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_distance_mode
:2;
64 unsigned separate_specular
:1;
65 unsigned point_attenuated
:1;
66 unsigned texture_enabled_global
:1;
67 unsigned fragprog_inputs_read
:12;
69 GLbitfield varying_vp_inputs
;
72 unsigned light_enabled
:1;
73 unsigned light_eyepos3_is_zero
:1;
74 unsigned light_spotcutoff_is_180
:1;
75 unsigned light_attenuated
:1;
76 unsigned texunit_really_enabled
:1;
77 unsigned texmat_enabled
:1;
78 unsigned coord_replace
:1;
79 unsigned texgen_enabled
:4;
80 unsigned texgen_mode0
:4;
81 unsigned texgen_mode1
:4;
82 unsigned texgen_mode2
:4;
83 unsigned texgen_mode3
:4;
89 #define TXG_OBJ_LINEAR 1
90 #define TXG_EYE_LINEAR 2
91 #define TXG_SPHERE_MAP 3
92 #define TXG_REFLECTION_MAP 4
93 #define TXG_NORMAL_MAP 5
95 static GLuint
translate_texgen( GLboolean enabled
, GLenum mode
)
101 case GL_OBJECT_LINEAR
: return TXG_OBJ_LINEAR
;
102 case GL_EYE_LINEAR
: return TXG_EYE_LINEAR
;
103 case GL_SPHERE_MAP
: return TXG_SPHERE_MAP
;
104 case GL_REFLECTION_MAP_NV
: return TXG_REFLECTION_MAP
;
105 case GL_NORMAL_MAP_NV
: return TXG_NORMAL_MAP
;
106 default: return TXG_NONE
;
110 #define FDM_EYE_RADIAL 0
111 #define FDM_EYE_PLANE 1
112 #define FDM_EYE_PLANE_ABS 2
113 #define FDM_FROM_ARRAY 3
115 static GLuint
translate_fog_distance_mode(GLenum source
, GLenum mode
)
117 if (source
== GL_FRAGMENT_DEPTH_EXT
) {
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
;
128 return FDM_FROM_ARRAY
;
132 static GLboolean
check_active_shininess( struct gl_context
*ctx
,
133 const struct state_key
*key
,
136 GLuint attr
= MAT_ATTRIB_FRONT_SHININESS
+ side
;
138 if ((key
->varying_vp_inputs
& VERT_BIT_COLOR0
) &&
139 (key
->light_color_material_mask
& (1 << attr
)))
142 if (key
->varying_vp_inputs
& VERT_BIT_GENERIC(attr
))
145 if (ctx
->Light
.Material
.Attrib
[attr
][0] != 0.0F
)
152 static void make_state_key( struct gl_context
*ctx
, struct state_key
*key
)
154 const struct gl_program
*fp
= ctx
->FragmentProgram
._Current
;
157 memset(key
, 0, sizeof(struct state_key
));
159 /* This now relies on texenvprogram.c being active:
163 key
->need_eye_coords
= ctx
->_NeedEyeCoords
;
165 key
->fragprog_inputs_read
= fp
->info
.inputs_read
;
166 key
->varying_vp_inputs
= ctx
->varying_vp_inputs
;
168 if (ctx
->RenderMode
== GL_FEEDBACK
) {
169 /* make sure the vertprog emits color and tex0 */
170 key
->fragprog_inputs_read
|= (VARYING_BIT_COL0
| VARYING_BIT_TEX0
);
173 if (ctx
->Light
.Enabled
) {
174 key
->light_global_enabled
= 1;
176 if (ctx
->Light
.Model
.LocalViewer
)
177 key
->light_local_viewer
= 1;
179 if (ctx
->Light
.Model
.TwoSide
)
180 key
->light_twoside
= 1;
182 if (ctx
->Light
.Model
.ColorControl
== GL_SEPARATE_SPECULAR_COLOR
)
183 key
->separate_specular
= 1;
185 if (ctx
->Light
.ColorMaterialEnabled
) {
186 key
->light_color_material_mask
= ctx
->Light
._ColorMaterialBitmask
;
189 mask
= ctx
->Light
._EnabledLights
;
191 const int i
= u_bit_scan(&mask
);
192 struct gl_light
*light
= &ctx
->Light
.Light
[i
];
194 key
->unit
[i
].light_enabled
= 1;
196 if (light
->EyePosition
[3] == 0.0F
)
197 key
->unit
[i
].light_eyepos3_is_zero
= 1;
199 if (light
->SpotCutoff
== 180.0F
)
200 key
->unit
[i
].light_spotcutoff_is_180
= 1;
202 if (light
->ConstantAttenuation
!= 1.0F
||
203 light
->LinearAttenuation
!= 0.0F
||
204 light
->QuadraticAttenuation
!= 0.0F
)
205 key
->unit
[i
].light_attenuated
= 1;
208 if (check_active_shininess(ctx
, key
, 0)) {
209 key
->material_shininess_is_zero
= 0;
211 else if (key
->light_twoside
&&
212 check_active_shininess(ctx
, key
, 1)) {
213 key
->material_shininess_is_zero
= 0;
216 key
->material_shininess_is_zero
= 1;
220 if (ctx
->Transform
.Normalize
)
223 if (ctx
->Transform
.RescaleNormals
)
224 key
->rescale_normals
= 1;
226 key
->fog_distance_mode
=
227 translate_fog_distance_mode(ctx
->Fog
.FogCoordinateSource
,
228 ctx
->Fog
.FogDistanceMode
);
230 if (ctx
->Point
._Attenuated
)
231 key
->point_attenuated
= 1;
233 if (ctx
->Texture
._TexGenEnabled
||
234 ctx
->Texture
._TexMatEnabled
||
235 ctx
->Texture
._MaxEnabledTexImageUnit
!= -1)
236 key
->texture_enabled_global
= 1;
238 mask
= ctx
->Texture
._EnabledCoordUnits
| ctx
->Texture
._TexGenEnabled
239 | ctx
->Texture
._TexMatEnabled
| ctx
->Point
.CoordReplace
;
241 const int i
= u_bit_scan(&mask
);
242 struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
244 if (texUnit
->_Current
)
245 key
->unit
[i
].texunit_really_enabled
= 1;
247 if (ctx
->Point
.PointSprite
)
248 if (ctx
->Point
.CoordReplace
& (1u << i
))
249 key
->unit
[i
].coord_replace
= 1;
251 if (ctx
->Texture
._TexMatEnabled
& ENABLE_TEXMAT(i
))
252 key
->unit
[i
].texmat_enabled
= 1;
254 if (texUnit
->TexGenEnabled
) {
255 key
->unit
[i
].texgen_enabled
= 1;
257 key
->unit
[i
].texgen_mode0
=
258 translate_texgen( texUnit
->TexGenEnabled
& (1<<0),
259 texUnit
->GenS
.Mode
);
260 key
->unit
[i
].texgen_mode1
=
261 translate_texgen( texUnit
->TexGenEnabled
& (1<<1),
262 texUnit
->GenT
.Mode
);
263 key
->unit
[i
].texgen_mode2
=
264 translate_texgen( texUnit
->TexGenEnabled
& (1<<2),
265 texUnit
->GenR
.Mode
);
266 key
->unit
[i
].texgen_mode3
=
267 translate_texgen( texUnit
->TexGenEnabled
& (1<<3),
268 texUnit
->GenQ
.Mode
);
275 /* Very useful debugging tool - produces annotated listing of
276 * generated program with line/function references for each
277 * instruction back into this file:
282 /* Use uregs to represent registers internally, translate to Mesa's
283 * expected formats on emit.
