• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright 2015 Advanced Micro Devices, Inc.
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 FROM,
20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21  * SOFTWARE.
22  */
23 
24 #ifndef R600_QUERY_H
25 #define R600_QUERY_H
26 
27 #include "util/u_threaded_context.h"
28 
29 struct pipe_context;
30 struct pipe_query;
31 struct pipe_resource;
32 
33 struct r600_common_context;
34 struct si_screen;
35 struct r600_query;
36 struct r600_query_hw;
37 struct r600_resource;
38 
39 enum {
40 	R600_QUERY_DRAW_CALLS = PIPE_QUERY_DRIVER_SPECIFIC,
41 	R600_QUERY_DECOMPRESS_CALLS,
42 	R600_QUERY_MRT_DRAW_CALLS,
43 	R600_QUERY_PRIM_RESTART_CALLS,
44 	R600_QUERY_SPILL_DRAW_CALLS,
45 	R600_QUERY_COMPUTE_CALLS,
46 	R600_QUERY_SPILL_COMPUTE_CALLS,
47 	R600_QUERY_DMA_CALLS,
48 	R600_QUERY_CP_DMA_CALLS,
49 	R600_QUERY_NUM_VS_FLUSHES,
50 	R600_QUERY_NUM_PS_FLUSHES,
51 	R600_QUERY_NUM_CS_FLUSHES,
52 	R600_QUERY_NUM_CB_CACHE_FLUSHES,
53 	R600_QUERY_NUM_DB_CACHE_FLUSHES,
54 	R600_QUERY_NUM_L2_INVALIDATES,
55 	R600_QUERY_NUM_L2_WRITEBACKS,
56 	R600_QUERY_NUM_RESIDENT_HANDLES,
57 	R600_QUERY_TC_OFFLOADED_SLOTS,
58 	R600_QUERY_TC_DIRECT_SLOTS,
59 	R600_QUERY_TC_NUM_SYNCS,
60 	R600_QUERY_CS_THREAD_BUSY,
61 	R600_QUERY_GALLIUM_THREAD_BUSY,
62 	R600_QUERY_REQUESTED_VRAM,
63 	R600_QUERY_REQUESTED_GTT,
64 	R600_QUERY_MAPPED_VRAM,
65 	R600_QUERY_MAPPED_GTT,
66 	R600_QUERY_BUFFER_WAIT_TIME,
67 	R600_QUERY_NUM_MAPPED_BUFFERS,
68 	R600_QUERY_NUM_GFX_IBS,
69 	R600_QUERY_NUM_SDMA_IBS,
70 	R600_QUERY_GFX_BO_LIST_SIZE,
71 	R600_QUERY_GFX_IB_SIZE,
72 	R600_QUERY_NUM_BYTES_MOVED,
73 	R600_QUERY_NUM_EVICTIONS,
74 	R600_QUERY_NUM_VRAM_CPU_PAGE_FAULTS,
75 	R600_QUERY_VRAM_USAGE,
76 	R600_QUERY_VRAM_VIS_USAGE,
77 	R600_QUERY_GTT_USAGE,
78 	R600_QUERY_GPU_TEMPERATURE,
79 	R600_QUERY_CURRENT_GPU_SCLK,
80 	R600_QUERY_CURRENT_GPU_MCLK,
81 	R600_QUERY_GPU_LOAD,
82 	R600_QUERY_GPU_SHADERS_BUSY,
83 	R600_QUERY_GPU_TA_BUSY,
84 	R600_QUERY_GPU_GDS_BUSY,
85 	R600_QUERY_GPU_VGT_BUSY,
86 	R600_QUERY_GPU_IA_BUSY,
87 	R600_QUERY_GPU_SX_BUSY,
88 	R600_QUERY_GPU_WD_BUSY,
89 	R600_QUERY_GPU_BCI_BUSY,
90 	R600_QUERY_GPU_SC_BUSY,
91 	R600_QUERY_GPU_PA_BUSY,
92 	R600_QUERY_GPU_DB_BUSY,
93 	R600_QUERY_GPU_CP_BUSY,
94 	R600_QUERY_GPU_CB_BUSY,
95 	R600_QUERY_GPU_SDMA_BUSY,
96 	R600_QUERY_GPU_PFP_BUSY,
97 	R600_QUERY_GPU_MEQ_BUSY,
98 	R600_QUERY_GPU_ME_BUSY,
99 	R600_QUERY_GPU_SURF_SYNC_BUSY,
100 	R600_QUERY_GPU_CP_DMA_BUSY,
101 	R600_QUERY_GPU_SCRATCH_RAM_BUSY,
102 	R600_QUERY_NUM_COMPILATIONS,
103 	R600_QUERY_NUM_SHADERS_CREATED,
104 	R600_QUERY_BACK_BUFFER_PS_DRAW_RATIO,
105 	R600_QUERY_NUM_SHADER_CACHE_HITS,
106 	R600_QUERY_GPIN_ASIC_ID,
107 	R600_QUERY_GPIN_NUM_SIMD,
108 	R600_QUERY_GPIN_NUM_RB,
109 	R600_QUERY_GPIN_NUM_SPI,
110 	R600_QUERY_GPIN_NUM_SE,
111 
112 	R600_QUERY_FIRST_PERFCOUNTER = PIPE_QUERY_DRIVER_SPECIFIC + 100,
113 };
114 
115 enum {
116 	R600_QUERY_GROUP_GPIN = 0,
117 	R600_NUM_SW_QUERY_GROUPS
118 };
119 
120 struct r600_query_ops {
121 	void (*destroy)(struct si_screen *, struct r600_query *);
122 	bool (*begin)(struct r600_common_context *, struct r600_query *);
123 	bool (*end)(struct r600_common_context *, struct r600_query *);
124 	bool (*get_result)(struct r600_common_context *,
125 			   struct r600_query *, bool wait,
126 			   union pipe_query_result *result);
127 	void (*get_result_resource)(struct r600_common_context *,
128 				    struct r600_query *, bool wait,
129 				    enum pipe_query_value_type result_type,
130 				    int index,
131 				    struct pipe_resource *resource,
132 				    unsigned offset);
133 };
134 
135 struct r600_query {
136 	struct threaded_query b;
137 	struct r600_query_ops *ops;
138 
139 	/* The type of query */
140 	unsigned type;
141 };
142 
