r300_render.c revision 5506f6ef966b8883e575a3f60ce96ad42ee6ffd2
1/*
2 * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2010 Marek Olšák <maraeo@gmail.com>
4 *
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:
11 *
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
14 * Software.
15 *
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. */
23
24/* r300_render: Vertex and index buffer primitive emission. Contains both
25 * HW TCL fastpath rendering, and SW TCL Draw-assisted rendering. */
26
27#include "draw/draw_context.h"
28#include "draw/draw_vbuf.h"
29
30#include "util/u_inlines.h"
31
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"
36
37#include "r300_cs.h"
38#include "r300_context.h"
39#include "r300_screen_buffer.h"
40#include "r300_emit.h"
41#include "r300_reg.h"
42
43#include <limits.h>
44
45#define IMMD_DWORDS 32
46
47static uint32_t r300_translate_primitive(unsigned prim)
48{
49    static const int prim_conv[] = {
50        R300_VAP_VF_CNTL__PRIM_POINTS,
51        R300_VAP_VF_CNTL__PRIM_LINES,
52        R300_VAP_VF_CNTL__PRIM_LINE_LOOP,
53        R300_VAP_VF_CNTL__PRIM_LINE_STRIP,
54        R300_VAP_VF_CNTL__PRIM_TRIANGLES,
55        R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP,
56        R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN,
57        R300_VAP_VF_CNTL__PRIM_QUADS,
58        R300_VAP_VF_CNTL__PRIM_QUAD_STRIP,
59        R300_VAP_VF_CNTL__PRIM_POLYGON,
60        -1,
61        -1,
62        -1,
63        -1
64    };
65    unsigned hwprim = prim_conv[prim];
66
67    assert(hwprim != -1);
68    return hwprim;
69}
70
71static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
72                                            unsigned mode)
73{
74    struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
75    uint32_t color_control = rs->color_control;
76
77    /* By default (see r300_state.c:r300_create_rs_state) color_control is
78     * initialized to provoking the first vertex.
79     *
80     * Triangle fans must be reduced to the second vertex, not the first, in
81     * Gallium flatshade-first mode, as per the GL spec.
82     * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
83     *
84     * Quads never provoke correctly in flatshade-first mode. The first
85     * vertex is never considered as provoking, so only the second, third,
86     * and fourth vertices can be selected, and both "third" and "last" modes
87     * select the fourth vertex. This is probably due to D3D lacking quads.
88     *
89     * Similarly, polygons reduce to the first, not the last, vertex, when in
90     * "last" mode, and all other modes start from the second vertex.
91     *
92     * ~ C.
93     */
94
95    if (rs->rs.flatshade_first) {
96        switch (mode) {
97            case PIPE_PRIM_TRIANGLE_FAN:
98                color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
99                break;
100            case PIPE_PRIM_QUADS:
101            case PIPE_PRIM_QUAD_STRIP:
102            case PIPE_PRIM_POLYGON:
103                color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
104                break;
105            default:
106                color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
107                break;
108        }
109    } else {
110        color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
111    }
112
113    return color_control;
114}
115
116void r500_emit_index_bias(struct r300_context *r300, int index_bias)
117{
118    CS_LOCALS(r300);
119
120    BEGIN_CS(2);
121    OUT_CS_REG(R500_VAP_INDEX_OFFSET,
122               (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0));
123    END_CS;
124}
125
126static void r300_emit_draw_init(struct r300_context *r300, unsigned mode,
127                                unsigned min_index, unsigned max_index)
128{
129    CS_LOCALS(r300);
130
131    BEGIN_CS(5);
132    OUT_CS_REG(R300_GA_COLOR_CONTROL,
133            r300_provoking_vertex_fixes(r300, mode));
134    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
135    OUT_CS(max_index);
136    OUT_CS(min_index);
137    END_CS;
138}
139
140/* This function splits the index bias value into two parts:
141 * - buffer_offset: the value that can be safely added to buffer offsets
142 *   in r300_emit_vertex_arrays (it must yield a positive offset when added to
143 *   a vertex buffer offset)
144 * - index_offset: the value that must be manually subtracted from indices
145 *   in an index buffer to achieve negative offsets. */
146static void r300_split_index_bias(struct r300_context *r300, int index_bias,
147                                  int *buffer_offset, int *index_offset)
148{
149    struct pipe_vertex_buffer *vb, *vbufs = r300->vbuf_mgr->real_vertex_buffer;
150    struct pipe_vertex_element *velem = r300->velems->velem;
151    unsigned i, size;
152    int max_neg_bias;
153
154    if (index_bias < 0) {
155        /* See how large index bias we may subtract. We must be careful
156         * here because negative buffer offsets are not allowed
157         * by the DRM API. */
158        max_neg_bias = INT_MAX;
159        for (i = 0; i < r300->velems->count; i++) {
160            vb = &vbufs[velem[i].vertex_buffer_index];
161            size = (vb->buffer_offset + velem[i].src_offset) / vb->stride;
162            max_neg_bias = MIN2(max_neg_bias, size);
163        }
164
165        /* Now set the minimum allowed value. */
166        *buffer_offset = MAX2(-max_neg_bias, index_bias);
167    } else {
168        /* A positive index bias is OK. */
169        *buffer_offset = index_bias;
170    }
171
172    *index_offset = index_bias - *buffer_offset;
173}
174
175enum r300_prepare_flags {
176    PREP_EMIT_STATES    = (1 << 0), /* call emit_dirty_state and friends? */
177    PREP_VALIDATE_VBOS  = (1 << 1), /* validate VBOs? */
178    PREP_EMIT_VARRAYS       = (1 << 2), /* call emit_vertex_arrays? */
179    PREP_EMIT_VARRAYS_SWTCL = (1 << 3), /* call emit_vertex_arrays_swtcl? */
180    PREP_INDEXED        = (1 << 4)  /* is this draw_elements? */
181};
182
183/**
184 * Check if the requested number of dwords is available in the CS and
185 * if not, flush.
