2 * Mesa 3-D graphics library
5 * Copyright (C) 2007 Tungsten Graphics All Rights Reserved.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * TUNGSTEN GRAPHICS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
22 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
23 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
28 * Create a vertex program to execute the current fixed function T&L pipeline.
29 * \author Keith Whitwell
36 #include "shader/program.h"
37 #include "shader/prog_instruction.h"
38 #include "shader/prog_parameter.h"
39 #include "shader/prog_print.h"
40 #include "shader/prog_statevars.h"
41 #include "t_context.h" /* NOTE: very light dependency on this */
42 #include "t_vp_build.h"
46 unsigned light_global_enabled
:1;
47 unsigned light_local_viewer
:1;
48 unsigned light_twoside
:1;
49 unsigned light_color_material
:1;
50 unsigned light_color_material_mask
:12;
51 unsigned light_material_mask
:12;
54 unsigned rescale_normals
:1;
55 unsigned fog_source_is_depth
:1;
56 unsigned tnl_do_vertex_fog
:1;
57 unsigned separate_specular
:1;
59 unsigned point_attenuated
:1;
60 unsigned texture_enabled_global
:1;
61 unsigned fragprog_inputs_read
:12;
64 unsigned light_enabled
:1;
65 unsigned light_eyepos3_is_zero
:1;
66 unsigned light_spotcutoff_is_180
:1;
67 unsigned light_attenuated
:1;
68 unsigned texunit_really_enabled
:1;
69 unsigned texmat_enabled
:1;
70 unsigned texgen_enabled
:4;
71 unsigned texgen_mode0
:4;
72 unsigned texgen_mode1
:4;
73 unsigned texgen_mode2
:4;
74 unsigned texgen_mode3
:4;
85 static GLuint
translate_fog_mode( GLenum mode
)
88 case GL_LINEAR
: return FOG_LINEAR
;
89 case GL_EXP
: return FOG_EXP
;
90 case GL_EXP2
: return FOG_EXP2
;
91 default: return FOG_NONE
;
96 #define TXG_OBJ_LINEAR 1
97 #define TXG_EYE_LINEAR 2
98 #define TXG_SPHERE_MAP 3
99 #define TXG_REFLECTION_MAP 4
100 #define TXG_NORMAL_MAP 5
102 static GLuint
translate_texgen( GLboolean enabled
, GLenum mode
)
108 case GL_OBJECT_LINEAR
: return TXG_OBJ_LINEAR
;
109 case GL_EYE_LINEAR
: return TXG_EYE_LINEAR
;
110 case GL_SPHERE_MAP
: return TXG_SPHERE_MAP
;
111 case GL_REFLECTION_MAP_NV
: return TXG_REFLECTION_MAP
;
112 case GL_NORMAL_MAP_NV
: return TXG_NORMAL_MAP
;
113 default: return TXG_NONE
;
117 static struct state_key
*make_state_key( GLcontext
*ctx
)
119 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
120 struct vertex_buffer
*VB
= &tnl
->vb
;
121 const struct gl_fragment_program
*fp
= ctx
->FragmentProgram
._Current
;
122 struct state_key
*key
= CALLOC_STRUCT(state_key
);
125 /* This now relies on texenvprogram.c being active:
129 key
->fragprog_inputs_read
= fp
->Base
.InputsRead
;
131 key
->separate_specular
= (ctx
->Light
.Model
.ColorControl
==
132 GL_SEPARATE_SPECULAR_COLOR
);
134 if (ctx
->Light
.Enabled
) {
135 key
->light_global_enabled
= 1;
137 if (ctx
->Light
.Model
.LocalViewer
)
138 key
->light_local_viewer
= 1;
140 if (ctx
->Light
.Model
.TwoSide
)
141 key
->light_twoside
= 1;
143 if (ctx
->Light
.ColorMaterialEnabled
) {
144 key
->light_color_material
= 1;
145 key
->light_color_material_mask
= ctx
->Light
.ColorMaterialBitmask
;
148 for (i
= _TNL_FIRST_MAT
; i
<= _TNL_LAST_MAT
; i
++)
149 if (VB
->AttribPtr
[i
]->stride
)
150 key
->light_material_mask
|= 1<<(i
-_TNL_ATTRIB_MAT_FRONT_AMBIENT
);
152 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
153 struct gl_light
*light
= &ctx
->Light
.Light
[i
];
155 if (light
->Enabled
) {
156 key
->unit
[i
].light_enabled
= 1;
158 if (light
->EyePosition
[3] == 0.0)
159 key
->unit
[i
].light_eyepos3_is_zero
= 1;
161 if (light
->SpotCutoff
== 180.0)
162 key
->unit
[i
].light_spotcutoff_is_180
= 1;
164 if (light
->ConstantAttenuation
!= 1.0 ||
165 light
->LinearAttenuation
!= 0.0 ||
166 light
->QuadraticAttenuation
!= 0.0)
167 key
->unit
[i
].light_attenuated
= 1;
172 if (ctx
->Transform
.Normalize
)
175 if (ctx
->Transform
.RescaleNormals
)
176 key
->rescale_normals
= 1;
178 key
->fog_mode
= translate_fog_mode(fp
->FogOption
);
180 if (ctx
->Fog
.FogCoordinateSource
== GL_FRAGMENT_DEPTH_EXT
)
181 key
->fog_source_is_depth
= 1;
183 if (tnl
->_DoVertexFog
)
184 key
->tnl_do_vertex_fog
= 1;
186 if (ctx
->Point
._Attenuated
)
187 key
->point_attenuated
= 1;
189 if (ctx
->Texture
._TexGenEnabled
||
190 ctx
->Texture
._TexMatEnabled
||
191 ctx
->Texture
._EnabledUnits
)
192 key
->texture_enabled_global
= 1;
194 for (i
= 0; i
< MAX_TEXTURE_UNITS
; i
++) {
195 struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
197 if (texUnit
->_ReallyEnabled
)
198 key
->unit
[i
].texunit_really_enabled
= 1;
200 if (ctx
->Texture
._TexMatEnabled
& ENABLE_TEXMAT(i
))
201 key
->unit
[i
].texmat_enabled
= 1;
203 if (texUnit
->TexGenEnabled
) {
204 key
->unit
[i
].texgen_enabled
= 1;
206 key
->unit
[i
].texgen_mode0
=
207 translate_texgen( texUnit
->TexGenEnabled
& (1<<0),
209 key
->unit
[i
].texgen_mode1
=
210 translate_texgen( texUnit
->TexGenEnabled
& (1<<1),
212 key
->unit
[i
].texgen_mode2
=
213 translate_texgen( texUnit
->TexGenEnabled
& (1<<2),
215 key
->unit
[i
].texgen_mode3
=
216 translate_texgen( texUnit
->TexGenEnabled
& (1<<3),
226 /* Very useful debugging tool - produces annotated listing of
227 * generated program with line/function references for each
228 * instruction back into this file:
230 #define DISASSEM (MESA_VERBOSE&VERBOSE_DISASSEM)
232 /* Should be tunable by the driver - do we want to do matrix
233 * multiplications with DP4's or with MUL/MAD's? SSE works better
234 * with the latter, drivers may differ.
