1/**************************************************************************
2
3Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
4                     VA Linux Systems Inc., Fremont, California.
5
6All Rights Reserved.
7
8Permission is hereby granted, free of charge, to any person obtaining
9a copy of this software and associated documentation files (the
10"Software"), to deal in the Software without restriction, including
11without limitation the rights to use, copy, modify, merge, publish,
12distribute, sublicense, and/or sell copies of the Software, and to
13permit persons to whom the Software is furnished to do so, subject to
14the following conditions:
15
16The above copyright notice and this permission notice (including the
17next paragraph) shall be included in all copies or substantial
18portions of the Software.
19
20THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
23IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
24LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
25OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
26WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27
28**************************************************************************/
29
30/*
31 * Authors:
32 *   Keith Whitwell <keith@tungstengraphics.com>
33 */
34
35#include "main/glheader.h"
36#include "main/mtypes.h"
37#include "main/colormac.h"
38#include "main/enums.h"
39#include "main/imports.h"
40#include "main/macros.h"
41#include "main/simple_list.h"
42
43#include "math/m_xform.h"
44
45#include "swrast_setup/swrast_setup.h"
46
47#include "tnl/tnl.h"
48#include "tnl/t_context.h"
49#include "tnl/t_pipeline.h"
50
51#include "radeon_context.h"
52#include "radeon_ioctl.h"
53#include "radeon_state.h"
54#include "radeon_swtcl.h"
55#include "radeon_tcl.h"
56#include "radeon_debug.h"
57
58
59/* R100: xyzw, c0, c1/fog, stq[0..2]  = 4+1+1+3*3 = 15  right? */
60/* R200: xyzw, c0, c1/fog, strq[0..5] = 4+1+1+4*6 = 30 */
61#define RADEON_MAX_TNL_VERTEX_SIZE (15 * sizeof(GLfloat))	/* for mesa _tnl stage */
62
63/***********************************************************************
64 *                         Initialization
65 ***********************************************************************/
66
67#define EMIT_ATTR( ATTR, STYLE, F0 )					\
68do {									\
69   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = (ATTR);	\
70   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = (STYLE);	\
71   rmesa->radeon.swtcl.vertex_attr_count++;					\
72   fmt_0 |= F0;								\
73} while (0)
74
75#define EMIT_PAD( N )							\
76do {									\
77   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = 0;		\
78   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = EMIT_PAD;	\
79   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].offset = (N);		\
80   rmesa->radeon.swtcl.vertex_attr_count++;					\
81} while (0)
82
83static GLuint radeon_cp_vc_frmts[3][2] =
84{
85   { RADEON_CP_VC_FRMT_ST0, RADEON_CP_VC_FRMT_ST0 | RADEON_CP_VC_FRMT_Q0 },
86   { RADEON_CP_VC_FRMT_ST1, RADEON_CP_VC_FRMT_ST1 | RADEON_CP_VC_FRMT_Q1 },
87   { RADEON_CP_VC_FRMT_ST2, RADEON_CP_VC_FRMT_ST2 | RADEON_CP_VC_FRMT_Q2 },
88};
89
90static void radeonSetVertexFormat( struct gl_context *ctx )
91{
92   r100ContextPtr rmesa = R100_CONTEXT( ctx );
93   TNLcontext *tnl = TNL_CONTEXT(ctx);
94   struct vertex_buffer *VB = &tnl->vb;
95   GLbitfield64 index_bitset = tnl->render_inputs_bitset;
96   int fmt_0 = 0;
97   int offset = 0;
98
99   /* Important:
100    */
101   if ( VB->NdcPtr != NULL ) {
102      VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
103   }
104   else {
105      VB->AttribPtr[VERT_ATTRIB_POS] = VB->ClipPtr;
106   }
107
108   assert( VB->AttribPtr[VERT_ATTRIB_POS] != NULL );
109   rmesa->radeon.swtcl.vertex_attr_count = 0;
110
111   /* EMIT_ATTR's must be in order as they tell t_vertex.c how to
112    * build up a hardware vertex.
