2 * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2010 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 /* r300_render: Vertex and index buffer primitive emission. Contains both
25 * HW TCL fastpath rendering, and SW TCL Draw-assisted rendering. */
27 #include "draw/draw_context.h"
28 #include "draw/draw_vbuf.h"
30 #include "util/u_inlines.h"
32 #include "util/u_format.h"
33 #include "util/u_memory.h"
34 #include "util/u_upload_mgr.h"
35 #include "util/u_prim.h"
38 #include "r300_context.h"
39 #include "r300_screen_buffer.h"
40 #include "r300_emit.h"
42 #include "r300_state_derived.h"
46 #define IMMD_DWORDS 32
48 static uint32_t r300_translate_primitive(unsigned prim
)
51 case PIPE_PRIM_POINTS
:
52 return R300_VAP_VF_CNTL__PRIM_POINTS
;
54 return R300_VAP_VF_CNTL__PRIM_LINES
;
55 case PIPE_PRIM_LINE_LOOP
:
56 return R300_VAP_VF_CNTL__PRIM_LINE_LOOP
;
57 case PIPE_PRIM_LINE_STRIP
:
58 return R300_VAP_VF_CNTL__PRIM_LINE_STRIP
;
59 case PIPE_PRIM_TRIANGLES
:
60 return R300_VAP_VF_CNTL__PRIM_TRIANGLES
;
61 case PIPE_PRIM_TRIANGLE_STRIP
:
62 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP
;
63 case PIPE_PRIM_TRIANGLE_FAN
:
64 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN
;
66 return R300_VAP_VF_CNTL__PRIM_QUADS
;
67 case PIPE_PRIM_QUAD_STRIP
:
68 return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP
;
69 case PIPE_PRIM_POLYGON
:
70 return R300_VAP_VF_CNTL__PRIM_POLYGON
;
76 static uint32_t r300_provoking_vertex_fixes(struct r300_context
*r300
,
79 struct r300_rs_state
* rs
= (struct r300_rs_state
*)r300
->rs_state
.state
;
80 uint32_t color_control
= rs
->color_control
;
82 /* By default (see r300_state.c:r300_create_rs_state) color_control is
83 * initialized to provoking the first vertex.
85 * Triangle fans must be reduced to the second vertex, not the first, in
86 * Gallium flatshade-first mode, as per the GL spec.
87 * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
89 * Quads never provoke correctly in flatshade-first mode. The first
90 * vertex is never considered as provoking, so only the second, third,
91 * and fourth vertices can be selected, and both "third" and "last" modes
92 * select the fourth vertex. This is probably due to D3D lacking quads.
94 * Similarly, polygons reduce to the first, not the last, vertex, when in
95 * "last" mode, and all other modes start from the second vertex.
100 if (rs
->rs
.flatshade_first
) {
102 case PIPE_PRIM_TRIANGLE_FAN
:
103 color_control
|= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND
;
105 case PIPE_PRIM_QUADS
:
106 case PIPE_PRIM_QUAD_STRIP
:
107 case PIPE_PRIM_POLYGON
:
108 color_control
|= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST
;
111 color_control
|= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST
;
115 color_control
|= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST
;
118 return color_control
;
121 boolean
r500_index_bias_supported(struct r300_context
*r300
)
123 return r300
->screen
->caps
.is_r500
&&
124 r300
->rws
->get_value(r300
->rws
, R300_VID_DRM_2_3_0
);
127 void r500_emit_index_bias(struct r300_context
*r300
, int index_bias
)
132 OUT_CS_REG(R500_VAP_INDEX_OFFSET
,
133 (index_bias
& 0xFFFFFF) | (index_bias
< 0 ? 1<<24 : 0));
137 /* This function splits the index bias value into two parts:
138 * - buffer_offset: the value that can be safely added to buffer offsets
139 * in r300_emit_aos (it must yield a positive offset when added to
140 * a vertex buffer offset)
141 * - index_offset: the value that must be manually subtracted from indices
142 * in an index buffer to achieve negative offsets. */
143 static void r300_split_index_bias(struct r300_context
*r300
, int index_bias
,
144 int *buffer_offset
, int *index_offset
)
146 struct pipe_vertex_buffer
*vb
, *vbufs
= r300
->vertex_buffer
;
147 struct pipe_vertex_element
*velem
= r300
->velems
->velem
;
151 if (index_bias
< 0) {
152 /* See how large index bias we may subtract. We must be careful
153 * here because negative buffer offsets are not allowed
155 max_neg_bias
= INT_MAX
;
156 for (i
= 0; i
< r300
->velems
->count
; i
++) {
157 vb
= &vbufs
[velem
[i
].vertex_buffer_index
];
158 size
= (vb
->buffer_offset
+ velem
[i
].src_offset
) / vb
->stride
;
159 max_neg_bias
= MIN2(max_neg_bias
, size
);
162 /* Now set the minimum allowed value. */
163 *buffer_offset
= MAX2(-max_neg_bias
, index_bias
);
165 /* A positive index bias is OK. */
166 *buffer_offset
= index_bias
;
169 *index_offset
= index_bias
- *buffer_offset
;
172 enum r300_prepare_flags
{
173 PREP_FIRST_DRAW
= (1 << 0), /* call emit_dirty_state and friends? */
174 PREP_VALIDATE_VBOS
= (1 << 1), /* validate VBOs? */
175 PREP_EMIT_AOS
= (1 << 2), /* call emit_aos? */
176 PREP_EMIT_AOS_SWTCL
= (1 << 3), /* call emit_aos_swtcl? */
177 PREP_INDEXED
= (1 << 4) /* is this draw_elements? */
181 * Check if the requested number of dwords is available in the CS and
183 * \param r300 The context.
