tests/test_vectorization.py

+import numpy as np
+
+import pytest
+
+import pystencils.config
+import sympy as sp
+
+import pystencils as ps
+import pystencils.astnodes as ast
+from pystencils.backends.simd_instruction_sets import get_supported_instruction_sets, get_vector_instruction_set
+from pystencils.cpu.vectorization import vectorize
+from pystencils.enums import Target
+from pystencils.transformations import replace_inner_stride_with_one
+
+supported_instruction_sets = get_supported_instruction_sets()
+if supported_instruction_sets:
+    instruction_set = supported_instruction_sets[-1]
+else:
+    instruction_set = None
+
+
+# TODO: Skip tests if no instruction set is available and check all codes if they are really vectorised !
+def test_vector_type_propagation1(instruction_set=instruction_set):
+    a, b, c, d, e = sp.symbols("a b c d e")
+    arr = np.ones((2 ** 2 + 2, 2 ** 3 + 2))
+    arr *= 10.0
+
+    f, g = ps.fields(f=arr, g=arr)
+    update_rule = [ps.Assignment(a, f[1, 0]),
+                   ps.Assignment(b, a),
+                   ps.Assignment(g[0, 0], b + 3 + f[0, 1])]
+
+    ast = ps.create_kernel(update_rule)
+    vectorize(ast, instruction_set=instruction_set)
+
+    # ps.show_code(ast)
+
+    func = ast.compile()
+    dst = np.zeros_like(arr)
+    func(g=dst, f=arr)
+    np.testing.assert_equal(dst[1:-1, 1:-1], 2 * 10.0 + 3)
+
+
+def test_vector_type_propagation2(instruction_set=instruction_set):
+    data_type = 'float32'
+    assume_aligned = True
+    assume_inner_stride_one = True
+    assume_sufficient_line_padding = True
+
+    domain_size = (24, 24)
+    dh = ps.create_data_handling(domain_size, default_target=Target.CPU)
+
+    # fields
+    g = dh.add_array("g", values_per_cell=1, dtype=data_type, alignment=True, ghost_layers=0)
+    f = dh.add_array("f", values_per_cell=1, dtype=data_type, alignment=True, ghost_layers=0)
+    dh.fill("g", 1.0, ghost_layers=True)
+    dh.fill("f", 0.5, ghost_layers=True)
+
+    collision = [
+        ps.Assignment(sp.Symbol("b"), sp.sqrt(sp.Symbol("a") * (1 - (1 - f.center)**2))),
+        ps.Assignment(g.center, sp.Symbol("b"))
+    ]
+
+    config = ps.CreateKernelConfig(
+        target=ps.Target.CPU,
+        data_type=data_type,
+        default_number_float=data_type,
+        cpu_vectorize_info={
+            'assume_aligned': assume_aligned,
+            'assume_inner_stride_one': assume_inner_stride_one,
+            'assume_sufficient_line_padding': assume_sufficient_line_padding,
+        },
+        ghost_layers=0
+    )
+
+    ast = ps.create_kernel(collision, config=config)
+    print(ast)
+    code = ps.get_code_str(ast)
+    print(code)
+
+    kernel = ast.compile()
+    dh.run_kernel(kernel, a=4)
+    g_arr = dh.cpu_arrays['g']
+    np.testing.assert_allclose(g_arr, np.full_like(g_arr, np.sqrt(3)))
+
+
+def test_vectorize_moved_constants1(instruction_set=instruction_set):
+    opt = {'instruction_set': instruction_set, 'assume_inner_stride_one': True}
+
+    f = ps.fields("f: [1D]")
+    x = ast.TypedSymbol("x", np.float64)
+
+    kernel_func = ps.create_kernel(
+        [ast.SympyAssignment(x, 2.0), ast.SympyAssignment(f[0], x)],