240 /* Use uregs to represent registers internally, translate to Mesa's
241 * expected formats on emit.
243 * NOTE: These are passed by value extensively in this file rather
244 * than as usual by pointer reference. If this disturbs you, try
245 * remembering they are just 32bits in size.
247 * GCC is smart enough to deal with these dword-sized structures in
248 * much the same way as if I had defined them as dwords and was using
249 * macros to access and set the fields. This is much nicer and easier
254 GLint idx
:8; /* relative addressing may be negative */
262 const struct state_key
*state
;
263 struct gl_vertex_program
*program
;
266 GLuint temp_reserved
;
268 struct ureg eye_position
;
269 struct ureg eye_position_normalized
;
270 struct ureg eye_normal
;
271 struct ureg identity
;
274 GLuint color_materials
;
278 static const struct ureg undef
= {
296 static struct ureg
make_ureg(GLuint file
, GLint idx
)
302 reg
.swz
= SWIZZLE_NOOP
;
309 static struct ureg
negate( struct ureg reg
)
316 static struct ureg
swizzle( struct ureg reg
, int x
, int y
, int z
, int w
)
318 reg
.swz
= MAKE_SWIZZLE4(GET_SWZ(reg
.swz
, x
),
321 GET_SWZ(reg
.swz
, w
));
326 static struct ureg
swizzle1( struct ureg reg
, int x
)
328 return swizzle(reg
, x
, x
, x
, x
);
331 static struct ureg
get_temp( struct tnl_program
*p
)
333 int bit
= _mesa_ffs( ~p
->temp_in_use
);
335 _mesa_problem(NULL
, "%s: out of temporaries\n", __FILE__
);
339 if ((GLuint
) bit
> p
->program
->Base
.NumTemporaries
)
340 p
->program
->Base
.NumTemporaries
= bit
;
342 p
->temp_in_use
|= 1<<(bit
-1);
343 return make_ureg(PROGRAM_TEMPORARY
, bit
-1);
346 static struct ureg
reserve_temp( struct tnl_program
*p
)
348 struct ureg temp
= get_temp( p
);
349 p
->temp_reserved
|= 1<<temp
.idx
;
353 static void release_temp( struct tnl_program
*p
, struct ureg reg
)
355 if (reg
.file
== PROGRAM_TEMPORARY
) {
356 p
->temp_in_use
&= ~(1<<reg
.idx
);
357 p
->temp_in_use
|= p
->temp_reserved
; /* can't release reserved temps */
361 static void release_temps( struct tnl_program
*p
)
363 p
->temp_in_use
= p
->temp_reserved
;
368 static struct ureg
register_input( struct tnl_program
*p
, GLuint input
)
370 p
->program
->Base
.InputsRead
|= (1<<input
);
371 return make_ureg(PROGRAM_INPUT
, input
);
374 static struct ureg
register_output( struct tnl_program
*p
, GLuint output
)
376 p
->program
->Base
.OutputsWritten
|= (1<<output
);
377 return make_ureg(PROGRAM_OUTPUT
, output
);
380 static struct ureg
register_const4f( struct tnl_program
*p
,
393 idx
= _mesa_add_unnamed_constant( p
->program
->Base
.Parameters
, values
, 4,
395 ASSERT(swizzle
== SWIZZLE_NOOP
);
396 return make_ureg(PROGRAM_STATE_VAR
, idx
);
399 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
400 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
401 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
402 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
404 static GLboolean
is_undef( struct ureg reg
)
406 return reg
.file
== PROGRAM_UNDEFINED
;
409 static struct ureg
get_identity_param( struct tnl_program
*p
)
411 if (is_undef(p
->identity
))
412 p
->identity
= register_const4f(p
, 0,0,0,1);
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
->Base
.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)
442 static void register_matrix_param5( struct tnl_program
*p
,
443 GLint s0
, /* modelview, projection, etc */
444 GLint s1
, /* texture matrix number */
445 GLint s2
, /* first row */
446 GLint s3
, /* last row */
447 GLint s4
, /* inverse, transpose, etc */
448 struct ureg
*matrix
)
452 /* This is a bit sad as the support is there to pull the whole
453 * matrix out in one go:
455 for (i
= 0; i
<= s3
- s2
; i
++)
456 matrix
[i
] = register_param5( p
, s0
, s1
, i
, i
, s4
);
461 * Convert a ureg source register to a prog_src_register.
463 static void emit_arg( struct prog_src_register
*src
,
466 assert(reg
.file
!= PROGRAM_OUTPUT
);
467 src
->File
= reg
.file
;
468 src
->Index
= reg
.idx
;
469 src
->Swizzle
= reg
.swz
;
470 src
->NegateBase
= reg
.negate
? NEGATE_XYZW
: 0;
477 * Convert a ureg dest register to a prog_dst_register.
479 static void emit_dst( struct prog_dst_register
*dst
,
480 struct ureg reg
, GLuint mask
)
482 /* Check for legal output register type. UNDEFINED will occur in
483 * instruction that don't produce a result (like END).
