Age | Commit message (Collapse) | Author |
|
Change-Id: I18bbb12d9a61ca8e6f0974712e1c35a99813e45b
|
|
A bug was introduced when stride support was added to strided slice
operators. Fix is to convert stride tensor to four dimensions before
extracting height and width strides.
Change-Id: Ief7159276980613331fa8b5b1df20158ba254f28
Signed-off-by: Rickard Bolin <rickard.bolin@arm.com>
|
|
Change-Id: Ibb1fb718fa35f2224351bef4eb48535f43f244fb
Signed-off-by: Alexander Bengtsson <Alexander.Bengtsson@arm.com>
|
|
Additional testing uncovered two more pieces of code with overflow
errors after updating to the latest version of NumPy. Solution is to
force larger type in the same way as for the other issues.
Change-Id: I40a04e115509055710a70a5767358a4bb8a9cd33
Signed-off-by: Rickard Bolin <rickard.bolin@arm.com>
|
|
Change-Id: Ie6f39d9c4125f7c16d27621de47cd76143c2e636
Signed-off-by: Rickard Bolin <rickard.bolin@arm.com>
|
|
- Added Ethos-U85 support via the --accelerator-config CLI option
Change-Id: Ia3c77dbf61c1b7fa9cb03f8f51d336de2f115a3a
Signed-off-by: Tim Hall <tim.hall@arm.com>
|
|
- Fully connected with batch shape will use the weights
more than once. Models with these type of fully connected
will benefit from weight buffering.
- If a fully connected op with this shape is detected it is
changed to a conv2d and the normal weight buffering
flow will be used.
Change-Id: I272741a32390e036d5e04bd5af41d4538162e86e
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
- When possible, a read slice from a split or stride is moved to
the following op. The problem in this case was that the following
op was a Maxpool op (from Softmax). The Maxpool op is using a
different input shape compared to the original Softmax op, and
this input shape was then changed when the read slice was applied
to the Maxpool op.
- The result is a faulty Maxpool op with an output diff.
- The fix is to prevent moving the slice read when the consumer
input shape differs from the Split/Stride ofm shape
Change-Id: I649d89c38645fa51c20c3602954e2b8af9372076
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
- When possible, a read slice from a split or stride is moved to
the following op. The problem in this case was that the following
op was an elementwise op where the ifm needed to be broadcasted
and that is not supported.
- The result is a faulty elementwise op with an output diff.
- The fix is to prevent moving the slice read to the elementwise op
if broadcasting is needed.
Change-Id: I89928c217510a822f91f051fd1ad6e34040c19de
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
Change-Id: I3c13118e14195a5fb8e522a38b205b75fb07b74b
Signed-off-by: Rickard Bolin <rickard.bolin@arm.com>
|
|
- Added graph optimiser function to convert TRANSPOSE op
into an AvgPool op with swapped stride for height and width
- Added TRANSPOSE supported op check
- Added unit tests for TRANSPOSE supported op check
- Updated SUPPORTED_OPS.md
- Fixed problem in pass packing when optimizing the pass list.
Old problem, but now seen when moving TRANSPOSE from cpu.
Change-Id: I0a0ef420b0fb8241090c2e2434622881105cde15
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
- A reshape followed by an activation function was converted
to a Memcpy with fused activation. The problem is that Memcpy
does not support activation so no activation was executed.
- Added logic to prevent activation functions to be fused
with the Memcpy.
Change-Id: Ibc7d985e5037146dd1f6cb2601407d0f8b865ac6
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
- Added support for stride_h > 3 when ofm height is 1
- Added support for stride_w > 3 when ofm width is 1
- Updated constraints
- Updated tests
- Updated SUPPORTED_OPS.md
Change-Id: I8f89909b05a0f052df5f03702966cee50da61cfc
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
- Added SQUARED_DIFFERENCE support
- Updated SUPPORTED_OPS.md
Change-Id: Id83d9d92129e645390c7979759dfdeff7a14c2ee
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
Only set stride to (1, 1) if kernel, stride and IFM shape all are
equal. And also set padding to VALID to handle ops with SAME padding.
Signed-off-by: Johan Gunnarsson <johan.gunnarsson@arm.com>
Change-Id: Id3cc34686f09667ea21541fac432351555344e3d
|
|
This fixup is not relevant for Resize ops.
