2 * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2010 Marek Olšák <maraeo@gmail.com>
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
24 /* r300_render: Vertex and index buffer primitive emission. Contains both
25 * HW TCL fastpath rendering, and SW TCL Draw-assisted rendering. */
27 #include "draw/draw_context.h"
28 #include "draw/draw_vbuf.h"
30 #include "util/u_inlines.h"
32 #include "util/u_format.h"
33 #include "util/u_memory.h"
34 #include "util/u_upload_mgr.h"
35 #include "util/u_prim.h"
38 #include "r300_context.h"
39 #include "r300_screen_buffer.h"
40 #include "r300_emit.h"
45 #define IMMD_DWORDS 32
47 static uint32_t r300_translate_primitive(unsigned prim)
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,
65 unsigned hwprim = prim_conv[prim];
71 static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
74 struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
75 uint32_t color_control = rs->color_control;
77 /* By default (see r300_state.c:r300_create_rs_state) color_control is
78 * initialized to provoking the first vertex.
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)
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.
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.
95 if (rs->rs.flatshade_first) {
97 case PIPE_PRIM_TRIANGLE_FAN:
98 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
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;
106 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
110 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
113 return color_control;
116 void r500_emit_index_bias(struct r300_context *r300, int index_bias)
121 OUT_CS_REG(R500_VAP_INDEX_OFFSET,
122 (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0));
126 static void r300_emit_draw_init(struct r300_context *r300, unsigned mode,
127 unsigned min_index, unsigned max_index)
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);
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. */
146 static void r300_split_index_bias(struct r300_context *r300, int index_bias,
147 int *buffer_offset, int *index_offset)
149 struct pipe_vertex_buffer *vb, *vbufs = r300->vbuf_mgr->real_vertex_buffer;
150 struct pipe_vertex_element *velem = r300->velems->velem;
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
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);
165 /* Now set the minimum allowed value. */
166 *buffer_offset = MAX2(-max_neg_bias, index_bias);
168 /* A positive index bias is OK. */
169 *buffer_offset = index_bias;
172 *index_offset = index_bias - *buffer_offset;
175 enum 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? */
184 * Check if the requested number of dwords is available in the CS and
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
191 static boolean r300_reserve_cs_dwords(struct r300_context *r300,
192 enum r300_prepare_flags flags,
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;
200 /* Add dirty state, index offset, and AOS. */
202 cs_dwords += r300_get_num_dirty_dwords(r300);
204 if (r300->screen->caps.is_r500)
205 cs_dwords += 2; /* emit_index_offset */
207 if (emit_vertex_arrays)
208 cs_dwords += 55; /* emit_vertex_arrays */
210 if (emit_vertex_arrays_swtcl)
211 cs_dwords += 7; /* emit_vertex_arrays_swtcl */
213 cs_dwords += r300_get_num_cs_end_dwords(r300);
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);
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
234 static boolean r300_emit_states(struct r300_context *r300,
235 enum r300_prepare_flags flags,
236 struct pipe_resource *index_buffer,
238 int index_bias, int instance_id)
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;
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,
250 fprintf(stderr, "r300: CS space validation failed. "
251 "(not enough memory?) Skipping rendering.\n");
257 r300_emit_dirty_state(r300);
259 if (r300->screen->caps.is_r500) {
260 if (r300->screen->caps.has_tcl)
261 r500_emit_index_bias(r300, index_bias);
263 r500_emit_index_bias(r300, 0);
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);
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;
279 if (emit_vertex_arrays_swtcl)
280 r300_emit_vertex_arrays_swtcl(r300, indexed);
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
297 static boolean r300_prepare_for_rendering(struct r300_context *r300,
298 enum r300_prepare_flags flags,
299 struct pipe_resource *index_buffer,
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;
