Merge branch 'dev' of https://github.com/Xilinx/finn into feature/remote_pynq_execution

AUTHORS.rst

@@ -2,4 +2,6 @@
 Contributors
 ============
 
-* Yaman Umuroglu <yamanu@xilinx.com>
+* Yaman Umuroglu
+* Jakoba Petri-Koenig
+* Andrea Rigoni

run-docker.sh

@@ -17,6 +17,9 @@ DOCKER_TAG="finn_${DOCKER_UNAME}"
 # uncomment to run multiple instances with different names
 # DOCKER_INST_NAME="finn_${DOCKER_UNAME}_${DOCKER_RND}"
 DOCKER_INST_NAME="finn_${DOCKER_UNAME}"
+# ensure Docker tag and inst. name are all lowercase
+DOCKER_TAG=$(echo "$DOCKER_TAG" | tr '[:upper:]' '[:lower:]')
+DOCKER_INST_NAME=$(echo "$DOCKER_INST_NAME" | tr '[:upper:]' '[:lower:]')
 # the settings below will be taken from environment variables if available,
 # otherwise the defaults below will be used
 : ${JUPYTER_PORT=8888}

tests/fpgadataflow/test_code_gen_trafo.py

@@ -5,7 +5,6 @@ from onnx import TensorProto, helper
 import finn.util.basic as util
 from finn.core.datatype import DataType
 from finn.core.modelwrapper import ModelWrapper
-from finn.transformation.fpgadataflow.cleanup import CleanUp
 from finn.transformation.fpgadataflow.codegen_npysim import CodeGen_npysim
 
 
@@ -66,4 +65,3 @@ def test_code_gen_trafo():
             op type {} is empty!""".format(
             node.op_type
         )
-    model = model.transform(CleanUp())

tests/fpgadataflow/test_compilation_trafo.py

@@ -5,7 +5,6 @@ from onnx import TensorProto, helper
 import finn.util.basic as util
 from finn.core.datatype import DataType
 from finn.core.modelwrapper import ModelWrapper
-from finn.transformation.fpgadataflow.cleanup import CleanUp
 from finn.transformation.fpgadataflow.codegen_npysim import CodeGen_npysim
 from finn.transformation.fpgadataflow.compile import Compile
 
@@ -63,4 +62,3 @@ def test_compilation_trafo():
             op type {} does not exist!""".format(
             node.op_type
         )
-    model = model.transform(CleanUp())

tests/fpgadataflow/test_fpgadataflow_convinputgenerator.py

@@ -6,7 +6,6 @@ from onnx import TensorProto, helper
 import finn.core.onnx_exec as oxe
 from finn.core.datatype import DataType
 from finn.core.modelwrapper import ModelWrapper
-from finn.transformation.fpgadataflow.cleanup import CleanUp
 from finn.transformation.fpgadataflow.codegen_ipgen import CodeGen_ipgen
 from finn.transformation.fpgadataflow.codegen_npysim import CodeGen_npysim
 from finn.transformation.fpgadataflow.compile import Compile
@@ -167,4 +166,3 @@ def test_fpgadataflow_slidingwindow(idt, k, ifm_dim, ifm_ch, stride):
     model = model.transform(HLSSynth_IPGen())
     y_produced = oxe.execute_onnx(model, input_dict)["outp"]
     assert (y_produced == y_expected).all(), "rtlsim failed"
-    model = model.transform(CleanUp())

tests/fpgadataflow/test_fpgadataflow_fclayer.py

@@ -9,7 +9,6 @@ from finn.analysis.fpgadataflow.hls_synth_res_estimation import hls_synth_res_es
 from finn.core.datatype import DataType
 from finn.core.modelwrapper import ModelWrapper
 from finn.custom_op.multithreshold import multithreshold
-from finn.transformation.fpgadataflow.cleanup import CleanUp
 from finn.transformation.fpgadataflow.codegen_ipgen import CodeGen_ipgen
 from finn.transformation.fpgadataflow.codegen_npysim import CodeGen_npysim
 from finn.transformation.fpgadataflow.compile import Compile
@@ -175,7 +174,6 @@ def test_fpgadataflow_fclayer_npysim(idt, wdt, act, nf, sf, mw, mh):
     # execute model
     y_produced = oxe.execute_onnx(model, input_dict)["outp"]
     assert (y_produced.reshape(y_expected.shape) == y_expected).all(), "npysim failed"
-    model = model.transform(CleanUp())
 
 
 # activation: None or DataType
@@ -257,5 +255,3 @@ def test_fpgadataflow_fclayer_rtlsim(idt, wdt, act, nf, sf, mw, mh):
 
     hls_synt_res_est = model.analysis(hls_synth_res_estimation)
     assert "StreamingFCLayer_Batch_0" in hls_synt_res_est
-
-    model = model.transform(CleanUp())

tests/fpgadataflow/test_layer_streaming_maxpool_batch.py

@@ -4,7 +4,6 @@ from onnx import TensorProto, helper
 import finn.core.onnx_exec as oxe
 from finn.core.datatype import DataType
 from finn.core.modelwrapper import ModelWrapper
-from finn.transformation.fpgadataflow.cleanup import CleanUp
 from finn.transformation.fpgadataflow.codegen_npysim import CodeGen_npysim
 from finn.transformation.fpgadataflow.compile import Compile
 from finn.transformation.fpgadataflow.set_exec_mode import SetExecMode
@@ -111,7 +110,6 @@ def test_layer_streaming_maxpool_batch():
         ],
         dtype=np.float32,
     ).reshape(2, 2, 4, 4)
-    print(input_tensor)
 
     model = model.transform(SetExecMode("npysim"))
     model = model.transform(CodeGen_npysim())
@@ -119,5 +117,3 @@ def test_layer_streaming_maxpool_batch():
 
     input_dict = {"in": input_tensor}
     output_dict = oxe.execute_onnx(model, input_dict)
-    print(output_dict)
-    model = model.transform(CleanUp())

tests/transformation/test_batchnorm_to_affine.py

@@ -30,16 +30,18 @@ def test_batchnorm_to_affine_lfc_w1a1():
     os.remove(export_onnx_path)
 
 
-def test_batchnorm_to_affine_cnv_w1a1():
-    lfc = get_test_model_trained("CNV", 1, 1)
-    bo.export_finn_onnx(lfc, (1, 3, 32, 32), export_onnx_path)
-    model = ModelWrapper(export_onnx_path)
-    model = model.transform(InferShapes())
-    model = model.transform(FoldConstants())
-    # TODO shape inference failing on transformed model below -- needs debug
-    new_model = model.transform(BatchNormToAffine())
-    # check that there are no BN nodes left
-    # TODO replace this with execution test
-    op_types = list(map(lambda x: x.op_type, new_model.graph.node))
-    assert "BatchNormalization" not in op_types
-    os.remove(export_onnx_path)
+# cnv batchnorm to affine not yet supported
+
+# def test_batchnorm_to_affine_cnv_w1a1():
+#    lfc = get_test_model_trained("CNV", 1, 1)
+#    bo.export_finn_onnx(lfc, (1, 3, 32, 32), export_onnx_path)
+#    model = ModelWrapper(export_onnx_path)
+#    model = model.transform(InferShapes())
+#    model = model.transform(FoldConstants())
+#    # TODO shape inference failing on transformed model below -- needs debug
+#    new_model = model.transform(BatchNormToAffine())
+#    # check that there are no BN nodes left
+#    # TODO replace this with execution test
+#    op_types = list(map(lambda x: x.op_type, new_model.graph.node))
+#    assert "BatchNormalization" not in op_types
+#    os.remove(export_onnx_path)