OpenCores
URL https://opencores.org/ocsvn/ft816float/ft816float/trunk

Subversion Repositories ft816float

Compare Revisions

  • This comparison shows the changes necessary to convert path 
    /
    from Rev 52 to Rev 53
    ↔ Reverse comparison
  •

Rev 52 → Rev 53

/ft816float/trunk/rtl/verilog2/BCDMath.v
369,6 → 369,78
 
endmodule
 
module BCDMul4(a,b,o);
input [15:0] a;
input [15:0] b;
output [31:0] o;
 
wire [15:0] p1,p2,p3,p4;
wire [31:0] s1;
 
BCDMul2 u1 (a[7:0],b[7:0],p1);
BCDMul2 u2 (a[15:8],b[7:0],p2);
BCDMul2 u3 (a[7:0],b[15:8],p3);
BCDMul2 u4 (a[15:8],b[15:8],p4);
 
BCDAddN #(.N(8)) u5 (1'b0,{p4,p1},{8'h0,p2,8'h0},s1);
BCDAddN #(.N(8)) u6 (1'b0,s1,{8'h0,p3,8'h0},o);
 
endmodule
 
module BCDMul8(a,b,o);
input [31:0] a;
input [31:0] b;
output [63:0] o;
 
wire [31:0] p1,p2,p3,p4;
wire [63:0] s1;
 
BCDMul4 u1 (a[15:0],b[15:0],p1);
BCDMul4 u2 (a[31:16],b[15:0],p2);
BCDMul4 u3 (a[15:0],b[31:16],p3);
BCDMul4 u4 (a[31:16],b[31:16],p4);
 
BCDAddN #(.N(16)) u5 (1'b0,{p4,p1},{16'h0,p2,16'h0},s1);
BCDAddN #(.N(16)) u6 (1'b0,s1,{16'h0,p3,16'h0},o);
 
endmodule
 
module BCDMul16(a,b,o);
input [63:0] a;
input [63:0] b;
output [127:0] o;
 
wire [63:0] p1,p2,p3,p4;
wire [127:0] s1;
 
BCDMul8 u1 (a[31:0],b[31:0],p1);
BCDMul8 u2 (a[63:32],b[31:0],p2);
BCDMul8 u3 (a[31:0],b[63:32],p3);
BCDMul8 u4 (a[63:32],b[63:32],p4);
 
BCDAddN #(.N(32)) u5 (1'b0,{p4,p1},{32'h0,p2,32'h0},s1);
BCDAddN #(.N(32)) u6 (1'b0,s1,{32'h0,p3,32'h0},o);
 
endmodule
 
module BCDMul32(a,b,o);
input [127:0] a;
input [127:0] b;
output [255:0] o;
 
wire [127:0] p1,p2,p3,p4;
wire [255:0] s1;
 
BCDMul16 u1 (a[63:0],b[63:0],p1);
BCDMul16 u2 (a[127:64],b[63:0],p2);
BCDMul16 u3 (a[63:0],b[127:64],p3);
BCDMul16 u4 (a[127:64],b[127:64],p4);
 
BCDAddN #(.N(64)) u5 (1'b0,{p4,p1},{64'h0,p2,64'h0},s1);
BCDAddN #(.N(64)) u6 (1'b0,s1,{64'h0,p3,64'h0},o);
 
endmodule
 
module BCDMul_tb();
 
wire [15:0] o1,o2,o3,o4;
/ft816float/trunk/rtl/verilog2/DFPDecompose.sv
40,7 → 40,7
output sgn;
output sx;
output [15:0] exp;
output [95:0] sig;
output [107:0] sig;
output xz;
output vz;
output inf;
/ft816float/trunk/rtl/verilog2/DFPMultiply.sv
0,0 → 1,239
// ============================================================================
// __
// \\__/ o\ (C) 2006-2020 Robert Finch, Waterloo
// \ __ / All rights reserved.
// \/_// robfinch<remove>@finitron.ca
// ||
//
// DFPMultiply.v
// - decimal floating point multiplier
// - can issue every clock cycle
// - parameterized width
//
//
// BSD 3-Clause License
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
//
// 1. Redistributions of source code must retain the above copyright notice, this
// list of conditions and the following disclaimer.
//
// 2. Redistributions in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// 3. Neither the name of the copyright holder nor the names of its
// contributors may be used to endorse or promote products derived from
// this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
// DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
// FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
// SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
// OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
//
//
// Floating Point Multiplier
//
// This multiplier handles denormalized numbers.
// The output format is of an internal expanded representation
// in preparation to be fed into a normalization unit, then
// rounding. Basically, it's the same as the regular format
// except the mantissa is doubled in size, the leading two
// bits of which are assumed to be whole bits.
//
//
// Floating Point Multiplier
//
// Properties:
// +-inf * +-inf = -+inf (this is handled by exOver)
// +-inf * 0 = QNaN
//
// ============================================================================
 
