svga: avoid emitting redundant SetIndexBuffer commands
[mesa.git] / src / gallium / drivers / svga / svga_draw.c
1 /**********************************************************
2 * Copyright 2008-2009 VMware, Inc. All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26 #include "pipe/p_compiler.h"
27 #include "util/u_inlines.h"
28 #include "pipe/p_defines.h"
29 #include "util/u_helpers.h"
30 #include "util/u_memory.h"
31 #include "util/u_math.h"
32
33 #include "svga_context.h"
34 #include "svga_draw.h"
35 #include "svga_draw_private.h"
36 #include "svga_debug.h"
37 #include "svga_screen.h"
38 #include "svga_resource.h"
39 #include "svga_resource_buffer.h"
40 #include "svga_resource_texture.h"
41 #include "svga_shader.h"
42 #include "svga_surface.h"
43 #include "svga_winsys.h"
44 #include "svga_cmd.h"
45
46
47 struct svga_hwtnl *
48 svga_hwtnl_create(struct svga_context *svga)
49 {
50 struct svga_hwtnl *hwtnl = CALLOC_STRUCT(svga_hwtnl);
51 if (!hwtnl)
52 goto fail;
53
54 hwtnl->svga = svga;
55
56 hwtnl->cmd.swc = svga->swc;
57
58 return hwtnl;
59
60 fail:
61 return NULL;
62 }
63
64
65 void
66 svga_hwtnl_destroy(struct svga_hwtnl *hwtnl)
67 {
68 unsigned i, j;
69
70 for (i = 0; i < PIPE_PRIM_MAX; i++) {
71 for (j = 0; j < IDX_CACHE_MAX; j++) {
72 pipe_resource_reference(&hwtnl->index_cache[i][j].buffer, NULL);
73 }
74 }
75
76 for (i = 0; i < hwtnl->cmd.vbuf_count; i++)
77 pipe_resource_reference(&hwtnl->cmd.vbufs[i].buffer, NULL);
78
79 for (i = 0; i < hwtnl->cmd.prim_count; i++)
80 pipe_resource_reference(&hwtnl->cmd.prim_ib[i], NULL);
81
82 FREE(hwtnl);
83 }
84
85
86 void
87 svga_hwtnl_set_flatshade(struct svga_hwtnl *hwtnl,
88 boolean flatshade, boolean flatshade_first)
89 {
90 struct svga_screen *svgascreen = svga_screen(hwtnl->svga->pipe.screen);
91
92 /* User-specified PV */
93 hwtnl->api_pv = (flatshade && !flatshade_first) ? PV_LAST : PV_FIRST;
94
95 /* Device supported PV */
96 if (svgascreen->haveProvokingVertex) {
97 /* use the mode specified by the user */
98 hwtnl->hw_pv = hwtnl->api_pv;
99 }
100 else {
101 /* the device only support first provoking vertex */
102 hwtnl->hw_pv = PV_FIRST;
103 }
104 }
105
106
107 void
108 svga_hwtnl_set_fillmode(struct svga_hwtnl *hwtnl, unsigned mode)
109 {
110 hwtnl->api_fillmode = mode;
111 }
112
113
114 void
115 svga_hwtnl_vertex_decls(struct svga_hwtnl *hwtnl,
116 unsigned count,
117 const SVGA3dVertexDecl * decls,
118 const unsigned *buffer_indexes,
119 SVGA3dElementLayoutId layout_id)
120 {
121 assert(hwtnl->cmd.prim_count == 0);
122 hwtnl->cmd.vdecl_count = count;
123 hwtnl->cmd.vdecl_layout_id = layout_id;
124 memcpy(hwtnl->cmd.vdecl, decls, count * sizeof(*decls));
125 memcpy(hwtnl->cmd.vdecl_buffer_index, buffer_indexes,
126 count * sizeof(unsigned));
127 }
128
129
130 /**
131 * Specify vertex buffers for hardware drawing.
