2 Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved.
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.
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:
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.
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.
28 **************************************************************************/
32 * Keith Whitwell <keith@tungstengraphics.com>
35 #include "main/glheader.h"
36 #include "main/imports.h"
37 #include "main/context.h"
38 #include "main/macros.h"
39 #include "main/teximage.h"
40 #include "main/texobj.h"
41 #include "main/enums.h"
43 #include "radeon_common.h"
44 #include "radeon_mipmap_tree.h"
45 #include "r200_context.h"
46 #include "r200_state.h"
47 #include "r200_ioctl.h"
48 #include "r200_swtcl.h"
53 #define R200_TXFORMAT_A8 R200_TXFORMAT_I8
54 #define R200_TXFORMAT_L8 R200_TXFORMAT_I8
55 #define R200_TXFORMAT_AL88 R200_TXFORMAT_AI88
56 #define R200_TXFORMAT_YCBCR R200_TXFORMAT_YVYU422
57 #define R200_TXFORMAT_YCBCR_REV R200_TXFORMAT_VYUY422
58 #define R200_TXFORMAT_RGB_DXT1 R200_TXFORMAT_DXT1
59 #define R200_TXFORMAT_RGBA_DXT1 R200_TXFORMAT_DXT1
60 #define R200_TXFORMAT_RGBA_DXT3 R200_TXFORMAT_DXT23
61 #define R200_TXFORMAT_RGBA_DXT5 R200_TXFORMAT_DXT45
64 [ MESA_FORMAT_ ## f ] = { R200_TXFORMAT_ ## f, 0 }
65 #define _COLOR_REV(f) \
66 [ MESA_FORMAT_ ## f ## _REV ] = { R200_TXFORMAT_ ## f, 0 }
68 [ MESA_FORMAT_ ## f ] = { R200_TXFORMAT_ ## f | R200_TXFORMAT_ALPHA_IN_MAP, 0 }
69 #define _ALPHA_REV(f) \
70 [ MESA_FORMAT_ ## f ## _REV ] = { R200_TXFORMAT_ ## f | R200_TXFORMAT_ALPHA_IN_MAP, 0 }
72 [ MESA_FORMAT_ ## f ] = { R200_TXFORMAT_ ## f, R200_YUV_TO_RGB }
74 [ MESA_FORMAT_ ## f ] = { 0xffffffff, 0 }
75 #define VALID_FORMAT(f) ( ((f) <= MESA_FORMAT_RGBA_DXT5) \
76 && (tx_table_be[f].format != 0xffffffff) )
79 GLuint format
, filter
;
82 static const struct tx_table tx_table_be
[] =
84 [ MESA_FORMAT_RGBA8888
] = { R200_TXFORMAT_ABGR8888
| R200_TXFORMAT_ALPHA_IN_MAP
, 0 },
111 static const struct tx_table tx_table_le
[] =
114 [ MESA_FORMAT_RGBA8888_REV
] = { R200_TXFORMAT_ABGR8888
| R200_TXFORMAT_ALPHA_IN_MAP
, 0 },
116 _ALPHA_REV(ARGB8888
),
117 [ MESA_FORMAT_RGB888
] = { R200_TXFORMAT_ARGB8888
, 0 },
121 _ALPHA_REV(ARGB4444
),
123 _ALPHA_REV(ARGB1555
),
144 /* ================================================================
145 * Texture combine functions
148 /* GL_ARB_texture_env_combine support
151 /* The color tables have combine functions for GL_SRC_COLOR,
152 * GL_ONE_MINUS_SRC_COLOR, GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
154 static GLuint r200_register_color
[][R200_MAX_TEXTURE_UNITS
] =
157 R200_TXC_ARG_A_R0_COLOR
,
158 R200_TXC_ARG_A_R1_COLOR
,
159 R200_TXC_ARG_A_R2_COLOR
,
160 R200_TXC_ARG_A_R3_COLOR
,
161 R200_TXC_ARG_A_R4_COLOR
,
162 R200_TXC_ARG_A_R5_COLOR
165 R200_TXC_ARG_A_R0_COLOR
| R200_TXC_COMP_ARG_A
,
166 R200_TXC_ARG_A_R1_COLOR
| R200_TXC_COMP_ARG_A
,
167 R200_TXC_ARG_A_R2_COLOR
| R200_TXC_COMP_ARG_A
,
168 R200_TXC_ARG_A_R3_COLOR
| R200_TXC_COMP_ARG_A
,
169 R200_TXC_ARG_A_R4_COLOR
| R200_TXC_COMP_ARG_A
,
170 R200_TXC_ARG_A_R5_COLOR
| R200_TXC_COMP_ARG_A
173 R200_TXC_ARG_A_R0_ALPHA
,
174 R200_TXC_ARG_A_R1_ALPHA
,
175 R200_TXC_ARG_A_R2_ALPHA
,
176 R200_TXC_ARG_A_R3_ALPHA
,
177 R200_TXC_ARG_A_R4_ALPHA
,
178 R200_TXC_ARG_A_R5_ALPHA
181 R200_TXC_ARG_A_R0_ALPHA
| R200_TXC_COMP_ARG_A
,
182 R200_TXC_ARG_A_R1_ALPHA
| R200_TXC_COMP_ARG_A
,
183 R200_TXC_ARG_A_R2_ALPHA
| R200_TXC_COMP_ARG_A
,
184 R200_TXC_ARG_A_R3_ALPHA
| R200_TXC_COMP_ARG_A
,
185 R200_TXC_ARG_A_R4_ALPHA
| R200_TXC_COMP_ARG_A
,
186 R200_TXC_ARG_A_R5_ALPHA
| R200_TXC_COMP_ARG_A
190 static GLuint r200_tfactor_color
[] =
192 R200_TXC_ARG_A_TFACTOR_COLOR
,
193 R200_TXC_ARG_A_TFACTOR_COLOR
| R200_TXC_COMP_ARG_A
,
194 R200_TXC_ARG_A_TFACTOR_ALPHA
,
195 R200_TXC_ARG_A_TFACTOR_ALPHA
| R200_TXC_COMP_ARG_A
198 static GLuint r200_tfactor1_color
[] =
200 R200_TXC_ARG_A_TFACTOR1_COLOR
,
201 R200_TXC_ARG_A_TFACTOR1_COLOR
| R200_TXC_COMP_ARG_A
,
202 R200_TXC_ARG_A_TFACTOR1_ALPHA
,
203 R200_TXC_ARG_A_TFACTOR1_ALPHA
| R200_TXC_COMP_ARG_A
206 static GLuint r200_primary_color
[] =
208 R200_TXC_ARG_A_DIFFUSE_COLOR
,
209 R200_TXC_ARG_A_DIFFUSE_COLOR
| R200_TXC_COMP_ARG_A
,
210 R200_TXC_ARG_A_DIFFUSE_ALPHA
,
211 R200_TXC_ARG_A_DIFFUSE_ALPHA
| R200_TXC_COMP_ARG_A
214 /* GL_ZERO table - indices 0-3
215 * GL_ONE table - indices 1-4
217 static GLuint r200_zero_color
[] =
220 R200_TXC_ARG_A_ZERO
| R200_TXC_COMP_ARG_A
,
222 R200_TXC_ARG_A_ZERO
| R200_TXC_COMP_ARG_A
,
226 /* The alpha tables only have GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
228 static GLuint r200_register_alpha
[][R200_MAX_TEXTURE_UNITS
] =
231 R200_TXA_ARG_A_R0_ALPHA
,
232 R200_TXA_ARG_A_R1_ALPHA
,
233 R200_TXA_ARG_A_R2_ALPHA
,
234 R200_TXA_ARG_A_R3_ALPHA
,
235 R200_TXA_ARG_A_R4_ALPHA
,
236 R200_TXA_ARG_A_R5_ALPHA
239 R200_TXA_ARG_A_R0_ALPHA
| R200_TXA_COMP_ARG_A
,
240 R200_TXA_ARG_A_R1_ALPHA
| R200_TXA_COMP_ARG_A
,
241 R200_TXA_ARG_A_R2_ALPHA
| R200_TXA_COMP_ARG_A
,
242 R200_TXA_ARG_A_R3_ALPHA
| R200_TXA_COMP_ARG_A
,
243 R200_TXA_ARG_A_R4_ALPHA
| R200_TXA_COMP_ARG_A
,
244 R200_TXA_ARG_A_R5_ALPHA
| R200_TXA_COMP_ARG_A
248 static GLuint r200_tfactor_alpha
[] =
250 R200_TXA_ARG_A_TFACTOR_ALPHA
,
251 R200_TXA_ARG_A_TFACTOR_ALPHA
| R200_TXA_COMP_ARG_A
254 static GLuint r200_tfactor1_alpha
[] =
256 R200_TXA_ARG_A_TFACTOR1_ALPHA
,
257 R200_TXA_ARG_A_TFACTOR1_ALPHA
| R200_TXA_COMP_ARG_A
260 static GLuint r200_primary_alpha
[] =
262 R200_TXA_ARG_A_DIFFUSE_ALPHA
,
263 R200_TXA_ARG_A_DIFFUSE_ALPHA
| R200_TXA_COMP_ARG_A
266 /* GL_ZERO table - indices 0-1
267 * GL_ONE table - indices 1-2
269 static GLuint r200_zero_alpha
[] =
272 R200_TXA_ARG_A_ZERO
| R200_TXA_COMP_ARG_A
,
277 /* Extract the arg from slot A, shift it into the correct argument slot
278 * and set the corresponding complement bit.
