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"
39 #include "r300_context.h"
40 #include "r300_screen_buffer.h"
41 #include "r300_emit.h"
43 #include "r300_state_derived.h"
47 #define IMMD_DWORDS 32
49 static uint32_t r300_translate_primitive(unsigned prim
)
52 case PIPE_PRIM_POINTS
:
53 return R300_VAP_VF_CNTL__PRIM_POINTS
;
55 return R300_VAP_VF_CNTL__PRIM_LINES
;
56 case PIPE_PRIM_LINE_LOOP
:
57 return R300_VAP_VF_CNTL__PRIM_LINE_LOOP
;
58 case PIPE_PRIM_LINE_STRIP
:
59 return R300_VAP_VF_CNTL__PRIM_LINE_STRIP
;
60 case PIPE_PRIM_TRIANGLES
:
61 return R300_VAP_VF_CNTL__PRIM_TRIANGLES
;
62 case PIPE_PRIM_TRIANGLE_STRIP
:
63 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP
;
64 case PIPE_PRIM_TRIANGLE_FAN
:
65 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN
;
67 return R300_VAP_VF_CNTL__PRIM_QUADS
;
68 case PIPE_PRIM_QUAD_STRIP
:
69 return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP
;
70 case PIPE_PRIM_POLYGON
:
71 return R300_VAP_VF_CNTL__PRIM_POLYGON
;
77 static uint32_t r300_provoking_vertex_fixes(struct r300_context
*r300
,
80 struct r300_rs_state
* rs
= (struct r300_rs_state
*)r300
->rs_state
.state
;
81 uint32_t color_control
= rs
->color_control
;
83 /* By default (see r300_state.c:r300_create_rs_state) color_control is
84 * initialized to provoking the first vertex.
86 * Triangle fans must be reduced to the second vertex, not the first, in
87 * Gallium flatshade-first mode, as per the GL spec.
88 * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
90 * Quads never provoke correctly in flatshade-first mode. The first
91 * vertex is never considered as provoking, so only the second, third,
92 * and fourth vertices can be selected, and both "third" and "last" modes
93 * select the fourth vertex. This is probably due to D3D lacking quads.
95 * Similarly, polygons reduce to the first, not the last, vertex, when in
96 * "last" mode, and all other modes start from the second vertex.
101 if (rs
->rs
.flatshade_first
) {
103 case PIPE_PRIM_TRIANGLE_FAN
:
104 color_control
|= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND
;
106 case PIPE_PRIM_QUADS
:
107 case PIPE_PRIM_QUAD_STRIP
:
108 case PIPE_PRIM_POLYGON
:
109 color_control
|= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST
;
112 color_control
|= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST
;
116 color_control
|= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST
;
119 return color_control
;
122 static boolean
index_bias_supported(struct r300_context
*r300
)
124 return r300
->screen
->caps
.is_r500
&&
125 r300
->rws
->get_value(r300
->rws
, R300_VID_DRM_2_3_0
);
128 static void r500_emit_index_bias(struct r300_context
*r300
, int index_bias
)
133 OUT_CS_REG(R500_VAP_INDEX_OFFSET
,
134 (index_bias
& 0xFFFFFF) | (index_bias
< 0 ? 1<<24 : 0));
138 /* This function splits the index bias value into two parts:
139 * - buffer_offset: the value that can be safely added to buffer offsets
140 * in r300_emit_aos (it must yield a positive offset when added to
141 * a vertex buffer offset)
142 * - index_offset: the value that must be manually subtracted from indices
143 * in an index buffer to achieve negative offsets. */
144 static void r300_split_index_bias(struct r300_context
*r300
, int index_bias
,
145 int *buffer_offset
, int *index_offset
)
147 struct pipe_vertex_buffer
*vb
, *vbufs
= r300
->vertex_buffer
;
148 struct pipe_vertex_element
*velem
= r300
->velems
->velem
;
152 if (index_bias
< 0) {
153 /* See how large index bias we may subtract. We must be careful
154 * here because negative buffer offsets are not allowed
156 max_neg_bias
= INT_MAX
;
157 for (i
= 0; i
< r300
->velems
->count
; i
++) {
158 vb
= &vbufs
[velem
[i
].vertex_buffer_index
];
159 size
= (vb
->buffer_offset
+ velem
[i
].src_offset
) / vb
->stride
;
160 max_neg_bias
= MIN2(max_neg_bias
, size
);
163 /* Now set the minimum allowed value. */
164 *buffer_offset
= MAX2(-max_neg_bias
, index_bias
);
166 /* A positive index bias is OK. */
167 *buffer_offset
= index_bias
;
170 *index_offset
= index_bias
- *buffer_offset
;
173 enum r300_prepare_flags
{
174 PREP_FIRST_DRAW
= (1 << 0), /* call emit_dirty_state and friends? */
175 PREP_VALIDATE_VBOS
= (1 << 1), /* validate VBOs? */
176 PREP_EMIT_AOS
= (1 << 2), /* call emit_aos? */
177 PREP_EMIT_AOS_SWTCL
= (1 << 3), /* call emit_aos_swtcl? */
178 PREP_INDEXED
= (1 << 4) /* is this draw_elements? */
182 * Check if the requested number of dwords is available in the CS and
183 * if not, flush. Then validate buffers and emit dirty state.
