mul2dsp: Lower partial products always have unsigned inputs

Signed-off-by: David Shah <dave@ds0.me>

1 files changed, 41 insertions, 31 deletions

diff --git a/techlibs/common/mul2dsp.v b/techlibs/common/mul2dsp.v
index 7344bc5fe..31469ddeb 100644
--- a/techlibs/common/mul2dsp.v
+++ b/techlibs/common/mul2dsp.v
@@ -92,20 +92,25 @@ module \$__mul_gen (A, B, Y);
 	genvar i;

 	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);

+`ifdef DSP_A_SIGNEDONLY

+			localparam sign_headroom = 1;

+`else 	

+			localparam sign_headroom = 0;

+`endif

+			localparam n_floored = A_WIDTH/(`DSP_A_MAXWIDTH - sign_headroom);

+			localparam n = n_floored + (n_floored*(`DSP_A_MAXWIDTH - sign_headroom) < 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];

 			localparam int_yw = `MIN(Y_WIDTH, B_WIDTH+`DSP_A_MAXWIDTH);

 

 			\$__mul_gen #(

-				.A_SIGNED(A_SIGNED),

+				.A_SIGNED(0),

 				.B_SIGNED(B_SIGNED),

 				.A_WIDTH(`DSP_A_MAXWIDTH),

 				.B_WIDTH(B_WIDTH),

 				.Y_WIDTH(int_yw)

 			) mul_slice_first (

-				.A(A[`DSP_A_MAXWIDTH-1:0]),

+				.A({{sign_headroom{1'b0}}, A[`DSP_A_MAXWIDTH-sign_headroom-1:0]}),

 				.B(B),

 				.Y(partial_sum[0][int_yw-1:0])

 			);

@@ -114,73 +119,78 @@ module \$__mul_gen (A, B, Y);
 

 			for (i = 1; i < n-1; i=i+1) begin:slice

 				\$__mul_gen #(

-					.A_SIGNED(A_SIGNED),

+					.A_SIGNED(0),

 					.B_SIGNED(B_SIGNED),

 					.A_WIDTH(`DSP_A_MAXWIDTH),

 					.B_WIDTH(B_WIDTH),

 					.Y_WIDTH(int_yw)

 				) mul_slice (

-					.A(A[(i+1)*`DSP_A_MAXWIDTH-1:i*`DSP_A_MAXWIDTH]),

+					.A({{sign_headroom{1'b0}}, A[(i+1)*(`DSP_A_MAXWIDTH-sign_headroom)-1:i*(`DSP_A_MAXWIDTH-sign_headroom)]}),

 					.B(B),

 					.Y(partial[i][int_yw-1:0])

 				);

 				//assign partial_sum[i] = (partial[i] << i*`DSP_A_MAXWIDTH) + partial_sum[i-1];

 				assign partial_sum[i] = {

 					partial[i][int_yw-1:0]

-					+ partial_sum[i-1][Y_WIDTH-1:(i*`DSP_A_MAXWIDTH)],

-					partial_sum[i-1][(i*`DSP_A_MAXWIDTH)-1:0]

+					+ partial_sum[i-1][Y_WIDTH-1:(i*(`DSP_A_MAXWIDTH-sign_headroom))],

+					partial_sum[i-1][(i*(`DSP_A_MAXWIDTH-sign_headroom))-1:0]

 				};

 			end

 

 			\$__mul_gen #(

 				.A_SIGNED(A_SIGNED),

 				.B_SIGNED(B_SIGNED),

-				.A_WIDTH(A_WIDTH-(n-1)*`DSP_A_MAXWIDTH),

+				.A_WIDTH(A_WIDTH-(n-1)*(`DSP_A_MAXWIDTH-sign_headroom)),

 				.B_WIDTH(B_WIDTH),

-				.Y_WIDTH(`MIN(Y_WIDTH, A_WIDTH-(n-1)*`DSP_A_MAXWIDTH+B_WIDTH)),

+				.Y_WIDTH(`MIN(Y_WIDTH, A_WIDTH-(n-1)*(`DSP_A_MAXWIDTH-sign_headroom)+B_WIDTH)),

 			) mul_slice_last (

-				.A(A[A_WIDTH-1:(n-1)*`DSP_A_MAXWIDTH]),

+				.A(A[A_WIDTH-1:(n-1)*(`DSP_A_MAXWIDTH-sign_headroom)]),

 				.B(B),

-				.Y(partial[n-1][`MIN(Y_WIDTH, A_WIDTH-(n-1)*`DSP_A_MAXWIDTH+B_WIDTH)-1:0])

+				.Y(partial[n-1][`MIN(Y_WIDTH, A_WIDTH-(n-1)*(`DSP_A_MAXWIDTH-sign_headroom)+B_WIDTH)-1:0])

 			);

 			//assign Y = (partial[n-1] << (n-1)*`DSP_A_MAXWIDTH) + partial_sum[n-2];

 			assign Y = {

-				partial[n-1][`MIN(Y_WIDTH, 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]

+				partial[n-1][`MIN(Y_WIDTH, A_WIDTH-(n-1)*(`DSP_A_MAXWIDTH-sign_headroom)+B_WIDTH):0]

+				+ partial_sum[n-2][Y_WIDTH-1:((n-1)*(`DSP_A_MAXWIDTH-sign_headroom))],

+				partial_sum[n-2][((n-1)*(`DSP_A_MAXWIDTH-sign_headroom))-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);

+`ifdef DSP_B_SIGNEDONLY

+			localparam sign_headroom = 1;

+`else 	

+			localparam sign_headroom = 0;

+`endif

+			localparam n_floored = B_WIDTH/(`DSP_B_MAXWIDTH - sign_headroom);

+			localparam n = n_floored + (n_floored*(`DSP_B_MAXWIDTH - sign_headroom) < 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];

 			localparam int_yw = `MIN(Y_WIDTH, A_WIDTH+`DSP_B_MAXWIDTH);

 

 			\$__mul_gen #(

 				.A_SIGNED(A_SIGNED),

-				.B_SIGNED(B_SIGNED),

+				.B_SIGNED(0),

 				.A_WIDTH(A_WIDTH),

 				.B_WIDTH(`DSP_B_MAXWIDTH),

 				.Y_WIDTH(int_yw)

 			) mul_first (

 				.A(A),

-				.B(B[`DSP_B_MAXWIDTH-1:0]),

+				.B({{sign_headroom{1'b0}}, B[(`DSP_B_MAXWIDTH - sign_headroom)-1:0]}),

 				.Y(partial_sum[0][int_yw-1:0])

 			);

 			if (Y_WIDTH > int_yw)

-				assign partial_sum[0][Y_WIDTH-1:A_WIDTH+`DSP_B_MAXWIDTH]=0;

+				assign partial_sum[0][Y_WIDTH-1:int_yw]=0;

 

 			for (i = 1; i < n-1; i=i+1) begin:slice

 				\$__mul_gen #(

 					.A_SIGNED(A_SIGNED),

-					.B_SIGNED(B_SIGNED),

+					.B_SIGNED(0),

 					.A_WIDTH(A_WIDTH),

 					.B_WIDTH(`DSP_B_MAXWIDTH),

 					.Y_WIDTH(int_yw)

 				) mul (

 					.A(A),

-					.B(B[(i+1)*`DSP_B_MAXWIDTH-1:i*`DSP_B_MAXWIDTH]),

+					.B({{sign_headroom{1'b0}}, B[(i+1)*(`DSP_B_MAXWIDTH - sign_headroom)-1:i*(`DSP_B_MAXWIDTH - sign_headroom)]}),

 					.Y(partial[i][int_yw-1:0])

 				);

 				//assign partial_sum[i] = (partial[i] << i*`DSP_B_MAXWIDTH) + partial_sum[i-1];

@@ -191,8 +201,8 @@ module \$__mul_gen (A, B, Y);
 				//	partial_sum[i-1][A_WIDTH+((i-1)*`DSP_B_MAXWIDTH):0]

 				assign partial_sum[i] = {

 					partial[i][int_yw-1:0]

-					+ partial_sum[i-1][Y_WIDTH-1:(i*`DSP_B_MAXWIDTH)],

-					partial_sum[i-1][(i*`DSP_B_MAXWIDTH)-1:0] 

+					+ partial_sum[i-1][Y_WIDTH-1:(i*(`DSP_B_MAXWIDTH - sign_headroom))],

+					partial_sum[i-1][(i*(`DSP_B_MAXWIDTH - sign_headroom))-1:0] 

 				};

 			end

 

@@ -200,12 +210,12 @@ module \$__mul_gen (A, B, Y);
 				.A_SIGNED(A_SIGNED),

 				.B_SIGNED(B_SIGNED),

 				.A_WIDTH(A_WIDTH),

-				.B_WIDTH(B_WIDTH-(n-1)*`DSP_B_MAXWIDTH),

-				.Y_WIDTH(`MIN(Y_WIDTH, A_WIDTH+B_WIDTH-(n-1)*`DSP_B_MAXWIDTH))

+				.B_WIDTH(B_WIDTH-(n-1)*(`DSP_B_MAXWIDTH - sign_headroom)),

+				.Y_WIDTH(`MIN(Y_WIDTH, A_WIDTH+B_WIDTH-(n-1)*(`DSP_B_MAXWIDTH - sign_headroom)))

 			) mul_last (

 				.A(A),

-				.B(B[B_WIDTH-1:(n-1)*`DSP_B_MAXWIDTH]),

-				.Y(partial[n-1][`MIN(Y_WIDTH, A_WIDTH+B_WIDTH-(n-1)*`DSP_B_MAXWIDTH)-1:0])

+				.B(B[B_WIDTH-1:(n-1)*(`DSP_B_MAXWIDTH - sign_headroom)]),

+				.Y(partial[n-1][`MIN(Y_WIDTH, A_WIDTH+B_WIDTH-(n-1)*(`DSP_B_MAXWIDTH - sign_headroom))-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];

@@ -215,9 +225,9 @@ module \$__mul_gen (A, B, Y);
 			//	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][`MIN(Y_WIDTH, 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]

+				partial[n-1][`MIN(Y_WIDTH, A_WIDTH+B_WIDTH-(n-1)*(`DSP_B_MAXWIDTH - sign_headroom))-1:0]

+				+ partial_sum[n-2][Y_WIDTH-1:((n-1)*(`DSP_B_MAXWIDTH - sign_headroom))],

+				partial_sum[n-2][((n-1)*(`DSP_B_MAXWIDTH - sign_headroom))-1:0]

 			};

 		end

 		else begin