Subversion Repositories fpu

/////////////////////////////////////////////////////////////////////
////                                                             ////
////  FPU                                                        ////
////  Floating Point Unit (Single precision)                     ////
////                                                             ////
////  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
 
/*
 
FPU Operations (fpu_op):
========================
 
0 = add
1 = sub
2 = mul
3 = div
4 =
5 =
6 =
7 =
 
Rounding Modes (rmode):
=======================
 
0 = round_nearest_even
1 = round_to_zero
2 = round_up
3 = round_down
 
*/
 
 
module fpu( clk, rmode, fpu_op, opa, opb, out, inf, snan, qnan, ine, overflow, underflow, zero, div_by_zero);
input		clk;
input	[1:0]	rmode;
input	[2:0]	fpu_op;
input	[31:0]	opa, opb;
output	[31:0]	out;
output		inf, snan, qnan;
output		ine;
output		overflow, underflow;
output		zero;
output		div_by_zero;
 
parameter	INF  = 31'h7f800000,
		QNAN = 31'h7fc00001,
		SNAN = 31'h7f800001;
 
////////////////////////////////////////////////////////////////////////
//
// Local Wires
//
reg		zero;
reg	[31:0]	opa_r, opb_r;		// Input operand registers
reg	[31:0]	out;			// Output register
reg		div_by_zero;		// Divide by zero output register
wire		signa, signb;		// alias to opX sign
wire		sign_fasu;		// sign output
wire	[26:0]	fracta, fractb;		// Fraction Outputs from EQU block
wire	[7:0]	exp_fasu;		// Exponent output from EQU block
reg	[7:0]	exp_r;			// Exponent output (registerd)
wire	[26:0]	fract_out_d;		// fraction output
wire		co;			// carry output
reg	[27:0]	fract_out_q;		// fraction output (registerd)
wire	[30:0]	out_d;			// Intermediate final result output
wire		overflow_d, underflow_d;// Overflow/Underflow Indicators
reg		overflow, underflow;	// Output registers for Overflow & Underflow
reg		inf, snan, qnan;	// Output Registers for INF, SNAN and QNAN
reg		ine;			// Output Registers for INE
reg	[1:0]	rmode_r1, rmode_r2, 	// Pipeline registers for rounding mode
		rmode_r3;
reg	[2:0]	fpu_op_r1, fpu_op_r2,	// Pipeline registers for fp opration
		fpu_op_r3;
wire		mul_inf, div_inf;
wire		mul_00, div_00;
 
////////////////////////////////////////////////////////////////////////
//
// Input Registers
//
 
always @(posedge clk)
	opa_r <= #1 opa;
 
always @(posedge clk)
	opb_r <= #1 opb;
 
always @(posedge clk)
	rmode_r1 <= #1 rmode;
 
always @(posedge clk)
	rmode_r2 <= #1 rmode_r1;
 
always @(posedge clk)
	rmode_r3 <= #1 rmode_r2;
 
always @(posedge clk)
	fpu_op_r1 <= #1 fpu_op;
 
always @(posedge clk)
	fpu_op_r2 <= #1 fpu_op_r1;
 
always @(posedge clk)
	fpu_op_r3 <= #1 fpu_op_r2;
 
////////////////////////////////////////////////////////////////////////
//
// Exceptions block
//
wire		inf_d, ind_d, qnan_d, snan_d, opa_nan, opb_nan;
wire		opa_00, opb_00;
wire		opa_inf, opb_inf;
wire		opa_dn, opb_dn;
 
except u0(	.clk(clk),
		.opa(opa_r), .opb(opb_r),
		.inf(inf_d), .ind(ind_d),
		.qnan(qnan_d), .snan(snan_d),
		.opa_nan(opa_nan), .opb_nan(opb_nan),
		.opa_00(opa_00), .opb_00(opb_00),
		.opa_inf(opa_inf), .opb_inf(opb_inf),
		.opa_dn(opa_dn), .opb_dn(opb_dn)
		);
 
