Merge remote-tracking branch 'origin/master' into xaig_dff

4 files changed, 240 insertions, 172 deletions

diff --git a/techlibs/xilinx/cells_sim.v b/techlibs/xilinx/cells_sim.v
index d7dff8975..5d589bd2f 100644
--- a/techlibs/xilinx/cells_sim.v
+++ b/techlibs/xilinx/cells_sim.v
@@ -227,6 +227,14 @@ module MUXCY(output O, input CI, DI, S);
   assign O = S ? CI : DI;
 endmodule
 
+module MUXF5(output O, input I0, I1, S);
+  assign O = S ? I1 : I0;
+endmodule
+
+module MUXF6(output O, input I0, I1, S);
+  assign O = S ? I1 : I0;
+endmodule
+
 (* abc9_box_id = 1, lib_whitebox *)
 module MUXF7(output O, input I0, I1, S);
   assign O = S ? I1 : I0;
@@ -237,6 +245,10 @@ module MUXF8(output O, input I0, I1, S);
   assign O = S ? I1 : I0;
 endmodule
 
+module MUXF9(output O, input I0, I1, S);
+  assign O = S ? I1 : I0;
+endmodule
+
 module XORCY(output O, input CI, LI);
   assign O = CI ^ LI;
 endmodule
@@ -258,6 +270,26 @@ module CARRY4(
   assign CO[3] = S[3] ? CO[2] : DI[3];
 endmodule
 
+module CARRY8(
+  output [7:0] CO,
+  output [7:0] O,
+  input        CI,
+  input        CI_TOP,
+  input  [7:0] DI, S
+);
+  parameter CARRY_TYPE = "SINGLE_CY8";
+  wire CI4 = (CARRY_TYPE == "DUAL_CY4" ? CI_TOP : CO[3]);
+  assign O = S ^ {CO[6:4], CI4, CO[2:0], CI};
+  assign CO[0] = S[0] ? CI : DI[0];
+  assign CO[1] = S[1] ? CO[0] : DI[1];
+  assign CO[2] = S[2] ? CO[1] : DI[2];
+  assign CO[3] = S[3] ? CO[2] : DI[3];
+  assign CO[4] = S[4] ? CI4 : DI[4];
+  assign CO[5] = S[5] ? CO[4] : DI[5];
+  assign CO[6] = S[6] ? CO[5] : DI[6];
+  assign CO[7] = S[7] ? CO[6] : DI[7];
+endmodule
+
 `ifdef _EXPLICIT_CARRY
 
 module CARRY0(output CO_CHAIN, CO_FABRIC, O, input CI, CI_INIT, DI, S);
@@ -281,6 +313,16 @@ endmodule
 
 `endif
 
+module ORCY (output O, input CI, I);
+  assign O = CI | I;
+endmodule
+
+module MULT_AND (output LO, input I0, I1);
+  assign LO = I0 & I1;
+endmodule
+
+// Flip-flops and latches.
+
 // Max delay from: https://github.com/SymbiFlow/prjxray-db/blob/34ea6eb08a63d21ec16264ad37a0a7b142ff6031/artix7/timings/CLBLL_L.sdf#L238-L250
 
 (* abc9_box_id=1100, lib_whitebox, abc9_flop *)
@@ -394,6 +436,51 @@ module FDCE_1 (
   always @(negedge C, posedge CLR) if (CLR) Q <= 1'b0; else if (CE) Q <= D;
 endmodule
 
