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  * Authors:
24  *  Nicolai Hähnle <nicolai.haehnle@amd.com>
25  *
26  */
27 
28 #ifndef R600_QUERY_H
29 #define R600_QUERY_H
30 
31 #include "util/u_threaded_context.h"
32 
33 struct pipe_context;
34 struct pipe_query;
35 struct pipe_resource;
36 
37 struct r600_common_context;
38 struct r600_common_screen;
39 struct r600_query;
40 struct r600_query_hw;
41 struct r600_resource;
42 
43 enum {
44 	R600_QUERY_DRAW_CALLS = PIPE_QUERY_DRIVER_SPECIFIC,
45 	R600_QUERY_DECOMPRESS_CALLS,
46 	R600_QUERY_MRT_DRAW_CALLS,
47 	R600_QUERY_PRIM_RESTART_CALLS,
48 	R600_QUERY_SPILL_DRAW_CALLS,
49 	R600_QUERY_COMPUTE_CALLS,
50 	R600_QUERY_SPILL_COMPUTE_CALLS,
51 	R600_QUERY_DMA_CALLS,
52 	R600_QUERY_CP_DMA_CALLS,
53 	R600_QUERY_NUM_VS_FLUSHES,
54 	R600_QUERY_NUM_PS_FLUSHES,
55 	R600_QUERY_NUM_CS_FLUSHES,
56 	R600_QUERY_NUM_CB_CACHE_FLUSHES,
57 	R600_QUERY_NUM_DB_CACHE_FLUSHES,
58 	R600_QUERY_NUM_RESIDENT_HANDLES,
59 	R600_QUERY_TC_OFFLOADED_SLOTS,
60 	R600_QUERY_TC_DIRECT_SLOTS,
61 	R600_QUERY_TC_NUM_SYNCS,
62 	R600_QUERY_CS_THREAD_BUSY,
63 	R600_QUERY_GALLIUM_THREAD_BUSY,
64 	R600_QUERY_REQUESTED_VRAM,
65 	R600_QUERY_REQUESTED_GTT,
66 	R600_QUERY_MAPPED_VRAM,
67 	R600_QUERY_MAPPED_GTT,
68 	R600_QUERY_BUFFER_WAIT_TIME,
69 	R600_QUERY_NUM_MAPPED_BUFFERS,
70 	R600_QUERY_NUM_GFX_IBS,
71 	R600_QUERY_NUM_SDMA_IBS,
72 	R600_QUERY_GFX_BO_LIST_SIZE,
73 	R600_QUERY_NUM_BYTES_MOVED,
74 	R600_QUERY_NUM_EVICTIONS,
75 	R600_QUERY_NUM_VRAM_CPU_PAGE_FAULTS,
76 	R600_QUERY_VRAM_USAGE,
77 	R600_QUERY_VRAM_VIS_USAGE,
78 	R600_QUERY_GTT_USAGE,
79 	R600_QUERY_GPU_TEMPERATURE,
80 	R600_QUERY_CURRENT_GPU_SCLK,
81 	R600_QUERY_CURRENT_GPU_MCLK,
82 	R600_QUERY_GPU_LOAD,
83 	R600_QUERY_GPU_SHADERS_BUSY,
84 	R600_QUERY_GPU_TA_BUSY,
85 	R600_QUERY_GPU_GDS_BUSY,
86 	R600_QUERY_GPU_VGT_BUSY,
87 	R600_QUERY_GPU_IA_BUSY,
88 	R600_QUERY_GPU_SX_BUSY,
89 	R600_QUERY_GPU_WD_BUSY,
90 	R600_QUERY_GPU_BCI_BUSY,
91 	R600_QUERY_GPU_SC_BUSY,
92 	R600_QUERY_GPU_PA_BUSY,
93 	R600_QUERY_GPU_DB_BUSY,
94 	R600_QUERY_GPU_CP_BUSY,
95 	R600_QUERY_GPU_CB_BUSY,
96 	R600_QUERY_GPU_SDMA_BUSY,
97 	R600_QUERY_GPU_PFP_BUSY,
98 	R600_QUERY_GPU_MEQ_BUSY,
99 	R600_QUERY_GPU_ME_BUSY,
100 	R600_QUERY_GPU_SURF_SYNC_BUSY,
101 	R600_QUERY_GPU_CP_DMA_BUSY,
102 	R600_QUERY_GPU_SCRATCH_RAM_BUSY,
103 	R600_QUERY_NUM_COMPILATIONS,
104 	R600_QUERY_NUM_SHADERS_CREATED,
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 r600_common_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 *,
129 				    enum pipe_query_flags flags,
130 				    enum pipe_query_value_type result_type,
131 				    int index,
132 				    struct pipe_resource *resource,
133 				    unsigned offset);
134 };
135 
136 struct r600_query {
137 	struct threaded_query b;
138 	struct r600_query_ops *ops;
139 
140 	/* The type of query */
141 	unsigned type;
142 };
143 
144 enum {
145 	R600_QUERY_HW_FLAG_NO_START = (1 << 0),
146 	/* gap */
147 	/* whether begin_query doesn't clear the result */
148 	R600_QUERY_HW_FLAG_BEGIN_RESUMES = (1 << 2),
149 };
150 
151 struct r600_query_hw_ops {
152 	bool (*prepare_buffer)(struct r600_common_screen *,
153 			       struct r600_query_hw *,
154 			       struct r600_resource *);
155 	void (*emit_start)(struct r600_common_context *,