////////////////////////////////////////////////////////////////////////
//
// Pre-Normalize block
// - Adjusts the numbers to equal exponents and sorts them
// - determine result sign
// - determine actual operation to perform (add or sub)
//
 
wire		nan_sign_d, result_zero_sign_d;
reg		sign_fasu_r;
wire	[7:0]	exp_mul;
wire		sign_mul;
reg		sign_mul_r;
wire	[23:0]	fracta_mul, fractb_mul;
wire		inf_mul;
reg		inf_mul_r;
wire	[1:0]	exp_ovf;
reg	[1:0]	exp_ovf_r;
wire		sign_exe;
reg		sign_exe_r;
wire	[2:0]	underflow_fmul_d;
 
 
pre_norm u1(.clk(clk),				// System Clock
	.rmode(rmode_r2),			// Roundin Mode
	.add(!fpu_op_r1[0]),			// Add/Sub Input
	.opa(opa_r),  .opb(opb_r),		// Registered OP Inputs
	.opa_nan(opa_nan),			// OpA is a NAN indicator
	.opb_nan(opb_nan),			// OpB is a NAN indicator
	.fracta_out(fracta),			// Equalized and sorted fraction
	.fractb_out(fractb),			// outputs (Registered)
	.exp_dn_out(exp_fasu),			// Selected exponent output (registered);
	.sign(sign_fasu),			// Encoded output Sign (registered)
	.nan_sign(nan_sign_d),			// Output Sign for NANs (registered)
	.result_zero_sign(result_zero_sign_d),	// Output Sign for zero result (registered)
	.fasu_op(fasu_op)			// Actual fasu operation output (registered)
	);
 
always @(posedge clk)
	sign_fasu_r <= #1 sign_fasu;
 
pre_norm_fmul u2(
		.clk(clk),
		.fpu_op(fpu_op_r1),
		.opa(opa_r), .opb(opb_r),
		.fracta(fracta_mul),
		.fractb(fractb_mul),
		.exp_out(exp_mul),	// FMUL exponent output (registered)
		.sign(sign_mul),	// FMUL sign output (registered)
		.sign_exe(sign_exe),	// FMUL exception sign output (registered)
		.inf(inf_mul),		// FMUL inf output (registered)
		.exp_ovf(exp_ovf),	// FMUL exponnent overflow output (registered)
		.underflow(underflow_fmul_d)
		);
 
 
always @(posedge clk)
	sign_mul_r <= #1 sign_mul;
 
always @(posedge clk)
	sign_exe_r <= #1 sign_exe;
 
always @(posedge clk)
	inf_mul_r <= #1 inf_mul;
 
always @(posedge clk)
	exp_ovf_r <= #1 exp_ovf;
 
 
////////////////////////////////////////////////////////////////////////
//
// Add/Sub
//
 
add_sub27 u3(
	.add(fasu_op),			// Add/Sub
	.opa(fracta),			// Fraction A input
	.opb(fractb),			// Fraction B Input
	.sum(fract_out_d),		// SUM output
	.co(co_d) );			// Carry Output
 
always @(posedge clk)
	fract_out_q <= #1 {co_d, fract_out_d};
 
////////////////////////////////////////////////////////////////////////
//
// Mul
//
wire	[47:0]	prod;
 
mul_r2 u5(.clk(clk), .opa(fracta_mul), .opb(fractb_mul), .prod(prod));
 
////////////////////////////////////////////////////////////////////////
//
// Divide
//
wire	[49:0]	quo;
wire	[49:0]	fdiv_opa;
wire	[49:0]	remainder;
wire		remainder_00;
reg	[4:0]	div_opa_ldz_d, div_opa_ldz_r1, div_opa_ldz_r2;
 
always @(fracta_mul)
	casex(fracta_mul[22:0])
	   23'b1??????????????????????: div_opa_ldz_d = 1;
	   23'b01?????????????????????: div_opa_ldz_d = 2;
	   23'b001????????????????????: div_opa_ldz_d = 3;
	   23'b0001???????????????????: div_opa_ldz_d = 4;
	   23'b00001??????????????????: div_opa_ldz_d = 5;
	   23'b000001?????????????????: div_opa_ldz_d = 6;
	   23'b0000001????????????????: div_opa_ldz_d = 7;
	   23'b00000001???????????????: div_opa_ldz_d = 8;
	   23'b000000001??????????????: div_opa_ldz_d = 9;
	   23'b0000000001?????????????: div_opa_ldz_d = 10;
	   23'b00000000001????????????: div_opa_ldz_d = 11;
	   23'b000000000001???????????: div_opa_ldz_d = 12;
	   23'b0000000000001??????????: div_opa_ldz_d = 13;
	   23'b00000000000001?????????: div_opa_ldz_d = 14;
	   23'b000000000000001????????: div_opa_ldz_d = 15;
	   23'b0000000000000001???????: div_opa_ldz_d = 16;
	   23'b00000000000000001??????: div_opa_ldz_d = 17;
	   23'b000000000000000001?????: div_opa_ldz_d = 18;
	   23'b0000000000000000001????: div_opa_ldz_d = 19;
	   23'b00000000000000000001???: div_opa_ldz_d = 20;
	   23'b000000000000000000001??: div_opa_ldz_d = 21;
	   23'b0000000000000000000001?: div_opa_ldz_d = 22;
	   23'b0000000000000000000000?: div_opa_ldz_d = 23;
	endcase
 
