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/mfeatures.h"
40 #include "main/enums.h"
41 #include "main/ffvertex_prog.h"
42 #include "program/program.h"
43 #include "program/prog_cache.h"
44 #include "program/prog_instruction.h"
45 #include "program/prog_parameter.h"
46 #include "program/prog_print.h"
47 #include "program/prog_statevars.h"
50 /** Max of number of lights and texture coord units */
51 #define NUM_UNITS MAX2(MAX_TEXTURE_COORD_UNITS, MAX_LIGHTS)
54 unsigned light_color_material_mask
:12;
55 unsigned light_global_enabled
:1;
56 unsigned light_local_viewer
:1;
57 unsigned light_twoside
:1;
58 unsigned material_shininess_is_zero
:1;
59 unsigned need_eye_coords
:1;
61 unsigned rescale_normals
:1;
63 unsigned fog_source_is_depth
:1;
64 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 coord_replace
:1;
80 unsigned texgen_enabled
:4;
81 unsigned texgen_mode0
:4;
82 unsigned texgen_mode1
:4;
83 unsigned texgen_mode2
:4;
84 unsigned texgen_mode3
:4;
90 #define TXG_OBJ_LINEAR 1
91 #define TXG_EYE_LINEAR 2
92 #define TXG_SPHERE_MAP 3
93 #define TXG_REFLECTION_MAP 4
94 #define TXG_NORMAL_MAP 5
96 static GLuint
translate_texgen( GLboolean enabled
, GLenum mode
)
102 case GL_OBJECT_LINEAR
: return TXG_OBJ_LINEAR
;
103 case GL_EYE_LINEAR
: return TXG_EYE_LINEAR
;
104 case GL_SPHERE_MAP
: return TXG_SPHERE_MAP
;
105 case GL_REFLECTION_MAP_NV
: return TXG_REFLECTION_MAP
;
106 case GL_NORMAL_MAP_NV
: return TXG_NORMAL_MAP
;
107 default: return TXG_NONE
;
113 static GLboolean
check_active_shininess( struct gl_context
*ctx
,
114 const struct state_key
*key
,
117 GLuint bit
= 1 << (MAT_ATTRIB_FRONT_SHININESS
+ side
);
119 if ((key
->varying_vp_inputs
& VERT_BIT_COLOR0
) &&
120 (key
->light_color_material_mask
& bit
))
123 if (key
->varying_vp_inputs
& (bit
<< 16))
126 if (ctx
->Light
.Material
.Attrib
[MAT_ATTRIB_FRONT_SHININESS
+ side
][0] != 0.0F
)
133 static void make_state_key( struct gl_context
*ctx
, struct state_key
*key
)
135 const struct gl_fragment_program
*fp
;
138 memset(key
, 0, sizeof(struct state_key
));
139 fp
= ctx
->FragmentProgram
._Current
;
141 /* This now relies on texenvprogram.c being active:
145 key
->need_eye_coords
= ctx
->_NeedEyeCoords
;
147 key
->fragprog_inputs_read
= fp
->Base
.InputsRead
;
148 key
->varying_vp_inputs
= ctx
->varying_vp_inputs
;
150 if (ctx
->RenderMode
== GL_FEEDBACK
) {
151 /* make sure the vertprog emits color and tex0 */
152 key
->fragprog_inputs_read
|= (FRAG_BIT_COL0
| FRAG_BIT_TEX0
);
155 key
->separate_specular
= (ctx
->Light
.Model
.ColorControl
==
156 GL_SEPARATE_SPECULAR_COLOR
);
158 if (ctx
->Light
.Enabled
) {
159 key
->light_global_enabled
= 1;
161 if (ctx
->Light
.Model
.LocalViewer
)
162 key
->light_local_viewer
= 1;
164 if (ctx
->Light
.Model
.TwoSide
)
165 key
->light_twoside
= 1;
167 if (ctx
->Light
.ColorMaterialEnabled
) {
168 key
->light_color_material_mask
= ctx
->Light
.ColorMaterialBitmask
;
171 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
172 struct gl_light
*light
= &ctx
->Light
.Light
[i
];
174 if (light
->Enabled
) {
175 key
->unit
[i
].light_enabled
= 1;
177 if (light
->EyePosition
[3] == 0.0)
178 key
->unit
[i
].light_eyepos3_is_zero
= 1;
180 if (light
->SpotCutoff
== 180.0)
181 key
->unit
[i
].light_spotcutoff_is_180
= 1;
183 if (light
->ConstantAttenuation
!= 1.0 ||
184 light
->LinearAttenuation
!= 0.0 ||
185 light
->QuadraticAttenuation
!= 0.0)
186 key
->unit
[i
].light_attenuated
= 1;
190 if (check_active_shininess(ctx
, key
, 0)) {
191 key
->material_shininess_is_zero
= 0;
193 else if (key
->light_twoside
&&
194 check_active_shininess(ctx
, key
, 1)) {
195 key
->material_shininess_is_zero
= 0;
198 key
->material_shininess_is_zero
= 1;
202 if (ctx
->Transform
.Normalize
)
205 if (ctx
->Transform
.RescaleNormals
)
206 key
->rescale_normals
= 1;
208 if (ctx
->Fog
.FogCoordinateSource
== GL_FRAGMENT_DEPTH_EXT
)
209 key
->fog_source_is_depth
= 1;
211 if (ctx
->Point
._Attenuated
)
212 key
->point_attenuated
= 1;
214 #if FEATURE_point_size_array
215 if (ctx
->Array
.ArrayObj
->PointSize
.Enabled
)
216 key
->point_array
= 1;
219 if (ctx
->Texture
._TexGenEnabled
||
220 ctx
->Texture
._TexMatEnabled
||
221 ctx
->Texture
._EnabledUnits
)
222 key
->texture_enabled_global
= 1;
224 for (i
= 0; i
< MAX_TEXTURE_COORD_UNITS
; i
++) {
225 struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
227 if (texUnit
->_ReallyEnabled
)
228 key
->unit
[i
].texunit_really_enabled
= 1;
230 if (ctx
->Point
.PointSprite
)
231 if (ctx
->Point
.CoordReplace
[i
])
232 key
->unit
[i
].coord_replace
= 1;
234 if (ctx
->Texture
._TexMatEnabled
& ENABLE_TEXMAT(i
))
235 key
->unit
[i
].texmat_enabled
= 1;
237 if (texUnit
->TexGenEnabled
) {
238 key
->unit
[i
].texgen_enabled
= 1;
240 key
->unit
[i
].texgen_mode0
=
241 translate_texgen( texUnit
->TexGenEnabled
& (1<<0),
242 texUnit
->GenS
.Mode
);
243 key
->unit
[i
].texgen_mode1
=
244 translate_texgen( texUnit
->TexGenEnabled
& (1<<1),
245 texUnit
->GenT
.Mode
);
246 key
->unit
[i
].texgen_mode2
=
247 translate_texgen( texUnit
->TexGenEnabled
& (1<<2),
248 texUnit
->GenR
.Mode
);
249 key
->unit
[i
].texgen_mode3
=
250 translate_texgen( texUnit
->TexGenEnabled
& (1<<3),
251 texUnit
->GenQ
.Mode
);
258 /* Very useful debugging tool - produces annotated listing of
259 * generated program with line/function references for each
260 * instruction back into this file:
265 /* Use uregs to represent registers internally, translate to Mesa's
266 * expected formats on emit.
