Merge branch 'main' into sep_to_dw_point

Author: jmitrevs

.pre-commit-config.yaml

@@ -30,7 +30,7 @@ repos:
     args: ["--profile", "black", --line-length=125]
 
 - repo: https://github.com/asottile/pyupgrade
-  rev: v3.15.2
+  rev: v3.16.0
   hooks:
   - id: pyupgrade
     args: ["--py36-plus"]
@@ -41,7 +41,7 @@ repos:
   - id: setup-cfg-fmt
 
 - repo: https://github.com/pycqa/flake8
-  rev: 7.0.0
+  rev: 7.1.0
   hooks:
   - id: flake8
     exclude: docs/conf.py

Jenkinsfile

@@ -1,7 +1,7 @@
 pipeline {
   agent {
     docker {
-      image 'vivado-el7:3'
+      image 'vivado-alma9:1'
       args  '-v /data/Xilinx:/data/Xilinx'
     }
   }
@@ -14,8 +14,9 @@ pipeline {
       steps {
         dir(path: 'test') {
           sh '''#!/bin/bash --login
-              conda activate hls4ml-py38
-              pip install tensorflow pyparsing
+              conda activate hls4ml-py310
+              conda install -y jupyterhub pydot graphviz pytest pytest-cov
+              pip install pytest-randomly jupyter onnx>=1.4.0 matplotlib pandas seaborn pydigitalwavetools==1.1 pyyaml tensorflow==2.14 qonnx torch git+https://github.com/google/qkeras.git pyparsing
               pip install -U ../ --user
               ./convert-keras-models.sh -x -f keras-models.txt
               pip uninstall hls4ml -y'''

hls4ml/backends/quartus/quartus_backend.py

@@ -274,7 +274,7 @@ def init_conv1d(self, layer):
         # - combination - at compile-time, the decision between Winograd and im2col is made
         # - im2col - specifically use im2col
         # - Winograd - use Winograd, if possible
-        layer.set_attr('implementation', layer.model.config.get_layer_config_value(layer, 'Implementation', 'combination'))
+        layer.set_attr('implementation', layer.model.config.get_layer_config_value(layer, 'Implementation', 'im2col'))
 
         layer.set_attr(
             'n_partitions', 1
@@ -305,7 +305,7 @@ def init_conv2d(self, layer):
         # - combination - at compile-time, the decision between Winograd and im2col is made
         # - im2col - specifically use im2col
         # - Winograd - use Winograd, if possible
-        layer.set_attr('implementation', layer.model.config.get_layer_config_value(layer, 'Implementation', 'combination'))
+        layer.set_attr('implementation', layer.model.config.get_layer_config_value(layer, 'Implementation', 'im2col'))
 
         layer.set_attr(
             'n_partitions', 1

hls4ml/backends/vivado/vivado_backend.py

@@ -19,8 +19,6 @@
     Embedding,
     GarNet,
     GarNetStack,
-    GlobalPooling1D,
-    GlobalPooling2D,
     Layer,
     Pooling1D,
     Pooling2D,
@@ -32,7 +30,6 @@
 from hls4ml.model.optimizer import get_backend_passes, layer_optimizer
 from hls4ml.model.types import FixedPrecisionType, IntegerPrecisionType, NamedType, PackedType
 from hls4ml.report import parse_vivado_report
-from hls4ml.utils.fixed_point_utils import ceil_log2
 
 
 class VivadoBackend(FPGABackend):
@@ -433,37 +430,14 @@ def init_depconv2d(self, layer):
 
         layer.set_attr('implementation', layer.model.config.get_conv_implementation(layer).lower())
 
-    def _set_pooling_accum_t(self, layer, pool_size):
-        extra_bits = ceil_log2(pool_size)
-        accum_t = layer.get_attr('accum_t')
-        accum_t.precision.width += extra_bits * 2
-        if isinstance(accum_t.precision, FixedPrecisionType):
-            accum_t.precision.integer += extra_bits
-
     @layer_optimizer(Pooling1D)
     def init_pooling1d(self, layer):
-        pool_size = layer.get_attr('pool_width')
-        self._set_pooling_accum_t(layer, pool_size)
-
         layer.set_attr('implementation', layer.model.config.get_conv_implementation(layer).lower())
 
     @layer_optimizer(Pooling2D)
     def init_pooling2d(self, layer):
-        pool_size = layer.get_attr('pool_height') * layer.get_attr('pool_width')
-        self._set_pooling_accum_t(layer, pool_size)
-
         layer.set_attr('implementation', layer.model.config.get_conv_implementation(layer).lower())
 