assign fdiv_opa = !(|opa_r[30:23]) ? {(fracta_mul<<div_opa_ldz_d), 26'h0} : {fracta_mul, 26'h0};
 
 
div_r2 u6(.clk(clk), .opa(fdiv_opa), .opb(fractb_mul), .quo(quo), .rem(remainder));
 
assign remainder_00 = !(|remainder);
 
always @(posedge clk)
	div_opa_ldz_r1 <= #1 div_opa_ldz_d;
 
always @(posedge clk)
	div_opa_ldz_r2 <= #1 div_opa_ldz_r1;
 
 
////////////////////////////////////////////////////////////////////////
//
// Normalize Result
//
wire		ine_d;
reg	[47:0]	fract_denorm;
wire	[47:0]	fract_div;
wire		sign_d;
reg		sign;
reg	[30:0]	opa_r1;
reg	[47:0]	fract_i2f;
reg		opas_r1, opas_r2;
wire		f2i_out_sign;
 
always @(posedge clk)			// Exponent must be once cycle delayed
	case(fpu_op_r2)
	  0,1:	exp_r <= #1 exp_fasu;
	  2,3:	exp_r <= #1 exp_mul;
	  4:	exp_r <= #1 0;
	  5:	exp_r <= #1 opa_r1[30:23];
	endcase
 
assign fract_div = (opb_dn ? quo[49:2] : {quo[26:0], 21'h0});
 
always @(posedge clk)
	opa_r1 <= #1 opa_r[30:0];
 
always @(posedge clk)
	fract_i2f <= #1 (fpu_op_r2==5) ?
			(sign_d ?  1-{24'h00, (|opa_r1[30:23]), opa_r1[22:0]}-1 : {24'h0, (|opa_r1[30:23]), opa_r1[22:0]}) :
			(sign_d ? 1 - {opa_r1, 17'h01} : {opa_r1, 17'h0});
 
always @(fpu_op_r3 or fract_out_q or prod or fract_div or fract_i2f)
	case(fpu_op_r3)
	   0,1:	fract_denorm = {fract_out_q, 20'h0};
	   2:	fract_denorm = prod;
	   3:	fract_denorm = fract_div;
	   4,5:	fract_denorm = fract_i2f;
	endcase
 
 
always @(posedge clk)
	opas_r1 <= #1 opa_r[31];
 
always @(posedge clk)
	opas_r2 <= #1 opas_r1;
 
assign sign_d = fpu_op_r2[1] ? sign_mul : sign_fasu;
 
always @(posedge clk)
	sign <= #1 (rmode_r2==2'h3) ? !sign_d : sign_d;
 
post_norm u4(.clk(clk),			// System Clock
	.fpu_op(fpu_op_r3),		// Floating Point Operation
	.opas(opas_r2),			// OPA Sign
	.sign(sign),			// Sign of the result
	.rmode(rmode_r3),		// Rounding mode
	.fract_in(fract_denorm),	// Fraction Input
	.exp_ovf(exp_ovf_r),		// Exponent Overflow
	.exp_in(exp_r),			// Exponent Input
	.opa_dn(opa_dn),		// Operand A Denormalized
	.opb_dn(opb_dn),		// Operand A Denormalized
	.rem_00(remainder_00),		// Diveide Remainder is zero
	.div_opa_ldz(div_opa_ldz_r2),	// Divide opa leading zeros count
	.output_zero(mul_00 | div_00),	// Force output to Zero
	.out(out_d),			// Normalized output (un-registered)
	.ine(ine_d),			// Result Inexact output (un-registered)
	.overflow(overflow_d),		// Overflow output (un-registered)
	.underflow(underflow_d),	// Underflow output (un-registered)
	.f2i_out_sign(f2i_out_sign)	// F2I Output Sign
	);
 
////////////////////////////////////////////////////////////////////////
//
// FPU Outputs
//
reg		fasu_op_r1, fasu_op_r2;
wire	[30:0]	out_fixed;
wire		output_zero_fasu;
wire		output_zero_fdiv;
wire		output_zero_fmul;
reg		inf_mul2;
wire		overflow_fasu;
wire		overflow_fmul;
wire		overflow_fdiv;
wire		inf_fmul;
wire		sign_mul_final;
wire		out_d_00;
wire		sign_div_final;
wire		ine_mul, ine_mula, ine_div, ine_fasu;
wire		underflow_fasu, underflow_fmul, underflow_fdiv;
wire		underflow_fmul1;
reg	[2:0]	underflow_fmul_r;
reg		opa_nan_r;
 
 
always @(posedge clk)
	fasu_op_r1 <= #1 fasu_op;
 
always @(posedge clk)
	fasu_op_r2 <= #1 fasu_op_r1;
 
always @(posedge clk)
	inf_mul2 <= #1 exp_mul == 8'hff;
 