485 assert(reg
.file
== PROGRAM_TEMPORARY
||
486 reg
.file
== PROGRAM_OUTPUT
||
487 reg
.file
== PROGRAM_UNDEFINED
);
488 dst
->File
= reg
.file
;
489 dst
->Index
= reg
.idx
;
490 /* allow zero as a shorthand for xyzw */
491 dst
->WriteMask
= mask
? mask
: WRITEMASK_XYZW
;
492 dst
->CondMask
= COND_TR
; /* always pass cond test */
493 dst
->CondSwizzle
= SWIZZLE_NOOP
;
498 static void debug_insn( struct prog_instruction
*inst
, const char *fn
,
502 static const char *last_fn
;
506 _mesa_printf("%s:\n", fn
);
509 _mesa_printf("%d:\t", line
);
510 _mesa_print_instruction(inst
);
515 static void emit_op3fn(struct tnl_program
*p
,
525 GLuint nr
= p
->program
->Base
.NumInstructions
++;
526 struct prog_instruction
*inst
= &p
->program
->Base
.Instructions
[nr
];
528 if (p
->program
->Base
.NumInstructions
> MAX_INSN
) {
529 _mesa_problem(0, "Out of instructions in emit_op3fn\n");
533 inst
->Opcode
= (enum prog_opcode
) op
;
537 emit_arg( &inst
->SrcReg
[0], src0
);
538 emit_arg( &inst
->SrcReg
[1], src1
);
539 emit_arg( &inst
->SrcReg
[2], src2
);
541 emit_dst( &inst
->DstReg
, dest
, mask
);
543 debug_insn(inst
, fn
, line
);
547 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
548 emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
550 #define emit_op2(p, op, dst, mask, src0, src1) \
551 emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
553 #define emit_op1(p, op, dst, mask, src0) \
554 emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
557 static struct ureg
make_temp( struct tnl_program
*p
, struct ureg reg
)
559 if (reg
.file
== PROGRAM_TEMPORARY
&&
560 !(p
->temp_reserved
& (1<<reg
.idx
)))
563 struct ureg temp
= get_temp(p
);
564 emit_op1(p
, OPCODE_MOV
, temp
, 0, reg
);
570 /* Currently no tracking performed of input/output/register size or
571 * active elements. Could be used to reduce these operations, as
572 * could the matrix type.
574 static void emit_matrix_transform_vec4( struct tnl_program
*p
,
576 const struct ureg
*mat
,
579 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_X
, src
, mat
[0]);
580 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Y
, src
, mat
[1]);
581 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_Z
, src
, mat
[2]);
582 emit_op2(p
, OPCODE_DP4
, dest
, WRITEMASK_W
, src
, mat
[3]);
585 /* This version is much easier to implement if writemasks are not
586 * supported natively on the target or (like SSE), the target doesn't
587 * have a clean/obvious dotproduct implementation.
589 static void emit_transpose_matrix_transform_vec4( struct tnl_program
*p
,
591 const struct ureg
*mat
,
596 if (dest
.file
!= PROGRAM_TEMPORARY
)
601 emit_op2(p
, OPCODE_MUL
, tmp
, 0, swizzle1(src
,X
), mat
[0]);
602 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Y
), mat
[1], tmp
);
603 emit_op3(p
, OPCODE_MAD
, tmp
, 0, swizzle1(src
,Z
), mat
[2], tmp
);
604 emit_op3(p
, OPCODE_MAD
, dest
, 0, swizzle1(src
,W
), mat
[3], tmp
);
606 if (dest
.file
!= PROGRAM_TEMPORARY
)
607 release_temp(p
, tmp
);
610 static void emit_matrix_transform_vec3( struct tnl_program
*p
,
612 const struct ureg
*mat
,
615 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_X
, src
, mat
[0]);
616 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Y
, src
, mat
[1]);
617 emit_op2(p
, OPCODE_DP3
, dest
, WRITEMASK_Z
, src
, mat
[2]);
621 static void emit_normalize_vec3( struct tnl_program
*p
,
625 struct ureg tmp
= get_temp(p
);
626 emit_op2(p
, OPCODE_DP3
, tmp
, 0, src
, src
);
627 emit_op1(p
, OPCODE_RSQ
, tmp
, 0, tmp
);
628 emit_op2(p
, OPCODE_MUL
, dest
, 0, src
, tmp
);
629 release_temp(p
, tmp
);
632 static void emit_passthrough( struct tnl_program
*p
,
636 struct ureg out
= register_output(p
, output
);
637 emit_op1(p
, OPCODE_MOV
, out
, 0, register_input(p
, input
));
640 static struct ureg
get_eye_position( struct tnl_program
*p
)
642 if (is_undef(p
->eye_position
)) {
643 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
644 struct ureg modelview
[4];
646 p
->eye_position
= reserve_temp(p
);
649 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
652 emit_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
655 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 3,
656 STATE_MATRIX_TRANSPOSE
, modelview
);
658 emit_transpose_matrix_transform_vec4(p
, p
->eye_position
, modelview
, pos
);
662 return p
->eye_position
;
666 static struct ureg
get_eye_position_normalized( struct tnl_program
*p
)
668 if (is_undef(p
->eye_position_normalized
)) {
669 struct ureg eye
= get_eye_position(p
);
670 p
->eye_position_normalized
= reserve_temp(p
);
671 emit_normalize_vec3(p
, p
->eye_position_normalized
, eye
);
674 return p
->eye_position_normalized
;
678 static struct ureg
get_eye_normal( struct tnl_program
*p
)
680 if (is_undef(p
->eye_normal
)) {
681 struct ureg normal
= register_input(p
, VERT_ATTRIB_NORMAL
);
682 struct ureg mvinv
[3];
684 register_matrix_param5( p
, STATE_MODELVIEW_MATRIX
, 0, 0, 2,
685 STATE_MATRIX_INVTRANS
, mvinv
);
687 p
->eye_normal
= reserve_temp(p
);
689 /* Transform to eye space:
691 emit_matrix_transform_vec3( p
, p
->eye_normal
, mvinv
, normal
);
693 /* Normalize/Rescale:
695 if (p
->state
->normalize
) {
696 emit_normalize_vec3( p
, p
->eye_normal
, p
->eye_normal
);
698 else if (p
->state
->rescale_normals
) {
699 struct ureg rescale
= register_param2(p
, STATE_INTERNAL
,
702 emit_op2( p
, OPCODE_MUL
, p
->eye_normal
, 0, p
->eye_normal
,
703 swizzle1(rescale
, X
));
707 return p
->eye_normal
;
712 static void build_hpos( struct tnl_program
*p
)
714 struct ureg pos
= register_input( p
, VERT_ATTRIB_POS
);
715 struct ureg hpos
= register_output( p
, VERT_RESULT_HPOS
);
719 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
721 emit_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
724 register_matrix_param5( p
, STATE_MVP_MATRIX
, 0, 0, 3,
725 STATE_MATRIX_TRANSPOSE
, mvp
);
726 emit_transpose_matrix_transform_vec4( p
, hpos
, mvp
, pos
);
731 static GLuint
material_attrib( GLuint side
, GLuint property
)
733 return ((property
- STATE_AMBIENT
) * 2 +
737 /* Get a bitmask of which material values vary on a per-vertex basis.
