1 /**************************************************************************
3 * Copyright 2007 VMware, Inc.
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 VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
29 * \file ffvertex_prog.c
31 * Create a vertex program to execute the current fixed function T&L pipeline.
32 * \author Keith Whitwell
36 #include "main/glheader.h"
37 #include "main/mtypes.h"
38 #include "main/macros.h"
39 #include "main/enums.h"
40 #include "main/ffvertex_prog.h"
41 #include "program/program.h"
42 #include "program/prog_cache.h"
43 #include "program/prog_instruction.h"
44 #include "program/prog_parameter.h"
45 #include "program/prog_print.h"
46 #include "program/prog_statevars.h"
47 #include "util/bitscan.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 fog_distance_mode
:2;
65 unsigned separate_specular
:1;
66 unsigned point_attenuated
:1;
67 unsigned texture_enabled_global
:1;
68 unsigned fragprog_inputs_read
:12;
70 GLbitfield 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
;
111 #define FDM_EYE_RADIAL 0
112 #define FDM_EYE_PLANE 1
113 #define FDM_EYE_PLANE_ABS 2
115 static GLuint
translate_fog_distance_mode( GLenum mode
)
118 case GL_EYE_RADIAL_NV
:
119 return FDM_EYE_RADIAL
;
121 return FDM_EYE_PLANE
;
122 default: /* shouldn't happen; fall through to a sensible default */
123 case GL_EYE_PLANE_ABSOLUTE_NV
:
124 return FDM_EYE_PLANE_ABS
;
128 static GLboolean
check_active_shininess( struct gl_context
*ctx
,
129 const struct state_key
*key
,
132 GLuint attr
= MAT_ATTRIB_FRONT_SHININESS
+ side
;
134 if ((key
->varying_vp_inputs
& VERT_BIT_COLOR0
) &&
135 (key
->light_color_material_mask
& (1 << attr
)))
138 if (key
->varying_vp_inputs
& VERT_BIT_GENERIC(attr
))
141 if (ctx
->Light
.Material
.Attrib
[attr
][0] != 0.0F
)
148 static void make_state_key( struct gl_context
*ctx
, struct state_key
*key
)
150 const struct gl_program
*fp
= ctx
->FragmentProgram
._Current
;
153 memset(key
, 0, sizeof(struct state_key
));
155 /* This now relies on texenvprogram.c being active:
159 key
->need_eye_coords
= ctx
->_NeedEyeCoords
;
161 key
->fragprog_inputs_read
= fp
->info
.inputs_read
;
162 key
->varying_vp_inputs
= ctx
->varying_vp_inputs
;
164 if (ctx
->RenderMode
== GL_FEEDBACK
) {
165 /* make sure the vertprog emits color and tex0 */
166 key
->fragprog_inputs_read
|= (VARYING_BIT_COL0
| VARYING_BIT_TEX0
);
169 key
->separate_specular
= (ctx
->Light
.Model
.ColorControl
==
170 GL_SEPARATE_SPECULAR_COLOR
);
172 if (ctx
->Light
.Enabled
) {
173 key
->light_global_enabled
= 1;
175 if (ctx
->Light
.Model
.LocalViewer
)
176 key
->light_local_viewer
= 1;
178 if (ctx
->Light
.Model
.TwoSide
)
179 key
->light_twoside
= 1;
181 if (ctx
->Light
.ColorMaterialEnabled
) {
182 key
->light_color_material_mask
= ctx
->Light
._ColorMaterialBitmask
;
185 mask
= ctx
->Light
._EnabledLights
;
187 const int i
= u_bit_scan(&mask
);
188 struct gl_light
*light
= &ctx
->Light
.Light
[i
];
190 key
->unit
[i
].light_enabled
= 1;
192 if (light
->EyePosition
[3] == 0.0F
)
193 key
->unit
[i
].light_eyepos3_is_zero
= 1;
195 if (light
->SpotCutoff
== 180.0F
)
196 key
->unit
[i
].light_spotcutoff_is_180
= 1;
198 if (light
->ConstantAttenuation
!= 1.0F
||
199 light
->LinearAttenuation
!= 0.0F
||
200 light
->QuadraticAttenuation
!= 0.0F
)
201 key
->unit
[i
].light_attenuated
= 1;
204 if (check_active_shininess(ctx
, key
, 0)) {
205 key
->material_shininess_is_zero
= 0;
207 else if (key
->light_twoside
&&
208 check_active_shininess(ctx
, key
, 1)) {
209 key
->material_shininess_is_zero
= 0;
212 key
->material_shininess_is_zero
= 1;
216 if (ctx
->Transform
.Normalize
)
219 if (ctx
->Transform
.RescaleNormals
)
220 key
->rescale_normals
= 1;
222 if (ctx
->Fog
.FogCoordinateSource
== GL_FRAGMENT_DEPTH_EXT
) {
223 key
->fog_source_is_depth
= 1;
224 key
->fog_distance_mode
= translate_fog_distance_mode(ctx
->Fog
.FogDistanceMode
);
227 if (ctx
->Point
._Attenuated
)
228 key
->point_attenuated
= 1;
230 if (ctx
->Texture
._TexGenEnabled
||
231 ctx
->Texture
._TexMatEnabled
||
232 ctx
->Texture
._MaxEnabledTexImageUnit
!= -1)
233 key
->texture_enabled_global
= 1;
235 mask
= ctx
->Texture
._EnabledCoordUnits
| ctx
->Texture
._TexGenEnabled
236 | ctx
->Texture
._TexMatEnabled
| ctx
->Point
.CoordReplace
;
238 const int i
= u_bit_scan(&mask
);
239 struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
241 if (texUnit
->_Current
)
242 key
->unit
[i
].texunit_really_enabled
= 1;
244 if (ctx
->Point
.PointSprite
)
245 if (ctx
->Point
.CoordReplace
& (1u << i
))
246 key
->unit
[i
].coord_replace
= 1;
248 if (ctx
->Texture
._TexMatEnabled
& ENABLE_TEXMAT(i
))
249 key
->unit
[i
].texmat_enabled
= 1;
251 if (texUnit
->TexGenEnabled
) {
252 key
->unit
[i
].texgen_enabled
= 1;
254 key
->unit
[i
].texgen_mode0
=
255 translate_texgen( texUnit
->TexGenEnabled
& (1<<0),
256 texUnit
->GenS
.Mode
);
257 key
->unit
[i
].texgen_mode1
=
258 translate_texgen( texUnit
->TexGenEnabled
& (1<<1),
259 texUnit
->GenT
.Mode
);
260 key
->unit
[i
].texgen_mode2
=
261 translate_texgen( texUnit
->TexGenEnabled
& (1<<2),
262 texUnit
->GenR
.Mode
);
263 key
->unit
[i
].texgen_mode3
=
264 translate_texgen( texUnit
->TexGenEnabled
& (1<<3),
265 texUnit
->GenQ
.Mode
);
272 /* Very useful debugging tool - produces annotated listing of
273 * generated program with line/function references for each
274 * instruction back into this file:
279 /* Use uregs to represent registers internally, translate to Mesa's
280 * expected formats on emit.
