Readyfix (#57)

Improved ready bit wiring in Terra. As a consequence, fanOut of HS modules without a broadcastStream now results in an error! (Technically, this may have been an error before, but wasn't checked). This makes ready bit wiring much simpler - no ANDing of ready bits in the wiring code.

examples/convpadcrop_wide_handshake_4_1.lua

@@ -1,4 +1,5 @@
 local R = require "rigel"
+local RS = require "rigelSimple"
 local RM = require "modules"
 local types = require("types")
 local S = require("systolic")
@@ -49,8 +50,9 @@ function MAKE(T,ConvWidth,size1080p,NOSTALL)
   local RW_TYPE = types.array2d( types.uint(8), 8 ) -- simulate axi bus
   local HST = types.tuple{RW_TYPE,TAP_TYPE}
   local hsfninp_raw = R.input( R.Handshake(HST) )
-  local hsfninp = R.apply( "idx0", RM.makeHandshake(C.index(HST,0)), hsfninp_raw )
-  local hsfn_taps = R.apply( "idx1", RM.makeHandshake(C.index(HST,1)), hsfninp_raw )
+  local inp0, inp1 = RS.fanOut{input=hsfninp_raw,branches=2}
+  local hsfninp = R.apply( "idx0", RM.makeHandshake(C.index(HST,0)), inp0 )
+  local hsfn_taps = R.apply( "idx1", RM.makeHandshake(C.index(HST,1)), inp1 )
   local out = hsfninp
 
   local out = R.apply("reducerate", RM.liftHandshake(RM.changeRate(types.uint(8),1,8,T)), out )

examples/descriptor_core.lua

@@ -43,9 +43,10 @@ descriptor.descriptor = sift.siftDescriptor(types.int(8))
 
 function norm()
   local inp = R.input( R.HS( R.tuple{ R.array(R.int32,1), R.array(R.float,1) } ) )
+  local inp0, inp1 = R.fanOut{input=inp, branches=2}
 
-  local desc_sum = R.index{input=R.index{input=inp, key=1 }, key=0}
-  local desc0 = rigel.apply("d0lift",RM.makeHandshake(sift.fixedLift(R.int32)), R.index{input=R.index{input=inp,key=0 },key=0} )
+  local desc_sum = R.index{input=R.index{input=inp0, key=1 }, key=0}
+  local desc0 = rigel.apply("d0lift",RM.makeHandshake(sift.fixedLift(R.int32)), R.index{input=R.index{input=inp1,key=0 },key=0} )
 
   local desc = rigel.apply("pt",RM.packTuple{R.float,R.float},rigel.concat("PTT",{desc0,desc_sum}))
   local desc = rigel.apply("ptt",RM.makeHandshake(sift.fixedDiv(R.float)),desc)

examples/edge.lua

@@ -95,12 +95,17 @@ else
 end
 
 local inp = R.input(R.Handshake(FNTYPE))
+local inpB = R.apply("inpB", RM.broadcastStream(FNTYPE,2), inp)
+local inp0 = R.selectStream("inp0",inpB,0)
+local inp1 = R.selectStream("inp1",inpB,1)
+--local inp0, inp1 = RS.fanOut{input=inp, branches=2}
+
 local out
 local inptaps
 
 if TAPS then
-  out = R.apply( "i0", RM.makeHandshake(C.index(FNTYPE,0)), inp)
-  inptaps = R.apply("idx1",RM.makeHandshake(C.index(FNTYPE,1)), inp)
+  out = R.apply( "i0", RM.makeHandshake(C.index(FNTYPE,0)), inp0)
+  inptaps = R.apply("idx1",RM.makeHandshake(C.index(FNTYPE,1)), inp1)
   inptaps = R.apply("idx11",RM.makeHandshake(C.index(TTYPE,0)), inptaps)
   inptaps = R.apply("IIT", RM.makeHandshake(C.broadcast(types.uint(32):makeConst(),2)), inptaps)
 else

examples/harnessTerra.t

@@ -17,9 +17,17 @@ local terraWrapper = J.memoize(function(fn,inputFilename,inputType,tapType,outpu
 
     local ITYPE = types.tuple{types.null(),fixedTapInputType}
     inpSymb = R.input( R.Handshake(ITYPE) )
+
+    local inpL, inpR = inpSymb, inpSymb
+
+    if tapType~=nil then
+      local O = R.apply("bstream",RM.broadcastStream(ITYPE,2),inpSymb)
+      inpL = R.selectStream("b0",O,0)
+      inpR = R.selectStream("b1",O,1)
+    end
 