186 * \param r300          The context.
187 * \param flags         See r300_prepare_flags.
188 * \param cs_dwords     The number of dwords to reserve in CS.
189 * \return TRUE if the CS was flushed
190 */
191static boolean r300_reserve_cs_dwords(struct r300_context *r300,
192                                      enum r300_prepare_flags flags,
193                                      unsigned cs_dwords)
194{
195    boolean flushed        = FALSE;
196    boolean emit_states    = flags & PREP_EMIT_STATES;
197    boolean emit_vertex_arrays       = flags & PREP_EMIT_VARRAYS;
198    boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL;
199
200    /* Add dirty state, index offset, and AOS. */
201    if (emit_states)
202        cs_dwords += r300_get_num_dirty_dwords(r300);
203
204    if (r300->screen->caps.is_r500)
205        cs_dwords += 2; /* emit_index_offset */
206
207    if (emit_vertex_arrays)
208        cs_dwords += 55; /* emit_vertex_arrays */
209
210    if (emit_vertex_arrays_swtcl)
211        cs_dwords += 7; /* emit_vertex_arrays_swtcl */
212
213    cs_dwords += r300_get_num_cs_end_dwords(r300);
214
215    /* Reserve requested CS space. */
216    if (cs_dwords > (RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
217        r300_flush(&r300->context, RADEON_FLUSH_ASYNC, NULL);
218        flushed = TRUE;
219    }
220
221    return flushed;
222}
223
224/**
225 * Validate buffers and emit dirty state.
226 * \param r300          The context.
227 * \param flags         See r300_prepare_flags.
228 * \param index_buffer  The index buffer to validate. The parameter may be NULL.
229 * \param buffer_offset The offset passed to emit_vertex_arrays.
230 * \param index_bias    The index bias to emit.
231 * \param instance_id   Index of instance to render
232 * \return TRUE if rendering should be skipped
233 */
234static boolean r300_emit_states(struct r300_context *r300,
235                                enum r300_prepare_flags flags,
236                                struct pipe_resource *index_buffer,
237                                int buffer_offset,
238                                int index_bias, int instance_id)
239{
240    boolean emit_states    = flags & PREP_EMIT_STATES;
241    boolean emit_vertex_arrays       = flags & PREP_EMIT_VARRAYS;
242    boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL;
243    boolean indexed        = flags & PREP_INDEXED;
244    boolean validate_vbos  = flags & PREP_VALIDATE_VBOS;
245
246    /* Validate buffers and emit dirty state if needed. */
247    if (emit_states || (emit_vertex_arrays && validate_vbos)) {
248        if (!r300_emit_buffer_validate(r300, validate_vbos,
249                                       index_buffer)) {
250           fprintf(stderr, "r300: CS space validation failed. "
251                   "(not enough memory?) Skipping rendering.\n");
252           return FALSE;
253        }
254    }
255
256    if (emit_states)
257        r300_emit_dirty_state(r300);
258
259    if (r300->screen->caps.is_r500) {
260        if (r300->screen->caps.has_tcl)
261            r500_emit_index_bias(r300, index_bias);
262        else
263            r500_emit_index_bias(r300, 0);
264    }
265
266    if (emit_vertex_arrays &&
267        (r300->vertex_arrays_dirty ||
268         r300->vertex_arrays_indexed != indexed ||
269         r300->vertex_arrays_offset != buffer_offset ||
270         r300->vertex_arrays_instance_id != instance_id)) {
271        r300_emit_vertex_arrays(r300, buffer_offset, indexed, instance_id);
272
273        r300->vertex_arrays_dirty = FALSE;
274        r300->vertex_arrays_indexed = indexed;
275        r300->vertex_arrays_offset = buffer_offset;
276        r300->vertex_arrays_instance_id = instance_id;
277    }
278
279    if (emit_vertex_arrays_swtcl)
280        r300_emit_vertex_arrays_swtcl(r300, indexed);
281
282    return TRUE;
283}
284
285/**
286 * Check if the requested number of dwords is available in the CS and
287 * if not, flush. Then validate buffers and emit dirty state.
288 * \param r300          The context.
289 * \param flags         See r300_prepare_flags.
290 * \param index_buffer  The index buffer to validate. The parameter may be NULL.
291 * \param cs_dwords     The number of dwords to reserve in CS.
292 * \param buffer_offset The offset passed to emit_vertex_arrays.
293 * \param index_bias    The index bias to emit.
294 * \param instance_id The instance to render.
