From: David Shah Date: Mon, 8 Jul 2019 14:40:12 +0000 (+0100) Subject: Add mul2dsp multiplier splitting rule and ECP5 mapping X-Git-Tag: working-ls180~1039^2~375 X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=269ff450f55f4354c82db1b98f8eb722317d9250;p=yosys.git Add mul2dsp multiplier splitting rule and ECP5 mapping Signed-off-by: David Shah --- diff --git a/techlibs/common/Makefile.inc b/techlibs/common/Makefile.inc index 0e05620bc..e6d1c2f29 100644 --- a/techlibs/common/Makefile.inc +++ b/techlibs/common/Makefile.inc @@ -28,3 +28,4 @@ $(eval $(call add_share_file,share,techlibs/common/dff2ff.v)) $(eval $(call add_share_file,share,techlibs/common/gate2lut.v)) $(eval $(call add_share_file,share,techlibs/common/cmp2lut.v)) $(eval $(call add_share_file,share,techlibs/common/cells.lib)) +$(eval $(call add_share_file,share,techlibs/common/mul2dsp.v)) diff --git a/techlibs/common/mul2dsp.v b/techlibs/common/mul2dsp.v new file mode 100644 index 000000000..37ce2e485 --- /dev/null +++ b/techlibs/common/mul2dsp.v @@ -0,0 +1,237 @@ +// From Eddie Hung +// extracted from: https://github.com/eddiehung/vtr-with-yosys/blob/vtr7-with-yosys/vtr_flow/misc/yosys_models.v#L220 +// revised by Andre DeHon +// further revised by David Shah +`ifndef DSP_A_MAXWIDTH +`define DSP_A_MAXWIDTH 18 +`endif +`ifndef DSP_A_MAXWIDTH +`define DSP_B_MAXWIDTH 25 +`endif + +`ifndef ADDER_MINWIDTH +`define ADDER_MINWIDTH AAA +`endif + +`ifndef DSP_NAME +`define DSP_NAME M18x25 +`endif + +`define MAX(a,b) (a > b ? a : b) +`define MIN(a,b) (a < b ? a : b) + +(* techmap_celltype = "$mul" *) +module \$mul (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; + + wire [1023:0] _TECHMAP_DO_ = "proc; clean"; + + generate + if (A_WIDTH `DSP_A_MAXWIDTH) begin + localparam n_floored = A_WIDTH/`DSP_A_MAXWIDTH; + localparam n = n_floored + (n_floored*`DSP_A_MAXWIDTH < A_WIDTH ? 1 : 0); + wire [`DSP_A_MAXWIDTH+B_WIDTH-1:0] partial [n-1:1]; + wire [Y_WIDTH-1:0] partial_sum [n-2:0]; + + \$__mul_gen #( + .A_SIGNED(A_SIGNED), + .B_SIGNED(B_SIGNED), + .A_WIDTH(`DSP_A_MAXWIDTH), + .B_WIDTH(B_WIDTH), + .Y_WIDTH(B_WIDTH+`DSP_A_MAXWIDTH) + ) mul_slice_first ( + .A(A[`DSP_A_MAXWIDTH-1:0]), + .B(B), + .Y(partial_sum[0][B_WIDTH+`DSP_A_MAXWIDTH-1:0]) + ); + assign partial_sum[0][Y_WIDTH-1:B_WIDTH+`DSP_A_MAXWIDTH]=0; + + genvar i; + generate + for (i = 1; i < n-1; i=i+1) begin:slice + \$__mul_gen #( + .A_SIGNED(A_SIGNED), + .B_SIGNED(B_SIGNED), + .A_WIDTH(`DSP_A_MAXWIDTH), + .B_WIDTH(B_WIDTH), + .Y_WIDTH(B_WIDTH+`DSP_A_MAXWIDTH) + ) mul_slice ( + .A(A[(i+1)*`DSP_A_MAXWIDTH-1:i*`DSP_A_MAXWIDTH]), + .B(B), + .Y(partial[i][B_WIDTH+`DSP_A_MAXWIDTH-1:0]) + ); + //assign