-    local inpdata = R.apply("inpdata", RM.makeHandshake(C.index(types.tuple{types.null(),fixedTapInputType},0),nil,true), inpSymb)
-    local inptaps = R.apply("inptaps", RM.makeHandshake(C.index(types.tuple{types.null(),fixedTapInputType},1)), inpSymb)
+    local inpdata = R.apply("inpdata", RM.makeHandshake(C.index(types.tuple{types.null(),fixedTapInputType},0),nil,true), inpL)
+    local inptaps = R.apply("inptaps", RM.makeHandshake(C.index(types.tuple{types.null(),fixedTapInputType},1)), inpR)
 
     out = R.apply("fread",RM.makeHandshake(RM.freadSeq(inputFilename,R.extractData(inputType)),nil,true),inpdata)
     local hsfninp = out

examples/lk_tr_core.lua

@@ -152,9 +152,21 @@ function makeA( T, dType, window, bits )
 
   local stt = types.array2d(dType, window*T, window)
   local inpt = types.tuple{ stt, stt }
-  local inp = R.input( R.Handshake(inpt) )
-  local Fdx = R.apply("fdx", RM.makeHandshake(C.index(inpt,0)), inp)
-  local Fdy = R.apply("fdy", RM.makeHandshake(C.index(inpt,1)), inp)
+  local input = R.input( R.Handshake(inpt) )
+  local inputB = R.apply("inpB", RM.broadcastStream(inpt,2),input)
+  local input0 = R.selectStream("input0",inputB,0)
+  local input1 = R.selectStream("input1",inputB,1)
+
+  local Fdx = R.apply("fdx", RM.makeHandshake(C.index(inpt,0)), input0)
+  local FdxB = R.apply("fdxB", RM.broadcastStream(stt,2),Fdx)
+  local Fdx0 = R.selectStream("fdx0",FdxB,0)
+  local Fdx1 = R.selectStream("fdx1",FdxB,1)
+
+  local Fdy = R.apply("fdy", RM.makeHandshake(C.index(inpt,1)), input1)
+  --local Fdy0, Fdy1 = RS.fanOut{input=Fdy, branches=2}
+  local FdyB = R.apply("fdyB", RM.broadcastStream(stt,2),Fdy)
+  local Fdy0 = R.selectStream("fdy0",FdyB,0)
+  local Fdy1 = R.selectStream("fdy1",FdyB,1)
 
   local partial = makePartial( dType, dType, bits.Apartial[1], bits.Apartial[2] )
   bits.Apartial[3] = partial[3]
@@ -165,19 +177,19 @@ function makeA( T, dType, window, bits )
   local rsumfn = makeSumReduce(partial_type,false)
   local rsumAsyncfn = makeSumReduce(partial_type,true)
 
-  local inp0 = R.apply("inp0", C.SoAtoAoSHandshake(window*T,window,{dType,dType}), R.concat("o0",{Fdx,Fdx}) )
+  local inp0 = R.apply("inp0", C.SoAtoAoSHandshake(window*T,window,{dType,dType}), R.concat("o0",{Fdx0,Fdx0}) )
   local out0 = R.apply("out0", RM.makeHandshake(RM.map(partialfn, window*T, window)), inp0 )
   local out0 = R.apply("out0red", RM.makeHandshake(RM.reduce(rsumfn, window*T, window)), out0 )
   local out0 = R.apply("out0redseq", RM.liftHandshake(RM.liftDecimate(RM.reduceSeq( rsumAsyncfn, T ))), out0 )
 