309 return r300_emit_states(r300, flags, index_buffer, buffer_offset,
310 index_bias, instance_id);
313 static boolean immd_is_good_idea(struct r300_context *r300,
316 struct pipe_vertex_element* velem;
317 struct pipe_resource *buf;
318 boolean checked[PIPE_MAX_ATTRIBS] = {0};
319 unsigned vertex_element_count = r300->velems->count;
322 if (DBG_ON(r300, DBG_NO_IMMD)) {
330 if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) {
334 /* We shouldn't map buffers referenced by CS, busy buffers,
335 * and ones placed in VRAM. */
336 for (i = 0; i < vertex_element_count; i++) {
337 velem = &r300->velems->velem[i];
338 vbi = velem->vertex_buffer_index;
341 buf = r300->vbuf_mgr->real_vertex_buffer[vbi].buffer;
343 if ((r300_resource(buf)->domain != RADEON_DOMAIN_GTT)) {
353 /*****************************************************************************
354 * The HWTCL draw functions. *
355 ****************************************************************************/
357 static void r300_draw_arrays_immediate(struct r300_context *r300,
358 const struct pipe_draw_info *info)
360 struct pipe_vertex_element* velem;
361 struct pipe_vertex_buffer* vbuf;
362 unsigned vertex_element_count = r300->velems->count;
365 /* Size of the vertex, in dwords. */
366 unsigned vertex_size = r300->velems->vertex_size_dwords;
368 /* The number of dwords for this draw operation. */
369 unsigned dwords = 4 + info->count * vertex_size;
371 /* Size of the vertex element, in dwords. */
372 unsigned size[PIPE_MAX_ATTRIBS];
374 /* Stride to the same attrib in the next vertex in the vertex buffer,
376 unsigned stride[PIPE_MAX_ATTRIBS];
378 /* Mapped vertex buffers. */
379 uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
380 uint32_t* mapelem[PIPE_MAX_ATTRIBS];
384 if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
387 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
388 for (i = 0; i < vertex_element_count; i++) {
389 velem = &r300->velems->velem[i];
390 size[i] = r300->velems->format_size[i] / 4;
391 vbi = velem->vertex_buffer_index;
392 vbuf = &r300->vbuf_mgr->real_vertex_buffer[vbi];
393 stride[i] = vbuf->stride / 4;
395 /* Map the buffer. */
397 map[vbi] = (uint32_t*)r300->rws->buffer_map(
398 r300_resource(vbuf->buffer)->buf,
399 r300->cs, PIPE_TRANSFER_READ | PIPE_TRANSFER_UNSYNCHRONIZED);
400 map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * info->start;
402 mapelem[i] = map[vbi] + (velem->src_offset / 4);
405 r300_emit_draw_init(r300, info->mode, 0, info->count-1);
408 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
409 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, info->count * vertex_size);
410 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (info->count << 16) |
411 r300_translate_primitive(info->mode));
414 for (v = 0; v < info->count; v++) {
415 for (i = 0; i < vertex_element_count; i++) {
416 OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
422 for (i = 0; i < vertex_element_count; i++) {
423 vbi = r300->velems->velem[i].vertex_buffer_index;
426 r300->rws->buffer_unmap(r300_resource(r300->vbuf_mgr->real_vertex_buffer[vbi].buffer)->buf);
432 static void r300_emit_draw_arrays(struct r300_context *r300,
436 boolean alt_num_verts = count > 65535;
439 if (count >= (1 << 24)) {
440 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
441 "refusing to render.\n", count);
445 r300_emit_draw_init(r300, mode, 0, count-1);
447 BEGIN_CS(2 + (alt_num_verts ? 2 : 0));
449 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
451 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
452 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
453 r300_translate_primitive(mode) |
454 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
458 static void r300_emit_draw_elements(struct r300_context *r300,
459 struct pipe_resource* indexBuffer,
466 uint16_t *imm_indices3)
468 uint32_t count_dwords, offset_dwords;
469 boolean alt_num_verts = count > 65535;
472 if (count >= (1 << 24) || max_index >= (1 << 24)) {
473 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
474 "refusing to render (max_index: %i).\n", count, max_index);
478 DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
479 count, min_index, max_index);
481 r300_emit_draw_init(r300, mode, min_index, max_index);
483 /* If start is odd, render the first triangle with indices embedded
484 * in the command stream. This will increase start by 3 and make it
485 * even. We can then proceed without a fallback. */
486 if (indexSize == 2 && (start & 1) &&
