r300_render.c revision 21e3c585f7c59fa909f6bb485a99bad33119a186
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, info->min_index, info->max_index);
525
526    BEGIN_CS(2 + count_dwords);
527    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords);
528
529    switch (index_size) {
530    case 1:
531        ptr1 = r300_resource(r300->index_buffer.buffer)->b.user_ptr;
532        ptr1 += info->start;
533
534        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
535               r300_translate_primitive(info->mode));
536
537        if (info->index_bias && !r300->screen->caps.is_r500) {
538            for (i = 0; i < info->count-1; i += 2)
539                OUT_CS(((ptr1[i+1] + info->index_bias) << 16) |
540                        (ptr1[i]   + info->index_bias));
541
542            if (info->count & 1)
543                OUT_CS(ptr1[i] + info->index_bias);
544        } else {
545            for (i = 0; i < info->count-1; i += 2)
546                OUT_CS(((ptr1[i+1]) << 16) |
547                        (ptr1[i]  ));
548
549            if (info->count & 1)
550                OUT_CS(ptr1[i]);
551        }
552        break;
553
554    case 2:
555        ptr2 = (uint16_t*)r300_resource(r300->index_buffer.buffer)->b.user_ptr;
556        ptr2 += info->start;
557
558        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
559               r300_translate_primitive(info->mode));
560
561        if (info->index_bias && !r300->screen->caps.is_r500) {
562            for (i = 0; i < info->count-1; i += 2)
563                OUT_CS(((ptr2[i+1] + info->index_bias) << 16) |
564                        (ptr2[i]   + info->index_bias));
565
566            if (info->count & 1)
567                OUT_CS(ptr2[i] + info->index_bias);
568        } else {
569            OUT_CS_TABLE(ptr2, count_dwords);
570        }
571        break;
572
573    case 4:
574        ptr4 = (uint32_t*)r300_resource(r300->index_buffer.buffer)->b.user_ptr;
575        ptr4 += info->start;
576
577        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
578               R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
579               r300_translate_primitive(info->mode));
580
581        if (info->index_bias && !r300->screen->caps.is_r500) {
582            for (i = 0; i < info->count; i++)
583                OUT_CS(ptr4[i] + info->index_bias);
584        } else {
585            OUT_CS_TABLE(ptr4, count_dwords);
586        }
587        break;
588    }
589    END_CS;
590}
591
592static void r300_draw_elements(struct r300_context *r300,
593                               const struct pipe_draw_info *info,
594                               int instance_id)
595{
596    struct pipe_resource *indexBuffer = r300->index_buffer.buffer;
597    unsigned indexSize = r300->index_buffer.index_size;
598    struct pipe_resource* orgIndexBuffer = indexBuffer;
599    unsigned start = info->start;
600    unsigned count = info->count;
601    boolean alt_num_verts = r300->screen->caps.is_r500 &&
602                            count > 65536;
603    unsigned short_count;
604    int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
605    uint16_t indices3[3];
606
607    if (info->index_bias && !r300->screen->caps.is_r500) {
608        r300_split_index_bias(r300, info->index_bias, &buffer_offset, &index_offset);
609    }
610
611    r300_translate_index_buffer(r300, &indexBuffer, &indexSize, index_offset,
612                                &start, count);
613
614    /* Fallback for misaligned ushort indices. */
615    if (indexSize == 2 && (start & 1) &&
616        !r300_resource(indexBuffer)->b.user_ptr) {
617        /* If we got here, then orgIndexBuffer == indexBuffer. */
618        uint16_t *ptr = r300->rws->buffer_map(r300_resource(orgIndexBuffer)->buf,
619                                              r300->cs,
620                                              PIPE_TRANSFER_READ |
621                                              PIPE_TRANSFER_UNSYNCHRONIZED);
622
623        if (info->mode == PIPE_PRIM_TRIANGLES) {
624           memcpy(indices3, ptr + start, 6);
625        } else {
626            /* Copy the mapped index buffer directly to the upload buffer.
