1/*
2 * Copyright 2010 Christoph Bumiller
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 */
22
23#define NVC0_PUSH_EXPLICIT_SPACE_CHECKING
24
25#include "pipe/p_context.h"
26#include "pipe/p_state.h"
27#include "util/u_draw.h"
28#include "util/u_inlines.h"
29#include "util/format/u_format.h"
30#include "translate/translate.h"
31
32#include "nvc0/nvc0_context.h"
33#include "nvc0/nvc0_query_hw.h"
34#include "nvc0/nvc0_resource.h"
35
36#include "nvc0/nvc0_3d.xml.h"
37
38void
39nvc0_vertex_state_delete(struct pipe_context *pipe,
40                         void *hwcso)
41{
42   struct nvc0_vertex_stateobj *so = hwcso;
43
44   if (so->translate)
45      so->translate->release(so->translate);
46   FREE(hwcso);
47}
48
49void *
50nvc0_vertex_state_create(struct pipe_context *pipe,
51                         unsigned num_elements,
52                         const struct pipe_vertex_element *elements)
53{
54    struct nvc0_vertex_stateobj *so;
55    struct translate_key transkey;
56    unsigned i;
57    unsigned src_offset_max = 0;
58
59    so = MALLOC(sizeof(*so) +
60                num_elements * sizeof(struct nvc0_vertex_element));
61    if (!so)
62        return NULL;
63    so->num_elements = num_elements;
64    so->instance_elts = 0;
65    so->instance_bufs = 0;
66    so->shared_slots = false;
67    so->need_conversion = false;
68
69    memset(so->vb_access_size, 0, sizeof(so->vb_access_size));
70
71    for (i = 0; i < PIPE_MAX_ATTRIBS; ++i)
72       so->min_instance_div[i] = 0xffffffff;
73
74    transkey.nr_elements = 0;
75    transkey.output_stride = 0;
76
77    for (i = 0; i < num_elements; ++i) {
78        const struct pipe_vertex_element *ve = &elements[i];
79        const unsigned vbi = ve->vertex_buffer_index;
80        unsigned size;
81        enum pipe_format fmt = ve->src_format;
82
83        so->element[i].pipe = elements[i];
84        so->element[i].state = nvc0_vertex_format[fmt].vtx;
85
86        if (!so->element[i].state) {
87            switch (util_format_get_nr_components(fmt)) {
88            case 1: fmt = PIPE_FORMAT_R32_FLOAT; break;
89            case 2: fmt = PIPE_FORMAT_R32G32_FLOAT; break;
90            case 3: fmt = PIPE_FORMAT_R32G32B32_FLOAT; break;
91            case 4: fmt = PIPE_FORMAT_R32G32B32A32_FLOAT; break;
92            default:
93                assert(0);
94                FREE(so);
95                return NULL;
96            }
97            so->element[i].state = nvc0_vertex_format[fmt].vtx;
98            so->need_conversion = true;
99            pipe_debug_message(&nouveau_context(pipe)->debug, FALLBACK,
100                               "Converting vertex element %d, no hw format %s",
101                               i, util_format_name(ve->src_format));
102        }
103        size = util_format_get_blocksize(fmt);
104
105        src_offset_max = MAX2(src_offset_max, ve->src_offset);
106
107        if (so->vb_access_size[vbi] < (ve->src_offset + size))
108           so->vb_access_size[vbi] = ve->src_offset + size;
109
110        if (unlikely(ve->instance_divisor)) {
111           so->instance_elts |= 1 << i;
112           so->instance_bufs |= 1 << vbi;
113           if (ve->instance_divisor < so->min_instance_div[vbi])
114              so->min_instance_div[vbi] = ve->instance_divisor;
115        }
116
117        if (1) {
118            unsigned ca;
119            unsigned j = transkey.nr_elements++;
120
121            ca = util_format_description(fmt)->channel[0].size / 8;
122            if (ca != 1 && ca != 2)
123               ca = 4;
124
125            transkey.element[j].type = TRANSLATE_ELEMENT_NORMAL;
126            transkey.element[j].input_format = ve->src_format;
127            transkey.element[j].input_buffer = vbi;
128            transkey.element[j].input_offset = ve->src_offset;
129            transkey.element[j].instance_divisor = ve->instance_divisor;
130
131            transkey.output_stride = align(transkey.output_stride, ca);
132            transkey.element[j].output_format = fmt;
133            transkey.element[j].output_offset = transkey.output_stride;
134            transkey.output_stride += size;
135
136            so->element[i].state_alt = so->element[i].state;
137            so->element[i].state_alt |= transkey.element[j].output_offset << 7;
138        }
139
140        so->element[i].state |= i << NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__SHIFT;
141    }
142    transkey.output_stride = align(transkey.output_stride, 4);
143
144    so->size = transkey.output_stride;
145    so->translate = translate_create(&transkey);
146
147    if (so->instance_elts || src_offset_max >= (1 << 14))
148       return so;
149    so->shared_slots = true;
150
151    for (i = 0; i < num_elements; ++i) {
152       const unsigned b = elements[i].vertex_buffer_index;
153       const unsigned s = elements[i].src_offset;
154       so->element[i].state &= ~NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__MASK;
155       so->element[i].state |= b << NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__SHIFT;
156       so->element[i].state |= s << NVC0_3D_VERTEX_ATTRIB_FORMAT_OFFSET__SHIFT;
157    }
158    return so;