+        cpu_prepend_optimizations=[ps.transformations.move_constants_before_loop],  # explicitly move constants
+        cpu_vectorize_info=opt,
+    )
+    ps.show_code(kernel_func)  # fails if `x` on rhs was not correctly vectorized
+    kernel = kernel_func.compile()
+
+    f_arr = np.zeros(9)
+    kernel(f=f_arr)
+
+    assert(np.all(f_arr == 2))
+
+
+def test_vectorize_moved_constants2(instruction_set=instruction_set):
+    opt = {'instruction_set': instruction_set, 'assume_inner_stride_one': True}
+
+    f = ps.fields("f: [1D]")
+    x = ast.TypedSymbol("x", np.float64)
+    y = ast.TypedSymbol("y", np.float64)
+
+    kernel_func = ps.create_kernel(
+        [ast.SympyAssignment(x, 2.0), ast.SympyAssignment(y, 3.0), ast.SympyAssignment(f[0], x + y)],
+        cpu_prepend_optimizations=[ps.transformations.move_constants_before_loop],  # explicitly move constants
+        cpu_vectorize_info=opt,
+    )
+    ps.show_code(kernel_func)  # fails if `x` on rhs was not correctly vectorized
+    kernel = kernel_func.compile()
+
+    f_arr = np.zeros(9)
+    kernel(f=f_arr)
+
+    assert(np.all(f_arr == 5))
+
+
+@pytest.mark.parametrize('openmp', [True, False])
+def test_aligned_and_nt_stores(openmp, instruction_set=instruction_set):
+    domain_size = (24, 24)
+    # create a datahandling object
+    dh = ps.create_data_handling(domain_size, periodicity=(True, True), parallel=False, default_target=Target.CPU)
+
+    # fields
+    alignment = 'cacheline' if openmp else True
+    g = dh.add_array("g", values_per_cell=1, alignment=alignment)
+    dh.fill("g", 1.0, ghost_layers=True)
+    f = dh.add_array("f", values_per_cell=1, alignment=alignment)
+    dh.fill("f", 0.0, ghost_layers=True)
+    opt = {'instruction_set': instruction_set, 'assume_aligned': True, 'nontemporal': True,
+           'assume_inner_stride_one': True}
+    update_rule = [ps.Assignment(f.center(), 0.25 * (g[-1, 0] + g[1, 0] + g[0, -1] + g[0, 1]))]
+    # Without the base pointer spec, the inner store is not aligned
+    config = pystencils.config.CreateKernelConfig(target=dh.default_target, cpu_vectorize_info=opt, cpu_openmp=openmp)
+    ast = ps.create_kernel(update_rule, config=config)
+    if instruction_set in ['sse'] or instruction_set.startswith('avx') or instruction_set.startswith('sve'):
+        assert 'stream' in ast.instruction_set
+        assert 'streamFence' in ast.instruction_set
+    if instruction_set in ['neon', 'vsx', 'rvv']:
+        assert 'cachelineZero' in ast.instruction_set
+    if instruction_set in ['vsx']:
+        assert 'storeAAndFlushCacheline' in ast.instruction_set
+    for instruction in ['stream', 'streamFence', 'cachelineZero', 'storeAAndFlushCacheline', 'flushCacheline']:
+        if instruction in ast.instruction_set:
+            assert ast.instruction_set[instruction].split('{')[0] in ps.get_code_str(ast)
+    kernel = ast.compile()
+
+    # ps.show_code(ast)
+
+    dh.run_kernel(kernel)
+    np.testing.assert_equal(np.sum(dh.cpu_arrays['f']), np.prod(domain_size))
+
+
+def test_nt_stores_symbolic_size(instruction_set=instruction_set):
+    f, g = ps.fields('f, g: [2D]', layout='fzyx')
+    update_rule = [ps.Assignment(f.center(), 0.0), ps.Assignment(g.center(), 0.0)]
