1 /*
2 * Mesa 3-D graphics library
3 *
4 * Copyright (C) 1999-2008 Brian Paul All Rights Reserved.
5 * Copyright (C) 2009 VMware, Inc. All Rights Reserved.
6 *
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:
13 *
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
16 *
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 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
21 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
22 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
23 * OTHER DEALINGS IN THE SOFTWARE.
24 */
25
26
27 /**
28 * \file tnl/t_vb_program.c
29 * \brief Pipeline stage for executing vertex programs.
30 * \author Brian Paul, Keith Whitwell
31 */
32
33
34 #include "main/glheader.h"
35 #include "main/macros.h"
36 #include "main/imports.h"
37 #include "main/samplerobj.h"
38 #include "main/state.h"
39 #include "math/m_xform.h"
40 #include "program/prog_instruction.h"
41 #include "program/prog_statevars.h"
42 #include "program/prog_execute.h"
43 #include "swrast/s_context.h"
44 #include "util/bitscan.h"
45
46 #include "tnl/tnl.h"
47 #include "tnl/t_context.h"
48 #include "tnl/t_pipeline.h"
49
50
51 #ifdef NAN_CHECK
52 /** Check for NaNs and very large values */
53 static inline void
check_float(float x)54 check_float(float x)
55 {
56 assert(!IS_INF_OR_NAN(x));
57 assert(1.0e-15 <= x && x <= 1.0e15);
58 }
59 #endif
60
61
62 /*!
63 * Private storage for the vertex program pipeline stage.
64 */
65 struct vp_stage_data {
66 /** The results of running the vertex program go into these arrays. */
67 GLvector4f results[VARYING_SLOT_MAX];
68
69 GLvector4f ndcCoords; /**< normalized device coords */
70 GLubyte *clipmask; /**< clip flags */
71 GLubyte ormask, andmask; /**< for clipping */
72
73 GLboolean vertex_textures;
74
75 struct gl_program_machine machine;
76 };
77
78
79 #define VP_STAGE_DATA(stage) ((struct vp_stage_data *)(stage->privatePtr))
80
81
82 static void
userclip(struct gl_context * ctx,GLvector4f * clip,GLubyte * clipmask,GLubyte * clipormask,GLubyte * clipandmask)83 userclip( struct gl_context *ctx,
84 GLvector4f *clip,
85 GLubyte *clipmask,
86 GLubyte *clipormask,
87 GLubyte *clipandmask )
88 {
89 GLbitfield mask = ctx->Transform.ClipPlanesEnabled;
90 while (mask) {
91 const int p = u_bit_scan(&mask);
92 GLuint nr, i;
93 const GLfloat a = ctx->Transform._ClipUserPlane[p][0];
94 const GLfloat b = ctx->Transform._ClipUserPlane[p][1];
95 const GLfloat c = ctx->Transform._ClipUserPlane[p][2];
96 const GLfloat d = ctx->Transform._ClipUserPlane[p][3];
97 GLfloat *coord = (GLfloat *)clip->data;
98 GLuint stride = clip->stride;
99 GLuint count = clip->count;
100
101 for (nr = 0, i = 0 ; i < count ; i++) {
102 GLfloat dp = (coord[0] * a +
103 coord[1] * b +
104 coord[2] * c +
105 coord[3] * d);
106
107 if (dp < 0) {
108 nr++;
109 clipmask[i] |= CLIP_USER_BIT;
110 }
111
112 STRIDE_F(coord, stride);
113 }
114
115 if (nr > 0) {
116 *clipormask |= CLIP_USER_BIT;
117 if (nr == count) {
118 *clipandmask |= CLIP_USER_BIT;
119 return;
120 }
121 }
122 }
123 }
124
125
126 static GLboolean
do_ndc_cliptest(struct gl_context * ctx,struct vp_stage_data * store)127 do_ndc_cliptest(struct gl_context *ctx, struct vp_stage_data *store)
128 {
129 TNLcontext *tnl = TNL_CONTEXT(ctx);
130 struct vertex_buffer *VB = &tnl->vb;
131 /* Cliptest and perspective divide. Clip functions must clear
132 * the clipmask.
133 */
134 store->ormask = 0;
135 store->andmask = CLIP_FRUSTUM_BITS;
136
137 tnl_clip_prepare(ctx);
138
139 if (tnl->NeedNdcCoords) {
140 VB->NdcPtr =
141 _mesa_clip_tab[VB->ClipPtr->size]( VB->ClipPtr,
142 &store->ndcCoords,
143 store->clipmask,
144 &store->ormask,
145 &store->andmask,
146 !ctx->Transform.DepthClamp );
147 }
148 else {
149 VB->NdcPtr = NULL;
150 _mesa_clip_np_tab[VB->ClipPtr->size]( VB->ClipPtr,
151 NULL,
152 store->clipmask,
153 &store->ormask,
154 &store->andmask,
155 !ctx->Transform.DepthClamp );
156 }
157
158 if (store->andmask) {
159 /* All vertices are outside the frustum */
160 return GL_FALSE;
161 }
162
163 /* Test userclip planes. This contributes to VB->ClipMask.
