[92da231] | 1 | r""" |
---|
[eafc9fa] | 2 | DLL driver for C kernels |
---|
[750ffa5] | 3 | |
---|
[3764ec1] | 4 | If the environment variable *SAS_OPENMP* is set, then sasmodels |
---|
| 5 | will attempt to compile with OpenMP flags so that the model can use all |
---|
| 6 | available kernels. This may or may not be available on your compiler |
---|
| 7 | toolchain. Depending on operating system and environment. |
---|
[92da231] | 8 | |
---|
[3764ec1] | 9 | Windows does not have provide a compiler with the operating system. |
---|
| 10 | Instead, we assume that TinyCC is installed and available. This can |
---|
| 11 | be done with a simple pip command if it is not already available:: |
---|
[92da231] | 12 | |
---|
[3764ec1] | 13 | pip install tinycc |
---|
[92da231] | 14 | |
---|
[3764ec1] | 15 | If Microsoft Visual C++ is available (because VCINSTALLDIR is |
---|
| 16 | defined in the environment), then that will be used instead. |
---|
| 17 | Microsoft Visual C++ for Python is available from Microsoft: |
---|
[92da231] | 18 | |
---|
[d138d43] | 19 | `<http://www.microsoft.com/en-us/download/details.aspx?id=44266>`_ |
---|
[92da231] | 20 | |
---|
[3764ec1] | 21 | If neither compiler is available, sasmodels will check for *MinGW*, |
---|
| 22 | the GNU compiler toolchain. This available in packages such as Anaconda |
---|
| 23 | and PythonXY, or available stand alone. This toolchain has had |
---|
| 24 | difficulties on some systems, and may or may not work for you. |
---|
| 25 | |
---|
| 26 | You can control which compiler to use by setting SAS_COMPILER in the |
---|
| 27 | environment: |
---|
| 28 | |
---|
| 29 | - tinycc (Windows): use the TinyCC compiler shipped with SasView |
---|
| 30 | - msvc (Windows): use the Microsoft Visual C++ compiler |
---|
| 31 | - mingw (Windows): use the MinGW GNU cc compiler |
---|
| 32 | - unix (Linux): use the system cc compiler. |
---|
| 33 | - unix (Mac): use the clang compiler. You will need XCode installed, and |
---|
| 34 | the XCode command line tools. Mac comes with OpenCL drivers, so generally |
---|
| 35 | this will not be needed. |
---|
| 36 | |
---|
| 37 | Both *msvc* and *mingw* require that the compiler is available on your path. |
---|
| 38 | For *msvc*, this can done by running vcvarsall.bat in a windows terminal. |
---|
| 39 | Install locations are system dependent, such as: |
---|
[92da231] | 40 | |
---|
[2d81cfe] | 41 | C:\Program Files (x86)\Common Files\Microsoft\Visual |
---|
| 42 | C++ for Python\9.0\vcvarsall.bat |
---|
[92da231] | 43 | |
---|
| 44 | or maybe |
---|
| 45 | |
---|
[2d81cfe] | 46 | C:\Users\yourname\AppData\Local\Programs\Common\Microsoft\Visual |
---|
| 47 | C++ for Python\9.0\vcvarsall.bat |
---|
[92da231] | 48 | |
---|
[3764ec1] | 49 | OpenMP for *msvc* requires the Microsoft vcomp90.dll library, which doesn't |
---|
| 50 | seem to be included with the compiler, nor does there appear to be a public |
---|
| 51 | download location. There may be one on your machine already in a location |
---|
| 52 | such as: |
---|
[92da231] | 53 | |
---|
| 54 | C:\Windows\winsxs\x86_microsoft.vc90.openmp*\vcomp90.dll |
---|
| 55 | |
---|
[3764ec1] | 56 | If you copy this to somewhere on your path, such as the python directory or |
---|
| 57 | the install directory for this application, then OpenMP should be supported. |
---|
| 58 | |
---|
