projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Merge branch 'wip/nir-vtn' into vulkan
[mesa.git] / src / mesa / drivers / dri / r200 / r200_swtcl.c
1 /*
2 Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved.
3
4 The Weather Channel (TM) funded Tungsten Graphics to develop the
5 initial release of the Radeon 8500 driver under the XFree86 license.
6 This notice must be preserved.
7
8 Permission is hereby granted, free of charge, to any person obtaining
9 a copy of this software and associated documentation files (the
10 "Software"), to deal in the Software without restriction, including
11 without limitation the rights to use, copy, modify, merge, publish,
12 distribute, sublicense, and/or sell copies of the Software, and to
13 permit persons to whom the Software is furnished to do so, subject to
14 the following conditions:
15
16 The above copyright notice and this permission notice (including the
17 next paragraph) shall be included in all copies or substantial
18 portions of the Software.
19
20 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
23 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
24 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
25 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
26 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27
28 **************************************************************************/
29
30 /*
31 * Authors:
32 * Keith Whitwell <keithw@vmware.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/image.h"
40 #include "main/imports.h"
41 #include "main/macros.h"
42 #include "util/simple_list.h"
43
44 #include "swrast/s_context.h"
45 #include "swrast/s_fog.h"
46 #include "swrast_setup/swrast_setup.h"
47 #include "tnl/tnl.h"
48 #include "tnl/t_context.h"
49 #include "tnl/t_pipeline.h"
50
51 #include "r200_context.h"
52 #include "r200_ioctl.h"
53 #include "r200_state.h"
54 #include "r200_swtcl.h"
55 #include "r200_tcl.h"
56
57
58 /***********************************************************************
59 * Initialization
60 ***********************************************************************/
61
62 #define EMIT_ATTR( ATTR, STYLE, F0 ) \
63 do { \
64 rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = (ATTR); \
65 rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = (STYLE); \
66 rmesa->radeon.swtcl.vertex_attr_count++; \
67 fmt_0 |= F0; \
68 } while (0)
69
70 #define EMIT_PAD( N ) \
71 do { \
72 rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = 0; \
73 rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = EMIT_PAD; \
74 rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].offset = (N); \
75 rmesa->radeon.swtcl.vertex_attr_count++; \
76 } while (0)
77
78 static void r200SetVertexFormat( struct gl_context *ctx )
79 {
80 r200ContextPtr rmesa = R200_CONTEXT( ctx );
81 TNLcontext *tnl = TNL_CONTEXT(ctx);
82 struct vertex_buffer *VB = &tnl->vb;
83 GLbitfield64 index_bitset = tnl->render_inputs_bitset;
84 int fmt_0 = 0;
85 int fmt_1 = 0;
86 int offset = 0;
87
88 /* Important:
89 */
90 if ( VB->NdcPtr != NULL ) {
91 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
92 }
93 else {
94 VB->AttribPtr[VERT_ATTRIB_POS] = VB->ClipPtr;
95 }
96
97 assert( VB->AttribPtr[VERT_ATTRIB_POS] != NULL );
98 rmesa->radeon.swtcl.vertex_attr_count = 0;
99
100 /* EMIT_ATTR's must be in order as they tell t_vertex.c how to
101 * build up a hardware vertex.
102 */
103 if ( !rmesa->swtcl.needproj ||
104 (index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)) ) {
105 /* need w coord for projected textures */
106 EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F, R200_VTX_XY | R200_VTX_Z0 | R200_VTX_W0 );
107 offset = 4;
108 }
109 else {
110 EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_3F, R200_VTX_XY | R200_VTX_Z0 );
111 offset = 3;
112 }
113
114 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_POINTSIZE)) {
115 EMIT_ATTR( _TNL_ATTRIB_POINTSIZE, EMIT_1F, R200_VTX_POINT_SIZE );
116 offset += 1;
117 }
118
119 rmesa->swtcl.coloroffset = offset;
120 #if MESA_LITTLE_ENDIAN
121 EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_RGBA, (R200_VTX_PK_RGBA << R200_VTX_COLOR_0_SHIFT) );
122 #else
123 EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_ABGR, (R200_VTX_PK_RGBA << R200_VTX_COLOR_0_SHIFT) );
124 #endif
125 offset += 1;
126
127 rmesa->swtcl.specoffset = 0;
128 if (index_bitset &
