URL
https://opencores.org/ocsvn/m32632/m32632/trunk
Subversion Repositories m32632
[/] [m32632/] [trunk/] [rtl/] [DP_FPU.v] - Rev 28
Go to most recent revision | Compare with Previous | Blame | View Log
// +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // This file is part of the M32632 project // http://opencores.org/project,m32632 // // Filename: DP_FPU.v // Version: 2.0 // History: 1.0 first release of 30 Mai 2015 // Date: 14 August 2016 // // Copyright (C) 2016 Udo Moeller // // 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 source file is free software; you can redistribute it // and/or modify it under the terms of the GNU Lesser General // Public License as published by the Free Software Foundation; // either version 2.1 of the License, or (at your option) any // later version. // // This source 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 Lesser General Public License for more // details. // // You should have received a copy of the GNU Lesser General // Public License along with this source; if not, download it // from http://www.opencores.org/lgpl.shtml // // +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // Modules contained in this file: // 1. PREPDATA Prepare data for the big multiplier // 2. BCDADDER 4 bit BCD adder // 3. DFPU_BCD Binary coded decimal (BCD) adder and subtractor // 4. DFPU_ADDSUB Double precision floating point adder and subtractor // 5. DFPU_MISC Double precision floating point miscellaneous operations // 6. DFPU_MUL Double precision floating point multiplier // 7. SCANDIG Scan digit for leading one // 8. DIVI_PREP Prepare data for the divider // 9. DFPU_DIV The divider for all divide opcodes : double, single and integer // 10. DP_LOGIK Control logic and result path for different functions // 11. DP_FPU Top level of long operations datapath // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // 1. PREPDATA Prepare data for the big multiplier // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ module PREPDATA ( START, MEI, DFLOAT, BWD, SRC1, SRC2, MSD_1, MSD_2, LSD_1, LSD_2, LOAD_MSD, LOAD_LSD1, LOAD_LSD2 ); input [1:0] START; input MEI,DFLOAT; input [1:0] BWD; input [31:0] SRC1,SRC2; output [52:32] MSD_1,MSD_2; output [31:0] LSD_1,LSD_2; output LOAD_MSD,LOAD_LSD1,LOAD_LSD2; reg [31:0] LSD_1,LSD_2; assign MSD_1 = MEI ? 21'h0 : {1'b1,SRC1[19:0]}; assign MSD_2 = MEI ? 21'h0 : {1'b1,SRC2[19:0]}; always @(MEI or BWD or SRC1) casex ({MEI,BWD}) 3'b100 : LSD_1 = {24'h000000,SRC1[7:0]}; 3'b101 : LSD_1 = {16'h0000,SRC1[15:0]}; default : LSD_1 = SRC1; endcase always @(MEI or BWD or SRC2) casex ({MEI,BWD}) 3'b100 : LSD_2 = {24'h000000,SRC2[7:0]}; 3'b101 : LSD_2 = {16'h0000,SRC2[15:0]}; default : LSD_2 = SRC2; endcase assign LOAD_MSD = (START[0] & MEI) | (START[0] & DFLOAT); // 1. step data load at DFLOAT assign LOAD_LSD1 = (START[0] & MEI) | (START[1] & DFLOAT); // 2. step execute at DFLOAT assign LOAD_LSD2 = (START[1] & MEI) | (START[1] & DFLOAT); // 2. step execute at DFLOAT endmodule // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // 2. BCDADDER 4 bit BCD adder // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ module BCDADDER ( A_IN, B_IN, CY_IN, SUBP, OUT, CY_OUT ); input [3:0] A_IN,B_IN; input CY_IN; input SUBP; output [3:0] OUT; output CY_OUT; reg [4:0] data; wire [4:0] result; wire over; always @(B_IN) case (B_IN) 4'h0 : data = 5'h00; 4'h1 : data = 5'h1F; 4'h2 : data = 5'h1E; 4'h3 : data = 5'h1D; 4'h4 : data = 5'h1C; 4'h5 : data = 5'h1B; 4'h6 : data = 5'h1A; 4'h7 : data = 5'h19; 4'h8 : data = 5'h18; 4'h9 : data = 5'h17; default : data = 5'hxx; endcase assign result = {1'b0,A_IN} + (SUBP ? data : {1'b0,B_IN}) + {{4{SUBP & CY_IN}},CY_IN}; assign over = result[4] | (result[3] & (result[2] | result[1])); // if result<0 : -6 if result>9 : -10 assign OUT = result[3:0] - (SUBP ? {1'b0,result[4],result[4],1'b0} : {over,1'b0,over,1'b0}); assign CY_OUT = SUBP ? result[4] : over; endmodule // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // 3. DFPU_BCD Binary coded decimal (BCD) adder and subtractor // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ module DFPU_BCD ( BCLK, BRESET, START, DO_BCD, BWD, SRC1, SRC2, CY_IN, SUBP, BCD_Q, CY_OUT, BCD_DONE ); // Byte : 3 cycles in shortest case REG-REG, Word : 4 cycles and Double : 6 cycles input BCLK; input BRESET; input START; // START[1] input DO_BCD; // BCD Opcode is valid input [1:0] BWD; input [31:0] SRC1,SRC2; // Source , Destination, data is stable during operation input CY_IN; // comes from PSR input SUBP; // SUBP = 1 : SUBP , 0 : ADDP output reg [31:0] BCD_Q; output reg CY_OUT; // went to PSR if DONE is valid output BCD_DONE; reg run_bcd; reg [1:0] byte_cou; reg [15:0] datain; wire [7:0] result; wire carry,carry_lsd,carry_msd; // START : _/---\________________ // byte_cou : xxxxxx 0 x 1 x 2 x 3 x // BCD_DONE : _____/---\____________ if BWD = Byte always @(posedge BCLK or negedge BRESET) if (!BRESET) run_bcd <= 1'b0; else run_bcd <= (START & DO_BCD & (BWD != 2'd0)) | (run_bcd & (BWD != byte_cou)); always @(posedge BCLK) byte_cou <= START ? 2'd1 : byte_cou + {1'b0,run_bcd}; always @(*) casex ({START,byte_cou}) 3'b1_xx : datain = {SRC1[7:0], SRC2[7:0]}; 3'b0_0x : datain = {SRC1[15:8], SRC2[15:8]}; 3'b0_10 : datain = {SRC1[23:16],SRC2[23:16]}; 3'b0_11 : datain = {SRC1[31:24],SRC2[31:24]}; endcase assign carry = START ? CY_IN : CY_OUT; BCDADDER lsd_inst ( .A_IN(datain[3:0]), .B_IN(datain[11:8]), .CY_IN(carry), .SUBP(SUBP), .OUT(result[3:0]), .CY_OUT(carry_lsd) ); BCDADDER msd_inst ( .A_IN(datain[7:4]), .B_IN(datain[15:12]), .CY_IN(carry_lsd), .SUBP(SUBP), .OUT(result[7:4]), .CY_OUT(carry_msd) ); always @(posedge BCLK) CY_OUT <= carry_msd; always @(posedge BCLK) if (START) BCD_Q[7:0] <= result; always @(posedge BCLK) if (~byte_cou[1]) BCD_Q[15:8] <= result; always @(posedge BCLK) if (byte_cou == 2'd2) BCD_Q[23:16] <= result; always @(posedge BCLK) if (byte_cou == 2'd3) BCD_Q[31:24] <= result; assign BCD_DONE = (START & DO_BCD & (BWD == 2'd0)) | (run_bcd & (BWD == byte_cou)); endmodule // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // 4. DFPU_ADDSUB Double precision floating point adder and subtractor // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ module DFPU_ADDSUB ( BCLK, START, SRC1, SRC2, MAN1, MAN2, SRCFLAGS, BWD, SELECT, OUT, IOUT, CMPRES ); input BCLK; input [1:0] START; input [31:0] SRC1,SRC2; // The input data input [20:0] MAN1,MAN2; input [5:0] SRCFLAGS; // NAN, ZERO and SIGN of operands input [1:0] BWD; // size of integer input [3:0] SELECT; // upper 2 bits : R.T.F. code output [69:0] OUT; output [31:0] IOUT; // result of ROUNDLi/TRUNCLi/FLOORLi = R.T.F. output [1:0] CMPRES; reg [69:0] outreg; reg [31:0] IOUT; // MOViL : 2 cycles // ROUNDLi : 3 cycles (+TRUNC & FLOOR) // ADD/SUB : 4 cycles // CMP : 2 cycles // ++++++++++++++++++++++++++++++++++ // MOViL : 1. Pipeline stage : needs 3 cycles reg [31:0] movdat; reg [31:0] movif; reg sign_movif; always @(BWD or SRC1) casex(BWD) 2'b00 : movdat = {{24{SRC1[7]}}, SRC1[7:0]}; // Byte 2'b01 : movdat = {{16{SRC1[15]}},SRC1[15:0]}; // Word default : movdat = SRC1[31:0]; // Double endcase // This pipeline stage for better timing always @(posedge BCLK) movif <= ({32{movdat[31]}} ^ movdat) + {31'h0,movdat[31]}; // -2^31 is kept ! always @(posedge BCLK) sign_movif <= movdat[31]; // ROUNDLi/TRUNCLi/FLOORLi : 1. pipeline stage : can Opcode-Decoder deliver direct the 64 bit operand ? From register "yes" reg ovflag,ovflag2; reg rovfl; reg minint; wire [11:0] rexdiff,rexo; wire ganzklein; // Flag for 0 assign rexdiff = 12'h41D - {1'b0,SRC1[30:20]}; // 4..0 is the right shift value : like Single FP same value space // ovflag2 at the end of rounding : Check for Overflow always @(posedge BCLK) rovfl <= (ovflag | ovflag2) & (SELECT[1:0] == 2'b11) & ~minint; // a large positiv difference is a very small number : assign ganzklein = (~rexdiff[11] & (rexdiff[10:5] != 6'b0)); // 0 is implicit via SRC1[30:20]=0 // Detection of Overflow assign rexo = ({1'b0,SRC1[30:20]} - {11'h1FF,~BWD[1]}); // subtract B/W = 3FF , D = 3FE always @(BWD or rexo) // 0 ist in implicitly casex (BWD) 2'b00 : ovflag = (~rexo[11] & (rexo[10:3] != 8'h0)); // Exponent 0..7 because -128.4 => -128 2'b01 : ovflag = (~rexo[11] & (rexo[10:4] != 7'h0)); // Exponent 0..15 look above default : ovflag = (~rexo[11] & (rexo[10:5] != 6'h0)); // but Exponent only 0..30 endcase always @(posedge BCLK) if (START[1]) minint <= (SRC1 == 32'hC1E0_0000) & (SRC2 == 32'h0) & BWD[1]; // detection of -2^31 // ++++++++++++++++++++++++++++++++++++ // ADD/SUB : 1. Pipeline Stage : which operand ist bigger ? Exchange if neccessary // SUB/CMP : SRC2 - SRC1 reg ex_null,ma_null,ex_msb,ma_msb; reg [10:0] expo1,expo2; wire [11:0] exdiff,exdiff12; wire [20:0] madiff; wire switch,nan,sign,sign1,sign2; reg [5:0] shift1,shift2; // Pipeline register : reg [63:0] muxsrc2; wire [55:3] pipe1; // Nummbers for right shifter wire [5:0] shift; reg [2:0] pshift; reg vorz,addflag; wire [52:0] muxsrc1; wire [32:0] lowdiff; assign nan = (SELECT[1:0] == 2'b11) ? SRCFLAGS[1] : (~SELECT[1] & (SRCFLAGS[3] | SRCFLAGS[1])); // used at the end assign exdiff = {1'b0,SRC2[30:20]} - {1'b0,SRC1[30:20]}; // Difference of Exponents assign madiff = {1'b0,SRC2[19:0]} - {1'b0,SRC1[19:0]}; // Difference of Mantissa assign exdiff12 = {1'b0,SRC1[30:20]} - {1'b0,SRC2[30:20]}; // Diff. Exponents exchanged always @(posedge BCLK) if (START[0]) begin ex_null <= (exdiff[10:0] == 11'h0); ma_null <= (madiff[19:0] == 20'h0); ex_msb <= exdiff[11]; ma_msb <= madiff[20]; shift1 <= (exdiff[10:6] != 5'h0) ? 6'h3F : exdiff[5:0]; shift2 <= (exdiff12[10:6] != 5'h0) ? 6'h3F : exdiff12[5:0]; expo1 <= SRC1[30:20]; expo2 <= SRC2[30:20]; end assign lowdiff = {1'b0,SRC2} - {1'b0,SRC1}; // LSD compare assign switch = ex_msb | (ex_null & (ma_msb | (ma_null & lowdiff[32]))); // exchange ? assign muxsrc1 = switch ? {MAN2,SRC2} : {MAN1,SRC1}; assign pipe1 = SELECT[1] ? (ganzklein ? 53'd0 : {1'b1,SRC1[19:0],SRC2}) : muxsrc1; // feeding of R.T.F. assign shift = SELECT[1] ? {1'b0,rexdiff[4:0]} : (switch ? shift2 : shift1); always @(posedge BCLK) // Pipeline Reg begin muxsrc2 <= switch ? {expo1,MAN1,SRC1} : {expo2,MAN2,SRC2}; // Incl. Exponent & "1" of mantisse pshift <= shift[2:0]; end // SRC2 SRC1 : switch = 0 SRC2 SRC1 : switch = 1 // 5 + 3 : +(5 + 3) = 8 3 + 5 : +(5 + 3) = 8 SELECT[0] = 0 // 5 + (-3) : +(5 - 3) = 2 3 + (-5) : -(5 - 3) = -2 // (-5) + 3 : -(5 - 3) = -2 (-3) + 5 : +(5 - 3) = 2 // (-5) + (-3) : -(5 + 3) = -8 (-3) + (-5) : -(5 + 3) = -8 // 5 - 3 : +(5 - 3) = 2 3 - 5 : -(5 - 3) = -2 SELECT[0] = 1 // 5 - (-3) : +(5 + 3) = 8 3 - (-5) : +(5 + 3) = 8 // (-5) - 3 : -(5 + 3) = -8 (-3) - 5 : -(5 + 3) = -8 // (-5) - (-3) : -(5 - 3) = -2 (-3) - (-5) : +(5 - 3) = 2 assign sign1 = SRCFLAGS[4]; assign sign2 = SRCFLAGS[5]; always @(posedge BCLK) // Pipeline Reg begin vorz <= switch ? (SELECT[0] ^ sign1) : sign2; addflag <= ~(SELECT[0] ^ (sign1 ^ sign2)); end // CMPF : 1. Pipeline Stage : first result : is stored one level higer in Reg assign CMPRES[1] = ~CMPRES[0] & (switch ? ~sign1 : sign2); // look table above assign CMPRES[0] = (ex_null & ma_null & (sign1 == sign2) & (lowdiff == 33'h0)) | (SRCFLAGS[2] & SRCFLAGS[0]); // ++++++++++++++++++++++++++++++++++ // ADD/SUB + ROUND/TRUNC : 2. Step : Barrelshifter to the right --> wire [55:0] brshifta,brshiftb,brshiftd,brshifte,brshiftf; reg [55:0] brshiftc; // 5..33322222222221111111111 is this picture still correct ? Took over from Single FP // 5..2109876543210987654321098765432-10 // 1..VVVVVVVVVVVVVVVVVVVVVVVV0000000-00 // last 2 bit for rounding assign brshifta = shift[5] ? {32'h0, pipe1[55:33], (pipe1[32:3] != 30'h0)} : {pipe1,3'h0}; assign brshiftb = shift[4] ? {16'h0,brshifta[55:17],(brshifta[16:0] != 17'h0)} : brshifta; always @(posedge BCLK) brshiftc <= shift[3] ? { 8'h0, brshiftb[55:9], (brshiftb[8:0] != 9'h0)} : brshiftb; assign brshiftd = pshift[2] ? { 4'h0, brshiftc[55:5], (brshiftc[4:0] != 5'h0)} : brshiftc; assign brshifte = pshift[1] ? { 2'h0, brshiftd[55:3], (brshiftd[2:0] != 3'h0)} : brshiftd; assign brshiftf = pshift[0] ? { 1'b0, brshifte[55:2], (brshifte[1:0] != 2'h0)} : brshifte; // ++++++++++++++++++++++++++++++++++ // ROUNDLi/TRUNCLi/FLOORLi : 3. Step : round to Integer reg car_ry; wire [1:0] inex; wire [32:0] iadder; wire restbits; assign restbits = (brshiftf[23:0] != 24'h0); assign inex = {brshiftf[24],restbits}; // Inexact-Flag-Data transfered to multiplexer at the end always @(SELECT or sign1 or brshiftf or restbits or inex or ganzklein) casex (SELECT[3:2]) 2'b00 : car_ry = sign1 ^ (((brshiftf[25:24] == 2'b11) & ~restbits) | (inex == 2'b11)); // ROUNDLi 2'b1x : car_ry = sign1 ? (~ganzklein & (inex == 2'b00)) : 1'b0; // +numbers like TRUNCLi, -numbers to "-infinity" round default : car_ry = sign1; // TRUNCLi , simple cut off endcase assign iadder = (sign1 ? {2'b11,~brshiftf[55:25]} : {2'b0,brshiftf[55:25]}) + {32'h0,car_ry}; always @(posedge BCLK) IOUT <= minint ? 