Forked from
pycodegen / pystencils
763 commits behind, 22 commits ahead of the upstream repository.
-
Stephan Seitz authoredStephan Seitz authored
kernelcreation.py 17.79 KiB
import functools
from itertools import combinations
from types import MappingProxyType
import sympy as sp
from pystencils.assignment import Assignment
from pystencils.astnodes import Block, Conditional, LoopOverCoordinate, SympyAssignment
from pystencils.cpu.vectorization import vectorize
from pystencils.field import Field, FieldType
from pystencils.gpucuda.indexing import indexing_creator_from_params
from pystencils.simp.assignment_collection import AssignmentCollection
from pystencils.stencil import direction_string_to_offset, inverse_direction_string
from pystencils.transformations import (
loop_blocking, move_constants_before_loop, remove_conditionals_in_staggered_kernel)
def create_kernel(assignments,
target='cpu',
data_type="double",
iteration_slice=None,
ghost_layers=None,
skip_independence_check=False,
cpu_openmp=False,
cpu_vectorize_info=None,
cpu_blocking=None,
gpu_indexing='block',
gpu_indexing_params=MappingProxyType({}),
use_textures_for_interpolation=True,
cpu_prepend_optimizations=[],
use_auto_for_assignments=False,
opencl_queue=None,
opencl_ctx=None):
"""
Creates abstract syntax tree (AST) of kernel, using a list of update equations.
Args:
assignments: can be a single assignment, sequence of assignments or an `AssignmentCollection`
target: 'cpu', 'llvm', 'gpu' or 'opencl'
data_type: data type used for all untyped symbols (i.e. non-fields), can also be a dict from symbol name
to type
iteration_slice: rectangular subset to iterate over, if not specified the complete non-ghost layer \
part of the field is iterated over
ghost_layers: a single integer specifies the ghost layer count at all borders, can also be a sequence of
pairs ``[(x_lower_gl, x_upper_gl), .... ]``. These layers are excluded from the iteration.
If left to default, the number of ghost layers is determined automatically.
skip_independence_check: don't check that loop iterations are independent. This is needed e.g. for
periodicity kernel, that access the field outside the iteration bounds. Use with care!
cpu_openmp: True or number of threads for OpenMP parallelization, False for no OpenMP
cpu_vectorize_info: a dictionary with keys, 'vector_instruction_set', 'assume_aligned' and 'nontemporal'
for documentation of these parameters see vectorize function. Example:
'{'instruction_set': 'avx512', 'assume_aligned': True, 'nontemporal':True}'
cpu_blocking: a tuple of block sizes or None if no blocking should be applied
gpu_indexing: either 'block' or 'line' , or custom indexing class, see `AbstractIndexing`
gpu_indexing_params: dict with indexing parameters (constructor parameters of indexing class)
e.g. for 'block' one can specify '{'block_size': (20, 20, 10) }'
cpu_prepend_optimizations: list of extra optimizations to perform first on the AST
Returns:
abstract syntax tree (AST) object, that can either be printed as source code with `show_code` or
can be compiled with through its 'compile()' member
Example:
>>> import pystencils as ps
>>> import numpy as np
>>> s, d = ps.fields('s, d: [2D]')
>>> assignment = ps.Assignment(d[0,0], s[0, 1] + s[0, -1] + s[1, 0] + s[-1, 0])
>>> ast = ps.create_kernel(assignment, target='cpu', cpu_openmp=True)
>>> kernel = ast.compile()
>>> d_arr = np.zeros([5, 5])
>>> kernel(d=d_arr, s=np.ones([5, 5]))
>>> d_arr
array([[0., 0., 0., 0., 0.],
[0., 4., 4., 4., 0.],
[0., 4., 4., 4., 0.],
[0., 4., 4., 4., 0.],
[0., 0., 0., 0., 0.]])