156 			   struct r600_query_hw *,
157 			   struct r600_resource *buffer, uint64_t va);
158 	void (*emit_stop)(struct r600_common_context *,
159 			  struct r600_query_hw *,
160 			  struct r600_resource *buffer, uint64_t va);
161 	void (*clear_result)(struct r600_query_hw *, union pipe_query_result *);
162 	void (*add_result)(struct r600_common_screen *screen,
163 			   struct r600_query_hw *, void *buffer,
164 			   union pipe_query_result *result);
165 };
166 
167 struct r600_query_buffer {
168 	/* The buffer where query results are stored. */
169 	struct r600_resource		*buf;
170 	/* Offset of the next free result after current query data */
171 	unsigned			results_end;
172 	/* If a query buffer is full, a new buffer is created and the old one
173 	 * is put in here. When we calculate the result, we sum up the samples
174 	 * from all buffers. */
175 	struct r600_query_buffer	*previous;
176 };
177 
178 struct r600_query_hw {
179 	struct r600_query b;
180 	struct r600_query_hw_ops *ops;
181 	unsigned flags;
182 
183 	/* The query buffer and how many results are in it. */
184 	struct r600_query_buffer buffer;
185 	/* Size of the result in memory for both begin_query and end_query,
186 	 * this can be one or two numbers, or it could even be a size of a structure. */
187 	unsigned result_size;
188 	/* The number of dwords for begin_query or end_query. */
189 	unsigned num_cs_dw_begin;
190 	unsigned num_cs_dw_end;
191 	/* Linked list of queries */
192 	struct list_head list;
193 	/* For transform feedback: which stream the query is for */
194 	unsigned stream;
195 };
196 
197 bool r600_query_hw_init(struct r600_common_screen *rscreen,
198 			struct r600_query_hw *query);
199 void r600_query_hw_destroy(struct r600_common_screen *rscreen,
200 			   struct r600_query *rquery);
201 bool r600_query_hw_begin(struct r600_common_context *rctx,
202 			 struct r600_query *rquery);
203 bool r600_query_hw_end(struct r600_common_context *rctx,
204 		       struct r600_query *rquery);
205 bool r600_query_hw_get_result(struct r600_common_context *rctx,
206 			      struct r600_query *rquery,
207 			      bool wait,
208 			      union pipe_query_result *result);
209 
210 /* Performance counters */
211 enum {
212 	/* This block is part of the shader engine */
213 	R600_PC_BLOCK_SE = (1 << 0),
214 
215 	/* Expose per-instance groups instead of summing all instances (within
216 	 * an SE). */
217 	R600_PC_BLOCK_INSTANCE_GROUPS = (1 << 1),
218 
219 	/* Expose per-SE groups instead of summing instances across SEs. */
220 	R600_PC_BLOCK_SE_GROUPS = (1 << 2),
221 
222 	/* Shader block */
223 	R600_PC_BLOCK_SHADER = (1 << 3),
224 
225 	/* Non-shader block with perfcounters windowed by shaders. */
226 	R600_PC_BLOCK_SHADER_WINDOWED = (1 << 4),
227 };
228 
229 /* Describes a hardware block with performance counters. Multiple instances of
230  * each block, possibly per-SE, may exist on the chip. Depending on the block
231  * and on the user's configuration, we either
232  *  (a) expose every instance as a performance counter group,
233  *  (b) expose a single performance counter group that reports the sum over all
234  *      instances, or
235  *  (c) expose one performance counter group per instance, but summed over all
236  *      shader engines.
