310 lines
10 KiB
Coq
310 lines
10 KiB
Coq
|
// ==================================================================
|
||
|
// >>>>>>>>>>>>>>>>>>>>>>> COPYRIGHT NOTICE <<<<<<<<<<<<<<<<<<<<<<<<<
|
||
|
// ------------------------------------------------------------------
|
||
|
// Copyright (c) 2006-2011 by Lattice Semiconductor Corporation
|
||
|
// ALL RIGHTS RESERVED
|
||
|
// ------------------------------------------------------------------
|
||
|
//
|
||
|
// IMPORTANT: THIS FILE IS AUTO-GENERATED BY THE LATTICEMICO SYSTEM.
|
||
|
//
|
||
|
// Permission:
|
||
|
//
|
||
|
// Lattice Semiconductor grants permission to use this code
|
||
|
// pursuant to the terms of the Lattice Semiconductor Corporation
|
||
|
// Open Source License Agreement.
|
||
|
//
|
||
|
// Disclaimer:
|
||
|
//
|
||
|
// Lattice Semiconductor provides no warranty regarding the use or
|
||
|
// functionality of this code. It is the user's responsibility to
|
||
|
// verify the user's design for consistency and functionality through
|
||
|
// the use of formal verification methods.
|
||
|
//
|
||
|
// --------------------------------------------------------------------
|
||
|
//
|
||
|
// Lattice Semiconductor Corporation
|
||
|
// 5555 NE Moore Court
|
||
|
// Hillsboro, OR 97214
|
||
|
// U.S.A
|
||
|
//
|
||
|
// TEL: 1-800-Lattice (USA and Canada)
|
||
|
// 503-286-8001 (other locations)
|
||
|
//
|
||
|
// web: http://www.latticesemi.com/
|
||
|
// email: techsupport@latticesemi.com
|
||
|
//
|
||
|
// --------------------------------------------------------------------
|
||
|
// FILE DETAILS
|
||
|
// Project : LatticeMico32
|
||
|
// File : lm_mc_arithmetic.v
|
||
|
// Title : Multi-cycle arithmetic unit.
|
||
|
// Dependencies : lm32_include.v
|
||
|
// Version : 6.1.17
|
||
|
// : Initial Release
|
||
|
// Version : 7.0SP2, 3.0
|
||
|
// : No Change
|
||
|
// Version : 3.1
|
||
|
// : No Change
|
||
|
// =============================================================================
|
||
|
|
||
|
`include "lm32_include.v"
|
||
|
|
||
|
`define LM32_MC_STATE_RNG 2:0
|
||
|
`define LM32_MC_STATE_IDLE 3'b000
|
||
|
`define LM32_MC_STATE_MULTIPLY 3'b001
|
||
|
`define LM32_MC_STATE_MODULUS 3'b010
|
||
|
`define LM32_MC_STATE_DIVIDE 3'b011
|
||
|
`define LM32_MC_STATE_SHIFT_LEFT 3'b100
|
||
|
`define LM32_MC_STATE_SHIFT_RIGHT 3'b101
|
||
|
|
||
|
/////////////////////////////////////////////////////
|
||
|
// Module interface
|
||
|
/////////////////////////////////////////////////////
|
||
