projects / libreriscv.git / blobdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | inline | side by side
(no commit message)
[libreriscv.git] / openpower / sv_analysis.py
diff --git a/openpower/sv_analysis.py b/openpower/sv_analysis.py
index 91a1dbb03cc3880ec247e4308d48c20fd67ea2dd..487602312b0680fd6820017bb1a15d2f147345e1 100644 (file)
--- a/openpower/sv_analysis.py
+++ b/openpower/sv_analysis.py
@@ -1,48 +1,108 @@
-#!/usr/bin/env python3
+#!/usr/bin/env python2
+#
+# NOTE that this program is python2 compatible, please do not stop it
+# from working by adding syntax that prevents that.
+#
+# Initial version written by lkcl Oct 2020
+# This program analyses the Power 9 op codes and looks at in/out register uses
+# The results are displayed:
+#      https://libre-soc.org/openpower/opcode_regs_deduped/
+#
+# It finds .csv files in the directory isatables/
+# then goes through the categories and creates svp64 CSV augmentation
+# tables on a per-opcode basis
+
 import csv
 import os
 from os.path import dirname, join
 from glob import glob
 from collections import OrderedDict
 
 import csv
 import os
 from os.path import dirname, join
 from glob import glob
 from collections import OrderedDict
 
-
+# Return absolute path (ie $PWD) + isatables + name
 def find_wiki_file(name):
     filedir = os.path.dirname(os.path.abspath(__file__))
     tabledir = join(filedir, 'isatables')
 def find_wiki_file(name):
     filedir = os.path.dirname(os.path.abspath(__file__))
     tabledir = join(filedir, 'isatables')
-    file_path = join(tabledir, name) 
+    file_path = join(tabledir, name)
     return file_path
 
     return file_path
 
+# Return an array of dictionaries from the CSV file name:
 def get_csv(name):
     file_path = find_wiki_file(name)
     with open(file_path, 'r') as csvfile:
         reader = csv.DictReader(csvfile)
         return list(reader)
 
 def get_csv(name):
     file_path = find_wiki_file(name)
     with open(file_path, 'r') as csvfile:
         reader = csv.DictReader(csvfile)
         return list(reader)
 
+# Write an array of dictionaries to the CSV file name:
+def write_csv(name, items, headers):
+    file_path = find_wiki_file(name)
+    with open(file_path, 'wb') as csvfile:
+        writer = csv.DictWriter(csvfile, headers, lineterminator="\n")
+        writer.writeheader()
+        writer.writerows(items)
+
+# This will return True if all values are true.
+# Not sure what this is about
 def blank_key(row):
 def blank_key(row):
-    for v in row.items():
+    #for v in row.values():
+    #    if 'SPR' in v: # skip all SPRs
+    #        return True
+    for v in row.values():
         if v:
             return False
     return True
 
