from __future__ import absolute_import, division, print_function
__copyright__ = "Copyright (C) 2018 Xiaoyu Wei"
__license__ = """
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
"""
__doc__ = """
.. autoclass:: NearFieldEvalBase
:members:
.. autoclass:: NearFieldFromCSR
:members:
"""
import numpy as np
import loopy
import pyopencl as cl
from volumential.tools import KernelCacheWrapper
import logging
logger = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO)
# {{{ near field eval base class
[docs]class NearFieldEvalBase(KernelCacheWrapper):
"""Base class of near-field evalulator.
"""
default_name = "near_field_eval_base"
def __init__(
self,
integral_kernel,
table_data_shapes,
potential_kind=1,
options=[],
name=None,
device=None,
**kwargs
):
"""potential_kind:
1 - The (weakly singular) volume potentials.
2 - The (hypersingular) inverse potentials, like the fractional Laplacian.
Here, the fractional Laplacian is the inverse of the (weakly singular)
Riesz potential operator.
The two kinds share the same far-field code, but the second kind requires
exterior_mode_nmlz when computing the list1 interactions.
"""
self.integral_kernel = integral_kernel
self.n_tables = table_data_shapes["n_tables"]
self.n_q_points = table_data_shapes["n_q_points"]
self.n_table_entries = table_data_shapes["n_table_entries"]
self.potential_kind = potential_kind
assert np.isreal(self.n_tables)
assert np.isreal(self.n_q_points)
assert np.isreal(self.n_table_entries)
self.options = options
self.name = name or self.default_name
self.divice = device
self.extra_kwargs = kwargs
# Allow user to pass more tables to force using multiple tables
# instead of performing kernel scaling
if "infer_kernel_scaling" not in self.extra_kwargs:
self.extra_kwargs["infer_kernel_scaling"] = self.n_tables == 1
# Do not infer scaling rules when user defined rules are present
if ("kernel_scaling_code" in self.extra_kwargs) or (
"kernel_displacement_code" in self.extra_kwargs):
self.extra_kwargs["infer_kernel_scaling"] = False
# the two codes must be simultaneously given
assert ("kernel_scaling_code" in self.extra_kwargs) and (
"kernel_displacement_code" in self.extra_kwargs)
self.kname = self.integral_kernel.__repr__()
self.dim = self.integral_kernel.dim
def get_cache_key(self):
return (
type(self).__name__,
self.name,
self.kname,
"infer_scaling=" + str(self.extra_kwargs["infer_kernel_scaling"]),
)
# }}} End near field eval base class
# {{{ eval from CSR data
[docs]class NearFieldFromCSR(NearFieldEvalBase):
"""Evaluate the near-field potentials from CSR representation of the tree.
The class supports auto-scaling of simple kernels.
"""
default_name = "near_field_from_csr"
def codegen_vec_component(self, d=None):
if d is None:
dimension = self.dim - 1
else:
dimension = d
return (
"("
+ "(box_centers["
+ str(dimension)
+ ", target_box_id]"
+ "- box_centers["
+ str(dimension)
+ ", source_box_id]"
+ ") / sbox_extent * 4.0 + encoding_shift"
+ ")"
)
def codegen_vec_id(self):
dim = self.dim
code = "0.0"
for d in range(dim):
code = "(" + code + ") * encoding_base"
code = code + "+" + self.codegen_vec_component(d)
return code
[docs] def codegen_compute_scaling(self, box_name='sbox'):
"""box_name: the name of the box whose extent is used.
"""
if ("infer_kernel_scaling" in self.extra_kwargs) and (
self.extra_kwargs["infer_kernel_scaling"]
):
# Laplace 2D
if self.kname == "LapKnl2D":
logger.info("scaling for LapKnl2D")
code = "BOX_extent * BOX_extent / \
(table_root_extent * table_root_extent)"
elif self.kname in (
"AxisTargetDerivative(0, LapKnl2D)",
"AxisTargetDerivative(1, LapKnl2D)"):
logger.info("scaling for Grad(LapKnl2D)")
code = "BOX_extent / table_root_extent"
# Constant 2D
elif self.kname == "CstKnl2D":
logger.info("scaling for CstKnl2D")
code = "BOX_extent * BOX_extent / \
(table_root_extent * table_root_extent)"
# Laplace 3D
elif self.kname == "LapKnl3D":
logger.info("scaling for Lapknl3D")
code = "BOX_extent * BOX_extent / \
(table_root_extent * table_root_extent)"
elif self.kname in (
"AxisTargetDerivative(0, LapKnl3D)",
"AxisTargetDerivative(1, LapKnl3D)",
"AxisTargetDerivative(2, LapKnl3D)"):
logger.info("scaling for Grad(LapKnl3D)")
code = "BOX_extent / table_root_extent"
# Constant 3D
elif self.kname == "CstKnl3D":
logger.info("scaling for CstKnl3D")
code = "BOX_extent * BOX_extent * BOX_extent / \
(table_root_extent * table_root_extent * table_root_extent)"
else:
logger.warn(
"Kernel not scalable and not using multiple tables, "
"to get correct results, please make sure that your "
"tree is uniform and only needs one table."