282 * NOTE: These are passed by value extensively in this file rather
283 * than as usual by pointer reference. If this disturbs you, try
284 * remembering they are just 32bits in size.
286 * GCC is smart enough to deal with these dword-sized structures in
287 * much the same way as if I had defined them as dwords and was using
288 * macros to access and set the fields. This is much nicer and easier
293 GLint idx
:9; /* relative addressing may be negative */
294 /* sizeof(idx) should == sizeof(prog_src_reg::Index) */
302 const struct state_key
*state
;
303 struct gl_program
*program
;
304 GLuint max_inst
; /** number of instructions allocated for program */
305 GLboolean mvp_with_dp4
;
308 GLuint temp_reserved
;
310 struct ureg eye_position
;
311 struct ureg eye_position_z
;
312 struct ureg eye_position_normalized
;
313 struct ureg transformed_normal
;
314 struct ureg identity
;
317 GLuint color_materials
;
321 static const struct ureg undef
= {
339 static struct ureg
make_ureg(GLuint file
, GLint idx
)
345 reg
.swz
= SWIZZLE_NOOP
;
351 static struct ureg
negate( struct ureg reg
)
358 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
360 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
363 GET_SWZ(reg
.swz
, w
));
368 static struct ureg
swizzle1( struct ureg reg
, int x
)
370 return swizzle(reg
, x
, x
, x
, x
);
374 static struct ureg
get_temp( struct tnl_program
*p
)
376 int bit
= ffs( ~p
->temp_in_use
);
378 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
382 if ((GLuint
) bit
> p
->program
->arb
.NumTemporaries
)
383 p
->program
->arb
.NumTemporaries
= bit
;
385 p
->temp_in_use
|= 1<<(bit
-1);
386 return make_ureg(PROGRAM_TEMPORARY
, bit
-1);
390 static struct ureg
reserve_temp( struct tnl_program
*p
)
392 struct ureg temp
= get_temp( p
);
393 p
->temp_reserved
|= 1<<temp
.idx
;
398 static void release_temp( struct tnl_program
*p
, struct ureg reg
)
400 if (reg
.file
== PROGRAM_TEMPORARY
) {
401 p
->temp_in_use
&= ~(1<<reg
.idx
);
402 p
->temp_in_use
|= p
->temp_reserved
; /* can't release reserved temps */
406 static void release_temps( struct tnl_program
*p
)
408 p
->temp_in_use
= p
->temp_reserved
;
412 static struct ureg
register_param5(struct tnl_program
*p
,
419 gl_state_index tokens
[STATE_LENGTH
];
426 idx
= _mesa_add_state_reference(p
->program
->Parameters
, tokens
);
427 return make_ureg(PROGRAM_STATE_VAR
, idx
);
431 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
432 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
433 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
434 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
439 * \param input one of VERT_ATTRIB_x tokens.
441 static struct ureg
register_input( struct tnl_program
*p
, GLuint input
)
443 assert(input
< VERT_ATTRIB_MAX
);
445 if (p
->state
->varying_vp_inputs
& VERT_BIT(input
)) {
446 p
->program
->info
.inputs_read
|= VERT_BIT(input
);
447 return make_ureg(PROGRAM_INPUT
, input
);
450 return register_param3( p
, STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, input
);
456 * \param input one of VARYING_SLOT_x tokens.
458 static struct ureg
register_output( struct tnl_program
*p
, GLuint output
)
460 p
->program
->info
.outputs_written
|= BITFIELD64_BIT(output
);
461 return make_ureg(PROGRAM_OUTPUT
, output
);
465 static struct ureg
register_const4f( struct tnl_program
*p
,
471 gl_constant_value values
[4];
478 idx
= _mesa_add_unnamed_constant(p
->program
->Parameters
, values
, 4,
480 assert(swizzle
== SWIZZLE_NOOP
);
481 return make_ureg(PROGRAM_CONSTANT
, idx
);
484 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
485 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
486 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
487 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
489 static GLboolean
is_undef( struct ureg reg
)
491 return reg
.file
== PROGRAM_UNDEFINED
;
495 static struct ureg
get_identity_param( struct tnl_program
*p
)
497 if (is_undef(p
->identity
))
498 p
->identity
= register_const4f(p
, 0,0,0,1);
503 static void register_matrix_param5( struct tnl_program
*p
,
504 GLint s0
, /* modelview, projection, etc */
505 GLint s1
, /* texture matrix number */
506 GLint s2
, /* first row */
507 GLint s3
, /* last row */
508 GLint s4
, /* inverse, transpose, etc */
509 struct ureg
*matrix
)
513 /* This is a bit sad as the support is there to pull the whole
514 * matrix out in one go:
516 for (i
= 0; i
<= s3
- s2
; i
++)
517 matrix
[i
] = register_param5( p
, s0
, s1
, i
, i
, s4
);
521 static void emit_arg( struct prog_src_register
*src
,
524 src
->File
= reg
.file
;
525 src
->Index
= reg
.idx
;
526 src
->Swizzle
= reg
.swz
;
527 src
->Negate
= reg
.negate
? NEGATE_XYZW
: NEGATE_NONE
;
529 /* Check that bitfield sizes aren't exceeded */
530 assert(src
->Index
== reg
.idx
);
534 static void emit_dst( struct prog_dst_register
*dst
,
535 struct ureg reg
, GLuint mask
)
537 dst
->File
= reg
.file
;
538 dst
->Index
= reg
.idx
;
539 /* allow zero as a shorthand for xyzw */
540 dst
->WriteMask
= mask
? mask
: WRITEMASK_XYZW
;
541 /* Check that bitfield sizes aren't exceeded */
542 assert(dst
->Index
== reg
.idx
);
546 static void debug_insn( struct prog_instruction
*inst
, const char *fn
,
550 static const char *last_fn
;
557 printf("%d:\t", line
);
558 _mesa_print_instruction(inst
);
563 static void emit_op3fn(struct tnl_program
*p
,
574 struct prog_instruction
*inst
;
576 assert(p
->program
->arb
.NumInstructions
<= p
->max_inst
);
578 if (p
->program
->arb
.NumInstructions
== p
->max_inst
) {
579 /* need to extend the program's instruction array */
580 struct prog_instruction
*newInst
;
582 /* double the size */
586 rzalloc_array(p