159}
160
161#define NVC0_3D_VERTEX_ATTRIB_INACTIVE                                       \
162   NVC0_3D_VERTEX_ATTRIB_FORMAT_TYPE_FLOAT |                                 \
163   NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_32 | NVC0_3D_VERTEX_ATTRIB_FORMAT_CONST
164
165#define VTX_ATTR(a, c, t, s)                            \
166   ((NVC0_3D_VTX_ATTR_DEFINE_TYPE_##t) |                \
167    (NVC0_3D_VTX_ATTR_DEFINE_SIZE_##s) |                \
168    ((a) << NVC0_3D_VTX_ATTR_DEFINE_ATTR__SHIFT) |      \
169    ((c) << NVC0_3D_VTX_ATTR_DEFINE_COMP__SHIFT))
170
171static void
172nvc0_set_constant_vertex_attrib(struct nvc0_context *nvc0, const unsigned a)
173{
174   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
175   struct pipe_vertex_element *ve = &nvc0->vertex->element[a].pipe;
176   struct pipe_vertex_buffer *vb = &nvc0->vtxbuf[ve->vertex_buffer_index];
177   uint32_t mode;
178   const struct util_format_description *desc;
179   void *dst;
180   const void *src = (const uint8_t *)vb->buffer.user + ve->src_offset;
181   assert(vb->is_user_buffer);
182
183   desc = util_format_description(ve->src_format);
184
185   PUSH_SPACE(push, 6);
186   BEGIN_NVC0(push, NVC0_3D(VTX_ATTR_DEFINE), 5);
187   dst = &push->cur[1];
188   util_format_unpack_rgba(ve->src_format, dst, src, 1);
189   if (desc->channel[0].pure_integer) {
190      if (desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED) {
191         mode = VTX_ATTR(a, 4, SINT, 32);
192      } else {
193         mode = VTX_ATTR(a, 4, UINT, 32);
194      }
195   } else {
196      mode = VTX_ATTR(a, 4, FLOAT, 32);
197   }
198   push->cur[0] = mode;
199   push->cur += 5;
200}
201
202static inline void
203nvc0_user_vbuf_range(struct nvc0_context *nvc0, int vbi,
204                     uint32_t *base, uint32_t *size)
205{
206   if (unlikely(nvc0->vertex->instance_bufs & (1 << vbi))) {
207      const uint32_t div = nvc0->vertex->min_instance_div[vbi];
208      *base = nvc0->instance_off * nvc0->vtxbuf[vbi].stride;
209      *size = (nvc0->instance_max / div) * nvc0->vtxbuf[vbi].stride +
210         nvc0->vertex->vb_access_size[vbi];
211   } else {
212      /* NOTE: if there are user buffers, we *must* have index bounds */
213      assert(nvc0->vb_elt_limit != ~0);
214      *base = nvc0->vb_elt_first * nvc0->vtxbuf[vbi].stride;
215      *size = nvc0->vb_elt_limit * nvc0->vtxbuf[vbi].stride +
216         nvc0->vertex->vb_access_size[vbi];
217   }
218}
219
220static inline void
221nvc0_release_user_vbufs(struct nvc0_context *nvc0)
222{
223   if (nvc0->vbo_user) {
224      nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_VTX_TMP);
225      nouveau_scratch_done(&nvc0->base);
226   }
227}
228
229static void
230nvc0_update_user_vbufs(struct nvc0_context *nvc0)
231{
232   uint64_t address[PIPE_MAX_ATTRIBS];
233   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
234   int i;
235   uint32_t written = 0;
236
237   PUSH_SPACE(push, nvc0->vertex->num_elements * 8);
238   for (i = 0; i < nvc0->vertex->num_elements; ++i) {
239      struct pipe_vertex_element *ve = &nvc0->vertex->element[i].pipe;
240      const unsigned b = ve->vertex_buffer_index;
241      struct pipe_vertex_buffer *vb = &nvc0->vtxbuf[b];
242      uint32_t base, size;
243
244      if (!(nvc0->vbo_user & (1 << b)))
245         continue;
246      if (nvc0->constant_vbos & (1 << b)) {
247         nvc0_set_constant_vertex_attrib(nvc0, i);
248         continue;
249      }
250      nvc0_user_vbuf_range(nvc0, b, &base, &size);
251
252      if (!(written & (1 << b))) {
253         struct nouveau_bo *bo;
254         const uint32_t bo_flags = NOUVEAU_BO_RD | NOUVEAU_BO_GART;
255         written |= 1 << b;
256         address[b] = nouveau_scratch_data(&nvc0->base, vb->buffer.user,
257                                           base, size, &bo);
258         if (bo)
259            BCTX_REFN_bo(nvc0->bufctx_3d, 3D_VTX_TMP, bo_flags, bo);
260
261         NOUVEAU_DRV_STAT(&nvc0->screen->base, user_buffer_upload_bytes, size);
262      }
263
264      BEGIN_1IC0(push, NVC0_3D(MACRO_VERTEX_ARRAY_SELECT), 5);
265      PUSH_DATA (push, i);
266      PUSH_DATAh(push, address[b] + base + size - 1);
267      PUSH_DATA (push, address[b] + base + size - 1);
268      PUSH_DATAh(push, address[b] + ve->src_offset);
269      PUSH_DATA (push, address[b] + ve->src_offset);
270   }
271   nvc0->base.vbo_dirty = true;
272}
273
274static void
275nvc0_update_user_vbufs_shared(struct nvc0_context *nvc0)
276{
277   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
278   uint32_t mask = nvc0->vbo_user & ~nvc0->constant_vbos;
279
280   PUSH_SPACE(push, nvc0->num_vtxbufs * 8);
281   while (mask) {
282      struct nouveau_bo *bo;
283      const uint32_t bo_flags = NOUVEAU_BO_RD | NOUVEAU_BO_GART;
284      uint64_t address;
285      uint32_t base, size;
286      const int b = ffs(mask) - 1;
287      mask &= ~(1 << b);
288
289      nvc0_user_vbuf_range(nvc0, b, &base, &size);
290
291      address = nouveau_scratch_data(&nvc0->base, nvc0->vtxbuf[b].buffer.user,
292                                     base, size, &bo);
293      if (bo)
294         BCTX_REFN_bo(nvc0->bufctx_3d, 3D_VTX_TMP, bo_flags, bo);
