1 /*
2 * Copyright © 2008 Intel Corporation
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 (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 * Eric Anholt <eric@anholt.net>
25 *
26 */
27
28 /** @file intel_syncobj.c
29 *
30 * Support for ARB_sync
31 *
32 * ARB_sync is implemented by flushing the current batchbuffer and keeping a
33 * reference on it. We can then check for completion or wait for completion
34 * using the normal buffer object mechanisms. This does mean that if an
35 * application is using many sync objects, it will emit small batchbuffers
36 * which may end up being a significant overhead. In other tests of removing
37 * gratuitous batchbuffer syncs in Mesa, it hasn't appeared to be a significant
38 * performance bottleneck, though.
39 */
40
41 #include "main/simple_list.h"
42 #include "main/imports.h"
43
44 #include "intel_context.h"
45 #include "intel_batchbuffer.h"
46 #include "intel_reg.h"
47
48 static struct gl_sync_object *
intel_new_sync_object(struct gl_context * ctx,GLuint id)49 intel_new_sync_object(struct gl_context *ctx, GLuint id)
50 {
51 struct intel_sync_object *sync;
52
53 sync = calloc(1, sizeof(struct intel_sync_object));
54
55 return &sync->Base;
56 }
57
58 static void
intel_delete_sync_object(struct gl_context * ctx,struct gl_sync_object * s)59 intel_delete_sync_object(struct gl_context *ctx, struct gl_sync_object *s)
60 {
61 struct intel_sync_object *sync = (struct intel_sync_object *)s;
62
63 drm_intel_bo_unreference(sync->bo);
64 free(sync);
65 }
66
67 static void
intel_fence_sync(struct gl_context * ctx,struct gl_sync_object * s,GLenum condition,GLbitfield flags)68 intel_fence_sync(struct gl_context *ctx, struct gl_sync_object *s,
69 GLenum condition, GLbitfield flags)
70 {
71 struct intel_context *intel = intel_context(ctx);
72 struct intel_sync_object *sync = (struct intel_sync_object *)s;
73
74 assert(condition == GL_SYNC_GPU_COMMANDS_COMPLETE);
75 intel_batchbuffer_emit_mi_flush(intel);
76
77 sync->bo = intel->batch.bo;
78 drm_intel_bo_reference(sync->bo);
79
80 intel_flush(ctx);
81 }
82
83 /* We ignore the user-supplied timeout. This is weaselly -- we're allowed to
84 * round to an implementation-dependent accuracy, and right now our
85 * implementation "rounds" to the wait-forever value.
86 *
87 * The fix would be a new kernel function to do the GTT transition with a
88 * timeout.
89 */
intel_client_wait_sync(struct gl_context * ctx,struct gl_sync_object * s,GLbitfield flags,GLuint64 timeout)90 static void intel_client_wait_sync(struct gl_context *ctx, struct gl_sync_object *s,
91 GLbitfield flags, GLuint64 timeout)
92 {
93 struct intel_sync_object *sync = (struct intel_sync_object *)s;
94
95 if (sync->bo) {
96 drm_intel_bo_wait_rendering(sync->bo);
97 s->StatusFlag = 1;
98 drm_intel_bo_unreference(sync->bo);
99 sync->bo = NULL;
100 }
101 }
102
103 /* We have nothing to do for WaitSync. Our GL command stream is sequential,
104 * so given that the sync object has already flushed the batchbuffer,
105 * any batchbuffers coming after this waitsync will naturally not occur until
106 * the previous one is done.
107 */
intel_server_wait_sync(struct gl_context * ctx,struct gl_sync_object * s,GLbitfield flags,GLuint64 timeout)108 static void intel_server_wait_sync(struct gl_context *ctx, struct gl_sync_object *s,
109 GLbitfield flags, GLuint64 timeout)
110 {
111 }
112
intel_check_sync(struct gl_context * ctx,struct gl_sync_object * s)113 static void intel_check_sync(struct gl_context *ctx, struct gl_sync_object *s)
114 {
115 struct intel_sync_object *sync = (struct intel_sync_object *)s;
116
117 if (sync->bo && !drm_intel_bo_busy(sync->bo)) {
118 drm_intel_bo_unreference(sync->bo);
119 sync->bo = NULL;
120 s->StatusFlag = 1;
121 }
122 }
123
intel_init_syncobj_functions(struct dd_function_table * functions)124 void intel_init_syncobj_functions(struct dd_function_table *functions)
125 {
126 functions->NewSyncObject = intel_new_sync_object;
127 functions->DeleteSyncObject = intel_delete_sync_object;
128 functions->FenceSync = intel_fence_sync;
129 functions->CheckSync = intel_check_sync;
130 functions->ClientWaitSync = intel_client_wait_sync;
131 functions->ServerWaitSync = intel_server_wait_sync;
132 }
133