184 * \param r300 The context.
185 * \param flags See r300_prepare_flags.
186 * \param index_buffer The index buffer to validate. The parameter may be NULL.
187 * \param cs_dwords The number of dwords to reserve in CS.
188 * \param aos_offset The offset passed to emit_aos.
189 * \param index_bias The index bias to emit.
190 * \param end_cs_dwords The number of free dwords which must be available
191 * at the end of CS after drawing in case the CS space
192 * management is performed by a draw_* function manually.
193 * The parameter may be NULL.
195 static void r300_prepare_for_rendering(struct r300_context
*r300
,
196 enum r300_prepare_flags flags
,
197 struct pipe_resource
*index_buffer
,
201 unsigned *end_cs_dwords
)
203 unsigned end_dwords
= 0;
204 boolean flushed
= FALSE
;
205 boolean first_draw
= flags
& PREP_FIRST_DRAW
;
206 boolean emit_aos
= flags
& PREP_EMIT_AOS
;
207 boolean emit_aos_swtcl
= flags
& PREP_EMIT_AOS_SWTCL
;
208 boolean indexed
= flags
& PREP_INDEXED
;
209 boolean hw_index_bias
= index_bias_supported(r300
);
211 /* Add dirty state, index offset, and AOS. */
213 cs_dwords
+= r300_get_num_dirty_dwords(r300
);
216 cs_dwords
+= 2; /* emit_index_offset */
219 cs_dwords
+= 55; /* emit_aos */
222 cs_dwords
+= 7; /* emit_aos_swtcl */
225 /* Emitted in flush. */
226 end_dwords
+= 26; /* emit_query_end */
228 cs_dwords
+= end_dwords
;
230 /* Reserve requested CS space. */
231 if (!r300_check_cs(r300
, cs_dwords
)) {
232 r300
->context
.flush(&r300
->context
, 0, NULL
);
236 /* Validate buffers and emit dirty state if needed. */
237 if (first_draw
|| flushed
) {
238 r300_emit_buffer_validate(r300
, flags
& PREP_VALIDATE_VBOS
, index_buffer
);
239 r300_emit_dirty_state(r300
);
241 if (r300
->screen
->caps
.has_tcl
)
242 r500_emit_index_bias(r300
, index_bias
);
244 r500_emit_index_bias(r300
, 0);
248 r300_emit_aos(r300
, aos_offset
, indexed
);
251 r300_emit_aos_swtcl(r300
, indexed
);
255 *end_cs_dwords
= end_dwords
;
258 static boolean
immd_is_good_idea(struct r300_context
*r300
,
261 struct pipe_vertex_element
* velem
;
262 struct pipe_vertex_buffer
* vbuf
;
263 boolean checked
[PIPE_MAX_ATTRIBS
] = {0};
264 unsigned vertex_element_count
= r300
->velems
->count
;
267 if (DBG_ON(r300
, DBG_NO_IMMD
)) {
275 if (count
* r300
->velems
->vertex_size_dwords
> IMMD_DWORDS
) {
279 /* We shouldn't map buffers referenced by CS, busy buffers,
280 * and ones placed in VRAM. */
281 for (i
= 0; i
< vertex_element_count
; i
++) {
282 velem
= &r300
->velems
->velem
[i
];
283 vbi
= velem
->vertex_buffer_index
;
286 vbuf
= &r300
->vertex_buffer
[vbi
];
288 if (!(r300_buffer(vbuf
->buffer
)->domain
& R300_DOMAIN_GTT
)) {
292 if (r300_buffer_is_referenced(&r300
->context
,
294 R300_REF_CS
| R300_REF_HW
)) {
295 /* It's a very bad idea to map it... */
304 /*****************************************************************************
305 * The HWTCL draw functions. *
306 ****************************************************************************/
308 static void r300_emit_draw_arrays_immediate(struct r300_context
*r300
,
313 struct pipe_vertex_element
* velem
;
314 struct pipe_vertex_buffer
* vbuf
;
315 unsigned vertex_element_count
= r300
->velems
->count
;
316 unsigned i
, v
, vbi
, dwords
;
318 /* Size of the vertex, in dwords. */
319 unsigned vertex_size
= r300
->velems
->vertex_size_dwords
;
321 /* Size of the vertex element, in dwords. */
322 unsigned size
[PIPE_MAX_ATTRIBS
];
324 /* Stride to the same attrib in the next vertex in the vertex buffer,
326 unsigned stride
[PIPE_MAX_ATTRIBS
];
328 /* Mapped vertex buffers. */
329 uint32_t* map
[PIPE_MAX_ATTRIBS
];
330 uint32_t* mapelem
[PIPE_MAX_ATTRIBS
];
331 struct pipe_transfer
* transfer
[PIPE_MAX_ATTRIBS
] = {0};
335 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
336 for (i
= 0; i
< vertex_element_count
; i
++) {