184 * \param flags See r300_prepare_flags.
185 * \param cs_dwords The number of dwords to reserve in CS.
186 * \return TRUE if the CS was flushed
188 static boolean
r300_reserve_cs_dwords(struct r300_context
*r300
,
189 enum r300_prepare_flags flags
,
192 boolean flushed
= FALSE
;
193 boolean first_draw
= flags
& PREP_FIRST_DRAW
;
194 boolean emit_aos
= flags
& PREP_EMIT_AOS
;
195 boolean emit_aos_swtcl
= flags
& PREP_EMIT_AOS_SWTCL
;
196 boolean hw_index_bias
= r500_index_bias_supported(r300
);
198 /* Add dirty state, index offset, and AOS. */
200 cs_dwords
+= r300_get_num_dirty_dwords(r300
);
203 cs_dwords
+= 2; /* emit_index_offset */
206 cs_dwords
+= 55; /* emit_aos */
209 cs_dwords
+= 7; /* emit_aos_swtcl */
212 cs_dwords
+= r300_get_num_cs_end_dwords(r300
);
214 /* Reserve requested CS space. */
215 if (cs_dwords
> (r300
->cs
->ndw
- r300
->cs
->cdw
)) {
216 r300
->context
.flush(&r300
->context
, 0, NULL
);
224 * Validate buffers and emit dirty state.
225 * \param r300 The context.
226 * \param flags See r300_prepare_flags.
227 * \param index_buffer The index buffer to validate. The parameter may be NULL.
228 * \param aos_offset The offset passed to emit_aos.
229 * \param index_bias The index bias to emit.
230 * \return TRUE if rendering should be skipped
232 static boolean
r300_emit_states(struct r300_context
*r300
,
233 enum r300_prepare_flags flags
,
234 struct pipe_resource
*index_buffer
,
238 boolean first_draw
= flags
& PREP_FIRST_DRAW
;
239 boolean emit_aos
= flags
& PREP_EMIT_AOS
;
240 boolean emit_aos_swtcl
= flags
& PREP_EMIT_AOS_SWTCL
;
241 boolean indexed
= flags
& PREP_INDEXED
;
242 boolean hw_index_bias
= r500_index_bias_supported(r300
);
244 /* Validate buffers and emit dirty state if needed. */
246 if (!r300_emit_buffer_validate(r300
, flags
& PREP_VALIDATE_VBOS
,
248 fprintf(stderr
, "r300: CS space validation failed. "
249 "(not enough memory?) Skipping rendering.\n");
253 r300_emit_dirty_state(r300
);
255 if (r300
->screen
->caps
.has_tcl
)
256 r500_emit_index_bias(r300
, index_bias
);
258 r500_emit_index_bias(r300
, 0);
262 r300_emit_aos(r300
, aos_offset
, indexed
);
265 r300_emit_aos_swtcl(r300
, indexed
);
272 * Check if the requested number of dwords is available in the CS and
273 * if not, flush. Then validate buffers and emit dirty state.
274 * \param r300 The context.
275 * \param flags See r300_prepare_flags.
276 * \param index_buffer The index buffer to validate. The parameter may be NULL.
277 * \param cs_dwords The number of dwords to reserve in CS.
278 * \param aos_offset The offset passed to emit_aos.
279 * \param index_bias The index bias to emit.