         if v:
             return False
     return True
 
-keycolumns = ['in1', 'in2', 'in3', 'out', 'CR in', 'CR out',
-                 'ldst len', 'rc', 'lk']
+# General purpose registers have names like: RA, RT, R1, ...
+# Floating point registers names like: FRT, FRA, FR1, ..., FRTp, ...
+# Return True if field is a register
+def isreg(field):
+    return (field.startswith('R') or field.startswith('FR') or
+            field == 'SPR')
+
+
+# These are the attributes of the instructions,
+# register names
+keycolumns = ['unit', 'in1', 'in2', 'in3', 'out', 'CR in', 'CR out',
+                 ] # don't think we need these: 'ldst len', 'rc', 'lk']
+
+tablecols = ['unit', 'in', 'outcnt', 'CR in', 'CR out', 'imm'
+                 ] # don't think we need these: 'ldst len', 'rc', 'lk']
+
 def create_key(row):
     res = OrderedDict()
     #print ("row", row)
     for key in keycolumns:
 def create_key(row):
     res = OrderedDict()
     #print ("row", row)
     for key in keycolumns:
-        # registers
-        if key in ['in1', 'in2', 'in3', 'out']:
-            if row[key].startswith('R'):
-                res[key] = 'R'
-            else:
-                res[key] = '0'
-        # CRs
+        # registers IN - special-case: count number of regs RA/RB/RC/RS
+        if key in ['in1', 'in2', 'in3']:
+            if 'in' not in res:
+                res['in'] = 0
+            if isreg(row[key]):
+                res['in'] += 1
+
+        # registers OUT
+        if key == 'out':
+            # If upd is 1 then increment the count of outputs
+            if 'outcnt' not in res:
+                res['outcnt'] = 0
+            if isreg(row[key]):
+                res['outcnt'] += 1
+            if row['upd'] == '1':
+                res['outcnt'] += 1
+
+        # CRs (Condition Register) (CR0 .. CR7)
         if key.startswith('CR'):
             if row[key].startswith('NONE'):
                 res[key] = '0'
             else:
                 res[key] = '1'
         if key.startswith('CR'):
             if row[key].startswith('NONE'):
                 res[key] = '0'
             else:
                 res[key] = '1'
-        # LDST len
+            if row['comment'].startswith('cr'):
+                res['crop'] = '1'
+        # unit
+        if key == 'unit':
+            if row[key] == 'LDST': # we care about LDST units
+                res[key] = row[key]
+            else:
+                res[key] = 'OTHER'
+        # LDST len (LoadStore length)
         if key.startswith('ldst'):
             if row[key].startswith('NONE'):
                 res[key] = '0'
         if key.startswith('ldst'):
             if row[key].startswith('NONE'):
                 res[key] = '0'
@@ -59,8 +119,20 @@ def create_key(row):
         if key == 'lk':
             res[key] = row[key]
 
         if key == 'lk':
             res[key] = row[key]
 
+    # Convert the numerics 'in' & 'outcnt' to strings
+    res['in'] = str(res['in'])
+    res['outcnt'] = str(res['outcnt'])
+
+
+    # constants
+    if row['in2'].startswith('CONST_'):
+        res['imm'] = "1" # row['in2'].split("_")[1]
+    else:
+        res['imm'] = ''
+
     return res
 
     return res
 
+#
 def dformat(d):
     res = []
     for k, v in d.items():
 def dformat(d):
     res = []
     for k, v in d.items():
@@ -68,55 +140,425 @@ def dformat(d):
     return ' '.join(res)
 
 def tformat(d):
     return ' '.join(res)
 
 def tformat(d):
-    return "|" + ' | '.join(d) + "|"
+    return ' | '.join(d) + " |"
+
+def keyname(row):
+    res = []
+    if row['unit'] != 'OTHER':
+        res.append(row['unit'])
+    if row['in'] != '0':
+        res.append('%sR' % row['in'])
+    if row['outcnt'] != '0':
+        res.append('%sW' % row['outcnt'])
+    if row['CR in'] == '1' and row['CR out'] == '1':
+        if 'crop' in row:
+            res.append("CR=2R1W")
+        else:
+            res.append("CRio")
+    elif row['CR in'] == '1':
+        res.append("CRi")
+    elif row['CR out'] == '1':
+        res.append("CRo")
+    elif 'imm' in row and row['imm']:
+        res.append("imm")
+    return '-'.join(res)
 
 
 def process_csvs():
     csvs = {}
     bykey = {}
     primarykeys = set()
 
 
 def process_csvs():
     csvs = {}
     bykey = {}
     primarykeys = set()
-    dictkeys = {}
+    dictkeys = OrderedDict()
+    immediates = {}
+    insns = {} # dictionary of CSV row, by instruction
+
+    print ("# OpenPOWER ISA register 'profile's")
+    print ('')
+    print ("this page is auto-generated, do not edit")
+    print ("created by http://libre-soc.org/openpower/sv_analysis.py")
+    print ('')
 