337 velem
= &r300
->velems
->velem
[i
];
338 size
[i
] = r300
->velems
->hw_format_size
[i
] / 4;
339 vbi
= velem
->vertex_buffer_index
;
340 vbuf
= &r300
->vertex_buffer
[vbi
];
341 stride
[i
] = vbuf
->stride
/ 4;
343 /* Map the buffer. */
344 if (!transfer
[vbi
]) {
345 map
[vbi
] = (uint32_t*)pipe_buffer_map(&r300
->context
,
349 map
[vbi
] += (vbuf
->buffer_offset
/ 4) + stride
[i
] * start
;
351 mapelem
[i
] = map
[vbi
] + (velem
->src_offset
/ 4);
354 dwords
= 9 + count
* vertex_size
;
356 r300_prepare_for_rendering(r300
, PREP_FIRST_DRAW
, NULL
, dwords
, 0, 0, NULL
);
358 BEGIN_CS_AS_CB(r300
, dwords
);
359 OUT_CB_REG(R300_GA_COLOR_CONTROL
,
360 r300_provoking_vertex_fixes(r300
, mode
));
361 OUT_CB_REG(R300_VAP_VTX_SIZE
, vertex_size
);
362 OUT_CB_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX
, 2);
365 OUT_CB_PKT3(R300_PACKET3_3D_DRAW_IMMD_2
, count
* vertex_size
);
366 OUT_CB(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED
| (count
<< 16) |
367 r300_translate_primitive(mode
));
370 for (v
= 0; v
< count
; v
++) {
371 for (i
= 0; i
< vertex_element_count
; i
++) {
372 OUT_CB_TABLE(&mapelem
[i
][stride
[i
] * v
], size
[i
]);
378 for (i
= 0; i
< vertex_element_count
; i
++) {
379 vbi
= r300
->velems
->velem
[i
].vertex_buffer_index
;
382 vbuf
= &r300
->vertex_buffer
[vbi
];
383 pipe_buffer_unmap(&r300
->context
, vbuf
->buffer
, transfer
[vbi
]);
384 transfer
[vbi
] = NULL
;
389 static void r300_emit_draw_arrays(struct r300_context
*r300
,
393 boolean alt_num_verts
= count
> 65535;
396 if (count
>= (1 << 24)) {
397 fprintf(stderr
, "r300: Got a huge number of vertices: %i, "
398 "refusing to render.\n", count
);
402 BEGIN_CS(7 + (alt_num_verts
? 2 : 0));
404 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES
, count
);
406 OUT_CS_REG(R300_GA_COLOR_CONTROL
,
407 r300_provoking_vertex_fixes(r300
, mode
));
408 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX
, 2);
411 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2
, 0);
412 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST
| (count
<< 16) |
413 r300_translate_primitive(mode
) |
414 (alt_num_verts
? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS
: 0));
418 static void r300_emit_draw_elements(struct r300_context
*r300
,
419 struct pipe_resource
* indexBuffer
,
427 uint32_t count_dwords
;
428 uint32_t offset_dwords
= indexSize
* start
/ sizeof(uint32_t);
429 boolean alt_num_verts
= count
> 65535;
432 if (count
>= (1 << 24)) {
433 fprintf(stderr
, "r300: Got a huge number of vertices: %i, "
434 "refusing to render.\n", count
);
438 maxIndex
= MIN2(maxIndex
, r300
->vertex_buffer_max_index
);
440 DBG(r300
, DBG_DRAW
, "r300: Indexbuf of %u indices, min %u max %u\n",
441 count
, minIndex
, maxIndex
);
443 BEGIN_CS(13 + (alt_num_verts
? 2 : 0));
445 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES
, count
);
447 OUT_CS_REG(R300_GA_COLOR_CONTROL
,
448 r300_provoking_vertex_fixes(r300
, mode
));
449 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX
, 2);
452 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2
, 0);
453 if (indexSize
== 4) {
454 count_dwords
= count
;
455 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES
| (count
<< 16) |
456 R300_VAP_VF_CNTL__INDEX_SIZE_32bit
|
457 r300_translate_primitive(mode
) |
458 (alt_num_verts
? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS
: 0));
460 count_dwords
= (count
+ 1) / 2;
461 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES
| (count
<< 16) |
462 r300_translate_primitive(mode
) |
463 (alt_num_verts
? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS
: 0));
466 /* INDX_BUFFER is a truly special packet3.
467 * Unlike most other packet3, where the offset is after the count,
468 * the order is reversed, so the relocation ends up carrying the
469 * size of the indexbuf instead of the offset.