 
// Force pre-set values for non numerical output
assign mul_inf = (fpu_op_r3==3'b010) & (inf_mul_r | inf_mul2) & (rmode_r3==2'h0);
assign div_inf = (fpu_op_r3==3'b011) & (opb_00 | opa_inf);
 
assign mul_00 = (fpu_op_r3==3'b010) & (opa_00 | opb_00);
assign div_00 = (fpu_op_r3==3'b011) & (opa_00 | opb_inf);
 
assign out_fixed = (	(qnan_d | snan_d) |
			(ind_d & !fasu_op_r2) | 
			((fpu_op_r3==3'b011) & opb_00 & opa_00) |
			(((opa_inf & opb_00) | (opb_inf & opa_00 )) & fpu_op_r3==3'b010)
		   )  ? QNAN : INF;
 
always @(posedge clk)
	out[30:0] <= #1 (mul_inf | div_inf | (inf_d & (fpu_op_r3!=3'b011) & (fpu_op_r3!=3'b101)) | snan_d | qnan_d) & fpu_op_r3!=3'b100 ? out_fixed :
			out_d;
 
assign out_d_00 = !(|out_d);
 
assign sign_mul_final = (sign_exe_r & ((opa_00 & opb_inf) | (opb_00 & opa_inf))) ? !sign_mul_r : sign_mul_r;
assign sign_div_final = (sign_exe_r & (opa_inf & opb_inf)) ? !sign_mul_r : sign_mul_r | (opa_00 & opb_00);
 
always @(posedge clk)
	out[31] <= #1	((fpu_op_r3==3'b101) & out_d_00) ? (f2i_out_sign & !(qnan_d | snan_d) ) :
			((fpu_op_r3==3'b010) & !(snan_d | qnan_d)) ?	sign_mul_final :
			((fpu_op_r3==3'b011) & !(snan_d | qnan_d)) ?	sign_div_final :
			(snan_d | qnan_d | ind_d) ?			nan_sign_d :
			output_zero_fasu ?				result_zero_sign_d :
									sign_fasu_r;
 
// Exception Outputs
assign ine_mula = ((inf_mul_r |  inf_mul2 | opa_inf | opb_inf) & (rmode_r3==2'h1) & 
		!((opa_inf & opb_00) | (opb_inf & opa_00 )) & fpu_op_r3[1]);
 
assign ine_mul  = (ine_mula | ine_d | inf_fmul | out_d_00 | overflow_d | underflow_d) &
		  !opa_00 & !opb_00 & !(snan_d | qnan_d | inf_d);
assign ine_div  = (ine_d | overflow_d | underflow_d) & !(opb_00 | snan_d | qnan_d | inf_d);
assign ine_fasu = (ine_d | overflow_d | underflow_d) & !(snan_d | qnan_d | inf_d);
 
always @(posedge  clk)
	ine <= #1	 fpu_op_r3[2] ? ine_d :
			!fpu_op_r3[1] ? ine_fasu :
			 fpu_op_r3[0] ? ine_div  : ine_mul;
 
 
assign overflow_fasu = overflow_d & !(snan_d | qnan_d | inf_d);
assign overflow_fmul = !inf_d & (inf_mul_r | inf_mul2 | overflow_d) & !(snan_d | qnan_d);
assign overflow_fdiv = (overflow_d & !(opb_00 | inf_d | snan_d | qnan_d));
 
always @(posedge clk)
	overflow <= #1	 fpu_op_r3[2] ? 0 :
			!fpu_op_r3[1] ? overflow_fasu :
			 fpu_op_r3[0] ? overflow_fdiv : overflow_fmul;
 
always @(posedge clk)
	underflow_fmul_r <= #1 underflow_fmul_d;
 
 
assign underflow_fmul1 = underflow_fmul_r[0] |
			(underflow_fmul_r[1] & underflow_d ) |
			((opa_dn | opb_dn) & out_d_00 & (prod!=0) & sign) |
			(underflow_fmul_r[2] & ((out_d[30:23]==0) | (out_d[22:0]==0)));
 
assign underflow_fasu = underflow_d & !(inf_d | snan_d | qnan_d);
assign underflow_fmul = underflow_fmul1 & !(snan_d | qnan_d | inf_mul_r);
assign underflow_fdiv = underflow_fasu & !opb_00;
 
always @(posedge clk)
	underflow <= #1  fpu_op_r3[2] ? 0 :
			!fpu_op_r3[1] ? underflow_fasu :
			 fpu_op_r3[0] ? underflow_fdiv : underflow_fmul;
 
always @(posedge clk)
	snan <= #1 snan_d;
 
