radeonsi: set better tessellation tunables on gfx9 and gfx10
[mesa.git] / src / gallium / drivers / radeonsi / si_query.h
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
30 struct pipe_context;
31 struct pipe_query;
32 struct pipe_resource;
33
34 struct si_screen;
35 struct si_context;
36 struct si_query;
37 struct si_query_buffer;
38 struct si_query_hw;
39 struct si_resource;
40
41 #define SI_MAX_STREAMS 4
42
43 enum {
44 SI_QUERY_DRAW_CALLS = PIPE_QUERY_DRIVER_SPECIFIC,
45 SI_QUERY_DECOMPRESS_CALLS,
46 SI_QUERY_MRT_DRAW_CALLS,
47 SI_QUERY_PRIM_RESTART_CALLS,
48 SI_QUERY_SPILL_DRAW_CALLS,
49 SI_QUERY_COMPUTE_CALLS,
50 SI_QUERY_SPILL_COMPUTE_CALLS,
51 SI_QUERY_DMA_CALLS,
52 SI_QUERY_CP_DMA_CALLS,
53 SI_QUERY_NUM_VS_FLUSHES,
54 SI_QUERY_NUM_PS_FLUSHES,
55 SI_QUERY_NUM_CS_FLUSHES,
56 SI_QUERY_NUM_CB_CACHE_FLUSHES,
57 SI_QUERY_NUM_DB_CACHE_FLUSHES,
58 SI_QUERY_NUM_L2_INVALIDATES,
59 SI_QUERY_NUM_L2_WRITEBACKS,
60 SI_QUERY_NUM_RESIDENT_HANDLES,
61 SI_QUERY_TC_OFFLOADED_SLOTS,
62 SI_QUERY_TC_DIRECT_SLOTS,
63 SI_QUERY_TC_NUM_SYNCS,
64 SI_QUERY_CS_THREAD_BUSY,
65 SI_QUERY_GALLIUM_THREAD_BUSY,
66 SI_QUERY_REQUESTED_VRAM,
67 SI_QUERY_REQUESTED_GTT,
68 SI_QUERY_MAPPED_VRAM,
69 SI_QUERY_MAPPED_GTT,
70 SI_QUERY_BUFFER_WAIT_TIME,
71 SI_QUERY_NUM_MAPPED_BUFFERS,
72 SI_QUERY_NUM_GFX_IBS,
73 SI_QUERY_NUM_SDMA_IBS,
74 SI_QUERY_GFX_BO_LIST_SIZE,
75 SI_QUERY_GFX_IB_SIZE,
76 SI_QUERY_NUM_BYTES_MOVED,
77 SI_QUERY_NUM_EVICTIONS,
78 SI_QUERY_NUM_VRAM_CPU_PAGE_FAULTS,
79 SI_QUERY_VRAM_USAGE,
80 SI_QUERY_VRAM_VIS_USAGE,
81 SI_QUERY_GTT_USAGE,
82 SI_QUERY_GPU_TEMPERATURE,
83 SI_QUERY_CURRENT_GPU_SCLK,
84 SI_QUERY_CURRENT_GPU_MCLK,
85 SI_QUERY_GPU_LOAD,
86 SI_QUERY_GPU_SHADERS_BUSY,
87 SI_QUERY_GPU_TA_BUSY,
88 SI_QUERY_GPU_GDS_BUSY,
89 SI_QUERY_GPU_VGT_BUSY,
90 SI_QUERY_GPU_IA_BUSY,
91 SI_QUERY_GPU_SX_BUSY,
92 SI_QUERY_GPU_WD_BUSY,
93 SI_QUERY_GPU_BCI_BUSY,
94 SI_QUERY_GPU_SC_BUSY,
95 SI_QUERY_GPU_PA_BUSY,
96 SI_QUERY_GPU_DB_BUSY,
97 SI_QUERY_GPU_CP_BUSY,
98 SI_QUERY_GPU_CB_BUSY,
99 SI_QUERY_GPU_SDMA_BUSY,
100 SI_QUERY_GPU_PFP_BUSY,
101 SI_QUERY_GPU_MEQ_BUSY,
102 SI_QUERY_GPU_ME_BUSY,
103 SI_QUERY_GPU_SURF_SYNC_BUSY,
104 SI_QUERY_GPU_CP_DMA_BUSY,
105 SI_QUERY_GPU_SCRATCH_RAM_BUSY,
106 SI_QUERY_NUM_COMPILATIONS,
107 SI_QUERY_NUM_SHADERS_CREATED,
108 SI_QUERY_BACK_BUFFER_PS_DRAW_RATIO,
109 SI_QUERY_GPIN_ASIC_ID,
110 SI_QUERY_GPIN_NUM_SIMD,
111 SI_QUERY_GPIN_NUM_RB,
112 SI_QUERY_GPIN_NUM_SPI,
113 SI_QUERY_GPIN_NUM_SE,
114 SI_QUERY_TIME_ELAPSED_SDMA,
115 SI_QUERY_TIME_ELAPSED_SDMA_SI, /* emulated, measured on the CPU */
116 SI_QUERY_PD_NUM_PRIMS_ACCEPTED,
117 SI_QUERY_PD_NUM_PRIMS_REJECTED,
118 SI_QUERY_PD_NUM_PRIMS_INELIGIBLE,
119 SI_QUERY_LIVE_SHADER_CACHE_HITS,
120 SI_QUERY_LIVE_SHADER_CACHE_MISSES,
121 SI_QUERY_MEMORY_SHADER_CACHE_HITS,
122 SI_QUERY_MEMORY_SHADER_CACHE_MISSES,
123 SI_QUERY_DISK_SHADER_CACHE_HITS,
124 SI_QUERY_DISK_SHADER_CACHE_MISSES,
125
126 SI_QUERY_FIRST_PERFCOUNTER = PIPE_QUERY_DRIVER_SPECIFIC + 100,
127 };
128
129 enum {
130 SI_QUERY_GROUP_GPIN = 0,
131 SI_NUM_SW_QUERY_GROUPS
132 };
133
134 struct si_query_ops {
135 void (*destroy)(struct si_context *, struct si_query *);
136 bool (*begin)(struct si_context *, struct si_query *);
137 bool (*end)(struct si_context *, struct si_query *);
138 bool (*get_result)(struct si_context *,
139 struct si_query *, bool wait,
140 union pipe_query_result *result);
141 void (*get_result_resource)(struct si_context *,
142 struct si_query *, bool wait,
143 enum pipe_query_value_type result_type,
144 int index,
145 struct pipe_resource *resource,
146 unsigned offset);
147
148 void (*suspend)(struct si_context *, struct si_query *);
149 void (*resume)(struct si_context *, struct si_query *);
150 };
151
152 struct si_query {
153 struct threaded_query b;
154 const struct si_query_ops *ops;
155
156 /* The PIPE_QUERY_xxx type of query */
157 unsigned type;
158
159 /* The number of dwords for suspend. */
