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 point_array
:1;
68 unsigned texture_enabled_global
:1;
69 unsigned fragprog_inputs_read
:12;
71 GLbitfield64 varying_vp_inputs
;
74 unsigned light_enabled
:1;
75 unsigned light_eyepos3_is_zero
:1;
76 unsigned light_spotcutoff_is_180
:1;
77 unsigned light_attenuated
:1;
78 unsigned texunit_really_enabled
:1;
79 unsigned texmat_enabled
:1;
80 unsigned coord_replace
:1;
81 unsigned texgen_enabled
:4;
82 unsigned texgen_mode0
:4;
83 unsigned texgen_mode1
:4;
84 unsigned texgen_mode2
:4;
85 unsigned texgen_mode3
:4;
91 #define TXG_OBJ_LINEAR 1
92 #define TXG_EYE_LINEAR 2
93 #define TXG_SPHERE_MAP 3
94 #define TXG_REFLECTION_MAP 4
95 #define TXG_NORMAL_MAP 5
97 static GLuint
translate_texgen( GLboolean enabled
, GLenum mode
)
103 case GL_OBJECT_LINEAR
: return TXG_OBJ_LINEAR
;
104 case GL_EYE_LINEAR
: return TXG_EYE_LINEAR
;
105 case GL_SPHERE_MAP
: return TXG_SPHERE_MAP
;
106 case GL_REFLECTION_MAP_NV
: return TXG_REFLECTION_MAP
;
107 case GL_NORMAL_MAP_NV
: return TXG_NORMAL_MAP
;
108 default: return TXG_NONE
;
112 #define FDM_EYE_RADIAL 0
113 #define FDM_EYE_PLANE 1
114 #define FDM_EYE_PLANE_ABS 2
116 static GLuint
translate_fog_distance_mode( GLenum mode
)
119 case GL_EYE_RADIAL_NV
:
120 return FDM_EYE_RADIAL
;
122 return FDM_EYE_PLANE
;
123 default: /* shouldn't happen; fall through to a sensible default */
124 case GL_EYE_PLANE_ABSOLUTE_NV
:
125 return FDM_EYE_PLANE_ABS
;
129 static GLboolean
check_active_shininess( struct gl_context
*ctx
,
130 const struct state_key
*key
,
133 GLuint attr
= MAT_ATTRIB_FRONT_SHININESS
+ side
;
135 if ((key
->varying_vp_inputs
& VERT_BIT_COLOR0
) &&
136 (key
->light_color_material_mask
& (1 << attr
)))
139 if (key
->varying_vp_inputs
& VERT_BIT_GENERIC(attr
))
142 if (ctx
->Light
.Material
.Attrib
[attr
][0] != 0.0F
)
149 static void make_state_key( struct gl_context
*ctx
, struct state_key
*key
)
151 const struct gl_fragment_program
*fp
;
154 memset(key
, 0, sizeof(struct state_key
));
155 fp
= ctx
->FragmentProgram
._Current
;
157 /* This now relies on texenvprogram.c being active:
161 key
->need_eye_coords
= ctx
->_NeedEyeCoords
;
163 key
->fragprog_inputs_read
= fp
->Base
.InputsRead
;
164 key
->varying_vp_inputs
= ctx
->varying_vp_inputs
;
166 if (ctx
->RenderMode
== GL_FEEDBACK
) {
167 /* make sure the vertprog emits color and tex0 */
168 key
->fragprog_inputs_read
|= (VARYING_BIT_COL0
| VARYING_BIT_TEX0
);
171 key
->separate_specular
= (ctx
->Light
.Model
.ColorControl
==
172 GL_SEPARATE_SPECULAR_COLOR
);
174 if (ctx
->Light
.Enabled
) {
175 key
->light_global_enabled
= 1;
177 if (ctx
->Light
.Model
.LocalViewer
)
178 key
->light_local_viewer
= 1;
180 if (ctx
->Light
.Model
.TwoSide
)
181 key
->light_twoside
= 1;
183 if (ctx
->Light
.ColorMaterialEnabled
) {
184 key
->light_color_material_mask
= ctx
->Light
._ColorMaterialBitmask
;
187 mask
= ctx
->Light
._EnabledLights
;
189 const int i
= u_bit_scan(&mask
);
190 struct gl_light
*light
= &ctx
->Light
.Light
[i
];
192 key
->unit
[i
].light_enabled
= 1;
194 if (light
->EyePosition
[3] == 0.0F
)
195 key
->unit
[i
].light_eyepos3_is_zero
= 1;
197 if (light
->SpotCutoff
== 180.0F
)
198 key
->unit
[i
].light_spotcutoff_is_180
= 1;
200 if (light
->ConstantAttenuation
!= 1.0F
||
201 light
->LinearAttenuation
!= 0.0F
||
202 light
->QuadraticAttenuation
!= 0.0F
)
203 key
->unit
[i
].light_attenuated
= 1;
206 if (check_active_shininess(ctx
, key
, 0)) {
207 key
->material_shininess_is_zero
= 0;
209 else if (key
->light_twoside
&&
210 check_active_shininess(ctx
, key
, 1)) {
211 key
->material_shininess_is_zero
= 0;
214 key
->material_shininess_is_zero
= 1;
218 if (ctx
->Transform
.Normalize
)
221 if (ctx
->Transform
.RescaleNormals
)
222 key
->rescale_normals
= 1;
224 if (ctx
->Fog
.FogCoordinateSource
== GL_FRAGMENT_DEPTH_EXT
) {
225 key
->fog_source_is_depth
= 1;
226 key
->fog_distance_mode
= translate_fog_distance_mode(ctx
->Fog
.FogDistanceMode
);
229 if (ctx
->Point
._Attenuated
)
230 key
->point_attenuated
= 1;
232 if (ctx
->Array
.VAO
->VertexAttrib
[VERT_ATTRIB_POINT_SIZE
].Enabled
)
233 key
->point_array
= 1;
235 if (ctx
->Texture
._TexGenEnabled
||
236 ctx
->Texture
._TexMatEnabled
||
237 ctx
->Texture
._MaxEnabledTexImageUnit
!= -1)
238 key
->texture_enabled_global
= 1;
240 mask
= ctx
->Texture
._EnabledCoordUnits
| ctx
->Texture
._TexGenEnabled
241 | ctx
->Texture
._TexMatEnabled
| ctx
->Point
.CoordReplace
;
243 const int i
= u_bit_scan(&mask
);
244 struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
246 if (texUnit
->_Current
)
247 key
->unit
[i
].texunit_really_enabled
= 1;
249 if (ctx
->Point
.PointSprite
)
250 if (ctx
->Point
.CoordReplace
& (1u << i
))
251 key
->unit
[i
].coord_replace
= 1;
253 if (ctx
->Texture
._TexMatEnabled
& ENABLE_TEXMAT(i
))
254 key
->unit
[i
].texmat_enabled
= 1;
256 if (texUnit
->TexGenEnabled
) {
257 key
->unit
[i
].texgen_enabled
= 1;
259 key
->unit
[i
].texgen_mode0
=
260 translate_texgen( texUnit
->TexGenEnabled
& (1<<0),
261 texUnit
->GenS
.Mode
);
262 key
->unit
[i
].texgen_mode1
=
263 translate_texgen( texUnit
->TexGenEnabled
& (1<<1),
264 texUnit
->GenT
.Mode
);
265 key
->unit
[i
].texgen_mode2
=
266 translate_texgen( texUnit
->TexGenEnabled
& (1<<2),
267 texUnit
->GenR
.Mode
);
268 key
->unit
[i
].texgen_mode3
=
269 translate_texgen( texUnit
->TexGenEnabled
& (1<<3),
270 texUnit
->GenQ
.Mode
);
277 /* Very useful debugging tool - produces annotated listing of
278 * generated program with line/function references for each
279 * instruction back into this file:
284 /* Use uregs to represent registers internally, translate to Mesa's
285 * expected formats on emit.
