Merge branch 'gallium-newclear'
[mesa.git] / src / mesa / drivers / dri / intel / intel_batchbuffer.c
1 /**************************************************************************
2 *
3 * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 #include "intel_context.h"
29 #include "intel_batchbuffer.h"
30 #include "intel_decode.h"
31 #include "intel_reg.h"
32 #include "intel_bufmgr.h"
33 #include "intel_buffers.h"
34
35 void
36 intel_batchbuffer_reset(struct intel_batchbuffer *batch)
37 {
38 struct intel_context *intel = batch->intel;
39
40 if (batch->buf != NULL) {
41 dri_bo_unreference(batch->buf);
42 batch->buf = NULL;
43 }
44
45 if (!batch->buffer)
46 batch->buffer = malloc (intel->maxBatchSize);
47
48 batch->buf = dri_bo_alloc(intel->bufmgr, "batchbuffer",
49 intel->maxBatchSize, 4096);
50 if (batch->buffer)
51 batch->map = batch->buffer;
52 else {
53 dri_bo_map(batch->buf, GL_TRUE);
54 batch->map = batch->buf->virtual;
55 }
56 batch->size = intel->maxBatchSize;
57 batch->ptr = batch->map;
58 batch->reserved_space = BATCH_RESERVED;
59 batch->dirty_state = ~0;
60 }
61
62 struct intel_batchbuffer *
63 intel_batchbuffer_alloc(struct intel_context *intel)
64 {
65 struct intel_batchbuffer *batch = calloc(sizeof(*batch), 1);
66
67 batch->intel = intel;
68 intel_batchbuffer_reset(batch);
69
70 return batch;
71 }
72
73 void
74 intel_batchbuffer_free(struct intel_batchbuffer *batch)
75 {
76 if (batch->buffer)
77 free (batch->buffer);
78 else {
79 if (batch->map) {
80 dri_bo_unmap(batch->buf);
81 batch->map = NULL;
82 }
83 }
84 dri_bo_unreference(batch->buf);
85 batch->buf = NULL;
86 free(batch);
87 }
88
89
90
91 /* TODO: Push this whole function into bufmgr.
92 */
93 static void
94 do_flush_locked(struct intel_batchbuffer *batch, GLuint used)
95 {
96 struct intel_context *intel = batch->intel;
97 int ret = 0;
98 int x_off = 0, y_off = 0;
99
100 if (batch->buffer)
101 dri_bo_subdata (batch->buf, 0, used, batch->buffer);
102 else
103 dri_bo_unmap(batch->buf);
104
105 batch->map = NULL;
106 batch->ptr = NULL;
107
108 if (!intel->no_hw)
109 dri_bo_exec(batch->buf, used, NULL, 0, (x_off & 0xffff) | (y_off << 16));
110
111 if (INTEL_DEBUG & DEBUG_BATCH) {
112 dri_bo_map(batch->buf, GL_FALSE);
113 intel_decode(batch->buf->virtual, used / 4, batch->buf->offset,
114 intel->intelScreen->deviceID);
115 dri_bo_unmap(batch->buf);
116
117 if (intel->vtbl.debug_batch != NULL)
118 intel->vtbl.debug_batch(intel);
119 }
120
121 if (ret != 0) {
122 exit(1);
123 }
124 intel->vtbl.new_batch(intel);
125 }
126
127 void
128 _intel_batchbuffer_flush(struct intel_batchbuffer *batch, const char *file,
129 int line)
130 {
131 struct intel_context *intel = batch->intel;
132 GLuint used = batch->ptr - batch->map;
133
134 if (intel->first_post_swapbuffers_batch == NULL) {
135 intel->first_post_swapbuffers_batch = intel->batch->buf;
136 drm_intel_bo_reference(intel->first_post_swapbuffers_batch);
137 }
138
139 if (used == 0)
140 return;
141
142 if (INTEL_DEBUG & DEBUG_BATCH)
143 fprintf(stderr, "%s:%d: Batchbuffer flush with %db used\n", file, line,
144 used);
145
146 batch->reserved_space = 0;
147 /* Emit a flush if the bufmgr doesn't do it for us. */
148 if (intel->always_flush_cache) {
149 intel_batchbuffer_emit_mi_flush(batch);
150 used = batch->ptr - batch->map;
151 }
152
153 /* Round batchbuffer usage to 2 DWORDs. */
154
155 if ((used & 4) == 0) {
156 *(GLuint *) (batch->ptr) = 0; /* noop */
157 batch->ptr += 4;
158 used = batch->ptr - batch->map;
159 }
160
161 /* Mark the end of the buffer. */
162 *(GLuint *) (batch->ptr) = MI_BATCH_BUFFER_END;
163 batch->ptr += 4;
164 used = batch->ptr - batch->map;
165 assert (used <= batch->buf->size);
166
167 /* Workaround for recursive batchbuffer flushing: If the window is
168 * moved, we can get into a case where we try to flush during a
169 * flush. What happens is that when we try to grab the lock for
170 * the first flush, we detect that the window moved which then
171 * causes another flush (from the intel_draw_buffer() call in
172 * intelUpdatePageFlipping()). To work around this we reset the
173 * batchbuffer tail pointer before trying to get the lock. This
174 * prevent the nested buffer flush, but a better fix would be to
175 * avoid that in the first place. */
176 batch->ptr = batch->map;
177
178 if (intel->vtbl.finish_batch)
179 intel->vtbl.finish_batch(intel);
180
181 /* Check that we didn't just wrap our batchbuffer at a bad time. */
182 assert(!intel->no_batch_wrap);
183
184 /* TODO: Just pass the relocation list and dma buffer up to the
185 * kernel.
