diff --git a/llvm/projects/keras/frontend/approxhpvm_translator.py b/llvm/projects/keras/frontend/approxhpvm_translator.py
index 21e109d9ed896a2f9e6e73433d227da20f84895b..77889280bf8cf3cefcd604a820aef9ff78d5ba73 100644
--- a/llvm/projects/keras/frontend/approxhpvm_translator.py
+++ b/llvm/projects/keras/frontend/approxhpvm_translator.py
@@ -404,7 +404,7 @@ class TensorRtTranslator:
- elif layer_type == "Dense":
+ if layer_type == "Dense":
input_var_name = self.getSingleInputName(cur_node)
weights = cur_node.weights
@@ -416,7 +416,7 @@ class TensorRtTranslator:
self.program_str += inst_str
- elif self.hasBiasAdd(cur_node):
+ if self.hasBiasAdd(cur_node):
out_var_name2 = self.getVariableName(cur_node)
inst_str = "void* " + out_var_name2 + " = "
@@ -430,14 +430,14 @@ class TensorRtTranslator:
out_var_name1 = out_var_name2
- elif layer_type == "Activation":
+ if layer_type == "Activation":
input_var_name = self.getSingleInputName(cur_node)
inst_str = genActivationCallStr(input_var_name, out_var_name1, cur_node.activation_type)
self.program_str += inst_str
- elif self.hasActivation(cur_node) and layer_type != "Activation":
+ if self.hasActivation(cur_node) and layer_type != "Activation":
activation_type = cur_node.activation_type
out_var_name3 = self.getVariableName(cur_node)
@@ -449,7 +449,7 @@ class TensorRtTranslator:
sys.exit(0)
- elif layer_type == "BatchNormalization":
+ if layer_type == "BatchNormalization":
input_var_name = self.getSingleInputName(cur_node)
inst_str = "void* " + out_var_name1 + " = "
@@ -464,7 +464,7 @@ class TensorRtTranslator:
self.program_str += inst_str
- elif layer_type == "Add":
+ if layer_type == "Add":
input_vars = self.getMultipleInputNames(cur_node)
inst_str = "void* " + out_var_name1 + " = "
@@ -472,7 +472,7 @@ class TensorRtTranslator:
self.program_str += inst_str
- elif layer_type == "MaxPooling2D" or layer_type == "AveragePooling2D":
+ if layer_type == "MaxPooling2D" or layer_type == "AveragePooling2D":
input_var_name = self.getSingleInputName(cur_node)
pool_size = cur_node.pool_size
@@ -491,8 +491,6 @@ class TensorRtTranslator:
inst_str += "," + str(padding) + "," + str(padding) + "," + str(strides[0]) + "," + str(strides[1])
inst_str += "); \n"
self.program_str += inst_str
- #else:
- # print ("ERROR: Operator = ", layer_type, " is NOT currently supported")
@@ -771,7 +769,8 @@ class TensorRtTranslator:
file_path_str += "labels.bin\"); \n"
self.weight_str += file_path_str
- self.input_str += "uint8_t* labels = readLabels("
+ #self.input_str += "uint8_t* labels = readLabels("
+ self.input_str += "uint32_t* labels = readLabels2("
self.input_str += file_path + ".c_str()," + str(test_labels.shape[0]) + "); \n"
@@ -805,11 +804,13 @@ class TensorRtTranslator:
def endBatchLoop(self):
end_loop_str = ""
- end_loop_str += "\nuint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end); \n"
+ #end_loop_str += "\nuint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end); \n"
+ end_loop_str += "\nuint32_t* labels = readLabelsBatch2(labels_path.c_str(),start,end); \n"
last_node = self.dfg.last_node
output_var = self.output_map[last_node.layer_name]
- accuracy_call = "\nfloat accuracy = computeAccuracy2(labels, batch_size, " + output_var + "); \n"
+ #accuracy_call = "\nfloat accuracy = computeAccuracy2(labels, batch_size, " + output_var + "); \n"
+ accuracy_call = "\nfloat accuracy = computeAccuracy3(labels, batch_size, " + output_var + "); \n"
end_loop_str += accuracy_call
#end_loop_str += "float accuracy = computeAccuracy2(labels, batch_size, var_60); "
diff --git a/llvm/projects/keras/frontend/hpvm_dfg_translator.py b/llvm/projects/keras/frontend/hpvm_dfg_translator.py