287 * NOTE: These are passed by value extensively in this file rather
288 * than as usual by pointer reference. If this disturbs you, try
289 * remembering they are just 32bits in size.
291 * GCC is smart enough to deal with these dword-sized structures in
292 * much the same way as if I had defined them as dwords and was using
293 * macros to access and set the fields. This is much nicer and easier
298 GLint idx
:9; /* relative addressing may be negative */
299 /* sizeof(idx) should == sizeof(prog_src_reg::Index) */
307 const struct state_key
*state
;
308 struct gl_program
*program
;
309 GLuint max_inst
; /** number of instructions allocated for program */
310 GLboolean mvp_with_dp4
;
313 GLuint temp_reserved
;
315 struct ureg eye_position
;
316 struct ureg eye_position_z
;
317 struct ureg eye_position_normalized
;
318 struct ureg transformed_normal
;
319 struct ureg identity
;
322 GLuint color_materials
;
326 static const struct ureg undef
= {
344 static struct ureg
make_ureg(GLuint file
, GLint idx
)
350 reg
.swz
= SWIZZLE_NOOP
;
356 static struct ureg
negate( struct ureg reg
)
363 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
365 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
368 GET_SWZ(reg
.swz
, w
));
373 static struct ureg
swizzle1( struct ureg reg
, int x
)
375 return swizzle(reg
, x
, x
, x
, x
);
379 static struct ureg
get_temp( struct tnl_program
*p
)
381 int bit
= ffs( ~p
->temp_in_use
);
383 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
387 if ((GLuint
) bit
> p
->program
->NumTemporaries
)
388 p
->program
->NumTemporaries
= bit
;
390 p
->temp_in_use
|= 1<<(bit
-1);
391 return make_ureg(PROGRAM_TEMPORARY
, bit
-1);
395 static struct ureg
reserve_temp( struct tnl_program
*p
)
397 struct ureg temp
= get_temp( p
);
398 p
->temp_reserved
|= 1<<temp
.idx
;
403 static void release_temp( struct tnl_program
*p
, struct ureg reg
)
405 if (reg
.file
== PROGRAM_TEMPORARY
) {
406 p
->temp_in_use
&= ~(1<<reg
.idx
);
407 p
->temp_in_use
|= p
->temp_reserved
; /* can't release reserved temps */
411 static void release_temps( struct tnl_program
*p
)
413 p
->temp_in_use
= p
->temp_reserved
;
417 static struct ureg
register_param5(struct tnl_program
*p
,
424 gl_state_index tokens
[STATE_LENGTH
];
431 idx
= _mesa_add_state_reference(p
->program
->Parameters
, tokens
);
432 return make_ureg(PROGRAM_STATE_VAR
, idx
);
436 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
437 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
438 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
439 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
444 * \param input one of VERT_ATTRIB_x tokens.
446 static struct ureg
register_input( struct tnl_program
*p
, GLuint input
)
448 assert(input
< VERT_ATTRIB_MAX
);
450 if (p
->state
->varying_vp_inputs
& VERT_BIT(input
)) {
451 p
->program
->InputsRead
|= VERT_BIT(input
);
452 return make_ureg(PROGRAM_INPUT
, input
);
455 return register_param3( p
, STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, input
);
461 * \param input one of VARYING_SLOT_x tokens.
