src/gallium/drivers/radeon/r600_query.h

   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_L2_INVALIDATES,
  59         R600_QUERY_NUM_L2_WRITEBACKS,
  60         R600_QUERY_NUM_RESIDENT_HANDLES,
  61         R600_QUERY_TC_OFFLOADED_SLOTS,
  62         R600_QUERY_TC_DIRECT_SLOTS,
  63         R600_QUERY_TC_NUM_SYNCS,
  64         R600_QUERY_CS_THREAD_BUSY,
  65         R600_QUERY_GALLIUM_THREAD_BUSY,
  66         R600_QUERY_REQUESTED_VRAM,
  67         R600_QUERY_REQUESTED_GTT,
  68         R600_QUERY_MAPPED_VRAM,
  69         R600_QUERY_MAPPED_GTT,
  70         R600_QUERY_BUFFER_WAIT_TIME,
  71         R600_QUERY_NUM_MAPPED_BUFFERS,
  72         R600_QUERY_NUM_GFX_IBS,
  73         R600_QUERY_NUM_SDMA_IBS,
  74         R600_QUERY_GFX_BO_LIST_SIZE,
  75         R600_QUERY_NUM_BYTES_MOVED,
  76         R600_QUERY_NUM_EVICTIONS,
  77         R600_QUERY_NUM_VRAM_CPU_PAGE_FAULTS,
  78         R600_QUERY_VRAM_USAGE,
  79         R600_QUERY_VRAM_VIS_USAGE,
  80         R600_QUERY_GTT_USAGE,
  81         R600_QUERY_GPU_TEMPERATURE,
  82         R600_QUERY_CURRENT_GPU_SCLK,
  83         R600_QUERY_CURRENT_GPU_MCLK,
  84         R600_QUERY_GPU_LOAD,
  85         R600_QUERY_GPU_SHADERS_BUSY,
  86         R600_QUERY_GPU_TA_BUSY,
  87         R600_QUERY_GPU_GDS_BUSY,
  88         R600_QUERY_GPU_VGT_BUSY,
  89         R600_QUERY_GPU_IA_BUSY,
  90         R600_QUERY_GPU_SX_BUSY,
  91         R600_QUERY_GPU_WD_BUSY,
  92         R600_QUERY_GPU_BCI_BUSY,
  93         R600_QUERY_GPU_SC_BUSY,
  94         R600_QUERY_GPU_PA_BUSY,
  95         R600_QUERY_GPU_DB_BUSY,
  96         R600_QUERY_GPU_CP_BUSY,
  97         R600_QUERY_GPU_CB_BUSY,
  98         R600_QUERY_GPU_SDMA_BUSY,
  99         R600_QUERY_GPU_PFP_BUSY,
 100         R600_QUERY_GPU_MEQ_BUSY,
 101         R600_QUERY_GPU_ME_BUSY,
 102         R600_QUERY_GPU_SURF_SYNC_BUSY,
 103         R600_QUERY_GPU_CP_DMA_BUSY,
 104         R600_QUERY_GPU_SCRATCH_RAM_BUSY,
 105         R600_QUERY_NUM_COMPILATIONS,
 106         R600_QUERY_NUM_SHADERS_CREATED,
 107         R600_QUERY_BACK_BUFFER_PS_DRAW_RATIO,
 108         R600_QUERY_NUM_SHADER_CACHE_HITS,
 109         R600_QUERY_GPIN_ASIC_ID,
 110         R600_QUERY_GPIN_NUM_SIMD,
 111         R600_QUERY_GPIN_NUM_RB,
 112         R600_QUERY_GPIN_NUM_SPI,
 113         R600_QUERY_GPIN_NUM_SE,
 114
 115         R600_QUERY_FIRST_PERFCOUNTER = PIPE_QUERY_DRIVER_SPECIFIC + 100,
 116 };
 117
 118 enum {
 119         R600_QUERY_GROUP_GPIN = 0,
 120         R600_NUM_SW_QUERY_GROUPS
 121 };
 122
 123 struct r600_query_ops {
 124         void (*destroy)(struct r600_common_screen *, struct r600_query *);
 125         bool (*begin)(struct r600_common_context *, struct r600_query *);
 126         bool (*end)(struct r600_common_context *, struct r600_query *);
 127         bool (*get_result)(struct r600_common_context *,
 128                            struct r600_query *, bool wait,
 129                            union pipe_query_result *result);
 130         void (*get_result_resource)(struct r600_common_context *,
 131                                     struct r600_query *, bool wait,
 132                                     enum pipe_query_value_type result_type,
 133                                     int index,
 134                                     struct pipe_resource *resource,
 135                                     unsigned offset);
 136 };
 137
 138 struct r600_query {
 139         struct threaded_query b;
 140         struct r600_query_ops *ops;
 141
 142         /* The type of query */
 143         unsigned type;
 144 };
 145
 146 enum {
 147         R600_QUERY_HW_FLAG_NO_START = (1 << 0),
 148         /* gap */
 149         /* whether begin_query doesn't clear the result */
 150         R600_QUERY_HW_FLAG_BEGIN_RESUMES = (1 << 2),
 151 };
 152
 153 struct r600_query_hw_ops {
 154         bool (*prepare_buffer)(struct r600_common_screen *,
 155                                struct r600_query_hw *,
 156                                struct r600_resource *);
 157         void (*emit_start)(struct r600_common_context *,