-  local inp1 = R.apply("inp1", C.SoAtoAoSHandshake(window*T,window,{dType,dType}), R.concat("o1",{Fdx,Fdy}) )
+  local inp1 = R.apply("inp1", C.SoAtoAoSHandshake(window*T,window,{dType,dType}), R.concat("o1",{Fdx1,Fdy0}) )
   local out1 = R.apply("out1", RM.makeHandshake(RM.map(partialfn, window*T, window)), inp1 )
   local out1 = R.apply("out1red", RM.makeHandshake(RM.reduce(rsumfn, window*T, window)), out1 )
   local out1 = R.apply("out1redseq", RM.liftHandshake(RM.liftDecimate(RM.reduceSeq( rsumAsyncfn, T ))), out1 )
 
   local out2 = out1
 
-  local inp3 = R.apply("inp3", C.SoAtoAoSHandshake(window*T,window,{dType,dType}), R.concat("o3",{Fdy,Fdy}) )
+  local inp3 = R.apply("inp3", C.SoAtoAoSHandshake(window*T,window,{dType,dType}), R.concat("o3",{Fdy1,Fdy1}) )
   local out3 = R.apply("out3", RM.makeHandshake(RM.map(partialfn, window*T, window)), inp3 )
   local out3 = R.apply("out3red", RM.makeHandshake(RM.reduce(rsumfn, window*T, window)), out3 )
   local out3 = R.apply("out3redseq", RM.liftHandshake(RM.liftDecimate(RM.reduceSeq( rsumAsyncfn, T ))), out3 )
@@ -186,7 +198,7 @@ function makeA( T, dType, window, bits )
   out = R.apply("PT",RM.packTuple({partial_type,partial_type,partial_type,partial_type}),out)
   out = R.apply("PTC",RM.makeHandshake(C.tupleToArray(partial_type,4)),out)
 
-  return RM.lambda("A", inp, out ), partial_type
+  return RM.lambda("A", input, out ), partial_type
 end
 
 function minus(ty)
@@ -232,7 +244,13 @@ function makeB( T, dtype, window, bits )
   --print("GMF type", gmf_type)
   cost = cost + gmf_cost*window*window
   gmf = R.apply("SSM",RM.makeHandshake(RM.map(m,window*T,window)), gmf)
-  gmf = FIFO( fifos, statements, types.array2d(gmf_type,window*T,window), gmf)
+  local GMFT = types.array2d(gmf_type,window*T,window)
+  gmf = FIFO( fifos, statements, GMFT, gmf)
+
+  local gmfB = R.apply("inpB", RM.broadcastStream(GMFT,2),gmf)
+  local gmf0 = R.selectStream("inp0",gmfB,0)
+  local gmf1 = R.selectStream("inp1",gmfB,1)
+
   ---------
 
   local partial = makePartial( dtype, gmf_type, bits.Bpartial[1], bits.Bpartial[2] )
@@ -244,13 +262,13 @@ function makeB( T, dtype, window, bits )
   local rsumfn = makeSumReduce(partial_type,false)
   local rsumAsyncfn = makeSumReduce(partial_type,true)
 
-  local out_0 = R.concat("o0tup",{Fdx, gmf})
+  local out_0 = R.concat("o0tup",{Fdx, gmf0})
   local out_0 = R.apply("o0P", C.SoAtoAoSHandshake(window*T,window,{dtype, gmf_type}), out_0)
   local out_0 = R.apply("o0", RM.makeHandshake(RM.map(partialfn, window*T, window)), out_0)
   local out_0 = R.apply("out0red", RM.makeHandshake(RM.reduce(rsumfn, window*T, window)), out_0 )
   local out_0 = R.apply("out0redseq", RM.liftHandshake(RM.liftDecimate(RM.reduceSeq(rsumAsyncfn, T))), out_0 )
 
