1# 2# Copyright © 2021 Intel Corporation 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 24import argparse 25import sys 26 27# List of the default tracepoints enabled. By default most tracepoints are 28# enabled, set tp_default=False to disable them by default. 29# 30# Currently only stall is disabled by default 31intel_default_tps = [] 32 33# 34# Tracepoint definitions: 35# 36def define_tracepoints(args): 37 from u_trace import Header, HeaderScope 38 from u_trace import ForwardDecl 39 from u_trace import Tracepoint 40 from u_trace import TracepointArg as Arg 41 from u_trace import TracepointArgStruct as ArgStruct 42 43 Header('intel_driver_ds.h', scope=HeaderScope.SOURCE) 44 Header('vulkan/vulkan_core.h', scope=HeaderScope.SOURCE|HeaderScope.PERFETTO) 45 Header('blorp/blorp_priv.h', scope=HeaderScope.HEADER) 46 Header('ds/intel_driver_ds.h', scope=HeaderScope.HEADER) 47 48 def begin_end_tp(name, tp_args=[], tp_struct=None, tp_print=None, 49 tp_default_enabled=True, end_pipelined=True, 50 compute=False, maybe_compute=False, 51 need_cs_param=False): 52 global intel_default_tps 53 if tp_default_enabled: 54 intel_default_tps.append(name) 55 56 # Preprocess arguments to handle display_as_hex 57 processed_args = [] 58 for arg in tp_args: 59 # Manually handle display_as_hex by modifying the format 60 if "hash" in arg.var: 61 arg = Arg(type=arg.type, var=arg.var, c_format='%#x') # Convert to hex format 62 processed_args.append(arg) 63 64 Tracepoint('intel_begin_{0}'.format(name), 65 toggle_name=name, 66 tp_perfetto='intel_ds_begin_{0}'.format(name), 67 need_cs_param=need_cs_param) 68 tp_flags = [] 69 if end_pipelined: 70 if compute: 71 tp_flags.append('INTEL_DS_TRACEPOINT_FLAG_END_CS') 72 elif maybe_compute: 73 tp_flags.append('INTEL_DS_TRACEPOINT_FLAG_END_CS_OR_NOOP') 74 else: 75 tp_flags.append('INTEL_DS_TRACEPOINT_FLAG_END_OF_PIPE') 76 Tracepoint('intel_end_{0}'.format(name), 77 toggle_name=name, 78 args=processed_args, 79 tp_struct=tp_struct, 80 tp_perfetto='intel_ds_end_{0}'.format(name), 81 tp_print=tp_print, 82 tp_flags=tp_flags, 83 need_cs_param=need_cs_param) 84 85 # Frame tracepoints 86 begin_end_tp('frame', 87 tp_args=[Arg(type='uint32_t', var='frame', c_format='%u'),], 88 end_pipelined=False, 89 need_cs_param=True) 90 91 # Annotations for Queue(Begin|End)DebugUtilsLabelEXT 92 begin_end_tp('queue_annotation', 93 tp_args=[Arg(type='unsigned', var='len'), 94 Arg(type='str', var='str', c_format='%s', length_arg='len + 1', copy_func='strncpy'),], 95 tp_struct=[Arg(type='uint8_t', name='dummy', var='0')], 96 end_pipelined=False, 97 need_cs_param=True) 98 99 # Batch buffer tracepoints, only for Iris 100 begin_end_tp('batch', 101 tp_args=[Arg(type='uint8_t', var='name', c_format='%hhu'),], 102 end_pipelined=False) 103 104 # Command buffer tracepoints, only for Anv 105 begin_end_tp('cmd_buffer', 106 tp_args=[Arg(type='uint8_t', var='level', c_format='%hhu'),], 107 end_pipelined=False) 108 109 # Annotations for Cmd(Begin|End)DebugUtilsLabelEXT 110 begin_end_tp('cmd_buffer_annotation', 111 tp_args=[Arg(type='unsigned', var='len'), 112 Arg(type='str', var='str', c_format='%s', length_arg='len + 