471 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER
, 2);
472 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR
| (R300_VAP_PORT_IDX0
>> 2) |
473 (0 << R300_INDX_BUFFER_SKIP_SHIFT
));
474 OUT_CS(offset_dwords
<< 2);
475 OUT_CS_BUF_RELOC(indexBuffer
, count_dwords
,
476 r300_buffer(indexBuffer
)->domain
, 0, 0);
481 /* This is the fast-path drawing & emission for HW TCL. */
482 static void r300_draw_range_elements(struct pipe_context
* pipe
,
483 struct pipe_resource
* indexBuffer
,
492 struct r300_context
* r300
= r300_context(pipe
);
493 struct pipe_resource
* orgIndexBuffer
= indexBuffer
;
494 boolean alt_num_verts
= r300
->screen
->caps
.is_r500
&&
496 r300
->rws
->get_value(r300
->rws
, R300_VID_DRM_2_3_0
);
497 unsigned short_count
;
498 int buffer_offset
= 0, index_offset
= 0; /* for index bias emulation */
499 boolean translate
= FALSE
;
501 if (r300
->skip_rendering
) {
505 if (!u_trim_pipe_prim(mode
, &count
)) {
509 /* Index buffer range checking. */
510 if ((start
+ count
) * indexSize
> indexBuffer
->width0
) {
511 fprintf(stderr
, "r300: Invalid index buffer range. Skipping rendering.\n");
515 /* Set up fallback for incompatible vertex layout if needed. */
516 if (r300
->incompatible_vb_layout
|| r300
->velems
->incompatible_layout
) {
517 r300_begin_vertex_translate(r300
);
521 if (indexBias
&& !index_bias_supported(r300
)) {
522 r300_split_index_bias(r300
, indexBias
, &buffer_offset
, &index_offset
);
525 r300_translate_index_buffer(r300
, &indexBuffer
, &indexSize
, index_offset
,
528 r300_update_derived_state(r300
);
529 r300_upload_index_buffer(r300
, &indexBuffer
, indexSize
, start
, count
);
531 /* 15 dwords for emit_draw_elements */
532 r300_prepare_for_rendering(r300
,
533 PREP_FIRST_DRAW
| PREP_VALIDATE_VBOS
| PREP_EMIT_AOS
| PREP_INDEXED
,
534 indexBuffer
, 15, buffer_offset
, indexBias
, NULL
);
536 u_upload_flush(r300
->upload_vb
);
537 u_upload_flush(r300
->upload_ib
);
538 if (alt_num_verts
|| count
<= 65535) {
539 r300_emit_draw_elements(r300
, indexBuffer
, indexSize
,
540 minIndex
, maxIndex
, mode
, start
, count
);
543 short_count
= MIN2(count
, 65534);
544 r300_emit_draw_elements(r300
, indexBuffer
, indexSize
,
546 mode
, start
, short_count
);
548 start
+= short_count
;
549 count
-= short_count
;
551 /* 15 dwords for emit_draw_elements */
553 r300_prepare_for_rendering(r300
,
554 PREP_VALIDATE_VBOS
| PREP_EMIT_AOS
| PREP_INDEXED
,
555 indexBuffer
, 15, buffer_offset
, indexBias
, NULL
);
560 if (indexBuffer
!= orgIndexBuffer
) {
561 pipe_resource_reference( &indexBuffer
, NULL
);
565 r300_end_vertex_translate(r300
);
569 /* Simple helpers for context setup. Should probably be moved to util. */
570 static void r300_draw_elements(struct pipe_context
* pipe
,
571 struct pipe_resource
* indexBuffer
,
572 unsigned indexSize
, int indexBias
, unsigned mode
,
573 unsigned start
, unsigned count
)
575 struct r300_context
*r300
= r300_context(pipe
);
577 pipe
->draw_range_elements(pipe
, indexBuffer
, indexSize
, indexBias
,
578 0, r300
->vertex_buffer_max_index
,
582 static void r300_draw_arrays(struct pipe_context
* pipe
, unsigned mode
,
583 unsigned start
, unsigned count
)
585 struct r300_context
* r300
= r300_context(pipe
);
586 boolean alt_num_verts
= r300
->screen
->caps
.is_r500
&&
588 r300
->rws
->get_value(r300
->rws
, R300_VID_DRM_2_3_0
);
589 unsigned short_count
;
590 boolean translate
= FALSE
;
592 if (r300
->skip_rendering
) {
596 if (!u_trim_pipe_prim(mode
, &count
)) {
600 /* Set up fallback for incompatible vertex layout if needed. */
601 if (r300
->incompatible_vb_layout
|| r300
->velems
->incompatible_layout
) {
602 r300_begin_vertex_translate(r300
);
606 r300_update_derived_state(r300
);
608 if (immd_is_good_idea(r300
, count
)) {
609 r300_emit_draw_arrays_immediate(r300
, mode
, start
, count
);
611 /* 9 spare dwords for emit_draw_arrays. */
612 r300_prepare_for_rendering(r300
, PREP_FIRST_DRAW
| PREP_VALIDATE_VBOS
| PREP_EMIT_AOS
,
613 NULL
, 9, start
, 0, NULL
);
615 if (alt_num_verts
|| count
<= 65535) {
616 r300_emit_draw_arrays(r300
, mode
, count
);
619 short_count
= MIN2(count
, 65535);
620 r300_emit_draw_arrays(r300
, mode
, short_count
);
622 start
+= short_count
;
623 count
-= short_count
;
625 /* 9 spare dwords for emit_draw_arrays. */
627 r300_prepare_for_rendering(r300
,
628 PREP_VALIDATE_VBOS
| PREP_EMIT_AOS
, NULL
, 9,
633 u_upload_flush(r300
->upload_vb
);
637 r300_end_vertex_translate(r300
);
641 /****************************************************************************
642 * The rest of this file is for SW TCL rendering only. Please be polite and *
643 * keep these functions separated so that they are easier to locate. ~C. *
644 ***************************************************************************/
646 /* SW TCL arrays, using Draw. */
647 static void r300_swtcl_draw_arrays(struct pipe_context
* pipe
,
652 struct r300_context
* r300
= r300_context(pipe
);
653 struct pipe_transfer
*vb_transfer
[PIPE_MAX_ATTRIBS
];
656 if (r300
->skip_rendering
) {
660 if (!u_trim_pipe_prim(mode
, &count
)) {
664 r300_update_derived_state(r300
);
666 for (i
= 0; i
< r300
->vertex_buffer_count
; i
++) {
667 void* buf
= pipe_buffer_map(pipe
,
668 r300
->vertex_buffer
[i
].buffer
,
671 draw_set_mapped_vertex_buffer(r300
->draw
, i
, buf
);
674 draw_set_mapped_element_buffer(r300
->draw
, 0, 0, NULL
);
676 draw_arrays(r300
->draw
, mode
, start
, count
);
678 /* XXX Not sure whether this is the best fix.
