litex/litescope/host/dump.py

452 lines
8.4 KiB
Python

import sys
import os
import math
import shutil
import datetime
import zipfile
import re
from collections import OrderedDict
def dec2bin(d, nb=0):
if d=="x":
return "x"*nb
elif d==0:
b="0"
else:
b=""
while d!=0:
b="01"[d&1]+b
d=d>>1
return b.zfill(nb)
def get_bits(values, width, low, high=None):
r = []
for val in values:
t = dec2bin(val, width)[::-1]
if high == None:
t = t[low]
else:
t = t[low:high]
t = t[::-1]
t = int(t,2)
r.append(t)
return r
class Dat(list):
def __init__(self, width):
self.width = width
def __getitem__(self, key):
if isinstance(key, int):
return get_bits(self, self.width, key)
elif isinstance(key, slice):
if key.start != None:
start = key.start
else:
start = 0
if key.stop != None:
stop = key.stop
else:
stop = self.width
if stop > self.width:
stop = self.width
if key.step != None:
raise KeyError
return get_bits(self, self.width, start, stop)
else:
raise KeyError
def decode_rle(self):
rle_bit = self[-1]
rle_dat = self[:self.width-1]
dat = Dat(self.width)
i=0
last = 0
for d in self:
if rle_bit[i]:
if len(dat) >= 1:
# FIX ME... why is rle_dat in reverse order...
for j in range(int(dec2bin(rle_dat[i])[::-1],2)):
dat.append(last)
else:
dat.append(d)
last = d
i +=1
return dat
class Var:
def __init__(self, name, width, values=[], type="wire", default="x"):
self.type = type
self.width = width
self.name = name
self.val = default
self.values = values
self.vcd_id = None
def set_vcd_id(self, s):
self.vcd_id = s
def __len__(self):
return len(self.values)
def change(self, cnt):
r = ""
try :
if self.values[cnt+1] != self.val:
r += "b"
r += dec2bin(self.values[cnt+1], self.width)
r += " "
r += self.vcd_id
r += "\n"
return r
except :
return r
return r
class Dump:
def __init__(self):
self.vars = []
self.vcd_id = "!"
def add(self, var):
var.set_vcd_id(self.vcd_id)
self.vcd_id = chr(ord(self.vcd_id)+1)
self.vars.append(var)
def add_from_layout(self, layout, var):
i=0
for s, n in layout:
self.add(Var(s, n, var[i:i+n]))
i += n
def __len__(self):
l = 0
for var in self.vars:
l = max(len(var),l)
return l
class VCDExport():
def __init__(self, dump, timescale="1ps", comment=""):
self.dump = dump
self.timescale = timescale
self.comment = comment
self.cnt = -1
def change(self):
r = ""
c = ""
for var in self.dump.vars:
c += var.change(self.cnt)
if c != "":
r += "#"
r += str(self.cnt+1)
r += "\n"
r += c
return r
def p_date(self):
now = datetime.datetime.now()
r = "$date\n"
r += "\t"
r += now.strftime("%Y-%m-%d %H:%M")
r += "\n"
r += "$end\n"
return r
def p_version(self):
r = "$version\n"
r += "\tmiscope VCD dump\n"
r += "$end\n"
return r
def p_comment(self):
r = "$comment\n"
r += self.comment
r += "\n$end\n"
return r
def p_timescale(self):
r = "$timescale "
r += self.timescale
r += " $end\n"
return r
def p_scope(self):
r = "$scope "
r += self.timescale
r += " $end\n"
return r
def p_vars(self):
r = ""
for var in self.dump.vars:
r += "$var "
r += var.type
r += " "
r += str(var.width)
r += " "
r += var.vcd_id
r += " "
r += var.name
r += " $end\n"
return r
def p_unscope(self):
r = "$unscope "
r += " $end\n"
return r
def p_enddefinitions(self):
r = "$enddefinitions "
r += " $end\n"
return r
def p_dumpvars(self):
r = "$dumpvars\n"
for var in self.dump.vars:
r += "b"
r += dec2bin(var.val, var.width)
r += " "
r += var.vcd_id
r+= "\n"
r += "$end\n"
return r
def p_valuechange(self):
r = ""
for i in range(len(self.dump)):
r += self.change()
self.cnt += 1
return r
def __repr__(self):
r = ""
r += self.p_date()
r += self.p_version()
r += self.p_comment()
r += self.p_timescale()
r += self.p_scope()
r += self.p_vars()
r += self.p_unscope()
r += self.p_enddefinitions()
r += self.p_dumpvars()
r += self.p_valuechange()
return r
def write(self, filename):
f = open(filename, "w")
f.write(str(self))
f.close()
class CSVExport():
def __init__(self, dump):
self.dump = dump
def p_vars(self):
