projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Mipmapping and other texture filters now work.
[mesa.git] / src / mesa / drivers / dri / r300 / r300_render.c
1 /**************************************************************************
2
3 Copyright (C) 2004 Nicolai Haehnle.
4
5 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 on the rights to use, copy, modify, merge, publish, distribute, sub
11 license, and/or sell copies of the Software, and to permit persons to whom
12 the Software is furnished to do so, subject to the following conditions:
13
14 The above copyright notice and this permission notice (including the next
15 paragraph) shall be included in all copies or substantial portions of the
16 Software.
17
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21 ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
22 DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23 OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24 USE OR OTHER DEALINGS IN THE SOFTWARE.
25
26 **************************************************************************/
27
28 /*
29 * Authors:
30 * Nicolai Haehnle <prefect_@gmx.net>
31 */
32
33 #include "glheader.h"
34 #include "state.h"
35 #include "imports.h"
36 #include "enums.h"
37 #include "macros.h"
38 #include "context.h"
39 #include "dd.h"
40 #include "simple_list.h"
41
42 #include "api_arrayelt.h"
43 #include "swrast/swrast.h"
44 #include "swrast_setup/swrast_setup.h"
45 #include "array_cache/acache.h"
46 #include "tnl/tnl.h"
47
48 #include "radeon_reg.h"
49 #include "radeon_macros.h"
50 #include "radeon_ioctl.h"
51 #include "radeon_state.h"
52 #include "r300_context.h"
53 #include "r300_ioctl.h"
54 #include "r300_state.h"
55 #include "r300_reg.h"
56 #include "r300_program.h"
57 #include "r300_tex.h"
58
59 #include "r300_emit.h"
60
61 /**********************************************************************
62 * Hardware rasterization
63 *
64 * When we fell back to software TCL, we still try to use the
65 * rasterization hardware for rendering.
66 **********************************************************************/
67
68 static int r300_get_primitive_type(r300ContextPtr rmesa,
69 GLcontext *ctx,
70 int start,
71 int end,
72 int prim)
73 {
74 TNLcontext *tnl = TNL_CONTEXT(ctx);
75 struct vertex_buffer *VB = &tnl->vb;
76 GLuint i;
77 int type=-1, min_vertices=0;
78 char *name="UNKNOWN";
79
80 if(end<=start)return -1; /* do we need to watch for this ? */
81
82 switch (prim & PRIM_MODE_MASK) {
83 case GL_POINTS:
84 name="P";
85 type=R300_VAP_VF_CNTL__PRIM_POINTS;
86 min_vertices=1;
87 break;
88 case GL_LINES:
89 name="L";
90 type=R300_VAP_VF_CNTL__PRIM_LINES;
91 min_vertices=2;
92 break;
93 case GL_LINE_STRIP:
94 name="LS";
95 type=R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
96 min_vertices=2;
97 break;
98 case GL_LINE_LOOP:
99 name="LL";
100 min_vertices=2;
101 return -1;
102 break;
103 case GL_TRIANGLES:
104 name="T";
105 type=R300_VAP_VF_CNTL__PRIM_TRIANGLES;
106 min_vertices=3;
107 break;
108 case GL_TRIANGLE_STRIP:
109 name="TS";
110 type=R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
111 min_vertices=3;
112 break;
113 case GL_TRIANGLE_FAN:
114 name="TF";
115 type=R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
116 min_vertices=3;
117 break;
118 case GL_QUADS:
119 name="Q";
120 type=R300_VAP_VF_CNTL__PRIM_QUADS;
121 min_vertices=4;
122 break;
123 case GL_QUAD_STRIP:
124 name="QS";
125 type=R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
126 min_vertices=4;
127 break;
128 case GL_POLYGON:
129 name="P";
130 type=R300_VAP_VF_CNTL__PRIM_POLYGON;
131 min_vertices=3;
132 break;
133 default:
134 fprintf(stderr, "%s:%s Do not know how to handle primitive %02x - help me !\n",
135 __FILE__, __FUNCTION__,
136 prim & PRIM_MODE_MASK);
137 return -1;
138 break;
139 }
140 #if 0
141 fprintf(stderr, "[%d-%d]%s ", start, end, name);
142 #endif
143 if(start+min_vertices>end){
144 static int warn_once=1;
145 if(warn_once){
146 fprintf(stderr, "%s:%s Not enough vertices to draw primitive %02x - help me !\n",
147 __FILE__, __FUNCTION__,
148 prim & PRIM_MODE_MASK);
149 warn_once=0;
150 }
151 return -1;
152 }
153 return type;
154 }
155
156 /* This function compiles GL context into state registers that
157 describe data routing inside of R300 pipeline.
