# Copyright (c) Meta Platforms, Inc. and affiliates.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
"""
GEMM Specialization for A[RowMajor], B[RowMajor], C[RowMajor]
This is special in template based gemm solution
This is used for `torch.nn.functional.linear`
When use for `linear`, need set A->Data, B->Weight
Special kernel for small K and N
K <= 16, N <= 8
A: [M, K] A can be ND with the first N - 1 dimensions as batch dimensions
B: [K, N]
C: [M, N]
"""
from aitemplate.compiler.base import IntImm, Tensor
from aitemplate.compiler.ops.gemm_universal import gemm_common as common
# pylint: disable=C0103,W0223,W0221,W0613
[docs]class gemm_rrr_small_nk(common.gemm):
"""Special gemm kernel for small K and N (K <= 8, N <= 8)
A: [M, K]
B: [K, N]
C: [M, N]
"""
def __init__(self):
super().__init__()
self._attrs["op"] = "gemm_rrr_small_nk"
self._attrs["f_ab_alignment"] = True
self._attrs["has_profiler"] = False
@staticmethod
def is_valid_shape(a: Tensor, b: Tensor):
valid = len(a.shape()) >= 2 and len(b.shape()) == 2
for idx in range(2):
dim = b.shape()[idx]
if not isinstance(dim, IntImm):
return False
if idx == 0:
# check for K <= 16
valid &= dim.value() <= 16
else:
# check for N <= 8
valid &= dim.value() <= 8
return valid
def _infer_shapes(self, a: Tensor, b: Tensor):
assert (
a.shape()[-1] == b.shape()[0]
), f"gemm_rrr operand A and B should have the same K dim! A shape: {a.shape()}, B shape: {b.shape()}"
assert gemm_rrr_small_nk.is_valid_shape(
a, b
), "shape (tensor a:{}, tensor b:{}) not valid for gemm_rrr_small_nk".format(
a.shape(), b.shape()
)
return a._attrs["shape"][:-1] + [b._attrs["shape"][1]]
[docs] def gen_profiler(
self, workdir: str = None, dynamic_profiling_strategy=None
) -> None:
"""This kernel does not require profiling"""
return
def _extract_dims(self, for_profiling=False):
A_rank = self._attrs["inputs"][0]._rank()
# (M, K) * (K, N) = (M, N)
return {
"M": [
common.DimInfo(
common.Source.INPUT, tensor_idx=0, dim_idx=list(range(A_rank - 1))
),
common.DimInfo(
common.Source.OUTPUT, tensor_idx=0, dim_idx=list(range(A_rank - 1))
),
],
"N": [
common.DimInfo(common.Source.INPUT, tensor_idx=1, dim_idx=1),
common.DimInfo(common.Source.OUTPUT, tensor_idx=0, dim_idx=1),
],
"K": [
common.DimInfo(common.Source.INPUT, tensor_idx=0, dim_idx=A_rank - 1),
common.DimInfo(common.Source.INPUT, tensor_idx=1, dim_idx=0),
],
}
def __call__(self, a: Tensor, b: Tensor) -> Tensor:
self._attrs["inputs"] = [a, b]
self._set_depth()
output_shape = self._infer_shapes(a, b)
output = Tensor(output_shape, src_ops={self}, dtype=a.dtype())
self._attrs["outputs"] = [output]
# self._attrs["output_accessors"] = [TensorAccessor(output)]
return output