268 * NOTE: These are passed by value extensively in this file rather
269 * than as usual by pointer reference. If this disturbs you, try
270 * remembering they are just 32bits in size.
272 * GCC is smart enough to deal with these dword-sized structures in
273 * much the same way as if I had defined them as dwords and was using
274 * macros to access and set the fields. This is much nicer and easier
279 GLint idx
:9; /* relative addressing may be negative */
280 /* sizeof(idx) should == sizeof(prog_src_reg::Index) */
288 const struct state_key
*state
;
289 struct gl_vertex_program
*program
;
290 GLint max_inst
; /** number of instructions allocated for program */
291 GLboolean mvp_with_dp4
;
294 GLuint temp_reserved
;
296 struct ureg eye_position
;
297 struct ureg eye_position_z
;
298 struct ureg eye_position_normalized
;
299 struct ureg transformed_normal
;
300 struct ureg identity
;
303 GLuint color_materials
;
307 static const struct ureg undef
= {
325 static struct ureg
make_ureg(GLuint file
, GLint idx
)
331 reg
.swz
= SWIZZLE_NOOP
;
338 static struct ureg
negate( struct ureg reg
)
345 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
347 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
350 GET_SWZ(reg
.swz
, w
));
355 static struct ureg
swizzle1( struct ureg reg
, int x
)
357 return swizzle(reg
, x
, x
, x
, x
);
361 static struct ureg
get_temp( struct tnl_program
*p
)
363 int bit
= _mesa_ffs( ~p
->temp_in_use
);
365 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
369 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
370 p
->program
->Base
.NumTemporaries
= bit
;
372 p
->temp_in_use
|= 1<<(bit
-1);
373 return make_ureg(PROGRAM_TEMPORARY
, bit
-1);
377 static struct ureg
reserve_temp( struct tnl_program
*p
)
379 struct ureg temp
= get_temp( p
);
380 p
->temp_reserved
|= 1<<temp
.idx
;
385 static void release_temp( struct tnl_program
*p
, struct ureg reg
)
387 if (reg
.file
== PROGRAM_TEMPORARY
) {
388 p
->temp_in_use
&= ~(1<<reg
.idx
);
389 p
->temp_in_use
|= p
->temp_reserved
; /* can't release reserved temps */
393 static void release_temps( struct tnl_program
*p
)
395 p
->temp_in_use
= p
->temp_reserved
;
399 static struct ureg
register_param5(struct tnl_program
*p
,
406 gl_state_index tokens
[STATE_LENGTH
];
413 idx
= _mesa_add_state_reference( p
->program
->Base
.Parameters
, tokens
);
414 return make_ureg(PROGRAM_STATE_VAR
, idx
);
418 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
419 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
420 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
421 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
426 * \param input one of VERT_ATTRIB_x tokens.
428 static struct ureg
register_input( struct tnl_program
*p
, GLuint input
)
432 if (p
->state
->varying_vp_inputs
& (1<<input
)) {
433 p
->program
->Base
.InputsRead
|= (1<<input
);
434 return make_ureg(PROGRAM_INPUT
, input
);
437 return register_param3( p
, STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, input
);
443 * \param input one of VERT_RESULT_x tokens.