679 * It prevents CS from being rejected and weird assertion failures. */
680 draw_flush(r300
->draw
);
682 for (i
= 0; i
< r300
->vertex_buffer_count
; i
++) {
683 pipe_buffer_unmap(pipe
, r300
->vertex_buffer
[i
].buffer
,
685 draw_set_mapped_vertex_buffer(r300
->draw
, i
, NULL
);
689 /* SW TCL elements, using Draw. */
690 static void r300_swtcl_draw_range_elements(struct pipe_context
* pipe
,
691 struct pipe_resource
* indexBuffer
,
700 struct r300_context
* r300
= r300_context(pipe
);
701 struct pipe_transfer
*vb_transfer
[PIPE_MAX_ATTRIBS
];
702 struct pipe_transfer
*ib_transfer
;
706 if (r300
->skip_rendering
) {
710 if (!u_trim_pipe_prim(mode
, &count
)) {
714 r300_update_derived_state(r300
);
716 for (i
= 0; i
< r300
->vertex_buffer_count
; i
++) {
717 void* buf
= pipe_buffer_map(pipe
,
718 r300
->vertex_buffer
[i
].buffer
,
721 draw_set_mapped_vertex_buffer(r300
->draw
, i
, buf
);
724 indices
= pipe_buffer_map(pipe
, indexBuffer
,
725 PIPE_TRANSFER_READ
, &ib_transfer
);
726 draw_set_mapped_element_buffer_range(r300
->draw
, indexSize
, indexBias
,
727 minIndex
, maxIndex
, indices
);
729 draw_arrays(r300
->draw
, mode
, start
, count
);
731 /* XXX Not sure whether this is the best fix.
732 * It prevents CS from being rejected and weird assertion failures. */
733 draw_flush(r300
->draw
);
735 for (i
= 0; i
< r300
->vertex_buffer_count
; i
++) {
736 pipe_buffer_unmap(pipe
, r300
->vertex_buffer
[i
].buffer
,
738 draw_set_mapped_vertex_buffer(r300
->draw
, i
, NULL
);
741 pipe_buffer_unmap(pipe
, indexBuffer
,
743 draw_set_mapped_element_buffer_range(r300
->draw
, 0, 0,
744 start
, start
+ count
- 1,
748 /* Object for rendering using Draw. */
751 struct vbuf_render base
;
754 struct r300_context
* r300
;
756 /* Vertex information */
762 struct pipe_resource
* vbo
;
768 struct pipe_transfer
*vbo_transfer
;
771 static INLINE
struct r300_render
*
772 r300_render(struct vbuf_render
* render
)
774 return (struct r300_render
*)render
;
777 static const struct vertex_info
*
778 r300_render_get_vertex_info(struct vbuf_render
* render
)
780 struct r300_render
* r300render
= r300_render(render
);
781 struct r300_context
* r300
= r300render
->r300
;
783 return &r300
->vertex_info
;
786 static boolean
r300_render_allocate_vertices(struct vbuf_render
* render
,
790 struct r300_render
* r300render
= r300_render(render
);
791 struct r300_context
* r300
= r300render
->r300
;
792 struct pipe_screen
* screen
= r300
->context
.screen
;
793 size_t size
= (size_t)vertex_size
* (size_t)count
;
795 if (size
+ r300render
->vbo_offset
> r300render
->vbo_size
)
797 pipe_resource_reference(&r300
->vbo
, NULL
);
798 r300render
->vbo
= pipe_buffer_create(screen
,
799 PIPE_BIND_VERTEX_BUFFER
,
800 R300_MAX_DRAW_VBO_SIZE
);
801 r300render
->vbo_offset
= 0;
802 r300render
->vbo_size
= R300_MAX_DRAW_VBO_SIZE
;
805 r300render
->vertex_size
= vertex_size
;
806 r300
->vbo
= r300render
->vbo
;
807 r300
->vbo_offset
= r300render
->vbo_offset
;
809 return (r300render
->vbo
) ? TRUE
: FALSE
;
812 static void* r300_render_map_vertices(struct vbuf_render
* render
)
814 struct r300_render
* r300render
= r300_render(render
);
816 assert(!r300render
->vbo_transfer
);
818 r300render
->vbo_ptr
= pipe_buffer_map(&r300render
->r300
->context
,
821 &r300render
->vbo_transfer
);
823 return ((uint8_t*)r300render
->vbo_ptr
+ r300render
->vbo_offset
);
826 static void r300_render_unmap_vertices(struct vbuf_render
* render
,
830 struct r300_render
* r300render
= r300_render(render
);
831 struct pipe_context
* context
= &r300render
->r300
->context
;
833 assert(r300render
->vbo_transfer
);
835 r300render
->vbo_max_used
= MAX2(r300render
->vbo_max_used
,
836 r300render
->vertex_size
* (max
+ 1));
837 pipe_buffer_unmap(context
, r300render
->vbo
, r300render
->vbo_transfer
);
839 r300render
->vbo_transfer
= NULL
;
842 static void r300_render_release_vertices(struct vbuf_render
