2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
24 #include "draw/draw_context.h"
26 #include "util/u_math.h"
27 #include "util/u_memory.h"
28 #include "util/u_pack_color.h"
30 #include "r300_context.h"
32 #include "r300_screen.h"
33 #include "r300_shader_semantics.h"
34 #include "r300_state_inlines.h"
35 #include "r300_texture.h"
38 /* r300_state_derived: Various bits of state which are dependent upon
39 * currently bound CSO data. */
41 enum r300_rs_swizzle
{
48 enum r300_rs_col_write_type
{
53 static void r300_draw_emit_attrib(struct r300_context
* r300
,
54 enum attrib_emit emit
,
55 enum interp_mode interp
,
58 struct r300_vertex_shader
* vs
= r300
->vs_state
.state
;
59 struct tgsi_shader_info
* info
= &vs
->info
;
62 output
= draw_find_shader_output(r300
->draw
,
63 info
->output_semantic_name
[index
],
64 info
->output_semantic_index
[index
]);
65 draw_emit_vertex_attr(&r300
->vertex_info
, emit
, interp
, output
);
68 static void r300_draw_emit_all_attribs(struct r300_context
* r300
)
70 struct r300_vertex_shader
* vs
= r300
->vs_state
.state
;
71 struct r300_shader_semantics
* vs_outputs
= &vs
->outputs
;
75 if (vs_outputs
->pos
!= ATTR_UNUSED
) {
76 r300_draw_emit_attrib(r300
, EMIT_4F
, INTERP_PERSPECTIVE
,
83 if (vs_outputs
->psize
!= ATTR_UNUSED
) {
84 r300_draw_emit_attrib(r300
, EMIT_1F_PSIZE
, INTERP_POS
,
89 for (i
= 0; i
< ATTR_COLOR_COUNT
; i
++) {
90 if (vs_outputs
->color
[i
] != ATTR_UNUSED
) {
91 r300_draw_emit_attrib(r300
, EMIT_4F
, INTERP_LINEAR
,
92 vs_outputs
->color
[i
]);
96 /* Back-face colors. */
97 for (i
= 0; i
< ATTR_COLOR_COUNT
; i
++) {
98 if (vs_outputs
->bcolor
[i
] != ATTR_UNUSED
) {
99 r300_draw_emit_attrib(r300
, EMIT_4F
, INTERP_LINEAR
,
100 vs_outputs
->bcolor
[i
]);
104 /* Texture coordinates. */
105 /* Only 8 generic vertex attributes can be used. If there are more,
106 * they won't be rasterized. */
108 for (i
= 0; i
< ATTR_GENERIC_COUNT
&& gen_count
< 8; i
++) {
109 if (vs_outputs
->generic
[i
] != ATTR_UNUSED
&&
110 !(r300
->sprite_coord_enable
& (1 << i
))) {
111 r300_draw_emit_attrib(r300
, EMIT_4F
, INTERP_PERSPECTIVE
,
112 vs_outputs
->generic
[i
]);
117 /* Fog coordinates. */
118 if (gen_count
< 8 && vs_outputs
->fog
!= ATTR_UNUSED
) {
119 r300_draw_emit_attrib(r300
, EMIT_4F
, INTERP_PERSPECTIVE
,
125 if (r300_fs(r300
)->shader
->inputs
.wpos
!= ATTR_UNUSED
&& gen_count
< 8) {
126 DBG(r300
, DBG_SWTCL
, "draw_emit_attrib: WPOS, index: %i\n",
128 r300_draw_emit_attrib(r300
, EMIT_4F
, INTERP_PERSPECTIVE
,
133 /* Update the PSC tables for SW TCL, using Draw. */
134 static void r300_swtcl_vertex_psc(struct r300_context
*r300
)
136 struct r300_vertex_stream_state
*vstream
= r300
->vertex_stream_state
.state
;
137 struct vertex_info
*vinfo
= &r300
->vertex_info
;
138 uint16_t type
, swizzle
;
139 enum pipe_format format
;
140 unsigned i
, attrib_count
;
141 int* vs_output_tab
= r300
->stream_loc_notcl
;
143 memset(vstream
, 0, sizeof(struct r300_vertex_stream_state
));
145 /* For each Draw attribute, route it to the fragment shader according
146 * to the vs_output_tab. */
147 attrib_count
= vinfo
->num_attribs
;
148 DBG(r300
, DBG_SWTCL
, "r300: attrib count: %d\n", attrib_count
);
149 for (i
= 0; i
< attrib_count
; i
++) {
150 if (vs_output_tab
[i
] == -1) {
155 format
= draw_translate_vinfo_format(vinfo
->attrib
[i
].emit
);
158 "r300: swtcl_vertex_psc [%i] <- %s\n",
159 vs_output_tab
[i
], util_format_short_name(format
));
161 /* Obtain the type of data in this attribute. */
162 type
= r300_translate_vertex_data_type(format
);
163 if (type
== R300_INVALID_FORMAT
) {
164 fprintf(stderr
, "r300: Bad vertex format %s.\n",
165 util_format_short_name(format
));
170 type
|= vs_output_tab
[i
] << R300_DST_VEC_LOC_SHIFT
;
172 /* Obtain the swizzle for this attribute. Note that the default
173 * swizzle in the hardware is not XYZW! */