Signed-off-by: Johan Gunnarsson <johan.gunnarsson@arm.com>
Change-Id: I81b9d3c8a6dd820b1e5d747d754100282b93c641
|
|
- Support for stride WxH 1x1
- Support for stride WxH 2x1 when IFM and KERNEL
is 1D shape with height 1
- Added test to supported operators
- Updated SUPPORTED_OPS.md
Change-Id: Ic1abead8399a5e14a78d962f8aded0d3b3dbfcc4
Signed-off-by: Johan Alfven <johan.alfven@arm.com>X
|
|
There are networks out there with Pool ops with filter (W, H) equals
IFM (W, H) equals stride (W, H). The stride is technically too large
for the NPU, but we can actually run these ops in the NPU since the
filter is large enough the window doesn't slide. To support these ops
we need to fix the stride so later checks don't put this op on CPU.
Change-Id: I8f0a46b26fb94ee76c33748589536cc5ba07ea59
Signed-off-by: Johan Gunnarsson <johan.gunnarsson@arm.com>
|
|
This convert is already done in the pass packing stage, but doing it
in the graph optimiser stage is better.
Change-Id: Ib9baa98d115cf88491ce39936972a93467a378ce
Signed-off-by: Johan Gunnarsson <johan.gunnarsson@arm.com>
|
|
- npu_performance now uses write/read shapes instead of using ifm/ofms
for memory cycle estimations.
- also fixes a would be bug in the tflite_graph_optimiser, where one
read shape is not Shape4D.
Change-Id: I2067069a713d2cf9e65a5cc227e803de79940fff
Signed-off-by: William Isaksson <william.isaksson@arm.com>
|
|
- Added graph optimiser function to convert convolution groups into
a split followed by separate convolutions and then a concat
- Added semantic check for convolution groups
- Added unit tests for convolution groups semantic checks
- Fixed a minor typing issue with test_constraint_stride_range
Change-Id: I78ade408aa23469a79c9f517c4751da8619b77a9
Signed-off-by: Tim Hall <tim.hall@arm.com>
|
|
If any of H,W axes have shape 1, the IFM can be reshaped to support
reduction over the depth axis.
Signed-off-by: Alexander Hansson <Alexander.Hansson@arm.com>
Change-Id: I432ff1c399b7cee4ca5f0a8f4461e9c0a936d804
|
|
- Add support for batch and depth channels when shape is 1
- Refactor reshaping in convert_mean_to_depthwise_conv
Signed-off-by: Alexander Hansson <Alexander.Hansson@arm.com>
Change-Id: If663395934ab58c76ba92b6ebaaf484a389ae699
|
|
Add function description and type annotations to the optimization
functions missing them.
Fix type annotation issue when re-assigning variable
value to a different type.
Change-Id: I1ee442ff7a29cc07708fdd013430131eff599dd5
Signed-off-by: Raul Farkas <raul.farkas@arm.com>
|
|
* Convert Means with large IFMs to several DeptwiseConv2DBias and Add
operations.
* Update tflite supported operator check with new height and width
constraints.
* Update unit-tests to verify supported operator changes.
* Fix output-diff for 2D IFMs (MLBEDSW-7772)
Signed-off-by: Alexander Hansson <Alexander.Hansson@arm.com>
Change-Id: Ifae6fb1cdac475ae7dac5116c5f13631ff82108a
|
|
* Convert AvgPool with stride_width > 3 and Valid padding to Conv2D to
optimize it to run on NPU.
Change-Id: I06ab412357f0b09b1498f9019a9d1963a324ad34
Signed-off-by: Raul Farkas <raul.farkas@arm.com>
|
|
* Fix bug that caused filter padding to not be added proportionally
compared to the hardware padding added to IFM.
* Update needed_total_padding function that calculates hardware padding
to also account for the cases in which IFM width is not divisible by
the stride width.
* Update supported ops constraint on strides for conv2d to mark ops with
stride width > 3 and IFM width that is not divisible by the
optimization resize factor as not supported.
* Update unit tests that verify correct functionality when checking
whether ops are supported or not.