 
+    # Expand that (all .csv files)
     pth = find_wiki_file("*.csv")
     pth = find_wiki_file("*.csv")
+
+    # Ignore those containing: valid test sprs
     for fname in glob(pth):
     for fname in glob(pth):
+        if '-' in fname:
+            continue
         if 'valid' in fname:
             continue
         if 'test' in fname:
             continue
         if 'valid' in fname:
             continue
         if 'test' in fname:
             continue
-        if 'sprs' in fname:
+        if fname.endswith('sprs.csv'):
+            continue
+        if fname.endswith('minor_19_valid.csv'):
+            continue
+        if 'RM' in fname:
             continue
         #print (fname)
         csvname = os.path.split(fname)[1]
             continue
         #print (fname)
         csvname = os.path.split(fname)[1]
+        # csvname is something like: minor_59.csv, fname the whole path
         csv = get_csv(fname)
         csvs[fname] = csv
         for row in csv:
             if blank_key(row):
                 continue
         csv = get_csv(fname)
         csvs[fname] = csv
         for row in csv:
             if blank_key(row):
                 continue
+            insn_name = row['comment']
+            # skip instructions that are not suitable
+            if insn_name in ['mcrxr', 'mcrxrx', 'darn']:
+                continue
+            if insn_name.startswith('bc') or 'rfid' in insn_name:
+                continue
+            insns[insn_name] = row # accumulate csv data by instruction
             dkey = create_key(row)
             key = tuple(dkey.values())
             dkey = create_key(row)
             key = tuple(dkey.values())
+            # print("key=", key)
             dictkeys[key] = dkey
             primarykeys.add(key)
             if key not in bykey:
                 bykey[key] = []
             dictkeys[key] = dkey
             primarykeys.add(key)
             if key not in bykey:
                 bykey[key] = []
-            bykey[key].append((csvname, row['opcode'], row['comment']))
+            bykey[key].append((csvname, row['opcode'], insn_name,
+                               row['form'].upper() + '-Form'))
+
+            # detect immediates, collate them (useful info)
+            if row['in2'].startswith('CONST_'):
+                imm = row['in2'].split("_")[1]
+                if key not in immediates:
+                    immediates[key] = set()
+                immediates[key].add(imm)
+
+    primarykeys = list(primarykeys)
+    primarykeys.sort()
+
+    # mapping to old SVPrefix "Forms"
+    mapsto = {'3R-1W-CRio': 'RM-1P-3S1D',
+              '2R-1W-CRio': 'RM-1P-2S1D',
+              '2R-1W-CRi': 'RM-1P-3S1D',
+              '2R-1W-CRo': 'RM-1P-2S1D',
+              '2R': 'non-SV',
+              '2R-1W': 'RM-1P-2S1D',
+              '1R-CRio': 'RM-2P-2S1D',
+              '2R-CRio': 'RM-1P-2S1D',
+              '2R-CRo': 'RM-1P-2S1D',
+              '1R': 'non-SV',
+              '1R-1W-CRio': 'RM-2P-1S1D',
+              '1R-1W-CRo': 'RM-2P-1S1D',
+              '1R-1W': 'RM-2P-1S1D',
+              '1R-1W-imm': 'RM-2P-1S1D',
+              '1R-CRo': 'RM-2P-1S1D',
+              '1R-imm': 'non-SV',
+              '1W': 'non-SV',
+              '1W-CRi': 'RM-2P-1S1D',
+              'CRio': 'RM-2P-1S1D',
+              'CR=2R1W': 'RM-1P-2S1D',
+              'CRi': 'non-SV',
+              'imm': 'non-SV',
+              '': 'non-SV',
+              'LDST-2R-imm': 'LDSTRM-2P-2S',
+              'LDST-2R-1W-imm': 'LDSTRM-2P-2S1D',
+              'LDST-2R-1W': 'LDSTRM-2P-2S1D',
+              'LDST-2R-2W': 'LDSTRM-2P-2S1D',
+              'LDST-1R-1W-imm': 'LDSTRM-2P-1S1D',
+              'LDST-1R-2W-imm': 'LDSTRM-2P-1S2D',
+              'LDST-3R': 'LDSTRM-2P-3S',
+              'LDST-3R-CRo': 'LDSTRM-2P-3S',  # st*x
+              'LDST-3R-1W': 'LDSTRM-2P-2S1D',  # st*x
+              }
+    print ("# map to old SV Prefix")
+    print ('')
+    print ('[[!table  data="""')
+    for key in primarykeys:
+        name = keyname(dictkeys[key])
+        value = mapsto.get(name, "-")
+        print (tformat([name, value+ " "]))
+    print ('"""]]')
+    print ('')
 