1', copy_func='strncpy'),], 113 tp_struct=[Arg(type='uint8_t', name='dummy', var='0'),], 114 end_pipelined=True) 115 116 # Transform feedback, only for Anv 117 begin_end_tp('xfb', 118 end_pipelined=False) 119 120 # Dynamic rendering tracepoints, only for Anv 121 begin_end_tp('render_pass', 122 tp_args=[Arg(type='uint16_t', var='width', c_format='%hu'), 123 Arg(type='uint16_t', var='height', c_format='%hu'), 124 Arg(type='uint8_t', var='att_count', c_format='%hhu'), 125 Arg(type='uint8_t', var='msaa', c_format='%hhu'),]) 126 127 # Blorp operations, Anv & Iris 128 begin_end_tp('blorp', 129 tp_args=[Arg(type='enum blorp_op', var='op', c_format='%s', to_prim_type='blorp_op_to_name({})'), 130 Arg(type='uint32_t', var='width', c_format='%u'), 131 Arg(type='uint32_t', var='height', c_format='%u'), 132 Arg(type='uint32_t', var='samples', c_format='%u'), 133 Arg(type='enum blorp_shader_pipeline', var='shader_pipe', c_format='%s', to_prim_type='blorp_shader_pipeline_to_name({})'), 134 Arg(type='enum isl_format', var='dst_fmt', c_format='%s', to_prim_type='isl_format_get_short_name({})'), 135 Arg(type='enum isl_format', var='src_fmt', c_format='%s', to_prim_type='isl_format_get_short_name({})'), 136 Arg(type='uint8_t', var='predicated', c_format='%hhu'), 137 ]) 138 139 # vkCmdWriteBufferMarker*, only for Anv 140 begin_end_tp('write_buffer_marker', 141 end_pipelined=False) 142 143 # Indirect draw generation, only for Anv 144 begin_end_tp('generate_draws') 145 begin_end_tp('generate_commands') 146 147 # vkCmdResetQuery, only for Anv 148 begin_end_tp('query_clear_blorp', 149 tp_args=[Arg(type='uint32_t', var='count', c_format='%u')]) 150 begin_end_tp('query_clear_cs', 151 tp_args=[Arg(type='uint32_t', var='count', c_format='%u')], 152 end_pipelined=False) 153 154 # vkCmdCopyQueryResults, only for Anv 155 begin_end_tp('query_copy_cs', 156 tp_args=[Arg(type='uint32_t', var='count', c_format='%u')], 157 end_pipelined=False) 158 begin_end_tp('query_copy_shader', 159 tp_args=[Arg(type='uint32_t', var='count', c_format='%u')]) 160 161 # Various draws/dispatch, Anv & Iris 162 begin_end_tp('draw', 163 tp_args=[Arg(type='uint32_t', var='count', c_format='%u'), 164 Arg(type='uint32_t', var='vs_hash', c_format='%u'), 165 Arg(type='uint32_t', var='fs_hash', c_format='%u')]) 166 begin_end_tp('draw_multi', 167 tp_args=[Arg(type='uint32_t', var='count', c_format='%u'), 168 Arg(type='uint32_t', var='vs_hash', c_format='%u'), 169 Arg(type='uint32_t', var='fs_hash', c_format='%u')]) 170 begin_end_tp('draw_indexed', 171 tp_args=[Arg(type='uint32_t', var='count', c_format='%u'), 172 Arg(type='uint32_t', var='vs_hash', c_format='%u'), 173 Arg(type='uint32_t', var='fs_hash', c_format='%u')]) 174 begin_end_tp('draw_indexed_multi', 175 tp_args=[Arg(type='uint32_t', var='count', c_format='%u'), 176 Arg(type='uint32_t', var='vs_hash', c_format='%u'), 177 Arg(type='uint32_t', var='fs_hash', c_format='%u')]) 178 begin_end_tp('draw_indirect_byte_count', 179 tp_args=[Arg(type='uint32_t', var='instance_count', c_format='%u'), 180 Arg(type='uint32_t', var='vs_hash', c_format='%u'), 181 Arg(type='uint32_t', var='fs_hash', c_format='%u')]) 182 