32'h8000_0000 : iadder[31:0]; always @(iadder or BWD or sign1) // special overflow detection i.e. -129 to -255 at Byte casex (BWD) // or 127.9 -> 128 = error ! 2'b00 : ovflag2 = (iadder[8] != iadder[7]); // Byte 2'b01 : ovflag2 = (iadder[16] != iadder[15]); // Word default : ovflag2 = (iadder[32] != iadder[31]); // Double endcase // ++++++++++++++++++++++++++++++++++ // ADD/SUB : 3. Step : Addition or Subtraction wire [67:0] result; wire [55:0] blshifti; wire [12:0] shiftl; wire shift_32; wire [65:0] add_q; // The central adder : the subtraction needs 3 Guard-Bits after LSB for correct rounding assign result = {1'b0,muxsrc2,3'b000} + (addflag ? {12'h0,brshiftf} : {12'hFFF,~brshiftf}) + {67'd0,~addflag}; assign blshifti = SELECT[1] ? {movif,24'h0} : result[55:0]; // Feeding of MOViL, comes from Register assign shiftl = SELECT[1] ? 13'h041E : {1'b0,result[67:56]}; // MOViL assign shift_32 = (blshifti[55:24] == 32'h0); // In case of ADD the result bypasses the barrelshifter : LSB of exponent has changed assign add_q = (muxsrc2[53] != result[56]) ? {result[67:3],(result[2:0] != 3'b000)} : {result[67:56],result[54:2],(result[1:0] != 2'b00)} ; // ++++++++++++++++++++++++++++++++++ // ADD/SUB : 4. Step : Barrelshifter left for SUB and MOViF : wire shift_16,shift_8,shift_4,shift_2,shift_1,zero; wire [1:0] lsb_bl; wire [55:0] blshifta,blshiftb,blshiftc,blshiftd,blshifte,blshiftf; wire [12:0] expol; assign blshifta = shift_32 ? {blshifti[23:0],32'h0} : blshifti; assign shift_16 = (blshifta[55:40] == 16'h0); assign blshiftb = shift_16 ? {blshifta[39:0],16'h0} : blshifta; assign shift_8 = (blshiftb[55:48] == 8'h00); assign blshiftc = shift_8 ? {blshiftb[47:0],8'h0} : blshiftb; assign shift_4 = (blshiftc[55:52] == 4'h0); assign blshiftd = shift_4 ? {blshiftc[51:0],4'h0} : blshiftc; assign shift_2 = (blshiftd[55:54] == 2'b00); assign blshifte = shift_2 ? {blshiftd[53:0],2'b0} : blshiftd; assign shift_1 = ~blshifte[55]; assign blshiftf = shift_1 ? {blshifte[54:0],1'b0} : blshifte; // Overflow at ROUNDLi/TRUNCLi/FLOORLi is shown in overflow of exponent , SELECT[1] is then 1 assign expol = shiftl - {7'h00,shift_32,shift_16,shift_8,shift_4,shift_2,shift_1}; // Inexact at ROUNDLi/TRUNCLi/FLOORLi : evaluation for all one level higher assign lsb_bl = (SELECT == 2'b11) ? inex : {blshiftf[2],(blshiftf[1:0] != 2'b0)}; assign zero = (~SELECT[1] & SRCFLAGS[2] & SRCFLAGS[0]) | ((blshifti == 56'h0) & ((~addflag & ~SELECT[1]) | (SELECT[1:0] == 2'b10))); assign sign = SELECT[1] ? sign_movif : (vorz & ~zero); // sign for MOViL // 2. Pipeline register for ADD , SUB and MOViL always @(posedge BCLK) outreg <= (addflag & ~SELECT[1]) ? {nan,zero,sign,1'b0,add_q} : {nan,zero,sign,expol,blshiftf[54:3],lsb_bl}; // ++++++++++++++++++++++++++++++++++ assign OUT = {outreg[69:67],(rovfl ? 2'b01 : outreg[66:65]),outreg[64:0]}; endmodule // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // 5. DFPU_MISC Double precision floating point miscellaneous operations // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ module DFPU_MISC ( BCLK, START, SRC1, SRC2, MAN2, SRCFLAGS, MODE, OUT ); input BCLK; input [1:0] START; input [31:0] SRC1,SRC2; input [19:0] MAN2; input [5:0] SRCFLAGS; input [3:0] MODE; output [69:0] OUT; reg [69:0] OUT; reg [63:0] daten; wire sign; wire [12:0] lexpo,sexpo; wire [69:0] scalb_res,logb_res,fl_lf; always @(posedge BCLK) if (START[1]) daten <= {SRC1,SRC2}; assign sign = daten[63]; // +++++++++++++++++++++++++++ MOVFL and MOVLF +++++++++++++++++++++++++++++++++++ assign lexpo = {5'b0,daten[62:55]} + 13'h0380; // -7F + 3FF assign sexpo = (daten[62:52] > 11'h47E) ? 13'h0FFF : ((daten[62:52] < 11'h381) ? 13'h0 : {2'b0,{4{daten[62]}},daten[58:52]}); assign fl_lf = MODE[0] ? {SRCFLAGS[1:0],sign,lexpo,daten[54:32],31'h0} // MOVFL : {SRCFLAGS[1:0],sign,sexpo,daten[51:29],28'h0,daten[29:28],(daten[27:0] != 28'h0)}; // MOVLF // +++++++++++++++++++++++++++ LOGBf +++++++++++++++++++++++++++++++++++ wire [9:0] sel_data,unbiased,shift_l8,shift_l4,shift_l2; wire [8:0] shift_l; wire posi_8,posi_4,posi_2,posi_1; wire [4:0] calc_exp; wire [6:0] logb_exp; assign sel_data = MODE[1] ? {{3{~daten[62]}},daten[61:55]} : daten[61:52]; assign unbiased = daten[62] ? (sel_data + 10'h001) : ~sel_data; // detection of leading "1" assign posi_8 = (unbiased[9:2] == 8'h00); assign shift_l8 = posi_8 ? {unbiased[1:0],8'h00} : unbiased; assign posi_4 = (shift_l8[9:6] == 4'h0); assign shift_l4 = posi_4 ? {shift_l8[5:0],4'h0} : shift_l8; assign posi_2 = (shift_l4[9:8] == 2'b00); assign shift_l2 = posi_2 ? {shift_l4[7:0],2'b0} : shift_l4; assign posi_1 = ~shift_l2[9]; assign shift_l = posi_1 ? {shift_l2[7:0],1'b0} : shift_l2[8:0]; // top bit is hidden "1" assign calc_exp = 5'h08 - {1'b0,posi_8,posi_4,posi_2,posi_1}; // Minimum is "F" = for exponent +/-1 <=> 2^0 // exponent is set one level higher for F and L assign logb_exp = MODE[1] ? {{4{~calc_exp[4]}},{3{calc_exp[4]}}} : {~calc_exp[4],{6{calc_exp[4]}}}; assign logb_res = {SRCFLAGS[1],1'b0,~daten[62],2'b00,logb_exp,calc_exp[3:0],shift_l,45'h0}; // ++++++++++++++++++++++++ SCALBf ++++++++++++++++++++++++++++++++++ reg [3:0] rshift; reg [10:0] shf_r0,dexpo; // dexpo = Exponent Destination reg huge; reg svorz,dvorz; wire [10:0] shf_r1,shf_r2,shf_r4,shf_r8; wire [12:0] addexp,newexp,finexp; wire nan; always @(posedge BCLK) // 2**0,9.. is transformed to 2**0 = 1 -> no change at SRC2 if (START[0]) begin shf_r0 <= ( SRC1[30] | ((SRC1[29:23] == 7'h7F) & (MODE[1] | (SRC1[22:20] == 3'd7))) ) ? (MODE[1] ? {4'd1,SRC1[22:16]} : {1'b1,SRC1[19:10]}) : 11'd0; rshift <= MODE[1] ? 