237  */
238 struct r600_perfcounter_block {
239 	const char *basename;
240 	unsigned flags;
241 	unsigned num_counters;
242 	unsigned num_selectors;
243 	unsigned num_instances;
244 
245 	unsigned num_groups;
246 	char *group_names;
247 	unsigned group_name_stride;
248 
249 	char *selector_names;
250 	unsigned selector_name_stride;
251 
252 	void *data;
253 };
254 
255 struct r600_perfcounters {
256 	unsigned num_groups;
257 	unsigned num_blocks;
258 	struct r600_perfcounter_block *blocks;
259 
260 	unsigned num_start_cs_dwords;
261 	unsigned num_stop_cs_dwords;
262 	unsigned num_instance_cs_dwords;
263 	unsigned num_shaders_cs_dwords;
264 
265 	unsigned num_shader_types;
266 	const char * const *shader_type_suffixes;
267 	const unsigned *shader_type_bits;
268 
269 	void (*get_size)(struct r600_perfcounter_block *,
270 			 unsigned count, unsigned *selectors,
271 			 unsigned *num_select_dw, unsigned *num_read_dw);
272 
273 	void (*emit_instance)(struct r600_common_context *,
274 			      int se, int instance);
275 	void (*emit_shaders)(struct r600_common_context *, unsigned shaders);
276 	void (*emit_select)(struct r600_common_context *,
277 			    struct r600_perfcounter_block *,
278 			    unsigned count, unsigned *selectors);
279 	void (*emit_start)(struct r600_common_context *,
280 			  struct r600_resource *buffer, uint64_t va);
281 	void (*emit_stop)(struct r600_common_context *,
282 			  struct r600_resource *buffer, uint64_t va);
283 	void (*emit_read)(struct r600_common_context *,
284 			  struct r600_perfcounter_block *,
285 			  unsigned count, unsigned *selectors,
286 			  struct r600_resource *buffer, uint64_t va);
287 
288 	void (*cleanup)(struct r600_common_screen *);
289 
290 	bool separate_se;
291 	bool separate_instance;
292 };
293 
294 struct pipe_query *r600_create_batch_query(struct pipe_context *ctx,
295 					   unsigned num_queries,
296 					   unsigned *query_types);
297 
298 int r600_get_perfcounter_info(struct r600_common_screen *,
299 			      unsigned index,
300 			      struct pipe_driver_query_info *info);
301 int r600_get_perfcounter_group_info(struct r600_common_screen *,
302 				    unsigned index,
303 				    struct pipe_driver_query_group_info *info);
304 
305 bool r600_perfcounters_init(struct r600_perfcounters *, unsigned num_blocks);
306 void r600_perfcounters_add_block(struct r600_common_screen *,
307 				 struct r600_perfcounters *,
308 				 const char *name, unsigned flags,
309 				 unsigned counters, unsigned selectors,
310 				 unsigned instances, void *data);
311 void r600_perfcounters_do_destroy(struct r600_perfcounters *);
312 void r600_query_hw_reset_buffers(struct r600_common_context *rctx,
313 				 struct r600_query_hw *query);
314 
315 struct r600_qbo_state {
316 	void *saved_compute;
317 	struct pipe_constant_buffer saved_const0;
318 	struct pipe_shader_buffer saved_ssbo[3];
319 };
320 
321 #endif /* R600_QUERY_H */
322