627             * The start index will be aligned simply from the fact that
628             * every sub-buffer in the upload buffer is aligned. */
629            r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start,
630                                     count, (uint8_t*)ptr);
631        }
632        r300->rws->buffer_unmap(r300_resource(orgIndexBuffer)->buf);
633    } else {
634        if (r300_resource(indexBuffer)->b.user_ptr)
635            r300_upload_index_buffer(r300, &indexBuffer, indexSize,
636                                     &start, count,
637                                     r300_resource(indexBuffer)->b.user_ptr);
638    }
639
640    /* 19 dwords for emit_draw_elements. Give up if the function fails. */
641    if (!r300_prepare_for_rendering(r300,
642            PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
643            PREP_INDEXED, indexBuffer, 19, buffer_offset, info->index_bias,
644            instance_id))
645        goto done;
646
647    if (alt_num_verts || count <= 65535) {
648        r300_emit_draw_elements(r300, indexBuffer, indexSize, info->min_index,
649                                info->max_index, info->mode, start, count,
650                                indices3);
651    } else {
652        do {
653            /* The maximum must be divisible by 4 and 3,
654             * so that quad and triangle lists are split correctly.
655             *
656             * Strips, loops, and fans won't work. */
657            short_count = MIN2(count, 65532);
658
659            r300_emit_draw_elements(r300, indexBuffer, indexSize,
660                                     info->min_index, info->max_index,
661                                     info->mode, start, short_count, indices3);
662
663            start += short_count;
664            count -= short_count;
665
666            /* 15 dwords for emit_draw_elements */
667            if (count) {
668                if (!r300_prepare_for_rendering(r300,
669                        PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS | PREP_INDEXED,
670                        indexBuffer, 19, buffer_offset, info->index_bias,
671                        instance_id))
672                    goto done;
673            }
674        } while (count);
675    }
676
677done:
678    if (indexBuffer != orgIndexBuffer) {
679        pipe_resource_reference( &indexBuffer, NULL );
680    }
681}
682
683static void r300_draw_arrays(struct r300_context *r300,
684                             const struct pipe_draw_info *info,
685                             int instance_id)
686{
687    boolean alt_num_verts = r300->screen->caps.is_r500 &&
688                            info->count > 65536;
689    unsigned start = info->start;
690    unsigned count = info->count;
691    unsigned short_count;
692
693    /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
694    if (!r300_prepare_for_rendering(r300,
695                                    PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS,
696                                    NULL, 9, start, 0, instance_id))
697        return;
698
699    if (alt_num_verts || count <= 65535) {
700        r300_emit_draw_arrays(r300, info->mode, count);
701    } else {
702        do {
703            /* The maximum must be divisible by 4 and 3,
704             * so that quad and triangle lists are split correctly.
