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 IssuerDecode2ToOperand
27 from soc
.decoder
.decode2execute1
import Data
28 from soc
.experiment
.testmem
import TestMemory
# test only for instructions
29 from soc
.regfile
.regfiles
import StateRegs
, FastRegs
30 from soc
.simple
.core
import NonProductionCore
31 from soc
.config
.test
.test_loadstore
import TestMemPspec
32 from soc
.config
.ifetch
import ConfigFetchUnit
33 from soc
.decoder
.power_enums
import MicrOp
34 from soc
.debug
.dmi
import CoreDebug
, DMIInterface
35 from soc
.config
.state
import CoreState
36 from soc
.interrupts
.xics
import XICS_ICP
, XICS_ICS
37 from soc
.bus
.simple_gpio
import SimpleGPIO
39 from nmutil
.util
import rising_edge
42 class TestIssuer(Elaboratable
):
43 """TestIssuer - reads instructions from TestMemory and issues them
45 efficiency and speed is not the main goal here: functional correctness is.
47 def __init__(self
, pspec
):
49 # add interrupt controller?
50 self
.xics
= hasattr(pspec
, "xics") and pspec
.xics
== True
52 self
.xics_icp
= XICS_ICP()
53 self
.xics_ics
= XICS_ICS()
54 self
.int_level_i
= self
.xics_ics
.int_level_i
56 # add GPIO peripheral?
57 self
.gpio
= hasattr(pspec
, "gpio") and pspec
.gpio
== True
59 self
.simple_gpio
= SimpleGPIO()
60 self
.gpio_o
= self
.simple_gpio
.gpio_o
62 # main instruction core25
63 self
.core
= core
= NonProductionCore(pspec
)
65 # instruction decoder. goes into Trap Record
66 pdecode
= create_pdecode()
67 self
.cur_state
= CoreState("cur") # current state (MSR/PC/EINT)
68 self
.pdecode2
= PowerDecode2(pdecode
, state
=self
.cur_state
,
69 opkls
=IssuerDecode2ToOperand
)
71 # Test Instruction memory
72 self
.imem
= ConfigFetchUnit(pspec
).fu
73 # one-row cache of instruction read
74 self
.iline
= Signal(64) # one instruction line
75 self
.iprev_adr
= Signal(64) # previous address: if different, do read
78 self
.dbg
= CoreDebug()
80 # instruction go/monitor
81 self
.pc_o
= Signal(64, reset_less
=True)
82 self
.pc_i
= Data(64, "pc_i") # set "ok" to indicate "please change me"
83 self
.core_bigendian_i
= Signal()
84 self
.busy_o
= Signal(reset_less
=True)
85 self
.memerr_o
= Signal(reset_less
=True)
87 # FAST regfile read /write ports for PC, MSR, DEC/TB
88 staterf
= self
.core
.regs
.rf
['state']
89 self
.state_r_pc
= staterf
.r_ports
['cia'] # PC rd
90 self
.state_w_pc
= staterf
.w_ports
['d_wr1'] # PC wr
91 self
.state_r_msr
= staterf
.r_ports
['msr'] # MSR rd
93 # DMI interface access
94 intrf
= self
.core
.regs
.rf
['int']
95 crrf
= self
.core
.regs
.rf
['cr']
96 xerrf
= self
.core
.regs
.rf
['xer']
97 self
.int_r
= intrf
.r_ports
['dmi'] # INT read
98 self
.cr_r
= crrf
.r_ports
['full_cr_dbg'] # CR read
99 self
.xer_r
= xerrf
.r_ports
['full_xer'] # XER read
101 # hack method of keeping an eye on whether branch/trap set the PC
102 self
.state_nia
= self
.core
.regs
.rf
['state'].w_ports
['nia']
103 self
.state_nia
.wen
.name
= 'state_nia_wen'
105 def elaborate(self
, platform
):
107 comb
, sync
= m
.d
.comb
, m
.d
.sync
109 m
.submodules
.core
= core
= DomainRenamer("coresync")(self
.core
)
110 m
.submodules
.imem
= imem
= self
.imem
111 m
.submodules
.dbg
= dbg
= self
.dbg
113 cur_state
= self
.cur_state
115 # XICS interrupt handler
117 m
.submodules
.xics_icp
= icp
= self
.xics_icp
118 m
.submodules
.xics_ics
= ics
= self
.xics_ics