 158                            struct r600_query_hw *,
 159                            struct r600_resource *buffer, uint64_t va);
 160         void (*emit_stop)(struct r600_common_context *,
 161                           struct r600_query_hw *,
 162                           struct r600_resource *buffer, uint64_t va);
 163         void (*clear_result)(struct r600_query_hw *, union pipe_query_result *);
 164         void (*add_result)(struct r600_common_screen *screen,
 165                            struct r600_query_hw *, void *buffer,
 166                            union pipe_query_result *result);
 167 };
 168
 169 struct r600_query_buffer {
 170         /* The buffer where query results are stored. */
 171         struct r600_resource            *buf;
 172         /* Offset of the next free result after current query data */
 173         unsigned                        results_end;
 174         /* If a query buffer is full, a new buffer is created and the old one
 175          * is put in here. When we calculate the result, we sum up the samples
 176          * from all buffers. */
 177         struct r600_query_buffer        *previous;
 178 };
 179
 180 struct r600_query_hw {
 181         struct r600_query b;
 182         struct r600_query_hw_ops *ops;
 183         unsigned flags;
 184
 185         /* The query buffer and how many results are in it. */
 186         struct r600_query_buffer buffer;
 187         /* Size of the result in memory for both begin_query and end_query,
 188          * this can be one or two numbers, or it could even be a size of a structure. */
 189         unsigned result_size;
 190         /* The number of dwords for begin_query or end_query. */
 191         unsigned num_cs_dw_begin;
 192         unsigned num_cs_dw_end;
 193         /* Linked list of queries */
 194         struct list_head list;
 195         /* For transform feedback: which stream the query is for */
 196         unsigned stream;
 197
 198         /* Workaround via compute shader */
 199         struct r600_resource *workaround_buf;
 200         unsigned workaround_offset;
 201 };
 202
 203 bool si_query_hw_init(struct r600_common_screen *rscreen,
 204                       struct r600_query_hw *query);
 205 void si_query_hw_destroy(struct r600_common_screen *rscreen,
 206                          struct r600_query *rquery);
 207 bool si_query_hw_begin(struct r600_common_context *rctx,
 208                        struct r600_query *rquery);
 209 bool si_query_hw_end(struct r600_common_context *rctx,
 210                      struct r600_query *rquery);
 211 bool si_query_hw_get_result(struct r600_common_context *rctx,
 212                             struct r600_query *rquery,
 213                             bool wait,
 214                             union pipe_query_result *result);
 215
 216 /* Performance counters */
 217 enum {
 218         /* This block is part of the shader engine */
 219         R600_PC_BLOCK_SE = (1 << 0),
 220
 221         /* Expose per-instance groups instead of summing all instances (within
 222          * an SE). */
 223         R600_PC_BLOCK_INSTANCE_GROUPS = (1 << 1),
 224
 225         /* Expose per-SE groups instead of summing instances across SEs. */
 226         R600_PC_BLOCK_SE_GROUPS = (1 << 2),
 227
 228         /* Shader block */
 229         R600_PC_BLOCK_SHADER = (1 << 3),
 230
 231         /* Non-shader block with perfcounters windowed by shaders. */
 232         R600_PC_BLOCK_SHADER_WINDOWED = (1 << 4),
 233 };
 234
 235 /* Describes a hardware block with performance counters. Multiple instances of
 236  * each block, possibly per-SE, may exist on the chip. Depending on the block
 237  * and on the user's configuration, we either
 238  *  (a) expose every instance as a performance counter group,
 239  *  (b) expose a single performance counter group that reports the sum over all
 240  *      instances, or
 241  *  (c) expose one performance counter group per instance, but summed over all
 242  *      shader engines.
