Merge remote branch 'main/master' into radeon-rewrite
[mesa.git] / src / mesa / drivers / dri / r200 / r200_texstate.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 <keith@tungstengraphics.com>
33 */
34
35 #include "main/glheader.h"
36 #include "main/imports.h"
37 #include "main/context.h"
38 #include "main/macros.h"
39 #include "main/texformat.h"
40 #include "main/teximage.h"
41 #include "main/texobj.h"
42 #include "main/enums.h"
43
44 #include "radeon_common.h"
45 #include "radeon_mipmap_tree.h"
46 #include "r200_context.h"
47 #include "r200_state.h"
48 #include "r200_ioctl.h"
49 #include "r200_swtcl.h"
50 #include "r200_tex.h"
51 #include "r200_tcl.h"
52
53
54 #define R200_TXFORMAT_A8 R200_TXFORMAT_I8
55 #define R200_TXFORMAT_L8 R200_TXFORMAT_I8
56 #define R200_TXFORMAT_AL88 R200_TXFORMAT_AI88
57 #define R200_TXFORMAT_YCBCR R200_TXFORMAT_YVYU422
58 #define R200_TXFORMAT_YCBCR_REV R200_TXFORMAT_VYUY422
59 #define R200_TXFORMAT_RGB_DXT1 R200_TXFORMAT_DXT1
60 #define R200_TXFORMAT_RGBA_DXT1 R200_TXFORMAT_DXT1
61 #define R200_TXFORMAT_RGBA_DXT3 R200_TXFORMAT_DXT23
62 #define R200_TXFORMAT_RGBA_DXT5 R200_TXFORMAT_DXT45
63
64 #define _COLOR(f) \
65 [ MESA_FORMAT_ ## f ] = { R200_TXFORMAT_ ## f, 0 }
66 #define _COLOR_REV(f) \
67 [ MESA_FORMAT_ ## f ## _REV ] = { R200_TXFORMAT_ ## f, 0 }
68 #define _ALPHA(f) \
69 [ MESA_FORMAT_ ## f ] = { R200_TXFORMAT_ ## f | R200_TXFORMAT_ALPHA_IN_MAP, 0 }
70 #define _ALPHA_REV(f) \
71 [ MESA_FORMAT_ ## f ## _REV ] = { R200_TXFORMAT_ ## f | R200_TXFORMAT_ALPHA_IN_MAP, 0 }
72 #define _YUV(f) \
73 [ MESA_FORMAT_ ## f ] = { R200_TXFORMAT_ ## f, R200_YUV_TO_RGB }
74 #define _INVALID(f) \
75 [ MESA_FORMAT_ ## f ] = { 0xffffffff, 0 }
76 #define VALID_FORMAT(f) ( ((f) <= MESA_FORMAT_RGBA_DXT5) \
77 && (tx_table_be[f].format != 0xffffffff) )
78
79 struct tx_table {
80 GLuint format, filter;
81 };
82
83 static const struct tx_table tx_table_be[] =
84 {
85 [ MESA_FORMAT_RGBA8888 ] = { R200_TXFORMAT_ABGR8888 | R200_TXFORMAT_ALPHA_IN_MAP, 0 },
86 _ALPHA_REV(RGBA8888),
87 _ALPHA(ARGB8888),
88 _ALPHA_REV(ARGB8888),
89 _INVALID(RGB888),
90 _COLOR(RGB565),
91 _COLOR_REV(RGB565),
92 _ALPHA(ARGB4444),
93 _ALPHA_REV(ARGB4444),
94 _ALPHA(ARGB1555),
95 _ALPHA_REV(ARGB1555),
96 _ALPHA(AL88),
97 _ALPHA_REV(AL88),
98 _ALPHA(A8),
99 _COLOR(L8),
100 _ALPHA(I8),
101 _INVALID(CI8),
102 _YUV(YCBCR),
103 _YUV(YCBCR_REV),
104 _INVALID(RGB_FXT1),
105 _INVALID(RGBA_FXT1),
106 _COLOR(RGB_DXT1),
107 _ALPHA(RGBA_DXT1),
108 _ALPHA(RGBA_DXT3),
109 _ALPHA(RGBA_DXT5),
110 };
111
112 static const struct tx_table tx_table_le[] =
113 {
114 _ALPHA(RGBA8888),
115 [ MESA_FORMAT_RGBA8888_REV ] = { R200_TXFORMAT_ABGR8888 | R200_TXFORMAT_ALPHA_IN_MAP, 0 },
116 _ALPHA(ARGB8888),
117 _ALPHA_REV(ARGB8888),
118 [ MESA_FORMAT_RGB888 ] = { R200_TXFORMAT_ARGB8888, 0 },
119 _COLOR(RGB565),
120 _COLOR_REV(RGB565),
121 _ALPHA(ARGB4444),
122 _ALPHA_REV(ARGB4444),
123 _ALPHA(ARGB1555),
124 _ALPHA_REV(ARGB1555),
125 _ALPHA(AL88),
126 _ALPHA_REV(AL88),
127 _ALPHA(A8),
128 _COLOR(L8),
129 _ALPHA(I8),
130 _INVALID(CI8),
131 _YUV(YCBCR),
132 _YUV(YCBCR_REV),
133 _INVALID(RGB_FXT1),
134 _INVALID(RGBA_FXT1),
135 _COLOR(RGB_DXT1),
136 _ALPHA(RGBA_DXT1),
137 _ALPHA(RGBA_DXT3),
138 _ALPHA(RGBA_DXT5),
139 };
140
141 #undef _COLOR
142 #undef _ALPHA
143 #undef _INVALID
144
145 /* ================================================================
146 * Texture combine functions
147 */
148
149 /* GL_ARB_texture_env_combine support
150 */
151
152 /* The color tables have combine functions for GL_SRC_COLOR,
153 * GL_ONE_MINUS_SRC_COLOR, GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
154 */
155 static GLuint r200_register_color[][R200_MAX_TEXTURE_UNITS] =
156 {
157 {
158 R200_TXC_ARG_A_R0_COLOR,
159 R200_TXC_ARG_A_R1_COLOR,
160 R200_TXC_ARG_A_R2_COLOR,
161 R200_TXC_ARG_A_R3_COLOR,
162 R200_TXC_ARG_A_R4_COLOR,
163 R200_TXC_ARG_A_R5_COLOR
164 },
165 {
166 R200_TXC_ARG_A_R0_COLOR | R200_TXC_COMP_ARG_A,
167 R200_TXC_ARG_A_R1_COLOR | R200_TXC_COMP_ARG_A,
168 R200_TXC_ARG_A_R2_COLOR | R200_TXC_COMP_ARG_A,
169 R200_TXC_ARG_A_R3_COLOR | R200_TXC_COMP_ARG_A,
170 R200_TXC_ARG_A_R4_COLOR | R200_TXC_COMP_ARG_A,
171 R200_TXC_ARG_A_R5_COLOR | R200_TXC_COMP_ARG_A
172 },
173 {
174 R200_TXC_ARG_A_R0_ALPHA,
175 R200_TXC_ARG_A_R1_ALPHA,
176 R200_TXC_ARG_A_R2_ALPHA,
177 R200_TXC_ARG_A_R3_ALPHA,
178 R200_TXC_ARG_A_R4_ALPHA,
179 R200_TXC_ARG_A_R5_ALPHA
180 },
181 {
182 R200_TXC_ARG_A_R0_ALPHA | R200_TXC_COMP_ARG_A,
183 R200_TXC_ARG_A_R1_ALPHA | R200_TXC_COMP_ARG_A,
184 R200_TXC_ARG_A_R2_ALPHA | R200_TXC_COMP_ARG_A,
185 R200_TXC_ARG_A_R3_ALPHA | R200_TXC_COMP_ARG_A,
186 R200_TXC_ARG_A_R4_ALPHA | R200_TXC_COMP_ARG_A,
187 R200_TXC_ARG_A_R5_ALPHA | R200_TXC_COMP_ARG_A
188 },
189 };
190
191 static GLuint r200_tfactor_color[] =
192 {
193 R200_TXC_ARG_A_TFACTOR_COLOR,
194 R200_TXC_ARG_A_TFACTOR_COLOR | R200_TXC_COMP_ARG_A,
195 R200_TXC_ARG_A_TFACTOR_ALPHA,
196 R200_TXC_ARG_A_TFACTOR_ALPHA | R200_TXC_COMP_ARG_A
197 };
198
199 static GLuint r200_tfactor1_color[] =
200 {
201 R200_TXC_ARG_A_TFACTOR1_COLOR,
202 R200_TXC_ARG_A_TFACTOR1_COLOR | R200_TXC_COMP_ARG_A,
203 R200_TXC_ARG_A_TFACTOR1_ALPHA,
204 R200_TXC_ARG_A_TFACTOR1_ALPHA | R200_TXC_COMP_ARG_A
205 };
206
207 static GLuint r200_primary_color[] =
208 {
209 R200_TXC_ARG_A_DIFFUSE_COLOR,
210 R200_TXC_ARG_A_DIFFUSE_COLOR | R200_TXC_COMP_ARG_A,
211 R200_TXC_ARG_A_DIFFUSE_ALPHA,
212 R200_TXC_ARG_A_DIFFUSE_ALPHA | R200_TXC_COMP_ARG_A
213 };
214
215 /* GL_ZERO table - indices 0-3
216 * GL_ONE table - indices 1-4
217 */
218 static GLuint r200_zero_color[] =
219 {
220 R200_TXC_ARG_A_ZERO,
221 R200_TXC_ARG_A_ZERO | R200_TXC_COMP_ARG_A,
222 R200_TXC_ARG_A_ZERO,
223 R200_TXC_ARG_A_ZERO | R200_TXC_COMP_ARG_A,
224 R200_TXC_ARG_A_ZERO
225 };
226
227 /* The alpha tables only have GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
228 */
229 static GLuint r200_register_alpha[][R200_MAX_TEXTURE_UNITS] =
230 {
231 {
232 R200_TXA_ARG_A_R0_ALPHA,
233 R200_TXA_ARG_A_R1_ALPHA,
234 R200_TXA_ARG_A_R2_ALPHA,
235 R200_TXA_ARG_A_R3_ALPHA,
236 R200_TXA_ARG_A_R4_ALPHA,
237 R200_TXA_ARG_A_R5_ALPHA
238 },
239 {
240 R200_TXA_ARG_A_R0_ALPHA | R200_TXA_COMP_ARG_A,
241 R200_TXA_ARG_A_R1_ALPHA | R200_TXA_COMP_ARG_A,
242 R200_TXA_ARG_A_R2_ALPHA | R200_TXA_COMP_ARG_A,
243 R200_TXA_ARG_A_R3_ALPHA | R200_TXA_COMP_ARG_A,
244 R200_TXA_ARG_A_R4_ALPHA | R200_TXA_COMP_ARG_A,
245 R200_TXA_ARG_A_R5_ALPHA | R200_TXA_COMP_ARG_A
246 },
247 };
248
249 static GLuint r200_tfactor_alpha[] =
250 {
251 R200_TXA_ARG_A_TFACTOR_ALPHA,
252 R200_TXA_ARG_A_TFACTOR_ALPHA | R200_TXA_COMP_ARG_A
253 };
254
255 static GLuint r200_tfactor1_alpha[] =
256 {
257 R200_TXA_ARG_A_TFACTOR1_ALPHA,
258 R200_TXA_ARG_A_TFACTOR1_ALPHA | R200_TXA_COMP_ARG_A
259 };
260
261 static GLuint r200_primary_alpha[] =
262 {
263 R200_TXA_ARG_A_DIFFUSE_ALPHA,
264 R200_TXA_ARG_A_DIFFUSE_ALPHA | R200_TXA_COMP_ARG_A
265 };
266
267 /* GL_ZERO table - indices 0-1
268 * GL_ONE table - indices 1-2
269 */
270 static GLuint r200_zero_alpha[] =
271 {
272 R200_TXA_ARG_A_ZERO,
273 R200_TXA_ARG_A_ZERO | R200_TXA_COMP_ARG_A,
274 R200_TXA_ARG_A_ZERO,
275 };
276
277
278 /* Extract the arg from slot A, shift it into the correct argument slot
279 * and set the corresponding complement bit.