119 comb
+= icp
.ics_i
.eq(ics
.icp_o
) # connect ICS to ICP
120 sync
+= cur_state
.eint
.eq(icp
.core_irq_o
) # connect ICP to core
122 # GPIO test peripheral
124 m
.submodules
.simple_gpio
= simple_gpio
= self
.simple_gpio
126 # connect one GPIO output to ICS bit 15 (like in microwatt soc.vhdl)
127 if self
.gpio
and self
.xics
:
128 comb
+= self
.int_level_i
[15].eq(simple_gpio
.gpio_o
[0])
130 # instruction decoder
131 pdecode
= create_pdecode()
132 m
.submodules
.dec2
= pdecode2
= self
.pdecode2
135 dmi
, d_reg
, d_cr
, d_xer
, = dbg
.dmi
, dbg
.d_gpr
, dbg
.d_cr
, dbg
.d_xer
136 intrf
= self
.core
.regs
.rf
['int']
138 # clock delay power-on reset
139 cd_por
= ClockDomain(reset_less
=True)
140 cd_sync
= ClockDomain()
141 core_sync
= ClockDomain("coresync")
142 m
.domains
+= cd_por
, cd_sync
, core_sync
144 delay
= Signal(range(4), reset
=3)
145 with m
.If(delay
!= 0):
146 m
.d
.por
+= delay
.eq(delay
- 1)
147 comb
+= cd_por
.clk
.eq(ClockSignal())
148 comb
+= core_sync
.clk
.eq(ClockSignal())
149 # power-on reset delay
150 comb
+= core
.core_reset_i
.eq(delay
!= 0 | dbg
.core_rst_o
)
152 # busy/halted signals from core
153 comb
+= self
.busy_o
.eq(core
.busy_o
)
154 comb
+= pdecode2
.dec
.bigendian
.eq(self
.core_bigendian_i
)
156 # temporary hack: says "go" immediately for both address gen and ST
158 ldst
= core
.fus
.fus
['ldst0']
159 st_go_edge
= rising_edge(m
, ldst
.st
.rel_o
)
160 m
.d
.comb
+= ldst
.ad
.go_i
.eq(ldst
.ad
.rel_o
) # link addr-go direct to rel
161 m
.d
.comb
+= ldst
.st
.go_i
.eq(st_go_edge
) # link store-go to rising rel
163 # PC and instruction from I-Memory
164 pc_changed
= Signal() # note write to PC
165 comb
+= self
.pc_o
.eq(cur_state
.pc
)
168 # next instruction (+4 on current)
169 nia
= Signal(64, reset_less
=True)
170 comb
+= nia
.eq(cur_state
.pc
+ 4)
173 pc
= Signal(64, reset_less
=True)
174 pc_ok_delay
= Signal()
175 sync
+= pc_ok_delay
.eq(~self
.pc_i
.ok
)
176 with m
.If(self
.pc_i
.ok
):
177 # incoming override (start from pc_i)
178 comb
+= pc
.eq(self
.pc_i
.data
)
180 # otherwise read StateRegs regfile for PC...
181 comb
+= self
.state_r_pc
.ren
.eq(1<<StateRegs
.PC
)
182 # ... but on a 1-clock delay
183 with m
.If(pc_ok_delay
):
184 comb
+= pc
.eq(self
.state_r_pc
.data_o
)
186 # don't write pc every cycle
187 comb
+= self
.state_w_pc
.wen
.eq(0)
188 comb
+= self
.state_w_pc
.data_i
.eq(0)
190 # don't read msr every cycle
191 comb
+= self
.state_r_msr
.ren
.eq(0)
192 msr_read
= Signal(reset
=1)
194 # connect up debug signals
195 # TODO comb += core.icache_rst_i.eq(dbg.icache_rst_o)
196 comb
+= dbg
.terminate_i
.eq(core
.core_terminate_o
)
197 comb
+= dbg
.state
.pc
.eq(pc
)
198 #comb += dbg.state.pc.eq(cur_state.pc)
199 comb
+= dbg
.state
.msr
.eq(cur_state
.msr
)
202 core_busy_o
= core
.busy_o
# core is busy
203 core_ivalid_i
= core
.ivalid_i
# instruction is valid
204 core_issue_i
= core
.issue_i
# instruction is issued
205 dec_opcode_i
= pdecode2
.dec
.raw_opcode_in
# raw opcode
207 insn_type
= core
.e
.do
.insn_type
209 # actually use a nmigen FSM for the first time (w00t)
210 # this FSM is perhaps unusual in that it detects conditions
211 # then "holds" information, combinatorially, for the core
212 # (as opposed to using sync - which would be on a clock's delay)
213 # this includes the actual opcode, valid flags and so on.