295 * \return TRUE if rendering should be skipped
296 */
297static boolean r300_prepare_for_rendering(struct r300_context *r300,
298                                          enum r300_prepare_flags flags,
299                                          struct pipe_resource *index_buffer,
300                                          unsigned cs_dwords,
301                                          int buffer_offset,
302                                          int index_bias,
303                                          int instance_id)
304{
305    /* Make sure there is enough space in the command stream and emit states. */
306    if (r300_reserve_cs_dwords(r300, flags, cs_dwords))
307        flags |= PREP_EMIT_STATES;
308
309    return r300_emit_states(r300, flags, index_buffer, buffer_offset,
310                            index_bias, instance_id);
311}
312
313static boolean immd_is_good_idea(struct r300_context *r300,
314                                 unsigned count)
315{
316    if (DBG_ON(r300, DBG_NO_IMMD)) {
317        return FALSE;
318    }
319
320    if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) {
321        return FALSE;
322    }
323
324    /* Buffers can only be used for read by r300 (except query buffers, but
325     * those can't be bound by a state tracker as vertex buffers). */
326    return TRUE;
327}
328
329/*****************************************************************************
330 * The HWTCL draw functions.                                                 *
331 ****************************************************************************/
332
333static void r300_draw_arrays_immediate(struct r300_context *r300,
334                                       const struct pipe_draw_info *info)
335{
336    struct pipe_vertex_element* velem;
337    struct pipe_vertex_buffer* vbuf;
338    unsigned vertex_element_count = r300->velems->count;
339    unsigned i, v, vbi;
340
341    /* Size of the vertex, in dwords. */
342    unsigned vertex_size = r300->velems->vertex_size_dwords;
343
344    /* The number of dwords for this draw operation. */
345    unsigned dwords = 4 + info->count * vertex_size;
346
347    /* Size of the vertex element, in dwords. */
348    unsigned size[PIPE_MAX_ATTRIBS];
349
350    /* Stride to the same attrib in the next vertex in the vertex buffer,
351     * in dwords. */
352    unsigned stride[PIPE_MAX_ATTRIBS];
353
354    /* Mapped vertex buffers. */
355    uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
356    uint32_t* mapelem[PIPE_MAX_ATTRIBS];
357
358    CS_LOCALS(r300);
359
360    if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
361        return;
362
363    /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
364    for (i = 0; i < vertex_element_count; i++) {
365        velem = &r300->velems->velem[i];
366        size[i] = r300->velems->format_size[i] / 4;
367        vbi = velem->vertex_buffer_index;
368        vbuf = &r300->vbuf_mgr->real_vertex_buffer[vbi];
369        stride[i] = vbuf->stride / 4;
370
371        /* Map the buffer. */
372        if (!map[vbi]) {
373            map[vbi] = (uint32_t*)r300->rws->buffer_map(
374                r300_resource(vbuf->buffer)->buf,
375                r300->cs, PIPE_TRANSFER_READ | PIPE_TRANSFER_UNSYNCHRONIZED);
376            map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * info->start;
377        }
378        mapelem[i] = map[vbi] + (velem->src_offset / 4);
379    }
380
381    r300_emit_draw_init(r300, info->mode, 0, info->count-1);
382
383    BEGIN_CS(dwords);
384    OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
385    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, info->count * vertex_size);
386    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (info->count << 16) |
387            r300_translate_primitive(info->mode));
388
389    /* Emit vertices. */
390    for (v = 0; v < info->count; v++) {
391        for (i = 0; i < vertex_element_count; i++) {
392            OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
393        }
394    }
395    END_CS;
396
397    /* Unmap buffers. */
398    for (i = 0; i < vertex_element_count; i++) {
399        vbi = r300->velems->velem[i].vertex_buffer_index;
400
401        if (map[vbi]) {
402            r300->rws->buffer_unmap(r300_resource(r300->vbuf_mgr->real_vertex_buffer[vbi].buffer)->buf);
403            map[vbi] = NULL;
404        }
405    }
406}
407
408static void r300_emit_draw_arrays(struct r300_context *r300,
409                                  unsigned mode,
410                                  unsigned count)
411{
412    boolean alt_num_verts = count > 65535;
413    CS_LOCALS(r300);
414
415    if (count >= (1 << 24)) {
416        fprintf(stderr, "r300: Got a huge number of vertices: %i, "
417                "refusing to render.\n", count);
418        return;
419    }
420
421    r300_emit_draw_init(r300, mode, 0, count-1);
422
423    BEGIN_CS(2 + (alt_num_verts ? 2 : 0));
424    if (alt_num_verts) {
425        OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
426    }
427    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
428    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
429           r300_translate_primitive(mode) |
430           (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
431    END_CS;
432}
433
434static void r300_emit_draw_elements(struct r300_context *r300,
435                                    struct pipe_resource* indexBuffer,
436                                    unsigned indexSize,
437                                    unsigned min_index,
438                                    unsigned max_index,
439                                    unsigned mode,
440                                    unsigned start,
441                                    unsigned count,
442                                    uint16_t *imm_indices3)
443{
444    uint32_t count_dwords, offset_dwords;
445    boolean alt_num_verts = count > 65535;
446    CS_LOCALS(r300);
447
448    if (count >= (1 << 24) || max_index >= (1 << 24)) {
449        fprintf(stderr, "r300: Got a huge number of vertices: %i, "
450                "refusing to render (max_index: %i).\n", count, max_index);
451        return;
452    }
453
454    DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
455        count, min_index, max_index);
456
457    r300_emit_draw_init(r300, mode, min_index, max_index);
458
459    /* If start is odd, render the first triangle with indices embedded
460     * in the command stream. This will increase start by 3 and make it
461     * even. We can then proceed without a fallback. */
462    if (indexSize == 2 && (start & 1) &&