132 */
133 void
134 svga_hwtnl_vertex_buffers(struct svga_hwtnl *hwtnl,
135 unsigned count, struct pipe_vertex_buffer *buffers)
136 {
137 util_set_vertex_buffers_count(hwtnl->cmd.vbufs,
138 &hwtnl->cmd.vbuf_count, buffers, 0, count);
139 }
140
141
142 /**
143 * Determine whether the specified buffer is referred in the primitive queue,
144 * for which no commands have been written yet.
145 */
146 boolean
147 svga_hwtnl_is_buffer_referred(struct svga_hwtnl *hwtnl,
148 struct pipe_resource *buffer)
149 {
150 unsigned i;
151
152 if (svga_buffer_is_user_buffer(buffer)) {
153 return FALSE;
154 }
155
156 if (!hwtnl->cmd.prim_count) {
157 return FALSE;
158 }
159
160 for (i = 0; i < hwtnl->cmd.vbuf_count; ++i) {
161 if (hwtnl->cmd.vbufs[i].buffer == buffer) {
162 return TRUE;
163 }
164 }
165
166 for (i = 0; i < hwtnl->cmd.prim_count; ++i) {
167 if (hwtnl->cmd.prim_ib[i] == buffer) {
168 return TRUE;
169 }
170 }
171
172 return FALSE;
173 }
174
175
176 static enum pipe_error
177 draw_vgpu9(struct svga_hwtnl *hwtnl)
178 {
179 struct svga_winsys_context *swc = hwtnl->cmd.swc;
180 struct svga_context *svga = hwtnl->svga;
181 enum pipe_error ret;
182 struct svga_winsys_surface *vb_handle[SVGA3D_INPUTREG_MAX];
183 struct svga_winsys_surface *ib_handle[QSZ];
184 struct svga_winsys_surface *handle;
185 SVGA3dVertexDecl *vdecl;
186 SVGA3dPrimitiveRange *prim;
187 unsigned i;
188
189 for (i = 0; i < hwtnl->cmd.vdecl_count; i++) {
190 unsigned j = hwtnl->cmd.vdecl_buffer_index[i];
191 handle = svga_buffer_handle(svga, hwtnl->cmd.vbufs[j].buffer);
192 if (!handle)
193 return PIPE_ERROR_OUT_OF_MEMORY;
194
195 vb_handle[i] = handle;
196 }
197
198 for (i = 0; i < hwtnl->cmd.prim_count; i++) {
199 if (hwtnl->cmd.prim_ib[i]) {
200 handle = svga_buffer_handle(svga, hwtnl->cmd.prim_ib[i]);
201 if (!handle)
202 return PIPE_ERROR_OUT_OF_MEMORY;
203 }
204 else
205 handle = NULL;
206
207 ib_handle[i] = handle;
208 }
209
210 if (svga->rebind.flags.rendertargets) {
211 ret = svga_reemit_framebuffer_bindings(svga);
212 if (ret != PIPE_OK) {
213 return ret;
214 }
215 }
216
217 if (svga->rebind.flags.texture_samplers) {
218 ret = svga_reemit_tss_bindings(svga);
219 if (ret != PIPE_OK) {
220 return ret;
221 }
222 }
223
224 if (svga->rebind.flags.vs) {
225 ret = svga_reemit_vs_bindings(svga);
226 if (ret != PIPE_OK) {
227 return ret;
228 }
229 }
230
231 if (svga->rebind.flags.fs) {
232 ret = svga_reemit_fs_bindings(svga);
233 if (ret != PIPE_OK) {
234 return ret;
235 }
236 }
237
238 SVGA_DBG(DEBUG_DMA, "draw to sid %p, %d prims\n",
239 svga->curr.framebuffer.cbufs[0] ?