158
159 In particular, it programs input_route, output_vtx_fmt, texture
160 unit configuration and gb_output_vtx_fmt
161
162 This function encompasses setup_AOS() from r300_lib.c
163 */
164
165
166
167
168 /* Immediate implementation - vertex data is sent via command stream */
169
170 static GLfloat default_vector[4]={0.0, 0.0, 0.0, 1.0};
171
172 #define output_vector(v, i) \
173 { \
174 int _i; \
175 for(_i=0;_i<v->size;_i++){ \
176 efloat(VEC_ELT(v, GLfloat, i)[_i]); \
177 } \
178 for(_i=v->size;_i<4;_i++){ \
179 efloat(default_vector[_i]); \
180 } \
181 }
182
183 /* Immediate implementation - vertex data is sent via command stream */
184
185 static void r300_render_immediate_primitive(r300ContextPtr rmesa,
186 GLcontext *ctx,
187 int start,
188 int end,
189 int prim)
190 {
191 TNLcontext *tnl = TNL_CONTEXT(ctx);
192 struct vertex_buffer *VB = &tnl->vb;
193 GLuint i;
194 int k, type;
195 LOCAL_VARS
196
197 type=r300_get_primitive_type(rmesa, ctx, start, end, prim);
198
199 #if 0
200 fprintf(stderr,"ObjPtr: size=%d stride=%d\n",
201 VB->ObjPtr->size, VB->ObjPtr->stride);
202 fprintf(stderr,"ColorPtr[0]: size=%d stride=%d\n",
203 VB->ColorPtr[0]->size, VB->ColorPtr[0]->stride);
204 fprintf(stderr,"TexCoordPtr[0]: size=%d stride=%d\n",
205 VB->TexCoordPtr[0]->size, VB->TexCoordPtr[0]->stride);
206 #endif
207
208 if(type<0)return;
209
210 /* A packet cannot have more than 16383 data words.. */
211 if(((end-start)*8+4*rmesa->state.texture.tc_count)>16380){
212 fprintf(stderr, "%s:%s: Too many vertices to paint. Fix me !\n");
213 return;
214 }
215
216 start_immediate_packet(end-start, type, 8+4*rmesa->state.texture.tc_count);
217
218 for(i=start;i<end;i++){
219 #if 0
220 fprintf(stderr, "* (%f %f %f %f) (%f %f %f %f)\n",
221 VEC_ELT(VB->ObjPtr, GLfloat, i)[0],
222 VEC_ELT(VB->ObjPtr, GLfloat, i)[1],
223 VEC_ELT(VB->ObjPtr, GLfloat, i)[2],
224 VEC_ELT(VB->ObjPtr, GLfloat, i)[3],
225
226 VEC_ELT(VB->ColorPtr[0], GLfloat, i)[0],
227 VEC_ELT(VB->ColorPtr[0], GLfloat, i)[1],
228 VEC_ELT(VB->ColorPtr[0], GLfloat, i)[2],
229 VEC_ELT(VB->ColorPtr[0], GLfloat, i)[3]
230 );
231 #endif
232
233
234 /* coordinates */
235 output_vector(VB->ObjPtr, i);
236
237 /* color components */
238 output_vector(VB->ColorPtr[0], i);
239
240 /* texture coordinates */
241 for(k=0;k < ctx->Const.MaxTextureUnits;k++)
242 if(ctx->Texture.Unit[k].Enabled)
243 output_vector(VB->TexCoordPtr[k], i);
244 }
245
246 }
247
248
249 static GLboolean r300_run_immediate_render(GLcontext *ctx,
250 struct tnl_pipeline_stage *stage)
251 {
252 r300ContextPtr rmesa = R300_CONTEXT(ctx);
253 TNLcontext *tnl = TNL_CONTEXT(ctx);
254 struct vertex_buffer *VB = &tnl->vb;
255 GLuint i;
256 /* Only do 2d textures */
257 struct gl_texture_object *to=ctx->Texture.Unit[0].Current2D;
258 r300TexObjPtr t=to->DriverData;
259 LOCAL_VARS
260
261
262 /* Update texture state - needs to be done only when actually changed..