129 (BITFIELD64_BIT(_TNL_ATTRIB_COLOR1) | BITFIELD64_BIT(_TNL_ATTRIB_FOG))) {
130
131 #if MESA_LITTLE_ENDIAN
132 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_COLOR1)) {
133 rmesa->swtcl.specoffset = offset;
134 EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_RGB, (R200_VTX_PK_RGBA << R200_VTX_COLOR_1_SHIFT) );
135 }
136 else {
137 EMIT_PAD( 3 );
138 }
139
140 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_FOG)) {
141 EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F, (R200_VTX_PK_RGBA << R200_VTX_COLOR_1_SHIFT) );
142 }
143 else {
144 EMIT_PAD( 1 );
145 }
146 #else
147 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_FOG)) {
148 EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F, (R200_VTX_PK_RGBA << R200_VTX_COLOR_1_SHIFT) );
149 }
150 else {
151 EMIT_PAD( 1 );
152 }
153
154 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_COLOR1)) {
155 rmesa->swtcl.specoffset = offset;
156 EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, (R200_VTX_PK_RGBA << R200_VTX_COLOR_1_SHIFT) );
157 }
158 else {
159 EMIT_PAD( 3 );
160 }
161 #endif
162 }
163
164 if (index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)) {
165 int i;
166
167 for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
168 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_TEX(i))) {
169 GLuint sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
170
171 fmt_1 |= sz << (3 * i);
172 EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_1F + sz - 1, 0 );
173 }
174 }
175 }
176
177 if ( (rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] & R200_FOG_USE_MASK)
178 != R200_FOG_USE_SPEC_ALPHA ) {
179 R200_STATECHANGE( rmesa, ctx );
180 rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] &= ~R200_FOG_USE_MASK;
181 rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] |= R200_FOG_USE_SPEC_ALPHA;
182 }
183
184 if (rmesa->radeon.tnl_index_bitset != index_bitset ||
185 (rmesa->hw.vtx.cmd[VTX_VTXFMT_0] != fmt_0) ||
186 (rmesa->hw.vtx.cmd[VTX_VTXFMT_1] != fmt_1) ) {
187 R200_NEWPRIM(rmesa);
188 R200_STATECHANGE( rmesa, vtx );
189 rmesa->hw.vtx.cmd[VTX_VTXFMT_0] = fmt_0;
190 rmesa->hw.vtx.cmd[VTX_VTXFMT_1] = fmt_1;
191
192 rmesa->radeon.swtcl.vertex_size =
193 _tnl_install_attrs( ctx,
194 rmesa->radeon.swtcl.vertex_attrs,
195 rmesa->radeon.swtcl.vertex_attr_count,
196 NULL, 0 );
197 rmesa->radeon.swtcl.vertex_size /= 4;
198 rmesa->radeon.tnl_index_bitset = index_bitset;
199 }
200 }
201
202 static void r200_predict_emit_size( r200ContextPtr rmesa )
203 {
204 if (RADEON_DEBUG & RADEON_VERTS)
205 fprintf(stderr, "%s\n", __func__);
206 const int vertex_array_size = 7;
207 const int prim_size = 3;
208 if (!rmesa->radeon.swtcl.emit_prediction) {
209 const int state_size = radeonCountStateEmitSize(&rmesa->radeon);
210 if (rcommonEnsureCmdBufSpace(&rmesa->radeon,
211 state_size +
212 vertex_array_size + prim_size,
213 __func__))
214 rmesa->radeon.swtcl.emit_prediction = radeonCountStateEmitSize(&rmesa->radeon);
215 else
216 rmesa->radeon.swtcl.emit_prediction = state_size;
217 rmesa->radeon.swtcl.emit_prediction += vertex_array_size + prim_size
218 + rmesa->radeon.cmdbuf.cs->cdw;
219 }
220 }
221
222
223 static void r200RenderStart( struct gl_context *ctx )
224 {
225 r200SetVertexFormat( ctx );
226 if (RADEON_DEBUG & RADEON_VERTS)
227 fprintf(stderr, "%s\n", __func__);
228 }
229
230
231 /**
232 * Set vertex state for SW TCL. The primary purpose of this function is to
233 * determine in advance whether or not the hardware can / should do the
234 * projection divide or Mesa should do it.
235 */
236 void r200ChooseVertexState( struct gl_context *ctx )
237 {
238 r200ContextPtr rmesa = R200_CONTEXT( ctx );
239 TNLcontext *tnl = TNL_CONTEXT(ctx);
240 GLuint vte;
241 GLuint vap;
242 GLboolean unfilled = (ctx->Polygon.FrontMode != GL_FILL ||
243 ctx->Polygon.BackMode != GL_FILL);
244 GLboolean twosided = ctx->Light.Enabled && ctx->Light.Model.TwoSide;
245
246 /* We must ensure that we don't do _tnl_need_projected_coords while in a
247 * rasterization fallback. As this function will be called again when we
248 * leave a rasterization fallback, we can just skip it for now.
249 */
250 if (rmesa->radeon.Fallback != 0)
251 return;
252
253 vte = rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL];
254 vap = rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL];
255
256 /* HW perspective divide is a win, but tiny vertex formats are a
257 * bigger one.
