André R. Brodtkorb daa175ef32 Checkpoint
2018-08-24 11:48:30 +02:00

531 lines
19 KiB
Python

# -*- coding: utf-8 -*-
"""
This python module implements the different helper functions and
classes
Copyright (C) 2018 SINTEF ICT
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
"""
import os
import numpy as np
import time
import re
import io
import hashlib
import logging
import gc
import pycuda.compiler as cuda_compiler
import pycuda.gpuarray
import pycuda.driver as cuda
from GPUSimulators import Autotuner
"""
Class which keeps track of time spent for a section of code
"""
class Timer(object):
def __init__(self, tag, log_level=logging.DEBUG):
self.tag = tag
self.log_level = log_level
self.logger = logging.getLogger(__name__)
def __enter__(self):
self.start = time.time()
return self
def __exit__(self, *args):
self.end = time.time()
self.secs = self.end - self.start
self.msecs = self.secs * 1000 # millisecs
self.logger.log(self.log_level, "%s: %f ms", self.tag, self.msecs)
"""
Class which keeps track of the CUDA context and some helper functions
"""
class CudaContext(object):
def __init__(self, blocking=False, use_cache=True, autotuning=True):
self.blocking = blocking
self.use_cache = use_cache
self.logger = logging.getLogger(__name__)
self.kernels = {}
self.module_path = os.path.dirname(os.path.realpath(__file__))
#Initialize cuda (must be first call to PyCUDA)
cuda.init(flags=0)
self.logger.info("PyCUDA version %s", str(pycuda.VERSION_TEXT))
#Print some info about CUDA
self.logger.info("CUDA version %s", str(cuda.get_version()))
self.logger.info("Driver version %s", str(cuda.get_driver_version()))
self.cuda_device = cuda.Device(0)
self.logger.info("Using '%s' GPU", self.cuda_device.name())
self.logger.debug(" => compute capability: %s", str(self.cuda_device.compute_capability()))
# Create the CUDA context
if (self.blocking):
self.cuda_context = self.cuda_device.make_context(flags=cuda.ctx_flags.SCHED_BLOCKING_SYNC)
self.logger.warning("Using blocking context")
else:
self.cuda_context = self.cuda_device.make_context(flags=cuda.ctx_flags.SCHED_AUTO)
free, total = cuda.mem_get_info()
self.logger.debug(" => memory: %d / %d MB available", int(free/(1024*1024)), int(total/(1024*1024)))
self.logger.info("Created context handle <%s>", str(self.cuda_context.handle))
#Create cache dir for cubin files
if (self.use_cache):
self.cache_path = os.path.join(self.module_path, "cuda_cache")
if not os.path.isdir(self.cache_path):
os.mkdir(self.cache_path)
self.logger.info("Using CUDA cache dir %s", self.cache_path)
self.autotuner = None
if (autotuning):
self.logger.info("Autotuning enabled. It may take several minutes to run the code the first time: have patience")
self.autotuner = Autotuner.Autotuner()
def __del__(self, *args):
self.logger.info("Cleaning up CUDA context handle <%s>", str(self.cuda_context.handle))
# Loop over all contexts in stack, and remove "this"
other_contexts = []
while (cuda.Context.get_current() != None):
context = cuda.Context.get_current()
if (context.handle != self.cuda_context.handle):
self.logger.debug("<%s> Popping <%s> (*not* ours)", str(self.cuda_context.handle), str(context.handle))
other_contexts = [context] + other_contexts
cuda.Context.pop()
else:
self.logger.debug("<%s> Popping <%s> (ours)", str(self.cuda_context.handle), str(context.handle))
cuda.Context.pop()
# Add all the contexts we popped that were not our own
for context in other_contexts:
self.logger.debug("<%s> Pushing <%s>", str(self.cuda_context.handle), str(context.handle))
cuda.Context.push(context)
self.logger.debug("<%s> Detaching", str(self.cuda_context.handle))
self.cuda_context.detach()
def __str__(self):
return "CudaContext id " + str(self.cuda_context.handle)
def hash_kernel(kernel_filename, include_dirs):
# Generate a kernel ID for our caches
num_includes = 0
max_includes = 100
kernel_hasher = hashlib.md5()
logger = logging.getLogger(__name__)
# Loop over file and includes, and check if something has changed
files = [kernel_filename]
while len(files):
if (num_includes > max_includes):
raise("Maximum number of includes reached - circular include in {:}?".format(kernel_filename))
filename = files.pop()
#logger.debug("Hashing %s", filename)
modified = os.path.getmtime(filename)
# Open the file
with io.open(filename, "r") as file:
# Search for #inclue <something> and also hash the file
file_str = file.read()
kernel_hasher.update(file_str.encode('utf-8'))
kernel_hasher.update(str(modified).encode('utf-8'))
#Find all includes
includes = re.findall('^\W*#include\W+(.+?)\W*$', file_str, re.M)
# Loop over everything that looks like an include
for include_file in includes:
#Search through include directories for the file
file_path = os.path.dirname(filename)
for include_path in [file_path] + include_dirs:
# If we find it, add it to list of files to check
temp_path = os.path.join(include_path, include_file)
if (os.path.isfile(temp_path)):
files = files + [temp_path]
num_includes = num_includes + 1 #For circular includes...
