From 953c53d065127ec688c07b0ac45362551328ce75 Mon Sep 17 00:00:00 2001
From: David Harris <David_Harris@hmc.edu>
Date: Fri, 10 Nov 2023 09:11:15 -0800
Subject: [PATCH] fdivsqrt parameter cleanup

---
 config/shared/config-shared.vh       | 13 ++++++-------
 config/shared/parameter-defs.vh      |  3 +--
 src/cvw.sv                           |  1 -
 src/fpu/fdivsqrt/fdivsqrtpostproc.sv |  3 ++-
 src/fpu/fdivsqrt/fdivsqrtpreproc.sv  | 18 +++++++++---------
 5 files changed, 18 insertions(+), 20 deletions(-)

diff --git a/config/shared/config-shared.vh b/config/shared/config-shared.vh
index acc7996cb..17b1ede83 100644
--- a/config/shared/config-shared.vh
+++ b/config/shared/config-shared.vh
@@ -94,15 +94,14 @@ localparam FMT2  = ((F_SUPPORTED & (LEN1 != S_LEN)) ? 2'd0    : 2'd2);
 localparam BIAS2 = ((F_SUPPORTED & (LEN1 != S_LEN)) ? S_BIAS : H_BIAS);
 
 // division constants
-localparam DIVN        = ((NF+2<XLEN) & IDIV_ON_FPU) ? XLEN : NF+2; // standard length of input
-localparam LOGR        = $clog2(RADIX);           // r = log(R)
-localparam RK          = LOGR*DIVCOPIES;         // r*k used for intdiv preproc
-localparam LOGRK       = $clog2(RK);               // log2(r*k)
-localparam FPDUR       = (DIVN+1)/RK + 1 + (RADIX/4);
+localparam DIVN        = ((NF+2<XLEN) & IDIV_ON_FPU) ? XLEN : NF+2; // standard length of input: max(XLEN, NF+2)
+localparam LOGR        = $clog2(RADIX);                             // r = log(R)
+localparam RK          = LOGR*DIVCOPIES;                            // r*k bits per cycle generated
+localparam LOGRK       = $clog2(RK);                                // log2(r*k)
+localparam FPDUR       = (DIVN+1)/RK + 1 + (RADIX/4);               // 
 localparam DURLEN      = $clog2(FPDUR+1);
 localparam DIVb        = FPDUR*RK - 1; // canonical fdiv size (b)
-localparam DIVBLEN     = $clog2(DIVb+1)-1;
-localparam DIVa        = DIVb+1-XLEN; // used for idiv on fpu: Shift residual right by b - (XLEN-1) to put remainder in lsbs of integer result
+localparam DIVBLEN     = $clog2(DIVb+2)-1;
 
 // largest length in IEU/FPU
 localparam CVTLEN = ((NF<XLEN) ? (XLEN) : (NF));  // max(XLEN, NF)
diff --git a/config/shared/parameter-defs.vh b/config/shared/parameter-defs.vh
index d04b35e56..0c377c02d 100644
--- a/config/shared/parameter-defs.vh
+++ b/config/shared/parameter-defs.vh
@@ -184,6 +184,5 @@ localparam cvw_t P = '{
   FPDUR       : FPDUR,
   DURLEN      : DURLEN,
   DIVb        : DIVb,
-  DIVBLEN     : DIVBLEN,
-  DIVa        : DIVa
+  DIVBLEN     : DIVBLEN
 };
diff --git a/src/cvw.sv b/src/cvw.sv
index 4cbf67b28..02105823e 100644
--- a/src/cvw.sv
+++ b/src/cvw.sv
@@ -277,7 +277,6 @@ typedef struct packed {
   int DURLEN     ;
   int DIVb       ;
   int DIVBLEN    ;
-  int DIVa       ;
 
 } cvw_t;
 
diff --git a/src/fpu/fdivsqrt/fdivsqrtpostproc.sv b/src/fpu/fdivsqrt/fdivsqrtpostproc.sv
index c3c558902..9f887d4ab 100644
--- a/src/fpu/fdivsqrt/fdivsqrtpostproc.sv
+++ b/src/fpu/fdivsqrt/fdivsqrtpostproc.sv
@@ -110,7 +110,8 @@ module fdivsqrtpostproc import cvw::*;  #(parameter cvw_t P) (
     mux2 #(P.DIVb+4) quotresmux(UnsignedQuotM, -UnsignedQuotM, NegQuotM, NormQuotM);
 
