openpower/sv_analysis.py

   1 #!/usr/bin/env python2
   2 #
   3 # NOTE that this program is python2 compatible, please do not stop it
   4 # from working by adding syntax that prevents that.
   5 #
   6 # Initial version written by lkcl Oct 2020
   7 # This program analyses the Power 9 op codes and looks at in/out register uses
   8 # The results are displayed:
   9 #       https://libre-soc.org/openpower/opcode_regs_deduped/
  10 #
  11 # It finds .csv files in the directory isatables/
  12 # then goes through the categories and creates svp64 CSV augmentation
  13 # tables on a per-opcode basis
  14
  15 import csv
  16 import os
  17 from os.path import dirname, join
  18 from glob import glob
  19 from collections import OrderedDict
  20
  21 # Return absolute path (ie $PWD) + isatables + name
  22 def find_wiki_file(name):
  23     filedir = os.path.dirname(os.path.abspath(__file__))
  24     tabledir = join(filedir, 'isatables')
  25     file_path = join(tabledir, name)
  26     return file_path
  27
  28 # Return an array of dictionaries from the CSV file name:
  29 def get_csv(name):
  30     file_path = find_wiki_file(name)
  31     with open(file_path, 'r') as csvfile:
  32         reader = csv.DictReader(csvfile)
  33         return list(reader)
  34
  35 # Write an array of dictionaries to the CSV file name:
  36 def write_csv(name, items, headers):
  37     file_path = find_wiki_file(name)
  38     with open(file_path, 'w') as csvfile:
  39         writer = csv.DictWriter(csvfile, fieldnames)
  40         writer.writeheader()
  41         writer.writerows(items)
  42
  43 # This will return True if all values are true.
  44 # Not sure what this is about
  45 def blank_key(row):
  46     #for v in row.values():
  47     #    if 'SPR' in v: # skip all SPRs
  48     #        return True
  49     for v in row.values():
  50         if v:
  51             return False
  52     return True
  53
  54 # General purpose registers have names like: RA, RT, R1, ...
  55 # Floating point registers names like: FRT, FRA, FR1, ..., FRTp, ...
  56 # Return True if field is a register
  57 def isreg(field):
  58     return field.startswith('R') or field.startswith('FR')
  59
  60
  61 # These are the attributes of the instructions,
  62 # register names
  63 keycolumns = ['unit', 'in1', 'in2', 'in3', 'out', 'CR in', 'CR out',
  64                  ] # don't think we need these: 'ldst len', 'rc', 'lk']
  65
  66 tablecols = ['unit', 'in', 'outcnt', 'CR in', 'CR out', 'imm'
  67                  ] # don't think we need these: 'ldst len', 'rc', 'lk']
  68
  69 def create_key(row):
  70     res = OrderedDict()
  71     #print ("row", row)
  72     for key in keycolumns:
  73         # registers IN - special-case: count number of regs RA/RB/RC/RS
  74         if key in ['in1', 'in2', 'in3']:
  75             if 'in' not in res:
  76                 res['in'] = 0
  77             if isreg(row[key]):
  78                 res['in'] += 1
  79
  80         # registers OUT
  81         if key == 'out':
  82             # If upd is 1 then increment the count of outputs
  83             if 'outcnt' not in res:
  84                 res['outcnt'] = 0
  85             if isreg(row[key]):
  86                 res['outcnt'] += 1
  87             if row['upd'] == '1':
  88                 res['outcnt'] += 1
  89
  90         # CRs (Condition Register) (CR0 .. CR7)
  91         if key.startswith('CR'):
  92             if row[key].startswith('NONE'):
  93                 res[key] = '0'
  94             else:
  95                 res[key] = '1'
  96             if row['comment'].startswith('cr'):
  97                 res['crop'] = '1'
  98         # unit
  99         if key == 'unit':