487 mode == PIPE_PRIM_TRIANGLES) {
489 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 2);
490 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (3 << 16) |
491 R300_VAP_VF_CNTL__PRIM_TRIANGLES);
492 OUT_CS(imm_indices3[1] << 16 | imm_indices3[0]);
493 OUT_CS(imm_indices3[2]);
502 offset_dwords = indexSize * start / sizeof(uint32_t);
504 BEGIN_CS(8 + (alt_num_verts ? 2 : 0));
506 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
508 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
509 if (indexSize == 4) {
510 count_dwords = count;
511 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
512 R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
513 r300_translate_primitive(mode) |
514 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
516 count_dwords = (count + 1) / 2;
517 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
518 r300_translate_primitive(mode) |
519 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
522 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
523 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
524 (0 << R300_INDX_BUFFER_SKIP_SHIFT));
525 OUT_CS(offset_dwords << 2);
526 OUT_CS(count_dwords);
527 OUT_CS_RELOC(r300_resource(indexBuffer));
531 static void r300_draw_elements_immediate(struct r300_context *r300,
532 const struct pipe_draw_info *info)
537 unsigned index_size = r300->index_buffer.index_size;
538 unsigned i, count_dwords = index_size == 4 ? info->count :
539 (info->count + 1) / 2;
542 /* 19 dwords for r300_draw_elements_immediate. Give up if the function fails. */
543 if (!r300_prepare_for_rendering(r300,
544 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
545 PREP_INDEXED, NULL, 2+count_dwords, 0, info->index_bias, -1))
548 r300_emit_draw_init(r300, info->mode, info->min_index, info->max_index);
550 BEGIN_CS(2 + count_dwords);
551 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords);
553 switch (index_size) {
555 ptr1 = r300_resource(r300->index_buffer.buffer)->b.user_ptr;
558 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
559 r300_translate_primitive(info->mode));
561 if (info->index_bias && !r300->screen->caps.is_r500) {
562 for (i = 0; i < info->count-1; i += 2)
563 OUT_CS(((ptr1[i+1] + info->index_bias) << 16) |
564 (ptr1[i] + info->index_bias));
567 OUT_CS(ptr1[i] + info->index_bias);
569 for (i = 0; i < info->count-1; i += 2)
570 OUT_CS(((ptr1[i+1]) << 16) |
579 ptr2 = (uint16_t*)r300_resource(r300->index_buffer.buffer)->b.user_ptr;
582 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
583 r300_translate_primitive(info->mode));
585 if (info->index_bias && !r300->screen->caps.is_r500) {
586 for (i = 0; i < info->count-1; i += 2)
587 OUT_CS(((ptr2[i+1] + info->index_bias) << 16) |
588 (ptr2[i] + info->index_bias));
591 OUT_CS(ptr2[i] + info->index_bias);
593 OUT_CS_TABLE(ptr2, count_dwords);
598 ptr4 = (uint32_t*)r300_resource(r300->index_buffer.buffer)->b.user_ptr;
601 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
602 R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
603 r300_translate_primitive(info->mode));
605 if (info->index_bias && !r300->screen->caps.is_r500) {
606 for (i = 0; i < info->count; i++)
607 OUT_CS(ptr4[i] + info->index_bias);
609 OUT_CS_TABLE(ptr4, count_dwords);
616 static void r300_draw_elements(struct r300_context *r300,
617 const struct pipe_draw_info *info,
620 struct pipe_resource *indexBuffer = r300->index_buffer.buffer;
621 unsigned indexSize = r300->index_buffer.index_size;
622 struct pipe_resource* orgIndexBuffer = indexBuffer;
623 unsigned start = info->start;
624 unsigned count = info->count;
625 boolean alt_num_verts = r300->screen->caps.is_r500 &&
627 unsigned short_count;
628 int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
629 uint16_t indices3[3];
631 if (info->index_bias && !r300->screen->caps.is_r500) {
632 r300_split_index_bias(r300, info->index_bias, &buffer_offset, &index_offset);
635 r300_translate_index_buffer(r300, &indexBuffer, &indexSize, index_offset,
638 /* Fallback for misaligned ushort indices. */
639 if (indexSize == 2 && (start & 1) &&
640 !r300_resource(indexBuffer)->b.user_ptr) {
641 /* If we got here, then orgIndexBuffer == indexBuffer. */
642 uint16_t *ptr = r300->rws->buffer_map(r300_resource(orgIndexBuffer)->buf,
645 PIPE_TRANSFER_UNSYNCHRONIZED);
647 if (info->mode == PIPE_PRIM_TRIANGLES) {
648 memcpy(indices3, ptr + start, 6);
650 /* Copy the mapped index buffer directly to the upload buffer.
