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 BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
18  * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
19  * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
20  * SOFTWARE.
21  */
22 
23 #include "pipe/p_context.h"
24 #include "pipe/p_state.h"
25 #include "util/u_inlines.h"
26 #include "util/u_format.h"
27 #include "translate/translate.h"
28 
29 #include "nv50_context.h"
30 #include "nv50_resource.h"
31 
32 #include "nv50_3d.xml.h"
33 
34 void
nv50_vertex_state_delete(struct pipe_context * pipe,void * hwcso)35 nv50_vertex_state_delete(struct pipe_context *pipe,
36                          void *hwcso)
37 {
38    struct nv50_vertex_stateobj *so = hwcso;
39 
40    if (so->translate)
41       so->translate->release(so->translate);
42    FREE(hwcso);
43 }
44 
45 void *
nv50_vertex_state_create(struct pipe_context * pipe,unsigned num_elements,const struct pipe_vertex_element * elements)46 nv50_vertex_state_create(struct pipe_context *pipe,
47                          unsigned num_elements,
48                          const struct pipe_vertex_element *elements)
49 {
50     struct nv50_vertex_stateobj *so;
51     struct translate_key transkey;
52     unsigned i;
53 
54     so = MALLOC(sizeof(*so) +
55                 num_elements * sizeof(struct nv50_vertex_element));
56     if (!so)
57         return NULL;
58     so->num_elements = num_elements;
59     so->instance_elts = 0;
60     so->instance_bufs = 0;
61     so->need_conversion = FALSE;
62 
63     memset(so->vb_access_size, 0, sizeof(so->vb_access_size));
64 
65     for (i = 0; i < PIPE_MAX_ATTRIBS; ++i)
66        so->min_instance_div[i] = 0xffffffff;
67 
68     transkey.nr_elements = 0;
69     transkey.output_stride = 0;
70 
71     for (i = 0; i < num_elements; ++i) {
72         const struct pipe_vertex_element *ve = &elements[i];
73         const unsigned vbi = ve->vertex_buffer_index;
74         unsigned size;
75         enum pipe_format fmt = ve->src_format;
76 
77         so->element[i].pipe = elements[i];
78         so->element[i].state = nv50_format_table[fmt].vtx;
79 
80         if (!so->element[i].state) {
81             switch (util_format_get_nr_components(fmt)) {
82             case 1: fmt = PIPE_FORMAT_R32_FLOAT; break;
83             case 2: fmt = PIPE_FORMAT_R32G32_FLOAT; break;
84             case 3: fmt = PIPE_FORMAT_R32G32B32_FLOAT; break;
85             case 4: fmt = PIPE_FORMAT_R32G32B32A32_FLOAT; break;
86             default:
87                 assert(0);
88                 return NULL;
89             }
90             so->element[i].state = nv50_format_table[fmt].vtx;
91             so->need_conversion = TRUE;
92         }
93         so->element[i].state |= i;
94 
95         size = util_format_get_blocksize(fmt);
96         if (so->vb_access_size[vbi] < (ve->src_offset + size))
97            so->vb_access_size[vbi] = ve->src_offset + size;
98 
99         if (1) {
100             unsigned j = transkey.nr_elements++;
101 
102             transkey.element[j].type = TRANSLATE_ELEMENT_NORMAL;
103             transkey.element[j].input_format = ve->src_format;
104             transkey.element[j].input_buffer = vbi;
105             transkey.element[j].input_offset = ve->src_offset;
106             transkey.element[j].instance_divisor = ve->instance_divisor;
107 
108             transkey.element[j].output_format = fmt;
109             transkey.element[j].output_offset = transkey.output_stride;
110             transkey.output_stride += (util_format_get_stride(fmt, 1) + 3) & ~3;
111 
112             if (unlikely(ve->instance_divisor)) {
113                so->instance_elts |= 1 << i;
114                so->instance_bufs |= 1 << vbi;
115                if (ve->instance_divisor < so->min_instance_div[vbi])
116                   so->min_instance_div[vbi] = ve->instance_divisor;
117             }
118         }
119     }
120 
