mps.py 1.87 KB
Newer Older
1 2 3 4 5
"""External function interface to MPS libraroes."""
from __future__ import absolute_import as _abs
from .. import api as _api
from .. import intrin as _intrin

Leyuan Wang committed
6
# pylint: disable=C0103,W0612
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

def matmul(lhs, rhs, transa=False, transb=False):
    """Create an extern op that compute matrix mult of A and rhs with CrhsLAS

    This function serves as an example on how to calle external libraries.

    Parameters
    ----------
    lhs : Tensor
        The left matrix operand
    rhs : Tensor
        The right matrix operand
    transa : bool
        Whether transpose lhs
    transb : bool
        Whether transpose rhs

    Returns
    -------
    C : Tensor
        The result tensor.
    """
Leyuan Wang committed
29 30 31 32 33 34
    m = lhs.shape[0] if transa is False else lhs.shape[1]
    n = rhs.shape[1] if transb is False else rhs.shape[0]
    if transa:
        m = b
    if transb:
        n = c
35
    return _api.extern(
Leyuan Wang committed
36
        (m, n), [lhs, rhs],
37 38 39
        lambda ins, outs: _intrin.call_packed(
            "tvm.contrib.mps.matmul", ins[0], ins[1], outs[0], transa, transb),
        name="C")
Leyuan Wang committed
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71

def conv2d(data, weight, pad='SAME', stride=1):
    """
    Create an extern op that compute data * weight and return result in output

    Parameters:
    ----------
    data: Tensor
        The input data, format NHWC
    weight: Tensor
        The conv weight, format output_feature * kH * kW * input_feature
    pad: str
        Padding method, 'SAME' or 'VALID'
    stride: int
        convolution stride

    Returns
    -------
    output: Tensor
        The result tensor
    """
    n, hi, wi, ci = data.shape
    co, kh, kw, ciw = weight.shape
    padding = 0 if pad == 'SAME' else 1
    ho = hi // stride
    wo = wi // stride

    return _api.extern(
        (n, ho, wo, co), [data, weight],
        lambda ins, outs: _intrin.call_packed(
            "tvm.contrib.mps.conv2d", ins[0], ins[1], outs[0], padding, stride),
        name="C")