+    opt = {'instruction_set': instruction_set, 'assume_aligned': True, 'nontemporal': True,
+           'assume_inner_stride_one': True}
+    config = pystencils.config.CreateKernelConfig(target=Target.CPU, cpu_vectorize_info=opt)
+    ast = ps.create_kernel(update_rule, config=config)
+    # ps.show_code(ast)
+    ast.compile()
+
+
+def test_inplace_update(instruction_set=instruction_set):
+    shape = (9, 9, 3)
+    arr = np.ones(shape, order='f')
+
+    @ps.kernel
+    def update_rule(s):
+        f = ps.fields("f(3) : [2D]", f=arr)
+        s.tmp0 @= f(0)
+        s.tmp1 @= f(1)
+        s.tmp2 @= f(2)
+        f0, f1, f2 = f(0), f(1), f(2)
+        f0 @= 2 * s.tmp0
+        f1 @= 2 * s.tmp0
+        f2 @= 2 * s.tmp0
+
+    config = pystencils.config.CreateKernelConfig(cpu_vectorize_info={'instruction_set': instruction_set})
+    ast = ps.create_kernel(update_rule, config=config)
+    kernel = ast.compile()
+    kernel(f=arr)
+    np.testing.assert_equal(arr, 2)
+
+
+def test_vectorization_fixed_size(instruction_set=instruction_set):
+    instructions = get_vector_instruction_set(instruction_set=instruction_set)
+    configurations = []
+    # Fixed size - multiple of four
+    arr = np.ones((20 + 2, 24 + 2)) * 5.0
+    f, g = ps.fields(f=arr, g=arr)
+    configurations.append((arr, f, g))
+    # Fixed size - no multiple of four
+    arr = np.ones((21 + 2, 25 + 2)) * 5.0
+    f, g = ps.fields(f=arr, g=arr)
+    configurations.append((arr, f, g))
+    # Fixed size - different remainder
+    arr = np.ones((23 + 2, 17 + 2)) * 5.0
+    f, g = ps.fields(f=arr, g=arr)
+    configurations.append((arr, f, g))
+
+    for arr, f, g in configurations:
+        update_rule = [ps.Assignment(g[0, 0], f[0, 0] + f[-1, 0] + f[1, 0] + f[0, 1] + f[0, -1] + 42.0)]
+
+        ast = ps.create_kernel(update_rule)
+        vectorize(ast, instruction_set=instruction_set)
+        code = ps.get_code_str(ast)
+        add_instruction = instructions["+"][:instructions["+"].find("(")]
+        assert add_instruction in code
+        # print(code)
+
+        func = ast.compile()
+        dst = np.zeros_like(arr)
+        func(g=dst, f=arr)
+        np.testing.assert_equal(dst[1:-1, 1:-1], 5 * 5.0 + 42.0)
+
+
+def test_vectorization_variable_size(instruction_set=instruction_set):
+    f, g = ps.fields("f, g : double[2D]")
+    update_rule = [ps.Assignment(g[0, 0], f[0, 0] + f[-1, 0] + f[1, 0] + f[0, 1] + f[0, -1] + 42.0)]
+    ast = ps.create_kernel(update_rule)
+
+    replace_inner_stride_with_one(ast)
+    vectorize(ast, instruction_set=instruction_set)
+    func = ast.compile()
+
+    arr = np.ones((23 + 2, 17 + 2)) * 5.0
+    dst = np.zeros_like(arr)
+
+    func(g=dst, f=arr)
+    np.testing.assert_equal(dst[1:-1, 1:-1], 5 * 5.0 + 42.0)
+
+
+def test_piecewise1(instruction_set=instruction_set):
+    a, b, c, d, e = sp.symbols("a b c d e")
+    arr = np.ones((2 ** 3 + 2, 2 ** 4 + 2)) * 5.0
+
+    f, g = ps.fields(f=arr, g=arr)
+    update_rule = [ps.Assignment(a, f[1, 0]),
+                   ps.Assignment(b, a),
+                   ps.Assignment(c, f[0, 0] > 0.0),
+                   ps.Assignment(g[0, 0], sp.Piecewise((b + 3 + f[0, 1], c), (0.0, True)))]
+
+    ast = ps.create_kernel(update_rule)
