1/*
2 * Copyright 2011 Nouveau Project
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 * Authors: Christoph Bumiller
23 */
24
25#define NV50_PUSH_EXPLICIT_SPACE_CHECKING
26
27#include "nv50_context.h"
28#include "nouveau/nv_object.xml.h"
29
30/* XXX: Nested queries, and simultaneous queries on multiple gallium contexts
31 * (since we use only a single GPU channel per screen) will not work properly.
32 *
33 * The first is not that big of an issue because OpenGL does not allow nested
34 * queries anyway.
35 */
36
37struct nv50_query {
38   uint32_t *data;
39   uint16_t type;
40   uint16_t index;
41   uint32_t sequence;
42   struct nouveau_bo *bo;
43   uint32_t base;
44   uint32_t offset; /* base + i * 16 */
45   boolean ready;
46   boolean flushed;
47   boolean is64bit;
48   struct nouveau_mm_allocation *mm;
49};
50
51#define NV50_QUERY_ALLOC_SPACE 128
52
53static INLINE struct nv50_query *
54nv50_query(struct pipe_query *pipe)
55{
56   return (struct nv50_query *)pipe;
57}
58
59static boolean
60nv50_query_allocate(struct nv50_context *nv50, struct nv50_query *q, int size)
61{
62   struct nv50_screen *screen = nv50->screen;
63   int ret;
64
65   if (q->bo) {
66      nouveau_bo_ref(NULL, &q->bo);
67      if (q->mm) {
68         if (q->ready)
69            nouveau_mm_free(q->mm);
70         else
71            nouveau_fence_work(screen->base.fence.current, nouveau_mm_free_work,
72                               q->mm);
73      }
74   }
75   if (size) {
76      q->mm = nouveau_mm_allocate(screen->base.mm_GART, size, &q->bo, &q->base);
77      if (!q->bo)
78         return FALSE;
79      q->offset = q->base;
80
81      ret = nouveau_bo_map(q->bo, 0, screen->base.client);
82      if (ret) {
83         nv50_query_allocate(nv50, q, 0);
84         return FALSE;
85      }
86      q->data = (uint32_t *)((uint8_t *)q->bo->map + q->base);
87   }
88   return TRUE;
89}
90
91static void
92nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
93{
94   nv50_query_allocate(nv50_context(pipe), nv50_query(pq), 0);
95   FREE(nv50_query(pq));
96}
97
98static struct pipe_query *
99nv50_query_create(struct pipe_context *pipe, unsigned type)
100{
101   struct nv50_context *nv50 = nv50_context(pipe);
102   struct nv50_query *q;
103
104   q = CALLOC_STRUCT(nv50_query);
105   if (!q)
106      return NULL;
107
108   if (!nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE)) {
109      FREE(q);
110      return NULL;
111   }
112
113   q->is64bit = (type == PIPE_QUERY_PRIMITIVES_GENERATED ||
114                 type == PIPE_QUERY_PRIMITIVES_EMITTED ||
115                 type == PIPE_QUERY_SO_STATISTICS);
116   q->type = type;
117
118   if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
119      q->offset -= 16;
120      q->data -= 16 / sizeof(*q->data); /* we advance before query_begin ! */
121   }
122
123   return (struct pipe_query *)q;
124}
125
126static void
127nv50_query_get(struct nouveau_pushbuf *push, struct nv50_query *q,
128               unsigned offset, uint32_t get)
129{
130   offset += q->offset;
131
132   PUSH_SPACE(push, 5);
133   PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_WR);
134   BEGIN_NV04(push, NV50_3D(QUERY_ADDRESS_HIGH), 4);
135   PUSH_DATAh(push, q->bo->offset + offset);
136   PUSH_DATA (push, q->bo->offset + offset);
137   PUSH_DATA (push, q->sequence);
138   PUSH_DATA (push, get);
139}
140
141static void
142nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
143{
144   struct nv50_context *nv50 = nv50_context(pipe);
145   struct nouveau_pushbuf *push = nv50->base.pushbuf;
146   struct nv50_query *q = nv50_query(pq);
147
148   /* For occlusion queries we have to change the storage, because a previous
149    * query might set the initial render conition to FALSE even *after* we re-
150    * initialized it to TRUE.
151    */
152   if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
153      q->offset += 16;
154      q->data += 16 / sizeof(*q->data);
155      if (q->offset - q->base == NV50_QUERY_ALLOC_SPACE)
156         nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE);
157
158      /* XXX: can we do this with the GPU, and sync with respect to a previous
159       *  query ?