240 svga_surface(svga->curr.framebuffer.cbufs[0])->handle : NULL,
241 hwtnl->cmd.prim_count);
242
243 ret = SVGA3D_BeginDrawPrimitives(swc,
244 &vdecl,
245 hwtnl->cmd.vdecl_count,
246 &prim, hwtnl->cmd.prim_count);
247 if (ret != PIPE_OK)
248 return ret;
249
250 memcpy(vdecl,
251 hwtnl->cmd.vdecl,
252 hwtnl->cmd.vdecl_count * sizeof hwtnl->cmd.vdecl[0]);
253
254 for (i = 0; i < hwtnl->cmd.vdecl_count; i++) {
255 /* check for 4-byte alignment */
256 assert(vdecl[i].array.offset % 4 == 0);
257 assert(vdecl[i].array.stride % 4 == 0);
258
259 /* Given rangeHint is considered to be relative to indexBias, and
260 * indexBias varies per primitive, we cannot accurately supply an
261 * rangeHint when emitting more than one primitive per draw command.
262 */
263 if (hwtnl->cmd.prim_count == 1) {
264 vdecl[i].rangeHint.first = hwtnl->cmd.min_index[0];
265 vdecl[i].rangeHint.last = hwtnl->cmd.max_index[0] + 1;
266 }
267 else {
268 vdecl[i].rangeHint.first = 0;
269 vdecl[i].rangeHint.last = 0;
270 }
271
272 swc->surface_relocation(swc,
273 &vdecl[i].array.surfaceId,
274 NULL, vb_handle[i], SVGA_RELOC_READ);
275 }
276
277 memcpy(prim,
278 hwtnl->cmd.prim, hwtnl->cmd.prim_count * sizeof hwtnl->cmd.prim[0]);
279
280 for (i = 0; i < hwtnl->cmd.prim_count; i++) {
281 swc->surface_relocation(swc,
282 &prim[i].indexArray.surfaceId,
283 NULL, ib_handle[i], SVGA_RELOC_READ);
284 pipe_resource_reference(&hwtnl->cmd.prim_ib[i], NULL);
285 }
286
287 SVGA_FIFOCommitAll(swc);
288
289 hwtnl->cmd.prim_count = 0;
290
291 return PIPE_OK;
292 }
293
294
295 static SVGA3dSurfaceFormat
296 xlate_index_format(unsigned indexWidth)
297 {
298 if (indexWidth == 2) {
299 return SVGA3D_R16_UINT;
300 }
301 else if (indexWidth == 4) {
302 return SVGA3D_R32_UINT;
303 }
304 else {
305 assert(!"Bad indexWidth");
306 return SVGA3D_R32_UINT;
307 }
308 }
309
310
311 static enum pipe_error
312 validate_sampler_resources(struct svga_context *svga)
313 {
314 unsigned shader;
315
316 assert(svga_have_vgpu10(svga));
317
318 for (shader = PIPE_SHADER_VERTEX; shader <= PIPE_SHADER_GEOMETRY; shader++) {
319 unsigned count = svga->curr.num_sampler_views[shader];
320 unsigned i;
321 struct svga_winsys_surface *surfaces[PIPE_MAX_SAMPLERS];
322 enum pipe_error ret;
323
324 /*
325 * Reference bound sampler resources to ensure pending updates are
326 * noticed by the device.
327 */
328 for (i = 0; i < count; i++) {
329 struct svga_pipe_sampler_view *sv =
330 svga_pipe_sampler_view(svga->curr.sampler_views[shader][i]);
331
332 if (sv) {
333 if (sv->base.texture->target == PIPE_BUFFER) {
334 surfaces[i] = svga_buffer_handle(svga, sv->base.texture);
335 }
336 else {
337 surfaces[i] = svga_texture(sv->base.texture)->handle;
338 }
339 }
340 else {
341 surfaces[i] = NULL;
342 }
343 }
344
345 if (shader == PIPE_SHADER_FRAGMENT &&
346 svga->curr.rast->templ.poly_stipple_enable) {
