svga: Fix multiple uploads of the same user-buffer.
[mesa.git] / src / gallium / drivers / svga / svga_pipe_draw.c
1 /**********************************************************
2 * Copyright 2008-2009 VMware, Inc. All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26 #include "svga_cmd.h"
27
28 #include "util/u_format.h"
29 #include "util/u_inlines.h"
30 #include "util/u_prim.h"
31 #include "util/u_time.h"
32 #include "indices/u_indices.h"
33
34 #include "svga_hw_reg.h"
35 #include "svga_context.h"
36 #include "svga_screen.h"
37 #include "svga_draw.h"
38 #include "svga_state.h"
39 #include "svga_swtnl.h"
40 #include "svga_debug.h"
41 #include "svga_resource_buffer.h"
42 #include "util/u_upload_mgr.h"
43
44 /**
45 * Determine the ranges to upload for the user-buffers referenced
46 * by the next draw command.
47 *
48 * TODO: It might be beneficial to support multiple ranges. In that case,
49 * the struct svga_buffer::uploaded member should be made an array or a
50 * list, since we need to account for the possibility that different ranges
51 * may be uploaded to different hardware buffers chosen by the utility
52 * upload manager.
53 */
54
55 static void
56 svga_user_buffer_range(struct svga_context *svga,
57 unsigned start,
58 unsigned count,
59 unsigned instance_count)
60 {
61 const struct pipe_vertex_element *ve = svga->curr.velems->velem;
62 int i;
63
64 /*
65 * Release old uploaded range (if not done already) and
66 * initialize new ranges.
67 */
68
69 for (i=0; i < svga->curr.velems->count; i++) {
70 struct pipe_vertex_buffer *vb =
71 &svga->curr.vb[ve[i].vertex_buffer_index];
72
73 if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
74 struct svga_buffer *buffer = svga_buffer(vb->buffer);
75
76 pipe_resource_reference(&buffer->uploaded.buffer, NULL);
77 buffer->uploaded.start = ~0;
78 buffer->uploaded.end = 0;
79 }
80 }
81
82 for (i=0; i < svga->curr.velems->count; i++) {
83 struct pipe_vertex_buffer *vb =
84 &svga->curr.vb[ve[i].vertex_buffer_index];
85
86 if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
87 struct svga_buffer *buffer = svga_buffer(vb->buffer);
88 unsigned first, size;
89 unsigned instance_div = ve[i].instance_divisor;
90 unsigned elemSize = util_format_get_blocksize(ve->src_format);
91
92 svga->dirty |= SVGA_NEW_VBUFFER;
93
94 if (instance_div) {
95 first = ve[i].src_offset;
96 count = (instance_count + instance_div - 1) / instance_div;
97 size = vb->stride * (count - 1) + elemSize;
98 } else if (vb->stride) {
99 first = vb->stride * start + ve[i].src_offset;
100 size = vb->stride * (count - 1) + elemSize;
101 } else {
102 /* Only a single vertex!
103 * Upload with the largest vertex size the hw supports,
104 * if possible.
105 */
106 first = ve[i].src_offset;
107 size = MIN2(16, vb->buffer->width0);
108 }
109
110 buffer->uploaded.start = MIN2(buffer->uploaded.start, first);
111 buffer->uploaded.end = MAX2(buffer->uploaded.end, first + size);
112 }
113 }
114 }
115
116 /**
117 * svga_upload_user_buffers - upload parts of user buffers
118 *
119 * This function streams a part of a user buffer to hw and fills
120 * svga_buffer::uploaded with information on the upload.
121 */
122
123 static int
124 svga_upload_user_buffers(struct svga_context *svga,
125 unsigned start,
126 unsigned count,
127 unsigned instance_count)
128 {
129 const struct pipe_vertex_element *ve = svga->curr.velems->velem;
130 unsigned i;
131 int ret;
132
133 svga_user_buffer_range(svga, start, count, instance_count);
134
135 for (i=0; i < svga->curr.velems->count; i++) {
136 struct pipe_vertex_buffer *vb =
137 &svga->curr.vb[ve[i].vertex_buffer_index];
138
139 if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
140 struct svga_buffer *buffer = svga_buffer(vb->buffer);
141 boolean flushed;
142
143 /*
144 * Check if already uploaded. Otherwise go ahead and upload.
