def get_function_definition(self, codegen_state, codegen_result,
schedule_index, function_decl, function_body):
from loopy.kernel.data import TemporaryVariable
args = (["_lpy_cl_kernels", "queue"] + [
idi.name for idi in codegen_state.implemented_data_info
if not issubclass(idi.arg_class, TemporaryVariable)
] + ["wait_for=None", "allocator=None"])
from genpy import (For, Function, Suite, Return, Line, Statement as S)
return Function(
codegen_result.current_program(codegen_state).name,
args,
Suite([
Line(),
] + [
Line(),
function_body,
Line(),
] + ([
For(
"_tv",
"_global_temporaries",
# free global temporaries
S("_tv.release()"))
] if self._get_global_temporaries(codegen_state) else []) + [
Line(),
Return("_lpy_evt"),
]))
def emit_sequential_loop(self, codegen_state, iname, iname_dtype, lbound,
ubound, inner):
ecm = codegen_state.expression_to_code_mapper
from pymbolic.mapper.stringifier import PREC_NONE
from genpy import For
return For((iname, ), "range(%s, %s + 1)" % (
ecm(lbound, PREC_NONE, "i"),
ecm(ubound, PREC_NONE, "i"),
), inner)
def get_function_definition(self, codegen_state, codegen_result,
schedule_index, function_decl, function_body):
from loopy.kernel.data import TemporaryVariable
args = (["_lpy_cl_kernels", "queue"] + [
idi.name for idi in codegen_state.implemented_data_info
if not issubclass(idi.arg_class, TemporaryVariable)
] + ["wait_for=None", "allocator=None"])
from genpy import (For, Function, Suite, Import, ImportAs, Return,
FromImport, If, Assign, Line, Statement as S)
return Function(
codegen_result.current_program(codegen_state).name,
args,
Suite([
FromImport("struct", ["pack as _lpy_pack"]),
ImportAs("pyopencl", "_lpy_cl"),
Import("pyopencl.tools"),
Line(),
If(
"allocator is None",
Assign("allocator",
"_lpy_cl_tools.DeferredAllocator(queue.context)")),
Line(),
] + [
Line(),
function_body,
Line(),
] + [
For(
"_tv",
"_global_temporaries",
# free global temporaries
S("_tv.release()"))
] + [
Line(),
Return("_lpy_evt"),
]))