280 #define R200_COLOR_ARG( n, arg ) \
283 ((color_arg[n] & R200_TXC_ARG_A_MASK) \
284 << R200_TXC_ARG_##arg##_SHIFT); \
286 ((color_arg[n] >> R200_TXC_COMP_ARG_A_SHIFT) \
287 << R200_TXC_COMP_ARG_##arg##_SHIFT); \
290 #define R200_ALPHA_ARG( n, arg ) \
293 ((alpha_arg[n] & R200_TXA_ARG_A_MASK) \
294 << R200_TXA_ARG_##arg##_SHIFT); \
296 ((alpha_arg[n] >> R200_TXA_COMP_ARG_A_SHIFT) \
297 << R200_TXA_COMP_ARG_##arg##_SHIFT); \
301 /* ================================================================
302 * Texture unit state management
305 static GLboolean
r200UpdateTextureEnv( GLcontext
*ctx
, int unit
, int slot
, GLuint replaceargs
)
307 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
308 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[unit
];
309 GLuint color_combine
, alpha_combine
;
310 GLuint color_scale
= rmesa
->hw
.pix
[slot
].cmd
[PIX_PP_TXCBLEND2
] &
311 ~(R200_TXC_SCALE_MASK
| R200_TXC_OUTPUT_REG_MASK
| R200_TXC_TFACTOR_SEL_MASK
|
312 R200_TXC_TFACTOR1_SEL_MASK
);
313 GLuint alpha_scale
= rmesa
->hw
.pix
[slot
].cmd
[PIX_PP_TXABLEND2
] &
314 ~(R200_TXA_DOT_ALPHA
| R200_TXA_SCALE_MASK
| R200_TXA_OUTPUT_REG_MASK
|
315 R200_TXA_TFACTOR_SEL_MASK
| R200_TXA_TFACTOR1_SEL_MASK
);
317 /* texUnit->_Current can be NULL if and only if the texture unit is
318 * not actually enabled.
320 assert( (texUnit
->_ReallyEnabled
== 0)
321 || (texUnit
->_Current
!= NULL
) );
323 if ( R200_DEBUG
& RADEON_TEXTURE
) {
324 fprintf( stderr
, "%s( %p, %d )\n", __FUNCTION__
, (void *)ctx
, unit
);
327 /* Set the texture environment state. Isn't this nice and clean?
328 * The chip will automagically set the texture alpha to 0xff when
329 * the texture format does not include an alpha component. This
330 * reduces the amount of special-casing we have to do, alpha-only
331 * textures being a notable exception.
334 color_scale
|= ((rmesa
->state
.texture
.unit
[unit
].outputreg
+ 1) << R200_TXC_OUTPUT_REG_SHIFT
) |
335 (unit
<< R200_TXC_TFACTOR_SEL_SHIFT
) |
336 (replaceargs
<< R200_TXC_TFACTOR1_SEL_SHIFT
);
337 alpha_scale
|= ((rmesa
->state
.texture
.unit
[unit
].outputreg
+ 1) << R200_TXA_OUTPUT_REG_SHIFT
) |
338 (unit
<< R200_TXA_TFACTOR_SEL_SHIFT
) |
339 (replaceargs
<< R200_TXA_TFACTOR1_SEL_SHIFT
);
341 if ( !texUnit
->_ReallyEnabled
) {
343 color_combine
= R200_TXC_ARG_A_ZERO
| R200_TXC_ARG_B_ZERO
344 | R200_TXC_ARG_C_DIFFUSE_COLOR
| R200_TXC_OP_MADD
;
345 alpha_combine
= R200_TXA_ARG_A_ZERO
| R200_TXA_ARG_B_ZERO
346 | R200_TXA_ARG_C_DIFFUSE_ALPHA
| R200_TXA_OP_MADD
;
349 GLuint color_arg
[3], alpha_arg
[3];
351 const GLuint numColorArgs
= texUnit
->_CurrentCombine
->_NumArgsRGB
;
352 const GLuint numAlphaArgs
= texUnit
->_CurrentCombine
->_NumArgsA
;
353 GLuint RGBshift
= texUnit
->_CurrentCombine
->ScaleShiftRGB
;
354 GLuint Ashift
= texUnit
->_CurrentCombine
->ScaleShiftA
;
357 const GLint replaceoprgb
=
358 ctx
->Texture
.Unit
[replaceargs
]._CurrentCombine
->OperandRGB
[0] - GL_SRC_COLOR
;
359 const GLint replaceopa
=
360 ctx
->Texture
.Unit
[replaceargs
]._CurrentCombine
->OperandA
[0] - GL_SRC_ALPHA
;
363 * Extract the color and alpha combine function arguments.
365 for ( i
= 0 ; i
< numColorArgs
; i
++ ) {
366 GLint op
= texUnit
->_CurrentCombine
->OperandRGB
[i
] - GL_SRC_COLOR
;
367 const GLint srcRGBi
= texUnit
->_CurrentCombine
->SourceRGB
[i
];
372 color_arg
[i
] = r200_register_color
[op
][unit
];
375 color_arg
[i
] = r200_tfactor_color
[op
];
377 case GL_PRIMARY_COLOR
:
378 color_arg
[i
] = r200_primary_color
[op
];
381 if (replaceargs
!= unit
) {
382 const GLint srcRGBreplace
=
383 ctx
->Texture
.Unit
[replaceargs
]._CurrentCombine
->SourceRGB
[0];
385 op
= op
^ replaceopa
;
388 op
= op
^ replaceoprgb
;
390 switch (srcRGBreplace
) {
392 color_arg
[i
] = r200_register_color
[op
][replaceargs
];
395 color_arg
[i
] = r200_tfactor1_color
[op
];
397 case GL_PRIMARY_COLOR
:
398 color_arg
[i
] = r200_primary_color
[op
];
402 color_arg
[i
] = r200_primary_color
[op
];
404 color_arg
[i
] = r200_register_color
[op
]
405 [rmesa
->state
.texture
.unit
[replaceargs
- 1].outputreg
];
408 color_arg
[i
] = r200_zero_color
[op
];
411 color_arg
[i
] = r200_zero_color
[op
+1];
419 color_arg
[i
] = r200_register_color
[op
][srcRGBreplace
- GL_TEXTURE0
];
427 color_arg
[i
] = r200_primary_color
[op
];
429 color_arg
[i
] = r200_register_color
[op
]
430 [rmesa
->state
.texture
.unit
[unit
- 1].outputreg
];
434 color_arg
[i
] = r200_zero_color
[op
];
437 color_arg
[i
] = r200_zero_color
[op
+1];
445 color_arg
[i
] = r200_register_color
[op
][srcRGBi
- GL_TEXTURE0
];
452 for ( i
= 0 ; i
< numAlphaArgs
; i
++ ) {
453 GLint op
= texUnit
->_CurrentCombine
->OperandA
[i
] - GL_SRC_ALPHA
;
454 const GLint srcAi
= texUnit
->_CurrentCombine
->SourceA
[i
];
459 alpha_arg
[i
] = r200_register_alpha
[op
][unit
];
462 alpha_arg
[i
] = r200_tfactor_alpha
[op
];
464 case GL_PRIMARY_COLOR
:
465 alpha_arg
[i
] = r200_primary_alpha
[op
];
468 if (replaceargs
!= unit
) {
469 const GLint srcAreplace
=
470 ctx
->Texture
.Unit
[replaceargs
]._CurrentCombine
->SourceA
[0];
471 op
= op
^ replaceopa
;
472 switch (srcAreplace
) {
474 alpha_arg
[i
] = r200_register_alpha
[op
][replaceargs
];
477 alpha_arg
[i
] = r200_tfactor1_alpha
[op
];
479 case GL_PRIMARY_COLOR
:
480 alpha_arg
[i
] = r200_primary_alpha
[op
];
484 alpha_arg
[i
] = r200_primary_alpha
[op
];
486 alpha_arg
[i
] = r200_register_alpha
[op
]
487 [rmesa
->state
.texture
.unit
[replaceargs
- 1].outputreg
];
490 alpha_arg
[i
] = r200_zero_alpha
[op
];
493 alpha_arg
[i
] = r200_zero_alpha
[op
+1];
501 alpha_arg
[i
] = r200_register_alpha
[op
][srcAreplace
- GL_TEXTURE0
];
509 alpha_arg
[i
] = r200_primary_alpha
[op
];
511 alpha_arg
[i
] = r200_register_alpha
[op
]
512 [rmesa
->state
.texture
.unit
[unit
- 1].outputreg
];
516 alpha_arg
[i
] = r200_zero_alpha
[op
];
519 alpha_arg
[i
] = r200_zero_alpha
[op
+1];
527 alpha_arg
[i
] = r200_register_alpha
[op
][srcAi
- GL_TEXTURE0
];
535 * Build up the color and alpha combine functions.