263 All the time for now.. */
264
265
266 if (RADEON_DEBUG == DEBUG_PRIMS)
267 fprintf(stderr, "%s\n", __FUNCTION__);
268
269 #if 1 /* we need this, somehow */
270 /* Flush state - make sure command buffer is nice and large */
271 r300Flush(ctx);
272 /* Make sure we have enough space */
273 #else 0
274 /* Count is very imprecize, but should be good upper bound */
275 r300EnsureCmdBufSpace(rmesa, rmesa->hw.max_state_size + 4+2+30
276 +VB->PrimitiveCount*(1+8)+VB->Count*4*rmesa->state.texture.tc_count+4, __FUNCTION__);
277 #endif
278
279 /* needed before starting 3d operation .. */
280 reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
281 e32(0x0000000a);
282
283 reg_start(0x4f18,0);
284 e32(0x00000003);
285
286
287 #if 0 /* looks like the Z offset issue got fixed */
288 rmesa->hw.vte.cmd[1] = R300_VPORT_X_SCALE_ENA
289 | R300_VPORT_X_OFFSET_ENA
290 | R300_VPORT_Y_SCALE_ENA
291 | R300_VPORT_Y_OFFSET_ENA
292 | R300_VTX_W0_FMT;
293 R300_STATECHANGE(rmesa, vte);
294 #endif
295
296
297 /* Magic register - note it is right after 20b0 */
298
299
300 if(rmesa->state.texture.tc_count>0){
301 reg_start(0x20b4,0);
302 e32(0x0000000c);
303
304 }
305
306 r300EmitState(rmesa);
307
308 #if 0
309 reg_start(R300_RB3D_COLORMASK, 0);
310 e32(0xf);
311 #endif
312 /* ----------------------------------- */
313
314 /* We need LOAD_VBPNTR to setup AOS_ATTR fields.. the offsets are irrelevant */
315 r300EmitLOAD_VBPNTR(rmesa, 0);
316
317 for(i=0; i < VB->PrimitiveCount; i++){
318 GLuint prim = VB->Primitive[i].mode;
319 GLuint start = VB->Primitive[i].start;
320 GLuint length = VB->Primitive[i].count;
321 r300_render_immediate_primitive(rmesa, ctx, start, start + length, prim);
322 }
323
324 /* This sequence is required after any 3d drawing packet
325 I suspect it work arounds a bug (or deficiency) in hardware */
326
327 reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
328 e32(0x0000000a);
329
330 reg_start(0x4f18,0);
331 e32(0x00000003);
332
333 return GL_FALSE;
334 }
335
336
337 /* vertex buffer implementation */
338
339 /* We use the start part of GART texture buffer for vertices */
340
341
342 static void upload_vertex_buffer(r300ContextPtr rmesa, GLcontext *ctx)
343 {
344 TNLcontext *tnl = TNL_CONTEXT(ctx);
345 struct vertex_buffer *VB = &tnl->vb;
346 int idx=0;
347 int i,j,k;
348 radeonScreenPtr rsp=rmesa->radeon.radeonScreen;
349
350 /* A hack - we don't want to overwrite vertex buffers, so we
351 just use AGP space for them.. Fix me ! */
352 static int offset=0;
353 if(offset>2*1024*1024){
354 //fprintf(stderr, "Wrapping agp vertex buffer offset\n");
355 offset=0;
356 }
357 /* Not the most efficient implementation, but, for now, I just want something that
358 works */
359 /* to do - make single memcpy per column (is it possible ?) */
360 /* to do - use dirty flags to avoid redundant copies */
361 #define UPLOAD_VECTOR(v)\
362 { \
363 /* Is the data dirty ? */ \
364 if (v->flags & ((1<<v->size)-1)) { \
365 /* fprintf(stderr, "size=%d vs stride=%d\n", v->size, v->stride); */ \
366 if(v->size*4==v->stride){\
367 /* fast path */ \
368 memcpy(rsp->gartTextures.map+offset, v->data, v->stride*VB->Count); \
369 } else { \
370 for(i=0;i<VB->Count;i++){ \
371 /* copy one vertex at a time*/ \
372 memcpy(rsp->gartTextures.map+offset+i*v->size*4, VEC_ELT(v, GLfloat, i), v->size*4); \
373 } \
374 } \
375 /* v->flags &= ~((1<<v->size)-1);*/ \
376 } \
377 rmesa->state.aos[idx].offset=rsp->gartTextures.handle+offset; \