463        mode == PIPE_PRIM_TRIANGLES) {
464        BEGIN_CS(4);
465        OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 2);
466        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (3 << 16) |
467               R300_VAP_VF_CNTL__PRIM_TRIANGLES);
468        OUT_CS(imm_indices3[1] << 16 | imm_indices3[0]);
469        OUT_CS(imm_indices3[2]);
470        END_CS;
471
472        start += 3;
473        count -= 3;
474        if (!count)
475           return;
476    }
477
478    offset_dwords = indexSize * start / sizeof(uint32_t);
479
480    BEGIN_CS(8 + (alt_num_verts ? 2 : 0));
481    if (alt_num_verts) {
482        OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
483    }
484    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
485    if (indexSize == 4) {
486        count_dwords = count;
487        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
488               R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
489               r300_translate_primitive(mode) |
490               (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
491    } else {
492        count_dwords = (count + 1) / 2;
493        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
494               r300_translate_primitive(mode) |
495               (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
496    }
497
498    OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
499    OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
500           (0 << R300_INDX_BUFFER_SKIP_SHIFT));
501    OUT_CS(offset_dwords << 2);
502    OUT_CS(count_dwords);
503    OUT_CS_RELOC(r300_resource(indexBuffer));
504    END_CS;
505}
506
507static void r300_draw_elements_immediate(struct r300_context *r300,
508                                         const struct pipe_draw_info *info)
509{
510    uint8_t *ptr1;
511    uint16_t *ptr2;
512    uint32_t *ptr4;
513    unsigned index_size = r300->index_buffer.index_size;
514    unsigned i, count_dwords = index_size == 4 ? info->count :
515                                                 (info->count + 1) / 2;
516    CS_LOCALS(r300);
517
518    /* 19 dwords for r300_draw_elements_immediate. Give up if the function fails. */
519    if (!r300_prepare_for_rendering(r300,
520            PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
521            PREP_INDEXED, NULL, 2+count_dwords, 0, info->index_bias, -1))
522        return;
523
524    r300_emit_draw_init(r300, info->mode,
525                        info->min_index + info->index_bias,
526                        info->max_index + info->index_bias);
527
528    BEGIN_CS(2 + count_dwords);
529    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords);
530
531    switch (index_size) {
532    case 1:
533        ptr1 = r300_resource(r300->index_buffer.buffer)->b.user_ptr;
534        ptr1 += info->start;
535
536        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
537               r300_translate_primitive(info->mode));
538
539        if (info->index_bias && !r300->screen->caps.is_r500) {
540            for (i = 0; i < info->count-1; i += 2)
541                OUT_CS(((ptr1[i+1] + info->index_bias) << 16) |
542                        (ptr1[i]   + info->index_bias));
543
544            if (info->count & 1)
545                OUT_CS(ptr1[i] + info->index_bias);
546        } else {
547            for (i = 0; i < info->count-1; i += 2)
548                OUT_CS(((ptr1[i+1]) << 16) |
549                        (ptr1[i]  ));
550
551            if (info->count & 1)
552                OUT_CS(ptr1[i]);
553        }
554        break;
555
556    case 2:
557        ptr2 = (uint16_t*)r300_resource(r300->index_buffer.buffer)->b.user_ptr;
558        ptr2 += info->start;
559
560        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
561               r300_translate_primitive(info->mode));
562
563        if (info->index_bias && !r300->screen->caps.is_r500) {
564            for (i = 0; i < info->count-1; i += 2)
565                OUT_CS(((ptr2[i+1] + info->index_bias) << 16) |
566                        (ptr2[i]   + info->index_bias));
567
568            if (info->count & 1)
569                OUT_CS(ptr2[i] + info->index_bias);
570        } else {
571            OUT_CS_TABLE(ptr2, count_dwords);
572        }
573        break;
574
575    case 4:
576        ptr4 = (uint32_t*)r300_resource(r300->index_buffer.buffer)->b.user_ptr;
577        ptr4 += info->start;
578
579        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
580               R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
581               r300_translate_primitive(info->mode));
582
583        if (info->index_bias && !r300->screen->caps.is_r500) {
584            for (i = 0; i < info->count; i++)
585                OUT_CS(ptr4[i] + info->index_bias);
586        } else {
587            OUT_CS_TABLE(ptr4, count_dwords);
588        }
589        break;
590    }
591    END_CS;
592}
593
594static void r300_draw_elements(struct r300_context *r300,
595                               const struct pipe_draw_info *info,
596                               int instance_id)
597{
598    struct pipe_resource *indexBuffer = r300->index_buffer.buffer;
599    unsigned indexSize = r300->index_buffer.index_size;
600    struct pipe_resource* orgIndexBuffer = indexBuffer;
601    unsigned start = info->start;
602    unsigned count = info->count;
603    boolean alt_num_verts = r300->screen->caps.is_r500 &&
604                            count > 65536;
605    unsigned short_count;
606    int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
607    uint16_t indices3[3];
608
609    if (info->index_bias && !r300->screen->caps.is_r500) {
610        r300_split_index_bias(r300, info->index_bias, &buffer_offset, &index_offset);
611    }
612
613    r300_translate_index_buffer(r300, &indexBuffer, &indexSize, index_offset,
614                                &start, count);
615
616    /* Fallback for misaligned ushort indices. */
617    if (indexSize == 2 && (start & 1) &&
618        !r300_resource(indexBuffer)->b.user_ptr) {
619        /* If we got here, then orgIndexBuffer == indexBuffer. */
620        uint16_t *ptr = r300->rws->buffer_map(r300_resource(orgIndexBuffer)->buf,
621                                              r300->cs,
622                                              PIPE_TRANSFER_READ |
623                                              PIPE_TRANSFER_UNSYNCHRONIZED);
624
625        if (info->mode == PIPE_PRIM_TRIANGLES) {
626           memcpy(indices3, ptr + start, 6);
627        } else {
628            /* Copy the mapped index buffer directly to the upload buffer.