280 */
281 #define R200_COLOR_ARG( n, arg ) \
282 do { \
283 color_combine |= \
284 ((color_arg[n] & R200_TXC_ARG_A_MASK) \
285 << R200_TXC_ARG_##arg##_SHIFT); \
286 color_combine |= \
287 ((color_arg[n] >> R200_TXC_COMP_ARG_A_SHIFT) \
288 << R200_TXC_COMP_ARG_##arg##_SHIFT); \
289 } while (0)
290
291 #define R200_ALPHA_ARG( n, arg ) \
292 do { \
293 alpha_combine |= \
294 ((alpha_arg[n] & R200_TXA_ARG_A_MASK) \
295 << R200_TXA_ARG_##arg##_SHIFT); \
296 alpha_combine |= \
297 ((alpha_arg[n] >> R200_TXA_COMP_ARG_A_SHIFT) \
298 << R200_TXA_COMP_ARG_##arg##_SHIFT); \
299 } while (0)
300
301
302 /* ================================================================
303 * Texture unit state management
304 */
305
306 static GLboolean r200UpdateTextureEnv( GLcontext *ctx, int unit, int slot, GLuint replaceargs )
307 {
308 r200ContextPtr rmesa = R200_CONTEXT(ctx);
309 const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
310 GLuint color_combine, alpha_combine;
311 GLuint color_scale = rmesa->hw.pix[slot].cmd[PIX_PP_TXCBLEND2] &
312 ~(R200_TXC_SCALE_MASK | R200_TXC_OUTPUT_REG_MASK | R200_TXC_TFACTOR_SEL_MASK |
313 R200_TXC_TFACTOR1_SEL_MASK);
314 GLuint alpha_scale = rmesa->hw.pix[slot].cmd[PIX_PP_TXABLEND2] &
315 ~(R200_TXA_DOT_ALPHA | R200_TXA_SCALE_MASK | R200_TXA_OUTPUT_REG_MASK |
316 R200_TXA_TFACTOR_SEL_MASK | R200_TXA_TFACTOR1_SEL_MASK);
317
318 /* texUnit->_Current can be NULL if and only if the texture unit is
319 * not actually enabled.
320 */
321 assert( (texUnit->_ReallyEnabled == 0)
322 || (texUnit->_Current != NULL) );
323
324 if ( R200_DEBUG & DEBUG_TEXTURE ) {
325 fprintf( stderr, "%s( %p, %d )\n", __FUNCTION__, (void *)ctx, unit );
326 }
327
328 /* Set the texture environment state. Isn't this nice and clean?
329 * The chip will automagically set the texture alpha to 0xff when
330 * the texture format does not include an alpha component. This
331 * reduces the amount of special-casing we have to do, alpha-only
332 * textures being a notable exception.
333 */
334
335 color_scale |= ((rmesa->state.texture.unit[unit].outputreg + 1) << R200_TXC_OUTPUT_REG_SHIFT) |
336 (unit << R200_TXC_TFACTOR_SEL_SHIFT) |
337 (replaceargs << R200_TXC_TFACTOR1_SEL_SHIFT);
338 alpha_scale |= ((rmesa->state.texture.unit[unit].outputreg + 1) << R200_TXA_OUTPUT_REG_SHIFT) |
339 (unit << R200_TXA_TFACTOR_SEL_SHIFT) |
340 (replaceargs << R200_TXA_TFACTOR1_SEL_SHIFT);
341
342 if ( !texUnit->_ReallyEnabled ) {
343 assert( unit == 0);
344 color_combine = R200_TXC_ARG_A_ZERO | R200_TXC_ARG_B_ZERO
345 | R200_TXC_ARG_C_DIFFUSE_COLOR | R200_TXC_OP_MADD;
346 alpha_combine = R200_TXA_ARG_A_ZERO | R200_TXA_ARG_B_ZERO
347 | R200_TXA_ARG_C_DIFFUSE_ALPHA | R200_TXA_OP_MADD;
348 }
349 else {
350 GLuint color_arg[3], alpha_arg[3];
351 GLuint i;
352 const GLuint numColorArgs = texUnit->_CurrentCombine->_NumArgsRGB;
353 const GLuint numAlphaArgs = texUnit->_CurrentCombine->_NumArgsA;
354 GLuint RGBshift = texUnit->_CurrentCombine->ScaleShiftRGB;
355 GLuint Ashift = texUnit->_CurrentCombine->ScaleShiftA;
356
357
358 const GLint replaceoprgb =
359 ctx->Texture.Unit[replaceargs]._CurrentCombine->OperandRGB[0] - GL_SRC_COLOR;
360 const GLint replaceopa =
361 ctx->Texture.Unit[replaceargs]._CurrentCombine->OperandA[0] - GL_SRC_ALPHA;
362
363 /* Step 1:
364 * Extract the color and alpha combine function arguments.