651 * The start index will be aligned simply from the fact that
652 * every sub-buffer in the upload buffer is aligned. */
653 r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start,
654 count, (uint8_t*)ptr);
656 r300->rws->buffer_unmap(r300_resource(orgIndexBuffer)->buf);
658 if (r300_resource(indexBuffer)->b.user_ptr)
659 r300_upload_index_buffer(r300, &indexBuffer, indexSize,
661 r300_resource(indexBuffer)->b.user_ptr);
664 /* 19 dwords for emit_draw_elements. Give up if the function fails. */
665 if (!r300_prepare_for_rendering(r300,
666 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
667 PREP_INDEXED, indexBuffer, 19, buffer_offset, info->index_bias,
671 if (alt_num_verts || count <= 65535) {
672 r300_emit_draw_elements(r300, indexBuffer, indexSize, info->min_index,
673 info->max_index, info->mode, start, count,
677 /* The maximum must be divisible by 4 and 3,
678 * so that quad and triangle lists are split correctly.
680 * Strips, loops, and fans won't work. */
681 short_count = MIN2(count, 65532);
683 r300_emit_draw_elements(r300, indexBuffer, indexSize,
684 info->min_index, info->max_index,
685 info->mode, start, short_count, indices3);
687 start += short_count;
688 count -= short_count;
690 /* 15 dwords for emit_draw_elements */
692 if (!r300_prepare_for_rendering(r300,
693 PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS | PREP_INDEXED,
694 indexBuffer, 19, buffer_offset, info->index_bias,
702 if (indexBuffer != orgIndexBuffer) {
703 pipe_resource_reference( &indexBuffer, NULL );
707 static void r300_draw_arrays(struct r300_context *r300,
708 const struct pipe_draw_info *info,
711 boolean alt_num_verts = r300->screen->caps.is_r500 &&
713 unsigned start = info->start;
714 unsigned count = info->count;
715 unsigned short_count;
717 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
718 if (!r300_prepare_for_rendering(r300,
719 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS,
720 NULL, 9, start, 0, instance_id))
723 if (alt_num_verts || count <= 65535) {
724 r300_emit_draw_arrays(r300, info->mode, count);
727 /* The maximum must be divisible by 4 and 3,
728 * so that quad and triangle lists are split correctly.
