wehub-resource-sync
102 lines
3.3 KiB
Python
102 lines
3.3 KiB
Python
# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# 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.
|
|
import unittest
|
|
|
|
import numpy as np
|
|
from op_test import get_cuda_version, is_custom_device
|
|
|
|
import paddle
|
|
from paddle.base import core
|
|
|
|
# define the e4m3/e5m2 constants
|
|
E4M3_MAX_POS = 448.0
|
|
E5M2_MAX_POS = 57344.0
|
|
|
|
is_sm_supported = (core.is_compiled_with_cuda() or is_custom_device()) and (
|
|
(
|
|
paddle.device.cuda.get_device_capability()[0] == 8
|
|
and paddle.device.cuda.get_device_capability()[1] == 9
|
|
)
|
|
or (paddle.device.cuda.get_device_capability()[0] >= 9)
|
|
)
|
|
|
|
|
|
def check_fp8_support() -> bool:
|
|
"""Return if fp8 support is available"""
|
|
gpu_arch = (
|
|
paddle.device.cuda.get_device_capability()[0] * 10
|
|
+ paddle.device.cuda.get_device_capability()[1]
|
|
)
|
|
if gpu_arch >= 90: # hopper and above
|
|
return True
|
|
# Device compute capability 8.9 or higher required for FP8 execution.
|
|
if gpu_arch < 89: # pre-ada
|
|
return False
|
|
if get_cuda_version() < 12010:
|
|
return False
|
|
return True
|
|
|
|
|
|
def _to_fp8_saturated(x: paddle.Tensor, float8_dtype) -> paddle.Tensor:
|
|
# The default behavior in Paddle for casting to `float8_e4m3fn`
|
|
# and `e5m2` is to not saturate. So we saturate here manually.
|
|
if float8_dtype == paddle.float8_e4m3fn:
|
|
x = x.clip(min=-1 * E4M3_MAX_POS, max=E4M3_MAX_POS)
|
|
else:
|
|
x = x.clip(min=-1 * E5M2_MAX_POS, max=E5M2_MAX_POS)
|
|
return x.to(float8_dtype)
|
|
|
|
|
|
@unittest.skipIf(
|
|
not (core.is_compiled_with_cuda() or is_custom_device())
|
|
or not check_fp8_support(),
|
|
"Fp8 matmul requires CUDA >= 12.1 on Ada arch or hopper arch",
|
|
)
|
|
class TestMatmulFp8(unittest.TestCase):
|
|
def config(self):
|
|
self.dtype = 'float8_e4m3fn'
|
|
self.rtol = 1e-5
|
|
self.atol = 1e-5
|
|
self.x_shape = (64, 64)
|
|
self.y_shape = (64, 64)
|
|
|
|
def setUp(self):
|
|
self.config()
|
|
paddle.seed(2024)
|
|
self.input_a = paddle.rand(self.x_shape)
|
|
self.input_a_fp8 = _to_fp8_saturated(self.input_a, paddle.float8_e4m3fn)
|
|
self.input_b = paddle.rand(self.y_shape)
|
|
self.input_b_fp8 = _to_fp8_saturated(self.input_b, paddle.float8_e4m3fn)
|
|
|
|
def get_reference_out(self):
|
|
self.input_a_fp16 = self.input_a_fp8.astype("float16")
|
|
self.input_b_fp16 = self.input_b_fp8.astype("float16")
|
|
out = paddle.matmul(self.input_a_fp16, self.input_b_fp16)
|
|
return out
|
|
|
|
def get_op_out(self):
|
|
out = paddle.matmul(self.input_a_fp8, self.input_b_fp8)
|
|
return out
|
|
|
|
def test_matmul_fp8(self):
|
|
out_real = self.get_op_out()
|
|
out_expect = self.get_reference_out()
|
|
np.testing.assert_allclose(
|
|
out_real, out_expect, rtol=self.rtol, atol=self.atol
|
|
)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|