link ST.go directly to ST.rel
[soc.git] / src / soc / fu / compunits / test / test_compunit.py
1 from nmigen import Module, Signal, ResetSignal
2 from nmigen.back.pysim import Simulator, Delay, Settle
3 from nmutil.formaltest import FHDLTestCase
4 from nmigen.cli import rtlil
5 import unittest
6 from soc.decoder.power_decoder import (create_pdecode)
7 from soc.decoder.power_decoder2 import (PowerDecode2)
8 from soc.decoder.power_enums import Function
9 from soc.decoder.isa.all import ISA
10
11 from soc.experiment.compalu_multi import find_ok # hack
12
13
14 def set_cu_input(cu, idx, data):
15 rdop = cu.get_in_name(idx)
16 yield cu.src_i[idx].eq(data)
17 while True:
18 rd_rel_o = yield cu.rd.rel[idx]
19 print ("rd_rel %d wait HI" % idx, rd_rel_o, rdop, hex(data))
20 if rd_rel_o:
21 break
22 yield
23 yield cu.rd.go[idx].eq(1)
24 while True:
25 yield
26 rd_rel_o = yield cu.rd.rel[idx]
27 if rd_rel_o:
28 break
29 print ("rd_rel %d wait HI" % idx, rd_rel_o)
30 yield
31 yield cu.rd.go[idx].eq(0)
32 yield cu.src_i[idx].eq(0)
33
34
35 def get_cu_output(cu, idx, code):
36 wrmask = yield cu.wrmask
37 wrop = cu.get_out_name(idx)
38 wrok = cu.get_out(idx)
39 fname = find_ok(wrok.fields)
40 wrok = yield getattr(wrok, fname)
41 print ("wr_rel mask", repr(code), idx, wrop, bin(wrmask), fname, wrok)
42 assert wrmask & (1<<idx), \
43 "get_cu_output '%s': mask bit %d not set\n" \
44 "write-operand '%s' Data.ok likely not set (%s)" \
45 % (code, idx, wrop, hex(wrok))
46 while True:
47 wr_relall_o = yield cu.wr.rel
48 wr_rel_o = yield cu.wr.rel[idx]
49 print ("wr_rel %d wait" % idx, hex(wr_relall_o), wr_rel_o)
50 if wr_rel_o:
51 break
52 yield
53 yield cu.wr.go[idx].eq(1)
54 yield Settle()
55 result = yield cu.dest[idx]
56 yield
57 yield cu.wr.go[idx].eq(0)
58 print ("result", repr(code), idx, wrop, wrok, hex(result))
59 return result
60
61
62 def set_cu_inputs(cu, inp):
63 for idx, data in inp.items():
64 yield from set_cu_input(cu, idx, data)
65
66
67 def set_operand(cu, dec2, sim):
68 yield from cu.oper_i.eq_from_execute1(dec2.e)
69 yield cu.issue_i.eq(1)
70 yield
71 yield cu.issue_i.eq(0)
72 yield
73
74
75 def get_cu_outputs(cu, code):
76 res = {}
77 wrmask = yield cu.wrmask
78 print ("get_cu_outputs", cu.n_dst, wrmask)
79 if not wrmask: # no point waiting (however really should doublecheck wr.rel)
80 return {}
81 # wait for at least one result
82 while True:
83 wr_rel_o = yield cu.wr.rel
84 if wr_rel_o:
85 break
86 yield
87 for i in range(cu.n_dst):
88 wr_rel_o = yield cu.wr.rel[i]
89 if wr_rel_o:
90 result = yield from get_cu_output(cu, i, code)
91 wrop = cu.get_out_name(i)
92 print ("output", i, wrop, hex(result))
93 res[wrop] = result
94 return res
95
96
97 def get_inp_indexed(cu, inp):
98 res = {}
99 for i in range(cu.n_src):
100 wrop = cu.get_in_name(i)
101 if wrop in inp:
102 res[i] = inp[wrop]
103 return res
104
105
106 class TestRunner(FHDLTestCase):
107 def __init__(self, test_data, fukls, iodef, funit):
108 super().__init__("run_all")
109 self.test_data = test_data
110 self.fukls = fukls
111 self.iodef = iodef
112 self.funit = funit
113
114 def run_all(self):
115 m = Module()
116 comb = m.d.comb
117 instruction = Signal(32)
118
119 pdecode = create_pdecode()
120
121 m.submodules.pdecode2 = pdecode2 = PowerDecode2(pdecode)
122 if self.funit == Function.LDST:
123 from soc.experiment.l0_cache import TstL0CacheBuffer