121     so->translate = translate_create(&transkey);
122     so->vertex_size = transkey.output_stride / 4;
123     so->packet_vertex_limit = NV04_PFIFO_MAX_PACKET_LEN /
124        MAX2(so->vertex_size, 1);
125 
126     return so;
127 }
128 
129 #define NV50_3D_VERTEX_ATTRIB_INACTIVE              \
130    NV50_3D_VERTEX_ARRAY_ATTRIB_TYPE_FLOAT |         \
131    NV50_3D_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32_32 | \
132    NV50_3D_VERTEX_ARRAY_ATTRIB_CONST
133 
134 static void
nv50_emit_vtxattr(struct nv50_context * nv50,struct pipe_vertex_buffer * vb,struct pipe_vertex_element * ve,unsigned attr)135 nv50_emit_vtxattr(struct nv50_context *nv50, struct pipe_vertex_buffer *vb,
136                   struct pipe_vertex_element *ve, unsigned attr)
137 {
138    struct nouveau_pushbuf *push = nv50->base.pushbuf;
139    const void *data = (const uint8_t *)vb->user_buffer + ve->src_offset;
140    float v[4];
141    const unsigned nc = util_format_get_nr_components(ve->src_format);
142 
143    assert(vb->user_buffer);
144 
145    util_format_read_4f(ve->src_format, v, 0, data, 0, 0, 0, 1, 1);
146 
147    switch (nc) {
148    case 4:
149       BEGIN_NV04(push, NV50_3D(VTX_ATTR_4F_X(attr)), 4);
150       PUSH_DATAf(push, v[0]);
151       PUSH_DATAf(push, v[1]);
152       PUSH_DATAf(push, v[2]);
153       PUSH_DATAf(push, v[3]);
154       break;
155    case 3:
156       BEGIN_NV04(push, NV50_3D(VTX_ATTR_3F_X(attr)), 3);
157       PUSH_DATAf(push, v[0]);
158       PUSH_DATAf(push, v[1]);
159       PUSH_DATAf(push, v[2]);
160       break;
161    case 2:
162       BEGIN_NV04(push, NV50_3D(VTX_ATTR_2F_X(attr)), 2);
163       PUSH_DATAf(push, v[0]);
164       PUSH_DATAf(push, v[1]);
165       break;
166    case 1:
167       if (attr == nv50->vertprog->vp.edgeflag) {
168          BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
169          PUSH_DATA (push, v[0] ? 1 : 0);
170       }
171       BEGIN_NV04(push, NV50_3D(VTX_ATTR_1F(attr)), 1);
172       PUSH_DATAf(push, v[0]);
173       break;
174    default:
175       assert(0);
176       break;
177    }
178 }
179 
180 static INLINE void
nv50_user_vbuf_range(struct nv50_context * nv50,int vbi,uint32_t * base,uint32_t * size)181 nv50_user_vbuf_range(struct nv50_context *nv50, int vbi,
182                      uint32_t *base, uint32_t *size)
183 {
184    if (unlikely(nv50->vertex->instance_bufs & (1 << vbi))) {
185       /* TODO: use min and max instance divisor to get a proper range */
186       *base = 0;
187       *size = nv50->vtxbuf[vbi].buffer->width0;
188    } else {
189       /* NOTE: if there are user buffers, we *must* have index bounds */
190       assert(nv50->vb_elt_limit != ~0);
191       *base = nv50->vb_elt_first * nv50->vtxbuf[vbi].stride;
192       *size = nv50->vb_elt_limit * nv50->vtxbuf[vbi].stride +
193          nv50->vertex->vb_access_size[vbi];
194    }
195 }
196 
197 static void
nv50_upload_user_buffers(struct nv50_context * nv50,uint64_t addrs[],uint32_t limits[])198 nv50_upload_user_buffers(struct nv50_context *nv50,
199                          uint64_t addrs[], uint32_t limits[])
200 {
201    unsigned b;
202 
203    for (b = 0; b < nv50->num_vtxbufs; ++b) {
204       struct nouveau_bo *bo;
205       const struct pipe_vertex_buffer *vb = &nv50->vtxbuf[b];
206       uint32_t base, size;
207 
208       if (!(nv50->vbo_user & (1 << b)) || !vb->stride)
209          continue;
210       nv50_user_vbuf_range(nv50, b, &base, &size);
211 
212       limits[b] = base + size - 1;
213       addrs[b] = nouveau_scratch_data(&nv50->base, vb->user_buffer, base, size,
214                                       &bo);
215       if (addrs[b])
216          BCTX_REFN_bo(nv50->bufctx_3d, VERTEX_TMP, NOUVEAU_BO_GART |
217                       NOUVEAU_BO_RD, bo);
218    }
219    nv50->base.vbo_dirty = TRUE;
220 }
221 
222 static void
nv50_update_user_vbufs(struct nv50_context * nv50)223 nv50_update_user_vbufs(struct nv50_context *nv50)