4'd6 - SRC1[26:23] : 4'd9 - SRC1[23:20]; huge <= MODE[1] ? ( SRC1[30] & ((SRC1[29:26] != 4'd0) | (SRC1[25:23] == 3'h7)) ) // >406 in Double Style : ( SRC1[30] & ((SRC1[29:24] != 6'd0) | (SRC1[23] & (SRC1[22] | SRC1[21]))) ); // >409 svorz <= SRC1[31]; dvorz <= SRC2[31]; dexpo <= MODE[1] ? {3'd0,SRC2[30:23]} : SRC2[30:20]; end assign shf_r1 = rshift[0] ? {1'b0,shf_r0[10:1]} : shf_r0; // a mini-TRUNC of 11 Bits assign shf_r2 = rshift[1] ? {2'd0,shf_r1[10:2]} : shf_r1; assign shf_r4 = rshift[2] ? {4'd0,shf_r2[10:4]} : shf_r2; assign shf_r8 = rshift[3] ? {8'd0,shf_r4[10:8]} : shf_r4; assign addexp = svorz ? {2'd0,dexpo} - {2'd0,shf_r8} : {2'd0,dexpo} + {2'd0,shf_r8}; assign newexp = MODE[1] ? {addexp[9:8],{3{addexp[7]}},addexp[7:0]} : addexp[12:0]; assign finexp = SRCFLAGS[2] ? {3'd0,newexp[9:0]} // never an Overflow if SRC2 = 0.0 ! : {(huge ? {svorz,1'b1} : newexp[12:11]),newexp[10:0]}; // Overflow or Underflow assign nan = SRCFLAGS[3] | SRCFLAGS[1]; assign scalb_res = MODE[1] ? // Mantisse doesn't change ! {nan,SRCFLAGS[2],daten[31],finexp,daten[22:0],daten[28:0],2'b00} : {nan,SRCFLAGS[2],dvorz,finexp,MAN2,daten[31:0],2'b00}; // ++++++++++++++++++++++++ Output ++++++++++++++++++++++++++++++++++++++++++++++++++++++ always @(posedge BCLK) OUT <= MODE[3] ? (MODE[2] ? logb_res : scalb_res) : fl_lf ; // LOGB/SCALB : MOVLF/MOVFL endmodule // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // 6. DFPU_MUL Double precision floating point multiplier // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ module DFPU_MUL ( BCLK, SRC1, SRC2, START, MRESULT, SRCFLAGS, OUT ); input BCLK; input [31:0] SRC1,SRC2; input START; // that is START[0] input [105:0] MRESULT; input [5:0] SRCFLAGS; // NAN and ZERO flags output [69:0] OUT; // The result reg [69:0] OUT; reg [12:0] exponent; wire orlow; wire [12:0] expoh,expol; wire [1:0] restlow,resthigh; wire zero,nan,sign; assign zero = SRCFLAGS[2] | SRCFLAGS[0]; // one is NULL -> NULL is the result assign nan = SRCFLAGS[3] | SRCFLAGS[1]; // one is NAN -> error assign sign = (SRCFLAGS[5] ^ SRCFLAGS[4]) & ~zero; assign orlow = (MRESULT[50:0] != 51'b0); assign restlow = {MRESULT[51],orlow}; assign resthigh = {MRESULT[52],(MRESULT[51] | orlow)}; always @(posedge BCLK) if (START) exponent <= {2'b00,SRC1[30:20]} + {2'b00,SRC2[30:20]}; assign expoh = exponent - 13'h03FE; assign expol = exponent - 13'h03FF; // for MSB if MRESULT=0 always @(posedge BCLK) OUT <= MRESULT[105] ? {nan,zero,sign,expoh,MRESULT[104:53],resthigh} // 52 Bit Mantissa : {nan,zero,sign,expol,MRESULT[103:52],restlow}; endmodule // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // 7. SCANDIG Scan digit for leading one // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ module SCANDIG (DIN, MBIT, LBIT, NONZ); input [3:0] DIN; output MBIT,LBIT,NONZ; assign MBIT = DIN[3] | DIN[2]; // 1xxx = 11 assign LBIT = DIN[3] | (DIN[3:1] == 3'b001); // 01xx = 10 assign NONZ = (DIN != 4'd0); // 001x = 01 endmodule // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // 8. DIVI_PREP Prepare data for the divider // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ module DIVI_PREP (SRC, BWD, NOT_DEI, EXTDATA, DOUT, MSB, NULL, MINUS); input [31:0] SRC; input [1:0] BWD; input NOT_DEI; input EXTDATA; output [31:0] DOUT; output [4:0] MSB; output NULL; output MINUS; reg [31:0] double; reg [2:0] select; wire [1:0] modus; wire [7:0] mbits,lbits,dnonz; assign modus = (NOT_DEI | EXTDATA) ? BWD : {(BWD[1] | BWD[0]),1'b1}; always @(modus or SRC or NOT_DEI) casex (modus) 2'b00 : double = {{24{SRC[7] & NOT_DEI}},SRC[7:0]}; 2'b01 : double = {{16{SRC[15] & NOT_DEI}},SRC[15:0]}; 2'b1x : double = SRC; endcase assign MINUS = double[31] & NOT_DEI; assign DOUT = ({32{MINUS}} ^ double) + {31'h0,MINUS}; // assign DOUT = MINUS ? (32'd0 - double) : double; // now find most significant set bit : FFS SCANDIG digit_0 (.DIN(DOUT[3:0]), .MBIT(mbits[0]), .LBIT(lbits[0]), .NONZ(dnonz[0]) ); SCANDIG digit_1 (.DIN(DOUT[7:4]), .MBIT(mbits[1]), .LBIT(lbits[1]), .NONZ(dnonz[1]) ); SCANDIG digit_2 (.DIN(DOUT[11:8]), .MBIT(mbits[2]), .LBIT(lbits[2]), .NONZ(dnonz[2]) ); SCANDIG digit_3 (.DIN(DOUT[15:12]), .MBIT(mbits[3]), .LBIT(lbits[3]), .NONZ(dnonz[3]) ); SCANDIG digit_4 (.DIN(DOUT[19:16]), .MBIT(mbits[4]), .LBIT(lbits[4]), .NONZ(dnonz[4]) ); SCANDIG digit_5 (.DIN(DOUT[23:20]), .MBIT(mbits[5]), .LBIT(lbits[5]), .NONZ(dnonz[5]) ); SCANDIG digit_6 (.DIN(DOUT[27:24]), .MBIT(mbits[6]), .LBIT(lbits[6]), .NONZ(dnonz[6]) ); SCANDIG digit_7 (.DIN(DOUT[31:28]), .MBIT(mbits[7]), .LBIT(lbits[7]), .NONZ(dnonz[7]) ); always @(dnonz) casex (dnonz[7:1]) 7'b1xxx_xxx : select = 3'b111; 7'b01xx_xxx : select = 3'b110; 7'b001x_xxx : select = 3'b101; 7'b0001_xxx : select = 3'b100; 7'b0000_1xx : select = 3'b011; 7'b0000_01x : select = 3'b010; 7'b0000_001 : select = 3'b001; default : select = 3'b000; endcase assign NULL = (dnonz == 8'd0); assign MSB = {select,mbits[select],lbits[select]}; endmodule // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // 9. DFPU_DIV The divider for all divide opcodes : double, single and integer // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ module DFPU_DIV ( BCLK, BRST, START, SRC1, SRC2, MAN1, MAN2, SRCFLAGS, FL, BWD, OPCODE, OUT, DONE, DIVI_OUT, DVZ_TRAP, DEI_OVF ); // This version needs for Double 28+1 cycles if MAN1<MAN2 otherwise 28+2. // For Single it needs 13+1 cyckes or 13+2. input BCLK,BRST; input [3:0] START; // START & recognized Divider Operation input [31:0] SRC1,SRC2; // input data input [20:0] MAN1,MAN2; input [5:0] SRCFLAGS; // NAN and ZERO input FL; input [1:0] BWD; input [2:0] OPCODE; // for all DIVi variants output [69:0] OUT; // the result output reg DONE; // Pipeline-Flag output [63:0] DIVI_OUT; // for Integer Division output DVZ_TRAP; // Divide by Zero Trap output [1:0] DEI_OVF; // DEI Overflow // ++++++++++++++ for Integer Division ++++++++++++++ reg run_divi; reg divi_pipe1,divi_pipe2,divi_pipe3,divi_pipe4; reg neg_src1,neg_src2,nul_src2; reg [4:0] msb_src1; reg [5:0] msb_src2; reg [31:0] ivalue,src2_reg,pipe_reg; reg [4:0] divi_counter; reg sub_case; reg negativ; reg [32:0] divi_result; reg [63:0] DIVI_OUT; reg DVZ_TRAP,dvz_pipe; reg sel_in; reg [62:0] din_mux; reg dei_pipe; reg extdata; // extended data : 2 data packets, only apply to DEI reg [2:0] addoff; reg