629             * The start index will be aligned simply from the fact that
630             * every sub-buffer in the upload buffer is aligned. */
631            r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start,
632                                     count, (uint8_t*)ptr);
633        }
634        r300->rws->buffer_unmap(r300_resource(orgIndexBuffer)->buf);
635    } else {
636        if (r300_resource(indexBuffer)->b.user_ptr)
637            r300_upload_index_buffer(r300, &indexBuffer, indexSize,
638                                     &start, count,
639                                     r300_resource(indexBuffer)->b.user_ptr);
640    }
641
642    /* 19 dwords for emit_draw_elements. Give up if the function fails. */
643    if (!r300_prepare_for_rendering(r300,
644            PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
645            PREP_INDEXED, indexBuffer, 19, buffer_offset, info->index_bias,
646            instance_id))
647        goto done;
648
649    if (alt_num_verts || count <= 65535) {
650        r300_emit_draw_elements(r300, indexBuffer, indexSize, info->min_index,
651                                info->max_index, info->mode, start, count,
652                                indices3);
653    } else {
654        do {
655            /* The maximum must be divisible by 4 and 3,
656             * so that quad and triangle lists are split correctly.
657             *
658             * Strips, loops, and fans won't work. */
659            short_count = MIN2(count, 65532);
660
661            r300_emit_draw_elements(r300, indexBuffer, indexSize,
662                                     info->min_index, info->max_index,
663                                     info->mode, start, short_count, indices3);
664
665            start += short_count;
666            count -= short_count;
667
668            /* 15 dwords for emit_draw_elements */
669            if (count) {
670                if (!r300_prepare_for_rendering(r300,
671                        PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS | PREP_INDEXED,
672                        indexBuffer, 19, buffer_offset, info->index_bias,
673                        instance_id))
674                    goto done;
675            }
676        } while (count);
677    }
678
679done:
680    if (indexBuffer != orgIndexBuffer) {
681        pipe_resource_reference( &indexBuffer, NULL );
682    }
683}
684
685static void r300_draw_arrays(struct r300_context *r300,
686                             const struct pipe_draw_info *info,
687                             int instance_id)
688{
689    boolean alt_num_verts = r300->screen->caps.is_r500 &&
690                            info->count > 65536;
691    unsigned start = info->start;
692    unsigned count = info->count;
693    unsigned short_count;
694
695    /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
696    if (!r300_prepare_for_rendering(r300,
697                                    PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS,
698                                    NULL, 9, start, 0, instance_id))
699        return;
700
701    if (alt_num_verts || count <= 65535) {
702        r300_emit_draw_arrays(r300, info->mode, count);
703    } else {
704        do {
705            /* The maximum must be divisible by 4 and 3,
706             * so that quad and triangle lists are split correctly.
707             *
708             * Strips, loops, and fans won't work. */
709            short_count = MIN2(count, 65532);
710            r300_emit_draw_arrays(r300, info->mode, short_count);
711
712            start += short_count;
713            count -= short_count;
714
715            /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
716            if (count) {
717                if (!r300_prepare_for_rendering(r300,
718                                                PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS, NULL, 9,
719                                                start, 0, instance_id))
720                    return;
721            }
722        } while (count);
723    }
724}
725
726static void r300_draw_arrays_instanced(struct r300_context *r300,
727                                       const struct pipe_draw_info *info)
728{
729    int i;
730
731    for (i = 0; i < info->instance_count; i++)
732        r300_draw_arrays(r300, info, i);
733}
734
735static void r300_draw_elements_instanced(struct r300_context *r300,
736                                         const struct pipe_draw_info *info)
737{
738    int i;
739
740    for (i = 0; i < info->instance_count; i++)
741        r300_draw_elements(r300, info, i);
742}
743
744static void r300_draw_vbo(struct pipe_context* pipe,
745                          const struct pipe_draw_info *dinfo)
746{
747    struct r300_context* r300 = r300_context(pipe);
748    struct pipe_draw_info info = *dinfo;
749
750    info.indexed = info.indexed && r300->index_buffer.buffer;
751
752    if (r300->skip_rendering ||
753        !u_trim_pipe_prim(info.mode, &info.count)) {
754        return;
755    }
756
757    r300_update_derived_state(r300);
758
759    /* Start the vbuf manager and update buffers if needed. */
760    if (u_vbuf_draw_begin(r300->vbuf_mgr, &info) & U_VBUF_BUFFERS_UPDATED) {
761        r300->vertex_arrays_dirty = TRUE;
762    }
763
764    /* Draw. */
765    if (info.indexed) {
766        info.start += r300->index_buffer.offset;
767        info.max_index = MIN2(r300->vbuf_mgr->max_index, info.max_index);
768
769        if (info.instance_count <= 1) {
770            if (info.count <= 8 &&
771                r300_resource(r300->index_buffer.buffer)->b.user_ptr) {
772                r300_draw_elements_immediate(r300, &info);
773            } else {
774                r300_draw_elements(r300, &info, -1);
775            }
776        } else {
777            r300_draw_elements_instanced(r300, &info);
778        }
779    } else {
780        if (info.instance_count <= 1) {
781            if (immd_is_good_idea(r300, info.count)) {
782                r300_draw_arrays_immediate(r300, &info);
783            } else {
784                r300_draw_arrays(r300, &info, -1);
785            }
786        } else {
787            r300_draw_arrays_instanced(r300, &info);
788        }
789    }
790
791    u_vbuf_draw_end(r300->vbuf_mgr);
792}
793
794/****************************************************************************