463 static struct ureg
register_output( struct tnl_program
*p
, GLuint output
)
465 p
->program
->OutputsWritten
|= BITFIELD64_BIT(output
);
466 return make_ureg(PROGRAM_OUTPUT
, output
);
470 static struct ureg
register_const4f( struct tnl_program
*p
,
476 gl_constant_value values
[4];
483 idx
= _mesa_add_unnamed_constant(p
->program
->Parameters
, values
, 4,
485 assert(swizzle
== SWIZZLE_NOOP
);
486 return make_ureg(PROGRAM_CONSTANT
, idx
);
489 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
490 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
491 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
492 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
494 static GLboolean
is_undef( struct ureg reg
)
496 return reg
.file
== PROGRAM_UNDEFINED
;
500 static struct ureg
get_identity_param( struct tnl_program
*p
)
502 if (is_undef(p
->identity
))
503 p
->identity
= register_const4f(p
, 0,0,0,1);
508 static void register_matrix_param5( struct tnl_program
*p
,
509 GLint s0
, /* modelview, projection, etc */
510 GLint s1
, /* texture matrix number */
511 GLint s2
, /* first row */
512 GLint s3
, /* last row */
513 GLint s4
, /* inverse, transpose, etc */
514 struct ureg
*matrix
)
518 /* This is a bit sad as the support is there to pull the whole
519 * matrix out in one go:
521 for (i
= 0; i
<= s3
- s2
; i
++)
522 matrix
[i
] = register_param5( p
, s0
, s1
, i
, i
, s4
);
526 static void emit_arg( struct prog_src_register
*src
,
529 src
->File
= reg
.file
;
530 src
->Index
= reg
.idx
;
531 src
->Swizzle
= reg
.swz
;
532 src
->Negate
= reg
.negate
? NEGATE_XYZW
: NEGATE_NONE
;
534 /* Check that bitfield sizes aren't exceeded */
535 assert(src
->Index
== reg
.idx
);
539 static void emit_dst( struct prog_dst_register
*dst
,
540 struct ureg reg
, GLuint mask
)
542 dst
->File
= reg
.file
;
543 dst
->Index
= reg
.idx
;
544 /* allow zero as a shorthand for xyzw */
545 dst
->WriteMask
= mask
? mask
: WRITEMASK_XYZW
;
546 /* Check that bitfield sizes aren't exceeded */
547 assert(dst
->Index
== reg
.idx
);
551 static void debug_insn( struct prog_instruction
*inst
, const char *fn
,
555 static const char *last_fn
;
562 printf("%d:\t", line
);
563 _mesa_print_instruction(inst
);
568 static void emit_op3fn(struct tnl_program
*p
,
579 struct prog_instruction
*inst
;
581 assert(p
->program
->NumInstructions
<= p
->max_inst
);
583 if (p
->program
->NumInstructions
== p
->max_inst
) {
584 /* need to extend the program's instruction array */
585 struct prog_instruction
*newInst
;
587 /* double the size */
590 newInst
= _mesa_alloc_instructions(p
->max_inst
);
592 _mesa_error(NULL
, GL_OUT_OF_MEMORY
, "vertex program build");
596 _mesa_copy_instructions(newInst
, p
->program
->Instructions
,
597 p
->program
->NumInstructions
);
599 _mesa_free_instructions(p
->program
->Instructions
,
600 p
->program
->NumInstructions
);
602 p
->program
->Instructions
= newInst
;
605 nr
= p
->program
->NumInstructions
++;
607 inst
= &p
->program
->Instructions
[nr
];
608 inst
->Opcode
= (enum prog_opcode
) op
;
610 emit_arg( &inst
->SrcReg
[0], src0
);
611 emit_arg( &inst
->SrcReg
[1], src1
);
612 emit_arg( &inst
->SrcReg
[2], src2
);
614 emit_dst( &inst
->DstReg
, dest
, mask
);
616 debug_insn(inst
, fn
, line
);
620 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
621 emit_op3fn(p, op, dst, mask, src0, src1, src2, __func__, __LINE__)
623 #define emit_op2(p, op, dst, mask, src0, src1) \
624 emit_op3fn(p, op, dst, mask, src0, src1, undef, __func__, __LINE__)
626 #define emit_op1(p, op, dst, mask, src0) \
627 emit_op3fn(p, op, dst, mask, src0, undef, undef, __func__, __LINE__)
630 static struct ureg
make_temp( struct tnl_program
*p
, struct ureg reg
)
632 if (reg
.file
== PROGRAM_TEMPORARY
&&
633 !(p
->temp_reserved
& (1<<reg
.idx
)))
636 struct ureg temp
= get_temp(p
);
637 emit_op1(p
, OPCODE_MOV
, temp
, 0, reg
);
643 /* Currently no tracking performed of input/output/register size or
644 * active elements. Could be used to reduce these operations, as
645 * could the matrix type.
647 static void emit_matrix_transform_vec4( struct tnl_program
*p
,
649 const struct ureg
*mat
,
652 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_X
, src
, mat
[0]);
653 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Y
, src
, mat
[1]);
654 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Z
, src
, mat
[2]);
655 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_W
, src
, mat
[3]);
659 /* This version is much easier to implement if writemasks are not
660 * supported natively on the target or (like SSE), the target doesn't
661 * have a clean/obvious dotproduct implementation.
663 static void emit_transpose_matrix_transform_vec4( struct tnl_program
*p
,
665 const struct ureg
*mat
,
670 if (dest
.file
!= PROGRAM_TEMPORARY
)
675 emit_op2(p
, OPCODE_MUL
, tmp
, 0, swizzle1(src
,X
), mat
[0]);
676 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Y
), mat
[1], tmp
);
677 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Z
), mat
[2], tmp
);
678 emit_op3(p
, OPCODE_MAD
, dest
, 0, swizzle1(src
,W
), mat
[3], tmp
);
680 if (dest
.file
!= PROGRAM_TEMPORARY
)
681 release_temp(p
, tmp
);
685 static void emit_matrix_transform_vec3( struct tnl_program
*p
,
687 const struct ureg
*mat
,
690 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_X
, src
, mat
[0]);
691 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Y
, src
, mat
[1]);
692 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Z
