Subversion Repositories spi_master_slave

-----------------------------------------------------------------------------------------------------------------------

-- Author:          Jonny Doin, jdoin@opencores.org, jonnydoin@gmail.com

-- 

-- Create Date:     12:18:12 04/25/2011 

-- Module Name:     SPI_MASTER - RTL

-- Project Name:    SPI MASTER / SLAVE INTERFACE

-- Target Devices:  Spartan-6

-- Tool versions:   ISE 13.1

-- Description: 

--

--      This block is the SPI master interface, implemented in one single entity.

--      All internal core operations are synchronous to the 'sclk_i', and a spi base clock is generated by dividing sclk_i downto

--      a frequency that is 2x the spi SCK line frequency. The divider value is passed as a generic parameter during instantiation.

--      All parallel i/o interface operations are synchronous to the 'pclk_i' high speed clock, that can be asynchronous to the serial

--      'sclk_i' clock.

--      For optimized use of longlines, connect 'sclk_i' and 'pclk_i' to the same global clock line.

--      Fully pipelined cross-clock circuitry guarantees that no setup artifacts occur on the buffers that are accessed by the two 

--      clock domains.

--      The block is very simple to use, and has parallel inputs and outputs that behave like a synchronous memory i/o.

--      It is parameterizable via generics for the data width ('N'), SPI mode (CPHA and CPOL), lookahead prefetch signaling 

--      ('PREFETCH'), and spi base clock division from sclk_i ('SPI_2X_CLK_DIV').

--

--      SPI CLOCK GENERATION

--      ====================

--

--      The clock generation for the SPI SCK is derived from the high-speed 'sclk_i' clock. The core divides this reference 

--      clock to form the SPI base clock, by the 'SPI_2X_CLK_DIV' generic parameter. The user must set the divider value for the 

--      SPI_2X clock, which is 2x the desired SCK frequency. 

--      All registers in the core are clocked by the high-speed clocks, and clock enables are used to run the FSM and other logic

--      at lower rates. This architecture preserves FPGA clock resources like global clock buffers, and avoids path delays caused

--      by combinatorial clock dividers outputs.

--      The core has async clock domain circuitry to handle asynchronous clocks for the SPI and parallel interfaces.

--

--      PARALLEL WRITE INTERFACE

--      ========================

--      The parallel interface has an input port 'di_i' and an output port 'do_o'.

--      Parallel load is controlled using 3 signals: 'di_i', 'di_req_o' and 'wren_i'. 'di_req_o' is a look ahead data request line,

--      that is set 'PREFETCH' clock cycles in advance to synchronize a pipelined memory or fifo to present the 

--      next input data at 'di_i' in time to have continuous clock at the spi bus, to allow back-to-back continuous load.

--      For a pipelined sync RAM, a PREFETCH of 2 cycles allows an address generator to present the new adress to the RAM in one

--      cycle, and the RAM to respond in one more cycle, in time for 'di_i' to be latched by the shifter.

--      If the user sequencer needs a different value for PREFETCH, the generic can be altered at instantiation time.

--      The 'wren_i' write enable strobe must be valid at least one setup time before the rising edge of the last SPI clock cycle,

--      if continuous transmission is intended. If 'wren_i' is not valid 2 SPI clock cycles after the last transmitted bit, the interface

--      enters idle state and deasserts SSEL.

--      When the interface is idle, 'wren_i' write strobe loads the data and starts transmission. 'di_req_o' will strobe when entering 

--      idle state, if a previously loaded data has already been transferred.

--

--      PARALLEL WRITE SEQUENCE

--      =======================

--                         __    __    __    __    __    __    __ 

--      pclk_i          __/  \__/  \__/  \__/  \__/  \__/  \__/  \...     -- parallel interface clock

--                               ___________                        

--      di_req_o        ________/           \_____________________...     -- 'di_req_o' asserted on rising edge of 'pclk_i'

--                      ______________ ___________________________...

--      di_i            __old_data____X______new_data_____________...     -- user circuit loads data on 'di_i' at next 'pclk_i' rising edge

--                                                 _______                        

--      wren_i          __________________________/       \_______...     -- user strobes 'wren_i' for one cycle of 'pclk_i'

--                      

--

--      PARALLEL READ INTERFACE

--      =======================

--      An internal buffer is used to copy the internal shift register data to drive the 'do_o' port. When a complete word is received,

--      the core shift register is transferred to the buffer, at the rising edge of the spi clock, 'spi_clk'.

--      The signal 'do_valid_o' is set one 'spi_clk' clock after, to directly drive a synchronous memory or fifo write enable.