next_msb2; reg [31:0] dei_result; reg [1:0] DEI_OVF; wire [31:0] i_in; wire [37:0] i_out; wire [6:0] diff_msb; wire [5:1] shift_r; wire [62:0] shift_2; wire [62:0] shift_4; wire [62:0] shift_8; wire [62:0] shift_16; wire [64:0] shift_32; wire stop_divi,neg_flag; wire rest_null,plus_1,ist_null; wire not_dei; wire valdata; // Data <> 0 at DEI // ++++++++++++++ Floating Point & calculation path ++++++++ reg [69:0] OUT; reg [32:0] save1; reg runflag; reg [55:0] dreimal; reg [56:0] divreg,divsr; reg [31:0] divreg_ext; reg [12:0] exponent; wire load_src1,load_src2; wire [56:0] sub1,sub2,sub3; wire [32:0] src_1; wire [20:0] man_1; wire [12:0] expoh,expol,offset; wire restlsb,restlow,resthigh; wire zero,nan,sign,ende; wire orlow_s,orlow_d; wire short; // +++++++++++++++++++++++++++ Integer Division, DEI +++++++++++++++++++++++++++ assign not_dei = OPCODE[2]; // 0 = DEI always @(posedge BCLK) if (START[3]) extdata <= ~START[1]; // during START[0] for SRC1 not valid always @(posedge BCLK or negedge BRST) if (!BRST) run_divi <= 1'b0; else run_divi <= (START[3] & ~ist_null) | (~divi_pipe4 & run_divi); // Abort at DVZ Trap always @(posedge BCLK) divi_pipe1 <= START[3] & ~ist_null; // no start if SRC1 = 0 : DVZ Trap always @(posedge BCLK) dei_pipe <= divi_pipe1 & extdata; always @(posedge BCLK) divi_pipe2 <= extdata ? dei_pipe : divi_pipe1; always @(posedge BCLK) src2_reg <= SRC2; always @(posedge BCLK) sel_in <= START[3] | divi_pipe1; // two times data for DEI assign i_in = sel_in ? src2_reg : SRC1; DIVI_PREP prep_inst ( .SRC(i_in), .BWD(BWD), .NOT_DEI(not_dei), .EXTDATA(extdata | START[0]), .DOUT(i_out[31:0]), .MSB(i_out[36:32]), .NULL(ist_null), .MINUS(i_out[37]) ); always @(posedge BCLK) dvz_pipe <= START[3] & ist_null; // Pulse 1 cycle long always @(posedge BCLK) DVZ_TRAP <= dvz_pipe; // one cycle later if DEI with extdata always @(posedge BCLK) if (START[3]) begin neg_src1 <= i_out[37]; msb_src1 <= i_out[36:32]; end always @(posedge BCLK) if (divi_pipe1) begin nul_src2 <= ist_null; neg_src2 <= i_out[37]; end always @(posedge BCLK) ivalue <= i_out[31:0]; // The following is only for DEI : always @(posedge BCLK) pipe_reg <= {32{extdata}} & ivalue; // Register must be 0 if not used assign valdata = extdata & ~ist_null; always @(BWD or valdata) casex (BWD) 2'b00 : addoff = { 1'b0, 1'b0,valdata}; 2'b01 : addoff = { 1'b0,valdata, 1'b0}; default : addoff = {valdata, 1'b0, 1'b0}; endcase always @(posedge BCLK) next_msb2 <= extdata & ist_null & divi_pipe1; // Special case at DEI : MSD = 0 always @(posedge BCLK) if (divi_pipe1) msb_src2 <= {addoff[2],(addoff[1:0] | i_out[36:35]),i_out[34:32]}; else if (next_msb2) msb_src2 <= {1'b0,i_out[36:32]}; // Shifter for Source2 assign diff_msb = {1'b0,msb_src2} - {2'b0,msb_src1}; // negativ shift limited to 0 : Source2=0 calculated without special handling, result always 0 assign shift_r = diff_msb[6] ? 5'd0 : diff_msb[5:1]; // LSB does not count always @(BWD or extdata or ivalue or pipe_reg) casex ({BWD,extdata}) 3'b0x0 : din_mux = {31'b0,ivalue}; // the normal case for all except DEI 3'b001 : din_mux = {23'b0,pipe_reg,ivalue[7:0]}; 3'b011 : din_mux = {15'b0,pipe_reg,ivalue[15:0]}; default : din_mux = {pipe_reg[30:0],ivalue}; // 63 Bit wide endcase assign shift_2 = shift_r[1] ? din_mux : {din_mux[60:0], 2'b0}; assign shift_4 = shift_r[2] ? shift_2 : {shift_2[58:0], 4'b0}; assign shift_8 = shift_r[3] ? shift_4 : {shift_4[54:0], 8'b0}; assign shift_16 = shift_r[4] ? shift_8 : {shift_8[46:0],16'b0}; // Result is 63 Bit wide // 65 Bit result because of DEI assign shift_32 = shift_r[5] ? {1'b0,pipe_reg,ivalue} : {shift_16,2'b00}; // special case DEI : 32 times shift always @(posedge BCLK or negedge BRST) // Flag for rounding, only if DEST <>0 if (!BRST) divi_pipe3 <= 1'b0; else divi_pipe3 <= divi_pipe2 | (divi_pipe3 & ~stop_divi); always @(posedge BCLK) if (divi_pipe2) divi_counter <= shift_r; else divi_counter <= divi_counter - {4'b000,~stop_divi}; // should stop at 0 assign stop_divi = (divi_counter == 5'h0); // caclulation ready always @(posedge BCLK) divi_pipe4 <= divi_pipe3 & stop_divi; assign neg_flag = neg_src1 ^ neg_src2; assign rest_null = (divreg[33:2] == 32'h0); always @(posedge BCLK) sub_case <= neg_flag & ~nul_src2; // little help for MODi opcode // Result preparation : // DEST SRC QUO REM / DIV MOD // +33 +13 : 2 7 / 2 7 // +33 -13 : -2 7 / -3 -6 // -33 +13 : -2 -7 / -3 6 // -33 -13 : 2 -7 / 2 -7 always @(*) case (OPCODE[1:0]) 2'b00 : divi_result = {neg_flag,divsr[31:0]}; // QUO 2'b01 : divi_result = {neg_src2,divreg[33:2]}; // REM 2'b10 : divi_result = {neg_src1,((sub_case & ~rest_null) ? (save1[31:0] - divreg[33:2]) : divreg[33:2])}; // MOD 2'b11 : divi_result = {neg_flag,divsr[31:0]}; // DIV endcase always @(posedge BCLK) negativ <= divi_result[32]; assign plus_1 = (OPCODE[1:0] == 2'b11) ? (negativ & rest_null) : negativ; // Special case Rest=0 at DIV always @(posedge BCLK) if (divi_pipe4) DIVI_OUT[63:32] <= not_dei ? (({32{negativ}} ^ divi_result[31:0]) + {31'd0,plus_1}) : dei_result; always @(posedge BCLK) if (divi_pipe4) DIVI_OUT[31:0] <= divreg[33:2]; always @(extdata or BWD or divsr or divreg) casex ({extdata,BWD}) 3'b000 : dei_result = {16'hxxxx,divsr[7:0],divreg[9:2]}; 3'b001 : dei_result = {divsr[15:0],divreg[17:2]}; default : dei_result = divsr[31:0]; endcase // +++++++++++++++++++++++++++ Calculation path for Division ++++++++++++++++++++++++++++ always @(posedge BCLK or negedge BRST) if (!BRST) runflag <= 1'b0; else runflag <= START[2] | (~ende & runflag); always @(posedge BCLK) DONE <= (ende & runflag) | divi_pipe4; assign man_1 = (FL | run_divi) ? 