186 */
187 do_flush_locked(batch, used);
188
189 if (INTEL_DEBUG & DEBUG_SYNC) {
190 fprintf(stderr, "waiting for idle\n");
191 dri_bo_map(batch->buf, GL_TRUE);
192 dri_bo_unmap(batch->buf);
193 }
194
195 /* Reset the buffer:
196 */
197 intel_batchbuffer_reset(batch);
198 }
199
200
201 /* This is the only way buffers get added to the validate list.
202 */
203 GLboolean
204 intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch,
205 dri_bo *buffer,
206 uint32_t read_domains, uint32_t write_domain,
207 uint32_t delta)
208 {
209 int ret;
210
211 assert(delta < buffer->size);
212
213 if (batch->ptr - batch->map > batch->buf->size)
214 printf ("bad relocation ptr %p map %p offset %d size %lu\n",
215 batch->ptr, batch->map, batch->ptr - batch->map, batch->buf->size);
216 ret = dri_bo_emit_reloc(batch->buf, read_domains, write_domain,
217 delta, batch->ptr - batch->map, buffer);
218
219 /*
220 * Using the old buffer offset, write in what the right data would be, in case
221 * the buffer doesn't move and we can short-circuit the relocation processing
222 * in the kernel
223 */
224 intel_batchbuffer_emit_dword (batch, buffer->offset + delta);
225
226 return GL_TRUE;
227 }
228
229 GLboolean
230 intel_batchbuffer_emit_reloc_fenced(struct intel_batchbuffer *batch,
231 drm_intel_bo *buffer,
232 uint32_t read_domains, uint32_t write_domain,
233 uint32_t delta)
234 {
235 int ret;
236
237 assert(delta < buffer->size);
238
239 if (batch->ptr - batch->map > batch->buf->size)
240 printf ("bad relocation ptr %p map %p offset %d size %lu\n",
241 batch->ptr, batch->map, batch->ptr - batch->map, batch->buf->size);
242 ret = drm_intel_bo_emit_reloc_fence(batch->buf, batch->ptr - batch->map,
243 buffer, delta,
244 read_domains, write_domain);
245
246 /*
247 * Using the old buffer offset, write in what the right data would
248 * be, in case the buffer doesn't move and we can short-circuit the
249 * relocation processing in the kernel
250 */
251 intel_batchbuffer_emit_dword (batch, buffer->offset + delta);
252
253 return GL_TRUE;
254 }
255
256 void
257 intel_batchbuffer_data(struct intel_batchbuffer *batch,
258 const void *data, GLuint bytes)
259 {
260 assert((bytes & 3) == 0);
261 intel_batchbuffer_require_space(batch, bytes);
262 __memcpy(batch->ptr, data, bytes);
263 batch->ptr += bytes;
264 }
265
266 /* Emit a pipelined flush to either flush render and texture cache for
267 * reading from a FBO-drawn texture, or flush so that frontbuffer
268 * render appears on the screen in DRI1.
269 *
270 * This is also used for the always_flush_cache driconf debug option.
271 */
272 void
273 intel_batchbuffer_emit_mi_flush(struct intel_batchbuffer *batch)
274 {
275 struct intel_context *intel = batch->intel;
276
277 if (intel->gen >= 6) {
278 BEGIN_BATCH(4);
279 OUT_BATCH(_3DSTATE_PIPE_CONTROL);
280 OUT_BATCH(PIPE_CONTROL_INSTRUCTION_FLUSH |
281 PIPE_CONTROL_WRITE_FLUSH |
282 PIPE_CONTROL_NO_WRITE);
283 OUT_BATCH(0); /* write address */
284 OUT_BATCH(0); /* write data */
285 ADVANCE_BATCH();
286 } else if (intel->gen >= 4) {
287 BEGIN_BATCH(4);
288 OUT_BATCH(_3DSTATE_PIPE_CONTROL |
289 PIPE_CONTROL_WRITE_FLUSH |
290 PIPE_CONTROL_NO_WRITE);
291 OUT_BATCH(0); /* write address */
292 OUT_BATCH(0); /* write data */
293 OUT_BATCH(0); /* write data */
294 ADVANCE_BATCH();
295 } else {
296 BEGIN_BATCH(1);
297 OUT_BATCH(MI_FLUSH);
298 ADVANCE_BATCH();
299 }
300 }