295
296      BEGIN_1IC0(push, NVC0_3D(MACRO_VERTEX_ARRAY_SELECT), 5);
297      PUSH_DATA (push, b);
298      PUSH_DATAh(push, address + base + size - 1);
299      PUSH_DATA (push, address + base + size - 1);
300      PUSH_DATAh(push, address);
301      PUSH_DATA (push, address);
302
303      NOUVEAU_DRV_STAT(&nvc0->screen->base, user_buffer_upload_bytes, size);
304   }
305
306   mask = nvc0->state.constant_elts;
307   while (mask) {
308      int i = ffs(mask) - 1;
309      mask &= ~(1 << i);
310      nvc0_set_constant_vertex_attrib(nvc0, i);
311   }
312}
313
314static void
315nvc0_validate_vertex_buffers(struct nvc0_context *nvc0)
316{
317   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
318   const struct nvc0_vertex_stateobj *vertex = nvc0->vertex;
319   uint32_t refd = 0;
320   unsigned i;
321
322   PUSH_SPACE(push, vertex->num_elements * 8);
323   for (i = 0; i < vertex->num_elements; ++i) {
324      const struct nvc0_vertex_element *ve;
325      const struct pipe_vertex_buffer *vb;
326      struct nv04_resource *res;
327      unsigned b;
328      unsigned limit, offset;
329
330      if (nvc0->state.constant_elts & (1 << i))
331         continue;
332      ve = &vertex->element[i];
333      b = ve->pipe.vertex_buffer_index;
334      vb = &nvc0->vtxbuf[b];
335
336      if (nvc0->vbo_user & (1 << b)) {
337         if (!(nvc0->constant_vbos & (1 << b))) {
338            if (ve->pipe.instance_divisor) {
339               BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_DIVISOR(i)), 1);
340               PUSH_DATA (push, ve->pipe.instance_divisor);
341            }
342            BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 1);
343            PUSH_DATA (push, (1 << 12) | vb->stride);
344         }
345         /* address/value set in nvc0_update_user_vbufs */
346         continue;
347      }
348      res = nv04_resource(vb->buffer.resource);
349      offset = ve->pipe.src_offset + vb->buffer_offset;
350      limit = vb->buffer.resource->width0 - 1;
351
352      if (unlikely(ve->pipe.instance_divisor)) {
353         BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 4);
354         PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | vb->stride);
355         PUSH_DATAh(push, res->address + offset);
356         PUSH_DATA (push, res->address + offset);
357         PUSH_DATA (push, ve->pipe.instance_divisor);
358      } else {
359         BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 3);
360         PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | vb->stride);
361         PUSH_DATAh(push, res->address + offset);
362         PUSH_DATA (push, res->address + offset);
363      }
364
365      if (nvc0->screen->eng3d->oclass < TU102_3D_CLASS)
366         BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_LIMIT_HIGH(i)), 2);
367      else
368         BEGIN_NVC0(push, SUBC_3D(TU102_3D_VERTEX_ARRAY_LIMIT_HIGH(i)), 2);
369      PUSH_DATAh(push, res->address + limit);
370      PUSH_DATA (push, res->address + limit);
371
372      if (!(refd & (1 << b))) {
373         refd |= 1 << b;
374         BCTX_REFN(nvc0->bufctx_3d, 3D_VTX, res, RD);
375      }
376   }
377   if (nvc0->vbo_user)
378      nvc0_update_user_vbufs(nvc0);
379}
380
381static void
382nvc0_validate_vertex_buffers_shared(struct nvc0_context *nvc0)
383{
384   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
385   unsigned b;
386   const uint32_t mask = nvc0->vbo_user;
387
388   PUSH_SPACE(push, nvc0->num_vtxbufs * 8 + nvc0->vertex->num_elements);
389   for (b = 0; b < nvc0->num_vtxbufs; ++b) {
390      struct pipe_vertex_buffer *vb = &nvc0->vtxbuf[b];
391      struct nv04_resource *buf;
392      uint32_t offset, limit;
393
394      if (mask & (1 << b)) {
395         if (!(nvc0->constant_vbos & (1 << b))) {
396            BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(b)), 1);
397            PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | vb->stride);
398         }
399         /* address/value set in nvc0_update_user_vbufs_shared */
400         continue;
401      } else if (!vb->buffer.resource) {
402         /* there can be holes in the vertex buffer lists */
403         IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(b)), 0);
404         continue;
405      }
406      buf = nv04_resource(vb->buffer.resource);
407      offset = vb->buffer_offset;
408      limit = buf->base.width0 - 1;
409
410      BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(b)), 3);
411      PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | vb->stride);
412      PUSH_DATAh(push, buf->address + offset);
413      PUSH_DATA (push, buf->address + offset);
414
415      if (nvc0->screen->eng3d->oclass < TU102_3D_CLASS)
416         BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_LIMIT_HIGH(b)), 2);
417      else
418         BEGIN_NVC0(push, SUBC_3D(TU102_3D_VERTEX_ARRAY_LIMIT_HIGH(b)), 2);
419      PUSH_DATAh(push, buf->address + limit);
420      PUSH_DATA (push, buf->address + limit);
421
422      BCTX_REFN(nvc0->bufctx_3d, 3D_VTX, buf, RD);
423   }
424   /* If there are more elements than buffers, we might not have unset
425    * fetching on the later elements.
