src/intel/perf/gen_perf.py

   1 # Copyright (c) 2015-2017 Intel Corporation
   2 #
   3 # Permission is hereby granted, free of charge, to any person obtaining a
   4 # copy of this software and associated documentation files (the "Software"),
   5 # to deal in the Software without restriction, including without limitation
   6 # the rights to use, copy, modify, merge, publish, distribute, sublicense,
   7 # and/or sell copies of the Software, and to permit persons to whom the
   8 # Software is furnished to do so, subject to the following conditions:
   9 #
  10 # The above copyright notice and this permission notice (including the next
  11 # paragraph) shall be included in all copies or substantial portions of the
  12 # Software.
  13 #
  14 # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15 # IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16 # FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17 # THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  18 # LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  19 # FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  20 # IN THE SOFTWARE.
  21
  22 import argparse
  23 import os
  24 import sys
  25 import textwrap
  26
  27 import xml.etree.cElementTree as et
  28
  29 hashed_funcs = {}
  30
  31 c_file = None
  32 _c_indent = 0
  33
  34 def c(*args):
  35     code = ' '.join(map(str,args))
  36     for line in code.splitlines():
  37         text = ''.rjust(_c_indent) + line
  38         c_file.write(text.rstrip() + "\n")
  39
  40 # indented, but no trailing newline...
  41 def c_line_start(code):
  42     c_file.write(''.rjust(_c_indent) + code)
  43 def c_raw(code):
  44     c_file.write(code)
  45
  46 def c_indent(n):
  47     global _c_indent
  48     _c_indent = _c_indent + n
  49 def c_outdent(n):
  50     global _c_indent
  51     _c_indent = _c_indent - n
  52
  53 header_file = None
  54 _h_indent = 0
  55
  56 def h(*args):
  57     code = ' '.join(map(str,args))
  58     for line in code.splitlines():
  59         text = ''.rjust(_h_indent) + line
  60         header_file.write(text.rstrip() + "\n")
  61
  62 def h_indent(n):
  63     global _c_indent
  64     _h_indent = _h_indent + n
  65 def h_outdent(n):
  66     global _c_indent
  67     _h_indent = _h_indent - n
  68
  69
  70 def emit_fadd(tmp_id, args):
  71     c("double tmp{0} = {1} + {2};".format(tmp_id, args[1], args[0]))
  72     return tmp_id + 1
  73
  74 # Be careful to check for divide by zero...
  75 def emit_fdiv(tmp_id, args):
  76     c("double tmp{0} = {1};".format(tmp_id, args[1]))
  77     c("double tmp{0} = {1};".format(tmp_id + 1, args[0]))
  78     c("double tmp{0} = tmp{1} ? tmp{2} / tmp{1} : 0;".format(tmp_id + 2, tmp_id + 1, tmp_id))
  79     return tmp_id + 3
  80
  81 def emit_fmax(tmp_id, args):
  82     c("double tmp{0} = {1};".format(tmp_id, args[1]))
  83     c("double tmp{0} = {1};".format(tmp_id + 1, args[0]))
  84     c("double tmp{0} = MAX(tmp{1}, tmp{2});".format(tmp_id + 2, tmp_id, tmp_id + 1))
  85     return tmp_id + 3
  86
  87 def emit_fmul(tmp_id, args):
  88     c("double tmp{0} = {1} * {2};".format(tmp_id, args[1], args[0]))
  89     return tmp_id + 1
  90
  91 def emit_fsub(tmp_id, args):
  92     c("double tmp{0} = {1} - {2};".format(tmp_id, args[1], args[0]))
  93     return tmp_id + 1
  94
  95 def emit_read(tmp_id, args):
  96     type = args[1].lower()
  97     c("uint64_t tmp{0} = accumulator[query->{1}_offset + {2}];".format(tmp_id, type, args[0]))
  98     return tmp_id + 1
  99
 100 def emit_uadd(tmp_id, args):
 101     c("uint64_t tmp{0} = {1} + {2};".format(tmp_id, args[1], args[0]))
 102     return tmp_id + 1
 103
 104 # Be careful to check for divide by zero...