174 swizzle
= r300_translate_vertex_data_swizzle(format
);
176 /* Add the attribute to the PSC table. */
178 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
<< 16;
179 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
<< 16;
181 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
;
182 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
;
186 /* Set the last vector in the PSC. */
190 vstream
->vap_prog_stream_cntl
[i
>> 1] |=
191 (R300_LAST_VEC
<< (i
& 1 ? 16 : 0));
193 vstream
->count
= (i
>> 1) + 1;
194 r300_mark_atom_dirty(r300
, &r300
->vertex_stream_state
);
195 r300
->vertex_stream_state
.size
= (1 + vstream
->count
) * 2;
198 static void r300_rs_col(struct r300_rs_block
* rs
, int id
, int ptr
,
199 enum r300_rs_swizzle swiz
)
201 rs
->ip
[id
] |= R300_RS_COL_PTR(ptr
);
202 if (swiz
== SWIZ_0001
) {
203 rs
->ip
[id
] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001
);
205 rs
->ip
[id
] |= R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA
);
207 rs
->inst
[id
] |= R300_RS_INST_COL_ID(id
);
210 static void r300_rs_col_write(struct r300_rs_block
* rs
, int id
, int fp_offset
,
211 enum r300_rs_col_write_type type
)
213 assert(type
== WRITE_COLOR
);
214 rs
->inst
[id
] |= R300_RS_INST_COL_CN_WRITE
|
215 R300_RS_INST_COL_ADDR(fp_offset
);
218 static void r300_rs_tex(struct r300_rs_block
* rs
, int id
, int ptr
,
219 enum r300_rs_swizzle swiz
)
221 if (swiz
== SWIZ_X001
) {
222 rs
->ip
[id
] |= R300_RS_TEX_PTR(ptr
) |
223 R300_RS_SEL_S(R300_RS_SEL_C0
) |
224 R300_RS_SEL_T(R300_RS_SEL_K0
) |
225 R300_RS_SEL_R(R300_RS_SEL_K0
) |
226 R300_RS_SEL_Q(R300_RS_SEL_K1
);
227 } else if (swiz
== SWIZ_XY01
) {
228 rs
->ip
[id
] |= R300_RS_TEX_PTR(ptr
) |
229 R300_RS_SEL_S(R300_RS_SEL_C0
) |
230 R300_RS_SEL_T(R300_RS_SEL_C1
) |
231 R300_RS_SEL_R(R300_RS_SEL_K0
) |
232 R300_RS_SEL_Q(R300_RS_SEL_K1
);
234 rs
->ip
[id
] |= R300_RS_TEX_PTR(ptr
) |
235 R300_RS_SEL_S(R300_RS_SEL_C0
) |
236 R300_RS_SEL_T(R300_RS_SEL_C1
) |
237 R300_RS_SEL_R(R300_RS_SEL_C2
) |
238 R300_RS_SEL_Q(R300_RS_SEL_C3
);
240 rs
->inst
[id
] |= R300_RS_INST_TEX_ID(id
);
243 static void r300_rs_tex_write(struct r300_rs_block
* rs
, int id
, int fp_offset
)
245 rs
->inst
[id
] |= R300_RS_INST_TEX_CN_WRITE
|
246 R300_RS_INST_TEX_ADDR(fp_offset
);
249 static void r500_rs_col(struct r300_rs_block
* rs
, int id
, int ptr
,
250 enum r300_rs_swizzle swiz
)
252 rs
->ip
[id
] |= R500_RS_COL_PTR(ptr
);
253 if (swiz
== SWIZ_0001
) {
254 rs
->ip
[id
] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001
);
256 rs
->ip
[id
] |= R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA
);
258 rs
->inst
[id
] |= R500_RS_INST_COL_ID(id
);
261 static void r500_rs_col_write(struct r300_rs_block
* rs
, int id
, int fp_offset
,
262 enum r300_rs_col_write_type type
)
264 if (type
== WRITE_FACE
)
265 rs
->inst
[id
] |= R500_RS_INST_COL_CN_WRITE_BACKFACE
|
266 R500_RS_INST_COL_ADDR(fp_offset
);
268 rs
->inst
[id
] |= R500_RS_INST_COL_CN_WRITE
|
269 R500_RS_INST_COL_ADDR(fp_offset
);
273 static void r500_rs_tex(struct r300_rs_block
* rs
, int id
, int ptr
,
274 enum r300_rs_swizzle swiz
)
276 if (swiz
== SWIZ_X001
) {
277 rs
->ip
[id
] |= R500_RS_SEL_S(ptr
) |
278 R500_RS_SEL_T(R500_RS_IP_PTR_K0
) |
279 R500_RS_SEL_R(R500_RS_IP_PTR_K0
) |
280 R500_RS_SEL_Q(R500_RS_IP_PTR_K1
);
281 } else if (swiz
== SWIZ_XY01
) {
282 rs
->ip
[id
] |= R500_RS_SEL_S(ptr
) |
283 R500_RS_SEL_T(ptr
+ 1) |
284 R500_RS_SEL_R(R500_RS_IP_PTR_K0
) |
285 R500_RS_SEL_Q(R500_RS_IP_PTR_K1
);
287 rs
->ip
[id
] |= R500_RS_SEL_S(ptr
) |
288 R500_RS_SEL_T(ptr
+ 1) |
289 R500_RS_SEL_R(ptr
+ 2) |
290 R500_RS_SEL_Q(ptr
+ 3);
292 rs
->inst
[id
] |= R500_RS_INST_TEX_ID(id
);
295 static void r500_rs_tex_write(struct r300_rs_block
* rs
, int id
, int fp_offset
)
297 rs
->inst
[id
] |= R500_RS_INST_TEX_CN_WRITE
|
298 R500_RS_INST_TEX_ADDR(fp_offset
);
301 /* Set up the RS block.