113    */
114   if ( !rmesa->swtcl.needproj ||
115        (index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX))) {
116      /* for projtex */
117      EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F,
118		 RADEON_CP_VC_FRMT_XY |	RADEON_CP_VC_FRMT_Z | RADEON_CP_VC_FRMT_W0 );
119      offset = 4;
120   }
121   else {
122      EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_3F,
123		 RADEON_CP_VC_FRMT_XY |	RADEON_CP_VC_FRMT_Z );
124      offset = 3;
125   }
126
127   rmesa->swtcl.coloroffset = offset;
128#if MESA_LITTLE_ENDIAN
129   EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_RGBA,
130	      RADEON_CP_VC_FRMT_PKCOLOR );
131#else
132   EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_ABGR,
133	      RADEON_CP_VC_FRMT_PKCOLOR );
134#endif
135   offset += 1;
136
137   rmesa->swtcl.specoffset = 0;
138   if (index_bitset &
139       (BITFIELD64_BIT(_TNL_ATTRIB_COLOR1) | BITFIELD64_BIT(_TNL_ATTRIB_FOG))) {
140
141#if MESA_LITTLE_ENDIAN
142      if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_COLOR1)) {
143	 rmesa->swtcl.specoffset = offset;
144	 EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_RGB,
145	 	    RADEON_CP_VC_FRMT_PKSPEC );
146      }
147      else {
148	 EMIT_PAD( 3 );
149      }
150
151      if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_FOG)) {
152	 EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F,
153	 	    RADEON_CP_VC_FRMT_PKSPEC );
154      }
155      else {
156	 EMIT_PAD( 1 );
157      }
158#else
159      if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_FOG)) {
160	 EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F,
161	 	    RADEON_CP_VC_FRMT_PKSPEC );
162      }
163      else {
164	 EMIT_PAD( 1 );
165      }
166
167      if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_COLOR1)) {
168	 rmesa->swtcl.specoffset = offset;
169	 EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR,
170	 	    RADEON_CP_VC_FRMT_PKSPEC );
171      }
172      else {
173	 EMIT_PAD( 3 );
174      }
175#endif
176   }
177
178   if (index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)) {
179      int i;
180
181      for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
182	 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_TEX(i))) {
183	    GLuint sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
184
185	    switch (sz) {
186	    case 1:
187	    case 2:
188	       EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_2F,
189			  radeon_cp_vc_frmts[i][0] );
190	       break;
191	    case 3:
192	       if (ctx->Texture.Unit[i]._ReallyEnabled & (TEXTURE_CUBE_BIT) ) {
193	           EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_3F,
194			      radeon_cp_vc_frmts[i][1] );
195               } else {
196	           EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_2F,
197			      radeon_cp_vc_frmts[i][0] );
198               }
199               break;
200	    case 4:
201	       if (ctx->Texture.Unit[i]._ReallyEnabled & (TEXTURE_CUBE_BIT) ) {
202		  EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_3F,
203			     radeon_cp_vc_frmts[i][1] );
204	       } else {
205		  EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_3F_XYW,
206			     radeon_cp_vc_frmts[i][1] );
207	       }
208	       break;
209	    default:
210	       continue;
211	    };
212	 }
213      }
214   }
215
216   if (rmesa->radeon.tnl_index_bitset != index_bitset ||
217       fmt_0 != rmesa->swtcl.vertex_format) {
218      RADEON_NEWPRIM(rmesa);
219      rmesa->swtcl.vertex_format = fmt_0;
220      rmesa->radeon.swtcl.vertex_size =
221	  _tnl_install_attrs( ctx,
222			      rmesa->radeon.swtcl.vertex_attrs,
223			      rmesa->radeon.swtcl.vertex_attr_count,
224			      NULL, 0 );
225      rmesa->radeon.swtcl.vertex_size /= 4;
226      rmesa->radeon.tnl_index_bitset = index_bitset;
227      radeon_print(RADEON_SWRENDER, RADEON_VERBOSE,
228	  "%s: vertex_size= %d floats\n",  __FUNCTION__, rmesa->radeon.swtcl.vertex_size);
229   }
230}
231
232static void radeon_predict_emit_size( r100ContextPtr rmesa )
233{
234
235    if (!rmesa->radeon.swtcl.emit_prediction) {
236        const int state_size = radeonCountStateEmitSize( &rmesa->radeon );
237        const int scissor_size = 8;
238        const int prims_size = 8;
239        const int vertex_size = 7;
240
241        if (rcommonEnsureCmdBufSpace(&rmesa->radeon,
242                    state_size +
243                    (scissor_size + prims_size + vertex_size),
244                    __FUNCTION__))
245            rmesa->radeon.swtcl.emit_prediction = radeonCountStateEmitSize( &rmesa->radeon );
246        else
247            rmesa->radeon.swtcl.emit_prediction = state_size;
248        rmesa->radeon.swtcl.emit_prediction += scissor_size + prims_size + vertex_size
249            + rmesa->radeon.cmdbuf.cs->cdw;
250    }
251}
252
253static void radeonRenderStart( struct gl_context *ctx )
254{
255    r100ContextPtr rmesa = R100_CONTEXT( ctx );
256
257    radeonSetVertexFormat( ctx );
258
259    if (rmesa->radeon.dma.flush != 0 &&
260            rmesa->radeon.dma.flush != rcommon_flush_last_swtcl_prim)
261        rmesa->radeon.dma.flush( ctx );
262}
263
264
265/**
266 * Set vertex state for SW TCL.  The primary purpose of this function is to
267 * determine in advance whether or not the hardware can / should do the
268 * projection divide or Mesa should do it.