-  local out_1 = R.concat("o1tup",{Fdy, gmf})
+  local out_1 = R.concat("o1tup",{Fdy, gmf1})
   local out_1 = R.apply("o1P", C.SoAtoAoSHandshake(window*T,window,{dtype,gmf_type}), out_1)
   local out_1 = R.apply("o1", RM.makeHandshake(RM.map(partialfn, window*T, window)), out_1)
   local out_1 = R.apply("out1red", RM.makeHandshake(RM.reduce(rsumfn, window*T, window)), out_1 )
@@ -318,8 +336,13 @@ function makeLK( internalT, internalW, internalH, window, bits )
 
   local INPTYPE = types.array2d(types.uint(8),2)
   local inp = R.input(R.Handshake(INPTYPE))
-  local frame0 = R.apply("f0", RM.makeHandshake(C.index(INPTYPE, 0 )), inp)
-  local frame1 = R.apply("f1", RM.makeHandshake(C.index(INPTYPE, 1 )), inp)
+
+  local inpB = R.apply("inpB", RM.broadcastStream(INPTYPE,2),inp)
+  local inp0 = R.selectStream("inp0",inpB,0)
+  local inp1 = R.selectStream("inp1",inpB,1)
+
+  local frame0 = R.apply("f0", RM.makeHandshake(C.index(INPTYPE, 0 )), inp0)
+  local frame1 = R.apply("f1", RM.makeHandshake(C.index(INPTYPE, 1 )), inp1)
 
 
   local frame0_arr = R.apply("f0_arr", RM.makeHandshake(C.arrayop(types.uint(8),1)), frame0)

examples/pyramid_core.lua

@@ -1,5 +1,6 @@
 local R = require "rigel"
 local RM = require "modules"
+local RS = require "rigelSimple"
 local types = require "types"
 local C = require "examplescommon"
 local P = {}
@@ -78,13 +79,14 @@ function P.pyramidIterTaps(i,doDownsample,internalT,W,H,ConvWidth,nofifo,DUMB_DO
   local TAP_TYPE = types.array2d( A, ConvWidth, ConvWidth):makeConst()
   local INP_TYPE = types.tuple{DATA_TYPE,TAP_TYPE}
   local inp = R.input( R.Handshake(INP_TYPE) )
+  local inpL, inpR = RS.fanOut{input=inp, branches=2}
   local borderValue = 0
 
   local fifos = {}
   local statements = {}
 
-  local out = R.apply("idx0",RM.makeHandshake(C.index(INP_TYPE,0)),inp)
-  local tapinp = R.apply("idx1",RM.makeHandshake(C.index(INP_TYPE,1)),inp)
+  local out = R.apply("idx0",RM.makeHandshake(C.index(INP_TYPE,0)),inpL)
+  local tapinp = R.apply("idx1",RM.makeHandshake(C.index(INP_TYPE,1)),inpR)
 
   local PadWidth = ConvWidth/2
   if PadWidth < internalT then PadWidth = internalT end

examples/pyramid_taps_1.lua

@@ -1,10 +1,12 @@
 local R = require "rigel"
 local RM = require "modules"
+local RS = require "rigelSimple"
 local types = require("types")
 local harness = require "harness"
 local C = require "examplescommon"
 local P = require "pyramid_core"
 local SDFRate = require "sdfrate"
+local J = require "common"
 
 internalT = 4 -- internal throughput
 outputT = 8
@@ -36,8 +38,8 @@ local DATA_TYPE = types.array2d(A,8)
 local HST = types.tuple{DATA_TYPE,TAP_TYPE}
 
 local inp = R.input( R.Handshake(HST) )
-local tapinp =  R.apply("idx1",RM.makeHandshake(C.index(HST,1)),inp)
-local out = R.apply("idx0",RM.makeHandshake(C.index(HST,0)),inp)
+local inpList = J.pack(RS.fanOut{input=inp,branches=1+TARGET_DEPTH})
+local out = R.apply("idx0",RM.makeHandshake(C.index(HST,0)),inpList[1])
 
 if internalT<8 then
   out = R.apply("CRtop",RM.liftHandshake(RM.changeRate(A,1,8,internalT)), out)
@@ -57,14 +59,13 @@ local statements = {}
 
 
 for depth=1,TARGET_DEPTH do
-  --print("DODEPTH",depth)
-  --local PI = P.pyramidIter(depth,depth>1,internalT,curW,curH,ConvWidth)
+
   local PI = P.pyramidIterTaps( depth, depth>1, internalT, curW, curH, ConvWidth, NOFIFO, true )
-  --print("PI",PI.inputType,PI.outputType)
-  --print(PI.sdfInput[1][1],PI.sdfInput[1][2])
-  --print(PI.sdfOutput[1][1],PI.sdfOutput[1][2])
 