     print ("# keys")
 
     print ("# keys")
-    print ()
-    print (tformat(keycolumns))
+    print ('')
+    print ('[[!table  data="""')
+    print (tformat(tablecols) + " imms | name |")
+
+    # print out the keys and the table from which they're derived
     for key in primarykeys:
     for key in primarykeys:
-        print (tformat(dictkeys[key].values()))
-    print ()
+        name = keyname(dictkeys[key])
+        row = tformat(dictkeys[key].values())
+        imms = list(immediates.get(key, ""))
+        imms.sort()
+        row += " %s | " % ("/".join(imms))
+        row += " %s |" % name
+        print (row)
+    print ('"""]]')
+    print ('')
 
 
+    # print out, by remap name, all the instructions under that category
     for key in primarykeys:
     for key in primarykeys:
-        print ("## ", dformat(dictkeys[key]))
-        print ()
-        for row in bykey[key]:
-            print (" * ", row)
-        print ()
+        name = keyname(dictkeys[key])
+        value = mapsto.get(name, "-")
+        print ("## %s (%s)" % (name, value))
+        print ('')
+        print ('[[!table  data="""')
+        print (tformat(['CSV', 'opcode', 'asm', 'form']))
+        rows = bykey[key]
+        rows.sort()
+        for row in rows:
+            print (tformat(row))
+        print ('"""]]')
+        print ('')
 