378 offset+=v->size*4*VB->Count; \
379 idx++; \
380 }
381
382 UPLOAD_VECTOR(VB->ObjPtr);
383 UPLOAD_VECTOR(VB->ColorPtr[0]);
384 /* texture coordinates */
385 for(k=0;k < ctx->Const.MaxTextureUnits;k++)
386 if(ctx->Texture.Unit[k].Enabled)
387 UPLOAD_VECTOR(VB->TexCoordPtr[k]);
388
389 if(idx>=R300_MAX_AOS_ARRAYS){
390 fprintf(stderr, "Aieee ! Maximum AOS arrays count exceeded.. \n");
391 exit(-1);
392 }
393 }
394
395 static void r300_render_vb_primitive(r300ContextPtr rmesa,
396 GLcontext *ctx,
397 int start,
398 int end,
399 int prim)
400 {
401 int type;
402 LOCAL_VARS
403
404 if(end<=start)return; /* do we need to watch for this ? */
405
406 type=r300_get_primitive_type(rmesa, ctx, start, end, prim);
407 if(type<0)return;
408
409 fire_AOS(PASS_PREFIX end-start, type);
410 }
411
412 static GLboolean r300_run_vb_render(GLcontext *ctx,
413 struct tnl_pipeline_stage *stage)
414 {
415 r300ContextPtr rmesa = R300_CONTEXT(ctx);
416 TNLcontext *tnl = TNL_CONTEXT(ctx);
417 struct vertex_buffer *VB = &tnl->vb;
418 int i, j;
419 LOCAL_VARS
420
421 if (RADEON_DEBUG == DEBUG_PRIMS)
422 fprintf(stderr, "%s\n", __FUNCTION__);
423
424
425 reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
426 e32(0x0000000a);
427
428 reg_start(0x4f18,0);
429 e32(0x00000003);
430
431 r300_setup_routing(ctx, GL_FALSE);
432
433 r300EmitState(rmesa);
434
435 /* setup array of structures data */
436 LOCK_HARDWARE(&(rmesa->radeon));
437
438 upload_vertex_buffer(rmesa, ctx);
439 //fprintf(stderr, "Using %d AOS arrays\n", n_arrays);
440
441 for(i=0; i < VB->PrimitiveCount; i++){
442 GLuint prim = VB->Primitive[i].mode;
443 GLuint start = VB->Primitive[i].start;
444 GLuint length = VB->Primitive[i].count;
445
446 /* We need LOAD_VBPNTR to setup AOS_ATTR fields.. */
447 r300EmitLOAD_VBPNTR(rmesa, start);
448
449 r300_render_vb_primitive(rmesa, ctx, start, start + length, prim);
450 }
451
452 /* This sequence is required after any 3d drawing packet
453 I suspect it works around a bug (or deficiency) in hardware */
454
455 reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
456 e32(0x0000000a);
457
458 reg_start(0x4f18,0);
459 e32(0x00000003);
460
461 end_3d(PASS_PREFIX_VOID);
462
463 /* Flush state - we are done drawing.. */
464 r300FlushCmdBufLocked(ctx, __FUNCTION__);
465 radeonWaitForIdleLocked(&(rmesa->radeon));
466
467 UNLOCK_HARDWARE(&(rmesa->radeon));
468 return GL_FALSE;
469 }
470
471
472 /**
473 * Called by the pipeline manager to render a batch of primitives.
474 * We can return true to pass on to the next stage (i.e. software
475 * rasterization) or false to indicate that the pipeline has finished
476 * after we render something.
477 */
478 static GLboolean r300_run_render(GLcontext *ctx,
479 struct tnl_pipeline_stage *stage)
480 {
481 r300ContextPtr rmesa = R300_CONTEXT(ctx);
482 TNLcontext *tnl = TNL_CONTEXT(ctx);
483 struct vertex_buffer *VB = &tnl->vb;
484 GLuint i;
485
486 if (RADEON_DEBUG == DEBUG_PRIMS)
487 fprintf(stderr, "%s\n", __FUNCTION__);
488
489
490 #if 1
491
492 #if 1
493 return r300_run_immediate_render(ctx, stage);
494 #else
495 return r300_run_vb_render(ctx, stage);
496 #endif
497 #else
498 return GL_TRUE;
499 #endif
500
501 #if 0
502 mgaContextPtr mmesa = MGA_CONTEXT(ctx);
503 TNLcontext *tnl = TNL_CONTEXT(ctx);
504 struct vertex_buffer *VB = &tnl->vb;
505 GLuint i;
506
507 /* Don't handle clipping or indexed vertices or vertex manipulations.
