kaushikcfd

---------------------------------------------------------------------------
KERNEL: resample_by_mat
---------------------------------------------------------------------------
ARGUMENTS:
ary: type: np:dtype('float64'), shape: (nelements_vec, n_from_nodes), dim_tags: (N1:stride:n_from_nodes, N0:stride:1), offset: <class 'loopy.kernel.data.auto'> aspace: global
from_element_indices: type: np:dtype('int64'), shape: (nelements), dim_tags: (N0:stride:1), offset: <class 'loopy.kernel.data.auto'> aspace: global
n_from_nodes: ValueArg, type: np:dtype('int64')
n_to_nodes: ValueArg, type: np:dtype('int64')
nelements: ValueArg, type: np:dtype('int64')
nelements_vec: ValueArg, type: np:dtype('int32')

kaushikcfd

from kanren import var, eq, run, facts, Relation
from pytools.tag import Tag, tag_dataclass, UniqueTag
from typing import FrozenSet
from dataclasses import dataclass
from unification import unifiable


@unifiable
@tag_dataclass
class AxisPropagatedFromInput(UniqueTag):

kaushikcfd

import loopy as lp
import numpy as np


_pt_kernel_knl = lp.make_kernel([
    "{ [_pt_sum_r0_1, _pt_sum_r0_17] : 0<=_pt_sum_r0_1<=2 and 0<=_pt_sum_r0_17<=2 }",
    "{ [_pt_sum_r0_0] : 0 <= _pt_sum_r0_0 <= 34 }",
    "{ [iel_cse_1, idof_cse_1] : 0<=iel_cse_1<=187199 and 0<=idof_cse_1<=14 }",
    "{ [iel_cse_0, idof_cse_0] : 0<=iel_cse_0<=187199 and 0<=idof_cse_0<=14 }",
    "{ [iel_cse_2, idof_cse_2] : 0<=iel_cse_2<=4799 and 0<=idof_cse_2<=14 }",

kaushikcfd

for idof_face_mass, iel_face_mass
    acc_face_f   = 0.0  {id=face_insn_face_f_init}
    acc_face_f_0 = 0.0  {id=face_insn_0_face_f_0_init}
    acc_face_f_1 = 0.0  {id=face_insn_1_face_f_1_init}
    acc_face_f_2 = 0.0  {id=face_insn_2_face_f_2_init}
    for face_f
      acc_face_jdof   = 0.0  {id=face_insn_face_f_update_face_jdof_init}
      acc_face_jdof_0 = 0.0  {id=face_insn_0_face_f_0_update_face_jdof_0_init}
      acc_face_jdof_1 = 0.0  {id=face_insn_1_face_f_1_update_face_jdof_1_init}
      acc_face_jdof_2 = 0.0  {id=face_insn_2_face_f_2_update_face_jdof_2_init}

kaushikcfd

#define lid(N) ((int) get_local_id(N))
#define gid(N) ((int) get_group_id(N))
#if __OPENCL_C_VERSION__ < 120
#pragma OPENCL EXTENSION cl_khr_fp64: enable
#endif
#define LOOPY_CALL_WITH_INTEGER_TYPES(MACRO_NAME) \
    MACRO_NAME(int8, char) \
    MACRO_NAME(int16, short) \
    MACRO_NAME(int32, int) \
    MACRO_NAME(int64, long)

kaushikcfd

---------------------------------------------------------------------------
KERNEL: _pt_kernel
---------------------------------------------------------------------------
ARGUMENTS:
_actx_in_0: type: np:dtype('float64'), shape: () aspace: global
_actx_in_1_0_0: type: np:dtype('float64'), shape: (48000, 35), dim_tags: (N1:stride:35, N0:stride:1) aspace: global
_actx_in_1_1_0: type: np:dtype('float64'), shape: (48000, 35), dim_tags: (N1:stride:35, N0:stride:1) aspace: global
_actx_in_1_2_0: type: np:dtype('float64'), shape: (48000, 35), dim_tags: (N1:stride:35, N0:stride:1) aspace: global
_actx_in_1_3_0: type: np:dtype('float64'), shape: (48000, 35), dim_tags: (N1:stride:35, N0:stride:1) aspace: global
_pt_in: type: np:dtype('float64'), shape: (48000, 35), dim_tags: (N1:stride:35, N0:stride:1) aspace: global

kaushikcfd

---------------------------------------------------------------------------
KERNEL: face_mass
---------------------------------------------------------------------------
ARGUMENTS:
jac_surf: type: np:dtype('float64'), shape: (nfaces, nelements, nface_nodes), dim_tags: (N2:stride:nface_nodes*nelements, N1:stride:nface_nodes, N0:stride:1), offset: <class 'loopy.kernel.data.auto'> aspace: global
mat: type: np:dtype('float64'), shape: (nvol_nodes, nfaces, nface_nodes), dim_tags: (N2:stride:nface_nodes*nfaces, N1:stride:nface_nodes, N0:stride:1), offset: <class 'loopy.kernel.data.auto'> aspace: global
nelements: ValueArg, type: np:dtype('int64')
nface_nodes: ValueArg, type: np:dtype('int64')
nfaces: ValueArg, type: np:dtype('int64')
nvol_nodes: ValueArg, type: np:dtype('int64')

kaushikcfd

import numpy as np
import pyopencl as cl
from grudge.discretization import DiscretizationCollection
import grudge.op as op
from meshmode.mesh.generation import generate_box_mesh
from meshmode.array_context import PyOpenCLArrayContext
from arraycontext import freeze, thaw, to_numpy
from grudge.dof_desc import DTAG_BOUNDARY
from meshmode.discretization.connection import FACE_RESTR_INTERIOR

kaushikcfd

diff --git a/pytato/analysis.py b/pytato/analysis.py
index 3bd2839..93147b1 100644
--- a/pytato/analysis.py
+++ b/pytato/analysis.py
@@ -24,13 +24,19 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 THE SOFTWARE.
 """
 
-from typing import Mapping, Dict, Union
+from typing import Mapping, Dict, Union, FrozenSet

kaushikcfd

#define lid(N) ((int) get_local_id(N))
#define gid(N) ((int) get_group_id(N))
#if __OPENCL_C_VERSION__ < 120
#pragma OPENCL EXTENSION cl_khr_fp64: enable
#endif

static void face_mass(__global double const *__restrict__ jac_surf, __global double const *__restrict__ mat, int const nelements, int const nface_nodes, int const nfaces, int const nvol_nodes, __global double *__restrict__ result, __global double const *__restrict__ vec);
static void resample_by_picking(__global double const *__restrict__ ary, int const nelements_vec, int const n_from_nodes, __global long const *__restrict__ from_element_indices, int const n_to_nodes, int const nelements, __global int const *__restrict__ pick_list, __global double *__restrict__ result);
static void resample_by_picking_0(__global double const *__restrict__ ary, int const nelements_vec, int const n_from_nodes, __global int const *__restrict__ from_element_indices, int const n_to_nodes, int const nelements, __global int const *__restrict__ pick_list, __global double *__