3 not in any way intended for production use. this runs a FSM that:
5 * reads the Program Counter from StateRegs
6 * reads an instruction from a fixed-size Test Memory
7 * issues it to the Simple Core
8 * waits for it to complete
10 * does it all over again
12 the purpose of this module is to verify the functional correctness
13 of the Function Units in the absolute simplest and clearest possible
14 way, and to at provide something that can be further incrementally
18 from nmigen
import (Elaboratable
, Module
, Signal
, ClockSignal
, ResetSignal
,
19 ClockDomain
, DomainRenamer
)
20 from nmigen
.cli
import rtlil
21 from nmigen
.cli
import main
24 from soc
.decoder
.power_decoder
import create_pdecode
25 from soc
.decoder
.power_decoder2
import PowerDecode2
26 from soc
.decoder
.decode2execute1
import Data
27 from soc
.experiment
.testmem
import TestMemory
# test only for instructions
28 from soc
.regfile
.regfiles
import StateRegs
29 from soc
.simple
.core
import NonProductionCore
30 from soc
.config
.test
.test_loadstore
import TestMemPspec
31 from soc
.config
.ifetch
import ConfigFetchUnit
32 from soc
.decoder
.power_enums
import MicrOp
33 from soc
.debug
.dmi
import CoreDebug
, DMIInterface
34 from soc
.config
.state
import CoreState
36 from nmutil
.util
import rising_edge
39 class TestIssuer(Elaboratable
):
40 """TestIssuer - reads instructions from TestMemory and issues them
42 efficiency and speed is not the main goal here: functional correctness is.
44 def __init__(self
, pspec
):
45 # main instruction core
46 self
.core
= core
= NonProductionCore(pspec
)
49 pdecode
= create_pdecode()
50 self
.pdecode2
= PowerDecode2(pdecode
) # decoder
52 # Test Instruction memory
53 self
.imem
= ConfigFetchUnit(pspec
).fu
54 # one-row cache of instruction read
55 self
.iline
= Signal(64) # one instruction line
56 self
.iprev_adr
= Signal(64) # previous address: if different, do read
59 self
.dbg
= CoreDebug()
61 # instruction go/monitor
62 self
.pc_o
= Signal(64, reset_less
=True)
63 self
.pc_i
= Data(64, "pc_i") # set "ok" to indicate "please change me"
64 self
.core_bigendian_i
= Signal()
65 self
.busy_o
= Signal(reset_less
=True)
66 self
.memerr_o
= Signal(reset_less
=True)
68 # FAST regfile read /write ports for PC and MSR
69 self
.state_r_pc
= self
.core
.regs
.rf
['state'].r_ports
['cia'] # PC rd
70 self
.state_w_pc
= self
.core
.regs
.rf
['state'].w_ports
['d_wr1'] # PC wr
71 self
.state_r_msr
= self
.core
.regs
.rf
['state'].r_ports
['msr'] # MSR rd
73 # DMI interface access
74 intrf
= self
.core
.regs
.rf
['int']
75 self
.int_r
= intrf
.r_ports
['dmi'] # INT read
77 # hack method of keeping an eye on whether branch/trap set the PC
78 self
.state_nia
= self
.core
.regs
.rf
['state'].w_ports
['nia']
79 self
.state_nia
.wen
.name
= 'state_nia_wen'
81 def elaborate(self
, platform
):
83 comb
, sync
= m
.d
.comb
, m
.d
.sync
85 m
.submodules
.core
= core
= DomainRenamer("coresync")(self
.core
)
86 m
.submodules
.imem
= imem
= self
.imem
87 m
.submodules
.dbg
= dbg
= self
.dbg
90 pdecode
= create_pdecode()
91 m
.submodules
.dec2
= pdecode2
= self
.pdecode2
96 intrf
= self
.core
.regs
.rf
['int']
98 # clock delay power-on reset
99 cd_por
= ClockDomain(reset_less
=True)
100 cd_sync
= ClockDomain()
101 core_sync
= ClockDomain("coresync")
102 m
.domains
+= cd_por
, cd_sync
, core_sync
104 delay
