X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fadd%2Fexample_buf_pipe.py;h=b83d5035cc01c7ec338c1589359323b213e3099f;hb=f3945d1317aa236b752c1e286801caa2c3a07703;hp=6678a6713f05280828d7b3a5ef4fd4f3adffaa96;hpb=af3ae7902ba4e5a26556eb4442c8351c95b267a4;p=ieee754fpu.git

diff --git a/src/add/example_buf_pipe.py b/src/add/example_buf_pipe.py
index 6678a671..b83d5035 100644
--- a/src/add/example_buf_pipe.py
+++ b/src/add/example_buf_pipe.py
@@ -12,12 +12,12 @@
     where data will flow on *every* clock when the conditions are right.
 
     input acceptance conditions are when:
-        * incoming previous-stage strobe (i_p_stb) is HIGH
-        * outgoing previous-stage busy   (o_p_busy) is LOW
+        * incoming previous-stage strobe (p.i_valid) is HIGH
+        * outgoing previous-stage ready   (p.o_ready) is LOW
 
     output transmission conditions are when:
-        * outgoing next-stage strobe (o_n_stb) is HIGH
-        * outgoing next-stage busy   (i_n_busy) is LOW
+        * outgoing next-stage strobe (n.o_valid) is HIGH
+        * outgoing next-stage ready   (n.i_ready) is LOW
 
     the tricky bit is when the input has valid data and the output is not
     ready to accept it.  if it wasn't for the clock synchronisation, it
@@ -33,7 +33,7 @@
     we now effectively have *two* possible pieces of data to "choose" from:
     the buffered data, and the incoming data.  the decision as to which
     to process and output is based on whether we are in "stall" or not.
-    i.e. when the next stage is no longer busy, the output comes from
+    i.e. when the next stage is no longer ready, the output comes from
     the buffer if a stall had previously occurred, otherwise it comes
     direct from processing the input.
 
@@ -45,154 +45,414 @@
 
 from nmigen import Signal, Cat, Const, Mux, Module
 from nmigen.cli import verilog, rtlil
+from nmigen.hdl.rec import Record, Layout
 
+from collections.abc import Sequence
 
-class ExampleStage:
-    """ an example of how to use the buffered pipeline.  actual names of
-        variables (i_data, r_data, o_data, result) below do not matter:
-        the functions however do.
 
-        input data i_data is read (only), is processed and goes into an
-        intermediate result store [process()].  this is updated combinatorially.
+class PrevControl:
+    """ contains signals that come *from* the previous stage (both in and out)
+        * i_valid: previous stage indicating all incoming data is valid.
+                   may be a multi-bit signal, where all bits are required
+                   to be asserted to indicate "valid".
+        * o_ready: output to next stage indicating readiness to accept data
+        * i_data : an input - added by the user of this class
+    """
 
-        in a non-stall condition, the intermediate result will go into the
-        output (update_output).  however if ever there is a stall, it goes
-        into r_data instead [update_buffer()].
+    def __init__(self, i_width=1):
+        self.i_valid = Signal(i_width, name="p_i_valid") # prev   >>in  self
+        self.o_ready = Signal(name="p_o_ready") # prev   <<out self
 
-        when the non-stall condition is released, r_data is the first
-        to be transferred to the output [flush_buffer()], and the stall
-        condition cleared.
+    def connect_in(self, prev):
+        """ helper function to connect stage to an input source.  do not
+            use to connect stage-to-stage!
+        """
+        return [self.i_valid.eq(prev.i_valid),
+                prev.o_ready.eq(self.o_ready),
+                eq(self.i_data, prev.i_data),
+               ]
+
+    def i_valid_logic(self):
+        vlen = len(self.i_valid)
+        if vlen > 1: # multi-bit case: valid only when i_valid is all 1s
+            all1s = Const(-1, (len(self.i_valid), False))
+            return self.i_valid == all1s
+        # single-bit i_valid case
+        return self.i_valid
 
-        on the next cycle (as long as stall is not raised again) the
-        input may begin to be processed and transferred directly to output.
-    """
 