217 with m
.State("IDLE"):
218 sync
+= pc_changed
.eq(0)
220 sync
+= core
.raw_insn_i
.eq(0)
221 sync
+= core
.bigendian_i
.eq(0)
222 with m
.If(~dbg
.core_stop_o
& ~core
.core_reset_i
):
223 # instruction allowed to go: start by reading the PC
224 # capture the PC and also drop it into Insn Memory
225 # we have joined a pair of combinatorial memory
226 # lookups together. this is Generally Bad.
227 comb
+= self
.imem
.a_pc_i
.eq(pc
)
228 comb
+= self
.imem
.a_valid_i
.eq(1)
229 comb
+= self
.imem
.f_valid_i
.eq(1)
230 sync
+= cur_state
.pc
.eq(pc
)
232 # initiate read of MSR. arrives one clock later
233 comb
+= self
.state_r_msr
.ren
.eq(1<<StateRegs
.MSR
)
234 sync
+= msr_read
.eq(0)
236 m
.next
= "INSN_READ" # move to "wait for bus" phase
238 comb
+= core
.core_stopped_i
.eq(1)
239 comb
+= dbg
.core_stopped_i
.eq(1)
241 # dummy pause to find out why simulation is not keeping up
242 with m
.State("INSN_READ"):
243 # one cycle later, msr read arrives. valid only once.
244 with m
.If(~msr_read
):
245 sync
+= msr_read
.eq(1) # yeah don't read it again
246 sync
+= cur_state
.msr
.eq(self
.state_r_msr
.data_o
)
247 with m
.If(self
.imem
.f_busy_o
): # zzz...
248 # busy: stay in wait-read
249 comb
+= self
.imem
.a_valid_i
.eq(1)
250 comb
+= self
.imem
.f_valid_i
.eq(1)
252 # not busy: instruction fetched
253 f_instr_o
= self
.imem
.f_instr_o
254 if f_instr_o
.width
== 32:
257 insn
= f_instr_o
.word_select(cur_state
.pc
[2], 32)
258 comb
+= dec_opcode_i
.eq(insn
) # actual opcode
259 sync
+= core
.e
.eq(pdecode2
.e
)
260 sync
+= core
.state
.eq(cur_state
)
261 sync
+= core
.raw_insn_i
.eq(dec_opcode_i
)
262 sync
+= core
.bigendian_i
.eq(self
.core_bigendian_i
)
263 sync
+= ilatch
.eq(insn
) # latch current insn
264 # also drop PC and MSR into decode "state"
265 m
.next
= "INSN_START" # move to "start"
267 # waiting for instruction bus (stays there until not busy)
268 with m
.State("INSN_START"):
269 comb
+= core_ivalid_i
.eq(1) # instruction is valid
270 comb
+= core_issue_i
.eq(1) # and issued
272 m
.next
= "INSN_ACTIVE" # move to "wait completion"
274 # instruction started: must wait till it finishes
275 with m
.State("INSN_ACTIVE"):
276 with m
.If(insn_type
!= MicrOp
.OP_NOP
):
277 comb
+= core_ivalid_i
.eq(1) # instruction is valid
278 with m
.If(self
.state_nia
.wen
& (1<<StateRegs
.PC
)):
279 sync
+= pc_changed
.eq(1)
280 with m
.If(~core_busy_o
): # instruction done!
281 # ok here we are not reading the branch unit. TODO
282 # this just blithely overwrites whatever pipeline
284 with m
.If(~pc_changed
):
285 comb
+= self
.state_w_pc
.wen
.eq(1<<StateRegs
.PC
)
286 comb
+= self
.state_w_pc
.data_i
.eq(nia
)
288 sync
+= core
.raw_insn_i
.eq(0)
289 sync
+= core
.bigendian_i
.eq(0)
290 m
.next
= "IDLE" # back to idle
292 # this bit doesn't have to be in the FSM: connect up to read
293 # regfiles on demand from DMI
294 with m
.If(d_reg
.req
): # request for regfile access being made
295 # TODO: error-check this
296 # XXX should this be combinatorial? sync better?
