How to use run_and_check method in avocado

Best Python code snippet using avocado_python

test_hybrid_script.py

Source:test_hybrid_script.py

...18from tvm.contrib import util19from tvm.hybrid import script20from tvm.hybrid.runtime import HYBRID_GLOBALS21@pytest.mark.skip22def run_and_check(func, args, var_dict={}, target='llvm', sch=None, outs=None):23    def tvm_val_2_py_val(val):24        val = tvm.ir_pass.Substitute(val, var_dict)25        val = tvm.ir_pass.Simplify(val)26        assert isinstance(val, (tvm.expr.IntImm,))27        return val.value28    ctx = tvm.context(target, 0)29    op = None30    if sch is None:31        outs = func(*tuple(tvm.convert(i) if isinstance(i, list) else i for i in args))32        op = outs[0].op if isinstance(outs, list) else outs.op33        sch = tvm.create_schedule(op)34    else:35        assert outs is not None36        assert isinstance(outs, list)37        op = outs[0].op38    emu_args = []39    nd_args = []40    for i in args:41        if isinstance(i, tvm.tensor.Tensor):42            shape = [tvm_val_2_py_val(j) for j in i.shape]43            emu_args.append(numpy.random.randn(*shape).astype(i.dtype))44            nd_args.append(tvm.nd.array(emu_args[-1], ctx))45        elif isinstance(i, tvm.expr.Var):46            emu_args.append(tvm_val_2_py_val(i))47            nd_args.append(emu_args[-1])48        else:49            assert isinstance(i, list)50            emu_args.append(numpy.array(i))51    compile_args = [i for i in args if isinstance(i, (tvm.tensor.Tensor, tvm.expr.Var))] + \52                   (outs if isinstance(outs, list) else [outs])53    module = tvm.build(sch,54                       compile_args,55                       target=target)56    assert module57    out_tensors = []58    for i in range(op.num_outputs):59        output = op.output(i)60        shape = [tvm_val_2_py_val(j) for j in output.shape]61        nd_args.append(tvm.nd.array(numpy.zeros(shape).astype(output.dtype), ctx))62        out_tensors.append(nd_args[-1])63    ref_data = func(*emu_args)64    if isinstance(ref_data, numpy.ndarray):65        ref_data = [ref_data]66    module(*nd_args)67    for nd, np in zip(out_tensors, ref_data):68        tvm.testing.assert_allclose(nd.asnumpy(), np, rtol=1e-5, atol=1e-5)69    module_args = [i for i in args if isinstance(i, (tvm.tensor.Tensor, tvm.expr.Var))]70    module_outs = [outs] if isinstance(outs, tvm.tensor.Tensor) else outs71    h_module = tvm.hybrid.build(sch, module_args, module_outs)72    return h_module, module_args, module_outs73@script74def outer_product(n, m, a, b):75    """This is a simple outer product.76    Actually this function is not required to be documented.77    I write this docstring to test skipping docstring functionality.78    """79    c = output_tensor((n, m), a.dtype)80    for i in range(n):81        for j in range(m):82            assert i < n and j < m, "index out of range!"83            c[i, j] = a[i] * b[j]84    return c85#Test global function86#Test bridge between frontend and backend87def test_outer_product():88    n = tvm.size_var('n')89    m = tvm.size_var('m')90    a = tvm.placeholder((n, ), name='a')91    b = tvm.placeholder((m, ), name='b')92    try:93        c = outer_product(n, m, a, b)94        ir = c.op.body95    except IOError as err:96        assert sys.version_info[0] == 2 and str(err) == 'could not get source code'97        return98    #Check for i in (0, n)99    assert isinstance(ir, tvm.stmt.For)100    assert ir.loop_var.name == 'i'101    assert ir.min.value == 0102    assert ir.extent.name == 'n'103    ibody = ir.body104    assert isinstance(ibody, tvm.stmt.For)105    #Check for j in (0, m)106    assert ibody.loop_var.name == 'j'107    assert ibody.min.value == 0108    assert ibody.extent.name == 'm'109    #Check loop body110    jblock = ibody.body111    