2 * Copyright © 2011 Marek Olšák <maraeo@gmail.com>
3 * Copyright © 2015 Advanced Micro Devices, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining
7 * a 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:
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
15 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
16 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
17 * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS
18 * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
23 * The above copyright notice and this permission notice (including the
24 * next paragraph) shall be included in all copies or substantial portions
29 * Marek Olšák <maraeo@gmail.com>
35 #include "amdgpu_bo.h"
36 #include "util/u_memory.h"
39 struct amdgpu_winsys
*ws
;
40 amdgpu_context_handle ctx
;
41 amdgpu_bo_handle user_fence_bo
;
42 uint64_t *user_fence_cpu_address_base
;
44 unsigned initial_num_total_rejected_cs
;
45 unsigned num_rejected_cs
;
48 struct amdgpu_cs_buffer
{
49 struct amdgpu_winsys_bo
*bo
;
52 uint64_t priority_usage
;
55 uint32_t real_idx
; /* index of underlying real BO */
58 enum radeon_bo_usage usage
;
67 struct radeon_winsys_cs base
;
69 /* A buffer out of which new IBs are allocated. */
70 struct pb_buffer
*big_ib_buffer
;
72 unsigned used_ib_space
;
74 uint32_t *ptr_ib_size
;
78 struct amdgpu_cs_context
{
79 struct amdgpu_cs_request request
;
80 struct amdgpu_cs_ib_info ib
[IB_NUM
];
83 unsigned max_real_buffers
;
84 unsigned num_real_buffers
;
85 struct amdgpu_cs_buffer
*real_buffers
;
87 unsigned max_real_submit
;
88 amdgpu_bo_handle
*handles
;
91 unsigned num_slab_buffers
;
92 unsigned max_slab_buffers
;
93 struct amdgpu_cs_buffer
*slab_buffers
;
95 unsigned num_sparse_buffers
;
96 unsigned max_sparse_buffers
;
97 struct amdgpu_cs_buffer
*sparse_buffers
;
99 int buffer_indices_hashlist
[4096];
101 struct amdgpu_winsys_bo
*last_added_bo
;
102 unsigned last_added_bo_index
;
103 unsigned last_added_bo_usage
;
104 uint64_t last_added_bo_priority_usage
;
106 struct pipe_fence_handle
**fence_dependencies
;
107 unsigned num_fence_dependencies
;
108 unsigned max_fence_dependencies
;
110 struct pipe_fence_handle
*fence
;
112 /* the error returned from cs_flush for non-async submissions */
117 struct amdgpu_ib main
; /* must be first because this is inherited */
118 struct amdgpu_ctx
*ctx
;
119 enum ring_type ring_type
;
121 /* We flip between these two CS. While one is being consumed
122 * by the kernel in another thread, the other one is being filled
123 * by the pipe driver. */
124 struct amdgpu_cs_context csc1
;
125 struct amdgpu_cs_context csc2
;
126 /* The currently-used CS. */
127 struct amdgpu_cs_context
*csc
;
128 /* The CS being currently-owned by the other thread. */
129 struct amdgpu_cs_context
*cst
;
132 void (*flush_cs
)(void *ctx
, unsigned flags
, struct pipe_fence_handle
**fence
);
135 struct util_queue_fence flush_completed
;
136 struct pipe_fence_handle
*next_fence
;
139 struct amdgpu_fence
{
140 struct pipe_reference reference
;
142 struct amdgpu_ctx
*ctx
; /* submission context */
143 struct amdgpu_cs_fence fence
;
144 uint64_t *user_fence_cpu_address
;
146 /* If the fence is unknown due to an IB still being submitted
147 * in the other thread. */
148 volatile int submission_in_progress
; /* bool (int for atomicity) */
149 volatile int signalled
; /* bool (int for atomicity) */
152 static inline void amdgpu_ctx_unref(struct amdgpu_ctx
*ctx
)
154 if (p_atomic_dec_zero(&ctx
->refcount
)) {
155 amdgpu_cs_ctx_free(ctx
->ctx
);
156 amdgpu_bo_free(ctx
->user_fence_bo
);
161 static inline void amdgpu_fence_reference(struct pipe_fence_handle
**dst
,
162 struct pipe_fence_handle
*src
)
164 struct amdgpu_fence
**rdst
= (struct amdgpu_fence
**)dst
;
165 struct amdgpu_fence
*rsrc
= (struct amdgpu_fence
*)src
;
167 if (pipe_reference(&(*rdst
)->reference
, &rsrc
->reference
)) {
168 amdgpu_ctx_unref((*rdst
)->ctx
);
174 int amdgpu_lookup_buffer(struct amdgpu_cs_context
*cs
, struct amdgpu_winsys_bo
*bo
);
176 static inline struct amdgpu_ib
*
177 amdgpu_ib(struct radeon_winsys_cs
*base
)
179 return (struct amdgpu_ib
*)base
;
182 static inline struct amdgpu_cs
*
183 amdgpu_cs(struct radeon_winsys_cs
*base
)
185 assert(amdgpu_ib(base
)->ib_type
== IB_MAIN
);
186 return (struct amdgpu_cs
*)base
;
189 #define get_container(member_ptr, container_type, container_member) \
190 (container_type *)((char *)(member_ptr) - offsetof(container_type, container_member))
192 static inline struct amdgpu_cs
*
193 amdgpu_cs_from_ib(struct amdgpu_ib
*ib
)
195 switch (ib
->ib_type
) {
197 return get_container(ib
, struct amdgpu_cs
, main
);
199 unreachable("bad ib_type");
204 amdgpu_bo_is_referenced_by_cs(struct amdgpu_cs
*cs
,
205 struct amdgpu_winsys_bo
*bo
)
207 int num_refs
= bo
->num_cs_references
;
208 return num_refs
== bo
->ws
->num_cs
||
209 (num_refs
&& amdgpu_lookup_buffer(cs
->csc
, bo
) != -1);
213 amdgpu_bo_is_referenced_by_cs_with_usage(struct amdgpu_cs
*cs
,
214 struct amdgpu_winsys_bo
*bo
,
215 enum radeon_bo_usage usage
)
218 struct amdgpu_cs_buffer
*buffer
;
220 if (!bo
->num_cs_references
)
223 index
= amdgpu_lookup_buffer(cs
->csc
, bo
);
227 buffer
= bo
->bo
? &cs
->csc
->real_buffers
[index
] :
228 bo
->sparse
? &cs
->csc
->sparse_buffers
[index
] :
229 &cs
->csc
->slab_buffers
[index
];
231 return (buffer
->usage
& usage
) != 0;
235 amdgpu_bo_is_referenced_by_any_cs(struct amdgpu_winsys_bo
*bo
)
237 return bo
->num_cs_references
!= 0;
240 bool amdgpu_fence_wait(struct pipe_fence_handle
*fence
, uint64_t timeout
,
242 void amdgpu_add_fences(struct amdgpu_winsys_bo
*bo
,
244 struct pipe_fence_handle
**fences
);
245 void amdgpu_cs_sync_flush(struct radeon_winsys_cs
*rcs
);
246 void amdgpu_cs_init_functions(struct amdgpu_winsys
*ws
);
247 void amdgpu_cs_submit_ib(void *job
, int thread_index
);