795 * The rest of this file is for SW TCL rendering only. Please be polite and *
796 * keep these functions separated so that they are easier to locate. ~C.    *
797 ***************************************************************************/
798
799/* SW TCL elements, using Draw. */
800static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
801                                const struct pipe_draw_info *info)
802{
803    struct r300_context* r300 = r300_context(pipe);
804    struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
805    struct pipe_transfer *ib_transfer = NULL;
806    unsigned count = info->count;
807    int i;
808    void *indices = NULL;
809    boolean indexed = info->indexed && r300->index_buffer.buffer;
810
811    if (r300->skip_rendering) {
812        return;
813    }
814
815    if (!u_trim_pipe_prim(info->mode, &count)) {
816        return;
817    }
818
819    r300_update_derived_state(r300);
820
821    r300_reserve_cs_dwords(r300,
822            PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL |
823            (indexed ? PREP_INDEXED : 0),
824            indexed ? 256 : 6);
825
826    for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) {
827        if (r300->vbuf_mgr->vertex_buffer[i].buffer) {
828            void *buf = pipe_buffer_map(pipe,
829                                  r300->vbuf_mgr->vertex_buffer[i].buffer,
830                                  PIPE_TRANSFER_READ |
831                                  PIPE_TRANSFER_UNSYNCHRONIZED,
832                                  &vb_transfer[i]);
833            draw_set_mapped_vertex_buffer(r300->draw, i, buf);
834        }
835    }
836
837    if (indexed) {
838        indices = pipe_buffer_map(pipe, r300->index_buffer.buffer,
839                                  PIPE_TRANSFER_READ |
840                                  PIPE_TRANSFER_UNSYNCHRONIZED, &ib_transfer);
841    }
842
843    draw_set_mapped_index_buffer(r300->draw, indices);
844
845    r300->draw_vbo_locked = TRUE;
846    r300->draw_first_emitted = FALSE;
847    draw_vbo(r300->draw, info);
848    draw_flush(r300->draw);
849    r300->draw_vbo_locked = FALSE;
850
851    for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) {
852        if (r300->vbuf_mgr->vertex_buffer[i].buffer) {
853            pipe_buffer_unmap(pipe, vb_transfer[i]);
854            draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
855        }
856    }
857
858    if (indexed) {
859        pipe_buffer_unmap(pipe, ib_transfer);
860        draw_set_mapped_index_buffer(r300->draw, NULL);
861    }
862}
863
864/* Object for rendering using Draw. */
865struct r300_render {
866    /* Parent class */
867    struct vbuf_render base;
868
869    /* Pipe context */
870    struct r300_context* r300;
871
872    /* Vertex information */
873    size_t vertex_size;
874    unsigned prim;
875    unsigned hwprim;
876
877    /* VBO */
878    size_t vbo_max_used;
879    void * vbo_ptr;
880
881    struct pipe_transfer *vbo_transfer;
882};
883
884static INLINE struct r300_render*
885r300_render(struct vbuf_render* render)
886{
887    return (struct r300_render*)render;
888}
889
890static const struct vertex_info*
891r300_render_get_vertex_info(struct vbuf_render* render)
892{
893    struct r300_render* r300render = r300_render(render);
894    struct r300_context* r300 = r300render->r300;
895
896    return &r300->vertex_info;
897}
898
899static boolean r300_render_allocate_vertices(struct vbuf_render* render,
900                                                   ushort vertex_size,
901                                                   ushort count)
902{
903    struct r300_render* r300render = r300_render(render);
904    struct r300_context* r300 = r300render->r300;
905    struct pipe_screen* screen = r300->context.screen;
906    size_t size = (size_t)vertex_size * (size_t)count;
907
908    DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size);
909
910    if (size + r300->draw_vbo_offset > r300->draw_vbo_size)
911    {
912	pipe_resource_reference(&r300->vbo, NULL);
913        r300->vbo = pipe_buffer_create(screen,
914				       PIPE_BIND_VERTEX_BUFFER,
915				       PIPE_USAGE_STREAM,
916				       R300_MAX_DRAW_VBO_SIZE);
917        r300->draw_vbo_offset = 0;
918        r300->draw_vbo_size = R300_MAX_DRAW_VBO_SIZE;
919    }
920
921    r300render->vertex_size = vertex_size;
922
923    return (r300->vbo) ? TRUE : FALSE;
924}
925
926static void* r300_render_map_vertices(struct vbuf_render* render)
927{
928    struct r300_render* r300render = r300_render(render);
929    struct r300_context* r300 = r300render->r300;
930
931    assert(!r300render->vbo_transfer);
932
933    DBG(r300, DBG_DRAW, "r300: render_map_vertices\n");
934
935    r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context,
936					  r300->vbo,
937                                          PIPE_TRANSFER_WRITE |
938                                          PIPE_TRANSFER_UNSYNCHRONIZED,
939					  &r300render->vbo_transfer);
940
941    assert(r300render->vbo_ptr);
942
943    return ((uint8_t*)r300render->vbo_ptr + r300->draw_vbo_offset);
944}
945
946static void r300_render_unmap_vertices(struct vbuf_render* render,
947                                             ushort min,
948                                             ushort max)
949{
950    struct r300_render* r300render = r300_render(render);
951    struct pipe_context* context = &r300render->r300->context;
952    struct r300_context* r300 = r300render->r300;
953
954    assert(r300render->vbo_transfer);
955
956    DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n");
957
958    r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
959                                    r300render->vertex_size * (max + 1));
960    pipe_buffer_unmap(context, r300render->vbo_transfer);
961
962    r300render->vbo_transfer = NULL;
963}
964
965static void r300_render_release_vertices(struct vbuf_render* render)
966{
967    struct r300_render* r300render = r300_render(render);
968    struct r300_context* r300 = r300render->r300;
969
970    DBG(r300, DBG_DRAW, "r300: render_release_vertices\n");
971
972    r300->draw_vbo_offset += r300render->vbo_max_used;
973    r300render->vbo_max_used = 0;
974}
975
976static boolean r300_render_set_primitive(struct vbuf_render* render,