303 * This is the part of the chipset that is responsible for linking vertex
304 * and fragment shaders and stuffed texture coordinates.
306 * The rasterizer reads data from VAP, which produces vertex shader outputs,
307 * and GA, which produces stuffed texture coordinates. VAP outputs have
308 * precedence over GA. All outputs must be rasterized otherwise it locks up.
309 * If there are more outputs rasterized than is set in VAP/GA, it locks up
310 * too. The funky part is that this info has been pretty much obtained by trial
312 static void r300_update_rs_block(struct r300_context
*r300
)
314 struct r300_vertex_shader
*vs
= r300
->vs_state
.state
;
315 struct r300_shader_semantics
*vs_outputs
= &vs
->outputs
;
316 struct r300_shader_semantics
*fs_inputs
= &r300_fs(r300
)->shader
->inputs
;
317 struct r300_rs_block rs
= {0};
318 int i
, col_count
= 0, tex_count
= 0, fp_offset
= 0, count
, loc
= 0, tex_ptr
= 0;
320 void (*rX00_rs_col
)(struct r300_rs_block
*, int, int, enum r300_rs_swizzle
);
321 void (*rX00_rs_col_write
)(struct r300_rs_block
*, int, int, enum r300_rs_col_write_type
);
322 void (*rX00_rs_tex
)(struct r300_rs_block
*, int, int, enum r300_rs_swizzle
);
323 void (*rX00_rs_tex_write
)(struct r300_rs_block
*, int, int);
324 boolean any_bcolor_used
= vs_outputs
->bcolor
[0] != ATTR_UNUSED
||
325 vs_outputs
->bcolor
[1] != ATTR_UNUSED
;
326 int *stream_loc_notcl
= r300
->stream_loc_notcl
;
327 uint32_t stuffing_enable
= 0;
329 if (r300
->screen
->caps
.is_r500
) {
330 rX00_rs_col
= r500_rs_col
;
331 rX00_rs_col_write
= r500_rs_col_write
;
332 rX00_rs_tex
= r500_rs_tex
;
333 rX00_rs_tex_write
= r500_rs_tex_write
;
335 rX00_rs_col
= r300_rs_col
;
336 rX00_rs_col_write
= r300_rs_col_write
;
337 rX00_rs_tex
= r300_rs_tex
;
338 rX00_rs_tex_write
= r300_rs_tex_write
;
341 /* 0x5555 copied from classic, which means:
342 * Select user color 0 for COLOR0 up to COLOR7.
343 * What the hell does that mean? */
344 rs
.vap_vtx_state_cntl
= 0x5555;
346 /* The position is always present in VAP. */
347 rs
.vap_vsm_vtx_assm
|= R300_INPUT_CNTL_POS
;
348 rs
.vap_out_vtx_fmt
[0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT
;
349 stream_loc_notcl
[loc
++] = 0;
351 /* Set up the point size in VAP. */
352 if (vs_outputs
->psize
!= ATTR_UNUSED
) {
353 rs
.vap_out_vtx_fmt
[0] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT
;
354 stream_loc_notcl
[loc
++] = 1;
357 /* Set up and rasterize colors. */
358 for (i
= 0; i
< ATTR_COLOR_COUNT
; i
++) {
359 if (vs_outputs
->color
[i
] != ATTR_UNUSED
|| any_bcolor_used
||
360 vs_outputs
->color
[1] != ATTR_UNUSED
) {
361 /* Set up the color in VAP. */
362 rs
.vap_vsm_vtx_assm
|= R300_INPUT_CNTL_COLOR
;
363 rs
.vap_out_vtx_fmt
[0] |=
364 R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT
<< i
;
365 stream_loc_notcl
[loc
++] = 2 + i
;
368 rX00_rs_col(&rs
, col_count
, col_count
, SWIZ_XYZW
);
370 /* Write it to the FS input register if it's needed by the FS. */
371 if (fs_inputs
->color
[i
] != ATTR_UNUSED
) {
372 rX00_rs_col_write(&rs
, col_count
, fp_offset
, WRITE_COLOR
);
376 "r300: Rasterized color %i written to FS.\n", i
);
378 DBG(r300
, DBG_RS
, "r300: Rasterized color %i unused.\n", i
);
382 /* Skip the FS input register, leave it uninitialized. */
383 /* If we try to set it to (0,0,0,1), it will lock up. */
384 if (fs_inputs
->color
[i
] != ATTR_UNUSED
) {
387 DBG(r300
, DBG_RS
, "r300: FS input color %i unassigned%s.\n",
393 /* Set up back-face colors. The rasterizer will do the color selection
395 if (any_bcolor_used
) {
396 if (r300
->two_sided_color
) {
397 /* Rasterize as back-face colors. */
398 for (i
= 0; i
< ATTR_COLOR_COUNT
; i
++) {
399 rs
.vap_vsm_vtx_assm
|= R300_INPUT_CNTL_COLOR
;
400 rs
.vap_out_vtx_fmt
[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT
<< (2+i
);
401 stream_loc_notcl
[loc
++] = 4 + i
;
404 /* Rasterize two fake texcoords to prevent from the two-sided color
406 /* XXX Consider recompiling the vertex shader to save 2 RS units. */
407 for (i
= 0; i
< 2; i
++) {
408 rs
.vap_vsm_vtx_assm
|= (R300_INPUT_CNTL_TC0
<< tex_count
);
409 rs
.vap_out_vtx_fmt
[1] |= (4 << (3 * tex_count
));
410 stream_loc_notcl
[loc
++] = 6 + tex_count
;
413 rX00_rs_tex(&rs
, tex_count
, tex_ptr
, SWIZ_XYZW
);
421 * Note that we can use either the two-sided color selection based on
422 * the front and back vertex shader colors, or gl_FrontFacing,
423 * but not both! It locks up otherwise.
