URL
https://opencores.org/ocsvn/sparc64soc/sparc64soc/trunk
Subversion Repositories sparc64soc
[/] [sparc64soc/] [trunk/] [T1-FPU/] [fpu_mul_ctl.v] - Rev 3
Go to most recent revision | Compare with Previous | Blame | View Log
// ========== Copyright Header Begin ========================================== // // OpenSPARC T1 Processor File: fpu_mul_ctl.v // Copyright (c) 2006 Sun Microsystems, Inc. All Rights Reserved. // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES. // // The above named program is free software; you can redistribute it and/or // modify it under the terms of the GNU General Public // License version 2 as published by the Free Software Foundation. // // The above named program is distributed in the hope that it will be // useful, but WITHOUT ANY WARRANTY; without even the implied warranty of // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU // General Public License for more details. // // You should have received a copy of the GNU General Public // License along with this work; if not, write to the Free Software // Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. // // ========== Copyright Header End ============================================ /////////////////////////////////////////////////////////////////////////////// // // Multiply pipeline synthesizable logic // - special input cases // - opcode pipeline // - sign logic // - exception logic // - datapath control- select lines and control logic // /////////////////////////////////////////////////////////////////////////////// module fpu_mul_ctl ( inq_in1_51, inq_in1_54, inq_in1_53_0_neq_0, inq_in1_50_0_neq_0, inq_in1_53_32_neq_0, inq_in1_exp_eq_0, inq_in1_exp_neq_ffs, inq_in2_51, inq_in2_54, inq_in2_53_0_neq_0, inq_in2_50_0_neq_0, inq_in2_53_32_neq_0, inq_in2_exp_eq_0, inq_in2_exp_neq_ffs, inq_op, inq_mul, inq_rnd_mode, inq_id, inq_in1_63, inq_in2_63, mul_dest_rdy, mul_dest_rdya, m5stg_exp, m5stg_fracadd_cout, m5stg_frac_neq_0, m5stg_frac_dbl_nx, m5stg_frac_sng_nx, m1stg_ld0_1, m1stg_ld0_2, m3stg_exp, m3stg_expadd_eq_0, m3stg_expadd_lte_0_inv, m3stg_ld0_inv, m4stg_exp, m4stg_frac_105, m5stg_frac, arst_l, grst_l, rclk, mul_pipe_active, m1stg_snan_sng_in1, m1stg_snan_dbl_in1, m1stg_snan_sng_in2, m1stg_snan_dbl_in2, m1stg_step, m1stg_sngop, m1stg_dblop, m1stg_dblop_inv, m1stg_fmul, m1stg_fsmuld, m2stg_fmuls, m2stg_fmuld, m2stg_fsmuld, m5stg_fmuls, m5stg_fmuld, m5stg_fmulda, m6stg_fmul_in, m6stg_id_in, m6stg_fmul_dbl_dst, m6stg_fmuls, m6stg_step, mul_sign_out, m5stg_in_of, mul_exc_out, m2stg_frac1_dbl_norm, m2stg_frac1_dbl_dnrm, m2stg_frac1_sng_norm, m2stg_frac1_sng_dnrm, m2stg_frac1_inf, m2stg_frac2_dbl_norm, m2stg_frac2_dbl_dnrm, m2stg_frac2_sng_norm, m2stg_frac2_sng_dnrm, m2stg_frac2_inf, m1stg_inf_zero_in, m1stg_inf_zero_in_dbl, m2stg_exp_expadd, m2stg_exp_0bff, m2stg_exp_017f, m2stg_exp_04ff, m2stg_exp_zero, m3bstg_ld0_inv, m4stg_sh_cnt_in, m4stg_inc_exp_54, m4stg_inc_exp_55, m4stg_inc_exp_105, m4stg_left_shift_step, m4stg_right_shift_step, m5stg_to_0, m5stg_to_0_inv, mul_frac_out_fracadd, mul_frac_out_frac, mul_exp_out_exp_plus1, mul_exp_out_exp, mula_rst_l, se, si, so ); parameter FMULS= 8'h49, FMULD= 8'h4a, FSMULD= 8'h69; input inq_in1_51; // request operand 1[51] input inq_in1_54; // request operand 1[54] input inq_in1_53_0_neq_0; // request operand 1[53:0]!=0 input inq_in1_50_0_neq_0; // request operand 1[50:0]!=0 input inq_in1_53_32_neq_0; // request operand 1[53:32]!=0 input inq_in1_exp_eq_0; // request operand 1[62:52]==0 input inq_in1_exp_neq_ffs; // request operand 1[62:52]!=0x7ff input inq_in2_51; // request operand 2[51] input inq_in2_54; // request operand 2[54] input inq_in2_53_0_neq_0; // request operand 2[53:0]!=0 input inq_in2_50_0_neq_0; // request operand 2[50:0]!=0 input inq_in2_53_32_neq_0; // request operand 2[53:32]!=0 input inq_in2_exp_eq_0; // request operand 2[62:52]==0 input inq_in2_exp_neq_ffs; // request operand 2[62:52]!=0x7ff input [7:0] inq_op; // request opcode to op pipes input inq_mul; // multiply pipe request input [1:0] inq_rnd_mode; // request rounding mode to op pipes input [4:0] inq_id; // request ID to the operation pipes input inq_in1_63; // request[63] operand 1 to op pipes input inq_in2_63; // request[63] operand 2 to op pipes input mul_dest_rdy; // multiply result req accepted for CPX input mul_dest_rdya; // multiply result req accepted for CPX input [12:0] m5stg_exp; // exponent input- multiply 5 stage input m5stg_fracadd_cout; // fraction rounding adder carry out input m5stg_frac_neq_0; // fraction input to mul 5 stage != 0 input m5stg_frac_dbl_nx; // double precision inexact result input m5stg_frac_sng_nx; // single precision inexact result input [5:0] m1stg_ld0_1; // denorm operand 1 leading 0's input [5:0] m1stg_ld0_2; // denorm operand 2 leading 0's input [12:0] m3stg_exp; // exponent input- multiply 3 stage input m3stg_expadd_eq_0; // mul stage 3 exponent adder sum == 0 input m3stg_expadd_lte_0_inv; // mul stage 3 exponent adder sum <= 0 input [5:0] m3stg_ld0_inv; // leading 0's in multiply operands input [12:0] m4stg_exp; // exponent input- multiply 4 stage input m4stg_frac_105; // multiply stage 4a fraction input[105] input [32:0] m5stg_frac; // multiply stage 5 fraction input input arst_l; // asynchronous global reset- asserted low input grst_l; // synchronous global reset- asserted low input rclk; // global clock output mul_pipe_active; // mul pipe is executing a valid instr output m1stg_snan_sng_in1; // operand 1 is single signalling NaN output