280 * \return TRUE if rendering should be skipped
282 static boolean
r300_prepare_for_rendering(struct r300_context
*r300
,
283 enum r300_prepare_flags flags
,
284 struct pipe_resource
*index_buffer
,
289 if (r300_reserve_cs_dwords(r300
, flags
, cs_dwords
))
290 flags
|= PREP_FIRST_DRAW
;
292 return r300_emit_states(r300
, flags
, index_buffer
, aos_offset
, index_bias
);
295 static boolean
immd_is_good_idea(struct r300_context
*r300
,
298 struct pipe_vertex_element
* velem
;
299 struct pipe_vertex_buffer
* vbuf
;
300 boolean checked
[PIPE_MAX_ATTRIBS
] = {0};
301 unsigned vertex_element_count
= r300
->velems
->count
;
304 if (DBG_ON(r300
, DBG_NO_IMMD
)) {
312 if (count
* r300
->velems
->vertex_size_dwords
> IMMD_DWORDS
) {
316 /* We shouldn't map buffers referenced by CS, busy buffers,
317 * and ones placed in VRAM. */
318 for (i
= 0; i
< vertex_element_count
; i
++) {
319 velem
= &r300
->velems
->velem
[i
];
320 vbi
= velem
->vertex_buffer_index
;
323 vbuf
= &r300
->vertex_buffer
[vbi
];
325 if (!(r300_buffer(vbuf
->buffer
)->domain
& R300_DOMAIN_GTT
)) {
329 if (r300_buffer_is_referenced(&r300
->context
,
331 R300_REF_CS
| R300_REF_HW
)) {
332 /* It's a very bad idea to map it... */
341 /*****************************************************************************
342 * The HWTCL draw functions. *
343 ****************************************************************************/
345 static void r300_emit_draw_arrays_immediate(struct r300_context
*r300
,
350 struct pipe_vertex_element
* velem
;
351 struct pipe_vertex_buffer
* vbuf
;
352 unsigned vertex_element_count
= r300
->velems
->count
;
355 /* Size of the vertex, in dwords. */
356 unsigned vertex_size
= r300
->velems
->vertex_size_dwords
;
358 /* The number of dwords for this draw operation. */
359 unsigned dwords
= 9 + count
* vertex_size
;
361 /* Size of the vertex element, in dwords. */
362 unsigned size
[PIPE_MAX_ATTRIBS
];
364 /* Stride to the same attrib in the next vertex in the vertex buffer,
366 unsigned stride
[PIPE_MAX_ATTRIBS
];
368 /* Mapped vertex buffers. */
369 uint32_t* map
[PIPE_MAX_ATTRIBS
];
370 uint32_t* mapelem
[PIPE_MAX_ATTRIBS
];
371 struct pipe_transfer
* transfer
[PIPE_MAX_ATTRIBS
] = {0};
375 if (!r300_prepare_for_rendering(r300
, PREP_FIRST_DRAW
, NULL
, dwords
, 0, 0))
378 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
379 for (i
= 0; i
< vertex_element_count
; i
++) {
380 velem
= &r300
->velems
->velem
[i
];
381 size
[i
] = r300
->velems
->hw_format_size
[i
] / 4;
382 vbi
= velem
->vertex_buffer_index
;
383 vbuf
= &r300
->vertex_buffer
[vbi
];
384 stride
[i
] = vbuf
->stride
/ 4;
386 /* Map the buffer. */
387 if (!transfer
[vbi
]) {
388 map
[vbi
] = (uint32_t*)pipe_buffer_map(&r300
->context
,
392 map
[vbi
] += (vbuf
->buffer_offset
/ 4) + stride
[i
] * start
;
394 mapelem
[i
] = map
[vbi
] + (velem
->src_offset
/ 4);
398 OUT_CS_REG(R300_GA_COLOR_CONTROL
,
399 r300_provoking_vertex_fixes(r300
, mode
));
400 OUT_CS_REG(R300_VAP_VTX_SIZE
, vertex_size
);
401 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX
, 2);
404 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2
, count
* vertex_size
);
405 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED
| (count
<< 16) |
406 r300_translate_primitive(mode
));
409 for (v
= 0; v
< count
; v
++) {
410 for (i
= 0; i
< vertex_element_count
; i
++) {
411 OUT_CS_TABLE(&mapelem
[i
][stride
[i
] * v
], size
[i
]);
417 for (i
= 0; i
< vertex_element_count
; i
++) {
418 vbi
= r300
->velems
->velem
[i
].vertex_buffer_index
;
421 vbuf
= &r300
->vertex_buffer
[vbi
];
422 pipe_buffer_unmap(&r300
->context
, vbuf
->buffer
, transfer
[vbi
]);
423 transfer
[vbi
] = NULL
;
428 static void r300_emit_draw_arrays(struct r300_context
*r300
,
432 boolean alt_num_verts
= count
> 65535;
435 if (count
>= (1 << 24)) {
436 fprintf(stderr
, "r300: Got a huge number of vertices: %i, "
437 "refusing to render.\n", count
);
441 BEGIN_CS(7 + (alt_num_verts
? 2 : 0));
443 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES
, count
);
445 OUT_CS_REG(R300_GA_COLOR_CONTROL
,
446 r300_provoking_vertex_fixes(r300
, mode
));
447 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX
, 2);
450 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2
, 0);
451 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST
| (count
<< 16) |
452 r300_translate_primitive(mode
) |
453 (alt_num_verts
? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS
: 0));
457 static void r300_emit_draw_elements(struct r300_context
*r300
,
458 struct pipe_resource
* indexBuffer
,
466 uint32_t count_dwords
;
467 uint32_t offset_dwords
= indexSize
* start
/ sizeof(uint32_t);
468 boolean alt_num_verts
= count
> 65535;
471 if (count
>= (1 << 24)) {
472 fprintf(stderr
, "r300: Got a huge number of vertices: %i, "
473 "refusing to render.\n", count
);
477 maxIndex
= MIN2(maxIndex
, r300
->vertex_buffer_max_index
);
479 DBG(r300
, DBG_DRAW
, "r300: Indexbuf of %u indices, min %u max %u\n",
480 count
, minIndex
, maxIndex
);
482 BEGIN_CS(13 + (alt_num_verts
? 2 : 0));
484 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES
, count
);
486 OUT_CS_REG(R300_GA_COLOR_CONTROL
,
487 r300_provoking_vertex_fixes(r300
, mode
));
488 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX
, 2);
491 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2
, 0);
492 if (indexSize
== 4) {
493 count_dwords
= count
;
494 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES
| (count
<< 16) |
495 R300_VAP_VF_CNTL__INDEX_SIZE_32bit
|
496 r300_translate_primitive(mode
) |
497 (alt_num_verts
? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS
: 0));
499 count_dwords
= (count
+ 1) / 2;
500 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES
| (count
<< 16) |
501 r300_translate_primitive(mode
) |
502 (alt_num_verts
? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS
: 0));
505 /* INDX_BUFFER is a truly special packet3.