Change-Id: I62f14cca890b779ca787a9603fa37c873ad522f8
Signed-off-by: Raul Farkas <raul.farkas@arm.com>
|
|
- Added RSQRT int8 support, implemented as LUT.
- Added test to supported operators
- Updated SUPPORTED_OPS.md
Change-Id: I34904772e044be8d22a6dfe426edf85358a205b7
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
- Fixed an issue with the fusing of PAD and AVERAGE_POOL_2D whereby
the rounding away from zero didn't work because it requires the zero
point to be at zero but the input padding required it to be set to the
desired zero point. This affected both int8 and int16. The solution
was to remove it by using the bias prior to the scaling
- Refactored the rounding away from zero mode
Change-Id: I8f2df69df06d2a9722315c346646e5a901cb2c3b
Signed-off-by: Tim Hall <tim.hall@arm.com>
|
|
* Implement a general optimization solution for strided CONV2D that
supports a stride_w with no upper bound.
* Implement filter zero padding to allow for optimization in those cases
in which the filter width is not divisible by the stride width.
E.g.: Filter width = 8, stride width = 3 ->
Filter width = 8 + 1 (0 padding) = 9, stride width = 3
* Implement partial optimization to reduce the stride to hw supported
strides (i.e. 2 and 3) when optimizing to reach a stride = 1 is not
possible due to the IFM width not being divisible by the stride width.
* Implement optimization for when SAME padding is used. If the pre-opt
and post-opt padding do not match, add zero padding to the filter so
that the post-opt IFM padding matches.
Change-Id: Ia66b0d107281fa9993f6bf4d0c26627ee743253b
Signed-off-by: Raul Farkas <raul.farkas@arm.com>
|
|
This reverts commit 72c6a2414205e033279f80b622cdf479c05a4f5b.
Reason for revert: Fix performance regression caused by breaking cascades in certain models
Change-Id: I5aba6e3c59ab27c5129f4a3f0c320ed18df78943
Signed-off-by: Raul Farkas <raul.farkas@arm.com>
|
|
- Added int8 and int16 Exp support, implemented as LUT.
- Added generic 8bit and 16bit LUT table functions following
the implementation in the latest reference. If new ops are added
by the reference, they can easily be implemented in Vela using
the generic functions.
- Moved convert_to_lut to lut.py to have all LUT related code in
one file.
- Updated SUPPORTED_OPS.md
Change-Id: I388e76ea4b39162313599a5341cfb9bad71a782c
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
Mean operators with height larger than 64 are reshaped but the IFM shape
was then reset to the original value, causing an output diff.
Signed-off-by: Rickard Bolin <rickard.bolin@arm.com>
Change-Id: I3a89d4efac53173cbd6fe0a5c0542e028bed42ad
|
|
- Added checking and reporting of missing operator attributes when
reading and writing TFLite file
- Added a TFLite semantic check to ensure that all required attribute
fields of builtin operators are read
- Added some sanity checks for RESHAPE operators that run on the
Ethos-U
- Stopped CPU operators from having their attributes modified
Change-Id: I05700681acdb09554f5945819717c08a9457295c
Signed-off-by: Tim Hall <tim.hall@arm.com>
|
|
- Latest reference has changed implementation for the Mean op
and now only contain one variant.
- Updated Vela implementation to match reference. The full sum
is first calculated and then divided by the numbers of elements.
- Removed the avg pool variant and test case.
- Updated SUPPORTED_OPS.md
Change-Id: I4275e36e3697fa837f119f2cefd7c0ff94231605
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
Added int8 and int16 UNIDIRECTIONAL_SEQUENCE_LSTM support.
The implementation does not include support for:
* CIFG
* Peephole
* Projection
* Normalisation
This change also:
* Removed unused Op.BlockLSTM operation type.
* Removed the only one consumer limitation on putting the SplitSliceRead
on the tensor consumer(s), if all consumers fullfills the requirements
* Added Op.VariableTensorWrite as a Operation.memory_function to make
sure writes to variable tensors:
* Always use linear mode
* Are not moved to fast scratch
* Are not fused with other elementwise operation tensor ranges
Change-Id: Ief831738924ac3d1f2ba6d41f10bd6dc969911f3
Signed-off-by: Fredrik Svedberg <fredrik.svedberg@arm.com>
|
|
- Added 64-bit support for ArgMax
- Updated constraints for ArgMax and regenerated SUPPORTED_OPS.md
Change-Id: I4ef7d2e6fccab0088b87757f6afe40a006c77bbd
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
- Quantization for the OFM was added for the ArgMax operator
as a workaround in order to avoid a crash in the weight compressor.
