r300_render.c revision bf469f4edc60bd1c5fd770cb231b8d5ab801427f
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 max_index)
128{
129    CS_LOCALS(r300);
130
131    assert(max_index < (1 << 24));
132
133    BEGIN_CS(5);
134    OUT_CS_REG(R300_GA_COLOR_CONTROL,
135            r300_provoking_vertex_fixes(r300, mode));
136    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
137    OUT_CS(max_index);
138    OUT_CS(0);
139    END_CS;
140}
141
142/* This function splits the index bias value into two parts:
143 * - buffer_offset: the value that can be safely added to buffer offsets
144 *   in r300_emit_vertex_arrays (it must yield a positive offset when added to
145 *   a vertex buffer offset)
146 * - index_offset: the value that must be manually subtracted from indices
147 *   in an index buffer to achieve negative offsets. */
148static void r300_split_index_bias(struct r300_context *r300, int index_bias,
149                                  int *buffer_offset, int *index_offset)
150{
151    struct pipe_vertex_buffer *vb, *vbufs = r300->vertex_buffer;
152    struct pipe_vertex_element *velem = r300->velems->velem;
153    unsigned i, size;
154    int max_neg_bias;
155
156    if (index_bias < 0) {
157        /* See how large index bias we may subtract. We must be careful
158         * here because negative buffer offsets are not allowed
159         * by the DRM API. */
160        max_neg_bias = INT_MAX;
161        for (i = 0; i < r300->velems->count; i++) {
162            vb = &vbufs[velem[i].vertex_buffer_index];
163            size = (vb->buffer_offset + velem[i].src_offset) / vb->stride;
164            max_neg_bias = MIN2(max_neg_bias, size);
165        }
166
167        /* Now set the minimum allowed value. */
168        *buffer_offset = MAX2(-max_neg_bias, index_bias);
169    } else {
170        /* A positive index bias is OK. */
171        *buffer_offset = index_bias;
172    }
173
174    *index_offset = index_bias - *buffer_offset;
175}
176
177enum r300_prepare_flags {
178    PREP_EMIT_STATES    = (1 << 0), /* call emit_dirty_state and friends? */
179    PREP_VALIDATE_VBOS  = (1 << 1), /* validate VBOs? */
180    PREP_EMIT_VARRAYS       = (1 << 2), /* call emit_vertex_arrays? */
181    PREP_EMIT_VARRAYS_SWTCL = (1 << 3), /* call emit_vertex_arrays_swtcl? */
182    PREP_INDEXED        = (1 << 4)  /* is this draw_elements? */
183};
184
185/**
186 * Check if the requested number of dwords is available in the CS and
187 * if not, flush.
188 * \param r300          The context.
189 * \param flags         See r300_prepare_flags.
190 * \param cs_dwords     The number of dwords to reserve in CS.
191 * \return TRUE if the CS was flushed
192 */
193static boolean r300_reserve_cs_dwords(struct r300_context *r300,
194                                      enum r300_prepare_flags flags,
195                                      unsigned cs_dwords)
196{
197    boolean flushed        = FALSE;
198    boolean emit_states    = flags & PREP_EMIT_STATES;
199    boolean emit_vertex_arrays       = flags & PREP_EMIT_VARRAYS;
200    boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL;
201
202    /* Add dirty state, index offset, and AOS. */
203    if (emit_states)
204        cs_dwords += r300_get_num_dirty_dwords(r300);
205
206    if (r300->screen->caps.is_r500)
207        cs_dwords += 2; /* emit_index_offset */
208
209    if (emit_vertex_arrays)
210        cs_dwords += 55; /* emit_vertex_arrays */
211
212    if (emit_vertex_arrays_swtcl)
213        cs_dwords += 7; /* emit_vertex_arrays_swtcl */
214
215    cs_dwords += r300_get_num_cs_end_dwords(r300);
216
217    /* Reserve requested CS space. */
218    if (cs_dwords > (RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
219        r300_flush(&r300->context, RADEON_FLUSH_ASYNC, NULL);
220        flushed = TRUE;
221    }
222
223    return flushed;
224}
225
226/**
227 * Validate buffers and emit dirty state.
228 * \param r300          The context.
229 * \param flags         See r300_prepare_flags.
230 * \param index_buffer  The index buffer to validate. The parameter may be NULL.
231 * \param buffer_offset The offset passed to emit_vertex_arrays.
232 * \param index_bias    The index bias to emit.
