Skip to content
Cannot retrieve contributors at this time
62 lines (45 sloc) 1.82 KB
from llvmlite import binding as ll
from llvmlite.llvmpy import core as lc
from numba.core.codegen import BaseCPUCodegen, CodeLibrary
from numba.core import utils
from .cudadrv import nvvm
CUDA_TRIPLE = {32: 'nvptx-nvidia-cuda',
64: 'nvptx64-nvidia-cuda'}
class CUDACodeLibrary(CodeLibrary):
# We don't optimize the IR at the function or module level because it is
# optimized by NVVM after we've passed it on.
def _optimize_functions(self, ll_module):
def _optimize_final_module(self):
def _finalize_specific(self):
# Fix global naming
for gv in self._final_module.global_variables:
if '.' in ='.', '_')
def get_asm_str(self):
# Return nothing: we can only dump assembly code when it is later
# generated (in numba.cuda.compiler).
return None
class JITCUDACodegen(BaseCPUCodegen):
This codegen implementation for CUDA actually only generates optimized LLVM
IR. Generation of PTX code is done separately (see numba.cuda.compiler).
_library_class = CUDACodeLibrary
def _init(self, llvm_module):
assert list(llvm_module.global_variables) == [], "Module isn't empty"
self._data_layout = nvvm.default_data_layout
self._target_data = ll.create_target_data(self._data_layout)
def _create_empty_module(self, name):
ir_module = lc.Module(name)
ir_module.triple = CUDA_TRIPLE[utils.MACHINE_BITS]
if self._data_layout:
ir_module.data_layout = self._data_layout
return ir_module
def _module_pass_manager(self):
raise NotImplementedError
def _function_pass_manager(self, llvm_module):
raise NotImplementedError
def _add_module(self, module):
You can’t perform that action at this time.