0, tmp
);
1450 emit_op2(p
, OPCODE_MUL
, inv_m
, 0, tmp
, half
);
1452 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, r
, inv_m
, half
);
1454 release_temp(p
, tmp
);
1456 release_temp(p
, inv_m
);
1460 static void build_texture_transform( struct tnl_program
*p
)
1464 for (i
= 0; i
< MAX_TEXTURE_UNITS
; i
++) {
1466 if (!(p
->state
->fragprog_inputs_read
& FRAG_BIT_TEX(i
)))
1469 if (p
->state
->unit
[i
].texgen_enabled
||
1470 p
->state
->unit
[i
].texmat_enabled
) {
1472 GLuint texmat_enabled
= p
->state
->unit
[i
].texmat_enabled
;
1473 struct ureg out
= register_output(p
, VERT_RESULT_TEX0
+ i
);
1474 struct ureg out_texgen
= undef
;
1476 if (p
->state
->unit
[i
].texgen_enabled
) {
1477 GLuint copy_mask
= 0;
1478 GLuint sphere_mask
= 0;
1479 GLuint reflect_mask
= 0;
1480 GLuint normal_mask
= 0;
1484 out_texgen
= get_temp(p
);
1488 modes
[0] = p
->state
->unit
[i
].texgen_mode0
;
1489 modes
[1] = p
->state
->unit
[i
].texgen_mode1
;
1490 modes
[2] = p
->state
->unit
[i
].texgen_mode2
;
1491 modes
[3] = p
->state
->unit
[i
].texgen_mode3
;
1493 for (j
= 0; j
< 4; j
++) {
1495 case TXG_OBJ_LINEAR
: {
1496 struct ureg obj
= register_input(p
, VERT_ATTRIB_POS
);
1498 register_param3(p
, STATE_TEXGEN
, i
,
1499 STATE_TEXGEN_OBJECT_S
+ j
);
1501 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1505 case TXG_EYE_LINEAR
: {
1506 struct ureg eye
= get_eye_position(p
);
1508 register_param3(p
, STATE_TEXGEN
, i
,
1509 STATE_TEXGEN_EYE_S
+ j
);
1511 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1515 case TXG_SPHERE_MAP
:
1516 sphere_mask
|= WRITEMASK_X
<< j
;
1518 case TXG_REFLECTION_MAP
:
1519 reflect_mask
|= WRITEMASK_X
<< j
;
1521 case TXG_NORMAL_MAP
:
1522 normal_mask
|= WRITEMASK_X
<< j
;
1525 copy_mask
|= WRITEMASK_X
<< j
;
1532 build_sphere_texgen(p
, out_texgen
, sphere_mask
);
1536 build_reflect_texgen(p
, out_texgen
, reflect_mask
);
1540 struct ureg normal
= get_transformed_normal(p
);
1541 emit_op1(p
, OPCODE_MOV
, out_texgen
, normal_mask
, normal
);
1545 struct ureg in
= register_input(p
, VERT_ATTRIB_TEX0
+i
);
1546 emit_op1(p
, OPCODE_MOV
, out_texgen
, copy_mask
, in
);
1550 if (texmat_enabled
) {
1551 struct ureg texmat
[4];
1552 struct ureg in
= (!is_undef(out_texgen
) ?
1554 register_input(p
, VERT_ATTRIB_TEX0
+i
));
1556 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1558 emit_matrix_transform_vec4( p
, out
, texmat
, in
);
1561 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1562 STATE_MATRIX_TRANSPOSE
, texmat
);
1563 emit_transpose_matrix_transform_vec4( p
, out
, texmat
, in
);
1570 emit_passthrough(p
, VERT_ATTRIB_TEX0
+i
, VERT_RESULT_TEX0
+i
);
1577 * Point size attenuation computation.
1579 static void build_atten_pointsize( struct tnl_program
*p
)
1581 struct ureg eye
= get_eye_position_z(p
);
1582 struct ureg state_size
= register_param1(p
, STATE_POINT_SIZE
);
1583 struct ureg state_attenuation
= register_param1(p
, STATE_POINT_ATTENUATION
);
1584 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1585 struct ureg ut
= get_temp(p
);
1588 emit_op1(p
, OPCODE_ABS
, ut
, WRITEMASK_Y
, swizzle1(eye
, Z
));
1589 /* p1 + dist * (p2 + dist * p3); */
1590 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1591 swizzle1(state_attenuation
, Z
), swizzle1(state_attenuation
, Y
));
1592 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1593 ut
, swizzle1(state_attenuation
, X
));
1595 /* 1 / sqrt(factor) */
1596 emit_op1(p
, OPCODE_RSQ
, ut
, WRITEMASK_X
, ut
);
1599 /* out = pointSize / sqrt(factor) */
1600 emit_op2(p
, OPCODE_MUL
, out
, WRITEMASK_X
, ut
, state_size
);
1602 /* this is a good place to clamp the point size since there's likely
1603 * no hardware registers to clamp point size at rasterization time.
