2 * Mesa 3-D graphics library
5 * Copyright (C) 1999-2008 Brian Paul All Rights Reserved.
6 * Copyright (C) 2009 VMware, Inc. All Rights Reserved.
8 * Permission is hereby granted, free of charge, to any person obtaining a
9 * copy of this software and associated documentation files (the "Software"),
10 * to deal in the Software without restriction, including without limitation
11 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
12 * and/or sell copies of the Software, and to permit persons to whom the
13 * Software is furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice shall be included
16 * in all copies or substantial portions of the Software.
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
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 * \file tnl/t_vb_program.c
29 * \brief Pipeline stage for executing vertex programs.
30 * \author Brian Paul, Keith Whitwell
34 #include "main/glheader.h"
35 #include "main/colormac.h"
36 #include "main/context.h"
37 #include "main/macros.h"
38 #include "main/imports.h"
39 #include "shader/prog_instruction.h"
40 #include "shader/prog_statevars.h"
41 #include "shader/prog_execute.h"
42 #include "swrast/s_context.h"
45 #include "tnl/t_context.h"
46 #include "tnl/t_pipeline.h"
50 /** Check for NaNs and very large values */
54 assert(!IS_INF_OR_NAN(x
));
55 assert(1.0e-15 <= x
&& x
<= 1.0e15
);
61 * Private storage for the vertex program pipeline stage.
63 struct vp_stage_data
{
64 /** The results of running the vertex program go into these arrays. */
65 GLvector4f results
[VERT_RESULT_MAX
];
67 GLvector4f ndcCoords
; /**< normalized device coords */
68 GLubyte
*clipmask
; /**< clip flags */
69 GLubyte ormask
, andmask
; /**< for clipping */
73 #define VP_STAGE_DATA(stage) ((struct vp_stage_data *)(stage->privatePtr))
77 userclip( GLcontext
*ctx
,
81 GLubyte
*clipandmask
)
85 for (p
= 0; p
< ctx
->Const
.MaxClipPlanes
; p
++) {
86 if (ctx
->Transform
.ClipPlanesEnabled
& (1 << p
)) {
88 const GLfloat a
= ctx
->Transform
._ClipUserPlane
[p
][0];
89 const GLfloat b
= ctx
->Transform
._ClipUserPlane
[p
][1];
90 const GLfloat c
= ctx
->Transform
._ClipUserPlane
[p
][2];
91 const GLfloat d
= ctx
->Transform
._ClipUserPlane
[p
][3];
92 GLfloat
*coord
= (GLfloat
*)clip
->data
;
93 GLuint stride
= clip
->stride
;
94 GLuint count
= clip
->count
;
96 for (nr
= 0, i
= 0 ; i
< count
; i
++) {
97 GLfloat dp
= (coord
[0] * a
+
104 clipmask
[i
] |= CLIP_USER_BIT
;
107 STRIDE_F(coord
, stride
);
111 *clipormask
|= CLIP_USER_BIT
;
113 *clipandmask
|= CLIP_USER_BIT
;
123 do_ndc_cliptest(GLcontext
*ctx
, struct vp_stage_data
*store
)
125 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
126 struct vertex_buffer
*VB
= &tnl
->vb
;
127 /* Cliptest and perspective divide. Clip functions must clear
131 store
->andmask
= CLIP_FRUSTUM_BITS
;
133 tnl_clip_prepare(ctx
);
135 if (tnl
->NeedNdcCoords
) {
137 _mesa_clip_tab
[VB
->ClipPtr
->size
]( VB
->ClipPtr
,
142 !ctx
->Transform
.DepthClamp
);
146 _mesa_clip_np_tab
[VB
->ClipPtr
->size
]( VB
->ClipPtr
,
151 !ctx
->Transform
.DepthClamp
);
154 if (store
->andmask
) {
155 /* All vertices are outside the frustum */
159 /* Test userclip planes. This contributes to VB->ClipMask.
161 /** XXX NEW_SLANG _Enabled ??? */
162 if (ctx
->Transform
.ClipPlanesEnabled
&& (!ctx
->VertexProgram
._Enabled
||
163 ctx
->VertexProgram
.Current
->IsPositionInvariant
)) {
170 if (store
->andmask
) {
175 VB
->ClipAndMask
= store
->andmask
;
176 VB
->ClipOrMask
= store
->ormask
;
177 VB
->ClipMask
= store
->clipmask
;
184 * XXX the texture sampling code in this module is a bit of a hack.
185 * The texture sampling code is in swrast, though it doesn't have any
186 * real dependencies on the rest of swrast. It should probably be
187 * moved into main/ someday.
