mirror-github-bspeice/cuburn
1
0
Fork 0
mirror of https://github.com/stevenrobertson/cuburn.git synced 2025-07-06 08:15:09 -04:00
Code Issues Actions Packages Projects Releases Wiki Activity

Populate arch by default; add --keep

Browse Source
This commit is contained in:
Steven Robertson
2017-04-24 16:39:15 -07:00
parent 582221dd0f
commit 9892acbc7f
3 changed files with 13 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
cuburn/code/util.py
View File

@ -96,7 +96,7 @@ def assemble_code(*libs):
DEFAULT_CMP_OPTIONS = ('-use_fast_math', '-lineinfo')
DEFAULT_SAVE_KERNEL = True
def compile(name, src, opts=DEFAULT_CMP_OPTIONS, save=DEFAULT_SAVE_KERNEL,
arch=None):
arch=None, keep=False):
"""
Compile a module. Returns a copy of the source (for inspection or
display) and the compiled cubin.
@ -105,7 +105,8 @@ def compile(name, src, opts=DEFAULT_CMP_OPTIONS, save=DEFAULT_SAVE_KERNEL,
if save:
with open(os.path.join(dir, name + '_kern.cu'), 'w') as fp:
fp.write(src)
cubin = pycuda.compiler.compile(src, options=list(opts), arch=arch)
cubin = pycuda.compiler.compile(src, options=list(opts), arch=arch,
keep=keep)
if save:
with open(os.path.join(dir, name + '_kern.cubin'), 'w') as fp:
fp.write(cubin)

8
cuburn/render.py
View File

@ -218,9 +218,9 @@ class Renderer(object):
_modrefs = {}
@classmethod
def compile(cls, gnm, arch=None):
def compile(cls, gnm, arch=None, keep=False):
packer, lib = iter.mkiterlib(gnm)
cubin = util.compile('iter', assemble_code(lib), arch=arch)
cubin = util.compile('iter', assemble_code(lib), arch=arch, keep=keep)
return packer, lib, cubin
def load(self, cubin):
@ -232,8 +232,8 @@ class Renderer(object):
self._modrefs[cubin] = mod
return mod
def __init__(self, gnm, gprof):
self.packer, self.lib, self.cubin = self.compile(gnm)
def __init__(self, gnm, gprof, keep=False, arch=None):
self.packer, self.lib, self.cubin = self.compile(gnm, keep=keep, arch=arch)
self.mod = self.load(self.cubin)
self.filts = filters.create(gprof)
self.out = output.get_output_for_profile(gprof)