1/*
2 * Copyright 2015 Advanced Micro Devices, Inc.
3 * All Rights Reserved.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24
25#ifndef SI_QUERY_H
26#define SI_QUERY_H
27
28#include "util/u_threaded_context.h"
29
30struct pipe_context;
31struct pipe_query;
32struct pipe_resource;
33
34struct si_screen;
35struct si_context;
36struct si_query;
37struct si_query_buffer;
38struct si_query_hw;
39struct si_resource;
40
41enum {
42	SI_QUERY_DRAW_CALLS = PIPE_QUERY_DRIVER_SPECIFIC,
43	SI_QUERY_DECOMPRESS_CALLS,
44	SI_QUERY_MRT_DRAW_CALLS,
45	SI_QUERY_PRIM_RESTART_CALLS,
46	SI_QUERY_SPILL_DRAW_CALLS,
47	SI_QUERY_COMPUTE_CALLS,
48	SI_QUERY_SPILL_COMPUTE_CALLS,
49	SI_QUERY_DMA_CALLS,
50	SI_QUERY_CP_DMA_CALLS,
51	SI_QUERY_NUM_VS_FLUSHES,
52	SI_QUERY_NUM_PS_FLUSHES,
53	SI_QUERY_NUM_CS_FLUSHES,
54	SI_QUERY_NUM_CB_CACHE_FLUSHES,
55	SI_QUERY_NUM_DB_CACHE_FLUSHES,
56	SI_QUERY_NUM_L2_INVALIDATES,
57	SI_QUERY_NUM_L2_WRITEBACKS,
58	SI_QUERY_NUM_RESIDENT_HANDLES,
59	SI_QUERY_TC_OFFLOADED_SLOTS,
60	SI_QUERY_TC_DIRECT_SLOTS,
61	SI_QUERY_TC_NUM_SYNCS,
62	SI_QUERY_CS_THREAD_BUSY,
63	SI_QUERY_GALLIUM_THREAD_BUSY,
64	SI_QUERY_REQUESTED_VRAM,
65	SI_QUERY_REQUESTED_GTT,
66	SI_QUERY_MAPPED_VRAM,
67	SI_QUERY_MAPPED_GTT,
68	SI_QUERY_BUFFER_WAIT_TIME,
69	SI_QUERY_NUM_MAPPED_BUFFERS,
70	SI_QUERY_NUM_GFX_IBS,
71	SI_QUERY_NUM_SDMA_IBS,
72	SI_QUERY_GFX_BO_LIST_SIZE,
73	SI_QUERY_GFX_IB_SIZE,
74	SI_QUERY_NUM_BYTES_MOVED,
75	SI_QUERY_NUM_EVICTIONS,
76	SI_QUERY_NUM_VRAM_CPU_PAGE_FAULTS,
77	SI_QUERY_VRAM_USAGE,
78	SI_QUERY_VRAM_VIS_USAGE,
79	SI_QUERY_GTT_USAGE,
80	SI_QUERY_GPU_TEMPERATURE,
81	SI_QUERY_CURRENT_GPU_SCLK,
82	SI_QUERY_CURRENT_GPU_MCLK,
83	SI_QUERY_GPU_LOAD,
84	SI_QUERY_GPU_SHADERS_BUSY,
85	SI_QUERY_GPU_TA_BUSY,
86	SI_QUERY_GPU_GDS_BUSY,
87	SI_QUERY_GPU_VGT_BUSY,
88	SI_QUERY_GPU_IA_BUSY,
89	SI_QUERY_GPU_SX_BUSY,
90	SI_QUERY_GPU_WD_BUSY,
91	SI_QUERY_GPU_BCI_BUSY,
92	SI_QUERY_GPU_SC_BUSY,
93	SI_QUERY_GPU_PA_BUSY,
94	SI_QUERY_GPU_DB_BUSY,
95	SI_QUERY_GPU_CP_BUSY,
96	SI_QUERY_GPU_CB_BUSY,
97	SI_QUERY_GPU_SDMA_BUSY,
98	SI_QUERY_GPU_PFP_BUSY,
99	SI_QUERY_GPU_MEQ_BUSY,
100	SI_QUERY_GPU_ME_BUSY,
101	SI_QUERY_GPU_SURF_SYNC_BUSY,
102	SI_QUERY_GPU_CP_DMA_BUSY,
103	SI_QUERY_GPU_SCRATCH_RAM_BUSY,
104	SI_QUERY_NUM_COMPILATIONS,
105	SI_QUERY_NUM_SHADERS_CREATED,
106	SI_QUERY_BACK_BUFFER_PS_DRAW_RATIO,
107	SI_QUERY_NUM_SHADER_CACHE_HITS,
108	SI_QUERY_GPIN_ASIC_ID,
109	SI_QUERY_GPIN_NUM_SIMD,
110	SI_QUERY_GPIN_NUM_RB,
111	SI_QUERY_GPIN_NUM_SPI,
112	SI_QUERY_GPIN_NUM_SE,
113	SI_QUERY_TIME_ELAPSED_SDMA,
114	SI_QUERY_TIME_ELAPSED_SDMA_SI, /* emulated, measured on the CPU */
115
116	SI_QUERY_FIRST_PERFCOUNTER = PIPE_QUERY_DRIVER_SPECIFIC + 100,
117};
118
119enum {
120	SI_QUERY_GROUP_GPIN = 0,
121	SI_NUM_SW_QUERY_GROUPS
122};
123
124struct si_query_ops {
125	void (*destroy)(struct si_screen *, struct si_query *);
126	bool (*begin)(struct si_context *, struct si_query *);
127	bool (*end)(struct si_context *, struct si_query *);
128	bool (*get_result)(struct si_context *,
129			   struct si_query *, bool wait,
130			   union pipe_query_result *result);
131	void (*get_result_resource)(struct si_context *,
132				    struct si_query *, bool wait,
133				    enum pipe_query_value_type result_type,
134				    int index,
135				    struct pipe_resource *resource,
136				    unsigned offset);
137
138	void (*suspend)(struct si_context *, struct si_query *);
139	void (*resume)(struct si_context *, struct si_query *);
140};
141
142struct si_query {
143	struct threaded_query b;
144	const struct si_query_ops *ops;
145
146	/* The PIPE_QUERY_xxx type of query */
147	unsigned type;
148
149	/* The number of dwords for suspend. */
150	unsigned num_cs_dw_suspend;
151
152	/* Linked list of queries that must be suspended at end of CS. */
