Skip to content

Commit a1edcac

Browse files
diaconuccalindiaconuccalin
committed
Post-rebase fix
1 parent 107b893 commit a1edcac

File tree

6 files changed

+0
-403
lines changed

6 files changed

+0
-403
lines changed

‎Deeploy/Targets/Generic/Templates/FloatReduceMeanTemplate.py‎

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,6 @@
88

99

1010
class _FloatReduceMeanTemplate(NodeTemplate):
11-
# WARNING: Currently only supports single axis reducing!
1211

1312
def __init__(self, templateStr):
1413
super().__init__(templateStr)

‎Deeploy/Targets/Generic/Templates/ReduceMeanTemplate.py‎

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,6 @@
88

99

1010
class _ReduceMeanTemplate(NodeTemplate):
11-
# WARNING: Currently only supports single axis reducing!
1211

1312
def __init__(self, templateStr):
1413
super().__init__(templateStr)

‎Deeploy/Targets/Generic/Templates/SliceTemplate.py‎

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -10,7 +10,6 @@
1010

1111

1212
class _SliceTemplate(NodeTemplate):
13-
# WARNING: Currently only supports single axis slicing!
1413

1514
def __init__(self, templateStr):
1615
super().__init__(templateStr)

‎Deeploy/Targets/PULPOpen/Parsers.py‎

Lines changed: 0 additions & 366 deletions
Original file line numberDiff line numberDiff line change
@@ -182,372 +182,6 @@ def parseNodeCtxt(self,
182182
return ctxt, False
183183

184184

185-
class PULPFPDWConv2DParser(Conv2DParser):
186-
187-
def __init__(self, noBiasHoisting = True):
188-
super().__init__(noBiasHoisting)
189-
190-
def parseNode(self, node: gs.Node) -> (bool):
191-
# Parse root conv 2D information
192-
wellFormed = super().parseNode(node)
193-
194-
if wellFormed:
195-
# Check if the node is a depthwise convolution
196-
ret = all([
197-
# Make sure padding is square
198-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][2],
199-
self.operatorRepresentation['pads'][1] == self.operatorRepresentation['pads'][3],
200-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][1],
201-
202-
# Check number of inputs
203-
# 2 inputs if no bias, 3 if layer has bias
204-
len(node.inputs) in [2, 3],
205-
])
206-
207-
# Extract additional attributes
208-
self.operatorRepresentation['padding_y_top'] = int(self.operatorRepresentation['pads'][0])
209-
self.operatorRepresentation['padding_x_left'] = int(self.operatorRepresentation['pads'][1])
210-
self.operatorRepresentation['padding_y_bottom'] = int(self.operatorRepresentation['pads'][2])
211-
self.operatorRepresentation['padding_x_right'] = int(self.operatorRepresentation['pads'][3])
212-
213-
return ret
214-
return False
215-
216-
def parseNodeCtxt(self,
217-
ctxt: NetworkContext,
218-
node: gs.Node,
219-
channels_first: bool = True) -> Tuple[NetworkContext, bool]:
220-
# Parse node context for 2D conv
221-
newCtxt, ret = super().parseNodeCtxt(ctxt, node, channels_first)
222-
223-
if ret:
224-
# Define input names
225-
inputs = ['data_in', 'weight']
226-
227-
# Handle bias, if present
228-
if len(node.inputs) == 2:
229-
self.operatorRepresentation["has_bias"] = "false"
230-
self.operatorRepresentation["bias"] = "NULL"
231-
else:
232-
inputs.append("bias")
233-
self.operatorRepresentation["has_bias"] = "true"
234-
235-
# Map input nodes to operator representation
236-
for idx, inputNode in enumerate(node.inputs):
237-
self.operatorRepresentation[inputs[idx]] = ctxt.lookup(inputNode.name).name
238-
239-
# Check if DW
240-
if self.operatorRepresentation['group'] == self.operatorRepresentation['ch_im_in']:
241-
return newCtxt, True
242-
243-
return ctxt, False
244-
245-
246-
class PULPFPDWConv2DParser(Conv2DParser):
247-
248-
def __init__(self, noBiasHoisting = True):
249-
super().__init__(noBiasHoisting)
250-
251-
def parseNode(self, node: gs.Node) -> (bool):
252-
# Parse root conv 2D information
253-
wellFormed = super().parseNode(node)
254-
255-
if wellFormed:
256-
# Check if the node is a depthwise convolution
257-
ret = all([
258-
# Make sure padding is square
259-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][2],
260-
self.operatorRepresentation['pads'][1] == self.operatorRepresentation['pads'][3],
261-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][1],
262-
263-
# Check number of inputs
264-
# 2 inputs if no bias, 3 if layer has bias
265-
len(node.inputs) in [2, 3],
266-
])
267-
268-
# Extract additional attributes
269-
self.operatorRepresentation['padding_y_top'] = int(self.operatorRepresentation['pads'][0])
270-
self.operatorRepresentation['padding_x_left'] = int(self.operatorRepresentation['pads'][1])
271-
self.operatorRepresentation['padding_y_bottom'] = int(self.operatorRepresentation['pads'][2])
272-
self.operatorRepresentation['padding_x_right'] = int(self.operatorRepresentation['pads'][3])
273-
274-
return ret
275-
return False
276-
277-
def parseNodeCtxt(self,
278-
ctxt: NetworkContext,
279-
node: gs.Node,
280-
channels_first: bool = True) -> Tuple[NetworkContext, bool]:
281-
# Parse node context for 2D conv
282-
newCtxt, ret = super().parseNodeCtxt(ctxt, node, channels_first)
283-
284-
if ret:
285-
# Define input names
286-
inputs = ['data_in', 'weight']
287-
288-
# Handle bias, if present
289-
if len(node.inputs) == 2:
290-
self.operatorRepresentation["has_bias"] = "false"
291-
self.operatorRepresentation["bias"] = "NULL"
292-
else:
293-
inputs.append("bias")
294-
self.operatorRepresentation["has_bias"] = "true"
295-
296-
# Map input nodes to operator representation
297-
for idx, inputNode in enumerate(node.inputs):
298-
self.operatorRepresentation[inputs[idx]] = ctxt.lookup(inputNode.name).name
299-
300-
# Check if DW
301-
if self.operatorRepresentation['group'] == self.operatorRepresentation['ch_im_in']:
302-
return newCtxt, True
303-
304-
return ctxt, False
305-
306-
307-
class PULPFPDWConv2DParser(Conv2DParser):
308-
309-
def __init__(self, noBiasHoisting = True):
310-
super().__init__(noBiasHoisting)
311-
312-
def parseNode(self, node: gs.Node) -> (bool):
313-
# Parse root conv 2D information
314-
wellFormed = super().parseNode(node)
315-
316-
if wellFormed:
317-
# Check if the node is a depthwise convolution
318-
ret = all([
319-
# Make sure padding is square
320-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][2],
321-
self.operatorRepresentation['pads'][1] == self.operatorRepresentation['pads'][3],
322-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][1],
323-
324-
# Check number of inputs
325-
# 2 inputs if no bias, 3 if layer has bias
326-
len(node.inputs) in [2, 3],
327-
])
328-
329-
# Extract additional attributes
330-
self.operatorRepresentation['padding_y_top'] = int(self.operatorRepresentation['pads'][0])
331-
self.operatorRepresentation['padding_x_left'] = int(self.operatorRepresentation['pads'][1])
332-
self.operatorRepresentation['padding_y_bottom'] = int(self.operatorRepresentation['pads'][2])
333-
self.operatorRepresentation['padding_x_right'] = int(self.operatorRepresentation['pads'][3])
334-
335-
return ret
336-
return False
337-
338-
def parseNodeCtxt(self,
339-
ctxt: NetworkContext,
340-
node: gs.Node,
341-
channels_first: bool = True) -> Tuple[NetworkContext, bool]:
342-
# Parse node context for 2D conv
343-
newCtxt, ret = super().parseNodeCtxt(ctxt, node, channels_first)
344-
345-
if ret:
346-
# Define input names
347-
inputs = ['data_in', 'weight']
348-
349-
# Handle bias, if present
350-
if len(node.inputs) == 2:
351-
self.operatorRepresentation["has_bias"] = "false"
352-
self.operatorRepresentation["bias"] = "NULL"
353-
else:
354-
inputs.append("bias")
355-
self.operatorRepresentation["has_bias"] = "true"
356-
357-
# Map input nodes to operator representation
358-
for idx, inputNode in enumerate(node.inputs):
359-
self.operatorRepresentation[inputs[idx]] = ctxt.lookup(inputNode.name).name
360-
361-
# Check if DW
362-
if self.operatorRepresentation['group'] == self.operatorRepresentation['ch_im_in']:
363-
return newCtxt, True
364-
365-
return ctxt, False
366-
367-
368-
class PULPFPDWConv2DParser(Conv2DParser):
369-
370-
def __init__(self, noBiasHoisting = True):
371-
super().__init__(noBiasHoisting)
372-
373-
def parseNode(self, node: gs.Node) -> (bool):
374-
# Parse root conv 2D information
375-
wellFormed = super().parseNode(node)
376-
377-
if wellFormed:
378-
# Check if the node is a depthwise convolution
379-
ret = all([
380-
# Make sure padding is square
381-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][2],
382-
self.operatorRepresentation['pads'][1] == self.operatorRepresentation['pads'][3],
383-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][1],
384-
385-
# Check number of inputs
386-
# 2 inputs if no bias, 3 if layer has bias
387-
len(node.inputs) in [2, 3],
388-
])
389-
390-
# Extract additional attributes
391-
self.operatorRepresentation['padding_y_top'] = int(self.operatorRepresentation['pads'][0])
392-
self.operatorRepresentation['padding_x_left'] = int(self.operatorRepresentation['pads'][1])
393-
self.operatorRepresentation['padding_y_bottom'] = int(self.operatorRepresentation['pads'][2])
394-
self.operatorRepresentation['padding_x_right'] = int(self.operatorRepresentation['pads'][3])
395-
396-
return ret
397-
return False
398-
399-
def parseNodeCtxt(self,
400-
ctxt: NetworkContext,
401-
node: gs.Node,
402-
channels_first: bool = True) -> Tuple[NetworkContext, bool]:
403-
# Parse node context for 2D conv
404-
newCtxt, ret = super().parseNodeCtxt(ctxt, node, channels_first)
405-
406-
if ret:
407-
# Define input names
408-
inputs = ['data_in', 'weight']
409-
410-
# Handle bias, if present
411-
if len(node.inputs) == 2:
412-
self.operatorRepresentation["has_bias"] = "false"
413-
self.operatorRepresentation["bias"] = "NULL"
414-
else:
415-
inputs.append("bias")
416-
self.operatorRepresentation["has_bias"] = "true"
417-
418-
# Map input nodes to operator representation
419-
for idx, inputNode in enumerate(node.inputs):
420-
self.operatorRepresentation[inputs[idx]] = ctxt.lookup(inputNode.name).name
421-
422-
# Check if DW
423-
if self.operatorRepresentation['group'] == self.operatorRepresentation['ch_im_in']:
424-
return newCtxt, True
425-
426-
return ctxt, False
427-
428-
429-
class PULPFPDWConv2DParser(Conv2DParser):
430-
431-
def __init__(self, noBiasHoisting = True):
432-