233 * \param instance_id   Index of instance to render
234 * \return TRUE if rendering should be skipped
235 */
236static boolean r300_emit_states(struct r300_context *r300,
237                                enum r300_prepare_flags flags,
238                                struct pipe_resource *index_buffer,
239                                int buffer_offset,
240                                int index_bias, int instance_id)
241{
242    boolean emit_states    = flags & PREP_EMIT_STATES;
243    boolean emit_vertex_arrays       = flags & PREP_EMIT_VARRAYS;
244    boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL;
245    boolean indexed        = flags & PREP_INDEXED;
246    boolean validate_vbos  = flags & PREP_VALIDATE_VBOS;
247
248    /* Validate buffers and emit dirty state if needed. */
249    if (emit_states || (emit_vertex_arrays && validate_vbos)) {
250        if (!r300_emit_buffer_validate(r300, validate_vbos,
251                                       index_buffer)) {
252           fprintf(stderr, "r300: CS space validation failed. "
253                   "(not enough memory?) Skipping rendering.\n");
254           return FALSE;
255        }
256    }
257
258    if (emit_states)
259        r300_emit_dirty_state(r300);
260
261    if (r300->screen->caps.is_r500) {
262        if (r300->screen->caps.has_tcl)
263            r500_emit_index_bias(r300, index_bias);
264        else
265            r500_emit_index_bias(r300, 0);
266    }
267
268    if (emit_vertex_arrays &&
269        (r300->vertex_arrays_dirty ||
270         r300->vertex_arrays_indexed != indexed ||
271         r300->vertex_arrays_offset != buffer_offset ||
272         r300->vertex_arrays_instance_id != instance_id)) {
273        r300_emit_vertex_arrays(r300, buffer_offset, indexed, instance_id);
274
275        r300->vertex_arrays_dirty = FALSE;
276        r300->vertex_arrays_indexed = indexed;
277        r300->vertex_arrays_offset = buffer_offset;
278        r300->vertex_arrays_instance_id = instance_id;
279    }
280
281    if (emit_vertex_arrays_swtcl)
282        r300_emit_vertex_arrays_swtcl(r300, indexed);
283
284    return TRUE;
285}
286
287/**
288 * Check if the requested number of dwords is available in the CS and
289 * if not, flush. Then validate buffers and emit dirty state.
290 * \param r300          The context.
291 * \param flags         See r300_prepare_flags.
292 * \param index_buffer  The index buffer to validate. The parameter may be NULL.
293 * \param cs_dwords     The number of dwords to reserve in CS.
294 * \param buffer_offset The offset passed to emit_vertex_arrays.
295 * \param index_bias    The index bias to emit.
296 * \param instance_id The instance to render.
297 * \return TRUE if rendering should be skipped
298 */
299static boolean r300_prepare_for_rendering(struct r300_context *r300,
300                                          enum r300_prepare_flags flags,
301                                          struct pipe_resource *index_buffer,
302                                          unsigned cs_dwords,
303                                          int buffer_offset,
304                                          int index_bias,
305                                          int instance_id)
306{
307    /* Make sure there is enough space in the command stream and emit states. */
308    if (r300_reserve_cs_dwords(r300, flags, cs_dwords))
309        flags |= PREP_EMIT_STATES;
310
311    return r300_emit_states(r300, flags, index_buffer, buffer_offset,
312                            index_bias, instance_id);
313}
314
315static boolean immd_is_good_idea(struct r300_context *r300,
316                                 unsigned count)
317{
318    if (DBG_ON(r300, DBG_NO_IMMD)) {
319        return FALSE;
320    }
321
322    if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) {
323        return FALSE;
324    }
325
326    /* Buffers can only be used for read by r300 (except query buffers, but
327     * those can't be bound by a state tracker as vertex buffers). */
328    return TRUE;
329}
330
331/*****************************************************************************
332 * The HWTCL draw functions.                                                 *
333 ****************************************************************************/
334
335static void r300_draw_arrays_immediate(struct r300_context *r300,
336                                       const struct pipe_draw_info *info)
337{
338    struct pipe_vertex_element* velem;
339    struct pipe_vertex_buffer* vbuf;
340    unsigned vertex_element_count = r300->velems->count;
341    unsigned i, v, vbi;
342
343    /* Size of the vertex, in dwords. */
344    unsigned vertex_size = r300->velems->vertex_size_dwords;
345
346    /* The number of dwords for this draw operation. */
347    unsigned dwords = 4 + info->count * vertex_size;
348
349    /* Size of the vertex element, in dwords. */
350    unsigned size[PIPE_MAX_ATTRIBS];
351
352    /* Stride to the same attrib in the next vertex in the vertex buffer,
353     * in dwords. */
354    unsigned stride[PIPE_MAX_ATTRIBS];
355
356    /* Mapped vertex buffers. */
357    uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
358    uint32_t* mapelem[PIPE_MAX_ATTRIBS];
359
360    CS_LOCALS(r300);
361
362    if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
363        return;
364
365    /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
366    for (i = 0; i < vertex_element_count; i++) {
367        velem = &r300->velems->velem[i];
368        size[i] = r300->velems->format_size[i] / 4;
369        vbi = velem->vertex_buffer_index;
370        vbuf = &r300->vertex_buffer[vbi];
371        stride[i] = vbuf->stride / 4;
372
373        /* Map the buffer. */
374        if (!map[vbi]) {
375            map[vbi] = (uint32_t*)r300->rws->buffer_map(
376                r300_resource(vbuf->buffer)->cs_buf,
377                r300->cs, PIPE_TRANSFER_READ | PIPE_TRANSFER_UNSYNCHRONIZED);
378            map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * info->start;
379        }
380        mapelem[i] = map[vbi] + (velem->src_offset / 4);
381    }
382
383    r300_emit_draw_init(r300, info->mode, info->count-1);
384
385    BEGIN_CS(dwords);
386    OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
387    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, info->count * vertex_size);