| 59 | For full control of the compiler, define a function |
---|
| 60 | *compile_command(source,output)* which takes the name of the source file |
---|
| 61 | and the name of the output file and returns a compile command that can be |
---|
| 62 | evaluated in the shell. For even more control, replace the entire |
---|
| 63 | *compile(source,output)* function. |
---|
| 64 | |
---|
| 65 | The global attribute *ALLOW_SINGLE_PRECISION_DLLS* should be set to *False* if |
---|
| 66 | you wish to prevent single precision floating point evaluation for the compiled |
---|
| 67 | models, otherwise set it defaults to *True*. |
---|
[14de349] | 68 | """ |
---|
[eafc9fa] | 69 | from __future__ import print_function |
---|
[750ffa5] | 70 | |
---|
[5d4777d] | 71 | import sys |
---|
| 72 | import os |
---|
[40a87fa] | 73 | from os.path import join as joinpath, splitext |
---|
[7e16db7] | 74 | import subprocess |
---|
[df4dc86] | 75 | import tempfile |
---|
[7ae2b7f] | 76 | import ctypes as ct # type: ignore |
---|
[886fa25] | 77 | import _ctypes as _ct |
---|
[821a9c6] | 78 | import logging |
---|
[14de349] | 79 | |
---|
[7ae2b7f] | 80 | import numpy as np # type: ignore |
---|
[14de349] | 81 | |
---|
[739aad4] | 82 | try: |
---|
| 83 | import tinycc |
---|
| 84 | except ImportError: |
---|
| 85 | tinycc = None |
---|
| 86 | |
---|
[cb6ecf4] | 87 | from . import generate |
---|
[f619de7] | 88 | from .kernel import KernelModel, Kernel |
---|
| 89 | from .kernelpy import PyInput |
---|
[2c801fe] | 90 | from .exception import annotate_exception |
---|
[f619de7] | 91 | from .generate import F16, F32, F64 |
---|
| 92 | |
---|
[2d81cfe] | 93 | # pylint: disable=unused-import |
---|
[f619de7] | 94 | try: |
---|
| 95 | from typing import Tuple, Callable, Any |
---|
| 96 | from .modelinfo import ModelInfo |
---|
| 97 | from .details import CallDetails |
---|
| 98 | except ImportError: |
---|
| 99 | pass |
---|
[2d81cfe] | 100 | # pylint: enable=unused-import |
---|
[14de349] | 101 | |
---|
[3199b17] | 102 | # Compiler output is a byte stream that needs to be decode in python 3. |
---|
[e44432d] | 103 | decode = (lambda s: s) if sys.version_info[0] < 3 else (lambda s: s.decode('utf8')) |
---|
| 104 | |
---|
[1a3559f] | 105 | if "SAS_DLL_PATH" in os.environ: |
---|
| 106 | SAS_DLL_PATH = os.environ["SAS_DLL_PATH"] |
---|
| 107 | else: |
---|
| 108 | # Assume the default location of module DLLs is in .sasmodels/compiled_models. |
---|
| 109 | SAS_DLL_PATH = os.path.join(os.path.expanduser("~"), ".sasmodels", "compiled_models") |
---|
| 110 | |
---|
[739aad4] | 111 | if "SAS_COMPILER" in os.environ: |
---|
[b2f1d2f] | 112 | COMPILER = os.environ["SAS_COMPILER"] |
---|
[739aad4] | 113 | elif os.name == 'nt': |
---|
[b2f1d2f] | 114 | if tinycc is not None: |
---|
| 115 | COMPILER = "tinycc" |
---|
| 116 | elif "VCINSTALLDIR" in os.environ: |
---|
[3199b17] | 117 | # If vcvarsall.bat has been called, then VCINSTALLDIR is in the |
---|
| 118 | # environment and we can use the MSVC compiler. Otherwise, if |
---|
| 119 | # tinycc is available then use it. Otherwise, hope that mingw |
---|
| 120 | # is available. |
---|
[b2f1d2f] | 121 | COMPILER = "msvc" |
---|
[fb69211] | 122 | else: |
---|
[b2f1d2f] | 123 | COMPILER = "mingw" |
---|
[5d4777d] | 124 | else: |
---|
[b2f1d2f] | 125 | COMPILER = "unix" |
---|
[739aad4] | 126 | |
---|
[3199b17] | 127 | ARCH = "" if ct.sizeof(ct.c_void_p) > 4 else "x86" # 4 byte pointers on x86. |