+module FDCPE (
+  output wire Q,
+  (* clkbuf_sink *)
+  (* invertible_pin = "IS_C_INVERTED" *)
+  input C,
+  input CE,
+  (* invertible_pin = "IS_CLR_INVERTED" *)
+  input CLR,
+  input D,
+  (* invertible_pin = "IS_PRE_INVERTED" *)
+  input PRE
+);
+  parameter [0:0] INIT = 1'b0;
+  parameter [0:0] IS_C_INVERTED = 1'b0;
+  parameter [0:0] IS_CLR_INVERTED = 1'b0;
+  parameter [0:0] IS_PRE_INVERTED = 1'b0;
+  wire c = C ^ IS_C_INVERTED;
+  wire clr = CLR ^ IS_CLR_INVERTED;
+  wire pre = PRE ^ IS_PRE_INVERTED;
+  // Hacky model to avoid simulation-synthesis mismatches.
+  reg qc, qp, qs;
+  initial qc = INIT;
+  initial qp = INIT;
+  initial qs = 0;
+  always @(posedge c, posedge clr) begin
+    if (clr)
+      qc <= 0;
+    else if (CE)
+      qc <= D;
+  end
+  always @(posedge c, posedge pre) begin
+    if (pre)
+      qp <= 1;
+    else if (CE)
+      qp <= D;
+  end
+  always @* begin
+    if (clr)
+      qs <= 0;
+    else if (pre)
+      qs <= 1;
+  end
+  assign Q = qs ? qp : qc;
+endmodule
+
 (* abc9_box_id=1104, lib_whitebox, abc9_flop *)
 module FDPE (
   (* abc9_arrival=303 *)
@@ -488,8 +575,8 @@ module LDCE (
   wire clr = CLR ^ IS_CLR_INVERTED;
   wire g = G ^ IS_G_INVERTED;
   always @*
-    if (clr) Q = 1'b0;
-    else if (GE && g) Q = D;
+    if (clr) Q <= 1'b0;
+    else if (GE && g) Q <= D;
 endmodule
 
 module LDPE (
@@ -510,8 +597,59 @@ module LDPE (
   wire g = G ^ IS_G_INVERTED;
   wire pre = PRE ^ IS_PRE_INVERTED;
   always @*
-    if (pre) Q = 1'b1;
-    else if (GE && g) Q = D;
+    if (pre) Q <= 1'b1;
+    else if (GE && g) Q <= D;
+endmodule
+
+module LDCPE (
+  output reg Q,
+  (* invertible_pin = "IS_CLR_INVERTED" *)
+  input CLR,
+  (* invertible_pin = "IS_D_INVERTED" *)
+  input D,
+  (* invertible_pin = "IS_G_INVERTED" *)
+  input G,
+  (* invertible_pin = "IS_GE_INVERTED" *)
+  input GE,
+  (* invertible_pin = "IS_PRE_INVERTED" *)
+  input PRE
+);
+  parameter [0:0] INIT = 1'b1;
+  parameter [0:0] IS_CLR_INVERTED = 1'b0;
+  parameter [0:0] IS_D_INVERTED = 1'b0;
+  parameter [0:0] IS_G_INVERTED = 1'b0;
+  parameter [0:0] IS_GE_INVERTED = 1'b0;
+  parameter [0:0] IS_PRE_INVERTED = 1'b0;
+  initial Q = INIT;
+  wire d = D ^ IS_D_INVERTED;
+  wire g = G ^ IS_G_INVERTED;
+  wire ge = GE ^ IS_GE_INVERTED;
+  wire clr = CLR ^ IS_CLR_INVERTED;
+  wire pre = PRE ^ IS_PRE_INVERTED;
+  always @*
+    if (clr) Q <= 1'b0;
+    else if (pre) Q <= 1'b1;
+    else if (ge && g) Q <= d;
+endmodule
+
+module AND2B1L (
+  output O,
+  input DI,
+  (* invertible_pin = "IS_SRI_INVERTED" *)
+  input SRI
+);
+  parameter [0:0] IS_SRI_INVERTED = 1'b0;
+  assign O = DI & ~(SRI ^ IS_SRI_INVERTED);
+endmodule
+
+module OR2L (
+  output O,
+  input DI,
+  (* invertible_pin = "IS_SRI_INVERTED" *)
+  input SRI
+);
+  parameter [0:0] IS_SRI_INVERTED = 1'b0;
+  assign O = DI | (SRI ^ IS_SRI_INVERTED);
 endmodule
 
 // LUTRAM.
@@ -1377,6 +1515,20 @@ endmodule
 
 // Shift registers.
 