->program
, struct prog_instruction
, p
->max_inst
);
588 _mesa_error(NULL
, GL_OUT_OF_MEMORY
, "vertex program build");
592 _mesa_copy_instructions(newInst
, p
->program
->arb
.Instructions
,
593 p
->program
->arb
.NumInstructions
);
595 ralloc_free(p
->program
->arb
.Instructions
);
597 p
->program
->arb
.Instructions
= newInst
;
600 nr
= p
->program
->arb
.NumInstructions
++;
602 inst
= &p
->program
->arb
.Instructions
[nr
];
603 inst
->Opcode
= (enum prog_opcode
) op
;
605 emit_arg( &inst
->SrcReg
[0], src0
);
606 emit_arg( &inst
->SrcReg
[1], src1
);
607 emit_arg( &inst
->SrcReg
[2], src2
);
609 emit_dst( &inst
->DstReg
, dest
, mask
);
611 debug_insn(inst
, fn
, line
);
615 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
616 emit_op3fn(p, op, dst, mask, src0, src1, src2, __func__, __LINE__)
618 #define emit_op2(p, op, dst, mask, src0, src1) \
619 emit_op3fn(p, op, dst, mask, src0, src1, undef, __func__, __LINE__)
621 #define emit_op1(p, op, dst, mask, src0) \
622 emit_op3fn(p, op, dst, mask, src0, undef, undef, __func__, __LINE__)
625 static struct ureg
make_temp( struct tnl_program
*p
, struct ureg reg
)
627 if (reg
.file
== PROGRAM_TEMPORARY
&&
628 !(p
->temp_reserved
& (1<<reg
.idx
)))
631 struct ureg temp
= get_temp(p
);
632 emit_op1(p
, OPCODE_MOV
, temp
, 0, reg
);
638 /* Currently no tracking performed of input/output/register size or
639 * active elements. Could be used to reduce these operations, as
640 * could the matrix type.
642 static void emit_matrix_transform_vec4( struct tnl_program
*p
,
644 const struct ureg
*mat
,
647 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_X
, src
, mat
[0]);
648 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Y
, src
, mat
[1]);
649 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Z
, src
, mat
[2]);
650 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_W
, src
, mat
[3]);
654 /* This version is much easier to implement if writemasks are not
655 * supported natively on the target or (like SSE), the target doesn't
656 * have a clean/obvious dotproduct implementation.
658 static void emit_transpose_matrix_transform_vec4( struct tnl_program
*p
,
660 const struct ureg
*mat
,
665 if (dest
.file
!= PROGRAM_TEMPORARY
)
670 emit_op2(p
, OPCODE_MUL
, tmp
, 0, swizzle1(src
,X
), mat
[0]);
671 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Y
), mat
[1], tmp
);
672 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Z
), mat
[2], tmp
);
673 emit_op3(p
, OPCODE_MAD
, dest
, 0, swizzle1(src
,W
), mat
[3], tmp
);
675 if (dest
.file
!= PROGRAM_TEMPORARY
)
676 release_temp(p
, tmp
);
680 static void emit_matrix_transform_vec3( struct tnl_program
*p
,
682 const struct ureg
*mat
,
685 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_X
, src
, mat
[0]);
686 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Y
, src
, mat
[1]);
687 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Z
, src
, mat
[2]);
691 static void emit_normalize_vec3( struct tnl_program
*p
,
695 struct ureg tmp
= get_temp(p
);
696 emit_op2(p
, OPCODE_DP3
, tmp
, WRITEMASK_X
, src
, src
);
697 emit_op1(p
, OPCODE_RSQ
, tmp
, WRITEMASK_X
, tmp
);
698 emit_op2(p
, OPCODE_MUL
, dest
, 0, src
, swizzle1(tmp
, X
));
699 release_temp(p
, tmp
);
703 static void emit_passthrough( struct tnl_program
*p
,
707 struct ureg out
= register_output(p
, output
);
708 emit_op1(p
, OPCODE_MOV
, out
, 0, register_input(p
, input
));
712 static struct ureg
get_eye_position( struct tnl_program
*p
)
714 if (is_undef(p
->eye_position
)) {
715 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
716 struct ureg modelview
[4];
718 p
->eye_position
= reserve_temp(p
);
720 if (p
->mvp_with_dp4
) {
721 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
724 emit_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
727 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
728 STATE_MATRIX_TRANSPOSE
, modelview
);
730 emit_transpose_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
734 return p
->eye_position
;
738 static struct ureg
get_eye_position_z( struct tnl_program
*p
)
740 if (!is_undef(p
->eye_position
))
741 return swizzle1(p
->eye_position
, Z
);
743 if (is_undef(p
->eye_position_z
)) {
744 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
745 struct ureg modelview
[4];
747 p
->eye_position_z
= reserve_temp(p
);
749 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
752 emit_op2(p
, OPCODE_DP4
, p
->eye_position_z
, 0, pos
, modelview
[2]);
755 return p
->eye_position_z
;
759 static struct ureg
get_eye_position_normalized( struct tnl_program
*p
)
761 if (is_undef(p
->eye_position_normalized
)) {
762 struct ureg eye
= get_eye_position(p
);
763 p
->eye_position_normalized
= reserve_temp(p
);
764 emit_normalize_vec3(p
, p
->eye_position_normalized
, eye
);
767 return p
->eye_position_normalized
;
771 static struct ureg
get_transformed_normal( struct tnl_program
*p
)
773 if (is_undef(p
->transformed_normal
) &&
774 !p
->state
->need_eye_coords
&&
775 !p
->state
->normalize
&&
776 !(p
->state
->need_eye_coords
== p
->state
->rescale_normals
))
778 p
->transformed_normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
780 else if (is_undef(p
->transformed_normal
))
782 struct ureg normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
783 struct ureg mvinv
[3];
784 struct ureg transformed_normal
= reserve_temp(p
);
786 if (p
->state
->need_eye_coords
) {
787 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 2,
788 STATE_MATRIX_INVTRANS
, mvinv
);
790 /* Transform to eye space:
792 emit_matrix_transform_vec3( p
, transformed_normal
, mvinv
, normal
);
793 normal
= transformed_normal
;
796 /* Normalize/Rescale:
798 if (p
->state
->normalize