21'h0 : MAN1; assign src_1 = run_divi ? {1'b0,ivalue} : ( FL ? {10'h001,SRC1[22:0]} : {SRC1,1'b0}); assign load_src1 = START[2] | divi_pipe1; // *2 + *1 always @(posedge BCLK) if (load_src1) dreimal <= {1'b0,man_1,src_1,1'b0} + {2'b00,man_1,src_1}; // 54 Bit Reg always @(posedge BCLK) if (load_src1) save1 <= src_1; assign sub1 = divreg - {3'b000, man_1,save1 }; assign sub2 = divreg - {2'b00 ,man_1,save1,1'b0}; assign sub3 = divreg - {1'b0, dreimal }; assign load_src2 = START[2] | divi_pipe2; always @(posedge BCLK) if (load_src2) divreg <= divi_pipe2 ? {23'h0,shift_32[64:32]} : ( FL ? {34'h0_0000_0001,SRC2[22:0]} : {3'b0,MAN2,SRC2,1'b0}); else begin casex ({sub3[56],sub2[56],sub1[56]}) 3'b0xx : divreg <= {sub3[54:0],divreg_ext[31:30]}; 3'b10x : divreg <= {sub2[54:0],divreg_ext[31:30]}; 3'b110 : divreg <= {sub1[54:0],divreg_ext[31:30]}; default : divreg <= {divreg[54:0],divreg_ext[31:30]}; endcase end always @(posedge BCLK) // Extension Register for Integer Division if (load_src2) divreg_ext <= divi_pipe2 ? shift_32[31:0] : 32'd0; else divreg_ext <= {divreg_ext[29:0],2'b0}; always @(posedge BCLK) if (load_src2) divsr <= 57'h0; else begin casex ({sub3[56],sub2[56],sub1[56]}) 3'b0xx : divsr <= {divsr[54:0],2'b11}; 3'b10x : divsr <= {divsr[54:0],2'b10}; 3'b110 : divsr <= {divsr[54:0],2'b01}; default : divsr <= {divsr[54:0],2'b00}; endcase end // Overflow Detection for DEI : serial calculation always @(posedge BCLK) if (load_src2) DEI_OVF[0] <= 1'b0; else DEI_OVF[0] <= DEI_OVF[0] | (BWD[1] ? |divsr[33:32] : (BWD[0] ? |divsr[17:16] : |divsr[9:8])); always @(posedge BCLK) DEI_OVF[1] <= divi_pipe4; // Timing pulse for OVF inclusiv for DIV and QUO assign short = (SRCFLAGS[3:0] != 4'h0) & runflag; assign ende = ((FL ? (divsr[26] | divsr[25]) : (divsr[56] | divsr[55])) & runflag) | short; assign sign = (SRCFLAGS[4] ^ SRCFLAGS[5]) & ~zero; assign zero = SRCFLAGS[2] & ~SRCFLAGS[0]; // SRC2 = NULL -> NULL as result assign nan = SRCFLAGS[3] | SRCFLAGS[1] | (SRCFLAGS[2] & SRCFLAGS[0]); // one of both NAN or both 0 -> invalid Operation assign orlow_d = (divreg[56:27] != 29'b0) & ~zero & ~FL; // is there Rest ? [1:0] are always 0. assign orlow_s = (divreg[26:2] != 25'b0) & ~zero; assign restlsb = divsr[0] | orlow_s; assign restlow = (divsr[1:0] != 2'b00) | orlow_s | orlow_d; assign resthigh = divsr[2] | restlow; always @(posedge BCLK) if (START[0]) exponent <= FL ? ({5'b00,SRC2[30:23]} - {5'b00,SRC1[30:23]}) : ({2'b00,SRC2[30:20]} - {2'b00,SRC1[30:20]}); assign offset = FL ? 13'h007E : 13'h03FE; assign expoh = exponent + {offset[12:1],1'b1}; // Double = 3FF/3FE Single = 7F/7E assign expol = exponent + offset; // in case of normalizing always @(posedge BCLK) if (ende && runflag) casex ({FL,divsr[26],divsr[56]}) 3'b11x : OUT <= {nan,zero,sign,expoh[9:8],expoh[7],expoh[7],expoh[7],expoh[7:0],divsr[25:3],28'b0,divsr[3:2],restlow}; 3'b10x : OUT <= {nan,zero,sign,expol[9:8],expol[7],expol[7],expol[7],expol[7:0],divsr[24:2],28'b0,divsr[2:1],restlsb}; 3'b0x1 : OUT <= {nan,zero,sign,expoh,divsr[55:3],resthigh}; 3'b0x0 : OUT <= {nan,zero,sign,expol,divsr[54:2],restlow}; endcase endmodule // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // 10. DP_LOGIK Control logic and result path for different functions // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ module DP_LOGIK ( BCLK, BRESET, OPCODE, SRC1, SRC2, FSR, START, MRESULT, BWD, FL, MAN1, MAN2, WR_REG, CY_IN, COP_DONE, COP_OP, COP_IN, DOUT, TT_DP, DP_CMP, OVF_BCD, MEI, DFLOAT, DONE, UP_DP, CLR_LSB, WREN_L, DVZ_TRAP, COP_GO ); // Definition of output word OUT of sub-moduls : the hidden-bit of the mantissa is already gone // // N Z S Exponent Mantissa Round // A E I Double : 13 Bit 52 Bit 2 Bit // N R G Single : 10 Bit 23 Bit 2 Bit // O N -mmmm.mmmm.mmmm.mmmm.mmmm.mmm-.-- -m. // -F-F-F-E.EEEE.EEEE.EEEE-MMMM.MMMM.MMMM.MMMM.MMMM.MMMM.MMMM.MMMM.MMMM.MMMM.MMMM.MMMM.MMMM.RR // // 6 6 6 6 6666 6655 5555 5555 4444 4444 4433 3333 3333 2222 2222 2211 1111 1111 0000 0000 00 // 9 8 7 6 5432 1098 7654 3210 9876 5432 1098 7654 3210 9876 5432 1098 7654 3210 9876 5432 10 // // Single FP delivers the exponent in a way, that it is identical for rounding : // // Exponent 61 - 54 => kept // Bits 64 - 62 are filled with bit 61 , carry should come through // Exponent 62 => Bit 65 , Overflow // Exponent 63 => Bit 66 , Underflow input BCLK,BRESET; input [7:0] OPCODE; input [31:0] SRC1,SRC2; // the input data input [20:0] MAN1,MAN2; // the MSB of mantissa input [8:3] FSR; // Floating Point Status Register input [1:0] START; input [105:0] MRESULT; // Result of multiplier input [1:0] BWD; // Size of integer input FL; input WR_REG; // from DECODER input CY_IN; input COP_DONE; // Coprozessor Interface input [23:0] COP_OP; input [63:0] COP_IN; output [63:0] DOUT; output [4:0] TT_DP; // Trap-Info to FSR output [2:0] DP_CMP; // CMPL result output [3:0] OVF_BCD; // Integer Division Overflow + BCD Carry update output MEI,DFLOAT; output DONE,UP_DP,WREN_L; output reg CLR_LSB; output DVZ_TRAP; output reg COP_GO; reg [63:0] DOUT; reg [2:0] DP_CMP; reg [5:0] preflags; reg [5:0] srcflags; reg [69:0] fpout; reg [2:0] tt; reg [7:0] select; reg [5:0] wctrl; reg [2:1] sequ; reg misc_op; reg car_ry; reg wr_part2; reg up_flag; reg ovf_div; reg late_bcd_done; wire zexp2,zman2,zexp1,zman1,znan1; wire make_i; wire op_cmp; wire [69:0] mulout,addout,divout,miscout; wire go_divf,go_divi,divi_ops,div_done; wire bcd_ops,man_ops; wire [31:0] i_out; wire [63:0] divi_out; wire [66:2] rund,cy_val; // Indexnumber like in xxxout wire div_zero,overflow,underflow,inexact; wire [1:0] cmpres; wire [63:0] fp_out,fp_res; wire wr_part1; wire done_i; wire later; wire [31:0] bcd_q; wire bcd_done; wire bcd_carry; wire [1:0] dei_ovf; wire quo_div; wire copop; wire copwr; // Control of datapath : together with START the Double Unit becomes activ always @(OPCODE or FL) casex (OPCODE) 8'b1001_000x : select = {5'b0000_0, ~FL ,2'b10}; // 0 1 0 : MOViL 8'b1001_010x : select = {5'b0001_1, 1'b1,2'b00}; // MOVLF 8'b1001_011x : select = {5'b0011_1, 1'b1,2'b00}; // MOVFL 8'b1001_100x : select = {5'b0001_0, ~FL ,2'b11}; // 0 1 1 : ROUNDLi 8'b1001_101x : select = {5'b0001_0, ~FL ,2'b11}; // 0 1 1 : TRUNCLi 8'b1001_111x : select = {5'b0001_0, ~FL ,2'b11}; // 0 1 1 : FLOORLi 8'b1011_0000 : select = {5'b1010_0, ~FL ,2'b00}; // 0 0 0 : ADDL Es werden Shifter wiederverwendet... 