assert isinstance(jblock, tvm.stmt.SeqStmt)112    jbody = jblock[0]113    assert isinstance(jbody, tvm.stmt.AssertStmt)114    assert isinstance(jbody.message, tvm.expr.StringImm)115    assert jbody.message.value == "index out of range!"116    jbody = jblock[1]117    assert isinstance(jbody, tvm.stmt.Provide)118    assert jbody.func.name == 'c'119    assert len(jbody.args) == 2120    assert jbody.args[0].name == 'i'121    assert jbody.args[1].name == 'j'122    assert isinstance(jbody.value, tvm.expr.Mul)123    mul = jbody.value124    assert isinstance(mul.a, tvm.expr.Call)125    assert mul.a.name == 'a'126    assert mul.b.name == 'b'127    func, ins, outs = run_and_check(outer_product, [n, m, a, b], {n: 99, m: 101})128    temp = util.tempdir()129    path = temp.relpath('%s.py' % func.name)130    func.save(path)131    func_ = tvm.hybrid.HybridModule()132    func_.load(path)133    run_and_check(func_, ins, {n: 99, m: 101}, outs=outs)134    for key, _ in HYBRID_GLOBALS.items():135        assert key not in globals().keys()136        assert key not in outer_product.__globals__.keys()137#Test local function138#Test allocation of local variable139def test_fanout():140    @script141    def fanout(n, a):142        three = 3.0143        b = output_tensor((a.shape[0] - 3, ), a.dtype)144        for i in range(a.shape[0] - 3):145            sigma = 0.0146            for j in range(3):147                sigma += a[i + j]148            sigma = sigma / three149            b[i] = sigma150        return b151    n = tvm.size_var('n')152    a = tvm.placeholder((n, ), 'float32', name='a')153    try:154        b = fanout(n, a)155        ir = b.op.body156    except IOError as err:157        assert sys.version_info[0] == 2 and str(err) == 'could not get source code'158        return159    #Check for i in (0, n-3)160    assert isinstance(ir, tvm.stmt.For)161    assert ir.loop_var.name == 'i'162    assert ir.min.value == 0163    assert tvm.ir_pass.Equal(ir.extent, n - 3)164    #Check loopbody165    ibody = ir.body166    assert isinstance(ibody, tvm.stmt.AttrStmt)167    abody = ibody.body168    assert isinstance(abody, tvm.stmt.Realize)169    assert abody.bounds[0].min.value == 0170    assert abody.bounds[0].extent.value == 1171    assert abody.func.name == 'sigma'172    #Check i loop body173    rbody = abody.body174    assert isinstance(rbody[0], tvm.stmt.Provide)175    assert rbody[0].func.name == 'sigma'176    assert len(rbody[0].args) == 1177    assert rbody[0].args[0].value == 0178    #Check fanout loop179    jloop = rbody[1]180    assert jloop.loop_var.name == 'j'181    assert jloop.min.value == 0182    assert jloop.extent.value == 3183    jbody = jloop.body184    assert isinstance(jbody, tvm.stmt.Provide)185    assert len(jbody.args) == 1186    assert jbody.args[0].value == 0187    assert jbody.func.name == 'sigma'188    assert isinstance(jbody.value, tvm.expr.Add)189    value = jbody.value190    assert isinstance(value.a, tvm.expr.Call)191    assert value.a.name == 'sigma'192    assert len(value.a.args) == 1193    assert value.a.args[0].value == 0194    assert value.b.name == 'a'195    assert len(value.b.args) == 1196    assert tvm.ir_pass.Equal(value.b.args[0], ir.loop_var + jloop.loop_var)197    divide= rbody[2]198    assert isinstance(divide, tvm.stmt.Provide)199    assert len(divide.args) == 1200    assert divide.args[0].value == 0201    value = divide.value202    assert isinstance(value, tvm.expr.Mul)203    assert value.a.name == 'sigma'204    assert len(value.a.args) == 1205    assert value.a.args[0].value == 0206    assert abs(value.b.value - (1 / 3.0)) < 1e-5207    write = rbody[3]208    assert isinstance(write, tvm.stmt.Provide)209    assert write.func.name == 'b'210    assert write.value.name == 'sigma'211    assert len(write.value.args) == 1212    assert write.value.args[0].value == 0213    func, ins, outs = run_and_check(fanout, [n, a], {n: 10})214    run_and_check(func, ins, {n: 10}, outs=outs)215def test_looptype():216    @script217    def looptype(a, b, c):218        d = output_tensor((16, ), 'int32')219        e = output_tensor((16, ), 'int32')220        f = output_tensor((16, ), 'int32')221        for i in parallel(16):222            d[i] = a[i]223        for j in vectorize(16):224            e[j] = b[j]225        for k in unroll(16):226            f[k] = c[k]227        return d, e, f228    a = tvm.placeholder((16, ), name='a', dtype='int32')229    b = tvm.placeholder((16, ), name='b', dtype='int32')230    c = tvm.placeholder((16, ), name='c', dtype='int32')231    try:232        d, e, f = looptype(a, b, c)233        ir = d.op.body234    except:235        return236    iloop = ir[0]237    jloop = ir[1]238    kloop = ir[2]239    assert iloop.for_type == tvm.stmt.For.Parallel240    assert jloop.for_type == tvm.stmt.For.Vectorized241    assert kloop.for_type == tvm.stmt.For.Unrolled242    func, ins, outs = run_and_check(looptype, [a, b, c])243    run_and_check(func, ins, outs=outs)244def test_if():245    @script246    def if_then_else(a):247        b = output_tensor((10, ), 'int32')248        c = output_tensor((10, ), 'int32')249        for i in range(10):250            if i % 2 == 0:251                c[i] = a[i]252            else:253                c[i] = b[i]254        for i in unroll(10):255            b[i] = -1 if i % 2 == 0 else 1256        return b, c257    a = tvm.placeholder((10, ), dtype='int32', name='a')258    func, ins, outs = run_and_check(if_then_else, [a])259    run_and_check(func, ins, outs=outs)260    @script261    def if_triple_condition(a):262        b = output_tensor((10, ), 'int32')263        for i in range(10):264            if 0 <= i < 5:265                b[i] = a[i]266            else:267                b[i] = a[i] + 1268        return b269    func, ins, outs = run_and_check(if_triple_condition, [a])270    run_and_check(func, ins, outs=outs)271    @script272    def if_and(a):273        b = output_tensor((10, ), 'int32')274        for i in range(10):275            if i >= 0 and i < 5:276                b[i] = a[i]277            else:278                b[i] = a[i] + 1279        return b280    func, ins, outs = run_and_check(if_and, [a])281    run_and_check(func, ins, outs=outs)282def test_bind():283    if not tvm.gpu(0).exist:284        print('[Warning] No GPU found! Skip bind test!')285        return286    @script287    def vec_add(a, b):288        c = output_tensor((1000, ), 'float32')289        for tx in bind('threadIdx.x', 1000):290            c[tx] = a[tx] + b[tx]291        return c292    a = tvm.placeholder((1000, ), dtype='float32', name='a')293    b = tvm.placeholder((1000, ), dtype='float32', name='b')294    func, ins, outs = run_and_check(vec_add, [a, b], target='cuda')295    run_and_check(func, ins, outs=outs, target='cuda')296    @script297    def raw(a, b):298        c = output_tensor((1000, ), 'float32')299        for i in range(1000):300            c[i] = a[i] + b[i]301        return c302    c = raw(a, b)303    sch = tvm.create_schedule(c.op)304    x = tvm.thread_axis('threadIdx.x')305    sch[c].bind(c.op.axis[0], x)306    func, ins, outs = run_and_check(raw, [a, b], sch=sch, outs=[c], target='cuda')307    run_and_check(func, ins, outs=outs, target='cuda')308    @tvm.hybrid.script309    def foo(a):310        c = output_tensor((a.shape[0],), a.dtype)311        total = allocate((1,), a.dtype, 'local')312        len_i = a.shape[0]313        len_j = a.shape[1]314        for i in bind('threadIdx.x', len_i):315            total[0] = 0.316            for k in const_range(len_j):317                total[0] += a[i, k]318            c[i] = total[0]319        return c320    a = tvm.placeholder((8, 4), 