445 static struct ureg
register_output( struct tnl_program
*p
, GLuint output
)
447 p
->program
->Base
.OutputsWritten
|= BITFIELD64_BIT(output
);
448 return make_ureg(PROGRAM_OUTPUT
, output
);
452 static struct ureg
register_const4f( struct tnl_program
*p
,
458 gl_constant_value values
[4];
465 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
467 ASSERT(swizzle
== SWIZZLE_NOOP
);
468 return make_ureg(PROGRAM_CONSTANT
, idx
);
471 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
472 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
473 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
474 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
476 static GLboolean
is_undef( struct ureg reg
)
478 return reg
.file
== PROGRAM_UNDEFINED
;
482 static struct ureg
get_identity_param( struct tnl_program
*p
)
484 if (is_undef(p
->identity
))
485 p
->identity
= register_const4f(p
, 0,0,0,1);
490 static void register_matrix_param5( struct tnl_program
*p
,
491 GLint s0
, /* modelview, projection, etc */
492 GLint s1
, /* texture matrix number */
493 GLint s2
, /* first row */
494 GLint s3
, /* last row */
495 GLint s4
, /* inverse, transpose, etc */
496 struct ureg
*matrix
)
500 /* This is a bit sad as the support is there to pull the whole
501 * matrix out in one go:
503 for (i
= 0; i
<= s3
- s2
; i
++)
504 matrix
[i
] = register_param5( p
, s0
, s1
, i
, i
, s4
);
508 static void emit_arg( struct prog_src_register
*src
,
511 src
->File
= reg
.file
;
512 src
->Index
= reg
.idx
;
513 src
->Swizzle
= reg
.swz
;
514 src
->Negate
= reg
.negate
? NEGATE_XYZW
: NEGATE_NONE
;
517 /* Check that bitfield sizes aren't exceeded */
518 ASSERT(src
->Index
== reg
.idx
);
522 static void emit_dst( struct prog_dst_register
*dst
,
523 struct ureg reg
, GLuint mask
)
525 dst
->File
= reg
.file
;
526 dst
->Index
= reg
.idx
;
527 /* allow zero as a shorthand for xyzw */
528 dst
->WriteMask
= mask
? mask
: WRITEMASK_XYZW
;
529 dst
->CondMask
= COND_TR
; /* always pass cond test */
530 dst
->CondSwizzle
= SWIZZLE_NOOP
;
532 /* Check that bitfield sizes aren't exceeded */
533 ASSERT(dst
->Index
== reg
.idx
);
537 static void debug_insn( struct prog_instruction
*inst
, const char *fn
,
541 static const char *last_fn
;
548 printf("%d:\t", line
);
549 _mesa_print_instruction(inst
);
554 static void emit_op3fn(struct tnl_program
*p
,
565 struct prog_instruction
*inst
;
567 assert((GLint
) p
->program
->Base
.NumInstructions
<= p
->max_inst
);
569 if (p
->program
->Base
.NumInstructions
== p
->max_inst
) {
570 /* need to extend the program's instruction array */
571 struct prog_instruction
*newInst
;
573 /* double the size */
576 newInst
= _mesa_alloc_instructions(p
->max_inst
);
578 _mesa_error(NULL
, GL_OUT_OF_MEMORY
, "vertex program build");
582 _mesa_copy_instructions(newInst
,
583 p
->program
->Base
.Instructions
,
584 p
->program
->Base
.NumInstructions
);
586 _mesa_free_instructions(p
->program
->Base
.Instructions
,
587 p
->program
->Base
.NumInstructions
);
589 p
->program
->Base
.Instructions
= newInst
;
592 nr
= p
->program
->Base
.NumInstructions
++;
594 inst
= &p
->program
->Base
.Instructions
[nr
];
595 inst
->Opcode
= (enum prog_opcode
) op
;
598 emit_arg( &inst
->SrcReg
[0], src0
);
599 emit_arg( &inst
->SrcReg
[1], src1
);
600 emit_arg( &inst
->SrcReg
[2], src2
);
602 emit_dst( &inst
->DstReg
, dest
, mask
);
604 debug_insn(inst
, fn
, line
);
608 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
609 emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
611 #define emit_op2(p, op, dst, mask, src0, src1) \
612 emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
614 #define emit_op1(p, op, dst, mask, src0) \
615 emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
618 static struct ureg
make_temp( struct tnl_program
*p
, struct ureg reg
)
620 if (reg
.file
== PROGRAM_TEMPORARY
&&
621 !(p
->temp_reserved
& (1<<reg
.idx
)))
624 struct ureg temp
= get_temp(p
);
625 emit_op1(p
, OPCODE_MOV
, temp
, 0, reg
);
631 /* Currently no tracking performed of input/output/register size or
632 * active elements. Could be used to reduce these operations, as
633 * could the matrix type.
635 static void emit_matrix_transform_vec4( struct tnl_program
*p
,
637 const struct ureg
*mat
,
640 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_X
, src
, mat
[0]);
641 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Y
, src
, mat
[1]);
642 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Z
, src
, mat
[2]);
643 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_W
, src
, mat
[3]);
647 /* This version is much easier to implement if writemasks are not
648 * supported natively on the target or (like SSE), the target doesn't
649 * have a clean/obvious dotproduct implementation.
651 static void emit_transpose_matrix_transform_vec4( struct tnl_program
*p
,
653 const struct ureg
*mat
,
658 if (dest
.file
!= PROGRAM_TEMPORARY
)
663 emit_op2(p
, OPCODE_MUL
, tmp
, 0, swizzle1(src
,X
), mat
[0]);
664 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Y
), mat
[1], tmp
);
665 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Z
), mat
[2], tmp
);
666 emit_op3(p
, OPCODE_MAD
, dest
, 0, swizzle1(src
,W
), mat
[3], tmp
);
668 if (dest
.file
!= PROGRAM_TEMPORARY
)
669 release_temp(p
, tmp
);
673 static void emit_matrix_transform_vec3( struct tnl_program
*p
,
675 const struct ureg
*mat
,
678 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_X
, src
, mat
[0]);
679 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Y
, src
, mat
[1]);
680 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Z