-  local piinp = R.apply("CPI"..depth, RM.packTuple({types.array2d(A,internalT),TAP_TYPE}), R.concat("CONVPIPEINP"..depth,{out,tapinp}))
+  local tapinp =  R.apply("idx1_"..tostring(depth),RM.makeHandshake(C.index(HST,1)),inpList[1+depth])
+
+  local CCT = R.concat("CONVPIPEINP"..depth,{out,tapinp})
+  local piinp = R.apply("CPI"..depth, RM.packTuple({types.array2d(A,internalT),TAP_TYPE}), CCT)
   out = R.apply("p"..depth, PI, piinp)
 
   local thisW = inputW*inputH/math.pow(4,depth-1)

examples/pyramid_tr_1.lua

@@ -1,10 +1,12 @@
 local R = require "rigel"
 local RM = require "modules"
+local RS = require "rigelSimple"
 local types = require("types")
 local harness = require "harness"
 local C = require "examplescommon"
 local P = require "pyramid_core"
 local SDFRate = require "sdfrate"
+local J = require "common"
 
 T = 4 -- throughput
 outputT = 8
@@ -35,14 +37,24 @@ local DATA_TYPE = types.array2d(A,8)
 local HST = types.tuple{DATA_TYPE,TAP_TYPE}
 
 local inp = R.input( R.Handshake(HST) )
+local inp0, inp1 = RS.fanOut{input=inp,branches=2}
 
-local out = R.apply("idx0", RM.makeHandshake( C.index(HST,0)),inp)
+local out = R.apply("idx0", RM.makeHandshake( C.index(HST,0)),inp0)
 if T<8 then
   out = R.apply("CRtop", RM.liftHandshake( RM.changeRate(A,1,8,T)), out)
 end
 
-local tapinp =  R.apply("idx1", RM.makeHandshake( C.index(HST,1)),inp)
+local totT = 0
+local totTT = T
+for depth=1,TARGET_DEPTH do
+  if totTT>1 then  totT=totT+1 end
+  if depth>1 then totTT=totTT/4; end
+end
 
+local tapinp =  R.apply("idx1", RM.makeHandshake( C.index(HST,1)),inp1)
+--print("TOT",totT)
+local tapinpL = J.pack(RS.fanOut{input=tapinp, branches=totT})
+
 curT = T
 --vecT = T
 curW = inputW
@@ -61,7 +73,7 @@ for depth=1,TARGET_DEPTH do
 
   if curT>1 then
     PI = P.pyramidIterTaps(depth,depth>1,curT,curW,curH,ConvWidth,NOFIFO,false)
-    local piinp = R.apply("CPI"..depth, RM.packTuple({types.array2d(A,curT),TAP_TYPE}), R.concat("CONVPIPEINP"..depth,{out,tapinp}))
+    local piinp = R.apply("CPI"..depth, RM.packTuple({types.array2d(A,curT),TAP_TYPE}), R.concat("CONVPIPEINP"..depth,{out,tapinpL[depth]}))
     out = R.apply("p"..depth, PI, piinp)
   else
     PI = P.pyramidIterTR(depth,curT,curW,curH,ConvWidth,NOFIFO)