347 const unsigned unit = svga->state.hw_draw.fs->pstipple_sampler_unit;
348 struct svga_pipe_sampler_view *sv =
349 svga->polygon_stipple.sampler_view;
350
351 assert(sv);
352 surfaces[unit] = svga_texture(sv->base.texture)->handle;
353 count = MAX2(count, unit+1);
354 }
355
356 /* rebind the shader resources if needed */
357 if (svga->rebind.flags.texture_samplers) {
358 for (i = 0; i < count; i++) {
359 if (surfaces[i]) {
360 ret = svga->swc->resource_rebind(svga->swc,
361 surfaces[i],
362 NULL,
363 SVGA_RELOC_READ);
364 if (ret != PIPE_OK)
365 return ret;
366 }
367 }
368 }
369 }
370 svga->rebind.flags.texture_samplers = FALSE;
371
372 return PIPE_OK;
373 }
374
375
376 static enum pipe_error
377 validate_constant_buffers(struct svga_context *svga)
378 {
379 unsigned shader;
380
381 assert(svga_have_vgpu10(svga));
382
383 for (shader = PIPE_SHADER_VERTEX; shader <= PIPE_SHADER_GEOMETRY; shader++) {
384 enum pipe_error ret;
385 struct svga_buffer *buffer;
386 struct svga_winsys_surface *handle;
387 unsigned enabled_constbufs;
388
389 /* Rebind the default constant buffer if needed */
390 if (svga->rebind.flags.constbufs) {
391 buffer = svga_buffer(svga->state.hw_draw.constbuf[shader]);
392 if (buffer) {
393 ret = svga->swc->resource_rebind(svga->swc,
394 buffer->handle,
395 NULL,
396 SVGA_RELOC_READ);
397 if (ret != PIPE_OK)
398 return ret;
399 }
400 }
401
402 /*
403 * Reference other bound constant buffers to ensure pending updates are
404 * noticed by the device.
405 */
406 enabled_constbufs = svga->state.hw_draw.enabled_constbufs[shader] & ~1u;
407 while (enabled_constbufs) {
408 unsigned i = u_bit_scan(&enabled_constbufs);
409 buffer = svga_buffer(svga->curr.constbufs[shader][i].buffer);
410 if (buffer) {
411 handle = svga_buffer_handle(svga, &buffer->b.b);
412
413 if (svga->rebind.flags.constbufs) {
414 ret = svga->swc->resource_rebind(svga->swc,
415 handle,
416 NULL,
417 SVGA_RELOC_READ);
418 if (ret != PIPE_OK)
419 return ret;
420 }
421 }
422 }
423 }
424 svga->rebind.flags.constbufs = FALSE;
425
426 return PIPE_OK;
427 }
428
429
430 static enum pipe_error
431 draw_vgpu10(struct svga_hwtnl *hwtnl,
432 const SVGA3dPrimitiveRange *range,
433 unsigned vcount,
434 unsigned min_index,
435 unsigned max_index, struct pipe_resource *ib,
436 unsigned start_instance, unsigned instance_count)
437 {
438 struct svga_context *svga = hwtnl->svga;
439 struct svga_winsys_surface *vb_handle[SVGA3D_INPUTREG_MAX];
440 struct svga_winsys_surface *ib_handle;
441 const unsigned vbuf_count = hwtnl->cmd.vbuf_count;
442 enum pipe_error ret;
443 unsigned i;
444
445 assert(svga_have_vgpu10(svga));
446 assert(hwtnl->cmd.prim_count == 0);
447
448 /* We need to reemit all the current resource bindings along with the Draw
449 * command to be sure that the referenced resources are available for the
450 * Draw command, just in case the surfaces associated with the resources
451 * are paged out.