+    vectorize(ast, instruction_set=instruction_set)
+    func = ast.compile()
+    dst = np.zeros_like(arr)
+    func(g=dst, f=arr)
+    np.testing.assert_equal(dst[1:-1, 1:-1], 5 + 3 + 5.0)
+
+
+def test_piecewise2(instruction_set=instruction_set):
+    arr = np.zeros((20, 20))
+
+    @ps.kernel
+    def test_kernel(s):
+        f, g = ps.fields(f=arr, g=arr)
+
+        s.condition @= f[0, 0] > 1
+        s.result    @= 0.0 if s.condition else 1.0
+        g[0, 0]     @= s.result
+
+    ast = ps.create_kernel(test_kernel)
+    # ps.show_code(ast)
+    vectorize(ast, instruction_set=instruction_set)
+    # ps.show_code(ast)
+    func = ast.compile()
+    func(f=arr, g=arr)
+    np.testing.assert_equal(arr, np.ones_like(arr))
+
+
+def test_piecewise3(instruction_set=instruction_set):
+    arr = np.zeros((22, 22))
+
+    @ps.kernel
+    def test_kernel(s):
+        f, g = ps.fields(f=arr, g=arr)
+        s.b     @= f[0, 1]
+        g[0, 0] @= 1.0 / (s.b + s.k) if f[0, 0] > 0.0 else 1.0
+
+    ast = ps.create_kernel(test_kernel)
+    # ps.show_code(ast)
+    vectorize(ast, instruction_set=instruction_set)
+    # ps.show_code(ast)
+    ast.compile()
+
+
+def test_logical_operators(instruction_set=instruction_set):
+    arr = np.zeros((22, 22))
+
+    @ps.kernel
+    def kernel_and(s):
+        f, g = ps.fields(f=arr, g=arr)
+        s.c @= sp.And(f[0, 1] < 0.0, f[1, 0] < 0.0)
+        g[0, 0] @= sp.Piecewise([1.0 / f[1, 0], s.c], [1.0, True])
+
+    ast = ps.create_kernel(kernel_and)
+    vectorize(ast, instruction_set=instruction_set)
+    ast.compile()
+
+    @ps.kernel
+    def kernel_or(s):
+        f, g = ps.fields(f=arr, g=arr)
+        s.c @= sp.Or(f[0, 1] < 0.0, f[1, 0] < 0.0)
+        g[0, 0] @= sp.Piecewise([1.0 / f[1, 0], s.c], [1.0, True])
+
+    ast = ps.create_kernel(kernel_or)
+    vectorize(ast, instruction_set=instruction_set)
+    ast.compile()
+
+    @ps.kernel
+    def kernel_equal(s):
+        f, g = ps.fields(f=arr, g=arr)
+        s.c @= sp.Eq(f[0, 1], 2.0)
+        g[0, 0] @= sp.Piecewise([1.0 / f[1, 0], s.c], [1.0, True])
+
+    ast = ps.create_kernel(kernel_equal)
+    vectorize(ast, instruction_set=instruction_set)
+    ast.compile()
+
+
+def test_hardware_query():
+    assert {'sse', 'neon', 'sve', 'sve2', 'sme', 'vsx', 'rvv'}.intersection(supported_instruction_sets)
+
+
+def test_vectorised_pow(instruction_set=instruction_set):
+    arr = np.zeros((24, 24))
+    f, g = ps.fields(f=arr, g=arr)
+
+    as1 = ps.Assignment(g[0, 0], sp.Pow(f[0, 0], 2))
+    as2 = ps.Assignment(g[0, 0], sp.Pow(f[0, 0], 0.5))
+    as3 = ps.Assignment(g[0, 0], sp.Pow(f[0, 0], -0.5))
+    as4 = ps.Assignment(g[0, 0], sp.Pow(f[0, 0], 4))
+    as5 = ps.Assignment(g[0, 0], sp.Pow(f[0, 0], -4))
+    as6 = ps.Assignment(g[0, 0], sp.Pow(f[0, 0], -1))
+
+    ast = ps.create_kernel(as1)
+    vectorize(ast, instruction_set=instruction_set)
+    ast.compile()
+
+    ast = ps.create_kernel(as2)
+    vectorize(ast, instruction_set=instruction_set)
+    ast.compile()
+
+    ast = ps.create_kernel(as3)
+    vectorize(ast, instruction_set=instruction_set)
+    ast.compile()
+
+    ast = ps.create_kernel(as4)
+    vectorize(ast, instruction_set=instruction_set)
+    ast.compile()
+
