URL
https://opencores.org/ocsvn/dblclockfft/dblclockfft/trunk
Subversion Repositories dblclockfft
[/] [dblclockfft/] [trunk/] [rtl/] [butterfly.v] - Rev 40
Go to most recent revision | Compare with Previous | Blame | View Log
//////////////////////////////////////////////////////////////////////////////// // // Filename: butterfly.v // // Project: A General Purpose Pipelined FFT Implementation // // Purpose: This routine caculates a butterfly for a decimation // in frequency version of an FFT. Specifically, given // complex Left and Right values together with a coefficient, the output // of this routine is given by: // // L' = L + R // R' = (L - R)*C // // The rest of the junk below handles timing (mostly), to make certain // that L' and R' reach the output at the same clock. Further, just to // make certain that is the case, an 'aux' input exists. This aux value // will come out of this routine synchronized to the values it came in // with. (i.e., both L', R', and aux all have the same delay.) Hence, // a caller of this routine may set aux on the first input with valid // data, and then wait to see aux set on the output to know when to find // the first output with valid data. // // All bits are preserved until the very last clock, where any more bits // than OWIDTH will be quietly discarded. // // This design features no overflow checking. // // Notes: // CORDIC: // Much as we might like, we can't use a cordic here. // The goal is to accomplish an FFT, as defined, and a // CORDIC places a scale factor onto the data. Removing // the scale factor would cost two multiplies, which // is precisely what we are trying to avoid. // // // 3-MULTIPLIES: // It should also be possible to do this with three multiplies // and an extra two addition cycles. // // We want // R+I = (a + jb) * (c + jd) // R+I = (ac-bd) + j(ad+bc) // We multiply // P1 = ac // P2 = bd // P3 = (a+b)(c+d) // Then // R+I=(P1-P2)+j(P3-P2-P1) // // WIDTHS: // On multiplying an X width number by an // Y width number, X>Y, the result should be (X+Y) // bits, right? // -2^(X-1) <= a <= 2^(X-1) - 1 // -2^(Y-1) <= b <= 2^(Y-1) - 1 // (2^(Y-1)-1)*(-2^(X-1)) <= ab <= 2^(X-1)2^(Y-1) // -2^(X+Y-2)+2^(X-1) <= ab <= 2^(X+Y-2) <= 2^(X+Y-1) - 1 // -2^(X+Y-1) <= ab <= 2^(X+Y-1)-1 // YUP! But just barely. Do this and you'll really want // to drop a bit, although you will risk overflow in so // doing. // // 20150602 -- The sync logic lines have been completely redone. The // synchronization lines no longer go through the FIFO with the // left hand sum, but are kept out of memory. This allows the // butterfly to use more optimal memory resources, while also // guaranteeing that the sync lines can be properly reset upon // any reset signal. // // // Creator: Dan Gisselquist, Ph.D. // Gisselquist Technology, LLC // //////////////////////////////////////////////////////////////////////////////// // // Copyright (C) 2015-2018, Gisselquist Technology, LLC // // This file is part of the general purpose pipelined FFT project. // // The pipelined FFT project is free software (firmware): 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 3 of the // License, or (at your option) any later version. // // The pipelined FFT project is distributed in the hope that it will be useful, // but WITHOUT ANY WARRANTY; without even the implied warranty of // MERCHANTIBILITY 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 program. (It's in the $(ROOT)/doc directory. Run make // with no target there if the PDF file isn't present.) If not, see // <http://www.gnu.org/licenses/> for a copy. // // License: LGPL, v3, as defined and