def preprocess_source(source, options, nvcc):
handle, source_path = mkstemp(suffix='.cu')
outf = open(source_path, 'w')
outf.write(source)
outf.close()
os.close(handle)
cmdline = [nvcc, '--preprocess'] + options + [source_path]
if 'win32' in sys.platform:
cmdline.extend(['--compiler-options', '-EP'])
else:
cmdline.extend(['--compiler-options', '-P'])
result, stdout, stderr = call_capture_output(cmdline,
error_on_nonzero=False)
if result != 0:
from pycuda.driver import CompileError
raise CompileError("nvcc preprocessing of %s failed" % source_path,
cmdline,
stderr=stderr)
# sanity check
if len(stdout) < 0.5 * len(source):
from pycuda.driver import CompileError
raise CompileError("nvcc preprocessing of %s failed with ridiculously "
"small code output - likely unsupported compiler." %
source_path,
cmdline,
stderr=stderr.decode("utf-8", "replace"))
unlink(source_path)
return stdout.decode("utf-8", "replace")
def preprocess_source(source, options, nvcc):
handle, source_path = mkstemp(suffix=".cu")
outf = open(source_path, "w")
outf.write(source)
outf.close()
os.close(handle)
cmdline = [nvcc, "--preprocess"] + options + [source_path]
if "win32" in sys.platform:
cmdline.extend(["--compiler-options", "-EP"])
else:
cmdline.extend(["--compiler-options", "-P"])
result, stdout, stderr = call_capture_output(cmdline,
error_on_nonzero=False)
if result != 0:
from pycuda.driver import CompileError
raise CompileError("nvcc preprocessing of %s failed" % source_path,
cmdline,
stderr=stderr)
# sanity check
if len(stdout) < 0.5 * len(source):
from pycuda.driver import CompileError
raise CompileError(
"nvcc preprocessing of %s failed with ridiculously "
"small code output - likely unsupported compiler." % source_path,
cmdline,
stderr=stderr.decode("utf-8", "replace"),
)
unlink(source_path)
preprocessed_str = stdout.decode("utf-8", "replace")
# remove the temporary filename from the preprocessed source code to get reproducible hashes
return preprocessed_str.replace(os.path.basename(source_path), "")
def compile_plain(source, options, keep, nvcc, cache_dir, target="cubin"):
from os.path import join
assert target in ["cubin", "ptx", "fatbin"]
if cache_dir:
checksum = _new_md5()
if '#include' in source:
checksum.update(
preprocess_source(source, options, nvcc).encode("utf-8"))
else:
checksum.update(source.encode("utf-8"))
for option in options:
checksum.update(option.encode("utf-8"))
checksum.update(get_nvcc_version(nvcc).encode("utf-8"))
from pycuda.characterize import platform_bits
checksum.update(str(platform_bits()).encode("utf-8"))
cache_file = checksum.hexdigest()
cache_path = join(cache_dir, cache_file + "." + target)
try:
cache_file = open(cache_path, "rb")
try:
return cache_file.read()
finally:
cache_file.close()
except:
pass
from tempfile import mkdtemp
file_dir = mkdtemp()
file_root = "kernel"
cu_file_name = file_root + ".cu"
cu_file_path = join(file_dir, cu_file_name)
outf = open(cu_file_path, "w")
outf.write(str(source))
outf.close()
if keep:
options = options[:]
options.append("--keep")
print("*** compiler output in %s" % file_dir)
cmdline = [nvcc, "--" + target] + options + [cu_file_name]
result, stdout, stderr = call_capture_output(cmdline,
cwd=file_dir,
error_on_nonzero=False)
try:
result_f = open(join(file_dir, file_root + "." + target), "rb")
except IOError:
no_output = True
else:
no_output = False
if result != 0 or (no_output and (stdout or stderr)):
if result == 0:
from warnings import warn
warn("PyCUDA: nvcc exited with status 0, but appears to have "
"encountered an error")
from pycuda.driver import CompileError
raise CompileError("nvcc compilation of %s failed" % cu_file_path,
cmdline,
stdout=stdout.decode("utf-8", "replace"),
stderr=stderr.decode("utf-8", "replace"))
if stdout or stderr:
lcase_err_text = (stdout + stderr).decode("utf-8", "replace").lower()
from warnings import warn
if "demoted" in lcase_err_text or "demoting" in lcase_err_text:
warn(
"nvcc said it demoted types in source code it "
"compiled--this is likely not what you want.",
stacklevel=4)
warn("The CUDA compiler succeeded, but said the following:\n" +
(stdout + stderr).decode("utf-8", "replace"),
stacklevel=4)
result_data = result_f.read()
result_f.close()
if cache_dir:
outf = open(cache_path, "wb")
outf.write(result_data)
outf.close()
if not keep:
from os import listdir, unlink, rmdir
for name in listdir(file_dir):
unlink(join(file_dir, name))
rmdir(file_dir)
return result_data
error_on_nonzero=False)
try:
cubin_f = open(join(file_dir, file_root + ".cubin"), "rb")
except IOError:
no_output = True
else:
no_output = False
if result != 0 or (no_output and (stdout or stderr)):
if result == 0:
from warnings import warn
warn("PyCUDA: nvcc exited with status 0, but appears to have "
"encountered an error")
from pycuda.driver import CompileError
raise CompileError("nvcc compilation of %s failed" % cu_file_path,
cmdline, stdout=stdout, stderr=stderr)
if stdout or stderr:
lcase_err_text = (stdout+stderr).lower()
if "demoted" in lcase_err_text or "demoting" in lcase_err_text:
from pycuda.driver import CompileError
raise CompileError("nvcc said it demoted types in source code it "
"compiled--this is likely not what you want.",
cmdline, stdout=stdout, stderr=stderr)
from warnings import warn
warn("The CUDA compiler suceeded, but said the following:\n"
+stdout+stderr)
cubin = cubin_f.read()
cubin_f.close()
try:
cubin_f = open(join(file_dir, file_root + ".cubin"), "rb")
except IOError:
no_output = True
else:
no_output = False
if result != 0 or (no_output and (stdout or stderr)):
if result == 0:
from warnings import warn
warn("PyCUDA: nvcc exited with status 0, but appears to have "
"encountered an error")
from pycuda.driver import CompileError
raise CompileError("nvcc compilation of %s failed" % cu_file_path,
cmdline,
stdout=stdout,
stderr=stderr)
if stdout or stderr:
from warnings import warn
warn("The CUDA compiler suceeded, but said the following:\n" + stdout +
stderr)
cubin = cubin_f.read()
cubin_f.close()
if cache_dir:
outf = open(cache_path, "wb")
outf.write(cubin)
outf.close()