Float Conv2D, Layernorm, Softmax, Reshape, Div, Relu #27
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
runwangdl
changed the title
runwangdl
force-pushed
the
FloatOperator
branch
3 times, most recently
from
aada6cd to
d6fdb16
Compare
runwangdl
changed the title
runwangdl
force-pushed
the
FloatOperator
branch
2 times, most recently
from
dd6c3d3 to
92efdff
Compare
…n PulpOpen - Float Conv2D, LayerNorm, Div, Reshape, Relu, Softmax basic classes - Add Float Conv2D, LayerNorm, Div, Reshape, Relu on Generic - Add Softmax on PulpOpen
runwangdl
force-pushed
the
FloatOperator
branch
from
92efdff to
a8dea26
Compare
runwangdl
changed the title
runwangdl
commented
Jan 24, 2025
DeeployTest/Platforms/Generic/main.c
Outdated
There was a problem hiding this comment.
Some float test can not pass with diff 1e-5.
There was a problem hiding this comment.
For which operator and with which shape?
Victor-Jung
reviewed
Jan 24, 2025
Deeploy/Targets/Generic/Parsers.py
Outdated
| self.operatorRepresentation['data_in'] = data_in.name | ||
| self.operatorRepresentation['data_out'] = data_out.name | ||
| self.operatorRepresentation['size'] = np.prod(data_in.shape) | ||
| self.operatorRepresentation['lastDimLength'] = data_in.shape[-1] |
There was a problem hiding this comment.
Why do you need to know the last dimension length for RELU?
Deeploy/Targets/Generic/Parsers.py
Outdated
| self.operatorRepresentation[outputs[idx]] = ctxt.lookup(outputNode.name).name | ||
|
|
||
| self.operatorRepresentation['size'] = np.prod(ctxt.lookup(self.operatorRepresentation['input1']).shape) | ||
| self.operatorRepresentation['lastDimLength'] = ctxt.lookup(self.operatorRepresentation['input1']).shape[-1] |
There was a problem hiding this comment.
Same question here, why would you need to know the length of the last dimension for a Div?
Comment on lines
36
to
42
| def alignToContext(self, ctxt: NetworkContext, | ||
| operatorRepresentation: OperatorRepresentation) -> Tuple[NetworkContext, Dict]: | ||
|
|
||
| data_in = ctxt.lookup(operatorRepresentation['data_in']) | ||
| data_out = ctxt.lookup(operatorRepresentation['data_out']) | ||
|
|
||
| return ctxt, operatorRepresentation, [] |
There was a problem hiding this comment.
This method does not do anything, so why do you need it?
Comment on lines
29
to
32
| class _reluTemplate(NodeTemplate): | ||
|
|
||
| def __init__(self, templateStr): | ||
| super().__init__(templateStr) |
There was a problem hiding this comment.
You don't need this, just do `referenceTemplate = NodeTemplate(""""Your template""")
Comment on lines
31
to
42
| class _FloatConvTemplate(NodeTemplate): | ||
|
|
||
| def __init__(self, templateStr): | ||
| super().__init__(templateStr) | ||
|
|
||
| def alignToContext(self, ctxt: NetworkContext, | ||
| operatorRepresentation: OperatorRepresentation) -> Tuple[NetworkContext, Dict, List[str]]: | ||
|
|
||
| data_in = ctxt.lookup(operatorRepresentation['data_in']) | ||
| data_out = ctxt.lookup(operatorRepresentation['data_out']) | ||
|
|
||
| return ctxt, operatorRepresentation, [] |
There was a problem hiding this comment.
Same here, this is class is useless
runwangdl
force-pushed
the
FloatOperator
branch
from
4ead546 to
8942dbd
Compare
1. Delete useless classes in Templates 2. Fix relu dimension bug
Victor-Jung
reviewed
Jan 24, 2025
Deeploy/Targets/Generic/Parsers.py
Outdated
| self.operatorRepresentation['data_in'] = data_in.name | ||
| self.operatorRepresentation['data_out'] = data_out.name | ||
| self.operatorRepresentation['size'] = np.prod(data_in.shape) | ||
| self.operatorRepresentation['batch'] = data_in.shape[0] |
There was a problem hiding this comment.
Why do we need to know the batch for the relu?
Victor-Jung
reviewed
Jan 24, 2025
Comment on lines
31
to
40
| void Relu_fp32_fp32(float32_t* input, float32_t* output, int32_t size, int32_t last_dim_length) { | ||
|
|
||
| int32_t batch_size = size / last_dim_length; | ||
|
|
||
| for (int b = 0; b < batch_size; b++) { | ||
| for (int i = 0; i < last_dim_length; i++) { | ||
| output[b * last_dim_length + i] = MAX(input[b * last_dim_length + i], 0.0f); | ||
| } | ||
| } | ||
| } |
There was a problem hiding this comment.
Relu is an element-wise function (unlike softmax or layernorm, which are row-wise). We don't need the last_dim_length here :)
There was a problem hiding this comment.
Don't forget to align the function signature in the header ;)
Victor-Jung
reviewed
Jan 24, 2025
|
|
||
| referenceTemplate = NodeTemplate(""" | ||
| // Relu (Name: ${nodeName}, Op: ${nodeOp}) | ||
| SINGLE_CORE Relu_fp${data_in_type.referencedType.typeWidth}_fp${data_out_type.referencedType.typeWidth}(${data_in}, ${data_out}, ${size}, ${batch}); |
There was a problem hiding this comment.
Remove the batch here please, we don't need this for relu.
Victor-Jung
reviewed
Jan 24, 2025
| @@ -112,7 +113,8 @@ | |||
| 'RQIntegerDiv': RQIntegerDivLayer([RQIntegerDivMapper]), | |||
| 'MatMul': MatMulLayer([MatMulMapper]), | |||
| 'IntegerMean': ReduceMeanLayer([ReduceMeanMapper]), | |||
| 'iSoftmax': iSoftmaxLayer([Softmax_int8_Mapper]), | |||
| 'iSoftmax': SoftmaxLayer([Softmax_int8_Mapper]), | |||
There was a problem hiding this comment.
FYI: For class name we use PascalCase and for variable names we use camelCase. This is a very NIT comment but try to stick to the syle as much as you can please 😁
runwangdl
force-pushed
the
FloatOperator
branch
from
8942dbd to
934456a
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add Float Conv2D, LayerNorm, Div, Softmax, Reshape, Relu on Generic Platform
Changed:
(Note: nlevel, sign not used for float)
Added:
Deleted:
Issue
PR Merge Checklist
develcommit and pointing todevel.CHANGELOG.mdfile has been updated.