 105 def emit_udiv(tmp_id, args):
 106     c("uint64_t tmp{0} = {1};".format(tmp_id, args[1]))
 107     c("uint64_t tmp{0} = {1};".format(tmp_id + 1, args[0]))
 108     c("uint64_t tmp{0} = tmp{1} ? tmp{2} / tmp{1} : 0;".format(tmp_id + 2, tmp_id + 1, tmp_id))
 109     return tmp_id + 3
 110
 111 def emit_umul(tmp_id, args):
 112     c("uint64_t tmp{0} = {1} * {2};".format(tmp_id, args[1], args[0]))
 113     return tmp_id + 1
 114
 115 def emit_usub(tmp_id, args):
 116     c("uint64_t tmp{0} = {1} - {2};".format(tmp_id, args[1], args[0]))
 117     return tmp_id + 1
 118
 119 def emit_umin(tmp_id, args):
 120     c("uint64_t tmp{0} = MIN({1}, {2});".format(tmp_id, args[1], args[0]))
 121     return tmp_id + 1
 122
 123 def emit_lshft(tmp_id, args):
 124     c("uint64_t tmp{0} = {1} << {2};".format(tmp_id, args[1], args[0]))
 125     return tmp_id + 1
 126
 127 def emit_rshft(tmp_id, args):
 128     c("uint64_t tmp{0} = {1} >> {2};".format(tmp_id, args[1], args[0]))
 129     return tmp_id + 1
 130
 131 def emit_and(tmp_id, args):
 132     c("uint64_t tmp{0} = {1} & {2};".format(tmp_id, args[1], args[0]))
 133     return tmp_id + 1
 134
 135 ops = {}
 136 #             (n operands, emitter)
 137 ops["FADD"] = (2, emit_fadd)
 138 ops["FDIV"] = (2, emit_fdiv)
 139 ops["FMAX"] = (2, emit_fmax)
 140 ops["FMUL"] = (2, emit_fmul)
 141 ops["FSUB"] = (2, emit_fsub)
 142 ops["READ"] = (2, emit_read)
 143 ops["UADD"] = (2, emit_uadd)
 144 ops["UDIV"] = (2, emit_udiv)
 145 ops["UMUL"] = (2, emit_umul)
 146 ops["USUB"] = (2, emit_usub)
 147 ops["UMIN"] = (2, emit_umin)
 148 ops["<<"]   = (2, emit_lshft)
 149 ops[">>"]   = (2, emit_rshft)
 150 ops["AND"]  = (2, emit_and)
 151
 152 def brkt(subexp):
 153     if " " in subexp:
 154         return "(" + subexp + ")"
 155     else:
 156         return subexp
 157
 158 def splice_bitwise_and(args):
 159     return brkt(args[1]) + " & " + brkt(args[0])
 160
 161 def splice_logical_and(args):
 162     return brkt(args[1]) + " && " + brkt(args[0])
 163
 164 def splice_ult(args):
 165     return brkt(args[1]) + " < " + brkt(args[0])
 166
 167 def splice_ugte(args):
 168     return brkt(args[1]) + " >= " + brkt(args[0])
 169
 170 exp_ops = {}
 171 #                 (n operands, splicer)
 172 exp_ops["AND"]  = (2, splice_bitwise_and)
 173 exp_ops["UGTE"] = (2, splice_ugte)
 174 exp_ops["ULT"]  = (2, splice_ult)
 175 exp_ops["&&"]   = (2, splice_logical_and)
 176
 177
 178 hw_vars = {}
 179 hw_vars["$EuCoresTotalCount"] = "perf->sys_vars.n_eus"
 180 hw_vars["$EuSlicesTotalCount"] = "perf->sys_vars.n_eu_slices"
 181 hw_vars["$EuSubslicesTotalCount"] = "perf->sys_vars.n_eu_sub_slices"
 182 hw_vars["$EuThreadsCount"] = "perf->sys_vars.eu_threads_count"
 183 hw_vars["$SliceMask"] = "perf->sys_vars.slice_mask"
 184 hw_vars["$SubsliceMask"] = "perf->sys_vars.subslice_mask"
 185 hw_vars["$GpuTimestampFrequency"] = "perf->sys_vars.timestamp_frequency"
 186 hw_vars["$GpuMinFrequency"] = "perf->sys_vars.gt_min_freq"
 187 hw_vars["$GpuMaxFrequency"] = "perf->sys_vars.gt_max_freq"
 188 hw_vars["$SkuRevisionId"] = "perf->sys_vars.revision"
 189
 190 def output_rpn_equation_code(set, counter, equation):
 191     c("/* RPN equation: " + equation + " */")
 192     tokens = equation.split()
 193     stack = []
 194     tmp_id = 0
 195     tmp = None
 196
 197     for token in tokens:
 198         stack.append(token)
 199         while stack and stack[-1] in ops:
 200             op = stack.pop()
 201             argc, callback = ops[op]
 202             args = []
 203             for i in range(0, argc):
 204                 operand = stack.pop()
 205                 if operand[0] == "$":
 206                     if operand in hw_vars:
 207                         operand = hw_vars[operand]
 208                     elif operand in set.counter_vars:
 209                         reference = set.counter_vars[operand]
 210                         operand = set.read_funcs[operand[1:]] + "(perf, query, accumulator)"
 211                     else:
 212                         raise Exception("Failed to resolve variable " + operand + " in equation " + equation + " for " + set.name + " :: " + counter.get('name'));
 213                 args.append(operand)
 214
 215             tmp_id = callback(tmp_id, args)
 216
 217             tmp = "tmp{0}".format(tmp_id - 1)
 218             stack.append(tmp)
 219
 220     if len(stack) != 1:
 221         raise Exception("Spurious empty rpn code for " + set.name + " :: " +
 222                 counter.get('name') + ".\nThis is probably due to some unhandled RPN function, in the equation \"" +
 223                 equation + "\"")
 224
 225     value = stack[-1]
 226
 227     if value in hw_vars:
 228         value = hw_vars[value]
 229     if value in set.counter_vars:
 230         value = set.read_funcs[value[1:]] + "(perf, query, accumulator)"
 231
 232     c("\nreturn " + value + ";")
 233
 234 def splice_rpn_expression(set, counter, expression):
 235     tokens = expression.split()
 236     stack = []
 237
 238     for token in tokens:
 239         stack.append(token)
 240         while stack and stack[-1] in exp_ops:
 241             op = stack.pop()
 242             argc, callback = exp_ops[op]
 243             args = []
 244             for i in range(0, argc):
 245                 operand = stack.pop()
 246                 if operand[0] == "$":
 247                     if operand in hw_vars:
 248                         operand = hw_vars[operand]
 249                     else:
 250                         raise Exception("Failed to resolve variable " + operand + " in expression " + expression + " for " + set.name + " :: " + counter.get('name'));
 251                 args.append(operand)
 252
 253             subexp = callback(args)
 254
 255             stack.append(subexp)
 256
 257     if len(stack) != 1:
 258         raise Exception("Spurious empty rpn expression for " + set.name + " :: " +
 259                 counter.get('name') + ".\nThis is probably due to some unhandled RPN operation, in the expression \"" +
 260                 expression + "\"")
 261
 262     return stack[-1]
 263
 264 def output_counter_read(gen, set, counter):
 265     c("\n")
 266     c("/* {0} :: {1} */".format(set.name, counter.get('name')))
 267
 268     if counter.read_hash in hashed_funcs:
 269         c("#define %s \\" % counter.read_sym)
 270         c_indent(3)
 271         c("%s" % hashed_funcs[counter.read_hash])
 272         c_outdent(3)
 273     else:
 274         ret_type = counter.get('data_type')
 275         if ret_type == "uint64":
 276             ret_type = "uint64_t"
 277
 278         read_eq = counter.get('equation')
 279
 280         c("static " + ret_type)
 281         c(counter.read_sym + "(UNUSED struct gen_perf_config *perf,\n")
 282         c_indent(len(counter.read_sym) + 1)
 283         c("const struct gen_perf_query_info *query,\n")
 284         c("const uint64_t *accumulator)\n")
 285         c_outdent(len(counter.read_sym) + 1)
 286
 287         c("{")
 288         c_indent(3)
 289         output_rpn_equation_code(set, counter, read_eq)
 290         c_outdent(3)
 291         c("}")
 292
 293         hashed_funcs[counter.read_hash] = counter.read_sym
 294
 295
 296 def output_counter_max(gen, set, counter):
 297     max_eq = counter.get('max_equation')
 298
 299     if not counter.has_max_func():
 300         return
 301
 302     c("\n")
 303     c("/* {0} :: {1} */".format(set.name, counter.get('name')))
 304
 305     if counter.max_hash in hashed_funcs:
 306         c("#define %s \\" % counter.max_sym())
 307         c_indent(3)
 308         c("%s" % hashed_funcs[counter.max_hash])
 309         c_outdent(3)
 310     else:
 311         ret_type = counter.get('data_type')
 312         if ret_type == "uint64":
 313             ret_type = "uint64_t"
 314
 315         c("static " + ret_type)
 316         c(counter.max_sym() + "(struct gen_perf_config *perf)\n")
 317         c("{")
 318         c_indent(3)
 319         output_rpn_equation_code(set, counter, max_eq)
 320         c_outdent(3)
