diff options
Diffstat (limited to 'techlibs/xilinx')
-rw-r--r-- | techlibs/xilinx/Makefile.inc | 3 | ||||
-rw-r--r-- | techlibs/xilinx/abc_xc7.box | 62 | ||||
-rw-r--r-- | techlibs/xilinx/abc_xc7.lut | 15 | ||||
-rw-r--r-- | techlibs/xilinx/arith_map.v | 4 | ||||
-rw-r--r-- | techlibs/xilinx/cells_map.v | 138 | ||||
-rw-r--r-- | techlibs/xilinx/cells_sim.v | 7 | ||||
-rw-r--r-- | techlibs/xilinx/cells_xtra.sh | 4 | ||||
-rw-r--r-- | techlibs/xilinx/cells_xtra.v | 18 | ||||
-rw-r--r-- | techlibs/xilinx/lut_map.v | 97 | ||||
-rw-r--r-- | techlibs/xilinx/mux_map.v | 52 | ||||
-rw-r--r-- | techlibs/xilinx/synth_xilinx.cc | 87 |
11 files changed, 401 insertions, 86 deletions
diff --git a/techlibs/xilinx/Makefile.inc b/techlibs/xilinx/Makefile.inc index d68f03bb4..59fd61cf0 100644 --- a/techlibs/xilinx/Makefile.inc +++ b/techlibs/xilinx/Makefile.inc @@ -30,6 +30,9 @@ $(eval $(call add_share_file,share/xilinx,techlibs/xilinx/drams_map.v)) $(eval $(call add_share_file,share/xilinx,techlibs/xilinx/arith_map.v)) $(eval $(call add_share_file,share/xilinx,techlibs/xilinx/ff_map.v)) $(eval $(call add_share_file,share/xilinx,techlibs/xilinx/lut_map.v)) +$(eval $(call add_share_file,share/xilinx,techlibs/xilinx/mux_map.v)) +$(eval $(call add_share_file,share/xilinx,techlibs/xilinx/abc_xc7.box)) +$(eval $(call add_share_file,share/xilinx,techlibs/xilinx/abc_xc7.lut)) $(eval $(call add_gen_share_file,share/xilinx,techlibs/xilinx/brams_init_36.vh)) $(eval $(call add_gen_share_file,share/xilinx,techlibs/xilinx/brams_init_32.vh)) diff --git a/techlibs/xilinx/abc_xc7.box b/techlibs/xilinx/abc_xc7.box new file mode 100644 index 000000000..8a48bad4e --- /dev/null +++ b/techlibs/xilinx/abc_xc7.box @@ -0,0 +1,62 @@ +# Max delays from https://github.com/SymbiFlow/prjxray-db/blob/34ea6eb08a63d21ec16264ad37a0a7b142ff6031/artix7/timings/CLBLL_L.sdf + +# F7BMUX slower than F7AMUX +# Inputs: I0 I1 S0 +# Outputs: O +F7BMUX 1 1 3 1 +217 223 296 + +# Inputs: I0 I1 S0 +# Outputs: O +MUXF8 2 1 3 1 +104 94 273 + +# CARRY4 + CARRY4_[ABCD]X +# Inputs: S0 S1 S2 S3 CYINIT DI0 DI1 DI2 DI3 CI +# Outputs: O0 O1 O2 O3 CO0 CO1 CO2 CO3 +# (NB: carry chain input/output must be last input/output, +# swapped with what normally would have been the last +# output, here: CI <-> S, CO <-> O +CARRY4 3 1 10 8 +223 - - - 482 - - - - 222 +400 205 - - 598 407 - - - 334 +523 558 226 - 584 556 537 - - 239 +582 618 330 227 642 615 596 438 - 313 +340 - - - 536 379 - - - 271 +433 469 - - 494 465 445 - - 157 +512 548 292 - 592 540 520 356 - 228 +508 528 378 380 580 526 507 398 385 114 + +# SLICEM/A6LUT +# Inputs: A0 A1 A2 A3 A4 A5 D DPRA0 DPRA1 DPRA2 DPRA3 DPRA4 DPRA5 WCLK WE +# Outputs: DPO SPO +RAM64X1D 4 0 15 2 +- - - - - - - 124 124 124 124 124 124 - - +124 124 124 124 124 124 - - - - - - 124 - - + +# SLICEM/A6LUT + F7[AB]MUX +# Inputs: A0 A1 A2 A3 A4 A5 A6 D DPRA0 DPRA1 DPRA2 DPRA3 DPRA4 DPRA5 DPRA6 WCLK WE +# Outputs: DPO SPO +RAM128X1D 5 0 17 2 +- - - - - - - - 314 314 314 314 314 314 292 - - +347 347 347 347 347 347 296 - - - - - - - - - - + +# Inputs: C CE D R +# Outputs: Q +FDRE 6 0 4 1 +- - - - + +# Inputs: C CE D S +# Outputs: Q +FDSE 7 0 4 1 +- - - - + +# Inputs: C CE CLR D +# Outputs: Q +FDCE 8 0 4 1 +- - - - + +# Inputs: C CE D PRE +# Outputs: Q +FDPE 9 0 4 1 +- - - - diff --git a/techlibs/xilinx/abc_xc7.lut b/techlibs/xilinx/abc_xc7.lut new file mode 100644 index 000000000..bcbdec127 --- /dev/null +++ b/techlibs/xilinx/abc_xc7.lut @@ -0,0 +1,15 @@ +# Max delays from https://github.com/SymbiFlow/prjxray-db/blob/82bf5f158cd8e9a11ac4d04f1aeef48ed1a528a5/artix7/timings/CLBLL_L.sdf +# and https://github.com/SymbiFlow/prjxray-db/blob/82bf5f158cd8e9a11ac4d04f1aeef48ed1a528a5/artix7/tile_type_CLBLL_L.json + +# K area delay +1 1 127 +2 2 127 238 +3 3 127 238 407 +4 3 127 238 407 472 +5 3 127 238 407 472 631 +6 5 127 238 407 472 631 642 + # (F7[AB]MUX.S + [AC]OUTMUX) / 2 +7 10 464 513 624 793 858 1017 1028 + # F8MUX.S+BOUTMUX + # F8MUX.I0+F7MUX.S+BOUTMUX +8 20 468 585 634 745 914 979 1138 1149 diff --git a/techlibs/xilinx/arith_map.v b/techlibs/xilinx/arith_map.v index 09a5f07e8..5c848d4e6 100644 --- a/techlibs/xilinx/arith_map.v +++ b/techlibs/xilinx/arith_map.v @@ -180,7 +180,7 @@ module _80_xilinx_alu (A, B, CI, BI, X, Y, CO); // First one if (i == 0) begin - CARRY4 #(.IS_INITIALIZED(1'd1)) carry4_1st_part + CARRY4 carry4_1st_part ( .CYINIT(CI), .CI (1'd0), @@ -207,7 +207,7 @@ module _80_xilinx_alu (A, B, CI, BI, X, Y, CO); // First one if (i == 0) begin - CARRY4 #(.IS_INITIALIZED(1'd1)) carry4_1st_full + CARRY4 carry4_1st_full ( .CYINIT(CI), .CI (1'd0), diff --git a/techlibs/xilinx/cells_map.v b/techlibs/xilinx/cells_map.v index 40789ddbe..f139dc5d5 100644 --- a/techlibs/xilinx/cells_map.v +++ b/techlibs/xilinx/cells_map.v @@ -2,6 +2,7 @@ * yosys -- Yosys Open SYnthesis Suite * * Copyright (C) 2012 Clifford Wolf <clifford@clifford.at> + * 2019 Eddie Hung <eddie@fpgeh.com> * * Permission to use, copy, modify, and/or distribute this software for any * purpose with or without fee is hereby granted, provided that the above @@ -19,14 +20,16 @@ // Convert negative-polarity reset to positive-polarity (* techmap_celltype = "$_DFF_NN0_" *) -module _90_dff_nn0_to_np0 (input D, C, R, output Q); \$_DFF_NP0_ _TECHMAP_REPLACE_ (.D(D), .Q(Q), .C(C), .R(~R)); endmodule +module _90_dff_nn0_to_np0(input D, C, R, output Q); \$_DFF_NP0_ _TECHMAP_REPLACE_ (.D(D), .Q(Q), .C(C), .R(~R)); endmodule (* techmap_celltype = "$_DFF_PN0_" *) -module _90_dff_pn0_to_pp0 (input D, C, R, output Q); \$_DFF_PP0_ _TECHMAP_REPLACE_ (.D(D), .Q(Q), .C(C), .R(~R)); endmodule +module _90_dff_pn0_to_pp0(input D, C, R, output Q); \$_DFF_PP0_ _TECHMAP_REPLACE_ (.D(D), .Q(Q), .C(C), .R(~R)); endmodule + (* techmap_celltype = "$_DFF_NN1_" *) module _90_dff_nn1_to_np1 (input D, C, R, output Q); \$_DFF_NP1 _TECHMAP_REPLACE_ (.D(D), .Q(Q), .C(C), .R(~R)); endmodule (* techmap_celltype = "$_DFF_PN1_" *) module _90_dff_pn1_to_pp1 (input D, C, R, output Q); \$_DFF_PP1 _TECHMAP_REPLACE_ (.D(D), .Q(Q), .C(C), .R(~R)); endmodule + module \$__SHREG_ (input C, input D, input E, output Q); parameter DEPTH = 0; parameter [DEPTH-1:0] INIT = 0; @@ -88,7 +91,7 @@ module \$__XILINX_SHREG_ (input C, input D, input [31:0] L, input E, output Q, o end else if (DEPTH > 65 && DEPTH <= 96) begin wire T0, T1, T2, T3, T4, T5, T6; - SRLC32E #(.INIT(INIT_R[32-1:0]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_0 (.A(L[4:0]), .CE(CE), .CLK(C), .D(D), .Q(T0), .Q31(T1)); + SRLC32E #(.INIT(INIT_R[32-1: 0]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_0 (.A(L[4:0]), .CE(CE), .CLK(C), .D( D), .Q(T0), .Q31(T1)); SRLC32E #(.INIT(INIT_R[64-1:32]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_1 (.A(L[4:0]), .CE(CE), .CLK(C), .D(T1), .Q(T2), .Q31(T3)); \$__XILINX_SHREG_ #(.DEPTH(DEPTH-64), .INIT(INIT[DEPTH-64-1:0]), .CLKPOL(CLKPOL), .ENPOL(ENPOL)) fpga_srl_2 (.C(C), .D(T3), .L(L[4:0]), .E(E), .Q(T4)); if (&_TECHMAP_CONSTMSK_L_) @@ -101,7 +104,7 @@ module \$__XILINX_SHREG_ (input C, input D, input [31:0] L, input E, output Q, o end else if (DEPTH > 97 && DEPTH < 128) begin wire T0, T1, T2, T3, T4, T5, T6, T7, T8; - SRLC32E #(.INIT(INIT_R[32-1:0]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_0 (.A(L[4:0]), .CE(CE), .CLK(C), .D(D), .Q(T0), .Q31(T1)); + SRLC32E #(.INIT(INIT_R[32-1: 0]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_0 (.A(L[4:0]), .CE(CE), .CLK(C), .D( D), .Q(T0), .Q31(T1)); SRLC32E #(.INIT(INIT_R[64-1:32]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_1 (.A(L[4:0]), .CE(CE), .CLK(C), .D(T1), .Q(T2), .Q31(T3)); SRLC32E #(.INIT(INIT_R[96-1:64]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_2 (.A(L[4:0]), .CE(CE), .CLK(C), .D(T3), .Q(T4), .Q31(T5)); \$__XILINX_SHREG_ #(.DEPTH(DEPTH-96), .INIT(INIT[DEPTH-96-1:0]), .CLKPOL(CLKPOL), .ENPOL(ENPOL)) fpga_srl_3 (.C(C), .D(T5), .L(L[4:0]), .E(E), .Q(T6)); @@ -115,9 +118,9 @@ module \$__XILINX_SHREG_ (input C, input D, input [31:0] L, input E, output Q, o end else if (DEPTH == 128) begin wire T0, T1, T2, T3, T4, T5, T6; - SRLC32E #(.INIT(INIT_R[32-1:0]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_0 (.A(L[4:0]), .CE(CE), .CLK(C), .D(D), .Q(T0), .Q31(T1)); - SRLC32E #(.INIT(INIT_R[64-1:32]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_1 (.A(L[4:0]), .CE(CE), .CLK(C), .D(T1), .Q(T2), .Q31(T3)); - SRLC32E #(.INIT(INIT_R[96-1:64]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_2 (.A(L[4:0]), .CE(CE), .CLK(C), .D(T3), .Q(T4), .Q31(T5)); + SRLC32E #(.INIT(INIT_R[ 32-1: 0]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_0 (.A(L[4:0]), .CE(CE), .CLK(C), .D( D), .Q(T0), .Q31(T1)); + SRLC32E #(.INIT(INIT_R[ 64-1:32]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_1 (.A(L[4:0]), .CE(CE), .CLK(C), .D(T1), .Q(T2), .Q31(T3)); + SRLC32E #(.INIT(INIT_R[ 96-1:64]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_2 (.A(L[4:0]), .CE(CE), .CLK(C), .D(T3), .Q(T4), .Q31(T5)); SRLC32E #(.INIT(INIT_R[128-1:96]), .IS_CLK_INVERTED(~CLKPOL[0])) fpga_srl_3 (.A(L[4:0]), .CE(CE), .CLK(C), .D(T5), .Q(T6), .Q31(SO)); if (&_TECHMAP_CONSTMSK_L_) assign Q = T6; @@ -152,5 +155,122 @@ module \$__XILINX_SHREG_ (input C, input D, input [31:0] L, input E, output Q, o endgenerate endmodule -`ifndef SRL_ONLY -`endif +module \$__XILINX_SHIFTX (A, B, Y); + parameter A_SIGNED = 0; + parameter B_SIGNED = 0; + parameter A_WIDTH = 1; + parameter B_WIDTH = 1; + parameter Y_WIDTH = 1; + + input [A_WIDTH-1:0] A; + input [B_WIDTH-1:0] B; + output [Y_WIDTH-1:0] Y; + + parameter [A_WIDTH-1:0] _TECHMAP_CONSTMSK_A_ = 0; + parameter [A_WIDTH-1:0] _TECHMAP_CONSTVAL_A_ = 0; + parameter [B_WIDTH-1:0] _TECHMAP_CONSTMSK_B_ = 0; + parameter [B_WIDTH-1:0] _TECHMAP_CONSTVAL_B_ = 0; + + function integer compute_num_leading_X_in_A; + integer i, c; + begin + compute_num_leading_X_in_A = 0; + c = 1; + for (i = A_WIDTH-1; i >= 0; i=i-1) begin + if (!_TECHMAP_CONSTMSK_A_[i] || _TECHMAP_CONSTVAL_A_[i] !== 1'bx) + c = 0; + compute_num_leading_X_in_A = compute_num_leading_X_in_A + c; + end + end + endfunction + localparam num_leading_X_in_A = compute_num_leading_X_in_A(); + + generate + genvar i, j; + // Bit-blast + if (Y_WIDTH > 1) begin + for (i = 0; i < Y_WIDTH; i++) + \$__XILINX_SHIFTX #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(A_WIDTH-Y_WIDTH+1), .B_WIDTH(B_WIDTH), .Y_WIDTH(1'd1)) bitblast (.A(A[A_WIDTH-Y_WIDTH+i:i]), .B(B), .Y(Y[i])); + end + // If the LSB of B is constant zero (and Y_WIDTH is 1) then + // we can optimise by removing every other entry from A + // and popping the constant zero from B + else if (_TECHMAP_CONSTMSK_B_[0] && !_TECHMAP_CONSTVAL_B_[0]) begin + wire [(A_WIDTH+1)/2-1:0] A_i; + for (i = 0; i < (A_WIDTH+1)/2; i++) + assign A_i[i] = A[i*2]; + \$__XILINX_SHIFTX #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH((A_WIDTH+1'd1)/2'd2), .B_WIDTH(B_WIDTH-1'd1), .Y_WIDTH(Y_WIDTH)) _TECHMAP_REPLACE_ (.A(A_i), .B(B[B_WIDTH-1:1]), .Y(Y)); + end + // Trim off any leading 1'bx -es in A, and resize B accordingly + else if (num_leading_X_in_A > 0) begin + localparam A_WIDTH_new = A_WIDTH - num_leading_X_in_A; + localparam B_WIDTH_new = $clog2(A_WIDTH_new); + \$__XILINX_SHIFTX #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(A_WIDTH_new), .B_WIDTH(B_WIDTH_new), .Y_WIDTH(Y_WIDTH)) _TECHMAP_REPLACE_ (.A(A[A_WIDTH_new-1:0]), .B(B[B_WIDTH_new-1:0]), .Y(Y)); + end + else if (B_WIDTH < 3 || A_WIDTH <= 4) begin + \$shiftx #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(A_WIDTH), .B_WIDTH(B_WIDTH), .Y_WIDTH(Y_WIDTH)) _TECHMAP_REPLACE_ (.A(A), .B(B), .Y(Y)); + end + else if (B_WIDTH == 3) begin + localparam a_width0 = 2 ** 2; + localparam a_widthN = A_WIDTH - a_width0; + wire T0, T1; + \$shiftx #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(a_width0), .B_WIDTH(2), .Y_WIDTH(Y_WIDTH)) fpga_soft_mux (.A(A[a_width0-1:0]), .B(B[2-1:0]), .Y(T0)); + if (a_widthN > 1) + \$shiftx #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(a_widthN), .B_WIDTH($clog2(a_widthN)), .Y_WIDTH(Y_WIDTH)) fpga_soft_mux_last (.A(A[A_WIDTH-1:a_width0]), .B(B[$clog2(a_widthN)-1:0]), .Y(T1)); + else + assign T1 = A[A_WIDTH-1]; + MUXF7 fpga_hard_mux (.I0(T0), .I1(T1), .S(B[B_WIDTH-1]), .O(Y)); + end + else if (B_WIDTH == 4) begin + localparam a_width0 = 2 ** 2; + localparam num_mux8 = A_WIDTH / a_width0; + localparam a_widthN = A_WIDTH - num_mux8*a_width0; + wire [4-1:0] T; + wire T0, T1; + for (i = 0; i < 4; i++) + if (i < num_mux8) + \$shiftx #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(a_width0), .B_WIDTH(2), .Y_WIDTH(Y_WIDTH)) fpga_soft_mux (.A(A[i*a_width0+:a_width0]), .B(B[2-1:0]), .Y(T[i])); + else if (i == num_mux8 && a_widthN > 0) begin + if (a_widthN > 1) + \$shiftx #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(a_widthN), .B_WIDTH($clog2(a_widthN)), .Y_WIDTH(Y_WIDTH)) fpga_soft_mux_last (.A(A[A_WIDTH-1:i*a_width0]), .B(B[$clog2(a_widthN)-1:0]), .Y(T[i])); + else + assign T[i] = A[A_WIDTH-1]; + end + else + assign T[i] = 1'bx; + MUXF7 fpga_hard_mux_0 (.I0(T[0]), .I1(T[1]), .S(B[2]), .O(T0)); + MUXF7 fpga_hard_mux_1 (.I0(T[2]), .I1(T[3]), .S(B[2]), .O(T1)); + MUXF8 fpga_hard_mux_2 (.I0(T0), .I1(T1), .S(B[3]), .O(Y)); + end + else begin + localparam a_width0 = 2 ** 4; + localparam num_mux16 = A_WIDTH / a_width0; + localparam a_widthN = A_WIDTH - num_mux16*a_width0; + wire [(2**(B_WIDTH-4))-1:0] T; + for (i = 0; i < 2 ** (B_WIDTH-4); i++) + if (i < num_mux16) + \$__XILINX_SHIFTX #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(a_width0), .B_WIDTH(4), .Y_WIDTH(Y_WIDTH)) fpga_soft_mux (.A(A[i*a_width0+:a_width0]), .B(B[4-1:0]), .Y(T[i])); + else if (i == num_mux16 && a_widthN > 0) begin + if (a_widthN > 1) + \$__XILINX_SHIFTX #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(a_widthN), .B_WIDTH($clog2(a_widthN)), .Y_WIDTH(Y_WIDTH)) fpga_soft_mux_last (.A(A[A_WIDTH-1:i*a_width0]), .B(B[$clog2(a_widthN)-1:0]), .Y(T[i])); + else + assign T[i] = A[A_WIDTH-1]; + end + else + assign T[i] = 1'bx; + \$__XILINX_SHIFTX #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(2**(B_WIDTH-4)), .B_WIDTH(B_WIDTH-4), .Y_WIDTH(Y_WIDTH)) _TECHMAP_REPLACE_ (.A(T), .B(B[B_WIDTH-1:4]), .Y(Y)); + end + endgenerate +endmodule + +module \$_MUX8_ (A, B, C, D, E, F, G, H, S, T, U, Y); +input A, B, C, D, E, F, G, H, S, T, U; +output Y; + \$__XILINX_SHIFTX #(.A_SIGNED(0), .B_SIGNED(0), .A_WIDTH(8), .B_WIDTH(3), .Y_WIDTH(1)) _TECHMAP_REPLACE_ (.A({H,G,F,E,D,C,B,A}), .B({U,T,S}), .Y(Y)); +endmodule + +module \$_MUX16_ (A, B, C, D, E, F, G, H, I, J, K, L, M, N, O, P, S, T, U, V, Y); +input A, B, C, D, E, F, G, H, I, J, K, L, M, N, O, P, S, T, U, V; +output Y; + \$__XILINX_SHIFTX #(.A_SIGNED(0), .B_SIGNED(0), .A_WIDTH(16), .B_WIDTH(4), .Y_WIDTH(1)) _TECHMAP_REPLACE_ (.A({P,O,N,M,L,K,J,I,H,G,F,E,D,C,B,A}), .B({V,U,T,S}), .Y(Y)); +endmodule diff --git a/techlibs/xilinx/cells_sim.v b/techlibs/xilinx/cells_sim.v index 3a4540b83..bf7a0ed44 100644 --- a/techlibs/xilinx/cells_sim.v +++ b/techlibs/xilinx/cells_sim.v @@ -159,10 +159,12 @@ module MUXCY(output O, input CI, DI, S); assign O = S ? CI : DI; endmodule +(* abc_box_id = 1, lib_whitebox *) module MUXF7(output O, input I0, I1, S); assign O = S ? I1 : I0; endmodule +(* abc_box_id = 2, lib_whitebox *) module MUXF8(output O, input I0, I1, S); assign O = S ? I1 : I0; endmodule @@ -171,7 +173,8 @@ module XORCY(output O, input CI, LI); assign O = CI ^ LI; endmodule -module CARRY4(output [3:0] CO, O, input CI, CYINIT, input [3:0] DI, S); +(* abc_box_id = 3, abc_carry, lib_whitebox *) +module CARRY4((* abc_carry_out *) output [3:0] CO, output [3:0] O, (* abc_carry_in *) input CI, input CYINIT, input [3:0] DI, S); assign O = S ^ {CO[2:0], CI | CYINIT}; assign CO[0] = S[0] ? CI | CYINIT : DI[0]; assign CO[1] = S[1] ? CO[0] : DI[1]; @@ -278,6 +281,7 @@ module FDPE_1 (output reg Q, input C, CE, D, PRE); always @(negedge C, posedge PRE) if (PRE) Q <= 1'b1; else if (CE) Q <= D; endmodule +//(* abc_box_id = 4 /*, lib_whitebox*/ *) module RAM64X1D ( output DPO, SPO, input D, WCLK, WE, @@ -295,6 +299,7 @@ module RAM64X1D ( always @(posedge clk) if (WE) mem[a] <= D; endmodule +//(* abc_box_id = 5 /*, lib_whitebox*/ *) module RAM128X1D ( output DPO, SPO, input D, WCLK, WE, diff --git a/techlibs/xilinx/cells_xtra.sh b/techlibs/xilinx/cells_xtra.sh index 8e39b440d..2b384f405 100644 --- a/techlibs/xilinx/cells_xtra.sh +++ b/techlibs/xilinx/cells_xtra.sh @@ -116,7 +116,7 @@ function xtract_cell_decl() xtract_cell_decl PS7 "(* keep *)" xtract_cell_decl PULLDOWN xtract_cell_decl PULLUP - xtract_cell_decl RAM128X1D + #xtract_cell_decl RAM128X1D xtract_cell_decl RAM128X1S xtract_cell_decl RAM256X1S xtract_cell_decl RAM32M @@ -125,7 +125,7 @@ function xtract_cell_decl() xtract_cell_decl RAM32X1S_1 xtract_cell_decl RAM32X2S xtract_cell_decl RAM64M - xtract_cell_decl RAM64X1D + #xtract_cell_decl RAM64X1D xtract_cell_decl RAM64X1S xtract_cell_decl RAM64X1S_1 xtract_cell_decl RAM64X2S diff --git a/techlibs/xilinx/cells_xtra.v b/techlibs/xilinx/cells_xtra.v index fbcc74682..0ec3d0df0 100644 --- a/techlibs/xilinx/cells_xtra.v +++ b/techlibs/xilinx/cells_xtra.v @@ -3655,17 +3655,6 @@ module PULLUP (...); output O; endmodule -module RAM128X1D (...); - parameter [127:0] INIT = 128'h00000000000000000000000000000000; - parameter [0:0] IS_WCLK_INVERTED = 1'b0; - output DPO, SPO; - input [6:0] A; - input [6:0] DPRA; - input D; - input WCLK; - input WE; -endmodule - module RAM128X1S (...); parameter [127:0] INIT = 128'h00000000000000000000000000000000; parameter [0:0] IS_WCLK_INVERTED = 1'b0; @@ -3756,13 +3745,6 @@ module RAM64M (...); input WE; endmodule -module RAM64X1D (...); - parameter [63:0] INIT = 64'h0000000000000000; - parameter [0:0] IS_WCLK_INVERTED = 1'b0; - output DPO, SPO; - input A0, A1, A2, A3, A4, A5, D, DPRA0, DPRA1, DPRA2, DPRA3, DPRA4, DPRA5, WCLK, WE; -endmodule - module RAM64X1S (...); parameter [63:0] INIT = 64'h0000000000000000; parameter [0:0] IS_WCLK_INVERTED = 1'b0; diff --git a/techlibs/xilinx/lut_map.v b/techlibs/xilinx/lut_map.v index d07c59dee..13d3c3268 100644 --- a/techlibs/xilinx/lut_map.v +++ b/techlibs/xilinx/lut_map.v @@ -29,61 +29,86 @@ module \$lut (A, Y); input [WIDTH-1:0] A; output Y; + // Need to swap input ordering, and fix init accordingly, + // to match ABC's expectation of LUT inputs in non-decreasing + // delay order + function [WIDTH-1:0] permute_index; + input [WIDTH-1:0] i; + integer j; + begin + permute_index = 0; + for (j = 0; j < WIDTH; j = j + 1) + permute_index[WIDTH-1 - j] = i[j]; + end + endfunction + + function [2**WIDTH-1:0] permute_init; + input [2**WIDTH-1:0] orig; + integer i; + begin + permute_init = 0; + for (i = 0; i < 2**WIDTH; i = i + 