+class NextControl:
+    """ contains the signals that go *to* the next stage (both in and out)
+        * o_valid: output indicating to next stage that data is valid
+        * i_ready: input from next stage indicating that it can accept data
+        * o_data : an output - added by the user of this class
+    """
     def __init__(self):
-        """ i_data can be a DIFFERENT type from everything else
-            o_data, r_data and result must be of the same type
+        self.o_valid = Signal(name="n_o_valid") # self out>>  next
+        self.i_ready = Signal(name="n_i_ready") # self <<in   next
+
+    def connect_to_next(self, nxt):
+        """ helper function to connect to the next stage data/valid/ready.
+            data/valid is passed *TO* nxt, and ready comes *IN* from nxt.
+        """
+        return [nxt.i_valid.eq(self.o_valid),
+                self.i_ready.eq(nxt.o_ready),
+                eq(nxt.i_data, self.o_data),
+               ]
+
+    def connect_out(self, nxt):
+        """ helper function to connect stage to an output source.  do not
+            use to connect stage-to-stage!
+        """
+        return [nxt.o_valid.eq(self.o_valid),
+                self.i_ready.eq(nxt.i_ready),
+                eq(nxt.o_data, self.o_data),
+               ]
+
+
+def eq(o, i):
+    """ makes signals equal: a helper routine which identifies if it is being
+        passed a list (or tuple) of objects, or signals, or Records, and calls
+        the objects' eq function.
+
+        complex objects (classes) can be used: they must follow the
+        convention of having an eq member function, which takes the
+        responsibility of further calling eq and returning a list of
+        eq assignments
+
+        Record is a special (unusual, recursive) case, where the input
+        is specified as a dictionary (which may contain further dictionaries,
+        recursively), where the field names of the dictionary must match
+        the Record's field spec.
+    """
+    if not isinstance(o, Sequence):
+        o, i = [o], [i]
+    res = []
+    for (ao, ai) in zip(o, i):
+        #print ("eq", ao, ai)
+        if isinstance(ao, Record):
+            for idx, (field_name, field_shape, _) in enumerate(ao.layout):
+                if isinstance(field_shape, Layout):
+                    rres = eq(ao.fields[field_name], ai.fields[field_name])
+                else:
+                    rres = eq(ao.fields[field_name], ai[field_name])
+                res += rres
+        else:
+            rres = ao.eq(ai)
+            if not isinstance(rres, Sequence):
+                rres = [rres]
+            res += rres
+    return res
+
+
+class StageChain:
+    """ pass in a list of stages, and they will automatically be
+        chained together via their input and output specs into a
+        combinatorial chain.
+
+        * input to this class will be the input of the first stage
+        * output of first stage goes into input of second
+        * output of second goes into input into third (etc. etc.)
+        * the output of this class will be the output of the last stage
+    """
+    def __init__(self, chain):
+        self.chain = chain
+
+    def ispec(self):
+        return self.chain[0].ispec()
+
+    def ospec(self):
+        return self.chain[-1].ospec()
+
+    def setup(self, m, i):
+        for (idx, c) in enumerate(self.chain):
+            if hasattr(c, "setup"):
+                c.setup(m, i)               # stage may have some module stuff
+            o = self.chain[idx].ospec()     # only the last assignment survives
+            m.d.comb += eq(o, c.process(i)) # process input into "o"
+            if idx != len(self.chain)-1:
+                ni = self.chain[idx+1].ispec() # becomes new input on next loop
+                m.d.comb += eq(ni, o)          # assign output to next input
+                i = ni
+        self.o = o                             # last loop is the output
+
+    def process(self, i):
+        return self.o
+
+
+class PipelineBase:
+    """ Common functions for Pipeline API
+    """
+    def __init__(self, stage, in_multi=None):
+        """ pass in a "stage" which may be either a static class or a class
+            instance, which has four functions (one optional):
+            * ispec: returns input signals according to the input specification
+            * ispec: returns output signals to the output specification
+            * process: takes an input instance and returns processed data
+            * setup: performs any module linkage if the stage uses one.
+
+            User must also:
+            * add i_data member to PrevControl and
+            * add o_data member to NextControl
         """
-        self.i_data = Signal(16)
-        self.r_data = Signal(16)
-        self.o_data = Signal(16)
-        self.result = Signal(16)
-
-    def process(self):
-        """ process the input data and store it in result.
-            (not needed to be known: result is combinatorial)
+        self.stage = stage
+
+        # set up input and output IO ACK (prev/next ready/valid)
+        self.p = PrevControl(in_multi)
+        self.n = NextControl()
+
+    def connect_to_next(self, nxt):
+        """ helper function to connect to the next stage data/valid/ready.
         """
-        return self.result.eq(self.i_data + 1)
+        return self.n.connect_to_next(nxt.p)
 
