1/*
2 * Mesa 3-D graphics library
3 * Version:  7.1
4 *
5 * Copyright (C) 1999-2007  Brian Paul   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 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
21 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
22 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
23 *
24 * Authors:
25 *    Keith Whitwell <keith@tungstengraphics.com>
26 */
27
28#include "main/glheader.h"
29#include "main/bufferobj.h"
30#include "main/condrender.h"
31#include "main/context.h"
32#include "main/imports.h"
33#include "main/mtypes.h"
34#include "main/macros.h"
35#include "main/enums.h"
36
37#include "t_context.h"
38#include "tnl.h"
39
40
41
42static GLubyte *get_space(struct gl_context *ctx, GLuint bytes)
43{
44   TNLcontext *tnl = TNL_CONTEXT(ctx);
45   GLubyte *space = malloc(bytes);
46
47   tnl->block[tnl->nr_blocks++] = space;
48   return space;
49}
50
51
52static void free_space(struct gl_context *ctx)
53{
54   TNLcontext *tnl = TNL_CONTEXT(ctx);
55   GLuint i;
56   for (i = 0; i < tnl->nr_blocks; i++)
57      free(tnl->block[i]);
58   tnl->nr_blocks = 0;
59}
60
61
62/* Convert the incoming array to GLfloats.  Understands the
63 * array->Normalized flag and selects the correct conversion method.
64 */
65#define CONVERT( TYPE, MACRO ) do {		\
66   GLuint i, j;					\
67   if (input->Normalized) {			\
68      for (i = 0; i < count; i++) {		\
69	 const TYPE *in = (TYPE *)ptr;		\
70	 for (j = 0; j < sz; j++) {		\
71	    *fptr++ = MACRO(*in);		\
72	    in++;				\
73	 }					\
74	 ptr += input->StrideB;			\
75      }						\
76   } else {					\
77      for (i = 0; i < count; i++) {		\
78	 const TYPE *in = (TYPE *)ptr;		\
79	 for (j = 0; j < sz; j++) {		\
80	    *fptr++ = (GLfloat)(*in);		\
81	    in++;				\
82	 }					\
83	 ptr += input->StrideB;			\
84      }						\
85   }						\
86} while (0)
87
88
89/**
90 * Convert array of BGRA/GLubyte[4] values to RGBA/float[4]
91 * \param ptr  input/ubyte array
92 * \param fptr  output/float array
93 */
94static void
95convert_bgra_to_float(const struct gl_client_array *input,
96                      const GLubyte *ptr, GLfloat *fptr,
97                      GLuint count )
98{
99   GLuint i;
100   assert(input->Normalized);
101   assert(input->Size == 4);
102   for (i = 0; i < count; i++) {
103      const GLubyte *in = (GLubyte *) ptr;  /* in is in BGRA order */
104      *fptr++ = UBYTE_TO_FLOAT(in[2]);  /* red */
105      *fptr++ = UBYTE_TO_FLOAT(in[1]);  /* green */
106      *fptr++ = UBYTE_TO_FLOAT(in[0]);  /* blue */
107      *fptr++ = UBYTE_TO_FLOAT(in[3]);  /* alpha */
108      ptr += input->StrideB;
109   }
110}
111
112static void
113convert_half_to_float(const struct gl_client_array *input,
114		      const GLubyte *ptr, GLfloat *fptr,
115		      GLuint count, GLuint sz)
116{
117   GLuint i, j;
118
119   for (i = 0; i < count; i++) {
120      GLhalfARB *in = (GLhalfARB *)ptr;
121
122      for (j = 0; j < sz; j++) {
123	 *fptr++ = _mesa_half_to_float(in[j]);
124      }
125      ptr += input->StrideB;
126   }
127}
128
129/**
130 * \brief Convert fixed-point to floating-point.
131 *
132 * In OpenGL, a fixed-point number is a "signed 2's complement 16.16 scaled
133 * integer" (Table 2.2 of the OpenGL ES 2.0 spec).
134 *
135 * If the buffer has the \c normalized flag set, the formula
136 *     \code normalize(x) := (2*x + 1) / (2^16 - 1) \endcode
137 * is used to map the fixed-point numbers into the range [-1, 1].
138 */
139static void
140convert_fixed_to_float(const struct gl_client_array *input,
141                       const GLubyte *ptr, GLfloat *fptr,
142                       GLuint count)
143{
144   GLuint i, j;
145   const GLint size = input->Size;
146
147   if (input->Normalized) {
148      for (i = 0; i < count; ++i) {
149         const GLfixed *in = (GLfixed *) ptr;
150         for (j = 0; j < size; ++j) {
151            *fptr++ = (GLfloat) (2 * in[j] + 1) / (GLfloat) ((1 << 16) - 1);
152         }
153         ptr += input->StrideB;
154      }
155   } else {
156      for (i = 0; i < count; ++i) {
157         const GLfixed *in = (GLfixed *) ptr;
158         for (j = 0; j < size; ++j) {
159            *fptr++ = in[j] / (GLfloat) (1 << 16);
160         }
161         ptr += input->StrideB;
162      }
163   }
164}
165
166/* Adjust pointer to point at first requested element, convert to
167 * floating point, populate VB->AttribPtr[].