+    ast = ps.create_kernel(as5)
+    vectorize(ast, instruction_set=instruction_set)
+    ast.compile()
+
+    ast = ps.create_kernel(as6)
+    vectorize(ast, instruction_set=instruction_set)
+    ast.compile()
+
+
+def test_issue40(*_):
+    """https://i10git.cs.fau.de/pycodegen/pystencils/-/issues/40"""
+    opt = {'instruction_set': "avx512", 'assume_aligned': False,
+           'nontemporal': False, 'assume_inner_stride_one': True}
+
+    src = ps.fields("src(1): double[2D]", layout='fzyx')
+    eq = [ps.Assignment(sp.Symbol('rho'), 1.0),
+          ps.Assignment(src[0, 0](0), sp.Rational(4, 9) * sp.Symbol('rho'))]
+
+    config = pystencils.config.CreateKernelConfig(target=Target.CPU, cpu_vectorize_info=opt, data_type='float64')
+    ast = ps.create_kernel(eq, config=config)
+
+    code = ps.get_code_str(ast)
+    assert 'epi32' not in code

tests/test_vectorization_specific.py

+import pytest
+
+import numpy as np
+
+import pystencils.config
+import sympy as sp
+
+import pystencils as ps
+from pystencils.backends.simd_instruction_sets import (get_cacheline_size, get_supported_instruction_sets,
+                                                       get_vector_instruction_set)
+from pystencils.enums import Target
+from pystencils.typing import CFunction
+from . import test_vectorization
+
+
+supported_instruction_sets = get_supported_instruction_sets() if get_supported_instruction_sets() else []
+
+
+@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+def test_vectorisation_varying_arch(instruction_set):
+    shape = (9, 9, 3)
+    arr = np.ones(shape, order='f')
+
+    @ps.kernel
+    def update_rule(s):
+        f = ps.fields("f(3) : [2D]", f=arr)
+        s.tmp0 @= f(0)
+        s.tmp1 @= f(1)
+        s.tmp2 @= f(2)
+        f0, f1, f2 = f(0), f(1), f(2)
+        f0 @= 2 * s.tmp0
+        f1 @= 2 * s.tmp0
+        f2 @= 2 * s.tmp0
+
+    config = pystencils.config.CreateKernelConfig(cpu_vectorize_info={'instruction_set': instruction_set})
+    ast = ps.create_kernel(update_rule, config=config)
+    kernel = ast.compile()
+    kernel(f=arr)
+    np.testing.assert_equal(arr, 2)
+
+
+@pytest.mark.parametrize('dtype', ('float32', 'float64'))
+@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+def test_vectorized_abs_field(instruction_set, dtype):
+    """Some instructions sets have abs, some don't.
+       Furthermore, the special treatment of unary minus makes this data type-sensitive too.
+    """
+    arr = np.ones((2 ** 2 + 2, 2 ** 3 + 2), dtype=dtype)
+    arr[-3:, :] = -1
+
+    f, g = ps.fields(f=arr, g=arr)
+    update_rule = [ps.Assignment(g.center(), sp.Abs(f.center()))]
+
+    config = pystencils.config.CreateKernelConfig(cpu_vectorize_info={'instruction_set': instruction_set})
+    ast = ps.create_kernel(update_rule, config=config)
+
+    func = ast.compile()
+    dst = np.zeros_like(arr)
+    func(g=dst, f=arr)
+    np.testing.assert_equal(np.sum(dst[1:-1, 1:-1]), 2 ** 2 * 2 ** 3)
+
+
+@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+def test_vectorized_abs_scalar(instruction_set):
+    """Some instructions sets have abs, some don't.
+       Furthermore, the special treatment of unary minus makes this data type-sensitive too.
+    """
+    arr = np.zeros((2 ** 2 + 2, 2 ** 3 + 2), dtype="float64")
+
+    f = ps.fields(f=arr)
+    update_rule = [ps.Assignment(f.center(), sp.Abs(sp.Symbol("a")))]
+
+    config = pystencils.config.CreateKernelConfig(cpu_vectorize_info={'instruction_set': instruction_set})
+    ast = ps.create_kernel(update_rule, config=config)
+
+    func = ast.compile()
+    func(f=arr, a=-1)
+    np.testing.assert_equal(np.sum(arr[1:-1, 1:-1]), 2 ** 2 * 2 ** 3)
+
+
+@pytest.mark.parametrize('dtype', ('float32', 'float64'))
+@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+@pytest.mark.parametrize('nontemporal', [False, True])
+def test_strided(instruction_set, dtype, nontemporal):
+    f, g = ps.fields(f"f, g : {dtype}[2D]")
+    update_rule = [ps.Assignment(g[0, 0], f[0, 0] + f[-1, 0] + f[1, 0] + f[0, 1] + f[0, -1] + 42.0)]
+    config = pystencils.config.CreateKernelConfig(cpu_vectorize_info={'instruction_set': instruction_set,
+                                                                      'nontemporal': nontemporal},
+                                                  default_number_float=dtype)
+    if 'storeS' not in get_vector_instruction_set(dtype, instruction_set) \
+            and instruction_set not in ['avx512', 'avx512vl', 'rvv'] and not instruction_set.startswith('sve'):
+        with pytest.warns(UserWarning) as warn:
+            ast = ps.create_kernel(update_rule, config=config)
+            assert 'Could not vectorize loop' in warn[0].message.args[0]
+    else:
+        with pytest.warns(None) as warn:
+            ast = ps.create_kernel(update_rule, config=config)
+            assert len(warn) == 0
+        instruction = 'streamS' if nontemporal and 'streamS' in ast.instruction_set else 'storeS'
+        assert ast.instruction_set[instruction].split('{')[0] in ps.get_code_str(ast)
+    instruction = 'cachelineZero'
+    if instruction in ast.instruction_set:
+        assert ast.instruction_set[instruction] not in ps.get_code_str(ast)
+
+    # ps.show_code(ast)
+    func = ast.compile()
+    ref_config = pystencils.config.CreateKernelConfig(default_number_float=dtype)
+    ref_func = ps.create_kernel(update_rule, config=ref_config).compile()
+
+    # For some reason other array creations fail on the emulated ppc pipeline
+    size = (25, 19)
+    arr = np.zeros(size).astype(dtype)
+    for i in range(size[0]):
+        for j in range(size[1]):
+            arr[i, j] = i * j
+
+    dst = np.zeros_like(arr, dtype=dtype)
+    ref = np.zeros_like(arr, dtype=dtype)
+
+    func(g=dst, f=arr)
+    ref_func(g=ref, f=arr)
+
+    # print("dst: ", dst)
+    # print("np array: ", arr)
+
+    np.testing.assert_almost_equal(dst[1:-1, 1:-1], ref[1:-1, 1:-1], 13 if dtype == 'float64' else 5)
+
+
+@pytest.mark.parametrize('dtype', ('float32', 'float64'))
+@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+@pytest.mark.parametrize('gl_field, gl_kernel', [(1, 0), (0, 1), (1, 1)])
+def test_alignment_and_correct_ghost_layers(gl_field, gl_kernel, instruction_set, dtype):
+    domain_size = (128, 128)
+    dh = ps.create_data_handling(domain_size, periodicity=(True, True), default_target=Target.CPU)
+    src = dh.add_array("src", values_per_cell=1, dtype=dtype, ghost_layers=gl_field, alignment=True)
+    dh.fill(src.name, 1.0, ghost_layers=True)
+    dst = dh.add_array("dst", values_per_cell=1, dtype=dtype, ghost_layers=gl_field, alignment=True)
+    dh.fill(dst.name, 1.0, ghost_layers=True)
+
+    update_rule = ps.Assignment(dst[0, 0], src[0, 0])
+    opt = {'instruction_set': instruction_set, 'assume_aligned': True,
+           'nontemporal': True, 'assume_inner_stride_one': True}
+    config = pystencils.config.CreateKernelConfig(target=dh.default_target,
+                                                  cpu_vectorize_info=opt, ghost_layers=gl_kernel)
+    ast = ps.create_kernel(update_rule, config=config)
+    kernel = ast.compile()
+    if ('loadA' in ast.instruction_set or 'storeA' in ast.instruction_set) and gl_kernel != gl_field:
+        with pytest.raises(ValueError):
+            dh.run_kernel(kernel)
+    else:
+        dh.run_kernel(kernel)
+
+
+@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+def test_cacheline_size(instruction_set):
+    cacheline_size = get_cacheline_size(instruction_set)