 100             if row[key] == 'LDST': # we care about LDST units
 101                 res[key] = row[key]
 102             else:
 103                 res[key] = 'OTHER'
 104         # LDST len (LoadStore length)
 105         if key.startswith('ldst'):
 106             if row[key].startswith('NONE'):
 107                 res[key] = '0'
 108             else:
 109                 res[key] = '1'
 110         # rc, lk
 111         if key in ['rc', 'lk']:
 112             if row[key] == 'ONE':
 113                 res[key] = '1'
 114             elif row[key] == 'NONE':
 115                 res[key] = '0'
 116             else:
 117                 res[key] = 'R'
 118         if key == 'lk':
 119             res[key] = row[key]
 120
 121     # Convert the numerics 'in' & 'outcnt' to strings
 122     res['in'] = str(res['in'])
 123     res['outcnt'] = str(res['outcnt'])
 124
 125
 126     # constants
 127     if row['in2'].startswith('CONST_'):
 128         res['imm'] = "1" # row['in2'].split("_")[1]
 129     else:
 130         res['imm'] = ''
 131
 132     return res
 133
 134 #
 135 def dformat(d):
 136     res = []
 137     for k, v in d.items():
 138         res.append("%s: %s" % (k, v))
 139     return ' '.join(res)
 140
 141 def tformat(d):
 142     return ' | '.join(d) + " |"
 143
 144 def keyname(row):
 145     res = []
 146     if row['unit'] != 'OTHER':
 147         res.append(row['unit'])
 148     if row['in'] != '0':
 149         res.append('%sR' % row['in'])
 150     if row['outcnt'] != '0':
 151         res.append('%sW' % row['outcnt'])
 152     if row['CR in'] == '1' and row['CR out'] == '1':
 153         if 'crop' in row:
 154             res.append("CR=2R1W")
 155         else:
 156             res.append("CRio")
 157     elif row['CR in'] == '1':
 158         res.append("CRi")
 159     elif row['CR out'] == '1':
 160         res.append("CRo")
 161     elif 'imm' in row and row['imm']:
 162         res.append("imm")
 163     return '-'.join(res)
 164
 165
 166 def process_csvs():
 167     csvs = {}
 168     bykey = {}
 169     primarykeys = set()
 170     dictkeys = OrderedDict()
 171     immediates = {}
 172     insns = {} # dictionary of CSV row, by instruction
 173
 174     print ("# OpenPOWER ISA register 'profile's")
 175     print ('')
 176     print ("this page is auto-generated, do not edit")
 177     print ("created by http://libre-soc.org/openpower/sv_analysis.py")
 178     print ('')
 179
 180     # Expand that (all .csv files)
 181     pth = find_wiki_file("*.csv")
 182
 183     # Ignore those containing: valid test sprs
 184     for fname in glob(pth):
 185         if 'valid' in fname:
 186             continue
 187         if 'test' in fname:
 188             continue
 189         if 'sprs' in fname:
 190             continue
 191
 192         #print (fname)
 193         csvname = os.path.split(fname)[1]
 194         # csvname is something like: minor_59.csv, fname the whole path
 195         csv = get_csv(fname)
 196         csvs[fname] = csv
 197         for row in csv:
 198             if blank_key(row):
 199                 continue
 200             insn_name = row['comment']
 201             # skip instructions that are not suitable
 202             if insn_name in ['mcrxr', 'mcrxrx', 'darn']:
 203                 continue
 204             insns[insn_name] = row # accumulate csv data by instruction
 205             dkey = create_key(row)
 206             key = tuple(dkey.values())
 207             # print("key=", key)
 208             dictkeys[key] = dkey
 209             primarykeys.add(key)
 210             if key not in bykey:
 211                 bykey[key] = []
 212             bykey[key].append((csvname, row['opcode'], insn_name,
 213                                row['form'].upper() + '-Form'))
