URL
https://opencores.org/ocsvn/jart/jart/trunk
Subversion Repositories jart
[/] [jart/] [trunk/] [BL02/] [block02.vhd] - Rev 87
Go to most recent revision | Compare with Previous | Blame | View Log
-- Author : Julian Andres Guarin Reyes. -- Project : JART, Just Another Ray Tracer. -- email : jguarin2002 at gmail.com, j.guarin at javeriana.edu.co -- This code was entirely written by Julian Andres Guarin Reyes. -- The following code is licensed under GNU Public License -- http://www.gnu.org/licenses/gpl-3.0.txt. -- This file is part of JART (Just Another Ray Tracer). -- JART (Just Another Ray Tracer) is free software: you can redistribute it and/or modify -- it under the terms of the GNU General Public License as published by -- the Free Software Foundation, either version 3 of the License, or -- (at your option) any later version. -- JART (Just Another Ray Tracer) 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 General Public License for more details. -- You should have received a copy of the GNU General Public License -- along with JART (Just Another Ray Tracer). If not, see <http://www.gnu.org/licenses/>. -- ***************************************************************************************** -- This Block is the entire ray sphere intersection unit. An intersection unit is governed by the expression: -- V.D +- ((V.D)^2 - (V^2-R^2))^2 -- So theres a root square.... if this root square has a non complex value, then a ray sphere intersection exists. -- After several ray tracing algorithm performance simulations, I found that the most repeated operation is dot product and comparison. This are two operations that are performed when solving the root square inner expression. This expression is called discriminant. -- That result can be interpreted as follows : the performance improvement or decay comes through when more or less discriminants are solved by time unit. -- We assume that V^2 - R^2 is a contant because V is the position of the shpere in terms of the observer and R is the sphere's radius neither of those change in the frame calculation, we are going to call the root square of this expression the constant Ks. -- We have then the inequality: -- (V.D)^2 - (V^2-R^2) >= 0 (1) -- (V.D)^2>=(V^2-R^2) (2) -- (V.D)>=(V^2-R ^2)^(1/2) (3) -- (V.D)>=Ks (4) -- Assuming that there are more "no intersections" than "intersections" for all the V.D>=Ks pairs, is then a good stategy to use the (4) inequality instead of the (2) inequality because this way thre is a saving in time and silicon space, due of the lack of need an additional multiplication V.D x V.D. -- ****************************************************************************************************** -- Using the altera quartus ii simulator, it was concluded that a combinatorial propagation time is larger than a 50 MHz frecuency period of about 25 - 32 ns. The obvious solution was to split the dot product and the comparison in a two stage pipe: -- (V.D)-> FF-> (V.D>=Ks?)-> 1/0 -- Stage1 Stage2 Result. -- This Block of code is devoted to instantiate the dot product, the comparison and to separate this operations in a two stage pipeline library ieee; use ieee.std_logic_1164.all; entity bl02 is port ( -- Global control. 50 MHZ clk. rst: in std_logic; clk: in std_logic; -- Centro Esfera Vx : in std_logic_vector (17 downto 0); Vy : in std_logic_vector (17 downto 0); Vz : in std_logic_vector (17 downto 0); -- Direccion Rayo Dx : in std_logic_vector (17 downto 0); Dy : in std_logic_vector (17 downto 0); Dz : in std_logic_vector (17 downto 0); -- Sqrt(V² - r²) K : in std_logic_vector (31 downto 0); -- Producto punto VD : out std_logic_vector (31 downto 0); -- Interseccion Iok : out std_logic ); end entity; architecture rtl of bl02 is -- Component Headers (perhaps they should be on a package file). component bl00 port ( -- <vx, vy, vz> y <dx, dy, dz> fixed point A(10,7) => 18 bits de representacion. vx_A7_10 : in std_logic_vector (17 downto 0); vy_A7_10 : in std_logic_vector (17 downto 0); vz_A7_10 : in std_logic_vector (17 downto 0); dx_A7_10 : in std_logic_vector (17 downto 0); dy_A7_10 : in std_logic_vector (17 downto 0); dz_A7_10 : in std_logic_vector (17 downto 0); -- <vxdx + vydy + vzdz> fixed point A(23,8) => 32 bits de representacion. vd_A15_16 : out std_logic_vector (31 downto 0) ); end component; component bl01 port ( vd : in std_logic_vector(31 downto 0); k : in std_logic_vector(31 downto 0); i : out std_logic ); end component; -- V.D pipe signals in order to achieve 50 MHZ. signal vd_d : std_logic_vector (31 downto 0); signal vd_q : std_logic_vector (31 downto 0); begin --Stage 1 : Instantiate Dot Product. dotprod : bl00 port map( vx_A7_10 => Vx, vy_A7_10 => Vy, vz_A7_10 => Vz, dx_A7_10 => Dx, dy_A7_10 => Dy, dz_A7_10 => Dz, vd_A15_16 => vd_d ); --Stage 2 : Instantiate Comparison. compare : bl01 port map(vd => vd_q, k => K, i => Iok); pipe : process (rst,clk) begin if rst = '1' then -- De esta manera siempre sera el numero mas pequeño cuando haya reset. vd_q (30 downto 0) <= (others => '0'); vd_q (31) <= '1'; elsif rising_edge(clk) then -- Pipe. vd_q <= vd_d; end if; end process; -- Conectar el Pipe 0 a la salida en VD. VD <= vd_d; end rtl;
Go to most recent revision | Compare with Previous | Blame | View Log