, src
, mat
[2]);
696 static void emit_normalize_vec3( struct tnl_program
*p
,
700 struct ureg tmp
= get_temp(p
);
701 emit_op2(p
, OPCODE_DP3
, tmp
, WRITEMASK_X
, src
, src
);
702 emit_op1(p
, OPCODE_RSQ
, tmp
, WRITEMASK_X
, tmp
);
703 emit_op2(p
, OPCODE_MUL
, dest
, 0, src
, swizzle1(tmp
, X
));
704 release_temp(p
, tmp
);
708 static void emit_passthrough( struct tnl_program
*p
,
712 struct ureg out
= register_output(p
, output
);
713 emit_op1(p
, OPCODE_MOV
, out
, 0, register_input(p
, input
));
717 static struct ureg
get_eye_position( struct tnl_program
*p
)
719 if (is_undef(p
->eye_position
)) {
720 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
721 struct ureg modelview
[4];
723 p
->eye_position
= reserve_temp(p
);
725 if (p
->mvp_with_dp4
) {
726 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
729 emit_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
732 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
733 STATE_MATRIX_TRANSPOSE
, modelview
);
735 emit_transpose_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
739 return p
->eye_position
;
743 static struct ureg
get_eye_position_z( struct tnl_program
*p
)
745 if (!is_undef(p
->eye_position
))
746 return swizzle1(p
->eye_position
, Z
);
748 if (is_undef(p
->eye_position_z
)) {
749 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
750 struct ureg modelview
[4];
752 p
->eye_position_z
= reserve_temp(p
);
754 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
757 emit_op2(p
, OPCODE_DP4
, p
->eye_position_z
, 0, pos
, modelview
[2]);
760 return p
->eye_position_z
;
764 static struct ureg
get_eye_position_normalized( struct tnl_program
*p
)
766 if (is_undef(p
->eye_position_normalized
)) {
767 struct ureg eye
= get_eye_position(p
);
768 p
->eye_position_normalized
= reserve_temp(p
);
769 emit_normalize_vec3(p
, p
->eye_position_normalized
, eye
);
772 return p
->eye_position_normalized
;
776 static struct ureg
get_transformed_normal( struct tnl_program
*p
)
778 if (is_undef(p
->transformed_normal
) &&
779 !p
->state
->need_eye_coords
&&
780 !p
->state
->normalize
&&
781 !(p
->state
->need_eye_coords
== p
->state
->rescale_normals
))
783 p
->transformed_normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
785 else if (is_undef(p
->transformed_normal
))
787 struct ureg normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
788 struct ureg mvinv
[3];
789 struct ureg transformed_normal
= reserve_temp(p
);
791 if (p
->state
->need_eye_coords
) {
792 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 2,
793 STATE_MATRIX_INVTRANS
, mvinv
);
795 /* Transform to eye space:
797 emit_matrix_transform_vec3( p
, transformed_normal
, mvinv
, normal
);
798 normal
= transformed_normal
;
801 /* Normalize/Rescale:
803 if (p
->state
->normalize
) {
804 emit_normalize_vec3( p
, transformed_normal
, normal
);
805 normal
= transformed_normal
;
807 else if (p
->state
->need_eye_coords
== p
->state
->rescale_normals
) {
808 /* This is already adjusted for eye/non-eye rendering:
810 struct ureg rescale
= register_param2(p
, STATE_INTERNAL
,
813 emit_op2( p
, OPCODE_MUL
, transformed_normal
, 0, normal
, rescale
);
814 normal
= transformed_normal
;
817 assert(normal
.file
== PROGRAM_TEMPORARY
);
818 p
->transformed_normal
= normal
;
821 return p
->transformed_normal
;
825 static void build_hpos( struct tnl_program
*p
)
827 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
828 struct ureg hpos
= register_output( p
, VARYING_SLOT_POS
);
831 if (p
->mvp_with_dp4
) {
832 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
834 emit_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
837 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
838 STATE_MATRIX_TRANSPOSE
, mvp
);
839 emit_transpose_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
844 static GLuint
material_attrib( GLuint side
, GLuint property
)
846 return (property
- STATE_AMBIENT
) * 2 + side
;
851 * Get a bitmask of which material values vary on a per-vertex basis.
853 static void set_material_flags( struct tnl_program
*p
)
855 p
->color_materials
= 0;
858 if (p
->state
->varying_vp_inputs
& VERT_BIT_COLOR0
) {
860 p
->color_materials
= p
->state
->light_color_material_mask
;
863 p
->materials
|= (p
->state
->varying_vp_inputs
>> VERT_ATTRIB_GENERIC0
);
867 static struct ureg
get_material( struct tnl_program
*p
, GLuint side
,
870 GLuint attrib
= material_attrib(side
, property
);
872 if (p
->color_materials
& (1<<attrib
))
873 return register_input(p
, VERT_ATTRIB_COLOR0
);
874 else if (p
->materials
& (1<<attrib
)) {
875 /* Put material values in the GENERIC slots -- they are not used
876 * for anything in fixed function mode.
878 return register_input( p
, attrib
+ VERT_ATTRIB_GENERIC0
);
881 return register_param3( p
, STATE_MATERIAL
, side
, property
);
884 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
885 MAT_BIT_FRONT_AMBIENT | \
886 MAT_BIT_FRONT_DIFFUSE) << (side))
890 * Either return a precalculated constant value or emit code to
891 * calculate these values dynamically in the case where material calls
892 * are present between begin/end pairs.
894 * Probably want to shift this to the program compilation phase - if
895 * we always emitted the calculation here, a smart compiler could
896 * detect that it was constant (given a certain set of inputs), and
897 * lift it out of the main loop. That way the programs created here
898 * would be independent of the vertex_buffer details.