269 */
270void radeonChooseVertexState( struct gl_context *ctx )
271{
272   r100ContextPtr rmesa = R100_CONTEXT( ctx );
273   TNLcontext *tnl = TNL_CONTEXT(ctx);
274
275   GLuint se_coord_fmt = rmesa->hw.set.cmd[SET_SE_COORDFMT];
276
277   se_coord_fmt &= ~(RADEON_VTX_XY_PRE_MULT_1_OVER_W0 |
278		     RADEON_VTX_Z_PRE_MULT_1_OVER_W0 |
279		     RADEON_VTX_W0_IS_NOT_1_OVER_W0);
280
281   /* We must ensure that we don't do _tnl_need_projected_coords while in a
282    * rasterization fallback.  As this function will be called again when we
283    * leave a rasterization fallback, we can just skip it for now.
284    */
285   if (rmesa->radeon.Fallback != 0)
286      return;
287
288   /* HW perspective divide is a win, but tiny vertex formats are a
289    * bigger one.
290    */
291
292   if ((0 == (tnl->render_inputs_bitset &
293        (BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)
294         | BITFIELD64_BIT(_TNL_ATTRIB_COLOR1))))
295        || (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {
296      rmesa->swtcl.needproj = GL_TRUE;
297      se_coord_fmt |= (RADEON_VTX_XY_PRE_MULT_1_OVER_W0 |
298		      RADEON_VTX_Z_PRE_MULT_1_OVER_W0);
299   }
300   else {
301      rmesa->swtcl.needproj = GL_FALSE;
302      se_coord_fmt |= (RADEON_VTX_W0_IS_NOT_1_OVER_W0);
303   }
304
305   _tnl_need_projected_coords( ctx, rmesa->swtcl.needproj );
306
307   if ( se_coord_fmt != rmesa->hw.set.cmd[SET_SE_COORDFMT] ) {
308      RADEON_STATECHANGE( rmesa, set );
309      rmesa->hw.set.cmd[SET_SE_COORDFMT] = se_coord_fmt;
310   }
311}
312
313void r100_swtcl_flush(struct gl_context *ctx, uint32_t current_offset)
314{
315   r100ContextPtr rmesa = R100_CONTEXT(ctx);
316
317
318
319   radeonEmitState(&rmesa->radeon);
320   radeonEmitVertexAOS( rmesa,
321			rmesa->radeon.swtcl.vertex_size,
322			rmesa->radeon.swtcl.bo,
323			current_offset);
324
325
326   radeonEmitVbufPrim( rmesa,
327		       rmesa->swtcl.vertex_format,
328		       rmesa->radeon.swtcl.hw_primitive,
329		       rmesa->radeon.swtcl.numverts);
330   if ( rmesa->radeon.swtcl.emit_prediction < rmesa->radeon.cmdbuf.cs->cdw )
331     WARN_ONCE("Rendering was %d commands larger than predicted size."
332	 " We might overflow  command buffer.\n",
333	 rmesa->radeon.cmdbuf.cs->cdw - rmesa->radeon.swtcl.emit_prediction );
334
335
336   rmesa->radeon.swtcl.emit_prediction = 0;
337
338}
339
340/*
341 * Render unclipped vertex buffers by emitting vertices directly to
342 * dma buffers.  Use strip/fan hardware primitives where possible.
343 * Try to simulate missing primitives with indexed vertices.
