yosys/tests/iwls2005/fpu/pre_norm.v

271 lines
9.2 KiB
Verilog

/////////////////////////////////////////////////////////////////////
//// ////
//// Pre Normalize ////
//// Pre Normalization Unit for Add/Sub Operations ////
//// ////
//// Author: Rudolf Usselmann ////
//// rudi@asics.ws ////
//// ////
/////////////////////////////////////////////////////////////////////
//// ////
//// Copyright (C) 2000 Rudolf Usselmann ////
//// rudi@asics.ws ////
//// ////
//// This source file may be used and distributed without ////
//// restriction provided that this copyright statement is not ////
//// removed from the file and that any derivative work contains ////
//// the original copyright notice and the associated disclaimer.////
//// ////
//// THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY ////
//// EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED ////
//// TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS ////
//// FOR A PARTICULAR PURPOSE. IN NO EVENT SHALL THE AUTHOR ////
//// 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. ////
//// ////
/////////////////////////////////////////////////////////////////////
`timescale 1ns / 100ps
module pre_norm(clk, rmode, add, opa, opb, opa_nan, opb_nan, fracta_out,
fractb_out, exp_dn_out, sign, nan_sign, result_zero_sign,
fasu_op);
input clk;
input [1:0] rmode;
input add;
input [31:0] opa, opb;
input opa_nan, opb_nan;
output [26:0] fracta_out, fractb_out;
output [7:0] exp_dn_out;
output sign;
output nan_sign, result_zero_sign;
output fasu_op; // Operation Output
////////////////////////////////////////////////////////////////////////
//
// Local Wires and registers
//
wire signa, signb; // alias to opX sign
wire [7:0] expa, expb; // alias to opX exponent
wire [22:0] fracta, fractb; // alias to opX fraction
wire expa_lt_expb; // expa is larger than expb indicator
wire fractb_lt_fracta; // fractb is larger than fracta indicator
reg [7:0] exp_dn_out; // de normalized exponent output
wire [7:0] exp_small, exp_large;
wire [7:0] exp_diff; // Numeric difference of the two exponents
wire [22:0] adj_op; // Fraction adjustment: input
wire [26:0] adj_op_tmp;
wire [26:0] adj_op_out; // Fraction adjustment: output
wire [26:0] fracta_n, fractb_n; // Fraction selection after normalizing
wire [26:0] fracta_s, fractb_s; // Fraction Sorting out
reg [26:0] fracta_out, fractb_out; // Fraction Output
reg sign, sign_d; // Sign Output
reg add_d; // operation (add/sub)
reg fasu_op; // operation (add/sub) register
wire expa_dn, expb_dn;
reg sticky;
reg result_zero_sign;
reg add_r, signa_r, signb_r;
wire [4:0] exp_diff_sft;
wire exp_lt_27;
wire op_dn;
wire [26:0] adj_op_out_sft;
reg fracta_lt_fractb, fracta_eq_fractb;
wire nan_sign1;
reg nan_sign;
////////////////////////////////////////////////////////////////////////
//
// Aliases
//
assign signa = opa[31];
assign signb = opb[31];
assign expa = opa[30:23];
assign expb = opb[30:23];
assign fracta = opa[22:0];
assign fractb = opb[22:0];
////////////////////////////////////////////////////////////////////////
//
// Pre-Normalize exponents (and fractions)
//
assign expa_lt_expb = expa > expb; // expa is larger than expb
// ---------------------------------------------------------------------
// Normalize
assign expa_dn = !(|expa); // opa denormalized
assign expb_dn = !(|expb); // opb denormalized
// ---------------------------------------------------------------------
// Calculate the difference between the smaller and larger exponent
wire [7:0] exp_diff1, exp_diff1a, exp_diff2;
assign exp_small = expa_lt_expb ? expb : expa;
assign exp_large = expa_lt_expb ? expa : expb;
assign exp_diff1 = exp_large - exp_small;
assign exp_diff1a = exp_diff1-1;
assign exp_diff2 = (expa_dn | expb_dn) ? exp_diff1a : exp_diff1;
assign exp_diff = (expa_dn & expb_dn) ? 8'h0 : exp_diff2;
always @(posedge clk) // If numbers are equal we should return zero
exp_dn_out <= #1 (!add_d & expa==expb & fracta==fractb) ? 8'h0 : exp_large;
// ---------------------------------------------------------------------