298 comb
+= self
.int_r
.ren
.eq(1<<d_reg
.addr
)
300 comb
+= self
.int_r
.addr
.eq(d_reg
.addr
)
301 comb
+= self
.int_r
.ren
.eq(1)
302 d_reg_delay
= Signal()
303 sync
+= d_reg_delay
.eq(d_reg
.req
)
304 with m
.If(d_reg_delay
):
305 # data arrives one clock later
306 comb
+= d_reg
.data
.eq(self
.int_r
.data_o
)
307 comb
+= d_reg
.ack
.eq(1)
309 # sigh same thing for CR debug
310 with m
.If(d_cr
.req
): # request for regfile access being made
311 comb
+= self
.cr_r
.ren
.eq(0b11111111) # enable all
312 d_cr_delay
= Signal()
313 sync
+= d_cr_delay
.eq(d_cr
.req
)
314 with m
.If(d_cr_delay
):
315 # data arrives one clock later
316 comb
+= d_cr
.data
.eq(self
.cr_r
.data_o
)
317 comb
+= d_cr
.ack
.eq(1)
320 with m
.If(d_xer
.req
): # request for regfile access being made
321 comb
+= self
.xer_r
.ren
.eq(0b111111) # enable all
322 d_xer_delay
= Signal()
323 sync
+= d_xer_delay
.eq(d_xer
.req
)
324 with m
.If(d_xer_delay
):
325 # data arrives one clock later
326 comb
+= d_xer
.data
.eq(self
.xer_r
.data_o
)
327 comb
+= d_xer
.ack
.eq(1)
329 # DEC and TB inc/dec FSM
330 self
.tb_dec_fsm(m
, cur_state
.dec
)
334 def tb_dec_fsm(self
, m
, spr_dec
):
337 this is a FSM for updating either dec or tb. it runs alternately
338 DEC, TB, DEC, TB. note that SPR pipeline could have written a new
339 value to DEC, however the regfile has "passthrough" on it so this
342 see v3.0B p1097-1099 for Timeer Resource and p1065 and p1076
345 comb
, sync
= m
.d
.comb
, m
.d
.sync
346 fast_rf
= self
.core
.regs
.rf
['fast']
347 fast_r_dectb
= fast_rf
.r_ports
['issue'] # DEC/TB
348 fast_w_dectb
= fast_rf
.w_ports
['issue'] # DEC/TB
352 # initiates read of current DEC
353 with m
.State("DEC_READ"):
354 comb
+= fast_r_dectb
.addr
.eq(FastRegs
.DEC
)
355 comb
+= fast_r_dectb
.ren
.eq(1)
358 # waits for DEC read to arrive (1 cycle), updates with new value
359 with m
.State("DEC_WRITE"):
361 # TODO: MSR.LPCR 32-bit decrement mode
362 comb
+= new_dec
.eq(fast_r_dectb
.data_o
- 1)
363 comb
+= fast_w_dectb
.addr
.eq(FastRegs
.DEC
)
364 comb
+= fast_w_dectb
.wen
.eq(1)
365 comb
+= fast_w_dectb
.data_i
.eq(new_dec
)
366 sync
+= spr_dec
.eq(new_dec
) # copy into cur_state for decoder
369 # initiates read of current TB
370 with m
.State("TB_READ"):
371 comb
+= fast_r_dectb
.addr
.eq(FastRegs
.TB
)
372 comb
+= fast_r_dectb
.ren
.eq(1)
375 # waits for read TB to arrive, initiates write of current TB
376 with m
.State("TB_WRITE"):
378 comb
+= new_tb
.eq(fast_r_dectb
.data_o
+ 1)
379 comb
+= fast_w_dectb
.addr
.eq(FastRegs
.TB
)
380 comb
+= fast_w_dectb
.wen
.eq(1)
381 comb
+= fast_w_dectb
.data_i
.eq(new_tb
)
387 yield from self
.pc_i
.ports()
390 yield from self
.core
.ports()
391 yield from self
.imem
.ports()
392 yield self
.core_bigendian_i
398 def external_ports(self
):
399 ports
= self
.pc_i
.ports()
400 ports
+= [self
.pc_o
, self
.memerr_o
, self
.core_bigendian_i
, self
.busy_o
,
401 ClockSignal(), ResetSignal(),
403 ports
+= list(self
.dbg
.dmi
.ports())
404 ports
+= list(self
.imem
.ibus
.fields
.values())
405 ports
+= list(self
.core
.l0
.cmpi
.lsmem
.lsi
.slavebus
.fields
.values())
408 ports
+= list(self
.xics_icp
.bus
.fields
.values())
409 ports
+= list(self
.xics_ics
.bus
.fields
.values())
410 ports
.append(self
.int_level_i
)
413 ports
+= list(self
.simple_gpio
.bus
.fields
.values())
414 ports
.append(self
.gpio_o
)
422 if __name__
== '__main__':
423 units
= {'alu': 1, 'cr': 1, 'branch': 1, 'trap': 1, 'logical': 1,
429 pspec
= TestMemPspec(ldst_ifacetype
='bare_wb',
430 imem_ifacetype
='bare_wb',
435 dut
= TestIssuer(pspec
)
436 vl
= main(dut
, ports
=dut
.ports(), name
="test_issuer")
438 if len(sys
.argv
) == 1:
439 vl
= rtlil
.convert(dut
, ports
=dut
.external_ports(), name
="test_issuer")
440 with
open("test_issuer.il", "w") as f
: