96 lines
5.3 KiB
Python
96 lines
5.3 KiB
Python
#################################################################################################
|
|
#
|
|
# Copyright (c) 2023 - 2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
|
# SPDX-License-Identifier: BSD-3-Clause
|
|
#
|
|
# Redistribution and use in source and binary forms, with or without
|
|
# modification, are permitted provided that the following conditions are met:
|
|
#
|
|
# 1. Redistributions of source code must retain the above copyright notice, this
|
|
# list of conditions and the following disclaimer.
|
|
#
|
|
# 2. Redistributions in binary form must reproduce the above copyright notice,
|
|
# this list of conditions and the following disclaimer in the documentation
|
|
# and/or other materials provided with the distribution.
|
|
#
|
|
# 3. Neither the name of the copyright holder nor the names of its
|
|
# contributors may be used to endorse or promote products derived from
|
|
# this software without specific prior written permission.
|
|
#
|
|
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
|
|
# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
|
|
# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
|
|
# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
|
|
# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
|
|
# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
|
|
# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
#
|
|
#################################################################################################
|
|
|
|
"""
|
|
Low-level functionality tests for GEMM with S8 operands on SM90
|
|
"""
|
|
|
|
from functools import partial
|
|
|
|
import cutlass
|
|
import logging
|
|
import unittest
|
|
|
|
from cutlass.backend.test.utils import LayoutCombination, add_test_gemm
|
|
from cutlass.backend.utils.device import device_cc
|
|
|
|
|
|
cutlass.set_log_level(logging.WARNING)
|
|
cc = 90
|
|
|
|
@unittest.skipIf(device_cc() < cc, 'Device compute capability is insufficient for SM90 tests.')
|
|
class GemmS8Sm90(unittest.TestCase):
|
|
"""
|
|
Wrapper class to which tests will be added dynamically in __main__
|
|
"""
|
|
pass
|
|
|
|
|
|
add_test_specialized = partial(add_test_gemm, cls=GemmS8Sm90, element=cutlass.DataType.s8, compilation_modes=['nvcc'])
|
|
|
|
add_test_tensorop = partial(add_test_specialized, opclass=cutlass.OpcodeClass.TensorOp)
|
|
|
|
# Tests with 1x1x1 clusters
|
|
add_test_tensorop(layouts=LayoutCombination.TNN, alignments=[16, 16, 16], element_output=cutlass.DataType.s8,
|
|
element_accumulator=cutlass.DataType.s32, cluster_shape=[1, 1, 1], threadblock_shape=[128, 128, 128], stages=3)
|
|
add_test_tensorop(layouts=LayoutCombination.TNT, alignments=[16, 16, 16], element_output=cutlass.DataType.s8,
|
|
element_accumulator=cutlass.DataType.s32, cluster_shape=[1, 1, 1], threadblock_shape=[128, 128, 128], stages=None)
|
|
add_test_tensorop(layouts=LayoutCombination.TNT, alignments=[16, 16, 8], element_output=cutlass.DataType.s8,
|
|
element_accumulator=cutlass.DataType.s32, cluster_shape=[1, 1, 1], threadblock_shape=[128, 128, 128], stages=None)
|
|
add_test_tensorop(layouts=LayoutCombination.TNT, alignments=[16, 16, 16], element_output=cutlass.DataType.s8,
|
|
element_accumulator=cutlass.DataType.s32, cluster_shape=[1, 1, 1], threadblock_shape=[64, 128, 128], stages=None)
|
|
add_test_tensorop(layouts=LayoutCombination.TNT, alignments=[16, 16, 16], element_output=cutlass.DataType.s8,
|
|
element_accumulator=cutlass.DataType.s32, cluster_shape=[1, 1, 1], threadblock_shape=[128, 64, 32], stages=None)
|
|
add_test_tensorop(layouts=LayoutCombination.TNT, alignments=[ 4, 4, 16], element_output=cutlass.DataType.s8,
|
|
element_accumulator=cutlass.DataType.s32, cluster_shape=[1, 1, 1], threadblock_shape=[128, 128, 128], stages=None)
|
|
|
|
# Tests with different cluster shapes
|
|
add_test_tensorop(layouts=LayoutCombination.TNT, alignments=[16, 16, 16], element_output=cutlass.DataType.s8,
|
|
element_accumulator=cutlass.DataType.s32, cluster_shape=[2, 2, 1], threadblock_shape=[128, 128, 128], stages=None)
|
|
add_test_tensorop(layouts=LayoutCombination.TNT, alignments=[16, 16, 16], element_output=cutlass.DataType.s8,
|
|
element_accumulator=cutlass.DataType.s32, cluster_shape=[1, 4, 1], threadblock_shape=[128, 128, 128], stages=None)
|
|
|
|
# Tests with warp-specialized ping-pong schedule
|
|
add_test_tensorop(layouts=LayoutCombination.TNT, alignments=[16, 16, 16], element_output=cutlass.DataType.s8,
|
|
element_accumulator=cutlass.DataType.s32, cluster_shape=[2, 1, 1], threadblock_shape=[128, 128, 128], stages=None,
|
|
kernel_schedule=cutlass.KernelScheduleType.TmaWarpSpecializedPingpong,
|
|
epilogue_schedule=cutlass.EpilogueScheduleType.TmaWarpSpecialized)
|
|
|
|
# Tests for SIMT
|
|
add_test_simt = partial(add_test_specialized, opclass=cutlass.OpcodeClass.Simt)
|
|
add_test_simt(layouts=LayoutCombination.TNN, alignments=[1, 1, 1], element_output=cutlass.DataType.s8,
|
|
element_accumulator=cutlass.DataType.s32, cluster_shape=[1, 1, 1], threadblock_shape=[64, 32, 8], stages=2)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|