190 vp_fetch_texel(GLcontext
*ctx
, const GLfloat texcoord
[4], GLfloat lambda
,
191 GLuint unit
, GLfloat color
[4])
193 SWcontext
*swrast
= SWRAST_CONTEXT(ctx
);
195 /* XXX use a float-valued TextureSample routine here!!! */
196 swrast
->TextureSample
[unit
](ctx
, ctx
->Texture
.Unit
[unit
]._Current
,
197 1, (const GLfloat (*)[4]) texcoord
,
198 &lambda
, (GLfloat (*)[4]) color
);
203 * Called via ctx->Driver.ProgramStringNotify() after a new vertex program
204 * string has been parsed.
207 _tnl_program_string(GLcontext
*ctx
, GLenum target
, struct gl_program
*program
)
210 * If we had derived anything from the program that was private to this
211 * stage we'd recompute/validate it here.
217 * Initialize virtual machine state prior to executing vertex program.
220 init_machine(GLcontext
*ctx
, struct gl_program_machine
*machine
)
222 /* Input registers get initialized from the current vertex attribs */
223 MEMCPY(machine
->VertAttribs
, ctx
->Current
.Attrib
,
224 MAX_VERTEX_GENERIC_ATTRIBS
* 4 * sizeof(GLfloat
));
226 if (ctx
->VertexProgram
._Current
->IsNVProgram
) {
228 /* Output/result regs are initialized to [0,0,0,1] */
229 for (i
= 0; i
< MAX_NV_VERTEX_PROGRAM_OUTPUTS
; i
++) {
230 ASSIGN_4V(machine
->Outputs
[i
], 0.0F
, 0.0F
, 0.0F
, 1.0F
);
232 /* Temp regs are initialized to [0,0,0,0] */
233 for (i
= 0; i
< MAX_NV_VERTEX_PROGRAM_TEMPS
; i
++) {
234 ASSIGN_4V(machine
->Temporaries
[i
], 0.0F
, 0.0F
, 0.0F
, 0.0F
);
236 for (i
= 0; i
< MAX_VERTEX_PROGRAM_ADDRESS_REGS
; i
++) {
237 ASSIGN_4V(machine
->AddressReg
[i
], 0, 0, 0, 0);
241 machine
->NumDeriv
= 0;
243 /* init condition codes */
244 machine
->CondCodes
[0] = COND_EQ
;
245 machine
->CondCodes
[1] = COND_EQ
;
246 machine
->CondCodes
[2] = COND_EQ
;
247 machine
->CondCodes
[3] = COND_EQ
;
249 /* init call stack */
250 machine
->StackDepth
= 0;
252 machine
->FetchTexelLod
= vp_fetch_texel
;
253 machine
->FetchTexelDeriv
= NULL
; /* not used by vertex programs */
255 machine
->Samplers
= ctx
->VertexProgram
._Current
->Base
.SamplerUnits
;
260 * Map the texture images which the vertex program will access (if any).
263 map_textures(GLcontext
*ctx
, const struct gl_vertex_program
*vp
)
267 if (!ctx
->Driver
.MapTexture
)
270 for (u
= 0; u
< ctx
->Const
.MaxVertexTextureImageUnits
; u
++) {
271 if (vp
->Base
.TexturesUsed
[u
]) {
272 /* Note: _Current *should* correspond to the target indicated
273 * in TexturesUsed[u].
275 ctx
->Driver
.MapTexture(ctx
, ctx
->Texture
.Unit
[u
]._Current
);
282 * Unmap the texture images which were used by the vertex program (if any).
285 unmap_textures(GLcontext
*ctx
, const struct gl_vertex_program
*vp
)
289 if (!ctx
->Driver
.MapTexture
)
292 for (u
= 0; u
< ctx
->Const
.MaxVertexTextureImageUnits
; u
++) {
293 if (vp
->Base
.TexturesUsed
[u
]) {
294 /* Note: _Current *should* correspond to the target indicated
295 * in TexturesUsed[u].
