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, 51 need_cs_param=False): 52 global intel_default_tps 53 if tp_default_enabled: 54 intel_default_tps.append(name) 55 Tracepoint('intel_begin_{0}'.format(name), 56 toggle_name=name, 57 tp_perfetto='intel_ds_begin_{0}'.format(name), 58 need_cs_param=need_cs_param) 59 tp_flags = [] 60 if end_pipelined: 61 if compute: 62 tp_flags.append('INTEL_DS_TRACEPOINT_FLAG_END_OF_PIPE_CS') 63 else: 64 tp_flags.append('INTEL_DS_TRACEPOINT_FLAG_END_OF_PIPE') 65 Tracepoint('intel_end_{0}'.format(name), 66 toggle_name=name, 67 args=tp_args, 68 tp_struct=tp_struct, 69 tp_perfetto='intel_ds_end_{0}'.format(name), 70 tp_print=tp_print, 71 tp_flags=tp_flags, 72 need_cs_param=need_cs_param) 73 74 # Frame tracepoints 75 begin_end_tp('frame', 76 tp_args=[Arg(type='uint32_t', var='frame', c_format='%u'),], 77 end_pipelined=False, 78 need_cs_param=True) 79 80 # Annotations for Queue(Begin|End)DebugUtilsLabelEXT 81 begin_end_tp('queue_annotation', 82 tp_args=[Arg(type='unsigned', var='len'), 83 Arg(type='str', var='str', c_format='%s', length_arg='len + 1', copy_func='strncpy'),], 84 tp_struct=[Arg(type='uint8_t', name='dummy', var='0')], 85 end_pipelined=False, 86 need_cs_param=True) 87 88 # Batch buffer tracepoints, only for Iris 89 begin_end_tp('batch', 90 tp_args=[Arg(type='uint8_t', var='name', c_format='%hhu'),], 91 end_pipelined=False) 92 93 # Command buffer tracepoints, only for Anv 94 begin_end_tp('cmd_buffer', 95 tp_args=[Arg(type='uint8_t', var='level', c_format='%hhu'),], 96 end_pipelined=False) 97 98 # Annotations for Cmd(Begin|End)DebugUtilsLabelEXT 99 begin_end_tp('cmd_buffer_annotation', 100 tp_args=[Arg(type='unsigned', var='len'), 101 Arg(type='str', var='str', c_format='%s', length_arg='len + 1', copy_func='strncpy'),], 102 tp_struct=[Arg(type='uint8_t', name='dummy', var='0'),], 103 end_pipelined=True) 104 105 # Transform feedback, only for Anv 106 begin_end_tp('xfb', 107 end_pipelined=False) 108 109 # Dynamic rendering tracepoints, only for Anv 110 begin_end_tp('render_pass', 111 tp_args=[Arg(type='uint16_t', var='width', c_format='%hu'), 112 Arg(type='uint16_t', var='height', c_format='%hu'), 113 Arg(type='uint8_t', var='att_count', c_format='%hhu'), 114 Arg(type='uint8_t', var='msaa', c_format='%hhu'),]) 115 116 # Blorp operations, Anv & Iris 117 begin_end_tp('blorp', 118 tp_args=[Arg(type='enum blorp_op', var='op', c_format='%s', to_prim_type='blorp_op_to_name({})'), 119 Arg(type='uint32_t', var='width', c_format='%u'), 120 Arg(type='uint32_t', var='height', c_format='%u'), 121 Arg(type='uint32_t', var='samples', c_format='%u'), 122 Arg(type='enum blorp_shader_pipeline', var='shader_pipe', c_format='%s', to_prim_type='blorp_shader_pipeline_to_name({})'), 123 Arg(type='enum isl_format', var='dst_fmt', c_format='%s', to_prim_type='isl_format_get_short_name({})'), 124 Arg(type='enum isl_format', var='src_fmt', c_format='%s', to_prim_type='isl_format_get_short_name({})'), 125 Arg(type='uint8_t', var='predicated', c_format='%hhu'), 126 ]) 127 128 # vkCmdWriteBufferMarker*, only for Anv 129 begin_end_tp('write_buffer_marker', 130 end_pipelined=False) 131 132 # Indirect draw generation, only for Anv 133 begin_end_tp('generate_draws') 134 