= Signal(range(4), reset
=1)
105 with m
.If(delay
!= 0):
106 m
.d
.por
+= delay
.eq(delay
- 1)
107 comb
+= cd_por
.clk
.eq(ClockSignal())
108 comb
+= core_sync
.clk
.eq(ClockSignal())
109 # XXX TODO: power-on reset delay (later)
110 #comb += core.core_reset_i.eq(delay != 0 | dbg.core_rst_o)
112 # busy/halted signals from core
113 comb
+= self
.busy_o
.eq(core
.busy_o
)
114 comb
+= pdecode2
.dec
.bigendian
.eq(self
.core_bigendian_i
)
116 # current state (MSR/PC at the moment
117 cur_state
= CoreState("cur")
119 # temporary hack: says "go" immediately for both address gen and ST
121 ldst
= core
.fus
.fus
['ldst0']
122 st_go_edge
= rising_edge(m
, ldst
.st
.rel_o
)
123 m
.d
.comb
+= ldst
.ad
.go_i
.eq(ldst
.ad
.rel_o
) # link addr-go direct to rel
124 m
.d
.comb
+= ldst
.st
.go_i
.eq(st_go_edge
) # link store-go to rising rel
126 # PC and instruction from I-Memory
127 pc_changed
= Signal() # note write to PC
128 comb
+= self
.pc_o
.eq(cur_state
.pc
)
131 # next instruction (+4 on current)
132 nia
= Signal(64, reset_less
=True)
133 comb
+= nia
.eq(cur_state
.pc
+ 4)
136 pc
= Signal(64, reset_less
=True)
137 with m
.If(self
.pc_i
.ok
):
138 # incoming override (start from pc_i)
139 comb
+= pc
.eq(self
.pc_i
.data
)
141 # otherwise read StateRegs regfile for PC
142 comb
+= self
.state_r_pc
.ren
.eq(1<<StateRegs
.PC
)
143 comb
+= pc
.eq(self
.state_r_pc
.data_o
)
145 # don't write pc every cycle
146 sync
+= self
.state_w_pc
.wen
.eq(0)
147 sync
+= self
.state_w_pc
.data_i
.eq(0)
149 # don't read msr every cycle
150 sync
+= self
.state_r_msr
.ren
.eq(0)
152 # connect up debug signals
153 # TODO comb += core.icache_rst_i.eq(dbg.icache_rst_o)
154 comb
+= core
.core_stopped_i
.eq(dbg
.core_stop_o
)
155 comb
+= core
.core_reset_i
.eq(dbg
.core_rst_o
)
156 comb
+= dbg
.terminate_i
.eq(core
.core_terminate_o
)
157 comb
+= dbg
.state
.pc
.eq(pc
)
158 comb
+= dbg
.state
.msr
.eq(cur_state
.msr
)
161 core_busy_o
= core
.busy_o
# core is busy
162 core_ivalid_i
= core
.ivalid_i
# instruction is valid
163 core_issue_i
= core
.issue_i
# instruction is issued
164 dec_opcode_i
= pdecode2
.dec
.raw_opcode_in
# raw opcode
166 insn_type
= pdecode2
.e
.do
.insn_type
167 insn_state
= pdecode2
.state
169 # actually use a nmigen FSM for the first time (w00t)
170 # this FSM is perhaps unusual in that it detects conditions
171 # then "holds" information, combinatorially, for the core
172 # (as opposed to using sync - which would be on a clock's delay)
173 # this includes the actual opcode, valid flags and so on.
177 with m
.State("IDLE"):
178 sync
+= pc_changed
.eq(0)
180 with m
.If(~dbg
.core_stop_o
):
181 # instruction allowed to go: start by reading the PC
182 # capture the PC and also drop it into Insn Memory
183 # we have joined a pair of combinatorial memory
184 # lookups together. this is Generally Bad.
185 comb
+= self
.imem
.a_pc_i
.eq(pc
)
186 comb
+= self
.imem
.a_valid_i
.eq(1)
187 comb
+= self
.imem
.f_valid_i
.eq(1)
188 sync
+= cur_state
.pc
.eq(pc
)
190 # read MSR, latch it, and put it in decode "state"
191 sync
+= self
.state_r_msr
.ren
.eq(1<<StateRegs
.MSR
)
192 sync
+= cur_state
.msr
.eq(self
.state_r_msr
.data_o
)
194 m
.next
= "INSN_READ" # move to "wait for bus" phase
196 # dummy pause to find out why simulation is not keeping up
197 with m
.State("INSN_READ"):
198 with m
.If(self
.imem
.f_busy_o
): # zzz...