739 static void set_material_flags( struct tnl_program
*p
)
741 p
->color_materials
= 0;
744 if (p
->state
->light_color_material
) {
746 p
->color_materials
= p
->state
->light_color_material_mask
;
749 p
->materials
|= p
->state
->light_material_mask
;
753 static struct ureg
get_material( struct tnl_program
*p
, GLuint side
,
756 GLuint attrib
= material_attrib(side
, property
);
758 if (p
->color_materials
& (1<<attrib
))
759 return register_input(p
, VERT_ATTRIB_COLOR0
);
760 else if (p
->materials
& (1<<attrib
))
761 return register_input( p
, attrib
+ _TNL_ATTRIB_MAT_FRONT_AMBIENT
);
763 return register_param3( p
, STATE_MATERIAL
, side
, property
);
766 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
767 MAT_BIT_FRONT_AMBIENT | \
768 MAT_BIT_FRONT_DIFFUSE) << (side))
770 /* Either return a precalculated constant value or emit code to
771 * calculate these values dynamically in the case where material calls
772 * are present between begin/end pairs.
774 * Probably want to shift this to the program compilation phase - if
775 * we always emitted the calculation here, a smart compiler could
776 * detect that it was constant (given a certain set of inputs), and
777 * lift it out of the main loop. That way the programs created here
778 * would be independent of the vertex_buffer details.
780 static struct ureg
get_scenecolor( struct tnl_program
*p
, GLuint side
)
782 if (p
->materials
& SCENE_COLOR_BITS(side
)) {
783 struct ureg lm_ambient
= register_param1(p
, STATE_LIGHTMODEL_AMBIENT
);
784 struct ureg material_emission
= get_material(p
, side
, STATE_EMISSION
);
785 struct ureg material_ambient
= get_material(p
, side
, STATE_AMBIENT
);
786 struct ureg material_diffuse
= get_material(p
, side
, STATE_DIFFUSE
);
787 struct ureg tmp
= make_temp(p
, material_diffuse
);
788 emit_op3(p
, OPCODE_MAD
, tmp
, WRITEMASK_XYZ
, lm_ambient
,
789 material_ambient
, material_emission
);
793 return register_param2( p
, STATE_LIGHTMODEL_SCENECOLOR
, side
);
797 static struct ureg
get_lightprod( struct tnl_program
*p
, GLuint light
,
798 GLuint side
, GLuint property
)
800 GLuint attrib
= material_attrib(side
, property
);
801 if (p
->materials
& (1<<attrib
)) {
802 struct ureg light_value
=
803 register_param3(p
, STATE_LIGHT
, light
, property
);
804 struct ureg material_value
= get_material(p
, side
, property
);
805 struct ureg tmp
= get_temp(p
);
806 emit_op2(p
, OPCODE_MUL
, tmp
, 0, light_value
, material_value
);
810 return register_param4(p
, STATE_LIGHTPROD
, light
, side
, property
);
813 static struct ureg
calculate_light_attenuation( struct tnl_program
*p
,
818 struct ureg attenuation
= register_param3(p
, STATE_LIGHT
, i
,
820 struct ureg att
= get_temp(p
);
822 /* Calculate spot attenuation:
824 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
825 struct ureg spot_dir_norm
= register_param3(p
, STATE_INTERNAL
,
826 STATE_SPOT_DIR_NORMALIZED
, i
);
827 struct ureg spot
= get_temp(p
);
828 struct ureg slt
= get_temp(p
);
830 emit_op2(p
, OPCODE_DP3
, spot
, 0, negate(VPpli
), spot_dir_norm
);
831 emit_op2(p
, OPCODE_SLT
, slt
, 0, swizzle1(spot_dir_norm
,W
), spot
);
832 emit_op2(p
, OPCODE_POW
, spot
, 0, spot
, swizzle1(attenuation
, W
));
833 emit_op2(p
, OPCODE_MUL
, att
, 0, slt
, spot
);
835 release_temp(p
, spot
);
836 release_temp(p
, slt
);
839 /* Calculate distance attenuation:
841 if (p
->state
->unit
[i
].light_attenuated
) {
844 emit_op1(p
, OPCODE_RCP
, dist
, WRITEMASK_YZ
, dist
);
846 emit_op2(p
, OPCODE_MUL
, dist
, WRITEMASK_XZ
, dist
, swizzle1(dist
,Y
));
848 emit_op2(p
, OPCODE_DP3
, dist
, 0, attenuation
, dist
);
850 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
) {
852 emit_op1(p
, OPCODE_RCP
, dist
, 0, dist
);
853 /* spot-atten * dist-atten */
854 emit_op2(p
, OPCODE_MUL
, att
, 0, dist
, att
);
857 emit_op1(p
, OPCODE_RCP
, att
, 0, dist
);
868 /* Need to add some addtional parameters to allow lighting in object
869 * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
872 static void build_lighting( struct tnl_program
*p
)
874 const GLboolean twoside
= p
->state
->light_twoside
;
875 const GLboolean separate
= p
->state
->separate_specular
;
876 GLuint nr_lights
= 0, count
= 0;
877 struct ureg normal
= get_eye_normal(p
);
878 struct ureg lit
= get_temp(p
);
879 struct ureg dots
= get_temp(p
);
880 struct ureg _col0
= undef
, _col1
= undef
;
881 struct ureg _bfc0
= undef
, _bfc1
= undef
;
884 for (i
= 0; i
< MAX_LIGHTS
; i
++)
885 if (p
->state
->unit
[i
].light_enabled
)
888 set_material_flags(p
);
891 struct ureg shininess
= get_material(p
, 0, STATE_SHININESS
);
892 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_W
, swizzle1(shininess
,X
));
894 _col0
= make_temp(p
, get_scenecolor(p
, 0));
896 _col1
= make_temp(p
, get_identity_param(p
));
903 struct ureg shininess
= get_material(p
, 1, STATE_SHININESS
);
904 emit_op1(p
, OPCODE_MOV
, dots
, WRITEMASK_Z
,
905 negate(swizzle1(shininess
,X
)));
907 _bfc0
= make_temp(p
, get_scenecolor(p
, 1));
909 _bfc1
= make_temp(p
, get_identity_param(p
));
915 /* If no lights, still need to emit the scenecolor.
918 struct ureg res0
= register_output( p
, VERT_RESULT_COL0
);
919 emit_op1(p
, OPCODE_MOV
, res0
, 0, _col0
);
923 struct ureg res1
= register_output( p
, VERT_RESULT_COL1
);
924 emit_op1(p
, OPCODE_MOV
, res1
, 0, _col1
);
928 struct ureg res0
= register_output( p
, VERT_RESULT_BFC0
);
929 emit_op1(p
, OPCODE_MOV
, res0
, 0, _bfc0
);
932 if (twoside
&& separate
) {
933 struct ureg res1
= register_output( p
, VERT_RESULT_BFC1
);
934 emit_op1(p
, OPCODE_MOV
, res1
, 0, _bfc1
);
937 if (nr_lights
== 0) {
943 for (i
= 0; i
< MAX_LIGHTS
; i
++) {
944 if (p
->state
->unit
[i
].light_enabled
) {
945 struct ureg half
= undef
;
946 struct ureg att
= undef
, VPpli
= undef
;
950 if (p
->state
->unit
[i
].light_eyepos3_is_zero
) {
951 /* Can used precomputed constants in this case.
952 * Attenuation never applies to infinite lights.
954 VPpli
= register_param3(p
, STATE_LIGHT
, i
,
955 STATE_POSITION_NORMALIZED
);
956 if (p
->state
->light_local_viewer
) {
957 struct ureg eye_hat
= get_eye_position_normalized(p
);
959 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
960 emit_normalize_vec3(p
, half
, half
);
962 half
= register_param3(p
, STATE_LIGHT
, i
, STATE_HALF_VECTOR
);
966 struct ureg Ppli
= register_param3(p
, STATE_LIGHT
, i
,
968 struct ureg V
= get_eye_position(p
);
969 struct ureg dist
= get_temp(p
);
970 struct ureg tmpPpli
= get_temp(p
);
974 /* In homogeneous object coordinates
976 emit_op1(p
, OPCODE_RCP
, dist
, 0, swizzle1(Ppli
, W
));
977 emit_op2(p
, OPCODE_MUL
, tmpPpli
, 0, Ppli
, dist
);
979 /* Calculate VPpli vector
981 emit_op2(p
, OPCODE_SUB
, VPpli
, 0, tmpPpli
, V
);
983 /* we're done with tmpPpli now */
984 release_temp(p
, tmpPpli
);
986 /* Normalize VPpli. The dist value also used in
989 emit_op2(p
, OPCODE_DP3
, dist
, 0, VPpli
, VPpli
);
990 emit_op1(p
, OPCODE_RSQ
, dist
, 0, dist
);
991 emit_op2(p
, OPCODE_MUL
, VPpli
, 0, VPpli
, dist
);
994 /* Calculate attenuation:
996 if (!p
->state
->unit
[i
].light_spotcutoff_is_180
||
997 p
->state
->unit
[i
].light_attenuated
) {
998 att
= calculate_light_attenuation(p
, i
, VPpli
, dist
);
1001 /* We're done with dist now */
1002 release_temp(p
, dist
);
1005 /* Calculate viewer direction, or use infinite viewer:
1008 if (p
->state
->light_local_viewer
) {
1009 struct ureg eye_hat
= get_eye_position_normalized(p
);
1010 emit_op2(p
, OPCODE_SUB
, half
, 0, VPpli
, eye_hat
);
1013 struct ureg z_dir
= swizzle(get_identity_param(p
),X
,Y
,W
,Z
);
1014 emit_op2(p
, OPCODE_ADD
, half
, 0, VPpli
, z_dir
);
1017 emit_normalize_vec3(p
, half
, half
);
1020 /* Calculate dot products:
1022 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_X
, normal
, VPpli
);
1023 emit_op2(p
, OPCODE_DP3
, dots
, WRITEMASK_Y
, normal
, half
);
1025 /* we're done with VPpli and half now, so free them as to not drive up
1026 our temp usage unnecessary */
1027 release_temp(p
, VPpli
);
1028 release_temp(p
, half
);
1030 /* Front face lighting:
1033 struct ureg ambient
= get_lightprod(p
, i
, 0, STATE_AMBIENT
);
1034 struct ureg diffuse
= get_lightprod(p
, i
, 0, STATE_DIFFUSE
);
1035 struct ureg specular
= get_lightprod(p
, i
, 0, STATE_SPECULAR
);
1036 struct ureg res0
, res1
;
1037 GLuint mask0
, mask1
;
1039 emit_op1(p
, OPCODE_LIT
, lit
, 0, dots
);
1042 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1045 if (count
== nr_lights
) {
1047 mask0
= WRITEMASK_XYZ
;
1048 mask1
= WRITEMASK_XYZ
;
1049 res0
= register_output( p
, VERT_RESULT_COL0
);
1050 res1
= register_output( p
, VERT_RESULT_COL1
);
1054 mask1
= WRITEMASK_XYZ
;
1056 res1
= register_output( p
, VERT_RESULT_COL0
);
1065 emit_op3(p
, OPCODE_MAD
, _col0
, 0, swizzle1(lit
,X
), ambient
, _col0
);
1066 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _col0
);
1067 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _col1
);
1069 release_temp(p
, ambient
);
1070 release_temp(p
, diffuse
);
1071 release_temp(p
, specular
);
1074 /* Back face lighting:
1077 struct ureg ambient
= get_lightprod(p
, i
, 1, STATE_AMBIENT
);
1078 struct ureg diffuse
= get_lightprod(p
, i
, 1, STATE_DIFFUSE
);
1079 struct ureg specular
= get_lightprod(p
, i
, 1, STATE_SPECULAR
);
1080 struct ureg res0
, res1
;
1081 GLuint mask0
, mask1
;
1083 emit_op1(p
, OPCODE_LIT
, lit
, 0, negate(swizzle(dots
,X
,Y
,W
,Z
)));
1086 emit_op2(p
, OPCODE_MUL
, lit
, 0, lit
, att
);
1088 if (count
== nr_lights
) {
1090 mask0
= WRITEMASK_XYZ
