from migen.fhdl.structure import * from migen.flow.actor import * from migen.flow.network import * from migen.flow import plumbing from migen.actorlib import ala, misc, dma_asmi, structuring from migen.bank.description import * from migen.bank import csrgen _hbits = 11 _vbits = 11 _bpp = 32 _bpc = 10 _pixel_layout = [ ("b", BV(_bpc)), ("g", BV(_bpc)), ("r", BV(_bpc)), ("pad", BV(_bpp-3*_bpc)) ] _bpc_dac = 8 _dac_layout = [ ("hsync", BV(1)), ("vsync", BV(1)), ("b", BV(_bpc_dac)), ("g", BV(_bpc_dac)), ("r", BV(_bpc_dac)) ] class _FrameInitiator(Actor): def __init__(self, asmi_bits, length_bits, alignment_bits): self._alignment_bits = alignment_bits self._enable = RegisterField("enable") self._hres = RegisterField("hres", _hbits, reset=640) self._hsync_start = RegisterField("hsync_start", _hbits, reset=656) self._hsync_end = RegisterField("hsync_end", _hbits, reset=752) self._hscan = RegisterField("hscan", _hbits, reset=799) self._vres = RegisterField("vres", _vbits, reset=480) self._vsync_start = RegisterField("vsync_start", _vbits, reset=492) self._vsync_end = RegisterField("vsync_end", _vbits, reset=494) self._vscan = RegisterField("vscan", _vbits, reset=524) self._base = RegisterField("base", asmi_bits + self._alignment_bits) self._length = RegisterField("length", length_bits + self._alignment_bits, reset=640*480*4) layout = [ ("hres", BV(_hbits)), ("hsync_start", BV(_hbits)), ("hsync_end", BV(_hbits)), ("hscan", BV(_hbits)), ("vres", BV(_vbits)), ("vsync_start", BV(_vbits)), ("vsync_end", BV(_vbits)), ("vscan", BV(_vbits)), ("base", BV(asmi_bits)), ("length", BV(length_bits)) ] super().__init__(("frame", Source, layout)) def get_registers(self): return [self._enable, self._hres, self._hsync_start, self._hsync_end, self._hscan, self._vres, self._vsync_start, self._vsync_end, self._vscan, self._base, self._length] def get_fragment(self): # TODO: make address updates atomic token = self.token("frame") comb = [ self.busy.eq(0), self.endpoints["frame"].stb.eq(self._enable.field.r), token.hres.eq(self._hres.field.r), token.hsync_start.eq(self._hsync_start.field.r), token.hsync_end.eq(self._hsync_end.field.r), token.hscan.eq(self._hscan.field.r), token.vres.eq(self._vres.field.r), token.vsync_start.eq(self._vsync_start.field.r), token.vsync_end.eq(self._vsync_end.field.r), token.vscan.eq(self._vscan.field.r), token.base.eq(self._base.field.r[self._alignment_bits:]), token.length.eq(self._length.field.r[self._alignment_bits:]) ] return Fragment(comb) class VTG(Actor): def __init__(self): super().__init__( ("timing", Sink, [ ("hres", BV(_hbits)), ("hsync_start", BV(_hbits)), ("hsync_end", BV(_hbits)), ("hscan", BV(_hbits)), ("vres", BV(_vbits)), ("vsync_start", BV(_vbits)), ("vsync_end", BV(_vbits)), ("vscan", BV(_vbits))]), ("pixels", Sink, _pixel_layout), ("dac", Source, _dac_layout) ) def get_fragment(self): hactive = Signal() vactive = Signal() active = Signal() generate_en = Signal() hcounter = Signal(BV(_hbits)) vcounter = Signal(BV(_vbits)) skip = _bpc - _bpc_dac comb = [ active.eq(hactive & vactive), If(active, self.token("dac").r.eq(self.token("pixels").r[skip:]), self.token("dac").g.eq(self.token("pixels").g[skip:]), self.token("dac").b.eq(self.token("pixels").b[skip:]) ), generate_en.eq(self.endpoints["timing"].stb & (~active | self.endpoints["pixels"].stb)), self.endpoints["pixels"].ack.eq(self.endpoints["dac"].ack & active), self.endpoints["dac"].stb.eq(generate_en) ] tp = self.token("timing") sync = [ self.endpoints["timing"].ack.eq(0), If(generate_en & self.endpoints["dac"].ack, hcounter.eq(hcounter + 1), If(hcounter == 0, hactive.eq(1)), If(hcounter == tp.hres, hactive.eq(0)), If(hcounter == tp.hsync_start, self.token("dac").hsync.eq(1)), If(hcounter == tp.hsync_end, self.token("dac").hsync.eq(0)), If(hcounter == tp.hscan, hcounter.eq(0), If(vcounter == tp.vscan, vcounter.eq(0) # FIXME: work around Flow bug #self.endpoints["timing"].ack.eq(1) ).Else( vcounter.eq(vcounter + 1) ) ), If(vcounter == 0, vactive.eq(1)), If(vcounter == tp.vres, vactive.eq(0)), If(vcounter == tp.vsync_start, self.token("dac").vsync.eq(1)), If(vcounter == tp.vsync_end, self.token("dac").vsync.eq(0)) ) ] return Fragment(comb, sync) class FIFO(Actor): def __init__(self): super().