365 */
366 for ( i = 0 ; i < numColorArgs ; i++ ) {
367 GLint op = texUnit->_CurrentCombine->OperandRGB[i] - GL_SRC_COLOR;
368 const GLint srcRGBi = texUnit->_CurrentCombine->SourceRGB[i];
369 assert(op >= 0);
370 assert(op <= 3);
371 switch ( srcRGBi ) {
372 case GL_TEXTURE:
373 color_arg[i] = r200_register_color[op][unit];
374 break;
375 case GL_CONSTANT:
376 color_arg[i] = r200_tfactor_color[op];
377 break;
378 case GL_PRIMARY_COLOR:
379 color_arg[i] = r200_primary_color[op];
380 break;
381 case GL_PREVIOUS:
382 if (replaceargs != unit) {
383 const GLint srcRGBreplace =
384 ctx->Texture.Unit[replaceargs]._CurrentCombine->SourceRGB[0];
385 if (op >= 2) {
386 op = op ^ replaceopa;
387 }
388 else {
389 op = op ^ replaceoprgb;
390 }
391 switch (srcRGBreplace) {
392 case GL_TEXTURE:
393 color_arg[i] = r200_register_color[op][replaceargs];
394 break;
395 case GL_CONSTANT:
396 color_arg[i] = r200_tfactor1_color[op];
397 break;
398 case GL_PRIMARY_COLOR:
399 color_arg[i] = r200_primary_color[op];
400 break;
401 case GL_PREVIOUS:
402 if (slot == 0)
403 color_arg[i] = r200_primary_color[op];
404 else
405 color_arg[i] = r200_register_color[op]
406 [rmesa->state.texture.unit[replaceargs - 1].outputreg];
407 break;
408 case GL_ZERO:
409 color_arg[i] = r200_zero_color[op];
410 break;
411 case GL_ONE:
412 color_arg[i] = r200_zero_color[op+1];
413 break;
414 case GL_TEXTURE0:
415 case GL_TEXTURE1:
416 case GL_TEXTURE2:
417 case GL_TEXTURE3:
418 case GL_TEXTURE4:
419 case GL_TEXTURE5:
420 color_arg[i] = r200_register_color[op][srcRGBreplace - GL_TEXTURE0];
421 break;
422 default:
423 return GL_FALSE;
424 }
425 }
426 else {
427 if (slot == 0)
428 color_arg[i] = r200_primary_color[op];
429 else
430 color_arg[i] = r200_register_color[op]
431 [rmesa->state.texture.unit[unit - 1].outputreg];
432 }
433 break;
434 case GL_ZERO:
435 color_arg[i] = r200_zero_color[op];
436 break;
437 case GL_ONE:
438 color_arg[i] = r200_zero_color[op+1];
439 break;
440 case GL_TEXTURE0:
441 case GL_TEXTURE1:
442 case GL_TEXTURE2:
443 case GL_TEXTURE3:
444 case GL_TEXTURE4:
445 case GL_TEXTURE5:
446 color_arg[i] = r200_register_color[op][srcRGBi - GL_TEXTURE0];
447 break;
448 default:
449 return GL_FALSE;
450 }
451 }
452
453 for ( i = 0 ; i < numAlphaArgs ; i++ ) {
454 GLint op = texUnit->_CurrentCombine->OperandA[i] - GL_SRC_ALPHA;
455 const GLint srcAi = texUnit->_CurrentCombine->SourceA[i];
456 assert(op >= 0);
457 assert(op <= 1);
458 switch ( srcAi ) {
459 case GL_TEXTURE:
460 alpha_arg[i] = r200_register_alpha[op][unit];
461 break;
462 case GL_CONSTANT:
463 alpha_arg[i] = r200_tfactor_alpha[op];
464 break;
465 case GL_PRIMARY_COLOR:
466 alpha_arg[i] = r200_primary_alpha[op];
467 break;
468 case GL_PREVIOUS:
469 if (replaceargs != unit) {
470 const GLint srcAreplace =
471 ctx->Texture.Unit[replaceargs]._CurrentCombine->SourceA[0];
472 op = op ^ replaceopa;
473 switch (srcAreplace) {
474 case GL_TEXTURE:
475 alpha_arg[i] = r200_register_alpha[op][replaceargs];
476 break;
477 case GL_CONSTANT:
478 alpha_arg[i] = r200_tfactor1_alpha[op];
479 break;
480 case GL_PRIMARY_COLOR:
481 alpha_arg[i] = r200_primary_alpha[op];
482 break;
483 case GL_PREVIOUS:
484 if (slot == 0)
485 alpha_arg[i] = r200_primary_alpha[op];
486 else
487 alpha_arg[i] = r200_register_alpha[op]
488 [rmesa->state.texture.unit[replaceargs - 1].outputreg];
489 break;
490 case GL_ZERO:
491 alpha_arg[i] = r200_zero_alpha[op];
492 break;
493 case GL_ONE:
494 alpha_arg[i] = r200_zero_alpha[op+1];
495 break;
496 case GL_TEXTURE0:
497 case GL_TEXTURE1:
498 case GL_TEXTURE2:
499 case GL_TEXTURE3:
500 case GL_TEXTURE4:
501 case GL_TEXTURE5:
502 alpha_arg[i] = r200_register_alpha[op][srcAreplace - GL_TEXTURE0];
503 break;
504 default:
505 return GL_FALSE;
506 }
507 }
508 else {
509 if (slot == 0)
510 alpha_arg[i] = r200_primary_alpha[op];
511 else
512 alpha_arg[i] = r200_register_alpha[op]
513 [rmesa->state.texture.unit[unit - 1].outputreg];
514 }
515 break;
516 case GL_ZERO:
517 alpha_arg[i] = r200_zero_alpha[op];
518 break;
519 case GL_ONE:
520 alpha_arg[i] = r200_zero_alpha[op+1];
521 break;
522 case GL_TEXTURE0:
523 case GL_TEXTURE1:
524 case GL_TEXTURE2:
525 case GL_TEXTURE3:
526 case GL_TEXTURE4:
527 case GL_TEXTURE5:
528 alpha_arg[i] = r200_register_alpha[op][srcAi - GL_TEXTURE0];
529 break;
530 default:
531 return GL_FALSE;
532 }
533 }
534
535 /* Step 2:
536 * Build up the color and alpha combine functions.
537 */
538 switch ( texUnit->_CurrentCombine->ModeRGB ) {
539 case GL_REPLACE:
540 color_combine = (R200_TXC_ARG_A_ZERO |
541 R200_TXC_ARG_B_ZERO |
542 R200_TXC_OP_MADD);
543 R200_COLOR_ARG( 0, C );
544 break;
545 case GL_MODULATE:
546 color_combine = (R200_TXC_ARG_C_ZERO |
547 R200_TXC_OP_MADD);
548 R200_COLOR_ARG( 0, A );
549 R200_COLOR_ARG( 1, B );
550 break;
551 case GL_ADD:
552 color_combine = (R200_TXC_ARG_B_ZERO |
553 R200_TXC_COMP_ARG_B |
554 R200_TXC_OP_MADD);
555 R200_COLOR_ARG( 0, A );
556 R200_COLOR_ARG( 1, C );
557 break;
558 case GL_ADD_SIGNED:
559 color_combine = (R200_TXC_ARG_B_ZERO |
560 R200_TXC_COMP_ARG_B |
561 R200_TXC_BIAS_ARG_C | /* new */
562 R200_TXC_OP_MADD); /* was ADDSIGNED */
563 R200_COLOR_ARG( 0, A );
564 R200_COLOR_ARG( 1, C );
565 break;
566 case GL_SUBTRACT:
567 color_combine = (R200_TXC_ARG_B_ZERO |
568 R200_TXC_COMP_ARG_B |
569 R200_TXC_NEG_ARG_C |
570 R200_TXC_OP_MADD);
571 R200_COLOR_ARG( 0, A );
572 R200_COLOR_ARG( 1, C );
573 break;
574 case GL_INTERPOLATE:
575 color_combine = (R200_TXC_OP_LERP);
576 R200_COLOR_ARG( 0, B );
577 R200_COLOR_ARG( 1, A );
578 R200_COLOR_ARG( 2, C );
579 break;
580
581 case GL_DOT3_RGB_EXT:
582 case GL_DOT3_RGBA_EXT:
583 /* The EXT version of the DOT3 extension does not support the
584 * scale factor, but the ARB version (and the version in OpenGL
585 * 1.3) does.
586 */
587 RGBshift = 0;
588 /* FALLTHROUGH */
589
590 case GL_DOT3_RGB:
591 case GL_DOT3_RGBA:
592 /* DOT3 works differently on R200 than on R100. On R100, just
593 * setting the DOT3 mode did everything for you. On R200, the
594 * driver has to enable the biasing and scale in the inputs to
595 * put them in the proper [-1,1] range. This is what the 4x and
596 * the -0.5 in the DOT3 spec do. The post-scale is then set
597 * normally.
598 */
599
600 color_combine = (R200_TXC_ARG_C_ZERO |
601 R200_TXC_OP_DOT3 |
602 R200_TXC_BIAS_ARG_A |
603 R200_TXC_BIAS_ARG_B |
604 R200_TXC_SCALE_ARG_A |
605 R200_TXC_SCALE_ARG_B);
606 R200_COLOR_ARG( 0, A );
607 R200_COLOR_ARG( 1, B );
608 break;
609
610 case GL_MODULATE_ADD_ATI:
611 color_combine = (R200_TXC_OP_MADD);
612 R200_COLOR_ARG( 0, A );
613 R200_COLOR_ARG( 1, C );
614 R200_COLOR_ARG( 2, B );
615 break;
616 case GL_MODULATE_SIGNED_ADD_ATI:
617 color_combine = (R200_TXC_BIAS_ARG_C | /* new */
618 R200_TXC_OP_MADD); /* was ADDSIGNED */
619 R200_COLOR_ARG( 0, A );
620 R200_COLOR_ARG( 1, C );
621 R200_COLOR_ARG( 2, B );
622 break;
623 case GL_MODULATE_SUBTRACT_ATI:
624 color_combine = (R200_TXC_NEG_ARG_C |
625 R200_TXC_OP_MADD);
626 R200_COLOR_ARG( 0, A );
627 R200_COLOR_ARG( 1, C );
628 R200_COLOR_ARG( 2, B );
629 break;
630 default:
631 return GL_FALSE;
632 }
633
634 switch ( texUnit->_CurrentCombine->ModeA ) {
635 case GL_REPLACE:
636 alpha_combine = (R200_TXA_ARG_A_ZERO |
637 R200_TXA_ARG_B_ZERO |
638 R200_TXA_OP_MADD);
639 R200_ALPHA_ARG( 0, C );
640 break;
641 case GL_MODULATE:
642 alpha_combine = (R200_TXA_ARG_C_ZERO |
643 R200_TXA_OP_MADD);
644 R200_ALPHA_ARG( 0, A );
645 R200_ALPHA_ARG( 1, B );
646 break;
647 case GL_ADD:
648 alpha_combine = (R200_TXA_ARG_B_ZERO |
649 R200_TXA_COMP_ARG_B |
650 R200_TXA_OP_MADD);
651 R200_ALPHA_ARG( 0, A );
652 R200_ALPHA_ARG( 1, C );
653 break;
654 case GL_ADD_SIGNED:
655 alpha_combine = (R200_TXA_ARG_B_ZERO |
656 R200_TXA_COMP_ARG_B |
657 R200_TXA_BIAS_ARG_C | /* new */
658 R200_TXA_OP_MADD); /* was ADDSIGNED */
659 R200_ALPHA_ARG( 0, A );
660 R200_ALPHA_ARG( 1, C );
661 break;
662 case GL_SUBTRACT:
663 alpha_combine = (R200_TXA_ARG_B_ZERO |
664 R200_TXA_COMP_ARG_B |
665 R200_TXA_NEG_ARG_C |
666 R200_TXA_OP_MADD);
667 R200_ALPHA_ARG( 0, A );
668 R200_ALPHA_ARG( 1, C );
669 break;
670 case GL_INTERPOLATE:
671 alpha_combine = (R200_TXA_OP_LERP);
672 R200_ALPHA_ARG( 0, B );
673 R200_ALPHA_ARG( 1, A );
674 R200_ALPHA_ARG( 2, C );
675 break;
676
677 case GL_MODULATE_ADD_ATI:
678 alpha_combine = (R200_TXA_OP_MADD);
679 R200_ALPHA_ARG( 0, A );
680 R200_ALPHA_ARG( 1, C );
681 R200_ALPHA_ARG( 2, B );
682 break;
683 case GL_MODULATE_SIGNED_ADD_ATI:
684 alpha_combine = (R200_TXA_BIAS_ARG_C | /* new */
685 R200_TXA_OP_MADD); /* was ADDSIGNED */
686 R200_ALPHA_ARG( 0, A );
687 R200_ALPHA_ARG( 1, C );
688 R200_ALPHA_ARG( 2, B );
689 break;
690 case GL_MODULATE_SUBTRACT_ATI:
691 alpha_combine = (R200_TXA_NEG_ARG_C |
692 R200_TXA_OP_MADD);
693 R200_ALPHA_ARG( 0, A );
694 R200_ALPHA_ARG( 1, C );
695 R200_ALPHA_ARG( 2, B );
696 break;
697 default:
698 return GL_FALSE;
699 }
700
701 if ( (texUnit->_CurrentCombine->ModeRGB == GL_DOT3_RGBA_EXT)
702 || (texUnit->_CurrentCombine->ModeRGB == GL_DOT3_RGBA) ) {
703 alpha_scale |= R200_TXA_DOT_ALPHA;
704 Ashift = RGBshift;
705 }
706
707 /* Step 3:
708 * Apply the scale factor.