153	struct list_head active_list;
154};
155
156enum {
157	SI_QUERY_HW_FLAG_NO_START = (1 << 0),
158	/* gap */
159	/* whether begin_query doesn't clear the result */
160	SI_QUERY_HW_FLAG_BEGIN_RESUMES = (1 << 2),
161};
162
163struct si_query_hw_ops {
164	bool (*prepare_buffer)(struct si_context *, struct si_query_buffer *);
165	void (*emit_start)(struct si_context *,
166			   struct si_query_hw *,
167			   struct si_resource *buffer, uint64_t va);
168	void (*emit_stop)(struct si_context *,
169			  struct si_query_hw *,
170			  struct si_resource *buffer, uint64_t va);
171	void (*clear_result)(struct si_query_hw *, union pipe_query_result *);
172	void (*add_result)(struct si_screen *screen,
173			   struct si_query_hw *, void *buffer,
174			   union pipe_query_result *result);
175};
176
177struct si_query_buffer {
178	/* The buffer where query results are stored. */
179	struct si_resource		*buf;
180	/* If a query buffer is full, a new buffer is created and the old one
181	 * is put in here. When we calculate the result, we sum up the samples
182	 * from all buffers. */
183	struct si_query_buffer	*previous;
184	/* Offset of the next free result after current query data */
185	unsigned			results_end;
186	bool unprepared;
187};
188
189void si_query_buffer_destroy(struct si_screen *sctx, struct si_query_buffer *buffer);
190void si_query_buffer_reset(struct si_context *sctx, struct si_query_buffer *buffer);
191bool si_query_buffer_alloc(struct si_context *sctx, struct si_query_buffer *buffer,
192			   bool (*prepare_buffer)(struct si_context *, struct si_query_buffer*),
193			   unsigned size);
194
195
196struct si_query_hw {
197	struct si_query b;
198	struct si_query_hw_ops *ops;
199	unsigned flags;
200
201	/* The query buffer and how many results are in it. */
202	struct si_query_buffer buffer;
203	/* Size of the result in memory for both begin_query and end_query,
204	 * this can be one or two numbers, or it could even be a size of a structure. */
205	unsigned result_size;
206	/* For transform feedback: which stream the query is for */
207	unsigned stream;
208
209	/* Workaround via compute shader */
210	struct si_resource *workaround_buf;
211	unsigned workaround_offset;
212};
213
214void si_query_hw_destroy(struct si_screen *sscreen,
215			 struct si_query *squery);
216bool si_query_hw_begin(struct si_context *sctx,
217		       struct si_query *squery);
218bool si_query_hw_end(struct si_context *sctx,
219		     struct si_query *squery);
220bool si_query_hw_get_result(struct si_context *sctx,
221			    struct si_query *squery,
222			    bool wait,
223			    union pipe_query_result *result);
224void si_query_hw_suspend(struct si_context *sctx, struct si_query *query);
225void si_query_hw_resume(struct si_context *sctx, struct si_query *query);
226
227
228/* Performance counters */
229struct si_perfcounters {
230	unsigned num_groups;
231	unsigned num_blocks;
232	struct si_pc_block *blocks;
233
234	unsigned num_stop_cs_dwords;
235	unsigned num_instance_cs_dwords;
236
237	bool separate_se;
238	bool separate_instance;
239};
240
241struct pipe_query *si_create_batch_query(struct pipe_context *ctx,
242					 unsigned num_queries,
243					 unsigned *query_types);
244
245int si_get_perfcounter_info(struct si_screen *,
246			    unsigned index,
247			    struct pipe_driver_query_info *info);
248int si_get_perfcounter_group_info(struct si_screen *,
249				  unsigned index,
250				  struct pipe_driver_query_group_info *info);
251
252struct si_qbo_state {
253	void *saved_compute;
254	struct pipe_constant_buffer saved_const0;
255	struct pipe_shader_buffer saved_ssbo[3];
256	unsigned saved_ssbo_writable_mask;
257};
258
259#endif /* SI_QUERY_H */
260