+module SRL16 (
+  output Q,
+  input A0, A1, A2, A3,
+  (* clkbuf_sink *)
+  input CLK,
+  input D
+);
+  parameter [15:0] INIT = 16'h0000;
+
+  reg [15:0] r = INIT;
+  assign Q = r[{A3,A2,A1,A0}];
+  always @(posedge CLK) r <= { r[14:0], D };
+endmodule
+
 module SRL16E (
   // Max delay from: https://github.com/SymbiFlow/prjxray-db/blob/34ea6eb08a63d21ec16264ad37a0a7b142ff6031/artix7/timings/CLBLM_R.sdf#L904-L905
   (* abc9_arrival=1472 *)
@@ -1401,6 +1553,22 @@ module SRL16E (
   endgenerate
 endmodule
 
+module SRLC16 (
+  output Q,
+  output Q15,
+  input A0, A1, A2, A3,
+  (* clkbuf_sink *)
+  input CLK,
+  input D
+);
+  parameter [15:0] INIT = 16'h0000;
+
+  reg [15:0] r = INIT;
+  assign Q15 = r[15];
+  assign Q = r[{A3,A2,A1,A0}];
+  always @(posedge CLK) r <= { r[14:0], D };
+endmodule
+
 module SRLC16E (
   output Q,
   output Q15,
@@ -1453,6 +1621,31 @@ module SRLC32E (
   endgenerate
 endmodule
 
+module CFGLUT5 (
+  output CDO,
+  output O5,
+  output O6,
+  input I4,
+  input I3,
+  input I2,
+  input I1,
+  input I0,
+  input CDI,
+  input CE,
+  (* clkbuf_sink *)
+  (* invertible_pin = "IS_CLK_INVERTED" *)
+  input CLK
+);
+  parameter [31:0] INIT = 32'h00000000;
+  parameter [0:0] IS_CLK_INVERTED = 1'b0;
+  wire clk = CLK ^ IS_CLK_INVERTED;
+  reg [31:0] r = INIT;
+  assign CDO = r[31];
+  assign O5 = r[{1'b0, I3, I2, I1, I0}];
+  assign O6 = r[{I4, I3, I2, I1, I0}];
+  always @(posedge clk) if (CE) r <= {r[30:0], CDI};
+endmodule
+
 // DSP
 
 // Virtex 2, Virtex 2 Pro, Spartan 3.
diff --git a/techlibs/xilinx/cells_xtra.py b/techlibs/xilinx/cells_xtra.py
index 6d5adf1aa..d5c58c5d7 100644
--- a/techlibs/xilinx/cells_xtra.py
+++ b/techlibs/xilinx/cells_xtra.py
@@ -65,9 +65,9 @@ CELLS = [
 
     # CLB -- registers/latches.
     # Virtex 1/2/4/5, Spartan 3.
-    Cell('FDCPE', port_attrs={'C': ['clkbuf_sink']}),
+    # Cell('FDCPE', port_attrs={'C': ['clkbuf_sink']}),
     # Cell('FDRSE', port_attrs={'C': ['clkbuf_sink']}),
-    Cell('LDCPE', port_attrs={'C': ['clkbuf_sink']}),
+    # Cell('LDCPE', port_attrs={'C': ['clkbuf_sink']}),
     # Virtex 6, Spartan 6, Series 7, Ultrascale.
     # Cell('FDCE'),
     # Cell('FDPE'),
@@ -75,8 +75,8 @@ CELLS = [
     # Cell('FDSE'),
     # Cell('LDCE'),
     # Cell('LDPE'),
-    Cell('AND2B1L'),
-    Cell('OR2L'),
+    # Cell('AND2B1L'),
+    # Cell('OR2L'),
 
     # CLB -- other.
     # Cell('LUT1'),
@@ -86,23 +86,23 @@ CELLS = [
     # Cell('LUT5'),
     # Cell('LUT6'),
     # Cell('LUT6_2'),
-    Cell('MUXF5'),
-    Cell('MUXF6'),
+    # Cell('MUXF5'),
+    # Cell('MUXF6'),
     # Cell('MUXF7'),
     # Cell('MUXF8'),
-    Cell('MUXF9'),
+    # Cell('MUXF9'),
     # Cell('CARRY4'),
-    Cell('CARRY8'),
+    # Cell('CARRY8'),
     # Cell('MUXCY'),
     # Cell('XORCY'),
-    Cell('ORCY'),
-    Cell('MULT_AND'),
-    Cell('SRL16', port_attrs={'CLK': ['clkbuf_sink']}),
+    # Cell('ORCY'),
+    # Cell('MULT_AND'),
+    # Cell('SRL16', port_attrs={'CLK': ['clkbuf_sink']}),
     # Cell('SRL16E', port_attrs={'CLK': ['clkbuf_sink']}),
-    Cell('SRLC16', port_attrs={'CLK': ['clkbuf_sink']}),
+    # Cell('SRLC16', port_attrs={'CLK': ['clkbuf_sink']}),
     # Cell('SRLC16E', port_attrs={'CLK': ['clkbuf_sink']}),
     # Cell('SRLC32E', port_attrs={'CLK': ['clkbuf_sink']}),
-    Cell('CFGLUT5', port_attrs={'CLK': ['clkbuf_sink']}),
+    # Cell('CFGLUT5', port_attrs={'CLK': ['clkbuf_sink']}),
 