709 */
710 color_scale |= (RGBshift << R200_TXC_SCALE_SHIFT);
711 alpha_scale |= (Ashift << R200_TXA_SCALE_SHIFT);
712
713 /* All done!
714 */
715 }
716
717 if ( rmesa->hw.pix[slot].cmd[PIX_PP_TXCBLEND] != color_combine ||
718 rmesa->hw.pix[slot].cmd[PIX_PP_TXABLEND] != alpha_combine ||
719 rmesa->hw.pix[slot].cmd[PIX_PP_TXCBLEND2] != color_scale ||
720 rmesa->hw.pix[slot].cmd[PIX_PP_TXABLEND2] != alpha_scale) {
721 R200_STATECHANGE( rmesa, pix[slot] );
722 rmesa->hw.pix[slot].cmd[PIX_PP_TXCBLEND] = color_combine;
723 rmesa->hw.pix[slot].cmd[PIX_PP_TXABLEND] = alpha_combine;
724 rmesa->hw.pix[slot].cmd[PIX_PP_TXCBLEND2] = color_scale;
725 rmesa->hw.pix[slot].cmd[PIX_PP_TXABLEND2] = alpha_scale;
726 }
727
728 return GL_TRUE;
729 }
730
731 void r200SetTexOffset(__DRIcontext * pDRICtx, GLint texname,
732 unsigned long long offset, GLint depth, GLuint pitch)
733 {
734 r200ContextPtr rmesa = pDRICtx->driverPrivate;
735 struct gl_texture_object *tObj =
736 _mesa_lookup_texture(rmesa->radeon.glCtx, texname);
737 radeonTexObjPtr t = radeon_tex_obj(tObj);
738
739 if (!tObj)
740 return;
741
742 t->image_override = GL_TRUE;
743
744 if (!offset)
745 return;
746
747 t->bo = NULL;
748 t->override_offset = offset;
749 t->pp_txpitch = pitch - 32;
750
751 switch (depth) {
752 case 32:
753 t->pp_txformat = tx_table_le[MESA_FORMAT_ARGB8888].format;
754 t->pp_txfilter |= tx_table_le[MESA_FORMAT_ARGB8888].filter;
755 break;
756 case 24:
757 default:
758 t->pp_txformat = tx_table_le[MESA_FORMAT_RGB888].format;
759 t->pp_txfilter |= tx_table_le[MESA_FORMAT_RGB888].filter;
760 break;
761 case 16:
762 t->pp_txformat = tx_table_le[MESA_FORMAT_RGB565].format;
763 t->pp_txfilter |= tx_table_le[MESA_FORMAT_RGB565].filter;
764 break;
765 }
766 }
767
768 void r200SetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv)
769 {
770 struct gl_texture_unit *texUnit;
771 struct gl_texture_object *texObj;
772 struct gl_texture_image *texImage;
773 struct radeon_renderbuffer *rb;
774 radeon_texture_image *rImage;
775 radeonContextPtr radeon;
776 r200ContextPtr rmesa;
777 struct radeon_framebuffer *rfb;
778 radeonTexObjPtr t;
779 uint32_t pitch_val;
780
781 target = GL_TEXTURE_RECTANGLE_ARB;
782
783 radeon = pDRICtx->driverPrivate;
784 rmesa = pDRICtx->driverPrivate;
785
786 rfb = dPriv->driverPrivate;
787 texUnit = &radeon->glCtx->Texture.Unit[radeon->glCtx->Texture.CurrentUnit];
788 texObj = _mesa_select_tex_object(radeon->glCtx, texUnit, target);
789 texImage = _mesa_get_tex_image(radeon->glCtx, texObj, target, 0);
790
791 rImage = get_radeon_texture_image(texImage);
792 t = radeon_tex_obj(texObj);
793 if (t == NULL) {
794 return;
795 }
796
797 radeon_update_renderbuffers(pDRICtx, dPriv);
798 /* back & depth buffer are useless free them right away */
799 rb = (void*)rfb->base.Attachment[BUFFER_DEPTH].Renderbuffer;
800 if (rb && rb->bo) {
801 radeon_bo_unref(rb->bo);
802 rb->bo = NULL;
803 }
804 rb = (void*)rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer;
805 if (rb && rb->bo) {
806 radeon_bo_unref(rb->bo);
807 rb->bo = NULL;
808 }
809 rb = rfb->color_rb[0];
810 if (rb->bo == NULL) {
811 /* Failed to BO for the buffer */
812 return;
813 }
814
815 _mesa_lock_texture(radeon->glCtx, texObj);
816 if (t->bo) {
817 radeon_bo_unref(t->bo);
818 t->bo = NULL;
819 }
820 if (rImage->bo) {
821 radeon_bo_unref(rImage->bo);
822 rImage->bo = NULL;
823 }
824 if (t->mt) {
825 radeon_miptree_unreference(t->mt);
826 t->mt = NULL;
827 }
828 if (rImage->mt) {
829 radeon_miptree_unreference(rImage->mt);
830 rImage->mt = NULL;
831 }
832 fprintf(stderr,"settexbuf %d %dx%d@%d\n", rb->pitch, rb->width, rb->height, rb->cpp);
833 _mesa_init_teximage_fields(radeon->glCtx, target, texImage,
834 rb->width, rb->height, 1, 0, rb->cpp);
835 texImage->TexFormat = &_mesa_texformat_rgba8888_rev;
836 rImage->bo = rb->bo;
837 radeon_bo_ref(rImage->bo);
838 t->bo = rb->bo;
839 radeon_bo_ref(t->bo);
840 t->tile_bits = 0;
841 t->image_override = GL_TRUE;
842 t->override_offset = 0;
843 t->pp_txpitch &= (1 << 13) -1;
844 pitch_val = rb->pitch;
845 switch (rb->cpp) {
846 case 4:
847 t->pp_txformat = tx_table_le[MESA_FORMAT_ARGB8888].format;
848 t->pp_txfilter |= tx_table_le[MESA_FORMAT_ARGB8888].filter;
849 break;
850 case 3:
851 default:
852 t->pp_txformat = tx_table_le[MESA_FORMAT_RGB888].format;
853 t->pp_txfilter |= tx_table_le[MESA_FORMAT_RGB888].filter;
854 break;
855 case 2:
856 t->pp_txformat = tx_table_le[MESA_FORMAT_RGB565].format;
857 t->pp_txfilter |= tx_table_le[MESA_FORMAT_RGB565].filter;
858 break;
859 }
860 t->pp_txsize = ((rb->width - 1) << RADEON_TEX_USIZE_SHIFT)
861 | ((rb->height - 1) << RADEON_TEX_VSIZE_SHIFT);
862 t->pp_txformat |= R200_TXFORMAT_NON_POWER2;
863 t->pp_txpitch = pitch_val;
864 t->pp_txpitch -= 32;
865
866 t->validated = GL_TRUE;
867 _mesa_unlock_texture(radeon->glCtx, texObj);
868 return;
869 }
870
871 #define REF_COLOR 1
872 #define REF_ALPHA 2
873
874 static GLboolean r200UpdateAllTexEnv( GLcontext *ctx )
875 {
876 r200ContextPtr rmesa = R200_CONTEXT(ctx);
877 GLint i, j, currslot;
878 GLint maxunitused = -1;
879 GLboolean texregfree[6] = {GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE};
880 GLubyte stageref[7] = {0, 0, 0, 0, 0, 0, 0};
881 GLint nextunit[R200_MAX_TEXTURE_UNITS] = {0, 0, 0, 0, 0, 0};
882 GLint currentnext = -1;
883 GLboolean ok;
884
885 /* find highest used unit */
886 for ( j = 0; j < R200_MAX_TEXTURE_UNITS; j++) {
887 if (ctx->Texture.Unit[j]._ReallyEnabled) {
888 maxunitused = j;
889 }
890 }
891 stageref[maxunitused + 1] = REF_COLOR | REF_ALPHA;
892
893 for ( j = maxunitused; j >= 0; j-- ) {
894 const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[j];
895
896 rmesa->state.texture.unit[j].outputreg = -1;
897
898 if (stageref[j + 1]) {
899
900 /* use the lowest available reg. That gets us automatically reg0 for the last stage.