import fp::*;
 
module DFPMultiply(clk, ce, a, b, o, sign_exe, inf, overflow, underflow);
input clk;
input ce;
input [127:0] a, b;
output [243:0] o;
output sign_exe;
output inf;
output overflow;
output underflow;
parameter DELAY =
(FPWID == 128 ? 17 :
FPWID == 80 ? 17 :
FPWID == 64 ? 13 :
FPWID == 40 ? 8 :
FPWID == 32 ? 2 :
FPWID == 16 ? 2 : 2);
 
reg [15:0] xo1; // extra bit for sign
reg [215:0] mo1;
 
// constants
wire [15:0] infXp = 16'h9999; // infinite / NaN - all ones
// The following is the value for an exponent of zero, with the offset
// eg. 8'h7f for eight bit exponent, 11'h7ff for eleven bit exponent, etc.
// The following is a template for a quiet nan. (MSB=1)
wire [107:0] qNaN = {4'h1,{104{1'b0}}};
 
// variables
reg [215:0] sig1;
wire [15:0] ex2;
 
// Decompose the operands
wire sa, sb; // sign bit
wire [15:0] xa, xb; // exponent bits
wire sxa, sxb;
wire [107:0] siga, sigb;
wire a_dn, b_dn; // a/b is denormalized
wire aNan, bNan, aNan1, bNan1;
wire az, bz;
wire aInf, bInf, aInf1, bInf1;
 
 
// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
// Clock #1
// - decode the input operands
// - derive basic information
// - calculate exponent
// - calculate fraction
// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 
// -----------------------------------------------------------
// First clock
// -----------------------------------------------------------
 
reg under, over;
reg [15:0] sum_ex;
reg sx0;
 
DFPDecompose u1a (.i(a), .sgn(sa), .sx(sxa), .exp(xa), .sig(siga), .xz(a_dn), .vz(az), .inf(aInf), .nan(aNan) );
DFPDecompose u1b (.i(b), .sgn(sb), .sx(sxb), .exp(xb), .sig(sigb), .xz(b_dn), .vz(bz), .inf(bInf), .nan(bNan) );
 
// Compute the sum of the exponents.
// Exponents are sign-magnitude.
wire [15:0] xapxb, xamxb, xbmxa;
wire xapxbc, xamxbc, xbmxac;
BCDAddN #(.N(4)) u1c (.ci(1'b0), .a(xa), .b(xb), .o(xapxb), .co(xapxbc));
BCDSubN #(.N(4)) u1d (.ci(1'b0), .a(xa), .b(xb), .o(xamxb), .co(xamxbc));
BCDSubN #(.N(4)) u1e (.ci(1'b0), .a(xb), .b(xa), .o(xbmxa), .co(xbmxac));
 
always @*
case({sxa,sxb})
2'b11: begin sum_ex <= xapxb; over <= xapxbc; under <= 1'b0; sx0 <= sxa; end
2'b01: begin sum_ex <= xbmxa; over <= 1'b0; under <= 1'b0; sx0 <= ~xbmxac; end
2'b10: begin sum_ex <= xamxb; over <= 1'b0; under <= 1'b0; sx0 <= ~xamxbc; end
2'b00: begin sum_ex <= xapxb; over <= 1'b0; under <= xapxbc; sx0 <= sxa; end
endcase
 
wire [255:0] sigoo;
BCDMul32 u1f (.a({20'h0,siga}),.b({20'h0,sigb}),.o(sigoo));
 
always @(posedge clk)
if (ce) sig1 <= sigoo[215:0];
 
// Status
wire under1, over1;
 
delay #(.WID(16),.DEP(DELAY)) u3 (.clk(clk), .ce(ce), .i(sum_ex), .o(ex2) );
delay #(.WID(1),.DEP(DELAY)) u2a (.clk(clk), .ce(ce), .i(aInf), .o(aInf1) );
delay #(.WID(1),.DEP(DELAY)) u2b (.clk(clk), .ce(ce), .i(bInf), .o(bInf1) );
delay #(.WID(1),.DEP(DELAY)) u6 (.clk(clk), .ce(ce), .i(under), .o(under1) );
delay #(.WID(1),.DEP(DELAY)) u7 (.clk(clk), .ce(ce), .i(over), .o(over1) );
 
