Initial gateware code for VGA capture

Files:
    new file:   gateware/vga/__init__.py
    new file:   gateware/vga/analysis.py
    new file:   gateware/vga/datacapture.py

__init__.py:
    Implements a module VGAIn which instatiates submodules
    for Datacapture, FrameExtrantion and DMA

analysis.py:
    Implements FrameExtraction module, which is reponsible
    for sof(start of frame) detection, color space conversion
    framing(packing) and also uses async fifo to move data
    from VGA pixel clock domain to sys_clk domain

datacapture.py:
    Implements DataCapture which is responsible for capturing
    pixel data at proper time, depending on HSYNC and VSYNC
    signals

Author: rohitk-singh

gateware/vga/__init__.py

@@ -0,0 +1,36 @@
+from migen.fhdl.std import *
+from migen.bank.description import *
+
+from gateware.hdmi_in.dma import DMA
+from gateware.vga.analysis import FrameExtraction
+from gateware.vga.vga_gen import VGAGenerator
+from gateware.vga.datacapture import DataCapture
+
+
+class VGAIn(Module, AutoCSR):
+    def __init__(self, pads, lasmim, n_dma_slots=2, fifo_depth=512):
+
+        self.clock_domains.cd_pix = ClockDomain()
+        self.comb += [
+          self.cd_pix.clk.eq(pads.datack),
+          self.cd_pix.rst.eq(ResetSignal())  # XXX FIXME
+        ]
+        self.cap = DataCapture(pads)
+        self.submodules += self.cap
+
+        self.submodules.frame = FrameExtraction(lasmim.dw, fifo_depth)
+
+        self.comb += [
+            self.frame.valid_i.eq(self.cap.valid),
+            self.frame.de.eq(self.cap.de),
+            self.frame.vsync.eq(self.cap.vsync),
+            self.frame.r.eq(self.cap.r),
+            self.frame.g.eq(self.cap.g),
+            self.frame.b.eq(self.cap.b)
+        ]
+
+        self.submodules.dma = DMA(lasmim, n_dma_slots)
+        self.comb += self.frame.frame.connect(self.dma.frame)
+        self.ev = self.dma.ev
+
+    autocsr_exclude = {"ev"}

gateware/vga/analysis.py

@@ -0,0 +1,146 @@
+import math
+
+from migen.fhdl.std import *
+from migen.genlib.cdc import MultiReg, PulseSynchronizer
+from migen.genlib.fifo import AsyncFIFO, SyncFIFO
+from migen.genlib.record import Record
+from migen.bank.description import *
+from migen.flow.actor import *
+
+from gateware.csc.rgb2ycbcr import RGB2YCbCr
+from gateware.csc.ycbcr444to422 import YCbCr444to422
+
+
+class FrameExtraction(Module, AutoCSR):
+    def __init__(self, word_width, fifo_depth):
+        # in pix clock domain
+        self.valid_i = Signal()
+        self.vsync = Signal()
+        self.de = Signal()
+        self.r = Signal(8)
+        self.g = Signal(8)
+        self.b = Signal(8)
+
+        self.counter = Signal(math.ceil(math.log2(1024*768)))
+
+        word_layout = [("sof", 1), ("pixels", word_width)]
+        self.frame = Source(word_layout)
+        self.busy = Signal()
+
+        self._overflow = CSR()
+        self._start_counter = CSRStorage(1, reset=0)
+
+        self.sync += [
+            If((self._start_counter.storage),
+               self.counter.eq(self.counter + 1)
+            )
+        ]
+
+        de_r = Signal()
+        self.sync.pix += de_r.eq(self.de)
+
+        rgb2ycbcr = RGB2YCbCr()
+        self.submodules += RenameClockDomains(rgb2ycbcr, "pix")
+        chroma_downsampler = YCbCr444to422()
+        self.submodules += RenameClockDomains(chroma_downsampler, "pix")
+        self.comb += [
+            rgb2ycbcr.sink.stb.eq(self.valid_i),
+            rgb2ycbcr.sink.sop.eq(self.de & ~de_r),
+            rgb2ycbcr.sink.r.eq(self.r),
+            rgb2ycbcr.sink.g.eq(self.g),
+            rgb2ycbcr.sink.b.eq(self.b),
+            Record.connect(rgb2ycbcr.source, chroma_downsampler.sink),
+            chroma_downsampler.source.ack.eq(1)
+        ]
+        # XXX need clean up
+        de = self.de
+        vsync = self.vsync
+        for i in range(rgb2ycbcr.latency + chroma_downsampler.latency):
+            next_de = Signal()
+            next_vsync = Signal()
+            self.sync.pix += [
+                next_de.eq(de),
+                next_vsync.eq(vsync)
+            ]
+            de = next_de
+            vsync = next_vsync
+
+        # start of frame detection
+        vsync_r = Signal()
+        new_frame = Signal()
+        self.comb += new_frame.eq(vsync & ~vsync_r)
+        self.sync.pix += vsync_r.eq(vsync)
+
+        # pack pixels into words
+        cur_word = Signal(word_width)
+        cur_word_valid = Signal()
+        encoded_pixel = Signal(16)
+        self.comb += encoded_pixel.eq(Cat(chroma_downsampler.source.y, chroma_downsampler.source.cb_cr)),
+        pack_factor = word_width//16
+        assert(pack_factor & (pack_factor - 1) == 0)  # only support powers of 2
+        pack_counter = Signal(max=pack_factor)
+
+        self.sync.pix += [
+            cur_word_valid.eq(0),
+            If(new_frame,
+                cur_word_valid.eq(pack_counter == (pack_factor - 1)),
+                pack_counter.eq(0),
+            ).Elif(chroma_downsampler.source.stb & de,
+                [If(pack_counter == (pack_factor-i-1),
+                    cur_word[16*i:16*(i+1)].eq(encoded_pixel)) for i in range(pack_factor)],
+                cur_word_valid.eq(pack_counter == (pack_factor - 1)),
+                pack_counter.eq(pack_counter + 1)
+            )
+        ]
+
+        # FIFO
+        fifo = RenameClockDomains(AsyncFIFO(word_layout, fifo_depth),
+            {"write": "pix", "read": "sys"})
+        self.submodules += fifo
+        self.comb += [
+            fifo.din.pixels.eq(cur_word),
+            fifo.we.eq(cur_word_valid)
+        ]
+
+        self.sync.pix += \
+            If(new_frame,
+                fifo.din.sof.eq(1)
+            ).Elif(cur_word_valid,
+                fifo.din.sof.eq(0)
+            )
+
+        self.comb += [
+            self.frame.stb.eq(fifo.readable),
+            self.frame.payload.eq(fifo.dout),
+            fifo.re.eq(self.frame.ack),
+            self.busy.eq(0)
+        ]
+
+        # overflow detection
+        pix_overflow = Signal()
+        pix_overflow_reset = Signal()
+
+        self.sync.pix += [
+            If(fifo.we & ~fifo.writable,
+                pix_overflow.eq(1)
+            ).Elif(pix_overflow_reset,
+                pix_overflow.eq(0)
+            )
+        ]
+
+        sys_overflow = Signal()
+        self.specials += MultiReg(pix_overflow, sys_overflow)
+        self.comb += [
+            pix_overflow_reset.eq(self._overflow.re),
+        ]
+
+        overflow_mask = Signal()
+        self.comb += [
+            self._overflow.w.eq(sys_overflow & ~overflow_mask),
+        ]
+        self.sync += \
+            If(self._overflow.re,
+                overflow_mask.eq(1)
+            ).Elif(pix_overflow_reset,
+                overflow_mask.eq(0)
+            )