---|
[b2f1d2f] | 128 | if COMPILER == "unix": |
---|
[3199b17] | 129 | # Generic unix compile. |
---|
| 130 | # On Mac users will need the X code command line tools installed. |
---|
[7e16db7] | 131 | #COMPILE = "gcc-mp-4.7 -shared -fPIC -std=c99 -fopenmp -O2 -Wall %s -o %s -lm -lgomp" |
---|
[aa343d6] | 132 | CC = "cc -shared -fPIC -std=c99 -O2 -Wall".split() |
---|
[3199b17] | 133 | # Add OpenMP support if not running on a Mac. |
---|
[7e16db7] | 134 | if sys.platform != "darwin": |
---|
[3199b17] | 135 | # OpenMP seems to be broken on gcc 5.4.0 (ubuntu 16.04.9). |
---|
[33969b6] | 136 | # Shut it off for all unix until we can investigate. |
---|
| 137 | #CC.append("-fopenmp") |
---|
| 138 | pass |
---|
[7e16db7] | 139 | def compile_command(source, output): |
---|
[40a87fa] | 140 | """unix compiler command""" |
---|
[15e74ad] | 141 | return CC + [source, "-o", output, "-lm"] |
---|
[b2f1d2f] | 142 | elif COMPILER == "msvc": |
---|
[739aad4] | 143 | # Call vcvarsall.bat before compiling to set path, headers, libs, etc. |
---|
| 144 | # MSVC compiler is available, so use it. OpenMP requires a copy of |
---|
| 145 | # vcomp90.dll on the path. One may be found here: |
---|
| 146 | # C:/Windows/winsxs/x86_microsoft.vc90.openmp*/vcomp90.dll |
---|
[3199b17] | 147 | # Copy this to the python directory and uncomment the OpenMP COMPILE. |
---|
| 148 | # TODO: Remove intermediate OBJ file created in the directory. |
---|
| 149 | # TODO: Maybe don't use randomized name for the c file. |
---|
| 150 | # TODO: Maybe ask distutils to find MSVC. |
---|
[739aad4] | 151 | CC = "cl /nologo /Ox /MD /W3 /GS- /DNDEBUG".split() |
---|
| 152 | if "SAS_OPENMP" in os.environ: |
---|
| 153 | CC.append("/openmp") |
---|
| 154 | LN = "/link /DLL /INCREMENTAL:NO /MANIFEST".split() |
---|
| 155 | def compile_command(source, output): |
---|
[40a87fa] | 156 | """MSVC compiler command""" |
---|
[739aad4] | 157 | return CC + ["/Tp%s"%source] + LN + ["/OUT:%s"%output] |
---|
[b2f1d2f] | 158 | elif COMPILER == "tinycc": |
---|
[739aad4] | 159 | # TinyCC compiler. |
---|
| 160 | CC = [tinycc.TCC] + "-shared -rdynamic -Wall".split() |
---|
| 161 | def compile_command(source, output): |
---|
[40a87fa] | 162 | """tinycc compiler command""" |
---|
[739aad4] | 163 | return CC + [source, "-o", output] |
---|
[b2f1d2f] | 164 | elif COMPILER == "mingw": |
---|
[739aad4] | 165 | # MinGW compiler. |
---|
| 166 | CC = "gcc -shared -std=c99 -O2 -Wall".split() |
---|
| 167 | if "SAS_OPENMP" in os.environ: |
---|
| 168 | CC.append("-fopenmp") |
---|
| 169 | def compile_command(source, output): |
---|
[40a87fa] | 170 | """mingw compiler command""" |
---|
[739aad4] | 171 | return CC + [source, "-o", output, "-lm"] |
---|
[df4dc86] | 172 | |
---|
[5d316e9] | 173 | ALLOW_SINGLE_PRECISION_DLLS = True |
---|
[5d4777d] | 174 | |
---|
[3199b17] | 175 | |
---|
[b297ba9] | 176 | def compile_model(source, output): |
---|
[40a87fa] | 177 | # type: (str, str) -> None |
---|
| 178 | """ |
---|
| 179 | Compile *source* producing *output*. |
---|
| 180 | |
---|
| 181 | Raises RuntimeError if the compile failed or the output wasn't produced. |
---|
| 182 | """ |
---|
[7e16db7] | 183 | command = compile_command(source=source, output=output) |
---|
| 184 | command_str = " ".join('"%s"'%p if ' ' in p else p for p in command) |
---|
| 185 | logging.info(command_str) |