// determine when a NaN is output
wire qNaNOut;
wire [127:0] a1,b1;
delay #(.WID(1),.DEP(DELAY)) u5 (.clk(clk), .ce(ce), .i((aInf&bz)|(bInf&az)), .o(qNaNOut) );
delay #(.WID(1),.DEP(DELAY)) u14 (.clk(clk), .ce(ce), .i(aNan), .o(aNan1) );
delay #(.WID(1),.DEP(DELAY)) u15 (.clk(clk), .ce(ce), .i(bNan), .o(bNan1) );
delay #(.WID(128),.DEP(DELAY)) u16 (.clk(clk), .ce(ce), .i(a), .o(a1) );
delay #(.WID(128),.DEP(DELAY)) u17 (.clk(clk), .ce(ce), .i(b), .o(b1) );
 
// -----------------------------------------------------------
// Second clock
// - correct xponent and mantissa for exceptional conditions
// -----------------------------------------------------------
 
wire so1, sx1;
reg [3:0] st;
 
delay #(.WID(1),.DEP(1)) u8 (.clk(clk), .ce(ce), .i(~(sa ^ sb)), .o(so1) );// two clock delay!
delay #(.WID(1),.DEP(1)) u9 (.clk(clk), .ce(ce), .i(sx0), .o(sx1) );// two clock delay!
 
always @(posedge clk)
if (ce)
casez({qNaNOut|aNan1|bNan1,aInf1,bInf1,over1,under1})
5'b1????: xo1 = infXp; // qNaN - infinity * zero
5'b01???: xo1 = infXp; // 'a' infinite
5'b001??: xo1 = infXp; // 'b' infinite
5'b0001?: xo1 = infXp; // result overflow
5'b00001: xo1 = ex2[15:0];//0; // underflow
default: xo1 = ex2[15:0]; // situation normal
endcase
 
// Force mantissa to zero when underflow or zero exponent when not supporting denormals.
always @(posedge clk)
if (ce)
casez({aNan1,bNan1,qNaNOut,aInf1,bInf1,over1|under1})
6'b1?????: mo1 = {4'h1,a1[103:0],108'b0};
6'b01????: mo1 = {4'h1,b1[103:0],108'b0};
6'b001???: mo1 = {4'h1,qNaN|3'd4,108'b0}; // multiply inf * zero
6'b0001??: mo1 = 0; // mul inf's
6'b00001?: mo1 = 0; // mul inf's
6'b000001: mo1 = 0; // mul overflow
default: mo1 = sig1;
endcase
 
always @(posedge clk)
if (ce) begin
st[3] <= aNan1|bNan1;
st[2] <= so1;
st[1] <= aInf|bInf|over;
st[0] <= sx1;
end
 
delay #(.WID(1),.DEP(DELAY+1)) u10 (.clk(clk), .ce(ce), .i(sa & sb), .o(sign_exe) );
delay1 u11 (.clk(clk), .ce(ce), .i(over1), .o(overflow) );
delay1 u12 (.clk(clk), .ce(ce), .i(over1), .o(inf) );
delay1 u13 (.clk(clk), .ce(ce), .i(under1), .o(underflow) );
 
assign o = {st,xo1,mo1,8'h00};
 
endmodule
 
 
// Multiplier with normalization and rounding.
 
module DFPMultiplynr(clk, ce, a, b, o, rm, sign_exe, inf, overflow, underflow);
input clk;
input ce;
input [127:0] a, b;
output [127:0] o;
input [2:0] rm;
output sign_exe;
output inf;
output overflow;
output underflow;
 
wire [243:0] o1;
wire sign_exe1, inf1, overflow1, underflow1;
wire [131:0] fpn0;
 
DFPMultiply u1 (clk, ce, a, b, o1, sign_exe1, inf1, overflow1, underflow1);
DFPNormalize u2(.clk(clk), .ce(ce), .under_i(underflow1), .i(o1), .o(fpn0) );
DFPRound u3(.clk(clk), .ce(ce), .rm(rm), .i(fpn0), .o(o) );
delay2 #(1) u4(.clk(clk), .ce(ce), .i(sign_exe1), .o(sign_exe));
delay2 #(1) u5(.clk(clk), .ce(ce), .i(inf1), .o(inf));
delay2 #(1) u6(.clk(clk), .ce(ce), .i(overflow1), .o(overflow));
delay2 #(1) u7(.clk(clk), .ce(ce), .i(underflow1), .o(underflow));
endmodule

powered by: WebSVN 2.1.0

© copyright 1999-2024 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.