"""
# save the original assignments
assign = assignments
# ---- Normalizing parameters
split_groups = ()
if isinstance(assignments, AssignmentCollection):
if 'split_groups' in assignments.simplification_hints:
split_groups = assignments.simplification_hints['split_groups']
assignments = assignments.all_assignments
if isinstance(assignments, Assignment):
assignments = [assignments]
# ---- Creating ast
if target == 'cpu':
from pystencils.cpu import create_kernel
from pystencils.cpu import add_openmp
ast = create_kernel(assignments, type_info=data_type, split_groups=split_groups,
iteration_slice=iteration_slice, ghost_layers=ghost_layers,
skip_independence_check=skip_independence_check)
for optimization in cpu_prepend_optimizations:
optimization(ast)
omp_collapse = None
if cpu_blocking:
omp_collapse = loop_blocking(ast, cpu_blocking)
if cpu_openmp:
add_openmp(ast, num_threads=cpu_openmp, collapse=omp_collapse)
if cpu_vectorize_info:
if cpu_vectorize_info is True:
vectorize(ast)
elif isinstance(cpu_vectorize_info, dict):
vectorize(ast, **cpu_vectorize_info)
else:
raise ValueError("Invalid value for cpu_vectorize_info")
elif target == 'llvm':
from pystencils.llvm import create_kernel
ast = create_kernel(assignments, type_info=data_type, split_groups=split_groups,
iteration_slice=iteration_slice, ghost_layers=ghost_layers)
elif target == 'gpu' or target == 'opencl':
from pystencils.gpucuda import create_cuda_kernel
ast = create_cuda_kernel(assignments, type_info=data_type,
indexing_creator=indexing_creator_from_params(gpu_indexing, gpu_indexing_params),
iteration_slice=iteration_slice, ghost_layers=ghost_layers,
skip_independence_check=skip_independence_check,
use_textures_for_interpolation=use_textures_for_interpolation)
if target == 'opencl':
from pystencils.opencl.opencljit import make_python_function
ast._backend = 'opencl'
ast.compile = functools.partial(make_python_function, ast, opencl_queue, opencl_ctx)
ast._target = 'opencl'
ast._backend = 'opencl'
else:
raise ValueError("Unknown target %s. Has to be one of 'cpu', 'gpu' or 'llvm' " % (target,))
if use_auto_for_assignments:
for a in ast.atoms(SympyAssignment):
a.use_auto = True
ast.assignments = assign
ast.indexFields = False
return ast
def create_indexed_kernel(assignments,
index_fields,
target='cpu',
data_type="double",
coordinate_names=('x', 'y', 'z'),
cpu_openmp=True,
gpu_indexing='block',
gpu_indexing_params=MappingProxyType({}),
use_textures_for_interpolation=True,
opencl_queue=None,
opencl_ctx=None):
"""
Similar to :func:`create_kernel`, but here not all cells of a field are updated but only cells with
coordinates which are stored in an index field. This traversal method can e.g. be used for boundary handling.
The coordinates are stored in a separated index_field, which is a one dimensional array with struct data type.
This struct has to contain fields named 'x', 'y' and for 3D fields ('z'). These names are configurable with the
'coordinate_names' parameter. The struct can have also other fields that can be read and written in the kernel, for
example boundary parameters.