258 */
259 if ((0 == (tnl->render_inputs_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)))
260 || twosided
261 || unfilled) {
262 rmesa->swtcl.needproj = GL_TRUE;
263 vte |= R200_VTX_XY_FMT | R200_VTX_Z_FMT;
264 vte &= ~R200_VTX_W0_FMT;
265 if (tnl->render_inputs_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)) {
266 vap &= ~R200_VAP_FORCE_W_TO_ONE;
267 }
268 else {
269 vap |= R200_VAP_FORCE_W_TO_ONE;
270 }
271 }
272 else {
273 rmesa->swtcl.needproj = GL_FALSE;
274 vte &= ~(R200_VTX_XY_FMT | R200_VTX_Z_FMT);
275 vte |= R200_VTX_W0_FMT;
276 vap &= ~R200_VAP_FORCE_W_TO_ONE;
277 }
278
279 _tnl_need_projected_coords( ctx, rmesa->swtcl.needproj );
280
281 if (vte != rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL]) {
282 R200_STATECHANGE( rmesa, vte );
283 rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] = vte;
284 }
285
286 if (vap != rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL]) {
287 R200_STATECHANGE( rmesa, vap );
288 rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] = vap;
289 }
290 }
291
292 void r200_swtcl_flush(struct gl_context *ctx, uint32_t current_offset)
293 {
294 r200ContextPtr rmesa = R200_CONTEXT(ctx);
295 if (RADEON_DEBUG & RADEON_VERTS)
296 fprintf(stderr, "%s\n", __func__);
297
298
299 radeonEmitState(&rmesa->radeon);
300 r200EmitVertexAOS( rmesa,
301 rmesa->radeon.swtcl.vertex_size,
302 rmesa->radeon.swtcl.bo,
303 current_offset);
304
305
306 r200EmitVbufPrim( rmesa,
307 rmesa->radeon.swtcl.hw_primitive,
308 rmesa->radeon.swtcl.numverts);
309 if ( rmesa->radeon.swtcl.emit_prediction < rmesa->radeon.cmdbuf.cs->cdw )
310 WARN_ONCE("Rendering was %d commands larger than predicted size."
311 " We might overflow command buffer.\n",
312 rmesa->radeon.cmdbuf.cs->cdw - rmesa->radeon.swtcl.emit_prediction );
313
314 rmesa->radeon.swtcl.emit_prediction = 0;
315
316 }
317
318 /**************************************************************************/
319
320
321 static inline GLuint reduced_hw_prim( struct gl_context *ctx, GLuint prim)
322 {
323 switch (prim) {
324 case GL_POINTS:
325 return ((!ctx->Point.SmoothFlag) ?
326 R200_VF_PRIM_POINT_SPRITES : R200_VF_PRIM_POINTS);
327 case GL_LINES:
328 /* fallthrough */
329 case GL_LINE_LOOP:
330 /* fallthrough */
331 case GL_LINE_STRIP:
332 return R200_VF_PRIM_LINES;
333 default:
334 /* all others reduced to triangles */
335 return R200_VF_PRIM_TRIANGLES;
336 }
337 }
338
339
340 static void r200RasterPrimitive( struct gl_context *ctx, GLuint hwprim );
341 static void r200RenderPrimitive( struct gl_context *ctx, GLenum prim );
342 static void r200ResetLineStipple( struct gl_context *ctx );
343
344 /***********************************************************************
345 * Emit primitives as inline vertices *
346 ***********************************************************************/
347
348 #define HAVE_POINTS 1
349 #define HAVE_LINES 1
350 #define HAVE_LINE_STRIPS 1
351 #define HAVE_TRIANGLES 1
352 #define HAVE_TRI_STRIPS 1
353 #define HAVE_TRI_STRIP_1 0
354 #define HAVE_TRI_FANS 1
355 #define HAVE_QUADS 0
356 #define HAVE_QUAD_STRIPS 0
357 #define HAVE_POLYGONS 1
358 #define HAVE_ELTS 0
359
360 static void* r200_alloc_verts( r200ContextPtr rmesa, GLuint n, GLuint size)
361 {
362 void *rv;
363 do {
364 r200_predict_emit_size( rmesa );
365 rv = rcommonAllocDmaLowVerts( &rmesa->radeon, n, size * 4 );
366 } while(!rv);
367 return rv;
368 }
369
370 #undef LOCAL_VARS
371 #undef ALLOC_VERTS
372 #define CTX_ARG r200ContextPtr rmesa