506 * Unlike most other packet3, where the offset is after the count,
507 * the order is reversed, so the relocation ends up carrying the
508 * size of the indexbuf instead of the offset.
510 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER
, 2);
511 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR
| (R300_VAP_PORT_IDX0
>> 2) |
512 (0 << R300_INDX_BUFFER_SKIP_SHIFT
));
513 OUT_CS(offset_dwords
<< 2);
514 OUT_CS_BUF_RELOC(indexBuffer
, count_dwords
,
515 r300_buffer(indexBuffer
)->domain
, 0);
520 /* This is the fast-path drawing & emission for HW TCL. */
521 static void r300_draw_range_elements(struct pipe_context
* pipe
,
522 struct pipe_resource
* indexBuffer
,
531 struct r300_context
* r300
= r300_context(pipe
);
532 struct pipe_resource
* orgIndexBuffer
= indexBuffer
;
533 boolean alt_num_verts
= r300
->screen
->caps
.is_r500
&&
535 r300
->rws
->get_value(r300
->rws
, R300_VID_DRM_2_3_0
);
536 unsigned short_count
;
537 int buffer_offset
= 0, index_offset
= 0; /* for index bias emulation */
538 boolean translate
= FALSE
;
541 if (r300
->skip_rendering
) {
545 if (!u_trim_pipe_prim(mode
, &count
)) {
549 /* Index buffer range checking. */
550 if ((start
+ count
) * indexSize
> indexBuffer
->width0
) {
551 fprintf(stderr
, "r300: Invalid index buffer range. Skipping rendering.\n");
555 /* Set up fallback for incompatible vertex layout if needed. */
556 if (r300
->incompatible_vb_layout
|| r300
->velems
->incompatible_layout
) {
557 r300_begin_vertex_translate(r300
);
561 if (indexBias
&& !r500_index_bias_supported(r300
)) {
562 r300_split_index_bias(r300
, indexBias
, &buffer_offset
, &index_offset
);
565 r300_translate_index_buffer(r300
, &indexBuffer
, &indexSize
, index_offset
,
568 r300_update_derived_state(r300
);
569 r300_upload_index_buffer(r300
, &indexBuffer
, indexSize
, start
, count
, &new_offset
);
573 /* 15 dwords for emit_draw_elements. Give up if the function fails. */
574 if (!r300_prepare_for_rendering(r300
,
575 PREP_FIRST_DRAW
| PREP_VALIDATE_VBOS
| PREP_EMIT_AOS
|
576 PREP_INDEXED
, indexBuffer
, 15, buffer_offset
, indexBias
))
579 if (alt_num_verts
|| count
<= 65535) {
580 r300_emit_draw_elements(r300
, indexBuffer
, indexSize
,
581 minIndex
, maxIndex
, mode
, start
, count
);
584 short_count
= MIN2(count
, 65534);
585 r300_emit_draw_elements(r300
, indexBuffer
, indexSize
,
587 mode
, start
, short_count
);
589 start
+= short_count
;
590 count
-= short_count
;
592 /* 15 dwords for emit_draw_elements */
594 if (!r300_prepare_for_rendering(r300
,
595 PREP_VALIDATE_VBOS
| PREP_EMIT_AOS
| PREP_INDEXED
,
596 indexBuffer
, 15, buffer_offset
, indexBias
))
603 if (indexBuffer
!= orgIndexBuffer
) {
604 pipe_resource_reference( &indexBuffer
, NULL
);
608 r300_end_vertex_translate(r300
);
612 static void r300_draw_arrays(struct pipe_context
* pipe
, unsigned mode
,
613 unsigned start
, unsigned count
)
615 struct r300_context
* r300
= r300_context(pipe
);
616 boolean alt_num_verts
= r300
->screen
->caps
.is_r500
&&
618 r300
->rws
->get_value(r300
->rws
, R300_VID_DRM_2_3_0
);
619 unsigned short_count
;
620 boolean translate
= FALSE
;
622 if (r300
->skip_rendering
) {
626 if (!u_trim_pipe_prim(mode
, &count
)) {
630 /* Set up fallback for incompatible vertex layout if needed. */
631 if (r300
->incompatible_vb_layout
|| r300
->velems
->incompatible_layout
