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"
29 #include "r300_context.h"
31 #include "r300_screen.h"
32 #include "r300_shader_semantics.h"
33 #include "r300_state_derived.h"
34 #include "r300_state_inlines.h"
37 /* r300_state_derived: Various bits of state which are dependent upon
38 * currently bound CSO data. */
40 static void r300_draw_emit_attrib(struct r300_context
* r300
,
41 enum attrib_emit emit
,
42 enum interp_mode interp
,
45 struct r300_vertex_shader
* vs
= r300
->vs_state
.state
;
46 struct tgsi_shader_info
* info
= &vs
->info
;
49 output
= draw_find_shader_output(r300
->draw
,
50 info
->output_semantic_name
[index
],
51 info
->output_semantic_index
[index
]);
52 draw_emit_vertex_attr(&r300
->vertex_info
, emit
, interp
, output
);
55 static void r300_draw_emit_all_attribs(struct r300_context
* r300
)
57 struct r300_vertex_shader
* vs
= r300
->vs_state
.state
;
58 struct r300_shader_semantics
* vs_outputs
= &vs
->outputs
;
62 if (vs_outputs
->pos
!= ATTR_UNUSED
) {
63 r300_draw_emit_attrib(r300
, EMIT_4F
, INTERP_PERSPECTIVE
,
70 if (vs_outputs
->psize
!= ATTR_UNUSED
) {
71 r300_draw_emit_attrib(r300
, EMIT_1F_PSIZE
, INTERP_POS
,
76 for (i
= 0; i
< ATTR_COLOR_COUNT
; i
++) {
77 if (vs_outputs
->color
[i
] != ATTR_UNUSED
) {
78 r300_draw_emit_attrib(r300
, EMIT_4F
, INTERP_LINEAR
,
79 vs_outputs
->color
[i
]);
83 /* XXX Back-face colors. */
85 /* Texture coordinates. */
87 for (i
= 0; i
< ATTR_GENERIC_COUNT
; i
++) {
88 if (vs_outputs
->generic
[i
] != ATTR_UNUSED
) {
89 r300_draw_emit_attrib(r300
, EMIT_4F
, INTERP_PERSPECTIVE
,
90 vs_outputs
->generic
[i
]);
95 /* Fog coordinates. */
96 if (vs_outputs
->fog
!= ATTR_UNUSED
) {
97 r300_draw_emit_attrib(r300
, EMIT_4F
, INTERP_PERSPECTIVE
,
103 assert(gen_count
<= 8);
106 /* Update the PSC tables for SW TCL, using Draw. */
107 static void r300_swtcl_vertex_psc(struct r300_context
*r300
)
109 struct r300_vertex_stream_state
*vstream
= r300
->vertex_stream_state
.state
;
110 struct r300_vertex_shader
* vs
= r300
->vs_state
.state
;
111 struct vertex_info
* vinfo
= &r300
->vertex_info
;
112 uint16_t type
, swizzle
;
113 enum pipe_format format
;
114 unsigned i
, attrib_count
;
115 int* vs_output_tab
= vs
->stream_loc_notcl
;
118 memset(vstream
, 0, sizeof(struct r300_vertex_stream_state
));
120 /* For each Draw attribute, route it to the fragment shader according
121 * to the vs_output_tab. */
122 attrib_count
= vinfo
->num_attribs
;
123 DBG(r300
, DBG_DRAW
, "r300: attrib count: %d\n", attrib_count
);
124 for (i
= 0; i
< attrib_count
; i
++) {
125 DBG(r300
, DBG_DRAW
, "r300: attrib: offset %d, interp %d, size %d,"
126 " vs_output_tab %d\n", vinfo
->attrib
[i
].src_index
,
127 vinfo
->attrib
[i
].interp_mode
, vinfo
->attrib
[i
].emit
,
130 /* Make sure we have a proper destination for our attribute. */
131 assert(vs_output_tab
[i
] != -1);
133 format
= draw_translate_vinfo_format(vinfo
->attrib
[i
].emit
);
135 /* Obtain the type of data in this attribute. */
136 type
= r300_translate_vertex_data_type(format
) |
137 vs_output_tab
[i
] << R300_DST_VEC_LOC_SHIFT
;
139 /* Obtain the swizzle for this attribute. Note that the default