426    */
427   for (; b < nvc0->vertex->num_elements; ++b)
428      IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(b)), 0);
429
430   if (nvc0->vbo_user)
431      nvc0_update_user_vbufs_shared(nvc0);
432}
433
434void
435nvc0_vertex_arrays_validate(struct nvc0_context *nvc0)
436{
437   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
438   struct nvc0_vertex_stateobj *vertex = nvc0->vertex;
439   struct nvc0_vertex_element *ve;
440   uint32_t const_vbos;
441   unsigned i;
442   uint8_t vbo_mode;
443   bool update_vertex;
444
445   nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_VTX);
446
447   assert(vertex);
448   if (unlikely(vertex->need_conversion) ||
449       unlikely(nvc0->vertprog->vp.edgeflag < PIPE_MAX_ATTRIBS)) {
450      vbo_mode = 3;
451   } else if (nvc0->vbo_user & ~nvc0->constant_vbos) {
452      vbo_mode = nvc0->vbo_push_hint ? 1 : 0;
453   } else {
454      vbo_mode = 0;
455   }
456   const_vbos = vbo_mode ? 0 : nvc0->constant_vbos;
457
458   update_vertex = (nvc0->dirty_3d & NVC0_NEW_3D_VERTEX) ||
459      (const_vbos != nvc0->state.constant_vbos) ||
460      (vbo_mode != nvc0->state.vbo_mode);
461
462   if (update_vertex) {
463      const unsigned n = MAX2(vertex->num_elements, nvc0->state.num_vtxelts);
464
465      nvc0->state.constant_vbos = const_vbos;
466      nvc0->state.constant_elts = 0;
467      nvc0->state.num_vtxelts = vertex->num_elements;
468      nvc0->state.vbo_mode = vbo_mode;
469
470      if (unlikely(vbo_mode)) {
471         if (unlikely(nvc0->state.instance_elts & 3)) {
472            /* translate mode uses only 2 vertex buffers */
473            nvc0->state.instance_elts &= ~3;
474            PUSH_SPACE(push, 3);
475            BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_PER_INSTANCE(0)), 2);
476            PUSH_DATA (push, 0);
477            PUSH_DATA (push, 0);
478         }
479
480         PUSH_SPACE(push, n * 2 + 4);
481
482         BEGIN_NVC0(push, NVC0_3D(VERTEX_ATTRIB_FORMAT(0)), n);
483         for (i = 0; i < vertex->num_elements; ++i)
484            PUSH_DATA(push, vertex->element[i].state_alt);
485         for (; i < n; ++i)
486            PUSH_DATA(push, NVC0_3D_VERTEX_ATTRIB_INACTIVE);
487
488         BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(0)), 1);
489         PUSH_DATA (push, (1 << 12) | vertex->size);
490         for (i = 1; i < n; ++i)
491            IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 0);
492      } else {
493         uint32_t *restrict data;
494
495         if (unlikely(vertex->instance_elts != nvc0->state.instance_elts)) {
496            nvc0->state.instance_elts = vertex->instance_elts;
497            assert(n); /* if (n == 0), both masks should be 0 */
498            PUSH_SPACE(push, 3);
499            BEGIN_NVC0(push, NVC0_3D(MACRO_VERTEX_ARRAY_PER_INSTANCE), 2);
500            PUSH_DATA (push, n);
501            PUSH_DATA (push, vertex->instance_elts);
502         }
503
504         PUSH_SPACE(push, n * 2 + 1);
505         BEGIN_NVC0(push, NVC0_3D(VERTEX_ATTRIB_FORMAT(0)), n);
506         data = push->cur;
507         push->cur += n;
508         for (i = 0; i < vertex->num_elements; ++i) {
509            ve = &vertex->element[i];
510            data[i] = ve->state;
511            if (unlikely(const_vbos & (1 << ve->pipe.vertex_buffer_index))) {
512               nvc0->state.constant_elts |= 1 << i;
513               data[i] |= NVC0_3D_VERTEX_ATTRIB_FORMAT_CONST;
514               IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 0);
515            }
516         }
517         for (; i < n; ++i) {
518            data[i] = NVC0_3D_VERTEX_ATTRIB_INACTIVE;
519            IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 0);
520         }
521      }
522   }
523   if (nvc0->state.vbo_mode) /* using translate, don't set up arrays here */
524      return;
525
526   if (vertex->shared_slots)
527      nvc0_validate_vertex_buffers_shared(nvc0);
528   else
529      nvc0_validate_vertex_buffers(nvc0);
530}
531
532#define NVC0_PRIM_GL_CASE(n) \
533   case PIPE_PRIM_##n: return NVC0_3D_VERTEX_BEGIN_GL_PRIMITIVE_##n
534
535static inline unsigned
536nvc0_prim_gl(unsigned prim)
537{
538   switch (prim) {
539   NVC0_PRIM_GL_CASE(POINTS);
540   NVC0_PRIM_GL_CASE(LINES);
541   NVC0_PRIM_GL_CASE(LINE_LOOP);
542   NVC0_PRIM_GL_CASE(LINE_STRIP);
543   NVC0_PRIM_GL_CASE(TRIANGLES);
544   NVC0_PRIM_GL_CASE(TRIANGLE_STRIP);
545   NVC0_PRIM_GL_CASE(TRIANGLE_FAN);
546   NVC0_PRIM_GL_CASE(QUADS);
547   NVC0_PRIM_GL_CASE(QUAD_STRIP);
548   NVC0_PRIM_GL_CASE(POLYGON);
549   NVC0_PRIM_GL_CASE(LINES_ADJACENCY);
550   NVC0_PRIM_GL_CASE(LINE_STRIP_ADJACENCY);
551   NVC0_PRIM_GL_CASE(TRIANGLES_ADJACENCY);
552   NVC0_PRIM_GL_CASE(TRIANGLE_STRIP_ADJACENCY);
553   NVC0_PRIM_GL_CASE(PATCHES);
554   default:
555      return NVC0_3D_VERTEX_BEGIN_GL_PRIMITIVE_POINTS;
556   }
557}
558
559static void
560nvc0_draw_vbo_kick_notify(struct nouveau_pushbuf *push)
561{
562   struct nvc0_screen *screen = push->user_priv;
563
564   nouveau_fence_update(&screen->base, true);
565
566   NOUVEAU_DRV_STAT(&screen->base, pushbuf_count, 1);
567}
568
569static void
570nvc0_draw_arrays(struct nvc0_context *nvc0,
571                 unsigned mode, unsigned start, unsigned count,
572                 unsigned instance_count)
573{