, src
, mat
[2]);
684 static void emit_normalize_vec3( struct tnl_program
*p
,
689 /* XXX use this when drivers are ready for NRM3 */
690 emit_op1(p
, OPCODE_NRM3
, dest
, WRITEMASK_XYZ
, src
);
692 struct ureg tmp
= get_temp(p
);
693 emit_op2(p
, OPCODE_DP3
, tmp
, WRITEMASK_X
, src
, src
);
694 emit_op1(p
, OPCODE_RSQ
, tmp
, WRITEMASK_X
, tmp
);
695 emit_op2(p
, OPCODE_MUL
, dest
, 0, src
, swizzle1(tmp
, X
));
696 release_temp(p
, tmp
);
701 static void emit_passthrough( struct tnl_program
*p
,
705 struct ureg out
= register_output(p
, output
);
706 emit_op1(p
, OPCODE_MOV
, out
, 0, register_input(p
, input
));
710 static struct ureg
get_eye_position( struct tnl_program
*p
)
712 if (is_undef(p
->eye_position
)) {
713 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
714 struct ureg modelview
[4];
716 p
->eye_position
= reserve_temp(p
);
718 if (p
->mvp_with_dp4
) {
719 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
722 emit_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
725 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
726 STATE_MATRIX_TRANSPOSE
, modelview
);
728 emit_transpose_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
732 return p
->eye_position
;
736 static struct ureg
get_eye_position_z( struct tnl_program
*p
)
738 if (!is_undef(p
->eye_position
))
739 return swizzle1(p
->eye_position
, Z
);
741 if (is_undef(p
->eye_position_z
)) {
742 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
743 struct ureg modelview
[4];
745 p
->eye_position_z
= reserve_temp(p
);
747 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
750 emit_op2(p
, OPCODE_DP4
, p
->eye_position_z
, 0, pos
, modelview
[2]);
753 return p
->eye_position_z
;
757 static struct ureg
get_eye_position_normalized( struct tnl_program
*p
)
759 if (is_undef(p
->eye_position_normalized
)) {
760 struct ureg eye
= get_eye_position(p
);
761 p
->eye_position_normalized
= reserve_temp(p
);
762 emit_normalize_vec3(p
, p
->eye_position_normalized
, eye
);
765 return p
->eye_position_normalized
;
769 static struct ureg
get_transformed_normal( struct tnl_program
*p
)
771 if (is_undef(p
->transformed_normal
) &&
772 !p
->state
->need_eye_coords
&&
773 !p
->state
->normalize
&&
774 !(p
->state
->need_eye_coords
== p
->state
->rescale_normals
))
776 p
->transformed_normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
778 else if (is_undef(p
->transformed_normal
))
780 struct ureg normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
781 struct ureg mvinv
[3];
782 struct ureg transformed_normal
= reserve_temp(p
);
784 if (p
->state
->need_eye_coords
) {
785 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 2,
786 STATE_MATRIX_INVTRANS
, mvinv
);
788 /* Transform to eye space:
790 emit_matrix_transform_vec3( p
, transformed_normal
, mvinv
, normal
);
791 normal
= transformed_normal
;
794 /* Normalize/Rescale:
796 if (p
->state
->normalize
) {
797 emit_normalize_vec3( p
, transformed_normal
, normal
);
798 normal
= transformed_normal
;
800 else if (p
->state
->need_eye_coords
== p
->state
->rescale_normals
) {
801 /* This is already adjusted for eye/non-eye rendering:
803 struct ureg rescale
= register_param2(p
, STATE_INTERNAL
,
806 emit_op2( p
, OPCODE_MUL
, transformed_normal
, 0, normal
, rescale
);
807 normal
= transformed_normal
;
810 assert(normal
.file
== PROGRAM_TEMPORARY
);
811 p
->transformed_normal
= normal
;
814 return p
->transformed_normal
;
818 static void build_hpos( struct tnl_program
*p
)
820 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
821 struct ureg hpos
= register_output( p
, VERT_RESULT_HPOS
);
824 if (p
->mvp_with_dp4
) {
825 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
827 emit_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
830 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
831 STATE_MATRIX_TRANSPOSE
, mvp
);
832 emit_transpose_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
837 static GLuint
material_attrib( GLuint side
, GLuint property
)
839 return (property
- STATE_AMBIENT
) * 2 + side
;
844 * Get a bitmask of which material values vary on a per-vertex basis.
846 static void set_material_flags( struct tnl_program
*p
)
848 p
->color_materials
= 0;
851 if (p
->state
->varying_vp_inputs
& VERT_BIT_COLOR0
) {
853 p
->color_materials
= p
->state
->light_color_material_mask
;
856 p
->materials
|= (p
->state
->varying_vp_inputs
>> 16);
860 static struct ureg
get_material( struct tnl_program
*p
, GLuint side
,
863 GLuint attrib
= material_attrib(side
, property
);
865 if (p
->color_materials
& (1<<attrib
))
866 return register_input(p
, VERT_ATTRIB_COLOR0
);
867 else if (p
->materials
& (1<<attrib
)) {
868 /* Put material values in the GENERIC slots -- they are not used
869 * for anything in fixed function mode.
871 return register_input( p
, attrib
+ VERT_ATTRIB_GENERIC0
);
874 return register_param3( p
, STATE_MATERIAL
, side
, property
);
877 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
878 MAT_BIT_FRONT_AMBIENT | \
879 MAT_BIT_FRONT_DIFFUSE) << (side))
883 * Either return a precalculated constant value or emit code to
884 * calculate these values dynamically in the case where material calls
885 * are present between begin/end pairs.
887 * Probably want to shift this to the program compilation phase - if
888 * we always emitted the calculation here, a smart compiler could
889 * detect that it was constant (given a certain set of inputs), and
890 * lift it out of the main loop. That way the programs created here
891 * would be independent of the vertex_buffer details.