373 #define GET_VERTEX_DWORDS() rmesa->radeon.swtcl.vertex_size
374 #define ALLOC_VERTS( n, size ) r200_alloc_verts(rmesa, n, size)
375 #define LOCAL_VARS \
376 r200ContextPtr rmesa = R200_CONTEXT(ctx); \
377 const char *r200verts = (char *)rmesa->radeon.swtcl.verts;
378 #define VERT(x) (radeonVertex *)(r200verts + ((x) * vertsize * sizeof(int)))
379 #define VERTEX radeonVertex
380 #define DO_DEBUG_VERTS (1 && (R200_DEBUG & RADEON_VERTS))
381
382 #undef TAG
383 #define TAG(x) r200_##x
384 #include "tnl_dd/t_dd_triemit.h"
385
386
387 /***********************************************************************
388 * Macros for t_dd_tritmp.h to draw basic primitives *
389 ***********************************************************************/
390
391 #define QUAD( a, b, c, d ) r200_quad( rmesa, a, b, c, d )
392 #define TRI( a, b, c ) r200_triangle( rmesa, a, b, c )
393 #define LINE( a, b ) r200_line( rmesa, a, b )
394 #define POINT( a ) r200_point( rmesa, a )
395
396 /***********************************************************************
397 * Build render functions from dd templates *
398 ***********************************************************************/
399
400 #define R200_TWOSIDE_BIT 0x01
401 #define R200_UNFILLED_BIT 0x02
402 #define R200_MAX_TRIFUNC 0x04
403
404
405 static struct {
406 tnl_points_func points;
407 tnl_line_func line;
408 tnl_triangle_func triangle;
409 tnl_quad_func quad;
410 } rast_tab[R200_MAX_TRIFUNC];
411
412
413 #define DO_FALLBACK 0
414 #define DO_UNFILLED ((IND & R200_UNFILLED_BIT) != 0)
415 #define DO_TWOSIDE ((IND & R200_TWOSIDE_BIT) != 0)
416 #define DO_FLAT 0
417 #define DO_OFFSET 0
418 #define DO_TRI 1
419 #define DO_QUAD 1
420 #define DO_LINE 1
421 #define DO_POINTS 1
422 #define DO_FULL_QUAD 1
423
424 #define HAVE_SPEC 1
425 #define HAVE_BACK_COLORS 0
426 #define HAVE_HW_FLATSHADE 1
427 #define TAB rast_tab
428
429 #define DEPTH_SCALE 1.0
430 #define UNFILLED_TRI unfilled_tri
431 #define UNFILLED_QUAD unfilled_quad
432 #define VERT_X(_v) _v->v.x
433 #define VERT_Y(_v) _v->v.y
434 #define VERT_Z(_v) _v->v.z
435 #define AREA_IS_CCW( a ) (a < 0)
436 #define GET_VERTEX(e) (rmesa->radeon.swtcl.verts + (e*rmesa->radeon.swtcl.vertex_size*sizeof(int)))
437
438 #define VERT_SET_RGBA( v, c ) \
439 do { \
440 radeon_color_t *color = (radeon_color_t *)&((v)->ui[coloroffset]); \
441 UNCLAMPED_FLOAT_TO_UBYTE(color->red, (c)[0]); \
442 UNCLAMPED_FLOAT_TO_UBYTE(color->green, (c)[1]); \
443 UNCLAMPED_FLOAT_TO_UBYTE(color->blue, (c)[2]); \
444 UNCLAMPED_FLOAT_TO_UBYTE(color->alpha, (c)[3]); \
445 } while (0)
446
447 #define VERT_COPY_RGBA( v0, v1 ) v0->ui[coloroffset] = v1->ui[coloroffset]
448
449 #define VERT_SET_SPEC( v, c ) \
450 do { \
451 if (specoffset) { \
452 radeon_color_t *spec = (radeon_color_t *)&((v)->ui[specoffset]); \
453 UNCLAMPED_FLOAT_TO_UBYTE(spec->red, (c)[0]); \
454 UNCLAMPED_FLOAT_TO_UBYTE(spec->green, (c)[1]); \
455 UNCLAMPED_FLOAT_TO_UBYTE(spec->blue, (c)[2]); \
456 } \
457 } while (0)
458 #define VERT_COPY_SPEC( v0, v1 ) \
459 do { \
460 if (specoffset) { \
461 radeon_color_t *spec0 = (radeon_color_t *)&((v0)->ui[specoffset]); \
462 radeon_color_t *spec1 = (radeon_color_t *)&((v1)->ui[specoffset]); \
463 spec0->red = spec1->red; \
464 spec0->green = spec1->green; \
465 spec0->blue = spec1->blue; \
466 } \
467 } while (0)
468
469 /* These don't need LE32_TO_CPU() as they used to save and restore
470 * colors which are already in the correct format.