     // Select quotient or remainder and do normalization shift
-    mux2 #(P.DIVBLEN+1) normshiftmux(((P.DIVBLEN+1)'(P.DIVb) - (nM * (P.DIVBLEN+1)'(P.LOGR))), (mM + (P.DIVBLEN+1)'(P.DIVa)), RemOpM, NormShiftM);
+    localparam DIVa        = (P.DIVb+1-P.XLEN); // used for idiv on fpu: Shift residual right by b - (XLEN-1) to put remainder in lsbs of integer result
+    mux2 #(P.DIVBLEN+1) normshiftmux(((P.DIVBLEN+1)'(P.DIVb) - (nM * (P.DIVBLEN+1)'(P.LOGR))), (mM + (P.DIVBLEN+1)'(DIVa)), RemOpM, NormShiftM);
     mux2 #(P.DIVb+4)    presresultmux(NormQuotM, NormRemM, RemOpM, PreResultM);
     assign PreIntResultM = $signed(PreResultM >>> NormShiftM); 
 
diff --git a/src/fpu/fdivsqrt/fdivsqrtpreproc.sv b/src/fpu/fdivsqrt/fdivsqrtpreproc.sv
index 8f3c477c4..0e716ac20 100644
--- a/src/fpu/fdivsqrt/fdivsqrtpreproc.sv
+++ b/src/fpu/fdivsqrt/fdivsqrtpreproc.sv
@@ -48,7 +48,7 @@ module fdivsqrtpreproc import cvw::*;  #(parameter cvw_t P) (
   output logic [P.XLEN-1:0]    AM
 );
 
-  logic [P.DIVb-1:0]           Xfract, Dfract;
+  logic [P.DIVb:0]             Xfract, Dfract;
   logic [P.DIVb:0]             PreSqrtX;
   logic [P.DIVb+3:0]           DivX, DivXShifted, SqrtX, PreShiftX; // Variations of dividend, to be muxed
   logic [P.NE+1:0]             QeE;                                 // Quotient Exponent (FP only)
@@ -103,12 +103,12 @@ module fdivsqrtpreproc import cvw::*;  #(parameter cvw_t P) (
   //////////////////////////////////////////////////////
 
   // count leading zeros for Subnorm FP and to normalize integer inputs
-  lzc #(P.DIVb) lzcX (IFX[P.DIVb:1], ell);
-  lzc #(P.DIVb) lzcY (IFD[P.DIVb:1], mE);
+  lzc #(P.DIVb+1) lzcX (IFX, ell);
+  lzc #(P.DIVb+1) lzcY (IFD, mE);
 
   // Normalization shift: shift off leading one
-  assign Xfract = (IFX[P.DIVb:1] << ell) << 1;
-  assign Dfract = (IFD[P.DIVb:1] << mE)  << 1; 
+  assign Xfract = (IFX << ell);
+  assign Dfract = (IFD << mE); 
 
   //////////////////////////////////////////////////////
   // Integer Right Shift to digit boundary
@@ -158,10 +158,10 @@ module fdivsqrtpreproc import cvw::*;  #(parameter cvw_t P) (
   //  it comes out in the wash and gives the right answer.  Investigate later if possible.
   //////////////////////////////////////////////////////
 
-  assign DivX = {3'b000, ~NumerZeroE, Xfract};
+  assign DivX = {3'b000, Xfract};
 
   // Sqrt is initialized on step one as R(X-1), so depends on Radix
-  mux2 #(P.DIVb+1) sqrtxmux({~XZeroE, Xfract}, {1'b0, ~XZeroE, Xfract[P.DIVb-1:1]}, (Xe[0] ^ ell[0]), PreSqrtX);
+  mux2 #(P.DIVb+1) sqrtxmux(Xfract, {1'b0, Xfract[P.DIVb:1]}, (Xe[0] ^ ell[0]), PreSqrtX);
   if (P.RADIX == 2) assign SqrtX = {3'b111, PreSqrtX};
   else              assign SqrtX = {2'b11, PreSqrtX, 1'b0};
   mux2 #(P.DIVb+4) prexmux(DivX, SqrtX, SqrtE, PreShiftX);
@@ -176,8 +176,8 @@ module fdivsqrtpreproc import cvw::*;  #(parameter cvw_t P) (
     assign X = PreShiftX;
   end
 
-   // Divisior register
-  flopen #(P.DIVb+4) dreg(clk, IFDivStartE, {4'b0001, Dfract}, D);
+  // Divisior register
+  flopen #(P.DIVb+4) dreg(clk, IFDivStartE, {3'b000, Dfract}, D);
  
   // Floating-point exponent
   fdivsqrtexpcalc #(P) expcalc(.Fmt(FmtE), .Xe, .Ye, .Sqrt(SqrtE), .XZero(XZeroE), .ell, .m(mE), .Qe(QeE));