 321         c("}")
 322
 323         hashed_funcs[counter.max_hash] = counter.max_sym()
 324
 325
 326 c_type_sizes = { "uint32_t": 4, "uint64_t": 8, "float": 4, "double": 8, "bool": 4 }
 327 def sizeof(c_type):
 328     return c_type_sizes[c_type]
 329
 330 def pot_align(base, pot_alignment):
 331     return (base + pot_alignment - 1) & ~(pot_alignment - 1);
 332
 333 semantic_type_map = {
 334     "duration": "raw",
 335     "ratio": "event"
 336     }
 337
 338 def output_availability(set, availability, counter_name):
 339     expression = splice_rpn_expression(set, counter_name, availability)
 340     lines = expression.split(' && ')
 341     n_lines = len(lines)
 342     if n_lines == 1:
 343         c("if (" + lines[0] + ") {")
 344     else:
 345         c("if (" + lines[0] + " &&")
 346         c_indent(4)
 347         for i in range(1, (n_lines - 1)):
 348             c(lines[i] + " &&")
 349         c(lines[(n_lines - 1)] + ") {")
 350         c_outdent(4)
 351
 352
 353 def output_counter_report(set, counter, current_offset):
 354     data_type = counter.get('data_type')
 355     data_type_uc = data_type.upper()
 356     c_type = data_type
 357
 358     if "uint" in c_type:
 359         c_type = c_type + "_t"
 360
 361     semantic_type = counter.get('semantic_type')
 362     if semantic_type in semantic_type_map:
 363         semantic_type = semantic_type_map[semantic_type]
 364
 365     semantic_type_uc = semantic_type.upper()
 366
 367     c("\n")
 368
 369     availability = counter.get('availability')
 370     if availability:
 371         output_availability(set, availability, counter.get('name'))
 372         c_indent(3)
 373
 374     c("counter = &query->counters[query->n_counters++];\n")
 375     c("counter->oa_counter_read_" + data_type + " = " + set.read_funcs[counter.get('symbol_name')] + ";\n")
 376     c("counter->name = \"" + counter.get('name') + "\";\n")
 377     c("counter->desc = \"" + counter.get('description') + "\";\n")
 378     c("counter->type = GEN_PERF_COUNTER_TYPE_" + semantic_type_uc + ";\n")
 379     c("counter->data_type = GEN_PERF_COUNTER_DATA_TYPE_" + data_type_uc + ";\n")
 380     c("counter->raw_max = " + set.max_values[counter.get('symbol_name')] + ";\n")
 381
 382     current_offset = pot_align(current_offset, sizeof(c_type))
 383     c("counter->offset = " + str(current_offset) + ";\n")
 384
 385     if availability:
 386         c_outdent(3);
 387         c("}")
 388
 389     return current_offset + sizeof(c_type)
 390
 391
 392 register_types = {
 393     'FLEX': 'flex_regs',
 394     'NOA': 'mux_regs',
 395     'OA': 'b_counter_regs',
 396 }
 397
 398 def compute_register_lengths(set):
 399     register_lengths = {}
 400     register_configs = set.findall('register_config')
 401     for register_config in register_configs:
 402         t = register_types[register_config.get('type')]
 403         if t not in register_lengths:
 404             register_lengths[t] = len(register_config.findall('register'))
 405         else:
 406             register_lengths[t] += len(register_config.findall('register'))
 407
 408     return register_lengths
 409
 410
 411 def generate_register_configs(set):
 412     register_configs = set.findall('register_config')
 413     for register_config in register_configs:
 414         t = register_types[register_config.get('type')]
 415
 416         availability = register_config.get('availability')
 417         if availability:
 418             output_availability(set, availability, register_config.get('type') + ' register config')
 419             c_indent(3)
 420
 421         for register in register_config.findall('register'):
 422             c("query->%s[query->n_%s++] = (struct gen_perf_query_register_prog) { .reg = %s, .val = %s };" %
 423               (t, t, register.get('address'), register.get('value')))
 424
 425         if availability:
 426             c_outdent(3)
 427             c("}")
 428         c("\n")
 429
 430
 431 # Wraps a <counter> element from the oa-*.xml files.
