# Vendored parts of the code from https://github.com/flababah/cpuid.py, # so we replicate its copyright license. # Copyright (c) 2014 Anders Høst # # Permission is hereby granted, free of charge, to any person obtaining a copy of # this software and associated documentation files (the "Software"), to deal in # the Software without restriction, including without limitation the rights to # use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of # the Software, and to permit persons to whom the Software is furnished to do so, # subject to the following conditions: # # The above copyright notice and this permission notice shall be included in all # copies or substantial portions of the Software. # # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR # IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS # FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR # COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER # IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN # CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. from __future__ import annotations import ctypes import os from ctypes import CFUNCTYPE, POINTER, c_long, c_size_t, c_uint32, c_ulong, c_void_p from typing import ClassVar """ Determine whether Polars can be run on the current CPU. This must be done in pure Python, before the Polars binary is imported. If we were to try it on the Rust side the compiler could emit illegal instructions before/during the CPU feature check code. """ # Replaced during the build process with the architecture of our wheel. _POLARS_ARCH = "x86-64" # Replaced during the build process with our list of required feature flags # enabled at compile time. _POLARS_FEATURE_FLAGS = "+sse3,+ssse3,+sse4.1,+sse4.2,+popcnt,+cmpxchg16b,+avx,+avx2,+fma,+bmi1,+bmi2,+lzcnt,+pclmulqdq,+movbe" # Set to True during the build process if we are building a LTS CPU version. _POLARS_LTS_CPU = False _IS_WINDOWS = os.name == "nt" _IS_64BIT = ctypes.sizeof(ctypes.c_void_p) == 8 def get_lts_cpu() -> bool: return _POLARS_LTS_CPU def _open_posix_libc() -> ctypes.CDLL: # Avoid importing ctypes.util if possible. try: if os.uname().sysname == "Darwin": return ctypes.CDLL("libc.dylib", use_errno=True) else: return ctypes.CDLL("libc.so.6", use_errno=True) except Exception: from ctypes import util as ctutil return ctypes.CDLL(ctutil.find_library("c"), use_errno=True) # Posix x86_64: # Three first call registers : RDI, RSI, RDX # Volatile registers : RAX, RCX, RDX, RSI, RDI, R8-11 # Windows x86_64: # Three first call registers : RCX, RDX, R8 # Volatile registers : RAX, RCX, RDX, R8-11 # cdecl 32 bit: # Three first call registers : Stack (%esp) # Volatile registers : EAX, ECX, EDX # fmt: off _POSIX_64_OPC = [ 0x53, # push %rbx 0x89, 0xf0, # mov %esi,%eax 0x89, 0xd1, # mov %edx,%ecx 0x0f, 0xa2, # cpuid 0x89, 0x07, # mov %eax,(%rdi) 0x89, 0x5f, 0x04, # mov %ebx,0x4(%rdi) 0x89, 0x4f, 0x08, # mov %ecx,0x8(%rdi) 0x89, 0x57, 0x0c, # mov %edx,0xc(%rdi) 0x5b, # pop %rbx 0xc3 # retq ] _WINDOWS_64_OPC = [ 0x53, # push %rbx 0x89, 0xd0, # mov %edx,%eax 0x49, 0x89, 0xc9, # mov %rcx,%r9 0x44, 0x89, 0xc1, # mov %r8d,%ecx 0x0f, 0xa2, # cpuid 0x41, 0x89, 0x01, # mov %eax,(%r9) 0x41, 0x89, 0x59, 0x04, # mov %ebx,0x4(%r9) 0x41, 0x89, 0x49, 0x08, # mov %ecx,0x8(%r9) 0x41, 0x89, 0x51, 0x0c, # mov %edx,0xc(%r9) 0x5b, # pop %rbx 0xc3 # retq ] _CDECL_32_OPC = [ 0x53, # push %ebx 0x57, # push %edi 0x8b, 0x7c, 0x24, 0x0c, # mov 0xc(%esp),%edi 0x8b, 0x44, 0x24, 0x10, # mov 0x10(%esp),%eax 0x8b, 0x4c, 0x24, 0x14, # mov 0x14(%esp),%ecx 0x0f, 0xa2, # cpuid 0x89, 0x07, # mov %eax,(%edi) 0x89, 0x5f, 0x04, # mov %ebx,0x4(%edi) 0x89, 0x4f, 0x08, # mov %ecx,0x8(%edi) 0x89, 0x57, 0x0c, # mov %edx,0xc(%edi) 0x5f, # pop %edi 0x5b, # pop %ebx 0xc3 # ret ] # fmt: on # From memoryapi.h _MEM_COMMIT = 0x1000 _MEM_RESERVE = 0x2000 _MEM_RELEASE = 0x8000 _PAGE_EXECUTE_READWRITE = 0x40 class CPUID_struct(ctypes.Structure): _fields_: ClassVar[list[tuple[str, type]]] = [ (r, c_uint32) for r in ("eax", "ebx", "ecx", "edx") ] class