977                                               unsigned prim)
978{
979    struct r300_render* r300render = r300_render(render);
980
981    r300render->prim = prim;
982    r300render->hwprim = r300_translate_primitive(prim);
983
984    return TRUE;
985}
986
987static void r300_render_draw_arrays(struct vbuf_render* render,
988                                    unsigned start,
989                                    unsigned count)
990{
991    struct r300_render* r300render = r300_render(render);
992    struct r300_context* r300 = r300render->r300;
993    uint8_t* ptr;
994    unsigned i;
995    unsigned dwords = 6;
996
997    CS_LOCALS(r300);
998    (void) i; (void) ptr;
999
1000    DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
1001
1002    if (r300->draw_first_emitted) {
1003        if (!r300_prepare_for_rendering(r300,
1004                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
1005                NULL, dwords, 0, 0, -1))
1006            return;
1007    } else {
1008        if (!r300_emit_states(r300,
1009                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
1010                NULL, 0, 0, -1))
1011            return;
1012    }
1013
1014    BEGIN_CS(dwords);
1015    OUT_CS_REG(R300_GA_COLOR_CONTROL,
1016            r300_provoking_vertex_fixes(r300, r300render->prim));
1017    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
1018    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
1019    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
1020           r300render->hwprim);
1021    END_CS;
1022
1023    r300->draw_first_emitted = TRUE;
1024}
1025
1026static void r300_render_draw_elements(struct vbuf_render* render,
1027                                      const ushort* indices,
1028                                      uint count)
1029{
1030    struct r300_render* r300render = r300_render(render);
1031    struct r300_context* r300 = r300render->r300;
1032    int i;
1033    unsigned end_cs_dwords;
1034    unsigned max_index = (r300->draw_vbo_size - r300->draw_vbo_offset) /
1035                         (r300render->r300->vertex_info.size * 4) - 1;
1036    unsigned short_count;
1037    unsigned free_dwords;
1038
1039    CS_LOCALS(r300);
1040    DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
1041
1042    if (r300->draw_first_emitted) {
1043        if (!r300_prepare_for_rendering(r300,
1044                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1045                NULL, 256, 0, 0, -1))
1046            return;
1047    } else {
1048        if (!r300_emit_states(r300,
1049                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1050                NULL, 0, 0, -1))
1051            return;
1052    }
1053
1054    /* Below we manage the CS space manually because there may be more
1055     * indices than it can fit in CS. */
1056
1057    end_cs_dwords = r300_get_num_cs_end_dwords(r300);
1058
1059    while (count) {
1060        free_dwords = RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw;
1061
1062        short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2);
1063
1064        BEGIN_CS(6 + (short_count+1)/2);
1065        OUT_CS_REG(R300_GA_COLOR_CONTROL,
1066                r300_provoking_vertex_fixes(r300, r300render->prim));
1067        OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
1068        OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2);
1069        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) |
1070               r300render->hwprim);
1071        for (i = 0; i < short_count-1; i += 2) {
1072            OUT_CS(indices[i+1] << 16 | indices[i]);
1073        }
1074        if (short_count % 2) {
1075            OUT_CS(indices[short_count-1]);
1076        }
1077        END_CS;
1078
1079        /* OK now subtract the emitted indices and see if we need to emit
1080         * another draw packet. */
1081        indices += short_count;
1082        count -= short_count;
1083
1084        if (count) {
1085            if (!r300_prepare_for_rendering(r300,
1086                    PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1087                    NULL, 256, 0, 0, -1))
1088                return;
1089
1090            end_cs_dwords = r300_get_num_cs_end_dwords(r300);
1091        }
1092    }
1093
1094    r300->draw_first_emitted = TRUE;
1095}
1096
1097static void r300_render_destroy(struct vbuf_render* render)
1098{
1099    FREE(render);
1100}
1101
1102static struct vbuf_render* r300_render_create(struct r300_context* r300)
1103{
1104    struct r300_render* r300render = CALLOC_STRUCT(r300_render);
1105
1106    r300render->r300 = r300;
1107
1108    r300render->base.max_vertex_buffer_bytes = 1024 * 1024;
1109    r300render->base.max_indices = 16 * 1024;
1110
1111    r300render->base.get_vertex_info = r300_render_get_vertex_info;
1112    r300render->base.allocate_vertices = r300_render_allocate_vertices;
1113    r300render->base.map_vertices = r300_render_map_vertices;
1114    r300render->base.unmap_vertices = r300_render_unmap_vertices;
1115    r300render->base.set_primitive = r300_render_set_primitive;
1116    r300render->base.draw_elements = r300_render_draw_elements;
1117    r300render->base.draw_arrays = r300_render_draw_arrays;
1118    r300render->base.release_vertices = r300_render_release_vertices;
1119    r300render->base.destroy = r300_render_destroy;
1120
1121    return &r300render->base;
1122}
1123
1124struct draw_stage* r300_draw_stage(struct r300_context* r300)
1125{
1126    struct vbuf_render* render;
1127    struct draw_stage* stage;
1128
1129    render = r300_render_create(r300);
1130
1131    if (!render) {
1132        return NULL;
1133    }
1134
1135    stage = draw_vbuf_stage(r300->draw, render);
1136
1137    if (!stage) {
1138        render->destroy(render);
1139        return NULL;
1140    }
1141
1142    draw_set_render(r300->draw, render);
1143
1144    return stage;
1145}
1146
1147void r300_draw_flush_vbuf(struct r300_context *r300)
1148{
1149    pipe_resource_reference(&r300->vbo, NULL);
1150    r300->draw_vbo_size = 0;
1151}
1152
1153/****************************************************************************
1154 *                         End of SW TCL functions                          *
1155 ***************************************************************************/
1156
1157/* This functions is used to draw a rectangle for the blitter module.