140 * swizzle in the hardware is not XYZW! */
141 swizzle
= r300_translate_vertex_data_swizzle(format
);
143 /* Add the attribute to the PSC table. */
145 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
<< 16;
146 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
<< 16;
148 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
;
149 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
;
153 /* Set the last vector in the PSC. */
157 vstream
->vap_prog_stream_cntl
[i
>> 1] |=
158 (R300_LAST_VEC
<< (i
& 1 ? 16 : 0));
160 vstream
->count
= (i
>> 1) + 1;
161 r300
->vertex_stream_state
.dirty
= TRUE
;
162 r300
->vertex_stream_state
.size
= (1 + vstream
->count
) * 2;
165 static void r300_rs_col(struct r300_rs_block
* rs
, int id
, int ptr
,
166 boolean swizzle_0001
)
168 rs
->ip
[id
] |= R300_RS_COL_PTR(ptr
);
170 rs
->ip
[id
] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001
);
172 rs
->ip
[id
] |= R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA
);
174 rs
->inst
[id
] |= R300_RS_INST_COL_ID(id
);
177 static void r300_rs_col_write(struct r300_rs_block
* rs
, int id
, int fp_offset
)
179 rs
->inst
[id
] |= R300_RS_INST_COL_CN_WRITE
|
180 R300_RS_INST_COL_ADDR(fp_offset
);
183 static void r300_rs_tex(struct r300_rs_block
* rs
, int id
, int ptr
,
184 boolean swizzle_X001
)
187 rs
->ip
[id
] |= R300_RS_TEX_PTR(ptr
*4) |
188 R300_RS_SEL_S(R300_RS_SEL_C0
) |
189 R300_RS_SEL_T(R300_RS_SEL_K0
) |
190 R300_RS_SEL_R(R300_RS_SEL_K0
) |
191 R300_RS_SEL_Q(R300_RS_SEL_K1
);
193 rs
->ip
[id
] |= R300_RS_TEX_PTR(ptr
*4) |
194 R300_RS_SEL_S(R300_RS_SEL_C0
) |
195 R300_RS_SEL_T(R300_RS_SEL_C1
) |
196 R300_RS_SEL_R(R300_RS_SEL_C2
) |
197 R300_RS_SEL_Q(R300_RS_SEL_C3
);
199 rs
->inst
[id
] |= R300_RS_INST_TEX_ID(id
);
202 static void r300_rs_tex_write(struct r300_rs_block
* rs
, int id
, int fp_offset
)
204 rs
->inst
[id
] |= R300_RS_INST_TEX_CN_WRITE
|
205 R300_RS_INST_TEX_ADDR(fp_offset
);
208 static void r500_rs_col(struct r300_rs_block
* rs
, int id
, int ptr
,
209 boolean swizzle_0001
)
211 rs
->ip
[id
] |= R500_RS_COL_PTR(ptr
);
213 rs
->ip
[id
] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001
);
215 rs
->ip
[id
] |= R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA
);
217 rs
->inst
[id
] |= R500_RS_INST_COL_ID(id
);
220 static void r500_rs_col_write(struct r300_rs_block
* rs
, int id
, int fp_offset
)
222 rs
->inst
[id
] |= R500_RS_INST_COL_CN_WRITE
|
223 R500_RS_INST_COL_ADDR(fp_offset
);
226 static void r500_rs_tex(struct r300_rs_block
* rs
, int id
, int ptr
,
227 boolean swizzle_X001
)
229 int rs_tex_comp
= ptr
*4;
232 rs
->ip
[id
] |= R500_RS_SEL_S(rs_tex_comp
) |
233 R500_RS_SEL_T(R500_RS_IP_PTR_K0
) |
234 R500_RS_SEL_R(R500_RS_IP_PTR_K0
) |
235 R500_RS_SEL_Q(R500_RS_IP_PTR_K1
);
237 rs
->ip
[id
] |= R500_RS_SEL_S(rs_tex_comp
) |
238 R500_RS_SEL_T(rs_tex_comp
+ 1) |
239 R500_RS_SEL_R(rs_tex_comp
+ 2) |
240 R500_RS_SEL_Q(rs_tex_comp
+ 3);
242 rs
->inst
[id
] |= R500_RS_INST_TEX_ID(id
);
245 static void r500_rs_tex_write(struct r300_rs_block
* rs
, int id
, int fp_offset
)
247 rs
->inst
[id
] |= R500_RS_INST_TEX_CN_WRITE
|
248 R500_RS_INST_TEX_ADDR(fp_offset
);
251 /* Set up the RS block.