---|
| 186 | try: |
---|
[3199b17] | 187 | # Need shell=True on windows to keep console box from popping up. |
---|
[15e74ad] | 188 | shell = (os.name == 'nt') |
---|
[aa343d6] | 189 | subprocess.check_output(command, shell=shell, stderr=subprocess.STDOUT) |
---|
[7e16db7] | 190 | except subprocess.CalledProcessError as exc: |
---|
[e44432d] | 191 | output = decode(exc.output) |
---|
| 192 | raise RuntimeError("compile failed.\n%s\n%s"%(command_str, output)) |
---|
[7e16db7] | 193 | if not os.path.exists(output): |
---|
| 194 | raise RuntimeError("compile failed. File is in %r"%source) |
---|
[750ffa5] | 195 | |
---|
[3199b17] | 196 | |
---|
[151f3bc] | 197 | def dll_name(model_info, dtype): |
---|
[f619de7] | 198 | # type: (ModelInfo, np.dtype) -> str |
---|
[5d4777d] | 199 | """ |
---|
[151f3bc] | 200 | Name of the dll containing the model. This is the base file name without |
---|
| 201 | any path or extension, with a form such as 'sas_sphere32'. |
---|
[5d4777d] | 202 | """ |
---|
[151f3bc] | 203 | bits = 8*dtype.itemsize |
---|
[56b2687] | 204 | basename = "sas%d_%s"%(bits, model_info.id) |
---|
[c7118f4] | 205 | basename += ARCH + ".so" |
---|
[01b8659] | 206 | |
---|
[3199b17] | 207 | # Hack to find precompiled dlls. |
---|
[c7118f4] | 208 | path = joinpath(generate.DATA_PATH, '..', 'compiled_models', basename) |
---|
[01b8659] | 209 | if os.path.exists(path): |
---|
| 210 | return path |
---|
| 211 | |
---|
[1a3559f] | 212 | return joinpath(SAS_DLL_PATH, basename) |
---|
[5d4777d] | 213 | |
---|
[f619de7] | 214 | |
---|
[151f3bc] | 215 | def dll_path(model_info, dtype): |
---|
[f619de7] | 216 | # type: (ModelInfo, np.dtype) -> str |
---|
[151f3bc] | 217 | """ |
---|
| 218 | Complete path to the dll for the model. Note that the dll may not |
---|
| 219 | exist yet if it hasn't been compiled. |
---|
| 220 | """ |
---|
[1a3559f] | 221 | return os.path.join(SAS_DLL_PATH, dll_name(model_info, dtype)) |
---|
[5d4777d] | 222 | |
---|
[f619de7] | 223 | |
---|
| 224 | def make_dll(source, model_info, dtype=F64): |
---|
| 225 | # type: (str, ModelInfo, np.dtype) -> str |
---|
[5d4777d] | 226 | """ |
---|
[f619de7] | 227 | Returns the path to the compiled model defined by *kernel_module*. |
---|
[5d4777d] | 228 | |
---|
[f619de7] | 229 | If the model has not been compiled, or if the source file(s) are newer |
---|
| 230 | than the dll, then *make_dll* will compile the model before returning. |
---|
| 231 | This routine does not load the resulting dll. |
---|
[5d4777d] | 232 | |
---|
[aa4946b] | 233 | *dtype* is a numpy floating point precision specifier indicating whether |
---|
[f619de7] | 234 | the model should be single, double or long double precision. The default |
---|
| 235 | is double precision, *np.dtype('d')*. |
---|
[5d4777d] | 236 | |
---|
[f619de7] | 237 | Set *sasmodels.ALLOW_SINGLE_PRECISION_DLLS* to False if single precision |
---|
| 238 | models are not allowed as DLLs. |
---|
[aa4946b] | 239 | |
---|
[1a3559f] | 240 | Set *sasmodels.kerneldll.SAS_DLL_PATH* to the compiled dll output path. |
---|
| 241 | Alternatively, set the environment variable *SAS_DLL_PATH*. |
---|
[bf94e6e] | 242 | The default is in ~/.sasmodels/compiled_models. |
---|
[5d4777d] | 243 | """ |
---|
[f619de7] | 244 | if dtype == F16: |
---|
[5d316e9] | 245 | raise ValueError("16 bit floats not supported") |
---|