705             *
706             * Strips, loops, and fans won't work. */
707            short_count = MIN2(count, 65532);
708            r300_emit_draw_arrays(r300, info->mode, short_count);
709
710            start += short_count;
711            count -= short_count;
712
713            /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
714            if (count) {
715                if (!r300_prepare_for_rendering(r300,
716                                                PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS, NULL, 9,
717                                                start, 0, instance_id))
718                    return;
719            }
720        } while (count);
721    }
722}
723
724static void r300_draw_arrays_instanced(struct r300_context *r300,
725                                       const struct pipe_draw_info *info)
726{
727    int i;
728
729    for (i = 0; i < info->instance_count; i++)
730        r300_draw_arrays(r300, info, i);
731}
732
733static void r300_draw_elements_instanced(struct r300_context *r300,
734                                         const struct pipe_draw_info *info)
735{
736    int i;
737
738    for (i = 0; i < info->instance_count; i++)
739        r300_draw_elements(r300, info, i);
740}
741
742static void r300_draw_vbo(struct pipe_context* pipe,
743                          const struct pipe_draw_info *dinfo)
744{
745    struct r300_context* r300 = r300_context(pipe);
746    struct pipe_draw_info info = *dinfo;
747
748    info.indexed = info.indexed && r300->index_buffer.buffer;
749
750    if (r300->skip_rendering ||
751        !u_trim_pipe_prim(info.mode, &info.count)) {
752        return;
753    }
754
755    r300_update_derived_state(r300);
756
757    /* Start the vbuf manager and update buffers if needed. */
758    if (u_vbuf_draw_begin(r300->vbuf_mgr, &info) & U_VBUF_BUFFERS_UPDATED) {
759        r300->vertex_arrays_dirty = TRUE;
760    }
761
762    /* Draw. */
763    if (info.indexed) {
764        info.start += r300->index_buffer.offset;
765        info.max_index = MIN2(r300->vbuf_mgr->max_index, info.max_index);
766
767        if (info.instance_count <= 1) {
768            if (info.count <= 8 &&
769                r300_resource(r300->index_buffer.buffer)->b.user_ptr) {
770                r300_draw_elements_immediate(r300, &info);
771            } else {
772                r300_draw_elements(r300, &info, -1);
773            }
774        } else {
775            r300_draw_elements_instanced(r300, &info);
776        }
777    } else {
778        if (info.instance_count <= 1) {
779            if (immd_is_good_idea(r300, info.count)) {
780                r300_draw_arrays_immediate(r300, &info);
781            } else {
782                r300_draw_arrays(r300, &info, -1);
783            }
784        } else {
785            r300_draw_arrays_instanced(r300, &info);
786        }
787    }
788
789    u_vbuf_draw_end(r300->vbuf_mgr);
790}
791
792/****************************************************************************
793 * The rest of this file is for SW TCL rendering only. Please be polite and *
794 * keep these functions separated so that they are easier to locate. ~C.    *
795 ***************************************************************************/
796
797/* SW TCL elements, using Draw. */
798static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
799                                const struct pipe_draw_info *info)
800{
801    struct r300_context* r300 = r300_context(pipe);
802    struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
803    struct pipe_transfer *ib_transfer = NULL;
804    int i;
805    void *indices = NULL;
806    boolean indexed = info->indexed && r300->index_buffer.buffer;
807
808    if (r300->skip_rendering) {
809        return;
810    }
811
812    r300_update_derived_state(r300);
813
814    r300_reserve_cs_dwords(r300,
815            PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL |
816            (indexed ? PREP_INDEXED : 0),
817            indexed ? 256 : 6);
818
819    for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) {
820        if (r300->vbuf_mgr->vertex_buffer[i].buffer) {
821            void *buf = pipe_buffer_map(pipe,
822                                  r300->vbuf_mgr->vertex_buffer[i].buffer,
823                                  PIPE_TRANSFER_READ |