 
-    bykey = {}
-    for fname, csv in csvs.items():
-        key
+    #for fname, csv in csvs.items():
+    #    print (fname)
+
+    #for insn, row in insns.items():
+    #    print (insn, row)
+
+    print ("# svp64 remaps")
+    svp64 = OrderedDict()
+    # create a CSV file, per category, with SV "augmentation" info
+    csvcols = ['insn', 'Ptype', 'Etype', '0', '1', '2', '3']
+    csvcols += ['in1', 'in2', 'in3', 'out', 'CR in', 'CR out'] # temporary
+    for key in primarykeys:
+        # get the decoded key containing row-analysis, and name/value
+        dkey = dictkeys[key]
+        name = keyname(dkey)
+        value = mapsto.get(name, "-")
+        if value == 'non-SV':
+            continue
+
+        # print out svp64 tables by category
+        print ("* **%s**: %s" % (name, value))
+
+        # store csv entries by svp64 RM category
+        if value not in svp64:
+            svp64[value] = []
+
+        rows = bykey[key]
+        rows.sort()
+
+        for row in rows:
+            #for idx in range(len(row)):
+            #    if row[idx] == 'NONE':
+            #        row[idx] = ''
+            # get the instruction
+            insn_name = row[2]
+            insn = insns[insn_name]
+            # start constructing svp64 CSV row
+            res = OrderedDict()
+            res['insn'] = insn_name
+            res['Ptype'] = value.split('-')[1] # predication type (RM-xN-xxx)
+            # get whether R_xxx_EXTRAn fields are 2-bit or 3-bit
+            res['Etype'] = 'EXTRA2'
+            # go through each register matching to Rxxxx_EXTRAx
+            for k in ['0', '1', '2', '3']:
+                res[k] = ''
+
+            # temporary useful info
+            regs = []
+            for k in ['in1', 'in2', 'in3', 'out', 'CR in', 'CR out']:
+                if insn[k].startswith('CONST'):
+                    res[k] = ''
+                    regs.append('')
+                else:
+                    res[k] = insn[k]
+                    if insn[k] == 'RA_OR_ZERO':
+                        regs.append('RA')
+                    elif insn[k] != 'NONE':
+                        regs.append(insn[k])
+                    else:
+                        regs.append('')
+
+            # sigh now the fun begins.  this isn't the sanest way to do it
+            # but the patterns are pretty regular.
+            if value == 'LDSTRM-2P-1S1D':
+                res['Etype'] = 'EXTRA3' # RM EXTRA3 type
+                res['0'] = 'd:RT' # RT: Rdest_EXTRA3
+                res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
+
+            elif value == 'LDSTRM-2P-1S2D':
+                res['Etype'] = 'EXTRA2' # RM EXTRA2 type
+                res['0'] = 'd:RT' # RT: Rdest1_EXTRA2
+                res['1'] = 'd:RA' # RA: Rdest2_EXTRA2
+                res['2'] = 's:RA' # RA: Rsrc1_EXTRA2
+
+            elif value == 'LDSTRM-2P-2S':
+                res['Etype'] = 'EXTRA3' # RM EXTRA2 type
+                res['0'] = 'd:RS' # RT: Rdest1_EXTRA2
+                res['1'] = 's:RA' # RA: Rsrc1_EXTRA2
+
+            elif value == 'LDSTRM-2P-2S1D':
+                if 'st' in insn_name and 'x' not in insn_name: # stwu/stbu etc
+                    res['Etype'] = 'EXTRA2' # RM EXTRA2 type
+                    res['0'] = 'd:RA' # RA: Rdest1_EXTRA2
+                    res['1'] = 's:RS' # RS: Rdsrc1_EXTRA2
+                    res['2'] = 's:RA' # RA: Rsrc2_EXTRA2
+                elif 'st' in insn_name and 'x' in insn_name: # stwux
+                    res['Etype'] = 'EXTRA2' # RM EXTRA2 type
+                    res['0'] = 'd:RA' # RA: Rdest1_EXTRA2
+                    res['1'] = 's:RS,s:RA' # RS: Rdest2_EXTRA2, RA: Rsrc1_EXTRA2
+                    res['2'] = 's:RB' # RB: Rsrc2_EXTRA2
+                elif 'u' in insn_name: # ldux etc.
+                    res['Etype'] = 'EXTRA2' # RM EXTRA2 type
+                    res['0'] = 'd:RT' # RT: Rdest1_EXTRA2
+                    res['1'] = 'd:RA' # RA: Rdest2_EXTRA2
+                    res['2'] = 's:RB' # RB: Rsrc1_EXTRA2
+                else:
+                    res['Etype'] = 'EXTRA2' # RM EXTRA2 type