'float32')321    c = foo(a)322    s = tvm.create_schedule(c.op)323    ir = tvm.lower(s, [a, c], simple_mode=True)324    assert not isinstance(ir, tvm.stmt.AttrStmt)325    func, ins, outs = run_and_check(foo, [a], target='cuda')326    run_and_check(func, ins, outs=outs, target='cuda')327    @tvm.hybrid.script328    def max_threads(a):329        b = output_tensor(a.shape, a.dtype)330        n = a.shape[0]331        m = max_num_threads(True)332        for i in bind('threadIdx.x', m):333            for j in bind('blockIdx.x', ceil_div(n, m)):334                if i * m + j < n:335                    b[i * m + j] = a[i * m + j] + a[i * m + j]336        return b337    a = tvm.placeholder((10000, ), 'float32')338    with tvm.target.create('cuda'):339        func, ins, outs = run_and_check(max_threads, [a], target='cuda')340        run_and_check(func, ins, outs=outs, target='cuda')341def test_math_intrin():342    @script343    def intrin_real(a):344        b = output_tensor((8, ), 'float32')345        b[0] = sqrt(a[0])346        b[1] = log(a[1])347        b[2] = exp(a[2])348        b[3] = sigmoid(a[3])349        b[4] = power(a[4], a[5])350        b[5] = tanh(a[5])351        b[6] = min(a[4], a[5])352        b[7] = max(a[5], a[6])353        return b354    a8 = tvm.placeholder((8, ), dtype='float32', name='a')355    b8 = intrin_real(a8)356    sch = tvm.create_schedule(b8.op)357    func = tvm.build(sch, [a8, b8])358    assert func359    a = numpy.arange(2, 10).astype('float32')360    tvm_a = tvm.ndarray.array(a)361    tvm_b = tvm.ndarray.array(numpy.zeros((8, ), dtype='float32'))362    b = intrin_real(a)363    func(tvm_a, tvm_b)364    tvm.testing.assert_allclose(b, tvm_b.asnumpy(), rtol=1e-5)365    @script366    def intrin_int(a):367        b = output_tensor((1, ), 'int32')368        b[0] = popcount(a[0])369        return b370    a1 = tvm.placeholder((1, ), dtype='int32')371    b1 = intrin_int(a1)372    sch = tvm.create_schedule(b1.op)373    func = tvm.build(sch, [a1, b1])374    assert func375    a = numpy.array([114514]).astype('int32')376    tvm_a = tvm.ndarray.array(a)377    tvm_b = tvm.ndarray.array(numpy.array([0]).astype('int32'))378    b = intrin_int(a)379    func(tvm_a, tvm_b)380    assert tvm_b.asnumpy()[0] == b[0]381# test non caconical loops382def test_non_zero():383    @tvm.hybrid.script384    def blur(a):385        b = output_tensor((30, 30), 'float32')386        for i in range(2, 32):387            for j in range(2, 32):388                s = 0.0389                for di in range(3):390                    for dj in range(3):391                        s += a[i-di, j-dj]392                b[i-2, j-2] = s / 9.0393        return b394    a = tvm.placeholder((32, 32), 'float32', 'a')395    func, ins, outs = run_and_check(blur, [a])396    run_and_check(func, ins, outs=outs)397    @tvm.hybrid.script398    def triangle(a, b):399        c = output_tensor((10, 10), dtype='float32')400        for i in range(10):401            for j in range(i, 10):402                c[i, j] = a[i] * b[j]403        return c404    a = tvm.placeholder((10, ), dtype='float32', name='a')405    b = tvm.placeholder((10, ), dtype='float32', name='b')406    func, ins, outs = run_and_check(triangle, [a, b])407    run_and_check(func, ins, outs=outs)408def test_allocate():409    @tvm.hybrid.script410    def blur2d(a):411        b = output_tensor((30, 30), 'float32')412        for i in range(30):413            ha = allocate((3, 30), 'float32')414            for j in range(3):415                for k in range(30):416                    ha[j, k] = a[i+j, k] + a[i+j, k+1] + a[i+j, k+2]417            for j in range(30):418                b[i, j] = (ha[0, j] + ha[1, j] + ha[2, j]) / 9.0419        return b420    a = tvm.placeholder((32, 32), 'float32', 'a')421    b = blur2d(a)422    sch = tvm.create_schedule(b.op)423    