begin_end_tp('generate_commands') 135 136 # vkCmdResetQuery, only for Anv 137 begin_end_tp('query_clear_blorp', 138 tp_args=[Arg(type='uint32_t', var='count', c_format='%u')]) 139 begin_end_tp('query_clear_cs', 140 tp_args=[Arg(type='uint32_t', var='count', c_format='%u')], 141 end_pipelined=False) 142 143 # vkCmdCopyQueryResults, only for Anv 144 begin_end_tp('query_copy_cs', 145 tp_args=[Arg(type='uint32_t', var='count', c_format='%u')], 146 end_pipelined=False) 147 begin_end_tp('query_copy_shader', 148 tp_args=[Arg(type='uint32_t', var='count', c_format='%u')]) 149 150 # Various draws/dispatch, Anv & Iris 151 begin_end_tp('draw', 152 tp_args=[Arg(type='uint32_t', var='count', c_format='%u')]) 153 begin_end_tp('draw_multi', 154 tp_args=[Arg(type='uint32_t', var='count', c_format='%u'),]) 155 begin_end_tp('draw_indexed', 156 tp_args=[Arg(type='uint32_t', var='count', c_format='%u'),]) 157 begin_end_tp('draw_indexed_multi', 158 tp_args=[Arg(type='uint32_t', var='count', c_format='%u'),]) 159 begin_end_tp('draw_indirect_byte_count', 160 tp_args=[Arg(type='uint32_t', var='instance_count', c_format='%u'),]) 161 begin_end_tp('draw_indirect', 162 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u'),]) 163 begin_end_tp('draw_indexed_indirect', 164 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u'),]) 165 begin_end_tp('draw_indirect_count', 166 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u', 167 is_indirect=True),]) 168 begin_end_tp('draw_indexed_indirect_count', 169 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u', 170 is_indirect=True),]) 171 172 begin_end_tp('draw_mesh', 173 tp_args=[Arg(type='uint32_t', var='group_x', c_format='%u'), 174 Arg(type='uint32_t', var='group_y', c_format='%u'), 175 Arg(type='uint32_t', var='group_z', c_format='%u'),]) 176 begin_end_tp('draw_mesh_indirect', 177 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u'),]) 178 begin_end_tp('draw_mesh_indirect_count', 179 tp_args=[Arg(type='uint32_t', var='draw_count', c_format='%u', 180 is_indirect=True),]) 181 182 begin_end_tp('compute', 183 tp_args=[Arg(type='uint32_t', var='group_x', c_format='%u'), 184 Arg(type='uint32_t', var='group_y', c_format='%u'), 185 Arg(type='uint32_t', var='group_z', c_format='%u'),], 186 compute=True) 187 188 begin_end_tp('compute_indirect', 189 tp_args=[ArgStruct(type='VkDispatchIndirectCommand', var='size', 190 is_indirect=True, c_format="%ux%ux%u", 191 fields=['x', 'y', 'z'])], 192 compute=True) 193 194 # Used to identify copies generated by utrace 195 begin_end_tp('trace_copy', 196 tp_args=[Arg(type='uint32_t', var='count', c_format='%u'),]) 197 begin_end_tp('trace_copy_cb', 198 tp_args=[Arg(type='uint32_t', var='count', c_format='%u'),], 199 need_cs_param=True) 200 201 begin_end_tp('as_build') 202 203 begin_end_tp('rays', 204 tp_args=[Arg(type='uint32_t', var='group_x', c_format='%u'), 205 Arg(type='uint32_t', var='group_y', c_format='%u'), 206 Arg(type='uint32_t', var='group_z', c_format='%u'),], 207 compute=True) 208 209 def flag_bits(args): 210 bits = [Arg(type='enum intel_ds_stall_flag', name='flags', var='decode_cb(flags)', c_format='0x%x')] 211 for a in args: 212 bits.append(Arg(type='bool', name=a[1], var='__entry->flags & INTEL_DS_{0}_BIT'.format(a[0]), c_format='%u')) 