901 need this even for disabled units, as it may get referenced due to the replace
902 optimization */
903 for ( i = 0 ; i < R200_MAX_TEXTURE_UNITS; i++ ) {
904 if (texregfree[i]) {
905 rmesa->state.texture.unit[j].outputreg = i;
906 break;
907 }
908 }
909 if (rmesa->state.texture.unit[j].outputreg == -1) {
910 /* no more free regs we can use. Need a fallback :-( */
911 return GL_FALSE;
912 }
913
914 nextunit[j] = currentnext;
915
916 if (!texUnit->_ReallyEnabled) {
917 /* the not enabled stages are referenced "indirectly",
918 must not cut off the lower stages */
919 stageref[j] = REF_COLOR | REF_ALPHA;
920 continue;
921 }
922 currentnext = j;
923
924 const GLuint numColorArgs = texUnit->_CurrentCombine->_NumArgsRGB;
925 const GLuint numAlphaArgs = texUnit->_CurrentCombine->_NumArgsA;
926 const GLboolean isdot3rgba = (texUnit->_CurrentCombine->ModeRGB == GL_DOT3_RGBA) ||
927 (texUnit->_CurrentCombine->ModeRGB == GL_DOT3_RGBA_EXT);
928
929
930 /* check if we need the color part, special case for dot3_rgba
931 as if only the alpha part is referenced later on it still is using the color part */
932 if ((stageref[j + 1] & REF_COLOR) || isdot3rgba) {
933 for ( i = 0 ; i < numColorArgs ; i++ ) {
934 const GLuint srcRGBi = texUnit->_CurrentCombine->SourceRGB[i];
935 const GLuint op = texUnit->_CurrentCombine->OperandRGB[i];
936 switch ( srcRGBi ) {
937 case GL_PREVIOUS:
938 /* op 0/1 are referencing color, op 2/3 alpha */
939 stageref[j] |= (op >> 1) + 1;
940 break;
941 case GL_TEXTURE:
942 texregfree[j] = GL_FALSE;
943 break;
944 case GL_TEXTURE0:
945 case GL_TEXTURE1:
946 case GL_TEXTURE2:
947 case GL_TEXTURE3:
948 case GL_TEXTURE4:
949 case GL_TEXTURE5:
950 texregfree[srcRGBi - GL_TEXTURE0] = GL_FALSE;
951 break;
952 default: /* don't care about other sources here */
953 break;
954 }
955 }
956 }
957
958 /* alpha args are ignored for dot3_rgba */
959 if ((stageref[j + 1] & REF_ALPHA) && !isdot3rgba) {
960
961 for ( i = 0 ; i < numAlphaArgs ; i++ ) {
962 const GLuint srcAi = texUnit->_CurrentCombine->SourceA[i];
963 switch ( srcAi ) {
964 case GL_PREVIOUS:
965 stageref[j] |= REF_ALPHA;
966 break;
967 case GL_TEXTURE:
968 texregfree[j] = GL_FALSE;
969 break;
970 case GL_TEXTURE0:
971 case GL_TEXTURE1:
972 case GL_TEXTURE2:
973 case GL_TEXTURE3:
974 case GL_TEXTURE4:
975 case GL_TEXTURE5:
976 texregfree[srcAi - GL_TEXTURE0] = GL_FALSE;
977 break;
978 default: /* don't care about other sources here */
979 break;
980 }
981 }
982 }
983 }
984 }
985
986 /* don't enable texture sampling for units if the result is not used */
987 for (i = 0; i < R200_MAX_TEXTURE_UNITS; i++) {
988 if (ctx->Texture.Unit[i]._ReallyEnabled && !texregfree[i])
989 rmesa->state.texture.unit[i].unitneeded = ctx->Texture.Unit[i]._ReallyEnabled;
990 else rmesa->state.texture.unit[i].unitneeded = 0;
991 }
992
993 ok = GL_TRUE;
994 currslot = 0;
995 rmesa->state.envneeded = 1;
996
997 i = 0;
998 while ((i <= maxunitused) && (i >= 0)) {
999 /* only output instruction if the results are referenced */
1000 if (ctx->Texture.Unit[i]._ReallyEnabled && stageref[i+1]) {
1001 GLuint replaceunit = i;
1002 /* try to optimize GL_REPLACE away (only one level deep though) */
1003 if ( (ctx->Texture.Unit[i]._CurrentCombine->ModeRGB == GL_REPLACE) &&
1004 (ctx->Texture.Unit[i]._CurrentCombine->ModeA == GL_REPLACE) &&
1005 (ctx->Texture.Unit[i]._CurrentCombine->ScaleShiftRGB == 0) &&
1006 (ctx->Texture.Unit[i]._CurrentCombine->ScaleShiftA == 0) &&
1007 (nextunit[i] > 0) ) {
1008 /* yippie! can optimize it away! */
1009 replaceunit = i;
1010 i = nextunit[i];
1011 }
1012
1013 /* need env instruction slot */
1014 rmesa->state.envneeded |= 1 << currslot;
1015 ok = r200UpdateTextureEnv( ctx, i, currslot, replaceunit );
1016 if (!ok) return GL_FALSE;
1017 currslot++;
1018 }
1019 i = i + 1;
1020 }
1021
1022 if (currslot == 0) {
1023 /* need one stage at least */
1024 rmesa->state.texture.unit[0].outputreg = 0;
1025 ok = r200UpdateTextureEnv( ctx, 0, 0, 0 );
1026 }
1027
1028 R200_STATECHANGE( rmesa, ctx );
1029 rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~(R200_TEX_BLEND_ENABLE_MASK | R200_MULTI_PASS_ENABLE);
1030 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= rmesa->state.envneeded << R200_TEX_BLEND_0_ENABLE_SHIFT;
1031
1032 return ok;
1033 }
1034
1035 #undef REF_COLOR
1036 #undef REF_ALPHA
1037
1038
1039 #define TEXOBJ_TXFILTER_MASK (R200_MAX_MIP_LEVEL_MASK | \
1040 R200_MIN_FILTER_MASK | \
1041 R200_MAG_FILTER_MASK | \
1042 R200_MAX_ANISO_MASK | \
1043 R200_YUV_TO_RGB | \
1044 R200_YUV_TEMPERATURE_MASK | \
1045 R200_CLAMP_S_MASK | \
1046 R200_CLAMP_T_MASK | \
1047 R200_BORDER_MODE_D3D )
1048
1049 #define TEXOBJ_TXFORMAT_MASK (R200_TXFORMAT_WIDTH_MASK | \
1050 R200_TXFORMAT_HEIGHT_MASK | \
1051 R200_TXFORMAT_FORMAT_MASK | \
1052 R200_TXFORMAT_F5_WIDTH_MASK | \
1053 R200_TXFORMAT_F5_HEIGHT_MASK | \
1054 R200_TXFORMAT_ALPHA_IN_MAP | \
1055 R200_TXFORMAT_CUBIC_MAP_ENABLE | \
1056 R200_TXFORMAT_NON_POWER2)
1057
1058 #define TEXOBJ_TXFORMAT_X_MASK (R200_DEPTH_LOG2_MASK | \
1059 R200_TEXCOORD_MASK | \
1060 R200_CLAMP_Q_MASK | \
1061 R200_VOLUME_FILTER_MASK)
1062
1063
1064 static void disable_tex_obj_state( r200ContextPtr rmesa,
1065 int unit )
1066 {
1067
1068 R200_STATECHANGE( rmesa, vtx );
1069 rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] &= ~(7 << (unit * 3));
1070
1071 if (rmesa->radeon.TclFallback & (R200_TCL_FALLBACK_TEXGEN_0<<unit)) {
1072 TCL_FALLBACK( rmesa->radeon.glCtx, (R200_TCL_FALLBACK_TEXGEN_0<<unit), GL_FALSE);
1073 }
1074
1075 /* Actually want to keep all units less than max active texture
1076 * enabled, right? Fix this for >2 texunits.