285 * NOTE: These are passed by value extensively in this file rather
286 * than as usual by pointer reference. If this disturbs you, try
287 * remembering they are just 32bits in size.
289 * GCC is smart enough to deal with these dword-sized structures in
290 * much the same way as if I had defined them as dwords and was using
291 * macros to access and set the fields. This is much nicer and easier
296 GLint idx
:9; /* relative addressing may be negative */
297 /* sizeof(idx) should == sizeof(prog_src_reg::Index) */
305 const struct state_key
*state
;
306 struct gl_program
*program
;
307 GLuint max_inst
; /** number of instructions allocated for program */
308 GLboolean mvp_with_dp4
;
311 GLuint temp_reserved
;
313 struct ureg eye_position
;
314 struct ureg eye_position_z
;
315 struct ureg eye_position_normalized
;
316 struct ureg transformed_normal
;
317 struct ureg identity
;
320 GLuint color_materials
;
324 static const struct ureg undef
= {
342 static struct ureg
make_ureg(GLuint file
, GLint idx
)
348 reg
.swz
= SWIZZLE_NOOP
;
354 static struct ureg
negate( struct ureg reg
)
361 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
363 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
366 GET_SWZ(reg
.swz
, w
));
371 static struct ureg
swizzle1( struct ureg reg
, int x
)
373 return swizzle(reg
, x
, x
, x
, x
);
377 static struct ureg
get_temp( struct tnl_program
*p
)
379 int bit
= ffs( ~p
->temp_in_use
);
381 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
385 if ((GLuint
) bit
> p
->program
->arb
.NumTemporaries
)
386 p
->program
->arb
.NumTemporaries
= bit
;
388 p
->temp_in_use
|= 1<<(bit
-1);
389 return make_ureg(PROGRAM_TEMPORARY
, bit
-1);
393 static struct ureg
reserve_temp( struct tnl_program
*p
)
395 struct ureg temp
= get_temp( p
);
396 p
->temp_reserved
|= 1<<temp
.idx
;
401 static void release_temp( struct tnl_program
*p
, struct ureg reg
)
403 if (reg
.file
== PROGRAM_TEMPORARY
) {
404 p
->temp_in_use
&= ~(1<<reg
.idx
);
405 p
->temp_in_use
|= p
->temp_reserved
; /* can't release reserved temps */
409 static void release_temps( struct tnl_program
*p
)
411 p
->temp_in_use
= p
->temp_reserved
;
415 static struct ureg
register_param5(struct tnl_program
*p
,
422 gl_state_index tokens
[STATE_LENGTH
];
429 idx
= _mesa_add_state_reference(p
->program
->Parameters
, tokens
);
430 return make_ureg(PROGRAM_STATE_VAR
, idx
);
434 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
435 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
436 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
437 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
442 * \param input one of VERT_ATTRIB_x tokens.
444 static struct ureg
register_input( struct tnl_program
*p
, GLuint input
)
446 assert(input
< VERT_ATTRIB_MAX
);
448 if (p
->state
->varying_vp_inputs
& VERT_BIT(input
)) {
449 p
->program
->info
.inputs_read
|= VERT_BIT(input
);
450 return make_ureg(PROGRAM_INPUT
, input
);
453 return register_param3( p
, STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, input
);
459 * \param input one of VARYING_SLOT_x tokens.