) {
632 r300_begin_vertex_translate(r300
);
636 r300_update_derived_state(r300
);
638 if (immd_is_good_idea(r300
, count
)) {
639 r300_emit_draw_arrays_immediate(r300
, mode
, start
, count
);
641 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
642 if (!r300_prepare_for_rendering(r300
,
643 PREP_FIRST_DRAW
| PREP_VALIDATE_VBOS
| PREP_EMIT_AOS
,
647 if (alt_num_verts
|| count
<= 65535) {
648 r300_emit_draw_arrays(r300
, mode
, count
);
651 short_count
= MIN2(count
, 65535);
652 r300_emit_draw_arrays(r300
, mode
, short_count
);
654 start
+= short_count
;
655 count
-= short_count
;
657 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
659 if (!r300_prepare_for_rendering(r300
,
660 PREP_VALIDATE_VBOS
| PREP_EMIT_AOS
, NULL
, 9,
670 r300_end_vertex_translate(r300
);
674 static void r300_draw_vbo(struct pipe_context
* pipe
,
675 const struct pipe_draw_info
*info
)
677 struct r300_context
* r300
= r300_context(pipe
);
679 if (!r300
->velems
->count
|| !r300
->vertex_buffer_count
)
682 if (info
->indexed
&& r300
->index_buffer
.buffer
) {
685 assert(r300
->index_buffer
.offset
% r300
->index_buffer
.index_size
== 0);
686 offset
= r300
->index_buffer
.offset
/ r300
->index_buffer
.index_size
;
688 r300_draw_range_elements(pipe
,
689 r300
->index_buffer
.buffer
,
690 r300
->index_buffer
.index_size
,
695 info
->start
+ offset
,
699 r300_draw_arrays(pipe
,
706 /****************************************************************************
707 * The rest of this file is for SW TCL rendering only. Please be polite and *
708 * keep these functions separated so that they are easier to locate. ~C. *
709 ***************************************************************************/
711 /* SW TCL elements, using Draw. */
712 static void r300_swtcl_draw_vbo(struct pipe_context
* pipe
,
713 const struct pipe_draw_info
*info
)
715 struct r300_context
* r300
= r300_context(pipe
);
716 struct pipe_transfer
*vb_transfer
[PIPE_MAX_ATTRIBS
];
717 struct pipe_transfer
*ib_transfer
= NULL
;
718 unsigned count
= info
->count
;
720 void *indices
= NULL
;
721 boolean indexed
= info
->indexed
&& r300
->index_buffer
.buffer
;
723 if (r300
->skip_rendering
) {
727 if (!u_trim_pipe_prim(info
->mode
, &count
)) {
731 r300_update_derived_state(r300
);
733 r300_reserve_cs_dwords(r300
,
734 PREP_FIRST_DRAW
| PREP_EMIT_AOS_SWTCL
|
735 (indexed
? PREP_INDEXED
: 0),
738 for (i
= 0; i
< r300
->vertex_buffer_count
; i
++) {
739 if (r300
->vertex_buffer
[i
].buffer
) {
740 void *buf
= pipe_buffer_map(pipe
,
741 r300
->vertex_buffer
[i
].buffer
,
744 draw_set_mapped_vertex_buffer(r300
->draw
, i
, buf
);
749 indices
= pipe_buffer_map(pipe
, r300
->index_buffer
.buffer
,
750 PIPE_TRANSFER_READ
, &ib_transfer
);
753 draw_set_mapped_index_buffer(r300
->draw
, indices
);
755 r300
->draw_vbo_locked
= TRUE
;
756 draw_vbo(r300
->draw
, info
);
757 draw_flush(r300
->draw
);
758 r300
->draw_vbo_locked
= FALSE
;
760 for (i
= 0; i
< r300
->vertex_buffer_count
; i
++) {
761 if (r300
->vertex_buffer
[i
].buffer
) {
762 pipe_buffer_unmap(pipe
, r300
->vertex_buffer
[i
].buffer
,
764 draw_set_mapped_vertex_buffer(r300
->draw
, i
, NULL
);
769 pipe_buffer_unmap(pipe
, r300
->index_buffer
.buffer
, ib_transfer
);
770 draw_set_mapped_index_buffer(r300
->draw
, NULL
);
774 /* Object for rendering using Draw. */
777 struct vbuf_render base
