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 from soc.config.test.test_loadstore import TestMemPspec
  13
  14
  15 def set_cu_input(cu, idx, data):
  16     rdop = cu.get_in_name(idx)
  17     yield cu.src_i[idx].eq(data)
  18     while True:
  19         rd_rel_o = yield cu.rd.rel[idx]
  20         print ("rd_rel %d wait HI" % idx, rd_rel_o, rdop, hex(data))
  21         if rd_rel_o:
  22             break
  23         yield
  24     yield cu.rd.go[idx].eq(1)
  25     while True:
  26         yield
  27         rd_rel_o = yield cu.rd.rel[idx]
  28         if rd_rel_o:
  29             break
  30         print ("rd_rel %d wait HI" % idx, rd_rel_o)
  31         yield
  32     yield cu.rd.go[idx].eq(0)
  33     yield cu.src_i[idx].eq(0)
  34
  35
  36 def get_cu_output(cu, idx, code):
  37     wrmask = yield cu.wrmask
  38     wrop = cu.get_out_name(idx)
  39     wrok = cu.get_out(idx)
  40     fname = find_ok(wrok.fields)
  41     wrok = yield getattr(wrok, fname)
  42     print ("wr_rel mask", repr(code), idx, wrop, bin(wrmask), fname, wrok)
  43     assert wrmask & (1<<idx), \
  44             "get_cu_output '%s': mask bit %d not set\n" \
  45             "write-operand '%s' Data.ok likely not set (%s)" \
  46             % (code, idx, wrop, hex(wrok))
  47     while True:
  48         wr_relall_o = yield cu.wr.rel
  49         wr_rel_o = yield cu.wr.rel[idx]
  50         print ("wr_rel %d wait" % idx, hex(wr_relall_o), wr_rel_o)
  51         if wr_rel_o:
  52             break
  53         yield
  54     yield cu.wr.go[idx].eq(1)
  55     yield Settle()
  56     result = yield cu.dest[idx]
  57     yield
  58     yield cu.wr.go[idx].eq(0)
  59     print ("result", repr(code), idx, wrop, wrok, hex(result))
  60
  61     return result
  62
  63
  64 def set_cu_inputs(cu, inp):
  65     print ("set_cu_inputs", inp)
  66     for idx, data in inp.items():
  67         yield from set_cu_input(cu, idx, data)
  68
  69
  70 def set_operand(cu, dec2, sim):
  71     yield from cu.oper_i.eq_from_execute1(dec2.e)
  72     yield cu.issue_i.eq(1)
  73     yield
  74     yield cu.issue_i.eq(0)
  75     yield
  76
  77
  78 def get_cu_outputs(cu, code):
  79     res = {}
  80     wrmask = yield cu.wrmask
  81     print ("get_cu_outputs", cu.n_dst, wrmask)
  82     if not wrmask: # no point waiting (however really should doublecheck wr.rel)
  83         return {}
  84     # wait for at least one result
  85     while True:
  86         wr_rel_o = yield cu.wr.rel
  87         if wr_rel_o:
  88             break
  89         yield
  90     for i in range(cu.n_dst):
  91         wr_rel_o = yield cu.wr.rel[i]
  92         if wr_rel_o:
  93             result = yield from get_cu_output(cu, i, code)
  94             wrop = cu.get_out_name(i)
  95             print ("output", i, wrop, hex(result))
  96             res[wrop] = result
  97     return res
  98
  99
 100 def get_inp_indexed(cu, inp):
 101     res = {}
 102     for i in range(cu.n_src):
 103         wrop = cu.get_in_name(i)
 104         if wrop in inp:
 105             res[i] = inp[wrop]
 106     return res
 107
 108 def get_l0_mem(l0): # BLECH!