224 {
225    uint64_t address[PIPE_MAX_ATTRIBS];
226    struct nouveau_pushbuf *push = nv50->base.pushbuf;
227    unsigned i;
228    uint32_t written = 0;
229 
230    for (i = 0; i < nv50->vertex->num_elements; ++i) {
231       struct pipe_vertex_element *ve = &nv50->vertex->element[i].pipe;
232       const unsigned b = ve->vertex_buffer_index;
233       struct pipe_vertex_buffer *vb = &nv50->vtxbuf[b];
234       uint32_t base, size;
235 
236       if (!(nv50->vbo_user & (1 << b)))
237          continue;
238 
239       if (!vb->stride) {
240          nv50_emit_vtxattr(nv50, vb, ve, i);
241          continue;
242       }
243       nv50_user_vbuf_range(nv50, b, &base, &size);
244 
245       if (!(written & (1 << b))) {
246          struct nouveau_bo *bo;
247          const uint32_t bo_flags = NOUVEAU_BO_GART | NOUVEAU_BO_RD;
248          written |= 1 << b;
249          address[b] = nouveau_scratch_data(&nv50->base, vb->user_buffer,
250                                            base, size, &bo);
251          if (address[b])
252             BCTX_REFN_bo(nv50->bufctx_3d, VERTEX_TMP, bo_flags, bo);
253       }
254 
255       BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_LIMIT_HIGH(i)), 2);
256       PUSH_DATAh(push, address[b] + base + size - 1);
257       PUSH_DATA (push, address[b] + base + size - 1);
258       BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_START_HIGH(i)), 2);
259       PUSH_DATAh(push, address[b] + ve->src_offset);
260       PUSH_DATA (push, address[b] + ve->src_offset);
261    }
262    nv50->base.vbo_dirty = TRUE;
263 }
264 
265 static INLINE void
nv50_release_user_vbufs(struct nv50_context * nv50)266 nv50_release_user_vbufs(struct nv50_context *nv50)
267 {
268    if (nv50->vbo_user) {
269       nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_VERTEX_TMP);
270       nouveau_scratch_done(&nv50->base);
271    }
272 }
273 
274 void
nv50_vertex_arrays_validate(struct nv50_context * nv50)275 nv50_vertex_arrays_validate(struct nv50_context *nv50)
276 {
277    uint64_t addrs[PIPE_MAX_ATTRIBS];
278    uint32_t limits[PIPE_MAX_ATTRIBS];
279    struct nouveau_pushbuf *push = nv50->base.pushbuf;
280    struct nv50_vertex_stateobj *vertex = nv50->vertex;
281    struct pipe_vertex_buffer *vb;
282    struct nv50_vertex_element *ve;
283    uint32_t mask;
284    uint32_t refd = 0;
285    unsigned i;
286    const unsigned n = MAX2(vertex->num_elements, nv50->state.num_vtxelts);
287 
288    if (unlikely(vertex->need_conversion))
289       nv50->vbo_fifo = ~0;
290    else
291    if (nv50->vbo_user & ~nv50->vbo_constant)
292       nv50->vbo_fifo = nv50->vbo_push_hint ? ~0 : 0;
293    else
294       nv50->vbo_fifo = 0;
295 
296    if (!nv50->vbo_fifo) {
297       /* if vertex buffer was written by GPU - flush VBO cache */
298       for (i = 0; i < nv50->num_vtxbufs; ++i) {
299          struct nv04_resource *buf = nv04_resource(nv50->vtxbuf[i].buffer);
300          if (buf && buf->status & NOUVEAU_BUFFER_STATUS_GPU_WRITING) {
301             buf->status &= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING;
302             nv50->base.vbo_dirty = TRUE;
303             break;
304          }
305       }
306    }
307 
308    /* update vertex format state */
309    BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_ATTRIB(0)), n);
310    if (nv50->vbo_fifo) {
311       nv50->state.num_vtxelts = vertex->num_elements;
312       for (i = 0; i < vertex->num_elements; ++i)
313          PUSH_DATA (push, vertex->element[i].state);
314       for (; i < n; ++i)
315          PUSH_DATA (push, NV50_3D_VERTEX_ATTRIB_INACTIVE);
316       for (i = 0; i < n; ++i) {
317          BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_FETCH(i)), 1);
318          PUSH_DATA (push, 0);
319       }
320       return;
321    }
322    for (i = 0; i < vertex->num_elements; ++i) {
323       const unsigned b = vertex->element[i].pipe.vertex_buffer_index;