730 * Strips, loops, and fans won't work. */
731 short_count = MIN2(count, 65532);
732 r300_emit_draw_arrays(r300, info->mode, short_count);
734 start += short_count;
735 count -= short_count;
737 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
739 if (!r300_prepare_for_rendering(r300,
740 PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS, NULL, 9,
741 start, 0, instance_id))
748 static void r300_draw_arrays_instanced(struct r300_context *r300,
749 const struct pipe_draw_info *info)
753 for (i = 0; i < info->instance_count; i++)
754 r300_draw_arrays(r300, info, i);
757 static void r300_draw_elements_instanced(struct r300_context *r300,
758 const struct pipe_draw_info *info)
762 for (i = 0; i < info->instance_count; i++)
763 r300_draw_elements(r300, info, i);
766 static void r300_draw_vbo(struct pipe_context* pipe,
767 const struct pipe_draw_info *dinfo)
769 struct r300_context* r300 = r300_context(pipe);
770 struct pipe_draw_info info = *dinfo;
772 info.indexed = info.indexed && r300->index_buffer.buffer;
774 if (r300->skip_rendering ||
775 !u_trim_pipe_prim(info.mode, &info.count)) {
779 r300_update_derived_state(r300);
781 /* Start the vbuf manager and update buffers if needed. */
782 if (u_vbuf_draw_begin(r300->vbuf_mgr, &info) & U_VBUF_BUFFERS_UPDATED) {
783 r300->vertex_arrays_dirty = TRUE;
788 info.start += r300->index_buffer.offset;
789 info.max_index = MIN2(r300->vbuf_mgr->max_index, info.max_index);
791 if (info.instance_count <= 1) {
792 if (info.count <= 8 &&
793 r300_resource(r300->index_buffer.buffer)->b.user_ptr) {
794 r300_draw_elements_immediate(r300, &info);
796 r300_draw_elements(r300, &info, -1);
799 r300_draw_elements_instanced(r300, &info);
802 if (info.instance_count <= 1) {
803 if (immd_is_good_idea(r300, info.count)) {
804 r300_draw_arrays_immediate(r300, &info);
806 r300_draw_arrays(r300, &info, -1);
809 r300_draw_arrays_instanced(r300, &info);
813 u_vbuf_draw_end(r300->vbuf_mgr);
816 /****************************************************************************
817 * The rest of this file is for SW TCL rendering only. Please be polite and *
818 * keep these functions separated so that they are easier to locate. ~C. *
819 ***************************************************************************/
821 /* SW TCL elements, using Draw. */
822 static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
823 const struct pipe_draw_info *info)
825 struct r300_context* r300 = r300_context(pipe);
826 struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
827 struct pipe_transfer *ib_transfer = NULL;
829 void *indices = NULL;
830 boolean indexed = info->indexed && r300->index_buffer.buffer;
832 if (r300->skip_rendering) {
836 r300_update_derived_state(r300);
838 r300_reserve_cs_dwords(r300,
839 PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL |
840 (indexed ? PREP_INDEXED : 0),
843 for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) {
844 if (r300->vbuf_mgr->vertex_buffer[i].buffer) {
845 void *buf = pipe_buffer_map(pipe,
846 r300->vbuf_mgr->vertex_buffer[i].buffer,
848 PIPE_TRANSFER_UNSYNCHRONIZED,
850 draw_set_mapped_vertex_buffer(r300->draw, i, buf);
855 indices = pipe_buffer_map(pipe, r300->index_buffer.buffer,
857 PIPE_TRANSFER_UNSYNCHRONIZED, &ib_transfer);
860 draw_set_mapped_index_buffer(r300->draw, indices);
862 r300->draw_vbo_locked = TRUE;
863 r300->draw_first_emitted = FALSE;
864 draw_vbo(r300->draw, info);
865 draw_flush(r300->draw);
866 r300->draw_vbo_locked = FALSE;
868 for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) {
869 if (r300->vbuf_mgr->vertex_buffer[i].buffer) {
870 pipe_buffer_unmap(pipe, vb_transfer[i]);
871 draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
876 pipe_buffer_unmap(pipe, ib_transfer);