160       */
161      q->data[1] = 1; /* initial render condition = TRUE */
162   }
163   if (!q->is64bit)
164      q->data[0] = q->sequence++; /* the previously used one */
165
166   switch (q->type) {
167   case PIPE_QUERY_OCCLUSION_COUNTER:
168      PUSH_SPACE(push, 4);
169      BEGIN_NV04(push, NV50_3D(COUNTER_RESET), 1);
170      PUSH_DATA (push, NV50_3D_COUNTER_RESET_SAMPLECNT);
171      BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
172      PUSH_DATA (push, 1);
173      break;
174   case PIPE_QUERY_PRIMITIVES_GENERATED:
175      nv50_query_get(push, q, 0x10, 0x06805002);
176      break;
177   case PIPE_QUERY_PRIMITIVES_EMITTED:
178      nv50_query_get(push, q, 0x10, 0x05805002);
179      break;
180   case PIPE_QUERY_SO_STATISTICS:
181      nv50_query_get(push, q, 0x20, 0x05805002);
182      nv50_query_get(push, q, 0x30, 0x06805002);
183      break;
184   case PIPE_QUERY_TIMESTAMP_DISJOINT:
185   case PIPE_QUERY_TIME_ELAPSED:
186      nv50_query_get(push, q, 0x10, 0x00005002);
187      break;
188   default:
189      break;
190   }
191   q->ready = FALSE;
192}
193
194static void
195nv50_query_end(struct pipe_context *pipe, struct pipe_query *pq)
196{
197   struct nv50_context *nv50 = nv50_context(pipe);
198   struct nouveau_pushbuf *push = nv50->base.pushbuf;
199   struct nv50_query *q = nv50_query(pq);
200
201   switch (q->type) {
202   case PIPE_QUERY_OCCLUSION_COUNTER:
203      nv50_query_get(push, q, 0, 0x0100f002);
204      PUSH_SPACE(push, 2);
205      BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
206      PUSH_DATA (push, 0);
207      break;
208   case PIPE_QUERY_PRIMITIVES_GENERATED:
209      nv50_query_get(push, q, 0, 0x06805002);
210      break;
211   case PIPE_QUERY_PRIMITIVES_EMITTED:
212      nv50_query_get(push, q, 0, 0x05805002);
213      break;
214   case PIPE_QUERY_SO_STATISTICS:
215      nv50_query_get(push, q, 0x00, 0x05805002);
216      nv50_query_get(push, q, 0x10, 0x06805002);
217      break;
218   case PIPE_QUERY_TIMESTAMP:
219      q->sequence++;
220      /* fall through */
221   case PIPE_QUERY_TIMESTAMP_DISJOINT:
222   case PIPE_QUERY_TIME_ELAPSED:
223      nv50_query_get(push, q, 0, 0x00005002);
224      break;
225   case PIPE_QUERY_GPU_FINISHED:
226      q->sequence++;
227      nv50_query_get(push, q, 0, 0x1000f010);
228      break;
229   case NVA0_QUERY_STREAM_OUTPUT_BUFFER_OFFSET:
230      nv50_query_get(push, q, 0, 0x0d005002 | (q->index << 5));
231      break;
232   default:
233      assert(0);
234      break;
235   }
236   q->ready = q->flushed = FALSE;
237}
238
239static INLINE boolean
240nv50_query_ready(struct nv50_query *q)
241{
242   return q->ready || (!q->is64bit && (q->data[0] == q->sequence));
243}
244
245static boolean
246nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq,
247                  boolean wait, union pipe_query_result *result)
248{
249   struct nv50_context *nv50 = nv50_context(pipe);
250   struct nv50_query *q = nv50_query(pq);
251   uint64_t *res64 = (uint64_t *)result;
252   uint32_t *res32 = (uint32_t *)result;
253   boolean *res8 = (boolean *)result;
254   uint64_t *data64 = (uint64_t *)q->data;
255
256   if (!q->ready) /* update ? */
257      q->ready = nv50_query_ready(q);
258   if (!q->ready) {
259      if (!wait) {
260         /* for broken apps that spin on GL_QUERY_RESULT_AVAILABLE */
261         if (!q->flushed) {
262            q->flushed = TRUE;
263            PUSH_KICK(nv50->base.pushbuf);
264         }
265         return FALSE;
266      }
267      if (nouveau_bo_wait(q->bo, NOUVEAU_BO_RD, nv50->screen->base.client))
268         return FALSE;
269   }
270   q->ready = TRUE;
271
272   switch (q->type) {
273   case PIPE_QUERY_GPU_FINISHED:
274      res8[0] = TRUE;
275      break;
276   case PIPE_QUERY_OCCLUSION_COUNTER: /* u32 sequence, u32 count, u64 time */
277      res64[0] = q->data[1];