m1stg_snan_dbl_in1; // operand 1 is double signalling NaN output m1stg_snan_sng_in2; // operand 2 is single signalling NaN output m1stg_snan_dbl_in2; // operand 2 is double signalling NaN output m1stg_step; // multiply pipe load output m1stg_sngop; // single precision operation- mul 1 stg output m1stg_dblop; // double precision operation- mul 1 stg output m1stg_dblop_inv; // single or int operation- mul 1 stg output m1stg_fmul; // multiply operation- mul 1 stage output m1stg_fsmuld; // fsmuld- multiply 1 stage output m2stg_fmuls; // fmuls- multiply 2 stage output m2stg_fmuld; // fmuld- multiply 2 stage output m2stg_fsmuld; // fsmuld- multiply 2 stage output m5stg_fmuls; // fmuls- multiply 5 stage output m5stg_fmuld; // fmuld- multiply 5 stage output m5stg_fmulda; // fmuld- multiply 5 stage copy output m6stg_fmul_in; // mul pipe output request next cycle output [9:0] m6stg_id_in; // mul pipe output ID next cycle output m6stg_fmul_dbl_dst; // double precision multiply result output m6stg_fmuls; // fmuls- multiply 6 stage output m6stg_step; // advance the multiply pipe output mul_sign_out; // multiply sign output output m5stg_in_of; // multiply overflow- select exp out output [4:0] mul_exc_out; // multiply pipe result- exception flags output m2stg_frac1_dbl_norm; // select line to m2stg_frac1 output m2stg_frac1_dbl_dnrm; // select line to m2stg_frac1 output m2stg_frac1_sng_norm; // select line to m2stg_frac1 output m2stg_frac1_sng_dnrm; // select line to m2stg_frac1 output m2stg_frac1_inf; // select line to m2stg_frac1 output m2stg_frac2_dbl_norm; // select line to m2stg_frac2 output m2stg_frac2_dbl_dnrm; // select line to m2stg_frac2 output m2stg_frac2_sng_norm; // select line to m2stg_frac2 output m2stg_frac2_sng_dnrm; // select line to m2stg_frac2 output m2stg_frac2_inf; // select line to m2stg_frac2 output m1stg_inf_zero_in; // 1 operand is infinity; other is 0 output m1stg_inf_zero_in_dbl; // 1 opnd is infinity; other is 0- dbl output m2stg_exp_expadd; // select line to m2stg_exp output m2stg_exp_0bff; // select line to m2stg_exp output m2stg_exp_017f; // select line to m2stg_exp output m2stg_exp_04ff; // select line to m2stg_exp output m2stg_exp_zero; // select line to m2stg_exp output [6:0] m3bstg_ld0_inv; // leading 0's in multiply operands output [5:0] m4stg_sh_cnt_in; // multiply normalization shift count output m4stg_inc_exp_54; // select line to m5stg_exp output m4stg_inc_exp_55; // select line to m5stg_exp output m4stg_inc_exp_105; // select line to m5stg_exp output m4stg_left_shift_step; // select line to m5stg_frac output m4stg_right_shift_step; // select line to m5stg_frac output m5stg_to_0; // result to max finite on overflow output m5stg_to_0_inv; // result to infinity on overflow output mul_frac_out_fracadd; // select line to mul_frac_out output mul_frac_out_frac; // select line to mul_frac_out output mul_exp_out_exp_plus1; // select line to mul_exp_out output mul_exp_out_exp; // select line to mul_exp_out output mula_rst_l; // reset for mul64 input se; // scan_enable input si; // scan in output so; // scan out wire reset; wire mul_frac_in1_51; wire mul_frac_in1_54; wire mul_frac_in1_53_0_neq_0; wire mul_frac_in1_50_0_neq_0; wire mul_frac_in1_53_32_neq_0; wire mul_exp_in1_exp_eq_0; wire mul_exp_in1_exp_neq_ffs; wire mul_frac_in2_51; wire mul_frac_in2_54; wire mul_frac_in2_53_0_neq_0; wire mul_frac_in2_50_0_neq_0; wire mul_frac_in2_53_32_neq_0; wire mul_exp_in2_exp_eq_0; wire mul_exp_in2_exp_neq_ffs; wire m1stg_denorm_sng_in1; wire m1stg_denorm_dbl_in1; wire m1stg_denorm_sng_in2; wire m1stg_denorm_dbl_in2; wire m1stg_denorm_in1; wire m1stg_denorm_in2; wire m1stg_norm_sng_in1; wire m1stg_norm_dbl_in1; wire m1stg_norm_sng_in2; wire m1stg_norm_dbl_in2; wire m1stg_snan_sng_in1; wire m1stg_snan_dbl_in1; wire m1stg_snan_sng_in2; wire m1stg_snan_dbl_in2; wire m1stg_qnan_sng_in1; wire m1stg_qnan_dbl_in1; wire m1stg_qnan_sng_in2; wire m1stg_qnan_dbl_in2; wire m1stg_snan_in1; wire m1stg_snan_in2; wire m1stg_qnan_in1; wire m1stg_qnan_in2; wire m2stg_snan_in1; wire m2stg_snan_in2; wire m2stg_qnan_in1; wire m2stg_qnan_in2; wire m1stg_nan_sng_in1; wire m1stg_nan_dbl_in1; wire m1stg_nan_sng_in2; wire m1stg_nan_dbl_in2; wire m1stg_nan_in1; wire m1stg_nan_in2; wire m2stg_nan_in2; wire m1stg_inf_sng_in1; wire m1stg_inf_dbl_in1; wire m1stg_inf_sng_in2; wire m1stg_inf_dbl_in2; wire m1stg_inf_in1; wire m1stg_inf_in2; wire m1stg_inf_in; wire m2stg_inf_in1; wire m2stg_inf_in2; wire m2stg_inf_in; wire m1stg_infnan_sng_in1; wire m1stg_infnan_dbl_in1; wire m1stg_infnan_sng_in2; wire m1stg_infnan_dbl_in2; wire m1stg_infnan_in1; wire m1stg_infnan_in2; wire m1stg_infnan_in; wire m1stg_zero_in1; wire m1stg_zero_in2; wire m1stg_zero_in; wire m2stg_zero_in1; wire m2stg_zero_in2; wire m2stg_zero_in; wire m1stg_step; wire [7:0] m1stg_op_in; wire [7:0] m1stg_op; wire m1stg_mul_in; wire m1stg_mul; wire m1stg_sngop; wire [3:0] m1stg_sngopa; wire m1stg_dblop; wire [3:0] m1stg_dblopa; wire m1stg_dblop_inv_in; wire m1stg_dblop_inv; wire [1:0] m1stg_rnd_mode; wire [4:0] m1stg_id; wire m1stg_fmul; wire m1stg_fmul_dbl_dst; wire m1stg_fmuls; wire m1stg_fmuld; wire m1stg_fsmuld; wire [4:0] m1stg_opdec; wire [4:0] m2stg_opdec; wire [1:0] m2stg_rnd_mode; wire [4:0] m2stg_id; wire m2stg_fmul; wire m2stg_fmuls; wire m2stg_fmuld; wire m2stg_fsmuld; wire [4:1] m3astg_opdec; wire [1:0] m3astg_rnd_mode; wire [4:0] m3astg_id; wire [4:1] m3bstg_opdec; wire [1:0] m3bstg_rnd_mode; wire [4:0] m3bstg_id; wire [4:1] m3stg_opdec; wire [1:0] m3stg_rnd_mode; wire [4:0] m3stg_id; wire m3stg_fmul; wire [4:1] m4stg_opdec; wire [1:0] m4stg_rnd_mode; wire [4:0] m4stg_id; wire m4stg_fmul; wire m4stg_fmuld; wire [4:1] m5stg_opdec; wire [1:0] m5stg_rnd_mode; wire [4:0] m5stg_id; wire m5stg_fmul; wire m5stg_fmuls; wire m5stg_fmuld; wire m5stg_fmulda; wire m6stg_fmul_in; wire [4:2] m6stg_opdec; wire [9:0] m6stg_id_in; wire [9:0] m6stg_id; wire m6stg_fmul; wire m6stg_fmul_dbl_dst; wire m6stg_fmuls; wire m6stg_hold; wire m6stg_holda; wire m6stg_step; wire m6stg_stepa; wire m1stg_sign1; wire m1stg_sign2; wire m2stg_sign1; wire m2stg_sign2; wire m1stg_of_mask; wire m2stg_of_mask; wire m2stg_sign; wire m3astg_sign; wire m2stg_nv; wire m3astg_nv; wire m3astg_of_mask; wire m3bstg_sign; wire m3bstg_nv; wire m3stg_sign; wire m3stg_nv; wire m3stg_of_mask; wire m4stg_sign; wire m4stg_nv; wire m4stg_of_mask; wire m5stg_sign; wire m5stg_nv; wire m5stg_of_mask; wire mul_sign_out; wire mul_nv_out; wire m5stg_in_of; wire mul_of_out_tmp1_in; wire mul_of_out_tmp1; wire mul_of_out_tmp2; wire mul_of_out_cout; wire mul_of_out; wire mul_uf_out_in; wire mul_uf_out; wire mul_nx_out_in; wire mul_nx_out; wire [4:0] mul_exc_out; wire m2stg_frac1_dbl_norm; wire m2stg_frac1_dbl_dnrm; wire m2stg_frac1_sng_norm; wire m2stg_frac1_sng_dnrm; wire m2stg_frac1_inf; wire m2stg_frac2_dbl_norm; wire m2stg_frac2_dbl_dnrm; wire m2stg_frac2_sng_norm; wire m2stg_frac2_sng_dnrm; wire m2stg_frac2_inf; wire m1stg_inf_zero_in; wire m1stg_inf_zero_in_dbl; wire [5:0] m2stg_ld0_1_in; wire [5:0] m2stg_ld0_1; wire [5:0] m2stg_ld0_2_in; wire [5:0] m2stg_ld0_2; wire m2stg_exp_expadd; wire m2stg_exp_0bff; wire m2stg_exp_017f; wire m2stg_exp_04ff; wire m2stg_exp_zero; wire [6:0] m2stg_ld0; wire [6:0] m2stg_ld0_inv; wire [6:0] m3astg_ld0_inv; wire [6:0] m3bstg_ld0_inv; wire m4stg_expadd_eq_0; wire m3stg_exp_lte_0; wire m4stg_right_shift_in; wire m4stg_right_shift; wire [5:0] m3stg_exp_minus1; wire [5:0] m3stg_exp_inv_plus2; wire m3stg_exp_lt_neg57; wire [5:0] m4stg_sh_cnt_in; wire m4stg_left_shift_step; wire m4stg_right_shift_step; wire m4stg_inc_exp_54; wire m4stg_inc_exp_55; wire m4stg_inc_exp_105; wire m5stg_rndup; wire m5stg_to_0; wire m5stg_to_0_inv; wire mul_frac_out_fracadd; wire mul_frac_out_frac; wire mul_exp_out_exp_plus1; wire mul_exp_out_exp; wire mul_pipe_active_in; wire mul_pipe_active; wire mula_rst_l; dffrl_async #(1) dffrl_mul_ctl ( .din (grst_l), .clk (rclk), .rst_l(arst_l), .q (mul_ctl_rst_l), .se (se), .si (), .so () ); assign reset= (!mul_ctl_rst_l); // 3/14/03 reset signal for mul64 assign mula_rst_l = mul_ctl_rst_l; /////////////////////////////////////////////////////////////////////////////// // // Multiply pipeline special input cases. // /////////////////////////////////////////////////////////////////////////////// dffe_s #(1) i_mul_frac_in1_51 ( .din (inq_in1_51), .en (m6stg_step), .clk (rclk), .q (mul_frac_in1_51), .se (se), .si (), .so () ); dffe_s #(1) i_mul_frac_in1_54 ( .din (inq_in1_54), .en (m6stg_step), .clk (rclk), .q (mul_frac_in1_54), .se (se), .si (), .so () ); dffe_s #(1) i_mul_frac_in1_53_0_neq_0 ( .din (inq_in1_53_0_neq_0), .en (m6stg_step), .clk (rclk), .q (mul_frac_in1_53_0_neq_0), .se (se), .si (), .so () ); dffe_s #(1) i_mul_frac_in1_50_0_neq_0 ( .din (inq_in1_50_0_neq_0), .en (m6stg_step), .clk (rclk), .q (mul_frac_in1_50_0_neq_0), .se (se), .si (), .so () ); dffe_s #(1) i_mul_frac_in1_53_32_neq_0 ( .din (inq_in1_53_32_neq_0), .en (m6stg_step), .clk (rclk), .q (mul_frac_in1_53_32_neq_0), .se (se), .si (), .so () ); dffe_s #(1) i_mul_exp_in1_exp_eq_0 ( .din (inq_in1_exp_eq_0), .en (m6stg_step), .clk (rclk), .q (mul_exp_in1_exp_eq_0), .se (se), .si (), .so () ); dffe_s #(1) i_mul_exp_in1_exp_neq_ffs ( .din (inq_in1_exp_neq_ffs), .en (m6stg_step), .clk (rclk), .q (mul_exp_in1_exp_neq_ffs), .se (se), .si (), .so () ); dffe_s #(1) i_mul_frac_in2_51 ( .din (inq_in2_51), .en (m6stg_step), .clk (rclk), .q (mul_frac_in2_51), .se (se), .si (), .so () ); dffe_s #(1) i_mul_frac_in2_54 ( .din (inq_in2_54), .en (m6stg_step), .clk (rclk), .q (mul_frac_in2_54), .se (se), .si (), .so () ); dffe_s #(1) i_mul_frac_in2_53_0_neq_0 ( .din (inq_in2_53_0_neq_0), .en (m6stg_step), .clk (rclk), .q (mul_frac_in2_53_0_neq_0), .se (se), .si (), .so () ); dffe_s #(1) i_mul_frac_in2_50_0_neq_0 ( .din (inq_in2_50_0_neq_0), .en (m6stg_step), .clk (rclk), .q (mul_frac_in2_50_0_neq_0), .se (se), .si (), .so () ); dffe_s #(1) i_mul_frac_in2_53_32_neq_0 ( .din (inq_in2_53_32_neq_0), .en (m6stg_step), .clk (rclk), .q (mul_frac_in2_53_32_neq_0), .se (se), .si (), .so () ); dffe_s #(1) i_mul_exp_in2_exp_eq_0 ( .din (inq_in2_exp_eq_0), .en (m6stg_step), .clk (rclk), .q (mul_exp_in2_exp_eq_0), .se (se), .si (), .so () ); dffe_s #(1) i_mul_exp_in2_exp_neq_ffs ( .din (inq_in2_exp_neq_ffs), .en (m6stg_step), .clk (rclk), .q (mul_exp_in2_exp_neq_ffs), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Denorm multiply inputs. // /////////////////////////////////////////////////////////////////////////////// assign m1stg_denorm_sng_in1= mul_exp_in1_exp_eq_0 && m1stg_sngopa[0]; assign m1stg_denorm_dbl_in1= mul_exp_in1_exp_eq_0 && m1stg_dblopa[0]; assign m1stg_denorm_sng_in2= mul_exp_in2_exp_eq_0 && m1stg_sngopa[0]; assign m1stg_denorm_dbl_in2= mul_exp_in2_exp_eq_0 && m1stg_dblopa[0]; assign m1stg_denorm_in1= m1stg_denorm_sng_in1 || m1stg_denorm_dbl_in1; assign m1stg_denorm_in2= m1stg_denorm_sng_in2 || m1stg_denorm_dbl_in2; /////////////////////////////////////////////////////////////////////////////// // // Non-denorm