;
1091 mask1
= WRITEMASK_XYZ
;
1092 res0
= register_output( p
, VERT_RESULT_BFC0
);
1093 res1
= register_output( p
, VERT_RESULT_BFC1
);
1097 mask1
= WRITEMASK_XYZ
;
1099 res1
= register_output( p
, VERT_RESULT_BFC0
);
1108 emit_op3(p
, OPCODE_MAD
, _bfc0
, 0, swizzle1(lit
,X
), ambient
, _bfc0
);
1109 emit_op3(p
, OPCODE_MAD
, res0
, mask0
, swizzle1(lit
,Y
), diffuse
, _bfc0
);
1110 emit_op3(p
, OPCODE_MAD
, res1
, mask1
, swizzle1(lit
,Z
), specular
, _bfc1
);
1112 release_temp(p
, ambient
);
1113 release_temp(p
, diffuse
);
1114 release_temp(p
, specular
);
1117 release_temp(p
, att
);
1125 static void build_fog( struct tnl_program
*p
)
1127 struct ureg fog
= register_output(p
, VERT_RESULT_FOGC
);
1130 if (p
->state
->fog_source_is_depth
) {
1131 input
= swizzle1(get_eye_position(p
), Z
);
1134 input
= swizzle1(register_input(p
, VERT_ATTRIB_FOG
), X
);
1137 if (p
->state
->fog_mode
&& p
->state
->tnl_do_vertex_fog
) {
1138 struct ureg params
= register_param2(p
, STATE_INTERNAL
,
1139 STATE_FOG_PARAMS_OPTIMIZED
);
1140 struct ureg tmp
= get_temp(p
);
1141 GLboolean useabs
= (p
->state
->fog_mode
!= FOG_EXP2
);
1144 emit_op1(p
, OPCODE_ABS
, tmp
, 0, input
);
1147 switch (p
->state
->fog_mode
) {
1149 struct ureg id
= get_identity_param(p
);
1150 emit_op3(p
, OPCODE_MAD
, tmp
, 0, useabs
? tmp
: input
,
1151 swizzle1(params
,X
), swizzle1(params
,Y
));
1152 emit_op2(p
, OPCODE_MAX
, tmp
, 0, tmp
, swizzle1(id
,X
)); /* saturate */
1153 emit_op2(p
, OPCODE_MIN
, fog
, WRITEMASK_X
, tmp
, swizzle1(id
,W
));
1157 emit_op2(p
, OPCODE_MUL
, tmp
, 0, useabs
? tmp
: input
,
1158 swizzle1(params
,Z
));
1159 emit_op1(p
, OPCODE_EX2
, fog
, WRITEMASK_X
, negate(tmp
));
1162 emit_op2(p
, OPCODE_MUL
, tmp
, 0, input
, swizzle1(params
,W
));
1163 emit_op2(p
, OPCODE_MUL
, tmp
, 0, tmp
, tmp
);
1164 emit_op1(p
, OPCODE_EX2
, fog
, WRITEMASK_X
, negate(tmp
));
1168 release_temp(p
, tmp
);
1171 /* results = incoming fog coords (compute fog per-fragment later)
1173 * KW: Is it really necessary to do anything in this case?
1174 * BP: Yes, we always need to compute the absolute value, unless
1175 * we want to push that down into the fragment program...
1177 GLboolean useabs
= GL_TRUE
;
1178 emit_op1(p
, useabs
? OPCODE_ABS
: OPCODE_MOV
, fog
, WRITEMASK_X
, input
);
1182 static void build_reflect_texgen( struct tnl_program
*p
,
1186 struct ureg normal
= get_eye_normal(p
);
1187 struct ureg eye_hat
= get_eye_position_normalized(p
);
1188 struct ureg tmp
= get_temp(p
);
1191 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1193 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1195 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, negate(tmp
), normal
, eye_hat
);
1197 release_temp(p
, tmp
);
1200 static void build_sphere_texgen( struct tnl_program
*p
,
1204 struct ureg normal
= get_eye_normal(p
);
1205 struct ureg eye_hat
= get_eye_position_normalized(p
);
1206 struct ureg tmp
= get_temp(p
);
1207 struct ureg half
= register_scalar_const(p
, .5);
1208 struct ureg r
= get_temp(p
);
1209 struct ureg inv_m
= get_temp(p
);
1210 struct ureg id
= get_identity_param(p
);
1212 /* Could share the above calculations, but it would be
1213 * a fairly odd state for someone to set (both sphere and
1214 * reflection active for different texture coordinate
1215 * components. Of course - if two texture units enable
1216 * reflect and/or sphere, things start to tilt in favour
1217 * of seperating this out:
1221 emit_op2(p
, OPCODE_DP3
, tmp
, 0, normal
, eye_hat
);
1223 emit_op2(p
, OPCODE_ADD
, tmp
, 0, tmp
, tmp
);
1225 emit_op3(p
, OPCODE_MAD
, r
, 0, negate(tmp
), normal
, eye_hat
);
1227 emit_op2(p
, OPCODE_ADD
, tmp
, 0, r
, swizzle(id
,X
,Y
,W
,Z
));
1228 /* rx^2 + ry^2 + (rz+1)^2 */
1229 emit_op2(p
, OPCODE_DP3
, tmp
, 0, tmp
, tmp
);
1231 emit_op1(p
, OPCODE_RSQ
, tmp
, 0, tmp
);
1233 emit_op2(p
, OPCODE_MUL
, inv_m
, 0, tmp
, half
);
1235 emit_op3(p
, OPCODE_MAD
, dest
, writemask
, r
, inv_m
, half
);
1237 release_temp(p
, tmp
);
1239 release_temp(p
, inv_m
);
1243 static void build_texture_transform( struct tnl_program
*p
)
1247 for (i
= 0; i
< MAX_TEXTURE_UNITS
; i
++) {
1249 if (!(p
->state
->fragprog_inputs_read
& FRAG_BIT_TEX(i
)))
1252 if (p
->state
->unit
[i
].texgen_enabled
||
1253 p
->state
->unit
[i
].texmat_enabled
) {
1255 GLuint texmat_enabled
= p
->state
->unit
[i
].texmat_enabled
;
1256 struct ureg out