164 */
165 /** XXX NEW_SLANG _Enabled ??? */
166 if (ctx->Transform.ClipPlanesEnabled &&
167 (!_mesa_arb_vertex_program_enabled(ctx) ||
168 ctx->VertexProgram.Current->arb.IsPositionInvariant)) {
169 userclip( ctx,
170 VB->ClipPtr,
171 store->clipmask,
172 &store->ormask,
173 &store->andmask );
174
175 if (store->andmask) {
176 return GL_FALSE;
177 }
178 }
179
180 VB->ClipAndMask = store->andmask;
181 VB->ClipOrMask = store->ormask;
182 VB->ClipMask = store->clipmask;
183
184 return GL_TRUE;
185 }
186
187
188 /**
189 * XXX the texture sampling code in this module is a bit of a hack.
190 * The texture sampling code is in swrast, though it doesn't have any
191 * real dependencies on the rest of swrast. It should probably be
192 * moved into main/ someday.
193 */
194 static void
vp_fetch_texel(struct gl_context * ctx,const GLfloat texcoord[4],GLfloat lambda,GLuint unit,GLfloat color[4])195 vp_fetch_texel(struct gl_context *ctx, const GLfloat texcoord[4], GLfloat lambda,
196 GLuint unit, GLfloat color[4])
197 {
198 SWcontext *swrast = SWRAST_CONTEXT(ctx);
199
200 /* XXX use a float-valued TextureSample routine here!!! */
201 swrast->TextureSample[unit](ctx, _mesa_get_samplerobj(ctx, unit),
202 ctx->Texture.Unit[unit]._Current,
203 1, (const GLfloat (*)[4]) texcoord,
204 &lambda, (GLfloat (*)[4]) color);
205 }
206
207
208 /**
209 * Called via ctx->Driver.ProgramStringNotify() after a new vertex program
210 * string has been parsed.
211 */
212 GLboolean
_tnl_program_string(struct gl_context * ctx,GLenum target,struct gl_program * program)213 _tnl_program_string(struct gl_context *ctx, GLenum target, struct gl_program *program)
214 {
215 /* No-op.
216 * If we had derived anything from the program that was private to this
217 * stage we'd recompute/validate it here.
218 */
219 return GL_TRUE;
220 }
221
222
223 /**
224 * Initialize virtual machine state prior to executing vertex program.
225 */
226 static void
init_machine(struct gl_context * ctx,struct gl_program_machine * machine,GLuint instID)227 init_machine(struct gl_context *ctx, struct gl_program_machine *machine,
228 GLuint instID)
229 {
230 /* Input registers get initialized from the current vertex attribs */
231 memcpy(machine->VertAttribs, ctx->Current.Attrib,
232 MAX_VERTEX_GENERIC_ATTRIBS * 4 * sizeof(GLfloat));
233
234 machine->NumDeriv = 0;
235
236 /* init call stack */
237 machine->StackDepth = 0;
238
239 machine->FetchTexelLod = vp_fetch_texel;
240 machine->FetchTexelDeriv = NULL; /* not used by vertex programs */
241
242 machine->Samplers = ctx->VertexProgram._Current->SamplerUnits;
243
244 machine->SystemValues[SYSTEM_VALUE_INSTANCE_ID][0] = (GLfloat) instID;
245 }
246
247
248 /**
249 * Map the texture images which the vertex program will access (if any).
250 */
251 static void
map_textures(struct gl_context * ctx,const struct gl_program * vp)252 map_textures(struct gl_context *ctx, const struct gl_program *vp)
253 {
254 GLuint u;
255
256 for (u = 0; u < ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits; u++) {
257 if (vp->TexturesUsed[u]) {
258 /* Note: _Current *should* correspond to the target indicated
259 * in TexturesUsed[u].
260 */
261 _swrast_map_texture(ctx, ctx->Texture.Unit[u]._Current);
262 }
263 }
264 }
265
266
267 /**
268 * Unmap the texture images which were used by the vertex program (if any).
269 */
270 static void
unmap_textures(struct gl_context * ctx,const struct gl_program * vp)271 unmap_textures(struct gl_context *ctx, const struct gl_program *vp)
272 {
273 GLuint u;
274
275 for (u = 0; u < ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits; u++) {
276 if (vp->TexturesUsed[u]) {
277 /* Note: _Current *should* correspond to the target indicated
278 * in TexturesUsed[u].