 109     if hasattr(l0.pimem, 'lsui'):
 110         return l0.pimem.lsui.mem
 111     return l0.pimem.mem.mem
 112
 113 def setup_test_memory(l0, sim):
 114     mem = get_l0_mem(l0)
 115     print ("before, init mem", mem.depth, mem.width, mem)
 116     for i in range(mem.depth):
 117         data = sim.mem.ld(i*8, 8, False)
 118         print ("init ", i, hex(data))
 119         yield mem._array[i].eq(data)
 120     yield Settle()
 121     for k, v in sim.mem.mem.items():
 122         print ("    %6x %016x" % (k, v))
 123     print ("before, nmigen mem dump")
 124     for i in range(mem.depth):
 125         actual_mem = yield mem._array[i]
 126         print ("    %6i %016x" % (i, actual_mem))
 127
 128
 129 def dump_sim_memory(dut, l0, sim, code):
 130     mem = get_l0_mem(l0)
 131     print ("sim mem dump")
 132     for k, v in sim.mem.mem.items():
 133         print ("    %6x %016x" % (k, v))
 134     print ("nmigen mem dump")
 135     for i in range(mem.depth):
 136         actual_mem = yield mem._array[i]
 137         print ("    %6i %016x" % (i, actual_mem))
 138
 139
 140 def check_sim_memory(dut, l0, sim, code):
 141     mem = get_l0_mem(l0)
 142
 143     for i in range(mem.depth):
 144         expected_mem = sim.mem.ld(i*8, 8, False)
 145         actual_mem = yield mem._array[i]
 146         dut.assertEqual(expected_mem, actual_mem,
 147                 "%s %d %x %x" % (code, i,
 148                                  expected_mem, actual_mem))
 149
 150 class TestRunner(FHDLTestCase):
 151     def __init__(self, test_data, fukls, iodef, funit, bigendian):
 152         super().__init__("run_all")
 153         self.test_data = test_data
 154         self.fukls = fukls
 155         self.iodef = iodef
 156         self.funit = funit
 157         self.bigendian = bigendian
 158
 159     def run_all(self):
 160         m = Module()
 161         comb = m.d.comb
 162         instruction = Signal(32)
 163
 164         pdecode = create_pdecode()
 165         m.submodules.pdecode2 = pdecode2 = PowerDecode2(pdecode)
 166
 167         # copy of the decoder for simulator
 168         simdec = create_pdecode()
 169         simdec2 = PowerDecode2(simdec)
 170         m.submodules.simdec2 = simdec2 # pain in the neck
 171
 172         if self.funit == Function.LDST:
 173             from soc.experiment.l0_cache import TstL0CacheBuffer
 174             pspec = TestMemPspec(ldst_ifacetype='test_bare_wb',
 175                                  addr_wid=48,
 176                                  mask_wid=8,
 177                                  reg_wid=64)
 178             m.submodules.l0 = l0 = TstL0CacheBuffer(pspec, n_units=1)
 179             pi = l0.l0.dports[0]
 180             m.submodules.cu = cu = self.fukls(pi, idx=0, awid=3)
 181             m.d.comb += cu.ad.go.eq(cu.ad.rel) # link addr-go direct to rel
 182             m.d.comb += cu.st.go.eq(cu.st.rel) # link store-go direct to rel
 183         else:
 184             m.submodules.cu = cu = self.fukls(0)
 185
 186         comb += pdecode2.dec.raw_opcode_in.eq(instruction)
 187         sim = Simulator(m)
 188
 189         sim.add_clock(1e-6)
 190
 191         def process():
 192             yield cu.issue_i.eq(0)
 193             yield
 194
 195             for test in self.test_data:
 196                 print(test.name)
 197                 program = test.program
 198                 self.subTest(test.name)
 199                 print ("test", test.name, test.mem)
 200                 gen = list(program.generate_instructions())
 201                 insncode = program.assembly.splitlines()
 202                 instructions = list(zip(gen, insncode))
 203                 sim = ISA(simdec2, test.regs, test.sprs, test.cr, test.mem,
 204                           test.msr,
 205                           initial_insns=gen, respect_pc=False,
 206                           disassembly=insncode,
 207                           bigendian=self.bigendian)
 208
 209                 # initialise memory
 210                 if self.funit == Function.LDST:
 211                     yield from setup_test_memory(l0, sim)
 212
 213                 index = sim.pc.CIA.value//4
 214                 while True:
 215                     try:
 216                         yield from sim.setup_one()
 217                     except KeyError: # indicates instruction not in imem: stop
 218                         break
 219                     yield Settle()
 220                     ins, code = instructions[index]
 221                     print(index, code)
 222
 223                     # ask the decoder to decode this binary data (endian'd)
 224                     yield pdecode2.dec.bigendian.eq(self.bigendian)  # le / be?