 214
 215             # detect immediates, collate them (useful info)
 216             if row['in2'].startswith('CONST_'):
 217                 imm = row['in2'].split("_")[1]
 218                 if key not in immediates:
 219                     immediates[key] = set()
 220                 immediates[key].add(imm)
 221
 222     primarykeys = list(primarykeys)
 223     primarykeys.sort()
 224
 225     # mapping to old SVPrefix "Forms"
 226     mapsto = {'3R-1W-CRio': 'RM-1P-3S1D',
 227               '2R-1W-CRio': 'RM-1P-2S1D',
 228               '2R-1W-CRi': 'RM-1P-3S1D',
 229               '2R-1W-CRo': 'RM-1P-2S1D',
 230               '2R': 'non-SV',
 231               '2R-1W': 'RM-1P-2S1D',
 232               '1R-CRio': 'RM-2P-2S1D',
 233               '2R-CRio': 'RM-1P-2S1D',
 234               '2R-CRo': 'RM-1P-2S1D',
 235               '1R': 'non-SV',
 236               '1R-1W-CRio': 'RM-2P-1S1D',
 237               '1R-1W-CRo': 'RM-2P-1S1D',
 238               '1R-1W': 'RM-2P-1S1D',
 239               '1R-1W-imm': 'RM-2P-1S1D',
 240               '1R-CRo': 'RM-2P-1S1D',
 241               '1R-imm': 'non-SV',
 242               '1W': 'non-SV',
 243               '1W-CRi': 'RM-2P-1S1D',
 244               'CRio': 'RM-2P-1S1D',
 245               'CR=2R1W': 'RM-1P-2S1D',
 246               'CRi': 'non-SV',
 247               'imm': 'non-SV',
 248               '': 'non-SV',
 249               'LDST-2R-imm': 'LDSTRM-2P-2S',
 250               'LDST-2R-1W-imm': 'LDSTRM-2P-2S1D',
 251               'LDST-2R-1W': 'LDSTRM-2P-2S1D',
 252               'LDST-2R-2W': 'LDSTRM-2P-2S1D',
 253               'LDST-1R-1W-imm': 'LDSTRM-2P-1S1D',
 254               'LDST-1R-2W-imm': 'LDSTRM-2P-1S2D',
 255               'LDST-3R': 'LDSTRM-2P-3S',
 256               'LDST-3R-CRo': 'LDSTRM-2P-3S',  # st*x
 257               'LDST-3R-1W': 'LDSTRM-2P-2S1D',  # st*x
 258               }
 259     print ("# map to old SV Prefix")
 260     print ('')
 261     print ('[[!table  data="""')
 262     for key in primarykeys:
 263         name = keyname(dictkeys[key])
 264         value = mapsto.get(name, "-")
 265         print (tformat([name, value+ " "]))
 266     print ('"""]]')
 267     print ('')
 268
 269     print ("# keys")
 270     print ('')
 271     print ('[[!table  data="""')
 272     print (tformat(tablecols) + " imms | name |")
 273
 274     # print out the keys and the table from which they're derived
 275     for key in primarykeys:
 276         name = keyname(dictkeys[key])
 277         row = tformat(dictkeys[key].values())
 278         imms = list(immediates.get(key, ""))
 279         imms.sort()
 280         row += " %s | " % ("/".join(imms))
 281         row += " %s |" % name
 282         print (row)
 283     print ('"""]]')
 284     print ('')
 285
 286     # print out, by remap name, all the instructions under that category
 287     for key in primarykeys:
 288         name = keyname(dictkeys[key])
 289         value = mapsto.get(name, "-")
 290         print ("## %s (%s)" % (name, value))
 291         print ('')
 292         print ('[[!table  data="""')
 293         print (tformat(['CSV', 'opcode', 'asm', 'form']))
 294         rows = bykey[key]
 295         rows.sort()
 296         for row in rows:
 297             print (tformat(row))
 298         print ('"""]]')
 299         print ('')
 300
 301     #for fname, csv in csvs.items():
 302     #    print (fname)
 303
 304     #for insn, row in insns.items():
 305     #    print (insn, row)
 306
 307     print ("# svp64 remaps")
 308     svp64 = OrderedDict()
 309     # create a CSV file, per category, with SV "augmentation" info
 310     csvcols = ['insn', 'Ptype', 'Etype', '0', '1', '2', '3']