297 ctx
->Driver
.UnmapTexture(ctx
, ctx
->Texture
.Unit
[u
]._Current
);
304 * This function executes vertex programs
307 run_vp( GLcontext
*ctx
, struct tnl_pipeline_stage
*stage
)
309 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
310 struct vp_stage_data
*store
= VP_STAGE_DATA(stage
);
311 struct vertex_buffer
*VB
= &tnl
->vb
;
312 struct gl_vertex_program
*program
= ctx
->VertexProgram
._Current
;
313 struct gl_program_machine machine
;
314 GLuint outputs
[VERT_RESULT_MAX
], numOutputs
;
320 if (program
->IsNVProgram
) {
321 _mesa_load_tracked_matrices(ctx
);
324 /* ARB program or vertex shader */
325 _mesa_load_state_parameters(ctx
, program
->Base
.Parameters
);
328 /* make list of outputs to save some time below */
330 for (i
= 0; i
< VERT_RESULT_MAX
; i
++) {
331 if (program
->Base
.OutputsWritten
& BITFIELD64_BIT(i
)) {
332 outputs
[numOutputs
++] = i
;
336 map_textures(ctx
, program
);
338 for (i
= 0; i
< VB
->Count
; i
++) {
341 init_machine(ctx
, &machine
);
344 printf("Input %d: %f, %f, %f, %f\n", i
,
345 VB
->AttribPtr
[0]->data
[i
][0],
346 VB
->AttribPtr
[0]->data
[i
][1],
347 VB
->AttribPtr
[0]->data
[i
][2],
348 VB
->AttribPtr
[0]->data
[i
][3]);
349 printf(" color: %f, %f, %f, %f\n",
350 VB
->AttribPtr
[3]->data
[i
][0],
351 VB
->AttribPtr
[3]->data
[i
][1],
352 VB
->AttribPtr
[3]->data
[i
][2],
353 VB
->AttribPtr
[3]->data
[i
][3]);
354 printf(" normal: %f, %f, %f, %f\n",
355 VB
->AttribPtr
[2]->data
[i
][0],
356 VB
->AttribPtr
[2]->data
[i
][1],
357 VB
->AttribPtr
[2]->data
[i
][2],
358 VB
->AttribPtr
[2]->data
[i
][3]);
361 /* the vertex array case */
362 for (attr
= 0; attr
< VERT_ATTRIB_MAX
; attr
++) {
363 if (program
->Base
.InputsRead
& (1 << attr
)) {
364 const GLubyte
*ptr
= (const GLubyte
*) VB
->AttribPtr
[attr
]->data
;
365 const GLuint size
= VB
->AttribPtr
[attr
]->size
;
366 const GLuint stride
= VB
->AttribPtr
[attr
]->stride
;
367 const GLfloat
*data
= (GLfloat
*) (ptr
+ stride
* i
);
369 check_float(data
[0]);
370 check_float(data
[1]);
371 check_float(data
[2]);
372 check_float(data
[3]);
374 COPY_CLEAN_4V(machine
.VertAttribs
[attr
], size
, data
);
378 /* execute the program */
379 _mesa_execute_program(ctx
, &program
->Base
, &machine
);
381 /* copy the output registers into the VB->attribs arrays */
382 for (j
= 0; j
< numOutputs
; j
++) {
383 const GLuint attr
= outputs
[j
];
385 check_float(machine
.Outputs
[attr
][0]);
386 check_float(machine
.Outputs
[attr
][1]);
387 check_float(machine
.Outputs
[attr
][2]);
388 check_float(machine
.Outputs
[attr
][3]);
390 COPY_4V(store
->results
[attr
].data
[i
], machine
.Outputs
[attr
]);
393 /* FOGC is a special case. Fragment shader expects (f,0,0,1) */
394 if (program
->Base
.OutputsWritten
& BITFIELD64_BIT(VERT_RESULT_FOGC
)) {
395 store
->results
[VERT_RESULT_FOGC
].data
[i
][1] = 0.0;
396 store
->results
[VERT_RESULT_FOGC
].data
[i
][2] = 0.0;
397 store
->results
[VERT_RESULT_FOGC
].data
[i
][3] = 1.0;
400 ASSERT(machine
.Outputs
[0][3] != 0.0F
);
403 printf("HPOS: %f %f %f %f\n",
404 machine
.Outputs
[0][0],
405 machine
.Outputs
[0][1],
406 machine
.Outputs
[0][2],
407 machine
.Outputs
[0][3]);
411 unmap_textures(ctx
, program
);
413 /* Fixup fog and point size results if needed */
414 if (program
->IsNVProgram
) {
415 if (ctx
->Fog
.Enabled
&&
416 (program
->Base
.OutputsWritten
& BITFIELD64_BIT(VERT_RESULT_FOGC
)) == 0) {
417 for (i
= 0; i
< VB
->Count
; i
++) {
418 store
->results
[VERT_RESULT_FOGC
].data
[i
][0] = 1.0;
422 if (ctx
->VertexProgram
.PointSizeEnabled
&&
423 (program
->Base
.OutputsWritten
& BITFIELD64_BIT(VERT_RESULT_PSIZ
)) == 0) {
424 for (i
= 0; i
< VB
->Count
; i
++) {
425 store
->results
[VERT_RESULT_PSIZ
].data
[i
][0] = ctx
->Point
.Size
;
430 if (program
->IsPositionInvariant
) {
431 /* We need the exact same transform as in the fixed function path here
432 * to guarantee invariance, depending on compiler optimization flags
433 * results could be different otherwise.