900 static struct ureg
get_scenecolor( struct tnl_program
*p
, GLuint side
)
902 if (p
->materials
& SCENE_COLOR_BITS(side
)) {
903 struct ureg lm_ambient
= register_param1(p
, STATE_LIGHTMODEL_AMBIENT
);
904 struct ureg material_emission
= get_material(p
, side
, STATE_EMISSION
);
905 struct ureg material_ambient
= get_material(p
, side
, STATE_AMBIENT
);
906 struct ureg material_diffuse
= get_material(p
, side
, STATE_DIFFUSE
);
907 struct ureg tmp
= make_temp(p
, material_diffuse
);
908 emit_op3(p
, OPCODE_MAD
, tmp
, WRITEMASK_XYZ
, lm_ambient
,
909 material_ambient
, material_emission
);
913 return register_param2( p
, STATE_LIGHTMODEL_SCENECOLOR
, side
);
917 static struct ureg
get_lightprod( struct tnl_program
*p
, GLuint light
,
918 GLuint side
, GLuint property
)
920 GLuint attrib
= material_attrib(side
, property
);
921 if (p
->materials
& (1<<attrib
)) {
922 struct ureg light_value
=
923 register_param3(p
, STATE_LIGHT
, light
, property
);
924 struct ureg material_value
= get_material(p
, side
, property
);
925 struct ureg tmp
= get_temp(p
);
926 emit_op2(p
, OPCODE_MUL
, tmp
, 0, light_value
, material_value
);
930 return register_param4(p
, STATE_LIGHTPROD
, light
, side
, property
);
934 static struct ureg
calculate_light_attenuation( struct tnl_program
*p
,
939 struct ureg attenuation
= register_param3(p
, STATE_LIGHT
, i
,
941 struct ureg att
= undef
;
943 /* Calculate spot attenuation:
945 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
946 struct ureg spot_dir_norm
= register_param3(p
, STATE_INTERNAL
,
947 STATE_LIGHT_SPOT_DIR_NORMALIZED
, i
);
948 struct ureg spot
= get_temp(p
);
949 struct ureg slt
= get_temp(p
);
953 emit_op2(p
, OPCODE_DP3
, spot
, 0, negate(VPpli
), spot_dir_norm
);
954 emit_op2(p
, OPCODE_SLT
, slt
, 0, swizzle1(spot_dir_norm
,W
), spot
);
955 emit_op1(p
, OPCODE_ABS
, spot
, 0, spot
);
956 emit_op2(p
, OPCODE_POW
, spot
, 0, spot
, swizzle1(attenuation
, W
));
957 emit_op2(p
, OPCODE_MUL
, att
, 0, slt
, spot
);
959 release_temp(p
, spot
);
960 release_temp(p
, slt
);
963 /* Calculate distance attenuation(See formula (2.4) at glspec 2.1 page 62):
965 * Skip the calucation when _dist_ is undefined(light_eyepos3_is_zero)
967 if (p
->state
->unit
[i
].light_attenuated
&& !is_undef(dist
)) {
971 emit_op1(p
, OPCODE_RCP
, dist
, WRITEMASK_YZ
, dist
);
973 emit_op2(p
, OPCODE_MUL
, dist
, WRITEMASK_XZ
, dist
, swizzle1(dist
,Y
));
975 emit_op2(p
, OPCODE_DP3
, dist
, 0, attenuation
, dist
);
977 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
979 emit_op1(p
, OPCODE_RCP
, dist
, 0, dist
);
980 /* spot-atten * dist-atten */
981 emit_op2(p
, OPCODE_MUL
, att
, 0, dist
, att
);
985 emit_op1(p
, OPCODE_RCP
, att
, 0, dist
);
995 * lit.y = MAX(0, dots.x)
996 * lit.z = SLT(0, dots.x)
998 static void emit_degenerate_lit( struct tnl_program
*p
,
1002 struct ureg id
= get_identity_param(p
); /* id = {0,0,0,1} */
1004 /* Note that lit.x & lit.w will not be examined. Note also that
1005 * dots.xyzw == dots.xxxx.
1008 /* MAX lit, id, dots;
1010 emit_op2(p
, OPCODE_MAX
, lit
, WRITEMASK_XYZW
, id
, dots
);
1012 /* result[2] = (in > 0 ? 1 : 0)
1013 * SLT lit.z, id.z, dots; # lit.z = (0 < dots.z) ? 1 : 0
1015 emit_op2(p
, OPCODE_SLT
, lit
, WRITEMASK_Z
, swizzle1(id
,Z
), dots
);
1019 /* Need to add some addtional parameters to allow lighting in object
1020 * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
1023 static void build_lighting( struct tnl_program
*p
)
1025 const GLboolean twoside
= p
->state
->light_twoside
;
1026 const GLboolean separate
= p
->state
->separate_specular
;
1027 GLuint nr_lights
= 0, count
= 0;
1028 struct ureg normal
= get_transformed_normal(p
);
1029 struct ureg lit
= get_temp(p
);
1030 struct ureg dots
= get_temp(p
);
1031 struct ureg _col0
= undef
, _col1
= undef
;
1032 struct ureg _bfc0
= undef
, _bfc1
= undef
;
1037 * dots.x = dot(normal, VPpli)
1038 * dots.y = dot(normal, halfAngle)
1039 * dots.z = back.shininess
1040 * dots.w = front.shininess
1043 for (i
= 0; i
< MAX_LIGHTS
; i
++)
1044 if (p
->state
->unit
[i
].light_enabled
)
1047 set_material_flags(p
);
1050 if (!p
->state
->material_shininess_is_zero
) {
1051 struct ureg shininess
= get_material(p
, 0, STATE_SHININESS
);
1052 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_W
, swizzle1(shininess
,X
));
1053 release_temp(p
, shininess
);
1056 _col0
= make_temp(p
, get_scenecolor(p
, 0));
1058 _col1
= make_temp(p
, get_identity_param(p
));
1064 if (!p
->state
->material_shininess_is_zero
) {
1065 /* Note that we negate the back-face specular exponent here.
1066 * The negation will be un-done later in the back-face code below.
1068 struct ureg shininess
= get_material(p
, 1, STATE_SHININESS
);
1069 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_Z
,
1070 negate(swizzle1(shininess
,X
)));
1071 release_temp(p
, shininess
);
1074 _bfc0
= make_temp(p
, get_scenecolor(p
, 1));
1076 _bfc1
= make_temp(p
, get_identity_param(p
));
1081 /* If no lights, still need to emit the scenecolor.
1084 struct ureg res0
= register_output( p
, VARYING_SLOT_COL0
);
1085 emit_op1(p
, OPCODE_MOV
, res0
, 0, _col0
);
1089 struct ureg res1
= register_output( p
, VARYING_SLOT_COL1
);
1090 emit_op1(p
, OPCODE_MOV
, res1
, 0, _col1
);
1094 struct ureg res0
= register_output( p
, VARYING_SLOT_BFC0
);
1095 emit_op1(p
, OPCODE_MOV
, res0
, 0, _bfc0
);
1098 if (twoside
&& separate
) {
1099 struct ureg res1
= register_output( p
, VARYING_SLOT_BFC1
);
1100 emit_op1(p
, OPCODE_MOV
, res1
, 0, _bfc1
);
1103 if (nr_lights
== 0) {
1108 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
1109 if (p
->state
->unit
[i
].light_enabled
) {
1110 struct ureg half
= undef
;
1111 struct ureg att
= undef
, VPpli
= undef
;
1112 struct ureg dist
= undef
;
1115 if (p
->state
->unit
[i
].light_eyepos3_is_zero
) {
1116 VPpli
= register_param3(p
, STATE_INTERNAL
,
1117 STATE_LIGHT_POSITION_NORMALIZED
, i
);
1119 struct ureg Ppli
= register_param3(p
, STATE_INTERNAL
,
1120 STATE_LIGHT_POSITION
, i
);
1121 struct ureg V
= get_eye_position(p
);
1123 VPpli
= get_temp(p
);
1126 /* Calculate VPpli vector
1128 emit_op2(p
, OPCODE_SUB
, VPpli
, 0, Ppli
, V
);
1130 /* Normalize VPpli. The dist value also used in
1131 * attenuation below.
