2
0
mirror of https://github.com/m-labs/artiq.git synced 2024-12-26 19:58:25 +08:00

rtio/sed: lane distributor fixes

This commit is contained in:
Sebastien Bourdeauducq 2017-09-13 17:50:06 +08:00
parent bdd96084c5
commit c74abccfd5

View File

@ -8,7 +8,7 @@ def layout_lane_io(seqn_width, layout_payload):
("we", 1, DIR_M_TO_S),
("writable", 1, DIR_S_TO_M),
("seqn", seqn_width, DIR_M_TO_S),
("payload", layout_payload, DIR_M_TO_S)
("payload", [(a, b, DIR_M_TO_S) for a, b in layout_payload])
]
@ -18,11 +18,11 @@ def layout_lane_io(seqn_width, layout_payload):
# 3. check status
class LaneDistributor(Module):
def __init__(self, lane_count, fifo_size, layout_payload, fine_ts_width):
def __init__(self, lane_count, fifo_size, layout_payload, fine_ts_width, enable_spread=False):
if lane_count & (lane_count - 1):
raise NotImplementedError("lane count must be a power of 2")
seqn_width = 4*bits_for(fifo_size-1)*lane_count
seqn_width = 4*bits_for(lane_count*fifo_size-1)
self.cri = cri.Interface()
self.minimum_coarse_timestamp = Signal(64-fine_ts_width)
@ -45,7 +45,7 @@ class LaneDistributor(Module):
seqn = Signal(seqn_width)
# distribute data to lanes
for lio in lane_io:
for lio in self.lane_io:
self.comb += [
lio.seqn.eq(seqn),
lio.payload.channel.eq(self.cri.chan_sel[:16]),
@ -57,16 +57,19 @@ class LaneDistributor(Module):
self.comb += lio.payload.data.eq(self.cri.data)
# when timestamp arrives in cycle #1, prepare computations
coarse_timestamp = self.cri.timestamp[fine_ts_width:]
coarse_timestamp = Signal(64-fine_ts_width)
self.comb += coarse_timestamp.eq(self.cri.timestamp[fine_ts_width:])
timestamp_above_min = Signal()
timestamp_above_laneA_min = Signal()
timestamp_above_laneB_min = Signal()
use_laneB = Signal()
use_lanen = Signal(max=lane_count)
current_lane_plus_one = Signal(max=lane_count)
self.comb += current_lane_plus_one.eq(current_lane + 1)
self.sync += [
timestamp_above_min.eq(coarse_timestamp > self.minimum_coarse_timestamp),
timestamp_above_laneA_min.eq(coarse_timestamp > last_lane_coarse_timestamps[current_lane]),
timestamp_above_laneB_min.eq(coarse_timestamp > last_lane_coarse_timestamps[current_lane + 1]),
timestamp_above_laneB_min.eq(coarse_timestamp > last_lane_coarse_timestamps[current_lane_plus_one]),
If(coarse_timestamp <= last_coarse_timestamp,
use_lanen.eq(current_lane + 1),
use_laneB.eq(1)
@ -85,12 +88,12 @@ class LaneDistributor(Module):
timestamp_above_lane_min.eq(Mux(use_laneB, timestamp_above_laneB_min, timestamp_above_laneA_min)),
do_write.eq((self.cri.cmd == cri.commands["write"]) & timestamp_above_min & timestamp_above_lane_min),
do_underflow.eq((self.cri.cmd == cri.commands["write"]) & ~timestamp_above_min),
do_sequence_error((self.cri.cmd == cri.commands["write"]) & timestamp_above_min & ~timestamp_above_lane_min),
Array(lio.we for lio in lane_io)[use_lanen].eq(do_write)
do_sequence_error.eq((self.cri.cmd == cri.commands["write"]) & timestamp_above_min & ~timestamp_above_lane_min),
Array(lio.we for lio in self.lane_io)[use_lanen].eq(do_write)
]
self.sync += [
If(do_write,
current_lane.eq(current_lane + 1),
If(use_laneB, current_lane.eq(current_lane + 1)),
last_coarse_timestamp.eq(coarse_timestamp),
last_lane_coarse_timestamps[use_lanen].eq(coarse_timestamp),
seqn.eq(seqn + 1),
@ -100,19 +103,20 @@ class LaneDistributor(Module):
# cycle #3, read status
current_lane_writable = Signal()
self.comb += [
current_lane_writable.eq((lio.writable for lio in lane_io)[current_lane]),
current_lane_writable.eq(Array(lio.writable for lio in self.lane_io)[current_lane]),
o_status_wait.eq(~current_lane_writable)
]
self.sync += [
o_status_underflow.eq(do_underflow),
o_status_sequence_error(do_sequence_error)
o_status_sequence_error.eq(do_sequence_error)
]
# current lane has been full, spread events by switching to the next.
current_lane_writable_r = Signal()
self.sync += [
current_lane_writable_r.eq(current_lane_writable),
If(~current_lane_writable_r & current_lane_writable,
current_lane.eq(current_lane + 1)
)
]
if enable_spread:
current_lane_writable_r = Signal()
self.sync += [
current_lane_writable_r.eq(current_lane_writable),
If(~current_lane_writable_r & current_lane_writable,
current_lane.eq(current_lane + 1)
)
]