388    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (info->count << 16) |
389            r300_translate_primitive(info->mode));
390
391    /* Emit vertices. */
392    for (v = 0; v < info->count; v++) {
393        for (i = 0; i < vertex_element_count; i++) {
394            OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
395        }
396    }
397    END_CS;
398
399    /* Unmap buffers. */
400    for (i = 0; i < vertex_element_count; i++) {
401        vbi = r300->velems->velem[i].vertex_buffer_index;
402
403        if (map[vbi]) {
404            r300->rws->buffer_unmap(r300_resource(r300->vertex_buffer[vbi].buffer)->cs_buf);
405            map[vbi] = NULL;
406        }
407    }
408}
409
410static void r300_emit_draw_arrays(struct r300_context *r300,
411                                  unsigned mode,
412                                  unsigned count)
413{
414    boolean alt_num_verts = count > 65535;
415    CS_LOCALS(r300);
416
417    if (count >= (1 << 24)) {
418        fprintf(stderr, "r300: Got a huge number of vertices: %i, "
419                "refusing to render.\n", count);
420        return;
421    }
422
423    r300_emit_draw_init(r300, mode, count-1);
424
425    BEGIN_CS(2 + (alt_num_verts ? 2 : 0));
426    if (alt_num_verts) {
427        OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
428    }
429    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
430    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
431           r300_translate_primitive(mode) |
432           (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
433    END_CS;
434}
435
436static void r300_emit_draw_elements(struct r300_context *r300,
437                                    struct pipe_resource* indexBuffer,
438                                    unsigned indexSize,
439                                    unsigned max_index,
440                                    unsigned mode,
441                                    unsigned start,
442                                    unsigned count,
443                                    uint16_t *imm_indices3)
444{
445    uint32_t count_dwords, offset_dwords;
446    boolean alt_num_verts = count > 65535;
447    CS_LOCALS(r300);
448
449    if (count >= (1 << 24)) {
450        fprintf(stderr, "r300: Got a huge number of vertices: %i, "
451                "refusing to render (max_index: %i).\n", count, max_index);
452        return;
453    }
454
455    DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, max %u\n",
456        count, max_index);
457
458    r300_emit_draw_init(r300, mode, max_index);
459
460    /* If start is odd, render the first triangle with indices embedded
461     * in the command stream. This will increase start by 3 and make it
462     * even. We can then proceed without a fallback. */
463    if (indexSize == 2 && (start & 1) &&
464        mode == PIPE_PRIM_TRIANGLES) {
465        BEGIN_CS(4);
466        OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 2);
467        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (3 << 16) |
468               R300_VAP_VF_CNTL__PRIM_TRIANGLES);
469        OUT_CS(imm_indices3[1] << 16 | imm_indices3[0]);
470        OUT_CS(imm_indices3[2]);
471        END_CS;
472
473        start += 3;
474        count -= 3;
475        if (!count)
476           return;
477    }
478
479    offset_dwords = indexSize * start / sizeof(uint32_t);
480
481    BEGIN_CS(8 + (alt_num_verts ? 2 : 0));
482    if (alt_num_verts) {
483        OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
484    }
485    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
486    if (indexSize == 4) {
487        count_dwords = count;
488        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
489               R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
490               r300_translate_primitive(mode) |
491               (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
492    } else {
493        count_dwords = (count + 1) / 2;
494        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
495               r300_translate_primitive(mode) |
496               (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
497    }
498
499    OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
500    OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
501           (0 << R300_INDX_BUFFER_SKIP_SHIFT));
502    OUT_CS(offset_dwords << 2);
503    OUT_CS(count_dwords);
504    OUT_CS_RELOC(r300_resource(indexBuffer));
505    END_CS;
506}
507
508static void r300_draw_elements_immediate(struct r300_context *r300,
509                                         const struct pipe_draw_info *info)
510{
511    const uint8_t *ptr1;
512    const uint16_t *ptr2;
513    const uint32_t *ptr4;
514    unsigned index_size = r300->index_buffer.index_size;
515    unsigned i, count_dwords = index_size == 4 ? info->count :
516                                                 (info->count + 1) / 2;
517    CS_LOCALS(r300);
518
519    /* 19 dwords for r300_draw_elements_immediate. Give up if the function fails. */
520    if (!r300_prepare_for_rendering(r300,
521            PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
522            PREP_INDEXED, NULL, 2+count_dwords, 0, info->index_bias, -1))
523        return;
524
525    r300_emit_draw_init(r300, info->mode, info->max_index);
526
527    BEGIN_CS(2 + count_dwords);
528    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords);
529
530    switch (index_size) {
531    case 1:
532        ptr1 = (uint8_t*)r300->index_buffer.user_buffer;
533        ptr1 += info->start;
534
535        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
536               r300_translate_primitive(info->mode));
537
538        if (info->index_bias && !r300->screen->caps.is_r500) {
539            for (i = 0; i < info->count-1; i += 2)
540                OUT_CS(((ptr1[i+1] + info->index_bias) << 16) |
541                        (ptr1[i]   + info->index_bias));
542
543            if (info->count & 1)
544                OUT_CS(ptr1[i] + info->index_bias);
545        } else {
546            for (i = 0; i < info->count-1; i += 2)
547                OUT_CS(((ptr1[i+1]) << 16) |
548                        (ptr1[i]  ));
549
550            if (info->count & 1)
551                OUT_CS(ptr1[i]);
552        }
553        break;
554
555    case 2:
556        ptr2 = (uint16_t*)r300->index_buffer.user_buffer;
557        ptr2 += info->start;
558
559        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
560               r300_translate_primitive(info->mode));
561
562        if (info->index_bias && !r300->screen->caps.is_r500) {
563            for (i = 0; i < info->count-1; i += 2)
564                OUT_CS(((ptr2[i+1] + info->index_bias) << 16) |
565                        (ptr2[i]   + info->index_bias));
566
567            if (info->count & 1)
568                OUT_CS(ptr2[i] + info->index_bias);
569        } else {
570            OUT_CS_TABLE(ptr2, count_dwords);
571        }
572        break;
573
574    case 4:
575        ptr4 = (uint32_t*)r300->index_buffer.user_buffer;
576        ptr4 += info->start;
577
578        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
579               R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
580               r300_translate_primitive(info->mode));
581
582        if (info->index_bias && !r300->screen->caps.is_r500) {
583            for (i = 0; i < info->count; i++)
584                OUT_CS(ptr4[i] + info->index_bias);
585        } else {
586            OUT_CS_TABLE(ptr4, count_dwords);
587        }
588        break;
589    }
590    END_CS;
591}
592
593static void r300_draw_elements(struct r300_context *r300,
594                               const struct pipe_draw_info *info,
595                               int instance_id)
596{
597    struct pipe_resource *indexBuffer = r300->index_buffer.buffer;
598    unsigned indexSize = r300->index_buffer.index_size;
599    struct pipe_resource* orgIndexBuffer = indexBuffer;
600    unsigned start = info->start;
601    unsigned count = info->count;
602    boolean alt_num_verts = r300->screen->caps.is_r500 &&
603                            count > 65536;
604    unsigned short_count;
605    int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
606    uint16_t indices3[3];
607
608    if (info->index_bias && !r300->screen->caps.is_r500) {
609        r300_split_index_bias(r300, info->index_bias, &buffer_offset,
610                              &index_offset);
611    }
612
613    r300_translate_index_buffer(r300, &r300->index_buffer, &indexBuffer,
614                                &indexSize, index_offset, &start, count);
615
616    /* Fallback for misaligned ushort indices. */
617    if (indexSize == 2 && (start & 1) && indexBuffer) {
618        /* If we got here, then orgIndexBuffer == indexBuffer. */
619        uint16_t *ptr = r300->rws->buffer_map(r300_resource(orgIndexBuffer)->cs_buf,
620                                              r300->cs,
621                                              PIPE_TRANSFER_READ |
622                                              PIPE_TRANSFER_UNSYNCHRONIZED);
623
624        if (info->mode == PIPE_PRIM_TRIANGLES) {
625           memcpy(indices3, ptr + start, 6);
626        } else {
627            /* Copy the mapped index buffer directly to the upload buffer.
628             * The start index will be aligned simply from the fact that
629             * every sub-buffer in the upload buffer is aligned. */
630            r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start,
631                                     count, (uint8_t*)ptr);
632        }
633        r300->rws->buffer_unmap(r300_resource(orgIndexBuffer)->cs_buf);
634    } else {
635        if (r300->index_buffer.user_buffer)
636            r300_upload_index_buffer(r300, &indexBuffer, indexSize,
637                                     &start, count,
638                                     r300->index_buffer.user_buffer);
639    }
640
641    /* 19 dwords for emit_draw_elements. Give up if the function fails. */
642    if (!r300_prepare_for_rendering(r300,
643            PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
644            PREP_INDEXED, indexBuffer, 19, buffer_offset, info->index_bias,
645            instance_id))
646        goto done;
647
648    if (alt_num_verts || count <= 65535) {
649        r300_emit_draw_elements(r300, indexBuffer, indexSize,
650                                info->max_index, info->mode, start, count,
651                                indices3);
652    } else {
653        do {
654            /* The maximum must be divisible by 4 and 3,
655             * so that quad and triangle lists are split correctly.
656             *
657             * Strips, loops, and fans won't work. */
658            short_count = MIN2(count, 65532);
659
660            r300_emit_draw_elements(r300, indexBuffer, indexSize,
661                                     info->max_index,
662                                     info->mode, start, short_count, indices3);
663
664            start += short_count;
665            count -= short_count;
666
667            /* 15 dwords for emit_draw_elements */
668            if (count) {
669                if (!r300_prepare_for_rendering(r300,
670                        PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS | PREP_INDEXED,
671                        indexBuffer, 19, buffer_offset, info->index_bias,
672                        instance_id))
673                    goto done;
674            }
675        } while (count);
676    }
677
678done:
679    if (indexBuffer != orgIndexBuffer) {
680        pipe_resource_reference( &indexBuffer, NULL );
681    }
682}
683
684static void r300_draw_arrays(struct r300_context *r300,
685                             const struct pipe_draw_info *info,
686                             int instance_id)
687{
688    boolean alt_num_verts = r300->screen->caps.is_r500 &&
689                            info->count > 65536;
690    unsigned start = info->start;
691    unsigned count = info->count;
692    unsigned short_count;
693
694    /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
695    if (!r300_prepare_for_rendering(r300,
696                                    PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS,
697                                    NULL, 9, start, 0, instance_id))
698        return;
699
700    if (alt_num_verts || count <= 65535) {
701        r300_emit_draw_arrays(r300, info->mode, count);
702    } else {
703        do {
704            /* The maximum must be divisible by 4 and 3,
705             * so that quad and triangle lists are split correctly.
706             *
707             * Strips, loops, and fans won't work. */
708            short_count = MIN2(count, 65532);
709            r300_emit_draw_arrays(r300, info->mode, short_count);
710
711            start += short_count;
712            count -= short_count;
713
714            /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
715            if (count) {
716                if (!r300_prepare_for_rendering(r300,
717                                                PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS, NULL, 9,
718                                                start, 0, instance_id))
719                    return;
720            }
721        } while (count);
722    }
723}
724
725static void r300_draw_arrays_instanced(struct r300_context *r300,
726                                       const struct pipe_draw_info *info)
727{
728    int i;
729
730    for (i = 0; i < info->instance_count; i++)
731        r300_draw_arrays(r300, info, i);
732}
733
734static void r300_draw_elements_instanced(struct r300_context *r300,
735                                         const struct pipe_draw_info *info)
736{
737    int i;
738
739    for (i = 0; i < info->instance_count; i++)
740        r300_draw_elements(r300, info, i);
741}
742
743static unsigned r300_max_vertex_count(struct r300_context *r300)
744{
745   unsigned i, nr = r300->velems->count;
746   struct pipe_vertex_element *velems = r300->velems->velem;
747   unsigned result = ~0;
748
749   for (i = 0; i < nr; i++) {
750      struct pipe_vertex_buffer *vb =
751            &r300->vertex_buffer[velems[i].vertex_buffer_index];
752      unsigned size, max_count, value;
753
754      /* We're not interested in constant and per-instance attribs. */
755      if (!vb->buffer ||
756          !vb->stride ||
757          velems[i].instance_divisor) {
758         continue;
759      }
760
761      size = vb->buffer->width0;
762
763      /* Subtract buffer_offset. */
764      value = vb->buffer_offset;
765      if (value >= size) {
766         return 0;
767      }
768      size -= value;
769
770      /* Subtract src_offset. */
771      value = velems[i].src_offset;
772      if (value >= size) {
773         return 0;
774      }
775      size -= value;
776
777      /* Subtract format_size. */
778      value = r300->velems->format_size[i];
779      if (value >= size) {
780         return 0;
781      }
782      size -= value;
783
784      /* Compute the max count. */
785      max_count = 1 + size / vb->stride;
786      result = MIN2(result, max_count);
787   }
788   return result;
789}
790
791
792static void r300_draw_vbo(struct pipe_context* pipe,
793                          const struct pipe_draw_info *dinfo)
794{
795    struct r300_context* r300 = r300_context(pipe);
796    struct pipe_draw_info info = *dinfo;
797
798    info.indexed = info.indexed;
799
800    if (r300->skip_rendering ||
801        !u_trim_pipe_prim(info.mode, &info.count)) {
802        return;
803    }
804
805    r300_update_derived_state(r300);
806
807    /* Draw. */
808    if (info.indexed) {
809        unsigned max_count = r300_max_vertex_count(r300);
810
811        if (!max_count) {
812           fprintf(stderr, "r300: Skipping a draw command. There is a buffer "
813                   " which is too small to be used for rendering.\n");
814           return;
815        }
816
817        if (max_count == ~0) {
818           /* There are no per-vertex vertex elements. Use the hardware maximum. */
819           max_count = 0xffffff;
820        }
821
822        info.max_index = max_count - 1;
823        info.start += r300->index_buffer.offset / r300->index_buffer.index_size;
824
825        if (info.instance_count <= 1) {
826            if (info.count <= 8 &&
827                r300->index_buffer.user_buffer) {
828                r300_draw_elements_immediate(r300, &info);
829            } else {
830                r300_draw_elements(r300, &info, -1);
831            }
832        } else {
833            r300_draw_elements_instanced(r300, &info);
834        }
835    } else {
836        if (info.instance_count <= 1) {
837            if (immd_is_good_idea(r300, info.count)) {
838                r300_draw_arrays_immediate(r300, &info);
839            } else {
840                r300_draw_arrays(r300, &info, -1);
841            }
842        } else {
843            r300_draw_arrays_instanced(r300, &info);
844        }
845    }
846}
847
848/****************************************************************************
849 * The rest of this file is for SW TCL rendering only. Please be polite and *
850 * keep these functions separated so that they are easier to locate. ~C.    *
851 ***************************************************************************/
852
853/* SW TCL elements, using Draw. */
854static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
855                                const struct pipe_draw_info *info)
856{
857    struct r300_context* r300 = r300_context(pipe);
858    struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
859    struct pipe_transfer *ib_transfer = NULL;
860    int i;
861    const void *indices = NULL;
862    boolean indexed = info->indexed;
863
864    if (r300->skip_rendering) {
865        return;
866    }
867
868    r300_update_derived_state(r300);
869
870    r300_reserve_cs_dwords(r300,
871            PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL |
872            (indexed ? PREP_INDEXED : 0),
873            indexed ? 256 : 6);
874
875    for (i = 0; i < r300->nr_vertex_buffers; i++) {
876        if (r300->vertex_buffer[i].user_buffer) {
877            draw_set_mapped_vertex_buffer(r300->draw, i,
878                                          r300->vertex_buffer[i].user_buffer);
879        } else if (r300->vertex_buffer[i].buffer) {
880            void *buf = pipe_buffer_map(pipe,
881                                  r300->vertex_buffer[i].buffer,
882                                  PIPE_TRANSFER_READ |
883                                  PIPE_TRANSFER_UNSYNCHRONIZED,
884                                  &vb_transfer[i]);
885            draw_set_mapped_vertex_buffer(r300->draw, i, buf);
886        }
887    }
888
889    if (indexed) {
890        if (r300->index_buffer.user_buffer) {
891            indices = r300->index_buffer.user_buffer;
892        } else {
893            indices = pipe_buffer_map(pipe, r300->index_buffer.buffer,
894                                      PIPE_TRANSFER_READ |
895                                      PIPE_TRANSFER_UNSYNCHRONIZED, &ib_transfer);
896        }
897    }
898
899    draw_set_mapped_index_buffer(r300->draw, indices);
900
901    r300->draw_vbo_locked = TRUE;
902    r300->draw_first_emitted = FALSE;
903    draw_vbo(r300->draw, info);
904    draw_flush(r300->draw);
905    r300->draw_vbo_locked = FALSE;
906
907    for (i = 0; i < r300->nr_vertex_buffers; i++) {
908        if (r300->vertex_buffer[i].buffer) {
909            if (vb_transfer[i])
910                pipe_buffer_unmap(pipe, vb_transfer[i]);
911            draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
912        }
913    }
914
915    if (indexed) {
916        if (ib_transfer)
917            pipe_buffer_unmap(pipe, ib_transfer);
918        draw_set_mapped_index_buffer(r300->draw, NULL);
919    }
920}
921
922/* Object for rendering using Draw. */
923struct r300_render {
924    /* Parent class */
925    struct vbuf_render base;
926
927    /* Pipe context */
928    struct r300_context* r300;
929
930    /* Vertex information */
931    size_t vertex_size;
932    unsigned prim;
933    unsigned hwprim;
934
935    /* VBO */
936    size_t vbo_max_used;
937    void * vbo_ptr;
938
939    struct pipe_transfer *vbo_transfer;
940};
941
942static INLINE struct r300_render*
943r300_render(struct vbuf_render* render)
944{
945    return (struct r300_render*)render;
946}
947
948static const struct vertex_info*
949r300_render_get_vertex_info(struct vbuf_render* render)
950{
951    struct r300_render* r300render = r300_render(render);
952    struct r300_context* r300 = r300render->r300;
953
954    return &r300->vertex_info;
955}
956
957static boolean r300_render_allocate_vertices(struct vbuf_render* render,
958                                                   ushort vertex_size,
959                                                   ushort count)
960{
961    struct r300_render* r300render = r300_render(render);
962    struct r300_context* r300 = r300render->r300;
963    struct pipe_screen* screen = r300->context.screen;
964    size_t size = (size_t)vertex_size * (size_t)count;
965
966    DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size);
967
968    if (size + r300->draw_vbo_offset > r300->draw_vbo_size)
969    {
970	pipe_resource_reference(&r300->vbo, NULL);
971        r300->vbo = pipe_buffer_create(screen,
972				       PIPE_BIND_VERTEX_BUFFER,
973				       PIPE_USAGE_STREAM,
974				       R300_MAX_DRAW_VBO_SIZE);
975        r300->draw_vbo_offset = 0;
976        r300->draw_vbo_size = R300_MAX_DRAW_VBO_SIZE;
977    }
978
979    r300render->vertex_size = vertex_size;
980
981    return (r300->vbo) ? TRUE : FALSE;
982}
983
984static void* r300_render_map_vertices(struct vbuf_render* render)
985{
986    struct r300_render* r300render = r300_render(render);
987    struct r300_context* r300 = r300render->r300;
988
989    assert(!r300render->vbo_transfer);
990
991    DBG(r300, DBG_DRAW, "r300: render_map_vertices\n");
992
993    r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context,
994					  r300->vbo,
995                                          PIPE_TRANSFER_WRITE |
996                                          PIPE_TRANSFER_UNSYNCHRONIZED,
997					  &r300render->vbo_transfer);
998
999    assert(r300render->vbo_ptr);
1000
1001    return ((uint8_t*)r300render->vbo_ptr + r300->draw_vbo_offset);
1002}
1003
1004static void r300_render_unmap_vertices(struct vbuf_render* render,
1005                                             ushort min,
1006                                             ushort max)
1007{
1008    struct r300_render* r300render = r300_render(render);
1009    struct pipe_context* context = &r300render->r300->context;
1010    struct r300_context* r300 = r300render->r300;
1011
1012    assert(r300render->vbo_transfer);
1013
1014    DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n");
1015
1016    r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
1017                                    r300render->vertex_size * (max + 1));
1018    pipe_buffer_unmap(context, r300render->vbo_transfer);
1019
1020    r300render->vbo_transfer = NULL;
1021}
1022
1023static void r300_render_release_vertices(struct vbuf_render* render)
1024{
1025    struct r300_render* r300render = r300_render(render);
1026    struct r300_context* r300 = r300render->r300;
1027
1028    DBG(r300, DBG_DRAW, "r300: render_release_vertices\n");
1029
1030    r300->draw_vbo_offset += r300render->vbo_max_used;
1031    r300render->vbo_max_used = 0;
1032}
1033
1034static void r300_render_set_primitive(struct vbuf_render* render,
1035                                      unsigned prim)
1036{
1037    struct r300_render* r300render = r300_render(render);
1038
1039    r300render->prim = prim;
1040    r300render->hwprim = r300_translate_primitive(prim);
1041}
1042
1043static void r300_render_draw_arrays(struct vbuf_render* render,
1044                                    unsigned start,
1045                                    unsigned count)
1046{
1047    struct r300_render* r300render = r300_render(render);
1048    struct r300_context* r300 = r300render->r300;
1049    uint8_t* ptr;
1050    unsigned i;
1051    unsigned dwords = 6;
1052
1053    CS_LOCALS(r300);
1054    (void) i; (void) ptr;
1055
1056    DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
1057
1058    if (r300->draw_first_emitted) {
1059        if (!r300_prepare_for_rendering(r300,
1060                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
1061                NULL, dwords, 0, 0, -1))
1062            return;
1063    } else {
1064        if (!r300_emit_states(r300,
1065                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
1066                NULL, 0, 0, -1))
1067            return;
1068    }
1069
1070    BEGIN_CS(dwords);
1071    OUT_CS_REG(R300_GA_COLOR_CONTROL,
1072            r300_provoking_vertex_fixes(r300, r300render->prim));
1073    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
1074    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
1075    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
1076           r300render->hwprim);
1077    END_CS;
1078
1079    r300->draw_first_emitted = TRUE;
1080}
1081
1082static void r300_render_draw_elements(struct vbuf_render* render,
1083                                      const ushort* indices,
1084                                      uint count)
1085{
1086    struct r300_render* r300render = r300_render(render);
1087    struct r300_context* r300 = r300render->r300;
1088    int i;
1089    unsigned end_cs_dwords;
1090    unsigned max_index = (r300->draw_vbo_size - r300->draw_vbo_offset) /
1091                         (r300render->r300->vertex_info.size * 4) - 1;
1092    unsigned short_count;
1093    unsigned free_dwords;
1094
1095    CS_LOCALS(r300);
1096    DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
1097
1098    if (r300->draw_first_emitted) {
1099        if (!r300_prepare_for_rendering(r300,
1100                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1101                NULL, 256, 0, 0, -1))
1102            return;
1103    } else {
1104        if (!r300_emit_states(r300,
1105                PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1106                NULL, 0, 0, -1))
1107            return;
1108    }
1109
1110    /* Below we manage the CS space manually because there may be more
1111     * indices than it can fit in CS. */
1112
1113    end_cs_dwords = r300_get_num_cs_end_dwords(r300);
1114
1115    while (count) {
1116        free_dwords = RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw;
1117
1118        short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2);
1119
1120        BEGIN_CS(6 + (short_count+1)/2);
1121        OUT_CS_REG(R300_GA_COLOR_CONTROL,
1122                r300_provoking_vertex_fixes(r300, r300render->prim));
1123        OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
1124        OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2);
1125        OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) |
1126               r300render->hwprim);
1127        for (i = 0; i < short_count-1; i += 2) {
1128            OUT_CS(indices[i+1] << 16 | indices[i]);
1129        }
1130        if (short_count % 2) {
1131            OUT_CS(indices[short_count-1]);
1132        }
1133        END_CS;
1134
1135        /* OK now subtract the emitted indices and see if we need to emit
1136         * another draw packet. */
1137        indices += short_count;
1138        count -= short_count;
1139
1140        if (count) {
1141            if (!r300_prepare_for_rendering(r300,
1142                    PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1143                    NULL, 256, 0, 0, -1))
1144                return;
1145
1146            end_cs_dwords = r300_get_num_cs_end_dwords(r300);
1147        }
1148    }
1149
1150    r300->draw_first_emitted = TRUE;
1151}
1152
1153static void r300_render_destroy(struct vbuf_render* render)
1154{
1155    FREE(render);
1156}
1157
1158static struct vbuf_render* r300_render_create(struct r300_context* r300)
1159{
1160    struct r300_render* r300render = CALLOC_STRUCT(r300_render);
1161
1162    r300render->r300 = r300;
1163
1164    r300render->base.max_vertex_buffer_bytes = 1024 * 1024;
1165    r300render->base.max_indices = 16 * 1024;
1166
1167    r300render->base.get_vertex_info = r300_render_get_vertex_info;
1168    r300render->base.allocate_vertices = r300_render_allocate_vertices;
1169    r300render->base.map_vertices = r300_render_map_vertices;
1170    r300render->base.unmap_vertices = r300_render_unmap_vertices;
1171    r300render->base.set_primitive = r300_render_set_primitive;
1172    r300render->base.draw_elements = r300_render_draw_elements;
1173    r300render->base.draw_arrays = r300_render_draw_arrays;
1174    r300render->base.release_vertices = r300_render_release_vertices;
1175    r300render->base.destroy = r300_render_destroy;
1176
1177    return &r300render->base;
1178}
1179
1180struct draw_stage* r300_draw_stage(struct r300_context* r300)
1181{
1182    struct vbuf_render* render;
1183    struct draw_stage* stage;
1184
1185    render = r300_render_create(r300);
1186
1187    if (!render) {
1188        return NULL;
1189    }
1190
1191    stage = draw_vbuf_stage(r300->draw, render);
1192
1193    if (!stage) {
1194        render->destroy(render);
1195        return NULL;
1196    }
1197
1198    draw_set_render(r300->draw, render);
1199
1200    return stage;
1201}
1202
1203void r300_draw_flush_vbuf(struct r300_context *r300)
1204{
1205    pipe_resource_reference(&r300->vbo, NULL);
1206    r300->draw_vbo_size = 0;
1207}
1208
1209/****************************************************************************
1210 *                         End of SW TCL functions                          *
1211 ***************************************************************************/
1212
1213/* This functions is used to draw a rectangle for the blitter module.
1214 *
1215 * If we rendered a quad, the pixels on the main diagonal
1216 * would be computed and stored twice, which makes the clear/copy codepaths
1217 * somewhat inefficient. Instead we use a rectangular point sprite. */
1218void r300_blitter_draw_rectangle(struct blitter_context *blitter,
1219                                 unsigned x1, unsigned y1,
1220                                 unsigned x2, unsigned y2,
1221                                 float depth,
1222                                 enum blitter_attrib_type type,
1223                                 const union pipe_color_union *attrib)
1224{
1225    struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
1226    unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
1227    unsigned width = x2 - x1;
1228    unsigned height = y2 - y1;
1229    unsigned vertex_size =
1230            type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
1231    unsigned dwords = 13 + vertex_size +
1232                      (type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0);
1233    static const union pipe_color_union zeros;
1234    CS_LOCALS(r300);
1235
1236    if (r300->skip_rendering)
1237        return;
1238
1239    r300->context.set_vertex_buffers(&r300->context, 0, NULL);
1240
1241    if (type == UTIL_BLITTER_ATTRIB_TEXCOORD)
1242        r300->sprite_coord_enable = 1;
1243
1244    r300_update_derived_state(r300);
1245
1246    /* Mark some states we don't care about as non-dirty. */
1247    r300->viewport_state.dirty = FALSE;
1248
1249    if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
1250        goto done;
1251
1252    DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
1253
1254    BEGIN_CS(dwords);
1255    /* Set up GA. */
1256    OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
1257
1258    if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) {
1259        /* Set up the GA to generate texcoords. */
1260        OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
1261                   (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
1262        OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
1263        OUT_CS_32F(attrib->f[0]);
1264        OUT_CS_32F(attrib->f[3]);
1265        OUT_CS_32F(attrib->f[2]);
1266        OUT_CS_32F(attrib->f[1]);
1267    }
1268
1269    /* Set up VAP controls. */
1270    OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
1271    OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
1272    OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
1273    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
1274    OUT_CS(1);
1275    OUT_CS(0);
1276
1277    /* Draw. */
1278    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
1279    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
1280           R300_VAP_VF_CNTL__PRIM_POINTS);
1281
1282    OUT_CS_32F(x1 + width * 0.5f);
1283    OUT_CS_32F(y1 + height * 0.5f);
1284    OUT_CS_32F(depth);
1285    OUT_CS_32F(1);
1286
1287    if (vertex_size == 8) {
1288        if (!attrib)
1289            attrib = &zeros;
1290        OUT_CS_TABLE(attrib->f, 4);
1291    }
1292    END_CS;
1293
1294done:
1295    /* Restore the state. */
1296    r300_mark_atom_dirty(r300, &r300->rs_state);
1297    r300_mark_atom_dirty(r300, &r300->viewport_state);
1298
1299    r300->sprite_coord_enable = last_sprite_coord_enable;
1300}
1301
1302static void r300_resource_resolve(struct pipe_context *pipe,
1303                                  const struct pipe_resolve_info *info)
1304{
1305    struct r300_context *r300 = r300_context(pipe);
1306    struct pipe_surface *srcsurf, *dstsurf, surf_tmpl;
1307    struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
1308    static const union pipe_color_union color;
1309
1310    assert(0 && "Resource resolve is unsupported, invalid call.");
1311
1312    memset(&surf_tmpl, 0, sizeof(surf_tmpl));
1313    surf_tmpl.format = info->src.res->format;
1314    surf_tmpl.u.tex.first_layer =
1315    surf_tmpl.u.tex.last_layer = info->src.layer;
1316    srcsurf = pipe->create_surface(pipe, info->src.res, &surf_tmpl);
1317    /* XXX Offset both surfaces by x0,y1. */
1318
1319    surf_tmpl.format = info->dst.res->format;
1320    surf_tmpl.u.tex.level = info->dst.level;
1321    surf_tmpl.u.tex.first_layer =
1322    surf_tmpl.u.tex.last_layer = info->dst.layer;
1323    dstsurf = pipe->create_surface(pipe, info->dst.res, &surf_tmpl);
1324
1325    DBG(r300, DBG_DRAW, "r300: Resolving resource...\n");
1326
1327    /* Enable AA resolve. */
1328    aa->dest = r300_surface(dstsurf);
1329    aa->aaresolve_ctl =
1330        R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE |
1331        R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE;
1332    r300->aa_state.size = 10;
1333    r300_mark_atom_dirty(r300, &r300->aa_state);
1334
1335    /* Resolve the surface. */
1336    /* XXX: y1 < 0 ==> Y flip */
1337    r300->context.clear_render_target(pipe,
1338                                      srcsurf, &color, 0, 0,
1339                                      info->dst.x1 - info->dst.x0,
1340                                      info->dst.y1 - info->dst.y0);
1341
1342    /* Disable AA resolve. */
1343    aa->dest = NULL;
1344    aa->aaresolve_ctl = 0;
1345    r300->aa_state.size = 4;
1346    r300_mark_atom_dirty(r300, &r300->aa_state);
1347
1348    pipe_surface_reference(&srcsurf, NULL);
1349    pipe_surface_reference(&dstsurf, NULL);
1350}
1351
1352void r300_init_render_functions(struct r300_context *r300)
1353{
1354    /* Set draw functions based on presence of HW TCL. */
1355    if (r300->screen->caps.has_tcl) {
1356        r300->context.draw_vbo = r300_draw_vbo;
1357    } else {
1358        r300->context.draw_vbo = r300_swtcl_draw_vbo;
1359    }
1360
1361    r300->context.resource_resolve = r300_resource_resolve;
1362
1363    /* Plug in the two-sided stencil reference value fallback if needed. */
1364    if (!r300->screen->caps.is_r500)
1365        r300_plug_in_stencil_ref_fallback(r300);
1366}
1367