     # Block RAM.
     # Virtex.
diff --git a/techlibs/xilinx/cells_xtra.v b/techlibs/xilinx/cells_xtra.v
index 66b7c583f..c3e5c72f9 100644
--- a/techlibs/xilinx/cells_xtra.v
+++ b/techlibs/xilinx/cells_xtra.v
@@ -1,144 +1,5 @@
 // Created by cells_xtra.py from Xilinx models
 
-module FDCPE (...);
-    parameter [0:0] INIT = 1'b0;
-    parameter [0:0] IS_C_INVERTED = 1'b0;
-    parameter [0:0] IS_CLR_INVERTED = 1'b0;
-    parameter [0:0] IS_PRE_INVERTED = 1'b0;
-    output Q;
-    (* clkbuf_sink *)
-    (* invertible_pin = "IS_C_INVERTED" *)
-    input C;
-    input CE;
-    (* invertible_pin = "IS_CLR_INVERTED" *)
-    input CLR;
-    input D;
-    (* invertible_pin = "IS_PRE_INVERTED" *)
-    input PRE;
-endmodule
-
-module LDCPE (...);
-    parameter [0:0] INIT = 1'b0;
-    parameter [0:0] IS_CLR_INVERTED = 1'b0;
-    parameter [0:0] IS_D_INVERTED = 1'b0;
-    parameter [0:0] IS_G_INVERTED = 1'b0;
-    parameter [0:0] IS_GE_INVERTED = 1'b0;
-    parameter [0:0] IS_PRE_INVERTED = 1'b0;
-    output Q;
-    (* invertible_pin = "IS_CLR_INVERTED" *)
-    input CLR;
-    (* invertible_pin = "IS_D_INVERTED" *)
-    input D;
-    (* invertible_pin = "IS_G_INVERTED" *)
-    input G;
-    (* invertible_pin = "IS_GE_INVERTED" *)
-    input GE;
-    (* invertible_pin = "IS_PRE_INVERTED" *)
-    input PRE;
-endmodule
-
-module AND2B1L (...);
-    parameter [0:0] IS_SRI_INVERTED = 1'b0;
-    output O;
-    input DI;
-    (* invertible_pin = "IS_SRI_INVERTED" *)
-    input SRI;
-endmodule
-
-module OR2L (...);
-    parameter [0:0] IS_SRI_INVERTED = 1'b0;
-    output O;
-    input DI;
-    (* invertible_pin = "IS_SRI_INVERTED" *)
-    input SRI;
-endmodule
-
-module MUXF5 (...);
-    output O;
-    input I0;
-    input I1;
-    input S;
-endmodule
-
-module MUXF6 (...);
-    output O;
-    input I0;
-    input I1;
-    input S;
-endmodule
-
-module MUXF9 (...);
-    output O;
-    input I0;
-    input I1;
-    input S;
-endmodule
-
-module CARRY8 (...);
-    parameter CARRY_TYPE = "SINGLE_CY8";
-    output [7:0] CO;
-    output [7:0] O;
-    input CI;
-    input CI_TOP;
-    input [7:0] DI;
-    input [7:0] S;
-endmodule
-
-module ORCY (...);
-    output O;
-    input CI;
-    input I;
-endmodule
-
-module MULT_AND (...);
-    output LO;
-    input I0;
-    input I1;
-endmodule
-
-module SRL16 (...);
-    parameter [15:0] INIT = 16'h0000;
-    output Q;
-    input A0;
-    input A1;
-    input A2;
-    input A3;
-    (* clkbuf_sink *)
-    input CLK;
-    input D;
-endmodule
-
-module SRLC16 (...);
-    parameter [15:0] INIT = 16'h0000;
-    output Q;
-    output Q15;
-    input A0;
-    input A1;
-    input A2;
-    input A3;
-    (* clkbuf_sink *)
-    input CLK;
-    input D;
-endmodule
-
-module CFGLUT5 (...);
-    parameter [31:0] INIT = 32'h00000000;
-    parameter [0:0] IS_CLK_INVERTED = 1'b0;
-    output CDO;
-    output O5;
-    output O6;
-    input I4;
-    input I3;
-    input I2;
-    input I1;
-    input I0;
-    input CDI;
-    input CE;
-    (* clkbuf_sink *)
-    (* invertible_pin = "IS_CLK_INVERTED" *)
-    input CLK;
-endmodule
-
 module RAMB16_S1 (...);
     parameter [0:0] INIT = 1'h0;
     parameter [0:0] SRVAL = 1'h0;
diff --git a/techlibs/xilinx/xilinx_dffopt.cc b/techlibs/xilinx/xilinx_dffopt.cc
index 1256a08cb..13a0b9b83 100644
--- a/techlibs/xilinx/xilinx_dffopt.cc
+++ b/techlibs/xilinx/xilinx_dffopt.cc
@@ -27,20 +27,13 @@ typedef std::pair<Const, std::vector<SigBit>> LutData;
 