425 * In Direct3D 9, the two-sided color selection can be used
426 * with shaders 2.0 only, while gl_FrontFacing can be used
427 * with shaders 3.0 only. The hardware apparently hasn't been designed
428 * to support both at the same time. */
429 if (r300
->screen
->caps
.is_r500
&& fs_inputs
->face
!= ATTR_UNUSED
&&
430 !(any_bcolor_used
&& r300
->two_sided_color
)) {
431 rX00_rs_col(&rs
, col_count
, col_count
, SWIZ_XYZW
);
432 rX00_rs_col_write(&rs
, col_count
, fp_offset
, WRITE_FACE
);
435 DBG(r300
, DBG_RS
, "r300: Rasterized FACE written to FS.\n");
436 } else if (fs_inputs
->face
!= ATTR_UNUSED
) {
437 fprintf(stderr
, "r300: ERROR: FS input FACE unassigned.\n");
440 /* Re-use color varyings for texcoords if possible.
442 * The colors are interpolated as 20-bit floats (reduced precision),
443 * Use this hack only if there are too many generic varyings.
444 * (number of generic varyings + fog + wpos > 8) */
445 if (r300
->screen
->caps
.is_r500
&& !any_bcolor_used
&& !r300
->flatshade
&&
446 fs_inputs
->face
== ATTR_UNUSED
&&
447 vs_outputs
->num_generic
+ (vs_outputs
->fog
!= ATTR_UNUSED
) +
448 (fs_inputs
->wpos
!= ATTR_UNUSED
) > 8) {
449 for (i
= 0; i
< ATTR_GENERIC_COUNT
&& col_count
< 2; i
++) {
450 /* Cannot use color varyings for sprite coords. */
451 if (fs_inputs
->generic
[i
] != ATTR_UNUSED
&&
452 (r300
->sprite_coord_enable
& (1 << i
))) {
456 if (vs_outputs
->generic
[i
] != ATTR_UNUSED
) {
457 /* Set up the color in VAP. */
458 rs
.vap_vsm_vtx_assm
|= R300_INPUT_CNTL_COLOR
;
459 rs
.vap_out_vtx_fmt
[0] |=
460 R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT
<< col_count
;
461 stream_loc_notcl
[loc
++] = 2 + col_count
;
464 rX00_rs_col(&rs
, col_count
, col_count
, SWIZ_XYZW
);
466 /* Write it to the FS input register if it's needed by the FS. */
467 if (fs_inputs
->generic
[i
] != ATTR_UNUSED
) {
468 rX00_rs_col_write(&rs
, col_count
, fp_offset
, WRITE_COLOR
);
472 "r300: Rasterized generic %i redirected to color %i and written to FS.\n",
475 DBG(r300
, DBG_RS
, "r300: Rasterized generic %i redirected to color %i unused.\n",
480 /* Skip the FS input register, leave it uninitialized. */
481 /* If we try to set it to (0,0,0,1), it will lock up. */
482 if (fs_inputs
->generic
[i
] != ATTR_UNUSED
) {
485 DBG(r300
, DBG_RS
, "r300: FS input generic %i unassigned%s.\n", i
);
492 /* Rasterize texture coordinates. */
493 for (i
= gen_offset
; i
< ATTR_GENERIC_COUNT
&& tex_count
< 8; i
++) {
494 boolean sprite_coord
= false;
496 if (fs_inputs
->generic
[i
] != ATTR_UNUSED
) {
497 sprite_coord
= !!(r300
->sprite_coord_enable
& (1 << i
));
500 if (vs_outputs
->generic
[i
] != ATTR_UNUSED
|| sprite_coord
) {
502 /* Set up the texture coordinates in VAP. */
503 rs
.vap_vsm_vtx_assm
|= (R300_INPUT_CNTL_TC0
<< tex_count
);
504 rs
.vap_out_vtx_fmt
[1] |= (4 << (3 * tex_count
));
505 stream_loc_notcl
[loc
++] = 6 + tex_count
;
508 R300_GB_TEX_ST
<< (R300_GB_TEX0_SOURCE_SHIFT
+ (tex_count
*2));
511 rX00_rs_tex(&rs
, tex_count
, tex_ptr
,
512 sprite_coord
? SWIZ_XY01
: SWIZ_XYZW
);
514 /* Write it to the FS input register if it's needed by the FS. */
515 if (fs_inputs
->generic
[i
] != ATTR_UNUSED
) {
516 rX00_rs_tex_write(&rs
, tex_count
, fp_offset
);
520 "r300: Rasterized generic %i written to FS%s in texcoord %d.\n",
521 i
, sprite_coord
? " (sprite coord)" : "", tex_count
);
524 "r300: Rasterized generic %i unused%s.\n",
525 i
, sprite_coord
? " (sprite coord)" : "");
528 tex_ptr
+= sprite_coord
? 2 : 4;
530 /* Skip the FS input register, leave it uninitialized. */
531 /* If we try to set it to (0,0,0,1), it will lock up. */
532 if (fs_inputs
->generic
[i
] != ATTR_UNUSED
) {
535 DBG(r300
, DBG_RS
, "r300: FS input generic %i unassigned%s.\n",
536 i
, sprite_coord
? " (sprite coord)" : "");
541 for (; i
< ATTR_GENERIC_COUNT
; i
++) {
542 if (fs_inputs
->generic
[i
] != ATTR_UNUSED
) {
543 fprintf(stderr
, "r300: ERROR: FS input generic %i unassigned, "