CPUID: def __init__(self) -> None: if _IS_WINDOWS: if _IS_64BIT: # VirtualAlloc seems to fail under some weird # circumstances when ctypes.windll.kernel32 is # used under 64 bit Python. CDLL fixes this. self.win = ctypes.CDLL("kernel32.dll") opc = _WINDOWS_64_OPC else: # Here ctypes.windll.kernel32 is needed to get the # right DLL. Otherwise it will fail when running # 32 bit Python on 64 bit Windows. self.win = ctypes.windll.kernel32 # type: ignore[attr-defined] opc = _CDECL_32_OPC else: opc = _POSIX_64_OPC if _IS_64BIT else _CDECL_32_OPC size = len(opc) code = (ctypes.c_ubyte * size)(*opc) if _IS_WINDOWS: self.win.VirtualAlloc.restype = c_void_p self.win.VirtualAlloc.argtypes = [ ctypes.c_void_p, ctypes.c_size_t, ctypes.c_ulong, ctypes.c_ulong, ] self.addr = self.win.VirtualAlloc( None, size, _MEM_COMMIT | _MEM_RESERVE, _PAGE_EXECUTE_READWRITE ) if not self.addr: msg = "could not allocate memory for CPUID check" raise MemoryError(msg) ctypes.memmove(self.addr, code, size) else: import mmap # Only import if necessary. # On some platforms PROT_WRITE + PROT_EXEC is forbidden, so we first # only write and then mprotect into PROT_EXEC. libc = _open_posix_libc() mprotect = libc.mprotect mprotect.argtypes = (ctypes.c_void_p, ctypes.c_size_t, ctypes.c_int) mprotect.restype = ctypes.c_int self.mmap = mmap.mmap( -1, size, mmap.MAP_PRIVATE | mmap.MAP_ANONYMOUS, mmap.PROT_READ | mmap.PROT_WRITE, ) self.addr = ctypes.addressof(ctypes.c_void_p.from_buffer(self.mmap)) self.mmap.write(code) if mprotect(self.addr, size, mmap.PROT_READ | mmap.PROT_EXEC) != 0: msg = "could not execute mprotect for CPUID check" raise RuntimeError(msg) func_type = CFUNCTYPE(None, POINTER(CPUID_struct), c_uint32, c_uint32) self.func_ptr = func_type(self.addr) def __call__(self, eax: int, ecx: int = 0) -> CPUID_struct: struct = CPUID_struct() self.func_ptr(struct, eax, ecx) return struct def __del__(self) -> None: if _IS_WINDOWS: self.win.VirtualFree.restype = c_long self.win.VirtualFree.argtypes = [c_void_p, c_size_t, c_ulong] self.win.VirtualFree(self.addr, 0, _MEM_RELEASE) def _read_cpu_flags() -> dict[str, bool]: # CPU flags from https://en.wikipedia.org/wiki/CPUID cpuid = CPUID() cpuid1 = cpuid(1, 0) cpuid7 = cpuid(7, 0) cpuid81h = cpuid(0x80000001, 0) return { "sse3": bool(cpuid1.ecx & (1 << 0)), "ssse3": bool(cpuid1.ecx & (1 << 9)), "fma": bool(cpuid1.ecx & (1 << 12)), "cmpxchg16b": bool(cpuid1.ecx & (1 << 13)), "sse4.1": bool(cpuid1.ecx & (1 << 19)), "sse4.2": bool(cpuid1.ecx & (1 << 20)), "movbe": bool(cpuid1.ecx & (1 << 22)), "popcnt": bool(cpuid1.ecx & (1 << 23)), "pclmulqdq": bool(cpuid1.ecx & (1 << 1)), "avx": bool(cpuid1.ecx & (1 << 28)), "bmi1": bool(cpuid7.ebx & (1 << 3)), "bmi2": bool(cpuid7.ebx & (1 << 8)), "avx2": bool(cpuid7.ebx & (1 << 5)), "lzcnt": bool(cpuid81h.ecx & (1 << 5)), } def check_cpu_flags() -> None: if not _POLARS_FEATURE_FLAGS or os.environ.get("POLARS_SKIP_CPU_CHECK"): return expected_cpu_flags = [f.lstrip("+") for f in _POLARS_FEATURE_FLAGS.split(",")] supported_cpu_flags = _read_cpu_flags() missing_features = [] for f in expected_cpu_flags: if f not in supported_cpu_flags: msg = f"unknown feature flag: {f!r}" raise RuntimeError(msg) if not supported_cpu_flags[f]: missing_features.append(f) if missing_features: import warnings # Only import if necessary. warnings.warn( f"""Missing required CPU features. The following required CPU features were not detected: {", ".join(missing_features)} Continuing to use this version of Polars on this processor will likely result in a crash. Install the `polars-lts-cpu` package instead of `polars` to run Polars with better compatibility. Hint: If you are on an Apple ARM machine (e.g. M1) this is likely due to running Python under Rosetta. It is recommended to install a native version of Python that does not run under Rosetta x86-64 emulation. If you believe this warning to be a false positive, you can set the `POLARS_SKIP_CPU_CHECK` environment variable to bypass this check. """, RuntimeWarning, stacklevel=1, )