--      'do_valid_o' is synchronous to the parallel interface clock, and changes only on rising edges of 'pclk_i'.

--      When the interface is idle, data at the 'do_o' port holds the last word received.

--

--      PARALLEL READ SEQUENCE

--      ======================

--                      ______        ______        ______        ______   

--      spi_clk          bit1 \______/ bitN \______/bitN-1\______/bitN-2\__...  -- internal spi 2x base clock

--                      _    __    __    __    __    __    __    __    __  

--      pclk_i           \__/  \__/  \__/  \__/  \__/  \__/  \__/  \__/  \_...  -- parallel interface clock (may be async to sclk_i)

--                      _____________ _____________________________________...  -- 1) rx data is transferred to 'do_buffer_reg'

--      do_o            ___old_data__X__________new_data___________________...  --    after last rx bit, at rising 'spi_clk'.

--                                                   ____________               

--      do_valid_o      ____________________________/            \_________...  -- 2) 'do_valid_o' strobed for 2 'pclk_i' cycles

--                                                                              --    on the 3rd 'pclk_i' rising edge.

--

--

--      The propagation delay of spi_sck_o and spi_mosi_o, referred to the internal clock, is balanced by similar path delays,

--      but the sampling delay of spi_miso_i imposes a setup time referred to the sck signal that limits the high frequency

--      of the interface, for full duplex operation.

--

--      This design was originally targeted to a Spartan-6 platform, synthesized with XST and normal constraints.

--      The VHDL dialect used is VHDL'93, accepted largely by all synthesis tools.

--

------------------------------ COPYRIGHT NOTICE -----------------------------------------------------------------------

--                                                                   

--      This file is part of the SPI MASTER/SLAVE INTERFACE project http://opencores.org/project,spi_master_slave

--                                                                   

--      Author(s):      Jonny Doin, jdoin@opencores.org

--                                                                   

--      Copyright (C) 2011 Authors and OPENCORES.ORG

--      --------------------------------------------

--                                                                   

--      This source file may be used and distributed without restriction provided that this copyright statement is not    

--      removed from the file and that any derivative work contains the original copyright notice and the associated 

--      disclaimer. 

--                                                                   

--      This source file is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser 

--      General Public License as published by the Free Software Foundation; either version 2.1 of the License, or 

--      (at your option) any later version.

--                                                                   

--      This source is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied

--      warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more  

--      details.

--

--      You should have received a copy of the GNU Lesser General Public License along with this source; if not, download 

--      it from http://www.opencores.org/lgpl.shtml

--                                                                   

------------------------------ REVISION HISTORY -----------------------------------------------------------------------

--

-- 2011/04/28   v0.01.0010  [JD]    shifter implemented as a sequential process. timing problems and async issues in synthesis.

-- 2011/05/01   v0.01.0030  [JD]    changed original shifter design to a fully pipelined RTL fsmd. solved all synthesis issues.

-- 2011/05/05   v0.01.0034  [JD]    added an internal buffer register for rx_data, to allow greater liberty in data load/store.

-- 2011/05/08   v0.10.0038  [JD]    increased one state to have SSEL start one cycle before SCK. Implemented full CPOL/CPHA

--                                  logic, based on generics, and do_valid_o signal.

-- 2011/05/13   v0.20.0045  [JD]    streamlined signal names, added PREFETCH parameter, added assertions.

-- 2011/05/17   v0.80.0049  [JD]    added explicit clock synchronization circuitry across clock boundaries.

-- 2011/05/18   v0.95.0050  [JD]    clock generation circuitry, with generators for all-rising-edge clock core.

-- 2011/06/05   v0.96.0053  [JD]    changed async clear to sync resets.

-- 2011/06/07   v0.97.0065  [JD]    added cross-clock buffers, fixed fsm async glitches.

-- 2011/06/09   v0.97.0068  [JD]    reduced control sets (resets, CE, presets) to the absolute minimum to operate, to reduce

--                                  synthesis LUT overhead in Spartan-6 architecture.

-- 2011/06/11   v0.97.0075  [JD]    redesigned all parallel data interfacing ports, and implemented cross-clock strobe logic.

-- 2011/06/12   v0.97.0079  [JD]    streamlined wr_ack for all cases and eliminated unnecessary register resets.

-- 2011/06/14   v0.97.0083  [JD]    (bug CPHA effect) : redesigned SCK output circuit.

--                                  (minor bug) : removed fsm registers from (not rst_i) chip enable.

-- 2011/06/15   v0.97.0086  [JD]    removed master MISO input register, to relax MISO data setup time (to get higher speed).