168 */
169static void _tnl_import_array( struct gl_context *ctx,
170			       GLuint attrib,
171			       GLuint count,
172			       const struct gl_client_array *input,
173			       const GLubyte *ptr )
174{
175   TNLcontext *tnl = TNL_CONTEXT(ctx);
176   struct vertex_buffer *VB = &tnl->vb;
177   GLuint stride = input->StrideB;
178
179   if (input->Type != GL_FLOAT) {
180      const GLuint sz = input->Size;
181      GLubyte *buf = get_space(ctx, count * sz * sizeof(GLfloat));
182      GLfloat *fptr = (GLfloat *)buf;
183
184      switch (input->Type) {
185      case GL_BYTE:
186	 CONVERT(GLbyte, BYTE_TO_FLOAT);
187	 break;
188      case GL_UNSIGNED_BYTE:
189         if (input->Format == GL_BGRA) {
190            /* See GL_EXT_vertex_array_bgra */
191            convert_bgra_to_float(input, ptr, fptr, count);
192         }
193         else {
194            CONVERT(GLubyte, UBYTE_TO_FLOAT);
195         }
196	 break;
197      case GL_SHORT:
198	 CONVERT(GLshort, SHORT_TO_FLOAT);
199	 break;
200      case GL_UNSIGNED_SHORT:
201	 CONVERT(GLushort, USHORT_TO_FLOAT);
202	 break;
203      case GL_INT:
204	 CONVERT(GLint, INT_TO_FLOAT);
205	 break;
206      case GL_UNSIGNED_INT:
207	 CONVERT(GLuint, UINT_TO_FLOAT);
208	 break;
209      case GL_DOUBLE:
210	 CONVERT(GLdouble, (GLfloat));
211	 break;
212      case GL_HALF_FLOAT:
213	 convert_half_to_float(input, ptr, fptr, count, sz);
214	 break;
215      case GL_FIXED:
216         convert_fixed_to_float(input, ptr, fptr, count);
217         break;
218      default:
219	 assert(0);
220	 break;
221      }
222
223      ptr = buf;
224      stride = sz * sizeof(GLfloat);
225   }
226
227   VB->AttribPtr[attrib] = &tnl->tmp_inputs[attrib];
228   VB->AttribPtr[attrib]->data = (GLfloat (*)[4])ptr;
229   VB->AttribPtr[attrib]->start = (GLfloat *)ptr;
230   VB->AttribPtr[attrib]->count = count;
231   VB->AttribPtr[attrib]->stride = stride;
232   VB->AttribPtr[attrib]->size = input->Size;
233
234   /* This should die, but so should the whole GLvector4f concept:
235    */
236   VB->AttribPtr[attrib]->flags = (((1<<input->Size)-1) |
237				   VEC_NOT_WRITEABLE |
238				   (stride == 4*sizeof(GLfloat) ? 0 : VEC_BAD_STRIDE));
239
240   VB->AttribPtr[attrib]->storage = NULL;
241}
242
243#define CLIPVERTS  ((6 + MAX_CLIP_PLANES) * 2)
244
245
246static GLboolean *_tnl_import_edgeflag( struct gl_context *ctx,
247					const GLvector4f *input,
248					GLuint count)
249{
250   const GLubyte *ptr = (const GLubyte *)input->data;
251   const GLuint stride = input->stride;
252   GLboolean *space = (GLboolean *)get_space(ctx, count + CLIPVERTS);
253   GLboolean *bptr = space;
254   GLuint i;
255
256   for (i = 0; i < count; i++) {
257      *bptr++ = ((GLfloat *)ptr)[0] == 1.0;
258      ptr += stride;
259   }
260
261   return space;
262}
263
264
265static void bind_inputs( struct gl_context *ctx,
266			 const struct gl_client_array *inputs[],
267			 GLint count,
268			 struct gl_buffer_object **bo,
269			 GLuint *nr_bo )
270{
271   TNLcontext *tnl = TNL_CONTEXT(ctx);
272   struct vertex_buffer *VB = &tnl->vb;
273   GLuint i;
274
275   /* Map all the VBOs
276    */
277   for (i = 0; i < VERT_ATTRIB_MAX; i++) {
278      const void *ptr;
279
280      if (inputs[i]->BufferObj->Name) {
281	 if (!inputs[i]->BufferObj->Pointer) {
282	    bo[*nr_bo] = inputs[i]->BufferObj;
283	    (*nr_bo)++;
284	    ctx->Driver.MapBufferRange(ctx, 0, inputs[i]->BufferObj->Size,
285				       GL_MAP_READ_BIT,
286				       inputs[i]->BufferObj);
287
288	    assert(inputs[i]->BufferObj->Pointer);
289	 }
290
291	 ptr = ADD_POINTERS(inputs[i]->BufferObj->Pointer,
292			    inputs[i]->Ptr);
293      }
294      else
295	 ptr = inputs[i]->Ptr;
296
297      /* Just make sure the array is floating point, otherwise convert to
298       * temporary storage.