found on www.gnu.org, // http://www.gnu.org/licenses/lgpl.html // // //////////////////////////////////////////////////////////////////////////////// // // `default_nettype none // module butterfly(i_clk, i_reset, i_ce, i_coef, i_left, i_right, i_aux, o_left, o_right, o_aux); // Public changeable parameters ... parameter IWIDTH=16,CWIDTH=20,OWIDTH=17; parameter SHIFT=0; // The number of clocks per each i_ce. The actual number can be // more, but the algorithm depends upon at least this many for // extra internal processing. parameter CKPCE=1; // // Local/derived parameters that are calculated from the above // params. Apart from algorithmic changes below, these should not // be adjusted // // The first step is to calculate how many clocks it takes our // multiply to come back with an answer within. The time in the // multiply depends upon the input value with the fewest number of // bits--to keep the pipeline depth short. So, let's find the // fewest number of bits here. localparam MXMPYBITS = ((IWIDTH+2)>(CWIDTH+1)) ? (CWIDTH+1) : (IWIDTH + 2); // // Given this "fewest" number of bits, we can calculate the // number of clocks the multiply itself will take. localparam MPYDELAY=((MXMPYBITS+1)/2)+2; // // In an environment when CKPCE > 1, the multiply delay isn't // necessarily the delay felt by this algorithm--measured in // i_ce's. In particular, if the multiply can operate with more // operations per clock, it can appear to finish "faster". // Since most of the logic in this core operates on the slower // clock, we'll need to map that speed into the number of slower // clock ticks that it takes. localparam LCLDELAY = (CKPCE == 1) ? MPYDELAY : (CKPCE == 2) ? (MPYDELAY/2+2) : (MPYDELAY/3 + 2); localparam LGDELAY = (MPYDELAY>64) ? 7 : (MPYDELAY > 32) ? 6 : (MPYDELAY > 16) ? 5 : (MPYDELAY > 8) ? 4 : (MPYDELAY > 4) ? 3 : 2; localparam AUXLEN=(LCLDELAY+3); localparam MPYREMAINDER = MPYDELAY - CKPCE*(MPYDELAY/CKPCE); input wire i_clk, i_reset, i_ce; input wire [(2*CWIDTH-1):0] i_coef; input wire [(2*IWIDTH-1):0] i_left, i_right; input wire i_aux; output wire [(2*OWIDTH-1):0] o_left, o_right; output reg o_aux; `ifdef FORMAL localparam F_LGDEPTH = (AUXLEN > 64) ? 7 : (AUXLEN > 32) ? 6 : (AUXLEN > 16) ? 5 : (AUXLEN > 8) ? 4 : (AUXLEN > 4) ? 3 : 2; localparam F_DEPTH = AUXLEN; localparam [F_LGDEPTH-1:0] F_D = F_DEPTH[F_LGDEPTH-1:0]-1; reg signed [IWIDTH-1:0] f_dlyleft_r [0:F_DEPTH-1]; reg signed [IWIDTH-1:0] f_dlyleft_i [0:F_DEPTH-1]; reg signed [IWIDTH-1:0] f_dlyright_r [0:F_DEPTH-1]; reg signed [IWIDTH-1:0] f_dlyright_i [0:F_DEPTH-1]; reg signed [CWIDTH-1:0] f_dlycoeff_r [0:F_DEPTH-1]; reg signed [CWIDTH-1:0] f_dlycoeff_i [0:F_DEPTH-1]; reg signed [F_DEPTH-1:0] f_dlyaux; wire signed [IWIDTH:0] f_predifr, f_predifi; wire signed [IWIDTH+CWIDTH+3-1:0] f_predifrx, f_predifix; wire signed [CWIDTH:0] f_sumcoef; wire signed [IWIDTH+1:0] f_sumdiff; wire signed [IWIDTH:0] f_sumr, f_sumi; wire signed [IWIDTH+CWIDTH+3-1:0] f_sumrx, f_sumix; wire signed [IWIDTH:0] f_difr, f_difi; wire signed [IWIDTH+CWIDTH+3-1:0] f_difrx, f_difix; wire signed [IWIDTH+CWIDTH+3-1:0] f_widecoeff_r, f_widecoeff_i; wire [(CWIDTH):0] fp_one_ic, fp_two_ic, fp_three_ic, f_p3c_in; wire [(IWIDTH+1):0] fp_one_id, fp_two_id, fp_three_id, f_p3d_in; `endif reg [(2*IWIDTH-1):0] r_left, r_right; reg [(2*CWIDTH-1):0] r_coef, r_coef_2; wire signed [(IWIDTH-1):0] r_left_r, r_left_i, r_right_r, r_right_i; assign r_left_r = r_left[ (2*IWIDTH-1):(IWIDTH)]; assign r_left_i = r_left[ (IWIDTH-1):0]; assign r_right_r = r_right[(2*IWIDTH-1):(IWIDTH)]; assign r_right_i = r_right[(IWIDTH-1):0]; reg signed [(IWIDTH):0] r_sum_r, r_sum_i, r_dif_r, r_dif_i; reg [(LGDELAY-1):0] fifo_addr; wire [(LGDELAY-1):0] fifo_read_addr; assign fifo_read_addr = fifo_addr - LCLDELAY[(LGDELAY-1):0]; reg [(2*IWIDTH+1):0] fifo_left [ 0:((1<<LGDELAY)-1)]; // Set up the input to the multiply always @(posedge i_clk) if (i_ce) begin // One clock just latches the inputs r_left <= i_left; // No change in # of bits r_right <= i_right; r_coef <= i_coef; // Next clock adds/subtracts r_sum_r <= r_left_r + r_right_r; // Now IWIDTH+1 bits r_sum_i <= r_left_i + r_right_i; r_dif_r <= r_left_r - r_right_r; r_dif_i <= r_left_i - r_right_i; // Other inputs are simply delayed on second clock r_coef_2<= r_coef; end // Don't forget to record the even side, since it doesn't need // to be multiplied, but yet we still need the results in sync // with the answer when it is ready. initial fifo_addr = 0; always @(posedge i_clk) if (i_reset) fifo_addr <= 0; else if (i_ce) // Need to delay the sum side--nothing else happens // to it, but it needs to stay synchronized with the // right side. fifo_addr <= fifo_addr + 1; always @(posedge i_clk) if (i_ce) fifo_left[fifo_addr] <= { r_sum_r, r_sum_i }; wire signed [(CWIDTH-1):0] ir_coef_r, ir_coef_i; assign ir_coef_r = r_coef_2[(2*CWIDTH-1):CWIDTH]; assign ir_coef_i = r_coef_2[(CWIDTH-1):0]; wire signed [((IWIDTH+2)+(CWIDTH+1)-1):0] p_one, p_two, p_three; // Multiply output is always a width of the sum of the widths of // the two inputs. ALWAYS. This is independent of the number of // bits in p_one, p_two, or p_three. These values needed to // accumulate a bit (or two) each. However, this approach to a // three multiply complex multiply cannot increase the total // number of bits in our final output. We'll take care of // dropping back down to the proper width, OWIDTH, in our routine // below. // We accomplish here "Karatsuba" multiplication. That is, // by doing three multiplies we accomplish the work of four. // Let's prove to ourselves that this works ... We wish to // multiply: (a+jb) * (c+jd), where a+jb is given by // a + jb = r_dif_r + j r_dif_i, and // c + jd = ir_coef_r + j ir_coef_i. // We do this by calculating the intermediate products P1, P2, // and P3 as // P1 = ac // P2 = bd // P3 = (a + b) * (c + d) // and then complete our final answer with // ac - bd = P1 - P2 (this checks) // ad + bc = P3 - P2 - P1 // = (ac + bc + ad + bd) - bd - ac // = bc + ad (this checks) // This should really be based upon an IF, such as in // if (IWIDTH < CWIDTH) then ... // However, this is the only (other) way I know to do it. generate if (CKPCE <= 1) begin wire [(CWIDTH):0] p3c_in; wire [(IWIDTH+1):0] p3d_in; assign p3c_in = ir_coef_i + ir_coef_r; assign p3d_in = r_dif_r + r_dif_i; // We need to pad these first two multiplies by an extra // bit just to keep them aligned with the third, // simpler, multiply. longbimpy #(CWIDTH+1,IWIDTH+2) p1(i_clk, i_ce, {ir_coef_r[CWIDTH-1],ir_coef_r}, {r_dif_r[IWIDTH],r_dif_r}, p_one `ifdef FORMAL , fp_one_ic, fp_one_id `endif ); longbimpy #(CWIDTH+1,IWIDTH+2) p2(i_clk, i_ce, {ir_coef_i[CWIDTH-1],ir_coef_i}, {r_dif_i[IWIDTH],r_dif_i}, p_two `ifdef FORMAL , fp_two_ic, fp_two_id `endif ); longbimpy #(CWIDTH+1,IWIDTH+2) p3(i_clk, i_ce, p3c_in, p3d_in, p_three `ifdef FORMAL , fp_three_ic, fp_three_id `endif ); end else if (CKPCE == 2) begin : CKPCE_TWO // Coefficient multiply inputs reg [2*(CWIDTH)-1:0] mpy_pipe_c; // Data multiply inputs reg [2*(IWIDTH+1)-1:0] mpy_pipe_d; wire signed [(CWIDTH-1):0] mpy_pipe_vc; wire signed [(IWIDTH):0] mpy_pipe_vd; // reg signed [(CWIDTH+1)-1:0] mpy_cof_sum; reg signed [(IWIDTH+2)-1:0] mpy_dif_sum; assign mpy_pipe_vc = mpy_pipe_c[2*(CWIDTH)-1:CWIDTH]; assign mpy_pipe_vd = mpy_pipe_d[2*(IWIDTH+1)-1:IWIDTH+1]; reg mpy_pipe_v; reg ce_phase; reg signed [(CWIDTH+IWIDTH+3)-1:0] mpy_pipe_out; reg signed [IWIDTH+CWIDTH+3-1:0] longmpy; `ifdef FORMAL wire [CWIDTH:0] f_past_ic; wire [IWIDTH+1:0] f_past_id; wire [CWIDTH:0] f_past_mux_ic; wire [IWIDTH+1:0] f_past_mux_id; reg [CWIDTH:0] f_rpone_ic, f_rptwo_ic, f_rpthree_ic, f_rp2one_ic, f_rp2two_ic, f_rp2three_ic; reg [IWIDTH+1:0] f_rpone_id, f_rptwo_id, f_rpthree_id, f_rp2one_id, f_rp2two_id, f_rp2three_id; `endif initial ce_phase = 1'b0; always @(posedge i_clk) if (i_reset) ce_phase <= 1'b0; else if (i_ce) ce_phase <= 1'b1; else ce_phase <= 1'b0; always @(*) mpy_pipe_v = (i_ce)||(ce_phase); always @(posedge i_clk) if (ce_phase) begin mpy_pipe_c[2*CWIDTH-1:0] <= { ir_coef_r, ir_coef_i }; mpy_pipe_d[2*(IWIDTH+1)-1:0] <= { r_dif_r, r_dif_i }; mpy_cof_sum <= ir_coef_i + ir_coef_r; mpy_dif_sum <= r_dif_r + r_dif_i; end else if (i_ce) begin mpy_pipe_c[2*(CWIDTH)-1:0] <= { mpy_pipe_c[(CWIDTH)-1:0], {(CWIDTH){1'b0}} }; mpy_pipe_d[2*(IWIDTH+1)-1:0] <= { mpy_pipe_d[(IWIDTH+1)-1:0], {(IWIDTH+1){1'b0}} }; end longbimpy #(CWIDTH+1,IWIDTH+2) mpy0(i_clk, mpy_pipe_v, mpy_cof_sum, mpy_dif_sum, longmpy `ifdef FORMAL , f_past_ic, f_past_id `endif ); longbimpy #(CWIDTH+1,IWIDTH+2) mpy1(i_clk, mpy_pipe_v, { mpy_pipe_vc[CWIDTH-1], mpy_pipe_vc }, { mpy_pipe_vd[IWIDTH ], mpy_pipe_vd }, mpy_pipe_out `ifdef FORMAL , f_past_mux_ic, f_past_mux_id `endif ); reg signed [((IWIDTH+2)+(CWIDTH+1)-1):0] rp_one, rp_two, rp_three, rp2_one, rp2_two, rp2_three; always @(posedge i_clk) if (((i_ce)&&(!MPYDELAY[0])) ||((ce_phase)&&(MPYDELAY[0]))) begin rp_one <= mpy_pipe_out; `ifdef FORMAL f_rpone_ic <= f_past_mux_ic; f_rpone_id <= f_past_mux_id; `endif end always @(posedge i_clk) if (((i_ce)&&(MPYDELAY[0])) ||((ce_phase)&&(!MPYDELAY[0]))) begin rp_two <= mpy_pipe_out; `ifdef FORMAL f_rptwo_ic <= f_past_mux_ic; f_rptwo_id <= f_past_mux_id; `endif end always @(posedge i_clk) if (i_ce) begin rp_three <= longmpy; `ifdef FORMAL f_rpthree_ic <= f_past_ic; f_rpthree_id <= f_past_id; `endif end // Our outputs *MUST* be set on a clock where i_ce is // true for the following logic to work. Make that // happen here. always @(posedge i_clk) if (i_ce) begin rp2_one<= rp_one; rp2_two <= rp_two; rp2_three<= rp_three; `ifdef FORMAL f_rp2one_ic <= f_rpone_ic; f_rp2one_id <= f_rpone_id; f_rp2two_ic <= f_rptwo_ic; f_rp2two_id <= f_rptwo_id; f_rp2three_ic <= f_rpthree_ic; f_rp2three_id <= f_rpthree_id; `endif end assign p_one = rp2_one; assign p_two = (!MPYDELAY[0])? rp2_two : rp_two; assign p_three = ( MPYDELAY[0])? rp_three : rp2_three; // verilator lint_off UNUSED wire [2*(IWIDTH+CWIDTH+3)-1:0] unused; assign unused = { rp2_two, rp2_three }; // verilator lint_on UNUSED `ifdef FORMAL assign fp_one_ic = f_rp2one_ic; assign fp_one_id = f_rp2one_id; assign fp_two_ic = (!MPYDELAY[0])? f_rp2two_ic : f_rptwo_ic; assign fp_two_id = (!MPYDELAY[0])? f_rp2two_id : f_rptwo_id; assign fp_three_ic= (MPYDELAY[0])? f_rpthree_ic : f_rp2three_ic; assign fp_three_id= (MPYDELAY[0])? f_rpthree_id : f_rp2three_id; `endif end else if (CKPCE <= 3) begin : CKPCE_THREE // Coefficient multiply inputs reg [3*(CWIDTH+1)-1:0] mpy_pipe_c; // Data multiply inputs reg [3*(IWIDTH+2)-1:0] mpy_pipe_d; wire signed [(CWIDTH):0] mpy_pipe_vc; wire signed [(IWIDTH+1):0] mpy_pipe_vd; assign mpy_pipe_vc = mpy_pipe_c[3*(CWIDTH+1)-1:2*(CWIDTH+1)]; assign mpy_pipe_vd = mpy_pipe_d[3*(IWIDTH+2)-1:2*(IWIDTH+2)]; reg mpy_pipe_v; reg [2:0] ce_phase; reg signed [ (CWIDTH+IWIDTH+3)-1:0] mpy_pipe_out; `ifdef FORMAL wire [CWIDTH:0] f_past_ic; wire [IWIDTH+1:0] f_past_id; reg [CWIDTH:0] f_rpone_ic, f_rptwo_ic, f_rpthree_ic, f_rp2one_ic, f_rp2two_ic, f_rp2three_ic, f_rp3one_ic; reg [IWIDTH+1:0] f_rpone_id, f_rptwo_id, f_rpthree_id, f_rp2one_id, f_rp2two_id, f_rp2three_id, f_rp3one_id; `endif initial ce_phase = 3'b011; always @(posedge i_clk) if (i_reset) ce_phase <= 3'b011; else if (i_ce) ce_phase <= 3'b000; else if (ce_phase != 3'b011) ce_phase <= ce_phase + 1'b1; always @(*) mpy_pipe_v = (i_ce)||(ce_phase < 3'b010); always @(posedge i_clk) if (ce_phase == 3'b000) begin // Second clock mpy_pipe_c[3*(CWIDTH+1)-1:(CWIDTH+1)] <= { ir_coef_r[CWIDTH-1], ir_coef_r, ir_coef_i[CWIDTH-1], ir_coef_i }; mpy_pipe_c[CWIDTH:0] <= ir_coef_i + ir_coef_r; mpy_pipe_d[3*(IWIDTH+2)-1:(IWIDTH+2)] <= { r_dif_r[IWIDTH], r_dif_r, r_dif_i[IWIDTH], r_dif_i }; mpy_pipe_d[(IWIDTH+2)-1:0] <= r_dif_r + r_dif_i; end else if (mpy_pipe_v) begin mpy_pipe_c[3*(CWIDTH+1)-1:0] <= { mpy_pipe_c[2*(CWIDTH+1)-1:0], {(CWIDTH+1){1'b0}} }; mpy_pipe_d[3*(IWIDTH+2)-1:0] <= { mpy_pipe_d[2*(IWIDTH+2)-1:0], {(IWIDTH+2){1'b0}} }; end longbimpy #(CWIDTH+1,IWIDTH+2) mpy(i_clk, mpy_pipe_v, mpy_pipe_vc, mpy_pipe_vd, mpy_pipe_out `ifdef FORMAL , f_past_ic, f_past_id `endif ); reg signed [((IWIDTH+2)+(CWIDTH+1)-1):0] rp_one, rp_two, rp_three, rp2_one, rp2_two, rp2_three, rp3_one; always @(posedge i_clk) if (MPYREMAINDER == 0) begin if (i_ce) begin rp_two <= mpy_pipe_out; `ifdef FORMAL f_rptwo_ic <= f_past_ic; f_rptwo_id <= f_past_id; `endif end else if (ce_phase == 3'b000) begin rp_three <= mpy_pipe_out; `ifdef FORMAL f_rpthree_ic <= f_past_ic; f_rpthree_id <= f_past_id; `endif end else if (ce_phase == 3'b001) begin rp_one <= mpy_pipe_out; `ifdef FORMAL f_rpone_ic <= f_past_ic; f_rpone_id <= f_past_id; `endif end end else if (MPYREMAINDER == 1) begin if (i_ce) begin rp_one <= mpy_pipe_out; `ifdef FORMAL f_rpone_ic <= f_past_ic; f_rpone_id <= f_past_id; `endif end else if (ce_phase == 3'b000) begin rp_two <= mpy_pipe_out; `ifdef FORMAL f_rptwo_ic <= f_past_ic; f_rptwo_id <= f_past_id; `endif end else if (ce_phase == 3'b001) begin rp_three <= mpy_pipe_out; `ifdef FORMAL f_rpthree_ic <= f_past_ic; f_rpthree_id <= f_past_id; `endif end end else // if (MPYREMAINDER == 2) begin if (i_ce) begin rp_three <= mpy_pipe_out; `ifdef FORMAL f_rpthree_ic <= f_past_ic; f_rpthree_id <= f_past_id; `endif end else if (ce_phase == 3'b000) begin rp_one <= mpy_pipe_out; `ifdef FORMAL f_rpone_ic <= f_past_ic; f_rpone_id <= f_past_id; `endif end else if (ce_phase == 3'b001) begin rp_two <= mpy_pipe_out; `ifdef FORMAL f_rptwo_ic <= f_past_ic; f_rptwo_id <= f_past_id; `endif end end always @(posedge i_clk) if (i_ce) begin rp2_one <= rp_one; rp2_two <= rp_two; rp2_three <= (MPYREMAINDER == 2) ? mpy_pipe_out : rp_three; rp3_one <= (MPYREMAINDER == 0) ? rp2_one : rp_one; `ifdef FORMAL f_rp2one_ic <= f_rpone_ic; f_rp2one_id <= f_rpone_id; f_rp2two_ic <= f_rptwo_ic; f_rp2two_id <= f_rptwo_id; f_rp2three_ic <= (MPYREMAINDER==2) ? f_past_ic : f_rpthree_ic; f_rp2three_id <= (MPYREMAINDER==2) ? f_past_id : f_rpthree_id; f_rp3one_ic <= (MPYREMAINDER==0) ? f_rp2one_ic : f_rpone_ic; f_rp3one_id <= (MPYREMAINDER==0) ? f_rp2one_id : f_rpone_id; `endif end assign p_one = rp3_one; assign p_two = rp2_two; assign p_three = rp2_three; `ifdef FORMAL assign fp_one_ic = f_rp3one_ic; assign fp_one_id = f_rp3one_id; assign fp_two_ic = f_rp2two_ic; assign fp_two_id = f_rp2two_id; assign fp_three_ic = f_rp2three_ic; assign fp_three_id = f_rp2three_id; `endif end endgenerate // These values are held in memory and delayed during the // multiply. Here, we recover them. During the multiply, // values were multiplied by 2^(CWIDTH-2)*exp{-j*2*pi*...