1) + permute_init[i] = orig[permute_index(i)]; + end + endfunction + + parameter [2**WIDTH-1:0] P_LUT = permute_init(LUT); + generate if (WIDTH == 1) begin - LUT1 #(.INIT(LUT)) _TECHMAP_REPLACE_ (.O(Y), + LUT1 #(.INIT(P_LUT)) _TECHMAP_REPLACE_ (.O(Y), .I0(A[0])); end else if (WIDTH == 2) begin - LUT2 #(.INIT(LUT)) _TECHMAP_REPLACE_ (.O(Y), - .I0(A[0]), .I1(A[1])); + LUT2 #(.INIT(P_LUT)) _TECHMAP_REPLACE_ (.O(Y), + .I0(A[1]), .I1(A[0])); end else if (WIDTH == 3) begin - LUT3 #(.INIT(LUT)) _TECHMAP_REPLACE_ (.O(Y), - .I0(A[0]), .I1(A[1]), .I2(A[2])); + LUT3 #(.INIT(P_LUT)) _TECHMAP_REPLACE_ (.O(Y), + .I0(A[2]), .I1(A[1]), .I2(A[0])); end else if (WIDTH == 4) begin - LUT4 #(.INIT(LUT)) _TECHMAP_REPLACE_ (.O(Y), - .I0(A[0]), .I1(A[1]), .I2(A[2]), - .I3(A[3])); + LUT4 #(.INIT(P_LUT)) _TECHMAP_REPLACE_ (.O(Y), + .I0(A[3]), .I1(A[2]), .I2(A[1]), + .I3(A[0])); end else if (WIDTH == 5) begin - LUT5 #(.INIT(LUT)) _TECHMAP_REPLACE_ (.O(Y), - .I0(A[0]), .I1(A[1]), .I2(A[2]), - .I3(A[3]), .I4(A[4])); + LUT5 #(.INIT(P_LUT)) _TECHMAP_REPLACE_ (.O(Y), + .I0(A[4]), .I1(A[3]), .I2(A[2]), + .I3(A[1]), .I4(A[0])); end else if (WIDTH == 6) begin - LUT6 #(.INIT(LUT)) _TECHMAP_REPLACE_ (.O(Y), - .I0(A[0]), .I1(A[1]), .I2(A[2]), - .I3(A[3]), .I4(A[4]), .I5(A[5])); + LUT6 #(.INIT(P_LUT)) _TECHMAP_REPLACE_ (.O(Y), + .I0(A[5]), .I1(A[4]), .I2(A[3]), + .I3(A[2]), .I4(A[1]), .I5(A[0])); end else if (WIDTH == 7) begin wire T0, T1; - LUT6 #(.INIT(LUT[63:0])) fpga_lut_0 (.O(T0), - .I0(A[0]), .I1(A[1]), .I2(A[2]), - .I3(A[3]), .I4(A[4]), .I5(A[5])); - LUT6 #(.INIT(LUT[127:64])) fpga_lut_1 (.O(T1), - .I0(A[0]), .I1(A[1]), .I2(A[2]), - .I3(A[3]), .I4(A[4]), .I5(A[5])); - MUXF7 fpga_mux_0 (.O(Y), .I0(T0), .I1(T1), .S(A[6])); + LUT6 #(.INIT(P_LUT[63:0])) fpga_lut_0 (.O(T0), + .I0(A[6]), .I1(A[5]), .I2(A[4]), + .I3(A[3]), .I4(A[2]), .I5(A[1])); + LUT6 #(.INIT(P_LUT[127:64])) fpga_lut_1 (.O(T1), + .I0(A[6]), .I1(A[5]), .I2(A[4]), + .I3(A[3]), .I4(A[2]), .I5(A[1])); + MUXF7 fpga_mux_0 (.O(Y), .I0(T0), .I1(T1), .S(A[0])); end else if (WIDTH == 8) begin wire T0, T1, T2, T3, T4, T5; - LUT6 #(.INIT(LUT[63:0])) fpga_lut_0 (.O(T0), - .I0(A[0]), .I1(A[1]), .I2(A[2]), - .I3(A[3]), .I4(A[4]), .I5(A[5])); - LUT6 #(.INIT(LUT[127:64])) fpga_lut_1 (.O(T1), - .I0(A[0]), .I1(A[1]), .I2(A[2]), - .I3(A[3]), .I4(A[4]), .I5(A[5])); - LUT6 #(.INIT(LUT[191:128])) fpga_lut_2 (.O(T2), - .I0(A[0]), .I1(A[1]), .I2(A[2]), - .I3(A[3]), .I4(A[4]), .I5(A[5])); - LUT6 #(.INIT(LUT[255:192])) fpga_lut_3 (.O(T3), - .I0(A[0]), .I1(A[1]), .I2(A[2]), - .I3(A[3]), .I4(A[4]), .I5(A[5])); - MUXF7 fpga_mux_0 (.O(T4), .I0(T0), .I1(T1), .S(A[6])); - MUXF7 fpga_mux_1 (.O(T5), .I0(T2), .I1(T3), .S(A[6])); - MUXF8 fpga_mux_2 (.O(Y), .I0(T4), .I1(T5), .S(A[7])); + LUT6 #(.INIT(P_LUT[63:0])) fpga_lut_0 (.O(T0), + .I0(A[7]), .I1(A[6]), .I2(A[5]), + .I3(A[4]), .I4(A[3]), .I5(A[2])); + LUT6 #(.INIT(P_LUT[127:64])) fpga_lut_1 (.O(T1), + .I0(A[7]), .I1(A[6]), .I2(A[5]), + .I3(A[4]), .I4(A[3]), .I5(A[2])); + LUT6 #(.INIT(P_LUT[191:128])) fpga_lut_2 (.O(T2), + .I0(A[7]), .I1(A[6]), .I2(A[5]), + .I3(A[4]), .I4(A[3]), .I5(A[2])); + LUT6 #(.INIT(P_LUT[255:192])) fpga_lut_3 (.O(T3), + .I0(A[7]), .I1(A[6]), .I2(A[5]), + .I3(A[4]), .I4(A[3]), .I5(A[2])); + MUXF7 fpga_mux_0 (.O(T4), .I0(T0), .I1(T1), .S(A[1])); + MUXF7 fpga_mux_1 (.O(T5), .I0(T2), .I1(T3), .S(A[1])); + MUXF8 fpga_mux_2 (.O(Y), .I0(T4), .I1(T5), .S(A[0])); end else begin wire _TECHMAP_FAIL_ = 1; end diff --git a/techlibs/xilinx/mux_map.v b/techlibs/xilinx/mux_map.v new file mode 100644 index 000000000..0fa8db736 --- /dev/null +++ b/techlibs/xilinx/mux_map.v @@ -0,0 +1,52 @@ +/* + * yosys -- Yosys Open SYnthesis Suite + * + * Copyright (C) 2012 Clifford Wolf <clifford@clifford.at> + * 