From bc78cba1f4bb379e60bfb2668aa3e9d9b3739977 Mon Sep 17 00:00:00 2001
From: Hendrik Borras <hendrikborras@web.de>
Date: Fri, 15 Oct 2021 16:18:12 +0100
Subject: [PATCH] Fixed test_brevitas_avg_pool_export to properly integrate the
sign.
---
.../brevitas/test_brevitas_avg_pool_export.py | 24 ++++++++++++++-----
1 file changed, 18 insertions(+), 6 deletions(-)
diff --git a/tests/brevitas/test_brevitas_avg_pool_export.py b/tests/brevitas/test_brevitas_avg_pool_export.py
index fc6d50f8c..e202b6328 100644
--- a/tests/brevitas/test_brevitas_avg_pool_export.py
+++ b/tests/brevitas/test_brevitas_avg_pool_export.py
@@ -32,7 +32,7 @@ import os
import torch
from brevitas.export import FINNManager
from brevitas.export.onnx.generic.manager import BrevitasONNXManager
-from brevitas.nn import QuantAvgPool2d, QuantIdentity
+from brevitas.nn import QuantAvgPool2d, QuantIdentity, QuantReLU
from qonnx.util.cleanup import cleanup as qonnx_cleanup
from torch import nn
@@ -44,17 +44,17 @@ from finn.transformation.infer_shapes import InferShapes
from finn.transformation.qonnx.convert_qonnx_to_finn import ConvertQONNXtoFINN
from finn.util.basic import gen_finn_dt_tensor
-export_onnx_path = "test_brevitas_avg_pool_export.onnx"
+base_export_onnx_path = "test_brevitas_avg_pool_export.onnx"
+@pytest.mark.parametrize("QONNX_export", [False, True])
+@pytest.mark.parametrize("signed", [True, False])
@pytest.mark.parametrize("kernel_size", [2, 3])
@pytest.mark.parametrize("stride", [1, 2])
-@pytest.mark.parametrize("signed", [True, False])
@pytest.mark.parametrize("bit_width", [2, 4])
@pytest.mark.parametrize("input_bit_width", [4, 8, 16])
@pytest.mark.parametrize("channels", [2, 4])
@pytest.mark.parametrize("idim", [7, 8])
-@pytest.mark.parametrize("QONNX_export", [False, True])
def test_brevitas_avg_pool_export(
kernel_size,
stride,
@@ -65,12 +65,24 @@ def test_brevitas_avg_pool_export(
idim,
QONNX_export,
):
-
+ export_onnx_path = base_export_onnx_path.replace(
+ ".onnx", f"test_QONNX-{QONNX_export}.onnx"
+ )
# To do a proper static export Brevitas requires a quantized input tensor.
# For the BrevitasONNXManager these requirements are even more stringent,
# such that in-model quantization and de-quantization at the end are required.
+ if signed:
+ # Signed is only supported by the QuantIdentity in FINN
+ act_layer = QuantIdentity(
+ signed=signed, bit_width=input_bit_width, return_quant_tensor=True
+ )
+ else:
+ # Unsigned is only supported by the QuantReLU in FINN
+ act_layer = QuantReLU(
+ signed=signed, bit_width=input_bit_width, return_quant_tensor=True
+ )
quant_avgpool = nn.Sequential(
- QuantIdentity(bit_width=input_bit_width, return_quant_tensor=True),
+ act_layer,
QuantAvgPool2d(
kernel_size=kernel_size,
stride=stride,
--
GitLab