func, ins, outs = run_and_check(blur2d, [a])424    run_and_check(func, ins, outs=outs)425    if tvm.gpu().exist:426        @tvm.hybrid.script427        def share_vec_add(a, b):428            c = output_tensor((256, ), 'float32')429            shared = allocate((256, ), 'float32', 'shared')430            for i in bind("threadIdx.x", 256):431                shared[i] = a[i]432            local = allocate((256, ), 'float32', 'local')433            for i in bind("threadIdx.x", 256):434                local[i] = b[i]435            for i in bind("threadIdx.x", 256):436                c[i] = shared[i] + local[i]437            return c438        a = tvm.placeholder((256, ), dtype='float32', name='a')439        b = tvm.placeholder((256, ), dtype='float32', name='b')440        c = share_vec_add(a, b)441        func, ins, outs = run_and_check(share_vec_add, [a, b], target='cuda')442        run_and_check(func, ins, outs=outs, target='cuda')443    else:444        print('[Warning] No GPU found! Skip shared mem test!')445def test_upstream():446    @tvm.hybrid.script447    def upstream(a):448        b = output_tensor((20, ), 'float32')449        for i in range(20):450            b[i] = a[i] * i451        return b452    a = tvm.placeholder((20, ), 'float32')453    b = tvm.placeholder((20, ), 'float32')454    c = tvm.compute((20, ), lambda x: a[x] + b[x])455    d = upstream(c)456    sch = tvm.create_schedule([c.op, d.op])457    ir = tvm.lower(sch, [a, b, d], simple_mode=True)458    func = tvm.build(sch, [a, b, d])459    assert(func)460    a = numpy.random.randn(20).astype('float32')461    b = numpy.random.randn(20).astype('float32')462    ref = numpy.zeros((20, ), 'float32')463    for i in range(20):464        ref[i] = (a[i] + b[i]) * i465    tvm_a = tvm.nd.array(a)466    tvm_b = tvm.nd.array(b)467    tvm_d = tvm.nd.array(numpy.zeros((20, )).astype('float32'))468    func(tvm_a, tvm_b, tvm_d)469    tvm.testing.assert_allclose(tvm_d.asnumpy(), ref, 1e-5, 1e-5)470def test_downstream():471    @tvm.hybrid.script472    def downstream(a):473        b = output_tensor((20, ), 'float32')474        for i in range(20):475            b[i] = a[i] * i476        return b477    a = tvm.placeholder((20, ), 'float32')478    b = downstream(a)479    c = tvm.compute((20, ), lambda x: b[x] + 1.0)480    sch = tvm.create_schedule(c.op)481    module = tvm.build(sch, [a, c])482    assert module483    a = numpy.random.randn(20).astype('float32')484    ref = numpy.zeros((20, )).astype('float32')485    for i in range(20):486        ref[i] = (a[i] * i) + 1.0487    tvm_a = tvm.nd.array(a)488    tvm_c = tvm.nd.array(numpy.zeros((20, )).astype('float32'))489    module(tvm_a, tvm_c)490    tvm.testing.assert_allclose(tvm_c.asnumpy(), ref, 1e-5, 1e-5)491def test_const_param():492    @tvm.hybrid.script493    def add_something(a, b):494        c = output_tensor((11, ), 'int32')495        for i in range(11):496            c[i] = a[i] + b497        return c498    a = tvm.placeholder((11, ), dtype='int32', name='a')499    b = tvm.const(11, 'int32')500    c = add_something(a, b)501    sch = tvm.create_schedule(c.op)502    module = tvm.build(sch, [a, c], 'llvm')503    assert(module)504    np_a = numpy.arange(11).astype('int32')505    np_b = 11506    np_c = numpy.zeros((11, )).astype('int32')507    nd_a = tvm.ndarray.array(np_a)508    nd_c = tvm.ndarray.array(numpy.zeros((11, )).astype('int32'))509    module(nd_a, nd_c)510    ref = add_something(np_a, 11)511    tvm.testing.assert_allclose(nd_c.asnumpy(), ref, 1e-5, 1e-5)512def test_value_index():513    @tvm.hybrid.script514    def kernel_a(a):515        b = output_tensor((16, ), 'int32')516        c = output_tensor((4, 4), 'int32')517        for i in range(16):518            b[i] = a[i] + 2519            c[i // 4, i % 4] = a[i] + 1520        return b, c521    @tvm.hybrid.script522    