gateware/vga/datacapture.py

@@ -0,0 +1,97 @@
+from migen.fhdl.std import *
+from migen.genlib.cdc import MultiReg, PulseSynchronizer
+from migen.genlib.fifo import AsyncFIFO, SyncFIFO
+from migen.genlib.record import Record
+from migen.bank.description import *
+from migen.flow.actor import *
+
+
+class DataCapture(Module, AutoCSR):
+    def __init__(self, pads):
+
+        self.counterX = Signal(16)
+        self.counterY = Signal(16)
+
+        self.r = Signal(8)
+        self.g = Signal(8)
+        self.b = Signal(8)
+        self.de = Signal()
+        self.vsync = Signal()
+        self.hsync = Signal()
+        self.valid = Signal()
+
+        hActive = Signal()
+        vActive = Signal()
+
+        vsout = Signal()
+        self.comb += vsout.eq(pads.vsout)
+        vsout_r = Signal()
+        vsout_rising_edge = Signal()
+        self.comb += vsout_rising_edge.eq(vsout & ~vsout_r)
+        self.sync.pix += vsout_r.eq(vsout)
+
+        hsout = Signal()
+        self.comb += hsout.eq(pads.hsout)
+        hsout_r = Signal()
+        hsout_rising_edge = Signal()
+        self.comb += hsout_rising_edge.eq(hsout & ~hsout_r)
+        self.sync.pix += hsout_r.eq(hsout)
+
+        r = Signal(8)
+        g = Signal(8)
+        b = Signal(8)
+
+        # Interchange Red and Blue channels due to PCB issue
+        # and instead of 0:8 we have to take 2:10 that is higher bits
+        self.comb += [
+            r.eq(pads.blue[2:]),
+            g.eq(pads.green[2:]),
+            b.eq(pads.red[2:]),
+            self.vsync.eq(vsout),
+            self.hsync.eq(hsout),
+        ]
+
+        self.sync.pix += [
+
+            self.r.eq(r),
+            self.g.eq(g),
+            self.b.eq(b),
+
+            self.counterX.eq(self.counterX + 1),
+
+            If(hsout_rising_edge,
+                self.counterX.eq(0),
+                self.counterY.eq(self.counterY + 1)
+            ),
+
+            If(vsout_rising_edge,
+               self.counterY.eq(0),
+            ),
+
+            #  VGA Scan Timing Values used below for 1024x768@60Hz
+            #   Source: http://hamsterworks.co.nz/mediawiki/index.php/VGA_timings
+            #
+            #   Horizontal Scan:
+            #       Hsync: 136; HBackPorch: 160, HActive: 1024
+            #
+            #   Vertical Scan:
+            #       Vsync: 6; VBackPorch: 29; VActive: 768
+            #
+            If((136+160 < self.counterX) & (self.counterX <= 136+160+1024),
+                hActive.eq(1)
+            ).Else(
+                hActive.eq(0)
+            ),
+
+            If((6+29 < self.counterY) & (self.counterY <= 6+29+768),
+                vActive.eq(1)
+            ).Else(
+                vActive.eq(0)
+            ),
+        ]
+
+        # FIXME : valid signal should be proper
+        self.comb += [
+            self.valid.eq(1),
+            self.de.eq(vActive & hActive),
+        ]