574   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
575   unsigned prim;
576
577   if (nvc0->state.index_bias) {
578      /* index_bias is implied 0 if !info->index_size (really ?) */
579      /* TODO: can we deactivate it for the VERTEX_BUFFER_FIRST command ? */
580      PUSH_SPACE(push, 2);
581      IMMED_NVC0(push, NVC0_3D(VB_ELEMENT_BASE), 0);
582      IMMED_NVC0(push, NVC0_3D(VERTEX_ID_BASE), 0);
583      nvc0->state.index_bias = 0;
584   }
585
586   prim = nvc0_prim_gl(mode);
587
588   while (instance_count--) {
589      PUSH_SPACE(push, 6);
590      BEGIN_NVC0(push, NVC0_3D(VERTEX_BEGIN_GL), 1);
591      PUSH_DATA (push, prim);
592      BEGIN_NVC0(push, NVC0_3D(VERTEX_BUFFER_FIRST), 2);
593      PUSH_DATA (push, start);
594      PUSH_DATA (push, count);
595      IMMED_NVC0(push, NVC0_3D(VERTEX_END_GL), 0);
596
597      prim |= NVC0_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
598   }
599   NOUVEAU_DRV_STAT(&nvc0->screen->base, draw_calls_array, 1);
600}
601
602static void
603nvc0_draw_elements_inline_u08(struct nouveau_pushbuf *push, const uint8_t *map,
604                              unsigned start, unsigned count)
605{
606   map += start;
607
608   if (count & 3) {
609      unsigned i;
610      PUSH_SPACE(push, 4);
611      BEGIN_NIC0(push, NVC0_3D(VB_ELEMENT_U32), count & 3);
612      for (i = 0; i < (count & 3); ++i)
613         PUSH_DATA(push, *map++);
614      count &= ~3;
615   }
616   while (count) {
617      unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 4) / 4;
618
619      PUSH_SPACE(push, nr + 1);
620      BEGIN_NIC0(push, NVC0_3D(VB_ELEMENT_U8), nr);
621      for (i = 0; i < nr; ++i) {
622         PUSH_DATA(push,
623                  (map[3] << 24) | (map[2] << 16) | (map[1] << 8) | map[0]);
624         map += 4;
625      }
626      count -= nr * 4;
627   }
628}
629
630static void
631nvc0_draw_elements_inline_u16(struct nouveau_pushbuf *push, const uint16_t *map,
632                              unsigned start, unsigned count)
633{
634   map += start;
635
636   if (count & 1) {
637      count &= ~1;
638      PUSH_SPACE(push, 2);
639      BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_U32), 1);
640      PUSH_DATA (push, *map++);
641   }
642   while (count) {
643      unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 2) / 2;
644
645      PUSH_SPACE(push, nr + 1);
646      BEGIN_NIC0(push, NVC0_3D(VB_ELEMENT_U16), nr);
647      for (i = 0; i < nr; ++i) {
648         PUSH_DATA(push, (map[1] << 16) | map[0]);
649         map += 2;
650      }
651      count -= nr * 2;
652   }
653}
654
655static void
656nvc0_draw_elements_inline_u32(struct nouveau_pushbuf *push, const uint32_t *map,
657                              unsigned start, unsigned count)
658{
659   map += start;
660
661   while (count) {
662      const unsigned nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN);
663
664      PUSH_SPACE(push, nr + 1);
665      BEGIN_NIC0(push, NVC0_3D(VB_ELEMENT_U32), nr);
666      PUSH_DATAp(push, map, nr);
667
668      map += nr;
669      count -= nr;
670   }
671}
672
673static void
674nvc0_draw_elements_inline_u32_short(struct nouveau_pushbuf *push,
675                                    const uint32_t *map,
676                                    unsigned start, unsigned count)
677{
678   map += start;
679
680   if (count & 1) {
681      count--;
682      PUSH_SPACE(push, 2);
683      BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_U32), 1);
684      PUSH_DATA (push, *map++);
685   }
686   while (count) {
687      unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 2) / 2;
688
689      PUSH_SPACE(push, nr + 1);
690      BEGIN_NIC0(push, NVC0_3D(VB_ELEMENT_U16), nr);
691      for (i = 0; i < nr; ++i) {
692         PUSH_DATA(push, (map[1] << 16) | map[0]);
693         map += 2;
694      }
695      count -= nr * 2;
696   }
697}
698
699static void
700nvc0_draw_elements(struct nvc0_context *nvc0, bool shorten,
701                   const struct pipe_draw_info *info,
702                   unsigned mode, unsigned start, unsigned count,
703                   unsigned instance_count, int32_t index_bias,
704		   unsigned index_size)
705{
706   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
707   unsigned prim;
708
709   prim = nvc0_prim_gl(mode);
710
711   if (index_bias != nvc0->state.index_bias) {
712      PUSH_SPACE(push, 4);
713      BEGIN_NVC0(push, NVC0_3D(VB_ELEMENT_BASE), 1);
714      PUSH_DATA (push, index_bias);
715      BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_BASE), 1);
716      PUSH_DATA (push, index_bias);
717      nvc0->state.index_bias = index_bias;
718   }
719
720   if (!info->has_user_indices) {
721      PUSH_SPACE(push, 1);
722      IMMED_NVC0(push, NVC0_3D(VERTEX_BEGIN_GL), prim);
723      do {
724         PUSH_SPACE(push, 7);
725         BEGIN_NVC0(push, NVC0_3D(INDEX_BATCH_FIRST), 2);
726         PUSH_DATA (push, start);
727         PUSH_DATA (push, count);
728         if (--instance_count) {
729            BEGIN_NVC0(push, NVC0_3D(VERTEX_END_GL), 2);
730            PUSH_DATA (push, 0);
731            PUSH_DATA (push, prim | NVC0_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT);
732         }
733      } while (instance_count);
734      IMMED_NVC0(push, NVC0_3D(VERTEX_END_GL), 0);
735   } else {