253 * This is the part of the chipset that actually does the rasterization
254 * of vertices into fragments. This is also the part of the chipset that
255 * locks up if any part of it is even slightly wrong. */
256 static void r300_update_rs_block(struct r300_context
* r300
,
257 struct r300_shader_semantics
* vs_outputs
,
258 struct r300_shader_semantics
* fs_inputs
)
260 struct r300_rs_block rs
= { { 0 } };
261 int i
, col_count
= 0, tex_count
= 0, fp_offset
= 0, count
;
262 void (*rX00_rs_col
)(struct r300_rs_block
*, int, int, boolean
);
263 void (*rX00_rs_col_write
)(struct r300_rs_block
*, int, int);
264 void (*rX00_rs_tex
)(struct r300_rs_block
*, int, int, boolean
);
265 void (*rX00_rs_tex_write
)(struct r300_rs_block
*, int, int);
266 boolean any_bcolor_used
= vs_outputs
->bcolor
[0] != ATTR_UNUSED
||
267 vs_outputs
->bcolor
[1] != ATTR_UNUSED
;
269 if (r300_screen(r300
->context
.screen
)->caps
->is_r500
) {
270 rX00_rs_col
= r500_rs_col
;
271 rX00_rs_col_write
= r500_rs_col_write
;
272 rX00_rs_tex
= r500_rs_tex
;
273 rX00_rs_tex_write
= r500_rs_tex_write
;
275 rX00_rs_col
= r300_rs_col
;
276 rX00_rs_col_write
= r300_rs_col_write
;
277 rX00_rs_tex
= r300_rs_tex
;
278 rX00_rs_tex_write
= r300_rs_tex_write
;
281 /* Rasterize colors. */
282 for (i
= 0; i
< ATTR_COLOR_COUNT
; i
++) {
283 if (vs_outputs
->color
[i
] != ATTR_UNUSED
|| any_bcolor_used
||
284 vs_outputs
->color
[1] != ATTR_UNUSED
) {
285 /* Always rasterize if it's written by the VS,
286 * otherwise it locks up. */
287 rX00_rs_col(&rs
, col_count
, i
, FALSE
);
289 /* Write it to the FS input register if it's used by the FS. */
290 if (fs_inputs
->color
[i
] != ATTR_UNUSED
) {
291 rX00_rs_col_write(&rs
, col_count
, fp_offset
);
296 /* Skip the FS input register, leave it uninitialized. */
297 /* If we try to set it to (0,0,0,1), it will lock up. */
298 if (fs_inputs
->color
[i
] != ATTR_UNUSED
) {
304 /* Rasterize texture coordinates. */
305 for (i
= 0; i
< ATTR_GENERIC_COUNT
; i
++) {
306 if (vs_outputs
->generic
[i
] != ATTR_UNUSED
) {
307 /* Always rasterize if it's written by the VS,
308 * otherwise it locks up. */
309 rX00_rs_tex(&rs
, tex_count
, tex_count
, FALSE
);
311 /* Write it to the FS input register if it's used by the FS. */
312 if (fs_inputs
->generic
[i
] != ATTR_UNUSED
) {
313 rX00_rs_tex_write(&rs
, tex_count
, fp_offset
);
318 /* Skip the FS input register, leave it uninitialized. */
319 /* If we try to set it to (0,0,0,1), it will lock up. */
320 if (fs_inputs
->generic
[i
] != ATTR_UNUSED
) {
326 /* Rasterize fog coordinates. */
327 if (vs_outputs
->fog
!= ATTR_UNUSED
) {
328 /* Always rasterize if it's written by the VS,
329 * otherwise it locks up. */
330 rX00_rs_tex(&rs
, tex_count
, tex_count
, TRUE
);
332 /* Write it to the FS input register if it's used by the FS. */
333 if (fs_inputs
->fog
!= ATTR_UNUSED
) {
334 rX00_rs_tex_write(&rs
, tex_count