index_fields: list of index fields, i.e. 1D fields with struct data type
coordinate_names: name of the coordinate fields in the struct data type
Example:
>>> import pystencils as ps
>>> import numpy as np
>>>
>>> # Index field stores the indices of the cell to visit together with optional values
>>> index_arr_dtype = np.dtype([('x', np.int32), ('y', np.int32), ('val', np.double)])
>>> index_arr = np.array([(1, 1, 0.1), (2, 2, 0.2), (3, 3, 0.3)], dtype=index_arr_dtype)
>>> idx_field = ps.fields(idx=index_arr)
>>>
>>> # Additional values stored in index field can be accessed in the kernel as well
>>> s, d = ps.fields('s, d: [2D]')
>>> assignment = ps.Assignment(d[0,0], 2 * s[0, 1] + 2 * s[1, 0] + idx_field('val'))
>>> ast = create_indexed_kernel(assignment, [idx_field], coordinate_names=('x', 'y'))
>>> kernel = ast.compile()
>>> d_arr = np.zeros([5, 5])
>>> kernel(s=np.ones([5, 5]), d=d_arr, idx=index_arr)
>>> d_arr
array([[0. , 0. , 0. , 0. , 0. ],
[0. , 4.1, 0. , 0. , 0. ],
[0. , 0. , 4.2, 0. , 0. ],
[0. , 0. , 0. , 4.3, 0. ],
[0. , 0. , 0. , 0. , 0. ]])
"""
assign = assignments
indF = index_fields
if isinstance(assignments, Assignment):
assignments = [assignments]
elif isinstance(assignments, AssignmentCollection):
assignments = assignments.all_assignments
if target == 'cpu':
from pystencils.cpu import create_indexed_kernel
from pystencils.cpu import add_openmp
ast = create_indexed_kernel(assignments, index_fields=index_fields, type_info=data_type,
coordinate_names=coordinate_names)
if cpu_openmp:
add_openmp(ast, num_threads=cpu_openmp)
elif target == 'llvm':
raise NotImplementedError("Indexed kernels are not yet supported in LLVM backend")
elif target == 'gpu' or target == 'opencl':
from pystencils.gpucuda import created_indexed_cuda_kernel
idx_creator = indexing_creator_from_params(gpu_indexing, gpu_indexing_params)
ast = created_indexed_cuda_kernel(assignments,
index_fields,
type_info=data_type,
coordinate_names=coordinate_names,
indexing_creator=idx_creator,
use_textures_for_interpolation=use_textures_for_interpolation)
if target == 'opencl':
from pystencils.opencl.opencljit import make_python_function
ast._backend = 'opencl'
ast.compile = functools.partial(make_python_function, ast, opencl_queue, opencl_ctx)
ast._target = 'opencl'
ast._backend = 'opencl'
else:
raise ValueError("Unknown target %s. Has to be either 'cpu' or 'gpu'" % (target,))
ast.assignments = assign
ast.indexFields = indF
return ast
def create_staggered_kernel(assignments, target='cpu', gpu_exclusive_conditions=False, **kwargs):
"""Kernel that updates a staggered field.
.. image:: /img/staggered_grid.svg
For a staggered field, the first index coordinate defines the location of the staggered value.
Further index coordinates can be used to store vectors/tensors at each point.
Args:
assignments: a sequence of assignments or an AssignmentCollection.
Assignments to staggered field are processed specially, while subexpressions and assignments to
regular fields are passed through to `create_kernel`. Multiple different staggered fields can be
used, but they all need to use the same stencil (i.e. the same number of staggered points) and
shape.
target: 'cpu', 'llvm' or 'gpu'
gpu_exclusive_conditions: disable the use of multiple conditionals inside the loop. The outer layers are then
handled in an else branch.