877 draw_set_mapped_index_buffer(r300->draw, NULL);
881 /* Object for rendering using Draw. */
884 struct vbuf_render base;
887 struct r300_context* r300;
889 /* Vertex information */
898 struct pipe_transfer *vbo_transfer;
901 static INLINE struct r300_render*
902 r300_render(struct vbuf_render* render)
904 return (struct r300_render*)render;
907 static const struct vertex_info*
908 r300_render_get_vertex_info(struct vbuf_render* render)
910 struct r300_render* r300render = r300_render(render);
911 struct r300_context* r300 = r300render->r300;
913 return &r300->vertex_info;
916 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
920 struct r300_render* r300render = r300_render(render);
921 struct r300_context* r300 = r300render->r300;
922 struct pipe_screen* screen = r300->context.screen;
923 size_t size = (size_t)vertex_size * (size_t)count;
925 DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size);
927 if (size + r300->draw_vbo_offset > r300->draw_vbo_size)
929 pipe_resource_reference(&r300->vbo, NULL);
930 r300->vbo = pipe_buffer_create(screen,
931 PIPE_BIND_VERTEX_BUFFER,
933 R300_MAX_DRAW_VBO_SIZE);
934 r300->draw_vbo_offset = 0;
935 r300->draw_vbo_size = R300_MAX_DRAW_VBO_SIZE;
938 r300render->vertex_size = vertex_size;
940 return (r300->vbo) ? TRUE : FALSE;
943 static void* r300_render_map_vertices(struct vbuf_render* render)
945 struct r300_render* r300render = r300_render(render);
946 struct r300_context* r300 = r300render->r300;
948 assert(!r300render->vbo_transfer);
950 DBG(r300, DBG_DRAW, "r300: render_map_vertices\n");
952 r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context,
954 PIPE_TRANSFER_WRITE |
955 PIPE_TRANSFER_UNSYNCHRONIZED,
956 &r300render->vbo_transfer);
958 assert(r300render->vbo_ptr);
960 return ((uint8_t*)r300render->vbo_ptr + r300->draw_vbo_offset);
963 static void r300_render_unmap_vertices(struct vbuf_render* render,
967 struct r300_render* r300render = r300_render(render);
968 struct pipe_context* context = &r300render->r300->context;
969 struct r300_context* r300 = r300render->r300;
971 assert(r300render->vbo_transfer);
973 DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n");
975 r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
976 r300render->vertex_size * (max + 1));
977 pipe_buffer_unmap(context, r300render->vbo_transfer);
979 r300render->vbo_transfer = NULL;
982 static void r300_render_release_vertices(struct vbuf_render* render)
984 struct r300_render* r300render = r300_render(render);
985 struct r300_context* r300 = r300render->r300;
987 DBG(r300, DBG_DRAW, "r300: render_release_vertices\n");
989 r300->draw_vbo_offset += r300render->vbo_max_used;
990 r300render->vbo_max_used = 0;
993 static boolean r300_render_set_primitive(struct vbuf_render* render,
996 struct r300_render* r300render = r300_render(render);
998 r300render->prim = prim;
999 r300render->hwprim = r300_translate_primitive(prim);
1004 static void r300_render_draw_arrays(struct vbuf_render* render,
1008 struct r300_render* r300render = r300_render(render);
1009 struct r300_context* r300 = r300render->r300;
1012 unsigned dwords = 6;
1015 (void) i; (void) ptr;
1017 DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
1019 if (r300->draw_first_emitted) {
1020 if (!r300_prepare_for_rendering(r300,
1021 PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
1022 NULL, dwords, 0, 0, -1))
1025 if (!r300_emit_states(r300,
1026 PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
1032 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1033 r300_provoking_vertex_fixes(r300, r300render->prim));
1034 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
1035 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