199 # busy: stay in wait-read
200 comb
+= self
.imem
.a_valid_i
.eq(1)
201 comb
+= self
.imem
.f_valid_i
.eq(1)
203 # not busy: instruction fetched
204 f_instr_o
= self
.imem
.f_instr_o
205 if f_instr_o
.width
== 32:
208 insn
= f_instr_o
.word_select(cur_state
.pc
[2], 32)
209 comb
+= dec_opcode_i
.eq(insn
) # actual opcode
210 sync
+= core
.e
.eq(pdecode2
.e
)
211 sync
+= ilatch
.eq(insn
) # latch current insn
212 m
.next
= "INSN_START" # move to "start"
214 # waiting for instruction bus (stays there until not busy)
215 with m
.State("INSN_START"):
216 comb
+= core_ivalid_i
.eq(1) # instruction is valid
217 comb
+= core_issue_i
.eq(1) # and issued
218 comb
+= dec_opcode_i
.eq(ilatch
) # actual opcode
220 # also drop PC and MSR into decode "state"
221 comb
+= insn_state
.eq(cur_state
)
223 m
.next
= "INSN_ACTIVE" # move to "wait completion"
225 # instruction started: must wait till it finishes
226 with m
.State("INSN_ACTIVE"):
227 with m
.If(insn_type
!= MicrOp
.OP_NOP
):
228 comb
+= core_ivalid_i
.eq(1) # instruction is valid
229 comb
+= dec_opcode_i
.eq(ilatch
) # actual opcode
230 comb
+= insn_state
.eq(cur_state
) # and MSR and PC
231 with m
.If(self
.state_nia
.wen
):
232 sync
+= pc_changed
.eq(1)
233 with m
.If(~core_busy_o
): # instruction done!
234 # ok here we are not reading the branch unit. TODO
235 # this just blithely overwrites whatever pipeline
237 with m
.If(~pc_changed
):
238 sync
+= self
.state_w_pc
.wen
.eq(1<<StateRegs
.PC
)
239 sync
+= self
.state_w_pc
.data_i
.eq(nia
)
241 m
.next
= "IDLE" # back to idle
243 # this bit doesn't have to be in the FSM: connect up to read
244 # regfiles on demand from DMI
246 with m
.If(d_reg
.req
): # request for regfile access being made
247 # TODO: error-check this
248 # XXX should this be combinatorial? sync better?
250 comb
+= self
.int_r
.ren
.eq(1<<d_reg
.addr
)
252 comb
+= self
.int_r
.addr
.eq(d_reg
.addr
)
253 comb
+= self
.int_r
.ren
.eq(1)
254 comb
+= d_reg
.data
.eq(self
.int_r
.data_o
)
255 comb
+= d_reg
.ack
.eq(1)
260 yield from self
.pc_i
.ports()
263 yield from self
.core
.ports()
264 yield from self
.imem
.ports()
265 yield self
.core_bigendian_i
271 def external_ports(self
):
272 return self
.pc_i
.ports() + [self
.pc_o
,
274 self
.core_bigendian_i
,
279 list(self
.dbg
.dmi
.ports()) + \
280 list(self
.imem
.ibus
.fields
.values()) + \
281 list(self
.core
.l0
.cmpi
.lsmem
.lsi
.dbus
.fields
.values())
287 if __name__
== '__main__':
288 units
= {'alu': 1, 'cr': 1, 'branch': 1, 'trap': 1, 'logical': 1,
294 pspec
= TestMemPspec(ldst_ifacetype
='bare_wb',
295 imem_ifacetype
='bare_wb',
300 dut
= TestIssuer(pspec
)
301 vl
= main(dut
, ports
=dut
.ports(), name
="test_issuer")
303 if len(sys
.argv
) == 1:
304 vl
= rtlil
.convert(dut
, ports
=dut
.external_ports(), name
="test_issuer")
305 with
open("test_issuer.il", "w") as f
: