techmap/shift_shiftx: Remove the "shiftx2mux" special path.

Our techmap rules for $shift and $shiftx cells contained a special path
that aimed to decompose the shift LSB-first instead of MSB-first in
select cases that come up in pmux lowering.  This path was needlessly
overcomplicated and contained bugs.

Instead of doing that, just switch over the main path to iterate
LSB-first (except for the specially-handled MSB for signed shifts
and overflow handling).  This also makes the code consistent with
shl/shr/sshl/sshr cells, which are already decomposed LSB-first.

Fixes #2346.

3 files changed, 38 insertions, 81 deletions

diff --git a/techlibs/common/techmap.v b/techlibs/common/techmap.v
index 03c27d49d..2ab28e6e6 100644
--- a/techlibs/common/techmap.v
+++ b/techlibs/common/techmap.v
@@ -143,78 +143,46 @@ module _90_shift_shiftx (A, B, Y);
 	localparam extbit = _TECHMAP_CELLTYPE_ == "$shift" ? 1'b0 : 1'bx;
 	wire a_padding = _TECHMAP_CELLTYPE_ == "$shiftx" ? extbit : (A_SIGNED ? A[A_WIDTH-1] : 1'b0);
 
-	generate
-`ifndef NO_LSB_FIRST_SHIFT_SHIFTX
-		// If $shift/$shiftx only shifts in units of Y_WIDTH
-		//   (a common pattern created by pmux2shiftx)
-		//   which is checked by ensuring that all that
-		//   the appropriate LSBs of B are constant zero,
-		//   then we can decompose LSB first instead of
-		//   MSB first
-		localparam CLOG2_Y_WIDTH = $clog2(Y_WIDTH);
-		if (B_WIDTH > CLOG2_Y_WIDTH+1 &&
-			_TECHMAP_CONSTMSK_B_[CLOG2_Y_WIDTH-1:0] == {CLOG2_Y_WIDTH{1'b1}} &&
-			_TECHMAP_CONSTVAL_B_[CLOG2_Y_WIDTH-1:0] == {CLOG2_Y_WIDTH{1'b0}}) begin
-			// Halve the size of $shift/$shiftx by $mux-ing A according to
-			//   the LSB of B, after discarding the zeroed bits
-			localparam Y_WIDTH2 = 2**CLOG2_Y_WIDTH;
-			localparam entries = (A_WIDTH+Y_WIDTH-1)/Y_WIDTH2;
-			localparam len = Y_WIDTH2 * ((entries+1)/2);
-			wire [len-1:0] AA;
-			wire [(A_WIDTH+Y_WIDTH2+Y_WIDTH-1)-1:0] Apad = {{(Y_WIDTH2+Y_WIDTH-1){a_padding}}, A};
-			genvar i;
-			for (i = 0; i < A_WIDTH; i=i+Y_WIDTH2*2)
-				assign AA[i/2 +: Y_WIDTH2] = B[CLOG2_Y_WIDTH] ? Apad[i+Y_WIDTH2 +: Y_WIDTH2] : Apad[i +: Y_WIDTH2];
-			wire [B_WIDTH-2:0] BB = {B[B_WIDTH-1:CLOG2_Y_WIDTH+1], {CLOG2_Y_WIDTH{1'b0}}};
-			if (_TECHMAP_CELLTYPE_ == "$shift")
-				$shift #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(len), .B_WIDTH(B_WIDTH-1), .Y_WIDTH(Y_WIDTH)) _TECHMAP_REPLACE_ (.A(AA), .B(BB), .Y(Y));
-			else
-				$shiftx #(.A_SIGNED(A_SIGNED), .B_SIGNED(B_SIGNED), .A_WIDTH(len), .B_WIDTH(B_WIDTH-1), .Y_WIDTH(Y_WIDTH)) _TECHMAP_REPLACE_ (.A(AA), .B(BB), .Y(Y));
-		end
-		else
-`endif
-		begin
-			localparam BB_WIDTH = `MIN($clog2(`MAX(A_WIDTH, Y_WIDTH)) + (B_SIGNED ? 2 : 1), B_WIDTH);
-			localparam WIDTH = `MAX(A_WIDTH, Y_WIDTH) + (B_SIGNED ? 2**(BB_WIDTH-1) : 0);
+	localparam BB_WIDTH = `MIN($clog2(`MAX(A_WIDTH, Y_WIDTH)) + (B_SIGNED ? 2 : 1), B_WIDTH);
+	localparam WIDTH = `MAX(A_WIDTH, Y_WIDTH) + (B_SIGNED ? 2**(BB_WIDTH-1) : 0);
 
-			wire [1023:0] _TECHMAP_DO_00_ = "proc;;";
-			wire [1023:0] _TECHMAP_DO_01_ = "CONSTMAP; opt_muxtree; opt_expr -mux_undef -mux_bool -fine;;;";
+	wire [1023:0] _TECHMAP_DO_00_ = "proc;;";
+	wire [1023:0] _TECHMAP_DO_01_ = "CONSTMAP; opt_muxtree; opt_expr -mux_undef -mux_bool -fine;;;";
 
-			integer i;
-			(* force_downto *)
-			reg [WIDTH-1:0] buffer;
-			reg overflow;
+	integer i;
+	(* force_downto *)
+	reg [WIDTH-1:0] buffer;
+	reg overflow;
 
-			always @* begin
-				overflow = 0;
+	always @* begin
+		overflow = 0;
+		buffer = {WIDTH{extbit}};
+		buffer[Y_WIDTH-1:0] = {Y_WIDTH{a_padding}};
+		buffer[A_WIDTH-1:0] = A;
+
+		if (B_WIDTH > BB_WIDTH) begin
+			if (B_SIGNED) begin
+				for (i = BB_WIDTH; i < B_WIDTH; i = i+1)
+					if (B[i] != B[BB_WIDTH-1])
+						overflow = 1;
+			end else
+				overflow = |B[B_WIDTH-1:BB_WIDTH];
+			if (overflow)
 				buffer = {WIDTH{extbit}};
-				buffer[Y_WIDTH-1:0] = {Y_WIDTH{a_padding}};
-				buffer[A_WIDTH-1:0] = A;
-
-				if (B_WIDTH > BB_WIDTH) begin
-					if (B_SIGNED) begin
-						for (i = BB_WIDTH; i < B_WIDTH; i = i+1)
-							if (B[i] != B[BB_WIDTH-1])
-								overflow = 1;
-					end else
-						overflow = |B[B_WIDTH-1:BB_WIDTH];
-					if (overflow)
-						buffer = {WIDTH{extbit}};
-				end
-
-				for (i = BB_WIDTH-1; i >= 0; i = i-1)
-					if (B[i]) begin
-						if (B_SIGNED && i == BB_WIDTH-1)
-							buffer = {buffer, {2**i{extbit}}};
-						else if (2**i < WIDTH)
-							buffer = {{2**i{extbit}}, buffer[WIDTH-1 : 2**i]};
-						else
-							buffer = {WIDTH{extbit}};
-					end
-			end
-			assign Y = buffer;
 		end
-	endgenerate
+
+		if (B_SIGNED && B[BB_WIDTH-1])
+			buffer = {buffer, {2**(BB_WIDTH-1){extbit}}};
+
+		for (i = 0; i < (B_SIGNED ? BB_WIDTH-1 : BB_WIDTH); i = i+1)
+			if (B[i]) begin
+				if (2**i < WIDTH)
+					buffer = {{2**i{extbit}}, buffer[WIDTH-1 : 2**i]};
+				else
+					buffer = {WIDTH{extbit}};
+			end
+	end
+	assign Y = buffer;
 endmodule
 
 
diff --git a/tests/arch/intel_alm/mux.ys b/tests/arch/intel_alm/mux.ys
index 01cc78e1b..ac3b9b08f 100644
--- a/tests/arch/intel_alm/mux.ys
+++ b/tests/arch/intel_alm/mux.ys
@@ -70,8 +70,9 @@ equiv_opt -assert -map +/intel_alm/common/alm_sim.v synth_intel_alm -family cycl
 design -load postopt # load the post-opt design (otherwise equiv_opt loads the pre-opt design)
 cd mux16 # Constrain all select calls below inside the top module
 select -assert-count 1 t:MISTRAL_ALUT3
-select -assert-count 5 t:MISTRAL_ALUT6
-select -assert-none t:MISTRAL_ALUT3 t:MISTRAL_ALUT6 %% t:* %D
+select -assert-max 2 t:MISTRAL_ALUT5
+select -assert-max 5 t:MISTRAL_ALUT6
+select -assert-none t:MISTRAL_ALUT3 t:MISTRAL_ALUT5 t:MISTRAL_ALUT6 %% t:* %D
 
 
 design -load read
diff --git a/tests/techmap/shiftx2mux.ys b/tests/techmap/shiftx2mux.ys
index eb29680f6..f749e79b2 100644
--- a/tests/techmap/shiftx2mux.ys
+++ b/tests/techmap/shiftx2mux.ys
@@ -74,12 +74,6 @@ design -save gold
 
 
 design -load gold
-techmap -D NO_LSB_FIRST_SHIFT_SHIFTX
-abc -lut 6
-select -assert-min 17 t:$lut
-
-
-design -load gold
 techmap
 abc -lut 6
 select -assert-count 16 t:$lut
@@ -92,12 +86,6 @@ sat -verify -prove-asserts -show-ports miter
 
 
 design -load gold
-techmap -D NO_LSB_FIRST_SHIFT_SHIFTX
-abc9 -lut 6
-select -assert-min 17 t:$lut
-
-
-design -load gold
 techmap
 abc9 -lut 6
 select -assert-count 16 t:$lut