461 static struct ureg
register_output( struct tnl_program
*p
, GLuint output
)
463 p
->program
->info
.outputs_written
|= BITFIELD64_BIT(output
);
464 return make_ureg(PROGRAM_OUTPUT
, output
);
468 static struct ureg
register_const4f( struct tnl_program
*p
,
474 gl_constant_value values
[4];
481 idx
= _mesa_add_unnamed_constant(p
->program
->Parameters
, values
, 4,
483 assert(swizzle
== SWIZZLE_NOOP
);
484 return make_ureg(PROGRAM_CONSTANT
, idx
);
487 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
488 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
489 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
490 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
492 static GLboolean
is_undef( struct ureg reg
)
494 return reg
.file
== PROGRAM_UNDEFINED
;
498 static struct ureg
get_identity_param( struct tnl_program
*p
)
500 if (is_undef(p
->identity
))
501 p
->identity
= register_const4f(p
, 0,0,0,1);
506 static void register_matrix_param5( struct tnl_program
*p
,
507 GLint s0
, /* modelview, projection, etc */
508 GLint s1
, /* texture matrix number */
509 GLint s2
, /* first row */
510 GLint s3
, /* last row */
511 GLint s4
, /* inverse, transpose, etc */
512 struct ureg
*matrix
)
516 /* This is a bit sad as the support is there to pull the whole
517 * matrix out in one go:
519 for (i
= 0; i
<= s3
- s2
; i
++)
520 matrix
[i
] = register_param5( p
, s0
, s1
, i
, i
, s4
);
524 static void emit_arg( struct prog_src_register
*src
,
527 src
->File
= reg
.file
;
528 src
->Index
= reg
.idx
;
529 src
->Swizzle
= reg
.swz
;
530 src
->Negate
= reg
.negate
? NEGATE_XYZW
: NEGATE_NONE
;
532 /* Check that bitfield sizes aren't exceeded */
533 assert(src
->Index
== reg
.idx
);
537 static void emit_dst( struct prog_dst_register
*dst
,
538 struct ureg reg
, GLuint mask
)
540 dst
->File
= reg
.file
;
541 dst
->Index
= reg
.idx
;
542 /* allow zero as a shorthand for xyzw */
543 dst
->WriteMask
= mask
? mask
: WRITEMASK_XYZW
;
544 /* Check that bitfield sizes aren't exceeded */
545 assert(dst
->Index
== reg
.idx
);
549 static void debug_insn( struct prog_instruction
*inst
, const char *fn
,
553 static const char *last_fn
;
560 printf("%d:\t", line
);
561 _mesa_print_instruction(inst
);
566 static void emit_op3fn(struct tnl_program
*p
,
577 struct prog_instruction
*inst
;
579 assert(p
->program
->arb
.NumInstructions
<= p
->max_inst
);
581 if (p
->program
->arb
.NumInstructions
== p
->max_inst
) {
582 /* need to extend the program's instruction array */
583 struct prog_instruction
*newInst
;
585 /* double the size */
589 rzalloc_array(p
->program
, struct prog_instruction
, p
->max_inst
);
591 _mesa_error(NULL
, GL_OUT_OF_MEMORY
, "vertex program build");
595 _mesa_copy_instructions(newInst
, p
->program
->arb
.Instructions
,
596 p
->program
->arb
.NumInstructions
);
598 ralloc_free(p
->program
->arb
.Instructions
);
600 p
->program
->arb
.Instructions
= newInst
;
603 nr
= p
->program
->arb
.NumInstructions
++;
605 inst
= &p
->program
->arb
.Instructions
[nr
];
606 inst
->Opcode
= (enum prog_opcode
) op
;
608 emit_arg( &inst
->SrcReg
[0], src0
);
609 emit_arg( &inst
->SrcReg
[1], src1
);
610 emit_arg( &inst
->SrcReg
[2], src2
);
612 emit_dst( &inst
->DstReg
, dest
, mask
);
614 debug_insn(inst
, fn
, line
);
618 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
619 emit_op3fn(p, op, dst, mask, src0, src1, src2, __func__, __LINE__)
621 #define emit_op2(p, op, dst, mask, src0, src1) \
622 emit_op3fn(p, op, dst, mask, src0, src1, undef, __func__, __LINE__)
624 #define emit_op1(p, op, dst, mask, src0) \
625 emit_op3fn(p, op, dst, mask, src0, undef, undef, __func__, __LINE__)
628 static struct ureg
make_temp( struct tnl_program
*p
, struct ureg reg
)
630 if (reg
.file
== PROGRAM_TEMPORARY
&&
631 !(p
->temp_reserved
& (1<<reg
.idx
)))
634 struct ureg temp
= get_temp(p
);
635 emit_op1(p
, OPCODE_MOV
, temp
, 0, reg
);
641 /* Currently no tracking performed of input/output/register size or
642 * active elements. Could be used to reduce these operations, as
643 * could the matrix type.
645 static void emit_matrix_transform_vec4( struct tnl_program
*p
,
647 const struct ureg
*mat
,
650 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_X
, src
, mat
[0]);
651 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Y
, src
, mat
[1]);
652 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Z
, src
, mat
[2]);
653 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_W
, src
, mat
[3]);
657 /* This version is much easier to implement if writemasks are not
658 * supported natively on the target or (like SSE), the target doesn't
659 * have a clean/obvious dotproduct implementation.
661 static void emit_transpose_matrix_transform_vec4( struct tnl_program
*p
,
663 const struct ureg
*mat
,
668 if (dest
.file
!= PROGRAM_TEMPORARY
)
673 emit_op2(p
, OPCODE_MUL
, tmp
, 0, swizzle1(src
,X
), mat
[0]);
674 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Y
), mat
[1], tmp
);
675 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Z
), mat
[2], tmp
);
676 emit_op3(p
, OPCODE_MAD
, dest
, 0, swizzle1(src
,W
), mat
[3], tmp
);
678 if (dest
.file
!= PROGRAM_TEMPORARY
)
679 release_temp(p
, tmp
);
683 static void emit_matrix_transform_vec3( struct tnl_program
*p
,
685 const struct ureg
*mat
,
688 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_X
, src
, mat
[0]);
689 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Y
, src
, mat
[1]);
690 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Z
, src
, mat
[2]);
694 static void emit_normalize_vec3( struct tnl_program
*p
,
698 struct ureg tmp
= get_temp(p
);
699 emit_op2(p
, OPCODE_DP3
, tmp
, WRITEMASK_X
, src
, src
);
700 emit_op1(p
, OPCODE_RSQ
, tmp
, WRITEMASK_X
, tmp
);