344 */
345#define HAVE_POINTS      1
346#define HAVE_LINES       1
347#define HAVE_LINE_STRIPS 1
348#define HAVE_TRIANGLES   1
349#define HAVE_TRI_STRIPS  1
350#define HAVE_TRI_STRIP_1 0
351#define HAVE_TRI_FANS    1
352#define HAVE_QUADS       0
353#define HAVE_QUAD_STRIPS 0
354#define HAVE_POLYGONS    0
355/* \todo: is it possible to make "ELTS" work with t_vertex code ? */
356#define HAVE_ELTS        0
357
358static const GLuint hw_prim[GL_POLYGON+1] = {
359   RADEON_CP_VC_CNTL_PRIM_TYPE_POINT,
360   RADEON_CP_VC_CNTL_PRIM_TYPE_LINE,
361   0,
362   RADEON_CP_VC_CNTL_PRIM_TYPE_LINE_STRIP,
363   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
364   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_STRIP,
365   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN,
366   0,
367   0,
368   0
369};
370
371static INLINE void
372radeonDmaPrimitive( r100ContextPtr rmesa, GLenum prim )
373{
374   RADEON_NEWPRIM( rmesa );
375   rmesa->radeon.swtcl.hw_primitive = hw_prim[prim];
376   //   assert(rmesa->radeon.dma.current.ptr == rmesa->radeon.dma.current.start);
377}
378
379static void* radeon_alloc_verts( r100ContextPtr rmesa , GLuint nr, GLuint size )
380{
381   void *rv;
382   do {
383     radeon_predict_emit_size( rmesa );
384     rv = rcommonAllocDmaLowVerts( &rmesa->radeon, nr, size );
385   } while (!rv);
386   return rv;
387}
388
389#define LOCAL_VARS r100ContextPtr rmesa = R100_CONTEXT(ctx)
390#define INIT( prim ) radeonDmaPrimitive( rmesa, prim )
391#define FLUSH()  RADEON_NEWPRIM( rmesa )
392#define GET_CURRENT_VB_MAX_VERTS()					10\
393//  (((int)rmesa->radeon.dma.current.end - (int)rmesa->radeon.dma.current.ptr) / (rmesa->radeon.swtcl.vertex_size*4))
394#define GET_SUBSEQUENT_VB_MAX_VERTS() \
395  ((RADEON_BUFFER_SIZE) / (rmesa->radeon.swtcl.vertex_size*4))
396#define ALLOC_VERTS( nr ) radeon_alloc_verts( rmesa, nr, rmesa->radeon.swtcl.vertex_size * 4 )
397#define EMIT_VERTS( ctx, j, nr, buf ) \
398  _tnl_emit_vertices_to_buffer(ctx, j, (j)+(nr), buf)
399
400#define TAG(x) radeon_dma_##x
401#include "tnl_dd/t_dd_dmatmp.h"
402
403
404/**********************************************************************/
405/*                          Render pipeline stage                     */
406/**********************************************************************/
407
408
409static GLboolean radeon_run_render( struct gl_context *ctx,
410				    struct tnl_pipeline_stage *stage )
411{
412   r100ContextPtr rmesa = R100_CONTEXT(ctx);
413   TNLcontext *tnl = TNL_CONTEXT(ctx);
414   struct vertex_buffer *VB = &tnl->vb;
415   tnl_render_func *tab = TAG(render_tab_verts);
416   GLuint i;
417
418   if (rmesa->radeon.swtcl.RenderIndex != 0 ||
419       !radeon_dma_validate_render( ctx, VB ))
420      return GL_TRUE;
421
422   radeon_prepare_render(&rmesa->radeon);
423   if (rmesa->radeon.NewGLState)
424      radeonValidateState( ctx );
425
426   tnl->Driver.Render.Start( ctx );
427
428   for (i = 0 ; i < VB->PrimitiveCount ; i++)
429   {
430      GLuint prim = VB->Primitive[i].mode;
431      GLuint start = VB->Primitive[i].start;
432      GLuint length = VB->Primitive[i].count;
433
434      if (!length)
435	 continue;
436
437      radeon_print(RADEON_SWRENDER, RADEON_NORMAL,
438	  "radeon_render.c: prim %s %d..%d\n",
439		 _mesa_lookup_enum_by_nr(prim & PRIM_MODE_MASK),
440		 start, start+length);
441
442      if (length)
443	 tab[prim & PRIM_MODE_MASK]( ctx, start, start + length, prim );
444   }
445
446   tnl->Driver.Render.Finish( ctx );
447
448   return GL_FALSE;		/* finished the pipe */
449}
450
451
452
453const struct tnl_pipeline_stage _radeon_render_stage =
454{
455   "radeon render",
456   NULL,
457   NULL,
458   NULL,
459   NULL,
460   radeon_run_render		/* run */
461};
462
463
464/**************************************************************************/
465
466
467static const GLuint reduced_hw_prim[GL_POLYGON+1] = {
468   RADEON_CP_VC_CNTL_PRIM_TYPE_POINT,
469   RADEON_CP_VC_CNTL_PRIM_TYPE_LINE,
470   RADEON_CP_VC_CNTL_PRIM_TYPE_LINE,
471   RADEON_CP_VC_CNTL_PRIM_TYPE_LINE,
472   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
473   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
474   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
475   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