824                                  PIPE_TRANSFER_UNSYNCHRONIZED,
825                                  &vb_transfer[i]);
826            draw_set_mapped_vertex_buffer(r300->draw, i, buf);
827        }
828    }
829
830    if (indexed) {
831        indices = pipe_buffer_map(pipe, r300->index_buffer.buffer,
832                                  PIPE_TRANSFER_READ |
833                                  PIPE_TRANSFER_UNSYNCHRONIZED, &ib_transfer);
834    }
835
836    draw_set_mapped_index_buffer(r300->draw, indices);
837
838    r300->draw_vbo_locked = TRUE;
839    r300->draw_first_emitted = FALSE;
840    draw_vbo(r300->draw, info);
841    draw_flush(r300->draw);
842    r300->draw_vbo_locked = FALSE;
843
844    for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) {
845        if (r300->vbuf_mgr->vertex_buffer[i].buffer) {
846            pipe_buffer_unmap(pipe, vb_transfer[i]);
847            draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
848        }
849    }
850
851    if (indexed) {
852        pipe_buffer_unmap(pipe, ib_transfer);
853        draw_set_mapped_index_buffer(r300->draw, NULL);
854    }
855}
856
857/* Object for rendering using Draw. */
858struct r300_render {
859    /* Parent class */
860    struct vbuf_render base;
861
862    /* Pipe context */
863    struct r300_context* r300;
864
865    /* Vertex information */
866    size_t vertex_size;
867    unsigned prim;
868    unsigned hwprim;
869
870    /* VBO */
871    size_t vbo_max_used;
872    void * vbo_ptr;
873
874    struct pipe_transfer *vbo_transfer;
875};
876
877static INLINE struct r300_render*
878r300_render(struct vbuf_render* render)
879{
880    return (struct r300_render*)render;
881}
882
883static const struct vertex_info*
884r300_render_get_vertex_info(struct vbuf_render* render)
885{
886    struct r300_render* r300render = r300_render(render);
887    struct r300_context* r300 = r300render->r300;
888
889    return &r300->vertex_info;
890}
891
892static boolean r300_render_allocate_vertices(struct vbuf_render* render,
893                                                   ushort vertex_size,
894                                                   ushort count)
895{
896    struct r300_render* r300render = r300_render(render);
897    struct r300_context* r300 = r300render->r300;
898    struct pipe_screen* screen = r300->context.screen;
899    size_t size = (size_t)vertex_size * (size_t)count;
900
901    DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size);
902
903    if (size + r300->draw_vbo_offset > r300->draw_vbo_size)
904    {
905	pipe_resource_reference(&r300->vbo, NULL);
906        r300->vbo = pipe_buffer_create(screen,
907				       PIPE_BIND_VERTEX_BUFFER,
908				       PIPE_USAGE_STREAM,
909				       R300_MAX_DRAW_VBO_SIZE);
910        r300->draw_vbo_offset = 0;
911        r300->draw_vbo_size = R300_MAX_DRAW_VBO_SIZE;
912    }
913
914    r300render->vertex_size = vertex_size;
915
916    return (r300->vbo) ? TRUE : FALSE;
917}
918
919static void* r300_render_map_vertices(struct vbuf_render* render)
920{
921    struct r300_render* r300render = r300_render(render);
922    struct r300_context* r300 = r300render->r300;
923
924    assert(!r300render->vbo_transfer);
925
926    DBG(r300, DBG_DRAW, "r300: render_map_vertices\n");
927
928    r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context,
929					  r300->vbo,
930                                          PIPE_TRANSFER_WRITE |
931                                          PIPE_TRANSFER_UNSYNCHRONIZED,
932					  &r300render->vbo_transfer);
933
934    assert(r300render->vbo_ptr);
935
936    return ((uint8_t*)r300render->vbo_ptr + r300->draw_vbo_offset);
937}
938
939static void r300_render_unmap_vertices(struct vbuf_render* render,
940                                             ushort min,
941                                             ushort max)
942{
943    struct r300_render* r300render = r300_render(render);
944    struct pipe_context* context = &r300render->r300->context;
945    struct r300_context* r300 = r300render->r300;
946
947    assert(r300render->vbo_transfer);
948
949    DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n");
950
951    r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
952                                    r300render->vertex_size * (max + 1));