160 unsigned num_cs_dw_suspend;
161
162 /* Linked list of queries that must be suspended at end of CS. */
163 struct list_head active_list;
164 };
165
166 enum {
167 SI_QUERY_HW_FLAG_NO_START = (1 << 0),
168 /* gap */
169 /* whether begin_query doesn't clear the result */
170 SI_QUERY_HW_FLAG_BEGIN_RESUMES = (1 << 2),
171 };
172
173 struct si_query_hw_ops {
174 bool (*prepare_buffer)(struct si_context *, struct si_query_buffer *);
175 void (*emit_start)(struct si_context *,
176 struct si_query_hw *,
177 struct si_resource *buffer, uint64_t va);
178 void (*emit_stop)(struct si_context *,
179 struct si_query_hw *,
180 struct si_resource *buffer, uint64_t va);
181 void (*clear_result)(struct si_query_hw *, union pipe_query_result *);
182 void (*add_result)(struct si_screen *screen,
183 struct si_query_hw *, void *buffer,
184 union pipe_query_result *result);
185 };
186
187 struct si_query_buffer {
188 /* The buffer where query results are stored. */
189 struct si_resource *buf;
190 /* If a query buffer is full, a new buffer is created and the old one
191 * is put in here. When we calculate the result, we sum up the samples
192 * from all buffers. */
193 struct si_query_buffer *previous;
194 /* Offset of the next free result after current query data */
195 unsigned results_end;
196 bool unprepared;
197 };
198
199 void si_query_buffer_destroy(struct si_screen *sctx, struct si_query_buffer *buffer);
200 void si_query_buffer_reset(struct si_context *sctx, struct si_query_buffer *buffer);
201 bool si_query_buffer_alloc(struct si_context *sctx, struct si_query_buffer *buffer,
202 bool (*prepare_buffer)(struct si_context *, struct si_query_buffer*),
203 unsigned size);
204
205
206 struct si_query_hw {
207 struct si_query b;
208 struct si_query_hw_ops *ops;
209 unsigned flags;
210
211 /* The query buffer and how many results are in it. */
212 struct si_query_buffer buffer;
213 /* Size of the result in memory for both begin_query and end_query,
214 * this can be one or two numbers, or it could even be a size of a structure. */
215 unsigned result_size;
216 /* For transform feedback: which stream the query is for */
217 unsigned stream;
218
219 /* Workaround via compute shader */
220 struct si_resource *workaround_buf;
221 unsigned workaround_offset;
222 };
223
224 void si_query_hw_destroy(struct si_context *sctx,
225 struct si_query *squery);
226 bool si_query_hw_begin(struct si_context *sctx,
227 struct si_query *squery);
228 bool si_query_hw_end(struct si_context *sctx,
229 struct si_query *squery);
230 bool si_query_hw_get_result(struct si_context *sctx,
231 struct si_query *squery,
232 bool wait,
233 union pipe_query_result *result);
234 void si_query_hw_suspend(struct si_context *sctx, struct si_query *query);
235 void si_query_hw_resume(struct si_context *sctx, struct si_query *query);
236
237
238 /* Shader-based queries */
239 struct pipe_query *gfx10_sh_query_create(struct si_screen *screen,
240 enum pipe_query_type query_type,
241 unsigned index);
242
243
244 /* Performance counters */
245 struct si_perfcounters {
246 unsigned num_groups;
247 unsigned num_blocks;
248 struct si_pc_block *blocks;
249
250 unsigned num_stop_cs_dwords;
251 unsigned num_instance_cs_dwords;
252
253 bool separate_se;
254 bool separate_instance;
255 };
256
257 struct pipe_query *si_create_batch_query(struct pipe_context *ctx,
258 unsigned num_queries,
259 unsigned *query_types);
260
261 int si_get_perfcounter_info(struct si_screen *,
262 unsigned index,
263 struct pipe_driver_query_info *info);
264 int si_get_perfcounter_group_info(struct si_screen *,
265 unsigned index,
266 struct pipe_driver_query_group_info *info);
267
268 struct si_qbo_state {
269 void *saved_compute;
270 struct pipe_constant_buffer saved_const0;
271 struct pipe_shader_buffer saved_ssbo[3];
272 unsigned saved_ssbo_writable_mask;
273 };
274
275 #endif /* SI_QUERY_H */