278      break;
279   case PIPE_QUERY_PRIMITIVES_GENERATED: /* u64 count, u64 time */
280   case PIPE_QUERY_PRIMITIVES_EMITTED: /* u64 count, u64 time */
281      res64[0] = data64[0] - data64[2];
282      break;
283   case PIPE_QUERY_SO_STATISTICS:
284      res64[0] = data64[0] - data64[4];
285      res64[1] = data64[2] - data64[6];
286      break;
287   case PIPE_QUERY_TIMESTAMP:
288      res64[0] = data64[1];
289      break;
290   case PIPE_QUERY_TIMESTAMP_DISJOINT: /* u32 sequence, u32 0, u64 time */
291      res64[0] = 1000000000;
292      res8[8] = (data64[1] == data64[3]) ? FALSE : TRUE;
293      break;
294   case PIPE_QUERY_TIME_ELAPSED:
295      res64[0] = data64[1] - data64[3];
296      break;
297   case NVA0_QUERY_STREAM_OUTPUT_BUFFER_OFFSET:
298      res32[0] = q->data[1];
299      break;
300   default:
301      return FALSE;
302   }
303
304   return TRUE;
305}
306
307void
308nv84_query_fifo_wait(struct nouveau_pushbuf *push, struct pipe_query *pq)
309{
310   struct nv50_query *q = nv50_query(pq);
311   unsigned offset = q->offset;
312
313   PUSH_SPACE(push, 5);
314   PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
315   BEGIN_NV04(push, SUBC_3D(NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH), 4);
316   PUSH_DATAh(push, q->bo->offset + offset);
317   PUSH_DATA (push, q->bo->offset + offset);
318   PUSH_DATA (push, q->sequence);
319   PUSH_DATA (push, NV84_SUBCHAN_SEMAPHORE_TRIGGER_ACQUIRE_EQUAL);
320}
321
322static void
323nv50_render_condition(struct pipe_context *pipe,
324                      struct pipe_query *pq, uint mode)
325{
326   struct nv50_context *nv50 = nv50_context(pipe);
327   struct nouveau_pushbuf *push = nv50->base.pushbuf;
328   struct nv50_query *q;
329
330   PUSH_SPACE(push, 6);
331
332   if (!pq) {
333      BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
334      PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
335      return;
336   }
337   q = nv50_query(pq);
338
339   if (mode == PIPE_RENDER_COND_WAIT ||
340       mode == PIPE_RENDER_COND_BY_REGION_WAIT) {
341      BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
342      PUSH_DATA (push, 0);
343   }
344
345   BEGIN_NV04(push, NV50_3D(COND_ADDRESS_HIGH), 3);
346   PUSH_DATAh(push, q->bo->offset + q->offset);
347   PUSH_DATA (push, q->bo->offset + q->offset);
348   PUSH_DATA (push, NV50_3D_COND_MODE_RES_NON_ZERO);
349}
350
351void
352nv50_query_pushbuf_submit(struct nouveau_pushbuf *push,
353                          struct pipe_query *pq, unsigned result_offset)
354{
355   struct nv50_query *q = nv50_query(pq);
356
357   /* XXX: does this exist ? */
358#define NV50_IB_ENTRY_1_NO_PREFETCH (0 << (31 - 8))
359
360   nouveau_pushbuf_space(push, 0, 0, 1);
361   nouveau_pushbuf_data(push, q->bo, q->offset + result_offset, 4 |
362                        NV50_IB_ENTRY_1_NO_PREFETCH);
363}
364
365void
366nva0_so_target_save_offset(struct pipe_context *pipe,
367                           struct pipe_stream_output_target *ptarg,
368                           unsigned index, boolean serialize)
369{
370   struct nv50_so_target *targ = nv50_so_target(ptarg);
371
372   if (serialize) {
373      struct nouveau_pushbuf *push = nv50_context(pipe)->base.pushbuf;
374      PUSH_SPACE(push, 2);
375      BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
376      PUSH_DATA (push, 0);
377   }
378
379   nv50_query(targ->pq)->index = index;
380   nv50_query_end(pipe, targ->pq);
381}
382
383void
384nv50_init_query_functions(struct nv50_context *nv50)
385{
386   struct pipe_context *pipe = &nv50->base.pipe;
387
388   pipe->create_query = nv50_query_create;
389   pipe->destroy_query = nv50_query_destroy;
390   pipe->begin_query = nv50_query_begin;
391   pipe->end_query = nv50_query_end;
392   pipe->get_query_result = nv50_query_result;
393   pipe->render_condition = nv50_render_condition;
394}
395