124 m.submodules.l0 = l0 = TstL0CacheBuffer(n_units=1, regwid=64)
125 pi = l0.l0.dports[0].pi
126 m.submodules.cu = cu = self.fukls(pi, awid=4)
127 m.d.comb += cu.ad.go.eq(cu.ad.rel) # link addr-go direct to rel
128 m.d.comb += cu.st.go.eq(cu.st.rel) # link store-go direct to rel
129 else:
130 m.submodules.cu = cu = self.fukls()
131
132 comb += pdecode2.dec.raw_opcode_in.eq(instruction)
133 sim = Simulator(m)
134
135 sim.add_clock(1e-6)
136
137 def process():
138 yield cu.issue_i.eq(0)
139 yield
140
141 for test in self.test_data:
142 print(test.name)
143 program = test.program
144 self.subTest(test.name)
145 print ("test", test.name, test.mem)
146 sim = ISA(pdecode2, test.regs, test.sprs, test.cr, test.mem,
147 test.msr)
148 gen = program.generate_instructions()
149 instructions = list(zip(gen, program.assembly.splitlines()))
150
151 # initialise memory
152 if self.funit == Function.LDST:
153 mem = l0.mem.mem
154 memlist = []
155 for i in range(mem.depth//2):
156 data = sim.mem.ld(i*16, 8)
157 data1 = sim.mem.ld(i*16+8, 8)
158 yield mem._array[i].eq(data | (data1<<32))
159 print (mem, mem.depth, mem.width)
160 print ("mem init", list(map(hex,memlist)))
161
162 index = sim.pc.CIA.value//4
163 while index < len(instructions):
164 ins, code = instructions[index]
165
166 print("0x{:X}".format(ins & 0xffffffff))
167 print(code)
168
169 # ask the decoder to decode this binary data (endian'd)
170 yield pdecode2.dec.bigendian.eq(0) # little / big?
171 yield instruction.eq(ins) # raw binary instr.
172 yield Settle()
173 fn_unit = yield pdecode2.e.fn_unit
174 fuval = self.funit.value
175 self.assertEqual(fn_unit & fuval, fuval)
176
177 # set operand and get inputs
178 yield from set_operand(cu, pdecode2, sim)
179 iname = yield from self.iodef.get_cu_inputs(pdecode2, sim)
180 inp = get_inp_indexed(cu, iname)
181
182 # reset read-operand mask
183 rdmask = pdecode2.rdflags(cu)
184 #print ("hardcoded rdmask", cu.rdflags(pdecode2.e))
185 #print ("decoder rdmask", rdmask)
186 yield cu.rdmaskn.eq(~rdmask)
187
188 # reset write-operand mask
189 for idx in range(cu.n_dst):
190 wrok = cu.get_out(idx)
191 fname = find_ok(wrok.fields)
192 yield getattr(wrok, fname).eq(0)
193
194 yield Settle()
195
196 # set inputs into CU
197 rd_rel_o = yield cu.rd.rel
198 wr_rel_o = yield cu.wr.rel
199 print ("before inputs, rd_rel, wr_rel: ",
200 bin(rd_rel_o), bin(wr_rel_o))
201 assert wr_rel_o == 0, "wr.rel %s must be zero. "\
202 "previous instr not written all regs\n"\
203 "respec %s" % \
204 (bin(wr_rel_o), cu.rwid[1])
205 yield from set_cu_inputs(cu, inp)
206 yield
207 rd_rel_o = yield cu.rd.rel
208 wr_rel_o = yield cu.wr.rel
209 wrmask = yield cu.wrmask
210 print ("after inputs, rd_rel, wr_rel, wrmask: ",
211 bin(rd_rel_o), bin(wr_rel_o), bin(wrmask))
212
213 # call simulated operation
214 opname = code.split(' ')[0]
215 yield from sim.call(opname)
216 index = sim.pc.CIA.value//4
217
218 yield Settle()
219 # get all outputs (one by one, just "because")
220 res = yield from get_cu_outputs(cu, code)
221
222 yield from self.iodef.check_cu_outputs(res, pdecode2,
223 sim, code)
224
225 # sigh. hard-coded. test memory
226 if self.funit == Function.LDST:
227 print ("mem dump", sim.mem.mem)
228
229 sim.add_sync_process(process)
230
231 name = self.funit.name.lower()
232 with sim.write_vcd("%s_simulator.vcd" % name,
233 "%s_simulator.gtkw" % name,
234 traces=[]):
235 sim.run()
236
237