145 */
146
147 if (buffer->uploaded.buffer)
148 continue;
149
150 ret = u_upload_buffer( svga->upload_vb,
151 0,
152 buffer->uploaded.start,
153 buffer->uploaded.end - buffer->uploaded.start,
154 &buffer->b.b,
155 &buffer->uploaded.offset,
156 &buffer->uploaded.buffer,
157 &flushed);
158
159 if (ret)
160 return ret;
161
162 if (0)
163 debug_printf("%s: %d: orig buf %p upl buf %p ofs %d sofs %d"
164 " sz %d\n",
165 __FUNCTION__,
166 i,
167 buffer,
168 buffer->uploaded.buffer,
169 buffer->uploaded.offset,
170 buffer->uploaded.start,
171 buffer->uploaded.end - buffer->uploaded.start);
172
173 vb->buffer_offset = buffer->uploaded.offset;
174 }
175 }
176
177 return PIPE_OK;
178 }
179
180 /**
181 * svga_release_user_upl_buffers - release uploaded parts of user buffers
182 *
183 * This function releases the hw copy of the uploaded fraction of the
184 * user-buffer. It's important to do this as soon as all draw calls
185 * affecting the uploaded fraction are issued, as this allows for
186 * efficient reuse of the hardware surface backing the uploaded fraction.
187 *
188 * svga_buffer::source_offset is set to 0, and svga_buffer::uploaded::buffer
189 * is set to 0.
190 */
191
192 static void
193 svga_release_user_upl_buffers(struct svga_context *svga)
194 {
195 unsigned i;
196 unsigned nr;
197
198 nr = svga->curr.num_vertex_buffers;
199
200 for (i = 0; i < nr; ++i) {
201 struct pipe_vertex_buffer *vb = &svga->curr.vb[i];
202
203 if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
204 struct svga_buffer *buffer = svga_buffer(vb->buffer);
205
206 buffer->uploaded.start = ~0;
207 buffer->uploaded.end = 0;
208 if (buffer->uploaded.buffer)
209 pipe_resource_reference(&buffer->uploaded.buffer, NULL);
210 }
211 }
212 }
213
214
215
216 static enum pipe_error
217 retry_draw_range_elements( struct svga_context *svga,
218 struct pipe_resource *index_buffer,
219 unsigned index_size,
220 int index_bias,
221 unsigned min_index,
222 unsigned max_index,
223 unsigned prim,
224 unsigned start,
225 unsigned count,
226 unsigned instance_count,
227 boolean do_retry )
228 {
229 enum pipe_error ret = 0;
230
231 svga_hwtnl_set_unfilled( svga->hwtnl,
232 svga->curr.rast->hw_unfilled );
233
234 svga_hwtnl_set_flatshade( svga->hwtnl,
235 svga->curr.rast->templ.flatshade,
236 svga->curr.rast->templ.flatshade_first );
237
238 ret = svga_upload_user_buffers( svga, min_index + index_bias,
239 max_index - min_index + 1, instance_count );
240 if (ret != PIPE_OK)
241 goto retry;
242
243 ret = svga_update_state( svga, SVGA_STATE_HW_DRAW );
244 if (ret)
245 goto retry;
246
247 ret = svga_hwtnl_draw_range_elements( svga->hwtnl,
248 index_buffer, index_size, index_bias,
249 min_index, max_index,
250 prim, start, count );
251 if (ret)
252 goto retry;
253
254 return PIPE_OK;
255
256 retry:
257 svga_context_flush( svga, NULL );
258
259 if (do_retry)
260 {
261 return retry_draw_range_elements( svga,
262 index_buffer, index_size, index_bias,
263 min_index, max_index,
264 prim, start, count,
265 instance_count, FALSE );
266 }
267
268 return ret;
269 }
270
271
272 static enum pipe_error
273 retry_draw_arrays( struct svga_context *svga,
274 unsigned prim,
275 unsigned start,
276 unsigned count,
277 unsigned instance_count,
278 boolean do_retry )
279 {
280 enum pipe_error ret;
281
282 svga_hwtnl_set_unfilled( svga->hwtnl,