* render
)
844 struct r300_render
* r300render
= r300_render(render
);
846 r300render
->vbo_offset
+= r300render
->vbo_max_used
;
847 r300render
->vbo_max_used
= 0;
850 static boolean
r300_render_set_primitive(struct vbuf_render
* render
,
853 struct r300_render
* r300render
= r300_render(render
);
855 r300render
->prim
= prim
;
856 r300render
->hwprim
= r300_translate_primitive(prim
);
861 static void r300_render_draw_arrays(struct vbuf_render
* render
,
865 struct r300_render
* r300render
= r300_render(render
);
866 struct r300_context
* r300
= r300render
->r300
;
873 (void) i
; (void) ptr
;
875 r300_prepare_for_rendering(r300
, PREP_FIRST_DRAW
| PREP_EMIT_AOS_SWTCL
,
876 NULL
, dwords
, 0, 0, NULL
);
878 DBG(r300
, DBG_DRAW
, "r300: Doing vbuf render, count %d\n", count
);
880 /* Uncomment to dump all VBOs rendered through this interface.
882 ptr = pipe_buffer_map(&r300render->r300->context,
883 r300render->vbo, PIPE_TRANSFER_READ,
884 &r300render->vbo_transfer);
886 for (i = 0; i < count; i++) {
887 printf("r300: Vertex %d\n", i);
888 draw_dump_emitted_vertex(&r300->vertex_info, ptr);
889 ptr += r300->vertex_info.size * 4;
893 pipe_buffer_unmap(&r300render->r300->context, r300render->vbo,
894 r300render->vbo_transfer);
898 OUT_CS_REG(R300_GA_COLOR_CONTROL
,
899 r300_provoking_vertex_fixes(r300
, r300render
->prim
));
900 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX
, count
- 1);
901 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2
, 0);
902 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST
| (count
<< 16) |
907 static void r300_render_draw_elements(struct vbuf_render
* render
,
908 const ushort
* indices
,
911 struct r300_render
* r300render
= r300_render(render
);
912 struct r300_context
* r300
= r300render
->r300
;
914 unsigned end_cs_dwords
;
915 unsigned max_index
= (r300render
->vbo_size
- r300render
->vbo_offset
) /
916 (r300render
->r300
->vertex_info
.size
* 4) - 1;
917 unsigned short_count
;
918 unsigned free_dwords
;
922 /* Reserve at least 256 dwords.
924 * Below we manage the CS space manually because there may be more
925 * indices than it can fit in CS. */
926 r300_prepare_for_rendering(r300
,
927 PREP_FIRST_DRAW
| PREP_EMIT_AOS_SWTCL
| PREP_INDEXED
,
928 NULL
, 256, 0, 0, &end_cs_dwords
);
931 free_dwords
= r300
->rws
->get_cs_free_dwords(r300
->rws
);
933 short_count
= MIN2(count
, (free_dwords
- end_cs_dwords
- 6) * 2);
935 BEGIN_CS(6 + (short_count
+1)/2);
936 OUT_CS_REG(R300_GA_COLOR_CONTROL
,
937 r300_provoking_vertex_fixes(r300
, r300render
->prim
));
938 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX
, max_index
);
939 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2
, (short_count
+1)/2);
940 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES
| (short_count
<< 16) |
942 for (i
= 0; i
< short_count
-1; i
+= 2) {
943 OUT_CS(indices
[i
+1] << 16 | indices
[i
]);
945 if (short_count
% 2) {
946 OUT_CS(indices
[short_count
-1]);
950 /* OK now subtract the emitted indices and see if we need to emit
951 * another draw packet. */
952 indices
+= short_count
;
953 count
-= short_count
;
956 r300_prepare_for_rendering(r300
,
957 PREP_EMIT_AOS_SWTCL
| PREP_INDEXED
,
958 NULL
, 256, 0, 0, &end_cs_dwords
);
963 static void r300_render_destroy(struct vbuf_render
* render
)
968 static struct vbuf_render
* r300_render_create(struct r300_context
* r300
)
970 struct r300_render
* r300render
= CALLOC_STRUCT(r300_render
);
972 r300render
->r300
= r300
;
974 /* XXX find real numbers plz */
975 r300render
->base
.max_vertex_buffer_bytes
= 128 * 1024;
976 r300render
->base
.max_indices
= 16 * 1024;
978 r300render
->base
.get_vertex_info
= r300_render_get_vertex_info
;
979 r300render
->base
.allocate_vertices
= r300_render_allocate_vertices
;
980 r300render
->base
.map_vertices
= r300_render_map_vertices
;
981 r300render
->base
.unmap_vertices