143 enum {
144 	R600_QUERY_HW_FLAG_NO_START = (1 << 0),
145 	/* gap */
146 	/* whether begin_query doesn't clear the result */
147 	R600_QUERY_HW_FLAG_BEGIN_RESUMES = (1 << 2),
148 };
149 
150 struct r600_query_hw_ops {
151 	bool (*prepare_buffer)(struct si_screen *,
152 			       struct r600_query_hw *,
153 			       struct r600_resource *);
154 	void (*emit_start)(struct r600_common_context *,
155 			   struct r600_query_hw *,
156 			   struct r600_resource *buffer, uint64_t va);
157 	void (*emit_stop)(struct r600_common_context *,
158 			  struct r600_query_hw *,
159 			  struct r600_resource *buffer, uint64_t va);
160 	void (*clear_result)(struct r600_query_hw *, union pipe_query_result *);
161 	void (*add_result)(struct si_screen *screen,
162 			   struct r600_query_hw *, void *buffer,
163 			   union pipe_query_result *result);
164 };
165 
166 struct r600_query_buffer {
167 	/* The buffer where query results are stored. */
168 	struct r600_resource		*buf;
169 	/* Offset of the next free result after current query data */
170 	unsigned			results_end;
171 	/* If a query buffer is full, a new buffer is created and the old one
172 	 * is put in here. When we calculate the result, we sum up the samples
173 	 * from all buffers. */
174 	struct r600_query_buffer	*previous;
175 };
176 
177 struct r600_query_hw {
178 	struct r600_query b;
179 	struct r600_query_hw_ops *ops;
180 	unsigned flags;
181 
182 	/* The query buffer and how many results are in it. */
183 	struct r600_query_buffer buffer;
184 	/* Size of the result in memory for both begin_query and end_query,
185 	 * this can be one or two numbers, or it could even be a size of a structure. */
186 	unsigned result_size;
187 	/* The number of dwords for begin_query or end_query. */
188 	unsigned num_cs_dw_begin;
189 	unsigned num_cs_dw_end;
190 	/* Linked list of queries */
191 	struct list_head list;
192 	/* For transform feedback: which stream the query is for */
193 	unsigned stream;
194 
195 	/* Workaround via compute shader */
196 	struct r600_resource *workaround_buf;
197 	unsigned workaround_offset;
198 };
199 
200 bool si_query_hw_init(struct si_screen *sscreen,
201 		      struct r600_query_hw *query);
202 void si_query_hw_destroy(struct si_screen *sscreen,
203 			 struct r600_query *rquery);
204 bool si_query_hw_begin(struct r600_common_context *rctx,
205 		       struct r600_query *rquery);
206 bool si_query_hw_end(struct r600_common_context *rctx,
207 		     struct r600_query *rquery);
208 bool si_query_hw_get_result(struct r600_common_context *rctx,
209 			    struct r600_query *rquery,
210 			    bool wait,
211 			    union pipe_query_result *result);
212 
213 /* Performance counters */
214 enum {
215 	/* This block is part of the shader engine */
216 	R600_PC_BLOCK_SE = (1 << 0),
217 
218 	/* Expose per-instance groups instead of summing all instances (within
219 	 * an SE). */
220 	R600_PC_BLOCK_INSTANCE_GROUPS = (1 << 1),
221 
222 	/* Expose per-SE groups instead of summing instances across SEs. */
223 	R600_PC_BLOCK_SE_GROUPS = (1 << 2),
224 
225 	/* Shader block */
226 	R600_PC_BLOCK_SHADER = (1 << 3),
227 
228 	/* Non-shader block with perfcounters windowed by shaders. */
229 	R600_PC_BLOCK_SHADER_WINDOWED = (1 << 4),
230 };
231 
232 /* Describes a hardware block with performance counters. Multiple instances of
233  * each block, possibly per-SE, may exist on the chip. Depending on the block
234  * and on the user's configuration, we either
235  *  (a) expose every instance as a performance counter group,
236  *  (b) expose a single performance counter group that reports the sum over all
237  *      instances, or
238  *  (c) expose one performance counter group per instance, but summed over all
239  *      shader engines.