701 emit_op2(p
, OPCODE_MUL
, dest
, 0, src
, swizzle1(tmp
, X
));
702 release_temp(p
, tmp
);
706 static void emit_passthrough( struct tnl_program
*p
,
710 struct ureg out
= register_output(p
, output
);
711 emit_op1(p
, OPCODE_MOV
, out
, 0, register_input(p
, input
));
715 static struct ureg
get_eye_position( struct tnl_program
*p
)
717 if (is_undef(p
->eye_position
)) {
718 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
719 struct ureg modelview
[4];
721 p
->eye_position
= reserve_temp(p
);
723 if (p
->mvp_with_dp4
) {
724 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
727 emit_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
730 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
731 STATE_MATRIX_TRANSPOSE
, modelview
);
733 emit_transpose_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
737 return p
->eye_position
;
741 static struct ureg
get_eye_position_z( struct tnl_program
*p
)
743 if (!is_undef(p
->eye_position
))
744 return swizzle1(p
->eye_position
, Z
);
746 if (is_undef(p
->eye_position_z
)) {
747 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
748 struct ureg modelview
[4];
750 p
->eye_position_z
= reserve_temp(p
);
752 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
755 emit_op2(p
, OPCODE_DP4
, p
->eye_position_z
, 0, pos
, modelview
[2]);
758 return p
->eye_position_z
;
762 static struct ureg
get_eye_position_normalized( struct tnl_program
*p
)
764 if (is_undef(p
->eye_position_normalized
)) {
765 struct ureg eye
= get_eye_position(p
);
766 p
->eye_position_normalized
= reserve_temp(p
);
767 emit_normalize_vec3(p
, p
->eye_position_normalized
, eye
);
770 return p
->eye_position_normalized
;
774 static struct ureg
get_transformed_normal( struct tnl_program
*p
)
776 if (is_undef(p
->transformed_normal
) &&
777 !p
->state
->need_eye_coords
&&
778 !p
->state
->normalize
&&
779 !(p
->state
->need_eye_coords
== p
->state
->rescale_normals
))
781 p
->transformed_normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
783 else if (is_undef(p
->transformed_normal
))
785 struct ureg normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
786 struct ureg mvinv
[3];
787 struct ureg transformed_normal
= reserve_temp(p
);
789 if (p
->state
->need_eye_coords
) {
790 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 2,
791 STATE_MATRIX_INVTRANS
, mvinv
);
793 /* Transform to eye space:
795 emit_matrix_transform_vec3( p
, transformed_normal
, mvinv
, normal
);
796 normal
= transformed_normal
;
799 /* Normalize/Rescale:
801 if (p
->state
->normalize
) {
802 emit_normalize_vec3( p
, transformed_normal
, normal
);
803 normal
= transformed_normal
;
805 else if (p
->state
->need_eye_coords
== p
->state
->rescale_normals
) {
806 /* This is already adjusted for eye/non-eye rendering:
808 struct ureg rescale
= register_param2(p
, STATE_INTERNAL
,
811 emit_op2( p
, OPCODE_MUL
, transformed_normal
, 0, normal
, rescale
);
812 normal
= transformed_normal
;
815 assert(normal
.file
== PROGRAM_TEMPORARY
);
816 p
->transformed_normal
= normal
;
819 return p
->transformed_normal
;
823 static void build_hpos( struct tnl_program
*p
)
825 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
826 struct ureg hpos
= register_output( p
, VARYING_SLOT_POS
);
829 if (p
->mvp_with_dp4
) {
830 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
832 emit_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
835 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
836 STATE_MATRIX_TRANSPOSE
, mvp
);
837 emit_transpose_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
842 static GLuint
material_attrib( GLuint side
, GLuint property
)
844 return (property
- STATE_AMBIENT
) * 2 + side
;
849 * Get a bitmask of which material values vary on a per-vertex basis.
851 static void set_material_flags( struct tnl_program
*p
)
853 p
->color_materials
= 0;
856 if (p
->state
->varying_vp_inputs
& VERT_BIT_COLOR0
) {
858 p
->color_materials
= p
->state
->light_color_material_mask
;
861 p
->materials
|= (p
->state
->varying_vp_inputs
>> VERT_ATTRIB_GENERIC0
);
865 static struct ureg
get_material( struct tnl_program
*p
, GLuint side
,
868 GLuint attrib
= material_attrib(side
, property
);
870 if (p
->color_materials
& (1<<attrib
))
871 return register_input(p
, VERT_ATTRIB_COLOR0
);
872 else if (p
->materials
& (1<<attrib
)) {
873 /* Put material values in the GENERIC slots -- they are not used
874 * for anything in fixed function mode.
876 return register_input( p
, attrib
+ VERT_ATTRIB_GENERIC0
);
879 return register_param3( p
, STATE_MATERIAL
, side
, property
);
882 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
883 MAT_BIT_FRONT_AMBIENT | \
884 MAT_BIT_FRONT_DIFFUSE) << (side))
888 * Either return a precalculated constant value or emit code to
889 * calculate these values dynamically in the case where material calls
890 * are present between begin/end pairs.
892 * Probably want to shift this to the program compilation phase - if
893 * we always emitted the calculation here, a smart compiler could
894 * detect that it was constant (given a certain set of inputs), and
895 * lift it out of the main loop. That way the programs created here
896 * would be independent of the vertex_buffer details.
898 static struct ureg
get_scenecolor( struct tnl_program
*p
, GLuint side
)
900 if (p
->materials
& SCENE_COLOR_BITS(side
)) {
901 struct ureg lm_ambient
= register_param1(p
, STATE_LIGHTMODEL_AMBIENT
);
902 struct ureg material_emission
= get_material(p
, side
, STATE_EMISSION
);
903 struct ureg material_ambient
= get_material(p
, side
, STATE_AMBIENT
);
904 struct ureg material_diffuse
= get_material(p
, side
, STATE_DIFFUSE
);
905 struct ureg tmp
= make_temp(p
, material_diffuse
);
906 emit_op3(p