begin_end_tp('draw_indirect', 183 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u'), 184 Arg(type='uint32_t', var='vs_hash', c_format='%u'), 185 Arg(type='uint32_t', var='fs_hash', c_format='%u')]) 186 begin_end_tp('draw_indexed_indirect', 187 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u'), 188 Arg(type='uint32_t', var='vs_hash', c_format='%u'), 189 Arg(type='uint32_t', var='fs_hash', c_format='%u')]) 190 begin_end_tp('draw_indirect_count', 191 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u', 192 is_indirect=True), 193 Arg(type='uint32_t', var='vs_hash', c_format='%u'), 194 Arg(type='uint32_t', var='fs_hash', c_format='%u')]) 195 begin_end_tp('draw_indexed_indirect_count', 196 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u', 197 is_indirect=True), 198 Arg(type='uint32_t', var='vs_hash', c_format='%u'), 199 Arg(type='uint32_t', var='fs_hash', c_format='%u')]) 200 201 begin_end_tp('draw_mesh', 202 tp_args=[Arg(type='uint32_t', var='group_x', c_format='%u'), 203 Arg(type='uint32_t', var='group_y', c_format='%u'), 204 Arg(type='uint32_t', var='group_z', c_format='%u'),]) 205 begin_end_tp('draw_mesh_indirect', 206 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u'),]) 207 begin_end_tp('draw_mesh_indirect_count', 208 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u', 209 is_indirect=True),]) 210 211 begin_end_tp('compute', 212 tp_args=[Arg(type='uint32_t', var='group_x', c_format='%u'), 213 Arg(type='uint32_t', var='group_y', c_format='%u'), 214 Arg(type='uint32_t', var='group_z', c_format='%u'), 215 Arg(type='uint32_t', var='cs_hash', c_format='%u')], 216 compute=True) 217 218 begin_end_tp('compute_indirect', 219 tp_args=[ArgStruct(type='VkDispatchIndirectCommand', var='size', 220 is_indirect=True, c_format="%ux%ux%u", 221 fields=['x', 'y', 'z']), 222 Arg(type='uint32_t', var='cs_hash', c_format='%u')], 223 compute=True) 224 225 # Used to identify copies generated by utrace 226 begin_end_tp('trace_copy', 227 tp_args=[Arg(type='uint32_t', var='count', c_format='%u'),]) 228 begin_end_tp('trace_copy_cb', 229 tp_args=[Arg(type='uint32_t', var='count', c_format='%u'),], 230 need_cs_param=True) 231 232 begin_end_tp('as_build') 233 begin_end_tp('as_build_leaves', maybe_compute=True) 234 begin_end_tp('as_morton_generate', maybe_compute=True) 235 begin_end_tp('as_morton_sort', maybe_compute=True) 236 begin_end_tp('as_lbvh_build_internal', maybe_compute=True) 237 begin_end_tp('as_ploc_build_internal', maybe_compute=True) 238 begin_end_tp('as_encode', maybe_compute=True) 239 begin_end_tp('as_copy', maybe_compute=True) 240 241 begin_end_tp('rays', 242 tp_args=[Arg(type='uint32_t', var='group_x', c_format='%u'), 243 Arg(type='uint32_t', var='group_y', c_format='%u'), 244 Arg(type='uint32_t', var='group_z', c_format='%u'),], 245 compute=True) 246 247 def flag_bits(args): 248 bits = [Arg(type='enum intel_ds_stall_flag', name='flags', var='decode_cb(flags)', c_format='0x%x')] 249 for a in args: 250 bits.append(Arg(type='bool', name=a[1], var='__entry->flags & INTEL_DS_{0}_BIT'.format(a[0]), c_format='%u')) 251 return bits 252 253 def stall_args(args): 254 fmt = '' 255 exprs = [] 