;
780 struct r300_context
* r300
;
782 /* Vertex information */
791 struct pipe_transfer
*vbo_transfer
;
794 static INLINE
struct r300_render
*
795 r300_render(struct vbuf_render
* render
)
797 return (struct r300_render
*)render
;
800 static const struct vertex_info
*
801 r300_render_get_vertex_info(struct vbuf_render
* render
)
803 struct r300_render
* r300render
= r300_render(render
);
804 struct r300_context
* r300
= r300render
->r300
;
806 return &r300
->vertex_info
;
809 static boolean
r300_render_allocate_vertices(struct vbuf_render
* render
,
813 struct r300_render
* r300render
= r300_render(render
);
814 struct r300_context
* r300
= r300render
->r300
;
815 struct pipe_screen
* screen
= r300
->context
.screen
;
816 size_t size
= (size_t)vertex_size
* (size_t)count
;
818 DBG(r300
, DBG_DRAW
, "r300: render_allocate_vertices (size: %d)\n", size
);
820 if (size
+ r300
->draw_vbo_offset
> r300
->draw_vbo_size
)
822 pipe_resource_reference(&r300
->vbo
, NULL
);
823 r300
->vbo
= pipe_buffer_create(screen
,
824 PIPE_BIND_VERTEX_BUFFER
,
825 R300_MAX_DRAW_VBO_SIZE
);
826 r300
->draw_vbo_offset
= 0;
827 r300
->draw_vbo_size
= R300_MAX_DRAW_VBO_SIZE
;
830 r300render
->vertex_size
= vertex_size
;
832 return (r300
->vbo
) ? TRUE
: FALSE
;
835 static void* r300_render_map_vertices(struct vbuf_render
* render
)
837 struct r300_render
* r300render
= r300_render(render
);
838 struct r300_context
* r300
= r300render
->r300
;
840 assert(!r300render
->vbo_transfer
);
842 DBG(r300
, DBG_DRAW
, "r300: render_map_vertices\n");
844 r300render
->vbo_ptr
= pipe_buffer_map(&r300render
->r300
->context
,
847 &r300render
->vbo_transfer
);
849 assert(r300render
->vbo_ptr
);
851 return ((uint8_t*)r300render
->vbo_ptr
+ r300
->draw_vbo_offset
);
854 static void r300_render_unmap_vertices(struct vbuf_render
* render
,
858 struct r300_render
* r300render
= r300_render(render
);
859 struct pipe_context
* context
= &r300render
->r300
->context
;
860 struct r300_context
* r300
= r300render
->r300
;
862 assert(r300render
->vbo_transfer
);
864 DBG(r300
, DBG_DRAW
, "r300: render_unmap_vertices\n");
866 r300render
->vbo_max_used
= MAX2(r300render
->vbo_max_used
,
867 r300render
->vertex_size
* (max
+ 1));
868 pipe_buffer_unmap(context
, r300
->vbo
, r300render
->vbo_transfer
);
870 r300render
->vbo_transfer
= NULL
;
873 static void r300_render_release_vertices(struct vbuf_render
* render
)
875 struct r300_render
* r300render
= r300_render(render
);
876 struct r300_context
* r300
= r300render
->r300
;
878 DBG(r300
, DBG_DRAW
, "r300: render_release_vertices\n");
880 r300
->draw_vbo_offset
+= r300render
->vbo_max_used
;
881 r300render
->vbo_max_used
= 0;
884 static boolean
r300_render_set_primitive(struct vbuf_render
* render
,
887 struct r300_render
* r300render
= r300_render(render
);
889 r300render
->prim
= prim
;
890 r300render
->hwprim
= r300_translate_primitive(prim
);
895 static void r300_render_draw_arrays(struct vbuf_render
* render
,
899 struct r300_render
* r300render
= r300_render(render
);
900 struct r300_context
* r300
= r300render
->r300
;
906 (void) i
; (void) ptr
;
908 DBG(r300
, DBG_DRAW
, "r300: render_draw_arrays (count: %d)\n", count
);
910 if (!r300_emit_states(r300
,
911 PREP_FIRST_DRAW
| PREP_EMIT_AOS_SWTCL
,
915 /* Uncomment to dump all VBOs rendered through this interface.