476   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
477   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST
478};
479
480static void radeonRasterPrimitive( struct gl_context *ctx, GLuint hwprim );
481static void radeonRenderPrimitive( struct gl_context *ctx, GLenum prim );
482static void radeonResetLineStipple( struct gl_context *ctx );
483
484
485/***********************************************************************
486 *                    Emit primitives as inline vertices               *
487 ***********************************************************************/
488
489#undef LOCAL_VARS
490#undef ALLOC_VERTS
491#define CTX_ARG r100ContextPtr rmesa
492#define GET_VERTEX_DWORDS() rmesa->radeon.swtcl.vertex_size
493#define ALLOC_VERTS( n, size ) radeon_alloc_verts( rmesa, n, (size) * 4 )
494#undef LOCAL_VARS
495#define LOCAL_VARS						\
496   r100ContextPtr rmesa = R100_CONTEXT(ctx);		\
497   const char *radeonverts = (char *)rmesa->radeon.swtcl.verts;
498#define VERT(x) (radeonVertex *)(radeonverts + ((x) * (vertsize) * sizeof(int)))
499#define VERTEX radeonVertex
500#undef TAG
501#define TAG(x) radeon_##x
502#include "tnl_dd/t_dd_triemit.h"
503
504
505/***********************************************************************
506 *          Macros for t_dd_tritmp.h to draw basic primitives          *
507 ***********************************************************************/
508
509#define QUAD( a, b, c, d ) radeon_quad( rmesa, a, b, c, d )
510#define TRI( a, b, c )     radeon_triangle( rmesa, a, b, c )
511#define LINE( a, b )       radeon_line( rmesa, a, b )
512#define POINT( a )         radeon_point( rmesa, a )
513
514/***********************************************************************
515 *              Build render functions from dd templates               *
516 ***********************************************************************/
517
518#define RADEON_TWOSIDE_BIT	0x01
519#define RADEON_UNFILLED_BIT	0x02
520#define RADEON_MAX_TRIFUNC	0x04
521
522
523static struct {
524   tnl_points_func	        points;
525   tnl_line_func		line;
526   tnl_triangle_func	triangle;
527   tnl_quad_func		quad;
528} rast_tab[RADEON_MAX_TRIFUNC];
529
530
531#define DO_FALLBACK  0
532#define DO_OFFSET    0
533#define DO_UNFILLED (IND & RADEON_UNFILLED_BIT)
534#define DO_TWOSIDE  (IND & RADEON_TWOSIDE_BIT)
535#define DO_FLAT      0
536#define DO_TRI       1
537#define DO_QUAD      1
538#define DO_LINE      1
539#define DO_POINTS    1
540#define DO_FULL_QUAD 1
541
542#define HAVE_SPEC   1
543#define HAVE_BACK_COLORS  0
544#define HAVE_HW_FLATSHADE 1
545#define TAB rast_tab
546
547#define DEPTH_SCALE 1.0
548#define UNFILLED_TRI unfilled_tri
549#define UNFILLED_QUAD unfilled_quad
550#define VERT_X(_v) _v->v.x
551#define VERT_Y(_v) _v->v.y
552#define VERT_Z(_v) _v->v.z
553#define AREA_IS_CCW( a ) (a < 0)
554#define GET_VERTEX(e) (rmesa->radeon.swtcl.verts + ((e) * rmesa->radeon.swtcl.vertex_size * sizeof(int)))
555
556#define VERT_SET_RGBA( v, c )  					\
557do {								\
558   radeon_color_t *color = (radeon_color_t *)&((v)->ui[coloroffset]);	\
559   UNCLAMPED_FLOAT_TO_UBYTE(color->red, (c)[0]);		\
560   UNCLAMPED_FLOAT_TO_UBYTE(color->green, (c)[1]);		\
561   UNCLAMPED_FLOAT_TO_UBYTE(color->blue, (c)[2]);		\
562   UNCLAMPED_FLOAT_TO_UBYTE(color->alpha, (c)[3]);		\
563} while (0)
564
565#define VERT_COPY_RGBA( v0, v1 ) v0->ui[coloroffset] = v1->ui[coloroffset]
566
567#define VERT_SET_SPEC( v, c )					\
568do {								\
569   if (specoffset) {						\
570      radeon_color_t *spec = (radeon_color_t *)&((v)->ui[specoffset]);	\
571      UNCLAMPED_FLOAT_TO_UBYTE(spec->red, (c)[0]);	\
572      UNCLAMPED_FLOAT_TO_UBYTE(spec->green, (c)[1]);	\
573      UNCLAMPED_FLOAT_TO_UBYTE(spec->blue, (c)[2]);	\
574   }								\
575} while (0)
576#define VERT_COPY_SPEC( v0, v1 )			\
577do {							\
578   if (specoffset) {					\
579      radeon_color_t *spec0 = (radeon_color_t *)&((v0)->ui[specoffset]);	\
580      radeon_color_t *spec1 = (radeon_color_t *)&((v1)->ui[specoffset]);	\
581      spec0->red   = spec1->red;	\
582      spec0->green = spec1->green;	\
583      spec0->blue  = spec1->blue; 	\
584   }							\
585} while (0)
586
587/* These don't need LE32_TO_CPU() as they used to save and restore
588 * colors which are already in the correct format.