This quantization is now removed.
- The weight compressor expects that all tensors have a quantization.
Updated code to use scale = 1.0 and zero point = 0 for tensor without
quantization.
Change-Id: I6816dce2db55f7d795d19f88d7fbe7ee419347fc
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
- Updated ARG_MAX to support IFM rank less than 4
- Regenerated SUPPORTED_OPS.md
Change-Id: Icd8e72733279413cbea49021325e1ab06fdc6011
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
Remove op_index constraint and force linear format for all Conv2D that
have strides that can be optimised.
Change-Id: Idef3508ab074ea9abeacac030eaaa15a00ad1211
Signed-off-by: Raul Farkas <raul.farkas@arm.com>
|
|
- Add support for ArgMax along depth dimension with a depth limit of 127.
- Only supports 8-bit input and 32-bit output
Signed-off-by: Rickard Bolin <rickard.bolin@arm.com>
Change-Id: I5f6f0503135bebabbb1ca637f9729587b7c60740
|
|
- The logic when bypassing memory only ops is
complicated and it still does not fix all corner cases.
- This patch simplifies the logic by always bypassing
the op by replacing the IFM with the OFM. If that is not
possible the memory only op is changed to an memcpy op.
- The bypassing was previously done in two steps but
is now reduced to one.
Change-Id: I545dd65e0ec77c70be479a5ada2d277cac3a027c
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
|
|
Changed default behaviour to place int8 ops with asymmetric quantization on cpu, and added an option to force symmetric quantization
Change-Id: Ib9b717aaf61eae78833254ca3dfa745f4f253dc6
Signed-off-by: wilisa01 <william.isaksson@arm.com>
|
|
Swapped order of ifms to add in tflite graph optimiser.
The output diff was caused by the second input tensor being placed on sram, despite there being no dma request to move it there.
Change-Id: I2e83b669ba226c7e96a0bb0d46ba811434cf7bb6
Signed-off-by: wilisa01 <william.isaksson@arm.com>
|
|
Sets second input tensor of resize op to be constant and refactored function.
Signed-off-by: wilisa01 <william.isaksson@arm.com>
Change-Id: I496764f18b4c1ae0fa1a828dd7a90e937a42d41b
|
|
* Extend stride range from (1,3) to (1,4)
* Add stride 4 support when optimising CONV_2D
* Add some tests for various strides
Change-Id: Iddaeb42c4a6e02695ecdd3740bc8b9dd59a7eb3c
Signed-off-by: Raul Farkas <raul.farkas@arm.com>
|
|
- The issue is due to undefined behaviour when casting a NumPy float
to a NumPy unsigned integer which occurs in create_const_tensor()
- The fix is to make sure that the values are first cast to a Python
float
- In addition, the values datatype argument has been removed from
create_const_tensor() to stop the tensor and values datatypes getting
out of sync
Change-Id: I134b9be8c941b361929a5ae7db8cb35f2e9728f2
Signed-off-by: Tim Hall <tim.hall@arm.com>
|
|
Deprecation of some data type aliases in NumPy version 1.24.0 caused Vela
to crash when using Python version 3.8 or above. Replaced the deprecated
aliases.
Signed-off-by: Rickard Bolin <rickard.bolin@arm.com>
Change-Id: Ide167ee864a340194ec5e69537c8718192c78ace
|
|
- Adds missing operators and type conversion recording to DebugDB
Change-Id: If76b0b430bbe73ae1469024c3160ecf0eea26abe
Signed-off-by: wilisa01 <william.isaksson@arm.com>
|
|
- Update copyright notices to use SPDX format and add OSS mail as contact.
- Update years on files where it had been missed.
Signed-off-by: Rickard Bolin <rickard.bolin@arm.com>
Change-Id: I7e9715ea4e17b76252728c708e46df12ad67ab1f
|