1133 emit_op2(p
, OPCODE_DP3
, dist
, 0, VPpli
, VPpli
);
1134 emit_op1(p
, OPCODE_RSQ
, dist
, 0, dist
);
1135 emit_op2(p
, OPCODE_MUL
, VPpli
, 0, VPpli
, dist
);
1138 /* Calculate attenuation:
1140 att
= calculate_light_attenuation(p
, i
, VPpli
, dist
);
1141 release_temp(p
, dist
);
1143 /* Calculate viewer direction, or use infinite viewer:
1145 if (!p
->state
->material_shininess_is_zero
) {
1146 if (p
->state
->light_local_viewer
) {
1147 struct ureg eye_hat
= get_eye_position_normalized(p
);
1149 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1150 emit_normalize_vec3(p
, half
, half
);
1151 } else if (p
->state
->unit
[i
].light_eyepos3_is_zero
) {
1152 half
= register_param3(p
, STATE_INTERNAL
,
1153 STATE_LIGHT_HALF_VECTOR
, i
);
1155 struct ureg z_dir
= swizzle(get_identity_param(p
),X
,Y
,W
,Z
);
1157 emit_op2(p
, OPCODE_ADD
, half
, 0, VPpli
, z_dir
);
1158 emit_normalize_vec3(p
, half
, half
);
1162 /* Calculate dot products:
1164 if (p
->state
->material_shininess_is_zero
) {
1165 emit_op2(p
, OPCODE_DP3
, dots
, 0, normal
, VPpli
);
1168 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_X
, normal
, VPpli
);
1169 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_Y
, normal
, half
);
1172 /* Front face lighting:
1175 struct ureg ambient
= get_lightprod(p
, i
, 0, STATE_AMBIENT
);
1176 struct ureg diffuse
= get_lightprod(p
, i
, 0, STATE_DIFFUSE
);
1177 struct ureg specular
= get_lightprod(p
, i
, 0, STATE_SPECULAR
);
1178 struct ureg res0
, res1
;
1179 GLuint mask0
, mask1
;
1181 if (count
== nr_lights
) {
1183 mask0
= WRITEMASK_XYZ
;
1184 mask1
= WRITEMASK_XYZ
;
1185 res0
= register_output( p
, VARYING_SLOT_COL0
);
1186 res1
= register_output( p
, VARYING_SLOT_COL1
);
1190 mask1
= WRITEMASK_XYZ
;
1192 res1
= register_output( p
, VARYING_SLOT_COL0
);
1202 if (!is_undef(att
)) {
1203 /* light is attenuated by distance */
1204 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1205 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1206 emit_op3(p
, OPCODE_MAD
, _col0
, 0, swizzle1(lit
,X
), ambient
, _col0
);
1208 else if (!p
->state
->material_shininess_is_zero
) {
1209 /* there's a non-zero specular term */
1210 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1211 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1214 /* no attenutation, no specular */
1215 emit_degenerate_lit(p
, lit
, dots
);
1216 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1219 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _col0
);
1220 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _col1
);
1222 release_temp(p
, ambient
);
1223 release_temp(p
, diffuse
);
1224 release_temp(p
, specular
);
1227 /* Back face lighting:
1230 struct ureg ambient
= get_lightprod(p
, i
, 1, STATE_AMBIENT
);
1231 struct ureg diffuse
= get_lightprod(p
, i
, 1, STATE_DIFFUSE
);
1232 struct ureg specular
= get_lightprod(p
, i
, 1, STATE_SPECULAR
);
1233 struct ureg res0
, res1
;
1234 GLuint mask0
, mask1
;
1236 if (count
== nr_lights
) {
1238 mask0
= WRITEMASK_XYZ
;
1239 mask1
= WRITEMASK_XYZ
;
1240 res0
= register_output( p
, VARYING_SLOT_BFC0
);
1241 res1
= register_output( p
, VARYING_SLOT_BFC1
);
1245 mask1
= WRITEMASK_XYZ
;
1247 res1
= register_output( p
, VARYING_SLOT_BFC0
);
1257 /* For the back face we need to negate the X and Y component
1258 * dot products. dots.Z has the negated back-face specular
1259 * exponent. We swizzle that into the W position. This
1260 * negation makes the back-face specular term positive again.
1262 dots
= negate(swizzle(dots
,X
,Y
,W
,Z
));
1264 if (!is_undef(att
)) {
1265 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1266 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1267 emit_op3(p
, OPCODE_MAD
, _bfc0
, 0, swizzle1(lit
,X
), ambient
, _bfc0
);
1269 else if (!p
->state
->material_shininess_is_zero
) {
1270 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1271 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
); /**/
1274 emit_degenerate_lit(p
, lit
, dots
);
1275 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
);
1278 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _bfc0
);
1279 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _bfc1
);
1280 /* restore dots to its original state for subsequent lights
1281 * by negating and swizzling again.