) {
799 emit_normalize_vec3( p
, transformed_normal
, normal
);
800 normal
= transformed_normal
;
802 else if (p
->state
->need_eye_coords
== p
->state
->rescale_normals
) {
803 /* This is already adjusted for eye/non-eye rendering:
805 struct ureg rescale
= register_param2(p
, STATE_INTERNAL
,
808 emit_op2( p
, OPCODE_MUL
, transformed_normal
, 0, normal
, rescale
);
809 normal
= transformed_normal
;
812 assert(normal
.file
== PROGRAM_TEMPORARY
);
813 p
->transformed_normal
= normal
;
816 return p
->transformed_normal
;
820 static void build_hpos( struct tnl_program
*p
)
822 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
823 struct ureg hpos
= register_output( p
, VARYING_SLOT_POS
);
826 if (p
->mvp_with_dp4
) {
827 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
829 emit_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
832 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
833 STATE_MATRIX_TRANSPOSE
, mvp
);
834 emit_transpose_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
839 static GLuint
material_attrib( GLuint side
, GLuint property
)
841 return (property
- STATE_AMBIENT
) * 2 + side
;
846 * Get a bitmask of which material values vary on a per-vertex basis.
848 static void set_material_flags( struct tnl_program
*p
)
850 p
->color_materials
= 0;
853 if (p
->state
->varying_vp_inputs
& VERT_BIT_COLOR0
) {
855 p
->color_materials
= p
->state
->light_color_material_mask
;
858 p
->materials
|= (p
->state
->varying_vp_inputs
>> VERT_ATTRIB_GENERIC0
);
862 static struct ureg
get_material( struct tnl_program
*p
, GLuint side
,
865 GLuint attrib
= material_attrib(side
, property
);
867 if (p
->color_materials
& (1<<attrib
))
868 return register_input(p
, VERT_ATTRIB_COLOR0
);
869 else if (p
->materials
& (1<<attrib
)) {
870 /* Put material values in the GENERIC slots -- they are not used
871 * for anything in fixed function mode.
873 return register_input( p
, attrib
+ VERT_ATTRIB_GENERIC0
);
876 return register_param3( p
, STATE_MATERIAL
, side
, property
);
879 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
880 MAT_BIT_FRONT_AMBIENT | \
881 MAT_BIT_FRONT_DIFFUSE) << (side))
885 * Either return a precalculated constant value or emit code to
886 * calculate these values dynamically in the case where material calls
887 * are present between begin/end pairs.
889 * Probably want to shift this to the program compilation phase - if
890 * we always emitted the calculation here, a smart compiler could
891 * detect that it was constant (given a certain set of inputs), and
892 * lift it out of the main loop. That way the programs created here
893 * would be independent of the vertex_buffer details.
895 static struct ureg
get_scenecolor( struct tnl_program
*p
, GLuint side
)
897 if (p
->materials
& SCENE_COLOR_BITS(side
)) {
898 struct ureg lm_ambient
= register_param1(p
, STATE_LIGHTMODEL_AMBIENT
);
899 struct ureg material_emission
= get_material(p
, side
, STATE_EMISSION
);
900 struct ureg material_ambient
= get_material(p
, side
, STATE_AMBIENT
);
901 struct ureg material_diffuse
= get_material(p
, side
, STATE_DIFFUSE
);
902 struct ureg tmp
= make_temp(p
, material_diffuse
);
903 emit_op3(p
, OPCODE_MAD
, tmp
, WRITEMASK_XYZ
, lm_ambient
,
904 material_ambient
, material_emission
);
908 return register_param2( p
, STATE_LIGHTMODEL_SCENECOLOR
, side
);
912 static struct ureg
get_lightprod( struct tnl_program
*p
, GLuint light
,
913 GLuint side
, GLuint property
)
915 GLuint attrib
= material_attrib(side
, property
);
916 if (p
->materials
& (1<<attrib
)) {
917 struct ureg light_value
=
918 register_param3(p
, STATE_LIGHT
, light
, property
);
919 struct ureg material_value
= get_material(p
, side
, property
);
920 struct ureg tmp
= get_temp(p
);
921 emit_op2(p
, OPCODE_MUL
, tmp
, 0, light_value
, material_value
);
925 return register_param4(p
, STATE_LIGHTPROD
, light
, side
, property
);
929 static struct ureg
calculate_light_attenuation( struct tnl_program
*p
,
934 struct ureg attenuation
= register_param3(p
, STATE_LIGHT
, i
,
936 struct ureg att
= undef
;
938 /* Calculate spot attenuation:
940 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
941 struct ureg spot_dir_norm
= register_param3(p
, STATE_INTERNAL
,
942 STATE_LIGHT_SPOT_DIR_NORMALIZED
, i
);
943 struct ureg spot
= get_temp(p
);
944 struct ureg slt
= get_temp(p
);
948 emit_op2(p
, OPCODE_DP3
, spot
, 0, negate(VPpli
), spot_dir_norm
);
949 emit_op2(p
, OPCODE_SLT
, slt
, 0, swizzle1(spot_dir_norm
,W
), spot
);
950 emit_op1(p
, OPCODE_ABS
, spot
, 0, spot
);
951 emit_op2(p
, OPCODE_POW
, spot
, 0, spot
, swizzle1(attenuation
, W
));
952 emit_op2(p
, OPCODE_MUL
, att
, 0, slt
, spot
);
954 release_temp(p
, spot
);
955 release_temp(p
, slt
);
958 /* Calculate distance attenuation(See formula (2.4) at glspec 2.1 page 62):
960 * Skip the calucation when _dist_ is undefined(light_eyepos3_is_zero)
962 if (p
->state
->unit
[i
].light_attenuated
&& !is_undef(dist
)) {
966 emit_op1(p
, OPCODE_RCP
, dist
, WRITEMASK_YZ
, dist
);
968 emit_op2(p
, OPCODE_MUL
, dist
, WRITEMASK_XZ
, dist
, swizzle1(dist
,Y
));
970 emit_op2(p
, OPCODE_DP3
, dist
, 0, attenuation
, dist
);
972 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
974 emit_op1(p
, OPCODE_RCP
, dist
, 0, dist
);
975 /* spot-atten * dist-atten */
976 emit_op2(p
, OPCODE_MUL
, att
, 0, dist
, att
);
980 emit_op1(p
, OPCODE_RCP
, att
, 0, dist
);
990 * lit.y = MAX(0, dots.x)
991 * lit.z = SLT(0, dots.x)
993 static void emit_degenerate_lit( struct tnl_program
*p
,
997 struct ureg id
= get_identity_param(p
); /* id = {0,0,0,1} */
999 /* Note that lit.x & lit.w will not be examined. Note also that
1000 * dots.xyzw == dots.xxxx.