, fp_offset
);
339 /* Skip the FS input register, leave it uninitialized. */
340 /* If we try to set it to (0,0,0,1), it will lock up. */
341 if (fs_inputs
->fog
!= ATTR_UNUSED
) {
346 /* Rasterize WPOS. */
347 /* If the FS doesn't need it, it's not written by the VS. */
348 if (fs_inputs
->wpos
!= ATTR_UNUSED
) {
349 rX00_rs_tex(&rs
, tex_count
, tex_count
, FALSE
);
350 rX00_rs_tex_write(&rs
, tex_count
, fp_offset
);
356 /* Rasterize at least one color, or bad things happen. */
357 if (col_count
== 0 && tex_count
== 0) {
358 rX00_rs_col(&rs
, 0, 0, TRUE
);
362 rs
.count
= (tex_count
*4) | (col_count
<< R300_IC_COUNT_SHIFT
) |
365 count
= MAX3(col_count
, tex_count
, 1);
366 rs
.inst_count
= count
- 1;
368 /* Now, after all that, see if we actually need to update the state. */
369 if (memcmp(r300
->rs_block_state
.state
, &rs
, sizeof(struct r300_rs_block
))) {
370 memcpy(r300
->rs_block_state
.state
, &rs
, sizeof(struct r300_rs_block
));
371 r300
->rs_block_state
.size
= 5 + count
*2;
375 /* Update the shader-dependant states. */
376 static void r300_update_derived_shader_state(struct r300_context
* r300
)
378 struct r300_vertex_shader
* vs
= r300
->vs_state
.state
;
380 r300_update_rs_block(r300
, &vs
->outputs
, &r300
->fs
->inputs
);
383 static boolean
r300_dsa_writes_depth_stencil(struct r300_dsa_state
* dsa
)
385 /* We are interested only in the cases when a new depth or stencil value
386 * can be written and changed. */
388 /* We might optionally check for [Z func: never] and inspect the stencil
389 * state in a similar fashion, but it's not terribly important. */
390 return (dsa
->z_buffer_control
& R300_Z_WRITE_ENABLE
) ||
391 (dsa
->stencil_ref_mask
& R300_STENCILWRITEMASK_MASK
) ||
392 ((dsa
->z_buffer_control
& R500_STENCIL_REFMASK_FRONT_BACK
) &&
393 (dsa
->stencil_ref_bf
& R300_STENCILWRITEMASK_MASK
));
396 static boolean
r300_dsa_alpha_test_enabled(struct r300_dsa_state
* dsa
)
398 /* We are interested only in the cases when alpha testing can kill
400 uint32_t af
= dsa
->alpha_function
;
402 return (af
& R300_FG_ALPHA_FUNC_ENABLE
) &&
403 (af
& R300_FG_ALPHA_FUNC_ALWAYS
) != R300_FG_ALPHA_FUNC_ALWAYS
;
406 static void r300_update_ztop(struct r300_context
* r300
)
408 struct r300_ztop_state
* ztop_state
=
409 (struct r300_ztop_state
*)r300
->ztop_state
.state
;
411 /* This is important enough that I felt it warranted a comment.
413 * According to the docs, these are the conditions where ZTOP must be
415 * 1) Alpha testing enabled
416 * 2) Texture kill instructions in fragment shader
417 * 3) Chroma key culling enabled
418 * 4) W-buffering enabled
420 * The docs claim that for the first three cases, if no ZS writes happen,
421 * then ZTOP can be used.
423 * (3) will never apply since we do not support chroma-keyed operations.
424 * (4) will need to be re-examined (and this comment updated) if/when
425 * Hyper-Z becomes supported.