+    if cacheline_size is None and instruction_set in ['sse', 'avx', 'avx512', 'avx512vl', 'rvv']:
+        pytest.skip()
+    instruction_set = get_vector_instruction_set('double', instruction_set)
+    vector_size = instruction_set['bytes']
+    assert 8 < cacheline_size < 0x100000, "Cache line size is implausible"
+    if type(vector_size) is int:
+        assert cacheline_size % vector_size == 0, "Cache line size should be multiple of vector size"
+    assert cacheline_size & (cacheline_size - 1) == 0, "Cache line size is not a power of 2"
+
+
+# TODO move to vectorise
+@pytest.mark.parametrize('instruction_set',
+                         sorted(set(supported_instruction_sets) - {test_vectorization.instruction_set}))
+@pytest.mark.parametrize('function',
+                         [f for f in test_vectorization.__dict__ if f.startswith('test_') and f not in ['test_hardware_query', 'test_aligned_and_nt_stores']])
+def test_vectorization_other(instruction_set, function):
+    test_vectorization.__dict__[function](instruction_set)
+
+
+@pytest.mark.parametrize('dtype', ('float32', 'float64'))
+@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+@pytest.mark.parametrize('field_layout', ('fzyx', 'zyxf'))
+def test_square_root(dtype, instruction_set, field_layout):
+    config = pystencils.config.CreateKernelConfig(data_type=dtype,
+                                                  default_number_float=dtype,
+                                                  cpu_vectorize_info={'instruction_set': instruction_set,
+                                                                      'assume_inner_stride_one': True,
+                                                                      'assume_aligned': False,
+                                                                      'nontemporal': False})
+
+    src_field = ps.Field.create_generic('pdfs', 2, dtype, index_dimensions=1, layout=field_layout, index_shape=(9,))
+
+    eq = [ps.Assignment(sp.Symbol("xi"), sum(src_field.center_vector)),
+          ps.Assignment(sp.Symbol("xi_2"), sp.Symbol("xi") * sp.sqrt(src_field.center))]
+
+    ast = ps.create_kernel(eq, config=config)
+    ast.compile()
+    code = ps.get_code_str(ast)
+    print(code)
+
+
+@pytest.mark.parametrize('dtype', ('float32', 'float64'))
+@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+@pytest.mark.parametrize('padding', (True, False))
+def test_square_root_2(dtype, instruction_set, padding):
+    x, y = sp.symbols("x y")
+    src = ps.fields(f"src: {dtype}[2D]", layout='fzyx')
+
+    up = ps.Assignment(src[0, 0], 1 / x + sp.sqrt(y * 0.52 + x ** 2))
+
+    cpu_vec = {'instruction_set': instruction_set, 'assume_inner_stride_one': True,
+               'assume_sufficient_line_padding': padding,
+               'assume_aligned': True}
+
+    config = ps.CreateKernelConfig(data_type=dtype, default_number_float=dtype, cpu_vectorize_info=cpu_vec)
+    ast = ps.create_kernel(up, config=config)
+    ast.compile()
+
+    code = ps.get_code_str(ast)
+    print(code)
+
+
+@pytest.mark.parametrize('dtype', ('float32', 'float64'))
+@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+@pytest.mark.parametrize('padding', (True, False))
+def test_pow(dtype, instruction_set, padding):
+    config = pystencils.config.CreateKernelConfig(data_type=dtype,
+                                                  default_number_float=dtype,
+                                                  cpu_vectorize_info={'instruction_set': instruction_set,
+                                                                      'assume_inner_stride_one': True,
+                                                                      'assume_sufficient_line_padding': padding,
+                                                                      'assume_aligned': False, 'nontemporal': False})
+
+    src_field = ps.Field.create_generic('pdfs', 2, dtype, index_dimensions=1, layout='fzyx', index_shape=(9,))
+
+    eq = [ps.Assignment(sp.Symbol("xi"), sum(src_field.center_vector)),
+          ps.Assignment(sp.Symbol("xi_2"), sp.Symbol("xi") * sp.Pow(src_field.center, 0.5))]
+
+    ast = ps.create_kernel(eq, config=config)
+    ast.compile()
+    code = ps.get_code_str(ast)
+
+    print(code)
+
+