 311     for key in primarykeys:
 312         # get the decoded key containing row-analysis, and name/value
 313         dkey = dictkeys[key]
 314         name = keyname(dkey)
 315         value = mapsto.get(name, "-")
 316         if value == 'non-SV':
 317             continue
 318
 319         # store csv entries by svp64 RM category
 320         if value not in svp64:
 321             svp64[value] = []
 322
 323         # print out svp64 tables by category
 324         print ("## %s (%s)" % (name, value))
 325         print ('')
 326         print ('[[!table  data="""')
 327         print (tformat(csvcols))
 328         rows = bykey[key]
 329         rows.sort()
 330
 331         for row in rows:
 332             #for idx in range(len(row)):
 333             #    if row[idx] == 'NONE':
 334             #        row[idx] = ''
 335             # get the instruction
 336             insn_name = row[2]
 337             insn = insns[insn_name]
 338             # start constructing svp64 CSV row
 339             res = OrderedDict()
 340             res['insn'] = insn_name
 341             res['Ptype'] = value.split('-')[1] # predication type (RM-xN-xxx)
 342             # get whether R_xxx_EXTRAn fields are 2-bit or 3-bit
 343             res['Etype'] = 'EXTRA2'
 344             # go through each register matching to Rxxxx_EXTRAx
 345             for k in ['0', '1', '2', '3']:
 346                 res[k] = ''
 347
 348             # temporary useful info
 349             regs = []
 350             for k in ['in1', 'in2', 'in3', 'out', 'CR in', 'CR out']:
 351                 if insn[k].startswith('CONST'):
 352                     res[k] = ''
 353                     regs.append('')
 354                 else:
 355                     res[k] = insn[k]
 356                     if insn[k] == 'RA_OR_ZERO':
 357                         regs.append('RA')
 358                     elif insn[k] != 'NONE':
 359                         regs.append(insn[k])
 360                     else:
 361                         regs.append('')
 362
 363
 364             # sigh now the fun begins.  this isn't the sanest way to do it
 365             # but the patterns are pretty regular.
 366             if value == 'LDSTRM-2P-1S1D':
 367                 res['Etype'] = 'EXTRA3' # RM EXTRA3 type
 368                 res['0'] = 'd:RT' # RT: Rdest_EXTRA3
 369                 res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
 370
 371             elif value == 'LDSTRM-2P-1S2D':
 372                 res['Etype'] = 'EXTRA2' # RM EXTRA2 type
 373                 res['0'] = 'd:RT' # RT: Rdest1_EXTRA2
 374                 res['1'] = 's:RA' # RA: Rsrc1_EXTRA2
 375                 res['2'] = 'd:RA' # RA: Rdest2_EXTRA2
 376
 377             elif value == 'LDSTRM-2P-2S':
 378                 res['Etype'] = 'EXTRA3' # RM EXTRA2 type
 379                 res['0'] = 'd:RS' # RT: Rdest1_EXTRA2
 380                 res['1'] = 's:RA' # RA: Rsrc1_EXTRA2
 381
 382             elif value == 'LDSTRM-2P-2S1D':
 383                 if 'st' in insn_name and 'x' not in insn_name: # stwu/stbu etc
 384                     res['Etype'] = 'EXTRA2' # RM EXTRA2 type
 385                     res['0'] = 'd:RS' # RS: Rdest1_EXTRA2
 386                     res['1'] = 'd:RA' # RA: Rdest2_EXTRA2
 387                     res['2'] = 's:RA' # RA: Rsrc1_EXTRA2
 388                 if 'st' in insn_name and 'x' in insn_name: # stwux
 389                     res['Etype'] = 'EXTRA2' # RM EXTRA2 type
 390                     res['0'] = 'd:RS' # RS: Rdest1_EXTRA2
 391                     res['1'] = 'd:RA' # RA: Rdest2_EXTRA2, RA: Rsrc1_EXTRA2
 392                     res['2'] = 's:RB' # RB: Rsrc2_EXTRA2
 393                 elif 'u' in insn_name: # ldux etc.