break
return kernel_hasher.hexdigest()
"""
Reads a text file and creates an OpenCL kernel from that
"""
def get_prepared_kernel(self, kernel_filename, kernel_function_name, \
prepared_call_args, \
include_dirs=[], no_extern_c=True,
**kwargs):
"""
Helper function to print compilation output
"""
def cuda_compile_message_handler(compile_success_bool, info_str, error_str):
self.logger.debug("Compilation returned %s", str(compile_success_bool))
if info_str:
self.logger.debug("Info: %s", info_str)
if error_str:
self.logger.debug("Error: %s", error_str)
#self.logger.debug("Getting %s", kernel_filename)
# Create a hash of the kernel (and its includes)
kwargs_hasher = hashlib.md5()
kwargs_hasher.update(str(kwargs).encode('utf-8'));
kwargs_hash = kwargs_hasher.hexdigest()
kwargs_hasher = None
root, ext = os.path.splitext(kernel_filename)
kernel_hash = root \
+ "_" + CudaContext.hash_kernel( \
os.path.join(self.module_path, kernel_filename), \
include_dirs=[self.module_path] + include_dirs) \
+ "_" + kwargs_hash \
+ ext
cached_kernel_filename = os.path.join(self.cache_path, kernel_hash)
# If we have the kernel in our hashmap, return it
if (kernel_hash in self.kernels.keys()):
self.logger.debug("Found kernel %s cached in hashmap (%s)", kernel_filename, kernel_hash)
return self.kernels[kernel_hash]
# If we have it on disk, return it
elif (self.use_cache and os.path.isfile(cached_kernel_filename)):
self.logger.debug("Found kernel %s cached on disk (%s)", kernel_filename, kernel_hash)
with io.open(cached_kernel_filename, "rb") as file:
file_str = file.read()
module = cuda.module_from_buffer(file_str, message_handler=cuda_compile_message_handler)
kernel = module.get_function(kernel_function_name)
kernel.prepare(prepared_call_args)
self.kernels[kernel_hash] = kernel
return kernel
# Otherwise, compile it from source
else:
self.logger.debug("Compiling %s (%s)", kernel_filename, kernel_hash)
#Create kernel string
kernel_string = ""
for key, value in kwargs.items():
kernel_string += "#define {:s} {:s}\n".format(str(key), str(value))
kernel_string += '#include "{:s}"'.format(os.path.join(self.module_path, kernel_filename))
if (self.use_cache):
with io.open(cached_kernel_filename + ".txt", "w") as file:
file.write(kernel_string)
with Timer("compiler") as timer:
cubin = cuda_compiler.compile(kernel_string, include_dirs=include_dirs, no_extern_c=no_extern_c, cache_dir=False)
module = cuda.module_from_buffer(cubin, message_handler=cuda_compile_message_handler)
if (self.use_cache):
with io.open(cached_kernel_filename, "wb") as file:
file.write(cubin)
kernel = module.get_function(kernel_function_name)
kernel.prepare(prepared_call_args)
self.kernels[kernel_hash] = kernel
return kernel
"""
Clears the kernel cache (useful for debugging & development)
"""
def clear_kernel_cache(self):
self.logger.debug("Clearing cache")
self.kernels = {}
gc.collect()
"""
Synchronizes all streams etc
"""
def synchronize(self):
self.cuda_context.synchronize()
"""
Class that holds 2D data
"""
class CudaArray2D:
"""
Uploads initial data to the CL device
"""
def __init__(self, stream, nx, ny, x_halo, y_halo, cpu_data=None, dtype=np.float32):