-- 2011/07/09   v1.00.0095  [JD]    changed all clocking scheme to use a single high-speed clock with clock enables to control lower 

--                                  frequency sequential circuits, to preserve clocking resources and avoid path delay glitches.

-- 2011/07/10   v1.00.0098  [JD]    implemented SCK clock divider circuit to generate spi clock directly from system clock.

-- 2011/07/10   v1.10.0075  [JD]    verified spi_master_slave in silicon at 50MHz, 25MHz, 16.666MHz, 12.5MHz, 10MHz, 8.333MHz, 

--                                  7.1428MHz, 6.25MHz, 1MHz and 500kHz. The core proved very robust at all tested frequencies.

-- 2011/07/16   v1.11.0080  [JD]    verified both spi_master and spi_slave in loopback at 50MHz SPI clock.

-- 2011/07/17   v1.11.0080  [JD]    BUG: CPOL='1', CPHA='1' @50MHz causes MOSI to be shifted one bit earlier.

--                                  BUG: CPOL='0', CPHA='1' causes SCK to have one extra pulse with one sclk_i width at the end.

-- 2011/07/18   v1.12.0105  [JD]    CHG: spi sck output register changed to remove glitch at last clock when CPHA='1'.

--                                  for CPHA='1', max spi clock is 25MHz. for CPHA= '0', max spi clock is >50MHz.

-- 2011/07/24   v1.13.0125  [JD]    FIX: 'sck_ena_ce' is on half-cycle advanced to 'fsm_ce', elliminating CPHA='1' glitches.

--                                  Core verified for all CPOL, CPHA at up to 50MHz, simulates to over 100MHz.

-- 2011/07/29   v1.14.0130  [JD]    Removed global signal setting at the FSM, implementing exhaustive explicit signal attributions

--                                  for each state, to avoid reported inference problems in some synthesis engines.

--                                  Streamlined port names and indentation blocks.

-- 2011/08/01   v1.15.0135  [JD]    Fixed latch inference for spi_mosi_o driver at the fsm.

--                                  The master and slave cores were verified in FPGA with continuous transmission, for all SPI modes.

--

-----------------------------------------------------------------------------------------------------------------------

--  TODO

--  ====

--

-----------------------------------------------------------------------------------------------------------------------

library ieee;

use ieee.std_logic_1164.all;

use ieee.numeric_std.all;

use ieee.std_logic_unsigned.all;

--================================================================================================================

-- SYNTHESIS CONSIDERATIONS

-- ========================

-- There are several output ports that are used to simulate and verify the core operation. 

-- Do not map any signals to the unused ports, and the synthesis tool will remove the related interfacing

-- circuitry. 

-- The same is valid for the transmit and receive ports. If the receive ports are not mapped, the

-- synthesis tool will remove the receive logic from the generated circuitry.

--================================================================================================================

entity spi_master is

    Generic (

        N : positive := 32;                                             -- 32bit serial word length is default

        CPOL : std_logic := '0';                                        -- SPI mode selection (mode 0 default)

        CPHA : std_logic := '0';                                        -- CPOL = clock polarity, CPHA = clock phase.

        PREFETCH : positive := 2;                                       -- prefetch lookahead cycles

        SPI_2X_CLK_DIV : positive := 5);                                -- for a 100MHz sclk_i, yields a 10MHz SCK

    Port (

        sclk_i : in std_logic := 'X';                                   -- high-speed serial interface system clock

        pclk_i : in std_logic := 'X';                                   -- high-speed parallel interface system clock

        rst_i : in std_logic := 'X';                                    -- reset core

        ---- serial interface ----

        spi_ssel_o : out std_logic;                                     -- spi bus slave select line

        spi_sck_o : out std_logic;                                      -- spi bus sck

        spi_mosi_o : out std_logic;                                     -- spi bus mosi output

        spi_miso_i : in std_logic := 'X';                               -- spi bus spi_miso_i input

        ---- parallel interface ----

        di_req_o : out std_logic;                                       -- preload lookahead data request line

        di_i : in  std_logic_vector (N-1 downto 0) := (others => 'X');  -- parallel data in (clocked on rising spi_clk after last bit)

        wren_i : in std_logic := 'X';                                   -- user data write enable, starts transmission when interface is idle

        wr_ack_o : out std_logic;                                       -- write acknowledge

        do_valid_o : out std_logic;                                     -- do_o data valid signal, valid during one spi_clk rising edge.

        do_o : out  std_logic_vector (N-1 downto 0);                    -- parallel output (clocked on rising spi_clk after last bit)

        --- debug ports: can be removed or left unconnected for the application circuit ---

        sck_ena_o : out std_logic;                                      -- debug: internal sck enable signal

        sck_ena_ce_o : out std_logic;                                   -- debug: internal sck clock enable signal

        do_transfer_o : out std_logic;                                  -- debug: internal transfer driver

        wren_o : out std_logic;                                         -- debug: internal state of the wren_i pulse stretcher

        rx_bit_reg_o : out std_logic;                                   -- debug: internal rx bit

        state_dbg_o : out std_logic_vector (3 downto 0);                -- debug: internal state register

        core_clk_o : out std_logic;

        core_n_clk_o : out std_logic;

        core_ce_o : out std_logic;

        core_n_ce_o : out std_logic;

        sh_reg_dbg_o : out std_logic_vector (N-1 downto 0)              -- debug: internal shift register

    );

end spi_master;

--================================================================================================================

-- this architecture is a pipelined register-transfer description.

-- all signals are clocked at the rising edge of the system clock 'sclk_i'.

--================================================================================================================

architecture rtl of spi_master is

    -- core clocks, generated from 'sclk_i': initialized to differential values

    signal core_clk     : std_logic := '0';     -- continuous core clock, positive logic

    signal core_n_clk   : std_logic := '1';     -- continuous core clock, negative logic

    signal core_ce      : std_logic := '0';     -- core clock enable, positive logic

    signal core_n_ce    : std_logic := '1';     -- core clock enable, negative logic

    -- spi bus clock, generated from the CPOL selected core clock polarity

    signal spi_2x_ce    : std_logic := '1';     -- spi_2x clock enable

    signal spi_clk      : std_logic := '0';     -- spi bus output clock

    signal spi_clk_reg  : std_logic;            -- output pipeline delay for spi sck (do NOT global initialize)

    -- core fsm clock enables

    signal fsm_ce       : std_logic := '1';     -- fsm clock enable

    signal sck_ena_ce   : std_logic := '1';     -- SCK clock enable

    signal samp_ce      : std_logic := '1';     -- data sampling clock enable

    --

    -- GLOBAL RESET: 

    --      all signals are initialized to zero at GSR (global set/reset) by giving explicit

    --      initialization values at declaration. This is needed for all Xilinx FPGAs, and 

    --      especially for the Spartan-6 and newer CLB architectures, where a async reset can

    --      reduce the usability of the slice registers, due to the need to share the control 

    --      set (RESET/PRESET, CLOCK ENABLE and CLOCK) by all 8 registers in a slice.

    --      By using GSR for the initialization, and reducing async RESET local init to the bare

    --      essential, the model achieves better LUT/FF packing and CLB usability.

    --

    -- internal state signals for register and combinatorial stages

    signal state_next : natural range N+1 downto 0 := 0;

    signal state_reg : natural range N+1 downto 0 := 0;

    -- shifter signals for register and combinatorial stages

    signal sh_next : std_logic_vector (N-1 downto 0);

    signal sh_reg : std_logic_vector (N-1 downto 0);

    -- input bit sampled buffer

    signal rx_bit_reg : std_logic := '0';

    -- buffered di_i data signals for register and combinatorial stages

    signal di_reg : std_logic_vector (N-1 downto 0);

    -- internal wren_i stretcher for fsm combinatorial stage

    signal wren : std_logic;

    signal wr_ack_next : std_logic := '0';

    signal wr_ack_reg : std_logic := '0';

    -- internal SSEL enable control signals

    signal ssel_ena_next : std_logic := '0';

    signal ssel_ena_reg : std_logic := '0';

    -- internal SCK enable control signals

    signal sck_ena_next : std_logic;

    signal sck_ena_reg : std_logic;

    -- buffered do_o data signals for register and combinatorial stages

    signal do_buffer_next : std_logic_vector (N-1 downto 0);

    signal do_buffer_reg : std_logic_vector (N-1 downto 0);

    -- internal signal to flag transfer to do_buffer_reg

    signal do_transfer_next : std_logic := '0';

    signal do_transfer_reg : std_logic := '0';

    -- internal input data request signal 

    signal di_req_next : std_logic := '0';

    signal di_req_reg : std_logic := '0';

    -- cross-clock do_transfer_reg -> do_valid_o_reg pipeline

    signal do_valid_A : std_logic := '0';

    signal do_valid_B : std_logic := '0';

    signal do_valid_C : std_logic := '0';

    signal do_valid_D : std_logic := '0';

    signal do_valid_next : std_logic := '0';

    signal do_valid_o_reg : std_logic := '0';

    -- cross-clock di_req_reg -> di_req_o_reg pipeline

    signal di_req_o_A : std_logic := '0';

    signal di_req_o_B : std_logic := '0';

    signal di_req_o_C : std_logic := '0';

    signal di_req_o_D : std_logic := '0';

    signal di_req_o_next : std_logic := '1';

    signal di_req_o_reg : std_logic := '1';

begin

    --=============================================================================================

    --  GENERICS CONSTRAINTS CHECKING

    --=============================================================================================

    -- minimum word width is 8 bits

    assert N >= 8

    report "Generic parameter 'N' (shift register size) needs to be 8 bits minimum"

    severity FAILURE;

    -- minimum prefetch lookahead check

    assert PREFETCH >= 2

    report "Generic parameter 'PREFETCH' (lookahead count) needs to be 1 minimum"

    severity FAILURE;

    -- maximum prefetch lookahead check

    assert PREFETCH <= N-5

    report "Generic parameter 'PREFETCH' (lookahead count) out of range, needs to be N-5 maximum"

    severity FAILURE;

    -- SPI_2X_CLK_DIV clock divider value must not be zero

    assert SPI_2X_CLK_DIV > 0

    report "Generic parameter 'SPI_2X_CLK_DIV' must not be zero"

    severity FAILURE;

    --=============================================================================================

    --  CLOCK GENERATION

    --=============================================================================================

    -- In order to preserve global clocking resources, the core clocking scheme is completely based 

    -- on using clock enables to process the serial high-speed clock at lower rates for the core fsm,

    -- the spi clock generator and the input sampling clock.

    -- The clock generation block derive 2 continuous antiphase signals from the 2x spi base clock 

    -- for the core clocking.

    -- The 2 clock phases are generated by sepparate and synchronous FFs, and should have only 

    -- differential interconnect delay skew.

    -- Clock enable signals are generated with the same phase as the 2 core clocks, and these clock 

    -- enables are used to control clocking of all internal synchronous circuitry. 

    -- The clock enable phase is selected for serial input sampling, fsm clocking, and spi SCK output, 

    -- based on the configuration of CPOL and CPHA.

    -- Each phase is selected so that all the registers can be clocked with a rising edge on all SPI

    -- modes, by a single high-speed global clock, preserving clock resources and clock to data skew.

    -----------------------------------------------------------------------------------------------

    -- generate the 2x spi base clock enable from the serial high-speed input clock

    spi_2x_ce_gen_proc: process (sclk_i) is

        variable clk_cnt : integer range SPI_2X_CLK_DIV-1 downto 0 := 0;

    begin

        if sclk_i'event and sclk_i = '1' then

            if clk_cnt = SPI_2X_CLK_DIV-1 then

                spi_2x_ce <= '1';

                clk_cnt := 0;

            else

                spi_2x_ce <= '0';

                clk_cnt := clk_cnt + 1;

            end if;

        end if;

    end process spi_2x_ce_gen_proc;

    -----------------------------------------------------------------------------------------------

    -- generate the core antiphase clocks and clock enables from the 2x base CE.

    core_clock_gen_proc : process (sclk_i) is

    begin

        if sclk_i'event and sclk_i = '1' then

            if spi_2x_ce = '1' then

                -- generate the 2 antiphase core clocks

                core_clk <= core_n_clk;

                core_n_clk <= not core_n_clk;

                -- generate the 2 phase core clock enables

                core_ce <= core_n_clk;

                core_n_ce <= not core_n_clk;

            else

                core_ce <= '0';

                core_n_ce <= '0';

            end if;

        end if;

    end process core_clock_gen_proc;

    --=============================================================================================

    --  GENERATE BLOCKS

    --=============================================================================================

    -- spi clk generator: generate spi_clk from core_clk depending on CPOL

    spi_sck_cpol_0_proc: if CPOL = '0' generate

    begin

        spi_clk <= core_clk;            -- for CPOL=0, spi clk has idle LOW

    end generate;

    spi_sck_cpol_1_proc: if CPOL = '1' generate

    begin

        spi_clk <= core_n_clk;          -- for CPOL=1, spi clk has idle HIGH

    end generate;

    -----------------------------------------------------------------------------------------------

    -- Sampling clock enable generation: generate 'samp_ce' from 'core_ce' or 'core_n_ce' depending on CPHA

    -- always sample data at the half-cycle of the fsm update cell

    samp_ce_cpha_0_proc: if CPHA = '0' generate

    begin

        samp_ce <= core_ce;

    end generate;

    samp_ce_cpha_1_proc: if CPHA = '1' generate

    begin

        samp_ce <= core_n_ce;

    end generate;

    -----------------------------------------------------------------------------------------------

    -- FSM clock enable generation: generate 'fsm_ce' from core_ce or core_n_ce depending on CPHA

    fsm_ce_cpha_0_proc: if CPHA = '0' generate

    begin

        fsm_ce <= core_n_ce;            -- for CPHA=0, latch registers at rising edge of negative core clock enable

    end generate;

    fsm_ce_cpha_1_proc: if CPHA = '1' generate

    begin

        fsm_ce <= core_ce;              -- for CPHA=1, latch registers at rising edge of positive core clock enable

    end generate;

    -----------------------------------------------------------------------------------------------

    -- sck enable control: control sck advance phase for CPHA='1' relative to fsm clock

    sck_ena_ce <= core_n_ce;            -- for CPHA=1, SCK is advanced one-half cycle

    --=============================================================================================

    --  REGISTERED INPUTS

    --=============================================================================================

    -- rx bit flop: capture rx bit after SAMPLE edge of sck

    rx_bit_proc : process (sclk_i, spi_miso_i) is

    begin

        if sclk_i'event and sclk_i = '1' then

            if samp_ce = '1' then

                rx_bit_reg <= spi_miso_i;

            end if;

        end if;

    end process rx_bit_proc;

    --=============================================================================================

    --  CROSS-CLOCK PIPELINE TRANSFER LOGIC

    --=============================================================================================

    -- do_valid_o and di_req_o strobe output logic

    -- this is a delayed pulse generator with a ripple-transfer FFD pipeline, that generates a 

    -- fixed-length delayed pulse for the output flags, at the parallel clock domain

    out_transfer_proc : process ( pclk_i, do_transfer_reg, di_req_reg,

                                  do_valid_A, do_valid_B, do_valid_D,

                                  di_req_o_A, di_req_o_B, di_req_o_D ) is

    begin

        if pclk_i'event and pclk_i = '1' then               -- clock at parallel port clock

            -- do_transfer_reg -> do_valid_o_reg

            do_valid_A <= do_transfer_reg;                  -- the input signal must be at least 2 clocks long

            do_valid_B <= do_valid_A;                       -- feed it to a ripple chain of FFDs

            do_valid_C <= do_valid_B;

            do_valid_D <= do_valid_C;

            do_valid_o_reg <= do_valid_next;                -- registered output pulse

            --------------------------------

            -- di_req_reg -> di_req_o_reg

            di_req_o_A <= di_req_reg;                       -- the input signal must be at least 2 clocks long

            di_req_o_B <= di_req_o_A;                       -- feed it to a ripple chain of FFDs

            di_req_o_C <= di_req_o_B;

            di_req_o_D <= di_req_o_C;

            di_req_o_reg <= di_req_o_next;                  -- registered output pulse

        end if;

        -- generate a 2-clocks pulse at the 3rd clock cycle

        do_valid_next <= do_valid_A and do_valid_B and not do_valid_D;

        di_req_o_next <= di_req_o_A and di_req_o_B and not di_req_o_D;

    end process out_transfer_proc;

    -- parallel load input registers: data register and write enable

    in_transfer_proc: process ( pclk_i, wren_i, wr_ack_reg ) is

    begin

        -- registered data input, input register with clock enable

        if pclk_i'event and pclk_i = '1' then

            if wren_i = '1' then

                di_reg <= di_i;                             -- parallel data input buffer register

            end if;

        end  if;

        -- stretch wren pulse to be detected by spi fsm (ffd with sync preset and sync reset)

        if pclk_i'event and pclk_i = '1' then

            if wren_i = '1' then                            -- wren_i is the sync preset for wren

                wren <= '1';

            elsif wr_ack_reg = '1' then                     -- wr_ack is the sync reset for wren

                wren <= '0';

            end if;

        end  if;

    end process in_transfer_proc;

    --=============================================================================================

    --  REGISTER TRANSFER PROCESSES

    --=============================================================================================

    -- fsm state and data registers: synchronous to the spi base reference clock

    core_reg_proc : process (sclk_i) is

    begin

        -- FF registers clocked on rising edge and cleared on sync rst_i

        if sclk_i'event and sclk_i = '1' then

            if rst_i = '1' then                             -- sync reset

                state_reg <= 0;                             -- only provide local reset for the state machine

            elsif fsm_ce = '1' then                         -- fsm_ce is clock enable for the fsm

                state_reg <= state_next;                    -- state register

            end if;

        end if;

        -- FF registers clocked synchronous to the fsm state

        if sclk_i'event and sclk_i = '1' then

            if fsm_ce = '1' then

                sh_reg <= sh_next;                          -- shift register

                ssel_ena_reg <= ssel_ena_next;              -- spi select enable

                do_buffer_reg <= do_buffer_next;            -- registered output data buffer 

                do_transfer_reg <= do_transfer_next;        -- output data transferred to buffer

                di_req_reg <= di_req_next;                  -- input data request

                wr_ack_reg <= wr_ack_next;                  -- write acknowledge for data load synchronization

            end if;

        end if;

        -- FF registers clocked one-half cycle earlier than the fsm state

        if sclk_i'event and sclk_i = '1' then

            if sck_ena_ce = '1' then

                sck_ena_reg <= sck_ena_next;                -- spi clock enable: look ahead logic

            end if;

        end if;

    end process core_reg_proc;

    --=============================================================================================

    --  COMBINATORIAL LOGIC PROCESSES

    --=============================================================================================

    -- state and datapath combinatorial logic

    core_combi_proc : process ( sh_reg, state_reg, rx_bit_reg, ssel_ena_reg, sck_ena_reg, do_buffer_reg,

                                do_transfer_reg, wr_ack_reg, di_req_reg, di_reg, wren ) is

    begin

        sh_next <= sh_reg;                                              -- all output signals are assigned to (avoid latches)

        ssel_ena_next <= ssel_ena_reg;                                  -- controls the slave select line

        sck_ena_next <= sck_ena_reg;                                    -- controls the clock enable of spi sck line

        do_buffer_next <= do_buffer_reg;                                -- output data buffer

        do_transfer_next <= do_transfer_reg;                            -- output data flag

        wr_ack_next <= wr_ack_reg;                                      -- write acknowledge

        di_req_next <= di_req_reg;                                      -- prefetch data request

        spi_mosi_o <= sh_reg(N-1);                                      -- default to avoid latch inference

        state_next <= state_reg;                                        -- next state 

        case state_reg is

            when (N+1) =>                                               -- this state is to enable SSEL before SCK

                -- slave select

                spi_mosi_o <= sh_reg(N-1);                              -- shift out tx bit from the MSb

                ssel_ena_next <= '1';                                   -- tx in progress: will assert SSEL

                sck_ena_next <= '1';                                    -- enable SCK on next cycle (stays off on first SSEL clock cycle)

                di_req_next <= '0';                                     -- prefetch data request: deassert when shifting data

                wr_ack_next <= '0';                                     -- remove write acknowledge for all but the load stages

                state_next <= state_reg - 1;                            -- update next state at each sck pulse

            when (N) =>                                                 -- deassert 'di_rdy'

                -- stretch do_valid

                spi_mosi_o <= sh_reg(N-1);                              -- shift out tx bit from the MSb

                di_req_next <= '0';                                     -- prefetch data request: deassert when shifting data

                sh_next(N-1 downto 1) <= sh_reg(N-2 downto 0);          -- shift inner bits

                sh_next(0) <= rx_bit_reg;                               -- shift in rx bit into LSb

                wr_ack_next <= '0';                                     -- remove write acknowledge for all but the load stages

                state_next <= state_reg - 1;                            -- update next state at each sck pulse

            when (N-1) downto (PREFETCH+3) =>                           -- if rx data is valid, raise 'do_valid'. remove 'do_transfer'

                -- send bit out and shif bit in

                spi_mosi_o <= sh_reg(N-1);                              -- shift out tx bit from the MSb

                di_req_next <= '0';                                     -- prefetch data request: deassert when shifting data

                do_transfer_next <= '0';                                -- reset transfer signal

                sh_next(N-1 downto 1) <= sh_reg(N-2 downto 0);          -- shift inner bits

                sh_next(0) <= rx_bit_reg;                               -- shift in rx bit into LSb

                wr_ack_next <= '0';                                     -- remove write acknowledge for all but the load stages

                state_next <= state_reg - 1;                            -- update next state at each sck pulse

            when (PREFETCH+2) downto 2 =>                               -- raise prefetch 'di_req_o_next' signal and remove 'do_valid'

                -- raise data prefetch request

                spi_mosi_o <= sh_reg(N-1);                              -- shift out tx bit from the MSb

                di_req_next <= '1';                                     -- request data in advance to allow for pipeline delays

                sh_next(N-1 downto 1) <= sh_reg(N-2 downto 0);          -- shift inner bits

                sh_next(0) <= rx_bit_reg;                               -- shift in rx bit into LSb

                wr_ack_next <= '0';                                     -- remove write acknowledge for all but the load stages

                state_next <= state_reg - 1;                            -- update next state at each sck pulse

            when 1 =>                                                   -- transfer rx data to do_buffer and restart if wren