589 */
590#define VERT_SAVE_RGBA( idx )    color[idx] = v[idx]->ui[coloroffset]
591#define VERT_RESTORE_RGBA( idx ) v[idx]->ui[coloroffset] = color[idx]
592#define VERT_SAVE_SPEC( idx )    if (specoffset) spec[idx] = v[idx]->ui[specoffset]
593#define VERT_RESTORE_SPEC( idx ) if (specoffset) v[idx]->ui[specoffset] = spec[idx]
594
595#undef LOCAL_VARS
596#undef TAG
597#undef INIT
598
599#define LOCAL_VARS(n)							\
600   r100ContextPtr rmesa = R100_CONTEXT(ctx);			\
601   GLuint color[n] = {0}, spec[n] = {0};						\
602   GLuint coloroffset = rmesa->swtcl.coloroffset;	\
603   GLuint specoffset = rmesa->swtcl.specoffset;			\
604   (void) color; (void) spec; (void) coloroffset; (void) specoffset;
605
606/***********************************************************************
607 *                Helpers for rendering unfilled primitives            *
608 ***********************************************************************/
609
610#define RASTERIZE(x) radeonRasterPrimitive( ctx, reduced_hw_prim[x] )
611#define RENDER_PRIMITIVE rmesa->radeon.swtcl.render_primitive
612#undef TAG
613#define TAG(x) x
614#include "tnl_dd/t_dd_unfilled.h"
615#undef IND
616
617
618/***********************************************************************
619 *                      Generate GL render functions                   *
620 ***********************************************************************/
621
622
623#define IND (0)
624#define TAG(x) x
625#include "tnl_dd/t_dd_tritmp.h"
626
627#define IND (RADEON_TWOSIDE_BIT)
628#define TAG(x) x##_twoside
629#include "tnl_dd/t_dd_tritmp.h"
630
631#define IND (RADEON_UNFILLED_BIT)
632#define TAG(x) x##_unfilled
633#include "tnl_dd/t_dd_tritmp.h"
634
635#define IND (RADEON_TWOSIDE_BIT|RADEON_UNFILLED_BIT)
636#define TAG(x) x##_twoside_unfilled
637#include "tnl_dd/t_dd_tritmp.h"
638
639
640static void init_rast_tab( void )
641{
642   init();
643   init_twoside();
644   init_unfilled();
645   init_twoside_unfilled();
646}
647
648/**********************************************************************/
649/*               Render unclipped begin/end objects                   */
650/**********************************************************************/
651
652#define RENDER_POINTS( start, count )		\
653   for ( ; start < count ; start++)		\
654      radeon_point( rmesa, VERT(start) )
655#define RENDER_LINE( v0, v1 ) \
656   radeon_line( rmesa, VERT(v0), VERT(v1) )
657#define RENDER_TRI( v0, v1, v2 )  \
658   radeon_triangle( rmesa, VERT(v0), VERT(v1), VERT(v2) )
659#define RENDER_QUAD( v0, v1, v2, v3 ) \
660   radeon_quad( rmesa, VERT(v0), VERT(v1), VERT(v2), VERT(v3) )
661#undef INIT
662#define INIT(x) do {					\
663   radeonRenderPrimitive( ctx, x );			\
664} while (0)
665#undef LOCAL_VARS
666#define LOCAL_VARS						\
667   r100ContextPtr rmesa = R100_CONTEXT(ctx);		\
668   const GLuint vertsize = rmesa->radeon.swtcl.vertex_size;		\
669   const char *radeonverts = (char *)rmesa->radeon.swtcl.verts;		\
670   const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts;	\
671   const GLboolean stipple = ctx->Line.StippleFlag;		\
672   (void) elt; (void) stipple;
673#define RESET_STIPPLE	if ( stipple ) radeonResetLineStipple( ctx );
674#define RESET_OCCLUSION
675#define PRESERVE_VB_DEFS
676#define ELT(x) (x)
677#define TAG(x) radeon_##x##_verts
678#include "tnl/t_vb_rendertmp.h"
679#undef ELT
680#undef TAG
681#define TAG(x) radeon_##x##_elts