8'b1011_0010 : select = {5'b1010_0, ~FL ,2'b01}; // 0 0 1 : CMPL 8'b1011_0100 : select = {5'b1010_0, ~FL ,2'b01}; // 0 0 1 : SUBL 8'b1011_1000 : select = {1'b1,FL,1'b1,FL,1'b0,1'b1,2'b00}; // 1 0 1 : DIVf , Default Float fuer srcflags 8'b1011_1100 : select = {5'b1010_0, ~FL ,2'b00}; // 1 0 0 : MULL 8'b1011_0110 : select = {1'b1,FL,1'b1,FL,1'b1,1'b1,2'b00}; // SCALBf 8'b1011_0111 : select = {2'b00,FL,2'b1_1, 1'b1,2'b00}; // LOGBf default : select = 8'b0; endcase assign MEI = (OPCODE == 8'h79); assign divi_ops = (OPCODE[7:2] == 6'b0111_11) | (OPCODE == 8'h7B); // QUO/REM/MOD/DIV & DEI assign go_divf = (OPCODE == 8'hB8) & START[1]; // because of runflag in DIV Unit assign go_divi = divi_ops & (OPCODE[2] ? START[1] : START[0]); // DEI starts with START[0] assign bcd_ops = (OPCODE == 8'h71) | (OPCODE == 8'h70); // ADDP , SUBP assign man_ops = (OPCODE == 8'hB1) | (OPCODE == 8'hB5) | (OPCODE == 8'hB9) | (OPCODE == 8'hBD); // MOVf,NEGf,XXXf,ABSf assign DFLOAT = (select[2] | copop) & ~FL; // all Double Floating Point Operations for PREPDATA assign make_i = (select[1:0] == 2'b11) | divi_ops | bcd_ops; // ROUND/TRUNC/FLOOR for output multiplexer assign op_cmp = (OPCODE == 8'hB2) & ~FL; always @(posedge BCLK) misc_op <= select[3]; // for OUT-Multiplexer assign copop = (OPCODE == 8'hDD); assign copwr = (COP_OP[18:17] == 2'd0) & (COP_OP[13:11] == 3'b111) & (COP_OP[7:5] == 3'b001); // Custom Convert // SRCFLAGS : special handling for operands is done locally assign zexp2 = (SRC2[30:20] == 11'd0); assign zman2 = (SRC2[19:0] == 20'd0); assign zexp1 = (SRC1[30:20] == 11'd0); assign zman1 = (SRC1[19:0] == 20'd0); assign znan1 = (SRC1[30:20] == 11'h7FF); always @(posedge BCLK) if (START[0]) begin srcflags[5] <= SRC2[31]; srcflags[4] <= SRC1[31]; preflags <= {(SRC2[30:20] == 11'h7FF),zexp2,zman2,znan1,zexp1,zman1}; end // case Definition : 00 : 0 , if START[i]=0 then there are always 2 long operands // 01 : 1 Float Operand SCR1 // 10 : 1 Long Operand SRC1+SRC2 // 11 : 2 Float Operands SRC1 , SRC2 always @(posedge BCLK) // NaN if (START[1]) case (select[7:6]) 2'b10 : srcflags[3] <= preflags[5] | (preflags[4] & (~preflags[3] | SRC2[31] | ~zexp2 | ~zman2)); 2'b11 : srcflags[3] <= (SRC2[30:23] == 8'hFF) | ((SRC2[30:23] == 8'd0) & ((SRC2[22:20] != 3'd0) | ~zman2)); // F:SRC2 = NaN default : srcflags[3] <= 1'b0; endcase always @(posedge BCLK) // Zero : only exponent ! If denormalized => NaN ! if (START[1]) case (select[7:6]) 2'b10 : srcflags[2] <= preflags[4]; // L:SRC2 = Zero , 2*SRC2 2'b11 : srcflags[2] <= (SRC2[30:23] == 8'd0); // F:SRC2 = Zero default : srcflags[2] <= 1'b0; endcase always @(posedge BCLK) // NaN if (START[1]) case (select[5:4]) 2'b01 : srcflags[1] <= znan1 | (zexp1 & (~zman1 | SRC2[31] | ~zexp2 | ~zman2)); // L:(SRC1,SRC2) = NaN , SRC1 = MSB 2'b10 : srcflags[1] <= preflags[2] | (preflags[1] & (~preflags[0] | SRC1[31] | ~zexp1 | ~zman1)); 2'b11 : srcflags[1] <= (SRC1[30:23] == 8'hFF) | ((SRC1[30:23] == 8'd0) & ((SRC1[22:20] != 3'd0) | ~zman1)); // F:SRC1 = NaN default : srcflags[1] <= 1'b0; endcase always @(posedge BCLK) // Zero : only exponent ! If denormalized => NaN ! if (START[1]) case (select[5:4]) 2'b01 : srcflags[0] <= zexp1; // L:(SRC1,SRC2) = Zero , SRC1 = MSB, Special Case ROUNDL,etc. 2'b10 : srcflags[0] <= preflags[1]; // L:SRC1 = Zero , 2*SRC1 2'b11 : srcflags[0] <= (SRC1[30:23] == 8'd0); // F:SRC1 = Zero default : srcflags[0] <= 1'b0; endcase // The Sub-moduls : DFPU_ADDSUB as_inst ( .BCLK(BCLK), .START(START), .SRC1(SRC1), .SRC2(SRC2), .MAN1({~preflags[1],MAN1[19:0]}), .MAN2({~preflags[4],MAN2[19:0]}), .SRCFLAGS(srcflags), .BWD(BWD), .SELECT({OPCODE[2:1],select[1:0]}), .OUT(addout), .IOUT(i_out), .CMPRES(cmpres) ); DFPU_MUL mul_inst ( .BCLK(BCLK), .SRC1(SRC1), .SRC2(SRC2), .START(START[0]), .MRESULT(MRESULT), .OUT(mulout), .SRCFLAGS(srcflags) ); DFPU_DIV div_inst ( .BCLK(BCLK), .BRST(BRESET), .START({go_divi,go_divf,START}), .SRC1(SRC1), .SRC2(SRC2), .MAN1(MAN1), .MAN2(MAN2), .SRCFLAGS(srcflags), .FL(FL), .OUT(divout), .DONE(div_done), .BWD(BWD), .OPCODE(OPCODE[2:0]), .DIVI_OUT(divi_out), .DVZ_TRAP(DVZ_TRAP), .DEI_OVF(dei_ovf) ); DFPU_MISC misc_inst ( .BCLK(BCLK), .START(START), .SRC1(SRC1), .SRC2(SRC2), .MAN2(MAN2[19:0]), .SRCFLAGS(srcflags), .MODE({OPCODE[5],OPCODE[0],FL,OPCODE[1]}), .OUT(miscout) ); DFPU_BCD bcd_inst ( .BCLK(BCLK), .BRESET(BRESET), .START(START[1]), .DO_BCD(bcd_ops), .BWD(BWD), .SRC1(SRC1), .SRC2(SRC2), .CY_IN(CY_IN), .SUBP(~OPCODE[0]), .BCD_Q(bcd_q), .CY_OUT(bcd_carry), .BCD_DONE(bcd_done) ); // FP - path : selection of result and rounding : always @(misc_op or OPCODE or mulout or addout or divout or miscout) casex ({misc_op,OPCODE[5],OPCODE[3:2]}) //OPCODE[5] only for Flags i.e. NAN 4'b1xxx : fpout = miscout; // for MOVLF,MOVFL,SCALB & LOGB 4'b0110 : fpout = divout; 4'b0111 : fpout = mulout; default : fpout = addout; endcase always @(FSR or fpout) // Calculation of Carry according to rounding mode, fpout[67] = sign bit casex (FSR[8:7]) 2'b00 : car_ry = ((fpout[1:0] == 2'b10) & fpout[2]) | (fpout[1:0] == 2'b11); // round to nearest 2'b10 : car_ry = ~fpout[67] & (fpout[1:0] != 2'b00); // round to positiv infinity 2'b11 : car_ry = fpout[67] & (fpout[1:0] != 2'b00); // round to negativ infinity default : car_ry = 1'b0; // round to zero endcase assign cy_val = {35'h0,(FL & car_ry),28'h0,(~FL & car_ry)}; assign rund = {fpout[66:2]} + cy_val; // Detection of Div-by-0, Overflow, Underflow and Inexact : Epxonent from [66:54] = 13 Bits assign div_zero = (srcflags[3:0] == 4'h1) & ((OPCODE == 8'hB8) | (OPCODE == 8'hB7)); // true FPU Divide by Zero also for LOGBf assign overflow = ~rund[66] & (rund[65] | (rund[64:54] == 11'h7FF)); assign underflow = (rund[66] | (rund[65:54] == 12'h0)) & ~fpout[68]; // Zero-Flag assign inexact = (fpout[1:0] != 2'b00); always @(fpout or op_cmp or div_zero or overflow or underflow or inexact or FSR) casex ({fpout[69],op_cmp,div_zero,overflow,FSR[3],underflow,FSR[5],inexact}) // [69] = NAN 8'b1xxxxxxx : tt = 3'b101; // Invalid operation 8'b001xxxxx : tt = 3'b011; // Divide by Zero 8'b0001xxxx : tt = 3'b010; // Overflow 8'b000011xx : tt = 3'b001; // Underflow 8'b00000011 : tt = 3'b110; // Inexact Result default : tt = 3'b000; // no error endcase assign TT_DP = man_ops ? 5'd0 : {(inexact & ~op_cmp),(underflow & ~op_cmp),tt}; // at ABSf/NEGf no error : different to NS32381 ! assign fp_res = FL ? {fpout[67],rund[61:31],rund[33:2]} : {fpout[67],rund[64:2]}; // lower 32 bits identical // Underflow special case and get ZERO assign fp_out = (underflow | fpout[68]) ? 