-    def update_buffer(self):
-        """ copies the result into the intermediate register r_data
+    def connect_in(self, prev):
+        """ helper function to connect stage to an input source.  do not
+            use to connect stage-to-stage!
         """
-        return self.r_data.eq(self.result)
+        return self.p.connect_in(prev.p)
 
-    def update_output(self):
-        """ copies the (combinatorial) result into the output
+    def connect_out(self, nxt):
+        """ helper function to connect stage to an output source.  do not
+            use to connect stage-to-stage!
         """
-        return self.o_data.eq(self.result)
+        return self.n.connect_out(nxt.n)
 
-    def flush_buffer(self):
-        """ copies the *intermediate* register r_data into the output
+    def set_input(self, i):
+        """ helper function to set the input data
         """
-        return self.o_data.eq(self.r_data)
+        return eq(self.p.i_data, i)
 
     def ports(self):
-        return [self.i_data, self.o_data]
+        return [self.p.i_valid, self.n.i_ready,
+                self.n.o_valid, self.p.o_ready,
+                self.p.i_data, self.n.o_data   # XXX need flattening!
+               ]
 
 
-class BufferedPipeline:
-    """ buffered pipeline stage
+class BufferedPipeline(PipelineBase):
+    """ buffered pipeline stage.  data and strobe signals travel in sync.
+        if ever the input is ready and the output is not, processed data
+        is stored in a temporary register.
 
-        stage-1   i_p_stb  >>in   stage   o_n_stb  out>>   stage+1
-        stage-1   o_p_busy <<out  stage   i_n_busy <<in    stage+1
-        stage-1   i_data   >>in   stage   o_data   out>>   stage+1
+        stage-1   p.i_valid >>in   stage   n.o_valid out>>   stage+1
+        stage-1   p.o_ready <<out  stage   n.i_ready <<in    stage+1
+        stage-1   p.i_data  >>in   stage   n.o_data  out>>   stage+1
                               |             |
-                              +------->  process
+                            process --->----^
                               |             |
-                              +-- r_data ---+
+                              +-- r_data ->-+
+
+        input data p.i_data is read (only), is processed and goes into an
+        intermediate result store [process()].  this is updated combinatorially.
+
+        in a non-stall condition, the intermediate result will go into the
+        output (update_output).  however if ever there is a stall, it goes
+        into r_data instead [update_buffer()].
+
+        when the non-stall condition is released, r_data is the first
+        to be transferred to the output [flush_buffer()], and the stall
+        condition cleared.
+
+        on the next cycle (as long as stall is not raised again) the
+        input may begin to be processed and transferred directly to output.
     """
-    def __init__(self):
-        # input: strobe comes in from previous stage, busy comes in from next
-        #self.i_p_rst = Signal()    # >>in - comes in from PREVIOUS stage
-        self.i_p_stb = Signal()    # >>in - comes in from PREVIOUS stage
-        self.i_n_busy = Signal()   # in<< - comes in from the NEXT stage
+    def __init__(self, stage):
+        PipelineBase.__init__(self, stage)
 
-        # output: strobe goes out to next stage, busy comes in from previous
-        self.o_n_stb = Signal()    # out>> - goes out to the NEXT stage
-        self.o_p_busy = Signal()   # <<out - goes out to the PREVIOUS stage
+        # set up the input and output data
+        self.p.i_data = stage.ispec() # input type
+        self.n.o_data = stage.ospec()
 
     def elaborate(self, platform):
         m = Module()
 