__init__(("dac", Sink, _dac_layout)) self.vga_clk = Signal() self.vga_hsync_n = Signal() self.vga_vsync_n = Signal() self.vga_r = Signal(BV(_bpc_dac)) self.vga_g = Signal(BV(_bpc_dac)) self.vga_b = Signal(BV(_bpc_dac)) def get_fragment(self): data_width = 2+3*_bpc_dac asfifo = Instance("asfifo", [ ("data_out", BV(data_width)), ("empty", BV(1)), ("full", BV(1)) ], [ ("read_en", BV(1)), ("clk_read", self.vga_clk), ("data_in", BV(data_width)), ("write_en", BV(1)), ("rst", BV(1)) ], parameters=[ ("data_width", data_width), ("address_width", 8) ], clkport="clk_write") t = self.token("dac") return Fragment( [ asfifo.ins["read_en"].eq(1), self.endpoints["dac"].ack.eq(~asfifo.outs["full"]), asfifo.ins["write_en"].eq(self.endpoints["dac"].stb), asfifo.ins["data_in"].eq(Cat(~t.hsync, ~t.vsync, t.r, t.g, t.b)), self.busy.eq(0), asfifo.ins["rst"].eq(0) ], [ Cat(self.vga_hsync_n, self.vga_vsync_n, self.vga_r, self.vga_g, self.vga_b).eq(asfifo.outs["data_out"]) ], instances=[asfifo]) class FakeDMA(Actor): def __init__(self, port): self.port = port super().__init__( ("address", Sink, [("a", BV(self.port.hub.aw))]), ("data", Source, [("d", BV(self.port.hub.dw))])) def get_fragment(self): pixel = Signal(BV(32)) comb = [ self.endpoints["address"].ack.eq(1), self.endpoints["data"].stb.eq(1), self.token("data").d.eq(Replicate(pixel, 4)) ] sync = [ If(self.endpoints["data"].ack, pixel.eq(pixel + 1)) ] return Fragment(comb, sync) class Framebuffer: def __init__(self, address, asmiport): asmi_bits = asmiport.hub.aw alignment_bits = bits_for(asmiport.hub.dw//8) length_bits = _hbits + _vbits + 2 - alignment_bits pack_factor = asmiport.hub.dw//_bpp packed_pixels = structuring.pack_layout(_pixel_layout, pack_factor) fi = ActorNode(_FrameInitiator(asmi_bits, length_bits, alignment_bits)) adrloop = ActorNode(misc.IntSequence(length_bits)) adrbase = ActorNode(ala.Add(BV(asmi_bits))) adrbuffer = ActorNode(plumbing.Buffer) #dma = ActorNode(dma_asmi.SequentialReader(asmiport)) dma = ActorNode(FakeDMA(asmiport)) cast = ActorNode(structuring.Cast(asmiport.hub.dw, packed_pixels)) unpack = ActorNode(structuring.Unpack(pack_factor, _pixel_layout)) vtg = ActorNode(VTG()) fifo = ActorNode(FIFO()) g = DataFlowGraph() g.add_connection(fi, adrloop, source_subr=["length"]) g.add_connection(adrloop, adrbase, sink_subr=["a"]) g.add_connection(fi, adrbase, source_subr=["base"], sink_subr=["b"]) g.add_connection(adrbase, adrbuffer) g.add_connection(adrbuffer, dma) g.add_connection(dma, cast) g.add_connection(cast, unpack) g.add_connection(unpack, vtg, sink_ep="pixels") g.add_connection(fi, vtg, sink_ep="timing", source_subr=[ "hres", "hsync_start", "hsync_end", "hscan", "vres", "vsync_start", "vsync_end", "vscan"]) g.add_connection(vtg, fifo) self._comp_actor = CompositeActor(g) self.bank = csrgen.Bank(fi.actor.get_registers(), address=address) # VGA clock input self.vga_clk = fifo.actor.vga_clk # Pads self.vga_psave_n = Signal() self.vga_hsync_n = fifo.actor.vga_hsync_n self.vga_vsync_n = fifo.actor.vga_vsync_n self.vga_sync_n = Signal() self.vga_blank_n = Signal() self.vga_r = fifo.actor.vga_r self.vga_g = fifo.actor.vga_g self.vga_b = fifo.actor.vga_b def get_fragment(self): comb = [ self.vga_sync_n.eq(0), self.vga_psave_n.eq(1), self.vga_blank_n.eq(1) ] return self.bank.get_fragment() \ + self._comp_actor.get_fragment() \ + Fragment(comb)