+                    res['0'] = 'd:RT' # RT: Rdest1_EXTRA2
+                    res['1'] = 's:RA' # RA: Rsrc1_EXTRA2
+                    res['2'] = 's:RB' # RB: Rsrc2_EXTRA2
+
+            elif value == 'LDSTRM-2P-3S':
+                res['Etype'] = 'EXTRA2' # RM EXTRA2 type
+                if 'cx' in insn_name:
+                    res['0'] = 's:RS,d:CR0' # RS: Rsrc1_EXTRA2 CR0: dest
+                else:
+                    res['0'] = 's:RS' # RS: Rsrc1_EXTRA2
+                res['1'] = 's:RA' # RA: Rsrc2_EXTRA2
+                res['2'] = 's:RB' # RA: Rsrc3_EXTRA2
+
+            elif value == 'RM-2P-1S1D':
+                res['Etype'] = 'EXTRA3' # RM EXTRA3 type
+                if insn_name == 'mtspr':
+                    res['0'] = 'd:SPR' # SPR: Rdest1_EXTRA3
+                    res['1'] = 's:RS' # RS: Rsrc1_EXTRA3
+                elif insn_name == 'mfspr':
+                    res['0'] = 'd:RS' # RS: Rdest1_EXTRA3
+                    res['1'] = 's:SPR' # SPR: Rsrc1_EXTRA3
+                elif name == 'CRio' and insn_name == 'mcrf':
+                    res['0'] = 'd:BF' # BFA: Rdest1_EXTRA3
+                    res['1'] = 's:BFA' # BFA: Rsrc1_EXTRA3
+                elif 'mfcr' in insn_name or 'mfocrf' in insn_name:
+                    res['0'] = 'd:RT' # RT: Rdest1_EXTRA3
+                    res['1'] = 's:CR' # CR: Rsrc1_EXTRA3
+                elif insn_name == 'setb':
+                    res['0'] = 'd:RT' # RT: Rdest1_EXTRA3
+                    res['1'] = 's:BFA' # BFA: Rsrc1_EXTRA3
+                elif insn_name.startswith('cmp'): # cmpi
+                    res['0'] = 'd:BF' # BF: Rdest1_EXTRA3
+                    res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
+                elif regs == ['RA','','','RT','','']:
+                    res['0'] = 'd:RT' # RT: Rdest1_EXTRA3
+                    res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
+                elif regs == ['RA','','','RT','','CR0']:
+                    res['0'] = 'd:RT,d:CR0' # RT,CR0: Rdest1_EXTRA3
+                    res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
+                elif (regs == ['RS','','','RA','','CR0'] or
+                      regs == ['','','RS','RA','','CR0']):
+                    res['0'] = 'd:RA,d:CR0' # RA,CR0: Rdest1_EXTRA3
+                    res['1'] = 's:RS' # RS: Rsrc1_EXTRA3
+                elif regs == ['RS','','','RA','','']:
+                    res['0'] = 'd:RA' # RA: Rdest1_EXTRA3
+                    res['1'] = 's:RS' # RS: Rsrc1_EXTRA3
+                elif regs == ['','FRB','','FRT','0','CR1']:
+                    res['0'] = 'd:FRT,d:CR1' # FRT,CR1: Rdest1_EXTRA3
+                    res['1'] = 's:FRA' # FRA: Rsrc1_EXTRA3
+                elif regs == ['','FRB','','','','CR1']:
+                    res['0'] = 'd:CR1' # CR1: Rdest1_EXTRA3
+                    res['1'] = 's:FRB' # FRA: Rsrc1_EXTRA3
+                elif regs == ['','FRB','','','','BF']:
+                    res['0'] = 'd:BF' # BF: Rdest1_EXTRA3
+                    res['1'] = 's:FRB' # FRA: Rsrc1_EXTRA3
+                elif regs == ['','FRB','','FRT','','CR1']:
+                    res['0'] = 'd:FRT,d:CR1' # FRT,CR1: Rdest1_EXTRA3
+                    res['1'] = 's:FRB' # FRB: Rsrc1_EXTRA3
+                else:
+                    res['0'] = 'TODO'
+
+            elif value == 'RM-1P-2S1D':
+                res['Etype'] = 'EXTRA3' # RM EXTRA3 type
+                if insn_name.startswith('cr'):
+                    res['0'] = 'd:BT' # BT: Rdest1_EXTRA3
+                    res['1'] = 's:BA' # BA: Rsrc1_EXTRA3
+                    res['2'] = 's:BB' # BB: Rsrc2_EXTRA3
+                elif regs == ['FRA','','FRC','FRT','','CR1']:
+                    res['0'] = 'd:FRT,d:CR1' # FRT,CR1: Rdest1_EXTRA3
+                    res['1'] = 's:FRA' # FRA: Rsrc1_EXTRA3
+                    res['2'] = 's:FRC' # FRC: Rsrc1_EXTRA3
+                # should be for fcmp
+                elif regs == ['FRA','FRB','','','','BF']:
+                    res['0'] = 'd:BF' # BF: Rdest1_EXTRA3
+                    res['1'] = 's:FRA' # FRA: Rsrc1_EXTRA3