|
|
||
|
module lm32_mc_arithmetic (
|
||
|
// ----- Inputs -----
|
||
|
clk_i,
|
||
|
rst_i,
|
||
|
stall_d,
|
||
|
kill_x,
|
||
|
`ifdef CFG_MC_DIVIDE_ENABLED
|
||
|
divide_d,
|
||
|
modulus_d,
|
||
|
`endif
|
||
|
`ifdef CFG_MC_MULTIPLY_ENABLED
|
||
|
multiply_d,
|
||
|
`endif
|
||
|
`ifdef CFG_MC_BARREL_SHIFT_ENABLED
|
||
|
shift_left_d,
|
||
|
shift_right_d,
|
||
|
sign_extend_d,
|
||
|
`endif
|
||
|
operand_0_d,
|
||
|
operand_1_d,
|
||
|
// ----- Ouputs -----
|
||
|
result_x,
|
||
|
`ifdef CFG_MC_DIVIDE_ENABLED
|
||
|
divide_by_zero_x,
|
||
|
`endif
|
||
|
stall_request_x
|
||
|
);
|
||
|
|
||
|
/////////////////////////////////////////////////////
|
||
|
// Inputs
|
||
|
/////////////////////////////////////////////////////
|
||
|
|
||
|
input clk_i; // Clock
|
||
|
input rst_i; // Reset
|
||
|
input stall_d; // Stall instruction in D stage
|
||
|
input kill_x; // Kill instruction in X stage
|
||
|
`ifdef CFG_MC_DIVIDE_ENABLED
|
||
|
input divide_d; // Perform divide
|
||
|
input modulus_d; // Perform modulus
|
||
|
`endif
|
||
|
`ifdef CFG_MC_MULTIPLY_ENABLED
|
||
|
input multiply_d; // Perform multiply
|
||
|
`endif
|
||
|
`ifdef CFG_MC_BARREL_SHIFT_ENABLED
|
||
|
input shift_left_d; // Perform left shift
|
||
|
input shift_right_d; // Perform right shift
|
||
|
input sign_extend_d; // Whether to sign-extend (arithmetic) or zero-extend (logical)
|
||
|
`endif
|
||
|
input [`LM32_WORD_RNG] operand_0_d;
|
||
|
input [`LM32_WORD_RNG] operand_1_d;
|
||
|
|
||
|
/////////////////////////////////////////////////////
|
||
|
// Outputs
|
||
|
/////////////////////////////////////////////////////
|
||
|
|
||
|
output [`LM32_WORD_RNG] result_x; // Result of operation
|
||
|
reg [`LM32_WORD_RNG] result_x;
|
||
|
`ifdef CFG_MC_DIVIDE_ENABLED
|
||
|
output divide_by_zero_x; // A divide by zero was attempted
|
||
|
reg divide_by_zero_x;
|
||
|
`endif
|
||
|
output stall_request_x; // Request to stall pipeline from X stage back
|
||
|
wire stall_request_x;
|
||
|
|
||
|
/////////////////////////////////////////////////////
|
||
|
// Internal nets and registers
|
||
|
/////////////////////////////////////////////////////
|
||
|
|
||
|
reg [`LM32_WORD_RNG] p; // Temporary registers
|
||
|
reg [`LM32_WORD_RNG] a;
|
||
|
reg [`LM32_WORD_RNG] b;
|
||
|
`ifdef CFG_MC_DIVIDE_ENABLED
|
||
|
wire [32:0] t;
|
||
|
`endif
|
||
|
|
||
|
reg [`LM32_MC_STATE_RNG] state; // Current state of FSM
|
||
|
reg [5:0] cycles; // Number of cycles remaining in the operation
|
||
|
|
||
|
`ifdef CFG_MC_BARREL_SHIFT_ENABLED
|
||
|
reg sign_extend_x; // Whether to sign extend of zero extend right shifts
|
||
|
wire fill_value; // Value to fill with for right barrel-shifts
|
||
|
`endif
|
||
|
|
||
|
/////////////////////////////////////////////////////
|
||
|
// Combinational logic
|
||
|
/////////////////////////////////////////////////////
|
||
|
|
||
|
// Stall pipeline while any operation is being performed
|
||
|
assign stall_request_x = state != `LM32_MC_STATE_IDLE;
|
||
|
|
||
|
`ifdef CFG_MC_DIVIDE_ENABLED
|
||
|
// Subtraction
|
||
|
assign t = {p[`LM32_WORD_WIDTH-2:0], a[`LM32_WORD_WIDTH-1]} - b;
|
||
|
`endif
|
||
|
|
||
|
`ifdef CFG_MC_BARREL_SHIFT_ENABLED
|
||
|
// Determine fill value for right shift - Sign bit for arithmetic shift, or zero for logical shift
|
||
|
assign fill_value = (sign_extend_x == `TRUE) & b[`LM32_WORD_WIDTH-1];
|
||
|
`endif
|
||
|
|
||
|
/////////////////////////////////////////////////////
|
||
|
// Sequential logic
|
||
|
/////////////////////////////////////////////////////
|
||
|
|
||
|
// Perform right shift
|
||
|
always @(posedge clk_i `CFG_RESET_SENSITIVITY)
|
||
|
begin
|
||
|
if (rst_i == `TRUE)
|
||
|
begin
|
||
|
cycles <= {6{1'b0}};
|
||
|
p <= {`LM32_WORD_WIDTH{1'b0}};
|
||
|
a <= {`LM32_WORD_WIDTH{1'b0}};
|
||
|
b <= {`LM32_WORD_WIDTH{1'b0}};
|
||
|
`ifdef CFG_MC_BARREL_SHIFT_ENABLED
|
||
|
sign_extend_x <= 1'b0;
|
||
|
`endif
|
||
|
`ifdef CFG_MC_DIVIDE_ENABLED
|
||
|
divide_by_zero_x <= `FALSE;
|
||
|
`endif
|
||
|
result_x <= {`LM32_WORD_WIDTH{1'b0}};
|
||
|
state <= `LM32_MC_STATE_IDLE;
|
||
|
end
|
||
|
else
|
||
|
begin
|
||
|
`ifdef CFG_MC_DIVIDE_ENABLED
|
||
|
divide_by_zero_x <= `FALSE;
|
||
|
`endif
|
||
|
case (state)
|
||
|
`LM32_MC_STATE_IDLE:
|
||
|
begin
|
||
|
if (stall_d == `FALSE)
|
||
|
begin
|
||
|
cycles <= `LM32_WORD_WIDTH;
|
||
|
p <= 32'b0;
|
||
|
a <= operand_0_d;
|
||
|
b <= operand_1_d;
|
||
|
`ifdef CFG_MC_DIVIDE_ENABLED
|
||
|
if (divide_d == `TRUE)
|
||
|
state <= `LM32_MC_STATE_DIVIDE;
|
||
|
if (modulus_d == `TRUE)
|
||
|
state <= `LM32_MC_STATE_MODULUS;
|
||
|
`endif
|
||
|
`ifdef CFG_MC_MULTIPLY_ENABLED
|
||
|
if (multiply_d == `TRUE)
|
||
|
state <= `LM32_MC_STATE_MULTIPLY;
|
||
|
`endif
|
||
|
`ifdef CFG_MC_BARREL_SHIFT_ENABLED
|
||
|
if (shift_left_d == `TRUE)
|
||
|
begin
|
||
|
state <= `LM32_MC_STATE_SHIFT_LEFT;
|
||
|
sign_extend_x <= sign_extend_d;
|
||
|
cycles <= operand_1_d[4:0];
|
||
|
a <= operand_0_d;
|
||
|
b <= operand_0_d;
|
||
|
end
|
||
|
if (shift_right_d == `TRUE)
|
||
|
begin
|
||
|
state <= `LM32_MC_STATE_SHIFT_RIGHT;
|
||
|
sign_extend_x <= sign_extend_d;
|
||
|
cycles <= operand_1_d[4:0];
|
||
|
a <= operand_0_d;
|
||
|
b <= operand_0_d;
|
||
|
end
|
||
|
`endif
|
||
|
end
|
||
|
end
|
||
|
`ifdef CFG_MC_DIVIDE_ENABLED
|
||
|
`LM32_MC_STATE_DIVIDE:
|
||
|
begin
|
||
|
if (t[32] == 1'b0)
|
||
|
begin
|
||
|
p <= t[31:0];
|
||
|
a <= {a[`LM32_WORD_WIDTH-2:0], 1'b1};
|
||
|
end
|
||
|
else
|
||
|
begin
|
||
|
p <= {p[`LM32_WORD_WIDTH-2:0], a[`LM32_WORD_WIDTH-1]};
|
||
|
a <= {a[`LM32_WORD_WIDTH-2:0], 1'b0};
|
||
|
end
|
||
|
result_x <= a;
|
||
|
if ((cycles == `LM32_WORD_WIDTH'd0) || (kill_x == `TRUE))
|
||
|
begin
|
||
|
// Check for divide by zero
|
||
|
divide_by_zero_x <= b == {`LM32_WORD_WIDTH{1'b0}};
|
||
|
state <= `LM32_MC_STATE_IDLE;
|
||
|
end
|
||
|
cycles <= cycles - 1'b1;
|
||
|
end
|
||
|
`LM32_MC_STATE_MODULUS:
|
||
|
begin
|
||
|
if (t[32] == 1'b0)
|
||
|
begin
|
||
|
p <= t[31:0];
|
||
|
a <= {a[`LM32_WORD_WIDTH-2:0], 1'b1};
|
||
|
end
|
||
|
else
|
||
|
begin
|
||
|
p <= {p[`LM32_WORD_WIDTH-2:0], a[`LM32_WORD_WIDTH-1]};
|
||
|
a <= {a[`LM32_WORD_WIDTH-2:0], 1'b0};
|
||
|
end
|
||
|
result_x <= p;
|
||
|
if ((cycles == `LM32_WORD_WIDTH'd0) || (kill_x == `TRUE))
|
||
|
begin
|
||
|
// Check for divide by zero
|
||
|
divide_by_zero_x <= b == {`LM32_WORD_WIDTH{1'b0}};
|
||
|
state <= `LM32_MC_STATE_IDLE;
|
||
|
end
|
||
|
cycles <= cycles - 1'b1;
|
||
|
end
|
||
|
`endif
|
||
|
`ifdef CFG_MC_MULTIPLY_ENABLED
|
||
|
`LM32_MC_STATE_MULTIPLY:
|
||
|
begin
|
||
|
if (b[0] == 1'b1)
|
||
|
p <= p + a;
|
||
|
b <= {1'b0, b[`LM32_WORD_WIDTH-1:1]};
|
||
|
a <= {a[`LM32_WORD_WIDTH-2:0], 1'b0};
|
||
|
result_x <= p;
|
||
|
if ((cycles == `LM32_WORD_WIDTH'd0) || (kill_x == `TRUE))
|
||
|
state <= `LM32_MC_STATE_IDLE;
|
||
|
cycles <= cycles - 1'b1;
|
||
|
end
|
||
|
`endif
|
||
|
`ifdef CFG_MC_BARREL_SHIFT_ENABLED
|
||
|
`LM32_MC_STATE_SHIFT_LEFT:
|
||
|
begin
|
||
|
a <= {a[`LM32_WORD_WIDTH-2:0], 1'b0};
|
||
|
result_x <= a;
|
||
|
if ((cycles == `LM32_WORD_WIDTH'd0) || (kill_x == `TRUE))
|
||
|
state <= `LM32_MC_STATE_IDLE;
|
||
|
cycles <= cycles - 1'b1;
|
||
|
end
|
||
|
`LM32_MC_STATE_SHIFT_RIGHT:
|
||
|
begin
|
||
|
b <= {fill_value, b[`LM32_WORD_WIDTH-1:1]};
|
||
|
result_x <= b;
|
||
|
if ((cycles == `LM32_WORD_WIDTH'd0) || (kill_x == `TRUE))
|
||
|
state <= `LM32_MC_STATE_IDLE;
|
||
|
cycles <= cycles - 1'b1;
|
||
|
end
|
||
|
`endif
|
||
|
endcase
|
||
|
end
|
||
|
end
|
||
|
|
||
|
endmodule
|