, OPCODE_MAD
, tmp
, WRITEMASK_XYZ
, lm_ambient
,
907 material_ambient
, material_emission
);
911 return register_param2( p
, STATE_LIGHTMODEL_SCENECOLOR
, side
);
915 static struct ureg
get_lightprod( struct tnl_program
*p
, GLuint light
,
916 GLuint side
, GLuint property
)
918 GLuint attrib
= material_attrib(side
, property
);
919 if (p
->materials
& (1<<attrib
)) {
920 struct ureg light_value
=
921 register_param3(p
, STATE_LIGHT
, light
, property
);
922 struct ureg material_value
= get_material(p
, side
, property
);
923 struct ureg tmp
= get_temp(p
);
924 emit_op2(p
, OPCODE_MUL
, tmp
, 0, light_value
, material_value
);
928 return register_param4(p
, STATE_LIGHTPROD
, light
, side
, property
);
932 static struct ureg
calculate_light_attenuation( struct tnl_program
*p
,
937 struct ureg attenuation
= register_param3(p
, STATE_LIGHT
, i
,
939 struct ureg att
= undef
;
941 /* Calculate spot attenuation:
943 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
944 struct ureg spot_dir_norm
= register_param3(p
, STATE_INTERNAL
,
945 STATE_LIGHT_SPOT_DIR_NORMALIZED
, i
);
946 struct ureg spot
= get_temp(p
);
947 struct ureg slt
= get_temp(p
);
951 emit_op2(p
, OPCODE_DP3
, spot
, 0, negate(VPpli
), spot_dir_norm
);
952 emit_op2(p
, OPCODE_SLT
, slt
, 0, swizzle1(spot_dir_norm
,W
), spot
);
953 emit_op1(p
, OPCODE_ABS
, spot
, 0, spot
);
954 emit_op2(p
, OPCODE_POW
, spot
, 0, spot
, swizzle1(attenuation
, W
));
955 emit_op2(p
, OPCODE_MUL
, att
, 0, slt
, spot
);
957 release_temp(p
, spot
);
958 release_temp(p
, slt
);
961 /* Calculate distance attenuation(See formula (2.4) at glspec 2.1 page 62):
963 * Skip the calucation when _dist_ is undefined(light_eyepos3_is_zero)
965 if (p
->state
->unit
[i
].light_attenuated
&& !is_undef(dist
)) {
969 emit_op1(p
, OPCODE_RCP
, dist
, WRITEMASK_YZ
, dist
);
971 emit_op2(p
, OPCODE_MUL
, dist
, WRITEMASK_XZ
, dist
, swizzle1(dist
,Y
));
973 emit_op2(p
, OPCODE_DP3
, dist
, 0, attenuation
, dist
);
975 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
977 emit_op1(p
, OPCODE_RCP
, dist
, 0, dist
);
978 /* spot-atten * dist-atten */
979 emit_op2(p
, OPCODE_MUL
, att
, 0, dist
, att
);
983 emit_op1(p
, OPCODE_RCP
, att
, 0, dist
);
993 * lit.y = MAX(0, dots.x)
994 * lit.z = SLT(0, dots.x)
996 static void emit_degenerate_lit( struct tnl_program
*p
,
1000 struct ureg id
= get_identity_param(p
); /* id = {0,0,0,1} */
1002 /* Note that lit.x & lit.w will not be examined. Note also that
1003 * dots.xyzw == dots.xxxx.
1006 /* MAX lit, id, dots;
1008 emit_op2(p
, OPCODE_MAX
, lit
, WRITEMASK_XYZW
, id
, dots
);
1010 /* result[2] = (in > 0 ? 1 : 0)
1011 * SLT lit.z, id.z, dots; # lit.z = (0 < dots.z) ? 1 : 0
1013 emit_op2(p
, OPCODE_SLT
, lit
, WRITEMASK_Z
, swizzle1(id
,Z
), dots
);
1017 /* Need to add some addtional parameters to allow lighting in object
1018 * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
1021 static void build_lighting( struct tnl_program
*p
)
1023 const GLboolean twoside
= p
->state
->light_twoside
;
1024 const GLboolean separate
= p
->state
->separate_specular
;
1025 GLuint nr_lights
= 0, count
= 0;
1026 struct ureg normal
= get_transformed_normal(p
);
1027 struct ureg lit
= get_temp(p
);
1028 struct ureg dots
= get_temp(p
);
1029 struct ureg _col0
= undef
, _col1
= undef
;
1030 struct ureg _bfc0
= undef
, _bfc1
= undef
;
1035 * dots.x = dot(normal, VPpli)
1036 * dots.y = dot(normal, halfAngle)
1037 * dots.z = back.shininess
1038 * dots.w = front.shininess
1041 for (i
= 0; i
< MAX_LIGHTS
; i
++)
1042 if (p
->state
->unit
[i
].light_enabled
)
1045 set_material_flags(p
);
1048 if (!p
->state
->material_shininess_is_zero
) {
1049 struct ureg shininess
= get_material(p
, 0, STATE_SHININESS
);
1050 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_W
, swizzle1(shininess
,X
));
1051 release_temp(p
, shininess
);
1054 _col0
= make_temp(p
, get_scenecolor(p
, 0));
1056 _col1
= make_temp(p
, get_identity_param(p
));
1062 if (!p
->state
->material_shininess_is_zero
) {
1063 /* Note that we negate the back-face specular exponent here.
1064 * The negation will be un-done later in the back-face code below.
1066 struct ureg shininess
= get_material(p
, 1, STATE_SHININESS
);
1067 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_Z
,
1068 negate(swizzle1(shininess
,X
)));
1069 release_temp(p
, shininess
);
1072 _bfc0
= make_temp(p
, get_scenecolor(p
, 1));
1074 _bfc1
= make_temp(p
, get_identity_param(p
));
1079 /* If no lights, still need to emit the scenecolor.
1082 struct ureg res0
= register_output( p
, VARYING_SLOT_COL0
);
1083 emit_op1(p
, OPCODE_MOV
, res0
, 0, _col0
);
1087 struct ureg res1
= register_output( p
, VARYING_SLOT_COL1
);
1088 emit_op1(p
, OPCODE_MOV
, res1
, 0, _col1
);
1092 struct ureg res0
= register_output( p
, VARYING_SLOT_BFC0
);
1093 emit_op1(p
, OPCODE_MOV
, res0
, 0, _bfc0
);
1096 if (twoside
&& separate
) {
1097 struct ureg res1
= register_output( p
, VARYING_SLOT_BFC1
);
1098 emit_op1(p
, OPCODE_MOV
, res1
, 0, _bfc1
);
1101 if (nr_lights
== 0) {
1106 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
1107 if (p
->state
->unit
[i
].light_enabled
) {
1108 struct ureg half
= undef
;
1109 struct ureg att
= undef
, VPpli
= undef
;
1110 struct ureg dist
= undef
;
1113 if (p
->state
->unit
[i
].light_eyepos3_is_zero
) {
1114 VPpli
= register_param3(p
, STATE_INTERNAL
,
1115 STATE_LIGHT_POSITION_NORMALIZED
, i
);
1117 struct ureg Ppli
= register_param3(p
, STATE_INTERNAL
,
1118 STATE_LIGHT_POSITION
, i
);
1119 struct ureg V
= get_eye_position(p
);
1121 VPpli
= get_temp(p
);
1124 /* Calculate VPpli vector
1126 emit_op2(p
, OPCODE_SUB
, VPpli
, 0, Ppli
, V
);
1128 /* Normalize VPpli. The dist value also used in
1129 * attenuation below.
