[Cryptech-Commits] [core/math/modexpa7] 01/04: Started porting generic multiplier to Xilinx primitives.

[git at cryptech.is]

This is an automated email from the git hooks/post-receive script.

meisterpaul1 at yandex.ru pushed a commit to branch systolic
in repository core/math/modexpa7.

commit caea5e361940c0d445a7d90446641a637c6d49cc
Author: Pavel V. Shatov (Meister) <meisterpaul1 at yandex.ru>
AuthorDate: Sat Jul 1 22:26:32 2017 +0300

    Started porting generic multiplier to Xilinx primitives.
---
 src/rtl/modexpa7_systolic_multiplier.v             |  6 +-
 src/rtl/pe/artix7/multiplier32_artix7.v            | 83 ++++++++++++++++++++++
 .../multiplier32_generic.v}                        | 19 ++++-
 src/rtl/pe/modexpa7_pe_mul.v                       | 33 +++++----
 src/rtl/pe/modexpa7_pe_settings.v                  |  2 +
 5 files changed, 124 insertions(+), 19 deletions(-)

diff --git a/src/rtl/modexpa7_systolic_multiplier.v b/src/rtl/modexpa7_systolic_multiplier.v
index 0849b61..fa99c54 100644
--- a/src/rtl/modexpa7_systolic_multiplier.v
+++ b/src/rtl/modexpa7_systolic_multiplier.v
@@ -400,7 +400,7 @@ module modexpa7_systolic_multiplier #
 	generate for (syst=0; syst<SYSTOLIC_ARRAY_LENGTH; syst=syst+1)
 		begin : gen_mul
 	
-			/*modexpa7_*/pe_mul mul_ab_inst
+			modexpa7_pe_mul mul_ab_inst
 			(
 				.clk		(clk),
 				.a			(mul_ab_a),
@@ -412,7 +412,7 @@ module modexpa7_systolic_multiplier #
 				.c_out	(mul_ab_c_out[syst])
 			);
 			
-			/*modexpa7_*/pe_mul mul_q_inst
+			modexpa7_pe_mul mul_q_inst
 			(
 				.clk		(clk),
 				.a			(mul_q_a),
@@ -425,7 +425,7 @@ module modexpa7_systolic_multiplier #
 			);
 			
 
-			/*modexpa7_*/pe_mul mul_qn_inst
+			modexpa7_pe_mul mul_qn_inst
 			(
 				.clk		(clk),
 				.a			(mul_qn_a),
diff --git a/src/rtl/pe/artix7/multiplier32_artix7.v b/src/rtl/pe/artix7/multiplier32_artix7.v
new file mode 100644
index 0000000..5cc6340
--- /dev/null
+++ b/src/rtl/pe/artix7/multiplier32_artix7.v
@@ -0,0 +1,83 @@
+`timescale 1ns / 1ps
+
+module multiplier32_artix7
+	(
+		input					clk,
+		input		[31: 0]	a,
+		input		[31: 0]	b,
+		input		[31: 0]	t,
+		input		[31: 0]	c_in,
+		output	[31: 0]	p,
+		output	[31: 0]	c_out
+	);
+	
+	reg	[31: 0]	t_dly;
+	reg	[31: 0]	c_in_dly;
+	
+	always @(posedge clk) t_dly <= t;
+	always @(posedge clk) c_in_dly <= c_in;
+	
+	wire	[31: 0]	t_c_in_s;
+	wire				t_c_in_c_out;
+	
+	reg				t_c_in_c_out_dly;
+	
+	always @(posedge clk) t_c_in_c_out_dly <= t_c_in_c_out;
+	
+	adder32_artix7 add_t_c_in
+	(
+		.clk		(clk),
+		.a			(t_dly),
+		.b			(c_in_dly),
+		.c_in		(1'b0),
+		.s			(t_c_in_s),
+		.c_out	(t_c_in_c_out)
+	);
+
+	wire	[63: 0]	a_b;
+	
+	wire	[31: 0]	a_b_lsb = a_b[31: 0];
+	wire	[31: 0]	a_b_msb = a_b[63:32];
+	
+	reg	[31: 0]	a_b_msb_dly;
+	
+	always @(posedge clk) a_b_msb_dly <= a_b_msb;
+	
+	ip_mul32 mul_a_b
+	(
+		.clk	(clk),
+		.a		(a),
+		.b		(b),
+		.p		(a_b)
+	);
+	
+	wire	[31: 0]	add_p_s;
+	wire				add_p_c_out;
+	
+	reg	[31: 0]	add_p_s_dly;
+	
+	always @(posedge clk) add_p_s_dly <= add_p_s;
+	
+	assign p = add_p_s_dly;
+	
+	adder32_artix7 add_p
+	(
+		.clk		(clk),
+		.a			(a_b_lsb),
+		.b			(t_c_in_s),
+		.c_in		(1'b0),
+		.s			(add_p_s),
+		.c_out	(add_p_c_out)
+	);
+
+	adder32_artix7 add_c_out
+	(
+		.clk		(clk),
+		.a			(a_b_msb_dly),
+		.b			({{31{1'b0}}, t_c_in_c_out_dly}),
+		.c_in		(add_p_c_out),
+		.s			(c_out),
+		.c_out	()
+	);
+
+endmodule
diff --git a/src/rtl/pe/modexpa7_pe_mul.v b/src/rtl/pe/generic/multiplier32_generic.v
similarity index 90%
copy from src/rtl/pe/modexpa7_pe_mul.v
copy to src/rtl/pe/generic/multiplier32_generic.v
index ff15981..794ce96 100644
--- a/src/rtl/pe/modexpa7_pe_mul.v
+++ b/src/rtl/pe/generic/multiplier32_generic.v
@@ -36,7 +36,7 @@
 //
 //======================================================================
 
-module modexpa7_pe_mul
+module multiplier32_generic
 	(
 		input					clk,
 		input		[31: 0]	a,
@@ -47,16 +47,31 @@ module modexpa7_pe_mul
 		output	[31: 0]	c_out
 	);
 
-	localparam LATENCY = 4;
+		//
+		// Customizable Latency
+		//
+	parameter LATENCY = 4;
 		
+		//
+		// Delay Line
+		//
 	reg	[63: 0]	abct[1:LATENCY];
 	
+		//
+		// Outputs
+		//
 	assign p			= abct[LATENCY][31: 0];
 	assign c_out	= abct[LATENCY][63:32];
 
+		//
+		// Sub-products
+		//
 	wire	[63: 0]	ab = {{32{1'b0}}, a}    * {{32{1'b0}}, b};
 	wire	[63: 0]	ct = {{32{1'b0}}, c_in} + {{32{1'b0}}, t};
 
+		//
+		// Delay
+		//
 	integer i;
 	always @(posedge clk)
 		//
diff --git a/src/rtl/pe/modexpa7_pe_mul.v b/src/rtl/pe/modexpa7_pe_mul.v
index ff15981..6453049 100644
--- a/src/rtl/pe/modexpa7_pe_mul.v
+++ b/src/rtl/pe/modexpa7_pe_mul.v
@@ -47,21 +47,26 @@ module modexpa7_pe_mul
 		output	[31: 0]	c_out
 	);
 
-	localparam LATENCY = 4;
-		
-	reg	[63: 0]	abct[1:LATENCY];
-	
-	assign p			= abct[LATENCY][31: 0];
-	assign c_out	= abct[LATENCY][63:32];
 
-	wire	[63: 0]	ab = {{32{1'b0}}, a}    * {{32{1'b0}}, b};
-	wire	[63: 0]	ct = {{32{1'b0}}, c_in} + {{32{1'b0}}, t};
-
-	integer i;
-	always @(posedge clk)
-		//
-		for (i=1; i<=LATENCY; i=i+1)
-			abct[i] <= (i == 1) ? ab + ct : abct[i-1];
+		//
+		// Include Primitive Selector
+		//
+	`include "modexpa7_pe_settings.v"
+
+
+		//
+		// Instantiate Vendor/Generic Primitive
+		//
+	`MULTIPLIER32_PRIMITIVE multiplier32_inst
+	(
+		.clk(clk),
+		.a(a),
+		.b(b),
+		.t(t),
+		.c_in(c_in),
+		.p(p),
+		.c_out(c_out)
+	);
 
 endmodule
 
diff --git a/src/rtl/pe/modexpa7_pe_settings.v b/src/rtl/pe/modexpa7_pe_settings.v
index 97b5b89..9c16d73 100644
--- a/src/rtl/pe/modexpa7_pe_settings.v
+++ b/src/rtl/pe/modexpa7_pe_settings.v
@@ -4,10 +4,12 @@
 
 `define ADDER32_PRIMITIVE			adder32_artix7
 `define SUBTRACTOR32_PRIMITIVE	subtractor32_artix7
+`define MULTIPLIER32_PRIMITIVE	multiplier32_artix7
 
 `else
 
 `define ADDER32_PRIMITIVE			adder32_generic
 `define SUBTRACTOR32_PRIMITIVE	subtractor32_generic
+`define MULTIPLIER32_PRIMITIVE	multiplier32_generic
 
 `endif