}, // therefore, the left_x values need to be right shifted by // CWIDTH-2 as well. The additional bits come from a sign // extension. wire signed [(IWIDTH+CWIDTH):0] fifo_i, fifo_r; reg [(2*IWIDTH+1):0] fifo_read; assign fifo_r = { {2{fifo_read[2*(IWIDTH+1)-1]}}, fifo_read[(2*(IWIDTH+1)-1):(IWIDTH+1)], {(CWIDTH-2){1'b0}} }; assign fifo_i = { {2{fifo_read[(IWIDTH+1)-1]}}, fifo_read[((IWIDTH+1)-1):0], {(CWIDTH-2){1'b0}} }; reg signed [(CWIDTH+IWIDTH+3-1):0] mpy_r, mpy_i; // Let's do some rounding and remove unnecessary bits. // We have (IWIDTH+CWIDTH+3) bits here, we need to drop down to // OWIDTH, and SHIFT by SHIFT bits in the process. The trick is // that we don't need (IWIDTH+CWIDTH+3) bits. We've accumulated // them, but the actual values will never fill all these bits. // In particular, we only need: // IWIDTH bits for the input // +1 bit for the add/subtract // +CWIDTH bits for the coefficient multiply // +1 bit for the add/subtract in the complex multiply // ------ // (IWIDTH+CWIDTH+2) bits at full precision. // // However, the coefficient multiply multiplied by a maximum value // of 2^(CWIDTH-2). Thus, we only have // IWIDTH bits for the input // +1 bit for the add/subtract // +CWIDTH-2 bits for the coefficient multiply // +1 (optional) bit for the add/subtract in the cpx mpy. // -------- ... multiply. (This last bit may be shifted out.) // (IWIDTH+CWIDTH) valid output bits. // Now, if the user wants to keep any extras of these (via OWIDTH), // or if he wishes to arbitrarily shift some of these off (via // SHIFT) we accomplish that here. wire signed [(OWIDTH-1):0] rnd_left_r, rnd_left_i, rnd_right_r, rnd_right_i; wire signed [(CWIDTH+IWIDTH+3-1):0] left_sr, left_si; assign left_sr = { {(2){fifo_r[(IWIDTH+CWIDTH)]}}, fifo_r }; assign left_si = { {(2){fifo_i[(IWIDTH+CWIDTH)]}}, fifo_i }; convround #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_left_r(i_clk, i_ce, left_sr, rnd_left_r); convround #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_left_i(i_clk, i_ce, left_si, rnd_left_i); convround #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_right_r(i_clk, i_ce, mpy_r, rnd_right_r); convround #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_right_i(i_clk, i_ce, mpy_i, rnd_right_i); always @(posedge i_clk) if (i_ce) begin // First clock, recover all values fifo_read <= fifo_left[fifo_read_addr]; // These values are IWIDTH+CWIDTH+3 bits wide // although they only need to be (IWIDTH+1) // + (CWIDTH) bits wide. (We've got two // extra bits we need to get rid of.) mpy_r <= p_one - p_two; mpy_i <= p_three - p_one - p_two; end reg [(AUXLEN-1):0] aux_pipeline; initial aux_pipeline = 0; always @(posedge i_clk) if (i_reset) aux_pipeline <= 0; else if (i_ce) aux_pipeline <= { aux_pipeline[(AUXLEN-2):0], i_aux }; initial o_aux = 1'b0; always @(posedge i_clk) if (i_reset) o_aux <= 1'b0; else if (i_ce) begin // Second clock, latch for final clock o_aux <= aux_pipeline[AUXLEN-1]; end // As a final step, we pack our outputs into two packed two's // complement numbers per output word, so that each output word // has (2*OWIDTH) bits in it, with the top half being the real // portion and the bottom half being the imaginary portion. assign o_left = { rnd_left_r, rnd_left_i }; assign o_right= { rnd_right_r,rnd_right_i}; `ifdef FORMAL initial f_dlyaux[0] = 0; always @(posedge i_clk) if (i_reset) f_dlyaux <= 0; else if (i_ce) f_dlyaux <= { f_dlyaux[F_DEPTH-2:0], i_aux }; always @(posedge i_clk) if (i_ce) begin f_dlyleft_r[0] <= i_left[ (2*IWIDTH-1):IWIDTH]; f_dlyleft_i[0] <= i_left[ ( IWIDTH-1):0]; f_dlyright_r[0] <= i_right[(2*IWIDTH-1):IWIDTH]; f_dlyright_i[0] <= i_right[( IWIDTH-1):0]; f_dlycoeff_r[0] <= i_coef[ (2*CWIDTH-1):CWIDTH]; f_dlycoeff_i[0] <= i_coef[ ( CWIDTH-1):0]; end genvar k; generate for(k=1; k<F_DEPTH; k=k+1) begin : F_PROPAGATE_DELAY_LINES always @(posedge i_clk) if (i_ce) begin f_dlyleft_r[k] <= f_dlyleft_r[ k-1]; f_dlyleft_i[k] <= f_dlyleft_i[ k-1]; f_dlyright_r[k] <= f_dlyright_r[k-1]; f_dlyright_i[k] <= f_dlyright_i[k-1]; f_dlycoeff_r[k] <= f_dlycoeff_r[k-1]; f_dlycoeff_i[k] <= f_dlycoeff_i[k-1]; end end endgenerate `ifndef VERILATOR // // Make some i_ce restraining assumptions. These are necessary // to get the design to pass induction. // generate if (CKPCE <= 1) begin // No primary i_ce assumption. i_ce can be anything // // First induction i_ce assumption: No more than one // empty cycle between used cycles. Without this // assumption, or one like it, induction would never // complete. always @(posedge i_clk) if ((!$past(i_ce))) assume(i_ce); // Second induction i_ce assumption: avoid skipping an // i_ce and thus stretching out the i_ce cycle two i_ce // cycles in a row. Without this assumption, induction // would still complete, it would just take longer always @(posedge i_clk) if (($past(i_ce))&&(!$past(i_ce,2))) assume(i_ce); end else if (CKPCE == 2) begin : F_CKPCE_TWO // Primary i_ce assumption: Every i_ce cycle is followed // by a non-i_ce cycle, so the multiplies can be // multiplexed always @(posedge i_clk) if ($past(i_ce)) assume(!i_ce); // First induction assumption: Don't let this stretch // out too far. This is necessary to pass induction always @(posedge i_clk) if ((!$past(i_ce))&&(!$past(i_ce,2))) assume(i_ce); always @(posedge i_clk) if ((!$past(i_ce))&&($past(i_ce,2)) &&(!$past(i_ce,3))&&(!$past(i_ce,4))) assume(i_ce); end else if (CKPCE == 3) begin : F_CKPCE_THREE // Primary i_ce assumption: Following any i_ce cycle, // there must be two clock cycles with i_ce de-asserted always @(posedge i_clk) if (($past(i_ce))||($past(i_ce,2))) assume(!i_ce); // Induction assumption: Allow i_ce's every third or // fourth clock, but don't allow them to be separated // further than that always @(posedge i_clk) if ((!$past(i_ce))&&(!$past(i_ce,2))&&(!$past(i_ce,3))) assume(i_ce); // Second induction assumption, to speed up the proof: // If it's the earliest possible opportunity for an // i_ce, and the last i_ce was late, don't let this one // be late as well. always @(posedge i_clk) if ((!$past(i_ce))&&(!$past(i_ce,2)) &&($past(i_ce,3))&&(!$past(i_ce,4)) &&(!$past(i_ce,5))&&(!$past(i_ce,6))) assume(i_ce); end endgenerate `endif reg [F_LGDEPTH:0] f_startup_counter; initial f_startup_counter = 0; always @(posedge i_clk) if (i_reset) f_startup_counter <= 0; else if ((i_ce)&&(!(&f_startup_counter))) f_startup_counter <= f_startup_counter + 1; always @(*) begin f_sumr = f_dlyleft_r[F_D] + f_dlyright_r[F_D]; f_sumi = f_dlyleft_i[F_D] + f_dlyright_i[F_D]; end assign f_sumrx = { {(4){f_sumr[IWIDTH]}}, f_sumr, {(CWIDTH-2){1'b0}} }; assign f_sumix = { {(4){f_sumi[IWIDTH]}}, f_sumi, {(CWIDTH-2){1'b0}} }; always @(*) begin f_difr = f_dlyleft_r[F_D] - f_dlyright_r[F_D]; f_difi = f_dlyleft_i[F_D] - f_dlyright_i[F_D]; end assign f_difrx = { {(CWIDTH+2){f_difr[IWIDTH]}}, f_difr }; assign f_difix = { {(CWIDTH+2){f_difi[IWIDTH]}}, f_difi }; assign f_widecoeff_r ={ {(IWIDTH+3){f_dlycoeff_r[F_D][CWIDTH-1]}}, f_dlycoeff_r[F_D] }; assign f_widecoeff_i ={ {(IWIDTH+3){f_dlycoeff_i[F_D][CWIDTH-1]}}, f_dlycoeff_i[F_D] }; always @(posedge i_clk) if (f_startup_counter > {1'b0, F_D}) begin assert(aux_pipeline == f_dlyaux); assert(left_sr == f_sumrx); assert(left_si == f_sumix); assert(aux_pipeline[AUXLEN-1] == f_dlyaux[F_D]); if ((f_difr == 0)&&(f_difi == 0)) begin assert(mpy_r == 0); assert(mpy_i == 0); end else if ((f_dlycoeff_r[F_D] == 0) &&(f_dlycoeff_i[F_D] == 0)) begin assert(mpy_r == 0); assert(mpy_i == 0); end if ((f_dlycoeff_r[F_D] == 1)&&(f_dlycoeff_i[F_D] == 0)) begin assert(mpy_r == f_difrx); assert(mpy_i == f_difix); end if ((f_dlycoeff_r[F_D] == 0)&&(f_dlycoeff_i[F_D] == 1)) begin assert(mpy_r == -f_difix); assert(mpy_i == f_difrx); end if ((f_difr == 1)&&(f_difi == 0)) begin assert(mpy_r == f_widecoeff_r); assert(mpy_i == f_widecoeff_i); end if ((f_difr == 0)&&(f_difi == 1)) begin assert(mpy_r == -f_widecoeff_i); assert(mpy_i == f_widecoeff_r); end end // Let's see if we can improve our performance at all by // moving our test one clock earlier. If nothing else, it should // help induction finish one (or more) clocks ealier than // otherwise always @(*) begin f_predifr = f_dlyleft_r[F_D-1] - f_dlyright_r[F_D-1]; f_predifi = f_dlyleft_i[F_D-1] - f_dlyright_i[F_D-1]; end assign f_predifrx = { {(CWIDTH+2){f_predifr[IWIDTH]}}, f_predifr }; assign f_predifix = { {(CWIDTH+2){f_predifi[IWIDTH]}}, f_predifi }; always @(*) begin f_sumcoef = f_dlycoeff_r[F_D-1] + f_dlycoeff_i[F_D-1]; f_sumdiff = f_predifr + f_predifi; end // Induction helpers always @(posedge i_clk) if (f_startup_counter >= { 1'b0, F_D }) begin if (f_dlycoeff_r[F_D-1] == 0) assert(p_one == 0); if (f_dlycoeff_i[F_D-1] == 0) assert(p_two == 0); if (f_dlycoeff_r[F_D-1] == 1) assert(p_one == f_predifrx); if (f_dlycoeff_i[F_D-1] == 1) assert(p_two == f_predifix); if (f_predifr == 0) assert(p_one == 0); if (f_predifi == 0) assert(p_two == 0); // verilator lint_off WIDTH if (f_predifr == 1) assert(p_one == f_dlycoeff_r[F_D-1]); if (f_predifi == 1) assert(p_two == f_dlycoeff_i[F_D-1]); // verilator lint_on WIDTH if (f_sumcoef == 0) assert(p_three == 0); if (f_sumdiff == 0) assert(p_three == 0); // verilator lint_off WIDTH if (f_sumcoef == 1) assert(p_three == f_sumdiff); if (f_sumdiff == 1) assert(p_three == f_sumcoef); // verilator lint_on WIDTH `ifdef VERILATOR // Check that the multiplies match--but *ONLY* if using // Verilator, and not if using formal proper assert(p_one == f_predifr * f_dlycoeff_r[F_D-1]); assert(p_two == f_predifi * f_dlycoeff_i[F_D-1]); assert(p_three == f_sumdiff * f_sumcoef); `endif // VERILATOR end // The following logic formally insists that our version of the // inputs to the multiply matches what the (multiclock) multiply // thinks its inputs were. While this may seem redundant, the // proof will not complete in any reasonable amount of time // without these assertions. assign f_p3c_in = f_dlycoeff_i[F_D-1] + f_dlycoeff_r[F_D-1]; assign f_p3d_in = f_predifi + f_predifr; always @(*) if (f_startup_counter >= { 1'b0, F_D }) begin assert(fp_one_ic == { f_dlycoeff_r[F_D-1][CWIDTH-1], f_dlycoeff_r[F_D-1][CWIDTH-1:0] }); assert(fp_two_ic == { f_dlycoeff_i[F_D-1][CWIDTH-1], f_dlycoeff_i[F_D-1][CWIDTH-1:0] }); assert(fp_one_id == { f_predifr[IWIDTH], f_predifr }); assert(fp_two_id == { f_predifi[IWIDTH], f_predifi }); assert(fp_three_ic == f_p3c_in); assert(fp_three_id == f_p3d_in); end // F_CHECK will be set externally by the solver, so that we can // double check that the solver is actually testing what we think // it is testing. We'll set it here to MPYREMAINDER, which will // essentially eliminate the check--unless overridden by the // solver. parameter F_CHECK = MPYREMAINDER; initial assert(MPYREMAINDER == F_CHECK); `endif // FORMAL endmodule
Go to most recent revision | Compare with Previous | Blame | View Log