256 for a in args: 257 fmt += '%s' 258 exprs.append('(__entry->flags & INTEL_DS_{0}_BIT) ? "+{1}" : ""'.format(a[0], a[1])) 259 fmt += ' : %s%s%s%s%s%s%s' 260 exprs.append('(__entry->reason1) ? __entry->reason1 : "unknown"') 261 exprs.append('(__entry->reason2) ? "; " : ""') 262 exprs.append('(__entry->reason2) ? __entry->reason2 : ""') 263 exprs.append('(__entry->reason3) ? "; " : ""') 264 exprs.append('(__entry->reason3) ? __entry->reason3 : ""') 265 exprs.append('(__entry->reason4) ? "; " : ""') 266 exprs.append('(__entry->reason4) ? __entry->reason4 : ""') 267 # To printout flags 268 # fmt += '(0x%08x)' 269 # exprs.append('__entry->flags') 270 fmt = [fmt] 271 fmt += exprs 272 return fmt 273 274 stall_flags = [['DEPTH_CACHE_FLUSH', 'depth_flush'], 275 ['DATA_CACHE_FLUSH', 'dc_flush'], 276 ['HDC_PIPELINE_FLUSH', 'hdc_flush'], 277 ['RENDER_TARGET_CACHE_FLUSH', 'rt_flush'], 278 ['TILE_CACHE_FLUSH', 'tile_flush'], 279 ['STATE_CACHE_INVALIDATE', 'state_inval'], 280 ['CONST_CACHE_INVALIDATE', 'const_inval'], 281 ['VF_CACHE_INVALIDATE', 'vf_inval'], 282 ['TEXTURE_CACHE_INVALIDATE', 'tex_inval'], 283 ['INST_CACHE_INVALIDATE', 'ic_inval'], 284 ['STALL_AT_SCOREBOARD', 'pb_stall'], 285 ['DEPTH_STALL', 'depth_stall'], 286 ['CS_STALL', 'cs_stall'], 287 ['UNTYPED_DATAPORT_CACHE_FLUSH', 'udp_flush'], 288 ['PSS_STALL_SYNC', 'pss_stall'], 289 ['END_OF_PIPE', 'eop'], 290 ['CCS_CACHE_FLUSH', 'ccs_flush']] 291 292 begin_end_tp('stall', 293 tp_args=[ArgStruct(type='uint32_t', var='flags'), 294 ArgStruct(type='intel_ds_stall_cb_t', var='decode_cb'), 295 ArgStruct(type='const char *', var='reason1'), 296 ArgStruct(type='const char *', var='reason2'), 297 ArgStruct(type='const char *', var='reason3'), 298 ArgStruct(type='const char *', var='reason4'),], 299 tp_struct=[Arg(type='uint32_t', name='flags', var='decode_cb(flags)', c_format='0x%x'), 300 Arg(type='const char *', name='reason1', var='reason1', c_format='%s'), 301 Arg(type='const char *', name='reason2', var='reason2', c_format='%s'), 302 Arg(type='const char *', name='reason3', var='reason3', c_format='%s'), 303 Arg(type='const char *', name='reason4', var='reason4', c_format='%s'),], 304 tp_print=stall_args(stall_flags), 305 tp_default_enabled=False, 306 end_pipelined=False) 307 308 309def generate_code(args): 310 from u_trace import utrace_generate 311 from u_trace import utrace_generate_perfetto_utils 312 313 utrace_generate(cpath=args.utrace_src, hpath=args.utrace_hdr, 314 ctx_param='struct intel_ds_device *dev', 315 trace_toggle_name='intel_gpu_tracepoint', 316 trace_toggle_defaults=intel_default_tps) 317 utrace_generate_perfetto_utils(hpath=args.perfetto_hdr, 318 basename="intel_tracepoint") 319 320 321def main(): 322 parser = argparse.ArgumentParser() 323 parser.add_argument('-p', '--import-path', required=True) 324 parser.add_argument('--utrace-src', required=True) 325 parser.add_argument('--utrace-hdr', required=True) 326 parser.add_argument('--perfetto-hdr', required=True) 327 args = parser.parse_args() 328 sys.path.insert(0, args.import_path) 329 define_tracepoints(args) 330 generate_code(args) 331 332 333if __name__ == '__main__': 334 main() 335