953    pipe_buffer_unmap(context, r300render->vbo_transfer);
954
955    r300render->vbo_transfer = NULL;
956}
957
958static void r300_render_release_vertices(struct vbuf_render* render)
959{
960    struct r300_render* r300render = r300_render(render);
961    struct r300_context* r300 = r300render->r300;
962
963    DBG(r300, DBG_DRAW, "r300: render_release_vertices\n");
964
965    r300->draw_vbo_offset += r300render->vbo_max_used;
966    r300render->vbo_max_used = 0;
967}
968
969static boolean r300_render_set_primitive(struct vbuf_render* render,
970                                               unsigned prim)
971{
972    struct r300_render* r300render = r300_render(render);
973
974    r300render->prim = prim;
975    r300render->hwprim = r300_translate_primitive(prim);
976
977    return TRUE;
978}
979
980static void r300_render_draw_arrays(struct vbuf_render* render,
981                                    unsigned start,
982                                    unsigned count)
983{
984    struct r300_render* r300render = r300_render(render);
985    struct r300_context* r300 = r300render->r300;
986    uint8_t* ptr;
987    unsigned i;
988    unsigned dwords = 6;
989
990    CS_LOCALS(r300);
991    (void) i; (void) ptr;
992
993    DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
994
995    if (r300->draw_first_emitted) {
996        if (!r300_prepare_for_rendering(r300,
997                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
998                NULL, dwords, 0, 0, -1))
999            return;
1000    } else {
1001        if (!r300_emit_states(r300,
1002                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
1003                NULL, 0, 0, -1))
1004            return;
1005    }
1006
1007    BEGIN_CS(dwords);
1008    OUT_CS_REG(R300_GA_COLOR_CONTROL,
1009            r300_provoking_vertex_fixes(r300, r300render->prim));
1010    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
1011    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
1012    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
1013           r300render->hwprim);
1014    END_CS;
1015
1016    r300->draw_first_emitted = TRUE;
1017}
1018
1019static void r300_render_draw_elements(struct vbuf_render* render,
1020                                      const ushort* indices,
1021                                      uint count)
1022{
1023    struct r300_render* r300render = r300_render(render);
1024    struct r300_context* r300 = r300render->r300;
1025    int i;
1026    unsigned end_cs_dwords;
1027    unsigned max_index = (r300->draw_vbo_size - r300->draw_vbo_offset) /
1028                         (r300render->r300->vertex_info.size * 4) - 1;
1029    unsigned short_count;
1030    unsigned free_dwords;
1031
1032    CS_LOCALS(r300);
1033    DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
1034
1035    if (r300->draw_first_emitted) {
1036        if (!r300_prepare_for_rendering(r300,
1037                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1038                NULL, 256, 0, 0, -1))
1039            return;
1040    } else {
1041        if (!r300_emit_states(r300,
1042                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1043                NULL, 0, 0, -1))
1044            return;
1045    }
1046
1047    /* Below we manage the CS space manually because there may be more
1048     * indices than it can fit in CS. */
1049
1050    end_cs_dwords = r300_get_num_cs_end_dwords(r300);
1051
1052    while (count) {
1053        free_dwords = RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw;
1054
1055        short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2);
1056
1057        BEGIN_CS(6 + (short_count+1)/2);
1058        OUT_CS_REG(R300_GA_COLOR_CONTROL,
1059                r300_provoking_vertex_fixes(r300, r300render->prim));
1060        OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
1061        OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2);
1062        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) |
1063               r300render->hwprim);
1064        for (i = 0; i < short_count-1; i += 2) {
1065            OUT_CS(indices[i+1] << 16 | indices[i]);
1066        }
1067        if (short_count % 2) {
1068            OUT_CS(indices[short_count-1]);
1069        }
1070        END_CS;
1071
1072        /* OK now subtract the emitted indices and see if we need to emit