 432 class Counter:
 433     def __init__(self, set, xml):
 434         self.xml = xml
 435         self.set = set
 436         self.read_hash = None
 437         self.max_hash = None
 438
 439         self.read_sym = "{0}__{1}__{2}__read".format(self.set.gen.chipset,
 440                                                      self.set.underscore_name,
 441                                                      self.xml.get('underscore_name'))
 442
 443     def get(self, prop):
 444         return self.xml.get(prop)
 445
 446     # Compute the hash of a counter's equation by expanding (including all the
 447     # sub-equations it depends on)
 448     def compute_hashes(self):
 449         if self.read_hash is not None:
 450             return
 451
 452         def replace_token(token):
 453             if token[0] != "$":
 454                 return token
 455             if token not in self.set.counter_vars:
 456                 return token
 457             self.set.counter_vars[token].compute_hashes()
 458             return self.set.counter_vars[token].read_hash
 459
 460         read_eq = self.xml.get('equation')
 461         self.read_hash = ' '.join(map(replace_token, read_eq.split()))
 462
 463         max_eq = self.xml.get('max_equation')
 464         if max_eq:
 465             self.max_hash = ' '.join(map(replace_token, max_eq.split()))
 466
 467     def has_max_func(self):
 468         max_eq = self.xml.get('max_equation')
 469         if not max_eq:
 470             return False
 471
 472         try:
 473             val = float(max_eq)
 474             return False
 475         except ValueError:
 476             pass
 477
 478         for token in max_eq.split():
 479             if token[0] == '$' and token not in hw_vars:
 480                 return False
 481         return True
 482
 483     def max_sym(self):
 484         assert self.has_max_func()
 485         return "{0}__{1}__{2}__max".format(self.set.gen.chipset,
 486                                            self.set.underscore_name,
 487                                            self.xml.get('underscore_name'))
 488
 489     def max_value(self):
 490         max_eq = self.xml.get('max_equation')
 491         if not max_eq:
 492             return "0 /* undefined */"
 493
 494         try:
 495             return "{0}".format(float(max_eq))
 496         except ValueError:
 497             pass
 498
 499         for token in max_eq.split():
 500             if token[0] == '$' and token not in hw_vars:
 501                 return "0 /* unsupported (varies over time) */"
 502
 503         return "{0}__{1}__{2}__max(perf)".format(self.set.gen.chipset,
 504                                                  self.set.underscore_name,
 505                                                  self.xml.get('underscore_name'))
 506
 507 # Wraps a <set> element from the oa-*.xml files.