 394                     res['Etype'] = 'EXTRA2' # RM EXTRA2 type
 395                     res['0'] = 'd:RT' # RT: Rdest1_EXTRA2
 396                     res['1'] = 'd:RA' # RA: Rdest2_EXTRA2
 397                     res['2'] = 's:RB' # RB: Rsrc1_EXTRA2
 398                 else:
 399                     res['Etype'] = 'EXTRA2' # RM EXTRA2 type
 400                     res['0'] = 'd:RT' # RT: Rdest1_EXTRA2
 401                     res['1'] = 's:RA' # RA: Rsrc1_EXTRA2
 402                     res['2'] = 's:RB' # RB: Rsrc2_EXTRA2
 403
 404             elif value == 'LDSTRM-2P-3S':
 405                 res['Etype'] = 'EXTRA2' # RM EXTRA2 type
 406                 res['0'] = 's:RS,d:CR0' # RS: Rsrc1_EXTRA2 CR0: dest
 407                 res['1'] = 's:RA' # RA: Rsrc2_EXTRA2
 408                 res['2'] = 's:RB' # RA: Rsrc3_EXTRA2
 409
 410             elif value == 'RM-2P-1S1D':
 411                 res['Etype'] = 'EXTRA3' # RM EXTRA3 type
 412                 if name == 'CRio' and insn_name == 'mcrf':
 413                     res['0'] = 'd:BF' # BFA: Rdest1_EXTRA3
 414                     res['1'] = 's:BFA' # BFA: Rsrc1_EXTRA3
 415                 elif 'mfcr' in insn_name or 'mfocrf' in insn_name:
 416                     res['0'] = 'd:RT' # RT: Rdest1_EXTRA3
 417                     res['1'] = 's:CR' # CR: Rsrc1_EXTRA3
 418                 elif insn_name == 'setb':
 419                     res['0'] = 'd:RT' # RT: Rdest1_EXTRA3
 420                     res['1'] = 's:BFA' # BFA: Rsrc1_EXTRA3
 421                 elif insn_name.startswith('cmp'): # cmpi
 422                     res['0'] = 'd:BF' # BF: Rdest1_EXTRA3
 423                     res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
 424                 elif regs == ['RA','','','RT','','']:
 425                     res['0'] = 'd:RT' # RT: Rdest1_EXTRA3
 426                     res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
 427                 elif regs == ['RA','','','RT','','CR0']:
 428                     res['0'] = 'd:RT,d:CR0' # RT,CR0: Rdest1_EXTRA3
 429                     res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
 430                 elif (regs == ['RS','','','RA','','CR0'] or
 431                       regs == ['','','RS','RA','','CR0']):
 432                     res['0'] = 'd:RA,d:CR0' # RA,CR0: Rdest1_EXTRA3
 433                     res['1'] = 's:RS' # RS: Rsrc1_EXTRA3
 434                 elif regs == ['RS','','','RA','','']:
 435                     res['0'] = 'd:RA' # RA: Rdest1_EXTRA3
 436                     res['1'] = 's:RS' # RS: Rsrc1_EXTRA3
 437                 else:
 438                     res['0'] = 'TODO'
 439
 440             elif value == 'RM-1P-2S1D':
 441                 res['Etype'] = 'EXTRA3' # RM EXTRA3 type
 442                 if insn_name.startswith('cr'):
 443                     res['0'] = 'd:BT' # BT: Rdest1_EXTRA3
 444                     res['1'] = 's:BA' # BA: Rsrc1_EXTRA3
 445                     res['2'] = 's:BB' # BB: Rsrc2_EXTRA3
 446                 elif name == '2R-1W' or insn_name == 'cmpb': # cmpb
