Subversion Repositories plasma

<b><font size=+2 face="Helvetica, Arial"

color=#bf0000>Project Name: MIPS-lite core</font></b>

<p>

<font size=+1><b>Description</b></font>

<P>

MIPS-lite is a "clean room" VHDL implementation of a MIPS CPU.

It supports a simplified MIPS III+ instruction set with a two-stage pipeline.

Only User Mode instructions are supported.

<p>

<font size=+1><b>Block Diagram</b></font>

<p>

<center>

<img src="cpu.gif">

</center>

<br>

<font size=+1><b>Example Instruction</b></font>

<p>

As an example, an ADD instruction would take the following steps:

<ol>

<li>The "pc_next" entity would pass the program

       counter (PC) to the "mem_ctrl" entity. [First Stage of Pipeline]</li>

<li>"Mem_ctrl" passes the opcode to the "control" entity.</li>

<li>"Control" converts the 32-bit opcode to a 60-bit VLWI opcode

       and sends control signals to the other entities.</li>

<li>Based on the rs_index and rt_index control signals, "reg_bank"

       sends the 32-bit reg_source and reg_target to "bus_mux".</li>

<li>Based on the a_source and b_source control signals, "bus_mux"

       multiplexes reg_source onto a_bus and reg_target onto b_bus.</li>

<li>Based on the alu_func control signals, "alu" adds the values

       from a_bus and b_bus and places the result on c_bus.</li>

<li>Based on the c_source control signals, "bus_bux" multiplexes

       c_bus onto reg_dest.</li>

<li>Based on the rd_index control signal, "reg_bank" saves

       reg_dest into the correct register.</li>

</ol>

<font size=+1><b>Features</b></font>

<p>

The CPU is implemented as a two-stage pipeline with step #1 in the

first stage and steps #2-8 occurring the second stage.

Each instruction takes one clock cycle, except memory accesses,

which take two clock cycles, and multiplication and division, which

can be accessed in 32 clock cycles.

<br>

<br>

There are several control lines not shown in the diagram.

A pause (wait-state) line will cause the pipeline to pause

if the multiplication results are accessed before the

multiplication is complete.

<p>

<font size=+1><b>Supporting Documentation</b></font>

<p>

The implementation is based on information found in:

<ul>

<li>"MIPS RISC Architecture" by Gerry Kane and Joe Heinrich and</li>

<li>"The Designer's Guide to VHDL" by Peter J. Ashenden</li>

</ul>

<font size=+1><b>Tools</b></font>

<p>

The tools used include VHDL Synopsys, ModelTech, and the Microsoft

MIPS C compiler.

<p>

<font size=+1><b>Registers</b></font>

<p>

All of the registers are clocked by the single master clock.

The registers used in the design are grouped by entity and listed below:

<pre width=80>

     mem_ctrl

     ===========================================

     |    Register Name    |   Type    | Width |

     ===========================================

     | next_opcode_reg_reg | Flip-flop |  32   |

     |   opcode_reg_reg    | Flip-flop |  32   |

     |   setup_done_reg    | Flip-flop |   1   |

     ===========================================

     mult

     ===========================================

     |    Register Name    |   Type    | Width |

     ===========================================

     |   answer_reg_reg    | Flip-flop |  32   |

     |    count_reg_reg    | Flip-flop |   6   |

     |   do_div_reg_reg    | Flip-flop |   1   |

     |  do_signed_reg_reg  | Flip-flop |   1   |

     |      reg_a_reg      | Flip-flop |  32   |

     |      reg_b_reg      | Flip-flop |  64   |

     ===========================================

     pc_next

     ===========================================

     |    Register Name    |   Type    | Width |

     ===========================================

     |     pc_reg_reg      | Flip-flop |  30   |

     ===========================================

     reg_bank

     ===========================================

     |    Register Name    |   Type    | Width |

     ===========================================

     |      reg01_reg      | Flip-flop |  32   |

     |      reg02_reg      | Flip-flop |  32   |

     |      reg03_reg      | Flip-flop |  32   |

     |      reg04_reg      | Flip-flop |  32   |

     |      reg05_reg      | Flip-flop |  32   |

     |      reg06_reg      | Flip-flop |  32   |

     |      reg07_reg      | Flip-flop |  32   |

     |      reg08_reg      | Flip-flop |  32   |

     |      reg09_reg      | Flip-flop |  32   |

     |      reg10_reg      | Flip-flop |  32   |

     |      reg11_reg      | Flip-flop |  32   |

     |      reg12_reg      | Flip-flop |  32   |

     |      reg13_reg      | Flip-flop |  32   |

     |      reg14_reg      | Flip-flop |  32   |

     |      reg15_reg      | Flip-flop |  32   |

     |      reg16_reg      | Flip-flop |  32   |

     |      reg17_reg      | Flip-flop |  32   |

     |      reg18_reg      | Flip-flop |  32   |

     |      reg19_reg      | Flip-flop |  32   |

     |      reg20_reg      | Flip-flop |  32   |

     |      reg21_reg      | Flip-flop |  32   |

     |      reg22_reg      | Flip-flop |  32   |

     |      reg23_reg      | Flip-flop |  32   |

     |      reg24_reg      | Flip-flop |  32   |

     |      reg25_reg      | Flip-flop |  32   |

     |      reg26_reg      | Flip-flop |  32   |

     |      reg27_reg      | Flip-flop |  32   |

     |      reg28_reg      | Flip-flop |  32   |

     |      reg29_reg      | Flip-flop |  32   |

     |      reg30_reg      | Flip-flop |  32   |

     |      reg31_reg      | Flip-flop |  32   |

     |     reg_epc_reg     | Flip-flop |  32   |

     |   reg_status_reg    | Flip-flop |   1   |

     ===========================================

</pre>

<font size=+1><b>Preliminary Synthesis</b></font>

<p>

The CPU core was synthesized for 0.13 um line widths with a predicted

area less than 0.2 millimeters squared.  The predicted maximum

latency was less than 6 ns for a maximum clock speed of 150 MHz.

<br>

<br>

A preliminary synthesis yields the following cells and die area.

I think that optimization caused the mips_cpu entity

to be smaller than the sum of its

components.

If one assumes that a standard cell is composed of three gates,

then this is approximately a 20K gate design. [Is this correct??]

It is interesting to note that the register bank requires over 60% of the area.

<pre width=80>

     Block    ports   nets  cells cell_area   ~%   delay(ns)

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

     alu        101    919    850      7503   12        1.11

     bus_mux    283    672    486      4906    8        0.35

     control     93    296    263      2250    4        0.29

     mem_ctrl   271    455    318      3299    5        0.95

     mult       101   1111   1043      9342   15        0.72 ??