// synopsys translate_off
wire		mul_uf_del;
wire		uf2_del, ufb2_del, ufc2_del,  underflow_d_del;
wire		co_del;
wire	[30:0]	out_d_del;
wire		ov_fasu_del, ov_fmul_del;
wire	[2:0]	fop;
wire	[4:0]	ldza_del;
wire	[49:0]	quo_del;
 
delay1  #0 ud000(clk, underflow_fmul1, mul_uf_del);
delay1  #0 ud001(clk, underflow_fmul_r[0], uf2_del);
delay1  #0 ud002(clk, underflow_fmul_r[1], ufb2_del);
delay1  #0 ud003(clk, underflow_d, underflow_d_del);
delay1  #0 ud004(clk, test.u0.u4.exp_out1_co, co_del);
delay1  #0 ud005(clk, underflow_fmul_r[2], ufc2_del);
delay1 #30 ud006(clk, out_d, out_d_del);
 
delay1  #0 ud007(clk, overflow_fasu, ov_fasu_del);
delay1  #0 ud008(clk, overflow_fmul, ov_fmul_del);
 
delay1  #2 ud009(clk, fpu_op_r3, fop);
 
delay3  #4 ud010(clk, div_opa_ldz_d, ldza_del);
 
delay1  #49 ud012(clk, quo, quo_del);
 
always @(test.error_event)
   begin
	#0.2
	$display("muf: %b uf0: %b uf1: %b uf2: %b, tx0: %b, co: %b, out_d: %h (%h %h), ov_fasu: %b, ov_fmul: %b, fop: %h",
			mul_uf_del, uf2_del, ufb2_del, ufc2_del, underflow_d_del, co_del, out_d_del, out_d_del[30:23], out_d_del[22:0],
			ov_fasu_del, ov_fmul_del, fop );
	$display("ldza: %h, quo: %b",
			ldza_del, quo_del);
   end
// synopsys translate_on
 
 
 
// Status Outputs
always @(posedge clk)
	qnan <= #1	fpu_op_r3[2] ? 0 : (
						snan_d | qnan_d | (ind_d & !fasu_op_r2) |
						(opa_00 & opb_00 & fpu_op_r3==3'b011) |
						(((opa_inf & opb_00) | (opb_inf & opa_00 )) & fpu_op_r3==3'b010)
					   );
 
assign inf_fmul = 	(((inf_mul_r | inf_mul2) & (rmode_r3==2'h0)) | opa_inf | opb_inf) & 
			!((opa_inf & opb_00) | (opb_inf & opa_00 )) &
			fpu_op_r3==3'b010;
 
always @(posedge clk)
	inf <= #1	fpu_op_r3[2] ? 0 :
			(!(qnan_d | snan_d) & (
						((&out_d[30:23]) & !(|out_d[22:0]) & !(opb_00 & fpu_op_r3==3'b011)) |
						(inf_d & !(ind_d & !fasu_op_r2) & !fpu_op_r3[1]) |
						inf_fmul |
						(!opa_00 & opb_00 & fpu_op_r3==3'b011) |
						(fpu_op_r3==3'b011 & opa_inf & !opb_inf)
					      )
			);
 
assign output_zero_fasu = out_d_00 & !(inf_d | snan_d | qnan_d);
assign output_zero_fdiv = (div_00 | (out_d_00 & !opb_00)) & !(opa_inf & opb_inf) &
			  !(opa_00 & opb_00) & !(qnan_d | snan_d);
assign output_zero_fmul = (out_d_00 | opa_00 | opb_00) &
			  !(inf_mul_r | inf_mul2 | opa_inf | opb_inf | snan_d | qnan_d) &
			  !(opa_inf & opb_00) & !(opb_inf & opa_00);
 
always @(posedge clk)
	zero <= #1	fpu_op_r3==3'b101 ?	out_d_00 & !(snan_d | qnan_d):
			fpu_op_r3==3'b011 ?	output_zero_fdiv :
			fpu_op_r3==3'b010 ?	output_zero_fmul :
						output_zero_fasu ;
 
always @(posedge clk)
	opa_nan_r <= #1 !opa_nan & fpu_op_r2==3'b011;
 
always @(posedge clk)
	div_by_zero <= #1 opa_nan_r & !opa_00 & !opa_inf & opb_00;
 
endmodule