471 */
472 #define VERT_SAVE_RGBA( idx ) color[idx] = v[idx]->ui[coloroffset]
473 #define VERT_RESTORE_RGBA( idx ) v[idx]->ui[coloroffset] = color[idx]
474 #define VERT_SAVE_SPEC( idx ) if (specoffset) spec[idx] = v[idx]->ui[specoffset]
475 #define VERT_RESTORE_SPEC( idx ) if (specoffset) v[idx]->ui[specoffset] = spec[idx]
476
477 #undef LOCAL_VARS
478 #undef TAG
479 #undef INIT
480
481 #define LOCAL_VARS(n) \
482 r200ContextPtr rmesa = R200_CONTEXT(ctx); \
483 GLuint color[n] = {0}, spec[n] = {0}; \
484 GLuint coloroffset = rmesa->swtcl.coloroffset; \
485 GLuint specoffset = rmesa->swtcl.specoffset; \
486 (void) color; (void) spec; (void) coloroffset; (void) specoffset;
487
488 /***********************************************************************
489 * Helpers for rendering unfilled primitives *
490 ***********************************************************************/
491
492 #define RASTERIZE(x) r200RasterPrimitive( ctx, reduced_hw_prim(ctx, x) )
493 #define RENDER_PRIMITIVE rmesa->radeon.swtcl.render_primitive
494 #undef TAG
495 #define TAG(x) x
496 #include "tnl_dd/t_dd_unfilled.h"
497 #undef IND
498
499
500 /***********************************************************************
501 * Generate GL render functions *
502 ***********************************************************************/
503
504
505 #define IND (0)
506 #define TAG(x) x
507 #include "tnl_dd/t_dd_tritmp.h"
508
509 #define IND (R200_TWOSIDE_BIT)
510 #define TAG(x) x##_twoside
511 #include "tnl_dd/t_dd_tritmp.h"
512
513 #define IND (R200_UNFILLED_BIT)
514 #define TAG(x) x##_unfilled
515 #include "tnl_dd/t_dd_tritmp.h"
516
517 #define IND (R200_TWOSIDE_BIT|R200_UNFILLED_BIT)
518 #define TAG(x) x##_twoside_unfilled
519 #include "tnl_dd/t_dd_tritmp.h"
520
521
522 static void init_rast_tab( void )
523 {
524 init();
525 init_twoside();
526 init_unfilled();
527 init_twoside_unfilled();
528 }
529
530 /**********************************************************************/
531 /* Render unclipped begin/end objects */
532 /**********************************************************************/
533
534 #define RENDER_POINTS( start, count ) \
535 for ( ; start < count ; start++) \
536 r200_point( rmesa, VERT(start) )
537 #define RENDER_LINE( v0, v1 ) \
538 r200_line( rmesa, VERT(v0), VERT(v1) )
539 #define RENDER_TRI( v0, v1, v2 ) \
540 r200_triangle( rmesa, VERT(v0), VERT(v1), VERT(v2) )
541 #define RENDER_QUAD( v0, v1, v2, v3 ) \
542 r200_quad( rmesa, VERT(v0), VERT(v1), VERT(v2), VERT(v3) )
543 #define INIT(x) do { \
544 r200RenderPrimitive( ctx, x ); \
545 } while (0)
546 #undef LOCAL_VARS
547 #define LOCAL_VARS \
548 r200ContextPtr rmesa = R200_CONTEXT(ctx); \
549 const GLuint vertsize = rmesa->radeon.swtcl.vertex_size; \
550 const char *r200verts = (char *)rmesa->radeon.swtcl.verts; \
551 const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts; \
552 const GLboolean stipple = ctx->Line.StippleFlag; \
553 (void) elt; (void) stipple;
554 #define RESET_STIPPLE if ( stipple ) r200ResetLineStipple( ctx );
555 #define RESET_OCCLUSION
556 #define PRESERVE_VB_DEFS
557 #define ELT(x) (x)
558 #define TAG(x) r200_##x##_verts
559 #include "tnl/t_vb_rendertmp.h"
560 #undef ELT
561 #undef TAG
562 #define TAG(x) r200_##x##_elts
563 #define ELT(x) elt[x]
564 #include "tnl/t_vb_rendertmp.h"
565
566
567
568 /**********************************************************************/
569 /* Choose render functions */
570 /**********************************************************************/
571
572 void r200ChooseRenderState( struct gl_context *ctx )
573 {
574 TNLcontext *tnl = TNL_CONTEXT(ctx);
575 r200ContextPtr rmesa = R200_CONTEXT(ctx);
576 GLuint index = 0;
577 GLboolean unfilled = (ctx->Polygon.FrontMode != GL_FILL ||
578 ctx->Polygon.BackMode != GL_FILL);
579 GLboolean twosided = ctx->Light.Enabled && ctx->Light.Model.TwoSide;
580
581 if (!rmesa->radeon.TclFallback || rmesa->radeon.Fallback)
582 return;
583
584 if (twosided)
585 index |= R200_TWOSIDE_BIT;
586 if (unfilled)
587 index |= R200_UNFILLED_BIT;
588
589 if (index != rmesa->radeon.swtcl.RenderIndex) {
590 tnl->Driver.Render.Points = rast_tab[index].points;
591 tnl->Driver.Render.Line = rast_tab[index].line;
592 tnl->Driver.Render.ClippedLine = rast_tab[index].line;
593 tnl->Driver.Render.Triangle = rast_tab[index].triangle;