64'h0 : fp_res; // 63..32 goes to memory if Word or Byte ! Also in ODD Register , 31..0 goes in EVEN Register // DEI comes without WR_REG information always @(make_i or copop or MEI or BWD or WR_REG or MRESULT or COP_IN or i_out or fp_out or divi_ops or divi_out or bcd_ops or bcd_q) casex ({make_i,copop,MEI,BWD}) 5'b00100 : DOUT = {MRESULT[31:8], (WR_REG ? MRESULT[15:8] : MRESULT[7:0]), MRESULT[31:0]}; // LSD always the same 5'b00101 : DOUT = {MRESULT[31:16],(WR_REG ? MRESULT[31:16] : MRESULT[15:0]),MRESULT[31:0]}; 5'b0011x : DOUT = MRESULT[63:0]; 5'b01xxx : DOUT = COP_IN; // true alignment in Coprocessor 5'b1xxxx : DOUT = divi_ops ? divi_out : {(bcd_ops ? bcd_q : i_out),fp_out[31:0]}; // MSD is written first default : DOUT = fp_out; endcase always @(posedge BCLK) DP_CMP <= {(srcflags[3] | srcflags[1]),cmpres}; // Only valid if not NaN // Pipeline Control + Registerfile write control always @(posedge BCLK or negedge BRESET) if (!BRESET) sequ <= 2'b00; else sequ <= {(sequ[1] & ~DONE),(START[1] & ~wctrl[5])}; always @(FL or OPCODE or copwr) casex ({FL,OPCODE}) // WRITE Control : [2] = clr_lsb, [1] = wr_part2, [0] = wr_part1 9'b0_1001_000x : wctrl = 6'b001_111; // MOViL 9'b1_1001_000x : wctrl = 6'b100_010; // MOViF <= SFPU 9'bx_1001_010x : wctrl = 6'b000_010; // MOVLF 9'bx_1001_011x : wctrl = 6'b001_111; // MOVFL 9'b0_1001_100x : wctrl = 6'b000_010; // ROUNDLi 9'b0_1001_101x : wctrl = 6'b000_010; // TRUNCLi 9'b0_1001_111x : wctrl = 6'b000_010; // FLOORLi 9'b1_1001_100x : wctrl = 6'b100_010; // ROUNDFi 9'b1_1001_101x : wctrl = 6'b100_010; // TRUNCFi 9'b1_1001_111x : wctrl = 6'b100_010; // FLOORFi 9'b0_1011_0000 : wctrl = 6'b001_111; // ADDL 9'b1_1011_0000 : wctrl = 6'b100_010; // ADDF <= SFPU 9'b0_1011_0010 : wctrl = 6'b000_000; // CMPL 9'b1_1011_0010 : wctrl = 6'b100_000; // CMPF 9'b0_1011_0100 : wctrl = 6'b001_111; // SUBL 9'b1_1011_0100 : wctrl = 6'b100_010; // SUBF <= SFPU 9'b1_1011_1000 : wctrl = 6'b010_001; // DIVF - measured 18 clocks Reg-Reg 9'b0_1011_1000 : wctrl = 6'b010_111; // DIVL - measured 34 clocks Reg-Reg 9'b0_1011_1100 : wctrl = 6'b001_111; // MULL 9'b1_1011_1100 : wctrl = 6'b100_010; // MULF <= SFPU 9'bx_0111_000x : wctrl = 6'b100_010; // ADDP,SUBP 9'bx_0111_1001 : wctrl = 6'b000_111; // MEIi 9'bx_0111_1011 : wctrl = 6'b010_111; // DEIi 9'bx_0111_11xx : wctrl = 6'b010_001; // QUOi,REMi,MODi,DIVi 9'b1_1011_011x : wctrl = 6'b000_010; // SCALBF/LOGBF 9'b0_1011_011x : wctrl = 6'b001_111; // SCALBL/LOGBL 9'bx_1101_1101 : wctrl = {5'b010_00,copwr}; // Coprocessor opcode default : wctrl = 6'b00; endcase assign later = wctrl[3] & WR_REG; // if DEST = Reg and 64 bit of data then DONE comes 1 clock later assign done_i = wctrl[4] ? (div_done | COP_DONE) : ( later ? sequ[2] : sequ[1] ); assign DONE = wctrl[5] ? (bcd_ops ? bcd_done : START[1]) : ~START[1] & done_i; // DONE is valid for all opcodes assign wr_part1 = DONE & WR_REG & wctrl[0]; always @(posedge BCLK) CLR_LSB <= DONE & WR_REG & wctrl[2]; always @(posedge BCLK) wr_part2 <= DONE & WR_REG & wctrl[1]; assign WREN_L = wr_part1 | wr_part2; always @(posedge BCLK) up_flag <= DONE & ~later; // DONE one cycle later assign UP_DP = (select[2] & (later ? DONE : up_flag)) | man_ops; // Update FSR Trap etc. : all FPU opcodes of DP_FPU // Overflow Trap for Division : DEI, QUO, DIV assign quo_div = (OPCODE == 8'h7C) | (OPCODE == 8'h7F); always @(*) casex ({OPCODE[2],BWD}) 3'b100 : ovf_div = (divi_out[39] & SRC1[7] & SRC2[7] ) & quo_div; 3'b101 : ovf_div = (divi_out[47] & SRC1[15] & SRC2[15]) & quo_div; 3'b11x : ovf_div = (divi_out[63] & SRC1[31] & SRC2[31]) & quo_div; default : ovf_div = dei_ovf[0] & (OPCODE == 8'h7B); // DEI endcase always @(posedge BCLK) late_bcd_done <= bcd_done; // parallel to data write assign OVF_BCD = {dei_ovf[1],ovf_div,late_bcd_done,bcd_carry}; // to I_PFAD always @(posedge BCLK) COP_GO <= START[1] & copop; endmodule // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ // // 11. DP_FPU Top level of long operations datapath // // ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ module DP_FPU( BCLK, FL, BRESET, LD_OUT, WR_REG, BWD, FSR, OPCODE, SRC1, SRC2, START, CY_IN, COP_DONE, COP_OP, COP_IN, DONE, UP_DP, WREN_L, CLR_LSB, DVZ_TRAP, DP_CMP, DP_OUT, DP_Q, TT_DP, OVF_BCD, COP_GO, COP_OUT ); input BCLK; input FL; input BRESET; input [1:0] LD_OUT; input WR_REG; input [1:0] BWD; input [8:3] FSR; input [7:0] OPCODE; input [31:0] SRC1; input [31:0] SRC2; input [1:0] START; input CY_IN; input COP_DONE; input [23:0] COP_OP; input [63:0] COP_IN; output DONE; output UP_DP; output WREN_L; output CLR_LSB; output DVZ_TRAP; output [2:0] DP_CMP; output [31:0] DP_OUT; output [31:0] DP_Q; output [4:0] TT_DP; output [3:0] OVF_BCD; output COP_GO; output [127:0] COP_OUT; reg [52:0] MDA; reg [52:0] MDB; reg [31:0] DP_Q; reg [31:20] RCOPA,RCOPB; wire [63:0] DOUT; wire [105:0] MRESULT; wire MEI; wire DFLOAT; wire LOAD_MSD; wire LOAD_LSD1; wire LOAD_LSD2; wire [31:0] LSD_1; wire [31:0] LSD_2; wire [52:32] MSD_1; wire [52:32] MSD_2; DP_LOGIK DOUBLE_U( .FL(FL), .BRESET(BRESET), .BCLK(BCLK), .WR_REG(WR_REG), .BWD(BWD), .FSR(FSR), .MAN1(MDA[52:32]), .MAN2(MDB[52:32]), .MRESULT(MRESULT), .OPCODE(OPCODE), .SRC1(SRC1), .SRC2(SRC2), .START(START), .MEI(MEI), .DFLOAT(DFLOAT), .DONE(DONE), .UP_DP(UP_DP), .CLR_LSB(CLR_LSB), .WREN_L(WREN_L), .DVZ_TRAP(DVZ_TRAP), .DOUT(DOUT), .DP_CMP(DP_CMP), .TT_DP(TT_DP), .CY_IN(CY_IN), .OVF_BCD(OVF_BCD), .COP_DONE(COP_DONE), .COP_OP(COP_OP), .COP_IN(COP_IN), .COP_GO(COP_GO)); PREPDATA DP_PREP( .MEI(MEI), .DFLOAT(DFLOAT), .BWD(BWD), .SRC1(SRC1), .SRC2(SRC2), .START(START), .LOAD_LSD1(LOAD_LSD1), .LOAD_LSD2(LOAD_LSD2), .LOAD_MSD(LOAD_MSD), .LSD_1(LSD_1), .LSD_2(LSD_2), .MSD_1(MSD_1), .MSD_2(MSD_2)); assign MRESULT = MDA * MDB; // unsigned multiplier 53 * 53 bits = 106 bits assign DP_OUT = CLR_LSB ? DP_Q : DOUT[63:32]; always@(posedge BCLK) if (LD_OUT[1] || LD_OUT[0] || WREN_L) DP_Q <= LD_OUT[0] ? SRC2 : DOUT[31:0]; always@(posedge BCLK) if (LOAD_LSD1) MDA[31:0] <= LSD_1; always@(posedge BCLK) if (LOAD_LSD2) MDB[31:0] <= LSD_2; always@(posedge BCLK) if (LOAD_MSD) begin MDA[52:32] <= MSD_1; MDB[52:32] <= MSD_2; RCOPA <= SRC1[31:20]; RCOPB <= SRC2[31:20]; end assign COP_OUT = {RCOPA,MDA[51:32],SRC1,RCOPB,MDB[51:32],SRC2}; endmodule
Go to most recent revision | Compare with Previous | Blame | View Log