1077 */
1078
1079 {
1080 GLuint tmp = rmesa->TexGenEnabled;
1081
1082 rmesa->TexGenEnabled &= ~(R200_TEXGEN_TEXMAT_0_ENABLE<<unit);
1083 rmesa->TexGenEnabled &= ~(R200_TEXMAT_0_ENABLE<<unit);
1084 rmesa->TexGenNeedNormals[unit] = GL_FALSE;
1085 rmesa->TexGenCompSel &= ~(R200_OUTPUT_TEX_0 << unit);
1086
1087 if (tmp != rmesa->TexGenEnabled) {
1088 rmesa->recheck_texgen[unit] = GL_TRUE;
1089 rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX;
1090 }
1091 }
1092 }
1093 static void import_tex_obj_state( r200ContextPtr rmesa,
1094 int unit,
1095 radeonTexObjPtr texobj )
1096 {
1097 /* do not use RADEON_DB_STATE to avoid stale texture caches */
1098 GLuint *cmd = &rmesa->hw.tex[unit].cmd[TEX_CMD_0];
1099
1100 R200_STATECHANGE( rmesa, tex[unit] );
1101
1102 cmd[TEX_PP_TXFILTER] &= ~TEXOBJ_TXFILTER_MASK;
1103 cmd[TEX_PP_TXFILTER] |= texobj->pp_txfilter & TEXOBJ_TXFILTER_MASK;
1104 cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK;
1105 cmd[TEX_PP_TXFORMAT] |= texobj->pp_txformat & TEXOBJ_TXFORMAT_MASK;
1106 cmd[TEX_PP_TXFORMAT_X] &= ~TEXOBJ_TXFORMAT_X_MASK;
1107 cmd[TEX_PP_TXFORMAT_X] |= texobj->pp_txformat_x & TEXOBJ_TXFORMAT_X_MASK;
1108 cmd[TEX_PP_TXSIZE] = texobj->pp_txsize; /* NPOT only! */
1109 cmd[TEX_PP_TXPITCH] = texobj->pp_txpitch; /* NPOT only! */
1110 cmd[TEX_PP_BORDER_COLOR] = texobj->pp_border_color;
1111
1112 if (texobj->base.Target == GL_TEXTURE_CUBE_MAP) {
1113 GLuint *cube_cmd = &rmesa->hw.cube[unit].cmd[CUBE_CMD_0];
1114
1115 R200_STATECHANGE( rmesa, cube[unit] );
1116 cube_cmd[CUBE_PP_CUBIC_FACES] = texobj->pp_cubic_faces;
1117 if (rmesa->radeon.radeonScreen->drmSupportsFragShader) {
1118 /* that value is submitted twice. could change cube atom
1119 to not include that command when new drm is used */
1120 cmd[TEX_PP_CUBIC_FACES] = texobj->pp_cubic_faces;
1121 }
1122 }
1123
1124 }
1125
1126 static void set_texgen_matrix( r200ContextPtr rmesa,
1127 GLuint unit,
1128 const GLfloat *s_plane,
1129 const GLfloat *t_plane,
1130 const GLfloat *r_plane,
1131 const GLfloat *q_plane )
1132 {
1133 GLfloat m[16];
1134
1135 m[0] = s_plane[0];
1136 m[4] = s_plane[1];
1137 m[8] = s_plane[2];
1138 m[12] = s_plane[3];
1139
1140 m[1] = t_plane[0];
1141 m[5] = t_plane[1];
1142 m[9] = t_plane[2];
1143 m[13] = t_plane[3];
1144
1145 m[2] = r_plane[0];
1146 m[6] = r_plane[1];
1147 m[10] = r_plane[2];
1148 m[14] = r_plane[3];
1149
1150 m[3] = q_plane[0];
1151 m[7] = q_plane[1];
1152 m[11] = q_plane[2];
1153 m[15] = q_plane[3];
1154
1155 _math_matrix_loadf( &(rmesa->TexGenMatrix[unit]), m);
1156 _math_matrix_analyse( &(rmesa->TexGenMatrix[unit]) );
1157 rmesa->TexGenEnabled |= R200_TEXMAT_0_ENABLE<<unit;
1158 }
1159
1160
1161 static GLuint r200_need_dis_texgen(const GLbitfield texGenEnabled,
1162 const GLfloat *planeS,
1163 const GLfloat *planeT,
1164 const GLfloat *planeR,
1165 const GLfloat *planeQ)
1166 {
1167 GLuint needtgenable = 0;
1168
1169 if (!(texGenEnabled & S_BIT)) {
1170 if (((texGenEnabled & T_BIT) && planeT[0] != 0.0) ||
1171 ((texGenEnabled & R_BIT) && planeR[0] != 0.0) ||
1172 ((texGenEnabled & Q_BIT) && planeQ[0] != 0.0)) {
1173 needtgenable |= S_BIT;
1174 }
1175 }
1176 if (!(texGenEnabled & T_BIT)) {
1177 if (((texGenEnabled & S_BIT) && planeS[1] != 0.0) ||
1178 ((texGenEnabled & R_BIT) && planeR[1] != 0.0) ||
1179 ((texGenEnabled & Q_BIT) && planeQ[1] != 0.0)) {
1180 needtgenable |= T_BIT;
1181 }
1182 }
1183 if (!(texGenEnabled & R_BIT)) {
1184 if (((texGenEnabled & S_BIT) && planeS[2] != 0.0) ||
1185 ((texGenEnabled & T_BIT) && planeT[2] != 0.0) ||
1186 ((texGenEnabled & Q_BIT) && planeQ[2] != 0.0)) {
1187 needtgenable |= R_BIT;
1188 }
1189 }
1190 if (!(texGenEnabled & Q_BIT)) {
1191 if (((texGenEnabled & S_BIT) && planeS[3] != 0.0) ||
1192 ((texGenEnabled & T_BIT) && planeT[3] != 0.0) ||
1193 ((texGenEnabled & R_BIT) && planeR[3] != 0.0)) {
1194 needtgenable |= Q_BIT;
1195 }
1196 }
1197
1198 return needtgenable;
1199 }
1200
1201
1202 /*
1203 * Returns GL_FALSE if fallback required.
1204 */
1205 static GLboolean r200_validate_texgen( GLcontext *ctx, GLuint unit )
1206 {
1207 r200ContextPtr rmesa = R200_CONTEXT(ctx);
1208 const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
1209 GLuint inputshift = R200_TEXGEN_0_INPUT_SHIFT + unit*4;
1210 GLuint tgi, tgcm;
1211 GLuint mode = 0;
1212 GLboolean mixed_fallback = GL_FALSE;
1213 static const GLfloat I[16] = {
1214 1, 0, 0, 0,
1215 0, 1, 0, 0,
1216 0, 0, 1, 0,
1217 0, 0, 0, 1 };
1218 static const GLfloat reflect[16] = {
1219 -1, 0, 0, 0,
1220 0, -1, 0, 0,
1221 0, 0, -1, 0,
1222 0, 0, 0, 1 };
1223
1224 rmesa->TexGenCompSel &= ~(R200_OUTPUT_TEX_0 << unit);
1225 rmesa->TexGenEnabled &= ~(R200_TEXGEN_TEXMAT_0_ENABLE<<unit);
1226 rmesa->TexGenEnabled &= ~(R200_TEXMAT_0_ENABLE<<unit);
1227 rmesa->TexGenNeedNormals[unit] = GL_FALSE;
1228 tgi = rmesa->hw.tcg.cmd[TCG_TEX_PROC_CTL_1] & ~(R200_TEXGEN_INPUT_MASK <<
1229 inputshift);
1230 tgcm = rmesa->hw.tcg.cmd[TCG_TEX_PROC_CTL_2] & ~(R200_TEXGEN_COMP_MASK <<
1231 (unit * 4));
1232
1233 if (0)
1234 fprintf(stderr, "%s unit %d\n", __FUNCTION__, unit);
1235
1236 if (texUnit->TexGenEnabled & S_BIT) {
1237 mode = texUnit->GenS.Mode;
1238 } else {
1239 tgcm |= R200_TEXGEN_COMP_S << (unit * 4);
1240 }
1241
1242 if (texUnit->TexGenEnabled & T_BIT) {
1243 if (texUnit->GenT.Mode != mode)
1244 mixed_fallback = GL_TRUE;
1245 } else {
1246 tgcm |= R200_TEXGEN_COMP_T << (unit * 4);
1247 }
1248 if (texUnit->TexGenEnabled & R_BIT) {
1249 if (texUnit->GenR.Mode != mode)
1250 mixed_fallback = GL_TRUE;
1251 } else {
1252 tgcm |= R200_TEXGEN_COMP_R << (unit * 4);
1253 }
1254
1255 if (texUnit->TexGenEnabled & Q_BIT) {
1256 if (texUnit->GenQ.Mode != mode)
1257 mixed_fallback = GL_TRUE;
1258 } else {
1259 tgcm |= R200_TEXGEN_COMP_Q << (unit * 4);
1260 }
1261
1262 if (mixed_fallback) {
1263 if (R200_DEBUG & DEBUG_FALLBACKS)
1264 fprintf(stderr, "fallback mixed texgen, 0x%x (0x%x 0x%x 0x%x 0x%x)\n",
1265 texUnit->TexGenEnabled, texUnit->GenS.Mode, texUnit->GenT.Mode,
1266 texUnit->GenR.Mode, texUnit->GenQ.Mode);
1267 return GL_FALSE;
1268 }
1269
1270 /* we CANNOT do mixed mode if the texgen mode requires a plane where the input
1271 is not enabled for texgen, since the planes are concatenated into texmat,
1272 and thus the input will come from texcoord rather than tex gen equation!
1273 Either fallback or just hope that those texcoords aren't really needed...
1274 Assuming the former will cause lots of unnecessary fallbacks, the latter will
1275 generate bogus results sometimes - it's pretty much impossible to really know
1276 when a fallback is needed, depends on texmat and what sort of texture is bound
1277 etc, - for now fallback if we're missing either S or T bits, there's a high
1278 probability we need the texcoords in that case.