)
code = "1.0"
return code.replace("BOX", box_name)
elif "kernel_scaling_code" in self.extra_kwargs:
# user-defined scaling rule
assert isinstance(self.extra_kwargs['kernel_scaling_code'], str)
logger.info("Using scaling rule %s for %s.",
self.extra_kwargs['kernel_scaling_code'], self.kname
)
return self.extra_kwargs['kernel_scaling_code']
else:
logger.info("not scaling for " + self.kname)
logger.info("(using multiple tables)")
return "1.0"
def codegen_compute_displacement(self, box_name='sbox'):
if ("infer_kernel_scaling" in self.extra_kwargs) and (
self.extra_kwargs["infer_kernel_scaling"]
):
# Laplace 2D
if self.kname == "LapKnl2D":
logger.info("displacement for laplace 2D")
s = "-0.5 / PI * scaling * \
log(BOX_extent / table_root_extent) * \
mode_nmlz[table_lev, sid]"
import math
code = s.replace("PI", str(math.pi))
# Constant 2D
elif self.kname == "CstKnl2D":
logger.info("no displacement for CstKnl2D")
code = "0.0"
# Laplace 3D
elif self.kname == "LapKnl3D":
logger.info("no displacement for LapKnl3D")
code = "0.0"
# Constant 3D
elif self.kname == "CstKnl3D":
logger.info("no displacement for CstKnl3D")
code = "0.0"
else:
logger.warn(
"Kernel not scalable and not using multiple tables, "
"to get correct results, please make sure that either "
"no displacement is needed, or the box "
"tree is uniform and only needs one table."
)
code = "0.0"
elif "kernel_displacement_code" in self.extra_kwargs:
# user-defined displacement rule
assert isinstance(
self.extra_kwargs['kernel_displacement_code'], str)
logger.info("Using displacement %s for %s.",
self.extra_kwargs['kernel_displacement_code'],
self.kname
)
code = self.extra_kwargs['kernel_displacement_code']
else:
logger.info("no displacement for " + self.kname)
logger.info("(using multiple tables)")
code = "0.0"
return code.replace("BOX", box_name)
def codegen_get_table_level(self, box_name='sbox'):
if ("infer_kernel_scaling" in self.extra_kwargs) and (
self.extra_kwargs["infer_kernel_scaling"]
):
if (
self.kname == "LapKnl2D"
or self.kname == "LapKnl3D"
or self.kname == "CstKnl2D"
or self.kname == "CstKnl3D"
):
logger.info("scaling from table[0] for " + self.kname)
code = "0.0"
else:
logger.warn(
"Kernel not scalable and not using multiple tables, "
"to get correct results, please make sure that your "
"tree is uniform and only needs one table."
)
code = "0.0"
elif "kernel_scaling_code" in self.extra_kwargs:
# Using custom scaling
code = "0.0"
else:
logger.info("computing table level from box size")
logger.info("(using multiple tables)")
code = "log(table_root_extent / BOX_extent) / log(2.0)"
return code.replace("BOX", box_name)
[docs] def codegen_exterior_part(self):
"""Computes the exterior contribution. This is nonzero for
inverse-type potentials like the fractional Laplacian.