1073         * another draw packet. */
1074        indices += short_count;
1075        count -= short_count;
1076
1077        if (count) {
1078            if (!r300_prepare_for_rendering(r300,
1079                    PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1080                    NULL, 256, 0, 0, -1))
1081                return;
1082
1083            end_cs_dwords = r300_get_num_cs_end_dwords(r300);
1084        }
1085    }
1086
1087    r300->draw_first_emitted = TRUE;
1088}
1089
1090static void r300_render_destroy(struct vbuf_render* render)
1091{
1092    FREE(render);
1093}
1094
1095static struct vbuf_render* r300_render_create(struct r300_context* r300)
1096{
1097    struct r300_render* r300render = CALLOC_STRUCT(r300_render);
1098
1099    r300render->r300 = r300;
1100
1101    r300render->base.max_vertex_buffer_bytes = 1024 * 1024;
1102    r300render->base.max_indices = 16 * 1024;
1103
1104    r300render->base.get_vertex_info = r300_render_get_vertex_info;
1105    r300render->base.allocate_vertices = r300_render_allocate_vertices;
1106    r300render->base.map_vertices = r300_render_map_vertices;
1107    r300render->base.unmap_vertices = r300_render_unmap_vertices;
1108    r300render->base.set_primitive = r300_render_set_primitive;
1109    r300render->base.draw_elements = r300_render_draw_elements;
1110    r300render->base.draw_arrays = r300_render_draw_arrays;
1111    r300render->base.release_vertices = r300_render_release_vertices;
1112    r300render->base.destroy = r300_render_destroy;
1113
1114    return &r300render->base;
1115}
1116
1117struct draw_stage* r300_draw_stage(struct r300_context* r300)
1118{
1119    struct vbuf_render* render;
1120    struct draw_stage* stage;
1121
1122    render = r300_render_create(r300);
1123
1124    if (!render) {
1125        return NULL;
1126    }
1127
1128    stage = draw_vbuf_stage(r300->draw, render);
1129
1130    if (!stage) {
1131        render->destroy(render);
1132        return NULL;
1133    }
1134
1135    draw_set_render(r300->draw, render);
1136
1137    return stage;
1138}
1139
1140void r300_draw_flush_vbuf(struct r300_context *r300)
1141{
1142    pipe_resource_reference(&r300->vbo, NULL);
1143    r300->draw_vbo_size = 0;
1144}
1145
1146/****************************************************************************
1147 *                         End of SW TCL functions                          *
1148 ***************************************************************************/
1149
1150/* This functions is used to draw a rectangle for the blitter module.
1151 *
1152 * If we rendered a quad, the pixels on the main diagonal
1153 * would be computed and stored twice, which makes the clear/copy codepaths
1154 * somewhat inefficient. Instead we use a rectangular point sprite. */
1155static void r300_blitter_draw_rectangle(struct blitter_context *blitter,
1156                                        unsigned x1, unsigned y1,
1157                                        unsigned x2, unsigned y2,
1158                                        float depth,
1159                                        enum blitter_attrib_type type,
1160                                        const union pipe_color_union *attrib)
1161{
1162    struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
1163    unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
1164    unsigned width = x2 - x1;
1165    unsigned height = y2 - y1;
1166    unsigned vertex_size =
1167            type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
1168    unsigned dwords = 13 + vertex_size +
1169                      (type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0);
1170    static const union pipe_color_union zeros;
1171    CS_LOCALS(r300);
1172
1173    if (r300->skip_rendering)
1174        return;
1175
1176    r300->context.set_vertex_buffers(&r300->context, 0, NULL);
1177
1178    if (type == UTIL_BLITTER_ATTRIB_TEXCOORD)
1179        r300->sprite_coord_enable = 1;
1180
1181    r300_update_derived_state(r300);
1182
1183    /* Mark some states we don't care about as non-dirty. */
1184    r300->clip_state.dirty = FALSE;
1185    r300->viewport_state.dirty = FALSE;
1186
1187    if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
1188        goto done;
1189
1190    DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