736      const void *data = info->index.user;
737
738      while (instance_count--) {
739         PUSH_SPACE(push, 2);
740         BEGIN_NVC0(push, NVC0_3D(VERTEX_BEGIN_GL), 1);
741         PUSH_DATA (push, prim);
742         switch (index_size) {
743         case 1:
744            nvc0_draw_elements_inline_u08(push, data, start, count);
745            break;
746         case 2:
747            nvc0_draw_elements_inline_u16(push, data, start, count);
748            break;
749         case 4:
750            if (shorten)
751               nvc0_draw_elements_inline_u32_short(push, data, start, count);
752            else
753               nvc0_draw_elements_inline_u32(push, data, start, count);
754            break;
755         default:
756            assert(0);
757            return;
758         }
759         PUSH_SPACE(push, 1);
760         IMMED_NVC0(push, NVC0_3D(VERTEX_END_GL), 0);
761
762         prim |= NVC0_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
763      }
764   }
765   NOUVEAU_DRV_STAT(&nvc0->screen->base, draw_calls_indexed, 1);
766}
767
768static void
769nvc0_draw_stream_output(struct nvc0_context *nvc0,
770                        const struct pipe_draw_info *info,
771                        const struct pipe_draw_indirect_info *indirect)
772{
773   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
774   struct nvc0_so_target *so = nvc0_so_target(indirect->count_from_stream_output);
775   struct nv04_resource *res = nv04_resource(so->pipe.buffer);
776   unsigned mode = nvc0_prim_gl(info->mode);
777   unsigned num_instances = info->instance_count;
778
779   if (res->status & NOUVEAU_BUFFER_STATUS_GPU_WRITING) {
780      res->status &= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING;
781      PUSH_SPACE(push, 2);
782      IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0);
783      nvc0_hw_query_fifo_wait(nvc0, nvc0_query(so->pq));
784      if (nvc0->screen->eng3d->oclass < GM107_3D_CLASS)
785         IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FLUSH), 0);
786
787      NOUVEAU_DRV_STAT(&nvc0->screen->base, gpu_serialize_count, 1);
788   }
789
790   while (num_instances--) {
791      nouveau_pushbuf_space(push, 16, 0, 1);
792      BEGIN_NVC0(push, NVC0_3D(VERTEX_BEGIN_GL), 1);
793      PUSH_DATA (push, mode);
794      BEGIN_NVC0(push, NVC0_3D(DRAW_TFB_BASE), 1);
795      PUSH_DATA (push, 0);
796      BEGIN_NVC0(push, NVC0_3D(DRAW_TFB_STRIDE), 1);
797      PUSH_DATA (push, so->stride);
798      BEGIN_NVC0(push, NVC0_3D(DRAW_TFB_BYTES), 1);
799      nvc0_hw_query_pushbuf_submit(push, nvc0_query(so->pq), 0x4);
800      IMMED_NVC0(push, NVC0_3D(VERTEX_END_GL), 0);
801
802      mode |= NVC0_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
803   }
804}
805
806static void
807nvc0_draw_indirect(struct nvc0_context *nvc0, const struct pipe_draw_info *info,
808                   unsigned drawid_offset,
809                   const struct pipe_draw_indirect_info *indirect)
810{
811   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
812   struct nv04_resource *buf = nv04_resource(indirect->buffer);
813   struct nv04_resource *buf_count = nv04_resource(indirect->indirect_draw_count);
814   unsigned size, macro, count = indirect->draw_count, drawid = drawid_offset;
815   uint32_t offset = buf->offset + indirect->offset;
816   struct nvc0_screen *screen = nvc0->screen;
817
818   PUSH_SPACE(push, 7);
819
820   /* must make FIFO wait for engines idle before continuing to process */
821   if ((buf->fence_wr && !nouveau_fence_signalled(buf->fence_wr)) ||
822       (buf_count && buf_count->fence_wr &&
823        !nouveau_fence_signalled(buf_count->fence_wr))) {
824      IMMED_NVC0(push, SUBC_3D(NV10_SUBCHAN_REF_CNT), 0);
825   }
826
827   /* Queue things up to let the macros write params to the driver constbuf */
828   BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
829   PUSH_DATA (push, NVC0_CB_AUX_SIZE);
830   PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(0));
831   PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(0));
832   BEGIN_NVC0(push, NVC0_3D(CB_POS), 1);
833   PUSH_DATA (push, NVC0_CB_AUX_DRAW_INFO);
834
835   if (info->index_size) {
836      assert(!info->has_user_indices);
837      assert(nouveau_resource_mapped_by_gpu(info->index.resource));
838      size = 5;
839      if (buf_count)
840         macro = NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT;
841      else
842         macro = NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT;
843   } else {
844      if (nvc0->state.index_bias) {
845         /* index_bias is implied 0 if !info->index_size (really ?) */
846         IMMED_NVC0(push, NVC0_3D(VB_ELEMENT_BASE), 0);
847         IMMED_NVC0(push, NVC0_3D(VERTEX_ID_BASE), 0);
848         nvc0->state.index_bias = 0;
849      }
850      size = 4;
851      if (buf_count)
852         macro = NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT;
853      else
854         macro = NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT;
855   }
856
857   /* If the stride is not the natural stride, we have to stick a separate
858    * push data reference for each draw. Otherwise it can all go in as one.
859    * Of course there is a maximum packet size, so we have to break things up
860    * along those borders as well.