+        result = self.stage.ospec()
+        r_data = self.stage.ospec()
+        if hasattr(self.stage, "setup"):
+            self.stage.setup(m, self.p.i_data)
+
         # establish some combinatorial temporaries
-        o_p_busyn = Signal(reset_less=True)
-        o_n_stbn = Signal(reset_less=True)
-        i_n_busyn = Signal(reset_less=True)
-        i_p_stb_o_p_busyn = Signal(reset_less=True)
-        m.d.comb += [i_n_busyn.eq(~self.i_n_busy),
-                     o_n_stbn.eq(~self.o_n_stb),
-                     o_p_busyn.eq(~self.o_p_busy),
-                     i_p_stb_o_p_busyn.eq(self.i_p_stb & o_p_busyn),
+        o_n_validn = Signal(reset_less=True)
+        i_p_valid_o_p_ready = Signal(reset_less=True)
+        p_i_valid = Signal(reset_less=True)
+        m.d.comb += [p_i_valid.eq(self.p.i_valid_logic()),
+                     o_n_validn.eq(~self.n.o_valid),
+                     i_p_valid_o_p_ready.eq(p_i_valid & self.p.o_ready),
         ]
 
         # store result of processing in combinatorial temporary
-        with m.If(self.i_p_stb): # input is valid: process it
-            m.d.comb += self.stage.process()
+        #with m.If(self.p.i_valid): # input is valid: process it
+        m.d.comb += eq(result, self.stage.process(self.p.i_data))
         # if not in stall condition, update the temporary register
-        with m.If(o_p_busyn): # not stalled
-            m.d.sync += self.stage.update_buffer()
-
-        #with m.If(self.i_p_rst): # reset
-        #    m.d.sync += self.o_n_stb.eq(0)
-        #    m.d.sync += self.o_p_busy.eq(0)
-        with m.If(i_n_busyn): # next stage is not busy
-            with m.If(o_p_busyn): # not stalled
+        with m.If(self.p.o_ready): # not stalled
+            m.d.sync += eq(r_data, result) # update buffer
+
+        #with m.If(self.p.i_rst): # reset
+        #    m.d.sync += self.n.o_valid.eq(0)
+        #    m.d.sync += self.p.o_ready.eq(0)
+        with m.If(self.n.i_ready): # next stage is ready
+            with m.If(self.p.o_ready): # not stalled
                 # nothing in buffer: send (processed) input direct to output
-                m.d.sync += [self.o_n_stb.eq(self.i_p_stb),
-                             self.stage.update_output(),
+                m.d.sync += [self.n.o_valid.eq(p_i_valid),
+                             eq(self.n.o_data, result), # update output
                             ]
-            with m.Else(): # o_p_busy is true, and something is in our buffer.
+            with m.Else(): # p.o_ready is false, and something is in buffer.
                 # Flush the [already processed] buffer to the output port.
-                m.d.sync += [self.o_n_stb.eq(1),
-                             self.stage.flush_buffer(),
+                m.d.sync += [self.n.o_valid.eq(1),
+                             eq(self.n.o_data, r_data), # flush buffer
                              # clear stall condition, declare register empty.
-                             self.o_p_busy.eq(0),
+                             self.p.o_ready.eq(1),
                             ]
-                # ignore input, since o_p_busy is also true.
+                # ignore input, since p.o_ready is also false.
 
-        # (i_n_busy) is true here: next stage is busy
-        with m.Elif(o_n_stbn): # next stage being told "not busy"
-            m.d.sync += [self.o_n_stb.eq(self.i_p_stb),
-                         self.o_p_busy.eq(0), # Keep the buffer empty
+        # (n.i_ready) is false here: next stage is ready
+        with m.Elif(o_n_validn): # next stage being told "ready"
+            m.d.sync += [self.n.o_valid.eq(p_i_valid),
+                         self.p.o_ready.eq(1), # Keep the buffer empty
                          # set the output data (from comb result)
-                         self.stage.update_output(),
+                         eq(self.n.o_data, result),
                         ]
-        # (i_n_busy) and (o_n_stb) both true:
-        with m.Elif(i_p_stb_o_p_busyn):
-            # If next stage *is* busy, and not stalled yet, accept input
-            m.d.sync += self.o_p_busy.eq(self.i_p_stb & self.o_n_stb)
+        # (n.i_ready) false and (n.o_valid) true:
+        with m.Elif(i_p_valid_o_p_ready):
+            # If next stage *is* ready, and not stalled yet, accept input
+            m.d.sync += self.p.o_ready.eq(~(p_i_valid & self.n.o_valid))
 
         return m
 
-    def ports(self):
-        return [self.i_p_stb, self.i_n_busy,
-                self.o_n_stb, self.o_p_busy,
-               ]
 