1003 /* MAX lit, id, dots;
1005 emit_op2(p
, OPCODE_MAX
, lit
, WRITEMASK_XYZW
, id
, dots
);
1007 /* result[2] = (in > 0 ? 1 : 0)
1008 * SLT lit.z, id.z, dots; # lit.z = (0 < dots.z) ? 1 : 0
1010 emit_op2(p
, OPCODE_SLT
, lit
, WRITEMASK_Z
, swizzle1(id
,Z
), dots
);
1014 /* Need to add some addtional parameters to allow lighting in object
1015 * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
1018 static void build_lighting( struct tnl_program
*p
)
1020 const GLboolean twoside
= p
->state
->light_twoside
;
1021 const GLboolean separate
= p
->state
->separate_specular
;
1022 GLuint nr_lights
= 0, count
= 0;
1023 struct ureg normal
= get_transformed_normal(p
);
1024 struct ureg lit
= get_temp(p
);
1025 struct ureg dots
= get_temp(p
);
1026 struct ureg _col0
= undef
, _col1
= undef
;
1027 struct ureg _bfc0
= undef
, _bfc1
= undef
;
1032 * dots.x = dot(normal, VPpli)
1033 * dots.y = dot(normal, halfAngle)
1034 * dots.z = back.shininess
1035 * dots.w = front.shininess
1038 for (i
= 0; i
< MAX_LIGHTS
; i
++)
1039 if (p
->state
->unit
[i
].light_enabled
)
1042 set_material_flags(p
);
1045 if (!p
->state
->material_shininess_is_zero
) {
1046 struct ureg shininess
= get_material(p
, 0, STATE_SHININESS
);
1047 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_W
, swizzle1(shininess
,X
));
1048 release_temp(p
, shininess
);
1051 _col0
= make_temp(p
, get_scenecolor(p
, 0));
1053 _col1
= make_temp(p
, get_identity_param(p
));
1059 if (!p
->state
->material_shininess_is_zero
) {
1060 /* Note that we negate the back-face specular exponent here.
1061 * The negation will be un-done later in the back-face code below.
1063 struct ureg shininess
= get_material(p
, 1, STATE_SHININESS
);
1064 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_Z
,
1065 negate(swizzle1(shininess
,X
)));
1066 release_temp(p
, shininess
);
1069 _bfc0
= make_temp(p
, get_scenecolor(p
, 1));
1071 _bfc1
= make_temp(p
, get_identity_param(p
));
1076 /* If no lights, still need to emit the scenecolor.
1079 struct ureg res0
= register_output( p
, VARYING_SLOT_COL0
);
1080 emit_op1(p
, OPCODE_MOV
, res0
, 0, _col0
);
1084 struct ureg res1
= register_output( p
, VARYING_SLOT_COL1
);
1085 emit_op1(p
, OPCODE_MOV
, res1
, 0, _col1
);
1089 struct ureg res0
= register_output( p
, VARYING_SLOT_BFC0
);
1090 emit_op1(p
, OPCODE_MOV
, res0
, 0, _bfc0
);
1093 if (twoside
&& separate
) {
1094 struct ureg res1
= register_output( p
, VARYING_SLOT_BFC1
);
1095 emit_op1(p
, OPCODE_MOV
, res1
, 0, _bfc1
);
1098 if (nr_lights
== 0) {
1103 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
1104 if (p
->state
->unit
[i
].light_enabled
) {
1105 struct ureg half
= undef
;
1106 struct ureg att
= undef
, VPpli
= undef
;
1107 struct ureg dist
= undef
;
1110 if (p
->state
->unit
[i
].light_eyepos3_is_zero
) {
1111 VPpli
= register_param3(p
, STATE_INTERNAL
,
1112 STATE_LIGHT_POSITION_NORMALIZED
, i
);
1114 struct ureg Ppli
= register_param3(p
, STATE_INTERNAL
,
1115 STATE_LIGHT_POSITION
, i
);
1116 struct ureg V
= get_eye_position(p
);
1118 VPpli
= get_temp(p
);
1121 /* Calculate VPpli vector
1123 emit_op2(p
, OPCODE_SUB
, VPpli
, 0, Ppli
, V
);
1125 /* Normalize VPpli. The dist value also used in
1126 * attenuation below.