[f619de7] | 246 | if dtype == F32 and not ALLOW_SINGLE_PRECISION_DLLS: |
---|
[3199b17] | 247 | dtype = F64 # Force 64-bit dll. |
---|
| 248 | # Note: dtype may be F128 for long double precision. |
---|
[750ffa5] | 249 | |
---|
[17bbadd] | 250 | dll = dll_path(model_info, dtype) |
---|
[e1454ab] | 251 | |
---|
| 252 | if not os.path.exists(dll): |
---|
[5a91c6b] | 253 | need_recompile = True |
---|
| 254 | else: |
---|
| 255 | dll_time = os.path.getmtime(dll) |
---|
[0dc34c3] | 256 | newest_source = generate.dll_timestamp(model_info) |
---|
[5a91c6b] | 257 | need_recompile = dll_time < newest_source |
---|
| 258 | if need_recompile: |
---|
[3199b17] | 259 | # Make sure the DLL path exists. |
---|
[1a3559f] | 260 | if not os.path.exists(SAS_DLL_PATH): |
---|
| 261 | os.makedirs(SAS_DLL_PATH) |
---|
[40a87fa] | 262 | basename = splitext(os.path.basename(dll))[0] + "_" |
---|
| 263 | system_fd, filename = tempfile.mkstemp(suffix=".c", prefix=basename) |
---|
[f619de7] | 264 | source = generate.convert_type(source, dtype) |
---|
[40a87fa] | 265 | with os.fdopen(system_fd, "w") as file_handle: |
---|
| 266 | file_handle.write(source) |
---|
[b297ba9] | 267 | compile_model(source=filename, output=dll) |
---|
[3199b17] | 268 | # Comment the following to keep the generated C file. |
---|
| 269 | # Note: If there is a syntax error then compile raises an error |
---|
[7e16db7] | 270 | # and the source file will not be deleted. |
---|
[c036ddb] | 271 | os.unlink(filename) |
---|
| 272 | #print("saving compiled file in %r"%filename) |
---|
[aa4946b] | 273 | return dll |
---|
| 274 | |
---|
| 275 | |
---|
[f619de7] | 276 | def load_dll(source, model_info, dtype=F64): |
---|
| 277 | # type: (str, ModelInfo, np.dtype) -> "DllModel" |
---|
[aa4946b] | 278 | """ |
---|
[823e620] | 279 | Create and load a dll corresponding to the source, info pair returned |
---|
[aa4946b] | 280 | from :func:`sasmodels.generate.make` compiled for the target precision. |
---|
| 281 | |
---|
| 282 | See :func:`make_dll` for details on controlling the dll path and the |
---|
| 283 | allowed floating point precision. |
---|
| 284 | """ |
---|
[17bbadd] | 285 | filename = make_dll(source, model_info, dtype=dtype) |
---|
| 286 | return DllModel(filename, model_info, dtype=dtype) |
---|
[5d4777d] | 287 | |
---|
[f619de7] | 288 | |
---|
| 289 | class DllModel(KernelModel): |
---|
[14de349] | 290 | """ |
---|
| 291 | ctypes wrapper for a single model. |
---|
| 292 | |
---|
[17bbadd] | 293 | *source* and *model_info* are the model source and interface as returned |
---|
[14de349] | 294 | from :func:`gen.make`. |
---|
| 295 | |
---|
| 296 | *dtype* is the desired model precision. Any numpy dtype for single |
---|
| 297 | or double precision floats will do, such as 'f', 'float32' or 'single' |
---|
| 298 | for single and 'd', 'float64' or 'double' for double. Double precision |
---|
| 299 | is an optional extension which may not be available on all devices. |
---|
[ff7119b] | 300 | |
---|
| 301 | Call :meth:`release` when done with the kernel. |
---|
[14de349] | 302 | """ |
---|
[17bbadd] | 303 | def __init__(self, dllpath, model_info, dtype=generate.F32): |
---|
[f619de7] | 304 | # type: (str, ModelInfo, np.dtype) -> None |
---|
[17bbadd] | 305 | self.info = model_info |
---|
[ce27e21] | 306 | self.dllpath = dllpath |
---|
[f619de7] | 307 | self._dll = None # type: ct.CDLL |
---|