+class ExampleAddStage:
+    """ an example of how to use the buffered pipeline, as a class instance
+    """
 
-class BufPipe(BufferedPipeline):
+    def ispec(self):
+        """ returns a tuple of input signals which will be the incoming data
+        """
+        return (Signal(16), Signal(16))
+
+    def ospec(self):
+        """ returns an output signal which will happen to contain the sum
+            of the two inputs
+        """
+        return Signal(16)
+
+    def process(self, i):
+        """ process the input data (sums the values in the tuple) and returns it
+        """
+        return i[0] + i[1]
+
+
+class ExampleBufPipeAdd(BufferedPipeline):
+    """ an example of how to use the buffered pipeline, using a class instance
+    """
 
     def __init__(self):
-        BufferedPipeline.__init__(self)
-        self.stage = ExampleStage()
+        addstage = ExampleAddStage()
+        BufferedPipeline.__init__(self, addstage)
 
-    def ports(self):
-        return self.stage.ports() + BufferedPipeline.ports(self)
+
+class ExampleStage:
+    """ an example of how to use the buffered pipeline, in a static class
+        fashion
+    """
+
+    def ispec():
+        return Signal(16, name="example_input_signal")
+
+    def ospec():
+        return Signal(16, name="example_output_signal")
+
+    def process(i):
+        """ process the input data and returns it (adds 1)
+        """
+        return i + 1
+
+
+class ExampleStageCls:
+    """ an example of how to use the buffered pipeline, in a static class
+        fashion
+    """
+
+    def ispec(self):
+        return Signal(16, name="example_input_signal")
+
+    def ospec(self):
+        return Signal(16, name="example_output_signal")
+
+    def process(self, i):
+        """ process the input data and returns it (adds 1)
+        """
+        return i + 1
+
+
+class ExampleBufPipe(BufferedPipeline):
+    """ an example of how to use the buffered pipeline.
+    """
+
+    def __init__(self):
+        BufferedPipeline.__init__(self, ExampleStage)
+
+
+class CombPipe(PipelineBase):
+    """A simple pipeline stage containing combinational logic that can execute
+    completely in one clock cycle.
+
+    Attributes:
+    -----------
+    input : StageInput
+        The pipeline input
+    output : StageOutput
+        The pipeline output
+    r_data : Signal, input_shape
+        A temporary (buffered) copy of a prior (valid) input
+    result: Signal, output_shape
+        The output of the combinatorial logic
+    """
+
+    def __init__(self, stage):
+        PipelineBase.__init__(self, stage)
+        self._data_valid = Signal()
+
+        # set up the input and output data
+        self.p.i_data = stage.ispec() # input type
+        self.n.o_data = stage.ospec() # output type
+
+    def elaborate(self, platform):
+        m = Module()
+
+        r_data = self.stage.ispec() # input type
+        result = self.stage.ospec() # output data
+        if hasattr(self.stage, "setup"):
+            self.stage.setup(m, r_data)
+
+        p_i_valid = Signal(reset_less=True)
+        m.d.comb += p_i_valid.eq(self.p.i_valid_logic())
+        m.d.comb += eq(result, self.stage.process(r_data))
+        m.d.comb += self.n.o_valid.eq(self._data_valid)
+        m.d.comb += self.p.o_ready.eq(~self._data_valid | self.n.i_ready)
+        m.d.sync += self._data_valid.eq(p_i_valid | \
+                                        (~self.n.i_ready & self._data_valid))
+        with m.If(self.p.i_valid & self.p.o_ready):
+            m.d.sync += eq(r_data, self.p.i_data)
+        m.d.comb += eq(self.n.o_data, result)
+        return m
+
+
+class ExampleCombPipe(CombPipe):
+    """ an example of how to use the combinatorial pipeline.
+    """
+
+    def __init__(self):
+        CombPipe.__init__(self, ExampleStage)
 
 
 if __name__ == '__main__':
-    dut = BufPipe()
+    dut = ExampleBufPipe()
     vl = rtlil.convert(dut, ports=dut.ports())
     with open("test_bufpipe.il", "w") as f:
         f.write(vl)
+
+    dut = ExampleCombPipe()
+    vl = rtlil.convert(dut, ports=dut.ports())
+    with open("test_combpipe.il", "w") as f:
+        f.write(vl)