1128 emit_op2(p
, OPCODE_DP3
, dist
, 0, VPpli
, VPpli
);
1129 emit_op1(p
, OPCODE_RSQ
, dist
, 0, dist
);
1130 emit_op2(p
, OPCODE_MUL
, VPpli
, 0, VPpli
, dist
);
1133 /* Calculate attenuation:
1135 att
= calculate_light_attenuation(p
, i
, VPpli
, dist
);
1136 release_temp(p
, dist
);
1138 /* Calculate viewer direction, or use infinite viewer:
1140 if (!p
->state
->material_shininess_is_zero
) {
1141 if (p
->state
->light_local_viewer
) {
1142 struct ureg eye_hat
= get_eye_position_normalized(p
);
1144 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1145 emit_normalize_vec3(p
, half
, half
);
1146 } else if (p
->state
->unit
[i
].light_eyepos3_is_zero
) {
1147 half
= register_param3(p
, STATE_INTERNAL
,
1148 STATE_LIGHT_HALF_VECTOR
, i
);
1150 struct ureg z_dir
= swizzle(get_identity_param(p
),X
,Y
,W
,Z
);
1152 emit_op2(p
, OPCODE_ADD
, half
, 0, VPpli
, z_dir
);
1153 emit_normalize_vec3(p
, half
, half
);
1157 /* Calculate dot products:
1159 if (p
->state
->material_shininess_is_zero
) {
1160 emit_op2(p
, OPCODE_DP3
, dots
, 0, normal
, VPpli
);
1163 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_X
, normal
, VPpli
);
1164 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_Y
, normal
, half
);
1167 /* Front face lighting:
1170 struct ureg ambient
= get_lightprod(p
, i
, 0, STATE_AMBIENT
);
1171 struct ureg diffuse
= get_lightprod(p
, i
, 0, STATE_DIFFUSE
);
1172 struct ureg specular
= get_lightprod(p
, i
, 0, STATE_SPECULAR
);
1173 struct ureg res0
, res1
;
1174 GLuint mask0
, mask1
;
1176 if (count
== nr_lights
) {
1178 mask0
= WRITEMASK_XYZ
;
1179 mask1
= WRITEMASK_XYZ
;
1180 res0
= register_output( p
, VARYING_SLOT_COL0
);
1181 res1
= register_output( p
, VARYING_SLOT_COL1
);
1185 mask1
= WRITEMASK_XYZ
;
1187 res1
= register_output( p
, VARYING_SLOT_COL0
);
1197 if (!is_undef(att
)) {
1198 /* light is attenuated by distance */
1199 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1200 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1201 emit_op3(p
, OPCODE_MAD
, _col0
, 0, swizzle1(lit
,X
), ambient
, _col0
);
1203 else if (!p
->state
->material_shininess_is_zero
) {
1204 /* there's a non-zero specular term */
1205 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1206 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1209 /* no attenutation, no specular */
1210 emit_degenerate_lit(p
, lit
, dots
);
1211 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1214 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _col0
);
1215 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _col1
);
1217 release_temp(p
, ambient
);
1218 release_temp(p
, diffuse
);
1219 release_temp(p
, specular
);
1222 /* Back face lighting:
1225 struct ureg ambient
= get_lightprod(p
, i
, 1, STATE_AMBIENT
);
1226 struct ureg diffuse
= get_lightprod(p
, i
, 1, STATE_DIFFUSE
);
1227 struct ureg specular
= get_lightprod(p
, i
, 1, STATE_SPECULAR
);
1228 struct ureg res0
, res1
;
1229 GLuint mask0
, mask1
;
1231 if (count
== nr_lights
) {
1233 mask0
= WRITEMASK_XYZ
;
1234 mask1
= WRITEMASK_XYZ
;
1235 res0
= register_output( p
, VARYING_SLOT_BFC0
);
1236 res1
= register_output( p
, VARYING_SLOT_BFC1
);
1240 mask1
= WRITEMASK_XYZ
;
1242 res1
= register_output( p
, VARYING_SLOT_BFC0
);
1252 /* For the back face we need to negate the X and Y component
1253 * dot products. dots.Z has the negated back-face specular
1254 * exponent. We swizzle that into the W position. This
1255 * negation makes the back-face specular term positive again.
1257 dots
= negate(swizzle(dots
,X
,Y
,W
,Z
));
1259 if (!is_undef(att
)) {
1260 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1261 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1262 emit_op3(p
, OPCODE_MAD
, _bfc0
, 0, swizzle1(lit
,X
), ambient
, _bfc0
);
1264 else if (!p
->state
->material_shininess_is_zero
) {
1265 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1266 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
); /**/
1269 emit_degenerate_lit(p
, lit
, dots
);
1270 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
);
1273 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _bfc0
);
1274 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _bfc1
);
1275 /* restore dots to its original state for subsequent lights
1276 * by negating and swizzling again.