[3199b17] | 308 | self._kernels = None # type: List[Callable, Callable] |
---|
[750ffa5] | 309 | self.dtype = np.dtype(dtype) |
---|
[14de349] | 310 | |
---|
[ce27e21] | 311 | def _load_dll(self): |
---|
[f619de7] | 312 | # type: () -> None |
---|
[2c801fe] | 313 | try: |
---|
[f619de7] | 314 | self._dll = ct.CDLL(self.dllpath) |
---|
[4d76711] | 315 | except: |
---|
| 316 | annotate_exception("while loading "+self.dllpath) |
---|
[2c801fe] | 317 | raise |
---|
[14de349] | 318 | |
---|
[886fa25] | 319 | float_type = (ct.c_float if self.dtype == generate.F32 |
---|
| 320 | else ct.c_double if self.dtype == generate.F64 |
---|
| 321 | else ct.c_longdouble) |
---|
[ce27e21] | 322 | |
---|
[a738209] | 323 | # int, int, int, int*, double*, double*, double*, double*, double |
---|
[6e7ba14] | 324 | argtypes = [ct.c_int32]*3 + [ct.c_void_p]*4 + [float_type, ct.c_int32] |
---|
[9eb3632] | 325 | names = [generate.kernel_name(self.info, variant) |
---|
| 326 | for variant in ("Iq", "Iqxy", "Imagnetic")] |
---|
| 327 | self._kernels = [self._dll[name] for name in names] |
---|
| 328 | for k in self._kernels: |
---|
| 329 | k.argtypes = argtypes |
---|
[ce27e21] | 330 | |
---|
| 331 | def __getstate__(self): |
---|
[f619de7] | 332 | # type: () -> Tuple[ModelInfo, str] |
---|
[eafc9fa] | 333 | return self.info, self.dllpath |
---|
[ce27e21] | 334 | |
---|
| 335 | def __setstate__(self, state): |
---|
[f619de7] | 336 | # type: (Tuple[ModelInfo, str]) -> None |
---|
[eafc9fa] | 337 | self.info, self.dllpath = state |
---|
[f619de7] | 338 | self._dll = None |
---|
[ce27e21] | 339 | |
---|
[48fbd50] | 340 | def make_kernel(self, q_vectors): |
---|
[f619de7] | 341 | # type: (List[np.ndarray]) -> DllKernel |
---|
[eafc9fa] | 342 | q_input = PyInput(q_vectors, self.dtype) |
---|
[3199b17] | 343 | # Note: DLL is lazy loaded. |
---|
[f619de7] | 344 | if self._dll is None: |
---|
| 345 | self._load_dll() |
---|
[9eb3632] | 346 | is_2d = len(q_vectors) == 2 |
---|
| 347 | kernel = self._kernels[1:3] if is_2d else [self._kernels[0]]*2 |
---|
[3c56da87] | 348 | return DllKernel(kernel, self.info, q_input) |
---|
[4d76711] | 349 | |
---|
[eafc9fa] | 350 | def release(self): |
---|
[f619de7] | 351 | # type: () -> None |
---|
[14de349] | 352 | """ |
---|
[eafc9fa] | 353 | Release any resources associated with the model. |
---|
[14de349] | 354 | """ |
---|
[886fa25] | 355 | dll_handle = self._dll._handle |
---|
[6ad0e87] | 356 | if os.name == 'nt': |
---|
[40a87fa] | 357 | ct.windll.kernel32.FreeLibrary(dll_handle) |
---|
| 358 | else: |
---|
[886fa25] | 359 | _ct.dlclose(dll_handle) |
---|
| 360 | del self._dll |
---|
| 361 | self._dll = None |
---|
[14de349] | 362 | |
---|
[3199b17] | 363 | |
---|
[f619de7] | 364 | class DllKernel(Kernel): |
---|
[ff7119b] | 365 | """ |
---|
| 366 | Callable SAS kernel. |
---|
| 367 | |
---|
[b3f6bc3] | 368 | *kernel* is the c function to call. |
---|
[ff7119b] | 369 | |
---|
[17bbadd] | 370 | *model_info* is the module information |
---|
[ff7119b] | 371 | |
---|
[3c56da87] | 372 | *q_input* is the DllInput q vectors at which the kernel should be |
---|
[ff7119b] | 373 | evaluated. |
---|
| 374 | |
---|
| 375 | The resulting call method takes the *pars*, a list of values for |
---|
[823e620] | 376 | the fixed parameters to the kernel, and *pd_pars*, a list of (value, weight) |
---|
[ff7119b] | 377 | vectors for the polydisperse parameters. *cutoff* determines the |