2019 Eddie Hung <eddie@fpgeh.com> + * + * Permission to use, copy, modify, and/or distribute this software for any + * purpose with or without fee is hereby granted, provided that the above + * copyright notice and this permission notice appear in all copies. + * + * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES + * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF + * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR + * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES + * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN + * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF + * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. + * + */ + +module \$shiftx (A, B, Y); + parameter A_SIGNED = 0; + parameter B_SIGNED = 0; + parameter A_WIDTH = 1; + parameter B_WIDTH = 1; + parameter Y_WIDTH = 1; + + input [A_WIDTH-1:0] A; + input [B_WIDTH-1:0] B; + output [Y_WIDTH-1:0] Y; + + parameter [B_WIDTH-1:0] _TECHMAP_CONSTMSK_B_ = 0; + parameter [B_WIDTH-1:0] _TECHMAP_CONSTVAL_B_ = 0; + + generate + genvar i, j; + // TODO: Check if this opt still necessary + if (B_SIGNED) begin + if (_TECHMAP_CONSTMSK_B_[B_WIDTH-1] && _TECHMAP_CONSTVAL_B_[B_WIDTH-1] == 1'b0) + // Optimisation to remove B_SIGNED if sign bit of B is constant-0 + \$__XILINX_SHIFTX #(.A_SIGNED(A_SIGNED), .B_SIGNED(0), .A_WIDTH(A_WIDTH), .B_WIDTH(B_WIDTH-1'd1), .Y_WIDTH(Y_WIDTH)) _TECHMAP_REPLACE_ (.A(A), .B(B[B_WIDTH-2:0]), .Y(Y)); + else + wire _TECHMAP_FAIL_ = 1; + end + else if (B_WIDTH < 3 || A_WIDTH <= 4) begin + wire _TECHMAP_FAIL_ = 1; + end + else begin + \$__XILINX_SHIFTX #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(A_WIDTH), .B_WIDTH(B_WIDTH), .Y_WIDTH(Y_WIDTH)) _TECHMAP_REPLACE_ (.A(A), .B(B), .Y(Y)); + end + endgenerate +endmodule diff --git a/techlibs/xilinx/synth_xilinx.cc b/techlibs/xilinx/synth_xilinx.cc index a293081f1..7b20a7132 100644 --- a/techlibs/xilinx/synth_xilinx.cc +++ b/techlibs/xilinx/synth_xilinx.cc @@ -25,6 +25,9 @@ USING_YOSYS_NAMESPACE PRIVATE_NAMESPACE_BEGIN +#define XC7_WIRE_DELAY "300" // Number with which ABC will map a 6-input gate + // to one LUT6 (instead of a LUT5 + LUT2) + struct SynthXilinxPass : public ScriptPass { SynthXilinxPass() : ScriptPass("synth_xilinx", "synthesis for Xilinx FPGAs") { } @@ -58,6 +61,9 @@ struct SynthXilinxPass : public ScriptPass log(" generate an output netlist (and BLIF file) suitable for VPR\n"); log(" (this feature is experimental and incomplete)\n"); log("\n"); + log(" -nocarry\n"); + log(" disable inference of carry chains\n"); + log("\n"); log(" -nobram\n"); log(" disable inference of block rams\n"); log("\n"); @@ -67,6 +73,9 @@ struct SynthXilinxPass : public ScriptPass log(" -nosrl\n"); log(" disable inference of shift registers\n"); log("\n"); + log(" -nomux\n"); + log(" disable inference of wide multiplexers\n"); + log("\n"); log(" -run <from_label>:<to_label>\n"); log(" only run the commands between the labels (see below). an empty\n"); log(" from label is synonymous to 'begin', and empty to label is\n"); @@ -78,26 +87,32 @@ struct SynthXilinxPass : public ScriptPass log(" -retime\n"); log(" run 'abc' with -dff option\n"); log("\n"); + log(" -abc9\n"); + log(" use new ABC9 flow (EXPERIMENTAL)\n"); + log("\n"); log("\n"); log("The following commands are executed by this synthesis command:\n"); help_script(); log("\n"); } - std::string top_opt, edif_file, blif_file, arch; - bool flatten, retime, vpr, nobram, nodram, nosrl; + std::string top_opt, edif_file, blif_file, abc, arch; + bool flatten, retime, vpr, nocarry, nobram, nodram, nosrl, nomux; void clear_flags() YS_OVERRIDE { top_opt = "-auto-top"; edif_file.clear(); blif_file.clear(); + abc = "abc"; flatten = false; retime = false; vpr = false; + nocarry = false; nobram = false; nodram = false; nosrl = false; + nomux = false; arch = "xc7"; } @@ -145,6 +160,10 @@ struct SynthXilinxPass : public ScriptPass vpr = true; continue; } + if (args[argidx] == "-nocarry") { + nocarry = true; + continue; + } if (args[argidx] == "-nobram") { nobram = true; continue; @@ -157,6 +176,14 @@ struct SynthXilinxPass : public ScriptPass nosrl = true; continue; } + if (args[argidx] == "-nomux") { + nomux = true; + continue; + } + if (args[argidx] == "-abc9") { + abc = "abc9"; + continue; + } break; } extra_args(args, argidx, design); @@ -179,9 +206,9 @@ struct SynthXilinxPass : public ScriptPass { if (check_label("begin")) { if (vpr) - run("read_verilog -lib -D_EXPLICIT_CARRY +/xilinx/cells_sim.v"); + run("read_verilog -lib -D _ABC -D_EXPLICIT_CARRY +/xilinx/cells_sim.v"); else - run("read_verilog -lib +/xilinx/cells_sim.v"); + run("read_verilog -lib -D _ABC +/xilinx/cells_sim.v"); run("read_verilog -lib +/xilinx/cells_xtra.v"); @@ -200,6 +227,21 @@ struct SynthXilinxPass : public ScriptPass if (check_label("coarse")) { run("synth -run coarse"); + + //if (!nomux || help_mode) + // run("muxpack", "(skip if '-nomux')"); + + // shregmap -tech xilinx can cope with $shiftx and $mux + // cells for identifying variable-length shift registers, + // so attempt to convert $pmux-es to the former + // Also: wide multiplexer inference benefits from this too + if (!(nosrl && nomux) || help_mode) + run("pmux2shiftx", "(skip if '-nosrl' and '-nomux')"); + + // Run a number of peephole optimisations, including one + // that optimises $mul cells driving $shiftx's B input + // and that aids wide mux analysis + run("peepopt"); } if (check_label("bram", "(skip if '-nobram')")) { @@ -217,12 +259,6 @@ struct SynthXilinxPass : public ScriptPass } if (check_label("fine")) { - // shregmap -tech xilinx can cope with $shiftx and $mux - // cells for identifiying variable-length shift registers, - // so attempt to convert $pmux-es to the former - if (!nosrl || help_mode) - run("pmux2shiftx", "(skip if '-nosrl')"); - run("opt -fast -full"); run("memory_map"); run("dffsr2dff"); @@ -237,25 +273,40 @@ struct SynthXilinxPass : public ScriptPass run("shregmap -tech xilinx -minlen 3", "(skip if '-nosrl')"); } - if (!vpr || help_mode) - run("techmap -map +/techmap.v -map +/xilinx/arith_map.v"); - else - run("techmap -map +/techmap.v +/xilinx/arith_map.v -D _EXPLICIT_CARRY"); - + std::string techmap_files = " -map +/techmap.v"; + if (help_mode) + techmap_files += " [-map +/xilinx/mux_map.v]"; + else if (!nomux) + techmap_files += " -map +/xilinx/mux_map.v"; + if (help_mode) + techmap_files += " [-map +/xilinx/arith_map.v]"; + else if (!nocarry) { + techmap_files += " -map +/xilinx/arith_map.v"; + if (vpr) + techmap_files += " -D _EXPLICIT_CARRY"; + else if (abc == "abc9") + techmap_files += " -D _CLB_CARRY"; + } + run("techmap " + techmap_files); run("opt -fast"); } if (check_label("map_cells")) { + if (!nomux || help_mode) + run("muxcover -mux8 -mux16", "(skip if '-nomux')"); run("techmap -map +/techmap.v -map +/xilinx/cells_map.v"); run("clean"); } if (check_label("map_luts")) { - if (help_mode) - run("abc -luts 2:2,3,6:5,10,20 [-dff]"); + if (abc == "abc9") + run(abc + " -lut +/xilinx/abc_xc7.lut -box +/xilinx/abc_xc7.box -W " + XC7_WIRE_DELAY + string(retime ? " -dff" : "")); + else if (help_mode) + run(abc + " -luts 2:2,3,6:5,10,20 [-dff]"); else - run("abc -luts 2:2,3,6:5,10,20" + string(retime ? " -dff" : "")); + run(abc + " -luts 2:2,3,6:5,10,20" + string(retime ? " -dff" : "")); run("clean"); + // This shregmap call infers fixed length shift registers after abc // has performed any necessary retiming if (!nosrl || help_mode) |