435 VB
->ClipPtr
= TransformRaw( &store
->results
[0],
436 &ctx
->_ModelProjectMatrix
,
439 /* Drivers expect this to be clean to element 4...
441 switch (VB
->ClipPtr
->size
) {
445 _mesa_vector4f_clean_elem( VB
->ClipPtr
, VB
->Count
, 2 );
448 _mesa_vector4f_clean_elem( VB
->ClipPtr
, VB
->Count
, 3 );
455 /* Setup the VB pointers so that the next pipeline stages get
456 * their data from the right place (the program output arrays).
458 VB
->ClipPtr
= &store
->results
[VERT_RESULT_HPOS
];
459 VB
->ClipPtr
->size
= 4;
460 VB
->ClipPtr
->count
= VB
->Count
;
463 VB
->AttribPtr
[VERT_ATTRIB_COLOR0
] = &store
->results
[VERT_RESULT_COL0
];
464 VB
->AttribPtr
[VERT_ATTRIB_COLOR1
] = &store
->results
[VERT_RESULT_COL1
];
465 VB
->AttribPtr
[VERT_ATTRIB_FOG
] = &store
->results
[VERT_RESULT_FOGC
];
466 VB
->AttribPtr
[_TNL_ATTRIB_POINTSIZE
] = &store
->results
[VERT_RESULT_PSIZ
];
467 VB
->BackfaceColorPtr
= &store
->results
[VERT_RESULT_BFC0
];
468 VB
->BackfaceSecondaryColorPtr
= &store
->results
[VERT_RESULT_BFC1
];
470 for (i
= 0; i
< ctx
->Const
.MaxTextureCoordUnits
; i
++) {
471 VB
->AttribPtr
[_TNL_ATTRIB_TEX0
+ i
]
472 = &store
->results
[VERT_RESULT_TEX0
+ i
];
475 for (i
= 0; i
< ctx
->Const
.MaxVarying
; i
++) {
476 if (program
->Base
.OutputsWritten
& BITFIELD64_BIT(VERT_RESULT_VAR0
+ i
)) {
477 /* Note: varying results get put into the generic attributes */
478 VB
->AttribPtr
[VERT_ATTRIB_GENERIC0
+i
]
479 = &store
->results
[VERT_RESULT_VAR0
+ i
];
484 /* Perform NDC and cliptest operations:
486 return do_ndc_cliptest(ctx
, store
);
491 * Called the first time stage->run is called. In effect, don't
492 * allocate data until the first time the stage is run.
495 init_vp(GLcontext
*ctx
, struct tnl_pipeline_stage
*stage
)
497 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
498 struct vertex_buffer
*VB
= &(tnl
->vb
);
499 struct vp_stage_data
*store
;
500 const GLuint size
= VB
->Size
;
503 stage
->privatePtr
= MALLOC(sizeof(*store
));
504 store
= VP_STAGE_DATA(stage
);
508 /* Allocate arrays of vertex output values */
509 for (i
= 0; i
< VERT_RESULT_MAX
; i
++) {
510 _mesa_vector4f_alloc( &store
->results
[i
], 0, size
, 32 );
511 store
->results
[i
].size
= 4;
514 /* a few other misc allocations */
515 _mesa_vector4f_alloc( &store
->ndcCoords
, 0, size
, 32 );
516 store
->clipmask
= (GLubyte
*) ALIGN_MALLOC(sizeof(GLubyte
)*size
, 32 );
523 * Destructor for this pipeline stage.
526 dtr(struct tnl_pipeline_stage
*stage
)
528 struct vp_stage_data
*store
= VP_STAGE_DATA(stage
);
533 /* free the vertex program result arrays */
534 for (i
= 0; i
< VERT_RESULT_MAX
; i
++)
535 _mesa_vector4f_free( &store
->results
[i
] );
537 /* free misc arrays */
538 _mesa_vector4f_free( &store
->ndcCoords
);
539 ALIGN_FREE( store
->clipmask
);
542 stage
->privatePtr
= NULL
;
548 validate_vp_stage(GLcontext
*ctx
, struct tnl_pipeline_stage
*stage
)
550 if (ctx
->VertexProgram
._Current
) {
551 _swrast_update_texture_samplers(ctx
);
558 * Public description of this pipeline stage.
560 const struct tnl_pipeline_stage _tnl_vertex_program_stage
=
563 NULL
, /* private_data */
564 init_vp
, /* create */
566 validate_vp_stage
, /* validate */
567 run_vp
/* run -- initially set to ctr */