299       *
300       * XXX: remove the GLvector4f type at some stage and just use
301       * client arrays.
302       */
303      _tnl_import_array(ctx, i, count, inputs[i], ptr);
304   }
305
306   /* We process only the vertices between min & max index:
307    */
308   VB->Count = count;
309
310   /* These should perhaps be part of _TNL_ATTRIB_* */
311   VB->BackfaceColorPtr = NULL;
312   VB->BackfaceIndexPtr = NULL;
313   VB->BackfaceSecondaryColorPtr = NULL;
314
315   /* Clipping and drawing code still requires this to be a packed
316    * array of ubytes which can be written into.  TODO: Fix and
317    * remove.
318    */
319   if (ctx->Polygon.FrontMode != GL_FILL ||
320       ctx->Polygon.BackMode != GL_FILL)
321   {
322      VB->EdgeFlag = _tnl_import_edgeflag( ctx,
323					   VB->AttribPtr[_TNL_ATTRIB_EDGEFLAG],
324					   VB->Count );
325   }
326   else {
327      /* the data previously pointed to by EdgeFlag may have been freed */
328      VB->EdgeFlag = NULL;
329   }
330}
331
332
333/* Translate indices to GLuints and store in VB->Elts.
334 */
335static void bind_indices( struct gl_context *ctx,
336			  const struct _mesa_index_buffer *ib,
337			  struct gl_buffer_object **bo,
338			  GLuint *nr_bo)
339{
340   TNLcontext *tnl = TNL_CONTEXT(ctx);
341   struct vertex_buffer *VB = &tnl->vb;
342   GLuint i;
343   const void *ptr;
344
345   if (!ib) {
346      VB->Elts = NULL;
347      return;
348   }
349
350   if (_mesa_is_bufferobj(ib->obj) && !_mesa_bufferobj_mapped(ib->obj)) {
351      /* if the buffer object isn't mapped yet, map it now */
352      bo[*nr_bo] = ib->obj;
353      (*nr_bo)++;
354      ptr = ctx->Driver.MapBufferRange(ctx, (GLsizeiptr) ib->ptr,
355                                       ib->count * vbo_sizeof_ib_type(ib->type),
356				       GL_MAP_READ_BIT, ib->obj);
357      assert(ib->obj->Pointer);
358   } else {
359      /* user-space elements, or buffer already mapped */
360      ptr = ADD_POINTERS(ib->obj->Pointer, ib->ptr);
361   }
362
363   if (ib->type == GL_UNSIGNED_INT && VB->Primitive[0].basevertex == 0) {
364      VB->Elts = (GLuint *) ptr;
365   }
366   else {
367      GLuint *elts = (GLuint *)get_space(ctx, ib->count * sizeof(GLuint));
368      VB->Elts = elts;
369
370      if (ib->type == GL_UNSIGNED_INT) {
371	 const GLuint *in = (GLuint *)ptr;
372	 for (i = 0; i < ib->count; i++)
373	    *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
374      }
375      else if (ib->type == GL_UNSIGNED_SHORT) {
376	 const GLushort *in = (GLushort *)ptr;
377	 for (i = 0; i < ib->count; i++)
378	    *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
379      }
380      else {
381	 const GLubyte *in = (GLubyte *)ptr;
382	 for (i = 0; i < ib->count; i++)
383	    *elts++ = (GLuint)(*in++) + VB->Primitive[0].basevertex;
384      }
385   }
386}
387
388static void bind_prims( struct gl_context *ctx,
389			const struct _mesa_prim *prim,
390			GLuint nr_prims )
391{
392   TNLcontext *tnl = TNL_CONTEXT(ctx);
393   struct vertex_buffer *VB = &tnl->vb;
394
395   VB->Primitive = prim;
396   VB->PrimitiveCount = nr_prims;
397}
398
399static void unmap_vbos( struct gl_context *ctx,
400			struct gl_buffer_object **bo,
401			GLuint nr_bo )
402{
403   GLuint i;
404   for (i = 0; i < nr_bo; i++) {
405      ctx->Driver.UnmapBuffer(ctx, bo[i]);
406   }
407}
408
409
410void _tnl_vbo_draw_prims(struct gl_context *ctx,
411			 const struct _mesa_prim *prim,
412			 GLuint nr_prims,
413			 const struct _mesa_index_buffer *ib,
414			 GLboolean index_bounds_valid,
415			 GLuint min_index,
416			 GLuint max_index,
417			 struct gl_transform_feedback_object *tfb_vertcount)
418{
419   const struct gl_client_array **arrays = ctx->Array._DrawArrays;
420
421   if (!index_bounds_valid)
422      vbo_get_minmax_indices(ctx, prim, ib, &min_index, &max_index, nr_prims);
423
424   _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
425}
426
427/* This is the main entrypoint into the slimmed-down software tnl
428 * module.  In a regular swtnl driver, this can be plugged straight
429 * into the vbo->Driver.DrawPrims() callback.