544 "not enough hardware slots (it's not a bug, do not "
549 /* Rasterize fog coordinates. */
550 if (vs_outputs
->fog
!= ATTR_UNUSED
&& tex_count
< 8) {
551 /* Set up the fog coordinates in VAP. */
552 rs
.vap_vsm_vtx_assm
|= (R300_INPUT_CNTL_TC0
<< tex_count
);
553 rs
.vap_out_vtx_fmt
[1] |= (4 << (3 * tex_count
));
554 stream_loc_notcl
[loc
++] = 6 + tex_count
;
557 rX00_rs_tex(&rs
, tex_count
, tex_ptr
, SWIZ_X001
);
559 /* Write it to the FS input register if it's needed by the FS. */
560 if (fs_inputs
->fog
!= ATTR_UNUSED
) {
561 rX00_rs_tex_write(&rs
, tex_count
, fp_offset
);
564 DBG(r300
, DBG_RS
, "r300: Rasterized fog written to FS.\n");
566 DBG(r300
, DBG_RS
, "r300: Rasterized fog unused.\n");
571 /* Skip the FS input register, leave it uninitialized. */
572 /* If we try to set it to (0,0,0,1), it will lock up. */
573 if (fs_inputs
->fog
!= ATTR_UNUSED
) {
577 DBG(r300
, DBG_RS
, "r300: FS input fog unassigned.\n");
579 fprintf(stderr
, "r300: ERROR: FS input fog unassigned, "
580 "not enough hardware slots. (it's not a bug, "
581 "do not report it)\n");
586 /* Rasterize WPOS. */
587 /* Don't set it in VAP if the FS doesn't need it. */
588 if (fs_inputs
->wpos
!= ATTR_UNUSED
&& tex_count
< 8) {
589 /* Set up the WPOS coordinates in VAP. */
590 rs
.vap_vsm_vtx_assm
|= (R300_INPUT_CNTL_TC0
<< tex_count
);
591 rs
.vap_out_vtx_fmt
[1] |= (4 << (3 * tex_count
));
592 stream_loc_notcl
[loc
++] = 6 + tex_count
;
595 rX00_rs_tex(&rs
, tex_count
, tex_ptr
, SWIZ_XYZW
);
597 /* Write it to the FS input register. */
598 rX00_rs_tex_write(&rs
, tex_count
, fp_offset
);
600 DBG(r300
, DBG_RS
, "r300: Rasterized WPOS written to FS.\n");
606 if (fs_inputs
->wpos
!= ATTR_UNUSED
&& tex_count
>= 8) {
607 fprintf(stderr
, "r300: ERROR: FS input WPOS unassigned, "
608 "not enough hardware slots. (it's not a bug, do not "
613 /* Invalidate the rest of the no-TCL (GA) stream locations. */
615 stream_loc_notcl
[loc
++] = -1;
618 /* Rasterize at least one color, or bad things happen. */
619 if (col_count
== 0 && tex_count
== 0) {
620 rX00_rs_col(&rs
, 0, 0, SWIZ_0001
);
623 DBG(r300
, DBG_RS
, "r300: Rasterized color 0 to prevent lockups.\n");
626 DBG(r300
, DBG_RS
, "r300: --- Rasterizer status ---: colors: %i, "
627 "generics: %i.\n", col_count
, tex_count
);
629 rs
.count
= MIN2(tex_ptr
, 32) | (col_count
<< R300_IC_COUNT_SHIFT
) |
632 count
= MAX3(col_count
, tex_count
, 1);
633 rs
.inst_count
= count
- 1;
635 /* set the GB enable flags */
636 if (r300
->sprite_coord_enable
)
637 stuffing_enable
|= R300_GB_POINT_STUFF_ENABLE
;
639 rs
.gb_enable
= stuffing_enable
;
641 /* Now, after all that, see if we actually need to update the state. */
642 if (memcmp(r300
->rs_block_state
.state
, &rs
, sizeof(struct r300_rs_block
))) {
643 memcpy(r300
->rs_block_state
.state
, &rs
, sizeof(struct r300_rs_block
));
644 r300
->rs_block_state
.size
= 13 + count
*2;
648 static void rgba_to_bgra(float color
[4])
655 static uint32_t r300_get_border_color(enum pipe_format format
,
656 const float border
[4],
659 const struct util_format_description
*desc
;
660 float border_swizzled
[4] = {0};
661 union util_color uc
= {0};
663 desc
= util_format_description(format
);
665 /* Do depth formats first. */
666 if (util_format_is_depth_or_stencil(format
)) {
668 case PIPE_FORMAT_Z16_UNORM
:
669 return util_pack_z(PIPE_FORMAT_Z16_UNORM
, border
[0]);
670 case PIPE_FORMAT_X8Z24_UNORM
:
671 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
673 return util_pack_z(PIPE_FORMAT_X8Z24_UNORM
, border
[0]);
675 return util_pack_z(PIPE_FORMAT_Z16_UNORM
, border
[0]) << 16;
683 /* Apply inverse swizzle of the format. */
684 util_format_unswizzle_4f(border_swizzled
, border
, desc
->swizzle
);
686 /* Compressed formats. */
687 if (util_format_is_compressed(format
)) {
689 case PIPE_FORMAT_RGTC1_SNORM
:
690 case PIPE_FORMAT_LATC1_SNORM
:
691 border_swizzled
[0] = border_swizzled
[0] < 0 ?