508 */
509 if (mmesa->RenderIndex != 0 ||
510 !mga_validate_render( ctx, VB )) {
511 return GL_TRUE;
512 }
513
514 tnl->Driver.Render.Start( ctx );
515 mmesa->SetupNewInputs = ~0;
516
517 for (i = 0 ; i < VB->PrimitiveCount ; i++)
518 {
519 GLuint prim = VB->Primitive[i].mode;
520 GLuint start = VB->Primitive[i].start;
521 GLuint length = VB->Primitive[i].count;
522
523 if (!length)
524 continue;
525
526 mga_render_tab_verts[prim & PRIM_MODE_MASK]( ctx, start, start + length,
527 prim);
528 }
529
530 tnl->Driver.Render.Finish( ctx );
531
532 return GL_FALSE; /* finished the pipe */
533 #endif
534 }
535
536
537 /**
538 * Called by the pipeline manager once before rendering.
539 * We check the GL state here to
540 * a) decide whether we can do the current state in hardware and
541 * b) update hardware registers
542 */
543 #define FALLBACK_IF(expr) \
544 do { \
545 if (expr) { \
546 if (1 || RADEON_DEBUG & DEBUG_FALLBACKS) \
547 fprintf(stderr, "%s: fallback:%s\n", \
548 __FUNCTION__, #expr); \
549 stage->active = GL_FALSE; \
550 return; \
551 } \
552 } while(0)
553
554 static void r300_check_render(GLcontext *ctx, struct tnl_pipeline_stage *stage)
555 {
556 r300ContextPtr r300 = R300_CONTEXT(ctx);
557 int i;
558
559 if (RADEON_DEBUG & DEBUG_STATE)
560 fprintf(stderr, "%s\n", __FUNCTION__);
561
562 /* We only support rendering in hardware for now */
563 if (ctx->RenderMode != GL_RENDER) {
564 stage->active = GL_FALSE;
565 return;
566 }
567
568 // I failed to figure out how dither works in hardware,
569 // let's just ignore it for now
570 //FALLBACK_IF(ctx->Color.DitherFlag);
571
572 /* I'm almost certain I forgot something here */
573 #if 0 /* This should work now.. */
574 FALLBACK_IF(ctx->Color.AlphaEnabled); // GL_ALPHA_TEST
575 #endif
576 //FALLBACK_IF(ctx->Color.BlendEnabled); // GL_BLEND
577 FALLBACK_IF(ctx->Fog.Enabled); // GL_FOG
578 FALLBACK_IF(ctx->Line.SmoothFlag); // GL_LINE_SMOOTH
579 FALLBACK_IF(ctx->Line.StippleFlag); // GL_LINE_STIPPLE
580 FALLBACK_IF(ctx->Point.SmoothFlag); // GL_POINT_SMOOTH
581 if (ctx->Extensions.NV_point_sprite || ctx->Extensions.ARB_point_sprite)
582 FALLBACK_IF(ctx->Point.PointSprite); // GL_POINT_SPRITE_NV
583 FALLBACK_IF(ctx->Polygon.OffsetPoint); // GL_POLYGON_OFFSET_POINT
584 FALLBACK_IF(ctx->Polygon.OffsetLine); // GL_POLYGON_OFFSET_LINE
585 FALLBACK_IF(ctx->Polygon.OffsetFill); // GL_POLYGON_OFFSET_FILL
586 FALLBACK_IF(ctx->Polygon.SmoothFlag); // GL_POLYGON_SMOOTH
587 FALLBACK_IF(ctx->Polygon.StippleFlag); // GL_POLYGON_STIPPLE
588 FALLBACK_IF(ctx->Stencil.Enabled); // GL_STENCIL_TEST
589 FALLBACK_IF(ctx->Multisample.Enabled); // GL_MULTISAMPLE_ARB
590
591 /* One step at a time - let one texture pass.. */
592 for (i = 2; i < ctx->Const.MaxTextureUnits; i++)
593 FALLBACK_IF(ctx->Texture.Unit[i].Enabled);
594
595
596 /* let r300_run_render do its job */
597 #if 0
598 stage->active = GL_FALSE;
599 #endif
600 }
601
602
603 static void dtr(struct tnl_pipeline_stage *stage)
604 {
605 (void)stage;
606 }
607
608 const struct tnl_pipeline_stage _r300_render_stage = {
609 "r300 hw rasterize",
610 _NEW_ALL, /* re-check (always re-check for now) */
611 0, /* re-run (always runs) */
612 GL_TRUE, /* active */
613 0, 0, /* inputs (set in check_render), outputs */
614 0, 0, /* changed_inputs, private */
615 dtr, /* destructor */
616 r300_check_render, /* check */
617 r300_run_render /* run */
618 };