 508 class Set:
 509     def __init__(self, gen, xml):
 510         self.gen = gen
 511         self.xml = xml
 512
 513         self.counter_vars = {}
 514         self.max_values = {}
 515         self.read_funcs = {}
 516
 517         xml_counters = self.xml.findall("counter")
 518         self.counters = []
 519         for xml_counter in xml_counters:
 520             counter = Counter(self, xml_counter)
 521             self.counters.append(counter)
 522             self.counter_vars["$" + counter.get('symbol_name')] = counter
 523             self.read_funcs[counter.get('symbol_name')] = counter.read_sym
 524             self.max_values[counter.get('symbol_name')] = counter.max_value()
 525
 526         for counter in self.counters:
 527             counter.compute_hashes()
 528
 529     @property
 530     def hw_config_guid(self):
 531         return self.xml.get('hw_config_guid')
 532
 533     @property
 534     def name(self):
 535         return self.xml.get('name')
 536
 537     @property
 538     def symbol_name(self):
 539         return self.xml.get('symbol_name')
 540
 541     @property
 542     def underscore_name(self):
 543         return self.xml.get('underscore_name')
 544
 545     def findall(self, path):
 546         return self.xml.findall(path)
 547
 548     def find(self, path):
 549         return self.xml.find(path)
 550
 551
 552 # Wraps an entire oa-*.xml file.
 553 class Gen:
 554     def __init__(self, filename):
 555         self.filename = filename
 556         self.xml = et.parse(self.filename)
 557         self.chipset = self.xml.find('.//set').get('chipset').lower()
 558         self.sets = []
 559
 560         for xml_set in self.xml.findall(".//set"):
 561             self.sets.append(Set(self, xml_set))
 562
 563
 564 def main():
 565     global c_file
 566     global header_file
 567
 568     parser = argparse.ArgumentParser()
 569     parser.add_argument("--header", help="Header file to write", required=True)
 570     parser.add_argument("--code", help="C file to write", required=True)
 571     parser.add_argument("xml_files", nargs='+', help="List of xml metrics files to process")
 572
 573     args = parser.parse_args()
 574
 575     c_file = open(args.code, 'w')
 576     header_file = open(args.header, 'w')
 577
 578     gens = []
 579     for xml_file in args.xml_files:
 580         gens.append(Gen(xml_file))
 581
 582
 583     copyright = textwrap.dedent("""\
 584         /* Autogenerated file, DO NOT EDIT manually! generated by {}
 585          *
 586          * Copyright (c) 2015 Intel Corporation
 587          *
 588          * Permission is hereby granted, free of charge, to any person obtaining a
 589          * copy of this software and associated documentation files (the "Software"),
 590          * to deal in the Software without restriction, including without limitation
 591          * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 592          * and/or sell copies of the Software, and to permit persons to whom the
 593          * Software is furnished to do so, subject to the following conditions:
 594          *
 595          * The above copyright notice and this permission notice (including the next
 596          * paragraph) shall be included in all copies or substantial portions of the
 597          * Software.
 598          *
 599          * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 600          * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 601          * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 602          * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 603          * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 604          * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 605          * DEALINGS IN THE SOFTWARE.
 606          */
 607
 608         """).format(os.path.basename(__file__))
 609
 610     h(copyright)
 611     h(textwrap.dedent("""\
 612         #pragma once
 613
 614         struct gen_perf_config;
 615
 616         """))
 617
 618     c(copyright)
 619     c(textwrap.dedent("""\
 620         #include <stdint.h>
 621         #include <stdbool.h>
 622
 623         #include <drm-uapi/i915_drm.h>
 624
 625         #include "util/hash_table.h"
 626
 627         """))
 628
 629     c("#include \"" + os.path.basename(args.header) + "\"")
 630
 631     c(textwrap.dedent("""\
 632         #include "perf/gen_perf.h"
 633
 634
 635         #define MIN(a, b) ((a < b) ? (a) : (b))
 636         #define MAX(a, b) ((a > b) ? (a) : (b))
 637
 638
 639         """))
 640
 641     # Print out all equation functions.
 642     for gen in gens:
 643         for set in gen.sets:
 644             for counter in set.counters:
 645                 output_counter_read(gen, set, counter)
 646                 output_counter_max(gen, set, counter)
 647
 648     # Print out all metric sets registration functions for each set in each
 649     # generation.