692 border_swizzled
[0]*0.5+1 :
693 border_swizzled
[0]*0.5;
696 case PIPE_FORMAT_RGTC1_UNORM
:
697 case PIPE_FORMAT_LATC1_UNORM
:
698 /* Add 1/32 to round the border color instead of truncating. */
699 /* The Y component is used for the border color. */
700 border_swizzled
[1] = border_swizzled
[0] + 1.0f
/32;
701 util_pack_color(border_swizzled
, PIPE_FORMAT_B4G4R4A4_UNORM
, &uc
);
703 case PIPE_FORMAT_RGTC2_SNORM
:
704 case PIPE_FORMAT_LATC2_SNORM
:
705 util_pack_color(border_swizzled
, PIPE_FORMAT_R8G8B8A8_SNORM
, &uc
);
707 case PIPE_FORMAT_RGTC2_UNORM
:
708 case PIPE_FORMAT_LATC2_UNORM
:
709 util_pack_color(border_swizzled
, PIPE_FORMAT_R8G8B8A8_UNORM
, &uc
);
711 case PIPE_FORMAT_DXT1_SRGB
:
712 case PIPE_FORMAT_DXT1_SRGBA
:
713 case PIPE_FORMAT_DXT3_SRGBA
:
714 case PIPE_FORMAT_DXT5_SRGBA
:
715 util_pack_color(border_swizzled
, PIPE_FORMAT_B8G8R8A8_SRGB
, &uc
);
718 util_pack_color(border_swizzled
, PIPE_FORMAT_B8G8R8A8_UNORM
, &uc
);
723 switch (desc
->channel
[0].size
) {
725 rgba_to_bgra(border_swizzled
);
726 util_pack_color(border_swizzled
, PIPE_FORMAT_B2G3R3_UNORM
, &uc
);
730 rgba_to_bgra(border_swizzled
);
731 util_pack_color(border_swizzled
, PIPE_FORMAT_B4G4R4A4_UNORM
, &uc
);
735 rgba_to_bgra(border_swizzled
);
736 if (desc
->channel
[1].size
== 5) {
737 util_pack_color(border_swizzled
, PIPE_FORMAT_B5G5R5A1_UNORM
, &uc
);
738 } else if (desc
->channel
[1].size
== 6) {
739 util_pack_color(border_swizzled
, PIPE_FORMAT_B5G6R5_UNORM
, &uc
);
747 if (desc
->channel
[0].type
== UTIL_FORMAT_TYPE_SIGNED
) {
748 util_pack_color(border_swizzled
, PIPE_FORMAT_R8G8B8A8_SNORM
, &uc
);
749 } else if (desc
->colorspace
== UTIL_FORMAT_COLORSPACE_SRGB
) {
750 if (desc
->nr_channels
== 2) {
751 border_swizzled
[3] = border_swizzled
[1];
752 util_pack_color(border_swizzled
, PIPE_FORMAT_L8A8_SRGB
, &uc
);
754 util_pack_color(border_swizzled
, PIPE_FORMAT_R8G8B8A8_SRGB
, &uc
);
757 util_pack_color(border_swizzled
, PIPE_FORMAT_R8G8B8A8_UNORM
, &uc
);
762 util_pack_color(border_swizzled
, PIPE_FORMAT_R10G10B10A2_UNORM
, &uc
);
766 if (desc
->nr_channels
<= 2) {
767 if (desc
->channel
[0].type
== UTIL_FORMAT_TYPE_FLOAT
) {
768 util_pack_color(border_swizzled
, PIPE_FORMAT_R16G16_FLOAT
, &uc
);
769 } else if (desc
->channel
[0].type
== UTIL_FORMAT_TYPE_SIGNED
) {
770 util_pack_color(border_swizzled
, PIPE_FORMAT_R16G16_SNORM
, &uc
);
772 util_pack_color(border_swizzled
, PIPE_FORMAT_R16G16_UNORM
, &uc
);
775 if (desc
->channel
[0].type
== UTIL_FORMAT_TYPE_SIGNED
) {
776 util_pack_color(border_swizzled
, PIPE_FORMAT_R8G8B8A8_SNORM
, &uc
);
778 util_pack_color(border_swizzled
, PIPE_FORMAT_R8G8B8A8_UNORM
, &uc
);
784 if (desc
->nr_channels
== 1) {
785 util_pack_color(border_swizzled
, PIPE_FORMAT_R32_FLOAT
, &uc
);
787 util_pack_color(border_swizzled
, PIPE_FORMAT_R8G8B8A8_UNORM
, &uc
);
795 static void r300_merge_textures_and_samplers(struct r300_context
* r300
)
797 struct r300_textures_state
*state
=
798 (struct r300_textures_state
*)r300
->textures_state
.state
;
799 struct r300_texture_sampler_state
*texstate
;
800 struct r300_sampler_state
*sampler
;
801 struct r300_sampler_view
*view
;
802 struct r300_resource
*tex
;
803 unsigned base_level
, min_level
, level_count
, i
, j
, size
;
804 unsigned count
= MIN2(state
->sampler_view_count
,
805 state
->sampler_state_count
);
806 boolean has_us_format
= r300
->screen
->caps
.has_us_format
;
808 /* The KIL opcode fix, see below. */
809 if (!count
&& !r300
->screen
->caps
.is_r500
)
812 state
->tx_enable
= 0;
816 for (i
= 0; i
< count
; i
++) {
817 if (state
->sampler_views
[i
] && state