// Adjust the smaller fraction
assign op_dn = expa_lt_expb ? expb_dn : expa_dn;
assign adj_op = expa_lt_expb ? fractb : fracta;
assign adj_op_tmp = { ~op_dn, adj_op, 3'b0 }; // recover hidden bit (op_dn)
// adj_op_out is 27 bits wide, so can only be shifted 27 bits to the right
assign exp_lt_27 = exp_diff > 8'd27;
assign exp_diff_sft = exp_lt_27 ? 5'd27 : exp_diff[4:0];
assign adj_op_out_sft = adj_op_tmp >> exp_diff_sft;
assign adj_op_out = {adj_op_out_sft[26:1], adj_op_out_sft[0] | sticky };
// ---------------------------------------------------------------------
// Get truncated portion (sticky bit)
always @(exp_diff_sft or adj_op_tmp)
case(exp_diff_sft) // synopsys full_case parallel_case
00: sticky = 1'h0;
01: sticky = adj_op_tmp[0];
02: sticky = |adj_op_tmp[01:0];
03: sticky = |adj_op_tmp[02:0];
04: sticky = |adj_op_tmp[03:0];
05: sticky = |adj_op_tmp[04:0];
06: sticky = |adj_op_tmp[05:0];
07: sticky = |adj_op_tmp[06:0];
08: sticky = |adj_op_tmp[07:0];
09: sticky = |adj_op_tmp[08:0];
10: sticky = |adj_op_tmp[09:0];
11: sticky = |adj_op_tmp[10:0];
12: sticky = |adj_op_tmp[11:0];
13: sticky = |adj_op_tmp[12:0];
14: sticky = |adj_op_tmp[13:0];
15: sticky = |adj_op_tmp[14:0];
16: sticky = |adj_op_tmp[15:0];
17: sticky = |adj_op_tmp[16:0];
18: sticky = |adj_op_tmp[17:0];
19: sticky = |adj_op_tmp[18:0];
20: sticky = |adj_op_tmp[19:0];
21: sticky = |adj_op_tmp[20:0];
22: sticky = |adj_op_tmp[21:0];
23: sticky = |adj_op_tmp[22:0];
24: sticky = |adj_op_tmp[23:0];
25: sticky = |adj_op_tmp[24:0];
26: sticky = |adj_op_tmp[25:0];
27: sticky = |adj_op_tmp[26:0];
endcase
// ---------------------------------------------------------------------
// Select operands for add/sub (recover hidden bit)
assign fracta_n = expa_lt_expb ? {~expa_dn, fracta, 3'b0} : adj_op_out;
assign fractb_n = expa_lt_expb ? adj_op_out : {~expb_dn, fractb, 3'b0};
// ---------------------------------------------------------------------
// Sort operands (for sub only)
assign fractb_lt_fracta = fractb_n > fracta_n; // fractb is larger than fracta
assign fracta_s = fractb_lt_fracta ? fractb_n : fracta_n;
assign fractb_s = fractb_lt_fracta ? fracta_n : fractb_n;
always @(posedge clk)
fracta_out <= #1 fracta_s;
always @(posedge clk)
fractb_out <= #1 fractb_s;
// ---------------------------------------------------------------------
// Determine sign for the output
// sign: 0=Positive Number; 1=Negative Number
always @(signa or signb or add or fractb_lt_fracta)
case({signa, signb, add}) // synopsys full_case parallel_case
// Add
3'b0_0_1: sign_d = 0;
3'b0_1_1: sign_d = fractb_lt_fracta;
3'b1_0_1: sign_d = !fractb_lt_fracta;
3'b1_1_1: sign_d = 1;
// Sub
3'b0_0_0: sign_d = fractb_lt_fracta;
3'b0_1_0: sign_d = 0;
3'b1_0_0: sign_d = 1;
3'b1_1_0: sign_d = !fractb_lt_fracta;
endcase
always @(posedge clk)
sign <= #1 sign_d;
// Fix sign for ZERO result
always @(posedge clk)
signa_r <= #1 signa;
always @(posedge clk)
signb_r <= #1 signb;
always @(posedge clk)
add_r <= #1 add;
always @(posedge clk)
result_zero_sign <= #1 ( add_r & signa_r & signb_r) |
(!add_r & signa_r & !signb_r) |
( add_r & (signa_r | signb_r) & (rmode==3)) |
(!add_r & (signa_r == signb_r) & (rmode==3));
// Fix sign for NAN result
always @(posedge clk)
fracta_lt_fractb <= #1 fracta < fractb;
always @(posedge clk)
fracta_eq_fractb <= #1 fracta == fractb;
assign nan_sign1 = fracta_eq_fractb ? (signa_r & signb_r) : fracta_lt_fractb ? signb_r : signa_r;
always @(posedge clk)
nan_sign <= #1 (opa_nan & opb_nan) ? nan_sign1 : opb_nan ? signb_r : signa_r;
////////////////////////////////////////////////////////////////////////
//
// Decode Add/Sub operation
//
// add: 1=Add; 0=Subtract
always @(signa or signb or add)
case({signa, signb, add}) // synopsys full_case parallel_case
// Add
3'b0_0_1: add_d = 1;
3'b0_1_1: add_d = 0;
3'b1_0_1: add_d = 0;
3'b1_1_1: add_d = 1;
// Sub
3'b0_0_0: add_d = 0;
3'b0_1_0: add_d = 1;
3'b1_0_0: add_d = 1;
3'b1_1_0: add_d = 0;
endcase
always @(posedge clk)
fasu_op <= #1 add_d;
endmodule