= register_output(p
, VERT_RESULT_TEX0
+ i
);
1257 struct ureg out_texgen
= undef
;
1259 if (p
->state
->unit
[i
].texgen_enabled
) {
1260 GLuint copy_mask
= 0;
1261 GLuint sphere_mask
= 0;
1262 GLuint reflect_mask
= 0;
1263 GLuint normal_mask
= 0;
1267 out_texgen
= get_temp(p
);
1271 modes
[0] = p
->state
->unit
[i
].texgen_mode0
;
1272 modes
[1] = p
->state
->unit
[i
].texgen_mode1
;
1273 modes
[2] = p
->state
->unit
[i
].texgen_mode2
;
1274 modes
[3] = p
->state
->unit
[i
].texgen_mode3
;
1276 for (j
= 0; j
< 4; j
++) {
1278 case TXG_OBJ_LINEAR
: {
1279 struct ureg obj
= register_input(p
, VERT_ATTRIB_POS
);
1281 register_param3(p
, STATE_TEXGEN
, i
,
1282 STATE_TEXGEN_OBJECT_S
+ j
);
1284 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1288 case TXG_EYE_LINEAR
: {
1289 struct ureg eye
= get_eye_position(p
);
1291 register_param3(p
, STATE_TEXGEN
, i
,
1292 STATE_TEXGEN_EYE_S
+ j
);
1294 emit_op2(p
, OPCODE_DP4
, out_texgen
, WRITEMASK_X
<< j
,
1298 case TXG_SPHERE_MAP
:
1299 sphere_mask
|= WRITEMASK_X
<< j
;
1301 case TXG_REFLECTION_MAP
:
1302 reflect_mask
|= WRITEMASK_X
<< j
;
1304 case TXG_NORMAL_MAP
:
1305 normal_mask
|= WRITEMASK_X
<< j
;
1308 copy_mask
|= WRITEMASK_X
<< j
;
1315 build_sphere_texgen(p
, out_texgen
, sphere_mask
);
1319 build_reflect_texgen(p
, out_texgen
, reflect_mask
);
1323 struct ureg normal
= get_eye_normal(p
);
1324 emit_op1(p
, OPCODE_MOV
, out_texgen
, normal_mask
, normal
);
1328 struct ureg in
= register_input(p
, VERT_ATTRIB_TEX0
+i
);
1329 emit_op1(p
, OPCODE_MOV
, out_texgen
, copy_mask
, in
);
1333 if (texmat_enabled
) {
1334 struct ureg texmat
[4];
1335 struct ureg in
= (!is_undef(out_texgen
) ?
1337 register_input(p
, VERT_ATTRIB_TEX0
+i
));
1339 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1341 emit_matrix_transform_vec4( p
, out
, texmat
, in
);
1344 register_matrix_param5( p
, STATE_TEXTURE_MATRIX
, i
, 0, 3,
1345 STATE_MATRIX_TRANSPOSE
, texmat
);
1346 emit_transpose_matrix_transform_vec4( p
, out
, texmat
, in
);
1353 emit_passthrough(p
, VERT_ATTRIB_TEX0
+i
, VERT_RESULT_TEX0
+i
);
1359 static void build_pointsize( struct tnl_program
*p
)
1361 struct ureg eye
= get_eye_position(p
);
1362 struct ureg state_size
= register_param1(p
, STATE_POINT_SIZE
);
1363 struct ureg state_attenuation
= register_param1(p
, STATE_POINT_ATTENUATION
);
1364 struct ureg out
= register_output(p
, VERT_RESULT_PSIZ
);
1365 struct ureg ut
= get_temp(p
);
1368 emit_op1(p
, OPCODE_ABS
, ut
, WRITEMASK_Y
, swizzle1(eye
, Z
));
1369 /* p1 + dist * (p2 + dist * p3); */
1370 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1371 swizzle1(state_attenuation
, Z
), swizzle1(state_attenuation
, Y
));
1372 emit_op3(p
, OPCODE_MAD
, ut
, WRITEMASK_X
, swizzle1(ut
, Y
),
1373 ut
, swizzle1(state_attenuation
, X
));
1375 /* 1 / sqrt(factor) */
1376 emit_op1(p
, OPCODE_RSQ
, ut
, WRITEMASK_X
, ut
);
1379 /* out = pointSize / sqrt(factor) */
1380 emit_op2(p
, OPCODE_MUL
, out
, WRITEMASK_X
, ut
, state_size
);
1382 /* not sure, might make sense to do clamping here,
1383 but it's not done in t_vb_points neither */
1384 emit_op2(p
, OPCODE_MUL
, ut
, WRITEMASK_X
, ut
, state_size
);
1385 emit_op2(p
, OPCODE_MAX
, ut
, WRITEMASK_X
, ut
, swizzle1(state_size
, Y
));
1386 emit_op2(p
, OPCODE_MIN
, out
, WRITEMASK_X
, ut
, swizzle1(state_size
, Z
));
1389 release_temp(p
, ut
);
1392 static void build_tnl_program( struct tnl_program
*p
)
1393 { /* Emit the program, starting with modelviewproject:
1397 /* Lighting calculations:
1399 if (p
->state
->fragprog_inputs_read
& (FRAG_BIT_COL0
|FRAG_BIT_COL1
)) {
1400 if (p
->state
->light_global_enabled
)
1403 if (p
->state
->fragprog_inputs_read
& FRAG_BIT_COL0
)
1404 emit_passthrough(p
, VERT_ATTRIB_COLOR0
, VERT_RESULT_COL0
);
1406 if (p
->state
->fragprog_inputs_read
& FRAG_BIT_COL1
)
1407 emit_passthrough(p
, VERT_ATTRIB_COLOR1
, VERT_RESULT_COL1
);
1411 if ((p
->state
->fragprog_inputs_read
& FRAG_BIT_FOGC
) ||
1412 p
->state
->fog_mode
!= FOG_NONE
)
1415 if (p
->state
->fragprog_inputs_read
& FRAG_BITS_TEX_ANY
)
1416 build_texture_transform(p
);
1418 if (p
->state
->point_attenuated
)
1423 emit_op1(p
, OPCODE_END
, undef
, 0, undef
);
1428 _mesa_printf ("\n");
1434 create_new_program( const struct state_key
*key
,
1435 struct gl_vertex_program
*program
,
1438 struct tnl_program p
;
1440 _mesa_memset(&p
, 0, sizeof(p
));
1442 p
.program
= program
;
1443 p
