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