summaryrefslogtreecommitdiff
path: root/decoders
diff options
context:
space:
mode:
Diffstat (limited to 'decoders')
-rw-r--r--decoders/parallel/pd.py139
1 files changed, 72 insertions, 67 deletions
diff --git a/decoders/parallel/pd.py b/decoders/parallel/pd.py
index e0623e7..b4dbdb3 100644
--- a/decoders/parallel/pd.py
+++ b/decoders/parallel/pd.py
@@ -65,7 +65,7 @@ class Pin:
RESET = DATA_N
class Ann:
- ITEM, WORD = range(2)
+ ITEM, WORD, WARN = range(3)
class ChannelError(Exception):
pass
@@ -101,84 +101,91 @@ class Decoder(srd.Decoder):
annotations = (
('item', 'Item'),
('word', 'Word'),
+ ('warning', 'Warning'),
)
annotation_rows = (
('items', 'Items', (Ann.ITEM,)),
('words', 'Words', (Ann.WORD,)),
+ ('warnings', 'Warnings', (Ann.WARN,)),
)
def __init__(self):
self.reset()
def reset(self):
- self.items = []
- self.saved_item = None
- self.ss_item = self.es_item = None
- self.saved_word = None
- self.ss_word = self.es_word = None
- self.first = True
+ self.pend_item = None
+ self.word_items = []
def start(self):
self.out_python = self.register(srd.OUTPUT_PYTHON)
self.out_ann = self.register(srd.OUTPUT_ANN)
- def putpb(self, data):
- self.put(self.ss_item, self.es_item, self.out_python, data)
-
- def putb(self, data):
- self.put(self.ss_item, self.es_item, self.out_ann, data)
-
- def putpw(self, data):
- self.put(self.ss_word, self.es_word, self.out_python, data)
-
- def putw(self, data):
- self.put(self.ss_word, self.es_word, self.out_ann, data)
-
- def handle_bits(self, item, used_pins):
-
- # If a word was previously accumulated, then emit its annotation
- # now after its end samplenumber became available.
- if self.saved_word is not None:
- if self.options['wordsize'] > 0:
- self.es_word = self.samplenum
- self.putw([Ann.WORD, [self.fmt_word.format(self.saved_word)]])
- self.putpw(['WORD', self.saved_word])
- self.saved_word = None
-
- # Defer annotations for individual items until the next sample
- # is taken, and the previous sample's end samplenumber has
- # become available.
- if self.first:
- # Save the start sample and item for later (no output yet).
- self.ss_item = self.samplenum
- self.first = False
- self.saved_item = item
- elif self.saved_item is None:
- pass
- else:
- # Output the saved item (from the last CLK edge to the current).
- self.es_item = self.samplenum
- self.putpb(['ITEM', self.saved_item])
- self.putb([Ann.ITEM, [self.fmt_item.format(self.saved_item)]])
- self.ss_item = self.samplenum
- self.saved_item = item
-
- # Get as many items as the configured wordsize specifies.
- if not self.items:
- self.ss_word = self.samplenum
- self.items.append(item)
- ws = self.options['wordsize']
- if len(self.items) < ws:
+ def putg(self, ss, es, ann, txts):
+ self.put(ss, es, self.out_ann, [ann, txts])
+
+ def putpy(self, ss, es, ann, data):
+ self.put(ss, es, self.out_python, [ann, data])
+
+ def flush_word(self, bus_width):
+ if not self.word_items:
+ return
+ word_size = self.options['wordsize']
+
+ items = self.word_items
+ ss, es = items[0][0], items[-1][1]
+ items = [i[2] for i in items]
+ if self.options['endianness'] == 'big':
+ items.reverse()
+ word = sum([d << (i * bus_width) for i, d in enumerate(items)])
+
+ txts = [self.fmt_word.format(word)]
+ self.putg(ss, es, Ann.WORD, txts)
+ self.putpy(ss, es, 'WORD', word)
+ # self.putpy(ss, es, 'WORD', (word, bus_width, word_size))
+
+ if len(items) != word_size:
+ txts = ['incomplete word size', 'word size', 'ws']
+ self.putg(ss, es, Ann.WARN, txts)
+
+ self.word_items.clear()
+
+ def queue_word(self, now, item, bus_width):
+ wordsize = self.options['wordsize']
+ if not wordsize:
return
- # Collect words and prepare annotation details, but defer emission
- # until the end samplenumber becomes available.
- endian = self.options['endianness']
- if endian == 'big':
- self.items.reverse()
- word = sum([self.items[i] << (i * used_pins) for i in range(ws)])
- self.saved_word = word
- self.items = []
+ # Terminate a previously seen item of a word first. Emit the
+ # word's annotation when the last item's end was seen.
+ if self.word_items:
+ ss, _, data = self.word_items[-1]
+ es = now
+ self.word_items[-1] = (ss, es, data)
+ if len(self.word_items) == wordsize:
+ self.flush_word(bus_width)
+
+ # Start tracking the currently seen item (yet unknown end time).
+ if item is not None:
+ pend = (now, None, item)
+ self.word_items.append(pend)
+
+ def handle_bits(self, now, item, bus_width):
+
+ # Optionally flush a previously started item.
+ if self.pend_item:
+ ss, _, data = self.pend_item
+ self.pend_item = None
+ es = now
+ txts = [self.fmt_item.format(data)]
+ self.putg(ss, es, Ann.ITEM, txts)
+ self.putpy(ss, es, 'ITEM', data)
+ # self.putpy(ss, es, 'ITEM', (data, bus_width))
+
+ # Optionally queue the currently seen item.
+ if item is not None:
+ self.pend_item = (now, None, item)
+
+ # Pass the current item to the word accumulation logic.
+ self.queue_word(now, item, bus_width)
def decode(self):
# Determine which (optional) channels have input data. Insist in
@@ -252,10 +259,8 @@ class Decoder(srd.Decoder):
if reset_edge:
in_reset = pins[Pin.RESET] == reset_active
if in_reset:
- self.handle_bits(None, num_item_bits)
- self.saved_item = None
- self.saved_word = None
- self.first = True
+ self.handle_bits(self.samplenum, None, num_item_bits)
+ self.flush_word(num_item_bits)
if in_reset:
continue
@@ -263,4 +268,4 @@ class Decoder(srd.Decoder):
data_bits = [0 if idx is None else pins[idx] for idx in data_indices]
data_bits = data_bits[:num_item_bits]
item = bitpack(data_bits)
- self.handle_bits(item, num_item_bits)
+ self.handle_bits(self.samplenum, item, num_item_bits)