1605 emit_op2(p
, OPCODE_MUL
, ut
, WRITEMASK_X
, ut
, state_size
);
1606 emit_op2(p
, OPCODE_MAX
, ut
, WRITEMASK_X
, ut
, swizzle1(state_size
, Y
));
1607 emit_op2(p
, OPCODE_MIN
, out
, WRITEMASK_X
, ut
, swizzle1(state_size
, Z
));
1610 release_temp(p
, ut
);
1614 * Emit constant point size.
1616 static void build_constant_pointsize( struct tnl_program
*p
)
1618 struct ureg state_size
= register_param1(p
, STATE_POINT_SIZE
);
1619 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1620 emit_op1(p
, OPCODE_MOV
, out
, WRITEMASK_X
, state_size
);
1624 * Pass-though per-vertex point size, from user's point size array.
1626 static void build_array_pointsize( struct tnl_program
*p
)
1628 struct ureg in
= register_input(p
, VERT_ATTRIB_POINT_SIZE
);
1629 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1630 emit_op1(p
, OPCODE_MOV
, out
, WRITEMASK_X
, in
);
1634 static void build_tnl_program( struct tnl_program
*p
)
1635 { /* Emit the program, starting with modelviewproject:
1639 /* Lighting calculations:
1641 if (p
->state
->fragprog_inputs_read
& (FRAG_BIT_COL0
|FRAG_BIT_COL1
)) {
1642 if (p
->state
->light_global_enabled
)
1645 if (p
->state
->fragprog_inputs_read
& FRAG_BIT_COL0
)
1646 emit_passthrough(p
, VERT_ATTRIB_COLOR0
, VERT_RESULT_COL0
);
1648 if (p
->state
->fragprog_inputs_read
& FRAG_BIT_COL1
)
1649 emit_passthrough(p
, VERT_ATTRIB_COLOR1
, VERT_RESULT_COL1
);
1653 if ((p
->state
->fragprog_inputs_read
& FRAG_BIT_FOGC
) ||
1654 p
->state
->fog_mode
!= FOG_NONE
)
1657 if (p
->state
->fragprog_inputs_read
& FRAG_BITS_TEX_ANY
)
1658 build_texture_transform(p
);
1660 if (p
->state
->point_attenuated
)
1661 build_atten_pointsize(p
);
1662 else if (p
->state
->point_array
)
1663 build_array_pointsize(p
);
1666 build_constant_pointsize(p
);
1668 (void) build_constant_pointsize
;
1673 emit_op1(p
, OPCODE_END
, undef
, 0, undef
);
1678 _mesa_printf ("\n");
1684 create_new_program( const struct state_key
*key
,
1685 struct gl_vertex_program
*program
,
1688 struct tnl_program p
;
1690 _mesa_memset(&p
, 0, sizeof(p
));
1692 p
.program
= program
;
1693 p
.eye_position
= undef
;
1694 p
.eye_position_z
= undef
;
1695 p
.eye_position_normalized
= undef
;
1696 p
.transformed_normal
= undef
;
1700 if (max_temps
>= sizeof(int) * 8)
1701 p
.temp_reserved
= 0;
1703 p
.temp_reserved
= ~((1<<max_temps
)-1);
1705 /* Start by allocating 32 instructions.
1706 * If we need more, we'll grow the instruction array as needed.
1709 p
.program
->Base
.Instructions
= _mesa_alloc_instructions(p
.max_inst
);
1710 p
.program
->Base
.String
= NULL
;
1711 p
.program
->Base
.NumInstructions
=
1712 p
.program
->Base
.NumTemporaries
=
1713 p
.program
->Base
.NumParameters
=
1714 p
.program
->Base
.NumAttributes
= p
.program
->Base
.NumAddressRegs
= 0;
1715 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1716 p
.program
->Base
.InputsRead
= 0;
1717 p
.program
->Base
.OutputsWritten
= 0;
1719 build_tnl_program( &p
);
1724 * Return a vertex program which implements the current fixed-function
1725 * transform/lighting/texgen operations.
1726 * XXX move this into core mesa (main/)
1728 struct gl_vertex_program
*
1729 _mesa_get_fixed_func_vertex_program(GLcontext
*ctx
)
1731 struct gl_vertex_program
*prog
;
1732 struct state_key key
;
1734 /* Grab all the relevent state and put it in a single structure:
1736 make_state_key(ctx
, &key
);
1738 /* Look for an already-prepared program for this state:
1740 prog
= (struct gl_vertex_program
*)
1741 _mesa_search_program_cache(ctx
->VertexProgram
.Cache
, &key
, sizeof(key
));
1744 /* OK, we'll have to build a new one */
1746 _mesa_printf("Build new TNL program\n");
1748 prog
= (struct gl_vertex_program
*)
1749 ctx
->Driver
.NewProgram(ctx
, GL_VERTEX_PROGRAM_ARB
, 0);
1753 create_new_program( &key
, prog
,
1754 ctx
->Const
.VertexProgram
.MaxTemps
);
1757 if (ctx
->Driver
.ProgramStringNotify
)
1758 ctx
->Driver
.ProgramStringNotify( ctx
, GL_VERTEX_PROGRAM_ARB
,
1761 _mesa_program_cache_insert(ctx
, ctx
->VertexProgram
.Cache
,
1762 &key
, sizeof(key
), &prog
->Base
);