 650     for gen in gens:
 651         for set in gen.sets:
 652             counters = set.counters
 653
 654             c("\n")
 655             register_lengths = compute_register_lengths(set);
 656             for reg_type, reg_length in register_lengths.items():
 657                 c("static struct gen_perf_query_register_prog {0}_{1}_{2}[{3}];".format(gen.chipset,
 658                                                                                         set.underscore_name,
 659                                                                                         reg_type, reg_length))
 660
 661             c("\nstatic struct gen_perf_query_counter {0}_{1}_query_counters[{2}];\n".format(gen.chipset, set.underscore_name, len(counters)))
 662             c("static struct gen_perf_query_info " + gen.chipset + "_" + set.underscore_name + "_query = {\n")
 663             c_indent(3)
 664
 665             c(".kind = GEN_PERF_QUERY_TYPE_OA,\n")
 666             c(".name = \"" + set.name + "\",\n")
 667             c(".guid = \"" + set.hw_config_guid + "\",\n")
 668
 669             c(".counters = {0}_{1}_query_counters,".format(gen.chipset, set.underscore_name))
 670             c(".n_counters = 0,")
 671             c(".oa_metrics_set_id = 0, /* determined at runtime, via sysfs */")
 672
 673             if gen.chipset == "hsw":
 674                 c(textwrap.dedent("""\
 675                     .oa_format = I915_OA_FORMAT_A45_B8_C8,
 676
 677                     /* Accumulation buffer offsets... */
 678                     .gpu_time_offset = 0,
 679                     .a_offset = 1,
 680                     .b_offset = 46,
 681                     .c_offset = 54,
 682                 """))
 683             else:
 684                 c(textwrap.dedent("""\
 685                     .oa_format = I915_OA_FORMAT_A32u40_A4u32_B8_C8,
 686
 687                     /* Accumulation buffer offsets... */
 688                     .gpu_time_offset = 0,
 689                     .gpu_clock_offset = 1,
 690                     .a_offset = 2,
 691                     .b_offset = 38,
 692                     .c_offset = 46,
 693                 """))
 694
 695             for reg_type, reg_length in register_lengths.items():
 696                 c(".{0} = {1}_{2}_{3},".format(reg_type, gen.chipset, set.underscore_name, reg_type))
 697                 c(".n_{0} = 0, /* Determined at runtime */".format(reg_type))
 698
 699             c_outdent(3)
 700             c("};\n")
 701
 702             c("\nstatic void\n")
 703             c("{0}_register_{1}_counter_query(struct gen_perf_config *perf)\n".format(gen.chipset, set.underscore_name))
 704             c("{\n")
 705             c_indent(3)
 706
 707             c("static struct gen_perf_query_info *query = &" + gen.chipset + "_" + set.underscore_name + "_query;\n")
 708             c("struct gen_perf_query_counter *counter;\n")
 709
 710             c("\n")
 711             c("/* Note: we're assuming there can't be any variation in the definition ")
 712             c(" * of a query between contexts so it's ok to describe a query within a ")
 713             c(" * global variable which only needs to be initialized once... */")
 714             c("\nif (!query->data_size) {")
 715             c_indent(3)
 716
 717             generate_register_configs(set)
 718
 719             offset = 0
 720             for counter in counters:
 721                 offset = output_counter_report(set, counter, offset)
 722
 723
 724             c("\nquery->data_size = counter->offset + gen_perf_query_counter_get_size(counter);\n")
 725
 726             c_outdent(3)
 727             c("}");
 728
 729             c("\n_mesa_hash_table_insert(perf->oa_metrics_table, query->guid, query);")
 730
 731             c_outdent(3)
 732             c("}\n")
 733
 734         h("void gen_oa_register_queries_" + gen.chipset + "(struct gen_perf_config *perf);\n")
 735
 736         c("\nvoid")
 737         c("gen_oa_register_queries_" + gen.chipset + "(struct gen_perf_config *perf)")
 738         c("{")
 739         c_indent(3)
 740
 741         for set in gen.sets:
 742             c("{0}_register_{1}_counter_query(perf);".format(gen.chipset, set.underscore_name))
 743
 744         c_outdent(3)
 745         c("}")
 746
 747
 748 if __name__ == '__main__':
 749     main()