1036 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
1037 r300render->hwprim);
1040 r300->draw_first_emitted = TRUE;
1043 static void r300_render_draw_elements(struct vbuf_render* render,
1044 const ushort* indices,
1047 struct r300_render* r300render = r300_render(render);
1048 struct r300_context* r300 = r300render->r300;
1050 unsigned end_cs_dwords;
1051 unsigned max_index = (r300->draw_vbo_size - r300->draw_vbo_offset) /
1052 (r300render->r300->vertex_info.size * 4) - 1;
1053 unsigned short_count;
1054 unsigned free_dwords;
1057 DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
1059 if (r300->draw_first_emitted) {
1060 if (!r300_prepare_for_rendering(r300,
1061 PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1062 NULL, 256, 0, 0, -1))
1065 if (!r300_emit_states(r300,
1066 PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1071 /* Below we manage the CS space manually because there may be more
1072 * indices than it can fit in CS. */
1074 end_cs_dwords = r300_get_num_cs_end_dwords(r300);
1077 free_dwords = RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw;
1079 short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2);
1081 BEGIN_CS(6 + (short_count+1)/2);
1082 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1083 r300_provoking_vertex_fixes(r300, r300render->prim));
1084 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
1085 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2);
1086 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) |
1087 r300render->hwprim);
1088 for (i = 0; i < short_count-1; i += 2) {
1089 OUT_CS(indices[i+1] << 16 | indices[i]);
1091 if (short_count % 2) {
1092 OUT_CS(indices[short_count-1]);
1096 /* OK now subtract the emitted indices and see if we need to emit
1097 * another draw packet. */
1098 indices += short_count;
1099 count -= short_count;
1102 if (!r300_prepare_for_rendering(r300,
1103 PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1104 NULL, 256, 0, 0, -1))
1107 end_cs_dwords = r300_get_num_cs_end_dwords(r300);
1111 r300->draw_first_emitted = TRUE;
1114 static void r300_render_destroy(struct vbuf_render* render)
1119 static struct vbuf_render* r300_render_create(struct r300_context* r300)
1121 struct r300_render* r300render = CALLOC_STRUCT(r300_render);
1123 r300render->r300 = r300;
1125 r300render->base.max_vertex_buffer_bytes = 1024 * 1024;
1126 r300render->base.max_indices = 16 * 1024;
1128 r300render->base.get_vertex_info = r300_render_get_vertex_info;
1129 r300render->base.allocate_vertices = r300_render_allocate_vertices;
1130 r300render->base.map_vertices = r300_render_map_vertices;
1131 r300render->base.unmap_vertices = r300_render_unmap_vertices;
1132 r300render->base.set_primitive = r300_render_set_primitive;
1133 r300render->base.draw_elements = r300_render_draw_elements;
1134 r300render->base.draw_arrays = r300_render_draw_arrays;
1135 r300render->base.release_vertices = r300_render_release_vertices;
1136 r300render->base.destroy = r300_render_destroy;
1138 return &r300render->base;
1141 struct draw_stage* r300_draw_stage(struct r300_context* r300)
1143 struct vbuf_render* render;
1144 struct draw_stage* stage;
1146 render = r300_render_create(r300);
1152 stage = draw_vbuf_stage(r300->draw, render);
1155 render->destroy(render);
1159 draw_set_render(r300->draw, render);
1164 void r300_draw_flush_vbuf(struct r300_context *r300)
1166 pipe_resource_reference(&r300->vbo, NULL);
1167 r300->draw_vbo_size = 0;
1170 /****************************************************************************
1171 * End of SW TCL functions *
1172 ***************************************************************************/
1174 /* This functions is used to draw a rectangle for the blitter module.