.eye_position
= undef
;
1444 p
.eye_position_normalized
= undef
;
1445 p
.eye_normal
= undef
;
1449 if (max_temps
>= sizeof(int) * 8)
1450 p
.temp_reserved
= 0;
1452 p
.temp_reserved
= ~((1<<max_temps
)-1);
1454 p
.program
->Base
.Instructions
= _mesa_alloc_instructions(MAX_INSN
);
1455 p
.program
->Base
.String
= NULL
;
1456 p
.program
->Base
.NumInstructions
=
1457 p
.program
->Base
.NumTemporaries
=
1458 p
.program
->Base
.NumParameters
=
1459 p
.program
->Base
.NumAttributes
= p
.program
->Base
.NumAddressRegs
= 0;
1460 p
.program
->Base
.Parameters
= _mesa_new_parameter_list();
1461 p
.program
->Base
.InputsRead
= 0;
1462 p
.program
->Base
.OutputsWritten
= 0;
1464 build_tnl_program( &p
);
1467 static void *search_cache( struct tnl_cache
*cache
,
1472 struct tnl_cache_item
*c
;
1474 for (c
= cache
->items
[hash
% cache
->size
]; c
; c
= c
->next
) {
1475 if (c
->hash
== hash
&& _mesa_memcmp(c
->key
, key
, keysize
) == 0)
1482 static void rehash( struct tnl_cache
*cache
)
1484 struct tnl_cache_item
**items
;
1485 struct tnl_cache_item
*c
, *next
;
1488 size
= cache
->size
* 3;
1489 items
= (struct tnl_cache_item
**) _mesa_malloc(size
* sizeof(*items
));
1490 _mesa_memset(items
, 0, size
* sizeof(*items
));
1492 for (i
= 0; i
< cache
->size
; i
++)
1493 for (c
= cache
->items
[i
]; c
; c
= next
) {
1495 c
->next
= items
[c
->hash
% size
];
1496 items
[c
->hash
% size
] = c
;
1500 cache
->items
= items
;
1504 static void cache_item( struct tnl_cache
*cache
,
1509 struct tnl_cache_item
*c
= (struct tnl_cache_item
*) _mesa_malloc(sizeof(*c
));
1514 if (++cache
->n_items
> cache
->size
* 1.5)
1517 c
->next
= cache
->items
[hash
% cache
->size
];
1518 cache
->items
[hash
% cache
->size
] = c
;
1521 static GLuint
hash_key( struct state_key
*key
)
1523 GLuint
*ikey
= (GLuint
*)key
;
1526 /* I'm sure this can be improved on, but speed is important:
1528 for (i
= 0; i
< sizeof(*key
)/sizeof(GLuint
); i
++)
1534 void _tnl_UpdateFixedFunctionProgram( GLcontext
*ctx
)
1536 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
1537 struct state_key
*key
;
1539 const struct gl_vertex_program
*prev
= ctx
->VertexProgram
._Current
;
1541 if (!ctx
->VertexProgram
._Current
||
1542 ctx
->VertexProgram
._Current
== ctx
->VertexProgram
._TnlProgram
) {
1543 /* Grab all the relevent state and put it in a single structure:
1545 key
= make_state_key(ctx
);
1546 hash
= hash_key(key
);
1548 /* Look for an already-prepared program for this state:
1550 ctx
->VertexProgram
._TnlProgram
= (struct gl_vertex_program
*)
1551 search_cache( tnl
->vp_cache
, hash
, key
, sizeof(*key
) );
1553 /* OK, we'll have to build a new one:
1555 if (!ctx
->VertexProgram
._TnlProgram
) {
1557 _mesa_printf("Build new TNL program\n");
1559 ctx
->VertexProgram
._TnlProgram
= (struct gl_vertex_program
*)
1560 ctx
->Driver
.NewProgram(ctx
, GL_VERTEX_PROGRAM_ARB
, 0);
1562 create_new_program( key
, ctx
->VertexProgram
._TnlProgram
,
1563 ctx
->Const
.VertexProgram
.MaxTemps
);
1565 if (ctx
->Driver
.ProgramStringNotify
)
1566 ctx
->Driver
.ProgramStringNotify( ctx
, GL_VERTEX_PROGRAM_ARB
,
1567 &ctx
->VertexProgram
._TnlProgram
->Base
);
1569 cache_item(tnl
->vp_cache
, hash
, key
, ctx
->VertexProgram
._TnlProgram
);
1574 _mesa_printf("Found existing TNL program for key %x\n", hash
);
1576 _mesa_reference_vertprog(ctx
, &ctx
->VertexProgram
._Current
,
1577 ctx
->VertexProgram
._TnlProgram
);
1580 /* Tell the driver about the change. Could define a new target for
1583 if (ctx
->VertexProgram
._Current
!= prev
&& ctx
->Driver
.BindProgram
) {
1584 ctx
->Driver
.BindProgram(ctx
, GL_VERTEX_PROGRAM_ARB
,
1585 (struct gl_program
*) ctx
->VertexProgram
._Current
);
1589 void _tnl_ProgramCacheInit( GLcontext
*ctx
)
1591 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
1593 tnl
->vp_cache
= (struct tnl_cache
*) MALLOC(sizeof(*tnl
->vp_cache
));
1594 tnl
->vp_cache
->size
= 17;
1595 tnl
->vp_cache
->n_items
= 0;
1596 tnl
->vp_cache
->items
= (struct tnl_cache_item
**)
1597 _mesa_calloc(tnl
->vp_cache
->size
* sizeof(*tnl
->vp_cache
->items
));
1600 void _tnl_ProgramCacheDestroy( GLcontext
*ctx
)
1602 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
1603 struct tnl_cache_item
*c
, *next
;
1606 for (i
= 0; i
< tnl
->vp_cache
->size
; i
++)
1607 for (c
= tnl
->vp_cache
->items
[i
]; c
; c
= next
) {
1614 FREE(tnl
->vp_cache
->items
);
1615 FREE(tnl
->vp_cache
);