1191
1192    BEGIN_CS(dwords);
1193    /* Set up GA. */
1194    OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
1195
1196    if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) {
1197        /* Set up the GA to generate texcoords. */
1198        OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
1199                   (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
1200        OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
1201        OUT_CS_32F(attrib->f[0]);
1202        OUT_CS_32F(attrib->f[3]);
1203        OUT_CS_32F(attrib->f[2]);
1204        OUT_CS_32F(attrib->f[1]);
1205    }
1206
1207    /* Set up VAP controls. */
1208    OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
1209    OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
1210    OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
1211    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
1212    OUT_CS(1);
1213    OUT_CS(0);
1214
1215    /* Draw. */
1216    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
1217    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
1218           R300_VAP_VF_CNTL__PRIM_POINTS);
1219
1220    OUT_CS_32F(x1 + width * 0.5f);
1221    OUT_CS_32F(y1 + height * 0.5f);
1222    OUT_CS_32F(depth);
1223    OUT_CS_32F(1);
1224
1225    if (vertex_size == 8) {
1226        if (!attrib)
1227            attrib = &zeros;
1228        OUT_CS_TABLE(attrib->f, 4);
1229    }
1230    END_CS;
1231
1232done:
1233    /* Restore the state. */
1234    r300_mark_atom_dirty(r300, &r300->clip_state);
1235    r300_mark_atom_dirty(r300, &r300->rs_state);
1236    r300_mark_atom_dirty(r300, &r300->viewport_state);
1237
1238    r300->sprite_coord_enable = last_sprite_coord_enable;
1239}
1240
1241static void r300_resource_resolve(struct pipe_context *pipe,
1242                                  const struct pipe_resolve_info *info)
1243{
1244    struct r300_context *r300 = r300_context(pipe);
1245    struct pipe_surface *srcsurf, *dstsurf, surf_tmpl;
1246    struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
1247    static const union pipe_color_union color;
1248
1249    memset(&surf_tmpl, 0, sizeof(surf_tmpl));
1250    surf_tmpl.format = info->src.res->format;
1251    surf_tmpl.u.tex.first_layer =
1252    surf_tmpl.u.tex.last_layer = info->src.layer;
1253    srcsurf = pipe->create_surface(pipe, info->src.res, &surf_tmpl);
1254    /* XXX Offset both surfaces by x0,y1. */
1255
1256    surf_tmpl.format = info->dst.res->format;
1257    surf_tmpl.u.tex.level = info->dst.level;
1258    surf_tmpl.u.tex.first_layer =
1259    surf_tmpl.u.tex.last_layer = info->dst.layer;
1260    dstsurf = pipe->create_surface(pipe, info->dst.res, &surf_tmpl);
1261
1262    DBG(r300, DBG_DRAW, "r300: Resolving resource...\n");
1263
1264    /* Enable AA resolve. */
1265    aa->dest = r300_surface(dstsurf);
1266    aa->aaresolve_ctl =
1267        R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE |
1268        R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE;
1269    r300->aa_state.size = 10;
1270    r300_mark_atom_dirty(r300, &r300->aa_state);
1271
1272    /* Resolve the surface. */
1273    /* XXX: y1 < 0 ==> Y flip */
1274    r300->context.clear_render_target(pipe,
1275                                      srcsurf, &color, 0, 0,
1276                                      info->dst.x1 - info->dst.x0,
1277                                      info->dst.y1 - info->dst.y0);
1278
1279    /* Disable AA resolve. */
1280    aa->aaresolve_ctl = 0;
1281    r300->aa_state.size = 4;
1282    r300_mark_atom_dirty(r300, &r300->aa_state);
1283
1284    pipe_surface_reference(&srcsurf, NULL);
1285    pipe_surface_reference(&dstsurf, NULL);
1286}
1287
1288void r300_init_render_functions(struct r300_context *r300)
1289{
1290    /* Set draw functions based on presence of HW TCL. */
1291    if (r300->screen->caps.has_tcl) {
1292        r300->context.draw_vbo = r300_draw_vbo;
1293    } else {
1294        r300->context.draw_vbo = r300_swtcl_draw_vbo;
1295    }
1296
1297    r300->context.resource_resolve = r300_resource_resolve;
1298    r300->blitter->draw_rectangle = r300_blitter_draw_rectangle;
1299
1300    /* Plug in the two-sided stencil reference value fallback if needed. */
1301    if (!r300->screen->caps.is_r500)
1302        r300_plug_in_stencil_ref_fallback(r300);
1303}
1304