dvisampler: new DMA engine (buggy)
This commit is contained in:
parent
89dbc37ece
commit
29efa85dec
|
@ -12,7 +12,7 @@ from milkymist.dvisampler.analysis import SyncPolarity, ResolutionDetection, Fra
|
|||
from milkymist.dvisampler.dma import DMA
|
||||
|
||||
class DVISampler(Module, AutoCSR):
|
||||
def __init__(self, pads, asmiport):
|
||||
def __init__(self, pads, asmiport, n_dma_slots=2):
|
||||
self.submodules.edid = EDID(pads)
|
||||
self.submodules.clocking = Clocking(pads)
|
||||
|
||||
|
@ -79,5 +79,8 @@ class DVISampler(Module, AutoCSR):
|
|||
self.frame.b.eq(self.syncpol.b)
|
||||
]
|
||||
|
||||
self.submodules.dma = DMA(asmiport)
|
||||
self.submodules.dma = DMA(asmiport, n_dma_slots)
|
||||
self.comb += self.frame.frame.connect(self.dma.frame)
|
||||
self.ev = self.dma.ev
|
||||
|
||||
autocsr_exclude = {"ev"}
|
||||
|
|
|
@ -1,48 +1,160 @@
|
|||
from migen.fhdl.structure import *
|
||||
from migen.fhdl.module import Module
|
||||
from migen.genlib.fsm import FSM
|
||||
from migen.bank.description import *
|
||||
from migen.bank.eventmanager import *
|
||||
from migen.flow.actor import *
|
||||
from migen.actorlib import structuring, dma_asmi, spi
|
||||
from migen.actorlib import dma_asmi
|
||||
|
||||
from milkymist.dvisampler.common import frame_layout
|
||||
|
||||
class DMA(Module):
|
||||
def __init__(self, asmiport):
|
||||
self.frame = Sink(frame_layout)
|
||||
self.shoot = CSR()
|
||||
# Slot status: EMPTY=0 LOADED=1 PENDING=2
|
||||
class _Slot(Module, AutoCSR):
|
||||
def __init__(self, addr_bits, alignment_bits):
|
||||
self.ev_source = EventSourceLevel()
|
||||
self.address = Signal(addr_bits)
|
||||
self.address_valid = Signal()
|
||||
self.address_done = Signal()
|
||||
|
||||
self._r_status = CSRStorage(2, write_from_dev=True)
|
||||
self._r_address = CSRStorage(addr_bits + alignment_bits, alignment_bits=alignment_bits)
|
||||
|
||||
###
|
||||
|
||||
self.comb += [
|
||||
self.address.eq(self._r_address.storage),
|
||||
self.address_valid.eq(self._r_status.storage[0]),
|
||||
self._r_status.dat_w.eq(2),
|
||||
self._r_status.we.eq(self.address_done),
|
||||
self.ev_source.trigger.eq(self._r_status.storage[1])
|
||||
]
|
||||
|
||||
class _SlotArray(Module, AutoCSR):
|
||||
def __init__(self, nslots, addr_bits, alignment_bits):
|
||||
self.ev = EventManager()
|
||||
self.address = Signal(addr_bits)
|
||||
self.address_valid = Signal()
|
||||
self.address_done = Signal()
|
||||
|
||||
###
|
||||
|
||||
slots = [_Slot(addr_bits, alignment_bits) for i in range(nslots)]
|
||||
for n, slot in enumerate(slots):
|
||||
setattr(self.submodules, "slot"+str(n), slot)
|
||||
setattr(self.ev, "slot"+str(n), slot.ev_source)
|
||||
self.ev.finalize()
|
||||
|
||||
change_slot = Signal()
|
||||
current_slot = Signal(max=nslots)
|
||||
self.sync += If(change_slot, [If(slot.address_valid, current_slot.eq(n)) for n, slot in enumerate(slots)])
|
||||
self.comb += change_slot.eq(~self.address_valid | self.address_done)
|
||||
|
||||
self.comb += [
|
||||
self.address.eq(Array(slot.address for slot in slots)[current_slot]),
|
||||
self.address_valid.eq(Array(slot.address_valid for slot in slots)[current_slot])
|
||||
]
|
||||
self.comb += [slot.address_done.eq(self.address_done & (current_slot == n)) for n, slot in enumerate(slots)]
|
||||
|
||||
class DMA(Module):
|
||||
def __init__(self, asmiport, nslots):
|
||||
bus_aw = asmiport.hub.aw
|
||||
bus_dw = asmiport.hub.dw
|
||||
alignment_bits = bits_for(bus_dw//8) - 1
|
||||
|
||||
self.frame = Sink(frame_layout)
|
||||
self._r_frame_size = CSRStorage(bus_aw + alignment_bits, alignment_bits=alignment_bits)
|
||||
self.submodules._slot_array = _SlotArray(nslots, bus_aw, alignment_bits)
|
||||
self.ev = self._slot_array.ev
|
||||
|
||||
###
|
||||
|
||||
# start of frame detection
|
||||
sof = Signal()
|
||||
parity_r = Signal()
|
||||
self.comb += sof.eq(self.frame.stb & (parity_r ^ self.frame.payload.parity))
|
||||
self.sync += If(self.frame.stb & self.frame.ack, parity_r.eq(self.frame.payload.parity))