893 static struct ureg
get_scenecolor( struct tnl_program
*p
, GLuint side
)
895 if (p
->materials
& SCENE_COLOR_BITS(side
)) {
896 struct ureg lm_ambient
= register_param1(p
, STATE_LIGHTMODEL_AMBIENT
);
897 struct ureg material_emission
= get_material(p
, side
, STATE_EMISSION
);
898 struct ureg material_ambient
= get_material(p
, side
, STATE_AMBIENT
);
899 struct ureg material_diffuse
= get_material(p
, side
, STATE_DIFFUSE
);
900 struct ureg tmp
= make_temp(p
, material_diffuse
);
901 emit_op3(p
, OPCODE_MAD
, tmp
, WRITEMASK_XYZ
, lm_ambient
,
902 material_ambient
, material_emission
);
906 return register_param2( p
, STATE_LIGHTMODEL_SCENECOLOR
, side
);
910 static struct ureg
get_lightprod( struct tnl_program
*p
, GLuint light
,
911 GLuint side
, GLuint property
)
913 GLuint attrib
= material_attrib(side
, property
);
914 if (p
->materials
& (1<<attrib
)) {
915 struct ureg light_value
=
916 register_param3(p
, STATE_LIGHT
, light
, property
);
917 struct ureg material_value
= get_material(p
, side
, property
);
918 struct ureg tmp
= get_temp(p
);
919 emit_op2(p
, OPCODE_MUL
, tmp
, 0, light_value
, material_value
);
923 return register_param4(p
, STATE_LIGHTPROD
, light
, side
, property
);
927 static struct ureg
calculate_light_attenuation( struct tnl_program
*p
,
932 struct ureg attenuation
= register_param3(p
, STATE_LIGHT
, i
,
934 struct ureg att
= get_temp(p
);
936 /* Calculate spot attenuation:
938 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
939 struct ureg spot_dir_norm
= register_param3(p
, STATE_INTERNAL
,
940 STATE_LIGHT_SPOT_DIR_NORMALIZED
, i
);
941 struct ureg spot
= get_temp(p
);
942 struct ureg slt
= get_temp(p
);
944 emit_op2(p
, OPCODE_DP3
, spot
, 0, negate(VPpli
), spot_dir_norm
);
945 emit_op2(p
, OPCODE_SLT
, slt
, 0, swizzle1(spot_dir_norm
,W
), spot
);
946 emit_op2(p
, OPCODE_POW
, spot
, 0, spot
, swizzle1(attenuation
, W
));
947 emit_op2(p
, OPCODE_MUL
, att
, 0, slt
, spot
);
949 release_temp(p
, spot
);
950 release_temp(p
, slt
);
953 /* Calculate distance attenuation:
955 if (p
->state
->unit
[i
].light_attenuated
) {
957 emit_op1(p
, OPCODE_RCP
, dist
, WRITEMASK_YZ
, dist
);
959 emit_op2(p
, OPCODE_MUL
, dist
, WRITEMASK_XZ
, dist
, swizzle1(dist
,Y
));
961 emit_op2(p
, OPCODE_DP3
, dist
, 0, attenuation
, dist
);
963 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
965 emit_op1(p
, OPCODE_RCP
, dist
, 0, dist
);
966 /* spot-atten * dist-atten */
967 emit_op2(p
, OPCODE_MUL
, att
, 0, dist
, att
);
971 emit_op1(p
, OPCODE_RCP
, att
, 0, dist
);
981 * lit.y = MAX(0, dots.x)
982 * lit.z = SLT(0, dots.x)
984 static void emit_degenerate_lit( struct tnl_program
*p
,
988 struct ureg id
= get_identity_param(p
); /* id = {0,0,0,1} */
990 /* Note that lit.x & lit.w will not be examined. Note also that
991 * dots.xyzw == dots.xxxx.
994 /* MAX lit, id, dots;
996 emit_op2(p
, OPCODE_MAX
, lit
, WRITEMASK_XYZW
, id
, dots
);
998 /* result[2] = (in > 0 ? 1 : 0)
999 * SLT lit.z, id.z, dots; # lit.z = (0 < dots.z) ? 1 : 0
1001 emit_op2(p
, OPCODE_SLT
, lit
, WRITEMASK_Z
, swizzle1(id
,Z
), dots
);
1005 /* Need to add some addtional parameters to allow lighting in object
1006 * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
1009 static void build_lighting( struct tnl_program
*p
)
1011 const GLboolean twoside
= p
->state
->light_twoside
;
1012 const GLboolean separate
= p
->state
->separate_specular
;
1013 GLuint nr_lights
= 0, count
= 0;
1014 struct ureg normal
= get_transformed_normal(p
);
1015 struct ureg lit
= get_temp(p
);
1016 struct ureg dots
= get_temp(p
);
1017 struct ureg _col0
= undef
, _col1
= undef
;
1018 struct ureg _bfc0
= undef
, _bfc1
= undef
;
1023 * dots.x = dot(normal, VPpli)
1024 * dots.y = dot(normal, halfAngle)
1025 * dots.z = back.shininess
1026 * dots.w = front.shininess
1029 for (i
= 0; i
< MAX_LIGHTS
; i
++)
1030 if (p
->state
->unit
[i
].light_enabled
)
1033 set_material_flags(p
);
1036 if (!p
->state
->material_shininess_is_zero
) {
1037 struct ureg shininess
= get_material(p
, 0, STATE_SHININESS
);
1038 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_W
, swizzle1(shininess
,X
));
1039 release_temp(p
, shininess
);
1042 _col0
= make_temp(p
, get_scenecolor(p
, 0));
1044 _col1
= make_temp(p
, get_identity_param(p
));
1050 if (!p
->state
->material_shininess_is_zero
) {
1051 /* Note that we negate the back-face specular exponent here.
1052 * The negation will be un-done later in the back-face code below.
1054 struct ureg shininess
= get_material(p
, 1, STATE_SHININESS
);
1055 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_Z
,
1056 negate(swizzle1(shininess
,X
)));
1057 release_temp(p
, shininess
);
1060 _bfc0
= make_temp(p
, get_scenecolor(p
, 1));
1062 _bfc1
= make_temp(p
, get_identity_param(p
));
1067 /* If no lights, still need to emit the scenecolor.
1070 struct ureg res0
= register_output( p
, VERT_RESULT_COL0
);
1071 emit_op1(p
, OPCODE_MOV
, res0
, 0, _col0
);
1075 struct ureg res1
= register_output( p
, VERT_RESULT_COL1
);
1076 emit_op1(p
, OPCODE_MOV
, res1
, 0, _col1
);
1080 struct ureg res0
= register_output( p
, VERT_RESULT_BFC0
);
1081 emit_op1(p
, OPCODE_MOV
, res0
, 0, _bfc0
);
1084 if (twoside
&& separate
) {
1085 struct ureg res1
= register_output( p
, VERT_RESULT_BFC1
);
1086 emit_op1(p
, OPCODE_MOV
, res1
, 0, _bfc1
);
1089 if (nr_lights
== 0) {
1094 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
1095 if (p
->state
->unit
[i
].light_enabled
) {
1096 struct ureg half
= undef
;
1097 struct ureg att
= undef
, VPpli
= undef
;
1101 if (p
->state
->unit
[i
].light_eyepos3_is_zero
) {
1102 /* Can used precomputed constants in this case.
1103 * Attenuation never applies to infinite lights.
