= Generic Architectures = == The concept == `Generic Architecture` means we will define some SoC hardware architecture with few variable parameters. The main goal is to change number of Cpus or memory banks, but not main architecture like interconnect choice. == Implementation == An architecture is defined in a python class inherited from `Architecture` (defined in `soclib`). Method `architecture` will create contents of the SoC: components creation, connections, utilities (like MappingTable). You'll have to tell the architecture framework where is the lowest-level component, most probably the base interconnect. This must be archieved by calling `self.setBase(component)`. == Parameters == Anywhere in `architecture` method, you may call `self.getParam(name, [default value])` to get an architecture instance parameter. * name is the parameter name * default value is optionnal * user-specified value overrides default value * if default value is not given when calling `getParam`, user-specified argument becomes mandatory. == Examples == === Basic example === We may define a basic architecture with no parameters: {{{ class OneLevel(GenericArchitecture): def architecture(self): vgmn = Vgmn( ’vgmn’ ) mips0 = Mips( ’mips0’ ) cache0 = Xcache( ’cache0’ ) # Cache ports connexion cache0.cache( mips0.cache ) vgmn.getTarget()( cache0.vci ) reset = Segment( ’reset’, address = 0xbfc00000, type = Cached ) code = Segment( ’code’, type = Cached ) data = Segment( ’data’, type = Uncached ) ram = MultiRam( ’ram’, reset, code, data ) # Connections vgmn.getInit()( ram.vci ) # Declare base component self.setBase( vgmn ) # Add configuration utilities self.setTimeBase() self.setConfig(’mapping_table’, MappingTable()) }}} As you may want to refer to some objects later on, you may set attributes to self: {{{ ... self.ram = MultiRam( ’ram’, reset, code, data ) }}} === Parametrized example === We will define a one-level architecture using a Vgmn, with two parameters: number of Cpus (`ncpu`) and number of Ram components (`nram`). By default, we will set `nram` to 1. We'll add in ram0 (which always exists) two more segments. We'll set those attributes: * `vgmn`: the interconnect * `cpu`: an array of all the cpus * `ram`: an array of all the ram chips * `cram`: an array of all the cached ram segments * `uram`: an array of all the uncached ram segments This way, we'll be able to refer to `platform.ram[2]` or `platform.cpu[0]`. {{{ class Parametrized(GenericArchitecture): defaults = { 'nram': 1, } def architecture(self): self.vgmn = Vgmn( ’vgmn’ ) self.cpu = [] for i in range(self.getParam('ncpu')): cpu = Mips( ’mips%d’%i ) cache = Xcache( ’cache%d’%i ) # Cache ports connexion cache.cache( cpu.cache ) vgmn.getTarget()( cache.vci ) self.cpu.append( cpu ) self.ram = [] self.cram = [] self.uram = [] for i in range(self.getParam('nram')): cram = Segment( ’cram%d’%i, type = Cached ) uram = Segment( ’uram%d’%i, type = Uncached ) ram = MultiRam( ’ram%d’%i, cram, uram) vgmn.getInit()( ram.vci ) self.ram.append( ram ) self.cram.append( cram ) self.uram.append( uram ) self.reset = Segment( ’reset', type = Cached ) self.excep = Segment( ’excep', type = Cached ) self.ram[0].addSegment( self.reset, self.excep ) self.setBase( self.vgmn ) self.setTimeBase() self.setConfig(’mapping_table’, MappingTable()) }}}