def test_cross(self): stim = FieldCompositeModel("stim", True) f = FieldScalarModel("a", 16, False, True) stim.add_field(f) f2 = FieldScalarModel("b", 16, False, True) stim.add_field(f2) cg = CovergroupModel("cg") cp = CoverpointModel(ExprFieldRefModel(f), "cp1", CoverageOptionsModel()) cg.add_coverpoint(cp) bn = CoverpointBinArrayModel("cp", 0, 1, 16) cp.add_bin_model(bn) cp2 = CoverpointModel(ExprFieldRefModel(f2), "cp2", CoverageOptionsModel()) cg.add_coverpoint(cp2) bn = CoverpointBinArrayModel("cp", 0, 1, 16) cp2.add_bin_model(bn) cr = CoverpointCrossModel("aXb", CoverageOptionsModel()) cr.add_coverpoint(cp) cr.add_coverpoint(cp2) cg.add_coverpoint(cr) gen = GeneratorModel("top") gen.add_field(stim) gen.add_covergroup(cg) gen.finalize() # Need a special randomizer to deal with generators r = Randomizer() count = 0 for i in range(1000): r.do_randomize([gen]) cg.sample() count += 1 cov = cg.get_coverage() print("Coverage: (" + str(i) + ") " + str(cov)) if cov == 100: break self.assertEqual(cg.get_coverage(), 100) # Ensure that we converge relatively quickly self.assertLessEqual(count, (256+16+16))
def test_coverpoint_bins(self): stim = FieldCompositeModel("stim", True) f = FieldScalarModel("a", 16, False, True) stim.add_field(f) f2 = FieldScalarModel("b", 16, False, True) stim.add_field(f2) cg = CovergroupModel("cg") cp = CoverpointModel(ExprFieldRefModel(f), "cp1", CoverageOptionsModel()) cg.add_coverpoint(cp) cp.add_bin_model(CoverpointBinArrayModel("bn1", 0, 1, 16)) cp.add_bin_model(CoverpointBinCollectionModel.mk_collection("bn2", RangelistModel([ [17,65535-16-1] ]), 16)) cp.add_bin_model(CoverpointBinArrayModel("bn3", 0, 65535-16, 65535)) cp2 = CoverpointModel(ExprFieldRefModel(f2), "cp2", CoverageOptionsModel()) cg.add_coverpoint(cp2) bn = CoverpointBinArrayModel("cp", 0, 1, 16) cp2.add_bin_model(bn) gen = GeneratorModel("top") gen.add_field(stim) gen.add_covergroup(cg) gen.finalize() # Need a special randomizer to deal with generators r = Randomizer() count = 0 for i in range(1000): r.do_randomize([gen]) cg.sample() count += 1 cov = cg.get_coverage() if cov == 100: break self.assertEqual(cg.get_coverage(), 100) # Ensure that we converge relatively quickly self.assertLessEqual(count, 64)
def test_smoke(self): stim = FieldCompositeModel("stim", True) f = FieldScalarModel("a", 16, False, True) stim.add_field(f) f2 = FieldScalarModel("b", 16, False, True) stim.add_field(f2) cg = CovergroupModel("cg") cp = CoverpointModel(ExprFieldRefModel(f), "cp1", CoverageOptionsModel()) cg.add_coverpoint(cp) bn = CoverpointBinArrayModel("cp", 1, 16) cp.add_bin_model(bn) cp2 = CoverpointModel(ExprFieldRefModel(f2), "cp2", CoverageOptionsModel()) cg.add_coverpoint(cp2) bn = CoverpointBinArrayModel("cp", 1, 16) cp2.add_bin_model(bn) gen = GeneratorModel("top") gen.add_field(stim) gen.add_covergroup(cg) gen.finalize() # Need a special randomizer to deal with generators r = Randomizer(RandState(0)) randstate = RandState(0) count = 0 for i in range(1000): r.do_randomize(randstate, SourceInfo("", -1), [gen]) cg.sample() count += 1 cov = cg.get_coverage() if cov == 100: break self.assertEqual(cg.get_coverage(), 100) # Ensure that we converge relatively quickly self.assertLessEqual(count, 32)