682#define ELT(x) elt[x]
683#include "tnl/t_vb_rendertmp.h"
684
685
686
687/**********************************************************************/
688/*                    Choose render functions                         */
689/**********************************************************************/
690
691void radeonChooseRenderState( struct gl_context *ctx )
692{
693   TNLcontext *tnl = TNL_CONTEXT(ctx);
694   r100ContextPtr rmesa = R100_CONTEXT(ctx);
695   GLuint index = 0;
696   GLuint flags = ctx->_TriangleCaps;
697
698   if (!rmesa->radeon.TclFallback || rmesa->radeon.Fallback)
699      return;
700
701   if (flags & DD_TRI_LIGHT_TWOSIDE) index |= RADEON_TWOSIDE_BIT;
702   if (flags & DD_TRI_UNFILLED)      index |= RADEON_UNFILLED_BIT;
703
704   if (index != rmesa->radeon.swtcl.RenderIndex) {
705      tnl->Driver.Render.Points = rast_tab[index].points;
706      tnl->Driver.Render.Line = rast_tab[index].line;
707      tnl->Driver.Render.ClippedLine = rast_tab[index].line;
708      tnl->Driver.Render.Triangle = rast_tab[index].triangle;
709      tnl->Driver.Render.Quad = rast_tab[index].quad;
710
711      if (index == 0) {
712	 tnl->Driver.Render.PrimTabVerts = radeon_render_tab_verts;
713	 tnl->Driver.Render.PrimTabElts = radeon_render_tab_elts;
714	 tnl->Driver.Render.ClippedPolygon = radeon_fast_clipped_poly;
715      } else {
716	 tnl->Driver.Render.PrimTabVerts = _tnl_render_tab_verts;
717	 tnl->Driver.Render.PrimTabElts = _tnl_render_tab_elts;
718	 tnl->Driver.Render.ClippedPolygon = _tnl_RenderClippedPolygon;
719      }
720
721      rmesa->radeon.swtcl.RenderIndex = index;
722   }
723}
724
725
726/**********************************************************************/
727/*                 High level hooks for t_vb_render.c                 */
728/**********************************************************************/
729
730
731static void radeonRasterPrimitive( struct gl_context *ctx, GLuint hwprim )
732{
733   r100ContextPtr rmesa = R100_CONTEXT(ctx);
734
735   if (rmesa->radeon.swtcl.hw_primitive != hwprim) {
736      RADEON_NEWPRIM( rmesa );
737      rmesa->radeon.swtcl.hw_primitive = hwprim;
738   }
739}
740
741static void radeonRenderPrimitive( struct gl_context *ctx, GLenum prim )
742{
743   r100ContextPtr rmesa = R100_CONTEXT(ctx);
744   rmesa->radeon.swtcl.render_primitive = prim;
745   if (prim < GL_TRIANGLES || !(ctx->_TriangleCaps & DD_TRI_UNFILLED))
746      radeonRasterPrimitive( ctx, reduced_hw_prim[prim] );
747}
748
749static void radeonRenderFinish( struct gl_context *ctx )
750{
751}
752
753static void radeonResetLineStipple( struct gl_context *ctx )
754{
755   r100ContextPtr rmesa = R100_CONTEXT(ctx);
756   RADEON_STATECHANGE( rmesa, lin );
757}
758
759
760/**********************************************************************/
761/*           Transition to/from hardware rasterization.               */
762/**********************************************************************/
763
764static const char * const fallbackStrings[] = {
765   "Texture mode",
766   "glDrawBuffer(GL_FRONT_AND_BACK)",
767   "glEnable(GL_STENCIL) without hw stencil buffer",
768   "glRenderMode(selection or feedback)",
769   "glBlendEquation",
770   "glBlendFunc",
771   "RADEON_NO_RAST",
772   "Mixing GL_CLAMP_TO_BORDER and GL_CLAMP (or GL_MIRROR_CLAMP_ATI)"
773};
774
775
776static const char *getFallbackString(GLuint bit)
777{