super().__init__(noBiasHoisting)
433-
434-
def parseNode(self, node: gs.Node) -> (bool):
435-
# Parse root conv 2D information
436-
wellFormed = super().parseNode(node)
437-
438-
if wellFormed:
439-
# Check if the node is a depthwise convolution
440-
ret = all([
441-
# Make sure padding is square
442-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][2],
443-
self.operatorRepresentation['pads'][1] == self.operatorRepresentation['pads'][3],
444-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][1],
445-
446-
# Check number of inputs
447-
# 2 inputs if no bias, 3 if layer has bias
448-
len(node.inputs) in [2, 3],
449-
])
450-
451-
# Extract additional attributes
452-
self.operatorRepresentation['padding_y_top'] = int(self.operatorRepresentation['pads'][0])
453-
self.operatorRepresentation['padding_x_left'] = int(self.operatorRepresentation['pads'][1])
454-
self.operatorRepresentation['padding_y_bottom'] = int(self.operatorRepresentation['pads'][2])
455-
self.operatorRepresentation['padding_x_right'] = int(self.operatorRepresentation['pads'][3])
456-
457-
return ret
458-
return False
459-
460-
def parseNodeCtxt(self,
461-
ctxt: NetworkContext,
462-
node: gs.Node,
463-
channels_first: bool = True) -> Tuple[NetworkContext, bool]:
464-
# Parse node context for 2D conv
465-
newCtxt, ret = super().parseNodeCtxt(ctxt, node, channels_first)
466-
467-
if ret:
468-
# Define input names
469-
inputs = ['data_in', 'weight']
470-
471-
# Handle bias, if present
472-
if len(node.inputs) == 2:
473-
self.operatorRepresentation["has_bias"] = "false"
474-
self.operatorRepresentation["bias"] = "NULL"
475-
else:
476-
inputs.append("bias")
477-
self.operatorRepresentation["has_bias"] = "true"
478-
479-
# Map input nodes to operator representation
480-
for idx, inputNode in enumerate(node.inputs):
481-
self.operatorRepresentation[inputs[idx]] = ctxt.lookup(inputNode.name).name
482-
483-
# Check if DW
484-
if self.operatorRepresentation['group'] == self.operatorRepresentation['ch_im_in']:
485-
return newCtxt, True
486-
487-
return ctxt, False
488-
489-
490-
class PULPFPDWConv2DParser(Conv2DParser):
491-
492-
def __init__(self, noBiasHoisting = True):
493-
super().__init__(noBiasHoisting)
494-
495-
def parseNode(self, node: gs.Node) -> (bool):
496-
# Parse root conv 2D information
497-
wellFormed = super().parseNode(node)
498-
499-
if wellFormed:
500-
# Check if the node is a depthwise convolution
501-
ret = all([
502-
# Make sure padding is square
503-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][2],
504-
self.operatorRepresentation['pads'][1] == self.operatorRepresentation['pads'][3],
505-
self.operatorRepresentation['pads'][0] == self.operatorRepresentation['pads'][1],
506-
507-
# Check number of inputs
508-
# 2 inputs if no bias, 3 if layer has bias
509-
len(node.inputs) in [2, 3],
510-
])
511-
512-
# Extract additional attributes
513-
self.operatorRepresentation['padding_y_top'] = int(self.operatorRepresentation['pads'][0])
514-
self.operatorRepresentation['padding_x_left'] = int(self.operatorRepresentation['pads'][1])
515-
self.operatorRepresentation['padding_y_bottom'] = int(self.operatorRepresentation['pads'][2])
516-
self.operatorRepresentation['padding_x_right'] = int(self.operatorRepresentation['pads'][3])
517-
518-
return ret
519-
return False
520-
521-
def parseNodeCtxt(self,
522-
ctxt: NetworkContext,
523-
node: gs.Node,
524-
channels_first: bool = True) -> Tuple[NetworkContext, bool]:
525-
# Parse node context for 2D conv
526-
newCtxt, ret = super().parseNodeCtxt(ctxt, node, channels_first)
527-
528-
if ret:
529-
# Define input names
530-
inputs = ['data_in', 'weight']
531-
532-
# Handle bias, if present
533-
if len(node.inputs) == 2:
534-
self.operatorRepresentation["has_bias"] = "false"
535-
self.operatorRepresentation["bias"] = "NULL"
536-
else:
537-
inputs.append("bias")
538-
self.operatorRepresentation["has_bias"] = "true"
539-
540-
# Map input nodes to operator representation
541-
for idx, inputNode in enumerate(node.inputs):
542-
self.operatorRepresentation[inputs[idx]] = ctxt.lookup(inputNode.name).name
543-
544-
# Check if DW
545-
if self.operatorRepresentation['group'] == self.operatorRepresentation['ch_im_in']:
546-
return newCtxt, True
547-
548-
return newCtxt, False
549-
550-
551185
class PULPDWConv1DParser(RQSConv1DParser):
552186

553187
def __init__(self, noBiasHoisting = True):

0 commit comments

Comments
 (0)