def kernel_b(b, a):523        c = output_tensor((4, 4), 'int32')524        for i in range(4):525            for j in range(4):526                c[i, j] = a[i * 4 + j] * b[i, j]527        return c528    a = tvm.placeholder((16, ), 'int32')529    b, c = kernel_a(a)530    d = kernel_b(c, b)531    sch = tvm.create_schedule(d.op)532    module = tvm.build(sch, [a, d])533    assert module534    np_a = numpy.arange(16).astype('int32')535    np_b, np_c = kernel_a(np_a)536    ref = kernel_b(np_c, np_b)537    res = tvm.ndarray.array(numpy.zeros((4, 4)).astype('int32'))538    module(tvm.ndarray.array(np_a), res)539    tvm.testing.assert_allclose(res.asnumpy(), ref)540def test_func_call():541    @tvm.hybrid.script542    def foo(a, b):543        for i in range(len(a)):544            a[i] = i + 1.0545        for i in range(len(a)):546            b[i] = i + 1.0547        c = outer_product(10, 10, a, b)548        d = output_tensor(c.shape, c.dtype)549        for i in range(10):550            for j in range(10):551                d[i, j] = c[i, j] + i * j552        return d553    a = tvm.placeholder((10, ), name='a')554    b = tvm.placeholder((10, ), name='b')555    func, ins, outs = run_and_check(foo, [a, b])556    run_and_check(func, ins, outs=outs)557def test_bool():558    @tvm.hybrid.script559    def foo(a):560        b = output_tensor(a.shape, a.dtype)561        b[0] = 1.2562        for i in range(1, a.shape[0] - 1):563            if a[i] * a[i - 1] < a[i] or a[i] * a[i - 1] < a[i - 1] or i * a[i] == a[i]:564                b[i] = a[i]565            else:566                b[i] = 0.0567        return b568    a = tvm.placeholder((10, ), name='a')569    func, ins, outs = run_and_check(foo, [a])570    run_and_check(func, ins, outs=outs)571def test_const_range():572    @tvm.hybrid.script573    def foo(a, b):574        c = output_tensor(a.shape, a.dtype)575        d = output_tensor(a.shape, 'int32')576        for i in const_range(2):577            for j in const_range(5):578                c[i, j] = float32(int32(a[i, j]) + b[i, j])579        for i in const_range(len(b)):580            for j in const_range(len(b[0])):581                d[i, j] = int32(a[i, j] + b[i, j])582        return c, d583    a = tvm.placeholder((2, 5), name='a', dtype='float32')584    b = [[1, 2, 3, 4, 5], [5, 4, 3, 2, 1]]585    func, ins, outs = run_and_check(foo, [a, b])586    run_and_check(func, ins, outs=outs)587    @tvm.hybrid.script588    def goo(a, b):589        c = output_tensor(a.shape, a.dtype)590        len_b = len(b)591        for i in const_range(len_b * 2):592            if i < len_b:593                c[i] = a[i] + b[i]594            else:595                c[i - len_b] = a[i - len_b] + b[i - len_b]596        return c597    a = tvm.placeholder((5, ), name='a', dtype='int32')598    b = [1, 2, 3, 4, 5]599    c = goo(a, tvm.convert(b))600    sch = tvm.create_schedule(c.op)601    func, ins, outs = run_and_check(goo, [a, b])602    run_and_check(func, ins, outs=outs)603    @tvm.hybrid.script604    def hoo(a, b):605        c = output_tensor(a.shape, a.dtype)606        len_b = len(b)607        for i in range(a.shape[0]):608            for j in const_range(len(b)):609                d = a[i] * b[j]610                d += a[i] + b[j]611                c[i] = d612        return c613    a = tvm.placeholder((5, ), name='a', dtype='int32')614    b = [1, 2, 3, 4, 5]615    func, ins, outs = run_and_check(hoo, [a, b])616    run_and_check(func, ins, outs=outs)617def test_schedule():618    @script619    def outer_product(a, b):620        c = output_tensor((64, 64), a.dtype)621        for i in range(64):622            for j in range(64):623                c[i, j] = a[i] * b[j]624        return c625    a = tvm.placeholder((64,), name='a', dtype='float32')626    b = tvm.placeholder((64,), name='b', dtype='float32')627    