452 */
453 if (svga->rebind.val) {
454 ret = svga_rebind_framebuffer_bindings(svga);
455 if (ret != PIPE_OK)
456 return ret;
457
458 ret = svga_rebind_shaders(svga);
459 if (ret != PIPE_OK)
460 return ret;
461 }
462
463 ret = validate_sampler_resources(svga);
464 if (ret != PIPE_OK)
465 return ret;
466
467 ret = validate_constant_buffers(svga);
468 if (ret != PIPE_OK)
469 return ret;
470
471 /* Get handle for each referenced vertex buffer */
472 for (i = 0; i < vbuf_count; i++) {
473 struct svga_buffer *sbuf = svga_buffer(hwtnl->cmd.vbufs[i].buffer);
474
475 if (sbuf) {
476 assert(sbuf->key.flags & SVGA3D_SURFACE_BIND_VERTEX_BUFFER);
477 vb_handle[i] = svga_buffer_handle(svga, &sbuf->b.b);
478 if (vb_handle[i] == NULL)
479 return PIPE_ERROR_OUT_OF_MEMORY;
480 }
481 else {
482 vb_handle[i] = NULL;
483 }
484 }
485
486 /* Get handles for the index buffers */
487 if (ib) {
488 struct svga_buffer *sbuf = svga_buffer(ib);
489
490 assert(sbuf->key.flags & SVGA3D_SURFACE_BIND_INDEX_BUFFER);
491 (void) sbuf; /* silence unused var warning */
492
493 ib_handle = svga_buffer_handle(svga, ib);
494 if (!ib_handle)
495 return PIPE_ERROR_OUT_OF_MEMORY;
496 }
497 else {
498 ib_handle = NULL;
499 }
500
501 /* setup vertex attribute input layout */
502 if (svga->state.hw_draw.layout_id != hwtnl->cmd.vdecl_layout_id) {
503 ret = SVGA3D_vgpu10_SetInputLayout(svga->swc,
504 hwtnl->cmd.vdecl_layout_id);
505 if (ret != PIPE_OK)
506 return ret;
507
508 svga->state.hw_draw.layout_id = hwtnl->cmd.vdecl_layout_id;
509 }
510
511 /* setup vertex buffers */
512 {
513 SVGA3dVertexBuffer buffers[PIPE_MAX_ATTRIBS];
514
515 for (i = 0; i < vbuf_count; i++) {
516 buffers[i].stride = hwtnl->cmd.vbufs[i].stride;
517 buffers[i].offset = hwtnl->cmd.vbufs[i].buffer_offset;
518 }
519 if (vbuf_count > 0) {
520 ret = SVGA3D_vgpu10_SetVertexBuffers(svga->swc, vbuf_count,
521 0, /* startBuffer */
522 buffers, vb_handle);
523 if (ret != PIPE_OK)
524 return ret;
525 }
526 }
527
528 /* Set primitive type (line, tri, etc) */
529 if (svga->state.hw_draw.topology != range->primType) {
530 ret = SVGA3D_vgpu10_SetTopology(svga->swc, range->primType);
531 if (ret != PIPE_OK)
532 return ret;
533
534 svga->state.hw_draw.topology = range->primType;
535 }
536
537 if (ib_handle) {
538 /* indexed drawing */
539 SVGA3dSurfaceFormat indexFormat = xlate_index_format(range->indexWidth);
540
541 /* setup index buffer */
542 if (ib_handle != svga->state.hw_draw.ib ||
543 indexFormat != svga->state.hw_draw.ib_format ||
544 range->indexArray.offset != svga->state.hw_draw.ib_offset) {
545 ret = SVGA3D_vgpu10_SetIndexBuffer(svga->swc, ib_handle,
546 indexFormat,
547 range->indexArray.offset);
548 if (ret != PIPE_OK)
549 return ret;
550 svga->state.hw_draw.ib = ib_handle;
551 svga->state.hw_draw.ib_format = indexFormat;
552 svga->state.hw_draw.ib_offset = range->indexArray.offset;
553 }
554
555 if (instance_count > 1) {
556 ret = SVGA3D_vgpu10_DrawIndexedInstanced(svga->swc,
557 vcount,
558 instance_count,
559 0, /* startIndexLocation */
560 range->indexBias,
561 start_instance);
562 if (ret != PIPE_OK)
563 return ret;
564 }
565 else {
566 /* non-instanced drawing */
567 ret = SVGA3D_vgpu10_DrawIndexed(svga->swc,
568 vcount,
569 0, /* startIndexLocation */
570 range->indexBias);
571 if (ret != PIPE_OK)
572 return ret;
573 }
574 }
575 else {
576 /* non-indexed drawing */
577 if (instance_count > 1) {
578 ret = SVGA3D_vgpu10_DrawInstanced(svga->swc,
579 vcount,
580 instance_count,
581 range->indexBias,
582 start_instance);
583 if (ret != PIPE_OK)
584 return ret;
585 }
586 else {
587 /* non-instanced */
588 ret = SVGA3D_vgpu10_Draw(svga->swc,
589 vcount,
590 range->indexBias);
591 if (ret != PIPE_OK)
592 return ret;
593 }
594 }
595
596 hwtnl->cmd.prim_count = 0;
597
598 return PIPE_OK;
599 }
600
601
602
603 /**
604 * Emit any pending drawing commands to the command buffer.