594 tnl->Driver.Render.Quad = rast_tab[index].quad;
595
596 if (index == 0) {
597 tnl->Driver.Render.PrimTabVerts = r200_render_tab_verts;
598 tnl->Driver.Render.PrimTabElts = r200_render_tab_elts;
599 tnl->Driver.Render.ClippedPolygon = r200_fast_clipped_poly;
600 } else {
601 tnl->Driver.Render.PrimTabVerts = _tnl_render_tab_verts;
602 tnl->Driver.Render.PrimTabElts = _tnl_render_tab_elts;
603 tnl->Driver.Render.ClippedPolygon = _tnl_RenderClippedPolygon;
604 }
605
606 rmesa->radeon.swtcl.RenderIndex = index;
607 }
608 }
609
610
611 /**********************************************************************/
612 /* High level hooks for t_vb_render.c */
613 /**********************************************************************/
614
615
616 static void r200RasterPrimitive( struct gl_context *ctx, GLuint hwprim )
617 {
618 r200ContextPtr rmesa = R200_CONTEXT(ctx);
619
620 radeon_prepare_render(&rmesa->radeon);
621 if (rmesa->radeon.NewGLState)
622 r200ValidateState( ctx );
623
624
625 if (rmesa->radeon.swtcl.hw_primitive != hwprim) {
626 /* need to disable perspective-correct texturing for point sprites */
627 if ((hwprim & 0xf) == R200_VF_PRIM_POINT_SPRITES && ctx->Point.PointSprite) {
628 if (rmesa->hw.set.cmd[SET_RE_CNTL] & R200_PERSPECTIVE_ENABLE) {
629 R200_STATECHANGE( rmesa, set );
630 rmesa->hw.set.cmd[SET_RE_CNTL] &= ~R200_PERSPECTIVE_ENABLE;
631 }
632 }
633 else if (!(rmesa->hw.set.cmd[SET_RE_CNTL] & R200_PERSPECTIVE_ENABLE)) {
634 R200_STATECHANGE( rmesa, set );
635 rmesa->hw.set.cmd[SET_RE_CNTL] |= R200_PERSPECTIVE_ENABLE;
636 }
637 R200_NEWPRIM( rmesa );
638 rmesa->radeon.swtcl.hw_primitive = hwprim;
639 }
640 }
641
642 static void r200RenderPrimitive( struct gl_context *ctx, GLenum prim )
643 {
644 r200ContextPtr rmesa = R200_CONTEXT(ctx);
645 GLboolean unfilled = (ctx->Polygon.FrontMode != GL_FILL ||
646 ctx->Polygon.BackMode != GL_FILL);
647
648 rmesa->radeon.swtcl.render_primitive = prim;
649 if (prim < GL_TRIANGLES || !unfilled)
650 r200RasterPrimitive( ctx, reduced_hw_prim(ctx, prim) );
651 }
652
653 static void r200RenderFinish( struct gl_context *ctx )
654 {
655 }
656
657 static void r200ResetLineStipple( struct gl_context *ctx )
658 {
659 r200ContextPtr rmesa = R200_CONTEXT(ctx);
660 R200_STATECHANGE( rmesa, lin );
661 }
662
663
664 /**********************************************************************/
665 /* Transition to/from hardware rasterization. */
666 /**********************************************************************/
667
668 static const char * const fallbackStrings[] = {
669 "Texture mode",
670 "glDrawBuffer(GL_FRONT_AND_BACK)",
671 "glEnable(GL_STENCIL) without hw stencil buffer",
672 "glRenderMode(selection or feedback)",
673 "R200_NO_RAST",
674 "Mixing GL_CLAMP_TO_BORDER and GL_CLAMP (or GL_MIRROR_CLAMP_ATI)"
675 };
676
677
678 static const char *getFallbackString(GLuint bit)
679 {
680 int i = 0;
681 while (bit > 1) {
682 i++;
683 bit >>= 1;
684 }
685 return fallbackStrings[i];
686 }
687
688
689 void r200Fallback( struct gl_context *ctx, GLuint bit, GLboolean mode )
690 {
691 r200ContextPtr rmesa = R200_CONTEXT(ctx);
692 TNLcontext *tnl = TNL_CONTEXT(ctx);
693 GLuint oldfallback = rmesa->radeon.Fallback;
694
695 if (mode) {
696 rmesa->radeon.Fallback |= bit;
697 if (oldfallback == 0) {
698 radeon_firevertices(&rmesa->radeon);
699 TCL_FALLBACK( ctx, R200_TCL_FALLBACK_RASTER, GL_TRUE );
700 _swsetup_Wakeup( ctx );
701 rmesa->radeon.swtcl.RenderIndex = ~0;
702 if (R200_DEBUG & RADEON_FALLBACKS) {
703 fprintf(stderr, "R200 begin rasterization fallback: 0x%x %s\n",
704 bit, getFallbackString(bit));
705 }
706 }
707 }
708 else {
709 rmesa->radeon.Fallback &= ~bit;
710 if (oldfallback == bit) {
711
712 _swrast_flush( ctx );
713 tnl->Driver.Render.Start = r200RenderStart;
714 tnl->Driver.Render.PrimitiveNotify = r200RenderPrimitive;
715 tnl->Driver.Render.Finish = r200RenderFinish;
716
717 tnl->Driver.Render.BuildVertices = _tnl_build_vertices;
718 tnl->Driver.Render.CopyPV = _tnl_copy_pv;
719 tnl->Driver.Render.Interp = _tnl_interp;
720
721 tnl->Driver.Render.ResetLineStipple = r200ResetLineStipple;
722 TCL_FALLBACK( ctx, R200_TCL_FALLBACK_RASTER, GL_FALSE );
723 if (rmesa->radeon.TclFallback) {
724 /* These are already done if rmesa->radeon.TclFallback goes to
725 * zero above. But not if it doesn't (R200_NO_TCL for
726 * example?)