"""
if self.potential_kind == 1:
return "0.0"
elif self.potential_kind == 2:
return "source_coefs[target_id] * ext_nmlz"
else:
raise ValueError("Unsupported potential kind %d" % self.potential_kind)
def get_kernel(self):
if self.integral_kernel.is_complex_valued:
potential_dtype = np.complex128
else:
potential_dtype = np.float64
lpknl = loopy.make_kernel( # NOQA
[
"{ [ tbox ] : 0 <= tbox < n_tgt_boxes }",
"{ [ tid, sbox ] : 0 <= tid < n_box_targets and \
sbox_begin <= sbox < sbox_end }",
"{ [ sid ] : 0 <= sid < n_box_sources }",
],
"""
for tbox
<> target_box_id = target_boxes[tbox]
<> box_target_beg = box_target_starts[target_box_id]
<> n_box_targets = box_target_counts_cumul[target_box_id]
<> sbox_begin = neighbor_source_boxes_starts[tbox]
<> sbox_end = neighbor_source_boxes_starts[tbox+1]
<> tbox_level = box_levels[target_box_id]
<> tbox_extent = root_extent * (1.0 / (2**tbox_level))
for tid
<> target_id = box_target_beg + tid
end
for tid, sbox
<> source_box_id = source_boxes[sbox]
<> n_box_sources = box_source_counts_cuml[source_box_id]
<> box_source_beg = box_source_starts[source_box_id]
<> sbox_level = box_levels[source_box_id]
<> sbox_extent = root_extent * (1.0 / (2**sbox_level))
table_lev_tmp = GET_TABLE_LEVEL {id=tab_lev_tmp}
table_lev = round(table_lev_tmp) {id=tab_lev,dep=tab_lev_tmp}
vec_id_tmp = COMPUTE_VEC_ID {id=vec_id_tmp}
vec_id = round(vec_id_tmp) {id=vec_id,dep=vec_id_tmp}
<> case_id = case_indices[vec_id] {dep=vec_id}
<> scaling = COMPUTE_SCALING
for sid
<> tgt_scaling = COMPUTE_TGT_SCALING
<> tgt_displacement = COMPUTE_TGT_DISPLACEMENT
tgt_table_lev_tmp = GET_TGT_TABLE_LEVEL {id=tgttab_lev_tmp}
tgt_table_lev = round(tgt_table_lev_tmp) \
{id=tgttab_lev,dep=tgttab_lev_tmp}
<> ext_nmlz = exterior_mode_nmlz[tgt_table_lev, tid] \
* tgt_scaling + tgt_displacement \
{id=extnmlz,dep=tgttab_lev}
<> source_id = box_source_beg + sid
<> pair_id = sid * n_box_targets + tid
<> entry_id = case_id * \
(n_box_targets * n_box_sources) \
+ pair_id
<> displacement = COMPUTE_DISPLACEMENT
<> integ = table_data[table_lev, entry_id] * scaling \
+ displacement {id=integ,dep=tab_lev}
# <> source_id_tree = user_source_ids[source_id]
<> coef = source_coefs[source_id] {id=coef}
# <> target_id_user = sorted_target_ids[target_id]
#db_table_lev[target_id] = table_lev_tmp {dep=tab_lev}
#db_case_id[target_id] = case_id
#db_vec_id[target_id] = vec_id
#db_n_box_targets[target_id] = n_box_targets
#db_n_box_sources[target_id] = n_box_sources
#db_entry_id[target_id] = entry_id
end
end
for tid
result[target_id] = sum((sbox, sid),
coef * integ) + EXTERIOR_PART {dep=integ:coef:extnmlz}
# Try inspecting case_id if something goes wrong
# (like segmentation fault) and look for -1's
# result[target_id] = min((sbox, sid), case_id)
# result[target_id] = vec_id_tmp
end
end
"""
.replace("COMPUTE_VEC_ID", self.codegen_vec_id())
.replace("COMPUTE_SCALING", self.codegen_compute_scaling())
.replace("COMPUTE_DISPLACEMENT", self.codegen_compute_displacement())
.replace("COMPUTE_TGT_SCALING", self.codegen_compute_scaling('tbox'))
.replace("COMPUTE_TGT_DISPLACEMENT",
self.codegen_compute_displacement('tbox'))
.replace("GET_TABLE_LEVEL", self.codegen_get_table_level())
.replace("GET_TGT_TABLE_LEVEL", self.codegen_get_table_level('tbox'))
.replace("EXTERIOR_PART", self.codegen_exterior_part()),
[
loopy.TemporaryVariable("vec_id", np.int32),
loopy.TemporaryVariable("vec_id_tmp", np.float64),
loopy.TemporaryVariable("table_lev", np.int32),
loopy.TemporaryVariable("table_lev_tmp", np.float64),
loopy.TemporaryVariable("tgt_table_lev", np.int32),
loopy.TemporaryVariable("tgt_table_lev_tmp", np.float64),
loopy.ValueArg("encoding_base", np.int32),
loopy.GlobalArg("mode_nmlz", potential_dtype,
"n_tables, n_q_points"),
loopy.GlobalArg("exterior_mode_nmlz", potential_dtype,
"n_tables, n_q_points"),
loopy.GlobalArg("table_data", potential_dtype,
"n_tables, n_table_entries"),
loopy.GlobalArg("source_boxes", np.int32, "n_source_boxes"),
loopy.GlobalArg("box_centers", None, "dim, aligned_nboxes"),
loopy.ValueArg("aligned_nboxes", np.int32),
loopy.ValueArg("table_root_extent", np.float64),
loopy.ValueArg(
"dim, n_source_boxes, n_tables, " "n_q_points, n_table_entries",
np.int32,
),
"...",
],
name="near_field",
lang_version=(2018, 2)
)
# lpknl = loopy.set_options(lpknl, write_code=True)
lpknl = loopy.set_options(lpknl, return_dict=True)
return lpknl
def get_cache_key(self):
return (
type(self).__name__,
self.name,
self.kname,
"complex_kernel=" + str(self.integral_kernel.is_complex_valued),
"potential_kind=%d" % self.potential_kind,
"infer_scaling=" + str(self.extra_kwargs["infer_kernel_scaling"]),
"scaling_policy=" + self.codegen_compute_scaling(),
"displacement_policy=" + self.codegen_compute_displacement(),
)
def get_optimized_kernel(self, ncpus=None):
if ncpus is None:
import multiprocessing
# NOTE: this detects the number of logical cores, disable hyperthreading
# for the optimal performance.
ncpus = multiprocessing.cpu_count()
knl = self.get_kernel()
knl = loopy.split_iname(knl, "tbox", ncpus, inner_tag="g.0")
return knl
def __call__(self, queue, **kwargs):
knl = self.get_cached_optimized_kernel()
result = kwargs.pop("result")
box_centers = kwargs.pop("box_centers")
box_levels = kwargs.pop("box_levels")
box_source_counts_cumul = kwargs.pop("box_source_counts_cumul")
box_source_starts = kwargs.pop("box_source_starts")
box_target_counts_cumul = kwargs.pop("box_target_counts_cumul")
box_target_starts = kwargs.pop("box_target_starts")
case_indices = kwargs.pop("case_indices")
encoding_base = kwargs.pop("encoding_base")
encoding_shift = kwargs.pop("encoding_shift")
mode_nmlz_combined = kwargs.pop("mode_nmlz_combined")
exterior_mode_nmlz_combined = kwargs.pop("exterior_mode_nmlz_combined")
root_extent = kwargs.pop("root_extent")
table_root_extent = kwargs.pop("table_root_extent")
neighbor_source_boxes_starts = kwargs.pop("neighbor_source_boxes_starts")
neighbor_source_boxes_lists = kwargs.pop("neighbor_source_boxes_lists")
mode_coefs = kwargs.pop("mode_coefs")
table_data_combined = kwargs.pop("table_data_combined")
target_boxes = kwargs.pop("target_boxes")
integral_kernel_init_kargs = {
name: val
for name, val in zip(
self.integral_kernel.init_arg_names,
self.integral_kernel.__getinitargs__())
}
# help loopy's type inference
for key, val in integral_kernel_init_kargs.items():
if isinstance(val, int):
integral_kernel_init_kargs[key] = np.int32(val)
if isinstance(val, float):
integral_kernel_init_kargs[key] = np.float64(val)
extra_knl_args_from_init = {}
for key, val in integral_kernel_init_kargs.items():
if key in knl.arg_dict:
extra_knl_args_from_init[key] = val
evt, res = knl(
queue,
result=result,
# db_table_lev=np.zeros(out_pot.shape),
# db_case_id=np.zeros(out_pot.shape),
# db_vec_id=np.zeros(out_pot.shape),
# db_n_box_sources=np.zeros(out_pot.shape),
# db_n_box_targets=np.zeros(out_pot.shape),
# db_entry_id=np.zeros(out_pot.shape),
box_centers=box_centers,
box_levels=box_levels,
box_source_counts_cuml=box_source_counts_cumul,
box_source_starts=box_source_starts,
box_target_counts_cumul=box_target_counts_cumul,
box_target_starts=box_target_starts,
case_indices=case_indices,
n_tables=self.n_tables,
n_table_entries=self.n_table_entries,
n_q_points=self.n_q_points,
encoding_base=encoding_base,
encoding_shift=encoding_shift,
mode_nmlz=mode_nmlz_combined,
exterior_mode_nmlz=exterior_mode_nmlz_combined,
n_tgt_boxes=len(target_boxes),
neighbor_source_boxes_starts=neighbor_source_boxes_starts,
root_extent=root_extent,
source_boxes=neighbor_source_boxes_lists,
n_source_boxes=len(neighbor_source_boxes_lists),
source_coefs=mode_coefs,
table_data=table_data_combined,
target_boxes=target_boxes,
table_root_extent=table_root_extent,
**extra_knl_args_from_init
)
res['result'].add_event(evt)
if isinstance(result, cl.array.Array):
assert result is res["result"]
else:
assert isinstance(result, np.ndarray)
result = res['result'].get(queue)
queue.finish()
logger.info("list1 evaluation finished")
# check for data integrity
# if not (np.max(np.abs(out_pot.get()))) < 100:
# import pudb; pu.db
return result, evt
# }}} End eval from CSR data