                -- load next word or end transmission

                spi_mosi_o <= sh_reg(N-1);                              -- shift out tx bit from the MSb

                di_req_next <= '1';                                     -- request data in advance to allow for pipeline delays

                do_buffer_next(N-1 downto 1) <= sh_reg(N-2 downto 0);   -- shift rx data directly into rx buffer

                do_buffer_next(0) <= rx_bit_reg;                        -- shift last rx bit into rx buffer

                do_transfer_next <= '1';                                -- signal transfer to do_buffer

                if wren = '1' then                                      -- load tx register if valid data present at di_i

                    state_next <= N;                                    -- next state is top bit of new data

                    sh_next <= di_reg;                                  -- load parallel data from di_reg into shifter

                    sck_ena_next <= '1';                                -- SCK enabled

                    wr_ack_next <= '1';                                 -- acknowledge data in transfer

                else

                    sck_ena_next <= '0';                                -- SCK disabled: tx empty, no data to send

                    wr_ack_next <= '0';                                 -- remove write acknowledge for all but the load stages

                    state_next <= state_reg - 1;                        -- update next state at each sck pulse

                end if;

            when 0 =>

                -- idle state: start and end of transmission

                di_req_next <= '1';                                     -- will request data if shifter empty

                sck_ena_next <= '0';                                    -- SCK disabled: tx empty, no data to send

                if wren = '1' then                                      -- load tx register if valid data present at di_i

                    spi_mosi_o <= di_reg(N-1);                          -- special case: shift out first tx bit from the MSb (look ahead)

                    ssel_ena_next <= '1';                               -- enable interface SSEL

                    state_next <= N+1;                                  -- start from idle: let one cycle for SSEL settling

                    sh_next <= di_reg;                                  -- load bits from di_reg into shifter

                    wr_ack_next <= '1';                                 -- acknowledge data in transfer

                else

                    spi_mosi_o <= sh_reg(N-1);                          -- shift out tx bit from the MSb

                    ssel_ena_next <= '0';                               -- deassert SSEL: interface is idle

                    wr_ack_next <= '0';                                 -- remove write acknowledge for all but the load stages

                    state_next <= 0;                                    -- when idle, keep this state

                end if;

            when others =>

                state_next <= 0;                                        -- state 0 is safe state

        end case;

    end process core_combi_proc;

    --=============================================================================================

    --  OUTPUT LOGIC PROCESSES

    --=============================================================================================

    -- data output processes

    spi_ssel_o_proc:    spi_ssel_o <= not ssel_ena_reg;                 -- active-low slave select line 

    do_o_proc:          do_o <= do_buffer_reg;                          -- parallel data out

    do_valid_o_proc:    do_valid_o <= do_valid_o_reg;                   -- data out valid

    di_req_o_proc:      di_req_o <= di_req_o_reg;                       -- input data request for next cycle

    wr_ack_o_proc:      wr_ack_o <= wr_ack_reg;                         -- write acknowledge

    -----------------------------------------------------------------------------------------------

    -- SCK out logic: pipeline phase compensation for the SCK line

    -----------------------------------------------------------------------------------------------

    -- This is a MUX with an output register. 

    -- The register gives us a pipeline delay for the SCK line, pairing with the state machine moore 

    -- output pipeline delay for the MOSI line, and thus enabling higher SCK frequency. 

    spi_sck_o_gen_proc : process (sclk_i, sck_ena_reg, spi_clk, spi_clk_reg) is

    begin

        if sclk_i'event and sclk_i = '1' then

            if sck_ena_reg = '1' then

                spi_clk_reg <= spi_clk;                                 -- copy the selected clock polarity

            else

                spi_clk_reg <= CPOL;                                    -- when clock disabled, set to idle polarity

            end if;

        end if;

        spi_sck_o <= spi_clk_reg;                                       -- connect register to output

    end process spi_sck_o_gen_proc;

    --=============================================================================================

    --  DEBUG LOGIC PROCESSES

    --=============================================================================================

    -- these signals are useful for verification, and can be deleted after debug.

    do_transfer_proc:   do_transfer_o <= do_transfer_reg;

    state_dbg_proc:     state_dbg_o <= std_logic_vector(to_unsigned(state_reg, 4));

    rx_bit_reg_proc:    rx_bit_reg_o <= rx_bit_reg;

    wren_o_proc:        wren_o <= wren;

    sh_reg_dbg_proc:    sh_reg_dbg_o <= sh_reg;

    core_clk_o_proc:    core_clk_o <= core_clk;

    core_n_clk_o_proc:  core_n_clk_o <= core_n_clk;

    core_ce_o_proc:     core_ce_o <= core_ce;

    core_n_ce_o_proc:   core_n_ce_o <= core_n_ce;

    sck_ena_o_proc:     sck_ena_o <= sck_ena_reg;

    sck_ena_ce_o_proc:  sck_ena_ce_o <= sck_ena_ce;

end architecture rtl;