279 */
280 _swrast_unmap_texture(ctx, ctx->Texture.Unit[u]._Current);
281 }
282 }
283 }
284
285
286 /**
287 * This function executes vertex programs
288 */
289 static GLboolean
run_vp(struct gl_context * ctx,struct tnl_pipeline_stage * stage)290 run_vp( struct gl_context *ctx, struct tnl_pipeline_stage *stage )
291 {
292 TNLcontext *tnl = TNL_CONTEXT(ctx);
293 struct vp_stage_data *store = VP_STAGE_DATA(stage);
294 struct vertex_buffer *VB = &tnl->vb;
295 struct gl_program *program = ctx->VertexProgram._Current;
296 struct gl_program_machine *machine = &store->machine;
297 GLuint outputs[VARYING_SLOT_MAX], numOutputs;
298 GLuint i, j;
299
300 if (!program)
301 return GL_TRUE;
302
303 /* ARB program or vertex shader */
304 _mesa_load_state_parameters(ctx, program->Parameters);
305
306 /* make list of outputs to save some time below */
307 numOutputs = 0;
308 for (i = 0; i < VARYING_SLOT_MAX; i++) {
309 if (program->info.outputs_written & BITFIELD64_BIT(i)) {
310 outputs[numOutputs++] = i;
311 }
312 }
313
314 /* Allocate result vectors. We delay this until now to avoid allocating
315 * memory that would never be used if we don't run the software tnl pipeline.
316 */
317 if (!store->results[0].storage) {
318 for (i = 0; i < VARYING_SLOT_MAX; i++) {
319 assert(!store->results[i].storage);
320 _mesa_vector4f_alloc( &store->results[i], 0, VB->Size, 32 );
321 store->results[i].size = 4;
322 }
323 }
324
325 map_textures(ctx, program);
326
327 for (i = 0; i < VB->Count; i++) {
328 GLuint attr;
329
330 init_machine(ctx, machine, tnl->CurInstance);
331
332 #if 0
333 printf("Input %d: %f, %f, %f, %f\n", i,
334 VB->AttribPtr[0]->data[i][0],
335 VB->AttribPtr[0]->data[i][1],
336 VB->AttribPtr[0]->data[i][2],
337 VB->AttribPtr[0]->data[i][3]);
338 printf(" color: %f, %f, %f, %f\n",
339 VB->AttribPtr[3]->data[i][0],
340 VB->AttribPtr[3]->data[i][1],
341 VB->AttribPtr[3]->data[i][2],
342 VB->AttribPtr[3]->data[i][3]);
343 printf(" normal: %f, %f, %f, %f\n",
344 VB->AttribPtr[2]->data[i][0],
345 VB->AttribPtr[2]->data[i][1],
346 VB->AttribPtr[2]->data[i][2],
347 VB->AttribPtr[2]->data[i][3]);
348 #endif
349
350 /* the vertex array case */
351 for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
352 if (program->info.inputs_read & BITFIELD64_BIT(attr)) {
353 const GLubyte *ptr = (const GLubyte*) VB->AttribPtr[attr]->data;
354 const GLuint size = VB->AttribPtr[attr]->size;
355 const GLuint stride = VB->AttribPtr[attr]->stride;
356 const GLfloat *data = (GLfloat *) (ptr + stride * i);
357 #ifdef NAN_CHECK
358 check_float(data[0]);
359 check_float(data[1]);
360 check_float(data[2]);
361 check_float(data[3]);
362 #endif
363 COPY_CLEAN_4V(machine->VertAttribs[attr], size, data);
364 }
365 }
366
367 /* execute the program */
368 _mesa_execute_program(ctx, program, machine);
369
370 /* copy the output registers into the VB->attribs arrays */
371 for (j = 0; j < numOutputs; j++) {
372 const GLuint attr = outputs[j];
373 #ifdef NAN_CHECK
374 check_float(machine->Outputs[attr][0]);
375 check_float(machine->Outputs[attr][1]);
376 check_float(machine->Outputs[attr][2]);
377 check_float(machine->Outputs[attr][3]);
378 #endif
379 COPY_4V(store->results[attr].data[i], machine->Outputs[attr]);
380 }
381
382 /* FOGC is a special case. Fragment shader expects (f,0,0,1) */
383 if (program->info.outputs_written & BITFIELD64_BIT(VARYING_SLOT_FOGC)) {
384 store->results[VARYING_SLOT_FOGC].data[i][1] = 0.0;
385 store->results[VARYING_SLOT_FOGC].data[i][2] = 0.0;
386 store->results[VARYING_SLOT_FOGC].data[i][3] = 1.0;
387 }
388 #ifdef NAN_CHECK
389 assert(machine->Outputs[0][3] != 0.0F);
390 #endif
391 #if 0
392 printf("HPOS: %f %f %f %f\n",
393 machine->Outputs[0][0],
394 machine->Outputs[0][1],
395 machine->Outputs[0][2],
396 machine->Outputs[0][3]);
397 #endif
398 }
399
400 unmap_textures(ctx, program);
401
402 if (program->arb.IsPositionInvariant) {
403 /* We need the exact same transform as in the fixed function path here
404 * to guarantee invariance, depending on compiler optimization flags
405 * results could be different otherwise.