427 * Additionally, the following conditions require disabled ZTOP:
428 * 5) Depth writes in fragment shader
429 * 6) Outstanding occlusion queries
431 * This register causes stalls all the way from SC to CB when changed,
432 * but it is buffered on-chip so it does not hurt to write it if it has
439 if (r300_dsa_writes_depth_stencil(r300
->dsa_state
.state
) &&
440 (r300_dsa_alpha_test_enabled(r300
->dsa_state
.state
) ||/* (1) */
441 r300
->fs
->info
.uses_kill
)) { /* (2) */
442 ztop_state
->z_buffer_top
= R300_ZTOP_DISABLE
;
443 } else if (r300_fragment_shader_writes_depth(r300
->fs
)) { /* (5) */
444 ztop_state
->z_buffer_top
= R300_ZTOP_DISABLE
;
445 } else if (r300
->query_current
) { /* (6) */
446 ztop_state
->z_buffer_top
= R300_ZTOP_DISABLE
;
448 ztop_state
->z_buffer_top
= R300_ZTOP_ENABLE
;
451 r300
->ztop_state
.dirty
= TRUE
;
454 static void r300_merge_textures_and_samplers(struct r300_context
* r300
)
456 struct r300_textures_state
*state
=
457 (struct r300_textures_state
*)r300
->textures_state
.state
;
458 struct r300_texture_sampler_state
*texstate
;
459 struct r300_sampler_state
*sampler
;
460 struct pipe_sampler_view
*view
;
461 struct r300_texture
*tex
;
462 unsigned min_level
, max_level
, i
, size
;
463 unsigned count
= MIN2(state
->texture_count
, state
->sampler_count
);
465 state
->tx_enable
= 0;
469 for (i
= 0; i
< count
; i
++) {
470 if (state
->fragment_sampler_views
[i
] && state
->sampler_states
[i
]) {
471 state
->tx_enable
|= 1 << i
;
473 view
= state
->fragment_sampler_views
[i
];
474 tex
= (struct r300_texture
*)view
->texture
;
475 sampler
= state
->sampler_states
[i
];
477 assert(view
->format
== tex
->tex
.format
);
479 texstate
= &state
->regs
[i
];
480 memcpy(texstate
->format
, &tex
->state
, sizeof(uint32_t)*3);
481 texstate
->filter
[0] = sampler
->filter0
;
482 texstate
->filter
[1] = sampler
->filter1
;
483 texstate
->border_color
= sampler
->border_color
;
484 texstate
->tile_config
= R300_TXO_MACRO_TILE(tex
->macrotile
) |
485 R300_TXO_MICRO_TILE(tex
->microtile
);
487 /* to emulate 1D textures through 2D ones correctly */
488 if (tex
->tex
.target
== PIPE_TEXTURE_1D
) {
489 texstate
->filter
[0] &= ~R300_TX_WRAP_T_MASK
;
490 texstate
->filter
[0] |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE
);
494 /* NPOT textures don't support mip filter, unfortunately.
495 * This prevents incorrect rendering. */
496 texstate
->filter
[0] &= ~R300_TX_MIN_FILTER_MIP_MASK
;
498 /* determine min/max levels */
499 /* the MAX_MIP level is the largest (finest) one */
500 max_level
= MIN3(sampler
->max_lod
+ view
->first_level
,
501 tex
->tex
.last_level
, view
->last_level
);
502 min_level
= MIN2(sampler
->min_lod
+ view
->first_level
,
504 texstate
->format
[0] |= R300_TX_NUM_LEVELS(max_level
);
505 texstate
->filter
[0] |= R300_TX_MAX_MIP_LEVEL(min_level
);
508 texstate
->filter
[0] |= i
<< 28;
515 r300
->textures_state
.size
= size
;
518 void r300_update_derived_state(struct r300_context
* r300
)
520 if (r300
->rs_block_state
.dirty
) {
521 r300_update_derived_shader_state(r300
);
524 if (r300
->textures_state
.dirty
) {
525 r300_merge_textures_and_samplers(r300
);
529 memset(&r300
->vertex_info
, 0, sizeof(struct vertex_info
));
530 r300_draw_emit_all_attribs(r300
);
531 draw_compute_vertex_size(&r300
->vertex_info
);
532 r300_swtcl_vertex_psc(r300
);
535 r300_update_ztop(r300
);