= r300_render_unmap_vertices
;
982 r300render
->base
.set_primitive
= r300_render_set_primitive
;
983 r300render
->base
.draw_elements
= r300_render_draw_elements
;
984 r300render
->base
.draw_arrays
= r300_render_draw_arrays
;
985 r300render
->base
.release_vertices
= r300_render_release_vertices
;
986 r300render
->base
.destroy
= r300_render_destroy
;
988 r300render
->vbo
= NULL
;
989 r300render
->vbo_size
= 0;
990 r300render
->vbo_offset
= 0;
992 return &r300render
->base
;
995 struct draw_stage
* r300_draw_stage(struct r300_context
* r300
)
997 struct vbuf_render
* render
;
998 struct draw_stage
* stage
;
1000 render
= r300_render_create(r300
);
1006 stage
= draw_vbuf_stage(r300
->draw
, render
);
1009 render
->destroy(render
);
1013 draw_set_render(r300
->draw
, render
);
1018 /****************************************************************************
1019 * End of SW TCL functions *
1020 ***************************************************************************/
1022 /* If we used a quad to draw a rectangle, the pixels on the main diagonal
1023 * would be computed and stored twice, which makes the clear/copy codepaths
1024 * somewhat inefficient. Instead we use a rectangular point sprite with TCL
1026 static void r300_blitter_draw_rectangle(struct blitter_context
*blitter
,
1027 unsigned x1
, unsigned y1
,
1028 unsigned x2
, unsigned y2
,
1030 enum blitter_attrib_type type
,
1031 const float attrib
[4])
1033 struct r300_context
*r300
= r300_context(util_blitter_get_pipe(blitter
));
1034 unsigned width
= x2
- x1
;
1035 unsigned height
= y2
- y1
;
1037 unsigned vertex_size
= type
== UTIL_BLITTER_ATTRIB_COLOR
? 7 : 3;
1038 unsigned last_sprite_coord_enable
= r300
->sprite_coord_enable
;
1039 uint32_t vap_cntl_status
;
1042 /* Compute the number of dwords. */
1043 dwords
= r300
->draw
? 0 : 4;
1045 case UTIL_BLITTER_ATTRIB_COLOR
:
1049 case UTIL_BLITTER_ATTRIB_TEXCOORD
:
1053 case UTIL_BLITTER_ATTRIB_NONE
:
1058 /* Initialize the VAP control. */
1059 vap_cntl_status
= R300_VAP_TCL_BYPASS
;
1060 #ifdef PIPE_ARCH_LITTLE_ENDIAN
1061 vap_cntl_status
|= R300_VC_NO_SWAP
;
1063 vap_cntl_status
|= R300_VC_32BIT_SWAP
);
1066 if (type
== UTIL_BLITTER_ATTRIB_TEXCOORD
)
1067 r300
->sprite_coord_enable
= 1;
1069 r300_update_derived_state(r300
);
1071 /* Mark some states we don't care about as non-dirty. */
1072 r300
->viewport_state
.dirty
= FALSE
;
1073 r300
->clip_state
.dirty
= FALSE
;
1074 r300
->vertex_stream_state
.dirty
= FALSE
;
1075 r300
->vs_state
.dirty
= FALSE
;
1076 r300
->vs_constants
.dirty
= FALSE
;
1078 r300_prepare_for_rendering(r300
, PREP_FIRST_DRAW
, NULL
, dwords
, 0, 0, NULL
);
1080 BEGIN_CS_AS_CB(r300
, dwords
);
1082 OUT_CB_REG(R300_GA_POINT_SIZE
, (height
* 6) | ((width
* 6) << 16));
1085 case UTIL_BLITTER_ATTRIB_COLOR
:
1086 /* Set up the VAP output. */
1087 OUT_CB_REG(R300_VAP_VSM_VTX_ASSM
,
1088 R300_INPUT_CNTL_POS
| R300_INPUT_CNTL_COLOR
);
1089 OUT_CB_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0
, 2);
1090 OUT_CB(R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT
|
1091 R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT
);
1094 OUT_CB_REG(R300_VAP_PROG_STREAM_CNTL_0
,
1095 R300_DATA_TYPE_FLOAT_3
|
1096 ((R300_DATA_TYPE_FLOAT_4
| (2 << R300_DST_VEC_LOC_SHIFT
) |
1097 R300_LAST_VEC
) << 16));
1098 OUT_CB_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0
,
1099 R300_VAP_SWIZZLE_XYZ1
| (R300_VAP_SWIZZLE_XYZW
<< 16));
1102 case UTIL_BLITTER_ATTRIB_TEXCOORD
:
1103 /* Set up the GA to generate texcoords. */
1104 OUT_CB_REG(R300_GB_ENABLE
, R300_GB_POINT_STUFF_ENABLE
|
1105 (R300_GB_TEX_STR
<< R300_GB_TEX0_SOURCE_SHIFT
));
1106 OUT_CB_REG_SEQ(R300_GA_POINT_S0
, 4);
1107 OUT_CB_32F(attrib
[0]);
1108 OUT_CB_32F(attrib
[3]);
1109 OUT_CB_32F(attrib
[2]);
1110 OUT_CB_32F(attrib
[1]);
1113 case UTIL_BLITTER_ATTRIB_NONE
:
1114 /* Set up the VAP output. */
1115 OUT_CB_REG(R300_VAP_VSM_VTX_ASSM
, R300_INPUT_CNTL_POS
);
1116 OUT_CB_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0
, 2);
1117 OUT_CB(R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT
);
1120 OUT_CB_REG(R300_VAP_PROG_STREAM_CNTL_0
,
1121 R300_DATA_TYPE_FLOAT_3
| R300_LAST_VEC
);
1122 OUT_CB_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0
, R300_VAP_SWIZZLE_XYZ1
);
1126 /* Set up VAP controls. */
1128 OUT_CB_REG(R300_VAP_CNTL_STATUS
, vap_cntl_status
);
1129 OUT_CB_REG(R300_VAP_CLIP_CNTL
, R300_CLIP_DISABLE
);
1130 OUT_CB_REG(R300_VAP_VTE_CNTL
, R300_VTX_XY_FMT
| R300_VTX_Z_FMT
);
1131 OUT_CB_REG(R300_VAP_VTX_SIZE
, vertex_size
);
1132 OUT_CB_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX
, 2);
1137 OUT_CB_PKT3(R300_PACKET3_3D_DRAW_IMMD_2
, vertex_size
);
1138 OUT_CB(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED
| (1 << 16) |
1139 R300_VAP_VF_CNTL__PRIM_POINTS
);
1141 OUT_CB_32F(x1
+ width
* 0.5f
);
1142 OUT_CB_32F(y1
+ height
* 0.5f
);
1145 if (type
== UTIL_BLITTER_ATTRIB_COLOR
)
1146 for (i
= 0; i
< 4; i
++)
1147 OUT_CB_32F(attrib
[i
]);
1149 /* If we do not re-enable VAP immediately after the draw packet,
1150 * it goes crazy. Sometimes I wish this hardware didn't do random shit. */
1152 OUT_CB_REG(R300_VAP_CNTL_STATUS
,
1153 vap_cntl_status
& ~R300_VAP_TCL_BYPASS
);
1156 /* Restore the state. */
1157 r300
->clip_state
.dirty
= TRUE
;
1158 r300
->rs_block_state
.dirty
= TRUE
;
1159 r300
->rs_state
.dirty
= TRUE
;
1160 r300
->vertex_stream_state
.dirty
= TRUE
;
1161 r300
->viewport_state
.dirty
= TRUE
;
1162 r300
->vs_state
.dirty
= TRUE
;
1163 r300
->vs_constants
.dirty
= TRUE
;
1165 r300
->sprite_coord_enable
= last_sprite_coord_enable
;
1168 static void r300_resource_resolve(struct pipe_context
* pipe
,
1169 struct pipe_resource
* dest
,
1170 struct pipe_subresource subdest
,
1171 struct pipe_resource
* src
,
1172 struct pipe_subresource subsrc
)
1174 struct r300_context
* r300
= r300_context(pipe
);
1175 struct r300_aa_state
*aa
= (struct r300_aa_state
*)r300
->aa_state
.state
;
1176 struct pipe_surface
* srcsurf
= src
->screen
->get_tex_surface(src
->screen
,
1177 src
, subsrc
.face
, subsrc
.level
, 0, 0);
1178 float color
[] = {0, 0, 0, 0};
1180 DBG(r300
, DBG_DRAW
, "r300: Resolving resource...\n");
1182 /* Enable AA resolve. */
1183 aa
->dest
= r300_surface(
1184 dest
->screen
->get_tex_surface(dest
->screen
, dest
, subdest
.face
,
1185 subdest
.level
, 0, 0));
1188 R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE
|
1189 R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE
;
1190 r300
->aa_state
.size
= 12;
1191 r300
->aa_state
.dirty
= TRUE
;
1193 /* Resolve the surface. */
1194 r300
->context
.clear_render_target(pipe
,
1195 srcsurf
, color
, 0, 0, src
->width0
, src
->height0
);
1197 /* Disable AA resolve. */
1198 aa
->aaresolve_ctl
= 0;
1199 r300
->aa_state
.size
= 4;
1200 r300
->aa_state
.dirty
= TRUE
;
1202 pipe_surface_reference((struct pipe_surface
**)&srcsurf
, NULL
);
1203 pipe_surface_reference((struct pipe_surface
**)&aa
->dest
, NULL
);
1206 void r300_init_render_functions(struct r300_context
*r300
)
1208 /* Set generic functions. */
1209 r300
->context
.draw_elements
= r300_draw_elements
;
1211 /* Set draw functions based on presence of HW TCL. */
1212 if (r300
->screen
->caps
.has_tcl
) {
1213 r300
->context
.draw_arrays
= r300_draw_arrays
;
1214 r300
->context
.draw_range_elements
= r300_draw_range_elements
;
1216 r300
->context
.draw_arrays
= r300_swtcl_draw_arrays
;
1217 r300
->context
.draw_range_elements
= r300_swtcl_draw_range_elements
;
1220 r300
->context
.resource_resolve
= r300_resource_resolve
;
1221 r300
->blitter
->draw_rectangle
= r300_blitter_draw_rectangle
;
1223 /* Plug in the two-sided stencil reference value fallback if needed. */
1224 if (!r300
->screen
->caps
.is_r500
)
1225 r300_plug_in_stencil_ref_fallback(r300
);