406 */
407 VB->ClipPtr = TransformRaw( &store->results[0],
408 &ctx->_ModelProjectMatrix,
409 VB->AttribPtr[0] );
410
411 /* Drivers expect this to be clean to element 4...
412 */
413 switch (VB->ClipPtr->size) {
414 case 1:
415 /* impossible */
416 case 2:
417 _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 2 );
418 /* fall-through */
419 case 3:
420 _mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 3 );
421 /* fall-through */
422 case 4:
423 break;
424 }
425 }
426 else {
427 /* Setup the VB pointers so that the next pipeline stages get
428 * their data from the right place (the program output arrays).
429 */
430 VB->ClipPtr = &store->results[VARYING_SLOT_POS];
431 VB->ClipPtr->size = 4;
432 VB->ClipPtr->count = VB->Count;
433 }
434
435 VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VARYING_SLOT_COL0];
436 VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VARYING_SLOT_COL1];
437 VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VARYING_SLOT_FOGC];
438 VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VARYING_SLOT_PSIZ];
439 VB->BackfaceColorPtr = &store->results[VARYING_SLOT_BFC0];
440 VB->BackfaceSecondaryColorPtr = &store->results[VARYING_SLOT_BFC1];
441
442 for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
443 VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
444 = &store->results[VARYING_SLOT_TEX0 + i];
445 }
446
447 for (i = 0; i < ctx->Const.MaxVarying; i++) {
448 if (program->info.outputs_written &
449 BITFIELD64_BIT(VARYING_SLOT_VAR0 + i)) {
450 /* Note: varying results get put into the generic attributes */
451 VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
452 = &store->results[VARYING_SLOT_VAR0 + i];
453 }
454 }
455
456
457 /* Perform NDC and cliptest operations:
458 */
459 return do_ndc_cliptest(ctx, store);
460 }
461
462
463 /**
464 * Called the first time stage->run is called. In effect, don't
465 * allocate data until the first time the stage is run.
466 */
467 static GLboolean
init_vp(struct gl_context * ctx,struct tnl_pipeline_stage * stage)468 init_vp(struct gl_context *ctx, struct tnl_pipeline_stage *stage)
469 {
470 TNLcontext *tnl = TNL_CONTEXT(ctx);
471 struct vertex_buffer *VB = &(tnl->vb);
472 struct vp_stage_data *store;
473 const GLuint size = VB->Size;
474
475 stage->privatePtr = calloc(1, sizeof(*store));
476 store = VP_STAGE_DATA(stage);
477 if (!store)
478 return GL_FALSE;
479
480 /* a few other misc allocations */
481 _mesa_vector4f_alloc( &store->ndcCoords, 0, size, 32 );
482 store->clipmask = _mesa_align_malloc(sizeof(GLubyte)*size, 32 );
483
484 return GL_TRUE;
485 }
486
487
488 /**
489 * Destructor for this pipeline stage.
490 */
491 static void
dtr(struct tnl_pipeline_stage * stage)492 dtr(struct tnl_pipeline_stage *stage)
493 {
494 struct vp_stage_data *store = VP_STAGE_DATA(stage);
495
496 if (store) {
497 GLuint i;
498
499 /* free the vertex program result arrays */
500 for (i = 0; i < VARYING_SLOT_MAX; i++)
501 _mesa_vector4f_free( &store->results[i] );
502
503 /* free misc arrays */
504 _mesa_vector4f_free( &store->ndcCoords );
505 _mesa_align_free( store->clipmask );
506
507 free( store );
508 stage->privatePtr = NULL;
509 }
510 }
511
512
513 static void
validate_vp_stage(struct gl_context * ctx,struct tnl_pipeline_stage * stage)514 validate_vp_stage(struct gl_context *ctx, struct tnl_pipeline_stage *stage)
515 {
516 if (ctx->VertexProgram._Current) {
517 _swrast_update_texture_samplers(ctx);
518 }
519 }
520
521
522
523 /**
524 * Public description of this pipeline stage.
525 */
526 const struct tnl_pipeline_stage _tnl_vertex_program_stage =
527 {
528 "vertex-program",
529 NULL, /* private_data */
530 init_vp, /* create */
531 dtr, /* destroy */
532 validate_vp_stage, /* validate */
533 run_vp /* run -- initially set to ctr */
534 };
535