430 */
431void _tnl_draw_prims( struct gl_context *ctx,
432		      const struct gl_client_array *arrays[],
433		      const struct _mesa_prim *prim,
434		      GLuint nr_prims,
435		      const struct _mesa_index_buffer *ib,
436		      GLuint min_index,
437		      GLuint max_index)
438{
439   TNLcontext *tnl = TNL_CONTEXT(ctx);
440   const GLuint TEST_SPLIT = 0;
441   const GLint max = TEST_SPLIT ? 8 : tnl->vb.Size - MAX_CLIPPED_VERTICES;
442   GLint max_basevertex = prim->basevertex;
443   GLuint i;
444
445   /* Mesa core state should have been validated already */
446   assert(ctx->NewState == 0x0);
447
448   if (!_mesa_check_conditional_render(ctx))
449      return; /* don't draw */
450
451   for (i = 1; i < nr_prims; i++)
452      max_basevertex = MAX2(max_basevertex, prim[i].basevertex);
453
454   if (0)
455   {
456      printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
457      for (i = 0; i < nr_prims; i++)
458	 printf("prim %d: %s start %d count %d\n", i,
459		_mesa_lookup_enum_by_nr(prim[i].mode),
460		prim[i].start,
461		prim[i].count);
462   }
463
464   if (min_index) {
465      /* We always translate away calls with min_index != 0.
466       */
467      vbo_rebase_prims( ctx, arrays, prim, nr_prims, ib,
468			min_index, max_index,
469			_tnl_vbo_draw_prims );
470      return;
471   }
472   else if ((GLint)max_index + max_basevertex > max) {
473      /* The software TNL pipeline has a fixed amount of storage for
474       * vertices and it is necessary to split incoming drawing commands
475       * if they exceed that limit.
476       */
477      struct split_limits limits;
478      limits.max_verts = max;
479      limits.max_vb_size = ~0;
480      limits.max_indices = ~0;
481
482      /* This will split the buffers one way or another and
483       * recursively call back into this function.
484       */
485      vbo_split_prims( ctx, arrays, prim, nr_prims, ib,
486		       0, max_index + prim->basevertex,
487		       _tnl_vbo_draw_prims,
488		       &limits );
489   }
490   else {
491      /* May need to map a vertex buffer object for every attribute plus
492       * one for the index buffer.
493       */
494      struct gl_buffer_object *bo[VERT_ATTRIB_MAX + 1];
495      GLuint nr_bo = 0;
496      GLuint inst;
497
498      for (i = 0; i < nr_prims;) {
499	 GLuint this_nr_prims;
500
501	 /* Our SW TNL pipeline doesn't handle basevertex yet, so bind_indices
502	  * will rebase the elements to the basevertex, and we'll only
503	  * emit strings of prims with the same basevertex in one draw call.
504	  */
505	 for (this_nr_prims = 1; i + this_nr_prims < nr_prims;
506	      this_nr_prims++) {
507	    if (prim[i].basevertex != prim[i + this_nr_prims].basevertex)
508	       break;
509	 }
510
511         assert(prim[i].num_instances > 0);
512
513	 /* Binding inputs may imply mapping some vertex buffer objects.
514	  * They will need to be unmapped below.
515	  */
516         for (inst = 0; inst < prim[i].num_instances; inst++) {
517
518            bind_prims(ctx, &prim[i], this_nr_prims);
519            bind_inputs(ctx, arrays, max_index + prim[i].basevertex + 1,
520                        bo, &nr_bo);
521            bind_indices(ctx, ib, bo, &nr_bo);
522
523            tnl->CurInstance = inst;
524            TNL_CONTEXT(ctx)->Driver.RunPipeline(ctx);
525
526            unmap_vbos(ctx, bo, nr_bo);
527            free_space(ctx);
528         }
529
530	 i += this_nr_prims;
531      }
532   }
533}
534
535