self.logger = logging.getLogger(__name__)
self.nx = nx
self.ny = ny
self.x_halo = x_halo
self.y_halo = y_halo
nx_halo = nx + 2*x_halo
ny_halo = ny + 2*y_halo
#self.logger.debug("Allocating [%dx%d] buffer", self.nx, self.ny)
#Should perhaps use pycuda.driver.mem_alloc_data.pitch() here
self.data = pycuda.gpuarray.empty((ny_halo, nx_halo), dtype)
#If we don't have any data, just allocate and return
if cpu_data is None:
return
#Make sure data is in proper format
assert cpu_data.shape == (ny_halo, nx_halo) or cpu_data.shape == (self.ny, self.nx), "Wrong shape of data %s vs %s / %s" % (str(cpu_data.shape), str((self.ny, self.nx)), str((ny_halo, nx_halo)))
assert cpu_data.itemsize == 4, "Wrong size of data type"
assert not np.isfortran(cpu_data), "Wrong datatype (Fortran, expected C)"
#Create copy object from host to device
copy = cuda.Memcpy2D()
copy.set_src_host(cpu_data)
copy.set_dst_device(self.data.gpudata)
#Set offsets of upload in destination
x_offset = (nx_halo - cpu_data.shape[1]) // 2
y_offset = (ny_halo - cpu_data.shape[0]) // 2
copy.dst_x_in_bytes = x_offset*self.data.strides[1]
copy.dst_y = y_offset
#Set destination pitch
copy.dst_pitch = self.data.strides[0]
#Set width in bytes to copy for each row and
#number of rows to copy
width = max(self.nx, cpu_data.shape[1])
height = max(self.ny, cpu_data.shape[0])
copy.width_in_bytes = width*cpu_data.itemsize
copy.height = height
#Perform the copy
copy(stream)
#self.logger.debug("Buffer <%s> [%dx%d]: Allocated ", int(self.data.gpudata), self.nx, self.ny)
def __del__(self, *args):
#self.logger.debug("Buffer <%s> [%dx%d]: Releasing ", int(self.data.gpudata), self.nx, self.ny)
self.data.gpudata.free()
self.data = None
"""
Enables downloading data from GPU to Python
"""
def download(self, stream, async=False):
#self.logger.debug("Downloading [%dx%d] buffer", self.nx, self.ny)
#Allocate host memory
#cpu_data = cuda.pagelocked_empty((self.ny, self.nx), np.float32)
cpu_data = np.empty((self.ny, self.nx), dtype=np.float32)
#Create copy object from device to host
copy = cuda.Memcpy2D()
copy.set_src_device(self.data.gpudata)
copy.set_dst_host(cpu_data)
#Set offsets and pitch of source
copy.src_x_in_bytes = self.x_halo*self.data.strides[1]
copy.src_y = self.y_halo
copy.src_pitch = self.data.strides[0]
#Set width in bytes to copy for each row and
#number of rows to copy
copy.width_in_bytes = self.nx*cpu_data.itemsize
copy.height = self.ny
copy(stream)
if async==False:
stream.synchronize()
return cpu_data
"""
Class that holds 2D data
"""
class CudaArray3D:
"""
Uploads initial data to the CL device
"""
def __init__(self, stream, nx, ny, nz, x_halo, y_halo, z_halo, cpu_data=None, dtype=np.float32):
self.logger = logging.getLogger(__name__)
self.nx = nx
self.ny = ny
self.nz = nz
self.x_halo = x_halo
self.y_halo = y_halo
self.z_halo = z_halo
nx_halo = nx + 2*x_halo
ny_halo = ny + 2*y_halo
nz_halo = nz + 2*z_halo
#self.logger.debug("Allocating [%dx%dx%d] buffer", self.nx, self.ny, self.nz)
#Should perhaps use pycuda.driver.mem_alloc_data.pitch() here
self.data = pycuda.gpuarray.empty((nz_halo, ny_halo, nx_halo), dtype)
#If we don't have any data, just allocate and return