multiply inputs. // /////////////////////////////////////////////////////////////////////////////// assign m1stg_norm_sng_in1= (!mul_exp_in1_exp_eq_0) && m1stg_sngopa[0]; assign m1stg_norm_dbl_in1= (!mul_exp_in1_exp_eq_0) && m1stg_dblopa[0]; assign m1stg_norm_sng_in2= (!mul_exp_in2_exp_eq_0) && m1stg_sngopa[0]; assign m1stg_norm_dbl_in2= (!mul_exp_in2_exp_eq_0) && m1stg_dblopa[0]; /////////////////////////////////////////////////////////////////////////////// // // Nan multiply inputs. // /////////////////////////////////////////////////////////////////////////////// assign m1stg_snan_sng_in1= (!mul_exp_in1_exp_neq_ffs) && (!mul_frac_in1_54) && (mul_frac_in1_53_32_neq_0) && m1stg_sngopa[1]; assign m1stg_snan_dbl_in1= (!mul_exp_in1_exp_neq_ffs) && (!mul_frac_in1_51) && mul_frac_in1_50_0_neq_0 && m1stg_dblopa[1]; assign m1stg_snan_sng_in2= (!mul_exp_in2_exp_neq_ffs) && (!mul_frac_in2_54) && (mul_frac_in2_53_32_neq_0) && m1stg_sngopa[1]; assign m1stg_snan_dbl_in2= (!mul_exp_in2_exp_neq_ffs) && (!mul_frac_in2_51) && mul_frac_in2_50_0_neq_0 && m1stg_dblopa[1]; assign m1stg_qnan_sng_in1= (!mul_exp_in1_exp_neq_ffs) && mul_frac_in1_54 && m1stg_sngopa[1]; assign m1stg_qnan_dbl_in1= (!mul_exp_in1_exp_neq_ffs) && mul_frac_in1_51 && m1stg_dblopa[1]; assign m1stg_qnan_sng_in2= (!mul_exp_in2_exp_neq_ffs) && mul_frac_in2_54 && m1stg_sngopa[1]; assign m1stg_qnan_dbl_in2= (!mul_exp_in2_exp_neq_ffs) && mul_frac_in2_51 && m1stg_dblopa[1]; assign m1stg_snan_in1= m1stg_snan_sng_in1 || m1stg_snan_dbl_in1; assign m1stg_snan_in2= m1stg_snan_sng_in2 || m1stg_snan_dbl_in2; assign m1stg_qnan_in1= m1stg_qnan_sng_in1 || m1stg_qnan_dbl_in1; assign m1stg_qnan_in2= m1stg_qnan_sng_in2 || m1stg_qnan_dbl_in2; dffe_s #(1) i_m2stg_snan_in1 ( .din (m1stg_snan_in1), .en (m6stg_step), .clk (rclk), .q (m2stg_snan_in1), .se (se), .si (), .so () ); dffe_s #(1) i_m2stg_snan_in2 ( .din (m1stg_snan_in2), .en (m6stg_step), .clk (rclk), .q (m2stg_snan_in2), .se (se), .si (), .so () ); dffe_s #(1) i_m2stg_qnan_in1 ( .din (m1stg_qnan_in1), .en (m6stg_step), .clk (rclk), .q (m2stg_qnan_in1), .se (se), .si (), .so () ); dffe_s #(1) i_m2stg_qnan_in2 ( .din (m1stg_qnan_in2), .en (m6stg_step), .clk (rclk), .q (m2stg_qnan_in2), .se (se), .si (), .so () ); assign m1stg_nan_sng_in1= (!mul_exp_in1_exp_neq_ffs) && (mul_frac_in1_54 || mul_frac_in1_53_32_neq_0) && m1stg_sngopa[2]; assign m1stg_nan_dbl_in1= (!mul_exp_in1_exp_neq_ffs) && (mul_frac_in1_51 || mul_frac_in1_50_0_neq_0) && m1stg_dblopa[2]; assign m1stg_nan_sng_in2= (!mul_exp_in2_exp_neq_ffs) && (mul_frac_in2_54 || mul_frac_in2_53_32_neq_0) && m1stg_sngopa[2]; assign m1stg_nan_dbl_in2= (!mul_exp_in2_exp_neq_ffs) && (mul_frac_in2_51 || mul_frac_in2_50_0_neq_0) && m1stg_dblopa[2]; assign m1stg_nan_in1= m1stg_nan_sng_in1 || m1stg_nan_dbl_in1; assign m1stg_nan_in2= m1stg_nan_sng_in2 || m1stg_nan_dbl_in2; dffe_s #(1) i_m2stg_nan_in2 ( .din (m1stg_nan_in2), .en (m6stg_step), .clk (rclk), .q (m2stg_nan_in2), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Infinity multiply inputs. // /////////////////////////////////////////////////////////////////////////////// assign m1stg_inf_sng_in1= (!mul_exp_in1_exp_neq_ffs) && (!mul_frac_in1_54) && (!mul_frac_in1_53_32_neq_0) && m1stg_sngopa[2]; assign m1stg_inf_dbl_in1= (!mul_exp_in1_exp_neq_ffs) && (!mul_frac_in1_51) && (!mul_frac_in1_50_0_neq_0) && m1stg_dblopa[2]; assign m1stg_inf_sng_in2= (!mul_exp_in2_exp_neq_ffs) && (!mul_frac_in2_54) && (!mul_frac_in2_53_32_neq_0) && m1stg_sngopa[2]; assign m1stg_inf_dbl_in2= (!mul_exp_in2_exp_neq_ffs) && (!mul_frac_in2_51) && (!mul_frac_in2_50_0_neq_0) && m1stg_dblopa[2]; assign m1stg_inf_in1= m1stg_inf_sng_in1 || m1stg_inf_dbl_in1; assign m1stg_inf_in2= m1stg_inf_sng_in2 || m1stg_inf_dbl_in2; assign m1stg_inf_in= m1stg_inf_in1 || m1stg_inf_in2; dffe_s #(1) i_m2stg_inf_in1 ( .din (m1stg_inf_in1), .en (m6stg_step), .clk (rclk), .q (m2stg_inf_in1), .se (se), .si (), .so () ); dffe_s #(1) i_m2stg_inf_in2 ( .din (m1stg_inf_in2), .en (m6stg_step), .clk (rclk), .q (m2stg_inf_in2), .se (se), .si (), .so () ); dffe_s #(1) i_m2stg_inf_in ( .din (m1stg_inf_in), .en (m6stg_step), .clk (rclk), .q (m2stg_inf_in), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Infinity/Nan multiply inputs. // /////////////////////////////////////////////////////////////////////////////// assign m1stg_infnan_sng_in1= (!mul_exp_in1_exp_neq_ffs) && m1stg_sngopa[3]; assign m1stg_infnan_dbl_in1= (!mul_exp_in1_exp_neq_ffs) && m1stg_dblopa[3]; assign m1stg_infnan_sng_in2= (!mul_exp_in2_exp_neq_ffs) && m1stg_sngopa[3]; assign m1stg_infnan_dbl_in2= (!mul_exp_in2_exp_neq_ffs) && m1stg_dblopa[3]; assign m1stg_infnan_in1= m1stg_infnan_sng_in1 || m1stg_infnan_dbl_in1; assign m1stg_infnan_in2= m1stg_infnan_sng_in2 || m1stg_infnan_dbl_in2; assign m1stg_infnan_in= m1stg_infnan_in1 || m1stg_infnan_in2; /////////////////////////////////////////////////////////////////////////////// // // Zero multiply inputs. // /////////////////////////////////////////////////////////////////////////////// assign m1stg_zero_in1= mul_exp_in1_exp_eq_0 && (!mul_frac_in1_53_0_neq_0) && (!mul_frac_in1_54); assign m1stg_zero_in2= mul_exp_in2_exp_eq_0 && (!mul_frac_in2_53_0_neq_0) && (!mul_frac_in2_54); assign m1stg_zero_in= m1stg_zero_in1 || m1stg_zero_in2; dffe_s #(1) i_m2stg_zero_in1 ( .din (m1stg_zero_in1), .en (m6stg_step), .clk (rclk), .q (m2stg_zero_in1), .se (se), .si (), .so () ); dffe_s #(1) i_m2stg_zero_in2 ( .din (m1stg_zero_in2), .en (m6stg_step), .clk (rclk), .q (m2stg_zero_in2), .se (se), .si (), .so () ); dffe_s #(1) i_m2stg_zero_in ( .din (m1stg_zero_in), .en (m6stg_step), .clk (rclk), .q (m2stg_zero_in), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Floating point multiply control pipeline. // /////////////////////////////////////////////////////////////////////////////// /////////////////////////////////////////////////////////////////////////////// // // Opcode pipeline- multiply input stage. // /////////////////////////////////////////////////////////////////////////////// assign m1stg_step= m6stg_stepa && (!m1stg_mul); assign m1stg_op_in[7:0]= ({8{(m1stg_step && (!reset))}} & (inq_op[7:0] & {8{inq_mul}})) | ({8{((!m6stg_step) && (!reset))}} & m1stg_op[7:0]); dff_s #(8) i_m1stg_op ( .din (m1stg_op_in[7:0]), .clk (rclk), .q (m1stg_op[7:0]), .se (se), .si (), .so () ); assign m1stg_mul_in= (m1stg_step && (!reset) && inq_mul) || ((!m6stg_step) && (!reset) && m1stg_mul); dff_s #(1) i_m1stg_mul ( .din (m1stg_mul_in), .clk (rclk), .q (m1stg_mul), .se (se), .si (), .so () ); dffe_s #(1) i_m1stg_sngop ( .din (inq_op[0]), .en (m6stg_step), .clk (rclk), .q (m1stg_sngop), .se (se), .si (), .so () ); dffe_s #(4) i_m1stg_sngopa ( .din ({4{inq_op[0]}}), .en (m6stg_step), .clk (rclk), .q (m1stg_sngopa[3:0]), .se (se), .si (), .so () ); dffe_s #(1) i_m1stg_dblop ( .din (inq_op[1]), .en (m6stg_step), .clk (rclk), .q (m1stg_dblop), .se (se), .si (), .so () ); dffe_s #(4) i_m1stg_dblopa ( .din ({4{inq_op[1]}}), .en (m6stg_step), .clk (rclk), .q (m1stg_dblopa[3:0]), .se (se), .si (), .so () ); assign m1stg_dblop_inv_in= (!inq_op[1]); dffe_s #(1) i_m1stg_dblop_inv ( .din (m1stg_dblop_inv_in), .en (m6stg_step), .clk (rclk), .q (m1stg_dblop_inv), .se (se), .si (), .so () ); dffe_s #(2) i_m1stg_rnd_mode ( .din (inq_rnd_mode[1:0]), .en (m6stg_step), .clk (rclk), .q (m1stg_rnd_mode[1:0]), .se (se), .si (), .so () ); dffe_s #(5) i_m1stg_id ( .din (inq_id[4:0]), .en (m6stg_step), .clk (rclk), .q (m1stg_id[4:0]), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Opcode decode- multiply stage 1. // /////////////////////////////////////////////////////////////////////////////// assign m1stg_fmul= (m1stg_op[7:0]==FMULS) || (m1stg_op[7:0]==FMULD) || (m1stg_op[7:0]==FSMULD); assign m1stg_fmul_dbl_dst= (m1stg_op[7:0]==FMULD) || (m1stg_op[7:0]==FSMULD); assign m1stg_fmuls= (m1stg_op[7:0]==FMULS); assign m1stg_fmuld= (m1stg_op[7:0]==FMULD); assign m1stg_fsmuld= (m1stg_op[7:0]==FSMULD); assign m1stg_opdec[4:0]= {m1stg_fmul, m1stg_fmul_dbl_dst, m1stg_fmuls, m1stg_fmuld, m1stg_fsmuld}; dffre_s #(5) i_m2stg_opdec ( .din (m1stg_opdec[4:0]), .en (m6stg_step), .rst (reset), .clk (rclk), .q (m2stg_opdec[4:0]), .se (se), .si (), .so () ); dffe_s #(2) i_m2stg_rnd_mode ( .din (m1stg_rnd_mode[1:0]), .en (m6stg_step), .clk (rclk), .q (m2stg_rnd_mode[1:0]), .se (se), .si (), .so () ); dffe_s #(5) i_m2stg_id ( .din (m1stg_id[4:0]), .en (m6stg_step), .clk (rclk), .q (m2stg_id[4:0]), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Opcode pipeline- multiply stage 2. // /////////////////////////////////////////////////////////////////////////////// assign m2stg_fmul= m2stg_opdec[4]; assign m2stg_fmuls= m2stg_opdec[2]; assign m2stg_fmuld= m2stg_opdec[1]; assign m2stg_fsmuld= m2stg_opdec[0]; dffre_s #(4) i_m3astg_opdec ( .din (m2stg_opdec[4:1]), .en (m6stg_step), .rst (reset), .clk (rclk), .q (m3astg_opdec[4:1]), .se (se), .si (), .so () ); dffe_s #(2) i_m3astg_rnd_mode ( .din (m2stg_rnd_mode[1:0]), .en (m6stg_step), .clk (rclk), .q (m3astg_rnd_mode[1:0]), .se (se), .si (), .so () ); dffe_s #(5) i_m3astg_id ( .din (m2stg_id[4:0]), .en (m6stg_step), .clk (rclk), .q (m3astg_id[4:0]), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Opcode pipeline- multiply stage 3a. // /////////////////////////////////////////////////////////////////////////////// dffre_s #(4) i_m3bstg_opdec ( .din (m3astg_opdec[4:1]), .en (m6stg_step), .rst (reset), .clk (rclk), .q (m3bstg_opdec[4:1]), .se (se), .si (), .so () ); dffe_s #(2) i_m3bstg_rnd_mode ( .din (m3astg_rnd_mode[1:0]), .en (m6stg_step), .clk (rclk), .q (m3bstg_rnd_mode[1:0]), .se (se), .si (), .so () ); dffe_s #(5) i_m3bstg_id ( .din (m3astg_id[4:0]), .en (m6stg_step), .clk (rclk), .q (m3bstg_id[4:0]), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Opcode pipeline- multiply stage 3b. // /////////////////////////////////////////////////////////////////////////////// dffre_s #(4) i_m3stg_opdec ( .din (m3bstg_opdec[4:1]), .en (m6stg_step), .rst (reset), .clk (rclk), .q (m3stg_opdec[4:1]), .se (se), .si (), .so () ); dffe_s #(2) i_m3stg_rnd_mode ( .din (m3bstg_rnd_mode[1:0]), .en (m6stg_step), .clk (rclk), .q (m3stg_rnd_mode[1:0]), .se (se), .si (), .so () ); dffe_s #(5) i_m3stg_id ( .din (m3bstg_id[4:0]), .en (m6stg_step), .clk (rclk), .q (m3stg_id[4:0]), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Opcode pipeline- multiply stage 3. // /////////////////////////////////////////////////////////////////////////////// assign m3stg_fmul= m3stg_opdec[4]; dffre_s #(4) i_m4stg_opdec ( .din (m3stg_opdec[4:1]), .en (m6stg_step), .rst (reset), .clk (rclk), .q (m4stg_opdec[4:1]), .se (se), .si (), .so () ); dffe_s #(2) i_m4stg_rnd_mode ( .din (m3stg_rnd_mode[1:0]), .en (m6stg_step), .clk (rclk), .q (m4stg_rnd_mode[1:0]), .se (se), .si (), .so () ); dffe_s #(5) i_m4stg_id ( .din (m3stg_id[4:0]), .en (m6stg_step), .clk (rclk), .q (m4stg_id[4:0]), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Opcode pipeline- multiply stage 