-    @layer_optimizer(GlobalPooling1D)
-    def init_global_pooling1d(self, layer):
-        pool_size = layer.get_attr('n_in')
-        self._set_pooling_accum_t(layer, pool_size)
-
-    @layer_optimizer(GlobalPooling2D)
-    def init_global_pooling2d(self, layer):
-        pool_size = layer.get_attr('in_height') * layer.get_attr('in_width')
-        self._set_pooling_accum_t(layer, pool_size)
-
     @layer_optimizer(Softmax)
     def init_softmax(self, layer):
         if layer.model.config.get_config_value('IOType') == 'io_parallel':

hls4ml/converters/keras/qkeras.py

@@ -9,7 +9,9 @@
 
 
 def get_quantizer_from_config(keras_layer, quantizer_var):
-    quantizer_config = keras_layer['config'][f'{quantizer_var}_quantizer']
+    quantizer_config = keras_layer['config'].get(f'{quantizer_var}_quantizer', None)
+    if quantizer_config is None:
+        return None  # No quantizer specified in the layer
     if keras_layer['class_name'] == 'QBatchNormalization':
         return QKerasQuantizer(quantizer_config)
     elif 'binary' in quantizer_config['class_name']:
@@ -25,10 +27,7 @@ def parse_qdense_layer(keras_layer, input_names, input_shapes, data_reader):
     layer, output_shape = parse_dense_layer(keras_layer, input_names, input_shapes, data_reader)
 
     layer['weight_quantizer'] = get_quantizer_from_config(keras_layer, 'kernel')
-    if keras_layer['config']['bias_quantizer'] is not None:
-        layer['bias_quantizer'] = get_quantizer_from_config(keras_layer, 'bias')
-    else:
-        layer['bias_quantizer'] = None
+    layer['bias_quantizer'] = get_quantizer_from_config(keras_layer, 'bias')
 
     return layer, output_shape
 
@@ -43,10 +42,7 @@ def parse_qconv_layer(keras_layer, input_names, input_shapes, data_reader):
         layer, output_shape = parse_conv2d_layer(keras_layer, input_names, input_shapes, data_reader)
 
     layer['weight_quantizer'] = get_quantizer_from_config(keras_layer, 'kernel')
-    if keras_layer['config']['bias_quantizer'] is not None:
-        layer['bias_quantizer'] = get_quantizer_from_config(keras_layer, 'bias')
-    else:
-        layer['bias_quantizer'] = None
+    layer['bias_quantizer'] = get_quantizer_from_config(keras_layer, 'bias')
 
     return layer, output_shape
 
@@ -56,11 +52,7 @@ def parse_qdepthwiseqconv_layer(keras_layer, input_names, input_shapes, data_rea
     layer, output_shape = parse_conv2d_layer(keras_layer, input_names, input_shapes, data_reader)
 
     layer['depthwise_quantizer'] = get_quantizer_from_config(keras_layer, 'depthwise')
-
-    if keras_layer['config']['bias_quantizer'] is not None:
-        layer['bias_quantizer'] = get_quantizer_from_config(keras_layer, 'bias')
-    else:
-        layer['bias_quantizer'] = None
+    layer['bias_quantizer'] = get_quantizer_from_config(keras_layer, 'bias')
 
     return layer, output_shape
 
@@ -76,11 +68,7 @@ def parse_qsepconv_layer(keras_layer, input_names, input_shapes, data_reader):
 
     layer['depthwise_quantizer'] = get_quantizer_from_config(keras_layer, 'depthwise')
     layer['pointwise_quantizer'] = get_quantizer_from_config(keras_layer, 'pointwise')
-
-    if keras_layer['config']['bias_quantizer'] is not None:
-        layer['bias_quantizer'] = get_quantizer_from_config(keras_layer, 'bias')
-    else:
-        layer['bias_quantizer'] = None
+    layer['bias_quantizer'] = get_quantizer_from_config(keras_layer, 'bias')
 
     return layer, output_shape
 
@@ -93,11 +81,7 @@ def parse_qrnn_layer(keras_layer, input_names, input_shapes, data_reader):
 
     layer['weight_quantizer'] = get_quantizer_from_config(keras_layer, 'kernel')
     layer['recurrent_quantizer'] = get_quantizer_from_config(keras_layer, 'recurrent')
-
-    if keras_layer['config']['bias_quantizer'] is not None:
-        layer['bias_quantizer'] = get_quantizer_from_config(keras_layer, 'bias')
-    else:
-        layer['bias_quantizer'] = None
+    layer['bias_quantizer'] = get_quantizer_from_config(keras_layer, 'bias')
 
     return layer, output_shape
 

hls4ml/templates/quartus/firmware/nnet_utils/nnet_conv1d_resource.h

@@ -3,6 +3,7 @@
 
 #include "nnet_common.h"
 #include "nnet_dense.h"
+#include <cstdint>
 
 namespace nnet {
 

hls4ml/templates/quartus/firmware/nnet_utils/nnet_conv2d_resource.h

@@ -4,6 +4,7 @@
 #include "nnet_common.h"
 #include "nnet_dense.h"
 #include "nnet_helpers.h"
+#include <cstdint>
 
 namespace nnet {