kwargs: passed directly to create_kernel, iteration_slice and ghost_layers parameters are not allowed
Returns:
AST, see `create_kernel`
"""
assert 'iteration_slice' not in kwargs and 'ghost_layers' not in kwargs
if isinstance(assignments, AssignmentCollection):
subexpressions = assignments.subexpressions + [a for a in assignments.main_assignments
if not hasattr(a, 'lhs')
or type(a.lhs) is not Field.Access
or not FieldType.is_staggered(a.lhs.field)]
assignments = [a for a in assignments.main_assignments if hasattr(a, 'lhs')
and type(a.lhs) is Field.Access
and FieldType.is_staggered(a.lhs.field)]
else:
subexpressions = [a for a in assignments if not hasattr(a, 'lhs')
or type(a.lhs) is not Field.Access
or not FieldType.is_staggered(a.lhs.field)]
assignments = [a for a in assignments if hasattr(a, 'lhs')
and type(a.lhs) is Field.Access
and FieldType.is_staggered(a.lhs.field)]
if len(set([tuple(a.lhs.field.staggered_stencil) for a in assignments])) != 1:
raise ValueError("All assignments need to be made to staggered fields with the same stencil")
if len(set([a.lhs.field.shape for a in assignments])) != 1:
raise ValueError("All assignments need to be made to staggered fields with the same shape")
staggered_field = assignments[0].lhs.field
stencil = staggered_field.staggered_stencil
dim = staggered_field.spatial_dimensions
shape = staggered_field.shape
counters = [LoopOverCoordinate.get_loop_counter_symbol(i) for i in range(dim)]
final_assignments = []
# find out whether any of the ghost layers is not needed
common_exclusions = set(["E", "W", "N", "S", "T", "B"][:2 * dim])
for direction in stencil:
exclusions = set(["E", "W", "N", "S", "T", "B"][:2 * dim])
for elementary_direction in direction:
exclusions.remove(inverse_direction_string(elementary_direction))
common_exclusions.intersection_update(exclusions)
ghost_layers = [[0, 0] for d in range(dim)]
for direction in common_exclusions:
direction = direction_string_to_offset(direction)
for d, s in enumerate(direction):
if s == 1:
ghost_layers[d][1] = 1
elif s == -1:
ghost_layers[d][0] = 1
def condition(direction):
"""exclude those staggered points that correspond to fluxes between ghost cells"""
exclusions = set(["E", "W", "N", "S", "T", "B"][:2 * dim])
for elementary_direction in direction:
exclusions.remove(inverse_direction_string(elementary_direction))
conditions = []
for e in exclusions:
if e in common_exclusions:
continue
offset = direction_string_to_offset(e)
for i, o in enumerate(offset):
if o == 1:
conditions.append(counters[i] < shape[i] - 1)
elif o == -1:
conditions.append(counters[i] > 0)
return sp.And(*conditions)
if gpu_exclusive_conditions:
outer_assignment = None
conditions = {direction: condition(direction) for direction in stencil}
for num_conditions in range(len(stencil)):
for combination in combinations(conditions.values(), num_conditions):
for assignment in assignments:
direction = stencil[assignment.lhs.index[0]]
if conditions[direction] in combination:
assignment = SympyAssignment(assignment.lhs, assignment.rhs)
outer_assignment = Conditional(sp.And(*combination), Block([assignment]), outer_assignment)
inner_assignment = []
for assignment in assignments:
direction = stencil[assignment.lhs.index[0]]
inner_assignment.append(SympyAssignment(assignment.lhs, assignment.rhs))
last_conditional = Conditional(sp.And(*[condition(d) for d in stencil]),
Block(inner_assignment), outer_assignment)
final_assignments = [s for s in subexpressions if not hasattr(s, 'lhs')] + \
[SympyAssignment(s.lhs, s.rhs) for s in subexpressions if hasattr(s, 'lhs')] + \
[last_conditional]
if target == 'cpu':
from pystencils.cpu import create_kernel as create_kernel_cpu
ast = create_kernel_cpu(final_assignments, ghost_layers=ghost_layers, **kwargs)
else:
ast = create_kernel(final_assignments, ghost_layers=ghost_layers, target=target, **kwargs)
return ast
for assignment in assignments:
direction = stencil[assignment.lhs.index[0]]
sp_assignments = [s for s in subexpressions if not hasattr(s, 'lhs')] + \
[SympyAssignment(s.lhs, s.rhs) for s in subexpressions if hasattr(s, 'lhs')] + \
[SympyAssignment(assignment.lhs, assignment.rhs)]
last_conditional = Conditional(condition(direction), Block(sp_assignments))
final_assignments.append(last_conditional)
remove_start_conditional = any([gl[0] == 0 for gl in ghost_layers])
prepend_optimizations = [lambda ast: remove_conditionals_in_staggered_kernel(ast, remove_start_conditional),
move_constants_before_loop]
ast = create_kernel(final_assignments, ghost_layers=ghost_layers, target=target,
cpu_prepend_optimizations=prepend_optimizations, **kwargs)
return ast