4. // /////////////////////////////////////////////////////////////////////////////// assign m4stg_fmul= m4stg_opdec[4]; assign m4stg_fmuld= m4stg_opdec[1]; dffre_s #(4) i_m5stg_opdec ( .din (m4stg_opdec[4:1]), .en (m6stg_step), .rst (reset), .clk (rclk), .q (m5stg_opdec[4:1]), .se (se), .si (), .so () ); dffe_s #(2) i_m5stg_rnd_mode ( .din (m4stg_rnd_mode[1:0]), .en (m6stg_step), .clk (rclk), .q (m5stg_rnd_mode[1:0]), .se (se), .si (), .so () ); dffe_s #(5) i_m5stg_id ( .din (m4stg_id[4:0]), .en (m6stg_step), .clk (rclk), .q (m5stg_id[4:0]), .se (se), .si (), .so () ); dffre_s #(1) i_m5stg_fmulda ( .din (m4stg_fmuld), .en (m6stg_step), .rst (reset), .clk (rclk), .q (m5stg_fmulda), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Opcode pipeline- multiply stage 5. // /////////////////////////////////////////////////////////////////////////////// assign m5stg_fmul= m5stg_opdec[4]; assign m5stg_fmuls= m5stg_opdec[2]; assign m5stg_fmuld= m5stg_opdec[1]; assign m6stg_fmul_in= (m6stg_stepa && (!reset) && m5stg_fmul) || ((!m6stg_stepa) && (!reset) && m6stg_fmul); dffre_s #(3) i_m6stg_opdec ( .din (m5stg_opdec[4:2]), .en (m6stg_step), .rst (reset), .clk (rclk), .q (m6stg_opdec[4:2]), .se (se), .si (), .so () ); assign m6stg_id_in[9:0]= ({10{m6stg_stepa}} & {(m5stg_id[4:2]==3'o7), (m5stg_id[4:2]==3'o6), (m5stg_id[4:2]==3'o5), (m5stg_id[4:2]==3'o4), (m5stg_id[4:2]==3'o3), (m5stg_id[4:2]==3'o2), (m5stg_id[4:2]==3'o1), (m5stg_id[4:2]==3'o0), m5stg_id[1:0]}) | ({10{(!m6stg_stepa)}} & m6stg_id[9:0]); dffe_s #(10) i_m6stg_id ( .din (m6stg_id_in[9:0]), .en (m6stg_step), .clk (rclk), .q (m6stg_id[9:0]), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Opcode pipeline- multiply pipeline output. // /////////////////////////////////////////////////////////////////////////////// assign m6stg_fmul= m6stg_opdec[4]; assign m6stg_fmul_dbl_dst= m6stg_opdec[3]; assign m6stg_fmuls= m6stg_opdec[2]; assign m6stg_hold= m6stg_fmul && (!mul_dest_rdy); assign m6stg_holda= m6stg_fmul && (!mul_dest_rdya); assign m6stg_step= (!m6stg_hold); assign m6stg_stepa= (!m6stg_holda); // Austin update // Power management update assign mul_pipe_active_in = // mul pipe is executing a valid instr m1stg_fmul || m2stg_fmul || m3astg_opdec[4] || m3bstg_opdec[4] || m3stg_fmul || m4stg_fmul || m5stg_fmul || m6stg_fmul; dffre_s #(1) i_mul_pipe_active ( .din (mul_pipe_active_in), .en (1'b1), .rst (reset), .clk (rclk), .q (mul_pipe_active), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Multiply sign and exception logic. // /////////////////////////////////////////////////////////////////////////////// /////////////////////////////////////////////////////////////////////////////// // // Multiply sign inputs. // /////////////////////////////////////////////////////////////////////////////// dffe_s #(1) i_m1stg_sign1 ( .din (inq_in1_63), .en (m6stg_step), .clk (rclk), .q (m1stg_sign1), .se (se), .si (), .so () ); dffe_s #(1) i_m1stg_sign2 ( .din (inq_in2_63), .en (m6stg_step), .clk (rclk), .q (m1stg_sign2), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Multiply sign and exceptions. // // Multiply stage 1. // /////////////////////////////////////////////////////////////////////////////// dffe_s #(1) i_m2stg_sign1 ( .din (m1stg_sign1), .en (m6stg_step), .clk (rclk), .q (m2stg_sign1), .se (se), .si (), .so () ); dffe_s #(1) i_m2stg_sign2 ( .din (m1stg_sign2), .en (m6stg_step), .clk (rclk), .q (m2stg_sign2), .se (se), .si (), .so () ); assign m1stg_of_mask= (!m1stg_infnan_in); dffe_s #(1) i_m2stg_of_mask ( .din (m1stg_of_mask), .en (m6stg_step), .clk (rclk), .q (m2stg_of_mask), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Multiply sign and exceptions. // // Multiply stage 2. // /////////////////////////////////////////////////////////////////////////////// assign m2stg_sign= ((m2stg_sign1 && (!m2stg_snan_in2) && (!(m2stg_qnan_in2 && (!m2stg_snan_in1)))) ^ (m2stg_sign2 && (!(m2stg_snan_in1 && (!m2stg_snan_in2))) && (!(m2stg_qnan_in1 && (!m2stg_nan_in2))))) && (!(m2stg_inf_in && m2stg_zero_in)); dffe_s #(1) i_m3astg_sign ( .din (m2stg_sign), .en (m6stg_step), .clk (rclk), .q (m3astg_sign), .se (se), .si (), .so () ); assign m2stg_nv= m2stg_snan_in1 || m2stg_snan_in2 || (m2stg_zero_in1 && m2stg_inf_in2) || (m2stg_inf_in1 && m2stg_zero_in2); dffe_s #(1) i_m3astg_nv ( .din (m2stg_nv), .en (m6stg_step), .clk (rclk), .q (m3astg_nv), .se (se), .si (), .so () ); dffe_s #(1) i_m3astg_of_mask ( .din (m2stg_of_mask), .en (m6stg_step), .clk (rclk), .q (m3astg_of_mask), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Multiply sign and exceptions. // // Multiply stage 3a. // /////////////////////////////////////////////////////////////////////////////// dffe_s #(1) i_m3bstg_sign ( .din (m3astg_sign), .en (m6stg_step), .clk (rclk), .q (m3bstg_sign), .se (se), .si (), .so () ); dffe_s #(1) i_m3bstg_nv ( .din (m3astg_nv), .en (m6stg_step), .clk (rclk), .q (m3bstg_nv), .se (se), .si (), .so () ); dffe_s #(1) i_m3bstg_of_mask ( .din (m3astg_of_mask), .en (m6stg_step), .clk (rclk), .q (m3bstg_of_mask), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Multiply sign and exceptions. // // Multiply stage 3b. // /////////////////////////////////////////////////////////////////////////////// dffe_s #(1) i_m3stg_sign ( .din (m3bstg_sign), .en (m6stg_step), .clk (rclk), .q (m3stg_sign), .se (se), .si (), .so () ); dffe_s #(1) i_m3stg_nv ( .din (m3bstg_nv), .en (m6stg_step), .clk (rclk), .q (m3stg_nv), .se (se), .si (), .so () ); dffe_s #(1) i_m3stg_of_mask ( .din (m3bstg_of_mask), .en (m6stg_step), .clk (rclk), .q (m3stg_of_mask), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Multiply sign and exceptions. // // Multiply stage 3. // /////////////////////////////////////////////////////////////////////////////// dffe_s #(1) i_m4stg_sign ( .din (m3stg_sign), .en (m6stg_step), .clk (rclk), .q (m4stg_sign), .se (se), .si (), .so () ); dffe_s #(1) i_m4stg_nv ( .din (m3stg_nv), .en (m6stg_step), .clk (rclk), .q (m4stg_nv), .se (se), .si (), .so () ); dffe_s #(1) i_m4stg_of_mask ( .din (m3stg_of_mask), .en (m6stg_step), .clk (rclk), .q (m4stg_of_mask), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Multiply sign and exceptions. // // Multiply stage 4. // /////////////////////////////////////////////////////////////////////////////// dffe_s #(1) i_m5stg_sign ( .din (m4stg_sign), .en (m6stg_step), .clk (rclk), .q (m5stg_sign), .se (se), .si (), .so () ); dffe_s #(1) i_m5stg_nv ( .din (m4stg_nv), .en (m6stg_step), .clk (rclk), .q (m5stg_nv), .se (se), .si (), .so () ); dffe_s #(1) i_m5stg_of_mask ( .din (m4stg_of_mask), .en (m6stg_step), .clk (rclk), .q (m5stg_of_mask), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Multiply sign and exceptions. // // Multiply stage 5. // /////////////////////////////////////////////////////////////////////////////// dffe_s #(1) i_mul_sign_out ( .din (m5stg_sign), .en (m6stg_step), .clk (rclk), .q (mul_sign_out), .se (se), .si (), .so () ); dffe_s #(1) i_mul_nv_out ( .din (m5stg_nv), .en (m6stg_step), .clk (rclk), .q (mul_nv_out), .se (se), .si (), .so () ); assign m5stg_in_of= ((!m5stg_exp[12]) && m5stg_fmuld && (m5stg_exp[11] || (&m5stg_exp[10:0])) && m5stg_of_mask) || ((!m5stg_exp[12]) && m5stg_fmuls && ((|m5stg_exp[11:8]) || (&m5stg_exp[7:0])) && m5stg_of_mask); assign mul_of_out_tmp1_in= ((!m5stg_exp[12]) && m5stg_fmuld && (&m5stg_exp[10:1]) && m5stg_rndup && m5stg_of_mask) || ((!m5stg_exp[12]) && m5stg_fmuls && (&m5stg_exp[7:1]) && m5stg_rndup && m5stg_of_mask); dffe_s #(1) i_mul_of_out_tmp1 ( .din (mul_of_out_tmp1_in), .en (m6stg_step), .clk (rclk), .q (mul_of_out_tmp1), .se (se), .si (), .so () ); dffe_s #(1) i_mul_of_out_tmp2 ( .din (m5stg_in_of), .en (m6stg_step), .clk (rclk), .q (mul_of_out_tmp2), .se (se), .si (), .so () ); dffe_s #(1) i_mul_of_out_cout ( .din (m5stg_fracadd_cout), .en (m6stg_step), .clk (rclk), .q (mul_of_out_cout), .se (se), .si (), .so () ); assign mul_of_out= mul_of_out_tmp2 || (mul_of_out_tmp1 && mul_of_out_cout); assign mul_uf_out_in= (m5stg_exp[12] || (!(|m5stg_exp[11:0]))) && m5stg_frac_neq_0; dffe_s #(1) i_mul_uf_out ( .din (mul_uf_out_in), .en (m6stg_step), .clk (rclk), .q (mul_uf_out), .se (se), .si (), .so () ); assign mul_nx_out_in= (m5stg_fmuld && m5stg_frac_dbl_nx) || (m5stg_fmuls && m5stg_frac_sng_nx); dffe_s #(1) i_mul_nx_out ( .din (mul_nx_out_in), .en (m6stg_step), .clk (rclk), .q (mul_nx_out), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Multiply exception output. // /////////////////////////////////////////////////////////////////////////////// // Austin update // Overflow is always accompanied by inexact. // Previously this was handled within the FFU. // assign mul_exc_out[4:0]= {mul_nv_out, mul_of_out, mul_uf_out, 1'b0, mul_nx_out}; assign mul_exc_out[4:0] = {mul_nv_out, mul_of_out, mul_uf_out, 1'b0, (mul_nx_out || mul_of_out)}; // Overflow is always accompanied by inexact /////////////////////////////////////////////////////////////////////////////// // // Multiply pipeline control logic. // /////////////////////////////////////////////////////////////////////////////// /////////////////////////////////////////////////////////////////////////////// // // Select lines- multiply normalization and special input injection. // // Multiply stage 1. // /////////////////////////////////////////////////////////////////////////////// assign m2stg_frac1_dbl_norm= m1stg_norm_dbl_in1 && ((!(m1stg_infnan_dbl_in1 || m1stg_infnan_dbl_in2)) || (m1stg_snan_dbl_in1 && (!m1stg_snan_dbl_in2)) || (m1stg_qnan_dbl_in1 && (!m1stg_nan_dbl_in2))); assign m2stg_frac1_dbl_dnrm= m1stg_denorm_dbl_in1 && (!(m1stg_infnan_dbl_in1 || m1stg_infnan_dbl_in2)); assign m2stg_frac1_sng_norm= m1stg_norm_sng_in1 && ((!(m1stg_infnan_sng_in1 || m1stg_infnan_sng_in2)) || (m1stg_snan_sng_in1 && (!m1stg_snan_sng_in2)) || (m1stg_qnan_sng_in1 && (!m1stg_nan_sng_in2))); assign m2stg_frac1_sng_dnrm= m1stg_denorm_sng_in1 && (!(m1stg_infnan_sng_in1 || m1stg_infnan_sng_in2)); assign m2stg_frac1_inf= (m1stg_inf_in && (!m1stg_nan_in1) && (!m1stg_nan_in2)) || m1stg_snan_in2 || (m1stg_qnan_in2 && (!m1stg_snan_in1)); assign m2stg_frac2_dbl_norm= m1stg_norm_dbl_in2 && ((!(m1stg_infnan_dbl_in1 || m1stg_infnan_dbl_in2)) || m1stg_snan_dbl_in2 || (m1stg_qnan_dbl_in2 && (!m1stg_snan_dbl_in1))); assign m2stg_frac2_dbl_dnrm= m1stg_denorm_dbl_in2 && (!(m1stg_infnan_dbl_in1 || m1stg_infnan_dbl_in2)); assign m2stg_frac2_sng_norm= m1stg_norm_sng_in2 && ((!(m1stg_infnan_sng_in1 || m1stg_infnan_sng_in2)) || m1stg_snan_sng_in2 || (m1stg_qnan_sng_in2 && (!m1stg_snan_sng_in1))); assign m2stg_frac2_sng_dnrm= m1stg_denorm_sng_in2 && (!