917 ptr = pipe_buffer_map(&r300render->r300->context,
918 r300render->vbo, PIPE_TRANSFER_READ,
919 &r300render->vbo_transfer);
921 for (i = 0; i < count; i++) {
922 printf("r300: Vertex %d\n", i);
923 draw_dump_emitted_vertex(&r300->vertex_info, ptr);
924 ptr += r300->vertex_info.size * 4;
928 pipe_buffer_unmap(&r300render->r300->context, r300render->vbo,
929 r300render->vbo_transfer);
933 OUT_CS_REG(R300_GA_COLOR_CONTROL
,
934 r300_provoking_vertex_fixes(r300
, r300render
->prim
));
935 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX
, count
- 1);
936 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2
, 0);
937 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST
| (count
<< 16) |
942 static void r300_render_draw_elements(struct vbuf_render
* render
,
943 const ushort
* indices
,
946 struct r300_render
* r300render
= r300_render(render
);
947 struct r300_context
* r300
= r300render
->r300
;
949 unsigned end_cs_dwords
;
950 unsigned max_index
= (r300
->draw_vbo_size
- r300
->draw_vbo_offset
) /
951 (r300render
->r300
->vertex_info
.size
* 4) - 1;
952 unsigned short_count
;
953 unsigned free_dwords
;
956 DBG(r300
, DBG_DRAW
, "r300: render_draw_elements (count: %d)\n", count
);
958 if (!r300_emit_states(r300
,
959 PREP_FIRST_DRAW
| PREP_EMIT_AOS_SWTCL
| PREP_INDEXED
,
963 /* Below we manage the CS space manually because there may be more
964 * indices than it can fit in CS. */
966 end_cs_dwords
= r300_get_num_cs_end_dwords(r300
);
969 free_dwords
= r300
->cs
->ndw
- r300
->cs
->cdw
;
971 short_count
= MIN2(count
, (free_dwords
- end_cs_dwords
- 6) * 2);
973 BEGIN_CS(6 + (short_count
+1)/2);
974 OUT_CS_REG(R300_GA_COLOR_CONTROL
,
975 r300_provoking_vertex_fixes(r300
, r300render
->prim
));
976 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX
, max_index
);
977 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2
, (short_count
+1)/2);
978 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES
| (short_count
<< 16) |
980 for (i
= 0; i
< short_count
-1; i
+= 2) {
981 OUT_CS(indices
[i
+1] << 16 | indices
[i
]);
983 if (short_count
% 2) {
984 OUT_CS(indices
[short_count
-1]);
988 /* OK now subtract the emitted indices and see if we need to emit
989 * another draw packet. */
990 indices
+= short_count
;
991 count
-= short_count
;
994 if (!r300_prepare_for_rendering(r300
,
995 PREP_EMIT_AOS_SWTCL
| PREP_INDEXED
,
999 end_cs_dwords
= r300_get_num_cs_end_dwords(r300
);
1004 static void r300_render_destroy(struct vbuf_render
* render
)
1009 static struct vbuf_render
* r300_render_create(struct r300_context
* r300
)
1011 struct r300_render
* r300render
= CALLOC_STRUCT(r300_render
);
1013 r300render
->r300
= r300
;
1015 /* XXX find real numbers plz */
1016 r300render
->base
.max_vertex_buffer_bytes
= 128 * 1024;
1017 r300render
->base
.max_indices
= 16 * 1024;
1019 r300render
->base
.get_vertex_info
= r300_render_get_vertex_info
;
1020 r300render
->base
.allocate_vertices
= r300_render_allocate_vertices
;
1021 r300render
->base
.map_vertices
= r300_render_map_vertices
;
1022 r300render
->base
.unmap_vertices
= r300_render_unmap_vertices
;
1023 r300render
->base
.set_primitive
= r300_render_set_primitive
;
1024 r300render
->base
.draw_elements
= r300_render_draw_elements
;
1025 r300render
->base
.draw_arrays
= r300_render_draw_arrays
;
1026 r300render
->base
.release_vertices
= r300_render_release_vertices
;
1027 r300render
->base
.destroy
= r300_render_destroy
;
1029 return &r300render
->base
;
1032 struct draw_stage
* r300_draw_stage(struct r300_context
* r300
)
1034 struct vbuf_render
* render
;
1035 struct draw_stage
* stage
;
1037 render
= r300_render_create(r300
);
1043 stage
= draw_vbuf_stage(r300
->draw
, render
);
1046 render
->destroy(render
);
1050 draw_set_render(r300
->draw
, render
);
1055 void r300_draw_flush_vbuf(struct r300_context
*r300
)
1057 pipe_resource_reference(&r300
->vbo
, NULL
);
1058 r300
->draw_vbo_size
= 0;
1061 /****************************************************************************
1062 * End of SW TCL functions *
1063 ***************************************************************************/
1065 /* This functions is used to draw a rectangle for the blitter module.