r = ""
for var in self.dump.vars:
r += var.name
r += ","
r += "\n"
for var in self.dump.vars:
r += str(var.width)
r += ","
r += "\n"
return r
def p_dumpvars(self):
r = ""
for i in range(len(self.dump)):
for var in self.dump.vars:
try:
var.val = var.values[i]
except:
pass
if var.val == "x":
r += "x"
else:
r += dec2bin(var.val, var.width)
r += ", "
r+= "\n"
return r
def __repr__(self):
r = ""
r += self.p_vars()
r += self.p_dumpvars()
return r
def write(self, filename):
f = open(filename, "w")
f.write(str(self))
f.close()
class PYExport():
def __init__(self, dump):
self.dump = dump
def __repr__(self):
r = "dump = {\n"
for var in self.dump.vars:
r += "\"" + var.name + "\""
r += " : "
r += str(var.values)
r += ",\n"
r += "}"
return r
def write(self, filename):
f = open(filename, "w")
f.write(str(self))
f.close()
class SRExport():
def __init__(self, dump):
self.dump = dump
def write_version(self):
f = open("version", "w")
f.write("1")
f.close()
def write_metadata(self, name):
f = open("metadata", "w")
r = """
[global]
sigrok version = 0.2.0
[device 1]
driver = litescope
capturefile = {}
unitsize = 1
total probes = {}
samplerate = {} MHz
""".format(
name,
len(self.dump.vars),
50, # XXX add parameter
)
for i, var in enumerate(self.dump.vars):
r += "probe{} = {}\n".format(i, var.name)
f.write(r)
f.close()
def write_data(self, name):
# XXX are probes limited to 1 bit?
data_bits = math.ceil(len(self.dump.vars)/8)*8
data_len = 0
for var in self.dump.vars:
data_len = max(data_len, len(var))
datas = []
for i in range(data_len):
data = 0
for j, var in enumerate(reversed(self.dump.vars)):
data = data << 1
try:
data |= var.values[i] %2
except:
pass
datas.append(data)
f = open(name, "wb")
for data in datas:
f.write(data.to_bytes(data_bits//8, "big"))
f.close()
def zip(self, name):
f = zipfile.ZipFile(name + ".sr", "w")
os.chdir(name)
f.write("version")
f.write("metadata")
f.write(name)
os.chdir("..")
f.close()
def write(self, filename):
name, ext = os.path.splitext(filename)
if os.path.exists(name):
shutil.rmtree(name)
os.makedirs(name)
os.chdir(name)
self.write_version()
self.write_metadata(name)
self.write_data(name)
os.chdir("..")
self.zip(name)
shutil.rmtree(name)
class SRImport():
def __init__(self, filename):
name, ext = os.path.splitext(filename)
self.unzip(filename, name)
os.chdir(name)
probes = self.read_metadata()
total_probes = len(probes.keys())
datas = self.read_data(name, total_probes)
os.chdir("..")
shutil.rmtree(name)
self.dump = self.generate_dump(probes, datas)
# XXX we can maybe avoid this
def unzip(self, filename, name):
f = open(filename, "rb")
z = zipfile.ZipFile(f)
if os.path.exists(name):
shutil.rmtree(name)
os.makedirs(name)
for file in z.namelist():
z.extract(file, name)
f.close()
def read_metadata(self):
probes = OrderedDict()
f = open("metadata", "r")
for l in f:
m = re.search("probe([0-9]+) = (\w+)", l, re.I)
if m is not None:
index = int(m.group(1))
name = m.group(2)
probes[name] = index
f.close()
return probes
def read_data(self, name, total_probes):
datas = []
f = open(name, "rb")
while True:
data = f.read(math.ceil(total_probes/8))
if data == bytes('', "utf-8"):
break
data = int.from_bytes(data, "big")
datas.append(data)
f.close()
return datas
def generate_dump(self, probes, datas):
dump = Dump()
for k, v in probes.items():
probe_data = []
for data in datas:
probe_data.append((data >> v) & 0x1)
dump.add(Var(k, 1, probe_data))
return dump
def main():
dump = Dump()
dump.add(Var("foo1", 1, [0,1,0,1,0,1]))
dump.add(Var("foo2", 2, [1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0]))
ramp = [i%128 for i in range(1024)]
dump.add(Var("ramp", 16, ramp))
VCDExport(dump).write("mydump.vcd")
CSVExport(dump).write("mydump.csv")
PYExport(dump).write("mydump.py")
SRExport(dump).write("dump.sr")
dump = SRImport("dump.sr").dump
VCDExport(dump).write("dump.vcd")
if __name__ == '__main__':
main()