605 * When we receive VGPU9 drawing commands we accumulate them and don't
606 * immediately emit them into the command buffer.
607 * This function needs to be called before we change state that could
608 * effect those pending draws.
609 */
610 enum pipe_error
611 svga_hwtnl_flush(struct svga_hwtnl *hwtnl)
612 {
613 if (!svga_have_vgpu10(hwtnl->svga) && hwtnl->cmd.prim_count) {
614 /* we only queue up primitive for VGPU9 */
615 return draw_vgpu9(hwtnl);
616 }
617 return PIPE_OK;
618 }
619
620
621 void
622 svga_hwtnl_set_index_bias(struct svga_hwtnl *hwtnl, int index_bias)
623 {
624 hwtnl->index_bias = index_bias;
625 }
626
627
628
629 /***********************************************************************
630 * Internal functions:
631 */
632
633 /**
634 * For debugging only.
635 */
636 static void
637 check_draw_params(struct svga_hwtnl *hwtnl,
638 const SVGA3dPrimitiveRange *range,
639 unsigned min_index, unsigned max_index,
640 struct pipe_resource *ib)
641 {
642 unsigned i;
643
644 assert(!svga_have_vgpu10(hwtnl->svga));
645
646 for (i = 0; i < hwtnl->cmd.vdecl_count; i++) {
647 unsigned j = hwtnl->cmd.vdecl_buffer_index[i];
648 const struct pipe_vertex_buffer *vb = &hwtnl->cmd.vbufs[j];
649 unsigned size = vb->buffer ? vb->buffer->width0 : 0;
650 unsigned offset = hwtnl->cmd.vdecl[i].array.offset;
651 unsigned stride = hwtnl->cmd.vdecl[i].array.stride;
652 int index_bias = (int) range->indexBias + hwtnl->index_bias;
653 unsigned width;
654
655 if (size == 0)
656 continue;
657
658 assert(vb);
659 assert(size);
660 assert(offset < size);
661 assert(min_index <= max_index);
662 (void) width;
663 (void) stride;
664 (void) offset;
665 (void) size;
666
667 switch (hwtnl->cmd.vdecl[i].identity.type) {
668 case SVGA3D_DECLTYPE_FLOAT1:
669 width = 4;
670 break;
671 case SVGA3D_DECLTYPE_FLOAT2:
672 width = 4 * 2;
673 break;
674 case SVGA3D_DECLTYPE_FLOAT3:
675 width = 4 * 3;
676 break;
677 case SVGA3D_DECLTYPE_FLOAT4:
678 width = 4 * 4;
679 break;
680 case SVGA3D_DECLTYPE_D3DCOLOR:
681 width = 4;
682 break;
683 case SVGA3D_DECLTYPE_UBYTE4:
684 width = 1 * 4;
685 break;
686 case SVGA3D_DECLTYPE_SHORT2:
687 width = 2 * 2;
688 break;
689 case SVGA3D_DECLTYPE_SHORT4:
690 width = 2 * 4;
691 break;
692 case SVGA3D_DECLTYPE_UBYTE4N:
693 width = 1 * 4;
694 break;
695 case SVGA3D_DECLTYPE_SHORT2N:
696 width = 2 * 2;
697 break;
698 case SVGA3D_DECLTYPE_SHORT4N:
699 width = 2 * 4;
700 break;
701 case SVGA3D_DECLTYPE_USHORT2N:
702 width = 2 * 2;
703 break;
704 case SVGA3D_DECLTYPE_USHORT4N:
705 width = 2 * 4;
706 break;
707 case SVGA3D_DECLTYPE_UDEC3:
708 width = 4;
709 break;
710 case SVGA3D_DECLTYPE_DEC3N:
711 width = 4;
712 break;
713 case SVGA3D_DECLTYPE_FLOAT16_2:
714 width = 2 * 2;
715 break;
716 case SVGA3D_DECLTYPE_FLOAT16_4:
717 width = 2 * 4;
718 break;
719 default:
720 assert(0);
721 width = 0;
722 break;
723 }
724
725 if (index_bias >= 0) {
726 assert(offset + index_bias * stride + width <= size);
727 }
728
729 /*
730 * min_index/max_index are merely conservative guesses, so we can't
731 * make buffer overflow detection based on their values.