1283 dots
= negate(swizzle(dots
,X
,Y
,W
,Z
));
1285 release_temp(p
, ambient
);
1286 release_temp(p
, diffuse
);
1287 release_temp(p
, specular
);
1290 release_temp(p
, half
);
1291 release_temp(p
, VPpli
);
1292 release_temp(p
, att
);
1300 static void build_fog( struct tnl_program
*p
)
1302 struct ureg fog
= register_output(p
, VARYING_SLOT_FOGC
);
1305 if (p
->state
->fog_source_is_depth
) {
1307 switch (p
->state
->fog_distance_mode
) {
1308 case FDM_EYE_RADIAL
: /* Z = sqrt(Xe*Xe + Ye*Ye + Ze*Ze) */
1309 input
= get_eye_position(p
);
1310 emit_op2(p
, OPCODE_DP3
, fog
, WRITEMASK_X
, input
, input
);
1311 emit_op1(p
, OPCODE_RSQ
, fog
, WRITEMASK_X
, fog
);
1312 emit_op1(p
, OPCODE_RCP
, fog
, WRITEMASK_X
, fog
);
1314 case FDM_EYE_PLANE
: /* Z = Ze */
1315 input
= get_eye_position_z(p
);
1316 emit_op1(p
, OPCODE_MOV
, fog
, WRITEMASK_X
, input
);
1318 case FDM_EYE_PLANE_ABS
: /* Z = abs(Ze) */
1319 input
= get_eye_position_z(p
);
1320 emit_op1(p
, OPCODE_ABS
, fog
, WRITEMASK_X
, input
);
1323 assert(!"Bad fog mode in build_fog()");
1329 input
= swizzle1(register_input(p
, VERT_ATTRIB_FOG
), X
);
1330 emit_op1(p
, OPCODE_ABS
, fog
, WRITEMASK_X
, input
);
1333 emit_op1(p
, OPCODE_MOV
, fog
, WRITEMASK_YZW
, get_identity_param(p
));
1337 static void build_reflect_texgen( struct tnl_program
*p
,
1341 struct ureg normal
= get_transformed_normal(p
);
1342 struct ureg eye_hat
= get_eye_position_normalized(p
);
1343 struct ureg tmp
= get_temp(p
);
1346 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1348 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1350 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, negate(tmp
), normal
, eye_hat
);
1352 release_temp(p
, tmp
);
1356 static void build_sphere_texgen( struct tnl_program
*p
,
1360 struct ureg normal
= get_transformed_normal(p
);
1361 struct ureg eye_hat
= get_eye_position_normalized(p
);
1362 struct ureg tmp
= get_temp(p
);
1363 struct ureg half
= register_scalar_const(p
, .5);
1364 struct ureg r
= get_temp(p
);
1365 struct ureg inv_m
= get_temp(p
);
1366 struct ureg id
= get_identity_param(p
);
1368 /* Could share the above calculations, but it would be
1369 * a fairly odd state for someone to set (both sphere and
1370 * reflection active for different texture coordinate
1371 * components. Of course - if two texture units enable
1372 * reflect and/or sphere, things start to tilt in favour
1373 * of seperating this out:
1377 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1379 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1381 emit_op3(p
, OPCODE_MAD
, r
, 0, negate(tmp
), normal
, eye_hat
);
1383 emit_op2(p
, OPCODE_ADD
, tmp
, 0, r
, swizzle(id
,X
,Y
,W
,Z
));
1384 /* rx^2 + ry^2 + (rz+1)^2 */
1385 emit_op2(p
, OPCODE_DP3
, tmp
, 0, tmp
, tmp
);
1387 emit_op1(p
, OPCODE_RSQ
, tmp
, 0, tmp
);
1389 emit_op2(p
, OPCODE_MUL
, inv_m
, 0, tmp
, half
);
1391 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, r
, inv_m
, half
);
1393 release_temp(p
, tmp
);
1395 release_temp(p
, inv_m
);
1399 static void build_texture_transform( struct tnl_program
*p
)
1403 for (i
= 0; i
< MAX_TEXTURE_COORD_UNITS
; i
++) {
1405 if (!(p
->state
->fragprog_inputs_read
& VARYING_BIT_TEX(i
)))
1408 if (p
->state
->unit
[i
].coord_replace
)
1411 if (p
->state
->unit
[i
].texgen_enabled
||
1412 p
->state
->unit
[i
].texmat_enabled
) {
1414 GLuint texmat_enabled
= p
->state
->unit
[i
].texmat_enabled
;
1415 struct ureg out
= register_output(p
, VARYING_SLOT_TEX0
+ i
);
1416 struct ureg out_texgen
= undef
;
1418 if (p
->state
->unit
[i
].texgen_enabled
) {
1419 GLuint copy_mask
= 0;
1420 GLuint sphere_mask
= 0;
1421 GLuint reflect_mask
= 0;
1422 GLuint normal_mask
= 0;
1426 out_texgen
= get_temp(p
);
1430 modes
[0] = p
->state
->unit
[i
].texgen_mode0
;
1431 modes
[1] = p
->state
->unit
[i
].texgen_mode1
;
1432 modes
[2] = p
->state
->unit
[i
].texgen_mode2
;
1433 modes
[3] = p
->state
->unit
[i
].texgen_mode3
;
1435 for (j
= 0; j
< 4; j
++) {
1437 case TXG_OBJ_LINEAR
: {
1438 struct ureg obj
= register_input(p
, VERT_ATTRIB_POS
);
1440 register_param3(p
, STATE_TEXGEN
, i
,
1441 STATE_TEXGEN_OBJECT_S
+ j
);
1443 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1447 case TXG_EYE_LINEAR
: {
1448 struct ureg eye
= get_eye_position(p
);
1450 register_param3(p
, STATE_TEXGEN
, i
,
1451 STATE_TEXGEN_EYE_S
+ j
);
1453 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1457 case TXG_SPHERE_MAP
:
1458 sphere_mask
|= WRITEMASK_X
<< j
;
1460 case TXG_REFLECTION_MAP
:
1461 reflect_mask
|= WRITEMASK_X
<< j
;
1463 case TXG_NORMAL_MAP
:
1464 normal_mask
|= WRITEMASK_X
<< j
;
1467 copy_mask
|= WRITEMASK_X
<< j
;
1472 build_sphere_texgen(p
, out_texgen
, sphere_mask
);
1476 build_reflect_texgen(p
, out_texgen
, reflect_mask
);
1480 struct ureg normal
= get_transformed_normal(p
);
1481 emit_op1(p
, OPCODE_MOV
, out_texgen
, normal_mask
, normal
);
1485 struct ureg in
= register_input(p
, VERT_ATTRIB_TEX0
+i
);
1486 emit_op1(p
, OPCODE_MOV
, out_texgen
, copy_mask
, in
);
1490 if (texmat_enabled
) {
1491 struct ureg texmat
[4];
1492 struct ureg in
= (!is_undef(out_texgen
) ?
1494 register_input(p
, VERT_ATTRIB_TEX0
+i
));
1495 if (p
->mvp_with_dp4
) {
1496 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1498 emit_matrix_transform_vec4( p
, out
, texmat
, in
);
1501 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1502 STATE_MATRIX_TRANSPOSE
, texmat
);
1503 emit_transpose_matrix_transform_vec4( p
, out
, texmat
, in
);
1510 emit_passthrough(p
, VERT_ATTRIB_TEX0
+i
, VARYING_SLOT_TEX0
+i
);
1517 * Point size attenuation computation.