1158 *
1159 * If we rendered a quad, the pixels on the main diagonal
1160 * would be computed and stored twice, which makes the clear/copy codepaths
1161 * somewhat inefficient. Instead we use a rectangular point sprite. */
1162static void r300_blitter_draw_rectangle(struct blitter_context *blitter,
1163                                        unsigned x1, unsigned y1,
1164                                        unsigned x2, unsigned y2,
1165                                        float depth,
1166                                        enum blitter_attrib_type type,
1167                                        const union pipe_color_union *attrib)
1168{
1169    struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
1170    unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
1171    unsigned width = x2 - x1;
1172    unsigned height = y2 - y1;
1173    unsigned vertex_size =
1174            type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
1175    unsigned dwords = 13 + vertex_size +
1176                      (type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0);
1177    static const union pipe_color_union zeros;
1178    CS_LOCALS(r300);
1179
1180    if (r300->skip_rendering)
1181        return;
1182
1183    r300->context.set_vertex_buffers(&r300->context, 0, NULL);
1184
1185    if (type == UTIL_BLITTER_ATTRIB_TEXCOORD)
1186        r300->sprite_coord_enable = 1;
1187
1188    r300_update_derived_state(r300);
1189
1190    /* Mark some states we don't care about as non-dirty. */
1191    r300->clip_state.dirty = FALSE;
1192    r300->viewport_state.dirty = FALSE;
1193
1194    if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
1195        goto done;
1196
1197    DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
1198
1199    BEGIN_CS(dwords);
1200    /* Set up GA. */
1201    OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
1202
1203    if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) {
1204        /* Set up the GA to generate texcoords. */
1205        OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
1206                   (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
1207        OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
1208        OUT_CS_32F(attrib->f[0]);
1209        OUT_CS_32F(attrib->f[3]);
1210        OUT_CS_32F(attrib->f[2]);
1211        OUT_CS_32F(attrib->f[1]);
1212    }
1213
1214    /* Set up VAP controls. */
1215    OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
1216    OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
1217    OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
1218    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
1219    OUT_CS(1);
1220    OUT_CS(0);
1221
1222    /* Draw. */
1223    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
1224    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
1225           R300_VAP_VF_CNTL__PRIM_POINTS);
1226
1227    OUT_CS_32F(x1 + width * 0.5f);
1228    OUT_CS_32F(y1 + height * 0.5f);
1229    OUT_CS_32F(depth);
1230    OUT_CS_32F(1);
1231
1232    if (vertex_size == 8) {
1233        if (!attrib)
1234            attrib = &zeros;
1235        OUT_CS_TABLE(attrib->f, 4);
1236    }
1237    END_CS;
1238
1239done:
1240    /* Restore the state. */
1241    r300_mark_atom_dirty(r300, &r300->clip_state);
1242    r300_mark_atom_dirty(r300, &r300->rs_state);
1243    r300_mark_atom_dirty(r300, &r300->viewport_state);
1244
1245    r300->sprite_coord_enable = last_sprite_coord_enable;
1246}
1247
1248static void r300_resource_resolve(struct pipe_context *pipe,
1249                                  const struct pipe_resolve_info *info)
1250{
1251    struct r300_context *r300 = r300_context(pipe);
1252    struct pipe_surface *srcsurf, *dstsurf, surf_tmpl;
1253    struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
1254    static const union pipe_color_union color;
1255
1256    memset(&surf_tmpl, 0, sizeof(surf_tmpl));
1257    surf_tmpl.format = info->src.res->format;
1258    surf_tmpl.u.tex.first_layer =
1259    surf_tmpl.u.tex.last_layer = info->src.layer;
1260    srcsurf = pipe->create_surface(pipe, info->src.res, &surf_tmpl);
1261    /* XXX Offset both surfaces by x0,y1. */
1262
1263    surf_tmpl.format = info->dst.res->format;
1264    surf_tmpl.u.tex.level = info->dst.level;
1265    surf_tmpl.u.tex.first_layer =
1266    surf_tmpl.u.tex.last_layer = info->dst.layer;
1267    dstsurf = pipe->create_surface(pipe, info->dst.res, &surf_tmpl);
1268
1269    DBG(r300, DBG_DRAW, "r300: Resolving resource...\n");
1270
1271    /* Enable AA resolve. */
1272    aa->dest = r300_surface(dstsurf);
1273    aa->aaresolve_ctl =
1274        R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE |
1275        R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE;
1276    r300->aa_state.size = 10;
1277    r300_mark_atom_dirty(r300, &r300->aa_state);
1278
1279    /* Resolve the surface. */
1280    /* XXX: y1 < 0 ==> Y flip */
1281    r300->context.clear_render_target(pipe,
1282                                      srcsurf, &color, 0, 0,
1283                                      info->dst.x1 - info->dst.x0,
1284                                      info->dst.y1 - info->dst.y0);
1285
1286    /* Disable AA resolve. */
1287    aa->aaresolve_ctl = 0;
1288    r300->aa_state.size = 4;
1289    r300_mark_atom_dirty(r300, &r300->aa_state);
1290
1291    pipe_surface_reference(&srcsurf, NULL);
1292    pipe_surface_reference(&dstsurf, NULL);
1293}
1294
1295void r300_init_render_functions(struct r300_context *r300)
1296{
1297    /* Set draw functions based on presence of HW TCL. */
1298    if (r300->screen->caps.has_tcl) {
1299        r300->context.draw_vbo = r300_draw_vbo;
1300    } else {
1301        r300->context.draw_vbo = r300_swtcl_draw_vbo;
1302    }
1303
1304    r300->context.resource_resolve = r300_resource_resolve;
1305    r300->blitter->draw_rectangle = r300_blitter_draw_rectangle;
1306
1307    /* Plug in the two-sided stencil reference value fallback if needed. */
1308    if (!r300->screen->caps.is_r500)
1309        r300_plug_in_stencil_ref_fallback(r300);
1310}
1311