537 switch ( texUnit
->_CurrentCombine
->ModeRGB
) {
539 color_combine
= (R200_TXC_ARG_A_ZERO
|
540 R200_TXC_ARG_B_ZERO
|
542 R200_COLOR_ARG( 0, C
);
545 color_combine
= (R200_TXC_ARG_C_ZERO
|
547 R200_COLOR_ARG( 0, A
);
548 R200_COLOR_ARG( 1, B
);
551 color_combine
= (R200_TXC_ARG_B_ZERO
|
552 R200_TXC_COMP_ARG_B
|
554 R200_COLOR_ARG( 0, A
);
555 R200_COLOR_ARG( 1, C
);
558 color_combine
= (R200_TXC_ARG_B_ZERO
|
559 R200_TXC_COMP_ARG_B
|
560 R200_TXC_BIAS_ARG_C
| /* new */
561 R200_TXC_OP_MADD
); /* was ADDSIGNED */
562 R200_COLOR_ARG( 0, A
);
563 R200_COLOR_ARG( 1, C
);
566 color_combine
= (R200_TXC_ARG_B_ZERO
|
567 R200_TXC_COMP_ARG_B
|
570 R200_COLOR_ARG( 0, A
);
571 R200_COLOR_ARG( 1, C
);
574 color_combine
= (R200_TXC_OP_LERP
);
575 R200_COLOR_ARG( 0, B
);
576 R200_COLOR_ARG( 1, A
);
577 R200_COLOR_ARG( 2, C
);
580 case GL_DOT3_RGB_EXT
:
581 case GL_DOT3_RGBA_EXT
:
582 /* The EXT version of the DOT3 extension does not support the
583 * scale factor, but the ARB version (and the version in OpenGL
591 /* DOT3 works differently on R200 than on R100. On R100, just
592 * setting the DOT3 mode did everything for you. On R200, the
593 * driver has to enable the biasing and scale in the inputs to
594 * put them in the proper [-1,1] range. This is what the 4x and
595 * the -0.5 in the DOT3 spec do. The post-scale is then set
599 color_combine
= (R200_TXC_ARG_C_ZERO
|
601 R200_TXC_BIAS_ARG_A
|
602 R200_TXC_BIAS_ARG_B
|
603 R200_TXC_SCALE_ARG_A
|
604 R200_TXC_SCALE_ARG_B
);
605 R200_COLOR_ARG( 0, A
);
606 R200_COLOR_ARG( 1, B
);
609 case GL_MODULATE_ADD_ATI
:
610 color_combine
= (R200_TXC_OP_MADD
);
611 R200_COLOR_ARG( 0, A
);
612 R200_COLOR_ARG( 1, C
);
613 R200_COLOR_ARG( 2, B
);
615 case GL_MODULATE_SIGNED_ADD_ATI
:
616 color_combine
= (R200_TXC_BIAS_ARG_C
| /* new */
617 R200_TXC_OP_MADD
); /* was ADDSIGNED */
618 R200_COLOR_ARG( 0, A
);
619 R200_COLOR_ARG( 1, C
);
620 R200_COLOR_ARG( 2, B
);
622 case GL_MODULATE_SUBTRACT_ATI
:
623 color_combine
= (R200_TXC_NEG_ARG_C
|
625 R200_COLOR_ARG( 0, A
);
626 R200_COLOR_ARG( 1, C
);
627 R200_COLOR_ARG( 2, B
);
633 switch ( texUnit
->_CurrentCombine
->ModeA
) {
635 alpha_combine
= (R200_TXA_ARG_A_ZERO
|
636 R200_TXA_ARG_B_ZERO
|
638 R200_ALPHA_ARG( 0, C
);
641 alpha_combine
= (R200_TXA_ARG_C_ZERO
|
643 R200_ALPHA_ARG( 0, A
);
644 R200_ALPHA_ARG( 1, B
);
647 alpha_combine
= (R200_TXA_ARG_B_ZERO
|
648 R200_TXA_COMP_ARG_B
|
650 R200_ALPHA_ARG( 0, A
);
651 R200_ALPHA_ARG( 1, C
);
654 alpha_combine
= (R200_TXA_ARG_B_ZERO
|
655 R200_TXA_COMP_ARG_B
|
656 R200_TXA_BIAS_ARG_C
| /* new */
657 R200_TXA_OP_MADD
); /* was ADDSIGNED */
658 R200_ALPHA_ARG( 0, A
);
659 R200_ALPHA_ARG( 1, C
);
662 alpha_combine
= (R200_TXA_ARG_B_ZERO
|
663 R200_TXA_COMP_ARG_B
|
666 R200_ALPHA_ARG( 0, A
);
667 R200_ALPHA_ARG( 1, C
);
670 alpha_combine
= (R200_TXA_OP_LERP
);
671 R200_ALPHA_ARG( 0, B
);
672 R200_ALPHA_ARG( 1, A
);
673 R200_ALPHA_ARG( 2, C
);
676 case GL_MODULATE_ADD_ATI
:
677 alpha_combine
= (R200_TXA_OP_MADD
);
678 R200_ALPHA_ARG( 0, A
);
679 R200_ALPHA_ARG( 1, C
);
680 R200_ALPHA_ARG( 2, B
);
682 case GL_MODULATE_SIGNED_ADD_ATI
:
683 alpha_combine
= (R200_TXA_BIAS_ARG_C
| /* new */
684 R200_TXA_OP_MADD
); /* was ADDSIGNED */
685 R200_ALPHA_ARG( 0, A
);
686 R200_ALPHA_ARG( 1, C
);
687 R200_ALPHA_ARG( 2, B
);
689 case GL_MODULATE_SUBTRACT_ATI
:
690 alpha_combine
= (R200_TXA_NEG_ARG_C
|
692 R200_ALPHA_ARG( 0, A
);
693 R200_ALPHA_ARG( 1, C
);
694 R200_ALPHA_ARG( 2, B
);
700 if ( (texUnit
->_CurrentCombine
->ModeRGB
== GL_DOT3_RGBA_EXT
)
701 || (texUnit
->_CurrentCombine
->ModeRGB
== GL_DOT3_RGBA
) ) {
702 alpha_scale
|= R200_TXA_DOT_ALPHA
;
707 * Apply the scale factor.