1105 VPpli
= register_param3(p
, STATE_INTERNAL
,
1106 STATE_LIGHT_POSITION_NORMALIZED
, i
);
1108 if (!p
->state
->material_shininess_is_zero
) {
1109 if (p
->state
->light_local_viewer
) {
1110 struct ureg eye_hat
= get_eye_position_normalized(p
);
1112 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1113 emit_normalize_vec3(p
, half
, half
);
1116 half
= register_param3(p
, STATE_INTERNAL
,
1117 STATE_LIGHT_HALF_VECTOR
, i
);
1122 struct ureg Ppli
= register_param3(p
, STATE_INTERNAL
,
1123 STATE_LIGHT_POSITION
, i
);
1124 struct ureg V
= get_eye_position(p
);
1125 struct ureg dist
= get_temp(p
);
1127 VPpli
= get_temp(p
);
1129 /* Calculate VPpli vector
1131 emit_op2(p
, OPCODE_SUB
, VPpli
, 0, Ppli
, V
);
1133 /* Normalize VPpli. The dist value also used in
1134 * attenuation below.
1136 emit_op2(p
, OPCODE_DP3
, dist
, 0, VPpli
, VPpli
);
1137 emit_op1(p
, OPCODE_RSQ
, dist
, 0, dist
);
1138 emit_op2(p
, OPCODE_MUL
, VPpli
, 0, VPpli
, dist
);
1140 /* Calculate attenuation:
1142 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
||
1143 p
->state
->unit
[i
].light_attenuated
) {
1144 att
= calculate_light_attenuation(p
, i
, VPpli
, dist
);
1147 /* Calculate viewer direction, or use infinite viewer:
1149 if (!p
->state
->material_shininess_is_zero
) {
1152 if (p
->state
->light_local_viewer
) {
1153 struct ureg eye_hat
= get_eye_position_normalized(p
);
1154 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1157 struct ureg z_dir
= swizzle(get_identity_param(p
),X
,Y
,W
,Z
);
1158 emit_op2(p
, OPCODE_ADD
, half
, 0, VPpli
, z_dir
);
1161 emit_normalize_vec3(p
, half
, half
);
1164 release_temp(p
, dist
);
1167 /* Calculate dot products:
1169 if (p
->state
->material_shininess_is_zero
) {
1170 emit_op2(p
, OPCODE_DP3
, dots
, 0, normal
, VPpli
);
1173 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_X
, normal
, VPpli
);
1174 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_Y
, normal
, half
);
1177 /* Front face lighting:
1180 struct ureg ambient
= get_lightprod(p
, i
, 0, STATE_AMBIENT
);
1181 struct ureg diffuse
= get_lightprod(p
, i
, 0, STATE_DIFFUSE
);
1182 struct ureg specular
= get_lightprod(p
, i
, 0, STATE_SPECULAR
);
1183 struct ureg res0
, res1
;
1184 GLuint mask0
, mask1
;
1186 if (count
== nr_lights
) {
1188 mask0
= WRITEMASK_XYZ
;
1189 mask1
= WRITEMASK_XYZ
;
1190 res0
= register_output( p
, VERT_RESULT_COL0
);
1191 res1
= register_output( p
, VERT_RESULT_COL1
);
1195 mask1
= WRITEMASK_XYZ
;
1197 res1
= register_output( p
, VERT_RESULT_COL0
);
1207 if (!is_undef(att
)) {
1208 /* light is attenuated by distance */
1209 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1210 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1211 emit_op3(p
, OPCODE_MAD
, _col0
, 0, swizzle1(lit
,X
), ambient
, _col0
);
1213 else if (!p
->state
->material_shininess_is_zero
) {
1214 /* there's a non-zero specular term */
1215 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1216 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1219 /* no attenutation, no specular */
1220 emit_degenerate_lit(p
, lit
, dots
);
1221 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1224 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _col0
);
1225 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _col1
);
1227 release_temp(p
, ambient
);
1228 release_temp(p
, diffuse
);
1229 release_temp(p
, specular
);
1232 /* Back face lighting:
1235 struct ureg ambient
= get_lightprod(p
, i
, 1, STATE_AMBIENT
);
1236 struct ureg diffuse
= get_lightprod(p
, i
, 1, STATE_DIFFUSE
);
1237 struct ureg specular
= get_lightprod(p
, i
, 1, STATE_SPECULAR
);
1238 struct ureg res0
, res1
;
1239 GLuint mask0
, mask1
;
1241 if (count
== nr_lights
) {
1243 mask0
= WRITEMASK_XYZ
;
1244 mask1
= WRITEMASK_XYZ
;
1245 res0
= register_output( p
, VERT_RESULT_BFC0
);
1246 res1
= register_output( p
, VERT_RESULT_BFC1
);
1250 mask1
= WRITEMASK_XYZ
;
1252 res1
= register_output( p
, VERT_RESULT_BFC0
);
1262 /* For the back face we need to negate the X and Y component
1263 * dot products. dots.Z has the negated back-face specular
1264 * exponent. We swizzle that into the W position. This
1265 * negation makes the back-face specular term positive again.
1267 dots
= negate(swizzle(dots
,X
,Y
,W
,Z
));
1269 if (!is_undef(att
)) {
1270 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1271 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1272 emit_op3(p
, OPCODE_MAD
, _bfc0
, 0, swizzle1(lit
,X
), ambient
, _bfc0
);
1274 else if (!p
->state
->material_shininess_is_zero
) {
1275 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1276 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
); /**/
1279 emit_degenerate_lit(p
, lit
, dots
);
1280 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
);
1283 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _bfc0
);
1284 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _bfc1
);
1285 /* restore dots to its original state for subsequent lights
1286 * by negating and swizzling again.