+@pytest.mark.parametrize('dtype', ('float32', 'float64'))
+@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+@pytest.mark.parametrize('padding', (True, False))
+def test_issue62(dtype, instruction_set, padding):
+    opt = {'instruction_set': instruction_set, 'assume_aligned': True,
+           'assume_inner_stride_one': True,
+           'assume_sufficient_line_padding': padding}
+
+    dx = sp.Symbol("dx")
+    dy = sp.Symbol("dy")
+    src, dst, rhs = ps.fields(f"src, src_tmp, rhs: {dtype}[2D]", layout='fzyx')
+
+    up = ps.Assignment(dst[0, 0], ((dy ** 2 * (src[1, 0] + src[-1, 0]))
+                                   + (dx ** 2 * (src[0, 1] + src[0, -1]))
+                                   - (rhs[0, 0] * dx ** 2 * dy ** 2)) / (2 * (dx ** 2 + dy ** 2)))
+
+    config = ps.CreateKernelConfig(data_type=dtype,
+                                   default_number_float=dtype,
+                                   cpu_vectorize_info=opt)
+
+    ast = ps.create_kernel(up, config=config)
+    ast.compile()
+    code = ps.get_code_str(ast)
+
+    print(code)
+
+    assert 'pow' not in code
+
+
+@pytest.mark.parametrize('dtype', ('float32', 'float64'))
+@pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+def test_div_and_unevaluated_expr(dtype, instruction_set):
+    opt = {'instruction_set': instruction_set, 'assume_aligned': True, 'assume_inner_stride_one': True,
+           'assume_sufficient_line_padding': False}
+
+    x, y, z = sp.symbols("x y z")
+    rhs = (-4 * x ** 4 * y ** 2 * z ** 2 + (x ** 4 * y ** 2 / 3) + (x ** 4 * z ** 2 / 3)) / x ** 3
+
+    src = ps.fields(f"src: {dtype}[2D]", layout='fzyx')
+
+    up = ps.Assignment(src[0, 0], rhs)
+
+    config = ps.CreateKernelConfig(data_type=dtype,
+                                   default_number_float=dtype,
+                                   cpu_vectorize_info=opt)
+
+    ast = ps.create_kernel(up, config=config)
+    code = ps.get_code_str(ast)
+    # print(code)
+
+    ast.compile()
+
+    assert 'pow' not in code
+
+
+@pytest.mark.parametrize('dtype', ('float32', 'float64'))
+@pytest.mark.parametrize('instruction_set', ('sve', 'sve2', 'sme', 'rvv'))
+def test_check_ast_parameters_sizeless(dtype, instruction_set):
+    f, g = ps.fields(f"f, g: {dtype}[3D]", layout='fzyx')
+
+    update_rule = [ps.Assignment(g.center(), 2 * f.center())]
+
+    config = pystencils.config.CreateKernelConfig(data_type=dtype,
+                                                  cpu_vectorize_info={'instruction_set': instruction_set})
+    ast = ps.create_kernel(update_rule, config=config)
+    ast_symbols = [p.symbol for p in ast.get_parameters()]
+    assert ast.instruction_set['width'] not in ast_symbols
+    assert ast.instruction_set['intwidth'] not in ast_symbols
+
+
+
+# TODO this test case needs a complete rework of the vectoriser. The reason is that the vectoriser does not
+# TODO vectorise symbols at the moment because they could be strides or field sizes, thus involved in pointer arithmetic
+# TODO This means that the vectoriser only works if fields are involved on the rhs.
+# @pytest.mark.parametrize('dtype', ('float32', 'float64'))
+# @pytest.mark.parametrize('instruction_set', supported_instruction_sets)
+# def test_vectorised_symbols(dtype, instruction_set):
+#     opt = {'instruction_set': instruction_set, 'assume_aligned': True, 'assume_inner_stride_one': True,
+#            'assume_sufficient_line_padding': False}
+#
+#     x, y, z = sp.symbols("x y z")
+#     rhs = 1 / x ** 2 * (x + y)
+#
+#     src = ps.fields(f"src: {dtype}[2D]", layout='fzyx')
+#
+#     up = ps.Assignment(src[0, 0], rhs)
+#
+#     config = ps.CreateKernelConfig(data_type=dtype,
+#                                    default_number_float=dtype,
+#                                    cpu_vectorize_info=opt)
+#
+#     ast = ps.create_kernel(up, config=config)
+#     code = ps.get_code_str(ast)
+#     print(code)
+#
+#     ast.compile()
+#
+#     assert 'pow' not in code

tests/test_version_string.py

+import pystencils as ps
+
+
+def test_version_string():
+    version = ps.__version__
+    print(version)
+
+    numeric_version = [int(x, 10) for x in version.split('.')[0:1]]
+    test_version = sum(x * (100 ** i) for i, x in enumerate(numeric_version))
+
+    assert test_version >= 1