1176 * If we rendered a quad, the pixels on the main diagonal
1177 * would be computed and stored twice, which makes the clear/copy codepaths
1178 * somewhat inefficient. Instead we use a rectangular point sprite. */
1179 static void r300_blitter_draw_rectangle(struct blitter_context *blitter,
1180 unsigned x1, unsigned y1,
1181 unsigned x2, unsigned y2,
1183 enum blitter_attrib_type type,
1184 const float attrib[4])
1186 struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
1187 unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
1188 unsigned width = x2 - x1;
1189 unsigned height = y2 - y1;
1190 unsigned vertex_size =
1191 type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
1192 unsigned dwords = 13 + vertex_size +
1193 (type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0);
1194 const float zeros[4] = {0, 0, 0, 0};
1197 if (r300->skip_rendering)
1200 r300->context.set_vertex_buffers(&r300->context, 0, NULL);
1202 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD)
1203 r300->sprite_coord_enable = 1;
1205 r300_update_derived_state(r300);
1207 /* Mark some states we don't care about as non-dirty. */
1208 r300->clip_state.dirty = FALSE;
1209 r300->viewport_state.dirty = FALSE;
1211 if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
1214 DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
1218 OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
1220 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) {
1221 /* Set up the GA to generate texcoords. */
1222 OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
1223 (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
1224 OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
1225 OUT_CS_32F(attrib[0]);
1226 OUT_CS_32F(attrib[3]);
1227 OUT_CS_32F(attrib[2]);
1228 OUT_CS_32F(attrib[1]);
1231 /* Set up VAP controls. */
1232 OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
1233 OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
1234 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
1235 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
1240 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
1241 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
1242 R300_VAP_VF_CNTL__PRIM_POINTS);
1244 OUT_CS_32F(x1 + width * 0.5f);
1245 OUT_CS_32F(y1 + height * 0.5f);
1249 if (vertex_size == 8) {
1252 OUT_CS_TABLE(attrib, 4);
1257 /* Restore the state. */
1258 r300_mark_atom_dirty(r300, &r300->clip_state);
1259 r300_mark_atom_dirty(r300, &r300->rs_state);
1260 r300_mark_atom_dirty(r300, &r300->viewport_state);
1262 r300->sprite_coord_enable = last_sprite_coord_enable;
1265 static void r300_resource_resolve(struct pipe_context* pipe,
1266 struct pipe_resource* dest,
1268 struct pipe_resource* src,
1271 struct r300_context* r300 = r300_context(pipe);
1272 struct pipe_surface* srcsurf, surf_tmpl;
1273 struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
1274 float color[] = {0, 0, 0, 0};
1276 memset(&surf_tmpl, 0, sizeof(surf_tmpl));
1277 surf_tmpl.format = src->format;
1278 surf_tmpl.usage = 0; /* not really a surface hence no bind flags */
1279 surf_tmpl.u.tex.level = 0; /* msaa resources cannot have mipmaps */
1280 surf_tmpl.u.tex.first_layer = src_layer;
1281 surf_tmpl.u.tex.last_layer = src_layer;
1282 srcsurf = pipe->create_surface(pipe, src, &surf_tmpl);
1283 surf_tmpl.format = dest->format;
1284 surf_tmpl.u.tex.first_layer = dst_layer;
1285 surf_tmpl.u.tex.last_layer = dst_layer;
1287 DBG(r300, DBG_DRAW, "r300: Resolving resource...\n");
1289 /* Enable AA resolve. */
1290 aa->dest = r300_surface(pipe->create_surface(pipe, dest, &surf_tmpl));
1293 R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE |
1294 R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE;
1295 r300->aa_state.size = 10;
1296 r300_mark_atom_dirty(r300, &r300->aa_state);
1298 /* Resolve the surface. */
1299 r300->context.clear_render_target(pipe,
1300 srcsurf, color, 0, 0, src->width0, src->height0);
1302 /* Disable AA resolve. */
1303 aa->aaresolve_ctl = 0;
1304 r300->aa_state.size = 4;
1305 r300_mark_atom_dirty(r300, &r300->aa_state);
1307 pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL);
1308 pipe_surface_reference((struct pipe_surface**)&aa->dest, NULL);
1311 void r300_init_render_functions(struct r300_context *r300)
1313 /* Set draw functions based on presence of HW TCL. */
1314 if (r300->screen->caps.has_tcl) {
1315 r300->context.draw_vbo = r300_draw_vbo;
1317 r300->context.draw_vbo = r300_swtcl_draw_vbo;
1320 r300->context.resource_resolve = r300_resource_resolve;
1321 r300->blitter->draw_rectangle = r300_blitter_draw_rectangle;
1323 /* Plug in the two-sided stencil reference value fallback if needed. */
1324 if (!r300->screen->caps.is_r500)
1325 r300_plug_in_stencil_ref_fallback(r300);