213 return bits 214 215 def stall_args(args): 216 fmt = '' 217 exprs = [] 218 for a in args: 219 fmt += '%s' 220 exprs.append('(__entry->flags & INTEL_DS_{0}_BIT) ? "+{1}" : ""'.format(a[0], a[1])) 221 fmt += ' : %s%s%s%s%s%s%s' 222 exprs.append('(__entry->reason1) ? __entry->reason1 : "unknown"') 223 exprs.append('(__entry->reason2) ? "; " : ""') 224 exprs.append('(__entry->reason2) ? __entry->reason2 : ""') 225 exprs.append('(__entry->reason3) ? "; " : ""') 226 exprs.append('(__entry->reason3) ? __entry->reason3 : ""') 227 exprs.append('(__entry->reason4) ? "; " : ""') 228 exprs.append('(__entry->reason4) ? __entry->reason4 : ""') 229 # To printout flags 230 # fmt += '(0x%08x)' 231 # exprs.append('__entry->flags') 232 fmt = [fmt] 233 fmt += exprs 234 return fmt 235 236 stall_flags = [['DEPTH_CACHE_FLUSH', 'depth_flush'], 237 ['DATA_CACHE_FLUSH', 'dc_flush'], 238 ['HDC_PIPELINE_FLUSH', 'hdc_flush'], 239 ['RENDER_TARGET_CACHE_FLUSH', 'rt_flush'], 240 ['TILE_CACHE_FLUSH', 'tile_flush'], 241 ['STATE_CACHE_INVALIDATE', 'state_inval'], 242 ['CONST_CACHE_INVALIDATE', 'const_inval'], 243 ['VF_CACHE_INVALIDATE', 'vf_inval'], 244 ['TEXTURE_CACHE_INVALIDATE', 'tex_inval'], 245 ['INST_CACHE_INVALIDATE', 'ic_inval'], 246 ['STALL_AT_SCOREBOARD', 'pb_stall'], 247 ['DEPTH_STALL', 'depth_stall'], 248 ['CS_STALL', 'cs_stall'], 249 ['UNTYPED_DATAPORT_CACHE_FLUSH', 'udp_flush'], 250 ['PSS_STALL_SYNC', 'pss_stall'], 251 ['END_OF_PIPE', 'eop'], 252 ['CCS_CACHE_FLUSH', 'ccs_flush']] 253 254 begin_end_tp('stall', 255 tp_args=[ArgStruct(type='uint32_t', var='flags'), 256 ArgStruct(type='intel_ds_stall_cb_t', var='decode_cb'), 257 ArgStruct(type='const char *', var='reason1'), 258 ArgStruct(type='const char *', var='reason2'), 259 ArgStruct(type='const char *', var='reason3'), 260 ArgStruct(type='const char *', var='reason4'),], 261 tp_struct=[Arg(type='uint32_t', name='flags', var='decode_cb(flags)', c_format='0x%x'), 262 Arg(type='const char *', name='reason1', var='reason1', c_format='%s'), 263 Arg(type='const char *', name='reason2', var='reason2', c_format='%s'), 264 Arg(type='const char *', name='reason3', var='reason3', c_format='%s'), 265 Arg(type='const char *', name='reason4', var='reason4', c_format='%s'),], 266 tp_print=stall_args(stall_flags), 267 tp_default_enabled=False, 268 end_pipelined=False) 269 270 271def generate_code(args): 272 from u_trace import utrace_generate 273 from u_trace import utrace_generate_perfetto_utils 274 275 utrace_generate(cpath=args.utrace_src, hpath=args.utrace_hdr, 276 ctx_param='struct intel_ds_device *dev', 277 trace_toggle_name='intel_gpu_tracepoint', 278 trace_toggle_defaults=intel_default_tps) 279 utrace_generate_perfetto_utils(hpath=args.perfetto_hdr, 280 basename="intel_tracepoint") 281 282 283def main(): 284 parser = argparse.ArgumentParser() 285 parser.add_argument('-p', '--import-path', required=True) 286 parser.add_argument('--utrace-src', required=True) 287 parser.add_argument('--utrace-hdr', required=True) 288 parser.add_argument('--perfetto-hdr', required=True) 289 args = parser.parse_args() 290 sys.path.insert(0, args.import_path) 291 define_tracepoints(args) 292 generate_code(args) 293 294 295if __name__ == '__main__': 296 main() 297