1131 emit_op2(p
, OPCODE_DP3
, dist
, 0, VPpli
, VPpli
);
1132 emit_op1(p
, OPCODE_RSQ
, dist
, 0, dist
);
1133 emit_op2(p
, OPCODE_MUL
, VPpli
, 0, VPpli
, dist
);
1136 /* Calculate attenuation:
1138 att
= calculate_light_attenuation(p
, i
, VPpli
, dist
);
1139 release_temp(p
, dist
);
1141 /* Calculate viewer direction, or use infinite viewer:
1143 if (!p
->state
->material_shininess_is_zero
) {
1144 if (p
->state
->light_local_viewer
) {
1145 struct ureg eye_hat
= get_eye_position_normalized(p
);
1147 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1148 emit_normalize_vec3(p
, half
, half
);
1149 } else if (p
->state
->unit
[i
].light_eyepos3_is_zero
) {
1150 half
= register_param3(p
, STATE_INTERNAL
,
1151 STATE_LIGHT_HALF_VECTOR
, i
);
1153 struct ureg z_dir
= swizzle(get_identity_param(p
),X
,Y
,W
,Z
);
1155 emit_op2(p
, OPCODE_ADD
, half
, 0, VPpli
, z_dir
);
1156 emit_normalize_vec3(p
, half
, half
);
1160 /* Calculate dot products:
1162 if (p
->state
->material_shininess_is_zero
) {
1163 emit_op2(p
, OPCODE_DP3
, dots
, 0, normal
, VPpli
);
1166 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_X
, normal
, VPpli
);
1167 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_Y
, normal
, half
);
1170 /* Front face lighting:
1173 struct ureg ambient
= get_lightprod(p
, i
, 0, STATE_AMBIENT
);
1174 struct ureg diffuse
= get_lightprod(p
, i
, 0, STATE_DIFFUSE
);
1175 struct ureg specular
= get_lightprod(p
, i
, 0, STATE_SPECULAR
);
1176 struct ureg res0
, res1
;
1177 GLuint mask0
, mask1
;
1179 if (count
== nr_lights
) {
1181 mask0
= WRITEMASK_XYZ
;
1182 mask1
= WRITEMASK_XYZ
;
1183 res0
= register_output( p
, VARYING_SLOT_COL0
);
1184 res1
= register_output( p
, VARYING_SLOT_COL1
);
1188 mask1
= WRITEMASK_XYZ
;
1190 res1
= register_output( p
, VARYING_SLOT_COL0
);
1200 if (!is_undef(att
)) {
1201 /* light is attenuated by distance */
1202 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1203 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1204 emit_op3(p
, OPCODE_MAD
, _col0
, 0, swizzle1(lit
,X
), ambient
, _col0
);
1206 else if (!p
->state
->material_shininess_is_zero
) {
1207 /* there's a non-zero specular term */
1208 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1209 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1212 /* no attenutation, no specular */
1213 emit_degenerate_lit(p
, lit
, dots
);
1214 emit_op2(p
, OPCODE_ADD
, _col0
, 0, ambient
, _col0
);
1217 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _col0
);
1218 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _col1
);
1220 release_temp(p
, ambient
);
1221 release_temp(p
, diffuse
);
1222 release_temp(p
, specular
);
1225 /* Back face lighting:
1228 struct ureg ambient
= get_lightprod(p
, i
, 1, STATE_AMBIENT
);
1229 struct ureg diffuse
= get_lightprod(p
, i
, 1, STATE_DIFFUSE
);
1230 struct ureg specular
= get_lightprod(p
, i
, 1, STATE_SPECULAR
);
1231 struct ureg res0
, res1
;
1232 GLuint mask0
, mask1
;
1234 if (count
== nr_lights
) {
1236 mask0
= WRITEMASK_XYZ
;
1237 mask1
= WRITEMASK_XYZ
;
1238 res0
= register_output( p
, VARYING_SLOT_BFC0
);
1239 res1
= register_output( p
, VARYING_SLOT_BFC1
);
1243 mask1
= WRITEMASK_XYZ
;
1245 res1
= register_output( p
, VARYING_SLOT_BFC0
);
1255 /* For the back face we need to negate the X and Y component
1256 * dot products. dots.Z has the negated back-face specular
1257 * exponent. We swizzle that into the W position. This
1258 * negation makes the back-face specular term positive again.
1260 dots
= negate(swizzle(dots
,X
,Y
,W
,Z
));
1262 if (!is_undef(att
)) {
1263 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1264 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1265 emit_op3(p
, OPCODE_MAD
, _bfc0
, 0, swizzle1(lit
,X
), ambient
, _bfc0
);
1267 else if (!p
->state
->material_shininess_is_zero
) {
1268 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1269 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
); /**/
1272 emit_degenerate_lit(p
, lit
, dots
);
1273 emit_op2(p
, OPCODE_ADD
, _bfc0
, 0, ambient
, _bfc0
);
1276 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _bfc0
);
1277 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _bfc1
);
1278 /* restore dots to its original state for subsequent lights
1279 * by negating and swizzling again.