 243  */
 244 struct r600_perfcounter_block {
 245         const char *basename;
 246         unsigned flags;
 247         unsigned num_counters;
 248         unsigned num_selectors;
 249         unsigned num_instances;
 250
 251         unsigned num_groups;
 252         char *group_names;
 253         unsigned group_name_stride;
 254
 255         char *selector_names;
 256         unsigned selector_name_stride;
 257
 258         void *data;
 259 };
 260
 261 struct r600_perfcounters {
 262         unsigned num_groups;
 263         unsigned num_blocks;
 264         struct r600_perfcounter_block *blocks;
 265
 266         unsigned num_start_cs_dwords;
 267         unsigned num_stop_cs_dwords;
 268         unsigned num_instance_cs_dwords;
 269         unsigned num_shaders_cs_dwords;
 270
 271         unsigned num_shader_types;
 272         const char * const *shader_type_suffixes;
 273         const unsigned *shader_type_bits;
 274
 275         void (*get_size)(struct r600_perfcounter_block *,
 276                          unsigned count, unsigned *selectors,
 277                          unsigned *num_select_dw, unsigned *num_read_dw);
 278
 279         void (*emit_instance)(struct r600_common_context *,
 280                               int se, int instance);
 281         void (*emit_shaders)(struct r600_common_context *, unsigned shaders);
 282         void (*emit_select)(struct r600_common_context *,
 283                             struct r600_perfcounter_block *,
 284                             unsigned count, unsigned *selectors);
 285         void (*emit_start)(struct r600_common_context *,
 286                           struct r600_resource *buffer, uint64_t va);
 287         void (*emit_stop)(struct r600_common_context *,
 288                           struct r600_resource *buffer, uint64_t va);
 289         void (*emit_read)(struct r600_common_context *,
 290                           struct r600_perfcounter_block *,
 291                           unsigned count, unsigned *selectors,
 292                           struct r600_resource *buffer, uint64_t va);
 293
 294         void (*cleanup)(struct r600_common_screen *);
 295
 296         bool separate_se;
 297         bool separate_instance;
 298 };
 299
 300 struct pipe_query *si_create_batch_query(struct pipe_context *ctx,
 301                                          unsigned num_queries,
 302                                          unsigned *query_types);
 303
 304 int si_get_perfcounter_info(struct r600_common_screen *,
 305                             unsigned index,
 306                             struct pipe_driver_query_info *info);
 307 int si_get_perfcounter_group_info(struct r600_common_screen *,
 308                                   unsigned index,
 309                                   struct pipe_driver_query_group_info *info);
 310
 311 bool si_perfcounters_init(struct r600_perfcounters *, unsigned num_blocks);
 312 void si_perfcounters_add_block(struct r600_common_screen *,
 313                                struct r600_perfcounters *,
 314                                const char *name, unsigned flags,
 315                                unsigned counters, unsigned selectors,
 316                                unsigned instances, void *data);
 317 void si_perfcounters_do_destroy(struct r600_perfcounters *);
 318 void si_query_hw_reset_buffers(struct r600_common_context *rctx,
 319                                struct r600_query_hw *query);
 320
 321 struct r600_qbo_state {
 322         void *saved_compute;
 323         struct pipe_constant_buffer saved_const0;
 324         struct pipe_shader_buffer saved_ssbo[3];
 325 };
 326
 327 #endif /* R600_QUERY_H */