727 */
728 _tnl_invalidate_vertex_state( ctx, ~0 );
729 _tnl_invalidate_vertices( ctx, ~0 );
730 rmesa->radeon.tnl_index_bitset = 0;
731 r200ChooseVertexState( ctx );
732 r200ChooseRenderState( ctx );
733 }
734 if (R200_DEBUG & RADEON_FALLBACKS) {
735 fprintf(stderr, "R200 end rasterization fallback: 0x%x %s\n",
736 bit, getFallbackString(bit));
737 }
738 }
739 }
740 }
741
742
743
744
745 /**
746 * Cope with depth operations by drawing individual pixels as points.
747 *
748 * \todo
749 * The way the vertex state is set in this routine is hokey. It seems to
750 * work, but it's very hackish. This whole routine is pretty hackish. If
751 * the bitmap is small enough, it seems like it would be faster to copy it
752 * to AGP memory and use it as a non-power-of-two texture (i.e.,
753 * NV_texture_rectangle).
754 */
755 void
756 r200PointsBitmap( struct gl_context *ctx, GLint px, GLint py,
757 GLsizei width, GLsizei height,
758 const struct gl_pixelstore_attrib *unpack,
759 const GLubyte *bitmap )
760 {
761 r200ContextPtr rmesa = R200_CONTEXT(ctx);
762 const GLfloat *rc = ctx->Current.RasterColor;
763 GLint row, col;
764 radeonVertex vert;
765 GLuint orig_vte;
766 GLuint h;
767
768
769 /* Turn off tcl.
770 */
771 TCL_FALLBACK( ctx, R200_TCL_FALLBACK_BITMAP, 1 );
772
773 /* Choose tiny vertex format
774 */
775 {
776 const GLuint fmt_0 = R200_VTX_XY | R200_VTX_Z0 | R200_VTX_W0
777 | (R200_VTX_PK_RGBA << R200_VTX_COLOR_0_SHIFT);
778 const GLuint fmt_1 = 0;
779 GLuint vte = rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL];
780 GLuint vap = rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL];
781
782 vte &= ~(R200_VTX_XY_FMT | R200_VTX_Z_FMT);
783 vte |= R200_VTX_W0_FMT;
784 vap &= ~R200_VAP_FORCE_W_TO_ONE;
785
786 rmesa->radeon.swtcl.vertex_size = 5;
787
788 if ( (rmesa->hw.vtx.cmd[VTX_VTXFMT_0] != fmt_0)
789 || (rmesa->hw.vtx.cmd[VTX_VTXFMT_1] != fmt_1) ) {
790 R200_NEWPRIM(rmesa);
791 R200_STATECHANGE( rmesa, vtx );
792 rmesa->hw.vtx.cmd[VTX_VTXFMT_0] = fmt_0;
793 rmesa->hw.vtx.cmd[VTX_VTXFMT_1] = fmt_1;
794 }
795
796 if (vte != rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL]) {
797 R200_STATECHANGE( rmesa, vte );
798 rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] = vte;
799 }
800
801 if (vap != rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL]) {
802 R200_STATECHANGE( rmesa, vap );
803 rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] = vap;
804 }
805 }
806
807 /* Ready for point primitives:
808 */
809 r200RenderPrimitive( ctx, GL_POINTS );
810
811 /* Turn off the hw viewport transformation:
812 */
813 R200_STATECHANGE( rmesa, vte );
814 orig_vte = rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL];
815 rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] &= ~(R200_VPORT_X_SCALE_ENA |
816 R200_VPORT_Y_SCALE_ENA |
817 R200_VPORT_Z_SCALE_ENA |
818 R200_VPORT_X_OFFSET_ENA |
819 R200_VPORT_Y_OFFSET_ENA |
820 R200_VPORT_Z_OFFSET_ENA);
821
822 /* Turn off other stuff: Stipple?, texture?, blending?, etc.