1067 * If we rendered a quad, the pixels on the main diagonal
1068 * would be computed and stored twice, which makes the clear/copy codepaths
1069 * somewhat inefficient. Instead we use a rectangular point sprite. */
1070 static void r300_blitter_draw_rectangle(struct blitter_context
*blitter
,
1071 unsigned x1
, unsigned y1
,
1072 unsigned x2
, unsigned y2
,
1074 enum blitter_attrib_type type
,
1075 const float attrib
[4])
1077 struct r300_context
*r300
= r300_context(util_blitter_get_pipe(blitter
));
1078 unsigned last_sprite_coord_enable
= r300
->sprite_coord_enable
;
1079 unsigned width
= x2
- x1
;
1080 unsigned height
= y2
- y1
;
1081 unsigned vertex_size
=
1082 type
== UTIL_BLITTER_ATTRIB_COLOR
|| !r300
->draw
? 8 : 4;
1083 unsigned dwords
= 13 + vertex_size
+
1084 (type
== UTIL_BLITTER_ATTRIB_TEXCOORD
? 7 : 0);
1085 const float zeros
[4] = {0, 0, 0, 0};
1088 if (type
== UTIL_BLITTER_ATTRIB_TEXCOORD
)
1089 r300
->sprite_coord_enable
= 1;
1091 r300_update_derived_state(r300
);
1093 /* Mark some states we don't care about as non-dirty. */
1094 r300
->clip_state
.dirty
= FALSE
;
1095 r300
->viewport_state
.dirty
= FALSE
;
1097 if (!r300_prepare_for_rendering(r300
, PREP_FIRST_DRAW
, NULL
, dwords
, 0, 0))
1100 DBG(r300
, DBG_DRAW
, "r300: draw_rectangle\n");
1104 OUT_CS_REG(R300_GA_POINT_SIZE
, (height
* 6) | ((width
* 6) << 16));
1106 if (type
== UTIL_BLITTER_ATTRIB_TEXCOORD
) {
1107 /* Set up the GA to generate texcoords. */
1108 OUT_CS_REG(R300_GB_ENABLE
, R300_GB_POINT_STUFF_ENABLE
|
1109 (R300_GB_TEX_STR
<< R300_GB_TEX0_SOURCE_SHIFT
));
1110 OUT_CS_REG_SEQ(R300_GA_POINT_S0
, 4);
1111 OUT_CS_32F(attrib
[0]);
1112 OUT_CS_32F(attrib
[3]);
1113 OUT_CS_32F(attrib
[2]);
1114 OUT_CS_32F(attrib
[1]);
1117 /* Set up VAP controls. */
1118 OUT_CS_REG(R300_VAP_CLIP_CNTL
, R300_CLIP_DISABLE
);
1119 OUT_CS_REG(R300_VAP_VTE_CNTL
, R300_VTX_XY_FMT
| R300_VTX_Z_FMT
);
1120 OUT_CS_REG(R300_VAP_VTX_SIZE
, vertex_size
);
1121 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX
, 2);
1126 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2
, vertex_size
);
1127 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED
| (1 << 16) |
1128 R300_VAP_VF_CNTL__PRIM_POINTS
);
1130 OUT_CS_32F(x1
+ width
* 0.5f
);
1131 OUT_CS_32F(y1
+ height
* 0.5f
);
1135 if (vertex_size
== 8) {
1138 OUT_CS_TABLE(attrib
, 4);
1143 /* Restore the state. */
1144 r300
->clip_state
.dirty
= TRUE
;
1145 r300
->rs_state
.dirty
= TRUE
;
1146 r300
->viewport_state
.dirty
= TRUE
;
1148 r300
->sprite_coord_enable
= last_sprite_coord_enable
;
1151 static void r300_resource_resolve(struct pipe_context
* pipe
,
1152 struct pipe_resource
* dest
,
1153 struct pipe_subresource subdest
,
1154 struct pipe_resource
* src
,
1155 struct pipe_subresource subsrc
)
1157 struct r300_context
* r300
= r300_context(pipe
);
1158 struct r300_aa_state
*aa
= (struct r300_aa_state
*)r300
->aa_state
.state
;
1159 struct pipe_surface
* srcsurf
= src
->screen
->get_tex_surface(src
->screen
,
1160 src
, subsrc
.face
, subsrc
.level
, 0, 0);
1161 float color
[] = {0, 0, 0, 0};
1163 DBG(r300
, DBG_DRAW
, "r300: Resolving resource...\n");
1165 /* Enable AA resolve. */
1166 aa
->dest
= r300_surface(
1167 dest
->screen
->get_tex_surface(dest
->screen
, dest
, subdest
.face
,
1168 subdest
.level
, 0, 0));
1171 R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE
|
1172 R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE
;
1173 r300
->aa_state
.size
= 12;
1174 r300
->aa_state
.dirty
= TRUE
;
1176 /* Resolve the surface. */
1177 r300
->context
.clear_render_target(pipe
,
1178 srcsurf
, color
, 0, 0, src
->width0
, src
->height0
);
1180 /* Disable AA resolve. */
1181 aa
->aaresolve_ctl
= 0;
1182 r300
->aa_state
.size
= 4;
1183 r300
->aa_state
.dirty
= TRUE
;
1185 pipe_surface_reference((struct pipe_surface
**)&srcsurf
, NULL
);
1186 pipe_surface_reference((struct pipe_surface
**)&aa
->dest
, NULL
);
1189 void r300_init_render_functions(struct r300_context
*r300
)
1191 /* Set draw functions based on presence of HW TCL. */
1192 if (r300
->screen
->caps
.has_tcl
) {
1193 r300
->context
.draw_vbo
= r300_draw_vbo
;
1195 r300
->context
.draw_vbo
= r300_swtcl_draw_vbo
;
1198 r300
->context
.resource_resolve
= r300_resource_resolve
;
1199 r300
->blitter
->draw_rectangle
= r300_blitter_draw_rectangle
;
1201 /* Plug in the two-sided stencil reference value fallback if needed. */
1202 if (!r300
->screen
->caps
.is_r500
)
1203 r300_plug_in_stencil_ref_fallback(r300
);