---|
| 378 | integration limits: any points with combined weight less than *cutoff* |
---|
| 379 | will not be calculated. |
---|
| 380 | |
---|
| 381 | Call :meth:`release` when done with the kernel instance. |
---|
| 382 | """ |
---|
[17bbadd] | 383 | def __init__(self, kernel, model_info, q_input): |
---|
[f619de7] | 384 | # type: (Callable[[], np.ndarray], ModelInfo, PyInput) -> None |
---|
[3199b17] | 385 | dtype = q_input.dtype |
---|
[3c56da87] | 386 | self.q_input = q_input |
---|
[3199b17] | 387 | self.kernel = kernel |
---|
| 388 | |
---|
| 389 | # Attributes accessed from the outside. |
---|
[445d1c0] | 390 | self.dim = '2d' if q_input.is_2d else '1d' |
---|
[3199b17] | 391 | self.info = model_info |
---|
| 392 | self.dtype = dtype |
---|
| 393 | |
---|
| 394 | # Converter to translate input to target type. |
---|
| 395 | self._as_dtype = (np.float32 if dtype == generate.F32 |
---|
| 396 | else np.float64 if dtype == generate.F64 |
---|
| 397 | else np.float128) |
---|
| 398 | |
---|
| 399 | # Holding place for the returned value. |
---|
[6e7ba14] | 400 | nout = 2 if self.info.have_Fq else 1 |
---|
[3199b17] | 401 | extra_q = 4 # Total weight, form volume, shell volume and R_eff. |
---|
| 402 | self.result = np.empty(self.q_input.nq*nout + extra_q, dtype) |
---|
| 403 | |
---|
| 404 | def _call_kernel(self, call_details, values, cutoff, magnetic, |
---|
[a34b811] | 405 | radius_effective_mode): |
---|
[3199b17] | 406 | # type: (CallDetails, np.ndarray, float, bool, int) -> np.ndarray |
---|
| 407 | |
---|
| 408 | # Setup kernel function and arguments. |
---|
[9eb3632] | 409 | kernel = self.kernel[1 if magnetic else 0] |
---|
[3199b17] | 410 | kernel_args = [ |
---|
| 411 | self.q_input.nq, # Number of inputs. |
---|
| 412 | None, # Placeholder for pd_start. |
---|
| 413 | None, # Placeholder for pd_stop. |
---|
| 414 | call_details.buffer.ctypes.data, # Problem definition. |
---|
| 415 | values.ctypes.data, # Parameter values. |
---|
| 416 | self.q_input.q.ctypes.data, # Q values. |
---|
| 417 | self.result.ctypes.data, # Result storage. |
---|
| 418 | self._as_dtype(cutoff), # Probability cutoff. |
---|
[a34b811] | 419 | radius_effective_mode, # R_eff mode. |
---|
[9eb3632] | 420 | ] |
---|
[3199b17] | 421 | |
---|
| 422 | # Call kernel and retrieve results. |
---|
[c036ddb] | 423 | #print("Calling DLL") |
---|
[bde38b5] | 424 | #call_details.show(values) |
---|
[9eb3632] | 425 | step = 100 |
---|
[3199b17] | 426 | # TODO: Do we need the explicit sleep like the OpenCL and CUDA loops? |
---|
[bde38b5] | 427 | for start in range(0, call_details.num_eval, step): |
---|
| 428 | stop = min(start + step, call_details.num_eval) |
---|
[3199b17] | 429 | kernel_args[1:3] = [start, stop] |
---|
| 430 | kernel(*kernel_args) # type: ignore |
---|
[9eb3632] | 431 | |
---|
[14de349] | 432 | def release(self): |
---|
[f619de7] | 433 | # type: () -> None |
---|
[eafc9fa] | 434 | """ |
---|
[3199b17] | 435 | Release resources associated with the kernel. |
---|
[eafc9fa] | 436 | """ |
---|
[3199b17] | 437 | # TODO: OpenCL/CUDA allocate q_input in __init__ and free it in release. |
---|
| 438 | # Should we be doing the same for DLL? |
---|
| 439 | #self.q_input.release() |
---|
| 440 | pass |
---|
| 441 | |
---|
| 442 | def __del__(self): |
---|
| 443 | # type: () -> None |
---|
| 444 | self.release() |
---|