+                    res['2'] = 's:FRB' # FRB: Rsrc1_EXTRA3
+                elif regs == ['FRA','FRB','','FRT','','']:
+                    res['0'] = 'd:FRT' # FRT: Rdest1_EXTRA3
+                    res['1'] = 's:FRA' # FRA: Rsrc1_EXTRA3
+                    res['2'] = 's:FRB' # FRB: Rsrc1_EXTRA3
+                elif regs == ['FRA','FRB','','FRT','','CR1']:
+                    res['0'] = 'd:FRT,d:CR1' # FRT,CR1: Rdest1_EXTRA3
+                    res['1'] = 's:FRA' # FRA: Rsrc1_EXTRA3
+                    res['2'] = 's:FRB' # FRB: Rsrc1_EXTRA3
+                elif name == '2R-1W' or insn_name == 'cmpb': # cmpb
+                    if insn_name in ['bpermd', 'cmpb']:
+                        res['0'] = 'd:RA' # RA: Rdest1_EXTRA3
+                        res['1'] = 's:RS' # RS: Rsrc1_EXTRA3
+                    else:
+                        res['0'] = 'd:RT' # RT: Rdest1_EXTRA3
+                        res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
+                    res['2'] = 's:RB' # RB: Rsrc1_EXTRA3
+                elif insn_name.startswith('cmp'): # cmp
+                    res['0'] = 'd:BF' # BF: Rdest1_EXTRA3
+                    res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
+                    res['2'] = 's:RB' # RB: Rsrc1_EXTRA3
+                elif (regs == ['','RB','RS','RA','','CR0'] or
+                      regs == ['RS','RB','','RA','','CR0']):
+                    res['0'] = 'd:RA,d:CR0' # RA,CR0: Rdest1_EXTRA3
+                    res['1'] = 's:RB' # RB: Rsrc1_EXTRA3
+                    res['2'] = 's:RS' # RS: Rsrc1_EXTRA3
+                elif regs == ['RA','RB','','RT','','CR0']:
+                    res['0'] = 'd:RT,d:CR0' # RT,CR0: Rdest1_EXTRA3
+                    res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
+                    res['2'] = 's:RB' # RB: Rsrc1_EXTRA3
+                elif regs == ['RA','','RS','RA','','CR0']:
+                    res['0'] = 'd:RA,d:CR0' # RA,CR0: Rdest1_EXTRA3
+                    res['1'] = 's:RA' # RA: Rsrc1_EXTRA3
+                    res['2'] = 's:RS' # RS: Rsrc1_EXTRA3
+                else:
+                    res['0'] = 'TODO'
+
+            elif value == 'RM-2P-2S1D':
+                res['Etype'] = 'EXTRA2' # RM EXTRA2 type
+                if insn_name.startswith('mt'): # mtcrf
+                    res['0'] = 'd:CR' # CR: Rdest1_EXTRA2
+                    res['1'] = 's:RS' # RS: Rsrc1_EXTRA2
+                    res['2'] = 's:CR' # CR: Rsrc2_EXTRA2
+                else:
+                    res['0'] = 'TODO'
+
+            elif value == 'RM-1P-3S1D':
+                res['Etype'] = 'EXTRA2' # RM EXTRA2 type
+                if insn_name == 'isel':
+                    res['0'] = 'd:RT' # RT: Rdest1_EXTRA2
+                    res['1'] = 's:RA' # RA: Rsrc1_EXTRA2
+                    res['2'] = 's:RB' # RT: Rsrc2_EXTRA2
+                    res['3'] = 's:BC' # BC: Rsrc3_EXTRA2
+                else:
+                    res['0'] = 'd:FRT,d:CR1' # FRT, CR1: Rdest1_EXTRA2
+                    res['1'] = 's:FRA' # FRA: Rsrc1_EXTRA2
+                    res['2'] = 's:FRB' # FRB: Rsrc2_EXTRA2
+                    res['3'] = 's:FRC' # FRC: Rsrc3_EXTRA2
+
+            # add to svp64 csvs
+            #for k in ['in1', 'in2', 'in3', 'out', 'CR in', 'CR out']:
+            #    del res[k]
+            #if res['0'] != 'TODO':
+            for k in res:
+                if res[k] == 'NONE' or res[k] == '':
+                    res[k] = '0'
+            svp64[value].append(res)
+
+    print ('')
+
+    # now write out the csv files
+    for value, csv in svp64.items():
+        # print out svp64 tables by category
+        print ("## %s" % value)
+        print ('')
+        print ('[[!table format=csv file="openpower/isatables/%s.csv"]]' % \
+                    value)
+        print ('')
+
+        #csvcols = ['insn', 'Ptype', 'Etype', '0', '1', '2', '3']
+        write_csv("%s.csv" % value, csv, csvcols)
 
 if __name__ == '__main__':
     process_csvs()
 
 if __name__ == '__main__':
     process_csvs()
http://libre-riscv.org ikiwiki