1278 dots
= negate(swizzle(dots
,X
,Y
,W
,Z
));
1280 release_temp(p
, ambient
);
1281 release_temp(p
, diffuse
);
1282 release_temp(p
, specular
);
1285 release_temp(p
, half
);
1286 release_temp(p
, VPpli
);
1287 release_temp(p
, att
);
1295 static void build_fog( struct tnl_program
*p
)
1297 struct ureg fog
= register_output(p
, VARYING_SLOT_FOGC
);
1300 if (p
->state
->fog_source_is_depth
) {
1302 switch (p
->state
->fog_distance_mode
) {
1303 case FDM_EYE_RADIAL
: /* Z = sqrt(Xe*Xe + Ye*Ye + Ze*Ze) */
1304 input
= get_eye_position(p
);
1305 emit_op2(p
, OPCODE_DP3
, fog
, WRITEMASK_X
, input
, input
);
1306 emit_op1(p
, OPCODE_RSQ
, fog
, WRITEMASK_X
, fog
);
1307 emit_op1(p
, OPCODE_RCP
, fog
, WRITEMASK_X
, fog
);
1309 case FDM_EYE_PLANE
: /* Z = Ze */
1310 input
= get_eye_position_z(p
);
1311 emit_op1(p
, OPCODE_MOV
, fog
, WRITEMASK_X
, input
);
1313 case FDM_EYE_PLANE_ABS
: /* Z = abs(Ze) */
1314 input
= get_eye_position_z(p
);
1315 emit_op1(p
, OPCODE_ABS
, fog
, WRITEMASK_X
, input
);
1318 assert(!"Bad fog mode in build_fog()");
1324 input
= swizzle1(register_input(p
, VERT_ATTRIB_FOG
), X
);
1325 emit_op1(p
, OPCODE_ABS
, fog
, WRITEMASK_X
, input
);
1328 emit_op1(p
, OPCODE_MOV
, fog
, WRITEMASK_YZW
, get_identity_param(p
));
1332 static void build_reflect_texgen( struct tnl_program
*p
,
1336 struct ureg normal
= get_transformed_normal(p
);
1337 struct ureg eye_hat
= get_eye_position_normalized(p
);
1338 struct ureg tmp
= get_temp(p
);
1341 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1343 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1345 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, negate(tmp
), normal
, eye_hat
);
1347 release_temp(p
, tmp
);
1351 static void build_sphere_texgen( struct tnl_program
*p
,
1355 struct ureg normal
= get_transformed_normal(p
);
1356 struct ureg eye_hat
= get_eye_position_normalized(p
);
1357 struct ureg tmp
= get_temp(p
);
1358 struct ureg half
= register_scalar_const(p
, .5);
1359 struct ureg r
= get_temp(p
);
1360 struct ureg inv_m
= get_temp(p
);
1361 struct ureg id
= get_identity_param(p
);
1363 /* Could share the above calculations, but it would be
1364 * a fairly odd state for someone to set (both sphere and
1365 * reflection active for different texture coordinate
1366 * components. Of course - if two texture units enable
1367 * reflect and/or sphere, things start to tilt in favour
1368 * of seperating this out:
1372 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1374 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1376 emit_op3(p
, OPCODE_MAD
, r
, 0, negate(tmp
), normal
, eye_hat
);
1378 emit_op2(p
, OPCODE_ADD
, tmp
, 0, r
, swizzle(id
,X
,Y
,W
,Z
));
1379 /* rx^2 + ry^2 + (rz+1)^2 */
1380 emit_op2(p
, OPCODE_DP3
, tmp
, 0, tmp
, tmp
);
1382 emit_op1(p
, OPCODE_RSQ
, tmp
, 0, tmp
);
1384 emit_op2(p
, OPCODE_MUL
, inv_m
, 0, tmp
, half
);
1386 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, r
, inv_m
, half
);
1388 release_temp(p
, tmp
);
1390 release_temp(p
, inv_m
);
1394 static void build_texture_transform( struct tnl_program
*p
)
1398 for (i
= 0; i
< MAX_TEXTURE_COORD_UNITS
; i
++) {
1400 if (!(p
->state
->fragprog_inputs_read
& VARYING_BIT_TEX(i
)))
1403 if (p
->state
->unit
[i
].coord_replace
)
1406 if (p
->state
->unit
[i
].texgen_enabled
||
1407 p
->state
->unit
[i
].texmat_enabled
) {
1409 GLuint texmat_enabled
= p
->state
->unit
[i
].texmat_enabled
;
1410 struct ureg out
= register_output(p
, VARYING_SLOT_TEX0
+ i
);
1411 struct ureg out_texgen
= undef
;
1413 if (p
->state
->unit
[i
].texgen_enabled
) {
1414 GLuint copy_mask
= 0;
1415 GLuint sphere_mask
= 0;
1416 GLuint reflect_mask
= 0;
1417 GLuint normal_mask
= 0;
1421 out_texgen
= get_temp(p
);
1425 modes
[0] = p
->state
->unit
[i
].texgen_mode0
;
1426 modes
[1] = p
->state
->unit
[i
].texgen_mode1
;
1427 modes
[2] = p
->state
->unit
[i
].texgen_mode2
;
1428 modes
[3] = p
->state
->unit
[i
].texgen_mode3
;
1430 for (j
= 0; j
< 4; j
++) {
1432 case TXG_OBJ_LINEAR
: {
1433 struct ureg obj
= register_input(p
, VERT_ATTRIB_POS
);
1435 register_param3(p
, STATE_TEXGEN
, i
,
1436 STATE_TEXGEN_OBJECT_S
+ j
);
1438 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1442 case TXG_EYE_LINEAR
: {
1443 struct ureg eye
= get_eye_position(p
);
1445 register_param3(p
, STATE_TEXGEN
, i
,
1446 STATE_TEXGEN_EYE_S
+ j
);
1448 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1452 case TXG_SPHERE_MAP
:
1453 sphere_mask
|= WRITEMASK_X
<< j
;
1455 case TXG_REFLECTION_MAP
:
1456 reflect_mask
|= WRITEMASK_X
<< j
;
1458 case TXG_NORMAL_MAP
:
1459 normal_mask
|= WRITEMASK_X
<< j
;
1462 copy_mask
|= WRITEMASK_X
<< j
;
1467 build_sphere_texgen(p
, out_texgen
, sphere_mask
);
1471 build_reflect_texgen(p
, out_texgen
, reflect_mask
);
1475 struct ureg normal
= get_transformed_normal(p
);
1476 emit_op1(p
, OPCODE_MOV
, out_texgen
, normal_mask
, normal
);
1480 struct ureg in
= register_input(p
, VERT_ATTRIB_TEX0
+i
);
1481 emit_op1(p
, OPCODE_MOV
, out_texgen
, copy_mask
, in
);
1485 if (texmat_enabled
) {
1486 struct ureg texmat
[4];
1487 struct ureg in
= (!is_undef(out_texgen
) ?
1489 register_input(p
, VERT_ATTRIB_TEX0
+i
));
1490 if (p
->mvp_with_dp4
) {
1491 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1493 emit_matrix_transform_vec4( p
, out
, texmat
, in
);
1496 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1497 STATE_MATRIX_TRANSPOSE
, texmat
);
1498 emit_transpose_matrix_transform_vec4( p
, out
, texmat
, in
);
1505 emit_passthrough(p
, VERT_ATTRIB_TEX0
+i
, VARYING_SLOT_TEX0
+i
);
1512 * Point size attenuation computation.