861    */
862   while (count) {
863      unsigned draws = count, pushes, i;
864      if (indirect->stride == size * 4) {
865         draws = MIN2(draws, (NV04_PFIFO_MAX_PACKET_LEN - 4) / size);
866         pushes = 1;
867      } else {
868         draws = MIN2(draws, 32);
869         pushes = draws;
870      }
871
872      nouveau_pushbuf_space(push, 16, 0, pushes + !!buf_count);
873      PUSH_REFN(push, buf->bo, NOUVEAU_BO_RD | buf->domain);
874      if (buf_count)
875         PUSH_REFN(push, buf_count->bo, NOUVEAU_BO_RD | buf_count->domain);
876      PUSH_DATA(push,
877                NVC0_FIFO_PKHDR_1I(0, macro, 3 + !!buf_count + draws * size));
878      PUSH_DATA(push, nvc0_prim_gl(info->mode));
879      PUSH_DATA(push, drawid);
880      PUSH_DATA(push, draws);
881      if (buf_count) {
882         nouveau_pushbuf_data(push,
883                              buf_count->bo,
884                              buf_count->offset + indirect->indirect_draw_count_offset,
885                              NVC0_IB_ENTRY_1_NO_PREFETCH | 4);
886      }
887      if (pushes == 1) {
888         nouveau_pushbuf_data(push,
889                              buf->bo, offset,
890                              NVC0_IB_ENTRY_1_NO_PREFETCH | (size * 4 * draws));
891         offset += draws * indirect->stride;
892      } else {
893         for (i = 0; i < pushes; i++) {
894            nouveau_pushbuf_data(push,
895                                 buf->bo, offset,
896                                 NVC0_IB_ENTRY_1_NO_PREFETCH | (size * 4));
897            offset += indirect->stride;
898         }
899      }
900      count -= draws;
901      drawid += draws;
902   }
903}
904
905static inline void
906nvc0_update_prim_restart(struct nvc0_context *nvc0, bool en, uint32_t index)
907{
908   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
909
910   if (en != nvc0->state.prim_restart) {
911      if (en) {
912         BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_ENABLE), 2);
913         PUSH_DATA (push, 1);
914         PUSH_DATA (push, index);
915      } else {
916         IMMED_NVC0(push, NVC0_3D(PRIM_RESTART_ENABLE), 0);
917      }
918      nvc0->state.prim_restart = en;
919   } else
920   if (en) {
921      BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_INDEX), 1);
922      PUSH_DATA (push, index);
923   }
924}
925
926void
927nvc0_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info,
928              unsigned drawid_offset,
929              const struct pipe_draw_indirect_info *indirect,
930              const struct pipe_draw_start_count_bias *draws,
931              unsigned num_draws)
932{
933   if (num_draws > 1) {
934      util_draw_multi(pipe, info, drawid_offset, indirect, draws, num_draws);
935      return;
936   }
937
938   if (!indirect && (!draws[0].count || !info->instance_count))
939      return;
940
941   struct nvc0_context *nvc0 = nvc0_context(pipe);
942   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
943   struct nvc0_screen *screen = nvc0->screen;
944   unsigned vram_domain = NV_VRAM_DOMAIN(&screen->base);
945   int s;
946
947   /* NOTE: caller must ensure that (min_index + index_bias) is >= 0 */
948   if (info->index_bounds_valid) {
949      nvc0->vb_elt_first = info->min_index + (info->index_size ? draws->index_bias : 0);
950      nvc0->vb_elt_limit = info->max_index - info->min_index;
951   } else {
952      nvc0->vb_elt_first = 0;
953      nvc0->vb_elt_limit = ~0;
954   }
955   nvc0->instance_off = info->start_instance;
956   nvc0->instance_max = info->instance_count - 1;
957
958   /* For picking only a few vertices from a large user buffer, push is better,
959    * if index count is larger and we expect repeated vertices, suggest upload.
960    */
961   nvc0->vbo_push_hint =
962      (!indirect || indirect->count_from_stream_output) && info->index_size &&
963      (nvc0->vb_elt_limit >= (draws[0].count * 2));
964
965   /* Check whether we want to switch vertex-submission mode. */
966   if (nvc0->vbo_user && !(nvc0->dirty_3d & (NVC0_NEW_3D_ARRAYS | NVC0_NEW_3D_VERTEX))) {
967      if (nvc0->vbo_push_hint != !!nvc0->state.vbo_mode)
968         if (nvc0->state.vbo_mode != 3)
969            nvc0->dirty_3d |= NVC0_NEW_3D_ARRAYS;
970
971      if (!(nvc0->dirty_3d & NVC0_NEW_3D_ARRAYS) && nvc0->state.vbo_mode == 0) {
972         if (nvc0->vertex->shared_slots)
973            nvc0_update_user_vbufs_shared(nvc0);
974         else
975            nvc0_update_user_vbufs(nvc0);
976      }
977   }
978
979   if (info->mode == PIPE_PRIM_PATCHES &&
980       nvc0->state.patch_vertices != nvc0->patch_vertices) {
981      nvc0->state.patch_vertices = nvc0->patch_vertices;
982      PUSH_SPACE(push, 1);
983      IMMED_NVC0(push, NVC0_3D(PATCH_VERTICES), nvc0->state.patch_vertices);
984   }
985
986   if (info->index_size && !info->has_user_indices) {
987      struct nv04_resource *buf = nv04_resource(info->index.resource);
988
989      assert(buf);
990      assert(nouveau_resource_mapped_by_gpu(&buf->base));
991
992      PUSH_SPACE(push, 6);
993      if (nvc0->screen->eng3d->oclass < TU102_3D_CLASS) {
994         BEGIN_NVC0(push, NVC0_3D(INDEX_ARRAY_START_HIGH), 5);
995         PUSH_DATAh(push, buf->address);
996         PUSH_DATA (push, buf->address);
997         PUSH_DATAh(push, buf->address + buf->base.width0 - 1);
998         PUSH_DATA (push, buf->address + buf->base.width0 - 1);
999         PUSH_DATA (push, info->index_size >> 1);
1000      } else {
1001         BEGIN_NVC0(push, NVC0_3D(INDEX_ARRAY_START_HIGH), 2);
1002         PUSH_DATAh(push, buf->address);
1003         PUSH_DATA (push, buf->address);