|
||||
self.comb += sof.eq(parity_r ^ self.frame.payload.parity)
|
||||
|
||||
pending = Signal()
|
||||
frame_en = Signal()
|
||||
# address generator + maximum memory word count to prevent DMA buffer overrun
|
||||
reset_words = Signal()
|
||||
count_word = Signal()
|
||||
last_word = Signal()
|
||||
current_address = Signal(bus_aw)
|
||||
mwords_remaining = Signal(bus_aw)
|
||||
self.comb += last_word.eq(mwords_remaining == 1)
|
||||
self.sync += [
|
||||
If(sof,
|
||||
frame_en.eq(0),
|
||||
If(pending, frame_en.eq(1)),
|
||||
pending.eq(0)
|
||||
),
|
||||
If(self.shoot.re, pending.eq(1))
|
||||
If(reset_words,
|
||||
current_address.eq(self._slot_array.address),
|
||||
mwords_remaining.eq(self._r_frame_size.storage)
|
||||
).Elif(count_word,
|
||||
current_address.eq(current_address + 1),
|
||||
mwords_remaining.eq(mwords_remaining - 1)
|
||||
)
|
||||
]
|
||||
|
||||
pack_factor = asmiport.hub.dw//32
|
||||
self.submodules.packer = structuring.Pack(list(reversed([("pad", 2), ("r", 10), ("g", 10), ("b", 10)])), pack_factor)
|
||||
self.submodules.cast = structuring.Cast(self.packer.source.payload.layout, asmiport.hub.dw, reverse_from=False)
|
||||
self.submodules.dma = spi.DMAWriteController(dma_asmi.Writer(asmiport), spi.MODE_EXTERNAL)
|
||||
# pack pixels into memory words
|
||||
write_pixel = Signal()
|
||||
last_pixel = Signal()
|
||||
cur_memory_word = Signal(bus_dw)
|
||||
encoded_pixel = Signal(32)
|
||||
self.comb += [
|
||||
self.dma.generator.trigger.eq(self.shoot.re),
|
||||
self.packer.sink.stb.eq(self.frame.stb & frame_en),
|
||||
self.frame.ack.eq(self.packer.sink.ack | (~frame_en & ~(pending & sof))),
|
||||
self.packer.sink.payload.r.eq(self.frame.payload.r << 2),
|
||||
self.packer.sink.payload.g.eq(self.frame.payload.g << 2),
|
||||
self.packer.sink.payload.b.eq(self.frame.payload.b << 2),
|
||||
self.packer.source.connect(self.cast.sink, match_by_position=True),
|
||||
self.cast.source.connect(self.dma.data, match_by_position=True)
|
||||
encoded_pixel.eq(Cat(
|
||||
Replicate(0, 2), self.frame.payload.b,
|
||||
Replicate(0, 2), self.frame.payload.g,
|
||||
Replicate(0, 2), self.frame.payload.r))
|
||||
]
|
||||
pack_factor = bus_dw//32
|
||||
assert(pack_factor & (pack_factor - 1) == 0) # only support powers of 2
|
||||
pack_counter = Signal(max=pack_factor)
|
||||
self.comb += last_pixel.eq(pack_counter == (pack_factor - 1))
|
||||
self.sync += If(write_pixel,
|
||||
[If(pack_counter == i, cur_memory_word[32*i:32*(i+1)].eq(encoded_pixel)) for i in range(pack_factor)],
|
||||
pack_counter.eq(pack_counter + 1)
|
||||
)
|
||||
|
||||
# bus accessor
|
||||
self.submodules._bus_accessor = dma_asmi.Writer(asmiport)
|
||||
self.comb += [
|
||||
self._bus_accessor.address_data.payload.a.eq(current_address),
|
||||
self._bus_accessor.address_data.payload.d.eq(cur_memory_word)
|
||||
]
|
||||
|
||||
# control FSM
|
||||
fsm = FSM("WAIT_SOF", "TRANSFER_PIXEL", "TO_MEMORY", "EOF")
|
||||
self.submodules += fsm
|
||||
|
||||
fsm.act(fsm.WAIT_SOF,
|
||||
self.frame.ack.eq(~sof),
|
||||
reset_words.eq(1),
|
||||
If(self._slot_array.address_valid,
|
||||
If(self.frame.stb & sof, fsm.next_state(fsm.TRANSFER_PIXEL))
|
||||
)
|
||||
)
|
||||
fsm.act(fsm.TRANSFER_PIXEL,
|
||||
self.frame.ack.eq(1),
|
||||
If(self.frame.stb,
|
||||
write_pixel.eq(1),
|
||||
If(last_pixel,
|
||||
fsm.next_state(fsm.TO_MEMORY)
|
||||
)
|
||||
)
|
||||
)
|
||||
fsm.act(fsm.TO_MEMORY,
|
||||
self._bus_accessor.address_data.stb.eq(1),
|
||||
If(self._bus_accessor.address_data.ack,
|
||||
count_word.eq(1),
|
||||
If(last_word,
|
||||
fsm.next_state(fsm.EOF)
|
||||
).Else(
|
||||
fsm.next_state(fsm.TRANSFER_PIXEL)
|
||||
)
|
||||
)
|
||||
)
|
||||
fsm.act(fsm.EOF,
|
||||
If(~self._bus_accessor.busy,
|
||||
self._slot_array.address_done.eq(1),
|
||||
fsm.next_state(fsm.WAIT_SOF)
|
||||
)
|
||||
)
|
||||
|
||||
def get_csrs(self):
|
||||
return [self.shoot] + self.dma.get_csrs()
|
||||
return [self._r_frame_size] + self._slot_array.get_csrs()
|
||||
|
|
Loading…
Reference in New Issue