URL
https://opencores.org/ocsvn/openarty/openarty/trunk
Subversion Repositories openarty
[/] [openarty/] [trunk/] [rtl/] [cpu/] [cpudefs.v] - Rev 20
Go to most recent revision | Compare with Previous | Blame | View Log
/////////////////////////////////////////////////////////////////////////////// // // Filename: cpudefs.v // // Project: OpenArty, an entirely open SoC based upon the Arty platform // // Purpose: Some architectures have some needs, others have other needs. // Some of my projects need a Zip CPU with pipelining, others // can't handle the timing required to get the answer from the ALU // back into the input for the ALU. As each different projects has // different needs, I can either 1) reconfigure my entire baseline prior // to building each project, or 2) host a configuration file which contains // the information regarding each baseline. This file is that // configuration file. It controls how the CPU (not the system, // peripherals, or other) is defined and implemented. Several options // are available within here, making the Zip CPU pipelined or not, // able to handle a faster clock with more stalls or a slower clock with // no stalls, etc. // // This file encapsulates those control options. // // The number of LUTs the Zip CPU uses varies dramatically with the // options defined in this file. // // // OpenArty comments: // My goal on the OpenArty is going to be using the CPU to its fullest // extent. All features shall be turned on if they exist, full pipelines, // multiplies, divides, and hopefully even the 200MHz clock. This file // reflects that purpose. // // // Creator: Dan Gisselquist, Ph.D. // Gisselquist Technology, LLC // /////////////////////////////////////////////////////////////////////////////// // // Copyright (C) 2015-2016, Gisselquist Technology, LLC // // This program is free software (firmware): 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. // // This program is distributed in the hope that it will be useful, but WITHOUT // ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or // FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License // for more details. // // License: GPL, v3, as defined and found on www.gnu.org, // http://www.gnu.org/licenses/gpl.html // // /////////////////////////////////////////////////////////////////////////////// `ifndef CPUDEFS_H `define CPUDEFS_H // // // The first couple options control the Zip CPU instruction set, and how // it handles various instructions within the set: // // // OPT_ILLEGAL_INSTRUCTION is part of a new section of code that is supposed // to recognize illegal instructions and interrupt the CPU whenever one such // instruction is encountered. The goal is to create a soft floating point // unit via this approach, that can then be replaced with a true floating point // unit. As I'm not there yet, it just catches illegal instructions and // interrupts the CPU on any such instruction--when defined. Otherwise, // illegal instructions are quietly ignored and their behaviour is ... // undefined. (Many get treated like NOOPs ...) // // I recommend setting this flag, although it can be taken out if area is // critical ... // `define OPT_ILLEGAL_INSTRUCTION // // // // OPT_MULTIPLY controls whether or not the multiply is built and included // in the ALU by default. Set this option and a parameter will be set that // includes the multiply. (This parameter may still be overridden, as with // any parameter ...) If the multiply is not included and // OPT_ILLEGAL_INSTRUCTION is set, then the multiply will create an illegal // instruction that will then trip the illegal instruction trap. // // `define OPT_MULTIPLY 1 // // // // OPT_DIVIDE controls whether or not the divide instruction is built and // included into the ZipCPU by default. Set this option and a parameter will // be set that causes the divide unit to be included. (This parameter may // still be overridden, as with any parameter ...) If the divide is not // included and OPT_ILLEGAL_INSTRUCTION is set, then the multiply will create // an illegal instruction exception that will send the CPU into supervisor // mode. // // // `define OPT_DIVIDE // // // // OPT_IMPLEMENT_FPU will (one day) control whether or not the floating point // unit (once I have one) is built and included into the ZipCPU by default. // At that time, if this option is set then a parameter will be set that // causes the floating point unit to be included. (This parameter may // still be overridden, as with any parameter ...) If the floating point unit // is not included and OPT_ILLEGAL_INSTRUCTION is set, then as with the // multiply and divide any floating point instruction will result in an illegal // instruction exception that will send the CPU into supervisor mode. // // // `define OPT_IMPLEMENT_FPU // // // // OPT_NEW_INSTRUCTION_SET controls whether or not the new instruction set // is in use. The new instruction set contains space for floating point // operations, signed and unsigned divide instructions, as well as bit reversal // and ... at least two other operations yet to be defined. The decoder alone // uses about 70 fewer LUTs, although in practice this works out to 12 fewer // when all works out in the wash. Further, floating point and divide // instructions will cause an illegal instruction exception if they are not // implemented--so software capability can be built to use these instructions // immediately, even if the hardware is not yet ready. // // This option is likely to go away in the future, obsoleting the previous // instruction set, so I recommend setting this option and switching to the // new instruction set as soon as possible. // `define OPT_NEW_INSTRUCTION_SET // // // // // // // OPT_SINGLE_FETCH controls whether or not the prefetch has a cache, and // whether or not it can issue one instruction per clock. When set, the // prefetch has no cache, and only one instruction is fetched at a time. // This effectively sets the CPU so that only one instruction is ever // in the pipeline at once, and hence you may think of this as a "kill // pipeline" option. However, since the pipelined fetch component uses so // much area on the FPGA, this is an important option to use in trimming down // used area if necessary. Hence, it needs to be maintained for that purpose. // Be aware, though, it will drop your performance by a factor between 2x and // 3x. // // We can either pipeline our fetches, or issue one fetch at a time. Pipelined // fetches are more complicated and therefore use more FPGA resources, while // single fetches will cause the CPU to stall for about 5 stalls each // instruction cycle, effectively reducing the instruction count per clock to // about 0.2. However, the area cost may be worth it. Consider: // // Slice LUTs ZipSystem ZipCPU // Single Fetching 2521 1734 // Pipelined fetching 2796 2046 // (These numbers may be dated, but should still be representative ...) // // I recommend only defining this if you "need" to, if area is tight and // speed isn't as important. Otherwise, just leave this undefined. // // `define OPT_SINGLE_FETCH // // // // The next several options are pipeline optimization options. They make no // sense in a single instruction fetch mode, hence we #ifndef them so they // are only defined if we are in a full pipelined mode (i.e. OPT_SINGLE_FETCH // is not defined). // `ifndef OPT_SINGLE_FETCH // // // // OPT_PIPELINED is the natural result and opposite of using the single // instruction fetch unit. If you are not using that unit, the ZipCPU will // be pipelined. The option is defined here more for readability than // anything else, since OPT_PIPELINED makes more sense than OPT_SINGLE_FETCH, // well ... that and it does a better job of explaining what is going on. // // In other words, leave this define alone--lest you break the ZipCPU. // `define OPT_PIPELINED // // // // OPT_TRADITIONAL_PFCACHE allows you to switch between one of two prefetch // caches. If enabled, a more traditional cache is implemented. This more // traditional cache (currently) uses many more LUTs, but it also reduces // the stall count tremendously over the alternative hacked pipeline cache. // (The traditional pfcache is also pipelined, whereas the pipeline cache // implements a windowed approach to caching.) // // If you have the fabric to support this option, I recommend including it. // `define OPT_TRADITIONAL_PFCACHE // // // // OPT_EARLY_BRANCHING is an attempt to execute a BRA statement as early // as possible, to avoid as many pipeline stalls on a branch as possible. // It's not tremendously successful yet--BRA's still suffer stalls, // but I intend to keep working on this approach until the number of stalls // gets down to one or (ideally) zero. (With the OPT_TRADITIONAL_PFCACHE, this // gets down to a single stall cycle ...) That way a "BRA" can be used as the // compiler's branch prediction optimizer: BRA's barely stall, while branches // on conditions will always suffer about 4 stall cycles or so. // // I recommend setting this flag, so as to turn early branching on. // `define OPT_EARLY_BRANCHING // // // // OPT_PIPELINED_BUS_ACCESS controls whether or not LOD/STO instructions // can take advantaged of pipelined bus instructions. To be eligible, the // operations must be identical (cannot pipeline loads and stores, just loads // only or stores only), and the addresses must either be identical or one up // from the previous address. Further, the load/store string must all have // the same conditional. This approach gains the must use, in my humble // opinion, when saving registers to or restoring registers from the stack // at the beginning/end of a procedure, or when doing a context swap. // // I recommend setting this flag, for performance reasons, especially if your // wishbone bus can handle pipelined bus accesses. // `define OPT_PIPELINED_BUS_ACCESS // // // `ifdef OPT_NEW_INSTRUCTION_SET // // // // The new instruction set also defines a set of very long instruction words. // Well, calling them "very long" instruction words is probably a misnomer, // although we're going to do it. They're really 2x16-bit instructions--- // instruction words that pack two instructions into one word. (2x14 bit // really--'cause you need a bit to note the instruction is a 2x instruction, // and then 3-bits for the condition codes ...) Set OPT_VLIW to include these // double instructions as part of the new instruction set. These allow a single // instruction to contain two instructions within. These instructions are // designed to get more code density from the instruction set, and to hopefully // take some pain off of the performance of the pre-fetch and instruction cache. // // These new instructions, however, also necessitate a change in the Zip // CPU--the Zip CPU can no longer execute instructions atomically. It must // now execute non-VLIW instructions, or VLIW instruction pairs, atomically. // This logic has been added into the ZipCPU, but it has not (yet) been // tested thoroughly. // // Oh, and the assembler, the debugger, and the object file dumper, and the // simulator all need to be updated as well .... // `define OPT_VLIW // // `endif // OPT_NEW_INSTRUCTION_SET // // `endif // OPT_SINGLE_FETCH // // // // Now let's talk about peripherals for a moment. These next two defines // control whether the DMA controller is included in the Zip System, and // whether or not the 8 accounting timers are also included. Set these to // include the respective peripherals, comment them out not to. // `define INCLUDE_DMA_CONTROLLER `define INCLUDE_ACCOUNTING_COUNTERS // // // `define DEBUG_SCOPE // `endif // CPUDEFS_H
Go to most recent revision | Compare with Previous | Blame | View Log