1004         BEGIN_NVC0(push, SUBC_3D(TU102_3D_INDEX_ARRAY_LIMIT_HIGH), 2);
1005         PUSH_DATAh(push, buf->address + buf->base.width0 - 1);
1006         PUSH_DATA (push, buf->address + buf->base.width0 - 1);
1007         BEGIN_NVC0(push, NVC0_3D(INDEX_FORMAT), 1);
1008         PUSH_DATA (push, info->index_size >> 1);
1009      }
1010
1011      BCTX_REFN(nvc0->bufctx_3d, 3D_IDX, buf, RD);
1012   }
1013
1014   list_for_each_entry(struct nvc0_resident, resident, &nvc0->tex_head, list) {
1015      nvc0_add_resident(nvc0->bufctx_3d, NVC0_BIND_3D_BINDLESS, resident->buf,
1016                        resident->flags);
1017   }
1018
1019   list_for_each_entry(struct nvc0_resident, resident, &nvc0->img_head, list) {
1020      nvc0_add_resident(nvc0->bufctx_3d, NVC0_BIND_3D_BINDLESS, resident->buf,
1021                        resident->flags);
1022   }
1023
1024   BCTX_REFN_bo(nvc0->bufctx_3d, 3D_TEXT, vram_domain | NOUVEAU_BO_RD,
1025                screen->text);
1026
1027   nvc0_state_validate_3d(nvc0, ~0);
1028
1029   if (nvc0->vertprog->vp.need_draw_parameters && (!indirect || indirect->count_from_stream_output)) {
1030      PUSH_SPACE(push, 9);
1031      BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1032      PUSH_DATA (push, NVC0_CB_AUX_SIZE);
1033      PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(0));
1034      PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(0));
1035      BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 3);
1036      PUSH_DATA (push, NVC0_CB_AUX_DRAW_INFO);
1037      PUSH_DATA (push, info->index_size ? draws->index_bias : 0);
1038      PUSH_DATA (push, info->start_instance);
1039      PUSH_DATA (push, drawid_offset);
1040   }
1041
1042   if (nvc0->screen->base.class_3d < NVE4_3D_CLASS &&
1043       nvc0->seamless_cube_map != nvc0->state.seamless_cube_map) {
1044      nvc0->state.seamless_cube_map = nvc0->seamless_cube_map;
1045      PUSH_SPACE(push, 1);
1046      IMMED_NVC0(push, NVC0_3D(TEX_MISC),
1047                 nvc0->seamless_cube_map ? NVC0_3D_TEX_MISC_SEAMLESS_CUBE_MAP : 0);
1048   }
1049
1050   push->kick_notify = nvc0_draw_vbo_kick_notify;
1051
1052   for (s = 0; s < 5 && !nvc0->cb_dirty; ++s) {
1053      if (nvc0->constbuf_coherent[s])
1054         nvc0->cb_dirty = true;
1055   }
1056
1057   if (nvc0->cb_dirty) {
1058      PUSH_SPACE(push, 1);
1059      IMMED_NVC0(push, NVC0_3D(MEM_BARRIER), 0x1011);
1060      nvc0->cb_dirty = false;
1061   }
1062
1063   for (s = 0; s < 5; ++s) {
1064      if (!nvc0->textures_coherent[s])
1065         continue;
1066
1067      PUSH_SPACE(push, nvc0->num_textures[s] * 2);
1068
1069      for (int i = 0; i < nvc0->num_textures[s]; ++i) {
1070         struct nv50_tic_entry *tic = nv50_tic_entry(nvc0->textures[s][i]);
1071         if (!(nvc0->textures_coherent[s] & (1 << i)))
1072            continue;
1073
1074         BEGIN_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 1);
1075         PUSH_DATA (push, (tic->id << 4) | 1);
1076         NOUVEAU_DRV_STAT(&nvc0->screen->base, tex_cache_flush_count, 1);
1077      }
1078   }
1079
1080   if (nvc0->state.vbo_mode) {
1081      if (indirect && indirect->buffer)
1082         nvc0_push_vbo_indirect(nvc0, info, drawid_offset, indirect, &draws[0]);
1083      else
1084         nvc0_push_vbo(nvc0, info, indirect, &draws[0]);
1085      goto cleanup;
1086   }
1087
1088   /* space for base instance, flush, and prim restart */
1089   PUSH_SPACE(push, 8);
1090
1091   if (nvc0->state.instance_base != info->start_instance) {
1092      nvc0->state.instance_base = info->start_instance;
1093      /* NOTE: this does not affect the shader input, should it ? */
1094      BEGIN_NVC0(push, NVC0_3D(VB_INSTANCE_BASE), 1);
1095      PUSH_DATA (push, info->start_instance);
1096   }
1097
1098   nvc0->base.vbo_dirty |= !!nvc0->vtxbufs_coherent;
1099
1100   if (!nvc0->base.vbo_dirty && info->index_size && !info->has_user_indices &&
1101       info->index.resource->flags & PIPE_RESOURCE_FLAG_MAP_COHERENT)
1102      nvc0->base.vbo_dirty = true;
1103
1104   nvc0_update_prim_restart(nvc0, info->primitive_restart, info->restart_index);
1105
1106   if (nvc0->base.vbo_dirty) {
1107      if (nvc0->screen->eng3d->oclass < GM107_3D_CLASS)
1108         IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FLUSH), 0);
1109      nvc0->base.vbo_dirty = false;
1110   }
1111
1112   if (unlikely(indirect && indirect->buffer)) {
1113      nvc0_draw_indirect(nvc0, info, drawid_offset, indirect);
1114   } else
1115   if (unlikely(indirect && indirect->count_from_stream_output)) {
1116      nvc0_draw_stream_output(nvc0, info, indirect);
1117   } else
1118   if (info->index_size) {
1119      bool shorten = info->index_bounds_valid && info->max_index <= 65535;
1120
1121      if (info->primitive_restart && info->restart_index > 65535)
1122         shorten = false;
1123
1124      nvc0_draw_elements(nvc0, shorten, info,
1125                         info->mode, draws[0].start, draws[0].count,
1126                         info->instance_count, draws->index_bias, info->index_size);
1127   } else {
1128      nvc0_draw_arrays(nvc0,
1129                       info->mode, draws[0].start, draws[0].count,
1130                       info->instance_count);
1131   }
1132
1133cleanup:
1134   push->kick_notify = nvc0_default_kick_notify;
1135
1136   nvc0_release_user_vbufs(nvc0);
1137
1138   nouveau_pushbuf_bufctx(push, NULL);
1139
1140   nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_TEXT);
1141   nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_IDX);
1142   nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_BINDLESS);
1143}
1144