1279 That's a lot of work for some obscure texgen mixed mode fixup - why oh why
1280 doesn't the chip just directly accept the plane parameters :-(. */
1281 switch (mode) {
1282 case GL_OBJECT_LINEAR: {
1283 GLuint needtgenable = r200_need_dis_texgen( texUnit->TexGenEnabled,
1284 texUnit->GenS.ObjectPlane,
1285 texUnit->GenT.ObjectPlane,
1286 texUnit->GenR.ObjectPlane,
1287 texUnit->GenQ.ObjectPlane );
1288 if (needtgenable & (S_BIT | T_BIT)) {
1289 if (R200_DEBUG & DEBUG_FALLBACKS)
1290 fprintf(stderr, "fallback mixed texgen / obj plane, 0x%x\n",
1291 texUnit->TexGenEnabled);
1292 return GL_FALSE;
1293 }
1294 if (needtgenable & (R_BIT)) {
1295 tgcm &= ~(R200_TEXGEN_COMP_R << (unit * 4));
1296 }
1297 if (needtgenable & (Q_BIT)) {
1298 tgcm &= ~(R200_TEXGEN_COMP_Q << (unit * 4));
1299 }
1300
1301 tgi |= R200_TEXGEN_INPUT_OBJ << inputshift;
1302 set_texgen_matrix( rmesa, unit,
1303 (texUnit->TexGenEnabled & S_BIT) ? texUnit->GenS.ObjectPlane : I,
1304 (texUnit->TexGenEnabled & T_BIT) ? texUnit->GenT.ObjectPlane : I + 4,
1305 (texUnit->TexGenEnabled & R_BIT) ? texUnit->GenR.ObjectPlane : I + 8,
1306 (texUnit->TexGenEnabled & Q_BIT) ? texUnit->GenQ.ObjectPlane : I + 12);
1307 }
1308 break;
1309
1310 case GL_EYE_LINEAR: {
1311 GLuint needtgenable = r200_need_dis_texgen( texUnit->TexGenEnabled,
1312 texUnit->GenS.EyePlane,
1313 texUnit->GenT.EyePlane,
1314 texUnit->GenR.EyePlane,
1315 texUnit->GenQ.EyePlane );
1316 if (needtgenable & (S_BIT | T_BIT)) {
1317 if (R200_DEBUG & DEBUG_FALLBACKS)
1318 fprintf(stderr, "fallback mixed texgen / eye plane, 0x%x\n",
1319 texUnit->TexGenEnabled);
1320 return GL_FALSE;
1321 }
1322 if (needtgenable & (R_BIT)) {
1323 tgcm &= ~(R200_TEXGEN_COMP_R << (unit * 4));
1324 }
1325 if (needtgenable & (Q_BIT)) {
1326 tgcm &= ~(R200_TEXGEN_COMP_Q << (unit * 4));
1327 }
1328 tgi |= R200_TEXGEN_INPUT_EYE << inputshift;
1329 set_texgen_matrix( rmesa, unit,
1330 (texUnit->TexGenEnabled & S_BIT) ? texUnit->GenS.EyePlane : I,
1331 (texUnit->TexGenEnabled & T_BIT) ? texUnit->GenT.EyePlane : I + 4,
1332 (texUnit->TexGenEnabled & R_BIT) ? texUnit->GenR.EyePlane : I + 8,
1333 (texUnit->TexGenEnabled & Q_BIT) ? texUnit->GenQ.EyePlane : I + 12);
1334 }
1335 break;
1336
1337 case GL_REFLECTION_MAP_NV:
1338 rmesa->TexGenNeedNormals[unit] = GL_TRUE;
1339 tgi |= R200_TEXGEN_INPUT_EYE_REFLECT << inputshift;
1340 /* pretty weird, must only negate when lighting is enabled? */
1341 if (ctx->Light.Enabled)
1342 set_texgen_matrix( rmesa, unit,
1343 (texUnit->TexGenEnabled & S_BIT) ? reflect : I,
1344 (texUnit->TexGenEnabled & T_BIT) ? reflect + 4 : I + 4,
1345 (texUnit->TexGenEnabled & R_BIT) ? reflect + 8 : I + 8,
1346 I + 12);
1347 break;
1348
1349 case GL_NORMAL_MAP_NV:
1350 rmesa->TexGenNeedNormals[unit] = GL_TRUE;
1351 tgi |= R200_TEXGEN_INPUT_EYE_NORMAL<<inputshift;
1352 break;
1353
1354 case GL_SPHERE_MAP:
1355 rmesa->TexGenNeedNormals[unit] = GL_TRUE;
1356 tgi |= R200_TEXGEN_INPUT_SPHERE<<inputshift;
1357 break;
1358
1359 case 0:
1360 /* All texgen units were disabled, so just pass coords through. */
1361 tgi |= unit << inputshift;
1362 break;
1363
1364 default:
1365 /* Unsupported mode, fallback:
1366 */
1367 if (R200_DEBUG & DEBUG_FALLBACKS)
1368 fprintf(stderr, "fallback unsupported texgen, %d\n",
1369 texUnit->GenS.Mode);
1370 return GL_FALSE;
1371 }
1372
1373 rmesa->TexGenEnabled |= R200_TEXGEN_TEXMAT_0_ENABLE << unit;
1374 rmesa->TexGenCompSel |= R200_OUTPUT_TEX_0 << unit;
1375
1376 if (tgi != rmesa->hw.tcg.cmd[TCG_TEX_PROC_CTL_1] ||
1377 tgcm != rmesa->hw.tcg.cmd[TCG_TEX_PROC_CTL_2])
1378 {
1379 R200_STATECHANGE(rmesa, tcg);
1380 rmesa->hw.tcg.cmd[TCG_TEX_PROC_CTL_1] = tgi;
1381 rmesa->hw.tcg.cmd[TCG_TEX_PROC_CTL_2] = tgcm;
1382 }
1383
1384 return GL_TRUE;
1385 }
1386
1387 void set_re_cntl_d3d( GLcontext *ctx, int unit, GLboolean use_d3d )
1388 {
1389 r200ContextPtr rmesa = R200_CONTEXT(ctx);
1390
1391 GLuint re_cntl;
1392
1393 re_cntl = rmesa->hw.set.cmd[SET_RE_CNTL] & ~(R200_VTX_STQ0_D3D << (2 * unit));
1394 if (use_d3d)
1395 re_cntl |= R200_VTX_STQ0_D3D << (2 * unit);
1396
1397 if ( re_cntl != rmesa->hw.set.cmd[SET_RE_CNTL] ) {
1398 R200_STATECHANGE( rmesa, set );
1399 rmesa->hw.set.cmd[SET_RE_CNTL] = re_cntl;
1400 }
1401 }
1402
1403 /**
1404 * Compute the cached hardware register values for the given texture object.
1405 *
1406 * \param rmesa Context pointer
1407 * \param t the r300 texture object
1408 */
1409 static void setup_hardware_state(r200ContextPtr rmesa, radeonTexObj *t)
1410 {
1411 const struct gl_texture_image *firstImage =
1412 t->base.Image[0][t->mt->firstLevel];
1413 GLint log2Width, log2Height, log2Depth, texelBytes;
1414
1415 if ( t->bo ) {
1416 return;
1417 }
1418
1419 log2Width = firstImage->WidthLog2;
1420 log2Height = firstImage->HeightLog2;
1421 log2Depth = firstImage->DepthLog2;
1422 texelBytes = firstImage->TexFormat->TexelBytes;
1423
1424
1425 if (!t->image_override) {
1426 if (VALID_FORMAT(firstImage->TexFormat->MesaFormat)) {
1427 const struct tx_table *table = _mesa_little_endian() ? tx_table_le :
1428 tx_table_be;
1429
1430 t->pp_txformat &= ~(R200_TXFORMAT_FORMAT_MASK |
1431 R200_TXFORMAT_ALPHA_IN_MAP);
1432 t->pp_txfilter &= ~R200_YUV_TO_RGB;
1433
1434 t->pp_txformat |= table[ firstImage->TexFormat->MesaFormat ].format;
1435 t->pp_txfilter |= table[ firstImage->TexFormat->MesaFormat ].filter;
1436 } else {
1437 _mesa_problem(NULL, "unexpected texture format in %s",
1438 __FUNCTION__);
1439 return;
1440 }
1441 }
1442
1443 t->pp_txfilter &= ~R200_MAX_MIP_LEVEL_MASK;
1444 t->pp_txfilter |= (t->mt->lastLevel - t->mt->firstLevel) << R200_MAX_MIP_LEVEL_SHIFT;
1445
1446 t->pp_txformat &= ~(R200_TXFORMAT_WIDTH_MASK |
1447 R200_TXFORMAT_HEIGHT_MASK |
1448 R200_TXFORMAT_CUBIC_MAP_ENABLE |
1449 R200_TXFORMAT_F5_WIDTH_MASK |
1450 R200_TXFORMAT_F5_HEIGHT_MASK);
1451 t->pp_txformat |= ((log2Width << R200_TXFORMAT_WIDTH_SHIFT) |
1452 (log2Height << R200_TXFORMAT_HEIGHT_SHIFT));
1453
1454 t->tile_bits = 0;
1455
1456 t->pp_txformat_x &= ~(R200_DEPTH_LOG2_MASK | R200_TEXCOORD_MASK);
1457 if (t->base.Target == GL_TEXTURE_3D) {
1458 t->pp_txformat_x |= (log2Depth << R200_DEPTH_LOG2_SHIFT);
1459 t->pp_txformat_x |= R200_TEXCOORD_VOLUME;
1460
1461 }
1462 else if (t->base.Target == GL_TEXTURE_CUBE_MAP) {
1463 ASSERT(log2Width == log2Height);
1464 t->pp_txformat |= ((log2Width << R200_TXFORMAT_F5_WIDTH_SHIFT) |
1465 (log2Height << R200_TXFORMAT_F5_HEIGHT_SHIFT) |
1466 /* don't think we need this bit, if it exists at all - fglrx does not set it */
1467 (R200_TXFORMAT_CUBIC_MAP_ENABLE));
1468 t->pp_txformat_x |= R200_TEXCOORD_CUBIC_ENV;
1469 t->pp_cubic_faces = ((log2Width << R200_FACE_WIDTH_1_SHIFT) |
1470 (log2Height << R200_FACE_HEIGHT_1_SHIFT) |
1471 (log2Width << R200_FACE_WIDTH_2_SHIFT) |
1472 (log2Height << R200_FACE_HEIGHT_2_SHIFT) |
1473 (log2Width << R200_FACE_WIDTH_3_SHIFT) |
1474 (log2Height << R200_FACE_HEIGHT_3_SHIFT) |
1475 (log2Width << R200_FACE_WIDTH_4_SHIFT) |
1476 (log2Height << R200_FACE_HEIGHT_4_SHIFT));
1477 }
1478 else {
1479 /* If we don't in fact send enough texture coordinates, q will be 1,
1480 * making TEXCOORD_PROJ act like TEXCOORD_NONPROJ (Right?)