     pc_next     94    277    215      1756    3        0.15

     reg_bank   116   2650   2599     39477   62        1.02

     shifter     71    423    384      3026    5        1.51

     mips_cpu   201    555     45     63888  100        5.61

     total     1331   7358   6203

</pre>

<font size=+1><b>List of Files</b></font>

<p>

<ul>

<table border="2" style="border-color:Black;border-collapse:collapse;">

<tr>

 <td>FILE</td>

 <td>PURPOSE</td>

</td><tr>

</tr><tr>

 <td>makefile</td>

 <td>Makefile for the HP workstation for Synopsys</td>

</tr><tr>

 <td>code.txt</td>

 <td>Input opcodes for the test bench -- test.exe "converted"</td>

</tr><tr>

 <td>alu.vhd</td>

 <td>Arithmetic Logic Unit</td>

</tr><tr>

 <td>bus_mux.vhd</td>

 <td>BUS Multiplex Unit</td>

</tr><tr>

 <td>control.vhd</td>

 <td>Opcode Decoder</td>

</tr><tr>

 <td>mem_ctrl.vhd</td>

 <td>Memory Controller</td>

</tr><tr>

 <td>mips_cpu.vhd</td>

 <td>Top Level VHDL for MIPS CPU</td>

</tr><tr>

 <td>mips_pack.vhd</td>

 <td>Constants and Functions Package</td>

</tr><tr>

 <td>mult.vhd</td>

 <td>Multiplication and Division Unit</td>

</tr><tr>

 <td>pc_next.vhd</td>

 <td>Program Counter Unit</td>

</tr><tr>

 <td>ram.vhd</td>

 <td>RAM for the Test Bench</td>

</tr><tr>

 <td>reg_bank.vhd</td>

 <td>Register Bank for 32, 32-bit Registers</td>

</tr><tr>

 <td>shifter.vhd</td>

 <td>Shifter Unit</td>

</tr><tr>

 <td>tbench.vhd</td>

 <td>Test Bench that uses mips_vpu.vhd and ram.vhd</td>

</tr><tr>

</tr><tr>

 <td>makefile</td>

 <td>Makefile for the PC for creating "code.txt"</td>

</tr><tr>

 <td>convert.c</td>

 <td>Converts test.exe to code.txt</td>

</tr><tr>

 <td>mips.c</td>

 <td>Simulates a MIPS CPU in software</td>

</tr><tr>

 <td>test.c</td>

 <td>Test program (opcodes) for the MIPS CPU</td>

</tr><tr>

 <td>output.txt</td>

 <td>Output from the test bench</td>

</tr><tr>

 <td>index.shtml</td>

 <td>This help file</td>

</tr><tr>

 <td>cpu.gif</td>

 <td>Block Diagram</td>

</tr>

</table>

</ul>

<p>

<font size=+1><b>ZIP File</b></font>

<p>

CVS is the only way to download the latest files.  However

for a quick look at an old version you can download

<a href="mipslite.zip">MIPSlite.zip</a>.

<p>

<font size=+1><b>Convert</b></font>

<p>

The program "convert" changes the file "test.exe" into the HEX file "code.txt".

The opcodes in "test.exe" are changed to Big Endian.

All absolute jumps are changed to relative jumps.

The first opcode is also changed to set up the stack pointer.

<p>

<font size=+1><b>Big/Little Endian</b></font>

<p>

The MIPS CPU operates in Big Endian mode by default.  To operate in

Little Endian mode, change "little_endian" from "00" to "11" in

the file mem_ctrl.vhd.

<p>

<font size=+1><b>Legal Notice</b></font>

<p>

<font color="#FF0000">

MIPS is a registered trademark of MIPS Technologies, Inc.

If you use this core you are responsible for all legal issues.

This "clean room" implementation of a MIPS CPU does not negate

MIPS Technologies, Inc. of their trademark, copyrights, or patents....

<p>

Free for commercial and non-commercial use as long as the author and

warning notices are maintained.

<br>

<br>

This software is provided by Steve Rhoads "as is" and

any express or implied warranties, including, but not limited to, the

implied warranties of merchantability and fitness for a particular purpose

are disclaimed.  In no event shall the author or contributors be liable

for any direct, indirect, incidental, special, exemplary, or consequential

damages (including, but not limited to, procurement of substitute goods

or services; loss of use, data, or profits; or business interruption)

however caused and on any theory of liability, whether in contract, strict

liability, or tort (including negligence or otherwise) arising in any way

out of the use of this software, even if advised of the possibility of

such damage.

</font>

<p>

<p><b><font size="+1">Bus Interface</font></b></p>

<p>

<pre width=80>

   port(clk         : in std_logic;

        reset_in    : in std_logic;

        intr_in     : in std_logic;  --interrupt line

        --memory access buses

        mem_address : out std_logic_vector(31 downto 0);

        mem_data_w  : out std_logic_vector(31 downto 0); --avoided tri-state

        mem_data_r  : in std_logic_vector(31 downto 0);

        mem_sel     : out std_logic_vector(3 downto 0);  --byte lines

        mem_write   : out std_logic;

        mem_pause   : in std_logic

        );

</pre>

<p>

<font size=+1><b>Current Status</b></font>

<ul>

 <li>The test bench needs to be strengthened.</li>

 <li>Need feedback on the design.</li>

 <li>Need feedback on the tools.</li>

 <li>Need to add simulation of a cache.</li>

</ul>

<p>

<font size=+1><b>Maintainer</b></font>

<ul>Steve Rhoads,

<a href="mailto:rhoads@opencores.org_NOSPAM">rhoads@opencores.org_NOSPAM</a></ul>

<p>

<ul>*** I am not an experienced VHDL designer ***

Please let me know of any incorrect statements in this document.</ul>

<p>

<font size=+1><b>Mailing-list</b></font>

<p>

<ul><a href=mailto:cores@opencores.org_NOSPAM>cores@opencores.org_NOSPAM</A></ul>