1281 dots
= negate(swizzle(dots
,X
,Y
,W
,Z
));
1283 release_temp(p
, ambient
);
1284 release_temp(p
, diffuse
);
1285 release_temp(p
, specular
);
1288 release_temp(p
, half
);
1289 release_temp(p
, VPpli
);
1290 release_temp(p
, att
);
1298 static void build_fog( struct tnl_program
*p
)
1300 struct ureg fog
= register_output(p
, VARYING_SLOT_FOGC
);
1303 switch (p
->state
->fog_distance_mode
) {
1304 case FDM_EYE_RADIAL
: /* Z = sqrt(Xe*Xe + Ye*Ye + Ze*Ze) */
1305 input
= get_eye_position(p
);
1306 emit_op2(p
, OPCODE_DP3
, fog
, WRITEMASK_X
, input
, input
);
1307 emit_op1(p
, OPCODE_RSQ
, fog
, WRITEMASK_X
, fog
);
1308 emit_op1(p
, OPCODE_RCP
, fog
, WRITEMASK_X
, fog
);
1310 case FDM_EYE_PLANE
: /* Z = Ze */
1311 input
= get_eye_position_z(p
);
1312 emit_op1(p
, OPCODE_MOV
, fog
, WRITEMASK_X
, input
);
1314 case FDM_EYE_PLANE_ABS
: /* Z = abs(Ze) */
1315 input
= get_eye_position_z(p
);
1316 emit_op1(p
, OPCODE_ABS
, fog
, WRITEMASK_X
, input
);
1318 case FDM_FROM_ARRAY
:
1319 input
= swizzle1(register_input(p
, VERT_ATTRIB_FOG
), X
);
1320 emit_op1(p
, OPCODE_ABS
, fog
, WRITEMASK_X
, input
);
1323 assert(!"Bad fog mode in build_fog()");
1327 emit_op1(p
, OPCODE_MOV
, fog
, WRITEMASK_YZW
, get_identity_param(p
));
1331 static void build_reflect_texgen( struct tnl_program
*p
,
1335 struct ureg normal
= get_transformed_normal(p
);
1336 struct ureg eye_hat
= get_eye_position_normalized(p
);
1337 struct ureg tmp
= get_temp(p
);
1340 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1342 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1344 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, negate(tmp
), normal
, eye_hat
);
1346 release_temp(p
, tmp
);
1350 static void build_sphere_texgen( struct tnl_program
*p
,
1354 struct ureg normal
= get_transformed_normal(p
);
1355 struct ureg eye_hat
= get_eye_position_normalized(p
);
1356 struct ureg tmp
= get_temp(p
);
1357 struct ureg half
= register_scalar_const(p
, .5);
1358 struct ureg r
= get_temp(p
);
1359 struct ureg inv_m
= get_temp(p
);
1360 struct ureg id
= get_identity_param(p
);
1362 /* Could share the above calculations, but it would be
1363 * a fairly odd state for someone to set (both sphere and
1364 * reflection active for different texture coordinate
1365 * components. Of course - if two texture units enable
1366 * reflect and/or sphere, things start to tilt in favour
1367 * of seperating this out:
1371 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1373 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1375 emit_op3(p
, OPCODE_MAD
, r
, 0, negate(tmp
), normal
, eye_hat
);
1377 emit_op2(p
, OPCODE_ADD
, tmp
, 0, r
, swizzle(id
,X
,Y
,W
,Z
));
1378 /* rx^2 + ry^2 + (rz+1)^2 */
1379 emit_op2(p
, OPCODE_DP3
, tmp
, 0, tmp
, tmp
);
1381 emit_op1(p
, OPCODE_RSQ
, tmp
, 0, tmp
);
1383 emit_op2(p
, OPCODE_MUL
, inv_m
, 0, tmp
, half
);
1385 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, r
, inv_m
, half
);
1387 release_temp(p
, tmp
);
1389 release_temp(p
, inv_m
);
1393 static void build_texture_transform( struct tnl_program
*p
)
1397 for (i
= 0; i
< MAX_TEXTURE_COORD_UNITS
; i
++) {
1399 if (!(p
->state
->fragprog_inputs_read
& VARYING_BIT_TEX(i
)))
1402 if (p
->state
->unit
[i
].coord_replace
)
1405 if (p
->state
->unit
[i
].texgen_enabled
||
1406 p
->state
->unit
[i
].texmat_enabled
) {
1408 GLuint texmat_enabled
= p
->state
->unit
[i
].texmat_enabled
;
1409 struct ureg out
= register_output(p
, VARYING_SLOT_TEX0
+ i
);
1410 struct ureg out_texgen
= undef
;
1412 if (p
->state
->unit
[i
].texgen_enabled
) {
1413 GLuint copy_mask
= 0;
1414 GLuint sphere_mask
= 0;
1415 GLuint reflect_mask
= 0;
1416 GLuint normal_mask
= 0;
1420 out_texgen
= get_temp(p
);
1424 modes
[0] = p
->state
->unit
[i
].texgen_mode0
;
1425 modes
[1] = p
->state
->unit
[i
].texgen_mode1
;
1426 modes
[2] = p
->state
->unit
[i
].texgen_mode2
;
1427 modes
[3] = p
->state
->unit
[i
].texgen_mode3
;
1429 for (j
= 0; j
< 4; j
++) {
1431 case TXG_OBJ_LINEAR
: {
1432 struct ureg obj
= register_input(p
, VERT_ATTRIB_POS
);
1434 register_param3(p
, STATE_TEXGEN
, i
,
1435 STATE_TEXGEN_OBJECT_S
+ j
);
1437 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1441 case TXG_EYE_LINEAR
: {
1442 struct ureg eye
= get_eye_position(p
);
1444 register_param3(p
, STATE_TEXGEN
, i
,
1445 STATE_TEXGEN_EYE_S
+ j
);
1447 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1451 case TXG_SPHERE_MAP
:
1452 sphere_mask
|= WRITEMASK_X
<< j
;
1454 case TXG_REFLECTION_MAP
:
1455 reflect_mask
|= WRITEMASK_X
<< j
;
1457 case TXG_NORMAL_MAP
:
1458 normal_mask
|= WRITEMASK_X
<< j
;
1461 copy_mask
|= WRITEMASK_X
<< j
;
1466 build_sphere_texgen(p
, out_texgen
, sphere_mask
);
1470 build_reflect_texgen(p
, out_texgen
, reflect_mask
);
1474 struct ureg normal
= get_transformed_normal(p
);
1475 emit_op1(p
, OPCODE_MOV
, out_texgen
, normal_mask
, normal
);
1479 struct ureg in
= register_input(p
, VERT_ATTRIB_TEX0
+i
);
1480 emit_op1(p
, OPCODE_MOV
, out_texgen
, copy_mask
, in
);
1484 if (texmat_enabled
) {
1485 struct ureg texmat
[4];
1486 struct ureg in
= (!is_undef(out_texgen
) ?
1488 register_input(p
, VERT_ATTRIB_TEX0
+i
));
1489 if (p
->mvp_with_dp4
) {
1490 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1492 emit_matrix_transform_vec4( p
, out
, texmat
, in
);
1495 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1496 STATE_MATRIX_TRANSPOSE
, texmat
);
1497 emit_transpose_matrix_transform_vec4( p
, out
, texmat
, in
);
1504 emit_passthrough(p
, VERT_ATTRIB_TEX0
+i
, VARYING_SLOT_TEX0
+i
);
1511 * Point size attenuation computation.