(m1stg_infnan_sng_in1 || m1stg_infnan_sng_in2)); assign m2stg_frac2_inf= (m1stg_inf_in && (!m1stg_nan_in1) && (!m1stg_nan_in2)) || (m1stg_snan_in1 && (!m1stg_snan_in2)) || (m1stg_qnan_in1 && (!m1stg_nan_in2)); assign m1stg_inf_zero_in= (m1stg_inf_in1 && m1stg_zero_in2) || (m1stg_zero_in1 && m1stg_inf_in2); assign m1stg_inf_zero_in_dbl= ((m1stg_inf_in1 && m1stg_zero_in2) || (m1stg_zero_in1 && m1stg_inf_in2)) && m1stg_fmul_dbl_dst; /////////////////////////////////////////////////////////////////////////////// // // Select lines and control logic- multiply leading 0 counts. // // Multiply stage 1. // /////////////////////////////////////////////////////////////////////////////// assign m2stg_ld0_1_in[5:0]= ({6{(m1stg_denorm_in1 && (!m1stg_infnan_in))}} & m1stg_ld0_1[5:0]); dffe_s #(6) i_m2stg_ld0_1 ( .din (m2stg_ld0_1_in[5:0]), .en (m6stg_step), .clk (rclk), .q (m2stg_ld0_1[5:0]), .se (se), .si (), .so () ); assign m2stg_ld0_2_in[5:0]= ({6{(m1stg_denorm_in2 && (!m1stg_infnan_in))}} & m1stg_ld0_2[5:0]); dffe_s #(6) i_m2stg_ld0_2 ( .din (m2stg_ld0_2_in[5:0]), .en (m6stg_step), .clk (rclk), .q (m2stg_ld0_2[5:0]), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Select lines- multiply exponent adder. // // Multiply stage 1. // /////////////////////////////////////////////////////////////////////////////// assign m2stg_exp_expadd= (!m1stg_infnan_in) && (!m1stg_zero_in); assign m2stg_exp_0bff= m1stg_fmuld && m1stg_infnan_in; assign m2stg_exp_017f= m1stg_fmuls && m1stg_infnan_in; assign m2stg_exp_04ff= m1stg_fsmuld && m1stg_infnan_in; assign m2stg_exp_zero= m1stg_zero_in && (!m1stg_infnan_in); /////////////////////////////////////////////////////////////////////////////// // // Total the leading 0's. // // Multiply stage 2. // /////////////////////////////////////////////////////////////////////////////// assign m2stg_ld0[6:0]= {1'b0, m2stg_ld0_1[5:0]} + {1'b0, m2stg_ld0_2[5:0]}; assign m2stg_ld0_inv[6:0]= (~m2stg_ld0[6:0]); dffe_s #(7) i_m3astg_ld0_inv ( .din (m2stg_ld0_inv[6:0]), .en (m6stg_step), .clk (rclk), .q (m3astg_ld0_inv[6:0]), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Leading 0's. // // Multiply stage 3a. // /////////////////////////////////////////////////////////////////////////////// dffe_s #(7) i_m3bstg_ld0_inv ( .din (m3astg_ld0_inv[6:0]), .en (m6stg_step), .clk (rclk), .q (m3bstg_ld0_inv[6:0]), .se (se), .si (), .so () ); /////////////////////////////////////////////////////////////////////////////// // // Post-normalization/denormalization shift count and direction. // // Multiply stage 3. // /////////////////////////////////////////////////////////////////////////////// dffe_s #(1) i_m4stg_expadd_eq_0 ( .din (m3stg_expadd_eq_0), .en (m6stg_step), .clk (rclk), .q (m4stg_expadd_eq_0), .se (se), .si (), .so () ); assign m3stg_exp_lte_0= (!(|m3stg_exp[11:0])) || m3stg_exp[12]; assign m4stg_right_shift_in= (!m3stg_expadd_lte_0_inv) && m3stg_exp_lte_0; dffe_s #(1) i_m4stg_right_shift ( .din (m4stg_right_shift_in), .en (m6stg_step), .clk (rclk), .q (m4stg_right_shift), .se (se), .si (), .so () ); assign m3stg_exp_minus1[5:0]= m3stg_exp[5:0] + 6'h3f; assign m3stg_exp_inv_plus2[5:0]= (~m3stg_exp[5:0]) + 6'h02; assign m3stg_exp_lt_neg57= ((!(&m3stg_exp[11:6])) || (!(|m3stg_exp[5:3]))) && m3stg_exp[12]; assign m4stg_sh_cnt_in[5:0]= ({6{((!m3stg_expadd_lte_0_inv) && (!m3stg_exp_lte_0))}} & m3stg_exp_minus1[5:0]) | ({6{((!m3stg_expadd_lte_0_inv) && m3stg_exp_lte_0 && m3stg_exp_lt_neg57)}} & 6'h39) | ({6{((!m3stg_expadd_lte_0_inv) && m3stg_exp_lte_0 && (!m3stg_exp_lt_neg57))}} & m3stg_exp_inv_plus2[5:0]) | ({6{m3stg_expadd_lte_0_inv}} & (~m3stg_ld0_inv[5:0])); /////////////////////////////////////////////////////////////////////////////// // // Select lines and control logic- multiply shifts for // post-normalization/denormalization. // // Multiply stage 4. // /////////////////////////////////////////////////////////////////////////////// assign m4stg_left_shift_step= (!m4stg_right_shift) && m6stg_step; assign m4stg_right_shift_step= m4stg_right_shift && m6stg_step; // Austin update // uarch timing fix // Endpoint: fpu_mul_exp_dp/i_m5stg_exp_pre2_10 // assign m4stg_inc_exp= (((!(|m4stg_exp[12:0])) && (!m4stg_right_shift) // && m4stg_shl_54) // || (m4stg_expadd_eq_0 && m4stg_right_shift // && m4stg_frac_105) // || ((!m4stg_right_shift) && m4stg_shl_55)) // && m6stg_step; // // assign m4stg_inc_exp_inv= (!m4stg_inc_exp) && m6stg_step; assign m4stg_inc_exp_54 = (!(|m4stg_exp[12:0])) && (!m4stg_right_shift); assign m4stg_inc_exp_55 = !m4stg_right_shift; assign m4stg_inc_exp_105 = m4stg_expadd_eq_0 && m4stg_right_shift && m4stg_frac_105; /////////////////////////////////////////////////////////////////////////////// // // Select lines and control logic- multiply rounding. // // Multiply stage 5. // /////////////////////////////////////////////////////////////////////////////// assign m5stg_rndup= ((((m5stg_rnd_mode[1:0]==2'b10) && (!m5stg_sign) && (m5stg_frac[2:0]!=3'b0)) || ((m5stg_rnd_mode[1:0]==2'b11) && m5stg_sign && (m5stg_frac[2:0]!=3'b0)) || ((m5stg_rnd_mode[1:0]==2'b00) && m5stg_frac[2] && ((m5stg_frac[1:0]!=2'b0) || m5stg_frac[3]))) && m5stg_fmuld) || ((((m5stg_rnd_mode[1:0]==2'b10) && (!m5stg_sign) && (m5stg_frac[31:0]!=32'b0)) || ((m5stg_rnd_mode[1:0]==2'b11) && m5stg_sign && (m5stg_frac[31:0]!=32'b0)) || ((m5stg_rnd_mode[1:0]==2'b00) && m5stg_frac[31] && ((m5stg_frac[30:0]!=31'b0) || m5stg_frac[32]))) && m5stg_fmuls); assign m5stg_to_0= (m5stg_rnd_mode[1:0]==2'b01) || ((m5stg_rnd_mode[1:0]==2'b10) && m5stg_sign) || ((m5stg_rnd_mode[1:0]==2'b11) && (!m5stg_sign)); assign m5stg_to_0_inv= (!m5stg_to_0); assign mul_frac_out_fracadd= m5stg_rndup && (!m5stg_in_of); assign mul_frac_out_frac= (!m5stg_rndup) && (!m5stg_in_of); assign mul_exp_out_exp_plus1= m5stg_rndup && (!m5stg_in_of); assign mul_exp_out_exp= (!m5stg_rndup) && (!m5stg_in_of); endmodule
Go to most recent revision | Compare with Previous | Blame | View Log