c = outer_product(a, b)628    # Test perfect loop split629    # Test loop reorder630    # Test loop annotation631    sch = tvm.create_schedule(c.op)632    i, j = c.op.axis633    io, ii = sch[c].split(i, 4)634    sch[c].parallel(ii)635    jo, ji = sch[c].split(j, 4)636    joo, joi = sch[c].split(jo, 4)637    sch[c].vectorize(ji)638    sch[c].reorder(ii, io, joo, joi, ji)639    ir = tvm.lower(sch, [a, b, c], simple_mode=True)640    assert isinstance(ir, tvm.stmt.ProducerConsumer)641    ir = ir.body642    assert isinstance(ir, tvm.stmt.AttrStmt)643    ir = ir.body644    assert isinstance(ir, tvm.stmt.For)645    assert ir.loop_var.name == 'i.inner'646    ir = ir.body647    assert isinstance(ir, tvm.stmt.For)648    assert ir.loop_var.name == 'i.outer'649    ir = ir.body650    assert isinstance(ir, tvm.stmt.For)651    assert ir.loop_var.name == 'j.outer.outer'652    ir = ir.body653    assert isinstance(ir, tvm.stmt.For)654    assert ir.loop_var.name == 'j.outer.inner'655    ir = ir.body656    func, ins, outs = run_and_check(outer_product, [a, b], sch=sch, outs=[c])657    run_and_check(func, ins, outs=outs)658    # Test fuse659    sch = tvm.create_schedule(c.op)660    sch[c].fuse(c.op.axis[0], c.op.axis[1])661    ir = tvm.lower(sch, [a, b, c], simple_mode=True)662    assert isinstance(ir, tvm.stmt.ProducerConsumer)663    ir = ir.body664    assert isinstance(ir, tvm.stmt.AttrStmt)665    ir = ir.body666    assert isinstance(ir, tvm.stmt.For)667    assert ir.loop_var.name == 'i.j.fused'668    func, ins, outs = run_and_check(outer_product, [a, b], sch=sch, outs=[c])669    run_and_check(func, ins, outs=outs)670    # Test imperfect loop split671    sch = tvm.create_schedule(c.op)672    sch[c].split(c.op.axis[0], 3)673    ir = tvm.lower(sch, [a, b, c], simple_mode=True)674    func, ins, outs = run_and_check(outer_product, [a, b], sch=sch, outs=[c])675    run_and_check(func, ins, outs=outs)676    # Test loop binds677def test_capture():678    n = 8679    constant_tuple = (10, n)680    constant_list = [[1, 2], [3, n]]681    const_value = 1682    @tvm.hybrid.script683    def add_something(a):684        c = output_tensor((constant_tuple[1],), 'int32')685        for i in range(constant_tuple[1]):686            c[i] = a[i] + constant_list[1][const_value]687        return c688    a = tvm.placeholder((n, ), dtype='int32', name='a')689    func, ins, outs = run_and_check(add_something, [a])690    run_and_check(func, ins, outs=outs)691def test_array_inputs():692    @script693    def sum_array(inputs):694        out = output_tensor((10,), inputs[0].dtype)695        n = len(inputs)696        for i in range(10):697            for j in const_range(n):698                out[i] += inputs[j][i]699        return out700    n = 5701    inputs = []702    for i in range(n):703        inputs.append(tvm.placeholder((10,), name='t%s' % i, dtype='float32'))704    out = sum_array(tvm.convert(inputs))...

test__refcount.py

Source:test__refcount.py

...43        eventlet.spawn(make_request, addr)44    eventlet.sleep(0.1 + SOCKET_TIMEOUT)45    print('run_interaction: refcount(s.fd)', sys.getrefcount(s.fd))46    return weakref.ref(s.fd)47def run_and_check(run_client):48    w = run_interaction(run_client=run_client)49    # clear_sys_exc_info()50    gc.collect()51    fd = w()52    print('run_and_check: weakref fd:', fd)53    if fd:54        print(pprint.pformat(gc.get_referrers(fd)))55        for x in gc.get_referrers(fd):56            print(pprint.pformat(x))57            for y in gc.get_referrers(x):58                print('- {0}'.format(pprint.pformat(y)))59        raise AssertionError('server should be dead by now')60def test_clean_exit():61    run_and_check(True)62    run_and_check(True)63def test_timeout_exit():64    run_and_check(False)...

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.