709 color_scale
|= (RGBshift
<< R200_TXC_SCALE_SHIFT
);
710 alpha_scale
|= (Ashift
<< R200_TXA_SCALE_SHIFT
);
716 if ( rmesa
->hw
.pix
[slot
].cmd
[PIX_PP_TXCBLEND
] != color_combine
||
717 rmesa
->hw
.pix
[slot
].cmd
[PIX_PP_TXABLEND
] != alpha_combine
||
718 rmesa
->hw
.pix
[slot
].cmd
[PIX_PP_TXCBLEND2
] != color_scale
||
719 rmesa
->hw
.pix
[slot
].cmd
[PIX_PP_TXABLEND2
] != alpha_scale
) {
720 R200_STATECHANGE( rmesa
, pix
[slot
] );
721 rmesa
->hw
.pix
[slot
].cmd
[PIX_PP_TXCBLEND
] = color_combine
;
722 rmesa
->hw
.pix
[slot
].cmd
[PIX_PP_TXABLEND
] = alpha_combine
;
723 rmesa
->hw
.pix
[slot
].cmd
[PIX_PP_TXCBLEND2
] = color_scale
;
724 rmesa
->hw
.pix
[slot
].cmd
[PIX_PP_TXABLEND2
] = alpha_scale
;
730 void r200SetTexOffset(__DRIcontext
* pDRICtx
, GLint texname
,
731 unsigned long long offset
, GLint depth
, GLuint pitch
)
733 r200ContextPtr rmesa
= pDRICtx
->driverPrivate
;
734 struct gl_texture_object
*tObj
=
735 _mesa_lookup_texture(rmesa
->radeon
.glCtx
, texname
);
736 radeonTexObjPtr t
= radeon_tex_obj(tObj
);
741 t
->image_override
= GL_TRUE
;
747 t
->override_offset
= offset
;
748 t
->pp_txpitch
= pitch
- 32;
752 t
->pp_txformat
= tx_table_le
[MESA_FORMAT_ARGB8888
].format
;
753 t
->pp_txfilter
|= tx_table_le
[MESA_FORMAT_ARGB8888
].filter
;
757 t
->pp_txformat
= tx_table_le
[MESA_FORMAT_RGB888
].format
;
758 t
->pp_txfilter
|= tx_table_le
[MESA_FORMAT_RGB888
].filter
;
761 t
->pp_txformat
= tx_table_le
[MESA_FORMAT_RGB565
].format
;
762 t
->pp_txfilter
|= tx_table_le
[MESA_FORMAT_RGB565
].filter
;
767 void r200SetTexBuffer2(__DRIcontext
*pDRICtx
, GLint target
, GLint glx_texture_format
,
768 __DRIdrawable
*dPriv
)
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
;
780 uint32_t internalFormat
, type
, format
;
783 format
= GL_UNSIGNED_BYTE
;
784 internalFormat
= (glx_texture_format
== GLX_TEXTURE_FORMAT_RGB_EXT
? 3 : 4);
786 radeon
= pDRICtx
->driverPrivate
;
787 rmesa
= pDRICtx
->driverPrivate
;
789 rfb
= dPriv
->driverPrivate
;
790 texUnit
= &radeon
->glCtx
->Texture
.Unit
[radeon
->glCtx
->Texture
.CurrentUnit
];
791 texObj
= _mesa_select_tex_object(radeon
->glCtx
, texUnit
, target
);
792 texImage
= _mesa_get_tex_image(radeon
->glCtx
, texObj
, target
, 0);
794 rImage
= get_radeon_texture_image(texImage
);
795 t
= radeon_tex_obj(texObj
);
800 radeon_update_renderbuffers(pDRICtx
, dPriv
);
801 /* back & depth buffer are useless free them right away */
802 rb
= (void*)rfb
->base
.Attachment
[BUFFER_DEPTH
].Renderbuffer
;
804 radeon_bo_unref(rb
->bo
);
807 rb
= (void*)rfb
->base
.Attachment
[BUFFER_BACK_LEFT
].Renderbuffer
;
809 radeon_bo_unref(rb
->bo
);
812 rb
= rfb
->color_rb
[0];
813 if (rb
->bo
== NULL
) {
814 /* Failed to BO for the buffer */
818 _mesa_lock_texture(radeon
->glCtx
, texObj
);
820 radeon_bo_unref(t
->bo
);
824 radeon_bo_unref(rImage
->bo
);
828 radeon_miptree_unreference(&t
->mt
);
829 radeon_miptree_unreference(&rImage
->mt
);
831 _mesa_init_teximage_fields(radeon
->glCtx
, target
, texImage
,
832 rb
->base
.Width
, rb
->base
.Height
, 1, 0, rb
->cpp
);
833 texImage
->RowStride
= rb
->pitch
/ rb
->cpp
;
836 radeon_bo_ref(rImage
->bo
);
838 radeon_bo_ref(t
->bo
);
840 t
->image_override
= GL_TRUE
;
841 t
->override_offset
= 0;
842 t
->pp_txpitch
&= (1 << 13) -1;
843 pitch_val
= rb
->pitch
;
846 if (glx_texture_format
== GLX_TEXTURE_FORMAT_RGB_EXT
)
847 t
->pp_txformat
= tx_table_le
[MESA_FORMAT_RGB888
].format
;
849 t
->pp_txformat
= tx_table_le
[MESA_FORMAT_ARGB8888
].format
;
850 t
->pp_txfilter
|= tx_table_le
[MESA_FORMAT_ARGB8888
].filter
;
854 t
->pp_txformat
= tx_table_le
[MESA_FORMAT_RGB888
].format
;
855 t
->pp_txfilter
|= tx_table_le
[MESA_FORMAT_RGB888
].filter
;
858 t
->pp_txformat
= tx_table_le
[MESA_FORMAT_RGB565
].format
;
859 t
->pp_txfilter
|= tx_table_le
[MESA_FORMAT_RGB565
].filter
;
862 t
->pp_txsize
= ((rb
->base
.Width
- 1) << RADEON_TEX_USIZE_SHIFT
)
863 | ((rb
->base
.Height
- 1) << RADEON_TEX_VSIZE_SHIFT
);
864 t
->pp_txformat
|= R200_TXFORMAT_NON_POWER2
;
865 t
->pp_txpitch
= pitch_val
;
868 t
->validated
= GL_TRUE
;
869 _mesa_unlock_texture(radeon
->glCtx
, texObj
);
874 void r200SetTexBuffer(__DRIcontext
*pDRICtx
, GLint target
, __DRIdrawable
*dPriv
)
876 r200SetTexBuffer2(pDRICtx
, target
, GLX_TEXTURE_FORMAT_RGBA_EXT
, dPriv
);
883 static GLboolean
r200UpdateAllTexEnv( GLcontext
*ctx
)
885 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
886 GLint i
, j
, currslot
;
887 GLint maxunitused
= -1;
888 GLboolean texregfree
[6] = {GL_TRUE
, GL_TRUE
, GL_TRUE
, GL_TRUE
, GL_TRUE
, GL_TRUE
};
889 GLubyte stageref
[7] = {0, 0, 0, 0, 0, 0, 0};
890 GLint nextunit
[R200_MAX_TEXTURE_UNITS
] = {0, 0, 0, 0, 0, 0};
891 GLint currentnext
= -1;
894 /* find highest used unit */
895 for ( j
= 0; j
< R200_MAX_TEXTURE_UNITS
; j
++) {
896 if (ctx
->Texture
.Unit
[j
]._ReallyEnabled
) {
900 stageref
[maxunitused
+ 1] = REF_COLOR
| REF_ALPHA
;
902 for ( j
= maxunitused
; j
>= 0; j
-- ) {
903 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[j
];
905 rmesa
->state
.texture
.unit
[j
].outputreg
= -1;
907 if (stageref
[j
+ 1]) {
909 /* use the lowest available reg. That gets us automatically reg0 for the last stage.
910 need this even for disabled units, as it may get referenced due to the replace
912 for ( i
= 0 ; i
< R200_MAX_TEXTURE_UNITS
; i
++ ) {
914 rmesa
->state
.texture
.unit
[j
].outputreg
= i
;
918 if (rmesa
->state
.texture
.unit
[j
].outputreg
== -1) {
919 /* no more free regs we can use. Need a fallback :-( */
923 nextunit
[j
] = currentnext
;
925 if (!texUnit
->_ReallyEnabled
) {
926 /* the not enabled stages are referenced "indirectly",
927 must not cut off the lower stages */
928 stageref
[j
] = REF_COLOR
| REF_ALPHA
;
933 const GLuint numColorArgs
= texUnit
->_CurrentCombine
->_NumArgsRGB
;
934 const GLuint numAlphaArgs
= texUnit
->_CurrentCombine
->_NumArgsA
;
935 const GLboolean isdot3rgba
= (texUnit
->_CurrentCombine
->ModeRGB
== GL_DOT3_RGBA
) ||
936 (texUnit
->_CurrentCombine
->ModeRGB
== GL_DOT3_RGBA_EXT
);
939 /* check if we need the color part, special case for dot3_rgba
940 as if only the alpha part is referenced later on it still is using the color part */
941 if ((stageref
[j
+ 1] & REF_COLOR
) || isdot3rgba
) {
942 for ( i
= 0 ; i
< numColorArgs
; i
++ ) {
943 const GLuint srcRGBi
= texUnit
->_CurrentCombine
->SourceRGB
[i
];
944 const GLuint op
= texUnit
->_CurrentCombine
->OperandRGB
[i
];
947 /* op 0/1 are referencing color, op 2/3 alpha */
948 stageref
[j
] |= (op
>> 1) + 1;
951 texregfree
[j
] = GL_FALSE
;
959 texregfree
[srcRGBi
- GL_TEXTURE0
] = GL_FALSE
;
961 default: /* don't care about other sources here */
967 /* alpha args are ignored for dot3_rgba */
968 if ((stageref
[j
+ 1] & REF_ALPHA
) && !isdot3rgba
) {
970 for ( i
= 0 ; i
< numAlphaArgs
; i
++ ) {
971 const GLuint srcAi
= texUnit
->_CurrentCombine
->SourceA
[i
];
974 stageref
[j
] |= REF_ALPHA
;
977 texregfree
[j
] = GL_FALSE
;
985 texregfree
[srcAi
- GL_TEXTURE0
] = GL_FALSE
;
987 default: /* don't care about other sources here */
995 /* don't enable texture sampling for units if the result is not used */
996 for (i
= 0; i
< R200_MAX_TEXTURE_UNITS
; i
++) {
997 if (ctx
->Texture
.Unit
[i
]._ReallyEnabled
&& !texregfree
[i
])
998 rmesa
->state
.texture
.unit
[i
].unitneeded
= ctx
->Texture
.Unit
[i
]._ReallyEnabled
;
999 else rmesa
->state
.texture
.unit
[i
].unitneeded
= 0;
1004 rmesa
->state
.envneeded
= 1;
1007 while ((i
<= maxunitused
) && (i
>= 0)) {
1008 /* only output instruction if the results are referenced */
1009 if (ctx
->Texture
.Unit
[i
]._ReallyEnabled
&& stageref
[i
+1]) {
1010 GLuint replaceunit
= i
;
1011 /* try to optimize GL_REPLACE away (only one level deep though) */
1012 if ( (ctx
->Texture
.Unit
[i
]._CurrentCombine
->ModeRGB
== GL_REPLACE
) &&
1013 (ctx
->Texture
.Unit
[i
]._CurrentCombine
->ModeA
== GL_REPLACE
) &&
1014 (ctx
->Texture
.Unit
[i
]._CurrentCombine
->ScaleShiftRGB
== 0) &&
1015 (ctx
->Texture
.Unit
[i
]._CurrentCombine
->ScaleShiftA
== 0) &&
1016 (nextunit
[i
] > 0) ) {
1017 /* yippie! can optimize it away! */
1022 /* need env instruction slot */
1023 rmesa
->state
.envneeded
|= 1 << currslot
;
1024 ok
= r200UpdateTextureEnv( ctx
, i
, currslot
, replaceunit
);
1025 if (!ok
) return GL_FALSE
;
1031 if (currslot
== 0) {
1032 /* need one stage at least */
1033 rmesa
->state
.texture
.unit
[0].outputreg
= 0;
1034 ok
= r200UpdateTextureEnv( ctx
, 0, 0, 0 );
1037 R200_STATECHANGE( rmesa
, ctx
);
1038 rmesa
->hw
.ctx
.cmd
[CTX_PP_CNTL
] &= ~(R200_TEX_BLEND_ENABLE_MASK
| R200_MULTI_PASS_ENABLE
);
1039 rmesa
->hw
.ctx
.cmd
[CTX_PP_CNTL
] |= rmesa
->state
.envneeded
<< R200_TEX_BLEND_0_ENABLE_SHIFT
;
1048 #define TEXOBJ_TXFILTER_MASK (R200_MAX_MIP_LEVEL_MASK | \
1049 R200_MIN_FILTER_MASK | \
1050 R200_MAG_FILTER_MASK | \
1051 R200_MAX_ANISO_MASK | \
1053 R200_YUV_TEMPERATURE_MASK | \
1054 R200_CLAMP_S_MASK | \
1055 R200_CLAMP_T_MASK | \
1056 R200_BORDER_MODE_D3D )
1058 #define TEXOBJ_TXFORMAT_MASK (R200_TXFORMAT_WIDTH_MASK | \
1059 R200_TXFORMAT_HEIGHT_MASK | \
1060 R200_TXFORMAT_FORMAT_MASK | \
1061 R200_TXFORMAT_F5_WIDTH_MASK | \
1062 R200_TXFORMAT_F5_HEIGHT_MASK | \
1063 R200_TXFORMAT_ALPHA_IN_MAP | \
1064 R200_TXFORMAT_CUBIC_MAP_ENABLE | \
1065 R200_TXFORMAT_NON_POWER2)
1067 #define TEXOBJ_TXFORMAT_X_MASK (R200_DEPTH_LOG2_MASK | \
1068 R200_TEXCOORD_MASK | \
1069 R200_CLAMP_Q_MASK | \
1070 R200_VOLUME_FILTER_MASK)
1073 static void disable_tex_obj_state( r200ContextPtr rmesa
,
1077 R200_STATECHANGE( rmesa
, vtx
);
1078 rmesa
->hw
.vtx
.cmd
[VTX_TCL_OUTPUT_VTXFMT_1
] &= ~(7 << (unit
* 3));
1080 R200_STATECHANGE( rmesa
, ctx
);
1081 rmesa
->hw
.ctx
.cmd
[CTX_PP_CNTL
] &= ~(R200_TEX_0_ENABLE
<< unit
);
1082 if (rmesa
->radeon
.TclFallback
& (R200_TCL_FALLBACK_TEXGEN_0
<<unit
)) {
1083 TCL_FALLBACK( rmesa
->radeon
.glCtx
, (R200_TCL_FALLBACK_TEXGEN_0
<<unit
), GL_FALSE
);
1086 /* Actually want to keep all units less than max active texture
1087 * enabled, right? Fix this for >2 texunits.
1091 GLuint tmp
= rmesa
->TexGenEnabled
;
1093 rmesa
->TexGenEnabled
&= ~(R200_TEXGEN_TEXMAT_0_ENABLE
<<unit
);
1094 rmesa
->TexGenEnabled
&= ~(R200_TEXMAT_0_ENABLE
<<unit
);
1095 rmesa
->TexGenNeedNormals
[unit
] = GL_FALSE
;
1096 rmesa
->TexGenCompSel
&= ~(R200_OUTPUT_TEX_0
<< unit
);
1098 if (tmp
!= rmesa
->TexGenEnabled
) {
1099 rmesa
->recheck_texgen
[unit
] = GL_TRUE
;
1100 rmesa
->radeon
.NewGLState
|= _NEW_TEXTURE_MATRIX
;
1104 static void import_tex_obj_state( r200ContextPtr rmesa
,
1106 radeonTexObjPtr texobj
)
1108 /* do not use RADEON_DB_STATE to avoid stale texture caches */
1109 GLuint
*cmd
= &rmesa
->hw
.tex
[unit
].cmd
[TEX_CMD_0
];
1111 R200_STATECHANGE( rmesa
, tex
[unit
] );
1113 cmd
[TEX_PP_TXFILTER
] &= ~TEXOBJ_TXFILTER_MASK
;
1114 cmd
[TEX_PP_TXFILTER
] |= texobj
->pp_txfilter
& TEXOBJ_TXFILTER_MASK
;
1115 cmd
[TEX_PP_TXFORMAT
] &= ~TEXOBJ_TXFORMAT_MASK
;
1116 cmd
[TEX_PP_TXFORMAT
] |= texobj
->pp_txformat
& TEXOBJ_TXFORMAT_MASK
;
1117 cmd
[TEX_PP_TXFORMAT_X
] &= ~TEXOBJ_TXFORMAT_X_MASK
;
1118 cmd
[TEX_PP_TXFORMAT_X
] |= texobj
->pp_txformat_x
& TEXOBJ_TXFORMAT_X_MASK
;
1119 cmd
[TEX_PP_TXSIZE
] = texobj
->pp_txsize
; /* NPOT only! */
1120 cmd
[TEX_PP_TXPITCH
] = texobj
->pp_txpitch
; /* NPOT only! */
1121 cmd
[TEX_PP_BORDER_COLOR
] = texobj
->pp_border_color
;
1123 if (texobj
->base
.Target
== GL_TEXTURE_CUBE_MAP
) {
1124 GLuint
*cube_cmd
= &rmesa
->hw
.cube
[unit
].cmd
[CUBE_CMD_0
];
1126 R200_STATECHANGE( rmesa
, cube
[unit
] );
1127 cube_cmd
[CUBE_PP_CUBIC_FACES
] = texobj
->pp_cubic_faces
;
1128 if (rmesa
->radeon
.radeonScreen
->drmSupportsFragShader
) {
1129 /* that value is submitted twice. could change cube atom
1130 to not include that command when new drm is used */
1131 cmd
[TEX_PP_CUBIC_FACES
] = texobj
->pp_cubic_faces
;
1137 static void set_texgen_matrix( r200ContextPtr rmesa
,
1139 const GLfloat
*s_plane
,
1140 const GLfloat
*t_plane
,
1141 const GLfloat
*r_plane
,
1142 const GLfloat
*q_plane
)
1166 _math_matrix_loadf( &(rmesa
->TexGenMatrix
[unit
]), m
);
1167 _math_matrix_analyse( &(rmesa
->TexGenMatrix
[unit
]) );
1168 rmesa
->TexGenEnabled
|= R200_TEXMAT_0_ENABLE
<<unit
;
1172 static GLuint
r200_need_dis_texgen(const GLbitfield texGenEnabled
,
1173 const GLfloat
*planeS
,
1174 const GLfloat
*planeT
,
1175 const GLfloat
*planeR
,
1176 const GLfloat
*planeQ
)
1178 GLuint needtgenable
= 0;
1180 if (!(texGenEnabled
& S_BIT
)) {
1181 if (((texGenEnabled
& T_BIT
) && planeT
[0] != 0.0) ||
1182 ((texGenEnabled
& R_BIT
) && planeR
[0] != 0.0) ||
1183 ((texGenEnabled
& Q_BIT
) && planeQ
[0] != 0.0)) {
1184 needtgenable
|= S_BIT
;
1187 if (!(texGenEnabled
& T_BIT
)) {
1188 if (((texGenEnabled
& S_BIT
) && planeS
[1] != 0.0) ||
1189 ((texGenEnabled
& R_BIT
) && planeR
[1] != 0.0) ||
1190 ((texGenEnabled
& Q_BIT
) && planeQ
[1] != 0.0)) {
1191 needtgenable
|= T_BIT
;
1194 if (!(texGenEnabled
& R_BIT
)) {
1195 if (((texGenEnabled
& S_BIT
) && planeS
[2] != 0.0) ||
1196 ((texGenEnabled
& T_BIT
) && planeT
[2] != 0.0) ||
1197 ((texGenEnabled
& Q_BIT
) && planeQ
[2] != 0.0)) {
1198 needtgenable
|= R_BIT
;
1201 if (!(texGenEnabled
& Q_BIT
)) {
1202 if (((texGenEnabled
& S_BIT
) && planeS
[3] != 0.0) ||
1203 ((texGenEnabled
& T_BIT
) && planeT
[3] != 0.0) ||
1204 ((texGenEnabled
& R_BIT
) && planeR
[3] != 0.0)) {
1205 needtgenable
|= Q_BIT
;
1209 return needtgenable
;
1214 * Returns GL_FALSE if fallback required.
1216 static GLboolean
r200_validate_texgen( GLcontext
*ctx
, GLuint unit
)
1218 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
1219 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[unit
];
1220 GLuint inputshift
= R200_TEXGEN_0_INPUT_SHIFT
+ unit
*4;
1223 GLboolean mixed_fallback
= GL_FALSE
;
1224 static const GLfloat I
[16] = {
1229 static const GLfloat reflect
[16] = {
1235 rmesa
->TexGenCompSel
&= ~(R200_OUTPUT_TEX_0
<< unit
);
1236 rmesa
->TexGenEnabled
&= ~(R200_TEXGEN_TEXMAT_0_ENABLE
<<unit
);
1237 rmesa
->TexGenEnabled
&= ~(R200_TEXMAT_0_ENABLE
<<unit
);
1238 rmesa
->TexGenNeedNormals
[unit
] = GL_FALSE
;
1239 tgi
= rmesa
->hw
.tcg
.cmd
[TCG_TEX_PROC_CTL_1
] & ~(R200_TEXGEN_INPUT_MASK
<<
1241 tgcm
= rmesa
->hw
.tcg
.cmd
[TCG_TEX_PROC_CTL_2
] & ~(R200_TEXGEN_COMP_MASK
<<
1245 fprintf(stderr
, "%s unit %d\n", __FUNCTION__
, unit
);
1247 if (texUnit
->TexGenEnabled
& S_BIT
) {
1248 mode
= texUnit
->GenS
.Mode
;
1250 tgcm
|= R200_TEXGEN_COMP_S
<< (unit
* 4);
1253 if (texUnit
->TexGenEnabled
& T_BIT
) {
1254 if (texUnit
->GenT
.Mode
!= mode
)
1255 mixed_fallback
= GL_TRUE
;
1257 tgcm
|= R200_TEXGEN_COMP_T
<< (unit
* 4);
1259 if (texUnit
->TexGenEnabled
& R_BIT
) {
1260 if (texUnit
->GenR
.Mode
!= mode
)
1261 mixed_fallback
= GL_TRUE
;
1263 tgcm
|= R200_TEXGEN_COMP_R
<< (unit
* 4);
1266 if (texUnit
->TexGenEnabled
& Q_BIT
) {
1267 if (texUnit
->GenQ
.Mode
!= mode
)
1268 mixed_fallback
= GL_TRUE
;
1270 tgcm
|= R200_TEXGEN_COMP_Q
<< (unit
* 4);
1273 if (mixed_fallback
) {
1274 if (R200_DEBUG
& RADEON_FALLBACKS
)
1275 fprintf(stderr
, "fallback mixed texgen, 0x%x (0x%x 0x%x 0x%x 0x%x)\n",
1276 texUnit
->TexGenEnabled
, texUnit
->GenS
.Mode
, texUnit
->GenT
.Mode
,
1277 texUnit
->GenR
.Mode
, texUnit
->GenQ
.Mode
);
1281 /* we CANNOT do mixed mode if the texgen mode requires a plane where the input
1282 is not enabled for texgen, since the planes are concatenated into texmat,
1283 and thus the input will come from texcoord rather than tex gen equation!
1284 Either fallback or just hope that those texcoords aren't really needed...
1285 Assuming the former will cause lots of unnecessary fallbacks, the latter will
1286 generate bogus results sometimes - it's pretty much impossible to really know
1287 when a fallback is needed, depends on texmat and what sort of texture is bound
1288 etc, - for now fallback if we're missing either S or T bits, there's a high
1289 probability we need the texcoords in that case.
1290 That's a lot of work for some obscure texgen mixed mode fixup - why oh why
1291 doesn't the chip just directly accept the plane parameters :-(. */
1293 case GL_OBJECT_LINEAR
: {
1294 GLuint needtgenable
= r200_need_dis_texgen( texUnit
->TexGenEnabled
,
1295 texUnit
->GenS
.ObjectPlane
,
1296 texUnit
->GenT
.ObjectPlane
,
1297 texUnit
->GenR
.ObjectPlane
,
1298 texUnit
->GenQ
.ObjectPlane
);
1299 if (needtgenable
& (S_BIT
| T_BIT
)) {
1300 if (R200_DEBUG
& RADEON_FALLBACKS
)
1301 fprintf(stderr
, "fallback mixed texgen / obj plane, 0x%x\n",
1302 texUnit
->TexGenEnabled
);
1305 if (needtgenable
& (R_BIT
)) {
1306 tgcm
&= ~(R200_TEXGEN_COMP_R
<< (unit
* 4));
1308 if (needtgenable
& (Q_BIT
)) {
1309 tgcm
&= ~(R200_TEXGEN_COMP_Q
<< (unit
* 4));
1312 tgi
|= R200_TEXGEN_INPUT_OBJ
<< inputshift
;
1313 set_texgen_matrix( rmesa
, unit
,
1314 (texUnit
->TexGenEnabled
& S_BIT
) ? texUnit
->GenS
.ObjectPlane
: I
,
1315 (texUnit
->TexGenEnabled
& T_BIT
) ? texUnit
->GenT
.ObjectPlane
: I
+ 4,
1316 (texUnit
->TexGenEnabled
& R_BIT
) ? texUnit
->GenR
.ObjectPlane
: I
+ 8,
1317 (texUnit
->TexGenEnabled
& Q_BIT
) ? texUnit
->GenQ
.ObjectPlane
: I
+ 12);
1321 case GL_EYE_LINEAR
: {
1322 GLuint needtgenable
= r200_need_dis_texgen( texUnit
->TexGenEnabled
,
1323 texUnit
->GenS
.EyePlane
,
1324 texUnit
->GenT
.EyePlane
,
1325 texUnit
->GenR
.EyePlane
,
1326 texUnit
->GenQ
.EyePlane
);
1327 if (needtgenable
& (S_BIT
| T_BIT
)) {
1328 if (R200_DEBUG
& RADEON_FALLBACKS
)
1329 fprintf(stderr
, "fallback mixed texgen / eye plane, 0x%x\n",
1330 texUnit
->TexGenEnabled
);
1333 if (needtgenable
& (R_BIT
)) {
1334 tgcm
&= ~(R200_TEXGEN_COMP_R
<< (unit
* 4));
1336 if (needtgenable
& (Q_BIT
)) {
1337 tgcm
&= ~(R200_TEXGEN_COMP_Q
<< (unit
* 4));
1339 tgi
|= R200_TEXGEN_INPUT_EYE
<< inputshift
;
1340 set_texgen_matrix( rmesa
, unit
,
1341 (texUnit
->TexGenEnabled
& S_BIT
) ? texUnit
->GenS
.EyePlane
: I
,
1342 (texUnit
->TexGenEnabled
& T_BIT
) ? texUnit
->GenT
.EyePlane
: I
+ 4,
1343 (texUnit
->TexGenEnabled
& R_BIT
) ? texUnit
->GenR
.EyePlane
: I
+ 8,
1344 (texUnit
->TexGenEnabled
& Q_BIT
) ? texUnit
->GenQ
.EyePlane
: I
+ 12);
1348 case GL_REFLECTION_MAP_NV
:
1349 rmesa
->TexGenNeedNormals
[unit
] = GL_TRUE
;
1350 tgi
|= R200_TEXGEN_INPUT_EYE_REFLECT
<< inputshift
;
1351 /* pretty weird, must only negate when lighting is enabled? */
1352 if (ctx
->Light
.Enabled
)
1353 set_texgen_matrix( rmesa
, unit
,
1354 (texUnit
->TexGenEnabled
& S_BIT
) ? reflect
: I
,
1355 (texUnit
->TexGenEnabled
& T_BIT
) ? reflect
+ 4 : I
+ 4,
1356 (texUnit
->TexGenEnabled
& R_BIT
) ? reflect
+ 8 : I
+ 8,
1360 case GL_NORMAL_MAP_NV
:
1361 rmesa
->TexGenNeedNormals
[unit
] = GL_TRUE
;
1362 tgi
|= R200_TEXGEN_INPUT_EYE_NORMAL
<<inputshift
;
1366 rmesa
->TexGenNeedNormals
[unit
] = GL_TRUE
;
1367 tgi
|= R200_TEXGEN_INPUT_SPHERE
<<inputshift
;
1371 /* All texgen units were disabled, so just pass coords through. */
1372 tgi
|= unit
<< inputshift
;
1376 /* Unsupported mode, fallback:
1378 if (R200_DEBUG
& RADEON_FALLBACKS
)
1379 fprintf(stderr
, "fallback unsupported texgen, %d\n",
1380 texUnit
->GenS
.Mode
);
1384 rmesa
->TexGenEnabled
|= R200_TEXGEN_TEXMAT_0_ENABLE
<< unit
;
1385 rmesa
->TexGenCompSel
|= R200_OUTPUT_TEX_0
<< unit
;
1387 if (tgi
!= rmesa
->hw
.tcg
.cmd
[TCG_TEX_PROC_CTL_1
] ||
1388 tgcm
!= rmesa
->hw
.tcg
.cmd
[TCG_TEX_PROC_CTL_2
])
1390 R200_STATECHANGE(rmesa
, tcg
);
1391 rmesa
->hw
.tcg
.cmd
[TCG_TEX_PROC_CTL_1
] = tgi
;
1392 rmesa
->hw
.tcg
.cmd
[TCG_TEX_PROC_CTL_2
] = tgcm
;
1398 void set_re_cntl_d3d( GLcontext
*ctx
, int unit
, GLboolean use_d3d
)
1400 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
1404 re_cntl
= rmesa
->hw
.set
.cmd
[SET_RE_CNTL
] & ~(R200_VTX_STQ0_D3D
<< (2 * unit
));
1406 re_cntl
|= R200_VTX_STQ0_D3D
<< (2 * unit
);
1408 if ( re_cntl
!= rmesa
->hw
.set
.cmd
[SET_RE_CNTL
] ) {
1409 R200_STATECHANGE( rmesa
, set
);
1410 rmesa
->hw
.set
.cmd
[SET_RE_CNTL
] = re_cntl
;
1415 * Compute the cached hardware register values for the given texture object.
1417 * \param rmesa Context pointer
1418 * \param t the r300 texture object
1420 static void setup_hardware_state(r200ContextPtr rmesa
, radeonTexObj
*t
)
1422 const struct gl_texture_image
*firstImage
= t
->base
.Image
[0][t
->minLod
];
1423 GLint log2Width
, log2Height
, log2Depth
, texelBytes
;
1429 log2Width
= firstImage
->WidthLog2
;
1430 log2Height
= firstImage
->HeightLog2
;
1431 log2Depth
= firstImage
->DepthLog2
;
1432 texelBytes
= _mesa_get_format_bytes(firstImage
->TexFormat
);
1435 if (!t
->image_override
) {
1436 if (VALID_FORMAT(firstImage
->TexFormat
)) {
1437 const struct tx_table
*table
= _mesa_little_endian() ? tx_table_le
:
1440 t
->pp_txformat
&= ~(R200_TXFORMAT_FORMAT_MASK
|
1441 R200_TXFORMAT_ALPHA_IN_MAP
);
1442 t
->pp_txfilter
&= ~R200_YUV_TO_RGB
;
1444 t
->pp_txformat
|= table
[ firstImage
->TexFormat
].format
;
1445 t
->pp_txfilter
|= table
[ firstImage
->TexFormat
].filter
;
1447 _mesa_problem(NULL
, "unexpected texture format in %s",
1453 t
->pp_txfilter
&= ~R200_MAX_MIP_LEVEL_MASK
;
1454 t
->pp_txfilter
|= (t
->maxLod
- t
->minLod
) << R200_MAX_MIP_LEVEL_SHIFT
;
1456 t
->pp_txformat
&= ~(R200_TXFORMAT_WIDTH_MASK
|
1457 R200_TXFORMAT_HEIGHT_MASK
|
1458 R200_TXFORMAT_CUBIC_MAP_ENABLE
|
1459 R200_TXFORMAT_F5_WIDTH_MASK
|
1460 R200_TXFORMAT_F5_HEIGHT_MASK
);
1461 t
->pp_txformat
|= ((log2Width
<< R200_TXFORMAT_WIDTH_SHIFT
) |
1462 (log2Height
<< R200_TXFORMAT_HEIGHT_SHIFT
));
1466 t
->pp_txformat_x
&= ~(R200_DEPTH_LOG2_MASK
| R200_TEXCOORD_MASK
);
1467 if (t
->base
.Target
== GL_TEXTURE_3D
) {
1468 t
->pp_txformat_x
|= (log2Depth
<< R200_DEPTH_LOG2_SHIFT
);
1469 t
->pp_txformat_x
|= R200_TEXCOORD_VOLUME
;
1472 else if (t
->base
.Target
== GL_TEXTURE_CUBE_MAP
) {
1473 ASSERT(log2Width
== log2Height
);
1474 t
->pp_txformat
|= ((log2Width
<< R200_TXFORMAT_F5_WIDTH_SHIFT
) |
1475 (log2Height
<< R200_TXFORMAT_F5_HEIGHT_SHIFT
) |
1476 /* don't think we need this bit, if it exists at all - fglrx does not set it */
1477 (R200_TXFORMAT_CUBIC_MAP_ENABLE
));
1478 t
->pp_txformat_x
|= R200_TEXCOORD_CUBIC_ENV
;
1479 t
->pp_cubic_faces
= ((log2Width
<< R200_FACE_WIDTH_1_SHIFT
) |
1480 (log2Height
<< R200_FACE_HEIGHT_1_SHIFT
) |
1481 (log2Width
<< R200_FACE_WIDTH_2_SHIFT
) |
1482 (log2Height
<< R200_FACE_HEIGHT_2_SHIFT
) |
1483 (log2Width
<< R200_FACE_WIDTH_3_SHIFT
) |
1484 (log2Height
<< R200_FACE_HEIGHT_3_SHIFT
) |
1485 (log2Width
<< R200_FACE_WIDTH_4_SHIFT
) |
1486 (log2Height
<< R200_FACE_HEIGHT_4_SHIFT
));
1489 /* If we don't in fact send enough texture coordinates, q will be 1,
1490 * making TEXCOORD_PROJ act like TEXCOORD_NONPROJ (Right?)
1492 t
->pp_txformat_x
|= R200_TEXCOORD_PROJ
;
1495 t
->pp_txsize
= (((firstImage
->Width
- 1) << R200_PP_TX_WIDTHMASK_SHIFT
)
1496 | ((firstImage
->Height
- 1) << R200_PP_TX_HEIGHTMASK_SHIFT
));
1498 if ( !t
->image_override
) {
1499 if (_mesa_is_format_compressed(firstImage
->TexFormat
))
1500 t
->pp_txpitch
= (firstImage
->Width
+ 63) & ~(63);
1502 t
->pp_txpitch
= ((firstImage
->Width
* texelBytes
) + 63) & ~(63);
1503 t
->pp_txpitch
-= 32;
1506 if (t
->base
.Target
== GL_TEXTURE_RECTANGLE_NV
) {
1507 t
->pp_txformat
|= R200_TXFORMAT_NON_POWER2
;
1512 static GLboolean
r200_validate_texture(GLcontext
*ctx
, struct gl_texture_object
*texObj
, int unit
)
1514 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
1515 radeonTexObj
*t
= radeon_tex_obj(texObj
);
1517 if (!radeon_validate_texture_miptree(ctx
, texObj
))
1520 r200_validate_texgen(ctx
, unit
);
1521 /* Configure the hardware registers (more precisely, the cached version
1522 * of the hardware registers). */
1523 setup_hardware_state(rmesa
, t
);
1525 if (texObj
->Target
== GL_TEXTURE_RECTANGLE_NV
||
1526 texObj
->Target
== GL_TEXTURE_2D
||
1527 texObj
->Target
== GL_TEXTURE_1D
)
1528 set_re_cntl_d3d( ctx
, unit
, GL_FALSE
);
1530 set_re_cntl_d3d( ctx
, unit
, GL_TRUE
);
1531 R200_STATECHANGE( rmesa
, ctx
);
1532 rmesa
->hw
.ctx
.cmd
[CTX_PP_CNTL
] |= R200_TEX_0_ENABLE
<< unit
;
1534 R200_STATECHANGE( rmesa
, vtx
);
1535 rmesa
->hw
.vtx
.cmd
[VTX_TCL_OUTPUT_VTXFMT_1
] &= ~(7 << (unit
* 3));
1536 rmesa
->hw
.vtx
.cmd
[VTX_TCL_OUTPUT_VTXFMT_1
] |= 4 << (unit
* 3);
1538 rmesa
->recheck_texgen
[unit
] = GL_TRUE
;
1539 import_tex_obj_state( rmesa
, unit
, t
);
1541 if (rmesa
->recheck_texgen
[unit
]) {
1542 GLboolean fallback
= !r200_validate_texgen( ctx
, unit
);
1543 TCL_FALLBACK( ctx
, (R200_TCL_FALLBACK_TEXGEN_0
<<unit
), fallback
);
1544 rmesa
->recheck_texgen
[unit
] = 0;
1545 rmesa
->radeon
.NewGLState
|= _NEW_TEXTURE_MATRIX
;
1548 t
->validated
= GL_TRUE
;
1550 FALLBACK( rmesa
, RADEON_FALLBACK_BORDER_MODE
, t
->border_fallback
);
1552 return !t
->border_fallback
;
1555 static GLboolean
r200UpdateTextureUnit(GLcontext
*ctx
, int unit
)
1557 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
1558 GLuint unitneeded
= rmesa
->state
.texture
.unit
[unit
].unitneeded
;
1561 /* disable the unit */
1562 disable_tex_obj_state(rmesa
, unit
);
1566 if (!r200_validate_texture(ctx
, ctx
->Texture
.Unit
[unit
]._Current
, unit
)) {
1568 "failed to validate texture for unit %d.\n",
1570 rmesa
->state
.texture
.unit
[unit
].texobj
= NULL
;
1574 rmesa
->state
.texture
.unit
[unit
].texobj
= radeon_tex_obj(ctx
->Texture
.Unit
[unit
]._Current
);
1579 void r200UpdateTextureState( GLcontext
*ctx
)
1581 r200ContextPtr rmesa
= R200_CONTEXT(ctx
);
1585 /* NOTE: must not manipulate rmesa->state.texture.unit[].unitneeded or
1586 rmesa->state.envneeded before a R200_STATECHANGE (or R200_NEWPRIM) since
1587 we use these to determine if we want to emit the corresponding state
1589 R200_NEWPRIM( rmesa
);
1591 if (ctx
->ATIFragmentShader
._Enabled
) {
1593 for (i
= 0; i
< R200_MAX_TEXTURE_UNITS
; i
++) {
1594 rmesa
->state
.texture
.unit
[i
].unitneeded
= ctx
->Texture
.Unit
[i
]._ReallyEnabled
;
1599 ok
= r200UpdateAllTexEnv( ctx
);
1602 ok
= (r200UpdateTextureUnit( ctx
, 0 ) &&
1603 r200UpdateTextureUnit( ctx
, 1 ) &&
1604 r200UpdateTextureUnit( ctx
, 2 ) &&
1605 r200UpdateTextureUnit( ctx
, 3 ) &&
1606 r200UpdateTextureUnit( ctx
, 4 ) &&
1607 r200UpdateTextureUnit( ctx
, 5 ));
1610 if (ok
&& ctx
->ATIFragmentShader
._Enabled
) {
1611 r200UpdateFragmentShader(ctx
);
1614 FALLBACK( rmesa
, R200_FALLBACK_TEXTURE
, !ok
);
1616 if (rmesa
->radeon
.TclFallback
)
1617 r200ChooseVertexState( ctx
);
1620 if (rmesa
->radeon
.radeonScreen
->chip_family
== CHIP_FAMILY_R200
) {
1623 * T0 hang workaround -------------
1624 * not needed for r200 derivatives
1626 if ((rmesa
->hw
.ctx
.cmd
[CTX_PP_CNTL
] & R200_TEX_ENABLE_MASK
) == R200_TEX_0_ENABLE
&&
1627 (rmesa
->hw
.tex
[0].cmd
[TEX_PP_TXFILTER
] & R200_MIN_FILTER_MASK
) > R200_MIN_FILTER_LINEAR
) {
1629 R200_STATECHANGE(rmesa
, ctx
);
1630 R200_STATECHANGE(rmesa
, tex
[1]);
1631 rmesa
->hw
.ctx
.cmd
[CTX_PP_CNTL
] |= R200_TEX_1_ENABLE
;
1632 if (!(rmesa
->hw
.cst
.cmd
[CST_PP_CNTL_X
] & R200_PPX_TEX_1_ENABLE
))
1633 rmesa
->hw
.tex
[1].cmd
[TEX_PP_TXFORMAT
] &= ~TEXOBJ_TXFORMAT_MASK
;
1634 rmesa
->hw
.tex
[1].cmd
[TEX_PP_TXFORMAT
] |= R200_TXFORMAT_LOOKUP_DISABLE
;
1636 else if (!ctx
->ATIFragmentShader
._Enabled
) {
1637 if ((rmesa
->hw
.ctx
.cmd
[CTX_PP_CNTL
] & R200_TEX_1_ENABLE
) &&
1638 (rmesa
->hw
.tex
[1].cmd
[TEX_PP_TXFORMAT
] & R200_TXFORMAT_LOOKUP_DISABLE
)) {
1639 R200_STATECHANGE(rmesa
, tex
[1]);
1640 rmesa
->hw
.tex
[1].cmd
[TEX_PP_TXFORMAT
] &= ~R200_TXFORMAT_LOOKUP_DISABLE
;
1643 /* do the same workaround for the first pass of a fragment shader.
1644 * completely unknown if necessary / sufficient.
1646 if ((rmesa
->hw
.cst
.cmd
[CST_PP_CNTL_X
] & R200_PPX_TEX_ENABLE_MASK
) == R200_PPX_TEX_0_ENABLE
&&
1647 (rmesa
->hw
.tex
[0].cmd
[TEX_PP_TXFILTER
] & R200_MIN_FILTER_MASK
) > R200_MIN_FILTER_LINEAR
) {
1649 R200_STATECHANGE(rmesa
, cst
);
1650 R200_STATECHANGE(rmesa
, tex
[1]);
1651 rmesa
->hw
.cst
.cmd
[CST_PP_CNTL_X
] |= R200_PPX_TEX_1_ENABLE
;
1652 if (!(rmesa
->hw
.ctx
.cmd
[CTX_PP_CNTL
] & R200_TEX_1_ENABLE
))
1653 rmesa
->hw
.tex
[1].cmd
[TEX_PP_TXFORMAT
] &= ~TEXOBJ_TXFORMAT_MASK
;
1654 rmesa
->hw
.tex
[1].cmd
[TEX_PP_TXMULTI_CTL
] |= R200_PASS1_TXFORMAT_LOOKUP_DISABLE
;
1657 /* maybe needs to be done pairwise due to 2 parallel (physical) tex units ?
1658 looks like that's not the case, if 8500/9100 owners don't complain remove this...
1659 for ( i = 0; i < ctx->Const.MaxTextureUnits; i += 2) {
1660 if (((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & ((R200_TEX_0_ENABLE |
1661 R200_TEX_1_ENABLE ) << i)) == (R200_TEX_0_ENABLE << i)) &&
1662 ((rmesa->hw.tex[i].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK) >
1663 R200_MIN_FILTER_LINEAR)) {
1664 R200_STATECHANGE(rmesa, ctx);
1665 R200_STATECHANGE(rmesa, tex[i+1]);
1666 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= (R200_TEX_1_ENABLE << i);
1667 rmesa->hw.tex[i+1].cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK;
1668 rmesa->hw.tex[i+1].cmd[TEX_PP_TXFORMAT] |= 0x08000000;
1671 if ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (R200_TEX_1_ENABLE << i)) &&
1672 (rmesa->hw.tex[i+1].cmd[TEX_PP_TXFORMAT] & 0x08000000)) {
1673 R200_STATECHANGE(rmesa, tex[i+1]);
1674 rmesa->hw.tex[i+1].cmd[TEX_PP_TXFORMAT] &= ~0x08000000;
1680 * Texture cache LRU hang workaround -------------
1681 * not needed for r200 derivatives
1682 * hopefully this covers first pass of a shader as well
1685 /* While the cases below attempt to only enable the workaround in the
1686 * specific cases necessary, they were insufficient. See bugzilla #1519,
1687 * #729, #814. Tests with quake3 showed no impact on performance.
1692 if (((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (R200_TEX_0_ENABLE )) &&
1693 ((((rmesa->hw.tex[0].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1695 ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & R200_TEX_2_ENABLE) &&
1696 ((((rmesa->hw.tex[2].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1698 ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & R200_TEX_4_ENABLE) &&
1699 ((((rmesa->hw.tex[4].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1705 if (((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (R200_TEX_1_ENABLE )) &&
1706 ((((rmesa->hw.tex[1].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1708 ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & R200_TEX_3_ENABLE) &&
1709 ((((rmesa->hw.tex[3].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1711 ((rmesa->hw.ctx.cmd[CTX_PP_CNTL] & R200_TEX_5_ENABLE) &&
1712 ((((rmesa->hw.tex[5].cmd[TEX_PP_TXFILTER] & R200_MIN_FILTER_MASK)) &
1718 if (dbg
!= rmesa
->hw
.tam
.cmd
[TAM_DEBUG3
]) {
1719 R200_STATECHANGE( rmesa
, tam
);
1720 rmesa
->hw
.tam
.cmd
[TAM_DEBUG3
] = dbg
;
1721 if (0) printf("TEXCACHE LRU HANG WORKAROUND %x\n", dbg
);