forked from tinygrad/tinygrad
wozeparrotandGitHub
2fd2fb6380
* feat: int8 support * feat: uint8 support * feat: int8 tests * fix: fix uint8 on clang * feat: test casting between int8/uint8/float16/float32 * clean: way cleaner dtype tests * feat: preprocess_imagenet using the correct dtype * feat: add test for overflow between uint8 and int8
52 lines
2.2 KiB
Python
52 lines
2.2 KiB
Python
import ctypes
|
|
import numpy as np
|
|
from typing import TypeVar, Type, Any
|
|
from tinygrad.helpers import DType, dtypes, prod, GlobalCounters
|
|
|
|
_T = TypeVar("_T")
|
|
class RawBuffer: # pylint: disable=abstract-method
|
|
def __init__(self, size:int, dtype:DType, buf:Any=None):
|
|
self.size: int = size
|
|
self.dtype: DType = dtype
|
|
self._buf = buf
|
|
self._memsz: int = size*dtype.itemsize
|
|
GlobalCounters.mem_used += self._memsz
|
|
def __del__(self): # NOTE: if it fails on init (bad dtype), it won't have a _memsz
|
|
if hasattr(self, '_memsz'): GlobalCounters.mem_used -= self._memsz
|
|
def __repr__(self): return f"buffer<{self.size}, {self.dtype}>"
|
|
|
|
# NOTE: this interface allows for 0 copy
|
|
@classmethod
|
|
def fromCPU(cls:Type[_T], x:np.ndarray) -> _T: raise NotImplementedError("must be implemented")
|
|
def toCPU(self) -> np.ndarray: raise NotImplementedError("must be implemented")
|
|
|
|
class RawBufferCopyIn(RawBuffer):
|
|
def _copyin(self, x:np.ndarray) -> None: raise NotImplementedError("must be implemented")
|
|
|
|
@classmethod
|
|
def fromCPU(cls, x:np.ndarray, **kwargs):
|
|
ret = cls(prod(x.shape), dtypes.from_np(x.dtype), **kwargs)
|
|
ret._copyin(x)
|
|
return ret
|
|
|
|
class RawBufferMapped(RawBufferCopyIn):
|
|
def _buffer(self) -> memoryview: raise NotImplementedError("must be implemented")
|
|
def toCPU(self) -> np.ndarray: return np.frombuffer(self._buffer(), dtype=self.dtype.np)
|
|
def _copyin(self, x:np.ndarray) -> None: np.copyto(self.toCPU(), x.reshape(-1))
|
|
|
|
# this one is simple enough that i moved it out of the runtimes
|
|
class RawMallocBuffer(RawBufferMapped):
|
|
def __init__(self, size, dtype: DType): super().__init__(size, dtype, ({dtypes.float32: ctypes.c_float, dtypes.float16: ctypes.c_int16, dtypes.int8: ctypes.c_int8, dtypes.uint8: ctypes.c_uint8}[dtype] * size)())
|
|
def _buffer(self): return memoryview(self._buf)
|
|
|
|
class RawBufferCopyInOut(RawBufferCopyIn):
|
|
def _copyout(self, x:np.ndarray) -> None: raise NotImplementedError("must be implemented")
|
|
|
|
def toCPU(self) -> np.ndarray:
|
|
x: np.ndarray = np.empty(self.size, dtype=self.dtype.np)
|
|
self._copyout(x)
|
|
return x
|
|
|
|
class RawConst(RawBuffer): # pylint: disable=abstract-method
|
|
def __repr__(self): return f"const<{self._buf}, {self.dtype}>"
|