partial_sum[i] = (partial[i] << i*`DSP_A_MAXWIDTH) + partial_sum[i-1]; + assign partial_sum[i] = { + partial[i][B_WIDTH+`DSP_A_MAXWIDTH-1:0] + + partial_sum[i-1][Y_WIDTH-1:(i*`DSP_A_MAXWIDTH)], + partial_sum[i-1][(i*`DSP_A_MAXWIDTH)-1:0] + }; + end + endgenerate + + \$__mul_gen #( + .A_SIGNED(A_SIGNED), + .B_SIGNED(B_SIGNED), + .A_WIDTH(A_WIDTH-(n-1)*`DSP_A_MAXWIDTH), + .B_WIDTH(B_WIDTH), + .Y_WIDTH(A_WIDTH-(n-1)*`DSP_A_MAXWIDTH+B_WIDTH), + ) mul_slice_last ( + .A(A[A_WIDTH-1:(n-1)*`DSP_A_MAXWIDTH]), + .B(B), + .Y(partial[n-1][A_WIDTH-(n-1)*`DSP_A_MAXWIDTH+B_WIDTH-1:0]) + ); + //assign Y = (partial[n-1] << (n-1)*`DSP_A_MAXWIDTH) + partial_sum[n-2]; + assign Y = { + partial[n-1][A_WIDTH-(n-1)*`DSP_A_MAXWIDTH+B_WIDTH:0] + + partial_sum[n-2][Y_WIDTH-1:((n-1)*`DSP_A_MAXWIDTH)], + partial_sum[n-2][((n-1)*`DSP_A_MAXWIDTH)-1:0] + }; + end + else if (B_WIDTH > `DSP_B_MAXWIDTH) begin + localparam n_floored = B_WIDTH/`DSP_B_MAXWIDTH; + localparam n = n_floored + (n_floored*`DSP_B_MAXWIDTH < B_WIDTH ? 1 : 0); + wire [A_WIDTH+`DSP_B_MAXWIDTH-1:0] partial [n-1:1]; + wire [Y_WIDTH-1:0] partial_sum [n-2:0]; + + \$__mul_gen #( + .A_SIGNED(A_SIGNED), + .B_SIGNED(B_SIGNED), + .A_WIDTH(A_WIDTH), + .B_WIDTH(`DSP_B_MAXWIDTH), + .Y_WIDTH(A_WIDTH+`DSP_B_MAXWIDTH) + ) mul_first ( + .A(A), + .B(B[`DSP_B_MAXWIDTH-1:0]), + .Y(partial_sum[0][A_WIDTH+`DSP_B_MAXWIDTH-1:0]) + ); + assign partial_sum[0][Y_WIDTH-1:A_WIDTH+`DSP_B_MAXWIDTH]=0; + + genvar i; + generate + for (i = 1; i < n-1; i=i+1) begin:slice + \$__mul_gen #( + .A_SIGNED(A_SIGNED), + .B_SIGNED(B_SIGNED), + .A_WIDTH(A_WIDTH), + .B_WIDTH(`DSP_B_MAXWIDTH), + .Y_WIDTH(A_WIDTH+`DSP_B_MAXWIDTH) + ) mul ( + .A(A), + .B(B[(i+1)*`DSP_B_MAXWIDTH-1:i*`DSP_B_MAXWIDTH]), + .Y(partial[i][A_WIDTH+`DSP_B_MAXWIDTH-1:0]) + ); + //assign partial_sum[i] = (partial[i] << i*`DSP_B_MAXWIDTH) + partial_sum[i-1]; + // was: + //assign partial_sum[i] = { + // partial[i][A_WIDTH+`DSP_B_MAXWIDTH-1:`DSP_B_MAXWIDTH], + // partial[i][`DSP_B_MAXWIDTH-1:0] + partial_sum[i-1][A_WIDTH+(i*`DSP_B_MAXWIDTH)-1:A_WIDTH+((i-1)*`DSP_B_MAXWIDTH)], + // partial_sum[i-1][A_WIDTH+((i-1)*`DSP_B_MAXWIDTH):0] + assign partial_sum[i] = { + partial[i][A_WIDTH+`DSP_B_MAXWIDTH-1:0] + + partial_sum[i-1][Y_WIDTH-1:(i*`DSP_B_MAXWIDTH)], + partial_sum[i-1][(i*`DSP_B_MAXWIDTH)-1:0] + }; + end + endgenerate + + \$__mul_gen #( + .A_SIGNED(A_SIGNED), + .B_SIGNED(B_SIGNED), + .A_WIDTH(A_WIDTH), + .B_WIDTH(B_WIDTH-(n-1)*`DSP_B_MAXWIDTH), + .Y_WIDTH(A_WIDTH+B_WIDTH-(n-1)*`DSP_B_MAXWIDTH) + ) mul_last ( + .A(A), + .B(B[B_WIDTH-1:(n-1)*`DSP_B_MAXWIDTH]), + .Y(partial[n-1][A_WIDTH+B_WIDTH-(n-1)*`DSP_B_MAXWIDTH-1:0]) + ); + // AMD: this came comment out -- looks closer to right answer + //assign Y = (partial[n-1] << (n-1)*`DSP_B_MAXWIDTH) + partial_sum[n-2]; + // was (looks broken) + //assign Y = { + // partial[n-1][A_WIDTH+`DSP_B_MAXWIDTH-1:`DSP_B_MAXWIDTH], + // partial[n-1][`DSP_B_MAXWIDTH-1:0] + partial_sum[n-2][A_WIDTH+((n-1)*`DSP_B_MAXWIDTH)-1:A_WIDTH+((n-2)*`DSP_B_MAXWIDTH)], + // partial_sum[n-2][A_WIDTH+((n-2)*`DSP_B_MAXWIDTH):0] + assign Y = { + partial[n-1][A_WIDTH+B_WIDTH-(n-1)*`DSP_B_MAXWIDTH-1:0] + + partial_sum[n-2][Y_WIDTH-1:((n-1)*`DSP_B_MAXWIDTH)], + partial_sum[n-2][((n-1)*`DSP_B_MAXWIDTH)-1:0] + }; + end + else begin + wire [A_WIDTH+B_WIDTH-1:0] out; + wire [(`DSP_A_MAXWIDTH+`DSP_B_MAXWIDTH)-(A_WIDTH+B_WIDTH)-1:0] dummy; + wire Asign, Bsign; + assign Asign = (A_SIGNED ? A[A_WIDTH-1] : 1'b0); + assign Bsign = (B_SIGNED ? B[B_WIDTH-1] : 1'b0); + `DSP_NAME _TECHMAP_REPLACE_ ( + .A({ {{`DSP_A_MAXWIDTH-A_WIDTH}{Asign}}, A }), + .B({ {{`DSP_B_MAXWIDTH-B_WIDTH}{Bsign}}, B }), + .OUT({dummy, out}) + ); + if (Y_WIDTH < A_WIDTH+B_WIDTH) + assign Y = out[Y_WIDTH-1:0]; + else begin + wire Ysign = (A_SIGNED || B_SIGNED ? out[A_WIDTH+BWIDTH-1] : 1'b0); + assign Y = { {{Y_WIDTH-(A_WIDTH+B_WIDTH)}{Ysign}}, out[A_WIDTH+B_WIDTH-1:0] }; + end + end + endgenerate +endmodule + + diff --git a/techlibs/ecp5/Makefile.inc b/techlibs/ecp5/Makefile.inc index ff39ba4fe..a2f5cadee 100644 --- a/techlibs/ecp5/Makefile.inc +++ b/techlibs/ecp5/Makefile.inc @@ -10,6 +10,7 @@ $(eval $(call add_share_file,share/ecp5,techlibs/ecp5/brams_map.v)) $(eval $(call add_share_file,share/ecp5,techlibs/ecp5/bram.txt)) $(eval $(call add_share_file,share/ecp5,techlibs/ecp5/arith_map.v)) $(eval $(call add_share_file,share/ecp5,techlibs/ecp5/latches_map.v)) +$(eval $(call add_share_file,share/ecp5,techlibs/ecp5/dsp_map.v)) $(eval $(call add_share_file,share/ecp5,techlibs/ecp5/abc_5g.box)) $(eval $(call add_share_file,share/ecp5,techlibs/ecp5/abc_5g.lut)) diff --git a/techlibs/ecp5/dsp_map.v b/techlibs/ecp5/dsp_map.v new file mode 100644 index 000000000..22e30574c --- /dev/null +++ b/techlibs/ecp5/dsp_map.v @@ -0,0 +1,10 @@ +module \$__MUL18X18 (input [17:0] A, input [17:0] B, output [35:0] OUT); + MULT18X18D mult_i( + .A0(A[0]), .A1(A[1]), .A2(A[2]), .A3(A[3]), .A4(A[4]), .A5(A[5]), .A6(A[6]), .A7(A[7]), .A8(A[8]), .A9(A[9]), .A10(A[10]), .A11(A[11]), .A12(A[12]), .A13(A[13]), .A14(A[14]), .A15(A[15]), .A16(A[16]), .A17(A[17]), + .B0(B[0]), .B1(B[1]), .B2(B[2]), .B3(B[3]), .B4(B[4]), .B5(B[5]), .B6(B[6]), .B7(B[7]), .B8(B[8]), .B9(B[9]), .B10(B[10]), .B11(B[11]), .B12(B[12]), .B13(B[13]), .B14(B[14]), .B15(B[15]), .B16(B[16]), .B17(B[17]), + .C17(1'b0), .C16(1'b0), .C15(1'b0), .C14(1'b0), .C13(1'b0), .C12(1'b0), .C11(1'b0), .C10(1'b0), .C9(1'b0), .C8(1'b0), .C7(1'b0), .C6(1'b0), .C5(1'b0), .C4(1'b0), .C3(1'b0), .C2(1'b0), .C1(1'b0), .C0(1'b0), + .SIGNEDA(1'b0), .SIGNEDB(1'b0), .SOURCEA(1'b0), .SOURCEB(1'b0), + + .P0(OUT[0]), .P1(OUT[1]), .P2(OUT[2]), .P3(OUT[3]), .P4(OUT[4]), .P5(OUT[5]), .P6(OUT[6]), .P7(OUT[7]), .P8(OUT[8]), .P9(OUT[9]), .P10(OUT[10]), .P11(OUT[11]), .P12(OUT[12]), .P13(OUT[13]), .P14(OUT[14]), .P15(OUT[15]), .P16(OUT[16]), .P17(OUT[17]), .P18(OUT[18]), .P19(OUT[19]), .P20(OUT[20]), .P21(OUT[21]), .P22(OUT[22]), .P23(OUT[23]), .P24(OUT[24]), .P25(OUT[25]), .P26(OUT[26]), .P27(OUT[27]), .P28(OUT[28]), .P29(OUT[29]), .P30(OUT[30]), .P31(OUT[31]), .P32(OUT[32]), .P33(OUT[33]), .P34(OUT[34]), .P35(OUT[35]) + ); +endmodule \ No newline at end of file diff --git a/techlibs/ecp5/synth_ecp5.cc b/techlibs/ecp5/synth_ecp5.cc index f16a47f01..3b4185930 100644 --- a/techlibs/ecp5/synth_ecp5.cc +++ b/techlibs/ecp5/synth_ecp5.cc @@ -89,6 +89,9 @@ struct SynthEcp5Pass : 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(" -dsp\n"); + log(" map multipliers to MULT18X18D (EXPERIMENTAL)\n"); + log("\n"); log("\n"); log("The following commands are executed by this synthesis command:\n"); help_script(); @@ -96,7 +99,7 @@ struct SynthEcp5Pass : public ScriptPass } string top_opt, blif_file, edif_file, json_file; - bool noccu2, nodffe, nobram, nodram, nowidelut, flatten, retime, abc2, abc9, vpr; + bool noccu2, nodffe, nobram, nodram, nowidelut, flatten, retime, abc2, abc9, dsp, vpr; void clear_flags() YS_OVERRIDE { @@ -114,6 +117,7 @@ struct SynthEcp5Pass : public ScriptPass abc2 = false; vpr = false; abc9 = false; + dsp = false; } void execute(std::vector args, RTLIL::Design *design) YS_OVERRIDE @@ -192,6 +196,10 @@ struct SynthEcp5Pass : public ScriptPass abc9 = true; continue; } + if (args[argidx] == "-dsp") { + dsp = true; + continue; + } break; } extra_args(args, argidx, design); @@ -225,7 +233,28 @@ struct SynthEcp5Pass : public ScriptPass if (check_label("coarse")) { - run("synth -run coarse"); + run("opt_expr"); + run("opt_clean"); + run("check"); + run("opt"); + run("wreduce"); + run("peepopt"); + run("opt_clean"); + run("share"); + run("techmap -map +/cmp2lut.v -D LUT_WIDTH=4"); + run("opt_expr"); + run("opt_clean"); + if (dsp) { + run("techmap -map +/mul2dsp.v -D DSP_A_MAXWIDTH=18 -D DSP_B_MAXWIDTH=18 -D DSP_NAME=$__MUL18X18"); + run("clean"); + run("techmap -map +/ecp5/dsp_map.v"); + } + run("alumacc"); + run("opt"); + run("fsm"); + run("opt -fast"); + run("memory -nomap"); + run("opt_clean"); } if (!nobram && check_label("bram", "(skip if -nobram)"))