 447                     if insn_name in ['bpermd', 'cmpb']:
 448                         res['0'] = 'd:RA' # RA: Rdest1_EXTRA3
 449                         res['1'] = 's:RS' # RS: Rsrc1_EXTRA3
 450                     else:
 451                         res['0'] = 'd:RT' # RT: Rdest1_EXTRA3
 452                         res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
 453                     res['2'] = 's:RB' # RB: Rsrc1_EXTRA3
 454                 elif insn_name.startswith('cmp'): # cmp
 455                     res['0'] = 'd:BF' # BF: Rdest1_EXTRA3
 456                     res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
 457                     res['2'] = 's:RB' # RB: Rsrc1_EXTRA3
 458                 elif (regs == ['','RB','RS','RA','','CR0'] or
 459                       regs == ['RS','RB','','RA','','CR0']):
 460                     res['0'] = 'd:RA,d:CR0' # RA,CR0: Rdest1_EXTRA3
 461                     res['1'] = 's:RB' # RB: Rsrc1_EXTRA3
 462                     res['2'] = 's:RS' # RS: Rsrc1_EXTRA3
 463                 elif regs == ['RA','RB','','RT','','CR0']:
 464                     res['0'] = 'd:RT,d:CR0' # RT,CR0: Rdest1_EXTRA3
 465                     res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
 466                     res['2'] = 's:RB' # RB: Rsrc1_EXTRA3
 467                 elif regs == ['RA','','RS','RA','','CR0']:
 468                     res['0'] = 'd:RA,d:CR0' # RA,CR0: Rdest1_EXTRA3
 469                     res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
 470                     res['2'] = 's:RS' # RS: Rsrc1_EXTRA3
 471                 else:
 472                     res['0'] = 'TODO'
 473
 474             elif value == 'RM-2P-2S1D':
 475                 res['Etype'] = 'EXTRA2' # RM EXTRA2 type
 476                 if insn_name.startswith('mt'): # mtcrf
 477                     res['0'] = 'd:CR' # CR: Rdest1_EXTRA2
 478                     res['1'] = 's:RS' # RS: Rsrc1_EXTRA2
 479                     res['2'] = 's:CR' # CR: Rsrc2_EXTRA2
 480                 else:
 481                     res['0'] = 'TODO'
 482
 483             elif value == 'RM-1P-3S1D':
 484                 res['Etype'] = 'EXTRA2' # RM EXTRA2 type
 485                 if insn_name == 'isel':
 486                     res['0'] = 'd:RT' # RT: Rdest1_EXTRA2
 487                     res['1'] = 's:RA' # RA: Rsrc1_EXTRA2
 488                     res['2'] = 's:RB' # RT: Rsrc2_EXTRA2
 489                     res['3'] = 's:BC' # BC: Rsrc3_EXTRA2
 490                 else:
 491                     res['0'] = 'd:FRT,d:CR1' # FRT, CR1: Rdest1_EXTRA2
 492                     res['1'] = 's:FRA' # FRA: Rsrc1_EXTRA2
 493                     res['2'] = 's:FRB' # FRB: Rsrc2_EXTRA2
 494                     res['3'] = 's:FRC' # FRC: Rsrc3_EXTRA2
 495
 496
 497             # print out the row
 498             print (tformat(res.values()))
 499             # add to svp64 csvs
 500             svp64[value].append(res)
 501
 502         print ('"""]]')
 503         print ('')
 504
 505 if __name__ == '__main__':
 506     process_csvs()