1514 static void build_atten_pointsize( struct tnl_program
*p
)
1516 struct ureg eye
= get_eye_position_z(p
);
1517 struct ureg state_size
= register_param2(p
, STATE_INTERNAL
, STATE_POINT_SIZE_CLAMPED
);
1518 struct ureg state_attenuation
= register_param1(p
, STATE_POINT_ATTENUATION
);
1519 struct ureg out
= register_output(p
, VARYING_SLOT_PSIZ
);
1520 struct ureg ut
= get_temp(p
);
1523 emit_op1(p
, OPCODE_ABS
, ut
, WRITEMASK_Y
, swizzle1(eye
, Z
));
1524 /* p1 + dist * (p2 + dist * p3); */
1525 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1526 swizzle1(state_attenuation
, Z
), swizzle1(state_attenuation
, Y
));
1527 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1528 ut
, swizzle1(state_attenuation
, X
));
1530 /* 1 / sqrt(factor) */
1531 emit_op1(p
, OPCODE_RSQ
, ut
, WRITEMASK_X
, ut
);
1534 /* out = pointSize / sqrt(factor) */
1535 emit_op2(p
, OPCODE_MUL
, out
, WRITEMASK_X
, ut
, state_size
);
1537 /* this is a good place to clamp the point size since there's likely
1538 * no hardware registers to clamp point size at rasterization time.
1540 emit_op2(p
, OPCODE_MUL
, ut
, WRITEMASK_X
, ut
, state_size
);
1541 emit_op2(p
, OPCODE_MAX
, ut
, WRITEMASK_X
, ut
, swizzle1(state_size
, Y
));
1542 emit_op2(p
, OPCODE_MIN
, out
, WRITEMASK_X
, ut
, swizzle1(state_size
, Z
));
1545 release_temp(p
, ut
);
1550 * Pass-though per-vertex point size, from user's point size array.
1552 static void build_array_pointsize( struct tnl_program
*p
)
1554 struct ureg in
= register_input(p
, VERT_ATTRIB_POINT_SIZE
);
1555 struct ureg out
= register_output(p
, VARYING_SLOT_PSIZ
);
1556 emit_op1(p
, OPCODE_MOV
, out
, WRITEMASK_X
, in
);
1560 static void build_tnl_program( struct tnl_program
*p
)
1562 /* Emit the program, starting with the modelview, projection transforms:
1566 /* Lighting calculations:
1568 if (p
->state
->fragprog_inputs_read
& (VARYING_BIT_COL0
|VARYING_BIT_COL1
)) {
1569 if (p
->state
->light_global_enabled
)
1572 if (p
->state
->fragprog_inputs_read
& VARYING_BIT_COL0
)
1573 emit_passthrough(p
, VERT_ATTRIB_COLOR0
, VARYING_SLOT_COL0
);
1575 if (p
->state
->fragprog_inputs_read
& VARYING_BIT_COL1
)
1576 emit_passthrough(p
, VERT_ATTRIB_COLOR1
, VARYING_SLOT_COL1
);
1580 if (p
->state
->fragprog_inputs_read
& VARYING_BIT_FOGC
)
1583 if (p
->state
->fragprog_inputs_read
& VARYING_BITS_TEX_ANY
)
1584 build_texture_transform(p
);
1586 if (p
->state
->point_attenuated
)
1587 build_atten_pointsize(p
);
1588 else if (p
->state
->varying_vp_inputs
& VERT_BIT_POINT_SIZE
)
1589 build_array_pointsize(p
);
1593 emit_op1(p
, OPCODE_END
, undef
, 0, undef
);
1604 create_new_program( const struct state_key
*key
,
1605 struct gl_program
*program
,
1606 GLboolean mvp_with_dp4
,
1609 struct tnl_program p
;
1611 memset(&p
, 0, sizeof(p
));
1613 p
.program
= program
;
1614 p
.eye_position
= undef
;
1615 p
.eye_position_z
= undef
;
1616 p
.eye_position_normalized
= undef
;
1617 p
.transformed_normal
= undef
;
1620 p
.mvp_with_dp4
= mvp_with_dp4
;
1622 if (max_temps
>= sizeof(int) * 8)
1623 p
.temp_reserved
= 0;
1625 p
.temp_reserved
= ~((1<<max_temps
)-1);
1627 /* Start by allocating 32 instructions.
1628 * If we need more, we'll grow the instruction array as needed.
1631 p
.program
->arb
.Instructions
=
1632 rzalloc_array(program
, struct prog_instruction
, p
.max_inst
);
1633 p
.program
->String
= NULL
;
1634 p
.program
->arb
.NumInstructions
=
1635 p
.program
->arb
.NumTemporaries
=
1636 p
.program
->arb
.NumParameters
=
1637 p
.program
->arb
.NumAttributes
= p
.program
->arb
.NumAddressRegs
= 0;
1638 p
.program
->Parameters
= _mesa_new_parameter_list();
1639 p
.program
->info
.inputs_read
= 0;
1640 p
.program
->info
.outputs_written
= 0;
1642 build_tnl_program( &p
);
1647 * Return a vertex program which implements the current fixed-function
1648 * transform/lighting/texgen operations.
1651 _mesa_get_fixed_func_vertex_program(struct gl_context
*ctx
)
1653 struct gl_program
*prog
;
1654 struct state_key key
;
1656 /* Grab all the relevant state and put it in a single structure:
1658 make_state_key(ctx
, &key
);
1660 /* Look for an already-prepared program for this state:
1662 prog
= _mesa_search_program_cache(ctx
->VertexProgram
.Cache
, &key
,
1666 /* OK, we'll have to build a new one */
1668 printf("Build new TNL program\n");
1670 prog
= ctx
->Driver
.NewProgram(ctx
, GL_VERTEX_PROGRAM_ARB
, 0, true);
1674 create_new_program( &key
, prog
,
1675 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
,
1676 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTemps
);
1678 if (ctx
->Driver
.ProgramStringNotify
)
1679 ctx
->Driver
.ProgramStringNotify(ctx
, GL_VERTEX_PROGRAM_ARB
, prog
);
1681 _mesa_program_cache_insert(ctx
, ctx
->VertexProgram
.Cache
, &key
,