283 svga->curr.rast->hw_unfilled );
284
285 svga_hwtnl_set_flatshade( svga->hwtnl,
286 svga->curr.rast->templ.flatshade,
287 svga->curr.rast->templ.flatshade_first );
288
289 ret = svga_upload_user_buffers( svga, start, count, instance_count );
290
291 if (ret != PIPE_OK)
292 goto retry;
293
294 ret = svga_update_state( svga, SVGA_STATE_HW_DRAW );
295 if (ret)
296 goto retry;
297
298 ret = svga_hwtnl_draw_arrays( svga->hwtnl, prim,
299 start, count );
300 if (ret)
301 goto retry;
302
303 return 0;
304
305 retry:
306 if (ret == PIPE_ERROR_OUT_OF_MEMORY && do_retry)
307 {
308 svga_context_flush( svga, NULL );
309
310 return retry_draw_arrays( svga,
311 prim,
312 start,
313 count,
314 instance_count,
315 FALSE );
316 }
317
318 return ret;
319 }
320
321
322 static void
323 svga_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
324 {
325 struct svga_context *svga = svga_context( pipe );
326 unsigned reduced_prim = u_reduced_prim( info->mode );
327 unsigned count = info->count;
328 enum pipe_error ret = 0;
329 boolean needed_swtnl;
330
331 if (!u_trim_pipe_prim( info->mode, &count ))
332 return;
333
334 /*
335 * Mark currently bound target surfaces as dirty
336 * doesn't really matter if it is done before drawing.
337 *
338 * TODO If we ever normaly return something other then
339 * true we should not mark it as dirty then.
340 */
341 svga_mark_surfaces_dirty(svga_context(pipe));
342
343 if (svga->curr.reduced_prim != reduced_prim) {
344 svga->curr.reduced_prim = reduced_prim;
345 svga->dirty |= SVGA_NEW_REDUCED_PRIMITIVE;
346 }
347
348 needed_swtnl = svga->state.sw.need_swtnl;
349
350 svga_update_state_retry( svga, SVGA_STATE_NEED_SWTNL );
351
352 #ifdef DEBUG
353 if (svga->curr.vs->base.id == svga->debug.disable_shader ||
354 svga->curr.fs->base.id == svga->debug.disable_shader)
355 return;
356 #endif
357
358 if (svga->state.sw.need_swtnl) {
359 if (!needed_swtnl) {
360 /*
361 * We're switching from HW to SW TNL. SW TNL will require mapping all
362 * currently bound vertex buffers, some of which may already be
363 * referenced in the current command buffer as result of previous HW
364 * TNL. So flush now, to prevent the context to flush while a referred
365 * vertex buffer is mapped.
366 */
367
368 svga_context_flush(svga, NULL);
369 }
370
371 /* Avoid leaking the previous hwtnl bias to swtnl */
372 svga_hwtnl_set_index_bias( svga->hwtnl, 0 );
373 ret = svga_swtnl_draw_vbo( svga, info );
374 }
375 else {
376 if (info->indexed && svga->curr.ib.buffer) {
377 unsigned offset;
378
379 assert(svga->curr.ib.offset % svga->curr.ib.index_size == 0);
380 offset = svga->curr.ib.offset / svga->curr.ib.index_size;
381
382 ret = retry_draw_range_elements( svga,
383 svga->curr.ib.buffer,
384 svga->curr.ib.index_size,
385 info->index_bias,
386 info->min_index,
387 info->max_index,
388 info->mode,
389 info->start + offset,
390 info->count,
391 info->instance_count,
392 TRUE );
393 }
394 else {
395 ret = retry_draw_arrays( svga,
396 info->mode,
397 info->start,
398 info->count,
399 info->instance_count,
400 TRUE );
401 }
402 }
403
404 svga_release_user_upl_buffers( svga );
405
406 if (SVGA_DEBUG & DEBUG_FLUSH) {
407 svga_hwtnl_flush_retry( svga );
408 svga_context_flush(svga, NULL);
409 }
410 }
411
412
413 void svga_init_draw_functions( struct svga_context *svga )
414 {
415 svga->pipe.draw_vbo = svga_draw_vbo;
416 }