240  */
241 struct r600_perfcounter_block {
242 	const char *basename;
243 	unsigned flags;
244 	unsigned num_counters;
245 	unsigned num_selectors;
246 	unsigned num_instances;
247 
248 	unsigned num_groups;
249 	char *group_names;
250 	unsigned group_name_stride;
251 
252 	char *selector_names;
253 	unsigned selector_name_stride;
254 
255 	void *data;
256 };
257 
258 struct r600_perfcounters {
259 	unsigned num_groups;
260 	unsigned num_blocks;
261 	struct r600_perfcounter_block *blocks;
262 
263 	unsigned num_start_cs_dwords;
264 	unsigned num_stop_cs_dwords;
265 	unsigned num_instance_cs_dwords;
266 	unsigned num_shaders_cs_dwords;
267 
268 	unsigned num_shader_types;
269 	const char * const *shader_type_suffixes;
270 	const unsigned *shader_type_bits;
271 
272 	void (*get_size)(struct r600_perfcounter_block *,
273 			 unsigned count, unsigned *selectors,
274 			 unsigned *num_select_dw, unsigned *num_read_dw);
275 
276 	void (*emit_instance)(struct r600_common_context *,
277 			      int se, int instance);
278 	void (*emit_shaders)(struct r600_common_context *, unsigned shaders);
279 	void (*emit_select)(struct r600_common_context *,
280 			    struct r600_perfcounter_block *,
281 			    unsigned count, unsigned *selectors);
282 	void (*emit_start)(struct r600_common_context *,
283 			  struct r600_resource *buffer, uint64_t va);
284 	void (*emit_stop)(struct r600_common_context *,
285 			  struct r600_resource *buffer, uint64_t va);
286 	void (*emit_read)(struct r600_common_context *,
287 			  struct r600_perfcounter_block *,
288 			  unsigned count, unsigned *selectors,
289 			  struct r600_resource *buffer, uint64_t va);
290 
291 	void (*cleanup)(struct si_screen *);
292 
293 	bool separate_se;
294 	bool separate_instance;
295 };
296 
297 struct pipe_query *si_create_batch_query(struct pipe_context *ctx,
298 					 unsigned num_queries,
299 					 unsigned *query_types);
300 
301 int si_get_perfcounter_info(struct si_screen *,
302 			    unsigned index,
303 			    struct pipe_driver_query_info *info);
304 int si_get_perfcounter_group_info(struct si_screen *,
305 				  unsigned index,
306 				  struct pipe_driver_query_group_info *info);
307 
308 bool si_perfcounters_init(struct r600_perfcounters *, unsigned num_blocks);
309 void si_perfcounters_add_block(struct si_screen *,
310 			       struct r600_perfcounters *,
311 			       const char *name, unsigned flags,
312 			       unsigned counters, unsigned selectors,
313 			       unsigned instances, void *data);
314 void si_perfcounters_do_destroy(struct r600_perfcounters *);
315 void si_query_hw_reset_buffers(struct r600_common_context *rctx,
316 			       struct r600_query_hw *query);
317 
318 struct r600_qbo_state {
319 	void *saved_compute;
320 	struct pipe_constant_buffer saved_const0;
321 	struct pipe_shader_buffer saved_ssbo[3];
322 };
323 
324 #endif /* R600_QUERY_H */
325