From b9ac818352efa713d15b101c073e1ade7e57e470 Mon Sep 17 00:00:00 2001
From: Yaman Umuroglu <maltanar@gmail.com>
Date: Tue, 26 Jul 2022 10:08:22 +0200
Subject: [PATCH] [Downsample] add 1D downsample support + conversion
---
.../custom_op/fpgadataflow/downsampler.py | 31 ++++++++++++++++---
.../fpgadataflow/convert_to_hls_layers.py | 23 ++++++++------
2 files changed, 40 insertions(+), 14 deletions(-)
diff --git a/src/finn/custom_op/fpgadataflow/downsampler.py b/src/finn/custom_op/fpgadataflow/downsampler.py
index b8645e04e..e9009e185 100644
--- a/src/finn/custom_op/fpgadataflow/downsampler.py
+++ b/src/finn/custom_op/fpgadataflow/downsampler.py
@@ -55,6 +55,10 @@ class DownSampler(HLSCustomOp):
"inputDataType": ("s", True, ""),
# Batch size
"numInputVectors": ("i", False, 1),
+ # 1D (True) or 2D (False) spatial data
+ "is1D": ("i", False, 0),
+ # for 1D only: (D, 1) (True) or (1, D) dims
+ "is1D_unitx": ("i", False, 1),
}
my_attrs.update(super().get_nodeattr_types())
return my_attrs
@@ -66,25 +70,43 @@ class DownSampler(HLSCustomOp):
return int(np.floor((idim - 1) / stride) + 1)
def get_exp_cycles(self):
+ is_1D = self.get_nodeattr("is1D")
idim = self.get_nodeattr("ImgDim")
+ idim_total = idim if is_1D else idim * idim
channels = self.get_nodeattr("NumChannels")
simd = self.get_nodeattr("SIMD")
batch_size = self.get_nodeattr("numInputVectors")
- exp_cycles = channels / simd * batch_size * idim * idim
+ exp_cycles = channels / simd * batch_size * idim_total
return int(exp_cycles)
def get_normal_input_shape(self):
+ is_1D = self.get_nodeattr("is1D")
+ is_1D_unitx = self.get_nodeattr("is1D_unitx")
idim = self.get_nodeattr("ImgDim")
num_ch = self.get_nodeattr("NumChannels")
batch = self.get_nodeattr("numInputVectors")
- ishape = (batch, idim, idim, num_ch)
+ if is_1D:
+ if is_1D_unitx:
+ ishape = (batch, idim, 1, num_ch)
+ else:
+ ishape = (batch, 1, idim, num_ch)
+ else:
+ ishape = (batch, idim, idim, num_ch)
return ishape
def get_normal_output_shape(self):
+ is_1D = self.get_nodeattr("is1D")
+ is_1D_unitx = self.get_nodeattr("is1D_unitx")
odim = self.get_downsampled_odim()
num_ch = self.get_nodeattr("NumChannels")
batch = self.get_nodeattr("numInputVectors")
- oshape = (batch, odim, odim, num_ch)
+ if is_1D:
+ if is_1D_unitx:
+ oshape = (batch, odim, 1, num_ch)
+ else:
+ oshape = (batch, 1, odim, num_ch)
+ else:
+ oshape = (batch, odim, odim, num_ch)
return oshape
def get_folded_input_shape(self):
@@ -204,8 +226,9 @@ class DownSampler(HLSCustomOp):
)
def docompute(self):
+ dim_var = "1D" if (self.get_nodeattr("is1D") == 1) else "2D"
self.code_gen_dict["$DOCOMPUTE$"] = [
- """ConvolutionInputGenerator_2D_kernel1<IFMChannels, Input_precision,
+ f"""ConvolutionInputGenerator_{dim_var}_kernel1<IFMChannels, Input_precision,
IFMDim, SIMD,Stride> (in0, out, numReps);"""
]
diff --git a/src/finn/transformation/fpgadataflow/convert_to_hls_layers.py b/src/finn/transformation/fpgadataflow/convert_to_hls_layers.py
index 429bc34ff..9059e023a 100644
--- a/src/finn/transformation/fpgadataflow/convert_to_hls_layers.py
+++ b/src/finn/transformation/fpgadataflow/convert_to_hls_layers.py
@@ -138,17 +138,18 @@ class InferConvInpGen(Transformation):
)
if (stride_h > 1 or stride_w > 1) and is_kernel_pointwise:
- assert is_square_image, (
- "%s : DownSampler currently only supports square input images."
- % n.name
- )
- assert is_equal_stride, (
- """%s : DownSampler currently only supports equal stride value
- along different axes."""
- % n.name
+ downsample_1D = (ifm_dim_h == 1) or (ifm_dim_w == 1)
+ is1D_unitx = ifm_dim_w == 1
+ downsample_2D = (
+ (not downsample_1D) and is_square_image and is_equal_stride
)
- ConvInpGen_idim = ConvInpGen_idim_h
- stride = stride_h
+ if not (downsample_1D or downsample_2D):
+ warnings.warn(
+ f"Couldn't infer Downsample from {n.name}, check config."
+ )
+ continue
+ ConvInpGen_idim = max(ConvInpGen_idim_h, ConvInpGen_idim_w)
+ stride = max(stride_h, stride_w)
# create DownSampler node
ConvInpGen_node = helper.make_node(
"DownSampler",
@@ -162,6 +163,8 @@ class InferConvInpGen(Transformation):
Stride=stride,
inputDataType=dt.name,
name="DownSampler_" + n.name,
+ is1D=downsample_1D,
+ is1D_unitx=is1D_unitx,
)
graph.node.insert(ConvInpGen_node_idx, ConvInpGen_node)
else:
--
GitLab