1513 static void build_atten_pointsize( struct tnl_program
*p
)
1515 struct ureg eye
= get_eye_position_z(p
);
1516 struct ureg state_size
= register_param2(p
, STATE_INTERNAL
, STATE_POINT_SIZE_CLAMPED
);
1517 struct ureg state_attenuation
= register_param1(p
, STATE_POINT_ATTENUATION
);
1518 struct ureg out
= register_output(p
, VARYING_SLOT_PSIZ
);
1519 struct ureg ut
= get_temp(p
);
1522 emit_op1(p
, OPCODE_ABS
, ut
, WRITEMASK_Y
, swizzle1(eye
, Z
));
1523 /* p1 + dist * (p2 + dist * p3); */
1524 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1525 swizzle1(state_attenuation
, Z
), swizzle1(state_attenuation
, Y
));
1526 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1527 ut
, swizzle1(state_attenuation
, X
));
1529 /* 1 / sqrt(factor) */
1530 emit_op1(p
, OPCODE_RSQ
, ut
, WRITEMASK_X
, ut
);
1533 /* out = pointSize / sqrt(factor) */
1534 emit_op2(p
, OPCODE_MUL
, out
, WRITEMASK_X
, ut
, state_size
);
1536 /* this is a good place to clamp the point size since there's likely
1537 * no hardware registers to clamp point size at rasterization time.
1539 emit_op2(p
, OPCODE_MUL
, ut
, WRITEMASK_X
, ut
, state_size
);
1540 emit_op2(p
, OPCODE_MAX
, ut
, WRITEMASK_X
, ut
, swizzle1(state_size
, Y
));
1541 emit_op2(p
, OPCODE_MIN
, out
, WRITEMASK_X
, ut
, swizzle1(state_size
, Z
));
1544 release_temp(p
, ut
);
1549 * Pass-though per-vertex point size, from user's point size array.
1551 static void build_array_pointsize( struct tnl_program
*p
)
1553 struct ureg in
= register_input(p
, VERT_ATTRIB_POINT_SIZE
);
1554 struct ureg out
= register_output(p
, VARYING_SLOT_PSIZ
);
1555 emit_op1(p
, OPCODE_MOV
, out
, WRITEMASK_X
, in
);
1559 static void build_tnl_program( struct tnl_program
*p
)
1561 /* Emit the program, starting with the modelview, projection transforms:
1565 /* Lighting calculations:
1567 if (p
->state
->fragprog_inputs_read
& (VARYING_BIT_COL0
|VARYING_BIT_COL1
)) {
1568 if (p
->state
->light_global_enabled
)
1571 if (p
->state
->fragprog_inputs_read
& VARYING_BIT_COL0
)
1572 emit_passthrough(p
, VERT_ATTRIB_COLOR0
, VARYING_SLOT_COL0
);
1574 if (p
->state
->fragprog_inputs_read
& VARYING_BIT_COL1
)
1575 emit_passthrough(p
, VERT_ATTRIB_COLOR1
, VARYING_SLOT_COL1
);
1579 if (p
->state
->fragprog_inputs_read
& VARYING_BIT_FOGC
)
1582 if (p
->state
->fragprog_inputs_read
& VARYING_BITS_TEX_ANY
)
1583 build_texture_transform(p
);
1585 if (p
->state
->point_attenuated
)
1586 build_atten_pointsize(p
);
1587 else if (p
->state
->varying_vp_inputs
& VERT_BIT_POINT_SIZE
)
1588 build_array_pointsize(p
);
1592 emit_op1(p
, OPCODE_END
, undef
, 0, undef
);
1603 create_new_program( const struct state_key
*key
,
1604 struct gl_program
*program
,
1605 GLboolean mvp_with_dp4
,
1608 struct tnl_program p
;
1610 memset(&p
, 0, sizeof(p
));
1612 p
.program
= program
;
1613 p
.eye_position
= undef
;
1614 p
.eye_position_z
= undef
;
1615 p
.eye_position_normalized
= undef
;
1616 p
.transformed_normal
= undef
;
1619 p
.mvp_with_dp4
= mvp_with_dp4
;
1621 if (max_temps
>= sizeof(int) * 8)
1622 p
.temp_reserved
= 0;
1624 p
.temp_reserved
= ~((1<<max_temps
)-1);
1626 /* Start by allocating 32 instructions.
1627 * If we need more, we'll grow the instruction array as needed.
1630 p
.program
->arb
.Instructions
=
1631 rzalloc_array(program
, struct prog_instruction
, p
.max_inst
);
1632 p
.program
->String
= NULL
;
1633 p
.program
->arb
.NumInstructions
=
1634 p
.program
->arb
.NumTemporaries
=
1635 p
.program
->arb
.NumParameters
=
1636 p
.program
->arb
.NumAttributes
= p
.program
->arb
.NumAddressRegs
= 0;
1637 p
.program
->Parameters
= _mesa_new_parameter_list();
1638 p
.program
->info
.inputs_read
= 0;
1639 p
.program
->info
.outputs_written
= 0;
1641 build_tnl_program( &p
);
1646 * Return a vertex program which implements the current fixed-function
1647 * transform/lighting/texgen operations.
1650 _mesa_get_fixed_func_vertex_program(struct gl_context
*ctx
)
1652 struct gl_program
*prog
;
1653 struct state_key key
;
1655 /* Grab all the relevant state and put it in a single structure:
1657 make_state_key(ctx
, &key
);
1659 /* Look for an already-prepared program for this state:
1661 prog
= _mesa_search_program_cache(ctx
->VertexProgram
.Cache
, &key
,
1665 /* OK, we'll have to build a new one */
1667 printf("Build new TNL program\n");
1669 prog
= ctx
->Driver
.NewProgram(ctx
, GL_VERTEX_PROGRAM_ARB
, 0, true);
1673 create_new_program( &key
, prog
,
1674 ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].OptimizeForAOS
,
1675 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTemps
);
1677 if (ctx
->Driver
.ProgramStringNotify
)
1678 ctx
->Driver
.ProgramStringNotify(ctx
, GL_VERTEX_PROGRAM_ARB
, prog
);
1680 _mesa_program_cache_insert(ctx
, ctx
->VertexProgram
.Cache
, &key
,