1481 */
1482 t->pp_txformat_x |= R200_TEXCOORD_PROJ;
1483 }
1484
1485 t->pp_txsize = (((firstImage->Width - 1) << R200_PP_TX_WIDTHMASK_SHIFT)
1486 | ((firstImage->Height - 1) << R200_PP_TX_HEIGHTMASK_SHIFT));
1487
1488 if ( !t->image_override ) {
1489 if (firstImage->IsCompressed)
1490 t->pp_txpitch = (firstImage->Width + 63) & ~(63);
1491 else
1492 t->pp_txpitch = ((firstImage->Width * texelBytes) + 63) & ~(63);
1493 t->pp_txpitch -= 32;
1494 }
1495
1496 if (t->base.Target == GL_TEXTURE_RECTANGLE_NV) {
1497 t->pp_txformat |= R200_TXFORMAT_NON_POWER2;
1498 }
1499
1500 }
1501
1502 static GLboolean r200_validate_texture(GLcontext *ctx, struct gl_texture_object *texObj, int unit)
1503 {
1504 r200ContextPtr rmesa = R200_CONTEXT(ctx);
1505 radeonTexObj *t = radeon_tex_obj(texObj);
1506
1507 if (!radeon_validate_texture_miptree(ctx, texObj))
1508 return GL_FALSE;
1509
1510 r200_validate_texgen(ctx, unit);
1511 /* Configure the hardware registers (more precisely, the cached version
1512 * of the hardware registers). */
1513 setup_hardware_state(rmesa, t);
1514
1515 if (texObj->Target == GL_TEXTURE_RECTANGLE_NV ||
1516 texObj->Target == GL_TEXTURE_2D ||
1517 texObj->Target == GL_TEXTURE_1D)
1518 set_re_cntl_d3d( ctx, unit, GL_FALSE );
1519 else
1520 set_re_cntl_d3d( ctx, unit, GL_TRUE );
1521 R200_STATECHANGE( rmesa, ctx );
1522 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= R200_TEX_0_ENABLE << unit;
1523
1524 R200_STATECHANGE( rmesa, vtx );
1525 rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] &= ~(7 << (unit * 3));
1526 rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] |= 4 << (unit * 3);
1527
1528 rmesa->recheck_texgen[unit] = GL_TRUE;
1529 import_tex_obj_state( rmesa, unit, t );
1530
1531 if (rmesa->recheck_texgen[unit]) {
1532 GLboolean fallback = !r200_validate_texgen( ctx, unit );
1533 TCL_FALLBACK( ctx, (R200_TCL_FALLBACK_TEXGEN_0<<unit), fallback);
1534 rmesa->recheck_texgen[unit] = 0;
1535 rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX;
1536 }
1537
1538 t->validated = GL_TRUE;
1539
1540 FALLBACK( rmesa, RADEON_FALLBACK_BORDER_MODE, t->border_fallback );
1541
1542 return !t->border_fallback;
1543 }
1544
1545 static GLboolean r200UpdateTextureUnit(GLcontext *ctx, int unit)
1546 {
1547 r200ContextPtr rmesa = R200_CONTEXT(ctx);
1548 GLuint unitneeded = rmesa->state.texture.unit[unit].unitneeded;
1549
1550 if (!unitneeded) {
1551 /* disable the unit */
1552 disable_tex_obj_state(rmesa, unit);
1553 return GL_TRUE;
1554 }
1555
1556 if (!r200_validate_texture(ctx, ctx->Texture.Unit[unit]._Current, unit)) {
1557 _mesa_warning(ctx,
1558 "failed to validate texture for unit %d.\n",
1559 unit);
1560 rmesa->state.texture.unit[unit].texobj = NULL;
1561 return GL_FALSE;
1562 }
1563
1564 rmesa->state.texture.unit[unit].texobj = radeon_tex_obj(ctx->Texture.Unit[unit]._Current);
1565 return GL_TRUE;
1566 }
1567
1568
1569 void r200UpdateTextureState( GLcontext *ctx )
1570 {
1571 r200ContextPtr rmesa = R200_CONTEXT(ctx);
1572 GLboolean ok;
1573 GLuint dbg;
1574
1575 /* NOTE: must not manipulate rmesa->state.texture.unit[].unitneeded or
1576 rmesa->state.envneeded before a R200_STATECHANGE (or R200_NEWPRIM) since
1577 we use these to determine if we want to emit the corresponding state
1578 atoms. */
1579 R200_NEWPRIM( rmesa );
1580
1581 if (ctx->ATIFragmentShader._Enabled) {
1582 GLuint i;
1583 for (i = 0; i < R200_MAX_TEXTURE_UNITS; i++) {
1584 rmesa->state.texture.unit[i].unitneeded = ctx->Texture.Unit[i]._ReallyEnabled;
1585 }
1586 ok = GL_TRUE;
1587 }
1588 else {
1589 ok = r200UpdateAllTexEnv( ctx );
1590 }
1591 if (ok) {
1592 ok = (r200UpdateTextureUnit( ctx, 0 ) &&
1593 r200UpdateTextureUnit( ctx, 1 ) &&
1594 r200UpdateTextureUnit( ctx, 2 ) &&
1595 r200UpdateTextureUnit( ctx, 3 ) &&
1596 r200UpdateTextureUnit( ctx, 4 ) &&
1597 r200UpdateTextureUnit( ctx, 5 ));
1598 }
1599
1600 if (ok && ctx->ATIFragmentShader._Enabled) {
1601 r200UpdateFragmentShader(ctx);
1602 }
1603
1604 FALLBACK( rmesa, R200_FALLBACK_TEXTURE, !ok );
1605
1606 if (rmesa->radeon.TclFallback)
1607 r200ChooseVertexState( ctx );
1608
1609
1610 if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R200) {
1611
1612 /*
1613 * T0 hang workaround -------------
1614 * not needed for r200 derivatives
1615 */
1616 if ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & R200_TEX_ENABLE_MASK) == R200_TEX_0_ENABLE &&
1617 (rmesa->hw.tex[0].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK) > R200_MIN_FILTER_LINEAR) {
1618
1619 R200_STATECHANGE(rmesa, ctx);
1620 R200_STATECHANGE(rmesa, tex[1]);
1621 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= R200_TEX_1_ENABLE;
1622 if (!(rmesa->hw.cst.cmd[CST_PP_CNTL_X] & R200_PPX_TEX_1_ENABLE))
1623 rmesa->hw.tex[1].cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK;
1624 rmesa->hw.tex[1].cmd[TEX_PP_TXFORMAT] |= R200_TXFORMAT_LOOKUP_DISABLE;
1625 }
1626 else if (!ctx->ATIFragmentShader._Enabled) {
1627 if ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & R200_TEX_1_ENABLE) &&
1628 (rmesa->hw.tex[1].cmd[TEX_PP_TXFORMAT] & R200_TXFORMAT_LOOKUP_DISABLE)) {
1629 R200_STATECHANGE(rmesa, tex[1]);
1630 rmesa->hw.tex[1].cmd[TEX_PP_TXFORMAT] &= ~R200_TXFORMAT_LOOKUP_DISABLE;
1631 }
1632 }
1633 /* do the same workaround for the first pass of a fragment shader.
1634 * completely unknown if necessary / sufficient.
1635 */
1636 if ((rmesa->hw.cst.cmd[CST_PP_CNTL_X] & R200_PPX_TEX_ENABLE_MASK) == R200_PPX_TEX_0_ENABLE &&
1637 (rmesa->hw.tex[0].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK) > R200_MIN_FILTER_LINEAR) {
1638
1639 R200_STATECHANGE(rmesa, cst);
1640 R200_STATECHANGE(rmesa, tex[1]);
1641 rmesa->hw.cst.cmd[CST_PP_CNTL_X] |= R200_PPX_TEX_1_ENABLE;
1642 if (!(rmesa->hw.ctx.cmd[CTX_PP_CNTL] & R200_TEX_1_ENABLE))
1643 rmesa->hw.tex[1].cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK;
1644 rmesa->hw.tex[1].cmd[TEX_PP_TXMULTI_CTL] |= R200_PASS1_TXFORMAT_LOOKUP_DISABLE;
1645 }
1646
1647 /* maybe needs to be done pairwise due to 2 parallel (physical) tex units ?
1648 looks like that's not the case, if 8500/9100 owners don't complain remove this...
1649 for ( i = 0; i < ctx->Const.MaxTextureUnits; i += 2) {
1650 if (((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & ((R200_TEX_0_ENABLE |
1651 R200_TEX_1_ENABLE ) << i)) == (R200_TEX_0_ENABLE << i)) &&
1652 ((rmesa->hw.tex[i].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK) >
1653 R200_MIN_FILTER_LINEAR)) {
1654 R200_STATECHANGE(rmesa, ctx);
1655 R200_STATECHANGE(rmesa, tex[i+1]);
1656 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= (R200_TEX_1_ENABLE << i);
1657 rmesa->hw.tex[i+1].cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK;
1658 rmesa->hw.tex[i+1].cmd[TEX_PP_TXFORMAT] |= 0x08000000;
1659 }
1660 else {
1661 if ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (R200_TEX_1_ENABLE << i)) &&
1662 (rmesa->hw.tex[i+1].cmd[TEX_PP_TXFORMAT] & 0x08000000)) {
1663 R200_STATECHANGE(rmesa, tex[i+1]);
1664 rmesa->hw.tex[i+1].cmd[TEX_PP_TXFORMAT] &= ~0x08000000;
1665 }
1666 }
1667 } */
1668
1669 /*
1670 * Texture cache LRU hang workaround -------------
1671 * not needed for r200 derivatives
1672 * hopefully this covers first pass of a shader as well
1673 */
1674
1675 /* While the cases below attempt to only enable the workaround in the
1676 * specific cases necessary, they were insufficient. See bugzilla #1519,
1677 * #729, #814. Tests with quake3 showed no impact on performance.
1678 */
1679 dbg = 0x6;
1680
1681 /*
1682 if (((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (R200_TEX_0_ENABLE )) &&
1683 ((((rmesa->hw.tex[0].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1684 0x04) == 0)) ||
1685 ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & R200_TEX_2_ENABLE) &&
1686 ((((rmesa->hw.tex[2].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1687 0x04) == 0)) ||
1688 ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & R200_TEX_4_ENABLE) &&
1689 ((((rmesa->hw.tex[4].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1690 0x04) == 0)))
1691 {
1692 dbg |= 0x02;
1693 }
1694
1695 if (((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (R200_TEX_1_ENABLE )) &&
1696 ((((rmesa->hw.tex[1].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1697 0x04) == 0)) ||
1698 ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & R200_TEX_3_ENABLE) &&
1699 ((((rmesa->hw.tex[3].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1700 0x04) == 0)) ||
1701 ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & R200_TEX_5_ENABLE) &&
1702 ((((rmesa->hw.tex[5].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1703 0x04) == 0)))
1704 {
1705 dbg |= 0x04;
1706 }*/
1707
1708 if (dbg != rmesa->hw.tam.cmd[TAM_DEBUG3]) {
1709 R200_STATECHANGE( rmesa, tam );
1710 rmesa->hw.tam.cmd[TAM_DEBUG3] = dbg;
1711 if (0) printf("TEXCACHE LRU HANG WORKAROUND %x\n", dbg);
1712 }
1713 }
1714 }