if cpu_data is None:
return
#Make sure data is in proper format
assert cpu_data.shape == (nz_halo, ny_halo, nx_halo) or cpu_data.shape == (self.nz, self.ny, self.nx), "Wrong shape of data %s vs %s / %s" % (str(cpu_data.shape), str((self.nz, self.ny, self.nx)), str((nz_halo, ny_halo, nx_halo)))
assert cpu_data.itemsize == 4, "Wrong size of data type"
assert not np.isfortran(cpu_data), "Wrong datatype (Fortran, expected C)"
#Create copy object from host to device
copy = cuda.Memcpy3D()
copy.set_src_host(cpu_data)
copy.set_dst_device(self.data.gpudata)
#Set offsets of destination
x_offset = (nx_halo - cpu_data.shape[2]) // 2
y_offset = (ny_halo - cpu_data.shape[1]) // 2
z_offset = (nz_halo - cpu_data.shape[0]) // 2
copy.dst_x_in_bytes = x_offset*self.data.strides[1]
copy.dst_y = y_offset
copy.dst_z = z_offset
#Set pitch of destination
copy.dst_pitch = self.data.strides[0]
#Set width in bytes to copy for each row and
#number of rows to copy
width = max(self.nx, cpu_data.shape[2])
height = max(self.ny, cpu_data.shape[1])
depth = max(self.nz, cpu-data.shape[0])
copy.width_in_bytes = width*cpu_data.itemsize
copy.height = height
copy.depth = depth
#Perform the copy
copy(stream)
#self.logger.debug("Buffer <%s> [%dx%d]: Allocated ", int(self.data.gpudata), self.nx, self.ny)
def __del__(self, *args):
#self.logger.debug("Buffer <%s> [%dx%d]: Releasing ", int(self.data.gpudata), self.nx, self.ny)
self.data.gpudata.free()
self.data = None
"""
Enables downloading data from GPU to Python
"""
def download(self, stream, async=False):
#self.logger.debug("Downloading [%dx%d] buffer", self.nx, self.ny)
#Allocate host memory
#cpu_data = cuda.pagelocked_empty((self.ny, self.nx), np.float32)
cpu_data = np.empty((self.nz, self.ny, self.nx), dtype=np.float32)
#Create copy object from device to host
copy = cuda.Memcpy2D()
copy.set_src_device(self.data.gpudata)
copy.set_dst_host(cpu_data)
#Set offsets and pitch of source
copy.src_x_in_bytes = self.x_halo*self.data.strides[1]
copy.src_y = self.y_halo
copy.src_z = self.z_halo
copy.src_pitch = self.data.strides[0]
#Set width in bytes to copy for each row and
#number of rows to copy
copy.width_in_bytes = self.nx*cpu_data.itemsize
copy.height = self.ny
copy.depth = self.nz
copy(stream)
if async==False:
stream.synchronize()
return cpu_data
"""
A class representing an Arakawa A type (unstaggered, logically Cartesian) grid
"""
class ArakawaA2D:
"""
Uploads initial data to the CL device
"""
def __init__(self, stream, nx, ny, halo_x, halo_y, cpu_variables):
self.logger = logging.getLogger(__name__)
self.gpu_variables = []
for cpu_variable in cpu_variables:
self.gpu_variables += [CudaArray2D(stream, nx, ny, halo_x, halo_y, cpu_variable)]
def __getitem__(self, key):
assert type(key) == int, "Indexing is int based"
if (key > len(self.gpu_variables) or key < 0):
raise IndexError("Out of bounds")
return self.gpu_variables[key]
"""
Enables downloading data from CL device to Python
"""
def download(self, stream):
cpu_variables = []
for gpu_variable in self.gpu_variables:
cpu_variables += [gpu_variable.download(stream, async=True)]
stream.synchronize()
return cpu_variables