examples/sift_core.lua

@@ -1,4 +1,5 @@
 local R = require "rigel"
+local RS = require "rigelSimple"
 local RM = require "modules"
 local types = require("types")
 local S = require("systolic")
@@ -258,10 +259,11 @@ local function siftKernel(dxdyType)
   local inp_pos = C.fifo( fifos, statements, ITYPE, R.selectStream("i0",inp_broad,0), 1, "p0", true)
   local pos = R.apply("p",RM.makeHandshake(C.index(ITYPE,1)), inp_pos)
   local pos = C.fifo( fifos, statements, PTYPE, pos, 1024, "posfifo")
+  local posL, posR = RS.fanOut{input=pos,branches=2}
 --  local pos = C.fifo( fifos, statements, PTYPE, pos, 1024, "p0")
-  local posX = R.apply("px",RM.makeHandshake(C.index(PTYPE,0)),pos)
+  local posX = R.apply("px",RM.makeHandshake(C.index(PTYPE,0)),posL)
   local posX = C.fifo( fifos, statements, posType, posX, 1024, "pxfifo" )
-  local posY = R.apply("py",RM.makeHandshake(C.index(PTYPE,1)),pos)
+  local posY = R.apply("py",RM.makeHandshake(C.index(PTYPE,1)),posR)
   local posY = C.fifo( fifos, statements, posType, posY, 1024, "pyfifo" )
 
   local inp_dxdy = C.fifo( fifos, statements, ITYPE, R.selectStream("i1",inp_broad,1), 1, "p1", true)

examples/sift_core_hw.lua

@@ -300,11 +300,12 @@ function sift.siftKernel(dxdyType)
   local inp_pos = C.fifo( fifos, statements, ITYPE, R.selectStream("i0",inp_broad,0), 1, "p0", true) -- fifo size can also be 1 (tested in SW)
   local pos = R.apply("p",RM.makeHandshake(C.index(ITYPE,1)), inp_pos)
   local pos = C.fifo( fifos, statements, PTYPE, pos, 1024, "posfifo")
+  local pos0, pos1 = RS.fanOut{input=pos,branches=2}
 
 --  local pos = C.fifo( fifos, statements, PTYPE, pos, 1024, "p0")
-  local posX = R.apply("px",RM.makeHandshake(C.index(PTYPE,0)),pos)
+  local posX = R.apply("px",RM.makeHandshake(C.index(PTYPE,0)),pos0)
   local posX = C.fifo( fifos, statements, posType, posX, 1024, "pxfifo" )
-  local posY = R.apply("py",RM.makeHandshake(C.index(PTYPE,1)),pos)
+  local posY = R.apply("py",RM.makeHandshake(C.index(PTYPE,1)),pos1)
   local posY = C.fifo( fifos, statements, posType, posY, 1024, "pyfifo" )
 
   local inp_dxdy = C.fifo( fifos, statements, ITYPE, R.selectStream("i1",inp_broad,1), 1, "p1", true) -- fifo size can also be 1 (tested in SW)

rigel.lua

@@ -117,7 +117,9 @@ function darkroom.hasReady(a)
   elseif darkroom.isRV(a) then return true
   elseif a:isTuple() and darkroom.isHandshake(a.list[1]) then
     return true
-  elseif darkroom.isHandshakeArray(a) then
+  elseif a:isArray() and darkroom.isHandshake(a:arrayOver()) then
+    return true
+  elseif darkroom.isHandshakeArray(a) or darkroom.isHandshakeTmuxed(a) then
     return true
   elseif darkroom.isBasic(a) or darkroom.isV(a) then
     return false
@@ -150,7 +152,9 @@ function darkroom.extractValid(a)
 end
 
 function darkroom.streamCount(A)
-  if darkroom.isHandshake(A) then
+  if darkroom.isBasic(A) or darkroom.isV(A) or darkroom.isRV(A) then
+    return 0
+  elseif darkroom.isHandshake(A) then
     return 1
   elseif A:isArray() and darkroom.isHandshake(A:arrayOver()) then
     return A:channels()
@@ -159,7 +163,7 @@ function darkroom.streamCount(A)
   elseif darkroom.isHandshakeTmuxed(A) or darkroom.isHandshakeArray(A) then
     return A.params.N
   else
-    return 0
+    err(false, "NYI streamCount "..tostring(A))
   end
 end
 

src/fixed.lua

@@ -29,7 +29,7 @@ function fixed.isFixedType(ty)
 end
 
 function fixed.expectFixed(ty)
-  err( fixed.isFixedType(ty), "Expected fixed point type")
+  err( fixed.isFixedType(ty), "Expected fixed point type but was "..tostring(ty))
 end
 
 function fixed.extract(ty)