1519 static void build_atten_pointsize( struct tnl_program
*p
)
1521 struct ureg eye
= get_eye_position_z(p
);
1522 struct ureg state_size
= register_param2(p
, STATE_INTERNAL
, STATE_POINT_SIZE_CLAMPED
);
1523 struct ureg state_attenuation
= register_param1(p
, STATE_POINT_ATTENUATION
);
1524 struct ureg out
= register_output(p
, VARYING_SLOT_PSIZ
);
1525 struct ureg ut
= get_temp(p
);
1528 emit_op1(p
, OPCODE_ABS
, ut
, WRITEMASK_Y
, swizzle1(eye
, Z
));
1529 /* p1 + dist * (p2 + dist * p3); */
1530 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1531 swizzle1(state_attenuation
, Z
), swizzle1(state_attenuation
, Y
));
1532 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1533 ut
, swizzle1(state_attenuation
, X
));
1535 /* 1 / sqrt(factor) */
1536 emit_op1(p
, OPCODE_RSQ
, ut
, WRITEMASK_X
, ut
);
1539 /* out = pointSize / sqrt(factor) */
1540 emit_op2(p
, OPCODE_MUL
, out
, WRITEMASK_X
, ut
, state_size
);
1542 /* this is a good place to clamp the point size since there's likely
1543 * no hardware registers to clamp point size at rasterization time.
1545 emit_op2(p
, OPCODE_MUL
, ut
, WRITEMASK_X
, ut
, state_size
);
1546 emit_op2(p
, OPCODE_MAX
, ut
, WRITEMASK_X
, ut
, swizzle1(state_size
, Y
));
1547 emit_op2(p
, OPCODE_MIN
, out
, WRITEMASK_X
, ut
, swizzle1(state_size
, Z
));
1550 release_temp(p
, ut
);
1555 * Pass-though per-vertex point size, from user's point size array.
1557 static void build_array_pointsize( struct tnl_program
*p
)
1559 struct ureg in
= register_input(p
, VERT_ATTRIB_POINT_SIZE
);
1560 struct ureg out
= register_output(p
, VARYING_SLOT_PSIZ
);
1561 emit_op1(p
, OPCODE_MOV
, out
, WRITEMASK_X
, in
);
1565 static void build_tnl_program( struct tnl_program
*p
)
1567 /* Emit the program, starting with the modelview, projection transforms:
1571 /* Lighting calculations:
1573 if (p
->state
->fragprog_inputs_read
& (VARYING_BIT_COL0
|VARYING_BIT_COL1
)) {
1574 if (p
->state
->light_global_enabled
)
1577 if (p
->state
->fragprog_inputs_read
& VARYING_BIT_COL0
)
1578 emit_passthrough(p
, VERT_ATTRIB_COLOR0
, VARYING_SLOT_COL0
);
1580 if (p
->state
->fragprog_inputs_read
& VARYING_BIT_COL1
)
1581 emit_passthrough(p
, VERT_ATTRIB_COLOR1
, VARYING_SLOT_COL1
);
1585 if (p
->state
->fragprog_inputs_read
& VARYING_BIT_FOGC
)
1588 if (p
->state
->fragprog_inputs_read
& VARYING_BITS_TEX_ANY
)
1589 build_texture_transform(p
);
1591 if (p
->state
->point_attenuated
)
1592 build_atten_pointsize(p
);
1593 else if (p
->state
->point_array
)
1594 build_array_pointsize(p
);
1598 emit_op1(p
, OPCODE_END
, undef
, 0, undef
);
1609 create_new_program( const struct state_key
*key
,
1610 struct gl_program
*program
,
1611 GLboolean mvp_with_dp4
,
1614 struct tnl_program p
;
1616 memset(&p
, 0, sizeof(p
));
1618 p
.program
= program
;
1619 p
.eye_position
= undef
;
1620 p
.eye_position_z
= undef
;
1621 p
.eye_position_normalized
= undef
;
1622 p
.transformed_normal
= undef
;
1625 p
.mvp_with_dp4
= mvp_with_dp4
;
1627 if (max_temps
>= sizeof(int) * 8)
1628 p
.temp_reserved
= 0;
1630 p
.temp_reserved
= ~((1<<max_temps
)-1);
1632 /* Start by allocating 32 instructions.
1633 * If we need more, we'll grow the instruction array as needed.
1636 p
.program
->Instructions
= _mesa_alloc_instructions(p
.max_inst
);
1637 p
.program
->String
= NULL
;
1638 p
.program
->NumInstructions
=
1639 p
.program
->NumTemporaries
=
1640 p
.program
->NumParameters
=
1641 p
.program
->NumAttributes
= p
.program
->NumAddressRegs
= 0;
1642 p
.program
->Parameters
= _mesa_new_parameter_list();
1643 p
.program
->InputsRead
= 0;
1644 p
.program
->OutputsWritten
= 0;
1646 build_tnl_program( &p
);
1651 * Return a vertex program which implements the current fixed-function
1652 * transform/lighting/texgen operations.
1655 _mesa_get_fixed_func_vertex_program(struct gl_context
*ctx
)
1657 struct gl_program
*prog
;
1658 struct state_key key
;
1660 /* Grab all the relevant state and put it in a single structure:
1662 make_state_key(ctx
, &key
);
1664 /* Look for an already-prepared program for this state:
1666 prog
= _mesa_search_program_cache(ctx
->VertexProgram
.Cache
, &key
,
1670 /* OK, we'll have to build a new one */
1672 printf("Build new TNL program\n");
1674 prog
= ctx
->Driver
.NewProgram(ctx
, GL_VERTEX_PROGRAM_ARB
, 0);
1678 create_new_program( &key
, prog
,
1679 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
,
1680 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTemps
);
1682 if (ctx
->Driver
.ProgramStringNotify
)
1683 ctx
->Driver
.ProgramStringNotify(ctx
, GL_VERTEX_PROGRAM_ARB
, prog
);
1685 _mesa_program_cache_insert(ctx
, ctx
->VertexProgram
.Cache
, &key
,