823 */
824
825
826 /* Populate the vertex
827 *
828 * Incorporate FOG into RGBA
829 */
830 if (ctx->Fog.Enabled) {
831 const GLfloat *fc = ctx->Fog.Color;
832 GLfloat color[4];
833 GLfloat f;
834
835 if (ctx->Fog.FogCoordinateSource == GL_FOG_COORDINATE_EXT)
836 f = _swrast_z_to_fogfactor(ctx, ctx->Current.Attrib[VERT_ATTRIB_FOG][0]);
837 else
838 f = _swrast_z_to_fogfactor(ctx, ctx->Current.RasterDistance);
839
840 color[0] = f * rc[0] + (1.F - f) * fc[0];
841 color[1] = f * rc[1] + (1.F - f) * fc[1];
842 color[2] = f * rc[2] + (1.F - f) * fc[2];
843 color[3] = rc[3];
844
845 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.red, color[0]);
846 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.green, color[1]);
847 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.blue, color[2]);
848 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.alpha, color[3]);
849 }
850 else {
851 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.red, rc[0]);
852 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.green, rc[1]);
853 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.blue, rc[2]);
854 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.alpha, rc[3]);
855 }
856
857
858 vert.tv.z = ctx->Current.RasterPos[2];
859
860
861 /* Update window height
862 */
863 h = radeon_get_drawable(&rmesa->radeon)->h;
864
865 /* Clipping handled by existing mechansims in r200_ioctl.c?
866 */
867 for (row=0; row<height; row++) {
868 const GLubyte *src = (const GLubyte *)
869 _mesa_image_address2d(unpack, bitmap, width, height,
870 GL_COLOR_INDEX, GL_BITMAP, row, 0 );
871
872 if (unpack->LsbFirst) {
873 /* Lsb first */
874 GLubyte mask = 1U << (unpack->SkipPixels & 0x7);
875 for (col=0; col<width; col++) {
876 if (*src & mask) {
877 vert.tv.x = px+col;
878 vert.tv.y = h - (py+row) - 1;
879 r200_point( rmesa, &vert );
880 }
881 src += (mask >> 7);
882 mask = ((mask << 1) & 0xff) | (mask >> 7);
883 }
884
885 /* get ready for next row */
886 if (mask != 1)
887 src++;
888 }
889 else {
890 /* Msb first */
891 GLubyte mask = 128U >> (unpack->SkipPixels & 0x7);
892 for (col=0; col<width; col++) {
893 if (*src & mask) {
894 vert.tv.x = px+col;
895 vert.tv.y = h - (py+row) - 1;
896 r200_point( rmesa, &vert );
897 }
898 src += mask & 1;
899 mask = ((mask << 7) & 0xff) | (mask >> 1);
900 }
901 /* get ready for next row */
902 if (mask != 128)
903 src++;
904 }
905 }
906
907 /* Fire outstanding vertices, restore state
908 */
909 R200_STATECHANGE( rmesa, vte );
910 rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] = orig_vte;
911
912 /* Unfallback
913 */
914 TCL_FALLBACK( ctx, R200_TCL_FALLBACK_BITMAP, 0 );
915
916 /* Need to restore vertexformat?
917 */
918 if (rmesa->radeon.TclFallback)
919 r200ChooseVertexState( ctx );
920 }
921
922
923
924 /**********************************************************************/
925 /* Initialization. */
926 /**********************************************************************/
927
928 void r200InitSwtcl( struct gl_context *ctx )
929 {
930 TNLcontext *tnl = TNL_CONTEXT(ctx);
931 r200ContextPtr rmesa = R200_CONTEXT(ctx);
932 static int firsttime = 1;
933
934 if (firsttime) {
935 init_rast_tab();
936 firsttime = 0;
937 }
938 rmesa->radeon.swtcl.emit_prediction = 0;
939
940 tnl->Driver.Render.Start = r200RenderStart;
941 tnl->Driver.Render.Finish = r200RenderFinish;
942 tnl->Driver.Render.PrimitiveNotify = r200RenderPrimitive;
943 tnl->Driver.Render.ResetLineStipple = r200ResetLineStipple;
944 tnl->Driver.Render.BuildVertices = _tnl_build_vertices;
945 tnl->Driver.Render.CopyPV = _tnl_copy_pv;
946 tnl->Driver.Render.Interp = _tnl_interp;
947
948 /* FIXME: what are these numbers? */
949 _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12,
950 36 * sizeof(GLfloat) );
951
952 rmesa->radeon.swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf;
953 rmesa->radeon.swtcl.RenderIndex = ~0;
954 rmesa->radeon.swtcl.render_primitive = GL_TRIANGLES;
955 rmesa->radeon.swtcl.hw_primitive = 0;
956 }
957