index 066d5d23b90fae0048036b69bcdd7556adec9c92..7467f316e98e36f6edc8924def93a216712d7ab3 100644
--- a/llvm/projects/keras/frontend/hpvm_dfg_translator.py
+++ b/llvm/projects/keras/frontend/hpvm_dfg_translator.py
@@ -637,7 +637,8 @@ class HPVMTranslator:
main_func_str += "hpvm_request_tensor(result, 0); \n\n"
main_func_str += "__visc__cleanup(); \n "
- main_func_str += "computeAccuracy2(labels, " + str(len(test_data)) + ", result); \n"
+ #main_func_str += "computeAccuracy2(labels, " + str(len(test_data)) + ", result); \n"
+ main_func_str += "computeAccuracy3(labels, " + str(len(test_data)) + ", result); \n"
main_func_str += "return 0; \n\n"
main_func_str += "} \n"
diff --git a/llvm/projects/keras/frontend/promise_translator.py b/llvm/projects/keras/frontend/promise_translator.py
index ed8f6edb7cade3a2a9e6b813c65873fc4a03d974..f33770d5ad1bcc3166d8cee5586847790efedcd4 100644
--- a/llvm/projects/keras/frontend/promise_translator.py
+++ b/llvm/projects/keras/frontend/promise_translator.py
@@ -1131,11 +1131,14 @@ class PromiseRtTranslator:
def endBatchLoop(self):
end_loop_str = ""
- end_loop_str += "\nuint8_t* labels = readLabelsBatch(labels_path.c_str(),start,end); \n"
+ #end_loop_str += "\nuint8_t* labels = readLabelsBatch2(labels_path.c_str(),start,end); \n"
+ end_loop_str += "\nuint32_t* labels = readLabelsBatch2(labels_path.c_str(),start,end); \n"
+
last_node = self.dfg.last_node
output_var = self.output_map[last_node.layer_name]
- accuracy_call = "\nfloat accuracy = computeAccuracy2(labels, batch_size, " + output_var + "); \n"
+ #accuracy_call = "\nfloat accuracy = computeAccuracy2(labels, batch_size, " + output_var + "); \n"
+ accuracy_call = "\nfloat accuracy = computeAccuracy3(labels, batch_size, " + output_var + "); \n"
end_loop_str += accuracy_call
#end_loop_str += "float accuracy = computeAccuracy2(labels, batch_size, var_60); "
diff --git a/llvm/projects/keras/frontend/utils.py b/llvm/projects/keras/frontend/utils.py
index c4bbe1cde84732f36fcac46e7f5094223bcbcef0..ffc338c19ea60df7c53430ac38b613c2daef402e 100644
--- a/llvm/projects/keras/frontend/utils.py
+++ b/llvm/projects/keras/frontend/utils.py
@@ -5,16 +5,27 @@
def nodeHasBias(cur_node):
if cur_node.layer_type == "Conv2D" or cur_node.layer_type == "DepthwiseConv2D" or cur_node.layer_type == "Dense":
+ #return True
+ return cur_node.use_bias
+ else:
+ return False
+
+
+def layerHasActivationAttr(cur_node):
+
+ if cur_node.layer_type == "Conv2D" or cur_node.layer_type == "DepthwiseConv2D" \
+ or cur_node.layer_type == "Dense" or cur_node.layer_type == "Activation":
return True
else:
return False
-
+
def nodeHasActivation(cur_node):
if cur_node.layer_type == "Conv2D" or cur_node.layer_type == "DepthwiseConv2D" \
or cur_node.layer_type == "Dense" or cur_node.layer_type == "Activation":
- return True
+ #return True
+ return cur_node.activation_type != "linear"
else:
return False
diff --git a/llvm/projects/keras/frontend/weight_utils.py b/llvm/projects/keras/frontend/weight_utils.py
index 78504bbf03d727211957f6b3d64db749c7093174..db81fa2d70664daef92e98e7230d842359912723 100644
--- a/llvm/projects/keras/frontend/weight_utils.py
+++ b/llvm/projects/keras/frontend/weight_utils.py
@@ -12,9 +12,11 @@ def dumpLabels(file_name, Y_test):
for label in Y_test:
label_val = 0
if len(Y_test.shape) > 1:
- label_val = np.int8(label[0])
+ #label_val = np.int8(label[0])
+ label_val = np.int32(label[0])
else:
- label_val = np.int8(label)
+ #label_val = np.int8(label)
+ label_val = np.int32(label)
if label_val not in labels_map:
labels_map[label_val] = 0