Subversion Repositories astron_r2sdf_fft

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

--

-- Copyright 2020

-- ASTRON (Netherlands Institute for Radio Astronomy) <http://www.astron.nl/>

-- P.O.Box 2, 7990 AA Dwingeloo, The Netherlands

-- 

-- Licensed under the Apache License, Version 2.0 (the "License");

-- you may not use this file except in compliance with the License.

-- You may obtain a copy of the License at

-- 

--     http://www.apache.org/licenses/LICENSE-2.0

-- 

-- Unless required by applicable law or agreed to in writing, software

-- distributed under the License is distributed on an "AS IS" BASIS,

-- WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.

-- See the License for the specific language governing permissions and

-- limitations under the License.

--

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

library ieee, common_pkg_lib, common_components_lib, astron_multiplier_lib;

use IEEE.std_logic_1164.all;

--USE technology_lib.technology_select_pkg.ALL;

use common_pkg_lib.common_pkg.all;

entity rTwoWMul is

  generic (

    g_technology : NATURAL := 0;

    g_stage      : natural := 1;

    g_lat        : natural := 3+1       -- 3 for mult, 1 for round

  );

  port (

    clk       : in  std_logic;

    rst       : in  std_logic;

    weight_re : in  std_logic_vector;

    weight_im : in  std_logic_vector;

    in_re     : in  std_logic_vector;

    in_im     : in  std_logic_vector;

    in_val    : in  std_logic;

    in_sel    : in  std_logic;

    out_re    : out std_logic_vector;

    out_im    : out std_logic_vector;

    out_val   : out std_logic

  );

end entity rTwoWMul;

architecture str of rTwoWMul is

  -- Use multiplier product truncate or signed rounding (= away from zero). On hardware for Fsub in

  -- Apertif and using the WG at various frequencies at subband or between subbands it appears that

  -- using truncate or sround does not make a noticable difference in the SST. Still choose to use

  -- signed rounding to preserve zero DC.

  constant c_use_truncate     : boolean := true; --false;

  -- Derive the common_complex_mult g_pipeline_* values from g_lat. The sum c_total_lat = g_lat, so that g_lat defines

  -- the total latency from in_* to out_*.

  -- DSP multiplier IP

  constant c_dsp_mult_lat     : natural := 3;

  -- Pipeline multiplier product rounding from c_prod_w via c_round_w to c_out_dat_w

  constant c_round_lat        : natural := sel_a_b(g_lat > c_dsp_mult_lat, 1, 0);  -- allocate 1 pipeline for round

  constant c_lat              : natural := g_lat - c_round_lat;                    -- allocate remaining pipeline to multiplier

  constant c_mult_input_lat   : natural := sel_a_b(c_lat>1, 1, 0);                     -- second priority use DSP pipeline input

  constant c_mult_product_lat : natural := 0;

  constant c_mult_adder_lat   : natural := sel_a_b(c_lat>2, 1, 0);                     -- third priority use DSP internal product-sum pipeline

  constant c_mult_extra_lat   : natural := sel_a_b(c_lat>3, c_lat-3, 0);               -- remaining extra pipelining in logic

  constant c_mult_output_lat  : natural := sel_a_b(c_lat>0, 1, 0) + c_mult_extra_lat;  -- first priority use DSP pipeline output

  constant c_mult_lat         : natural := c_mult_input_lat + c_mult_product_lat + c_mult_adder_lat + c_mult_output_lat;

  -- Total input to output latency

  constant c_total_lat   : natural := c_mult_lat + c_round_lat;

  -- Quantization

  constant c_in_dat_w         : natural:= in_re'length;

  constant c_weight_w         : natural:= weight_re'length;

  constant c_prod_w           : natural:= c_in_dat_w+c_weight_w+1;

  constant c_round_w          : natural:= c_weight_w-c_sign_w;   -- the weights are normalized

  constant c_out_dat_w        : natural:= out_re'length;

  signal in_re_dly        : std_logic_vector(in_re'range);

  signal in_im_dly        : std_logic_vector(in_re'range);

  signal product_re       : std_logic_vector(c_prod_w-1 downto 0);

  signal product_im       : std_logic_vector(c_prod_w-1 downto 0);

  signal round_re         : std_logic_vector(out_re'range);

  signal round_im         : std_logic_vector(out_re'range);

  signal out_sel          : std_logic;

begin

  -- Total latency check

  ASSERT c_total_lat = g_lat

    REPORT "rTwoWMul: total pipeline error"

    SEVERITY FAILURE;

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

  -- Complex multiplication

  -- . use the common_complex_mult(rtl) for the output stage 1 because then

  --   the multiplier instance can get optimized away for the constant

  --   weight_re = 1 and weight_im = 0 inputs.

  -- . the IP in common_complex_mult(stratix4) only supports up to 18b wide

  --   inputs.

  --   . for c_lat = 0,1,2 use the RTL multiplier

  --   . for c_lat >= 3 default best use the FPGA multiplier IP block.