778   int i = 0;
779   while (bit > 1) {
780      i++;
781      bit >>= 1;
782   }
783   return fallbackStrings[i];
784}
785
786
787void radeonFallback( struct gl_context *ctx, GLuint bit, GLboolean mode )
788{
789   r100ContextPtr rmesa = R100_CONTEXT(ctx);
790   TNLcontext *tnl = TNL_CONTEXT(ctx);
791   GLuint oldfallback = rmesa->radeon.Fallback;
792
793   if (mode) {
794      rmesa->radeon.Fallback |= bit;
795      if (oldfallback == 0) {
796	 radeon_firevertices(&rmesa->radeon);
797	 TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_RASTER, GL_TRUE );
798	 _swsetup_Wakeup( ctx );
799	 rmesa->radeon.swtcl.RenderIndex = ~0;
800         if (RADEON_DEBUG & RADEON_FALLBACKS) {
801            fprintf(stderr, "Radeon begin rasterization fallback: 0x%x %s\n",
802                    bit, getFallbackString(bit));
803         }
804      }
805   }
806   else {
807      rmesa->radeon.Fallback &= ~bit;
808      if (oldfallback == bit) {
809	 _swrast_flush( ctx );
810	 tnl->Driver.Render.Start = radeonRenderStart;
811	 tnl->Driver.Render.PrimitiveNotify = radeonRenderPrimitive;
812	 tnl->Driver.Render.Finish = radeonRenderFinish;
813
814	 tnl->Driver.Render.BuildVertices = _tnl_build_vertices;
815	 tnl->Driver.Render.CopyPV = _tnl_copy_pv;
816	 tnl->Driver.Render.Interp = _tnl_interp;
817
818	 tnl->Driver.Render.ResetLineStipple = radeonResetLineStipple;
819	 TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_RASTER, GL_FALSE );
820	 if (rmesa->radeon.TclFallback) {
821	    /* These are already done if rmesa->radeon.TclFallback goes to
822	     * zero above. But not if it doesn't (RADEON_NO_TCL for
823	     * example?)
824	     */
825	    _tnl_invalidate_vertex_state( ctx, ~0 );
826	    _tnl_invalidate_vertices( ctx, ~0 );
827	    rmesa->radeon.tnl_index_bitset = 0;
828	    radeonChooseVertexState( ctx );
829	    radeonChooseRenderState( ctx );
830	 }
831         if (RADEON_DEBUG & RADEON_FALLBACKS) {
832            fprintf(stderr, "Radeon end rasterization fallback: 0x%x %s\n",
833                    bit, getFallbackString(bit));
834         }
835      }
836   }
837}
838
839
840/**********************************************************************/
841/*                            Initialization.                         */
842/**********************************************************************/
843
844void radeonInitSwtcl( struct gl_context *ctx )
845{
846   TNLcontext *tnl = TNL_CONTEXT(ctx);
847   r100ContextPtr rmesa = R100_CONTEXT(ctx);
848   static int firsttime = 1;
849
850   if (firsttime) {
851      init_rast_tab();
852      firsttime = 0;
853   }
854   rmesa->radeon.swtcl.emit_prediction = 0;
855
856   tnl->Driver.Render.Start = radeonRenderStart;
857   tnl->Driver.Render.Finish = radeonRenderFinish;
858   tnl->Driver.Render.PrimitiveNotify = radeonRenderPrimitive;
859   tnl->Driver.Render.ResetLineStipple = radeonResetLineStipple;
860   tnl->Driver.Render.BuildVertices = _tnl_build_vertices;
861   tnl->Driver.Render.CopyPV = _tnl_copy_pv;
862   tnl->Driver.Render.Interp = _tnl_interp;
863
864   _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12,
865		       RADEON_MAX_TNL_VERTEX_SIZE);
866
867   rmesa->radeon.swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf;
868   rmesa->radeon.swtcl.RenderIndex = ~0;
869   rmesa->radeon.swtcl.render_primitive = GL_TRIANGLES;
870   rmesa->radeon.swtcl.hw_primitive = 0;
871}
872
873