->sampler_states
[i
]) {
818 state
->tx_enable
|= 1 << i
;
820 view
= state
->sampler_views
[i
];
821 tex
= r300_resource(view
->base
.texture
);
822 sampler
= state
->sampler_states
[i
];
824 texstate
= &state
->regs
[i
];
825 texstate
->format
= view
->format
;
826 texstate
->filter0
= sampler
->filter0
;
827 texstate
->filter1
= sampler
->filter1
;
829 /* Set the border color. */
830 texstate
->border_color
=
831 r300_get_border_color(view
->base
.format
,
832 sampler
->state
.border_color
.f
,
833 r300
->screen
->caps
.is_r500
);
835 /* determine min/max levels */
836 base_level
= view
->base
.u
.tex
.first_level
;
837 min_level
= sampler
->min_lod
;
838 level_count
= MIN3(sampler
->max_lod
,
839 tex
->b
.b
.last_level
- base_level
,
840 view
->base
.u
.tex
.last_level
- base_level
);
842 if (base_level
+ min_level
) {
845 if (tex
->tex
.is_npot
) {
846 /* Even though we do not implement mipmapping for NPOT
847 * textures, we should at least honor the minimum level
848 * which is allowed to be displayed. We do this by setting up
849 * an i-th mipmap level as the zero level. */
850 base_level
+= min_level
;
852 offset
= tex
->tex
.offset_in_bytes
[base_level
];
854 r300_texture_setup_format_state(r300
->screen
, tex
,
857 view
->width0_override
,
858 view
->height0_override
,
860 texstate
->format
.tile_config
|= offset
& 0xffffffe0;
861 assert((offset
& 0x1f) == 0);
864 /* Assign a texture cache region. */
865 texstate
->format
.format1
|= view
->texcache_region
;
867 /* Depth textures are kinda special. */
868 if (util_format_is_depth_or_stencil(view
->base
.format
)) {
869 unsigned char depth_swizzle
[4];
871 if (!r300
->screen
->caps
.is_r500
&&
872 util_format_get_blocksizebits(view
->base
.format
) == 32) {
873 /* X24x8 is sampled as Y16X16 on r3xx-r4xx.
874 * The depth here is at the Y component. */
875 for (j
= 0; j
< 4; j
++)
876 depth_swizzle
[j
] = UTIL_FORMAT_SWIZZLE_Y
;
878 for (j
= 0; j
< 4; j
++)
879 depth_swizzle
[j
] = UTIL_FORMAT_SWIZZLE_X
;
882 /* If compare mode is disabled, sampler view swizzles
883 * are stored in the format.
884 * Otherwise, the swizzles must be applied after the compare
885 * mode in the fragment shader. */
886 if (sampler
->state
.compare_mode
== PIPE_TEX_COMPARE_NONE
) {
887 texstate
->format
.format1
|=
888 r300_get_swizzle_combined(depth_swizzle
,
889 view
->swizzle
, FALSE
);
891 texstate
->format
.format1
|=
892 r300_get_swizzle_combined(depth_swizzle
, 0, FALSE
);
896 if (r300
->screen
->caps
.dxtc_swizzle
&&
897 util_format_is_compressed(view
->base
.format
)) {
898 texstate
->filter1
|= R400_DXTC_SWIZZLE_ENABLE
;
901 /* to emulate 1D textures through 2D ones correctly */
902 if (tex
->b
.b
.target
== PIPE_TEXTURE_1D
) {
903 texstate
->filter0
&= ~R300_TX_WRAP_T_MASK
;
904 texstate
->filter0
|= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE
);
907 /* The hardware doesn't like CLAMP and CLAMP_TO_BORDER
908 * for the 3rd coordinate if the texture isn't 3D. */
909 if (tex
->b
.b
.target
!= PIPE_TEXTURE_3D
) {
910 texstate
->filter0
&= ~R300_TX_WRAP_R_MASK
;
913 if (tex
->tex
.is_npot
) {
914 /* NPOT textures don't support mip filter, unfortunately.
915 * This prevents incorrect rendering. */
916 texstate
->filter0
&= ~R300_TX_MIN_FILTER_MIP_MASK
;
918 /* Mask out the mirrored flag. */
919 if (texstate
->filter0
& R300_TX_WRAP_S(R300_TX_MIRRORED
)) {
920 texstate
->filter0
&= ~R300_TX_WRAP_S(R300_TX_MIRRORED
);
922 if (texstate
->filter0
& R300_TX_WRAP_T(R300_TX_MIRRORED
)) {
923 texstate
->filter0
&= ~R300_TX_WRAP_T(R300_TX_MIRRORED
);
926 /* Change repeat to clamp-to-edge.