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

  gen_rtl : if g_stage=1 or c_in_dat_w>c_dsp_mult_w or c_lat<c_dsp_mult_lat generate

    u_CmplxMul : entity astron_multiplier_lib.common_complex_mult

    generic map (

      g_technology       => g_technology,

      g_variant          => "RTL",

      g_in_a_w           => c_in_dat_w,

      g_in_b_w           => c_weight_w,

      g_out_p_w          => c_prod_w,

      g_conjugate_b      => false,

      g_pipeline_input   => c_mult_input_lat,

      g_pipeline_product => c_mult_product_lat,

      g_pipeline_adder   => c_mult_adder_lat,

      g_pipeline_output  => c_mult_output_lat

    )

    port map (

      rst       => rst,

      clk       => clk,

      in_ar     => in_re,

      in_ai     => in_im,

      in_br     => weight_re,

      in_bi     => weight_im,

      in_val    => in_val,

      out_pr    => product_re,

      out_pi    => product_im,

      out_val   => OPEN

    );

  end generate;

  gen_ip : if g_stage>1 and c_in_dat_w<=c_dsp_mult_w and c_lat>=c_dsp_mult_lat generate

    u_cmplx_mul : entity astron_multiplier_lib.common_complex_mult

    generic map (

      g_technology       => g_technology,

      g_variant          => "IP",

      g_in_a_w           => in_re'length,

      g_in_b_w           => weight_re'length,

      g_out_p_w          => product_re'length,

      g_conjugate_b      => false,

      g_pipeline_input   => c_mult_input_lat,

      g_pipeline_product => c_mult_product_lat,

      g_pipeline_adder   => c_mult_adder_lat,

      g_pipeline_output  => c_mult_output_lat

    )

    port map (

      rst       => rst,

      clk       => clk,

      in_ar     => in_re,

      in_ai     => in_im,

      in_br     => weight_re,

      in_bi     => weight_im,

      in_val    => in_val,

      out_pr    => product_re,

      out_pi    => product_im,

      out_val   => OPEN

    );

  end generate;

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

  -- Round WMult output

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

  gen_truncate : if c_use_truncate=true GENERATE

    -- use truncate    that throws away the c_round_w lower bits as rounding function

    -- use resize_svec that keeps the c_out_dat_w lower bits to get to the output width

    gen_comb : if c_round_lat=0 generate

      round_re <= truncate_and_resize_svec(product_re, c_round_w, c_out_dat_w);

      round_im <= truncate_and_resize_svec(product_im, c_round_w, c_out_dat_w);

    end generate;

    gen_reg : if c_round_lat=1 generate

      round_re <= truncate_and_resize_svec(product_re, c_round_w, c_out_dat_w) when rising_edge(clk);

      round_im <= truncate_and_resize_svec(product_im, c_round_w, c_out_dat_w) when rising_edge(clk);

    end generate;

  end generate;

  gen_sround : if c_use_truncate=false GENERATE

    -- Use resize_svec(s_round()) instead of truncate_and_resize_svec() to have symmetrical rounding around 0

    -- Rounding takes logic due to adding 0.5 therefore need to use c_round_lat=1 to achieve timing

    gen_comb : if c_round_lat=0 generate

      ASSERT false REPORT "rTwoWMul: can probably not achieve timing for sround without pipeline" SEVERITY FAILURE;

      round_re <= RESIZE_SVEC(s_round(product_re, c_round_w), c_out_dat_w);

      round_im <= RESIZE_SVEC(s_round(product_im, c_round_w), c_out_dat_w);

    end generate;

    gen_reg : if c_round_lat=1 generate

      round_re <= RESIZE_SVEC(s_round(product_re, c_round_w), c_out_dat_w) when rising_edge(clk);

      round_im <= RESIZE_SVEC(s_round(product_im, c_round_w), c_out_dat_w) when rising_edge(clk);

    end generate;

  end generate;

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

  -- Propagate data and control signals for input/output choice at WMult output

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

  -- No need to use rst for data, because initial data value is don't care

  u_re_lat : entity common_components_lib.common_pipeline

  generic map (

    g_pipeline  => g_lat,

    g_in_dat_w  => in_re'length,

    g_out_dat_w => in_re'length

  )

  port map (

    clk     => clk,

    in_dat  => in_re,

    out_dat => in_re_dly

  );

  u_im_lat : entity common_components_lib.common_pipeline

  generic map (

    g_pipeline  => g_lat,

    g_in_dat_w  => in_im'length,

    g_out_dat_w => in_im'length

  )

  port map (

    clk     => clk,

    in_dat  => in_im,

    out_dat => in_im_dly

  );

  -- Use rst for control to ensure initial low

  u_sel_lat : entity common_components_lib.common_pipeline_sl

  generic map (

    g_pipeline => g_lat

  )

  port map (

    rst     => rst,

    clk     => clk,

    in_dat  => in_sel,

    out_dat => out_sel

  );

  u_pipeline_out_val : entity common_components_lib.common_pipeline_sl

  generic map (

    g_pipeline  => g_lat

  )

  port map (

    rst     => rst,

    clk     => clk,

    in_dat  => in_val,

    out_dat => out_val

  );

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

  -- Output real and imaginary, switch between input and product

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

  out_re <= round_re when out_sel = '1' else in_re_dly;

  out_im <= round_im when out_sel = '1' else in_im_dly;

end str;