732 */
733 }
734
735 assert(range->indexWidth == range->indexArray.stride);
736
737 if (ib) {
738 unsigned size = ib->width0;
739 unsigned offset = range->indexArray.offset;
740 unsigned stride = range->indexArray.stride;
741 unsigned count;
742
743 assert(size);
744 assert(offset < size);
745 assert(stride);
746 (void) size;
747 (void) offset;
748 (void) stride;
749
750 switch (range->primType) {
751 case SVGA3D_PRIMITIVE_POINTLIST:
752 count = range->primitiveCount;
753 break;
754 case SVGA3D_PRIMITIVE_LINELIST:
755 count = range->primitiveCount * 2;
756 break;
757 case SVGA3D_PRIMITIVE_LINESTRIP:
758 count = range->primitiveCount + 1;
759 break;
760 case SVGA3D_PRIMITIVE_TRIANGLELIST:
761 count = range->primitiveCount * 3;
762 break;
763 case SVGA3D_PRIMITIVE_TRIANGLESTRIP:
764 count = range->primitiveCount + 2;
765 break;
766 case SVGA3D_PRIMITIVE_TRIANGLEFAN:
767 count = range->primitiveCount + 2;
768 break;
769 default:
770 assert(0);
771 count = 0;
772 break;
773 }
774
775 assert(offset + count * stride <= size);
776 }
777 }
778
779
780 /**
781 * All drawing filters down into this function, either directly
782 * on the hardware path or after doing software vertex processing.
783 */
784 enum pipe_error
785 svga_hwtnl_prim(struct svga_hwtnl *hwtnl,
786 const SVGA3dPrimitiveRange * range,
787 unsigned vcount,
788 unsigned min_index,
789 unsigned max_index, struct pipe_resource *ib,
790 unsigned start_instance, unsigned instance_count)
791 {
792 enum pipe_error ret = PIPE_OK;
793
794 if (svga_have_vgpu10(hwtnl->svga)) {
795 /* draw immediately */
796 ret = draw_vgpu10(hwtnl, range, vcount, min_index, max_index, ib,
797 start_instance, instance_count);
798 if (ret != PIPE_OK) {
799 svga_context_flush(hwtnl->svga, NULL);
800 ret = draw_vgpu10(hwtnl, range, vcount, min_index, max_index, ib,
801 start_instance, instance_count);
802 assert(ret == PIPE_OK);
803 }
804 }
805 else {
806 /* batch up drawing commands */
807 #ifdef DEBUG
808 check_draw_params(hwtnl, range, min_index, max_index, ib);
809 assert(start_instance == 0);
810 assert(instance_count <= 1);
811 #else
812 (void) check_draw_params;
813 #endif
814
815 if (hwtnl->cmd.prim_count + 1 >= QSZ) {
816 ret = svga_hwtnl_flush(hwtnl);
817 if (ret != PIPE_OK)
818 return ret;
819 }
820
821 /* min/max indices are relative to bias */
822 hwtnl->cmd.min_index[hwtnl->cmd.prim_count] = min_index;
823 hwtnl->cmd.max_index[hwtnl->cmd.prim_count] = max_index;
824
825 hwtnl->cmd.prim[hwtnl->cmd.prim_count] = *range;
826 hwtnl->cmd.prim[hwtnl->cmd.prim_count].indexBias += hwtnl->index_bias;
827
828 pipe_resource_reference(&hwtnl->cmd.prim_ib[hwtnl->cmd.prim_count], ib);
829 hwtnl->cmd.prim_count++;
830 }
831
832 return ret;
833 }