1288 dots
= negate(swizzle(dots
,X
,Y
,W
,Z
));
1290 release_temp(p
, ambient
);
1291 release_temp(p
, diffuse
);
1292 release_temp(p
, specular
);
1295 release_temp(p
, half
);
1296 release_temp(p
, VPpli
);
1297 release_temp(p
, att
);
1305 static void build_fog( struct tnl_program
*p
)
1307 struct ureg fog
= register_output(p
, VERT_RESULT_FOGC
);
1310 if (p
->state
->fog_source_is_depth
) {
1311 input
= get_eye_position_z(p
);
1314 input
= swizzle1(register_input(p
, VERT_ATTRIB_FOG
), X
);
1317 /* result.fog = {abs(f),0,0,1}; */
1318 emit_op1(p
, OPCODE_ABS
, fog
, WRITEMASK_X
, input
);
1319 emit_op1(p
, OPCODE_MOV
, fog
, WRITEMASK_YZW
, get_identity_param(p
));
1323 static void build_reflect_texgen( struct tnl_program
*p
,
1327 struct ureg normal
= get_transformed_normal(p
);
1328 struct ureg eye_hat
= get_eye_position_normalized(p
);
1329 struct ureg tmp
= get_temp(p
);
1332 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1334 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1336 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, negate(tmp
), normal
, eye_hat
);
1338 release_temp(p
, tmp
);
1342 static void build_sphere_texgen( struct tnl_program
*p
,
1346 struct ureg normal
= get_transformed_normal(p
);
1347 struct ureg eye_hat
= get_eye_position_normalized(p
);
1348 struct ureg tmp
= get_temp(p
);
1349 struct ureg half
= register_scalar_const(p
, .5);
1350 struct ureg r
= get_temp(p
);
1351 struct ureg inv_m
= get_temp(p
);
1352 struct ureg id
= get_identity_param(p
);
1354 /* Could share the above calculations, but it would be
1355 * a fairly odd state for someone to set (both sphere and
1356 * reflection active for different texture coordinate
1357 * components. Of course - if two texture units enable
1358 * reflect and/or sphere, things start to tilt in favour
1359 * of seperating this out:
1363 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1365 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1367 emit_op3(p
, OPCODE_MAD
, r
, 0, negate(tmp
), normal
, eye_hat
);
1369 emit_op2(p
, OPCODE_ADD
, tmp
, 0, r
, swizzle(id
,X
,Y
,W
,Z
));
1370 /* rx^2 + ry^2 + (rz+1)^2 */
1371 emit_op2(p
, OPCODE_DP3
, tmp
, 0, tmp
, tmp
);
1373 emit_op1(p
, OPCODE_RSQ
, tmp
, 0, tmp
);
1375 emit_op2(p
, OPCODE_MUL
, inv_m
, 0, tmp
, half
);
1377 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, r
, inv_m
, half
);
1379 release_temp(p
, tmp
);
1381 release_temp(p
, inv_m
);
1385 static void build_texture_transform( struct tnl_program
*p
)
1389 for (i
= 0; i
< MAX_TEXTURE_COORD_UNITS
; i
++) {
1391 if (!(p
->state
->fragprog_inputs_read
& FRAG_BIT_TEX(i
)))
1394 if (p
->state
->unit
[i
].coord_replace
)
1397 if (p
->state
->unit
[i
].texgen_enabled
||
1398 p
->state
->unit
[i
].texmat_enabled
) {
1400 GLuint texmat_enabled
= p
->state
->unit
[i
].texmat_enabled
;
1401 struct ureg out
= register_output(p
, VERT_RESULT_TEX0
+ i
);
1402 struct ureg out_texgen
= undef
;
1404 if (p
->state
->unit
[i
].texgen_enabled
) {
1405 GLuint copy_mask
= 0;
1406 GLuint sphere_mask
= 0;
1407 GLuint reflect_mask
= 0;
1408 GLuint normal_mask
= 0;
1412 out_texgen
= get_temp(p
);
1416 modes
[0] = p
->state
->unit
[i
].texgen_mode0
;
1417 modes
[1] = p
->state
->unit
[i
].texgen_mode1
;
1418 modes
[2] = p
->state
->unit
[i
].texgen_mode2
;
1419 modes
[3] = p
->state
->unit
[i
].texgen_mode3
;
1421 for (j
= 0; j
< 4; j
++) {
1423 case TXG_OBJ_LINEAR
: {
1424 struct ureg obj
= register_input(p
, VERT_ATTRIB_POS
);
1426 register_param3(p
, STATE_TEXGEN
, i
,
1427 STATE_TEXGEN_OBJECT_S
+ j
);
1429 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1433 case TXG_EYE_LINEAR
: {
1434 struct ureg eye
= get_eye_position(p
);
1436 register_param3(p
, STATE_TEXGEN
, i
,
1437 STATE_TEXGEN_EYE_S
+ j
);
1439 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1443 case TXG_SPHERE_MAP
:
1444 sphere_mask
|= WRITEMASK_X
<< j
;
1446 case TXG_REFLECTION_MAP
:
1447 reflect_mask
|= WRITEMASK_X
<< j
;
1449 case TXG_NORMAL_MAP
:
1450 normal_mask
|= WRITEMASK_X
<< j
;
1453 copy_mask
|= WRITEMASK_X
<< j
;
1458 build_sphere_texgen(p
, out_texgen
, sphere_mask
);
1462 build_reflect_texgen(p
, out_texgen
, reflect_mask
);
1466 struct ureg normal
= get_transformed_normal(p
);
1467 emit_op1(p
, OPCODE_MOV
, out_texgen
, normal_mask
, normal
);
1471 struct ureg in
= register_input(p
, VERT_ATTRIB_TEX0
+i
);
1472 emit_op1(p
, OPCODE_MOV
, out_texgen
, copy_mask
, in
);
1476 if (texmat_enabled
) {
1477 struct ureg texmat
[4];
1478 struct ureg in
= (!is_undef(out_texgen
) ?
1480 register_input(p
, VERT_ATTRIB_TEX0
+i
));
1481 if (p
->mvp_with_dp4
) {
1482 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1484 emit_matrix_transform_vec4( p
, out
, texmat
, in
);
1487 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1488 STATE_MATRIX_TRANSPOSE
, texmat
);
1489 emit_transpose_matrix_transform_vec4( p
, out
, texmat
, in
);
1496 emit_passthrough(p
, VERT_ATTRIB_TEX0
+i
, VERT_RESULT_TEX0
+i
);
1503 * Point size attenuation computation.