 225                     yield instruction.eq(ins)          # raw binary instr.
 226                     yield Settle()
 227                     fn_unit = yield pdecode2.e.do.fn_unit
 228                     fuval = self.funit.value
 229                     self.assertEqual(fn_unit & fuval, fuval)
 230
 231                     # set operand and get inputs
 232                     yield from set_operand(cu, pdecode2, sim)
 233                     yield Settle()
 234                     iname = yield from self.iodef.get_cu_inputs(pdecode2, sim)
 235                     inp = get_inp_indexed(cu, iname)
 236
 237                     # reset read-operand mask
 238                     rdmask = pdecode2.rdflags(cu)
 239                     #print ("hardcoded rdmask", cu.rdflags(pdecode2.e))
 240                     #print ("decoder rdmask", rdmask)
 241                     yield cu.rdmaskn.eq(~rdmask)
 242
 243                     # reset write-operand mask
 244                     for idx in range(cu.n_dst):
 245                         wrok = cu.get_out(idx)
 246                         fname = find_ok(wrok.fields)
 247                         yield getattr(wrok, fname).eq(0)
 248
 249                     yield Settle()
 250
 251                     # set inputs into CU
 252                     rd_rel_o = yield cu.rd.rel
 253                     wr_rel_o = yield cu.wr.rel
 254                     print ("before inputs, rd_rel, wr_rel: ",
 255                             bin(rd_rel_o), bin(wr_rel_o))
 256                     assert wr_rel_o == 0, "wr.rel %s must be zero. "\
 257                                 "previous instr not written all regs\n"\
 258                                 "respec %s" % \
 259                                 (bin(wr_rel_o), cu.rwid[1])
 260                     yield from set_cu_inputs(cu, inp)
 261                     rd_rel_o = yield cu.rd.rel
 262                     wr_rel_o = yield cu.wr.rel
 263                     wrmask = yield cu.wrmask
 264                     print ("after inputs, rd_rel, wr_rel, wrmask: ",
 265                             bin(rd_rel_o), bin(wr_rel_o), bin(wrmask))
 266
 267                     # call simulated operation
 268                     yield from sim.execute_one()
 269                     yield Settle()
 270                     index = sim.pc.CIA.value//4
 271
 272                     # get all outputs (one by one, just "because")
 273                     res = yield from get_cu_outputs(cu, code)
 274                     wrmask = yield cu.wrmask
 275                     rd_rel_o = yield cu.rd.rel
 276                     wr_rel_o = yield cu.wr.rel
 277                     print ("after got outputs, rd_rel, wr_rel, wrmask: ",
 278                             bin(rd_rel_o), bin(wr_rel_o), bin(wrmask))
 279
 280                     # wait for busy to go low
 281                     while True:
 282                         busy_o = yield cu.busy_o
 283                         print ("busy", busy_o)
 284                         if not busy_o:
 285                             break
 286                         yield
 287
 288                     if self.funit == Function.LDST:
 289                         yield from dump_sim_memory(self, l0, sim, code)
 290
 291
 292                     # sigh.  hard-coded.  test memory
 293                     if self.funit == Function.LDST:
 294                         yield from check_sim_memory(self, l0, sim, code)
 295                         yield from self.iodef.check_cu_outputs(res, pdecode2,
 296                                                                 sim, cu,
 297                                                                 code)
 298                     else:
 299                         yield from self.iodef.check_cu_outputs(res, pdecode2,
 300                                                                 sim, cu.alu,
 301                                                                 code)
 302
 303
 304         sim.add_sync_process(process)
 305
 306         name = self.funit.name.lower()
 307         with sim.write_vcd("%s_simulator.vcd" % name,
 308                            "%s_simulator.gtkw" % name,
 309                             traces=[]):
 310             sim.run()
 311
 312