324       ve = &vertex->element[i];
325       vb = &nv50->vtxbuf[b];
326 
327       if (likely(vb->stride) || !(nv50->vbo_user & (1 << b)))
328          PUSH_DATA(push, ve->state);
329       else
330          PUSH_DATA(push, ve->state | NV50_3D_VERTEX_ARRAY_ATTRIB_CONST);
331    }
332    for (; i < n; ++i)
333       PUSH_DATA(push, NV50_3D_VERTEX_ATTRIB_INACTIVE);
334 
335    /* update per-instance enables */
336    mask = vertex->instance_elts ^ nv50->state.instance_elts;
337    while (mask) {
338       const int i = ffs(mask) - 1;
339       mask &= ~(1 << i);
340       BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_PER_INSTANCE(i)), 1);
341       PUSH_DATA (push, (vertex->instance_elts >> i) & 1);
342    }
343    nv50->state.instance_elts = vertex->instance_elts;
344 
345    if (nv50->vbo_user & ~nv50->vbo_constant)
346       nv50_upload_user_buffers(nv50, addrs, limits);
347 
348    /* update buffers and set constant attributes */
349    for (i = 0; i < vertex->num_elements; ++i) {
350       uint64_t address, limit;
351       const unsigned b = vertex->element[i].pipe.vertex_buffer_index;
352       ve = &vertex->element[i];
353       vb = &nv50->vtxbuf[b];
354 
355       if (unlikely(nv50->vbo_constant & (1 << b))) {
356          BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_FETCH(i)), 1);
357          PUSH_DATA (push, 0);
358          nv50_emit_vtxattr(nv50, vb, &ve->pipe, i);
359          continue;
360       } else
361       if (nv50->vbo_user & (1 << b)) {
362          address = addrs[b] + ve->pipe.src_offset;
363          limit = addrs[b] + limits[b];
364       } else {
365          struct nv04_resource *buf = nv04_resource(vb->buffer);
366          if (!(refd & (1 << b))) {
367             refd |= 1 << b;
368             BCTX_REFN(nv50->bufctx_3d, VERTEX, buf, RD);
369          }
370          address = buf->address + vb->buffer_offset + ve->pipe.src_offset;
371          limit = buf->address + buf->base.width0 - 1;
372       }
373 
374       if (unlikely(ve->pipe.instance_divisor)) {
375          BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_FETCH(i)), 4);
376          PUSH_DATA (push, NV50_3D_VERTEX_ARRAY_FETCH_ENABLE | vb->stride);
377          PUSH_DATAh(push, address);
378          PUSH_DATA (push, address);
379          PUSH_DATA (push, ve->pipe.instance_divisor);
380       } else {
381          BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_FETCH(i)), 3);
382          PUSH_DATA (push, NV50_3D_VERTEX_ARRAY_FETCH_ENABLE | vb->stride);
383          PUSH_DATAh(push, address);
384          PUSH_DATA (push, address);
385       }
386       BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_LIMIT_HIGH(i)), 2);
387       PUSH_DATAh(push, limit);
388       PUSH_DATA (push, limit);
389    }
390    for (; i < nv50->state.num_vtxelts; ++i) {
391       BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_FETCH(i)), 1);
392       PUSH_DATA (push, 0);
393    }
394    nv50->state.num_vtxelts = vertex->num_elements;
395 }
396 
397 #define NV50_PRIM_GL_CASE(n) \
398    case PIPE_PRIM_##n: return NV50_3D_VERTEX_BEGIN_GL_PRIMITIVE_##n
399 
400 static INLINE unsigned
nv50_prim_gl(unsigned prim)401 nv50_prim_gl(unsigned prim)
402 {
403    switch (prim) {
404    NV50_PRIM_GL_CASE(POINTS);
405    NV50_PRIM_GL_CASE(LINES);
406    NV50_PRIM_GL_CASE(LINE_LOOP);
407    NV50_PRIM_GL_CASE(LINE_STRIP);
408    NV50_PRIM_GL_CASE(TRIANGLES);
409    NV50_PRIM_GL_CASE(TRIANGLE_STRIP);
410    NV50_PRIM_GL_CASE(TRIANGLE_FAN);
411    NV50_PRIM_GL_CASE(QUADS);
412    NV50_PRIM_GL_CASE(QUAD_STRIP);
413    NV50_PRIM_GL_CASE(POLYGON);
414    NV50_PRIM_GL_CASE(LINES_ADJACENCY);
415    NV50_PRIM_GL_CASE(LINE_STRIP_ADJACENCY);
416    NV50_PRIM_GL_CASE(TRIANGLES_ADJACENCY);
417    NV50_PRIM_GL_CASE(TRIANGLE_STRIP_ADJACENCY);
418    default:
419       return NV50_3D_VERTEX_BEGIN_GL_PRIMITIVE_POINTS;
420       break;
421    }
422 }
423 
424 /* For pre-nva0 transform feedback. */
425 static const uint8_t nv50_pipe_prim_to_prim_size[PIPE_PRIM_MAX + 1] =
426 {
427    [PIPE_PRIM_POINTS] = 1,
428    [PIPE_PRIM_LINES] = 2,
429    [PIPE_PRIM_LINE_LOOP] = 2,
430    [PIPE_PRIM_LINE_STRIP] = 2,
431    [PIPE_PRIM_TRIANGLES] = 3,
432    [PIPE_PRIM_TRIANGLE_STRIP] = 3,
433    [PIPE_PRIM_TRIANGLE_FAN] = 3,
434    [PIPE_PRIM_QUADS] = 3,
435    [PIPE_PRIM_QUAD_STRIP] = 3,
436    [PIPE_PRIM_POLYGON] = 3,
437    [PIPE_PRIM_LINES_ADJACENCY] = 2,
438    [PIPE_PRIM_LINE_STRIP_ADJACENCY] = 2,
439    [PIPE_PRIM_TRIANGLES_ADJACENCY] = 3,
440    [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY] = 3
441 };
442 
443 static void
nv50_draw_arrays(struct nv50_context * nv50,unsigned mode,unsigned start,unsigned count,unsigned instance_count)444 nv50_draw_arrays(struct nv50_context *nv50,
445                  unsigned mode, unsigned start, unsigned count,
446                  unsigned instance_count)
447 {
448    struct nouveau_pushbuf *push = nv50->base.pushbuf;
449    unsigned prim;
450 
451    if (nv50->state.index_bias) {
452       BEGIN_NV04(push, NV50_3D(VB_ELEMENT_BASE), 1);
453       PUSH_DATA (push, 0);
454       nv50->state.index_bias = 0;
455    }
456 
457    prim = nv50_prim_gl(mode);
458 
459    while (instance_count--) {
460       BEGIN_NV04(push, NV50_3D(VERTEX_BEGIN_GL), 1);
461       PUSH_DATA (push, prim);
462       BEGIN_NV04(push, NV50_3D(VERTEX_BUFFER_FIRST), 2);
463       PUSH_DATA (push, start);
464       PUSH_DATA (push, count);
465       BEGIN_NV04(push, NV50_3D(VERTEX_END_GL), 1);
466       PUSH_DATA (push, 0);
467 
468       prim |= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
469    }
470 }
471 
472 static void
nv50_draw_elements_inline_u08(struct nouveau_pushbuf * push,const uint8_t * map,unsigned start,unsigned count)473 nv50_draw_elements_inline_u08(struct nouveau_pushbuf *push, const uint8_t *map,
474                               unsigned start, unsigned count)
475 {
476    map += start;
477 
478    if (count & 3) {
479       unsigned i;
480       BEGIN_NI04(push, NV50_3D(VB_ELEMENT_U32), count & 3);
481       for (i = 0; i < (count & 3); ++i)
482          PUSH_DATA(push, *map++);
483       count &= ~3;
484    }
485    while (count) {
486       unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 4) / 4;
487 
488       BEGIN_NI04(push, NV50_3D(VB_ELEMENT_U8), nr);
489       for (i = 0; i < nr; ++i) {
490          PUSH_DATA(push,
491                    (map[3] << 24) | (map[2] << 16) | (map[1] << 8) | map[0]);
492          map += 4;
493       }
494       count -= nr * 4;
495    }
496 }
497 
498 static void
nv50_draw_elements_inline_u16(struct nouveau_pushbuf * push,const uint16_t * map,unsigned start,unsigned count)499 nv50_draw_elements_inline_u16(struct nouveau_pushbuf *push, const uint16_t *map,
500                               unsigned start, unsigned count)
501 {
502    map += start;
503 
504    if (count & 1) {
505       count &= ~1;
506       BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U32), 1);
507       PUSH_DATA (push, *map++);
508    }
509    while (count) {
510       unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 2) / 2;
511 
512       BEGIN_NI04(push, NV50_3D(VB_ELEMENT_U16), nr);
513       for (i = 0; i < nr; ++i) {
514          PUSH_DATA(push, (map[1] << 16) | map[0]);
515          map += 2;
516       }
517       count -= nr * 2;
518    }
519 }
520 
521 static void
nv50_draw_elements_inline_u32(struct nouveau_pushbuf * push,const uint32_t * map,unsigned start,unsigned count)522 nv50_draw_elements_inline_u32(struct nouveau_pushbuf *push, const uint32_t *map,
523                               unsigned start, unsigned count)
524 {
525    map += start;
526 
527    while (count) {
528       const unsigned nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN);
529 
530       BEGIN_NI04(push, NV50_3D(VB_ELEMENT_U32), nr);
531       PUSH_DATAp(push, map, nr);
532 
533       map += nr;
534       count -= nr;
535    }
536 }
537 
538 static void
nv50_draw_elements_inline_u32_short(struct nouveau_pushbuf * push,const uint32_t * map,unsigned start,unsigned count)539 nv50_draw_elements_inline_u32_short(struct nouveau_pushbuf *push,
540                                     const uint32_t *map,
541                                     unsigned start, unsigned count)
542 {
543    map += start;
544 
545    if (count & 1) {
546       count--;
547       BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U32), 1);
548       PUSH_DATA (push, *map++);
549    }
550    while (count) {
551       unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 2) / 2;
552 
553       BEGIN_NI04(push, NV50_3D(VB_ELEMENT_U16), nr);
554       for (i = 0; i < nr; ++i) {
555          PUSH_DATA(push, (map[1] << 16) | map[0]);
556          map += 2;
557       }
558       count -= nr * 2;
559    }
560 }
561 
562 static void
nv50_draw_elements(struct nv50_context * nv50,boolean shorten,unsigned mode,unsigned start,unsigned count,unsigned instance_count,int32_t index_bias)563 nv50_draw_elements(struct nv50_context *nv50, boolean shorten,
564                    unsigned mode, unsigned start, unsigned count,
565                    unsigned instance_count, int32_t index_bias)
566 {
567    struct nouveau_pushbuf *push = nv50->base.pushbuf;
568    unsigned prim;
569    const unsigned index_size = nv50->idxbuf.index_size;
570 
571    prim = nv50_prim_gl(mode);
572 
573    if (index_bias != nv50->state.index_bias) {
574       BEGIN_NV04(push, NV50_3D(VB_ELEMENT_BASE), 1);
575       PUSH_DATA (push, index_bias);
576       nv50->state.index_bias = index_bias;
577    }
578 
579    if (nv50->idxbuf.buffer) {
580       struct nv04_resource *buf = nv04_resource(nv50->idxbuf.buffer);
581       unsigned pb_start;
582       unsigned pb_bytes;
583       const unsigned base = (buf->offset + nv50->idxbuf.offset) & ~3;
584 
585       start += ((buf->offset + nv50->idxbuf.offset) & 3) >> (index_size >> 1);
586 
587       assert(nouveau_resource_mapped_by_gpu(nv50->idxbuf.buffer));
588 
589       while (instance_count--) {
590          BEGIN_NV04(push, NV50_3D(VERTEX_BEGIN_GL), 1);
591          PUSH_DATA (push, prim);
592 
593          nouveau_pushbuf_space(push, 8, 0, 1);
594 
595          switch (index_size) {
596          case 4:
597             BEGIN_NL50(push, NV50_3D(VB_ELEMENT_U32), count);
598             nouveau_pushbuf_data(push, buf->bo, base + start * 4, count * 4);
599             break;
600          case 2:
601             pb_start = (start & ~1) * 2;
602             pb_bytes = ((start + count + 1) & ~1) * 2 - pb_start;
603 
604             BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U16_SETUP), 1);
605             PUSH_DATA (push, (start << 31) | count);
606             BEGIN_NL50(push, NV50_3D(VB_ELEMENT_U16), pb_bytes / 4);
607             nouveau_pushbuf_data(push, buf->bo, base + pb_start, pb_bytes);
608             BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U16_SETUP), 1);
609             PUSH_DATA (push, 0);
610             break;
611          default:
612             assert(index_size == 1);
613             pb_start = start & ~3;
614             pb_bytes = ((start + count + 3) & ~3) - pb_start;
615 
616             BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U8_SETUP), 1);
617             PUSH_DATA (push, (start << 30) | count);
618             BEGIN_NL50(push, NV50_3D(VB_ELEMENT_U8), pb_bytes / 4);
619             nouveau_pushbuf_data(push, buf->bo, base + pb_start, pb_bytes);
620             BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U8_SETUP), 1);
621             PUSH_DATA (push, 0);
622             break;
623          }
624          BEGIN_NV04(push, NV50_3D(VERTEX_END_GL), 1);
625          PUSH_DATA (push, 0);
626 
627          prim |= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
628       }
629    } else {
630       const void *data = nv50->idxbuf.user_buffer;
631 
632       while (instance_count--) {
633          BEGIN_NV04(push, NV50_3D(VERTEX_BEGIN_GL), 1);
634          PUSH_DATA (push, prim);
635          switch (index_size) {
636          case 1:
637             nv50_draw_elements_inline_u08(push, data, start, count);
638             break;
639          case 2:
640             nv50_draw_elements_inline_u16(push, data, start, count);
641             break;
642          case 4:
643             if (shorten)
644                nv50_draw_elements_inline_u32_short(push, data, start, count);
645             else
646                nv50_draw_elements_inline_u32(push, data, start, count);
647             break;
648          default:
649             assert(0);
650             return;
651          }
652          BEGIN_NV04(push, NV50_3D(VERTEX_END_GL), 1);
653          PUSH_DATA (push, 0);
654 
655          prim |= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
656       }
657    }
658 }
659 
660 static void
nva0_draw_stream_output(struct nv50_context * nv50,const struct pipe_draw_info * info)661 nva0_draw_stream_output(struct nv50_context *nv50,
662                         const struct pipe_draw_info *info)
663 {
664    struct nouveau_pushbuf *push = nv50->base.pushbuf;
665    struct nv50_so_target *so = nv50_so_target(info->count_from_stream_output);
666    struct nv04_resource *res = nv04_resource(so->pipe.buffer);
667    unsigned num_instances = info->instance_count;
668    unsigned mode = nv50_prim_gl(info->mode);
669 
670    if (unlikely(nv50->screen->base.class_3d < NVA0_3D_CLASS)) {
671       /* A proper implementation without waiting doesn't seem possible,
672        * so don't bother.
673        */
674       NOUVEAU_ERR("draw_stream_output not supported on pre-NVA0 cards\n");
675       return;
676    }
677 
678    if (res->status & NOUVEAU_BUFFER_STATUS_GPU_WRITING) {
679       res->status &= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING;
680       PUSH_SPACE(push, 4);
681       BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
682       PUSH_DATA (push, 0);
683       BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_FLUSH), 1);
684       PUSH_DATA (push, 0);
685    }
686 
687    assert(num_instances);
688    do {
689       PUSH_SPACE(push, 8);
690       BEGIN_NV04(push, NV50_3D(VERTEX_BEGIN_GL), 1);
691       PUSH_DATA (push, mode);
692       BEGIN_NV04(push, NVA0_3D(DRAW_TFB_BASE), 1);
693       PUSH_DATA (push, 0);
694       BEGIN_NV04(push, NVA0_3D(DRAW_TFB_STRIDE), 1);
695       PUSH_DATA (push, 0);
696       BEGIN_NV04(push, NVA0_3D(DRAW_TFB_BYTES), 1);
697       nv50_query_pushbuf_submit(push, so->pq, 0x4);
698       BEGIN_NV04(push, NV50_3D(VERTEX_END_GL), 1);
699       PUSH_DATA (push, 0);
700 
701       mode |= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
702    } while (--num_instances);
703 }
704 
705 static void
nv50_draw_vbo_kick_notify(struct nouveau_pushbuf * chan)706 nv50_draw_vbo_kick_notify(struct nouveau_pushbuf *chan)
707 {
708    struct nv50_screen *screen = chan->user_priv;
709 
710    nouveau_fence_update(&screen->base, TRUE);
711 
712    nv50_bufctx_fence(screen->cur_ctx->bufctx_3d, TRUE);
713 }
714 
715 void
nv50_draw_vbo(struct pipe_context * pipe,const struct pipe_draw_info * info)716 nv50_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
717 {
718    struct nv50_context *nv50 = nv50_context(pipe);
719    struct nouveau_pushbuf *push = nv50->base.pushbuf;
720 
721    /* NOTE: caller must ensure that (min_index + index_bias) is >= 0 */
722    nv50->vb_elt_first = info->min_index + info->index_bias;
723    nv50->vb_elt_limit = info->max_index - info->min_index;
724    nv50->instance_off = info->start_instance;
725    nv50->instance_max = info->instance_count - 1;
726 
727    /* For picking only a few vertices from a large user buffer, push is better,
728     * if index count is larger and we expect repeated vertices, suggest upload.
729     */
730    nv50->vbo_push_hint = /* the 64 is heuristic */
731       !(info->indexed && ((nv50->vb_elt_limit + 64) < info->count));
732 
733    if (nv50->vbo_user && !(nv50->dirty & (NV50_NEW_ARRAYS | NV50_NEW_VERTEX))) {
734       if (!!nv50->vbo_fifo != nv50->vbo_push_hint)
735          nv50->dirty |= NV50_NEW_ARRAYS;
736       else
737       if (!nv50->vbo_fifo)
738          nv50_update_user_vbufs(nv50);
739    }
740 
741    if (unlikely(nv50->num_so_targets && !nv50->gmtyprog))
742       nv50->state.prim_size = nv50_pipe_prim_to_prim_size[info->mode];
743 
744    nv50_state_validate(nv50, ~0, 8); /* 8 as minimum, we use flush_notify */
745 
746    push->kick_notify = nv50_draw_vbo_kick_notify;
747 
748    if (nv50->vbo_fifo) {
749       nv50_push_vbo(nv50, info);
750       push->kick_notify = nv50_default_kick_notify;
751       nouveau_pushbuf_bufctx(push, NULL);
752       return;
753    }
754 
755    if (nv50->state.instance_base != info->start_instance) {
756       nv50->state.instance_base = info->start_instance;
757       /* NOTE: this does not affect the shader input, should it ? */
758       BEGIN_NV04(push, NV50_3D(VB_INSTANCE_BASE), 1);
759       PUSH_DATA (push, info->start_instance);
760    }
761 
762    if (nv50->base.vbo_dirty) {
763       BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_FLUSH), 1);
764       PUSH_DATA (push, 0);
765       nv50->base.vbo_dirty = FALSE;
766    }
767 
768    if (info->indexed) {
769       boolean shorten = info->max_index <= 65535;
770 
771       if (info->primitive_restart != nv50->state.prim_restart) {
772          if (info->primitive_restart) {
773             BEGIN_NV04(push, NV50_3D(PRIM_RESTART_ENABLE), 2);
774             PUSH_DATA (push, 1);
775             PUSH_DATA (push, info->restart_index);
776 
777             if (info->restart_index > 65535)
778                shorten = FALSE;
779          } else {
780             BEGIN_NV04(push, NV50_3D(PRIM_RESTART_ENABLE), 1);
781             PUSH_DATA (push, 0);
782          }
783          nv50->state.prim_restart = info->primitive_restart;
784       } else
785       if (info->primitive_restart) {
786          BEGIN_NV04(push, NV50_3D(PRIM_RESTART_INDEX), 1);
787          PUSH_DATA (push, info->restart_index);
788 
789          if (info->restart_index > 65535)
790             shorten = FALSE;
791       }
792 
793       nv50_draw_elements(nv50, shorten,
794                          info->mode, info->start, info->count,
795                          info->instance_count, info->index_bias);
796    } else
797    if (unlikely(info->count_from_stream_output)) {
798       nva0_draw_stream_output(nv50, info);
799    } else {
800       nv50_draw_arrays(nv50,
801                        info->mode, info->start, info->count,
802                        info->instance_count);
803    }
804    push->kick_notify = nv50_default_kick_notify;
805 
806    nv50_release_user_vbufs(nv50);
807 
808    nouveau_pushbuf_bufctx(push, NULL);
809 }
810