1505 static void build_atten_pointsize( struct tnl_program
*p
)
1507 struct ureg eye
= get_eye_position_z(p
);
1508 struct ureg state_size
= register_param2(p
, STATE_INTERNAL
, STATE_POINT_SIZE_CLAMPED
);
1509 struct ureg state_attenuation
= register_param1(p
, STATE_POINT_ATTENUATION
);
1510 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1511 struct ureg ut
= get_temp(p
);
1514 emit_op1(p
, OPCODE_ABS
, ut
, WRITEMASK_Y
, swizzle1(eye
, Z
));
1515 /* p1 + dist * (p2 + dist * p3); */
1516 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1517 swizzle1(state_attenuation
, Z
), swizzle1(state_attenuation
, Y
));
1518 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1519 ut
, swizzle1(state_attenuation
, X
));
1521 /* 1 / sqrt(factor) */
1522 emit_op1(p
, OPCODE_RSQ
, ut
, WRITEMASK_X
, ut
);
1525 /* out = pointSize / sqrt(factor) */
1526 emit_op2(p
, OPCODE_MUL
, out
, WRITEMASK_X
, ut
, state_size
);
1528 /* this is a good place to clamp the point size since there's likely
1529 * no hardware registers to clamp point size at rasterization time.
1531 emit_op2(p
, OPCODE_MUL
, ut
, WRITEMASK_X
, ut
, state_size
);
1532 emit_op2(p
, OPCODE_MAX
, ut
, WRITEMASK_X
, ut
, swizzle1(state_size
, Y
));
1533 emit_op2(p
, OPCODE_MIN
, out
, WRITEMASK_X
, ut
, swizzle1(state_size
, Z
));
1536 release_temp(p
, ut
);
1541 * Pass-though per-vertex point size, from user's point size array.
1543 static void build_array_pointsize( struct tnl_program
*p
)
1545 struct ureg in
= register_input(p
, VERT_ATTRIB_POINT_SIZE
);
1546 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1547 emit_op1(p
, OPCODE_MOV
, out
, WRITEMASK_X
, in
);
1551 static void build_tnl_program( struct tnl_program
*p
)
1553 /* Emit the program, starting with modelviewproject:
1557 /* Lighting calculations:
1559 if (p
->state
->fragprog_inputs_read
& (FRAG_BIT_COL0
|FRAG_BIT_COL1
)) {
1560 if (p
->state
->light_global_enabled
)
1563 if (p
->state
->fragprog_inputs_read
& FRAG_BIT_COL0
)
1564 emit_passthrough(p
, VERT_ATTRIB_COLOR0
, VERT_RESULT_COL0
);
1566 if (p
->state
->fragprog_inputs_read
& FRAG_BIT_COL1
)
1567 emit_passthrough(p
, VERT_ATTRIB_COLOR1
, VERT_RESULT_COL1
);
1571 if (p
->state
->fragprog_inputs_read
& FRAG_BIT_FOGC
)
1574 if (p
->state
->fragprog_inputs_read
& FRAG_BITS_TEX_ANY
)
1575 build_texture_transform(p
);
1577 if (p
->state
->point_attenuated
)
1578 build_atten_pointsize(p
);
1579 else if (p
->state
->point_array
)
1580 build_array_pointsize(p
);
1584 emit_op1(p
, OPCODE_END
, undef
, 0, undef
);
1595 create_new_program( const struct state_key
*key
,
1596 struct gl_vertex_program
*program
,
1597 GLboolean mvp_with_dp4
,
1600 struct tnl_program p
;
1602 memset(&p
, 0, sizeof(p
));
1604 p
.program
= program
;
1605 p
.eye_position
= undef
;
1606 p
.eye_position_z
= undef
;
1607 p
.eye_position_normalized
= undef
;
1608 p
.transformed_normal
= undef
;
1611 p
.mvp_with_dp4
= mvp_with_dp4
;
1613 if (max_temps
>= sizeof(int) * 8)
1614 p
.temp_reserved
= 0;
1616 p
.temp_reserved
= ~((1<<max_temps
)-1);
1618 /* Start by allocating 32 instructions.
1619 * If we need more, we'll grow the instruction array as needed.
1622 p
.program
->Base
.Instructions
= _mesa_alloc_instructions(p
.max_inst
);
1623 p
.program
->Base
.String
= NULL
;
1624 p
.program
->Base
.NumInstructions
=
1625 p
.program
->Base
.NumTemporaries
=
1626 p
.program
->Base
.NumParameters
=
1627 p
.program
->Base
.NumAttributes
= p
.program
->Base
.NumAddressRegs
= 0;
1628 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1629 p
.program
->Base
.InputsRead
= 0;
1630 p
.program
->Base
.OutputsWritten
= 0;
1632 build_tnl_program( &p
);
1637 * Return a vertex program which implements the current fixed-function
1638 * transform/lighting/texgen operations.
1639 * XXX move this into core mesa (main/)
1641 struct gl_vertex_program
*
1642 _mesa_get_fixed_func_vertex_program(struct gl_context
*ctx
)
1644 struct gl_vertex_program
*prog
;
1645 struct state_key key
;
1647 /* Grab all the relevent state and put it in a single structure:
1649 make_state_key(ctx
, &key
);
1651 /* Look for an already-prepared program for this state:
1653 prog
= (struct gl_vertex_program
*)
1654 _mesa_search_program_cache(ctx
->VertexProgram
.Cache
, &key
, sizeof(key
));
1657 /* OK, we'll have to build a new one */
1659 printf("Build new TNL program\n");
1661 prog
= (struct gl_vertex_program
*)
1662 ctx
->Driver
.NewProgram(ctx
, GL_VERTEX_PROGRAM_ARB
, 0);
1666 create_new_program( &key
, prog
,
1668 ctx
->Const
.VertexProgram
.MaxTemps
);
1671 if (ctx
->Driver
.ProgramStringNotify
)
1672 ctx
->Driver
.ProgramStringNotify( ctx
, GL_VERTEX_PROGRAM_ARB
,
1675 _mesa_program_cache_insert(ctx
, ctx
->VertexProgram
.Cache
,
1676 &key
, sizeof(key
), &prog
->Base
);