 // Compute a LUT implementing (select ^ select_inv) ? alt_data : data.  Returns true if successful.
 bool merge_lut(LutData &result, const LutData &data, const LutData select, bool select_inv, SigBit alt_data, int max_lut_size) {
-	// First, gather input signals.
+	// First, gather input signals -- insert new signals at the beginning
+	// of the vector, so they don't disturb the likely-critical D LUT input
+	// timings.
 	result.second = data.second;
-	int idx_alt = -1;
-	if (alt_data.wire) {
-		// Check if we already have it.
-		for (int i = 0; i < GetSize(result.second); i++)
-			if (result.second[i] == alt_data)
-				idx_alt = i;
-		// If not, add it.
-		if (idx_alt == -1) {
-			idx_alt = GetSize(result.second);
-			result.second.push_back(alt_data);
-		}
-	}
+	// D lut inputs initially start at 0.
+	int idx_data = 0;
+	// Now add the control input LUT inputs.
 	std::vector<int> idx_sel;
 	for (auto bit : select.second) {
 		int idx = -1;
@@ -48,11 +41,32 @@ bool merge_lut(LutData &result, const LutData &data, const LutData select, bool
 			if (result.second[i] == bit)
 				idx = i;
 		if (idx == -1) {
-			idx = GetSize(result.second);
-			result.second.push_back(bit);
+			idx = 0;
+			// Insert new signal at the beginning and bump all indices.
+			result.second.insert(result.second.begin(), bit);
+			idx_data++;
+			for (int &sidx : idx_sel)
+				sidx++;
 		}
 		idx_sel.push_back(idx);
 	}
+	// Insert the Q signal, if any, to the slowest input -- it will have
+	// no problem meeting timing.
+	int idx_alt = -1;
+	if (alt_data.wire) {
+		// Check if we already have it.
+		for (int i = 0; i < GetSize(result.second); i++)
+			if (result.second[i] == alt_data)
+				idx_alt = i;
+		// If not, add it.
+		if (idx_alt == -1) {
+			idx_alt = 0;
+			result.second.insert(result.second.begin(), alt_data);
+			idx_data++;
+			for (int &sidx : idx_sel)
+				sidx++;
+		}
+	}
 
 	// If LUT would be too large, bail.
 	if (GetSize(result.second) > max_lut_size)
@@ -75,7 +89,7 @@ bool merge_lut(LutData &result, const LutData &data, const LutData select, bool
 				new_bit = alt_data.data == State::S1;
 		} else {
 			// Use original LUT.
-			int lut_idx = i & ((1 << GetSize(data.second)) - 1);
+			int lut_idx = i >> idx_data & ((1 << GetSize(data.second)) - 1);
 			new_bit = data.first.bits[lut_idx] == State::S1;
 		}
 		result.first.bits[i] = new_bit ? State::S1 : State::S0;