927 * (the repeat bit has a value of 0, no masking needed). */
928 if ((texstate
->filter0
& R300_TX_WRAP_S_MASK
) ==
929 R300_TX_WRAP_S(R300_TX_REPEAT
)) {
930 texstate
->filter0
|= R300_TX_WRAP_S(R300_TX_CLAMP_TO_EDGE
);
932 if ((texstate
->filter0
& R300_TX_WRAP_T_MASK
) ==
933 R300_TX_WRAP_T(R300_TX_REPEAT
)) {
934 texstate
->filter0
|= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE
);
937 /* the MAX_MIP level is the largest (finest) one */
938 texstate
->format
.format0
|= R300_TX_NUM_LEVELS(level_count
);
939 texstate
->filter0
|= R300_TX_MAX_MIP_LEVEL(min_level
);
942 /* Float textures only support nearest and mip-nearest filtering. */
943 if (util_format_is_float(view
->base
.format
)) {
944 /* No MAG linear filtering. */
945 if ((texstate
->filter0
& R300_TX_MAG_FILTER_MASK
) ==
946 R300_TX_MAG_FILTER_LINEAR
) {
947 texstate
->filter0
&= ~R300_TX_MAG_FILTER_MASK
;
948 texstate
->filter0
|= R300_TX_MAG_FILTER_NEAREST
;
950 /* No MIN linear filtering. */
951 if ((texstate
->filter0
& R300_TX_MIN_FILTER_MASK
) ==
952 R300_TX_MIN_FILTER_LINEAR
) {
953 texstate
->filter0
&= ~R300_TX_MIN_FILTER_MASK
;
954 texstate
->filter0
|= R300_TX_MIN_FILTER_NEAREST
;
956 /* No mipmap linear filtering. */
957 if ((texstate
->filter0
& R300_TX_MIN_FILTER_MIP_MASK
) ==
958 R300_TX_MIN_FILTER_MIP_LINEAR
) {
959 texstate
->filter0
&= ~R300_TX_MIN_FILTER_MIP_MASK
;
960 texstate
->filter0
|= R300_TX_MIN_FILTER_MIP_NEAREST
;
962 /* No anisotropic filtering. */
963 texstate
->filter0
&= ~R300_TX_MAX_ANISO_MASK
;
964 texstate
->filter1
&= ~R500_TX_MAX_ANISO_MASK
;
965 texstate
->filter1
&= ~R500_TX_ANISO_HIGH_QUALITY
;
968 texstate
->filter0
|= i
<< 28;
970 size
+= 16 + (has_us_format
? 2 : 0);
973 /* For the KIL opcode to work on r3xx-r4xx, the texture unit
974 * assigned to this opcode (it's always the first one) must be
975 * enabled. Otherwise the opcode doesn't work.
977 * In order to not depend on the fragment shader, we just make
978 * the first unit enabled all the time. */
979 if (i
== 0 && !r300
->screen
->caps
.is_r500
) {
980 pipe_sampler_view_reference(
981 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
982 &r300
->texkill_sampler
->base
);
984 state
->tx_enable
|= 1 << i
;
986 texstate
= &state
->regs
[i
];
988 /* Just set some valid state. */
989 texstate
->format
= r300
->texkill_sampler
->format
;
991 r300_translate_tex_filters(PIPE_TEX_FILTER_NEAREST
,
992 PIPE_TEX_FILTER_NEAREST
,
993 PIPE_TEX_FILTER_NEAREST
,
995 texstate
->filter1
= 0;
996 texstate
->border_color
= 0;
998 texstate
->filter0
|= i
<< 28;
999 size
+= 16 + (has_us_format
? 2 : 0);
1005 r300
->textures_state
.size
= size
;
1007 /* Pick a fragment shader based on either the texture compare state
1008 * or the uses_pitch flag or some other external state. */
1010 r300
->fs_status
== FRAGMENT_SHADER_VALID
) {
1011 r300
->fs_status
= FRAGMENT_SHADER_MAYBE_DIRTY
;
1015 static void r300_decompress_depth_textures(struct r300_context
*r300
)
1017 struct r300_textures_state
*state
=
1018 (struct r300_textures_state
*)r300
->textures_state
.state
;
1019 struct pipe_resource
*tex
;
1020 unsigned count
= MIN2(state
->sampler_view_count
,
1021 state
->sampler_state_count
);
1024 if (!r300
->locked_zbuffer
) {
1028 for (i
= 0; i
< count
; i
++) {
1029 if (state
->sampler_views
[i
] && state
->sampler_states
[i
]) {
1030 tex
= state
->sampler_views
[i
]->base
.texture
;
1032 if (tex
== r300
->locked_zbuffer
->texture
) {
1033 r300_decompress_zmask_locked(r300
);
1040 static void r300_validate_fragment_shader(struct r300_context
*r300
)
1042 struct pipe_framebuffer_state
*fb
= r300
->fb_state
.state
;
1044 if (r300
->fs
.state
&& r300
->fs_status
!= FRAGMENT_SHADER_VALID
) {
1045 /* Pick the fragment shader based on external states.
1046 * Then mark the state dirty if the fragment shader is either dirty
1047 * or the function r300_pick_fragment_shader changed the shader. */
1048 if (r300_pick_fragment_shader(r300
) ||
1049 r300
->fs_status
== FRAGMENT_SHADER_DIRTY
) {
1050 /* Mark the state atom as dirty. */
1051 r300_mark_fs_code_dirty(r300
);
1053 /* Does Multiwrite need to be changed? */
1054 if (fb
->nr_cbufs
> 1) {
1055 boolean new_multiwrite
=
1056 r300_fragment_shader_writes_all(r300_fs(r300
));
1058 if (r300
->fb_multiwrite
!= new_multiwrite
) {
1059 r300
->fb_multiwrite
= new_multiwrite
;
1060 r300_mark_fb_state_dirty(r300
, R300_CHANGED_MULTIWRITE
);
1064 r300
->fs_status
= FRAGMENT_SHADER_VALID
;
1068 void r300_update_derived_state(struct r300_context
* r300
)
1070 if (r300
->textures_state
.dirty
) {
1071 r300_decompress_depth_textures(r300
);
1072 r300_merge_textures_and_samplers(r300
);
1075 r300_validate_fragment_shader(r300
);
1077 if (r300
->rs_block_state
.dirty
) {
1078 r300_update_rs_block(r300
);
1081 memset(&r300
->vertex_info
, 0, sizeof(struct vertex_info
));
1082 r300_draw_emit_all_attribs(r300
);
1083 draw_compute_vertex_size(&r300
->vertex_info
);
1084 r300_swtcl_vertex_psc(r300
);
1088 r300_update_hyperz_state(r300
);